Geometry, Rigidity, and Group Actions
chicago lectures in mathematics series Editors: Spencer J. Bloch, Peter Constan...

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Geometry, Rigidity, and Group Actions

chicago lectures in mathematics series Editors: Spencer J. Bloch, Peter Constantin, Benson Farb, Norman R. Lebovitz, Carlos Kenig, and J. P. May Other Chicago Lectures in Mathematics titles available from the University of Chicago Press Simplical Objects in Algebraic Topology, by J. Peter May (1967, 1993) Fields and Rings, Second Edition, by Irving Kaplansky (1969, 1972) Lie Algebras and Locally Compact Groups, by Irving Kaplansky (1971) Several Complex Variables, by Raghavan Narasimhan (1971) Torsion-Free Modules, by Eben Matlis (1973) Stable Homotopy and Generalised Homology, by J. F. Adams (1974) Rings with Involution, by I. N. Herstein (1976) Theory of Unitary Group Representation, by George V. Mackey (1976) Inﬁnite-Dimensional Optimization and Convexity, by Ivar Ekeland and Thomas Turnbull (1983) Commutative Semigroup Rings, by Robert Gilmer (1984) Navier-Stokes Equations, by Peter Constantin and Ciprian Foias (1988) Essential Results of Functional Analysis, by Robert J. Zimmer (1990) Fuchsian Groups, by Svetlana Katok (1992) Topological Classiﬁcation of Stratiﬁed Spaces, by Shmuel Weinberger (1994) Unstable Modules over the Steenrod Algebra and Sullivan’s Fixed Point Set Conjecture, by Lionel Schwartz (1994) Geometry of Nonpositively Curved Manifolds, by Patrick B. Eberlein (1996) Lectures on Exceptional Lie Groups, by J. F. Adams (1996) Dimension Theory in Dynamical Systems: Contemporary Views and Applications, by Yakov B. Pesin (1997) A Concise Course in Algebraic Topology, by J. P. May (1999) Harmonic Analysis and Partial Differential Equations: Essays in Honor of Alberto Calder´on, edited by Michael Christ, Carlos Kenig, and Cora Sadosky (1999) Topics in Geometric Group Theory, by Pierre de la Harpe (2000) Exterior Differential Systems and Euler-Lagrange Partial Differential Equations, by Robert Bryant, Phillip Grifﬁths, and Daniel Grossman (2003) Ratner’s Theorems on Unipotent Flows, by Dave Witte Morris (2005)

Geometry, Rigidity, and Group Actions Edited by Benson Farb and David Fisher

the university of chicago press • chicago and london

Benson Farb is professor of mathematics at the University of Chicago. He is the author of Problems on Mapping Class Groups and Related Topics and coauthor of Noncommutative Algebra. David Fisher is professor of mathematics at Indiana University.

The University of Chicago Press, Chicago 60637 The University of Chicago Press, Ltd., London © 2011 by The University of Chicago All rights reserved. Published 2011 Printed in the United States of America 20 19 18 17 16 15 14 13 12 11 12345 ISBN-13: 978-0-226-23788-6 (cloth) ISBN-10: 0-226-23788-5 (cloth) Library of Congress Cataloging-in-Publication Data Geometry, rigidity, and group actions/edited by Benson Farb and David Fisher. p. cm.—(Chicago lectures in mathematics series) Festschrift for Robert Zimmer on the occasion of his 60th birthday. ISBN-13: 978-0-226-23788-6 (cloth : alk. paper) ISBN-10: 0-226-23788-5 (cloth : alk. paper) 1. Rigidity (Geometry) 2. Group actions (Mathematics) 3. Manifolds (Mathematics) I. Zimmer, Robert J., 1947– II. Farb, Benson. III. Fisher, David. IV. Series: Chicago lectures in mathematics series. QA613.G465 2011 516 .11—dc22 2010026631 ∞ The paper used in this publication meets the minimum requirements of the American National Standard for Information Sciences—Permanence of Paper for Printed Library Materials, ANSI Z39.48-1992.

To Bob, mentor and friend

Contents

Preface

ix

PART 1 || Group Actions on Manifolds 1.

An Extension Criterion for Lattice Actions on the Circle 3 Marc Burger

2.

Meromorphic Almost Rigid Geometric Structures 32 Sorin Dumitrescu

3.

Harmonic Functions over Group Actions 59 Renato Feres and Emily Ronshausen

4.

Groups Acting on Manifolds: Around the Zimmer Program 72 David Fisher

5.

Can Lattices in SL (n, R) Act on the Circle? 158 Dave Witte Morris

6.

Some Remarks on Area-Preserving Actions of Lattices 208 Pierre Py

7.

Isometric Actions of Simple Groups and Transverse Structures: The Integrable Normal Case 229 Raul Quiroga-Barranco

8.

Some Remarks Inspired by the C0 Zimmer Program 262 Shmuel Weinberger

PART 2 || Analytic, Ergodic, and Measurable Group Theory 9.

Calculus on Nilpotent Lie Groups 285 Michael G. Cowling

viii / contents

10. A Survey of Measured Group Theory 296 Alex Furman

11. On Relative Property (T) 375 Alessandra Iozzi

12. Noncommutative Ergodic Theorems 396 Anders Karlsson and François Ledrappier

13. Cocycle and Orbit Superrigidity for Lattices in SL (n, R) Acting on Homogeneous Spaces 419 Sorin Popa and Stefaan Vaes

PART 3 || Geometric Group Theory 14. Heights on SL2 and Free Subgroups 455 Emmanuel Breuillard

15. Displacing Representations and Orbit Maps 494 Thomas Delzant, Olivier Guichard, François Labourie, and Shahar Mozes

16. Problems on Automorphism Groups of Nonpositively Curved Polyhedral Complexes and Their Lattices 515 Benson Farb, Chris Hruska, and Anne Thomas

17. The Geometry of Twisted Conjugacy Classes in Wreath Products 561 Jennifer Taback and Peter Wong

PART 4 || Group Actions on Representations Varieties 18. Ergodicity of Mapping Class Group Actions on SU(2)-Character Varieties 591 William M. Goldman and Eugene Z. Xia

19. Dynamics of Aut (Fn ) Actions on Group Presentations and Representations 609 Alexander Lubotzky

List of Contributors

645

Preface

In September 2007, a conference was held at the University of Chicago in honor of the sixtieth birthday of Robert J. Zimmer. The conference was a testament to and a celebration of Zimmer’s continuing and lasting inﬂuence on the ﬁelds of geometry, rigidity, and group actions. The wide variety of papers submitted to this volume are just one indication of that inﬂuence. This volume contains both survey papers and research papers. The common theme throughout concerns group actions. The general setup is that we have a group G, which is either a Lie group or a discrete group, acting on some space X . The examples are quite varied, and include, in (informal) order of decreasing “rigidity”: 1. algebraic actions on varieties 2. isometric actions on metric spaces 3. smooth actions on manifolds 4. measurable actions on measure spaces In each instance the goal is to relate algebraic/combinatorial properties of G to geometric/topological/measure-theoretic properties of X and the G action on X . In many cases, properties of G impose surprising restrictions on the possible choices of X and the possible actions on X . The papers by Goldman-Xia and Lubotzky concentrate on (1). The varieties here are representation and character varieties of certain ﬁnitely generated groups , with the automorphism group G = Aut() acting by algebraic automorphisms. This setup can be viewed as a dynamical system, where one can study its dynamical properties (as in Goldman-Xia’s paper), or as an algebraic system (as in Lubotzky’s paper), where information about G can be gleaned from properties of the action. A number of the papers in this volume are concerned with (2). The paper by Farb-Hruska-Thomas considers the case where X is a nonpositively curved

x / preface

simplicial complex and the group G is a lattice in the simplicial automorphism group of X ; here the goal is to extend the classical theory of lattices in algebraic groups over ﬁelds with a discrete valuation to this more general context. Breuillard’s paper provides a strengthening of the famous Tits alternative, and gives further applications to spectral gaps, group growth, and more. Here the G is an arbitrary, ﬁnitely generated subgroup of GL(2, C), and X is a projective space over various ﬁelds. The paper by Delzant-Guichard-Labourie-Mozes considers the general setup of groups acting isometrically on metric spaces. They relate the notion of a “displacing action” to the geometric property of an orbit being quasi-isometrically embedded. This setup is widely applicable, although the case of S-arithmetic groups is emphasized here. Iozzi considers isometric actions of semidirect products G of groups H with (normal) abelian groups A, acting unitarily on a Hilbert space X . The main goal is to give necessary and sufﬁcient conditions for the pair (G, A) to have relative property (T). Taback-Wong use the large-scale geometry of wreath products to study twisted conjugacy classes in these groups. Cowling’s paper develops some calculus on nilpotent groups with a view toward generalizing Mostow’s rigidity theorem concerning groups acting on symmetric spaces. The general study of smooth (usually volume-preserving) actions of (noncompact) Lie groups and their lattices on smooth manifolds goes by the name of the Zimmer program. The paper by Fisher gives a broad survey of this area, while the paper by Morris surveys the already rich case of actions on the circle. Burger’s paper also concerns actions of various large groups on the circle and provides a uniﬁed approach to many of the major results in this area. The Feres-Ronshausen paper studies general properties of groups acting on manifolds without an invariant volume by studying a certain type of harmonic function related to these actions. The primary applications in this paper also concern groups of diffeomorphisms of the circle. Moving up one dimension, Py’s paper provides a brief introduction to an interesting approach to studying volume-preserving diffeomorphisms of compact surfaces using quasi-morphisms. A major theme in the Zimmer program is the study of group actions that preserve geometric structures, a topic represented here in the papers by Dumitrescu and Quiroga-Barranco. While Zimmer’s focus was primarily on rigidity for smooth actions, Weinberger’s paper points to several interesting rigidity phenomena for actions by homeomorphisms. Another major topic of Zimmer’s research concerned cocycles over ergodic actions and orbit equivalence rigidity. Furman’s survey provides a comprehensive overview of this area. The Popa-Vaes paper illustrates some recent developments deepening connections between orbit equivalence rigidity and

preface / xi

operator algebras, and proves some striking new rigidity results. In a very different direction, Karlsson-Ledrappier write a survey on noncommutative ergodic theorems, an area in which cocycles over ergodic actions have played an interesting and important role. Acknowledgment. We would like thank the contributors to this volume, as well as the referees, who generously gave their time to this project.

PART 1

Group Actions on Manifolds

1

marc burger

AN EXTENSION CRITERION FOR LATTICE ACTIONS ON THE CIRCLE to bob zimmer

1. Introduction Let < G be a lattice in a locally compact second countable group G. The aim of this paper is to establish a necessary and sufﬁcient condition for a -action by homeomorphisms of the circle to extend continuously to G. This condition will be expressed in terms of the real bounded Euler class of this action. Combined with classical vanishing theorems in bounded cohomology, one recovers rigidity results of Ghys, Witte-Zimmer, Navas, and Bader-FurmanShaker in a uniﬁed manner. For a survey of various approaches to the problem of classifying lattice actions on the circle we refer to the paper by David Witte Morris [25] in this volume. Let Homeo+ (S1 ) be the group of orientation-preserving homeomorphisms of the circle and e ∈ H 2 (Homeo+ (S1 ), Z) the Euler class; recall that e corresponds to the central extension deﬁned by the universal covering of Homeo+ (S1 ). The Euler class admits a representing cocycle that is bounded and this deﬁnes a bounded class eb ∈ Hb2 (Homeo+ (S1 ), Z) called the bounded Euler class. The relevance of bounded cohomology to the study of group actions on the circle comes from a result of Ghys [14], namely that the bounded Euler class ρ ∗ (eb ) ∈ Hb2 (, Z) of an action ρ : → Homeo+ (S1 ) determines ρ up to quasiconjugation; a quasiconjugation is a self-map of the circle that is weakly cyclic order preserving and in particular not necessarily continuous; see Section 3 for details. If ebR denotes then the bounded class obtained by considering the bounded cocycle deﬁning eb as real valued, we call the invariant ρ ∗ (ebR ) ∈ Hb2 (, R) the real bounded Euler class of ρ. From this point of view we have the following dichotomy (see Proposition 3.2): (E) ρ ∗ (ebR ) = 0: in this case, ρ is quasiconjugated to an action of by rotations; as far as the extension problem is concerned, it reduces to the 3

4 / m. burger

properties of the restriction map Homc (G, R/Z) → Hom(, R/Z). (NE) ρ ∗ (ebR ) = 0: in this case, ρ is quasiconjugated to a minimal unbounded action; that is, every orbit is dense and the group of homeomorphisms ρ() is not equicontinuous. In the ﬁrst case (E) we call ρ elementary and in the second (NE) nonelementary; nonelementary actions are our main object of study in this paper. Concerning the extension problem, an issue that has to be taken care of is the existence of a nontrivial centralizer of the action under consideration. This is illustrated by the following:

example 1.1. Let < PSL(2, R) be a lattice that is nonuniform and torsion free. Since is a free group we can lift the identity to a homomorphism ρk : → PSL(2, R)k ⊂ Homeo+ (S1 ) into the k-fold cyclic covering of PSL(2, R), and this for every k ≥ 1. In this way we get an action that is minimal, unbounded, but for k ≥ 2 does not extend continuously to PSL(2, R). If is torsion-free cocompact, this construction applies provided k divides the Euler characteristic of , which is always the case for k = 2. Thus given a minimal unbounded action one is led to consider its topological S1 -factors; those are easily classiﬁed and in particular there is, up to conjugation, a unique factor ρsp : → Homeo+ (S1 ) that is strongly proximal (Proposition 3.7). This relies on arguments of Ghys that establish that the centralizer of ρ() is a ﬁnite cyclic group; the strongly proximal quotient is then obtained by passing to the quotient by this cyclic group. Our main result is:

theorem 1.2. Let < G be a lattice in a locally compact, second countable group G and ρ : → Homeo+ (S1 ) be a minimal unbounded action. Then the following are equivalent:

an extension criterion for lattice actions on the circle / 5

1) The real bounded Euler class ρ ∗ (ebR ) of ρ is in the image of the restriction map 2 (G, R) → Hb2 (, R). Hbc

2) The strongly proximal factor ρsp of ρ extends continuously to G. The main ingredient in the proof of Theorem 1.2 is a result that, for any countable group , characterizes the bounded classes in Hb2 (, R) obtained from minimal strongly proximal actions in terms of certain cocycles deﬁned on an appropriate Poisson boundary of ; see Theorem 4.5 in Section 4, where the result is proven in the more general context of locally compact, second countable groups. This leads to results of independent interest concerning the extent to which an action is determined by its real bounded Euler class and to the question of the possible values of its norm. These results are summarized in the following theorem and corollaries.

theorem 1.3. Let be a countable group. 1) For a homomorphism ρ : → Homeo+ (S1 ) we have ρ ∗ (ebR ) ≤

1 2

with equality if and only if ρ is quasiconjugated to a minimal strongly proximal action. 2) If two minimal strongly proximal actions ρ1 , ρ2 are not conjugated, then ρ1∗ (ebR ) − ρ2∗ (ebR ) = 1.

remark 1.4. 1) The ﬁrst assertion in Theorem 1.3 echoes results obtained in [7] concerning tight homomorphisms with values in a Lie group of Hermitian type. 2) Let ESP() ⊂ Hb2 (, R) denote the subset consisting of the real bounded Euler classes of minimal strongly proximal -actions and Z[ESP()] its Z-span. We will show (see Section 5) that the norm takes half-integral values on Z[EPS()]. In this context the following question arises, namely if Z[SP()] denotes the free abelian group on the set of conjugacy classes of minimal strongly proximal actions, can one determine the kernel of the homomorphism

6 / m. burger Z[SP()] → Z[ESP()]

ni [ρi ] → ni ρi (ebR ) and what is its signiﬁcance for the dynamics of -actions on S1 ? The following immediate corollary is another instance of the general principle that groups whose second bounded cohomology is ﬁnite-dimensional exhibit rigidity phenomena; compare, for example, with the case of actions by isometries on Hermitian symmetric spaces (see [4], [6]).

corollary 1.5. Let be a countable group and assume that Hb2 (, R) is ﬁnitedimensional. Then there are, up to conjugation, only ﬁnitely many minimal strongly proximal -actions on S1 . Together with the information concerning centralizers of minimal actions we deduce from Theorem 1.3,

corollary 1.6. 1) For a homomorphism ρ : → Homeo+ (S1 ) we have 1 : k∈N , ρ ∗ (ebR ) ∈ {0} ∪ 2k and this value equals (2k)−1 if and only if ρ is quasiconjugated to a minimal unbounded action whose centralizer is of order k. 2) For minimal unbounded actions ρ1 , ρ2 we have ρ1∗ (ebR ) = ρ2∗ (ebR ) if and only if, up to conjugation, ρ2 (γ ) = h(γ ) ρ1 (γ ), γ ∈ where h is a homomorphism with values in the centralizer of ρ1 (). The extension criterion in Theorem 1.2 leads to rigidity theorems when combined with the following two ingredients, namely vanishing theorems in bounded cohomology and the description of continuous homomorphisms from a locally compact group into Homeo+ (S1 ). Concerning the ﬁrst ingredient, we know that the restriction map 2 (G, R) → H 2 (, R) is a isomorphism in the following cases: Hbc b 1) Products (see [8], [21]): G = G1 × · · · × Gn is a Cartesian product of locally compact second countable groups and has dense projection in every factor Gi .

an extension criterion for lattice actions on the circle / 7

2) Higher-rank Lie groups (see [8]): G = G(k) where G is a connected almost simple k-group, k is a local ﬁeld and rankk G ≥ 2. Since it is elementary to classify continuous homomorphisms from a semisimple Lie group over a local ﬁeld into Homeo+ (S1 ) one obtains by combining (1) and (2),

corollary 1.7. ([15], [27]) Let < G := α∈A Gα (kα ) be an irreducible lattice where kα are local ﬁelds, Gα is a connected, simply connected, almost simple kα -group of positive rank. Assume that the sum of the kα -ranks of Gα is at least 2. For a homomorphism ρ : → Homeo+ (S1 ) one of the following holds up to quasiconjugation: 1) ρ has a ﬁnite orbit. 2) ρ is minimal unbounded and its strongly proximal factor ρsp extends continuously to G factoring via the projection on a factor of the form Gα (kα ) = SL(2, R). Concerning locally compact subgroups of Homeo+ (S1 ), one has, owing to the solution of Hilbert’s ﬁfth problem, a wealth of information and in particular, those that are connected and minimal have a simple classiﬁcation; they are up to conjugation, either Rot the subgroup of rotations or PSL(2, R)k , the k-fold cyclic covering of PSL(2, R) (see [10], [16]). When studying continuous homomorphisms from a locally compact group G into Homeo+ (S1 ) one has to deal with the fact that the image is not necessarily closed. In any case we have,

theorem 1.8. Let G be a locally compact group and π : G → Homeo+ (S1 ) a continuous and minimal action. Then one of the following holds: 1) π is conjugated into the group Rot of rotations and has a dense image in it. 2) π surjects onto PSL(2, R)k for some k ≥ 1, up to conjugation. 3) Ker π is an open subgroup of G. From Theorem 1.8 and the vanishing result for products mentioned above we obtain,

corollary 1.9. ([26], [2]) Let G = G1 × · · · × Gn be a product of locally compact second countable groups and < G a lattice with dense projections on each factor Gi . Assume that ρ : → Homeo+ (S1 ) is minimal unbounded. Then the strongly proximal quotient ρsp extends continuously to G,

8 / m. burger

(ρsp )ext : G → Homeo+ (S1 ) and we have one of the following: 1) Ker(ρsp )ext is open in G. 2) Up to conjugation (ρsp )ext factors via a projection onto some factor Gi followed by a continuous surjection onto PSL(2, R) ⊂ Homeo+ (S1 ). Finally, as shown by Bader, Furman, and Shaker, there is also in the context of actions on the circle a commensurator superrigidity theorem, which we state in a way that is somewhat different, but equivalent to their thm. C in [2].

theorem 1.10. ([2]) Let G be a locally compact second countable group, < G a lattice, and < G a subgroup such that ⊂ ⊂ CommG and is dense in G. Let ρ : → Homeo+ (S1 ) be a homomorphism such that ρ() is minimal and unbounded. Then the strongly proximal quotient ρsp extends continuously to G, (ρsp )ext : G → Homeo+ (S1 ) and we have one of the following: 1) Ker(ρsp )ext is open of inﬁnite index in G. 2) (ρsp )ext surjects onto PSL(2, R) ⊂ Homeo+ (S1 ). Let us make the following comment about the hypothesis of Theorem 1.10. If ρ : → Homeo+ (S1 ) is a homomorphism such that ρ() is unbounded, then either ρ() is minimal or there is an exceptional minimal set K ⊂ S1 (see Section 3) that is easily seen to be ρ()-invariant. Thus ρ is quasiconjugated to an action of for which is minimal and unbounded. The above result follows easily from the extension criterion (Theorem 1.2) and the following general fact concerning bounded cohomology:

theorem 1.11. ([20]) Let < G and < < CommG be as in Theorem 1.10, in particular is dense in G. Then the image of the restriction map Hb2 (, R) → Hb2 (, R) 2 (G, R) in H 2 (, R). coincides with the image of Hbc b

We will leave to the interested reader the exercise of deducing Theorem 1.10 from Theorem 1.11 and refer to [20] for elementary proofs of the isomor-

an extension criterion for lattice actions on the circle / 9

phism results for products and higher-rank groups mentioned above, as well as the proof of Theorem 1.11.

acknowledgments: Thanks to Luis Hernandez for his kind invitation to CIMAT where this work was completed. Thanks to Uri Bader and Alex Furman for sharing with me their result on boundary maps. 2. Boundary Maps In this section we present a general existence and uniqueness result concerning measurable equivariant maps that is due to Bader-Furman and is of general interest in rigidity theory. Here and in the sequel G is a second countable locally compact group, G × M → M is a continuous action on a compact metrisable space M, μ ∈ M1 (G) is a spread-out probability measure on G, and (B, νB ) is a standard Lebesgue G-space such that νB is μ-stationary.

theorem 2.1. ([1]) Assume that 1) the G-action is minimal and strongly proximal. 2) For every sequence (gn )n≥1 in G there exists a subsequence (nk )k≥1 such that M→M m −→ gnk (m) converges pointwise. Then every measurable G-equivariant map ϕ : B → M1 (M) takes values in the subset of Dirac measures.

remark 2.2. When G is discrete countable and M = S1 , a result essentially equivalent to Theorem 2.1 has been obtained by Deroin, Kleptsyn, and Navas (see [9]). We recall here that a G-action is strongly proximal if for every μ ∈ M1 (M) the closure Gμ ⊂ M1 (M) contains a Dirac mass. Proof. The main step consists in showing that the G-space M is μ-proximal in the sense of (2.6), def. 3 in [22] VI: this means that for every μ-stationary

10 / m. burger

measure ν ∈ M1 (M) the map 2.1

φν (ω) := lim (x1 , . . . xn ) ν, ω = (xn )n≥1 n→∞

with values in M1 (M), deﬁned almost everywhere on the inﬁnite product (S, μS ) := ∞ n=1 (G, μ), takes values in the set of Dirac measures; this we proceed to show now. The existence of the limit (2.1) follows from the martingale convergence theorem (prop. 2.4 in [22] VI); in addition one has the following remarkable fact ([13], [17] and lemma 1.33 in [11]): lim (x1 . . . xn g) ν = φν (ω)

2.2

n→∞

for μS -almost every ω ∈ S and λ-almost every g ∈ G, where λ is the left Haar measure on G. Fix now a sequence ω = (xn )n≥1 and a subset E ⊂ G of full λ-measure such that (2.2) holds for ω and all g ∈ E. Passing to a subsequence we may assume that the limit s(m) := lim (x1 . . . xn )(m) n→∞

exists for every m ∈ M. Then the map s : M → M, although in general not continuous, admits points of continuity; in fact, they form a Gδ -subset of M. Let then m ∈ M be a point of continuity of s. Since the G-action is minimal and strongly proximal, there exists a sequence (gn )n≥1 in G such that limn→∞ gn ν = δm . By the continuity of the G-action on M there exists for each n ≥ 1 an open neighborhood Un gn such that whenever (hn )n≥1 ∈ n≥1 Un , limn→∞ hn ν = δm . Since λ(Un ) > 0, ∀n ≥ 1, we may choose hn ∈ Un ∩ E and obtain a sequence (hn )n≥1 such that 2.3

a) lim hn ν = δm n→∞

b) lim (x1 . . . xn hk ) ν = φν (ω), for every k ≥ 1 . n→∞

Let V be a neighborhood of s(m) and W m an open neighborhood such that W ⊂ s−1 (V ). Taking into account that s : M → M is a Borel map we have (shk )ν(V ) = (hk ν)(s−1 (V )) ≥ hk ν(W ), which together with lim hk ν(W ) = 1 (see (2.3 a)) implies that k→∞

2.4

lim (shk )(ν) = δs(m) .

k→∞

For every k ≥ 1 and every continuous function f ∈ C(M) we have

an extension criterion for lattice actions on the circle / 11

(shk )ν(f ) =

M

f (shk x) dν(x)

=

lim f (x1 . . . xn hk x) dν(x)

M n→∞

= φν (ω)( f ) where in the second equality we have used the deﬁnition of s and the continuity of f , while in the third we have used dominated convergence and (2.3 b). Thus shk ν = φν (ω) for every k ≥ 1, which together with (2.4) implies that φν (ω) = δs(m) , and hence since ω could be chosen from a subset of full measure, φν takes values in the set of Dirac masses almost everywhere; thus M is μ-proximal. Let now (B, νB ) and ϕ : B → M1 (M) be as in Theorem 2.1. Then ϕ∗ (νB ) ∈ M1 (M1 (M)) is a μ-stationary probability measure on M1 (M) and since M is μ-proximal, prop. 2.9 in [22] VI implies that the support of ϕ∗ (νB ) is contained in the subset of M1 (M) consisting of Dirac masses. This concludes the proof of Theorem 2.1. In the following proposition we give examples of groups of homeomorphisms with the property in Theorem 2.1 (2).

proposition 2.3. Let (G, M) be one of the following pairs consisting of a compact metric space M and a subgroup G of the group of homeomorphisms of M: 1) M = PVk , G = PGL(Vk ) where Vk is a ﬁnite-dimensional vector space over a local ﬁeld k. 2) M = S1 , G = Homeo+ (S1 ). Then for any sequence (gn )n≥1 there is a subsequence such that (gnk )k≥1 converges pointwise. Question 2.4

Let X be a proper complete CAT(0)-space, M = X (∞) its visual boundary, and G = Isom(X ). Does (G, M) satisfy the conclusion of Proposition 2.3? The ﬁrst statement in Proposition 2.3 follows easily by recurrence on dim Vk using Furstenberg’s lemma [12]. Concerning the second, we may compose gn by an appropriate rotation rn such that rn ◦ gn ﬁxes a point in S1 and, passing to a subsequence we may assume that (rn )n≥1 converges uniformly; Proposition 2.3 (2) will then follow from

12 / m. burger

lemma 2.5. Let fn : [0, 1] → [0, 1], n ≥ 1, be a sequence of monotone increasing maps. Then there exists a subsequence converging pointwise. Proof. We may assume that f (r) := limn→∞ fn (r) exists ∀r ∈ Q ∩ [0, 1]. Then f is monotone increasing on Q ∩ [0, 1] and thus F(x) := sup{f (r) : r ≤ x, r ∈ Q ∩ [0, 1]} is a well-deﬁned function on [0, 1] that is monotone increasing and extends f . The set D ⊂ [0, 1] of discontinuities of F is at most countable; passing to a subsequence we may assume then that limn→∞ fn (d) exists for every d ∈ D. Let now t ∈ / D be a point of continuity of F. For every t , t ∈ Q with t ≤ t ≤ t we have F(t ) ≤ ξ ≤ F(t ) where ξ is any accumulation point of the sequence (fn (t))n≥1 . Letting t , t approach t and using the continuity of F we get ξ = F(t) and hence limn→∞ fn (t) = F(t).

3. Elementary Properties of Actions on the Circle In this section we establish the basic dichotomy concerning actions on the circle in terms of their real bounded Euler class, announced in the introduction, and show the uniqueness of the strongly proximal quotient. Along the way we will take the opportunity to recall some well-known facts that will be used later on. Let S1 = Z\R and consider the central extension i

p

+ 1 0 → Z → Homeo+ Z (R) → Homeo (S ) → (e)

where Homeo+ Z (R) is the group of increasing homeomorphisms of R commuting with integer translations and i(n) = T n , where T (x) = x + 1, x ∈ R. A section of p is obtained by associating to every f ∈ Homeo+ (S1 ) the unique lift f : R → R satisfying f (0) ∈ [0, 1). Then we have 3.1

fg T c(f ,g) = f g, f , g ∈ Homeo+ (S1 )

and c(f , g) ∈ {0, 1} is an inhomogeneous 2-cocycle called the Euler cocycle whose class e ∈ H 2 (Homeo+ (S1 ), Z) is called the Euler class. Considering c as a Z-valued, respectively R-valued bounded cocycle leads respectively to the bounded Euler class eb ∈ Hb2 (Homeo+ (S1 ), Z) and the real bounded Euler class ebR ∈ Hb2 (Homeo+ (S1 ), R). Given a group and a homomorphism ρ : → Homeo+ (S1 ) we obtain accordingly three invariants associated to ρ, namely, the Euler class ρ ∗ (e) ∈ H 2 (, Z), the bounded Euler class ρ ∗ (eb ) ∈ Hb2 (, Z), and the real bounded Euler class ρ ∗ (ebR ) ∈ Hb2 (, R). The information

an extension criterion for lattice actions on the circle / 13

contained in ρ ∗ (e) is of purely algebraic nature; ρ ∗ (e) = 0 if and only if ρ lifts to a homomorphism with values in Homeo+ Z (R), which for example is the case when is a free group. Concerning the bounded Euler class we have on the other hand,

theorem 3.1. ([14]) Two actions ρ0 , ρ1 are quasiconjugated if and only if ρ0∗ (eb ) = ρ1∗ (eb ). There are several equivalent ways to look at quasiconjugations. Ghys’s original deﬁnition is as follows: a quasiconjugation is a map h : S1 → S1 induced by a monotone increasing map h : R → R commuting with integer translations; observe that a quasiconjugation is not necessarily continuous. Two actions ρ0 , ρ1 are then quasiconjugated if there is such a map h with 3.2

hρ0 (γ ) = ρ1 (γ ) h, ∀γ ∈ .

Quasiconjugation is an equivalence relation. Observe that if we set (h)+ (x) = limy>x h(y), (h)− (x) = limy<x h(y), then (h)+ and (h)− induce respectively a right continuous h+ and a left continuous h− quasiconjugacy and if (3.2) holds, then we have h± ρ0 (γ ) = ρ1 (γ ) h± , ∀γ ∈ as well. An intrinsic way to deﬁne quasiconjugation rests on the consideration of the orientation cocycle o : (S1 )3 → {−1, 0, 1} deﬁned as follows: ⎧ 1 if x, y, z are positively oriented, ⎪ ⎪ ⎨ 3.3 o(x, y, z) = −1 if x, y, z are negatively oriented, ⎪ ⎪ ⎩ 0 if x, y, z are not pairwise distinct. Then a map h : S1 → S1 is a quasiconjugacy if and only if o(h(x), h(y), h(z)) ≥ 0 whenever o(x, y, z) ≥ 0 . A third characterization is obtained as follows. Deﬁne for x, y ∈ S1 the open interval (x, y) := {z ∈ S1 : o(x, z, y) = 1} and [x, y], (x, y], [x, y) in the obvious way. Given a point b ∈ S1 and a probability measure μ ∈ M1 (S1 ), the map 3.4

hb,μ (x) := μ([b, x)) mod Z

is a left continuous quasiconjugacy and every such quasiconjugacy is obtained this way for a unique pair (b, μ). The behavior under composition with homeomorphisms is particularly simple,

14 / m. burger

hb,μ (ϕ −1 (x)) = hϕ(b),ϕ∗ (μ) (x), ϕ ∈ Homeo+ (S1 ) .

3.5

A very useful tool in the study of group actions on S1 is the following trichotomy for ρ : → Homeo+ (S1 ), see ([18], chap. IV-3)

3.6

1)

There is a ﬁnite orbit and all ﬁnite orbits have the same cardinality.

2)

The action is minimal.

3)

There is a unique proper minimal invariant closed subset F ⊂ S1 , which is a Cantor set. +

In the ﬁrst case ρ is quasiconjugated to an action ρ0 by rotations; a quasiconjugacy is given by hb,μ where μ is the uniform measure along a ﬁnite orbit. Then ρ0 () is ﬁnite cyclic. In the second case ρ() is either bounded (equicontinuous), preserves hence a probability measure μ necessarily with full support and no atoms, and can be conjugated into the subgroup of rotations by the homeomorphism hb,μ ; or ρ() is unbounded. In the third case one can collapse each connected component of S1 \F to a point and obtain this way a continuous quasiconjugacy between ρ and a minimal action. In fact such a quasiconjugacy is given by hb,μ where μ is a probability measure without atoms, and with support, precisely the Cantor set F. With this at hand and denoting as usual rot : Homeo+ (S1 ) → Z\R, the rotation number function, we have

proposition 3.2. Let ρ : → Homeo+ (S1 ) be a homomorphism. 1) ρ ∗ (ebR ) = 0 if and only if ρ is quasiconjugated to a subgroup of the group of rotations. When this is the case, → Z\R γ −→ rot ρ(γ ) is a homomorphism and ρ is quasiconjugated to the action by rotations deﬁned by it. 2) ρ ∗ (ebR ) = 0 if and only if ρ is quasiconjugated to a minimal unbounded action.

an extension criterion for lattice actions on the circle / 15

Proof. 1) Assume ρ ∗ (ebR ) = 0. Then there exists β : → R bounded with dβ(γ , η) := β(γ ) − β(γ η) + β(η) = c(ρ(γ ), s(η)),

∀γ , η ∈ .

Since c( , ) takes values in Z, the map f : → Z\R, γ −→ f (γ ) = β(γ ) mod Z, is a homomorphism. Let ρ0 (γ )(x) := x + f (γ ), x ∈ Z\R be the corresponding action by rotations. Denoting by { } the fractional part of a real number, we then have ρ0 (γ ) (x) = x + {β(x)}, ∀γ ∈ and hence c(ρ0 (γ ), ρ0 (η)) = d({β})(γ , η), ∀γ , η ∈ . This implies for all γ , η ∈ : c(ρ0 (γ ), ρ0 (η)) + d([β])(γ , η) = c(ρ(γ ), ρ(η)) where [β] : → Z is the bounded function associating to γ the integral part [β(γ )] of β(γ ). As a result ρ0∗ (eb ) = ρ ∗ (eb ) and by Ghys’s theorem (Theorem 3.1) ρ and ρ0 are quasiconjugated. Since the rotation number is an invariant of quasiconjugation we have f (γ ) = rot ρ(γ ), ∀γ ∈ and hence γ −→ rot ρ(γ ) is a homomorphism. Conversely, if ρ is quasiconjugated to a homomorphism ρ0 : → Rot, then ρ ∗ (eb ) = ρ0∗ (eb ) and hence ρ ∗ (ebR ) = ρ0∗ (ebR ). But since Rot is compact, ebR |Rot = 0, which implies 0 = ρ0∗ (ebR ) = ρ ∗ (ebR ). 2) If ρ ∗ (ebR ) = 0 then it follows from 1) that ρ is not quasiconjugated into Rot and hence by the trichotomy (3.6) and the discussion following it, ρ is quasiconjugated to a minimal unbounded action. Since bounded cohomology with R-coefﬁcients vanishes for amenable groups we conclude that

corollary 3.3. If is amenable then for any homomorphism ρ : → Homeo+ (S1 ), the map → Z\R γ −→ rot ρ(γ ) is a homomorphism. A simpler proof of the last corollary consists in observing that the rotation number with respect to a -invariant measure coincides with Poincaré’s rotation number and is clearly a group homomorphism. Now we turn to a closer study of minimal unbounded actions. In this context the following deﬁnition will be useful:

16 / m. burger

definition 3.4. A nontrivial S1 -factor of an action ρ : → Homeo+ (S1 ) is a pair (ϕ, ρ0 ) consisting of a nonconstant continuous map ϕ : S1 → S1 and an action ρ0 : → Homeo+ (S1 ) with ϕρ(γ ) = ρ0 (γ )ϕ, ∀γ ∈ . Denoting by ZH+ (L) the centralizer in Homeo+ (S1 ) of a subgroup L we have

theorem 3.5. ([23], [16]) Let ρ : → Homeo+ (S1 ) be minimal unbounded. Then the centralizer Cρ := ZH+ (ρ()) of ρ() in Homeo+ (S1 ) is ﬁnite cyclic and the S1 -factor (ϕ, ρ0 ), where ϕ : S1 → Cρ \S1 is the quotient map and ρ0 the quotient action, is strongly proximal. For minimal circle actions one observes that strong proximality is equivalent to the property that every proper interval can be contracted; we recall that a subset F ⊂ S1 can be contracted if there exists a sequence (γn )n≥1 such that limn→∞ diam(ρ(γn )F) = 0. The idea of the proof of Theorem 3.5 is then as follows: Since ρ is minimal unbounded, in particular not equicontinuous, every point x ∈ S1 has a neighborhood that can be contracted. For every x ∈ S1 the set {I = [x, y)|I can be contracted} is then totally ordered by inclusion, nonempty, and hence contains a unique maximal element that we denote [x, θ(x)). Then one veriﬁes that θ : S1 → S1 is a homeomorphism that is periodic and Cρ = ZH+ (ρ()) = θ . By construction, every proper interval in the quotient Cρ \S1 can be contracted. We record the following immediate

corollary 3.6. Let ρ : → Homeo+ (S1 ) be minimal unbounded. Then ρ is strongly proximal if and only if ZH+ (ρ()) = (e). We now establish the uniqueness of a proximal S1 -factor:

proposition 3.7. Let ρ : → Homeo+ (S1 ) be minimal unbounded and let (ϕ, ρ0 ) be a nontrivial S1 -factor. If ρ0 is strongly proximal, then ϕ is up to conjugation the quotient map by ZH+ (ρ()). First we will need

lemma 3.8. If ρ is minimal and (ρ0 , ϕ) is a nontrivial S1 -factor, then deg ϕ = 0 and ϕ is surjective. Proof. Assume that deg ϕ = 0 and let

ϕ : S1 → R be a lift of ϕ. Then there is for every γ ∈ a c(γ ) ∈ Z such that ϕ.

ϕ ρ(γ ) = T c(γ ) ρ0 (γ )

an extension criterion for lattice actions on the circle / 17

Let M = sup

ϕ (x) and x0 ∈ S1 with M =

ϕ (x0 ). Then x∈S1

ϕ (x) = sup

ϕ (ρ(γ )x)

ϕ (x0 ) = sup

x

x

= c(γ ) + ρ0 (γ ) ( sup

ϕ (x)) x

ϕ (x0 )), = c(γ ) + ρ0 (γ ) (

which implies that ϕ(x0 ) ∈ S1 is ﬁxed by ρ0 (γ ), ∀γ ∈ . The nonempty closed subset ϕ −1 (ϕ(x0 )) being ρ()-invariant equals S1 and thus ϕ is constant. Proof of Proposition 3.7. Let θ be a generator of ZH+ (ρ()) constructed as indicated in the discussion following Theorem 3.5. We claim that ϕθ = ϕ. For x ∈ S1 we consider the closed connected subset Ix := ϕ([x, θ (x)]) and, for a ﬁxed x, distinguish two cases: a) Ix = S1 : then either ϕ(θ(x)) = ϕ(x), which implies by equivariance that ϕθ and ϕ coincide on ρ() · x and hence everywhere, or there is y ∈ (x, θ (x)) with ϕ(x) = ϕ(y) and ϕ([x, y]) = S1 . But by the construction of θ, [x, y] can be contracted, which by continuity of ϕ implies the same for ϕ([x, y]) = S1 ; this is absurd. b) Ix is a proper subset of S1 : Observe that for all γ ∈ , ρ0 (γ )Ix = Iρ(γ )x . Since Ix is a proper interval and ρ0 is strongly proximal we can choose a sequence (γn )n≥1 such that diam (ρ0 (γn )Ix ) tends to zero for n → ∞ and (ρ(γn )x)n≥1 converges to some point y, which by continuity of ϕ implies that ϕ(y) = ϕθ(y). Again by using equivariance and minimality we get ϕ = ϕθ. Thus the claim is established and therefore by passing to the quotient by ZH+ (ρ()) we may assume that ρ is strongly proximal as well. We already

know (Lemma 3.8) that deg ϕ = 0 and ϕ is surjective. Assume then that there exists s = t in S1 with ϕ(s) = ϕ(t). Since deg ϕ = 0 we have either ϕ([s, t]) = S1 or ϕ([t, s]) = S1 . Assume without loss of generality that the former occurs. Since [s, t] is a proper interval in S1 it can be contracted by ρ, which clearly implies a contradiction. Thus ϕ is injective as well and hence a homeomorphism.

4. Strongly Proximal Actions, Boundary Maps, and Cocycles In this section we study continuous group actions of a locally compact group G on the circle using boundary maps. The main goal is Theorem 4.5, which

18 / m. burger

gives a description of the space of conjugacy classes of minimal strongly proximal actions in terms of an explicit space of cocycles on appropriate Poisson boundaries of G. 4.1. Cohomological Preliminaries

For a nondiscrete group some care must be taken when deﬁning the analogues of the various Euler classes in the continuous context. Let G be locally compact, • (G, A) denotes the second countable, and let A be either Z or R. Then Hbc cohomology deﬁned via A-valued bounded Borel cochains on G; in the case A = R these cohomology groups coincide with those obtained by taking the subcomplex of bounded continuous cochains on G; for more details see [5], §2.3. Given a continuous action ρ : G → Homeo+ (S1 ) the function (g, h) −→ c(ρ(g), ρ(h)) is a bounded Borel (inhomogeneous) 2-cocycle on G and leads 2 (G, Z) and ρ ∗ (eb ) ∈ H 2 (G, R), which are respectively to classes ρ ∗ (eb ) ∈ Hbc R bc the bounded Euler class and the real bounded Euler class of the continuous action. Recall that if (B, νB ) is a standard Lebesgue G-space that is doubly ergodic and amenable, there is a canonical isometric isomorphism; see [24] thm. 7.5.3, 4.1

2 ∞ 3 Hbc (G, R) Z Lalt (B , R)G

where the right-hand side is the space of measurable, essentially bounded, alternating G-invariant cocycles on B3 .

proposition 4.1. Let ρ : G → Homeo+ (S1 ) be a continuous action and assume that there exists a measurable G-equivariant map ϕ : B → S1 . Then under the isomorphism in (4.1) the class ρ ∗ (ebR ) corresponds to the cocycle B3 → R 1 2

(x, y, z) −→ − o(ϕ(x), ϕ(y), ϕ(z)) . If in addition the essential image of ϕ contains at least three points, we have ρ ∗ (ebR ) =

1 . 2

We will need the following explicit relationship between the Euler and orientation cocycle; see for instance [19] Lemma 2.1.

lemma 4.2. For every f , g ∈ Homeo+ (S1 ) we have

an extension criterion for lattice actions on the circle / 19

2c( f , g) = − o(˙o, f (˙o), f g(˙o)) + 1 + (δo˙ ( f g(˙o)) − δo˙ ( f (˙o)) − δo˙ (g(˙o))), where o˙ ∈ Z\R. Proof of Proposition 4.1. This is a standard argument using [3]. Since G acts ∞ ((S1 )•+1 , R), d) of bounded alternatcontinuously on S1 , the G-complex (Balt 1 ing Borel cochains on S is a strong resolution of the trivial G-module R (see [3] Prop. 2.1). Then, owing to the properness of the G-action on G and the amenability of the G-action on B, the two complexes of G-modules ∞ (G•+1 , R), d) and (L∞ (B•+1 , R), d) are strong resolutions of R by relatively (Lalt alt injective G-Banach modules [24] section 7.5. Then there is a morphism of G-complexes 4.2

∞ •+1 ∞ •+1 c • : Lalt (G , R) → Lalt (B , R)

extending the identity and any two such are G-equivariantly homotopic; the canonical map induced in cohomology is then an isometric isomorphism that in degree two gives the one mentioned in (4.1). Consider the morphisms of G-resolutions ∞ 1 n+1 ∞ n+1 (G , R) ρ (n) : Balt (S ) , R) → Lalt

and ∞ ∞ n+1 (B , R) ϕ (n) : Balt ((S1 )n+1 , R) → Lalt

deﬁned respectively for α : (S1 )n+1 → R by ρ (n) (α)(g0 , . . . , gn ) := α(ρ(g0 )˙o, . . . , ρ(gn ) o˙ ) and ϕ (n) (α)(x0 , . . . , xn ) := α(ϕ(x0 ), . . . , ϕ(xn )) . Then c (•) ρ (•) and ϕ (•) are both morphisms of G-complexes extending the iden∞ (B•+1 , R), d) is a resolution by relatively injective tity and therefore, since (Lalt G-modules, induce the same map in cohomology. Now we specialize this ∞ (S•+1 )) be the class deﬁned by the orito degree n = 2. Let [o] ∈ H 2 (Balt entation cocycle. Then we have by Lemma 4.2: ρ (2) ([o]) = 2ρ ∗ (ebR ) and in addition that under the isometric isomorphism c (2) , ρ (2) ([o]) is represented by (x, y, z) −→ o(ϕ(x), ϕ(y), ϕ(z)). In particular 2ρ ∗ (ebR ) is the essential supremum of the latter cocycle.

20 / m. burger

We will see below that the existence of a map as in Proposition 4.1 imposes strong conditions on the type of action considered. Before we turn to this we will need

lemma 4.3. Let ρ0 , ρ1 : G → Homeo+ (S1 ) be continuous actions and ϕ : S1 → S1 a degree k covering such that ρ0 (g) ϕ = ϕ ρ1 (g),

∀g ∈ G .

Then we have the following equality ρ0∗ (eb ) = k ρ1∗ (eb ) 2 (G, Z). in Hbc

Proof. Let p : R → Z\R denote projection and Mk : R → R, x −→ k · x, so that up to conjugation we have p Mk = ϕp. Then we have ∀g ∈ G: p Mk ρ1 (g) = ϕ p ρ1 (g) = ϕ ρ1 (g) p = ρ0 (g) ϕp = ρ0 (g) p Mk = p ρ0 (g) Mk . Therefore there is α(g) ∈ Z such that 4.3

Mk ρ1 (g) = T α(g) ρ0 (g) Mk .

Evaluation at 0 ∈ R gives k ρ1 (g) (0) = α(g) + ρ0 (g) (0), which ﬁrst shows that α : G → Z is a Borel function and furthermore implies that supg∈G |α(g)| ≤ k + 1. Applying (4.3) to products and using the relation deﬁning the Euler cocycle we get k c(ρ1 (g), ρ1 (h)) + α(gh) − α(g) − α(h) = c(ρ0 (g), ρ0 (h)), which proves the lemma.

Now we come to our ﬁrst application.

corollary 4.4. Let μ ∈ M1 (G) be a spread-out probability measure on G and (B, νB ) a standard Lebesgue G-space, where νB is μ-stationary. Assume that B is doubly ergodic and amenable. Given a minimal unbounded action ρ : G → Homeo+ (S1 ) the following are equivalent:

an extension criterion for lattice actions on the circle / 21

1) ρ is strongly proximal. 2) There exists a G-equivariant measurable map ϕ : B → S1 . 3) ρ ∗ (ebR ) = 12 . Proof. (1) =⇒ (2): follows immediately from Theorem 2.1 and Proposition 2.3 (3). (2) =⇒ (3): follows from Proposition 4.1. (3) =⇒ (1): Let (ψ, ρ0 ) be the strongly proximal factor of ρ. Applying the implication (1) =⇒ (3) to ρ0 we get ρ0∗ (ebR ) = 12 , and for k = deg ψ, we get from Lemma 4.3 that ρ0∗ (ebR ) = k · ρ ∗ (ebR ), which implies k = 1 and hence that ρ is strongly proximal. 4.2. Minimal Strongly Proximal Actions and Cocycles

The main result of this section is

theorem 4.5. Let G be locally compact second countable and let μ ∈ M1 (G), (B, νB ) be as in Corollary 4.4. Then there is a bijection between a) The set of conjugacy classes of continuous G-actions on S1 that are minimal and strongly proximal. b) The set of measurable functions, up to equality almost everywhere, ω : B3 → R such that 1) ω is an alternating strict cocycle. 2) ω(gx, gy, gz) = ω(x, y, z) for every g ∈ G and almost every (x, y, z) ∈ B3 . 3) ω takes values in {±1} almost everywhere. This bijection is implemented by the map to which a continuous minimal strongly proximal action ρ : G → Homeo+ (S1 ) associates the cocycle ω(x, y, z) = o(ϕ(x), ϕ(y), ϕ(z)) where ϕ : B → S1 is the map given by Corollary 4.4. The proof of the theorem is divided into two steps. Fix ω : B3 → R satisfying the properties (1), (2), and (3) above. Step 1: We show that there exists a measurable map ϕ : B → S1 satisfying the following properties:

22 / m. burger

a) ϕ∗ (νB ) ∈ M1 (S1 ) has no atoms. b) the essential image Ess Im ϕ of ϕ equals S1 . c) o(ϕ(x), ϕ(y), ϕ(z)) = ω(x, y, z) for almost every (x, y, z) ∈ B3 . Step 2: Given a measurable function ϕ : B → S1 satisfying (a), (b), and (c) we construct a continuous homomorphism πϕ : G → Homeo+ (S1 ) such that ϕ(gx) = πϕ (g)(ϕ(x)), for every g ∈ G and almost every x ∈ B, and conclude that πϕ is minimal strongly proximal. Finally we show that given ϕ, ψ satisfying (a), (b), and (c), then πϕ and πψ are conjugate. Combining Corollary 4.4, Step 1, and Step 2 clearly completes the proof of Theorem 4.5. 4.3. Step 1: The Construction of the Measurable Map

We ﬁx once and for all a measurable map ω : B3 → R satisfying properties (1), (2), and (3) in Theorem 4.5. For every (x, y) ∈ B2 deﬁne I(x, y) = {z ∈ B : ω(x, z, y) = 1}, which is measurable. The next few lemmas are intended to show that the sets I(x, y) behave like intervals on S1 . We will use the notation E ≡ F to indicate that two sets E, F differ by a set of measure zero; similarly E ⊂ F means that E is contained in F up to a set of measure zero.

•

lemma 4.6. 1) I(x, y) ∪ I(y, x) ≡ B for a.e. (x, y) ∈ B2 . 2) I(x, y) ∩ I(y, x) = φ, ∀(x, y) ∈ B2 . 3) ∀g ∈ G and a.e. (x, y) ∈ B2 , g(I(x, y)) ≡ I(gx, gy). 4) For a.e. (x, y) ∈ B2 , 0 < νB (I(x, y)) < 1. Proof. 1) follows from the fact that ω is alternating and ω(x, y, z) ∈ {±1} for a.e. (x, y, z) ∈ B3 . 2) is obvious. 3) follows from the G-invariance of ω and the fact that the G-action on B preserves the measure class of νB . 4) Consider A1 = {(x, y) ∈ B2 : νB (I(x, y)) = 1} A0 = {(x, y) ∈ B2 : νB (I(x, y)) = 0} .

an extension criterion for lattice actions on the circle / 23

Denoting ∧ the map (x, y) → (y, x), we have from (1) that A1 ≡ A0 ; in addition we have that A1 ∩ A0 = φ and both sets are measurable and G-invariant by (3). Thus if νB2 (A1 ) > 0, then νB2 (A0 ) = νB2 ( A0 ) = νB2 (A1 ) > 0, which contradicts 2 the ergodicity of the G-action on B . Thus A1 and A0 are both of measure zero.

lemma 4.7. 1) For (a, c) ∈ B2 and b ∈ I(a, c) we have 4.4

I(a, c) ⊃ I(a, b) ∪ I(b, c)

4.5

I(a, b) ∩ I(b, c) = φ .

2) For a.e. (a, c) ∈ B2 and a.e. b ∈ I(a, c) 4.6

I(a, c) ⊂ I(a, b) ∪ I(b, c) . •

In particular, for a.e. (a, c) ∈ 4.7

B2

and a.e. b ∈ I(a, c)

I(a, c) ≡ I(a, b) ∪ I(b, c) .

Proof. Applying the cocycle identity to a, x, b, c ∈ B we have 4.8

ω(x, b, c) − ω(a, b, c) + ω(a, x, c) − ω(a, x, b) = 0 .

If b ∈ I(a, c) and x ∈ I(a, b), we have ω(a, b, c) = 1, ω(a, x, b) = 1, which with (4.8) implies ω(x, b, c) + ω(a, x, c) = 2 and hence x ∈ I(a, c). If x ∈ I(b, c) then (4.8) gives ω(a, x, c) − ω(a, x, b) = 2 and hence x ∈ I(a, c). This shows the inclusion (4.4). Concerning (4.5), we apply (4.8) to x ∈ I(a, b) ∩ I(b, c) to get a contradiction. Concerning (4.6), assume that b ∈ I(a, c) and x ∈ I(a, c) and apply (4.8) to get 4.9

ω(x, b, c) = ω(a, x, b).

For a.e. (a, b) ∈ B2 and a.e. x ∈ B we have either ω(a, x, b) = 1 and hence x ∈ I(a, b) or ω(a, x, b) = −1, which with (4.9) gives x ∈ I(b, c). Lemma 4.6 (4) and Lemma 4.7 imply then immediately

lemma 4.8. For a.e. (a, x, y) ∈ B3 we have the following dichotomy: 1) x ∈ I(a, y), I(a, y) ≡ I(a, x) ∪ I(x, y) and νB (I(a, x)) < νB ((a, y)). 2) y ∈ I(a, x), I(a, x) ≡ I(a, y) ∪ I(y, x) and νB (I(a, y)) < νB ((a, x)).

24 / m. burger

Given a ∈ B, we deﬁne fa : B → Z\R by fa (x) := νB (I(a, x)) mod Z . Then we have

lemma 4.9. For a.e. a ∈ B we have 1) E = {(x, y) ∈ B2 : fa (x) = fa (y)} is of measure zero. 2) (fa )∗ (νB ) has no atoms. Proof. 1) If (x, y) ∈ E then either νB (I(a, x)) = 0 and νB (I(a, y)) = 1, or νB (I(a, x)) = νB ((a, y)); in both cases these equalities hold only for a set of (a, x, y)’s of measure zero; in the ﬁrst case this follows from Lemma 4.6 (4) and in the second from Lemma 4.8. 2) follows from (1).

lemma 4.10. For a.e. a ∈ B and a.e. (x, y, z) ∈ B3 o( fa (x), fa (y), fa (z)) = ω(x, y, z) . Proof. Observe that the left-hand side gives a measurable alternating cocycle taking values in {−1, 0, 1}; it follows from Lemma 4.9 that it takes values in {−1, 1} almost everywhere. It therefore sufﬁces to show that if the left-hand side equals 1, then so does the right-hand side. If o( fa (x), fa (y), fa (z)) = 1 then up to cyclically permuting the variables we may assume, using the deﬁnition of fa , that νB (I(a, x)) < νB (I(a, y)) < νB (I(a, z)). Together with Lemma 4.8 this implies x ∈ I(a, y) and y ∈ I(a, z) except possibly for a set of (a, x, y, z)’s of measure zero. Using the cocycle identity for ω applied to (a, x, y, z) we get ω(x, y, z) + ω(a, x, z) = 2 and hence ω(x, y, z) = 1. The following lemma will be left to the reader:

lemma 4.11. Let ξ ∈ M1 (S1 ) be a probability measure without atoms and hξ : S1 → S1 x −→ ξ ([˙o, x)) mod Z the associated quasiconjugacy. Then, 1) hξ is continuous. 2) the measure (hξ )∗ (ξ ) has no atoms and its support equals S1 .

an extension criterion for lattice actions on the circle / 25

3) for ξ 3 -almost every (x, y, z) ∈ (S1 )3 we have o(hξ (x), hξ (y), hξ (z)) = o(x, y, z). Combining Lemma 4.9, 4.10, and 4.11 we obtain the following lemma, which completes Step 1:

lemma 4.12. For a.e. a ∈ B, let ξ = (fa )∗ (νB ) and deﬁne ϕa := hξ ◦ fa . Then 1) (ϕa )∗ (νB ) has no atoms. 2) Ess Im ϕa = S1 . 3) o(ϕa (x), ϕa (y), ϕa (z)) = ω(x, y, z) for a.e. (x, y, z) ∈ B3 . 4.4. Step 2: The Construction of the Action

The following result is the basis for our construction:

lemma 4.13. Let ϕ, ψ : B → S1 be measurable maps such that 1) ϕ∗ (νB ) and ψ∗ (νB ) have no atoms. 2) Ess Im ϕ = Ess Im ψ = S1 . 3) o(ϕ(x), ϕ(y), ϕ(z)) = o(ψ(x), ψ(y), ψ(z)) for a.e. (x, y, z) ∈ B3 . Then the essential image F ⊂ S1 × S1 of the map B → S1 × S1 x −→ (ϕ(x), ψ(x)) is the graph of an orientation-preserving homeomorphism h : S1 → S1 and, h(ϕ(x)) = ψ(x) for a.e. x ∈ B . Proof. The proof is in two steps.

claim 1: Given (ξ1 , η1 ), (ξ2 , η2 ), (ξ3 , η3 ) in F such that ξ1 , ξ2 , ξ3 are pairwise distinct and the same holds for η1 , η2 , η3 , we have o(ξ1 , ξ2 , ξ3 ) = o(η1 , η2 , η3 ) . Pick Vi ξi and Wi ηi open intervals such that V1 , V2 , V3 are pairwise disjoint and the same holds for W1 , W2 , W3 . Then Bi = {x ∈ B : ϕ(x) ∈ Vi and ψ(x) ∈ Wi } is of positive measure, i = 1, 2, 3.

26 / m. burger

We then have for a.e. (x, y, z) ∈ B1 × B2 × B3 , o(ξ1 , ξ2 , ξ3 ) = o(ϕ(x1 ), ϕ(x2 ), ϕ(x3 )) = o(ψ(x1 ), ψ(x2 ), ψ(x3 )) = o(η1 , η2 , η3 ) .

claim 2: Let (ξ1 , η1 ), (ξ2 , η2 ) be in F. Then ξ1 = ξ2 iff η1 = η2 . Assume that ξ1 = ξ2 . We will repeatedly use the fact that (ϕ(x), ψ(x)) ∈ F for a.e. x ∈ B .

4.10

Since Ess Im ϕ = S1 and ξ1 = ξ2 the set ϕ −1 ((ξ1 , ξ2 )) is of positive measure; then in virtue of (4.10) and the hypothesis that ψ∗ (νB ) has no atoms, we can pick t ∈ ϕ −1 ((ξ1 , ξ2 )) with (ϕ(t), ψ(t)) ∈ F and ψ(t) ∈ / {η1 , η2 }. Similarly we can ﬁnd s ∈ ϕ −1 ((ξ2 , ξ1 )) such that (ϕ(s), ψ(s)) ∈ F and ψ(s) ∈ / {ψ(t), η1 , η2 }. Then the cocycle identity applied to (η1 , ψ(t), η2 , ψ(s)} gives o(ψ(t), η2 , ψ(s)) − o(η1 , η2 , ψ(s)) + o(η1 , ψ(t), ψ(s)) − o(η1 , ψ(t), η2 ) = 0 . Applying Claim 1 to the ﬁrst and third terms we get o(ϕ(t), ξ2 , ϕ(s)) − o(η1 , η2 , ψ(s)) + o(ξ1 , ϕ(t), ϕ(s)) − o(η1 , ψ(t), η2 ) = 0 and by our choices we have o(ϕ(t), ξ2 , ϕ(s)) = 1 o(ξ1 , ϕ(t), ϕ(s)) = 1, which implies that o(η1 , η2 , ψ(s)) + o(η1 , ψ(t), η2 ) = 2 and hence η1 = η2 . This shows Claim 2 modulo interchanging the roles of ϕ and ψ. Since pri (F) = S1 we get from Claim 2 that F is the graph of a homeomorphism h, which by Claim 1 is orientation preserving. Finally (4.10) says precisely that h(ϕ(x)) = ψ(x) for a.e. x ∈ B and this completes the proof of the lemma. Now we return to our cocycle ω : B3 → R satisfying the hypothesis of Theorem 4.5 and let ϕ : B → S1 be a measurable map as given by Lemma 4.12. Since ω is G-invariant we can apply Lemma 4.13 to ϕ and ϕg, g ∈ G to obtain an orientation-preserving homeomorphism denoted πϕ (g) ∈ Homeo+ (S1 ) satisfying πϕ (g)(ϕ(x)) = ϕ(gx) for a.e. x ∈ B .

an extension criterion for lattice actions on the circle / 27

This equivariance property implies that πϕ : G → Homeo+ (S1 ) is a homomorphism.

lemma 4.14. 1) πϕ : G → Homeo+ (S1 ) is continuous. It is minimal, unbounded, and strongly proximal. 2) Let ϕ, ψ be as in Lemma 4.13, and πϕ , πψ the corresponding homomorphisms. Let h : S1 → S1 be the homeomorphism given by Lemma 4.13. Then, h πϕ (g) = πψ (g) h

∀g ∈ G.

Proof. Assertion 2) is clear and follows from the various equivariance properties and thus we concentrate on 1). For a.e. x ∈ B, the map g −→ πϕ (g)(ϕ(x)) = ϕ(gx) is measurable and hence the homomorphism πϕ : G → Homeo+ (S1 ) is measurable. Since G is locally compact second countable and since Homeo+ (S1 ) is second countable we deduce that πϕ is continuous. Using Proposition 4.1 we see that πϕ∗ (ebR ) = 0 and hence πϕ is nonelementary. Assume that there is an exceptional minimal set K ⊂ S1 . Let J be a connected component if S1 \K ; since Ess Im ϕ = S1 , +

ϕ −1 ( J) is of positive measure but not of full measure in B. Since ∀g ∈ B we have either πϕ (g) J = J or πϕ (g) J ∩ J = φ, we conclude that either gϕ −1 ( J) ≡ ϕ −1 ( J) or gϕ −1 ( J) ∩ ϕ −1 ( J) = φ. Since the G-action on B is ergodic we have −1 −1 −1 −1 1 g∈G gϕ ( J) = ϕ (S \K ) ≡ B and hence there is g0 ∈ B with g0 ϕ ( J) ∩ −1 ϕ ( J) = φ. Thus h∈G h(ϕ −1 ( J) × ϕ −1 ( J)) is of positive measure in B2 , G-invariant, and does not meet g0 ϕ −1 ( J) × ϕ −1 ( J), itself also of positive measure, this contradicts the ergodicity of the G-action on B × B. Thus πϕ is minimal unbounded. Finally it follows from Corollary 4.4 that πϕ is strongly proximal.

5. Proofs of Theorems 1.2 and 1.3, Remark 1.4(2), and Corollaries 1.5 and 1.6 In order to apply the results obtained so far we recall

theorem 5.1. ([21]) Let G be a locally compact second countable group and μ ∈ M1 (G) a symmetric spread-out probability measure on G. Then the G-action on the associated Poisson

28 / m. burger

boundary (B, νB ) is amenable, doubly ergodic; the same properties hold for the action of a lattice < G. Using Theorem 5.1 for = G and applying Corollary 4.4, Theorem 4.5, and Proposition 4.1 implies readily Theorem 1.3. Concerning Remark 1.4 (2), if ρ1 , . . . , ρn are minimal strongly proximal actions and ωi : B3 → {−1, 1}

are the associated cocycle, then for any ni ∈ Z, ni=1 ni ωi takes integral values and hence by the isomorphism in (4.1) and Proposition 4.1 the norm of

n ∗ b i=1 ni ρi (eR ) is half integral. Corollary 1.6 (1) follows from Proposition 3.2, Proposition 4.1, and Lemma 4.3. Corollary 1.6 (2) follows from Theorem 1.3 (2), which implies that ρ1 , ρ2 have modulo conjugation the same strongly proximal quotient. We turn now to the proof of Theorem 1.2. First assume that for the given minimal unbounded action ρ, ρsp extends continuously. Let k = |ZH+ (ρ())|. Then we have that (ρsp )∗ (ebR ) = k · ρ ∗ (ebR ) (Lemma 4.3) and since (ρsp )∗ (ebR ) is in the image of the restriction map, so is ρ ∗ (ebR ). Conversely, apply Theorem 5.1 and let μ ∈ M1 () be a probability measure of full support such that νB is μ -stationary (see [22]). On the space (B, νB ) the restriction map Hb2 (G, R) → Hb2 (, R) is realized by the inclusion 5.1

∞ 3 ∞ 3 Z Lalt (B , R)G → Z Lalt (B , R) .

Now if ρ ∗ (ebR ) is in the image of the restriction map then so is (ρsp )∗ (ebR ). Let ωρsp : B3 → R be the cocycle associated to ρsp ; using (5.1) we get that ωρsp is G-invariant, satisﬁes all hypotheses of Theorem 4.5 and hence corresponds to a continuous minimal strongly proximal homomorphism π : G → Homeo+ (S1 ). Since ωπ | clearly coincides with ωρsp we conclude from Theorem 4.5 that π | and ρsp are conjugate.

6. Locally Compact Groups Acting on S1 In this section we prove Theorem 1.8 using some results from [16] and [10]. In the sequel, Rot denotes as usual the group of rotations of S1 and, for every k ≥ 1, PSL(2, R)k ⊂ Homeo+ (S1 ) is the k-fold cyclic covering group of PSL(2, R). We record the following

an extension criterion for lattice actions on the circle / 29

lemma 6.1. Let H be connected, semisimple, with ﬁnite center and no compact factors, and π : H → Homeo+ (S1 )

6.1

a continuous nontrivial homomorphism. Then there exists k ∈ N such that up to conjugation, π (H) = PSL(2, R)k . For the following lemma we recall that given a subgroup L < Homeo+ (S1 ), NH+ (L) and ZH+ (L) denote respectively its normalizer and centralizer in Homeo+ (S1 ).

lemma 6.2. 1) NH+ (Rot) = ZH+ (Rot) = Rot. 2) NH+ (PSL(2, R)k ) = PSL(2, R)k . 3) ZH+ (PSL(2, R)k ) = Z (PSL(2, R)k ), which is cyclic of order k. Proof. 1) Let g ∈ NH+ (Rot); if g r g −1 = r −1 ∀r ∈ Rot, then g cannot preserve orientation; hence g ∈ ZH+ (Rot). Composing g with a rotation we may assume g(0) = 0, which on applying r ∈ Rot yields g(r(0)) = r(0) and thus g = e. 2) For g ∈ NH+ (PSL(2, R)k ) and Kk < PSL(2, R)k maximal compact subgroup, g −1 Kk g is maximal compact as well and hence there is h ∈ PSL(2, R)k with g Kk g −1 = h Kk h −1 that is h −1 g ∈ NH+ (Kk ); since Kk is conjugate to Rot we deduce from 1) that h −1 g ∈ Kk . 3) follows from (2). Finally we record the following consequence of Hilbert’s ﬁfth problem:

theorem 6.3. ([8] thm. 3.3.3) Let G be a locally compact group and A(G) G its amenable radical. Let Ga := G/A(G). Then 1) (Ga )0 is connected, semisimple, with trivial center and no compact factors. 2) The centralizer L of (Ga )0 in Ga is totally disconnected; we have L ∩ (Ga )0 = (e) and the product L · (Ga )0 is open of ﬁnite index in Ga . Proof of Theorem 1.8. Let π : G → Homeo+ (S1 ) be a continuous minimal action. We consider π |A(G) and apply our trichotomy:

30 / m. burger

Case 1: π(A(G)) has a ﬁnite orbit; then the set of ﬁnite π (A(G))-orbits with ﬁxed cardinality is nonempty closed and π(G)-invariant, hence coincides with S1 . As a result, π (A(G)) is ﬁnite cyclic. Case 2: An exceptional minimal set for π (A(G)) being unique would also be π (G)-invariant contradicting minimality of π(G). Case 3: π(A(G)) is minimal; since A(G) is amenable (π |A(G) )∗ (ebR ) = 0 and hence, modulo conjugating π , we may assume that π(A(G)) is a dense subgroup of Rot. In particular, π (G) ⊂ NH+ (Rot) = Rot by Lemma 6.2. This gives the ﬁrst alternative in Theorem 1.8. Now we return to Case 1 and consider Gπ := G/(A(G) ∩ Ker π ), which is a ﬁnite extension q : Gπ → Ga of Ga . Observe that π factors through Ga . Let Lπ := q−1 (L); then (Gπ )0 is connected, semisimple with ﬁnite center and no compact factors, Lπ is totally disconnected, and Lπ · (Gπ )0 is open of ﬁnite index in Gπ . We then have two cases: 1) π |(Gπ )0 is trivial. Since Homeo+ (S1 ) does not contain small subgroups, π sends some compact open subgroup of Lπ to the identity, which implies that Ker π is open. 2) π|(Gπ )0 is nontrivial and hence by Lemma 6.1, π ((Gπ )0 ) = PSL(2, R)k ; since π (G) normalizes π ((Gπ )0 ) we get π(G) = PSL(2, R)k by Lemma 6.2 (2).

References [1] [2] [3]

[4] [5] [6] [7]

U. Bader, A. Furman: e-mail exchange, private communication, September 9–October 6, 2006. U. Bader, A. Furman, A. Shaker: Superrigidity, Weyl groups and actions on the circle. ArXiv: math/0605276v4. M. Burger, A. Iozzi: Boundary maps in bounded cohomology. Appendix to: “Continuous bounded cohomology and applications to rigidity theory.” Geom. Funct. Anal., 12(2):281–292 (2002). M. Burger, A. Iozzi: Bounded Kähler class rigidity of actions on Hermitian symmetric spaces. Ann. Sci. Ecole Norm. Sup. (4), 37(1):77–103 (2004). M. Burger, A. Iozzi, A. Wienhard: Surface group representations with maximal Toledo invariant. Annals of Math., 172(1):517–566 (2010). M. Burger, A. Iozzi, A. Wienhard: Hermitian symmetric spaces and Kähler rigidity. Transform. Groups, 12(1):5–32 (2007). M. Burger, A. Iozzi, A. Wienhard: Tight homomorphisms and Hermitian symmetric spaces. Geom. Funct. Anal., 19(3):678–721 (2009).

an extension criterion for lattice actions on the circle / 31 [8] [9] [10] [11] [12] [13] [14]

[15] [16] [17]

[18]

[19]

[20] [21] [22] [23] [24] [25] [26] [27]

M. Burger, N. Monod: Continuous bounded cohomology and applications to rigidity theory. Geom. Funct. Anal., 12(2):219–280 (2002). B. Deroin, V. Kleptsyn, A. Navas: Sur la dynamique unidimensionnelle en régularité intermédiaire. Acta Math., 199(2):199–262 (2007). A. Furman: Mostow-Margulis rigidity with locally compact targets. Geom. Funct. Anal., 11(1):30–59 (2001). A. Furman: Random walks on groups and random transformations. Handbook of dynamical systems, Vol. 1A, 931–1014, North-Holland, Amsterdam, 2002. H. Furstenberg: A Poisson formula for semi-simple Lie groups. Ann. of Math. (2), 77:335–386 (1963). H. Furstenberg: Noncommuting random products. Trans. Amer. Math. Soc., 108:377– 428 (1963). E. Ghys: Groupes d’homéomorphismes du cercle et cohomologie bornée. The Lefschetz centennial conference, Part III (Mexico City, 1984), 81–106, Contemp. Math., 58, III, American Mathematical Society, Providence, RI, 1987. E. Ghys: Actions de réseaux sur le cercle. Invent. Math., 137(1):199–231 (1999). E. Ghys: Groups acting on the circle. Enseign. Math. (2), 47(3–4):329–407 (2001). Y. Guivarc’h, A. Raugi: Products of random matrices: convergence theorems. Random matrices and their applications (Brunswick, ME, 1984), 31–54, Contemp. Math., 50, American Mathematical Society, Providence, RI, 1986. G. Hector, U. Hirsch: Introduction to the geometry of foliations. Part B. Foliations of codimension one. Aspects of Mathematics, E3. Friedr. Vieweg & Sohn, Braunschweig, 1983. A. Iozzi: Bounded cohomology, boundary maps, and rigidity of representations into Homeo+ (S1 ) and SU(1, n). Rigidity in dynamics and geometry (Cambridge, 2000), 237– 260, Springer, Berlin, 2002. A. Iozzi: Vanishing theorems in bounded cohomology, in preparation. V. Kaimanovich: Double ergodicity of the Poisson boundary and applications to bounded cohomology. Geom. Funct. Anal., 13(4):852–861 (2003). G. Margulis: Discrete subgroups of semisimple Lie groups. Ergebnisse der Mathematik und ihrer Grenzgebiete (3), Springer-Verlag, Berlin, 1991. G. Margulis: Free subgroups of the homeomorphism group of the circle. C. R. Acad. Sci. Paris Sr. I Math., 331(9): 669–674 (2000). N. Monod: Continuous bounded cohomology of locally compact groups. Lecture Notes in Mathematics, 1758. Springer-Verlag, Berlin, 2001. D. Witte Morris: Can lattices in SL(n, R) act on the circle? This volume. A. Navas: Quelques nouveaux phénomènes de rang 1 pour les groupes de difféomorphismes du cercle. Comment. Math. Helv., 80(2): 355–375 (2005). D. Witte, R. Zimmer: Actions of semisimple Lie groups on circle bundles. Geom. Dedicata, 87(1–3): 91–121 (2001).

2

sorin dumitrescu

MEROMORPHIC ALMOST RIGID GEOMETRIC STRUCTURES dedicated to bob zimmer Abstract We study the local Killing Lie algebra of meromorphic almost rigid geometric structures on complex manifolds. This leads to classiﬁcation results for compact complex manifolds admitting holomorphic rigid geometric structures.

1. Introduction Zimmer and Gromov have conjectured that “big” actions of Lie groups that preserve unimodular rigid geometric structures (for example, a pseudoRiemannian metric or an afﬁne connection together with a volume form) are “essentially classiﬁable” [1, 28, 64, 65, 66, 67]. We try to address here this general question in the framework of the complex geometry. We study holomorphic (and meromorphic) unimodular geometric structures on compact complex manifolds, and we show that in many cases the holomorphic rigidity implies that the situation is classiﬁable (even without assuming any isometric group action). This is a survey paper, but some results are new (see Theorem 2.1, Corollary 2.2, and Theorem 5.10). In the sequel all complex manifolds are supposed to be smooth and connected. Consider a complex n-manifold M and, for all integers r ≥ 1, consider the associated bundle Rr (M) of r-frames, which is a Dr (Cn )-principal bundle over M, with Dr (Cn )-the (algebraic) group of r-jets at the origin of local biholomorphisms of Cn ﬁxing 0 [1, 28, 7, 12, 22, 63]. Let us consider, as in [1, 28], the following:

definition 1.1. A meromorphic geometric structure (of order r) φ on a complex manifold M is a Dr (Cn )-equivariant meromorphic map from Rr (M) to a 1991 Mathematics Subject Classiﬁcation 53B21, 53C56, 53A55. Keywords and phrases: complex manifolds, rigid geometric structures, transitive Killing Lie algebras This work was partially supported by the ANR Grant Symplexe BLAN 06-3-137237.

32

meromorphic almost rigid geometric structures / 33

quasiprojective variety Z endowed with an algebraic action of Dr (Cn ). If Z is an afﬁne variety, we say that φ is of afﬁne type. In holomorphic local coordinates U on M, the geometric structure φ is given by a meromorphic map φU : U → Z. Denote by P ⊂ M the analytic subset of poles of φ and by M ∗ = M \ P the open dense subset of M where φ is holomorphic. If P is the empty set, then φ is a holomorphic geometric structure on M. If φ uniquely determines a holomorphic volume form on M ∗ , then φ is called unimodular.

definition 1.2. The meromorphic geometric structure φ is called almost rigid if, away from an analytic subset of M of positive codimension, φ is (holomorphic and) rigid in Gromov’s sense [1, 28]. A holomorphic geometric structure on M is called rigid if it is rigid in Gromov’s sense on all of M. Notice that my deﬁniton of almost rigidity is not exactly the same as in [9].

examples On a complex connected n-manifold M the following meromorphic geometric structures are almost rigid. • A meromorphic map φ from M to a projective space P N (C), which is an

embedding on some nontrivial open subset of M. In this case M is of algebraic dimension n (see Section 2). If φ is a holomorphic embedding, then φ is rigid. • A family X1 , X2 , . . . , Xn of meromorphic vector ﬁelds such that there exists an open (dense) subset of M where the Xi span the holomorphic tangent bundle TM. Any homogeneous manifold admits a holomorphic family of such vector ﬁelds. They pull-back after the blowup of a point (or a submanifold) in the homogeneous manifold in a meromorphic family of such vector ﬁelds. In the special case where X1 , X2 , . . . , Xn are holomorphic and span TM on all of M, the corresponding geometric structure is called a holomorphic parallelization of the tangent bundle and it is a (unimodular) holomorphic rigid geometric structure of afﬁne type. • A meromorphic section g of the bundle S2 (T ∗ M) of complex quadratic forms such that g is nondegenerate on an open (dense) subset. If g is holomorphic and nondegenerate on M, then g is a holomorphic rigid geometric structure of afﬁne type called holomorphic Riemannian

34 / s. dumitrescu

metric. Up to a double cover of M, a holomorphic Riemannian metric is unimodular. A holomorphic Riemannian metric has nothing to do with the more usual Hermitian metric. It is in fact nothing but the complex version of a (pseudo-)Riemannian metric. Observe that since complex quadratic forms have no signature, there is no distinction here between the Riemannian and pseudo-Riemannian cases. This observation was at the origin of the niceuse by F. Gauß of the complexiﬁcation technic of (analytic) Riemannian metrics on surfaces, in order to ﬁnd conformal coordinates for them. Actually, the complexiﬁcation of analytic Riemannian metrics leading to holomorphic ones is becoming a standard trick (see for instance [23]). In the blow-up process, the pull-back of a holomorphic Riemannian metric will stay nondegenerate away from the exceptional set and will vanish on the exceptional set. More general, if g is a meromorpic section of S2 (T ∗ M) nondegenerate ˜ bimeromorphic to on an open dense set, then any complex manifold M M inherits a similar section g˜ . Moreover, if g is holomorphic, then g˜ is also holomorphic. This comes from the fact that indeterminacy points of a meromorphic map are of codimension at least 2 (see theorem 2.5 in [57]) and from Levi’s extension principle (see [57] for the details). • Meromorphic afﬁne or projective connections or meromorphic conformal structures in dimension ≥ 3. For example, in local holomorphic coordinates (z1 , . . . , zn ) on the manifold M, a meromorphic afﬁne connection ∇ is determined by the meromorphic functions ijk such that ∇

∂ ∂zi

∂ ∂zj

= ijk ∂z∂ , for all i, j, k ∈ {1, . . . , n}. k

Afﬁne connections are geometric structures of afﬁne type, but projective connections or conformal structures are not of afﬁne type. • If φ is a meromorphic almost rigid geometric structure of order r1 and g is a meromorphic geometric structure of order r2 , then we can put together φ and g in some meromorphic almost rigid geometric structure of order max(r1 , r2 ), denoted by (φ, g) [28, 1, 7, 12, 22, 63]. • The s-jet (prolongation) φ (s) of a meromorphic almost rigid geometric structure φ is still a meromorphic almost rigid geometric structure [28, 1, 7, 12, 22, 63]. Recall that local biholomorphisms of M that preserve a meromorphic geometric structure φ are called local isometries. Note that local isometries of (φ, g) are the local isometries of φ, which also preserve g.

meromorphic almost rigid geometric structures / 35

The set of local isometries of a holomorphic rigid geometric structure φ is a Lie pseudogroup Isloc (φ) generated by a Lie algebra of local vector ﬁelds called (local) Killing Lie algebra. If the local Killing Lie algebra is transitive on M, then φ is called locally homogeneous.

2. Local Isometries and Meromorphic Functions The maximal number of algebraically independent meromorphic functions on a complex manifold M is called the algebraic dimension a(M) of M. Recall that a theorem of Siegel proves that a complex n-manifold M admits at most n algebraically independent meromorphic functions [57]. Then a(M) ∈ {0, 1, . . . , n} and for algebraic manifolds a(M) = n. We will say that two points in M are in the same ﬁber of the algebraic reduction of M if any meromorphic function on M takes the same value at the two points. There exists some open dense set in M where the ﬁbers of the algebraic reduction are the ﬁbers of a holomorphic ﬁbration on an algebraic manifold of dimension a(M) and any meromorphic function on M is the pull-back of a meromorphic function on the basis [57]. The following theorem shows that the ﬁbers of the algebraic reduction are in the same orbit of the pseudogroup of local isometries for any meromorphic almost rigid geometric structure on M. This is a meromorphic version of the celebrated Gromov’s open dense orbit theorem [1, 28] (see also [7, 12, 22, 63]).

theorem 2.1. Let M be a connected complex manifold of dimension n, which admits a meromorphic almost rigid geometric structure φ. Then, there exists a nowhere-dense analytic subset S in M, such that M \ S is Isloc (φ)-invariant and the orbits of Isloc (φ) in M \ S are the ﬁbers of a holomorphic ﬁbration of constant rank. The dimension of the ﬁbers is ≥ n − a(M), where a(M) is the algebraic dimension of M. Recall that g = (φ, g) is still almost rigid for any meromorphic (not necessarily almost rigid) geometric structure g on M. This yields to the following:

corollary 2.2. If g is a meromorphic geometric structure on M, then, away from a nowhere dense analytic subset, the Isloc (g)-orbits are of dimension ≥ n − a(M). In particular, if a(M) = 0, then g is locally homogeneous outside a nowhere dense analytic subset of M (the Killing Lie algebra G of g = (φ, g) is transitive on

36 / s. dumitrescu

an open dense set). Moreover, if φ is unimodular and G is unimodular and simply transitive, then g is locally homogeneous on all of the open dense set where φ and g are holomorphic and φ is rigid. In this case, if φ is unimodular holomorphic and rigid and g is holomorphic, then g is locally homogeneous on M. Proof. Let φ be of order r, given by a map φ : Rr (M) → Z. For each positive integer s we consider the s-jet φ (s) of φ. This is a D(r+s) (Cn )-equivariant meromorphic map R(r+s) (M) → Z (s) , where Z (s) is the algebraic variety of the s-jets at the origin of holomorphic maps from Cn to Z. One can ﬁnd the expression of the (algebraic) D(r+s) (Cn )-action on Z (s) in [12, 22]. Since φ is almost rigid, there exists a nowhere dense analytic subset S in M, containing the poles of φ, and a positive integer s such that two points m, m in M\S are in the same orbit of Isloc (φ) if and only if φ (s) sends the ﬁbers of R(r+s) (M) above m and m on the same D(r+s) (Cn )-orbit in Z (s) [28, 1]. Rosenlicht’s theorem (see [49]) shows that there exists a D(r+s) (Cn )-invariant stratiﬁcation Z (s) = Z0 ⊃ · · · ⊃ Zl , such that Zi+1 is Zariski closed in Zi , the quotient of Zi \ Zi+1 by D(r+s) (Cn ) is a complex manifold, and rational D(r+s) (Cn )-invariant functions on Zi separate orbits in Zi \ Zi+1 . Consider the open dense Isloc (φ)-invariant subset U in M \ S , such that (s) φ is of constant rank above U and the image of R(r+s) (M)|U through φ (s) lies in Zi \ Zi+1 but not in Zi+1 . Then the orbits of Isloc (φ) in U are the ﬁbers of a ﬁbration of constant rank (on the quotient of Zi \ Zi+1 by D(r+s) (Cn )). Obviously, U = M \ S, where S is a nowhere dense analytic subset in M. If m and n are two points in U that are not in the same Isloc (φ)-orbit, then the corresponding ﬁbers of R(r+s) (M)|U are sent by φ (s) on two distinct D(r+s) (Cn )orbits in Zi \ Zi+1 . By Rosenlicht’s theorem there exists a D(r+s) (Cn )-invariant rational function F : Zi \ Zi+1 → C, which takes distinct values at these two orbits. The meromorphic function F ◦ φ (s) : R(r+s) (M) → C is D(r+s) (Cn )-invariant and descends in a Isloc (φ)-invariant meromorphic function on M that takes distinct values at m and at n. Consequently, the complex codimension in U of the Isloc (φ)-orbits is ≤ a(M), which ﬁnishes the proof.

meromorphic almost rigid geometric structures / 37

We now deduce the corollary: Proof. It is convenient to put together φ and g in some extra geometric structure g = (φ, g). Now g is a meromorphic almost rigid geometric structure and Theorem 2.1 shows that the complex dimension of a generic Isloc (g )-orbit is ≥ n − a(M). As each Isloc (g )-orbit is contained in a Isloc (g)-orbit, the result follows also for g. If a(M) = 0, the Killing Lie algebra G of g is transitive on a maximal open dense subset U in M. Suppose now that φ is unimodular and also that G is unimodular. The other inclusion being trivial, it is enough to show that U contains the maximal subset of M where φ and g are holomorphic and φ is rigid. Pick up a point m in the previous subset. We want to show that m is in U. Since g = (φ, g) is holomorphic and rigid in the neighborhood of m, it follows that m admits an open neighborhood Um in M such that any local holomorphic Killing ﬁeld of g deﬁned on a connected open subset in Um extends on all of Um [48, 2]. Since G acts transitively on U, choose local linearly independent Killing ﬁelds X1 , . . . , Xn on a connected open set included in U ∩ Um . As φ is unimodular, it determines a holomorphic volume form vol on Um (if necessary we restrict to a smaller Um ), which is preserved by Isloc (g ). But Isloc (g ) acts transitively on U ∩ Um and G is supposed to be unimodular. This implies that the function vol(X1 , . . . , Xn ) is G -invariant and, consequently, a (nonzero) constant on U ∩ Um . On the other hand, X1 , . . . , Xn extend in some holomorphic Killing ﬁelds X˜ 1 , . . . , X˜ n deﬁned on all of Um . The holomorphic function vol(X˜ 1 , . . . , X˜ n ) is a nonzero constant on Um : in particular, X˜ 1 (m), . . . , X˜ n (m) are linearly independent. We proved that G acts transitively in the neighborhood of m and thus m ∈ U.

3. Classification Results Parallelizations of the Tangent Bundle

We begin with the classiﬁcation of compact complex manifolds that admit a holomorphic parallelization of the tangent bundle [60].

theorem 3.1. (wang) Let M be a compact connected complex manifold with a holomorphic parallelization of the tangent bundle. Then M is a quotient \G, where G is a connected simply connected complex Lie group and is a uniform lattice in G. Moreover, M is Kaehler if and only if G is abelian (and M is a complex torus).

38 / s. dumitrescu

Proof. Let n be the complex dimension of M and consider X1 , X2 , . . . , Xn global holomorphic vector ﬁelds on M that span TM. Then, for all 1 ≤ i, j ≤ n, we have ij

ij

ij

[Xi , Xj ] = f1 X1 + f2 X2 + · · · + fn Xn , ij

with fk holomorphic functions on M. Since M is compact, these functions have to be constant and, consequently, X1 , X2 , . . . , Xn generate a n-dimensional Lie algebra G that acts simply transitively on M. By Lie’s theorem there exists a unique connected simply connected complex Lie group G corresponding to G . In particular, the holomorphic parallelization is locally homogeneous, locally modeled on the parallelization given by translation-invariant vector ﬁelds on the Lie group G. Since M is compact, the Xi are complete and they deﬁne a holomorphic simply transitive action of G on M. Hence M is a quotient of G by a cocompact discrete subgroup in G. Assume now M is Kaehler. Then, any holomorphic form on M has to be closed [27]. Consider ω1 , ω2 , . . . , ωn the dual basis with respect to X1 , X2 , . . . , Xn . The one forms ωi are holomorphic and translation-invariant on G (they descend on \G). The Lie-Cartan formula dωi (Xj , Xk ) = −ωi ([Xj , Xk ]), shows that the one forms ωi are all closed if and only if G is abelian and thus M is a complex torus. Holomorphic Riemannian Metrics

As in the real case, a holomorphic Riemannian metric on M gives rise to a covariant differential calculus, that is, a Levi-Civita (holomorphic) afﬁne connection, and to geometric features: curvature tensors and geodesic (complex) curves [40, 41]. Locally, a holomorphic Riemannian metric has the expression gij (z) dzi dzj , where (gij (z)) is a complex inversible symmetric matrix depending holomorphically on z. The standard example is that of the global ﬂat holomorphic Riemannian metric dz21 + dz22 + · · · + dz2n on Cn . This metric is translationinvariant and thus descends to any quotient of Cn by a lattice. Hence, complex tori possess (ﬂat) holomorphic Riemannian metrics. This is, however, a very special situation since, contrary to real case, only few compact complex manifolds admit holomorphic Riemannian metrics. In fact, Yau’s proof of the Calabi conjecture shows that, up to ﬁnite unramiﬁed covers, complex tori

meromorphic almost rigid geometric structures / 39

are the only compact Kaehler manifolds admitting holomorphic Riemannian metrics [34]. However, very interesting examples, constructed by Ghys in [25], do exist on 3-dimensional complex non-Kaehler manifolds and deserve classiﬁcation. Notice that parallelizable manifolds admit holomorphic Riemannian metrics coming from left-invariant holomorphic Riemannian metrics on G (which can be constructed by left translating any complex nondegenerate quadratic form deﬁned on the Lie algebra G ). Ghys’s examples of 3-dimensional compact complex manifolds endowed with holomorphic Riemannian metrics are obtained by deformation of the complex structure on parallelizable manifolds \SL(2, C) [25]. They are nonstandard, meaning they do not admit parallelizable manifolds as ﬁnite unramiﬁed covers. Those nonstandard examples will be described later on. A ﬁrst obstruction to the existence of a holomorphic Riemannian metric on a compact complex manifold is the vanishing of its ﬁrst Chern class. Indeed, a holomorphic Riemannian metric on M provides an isomorphism between TM and T ∗ M. In particular, the canonical bundle K is isomorphic to the anticanonical bundle K −1 and thus K 2 is trivial. This means that the ﬁrst Chern class of M vanishes and, up to a double unramiﬁed cover, M possesses a holomorphic volume form. The following proposition describes holomorphic Riemannian metrics on parallelizable manifolds:

proposition 3.2. Let M = \G a compact parallelizable manifold, with G a simply connected complex Lie group and a uniform lattice in G. Then, any holomorphic Riemannian metric g on M comes from a nondegenerate complex quadratic form on the Lie algebra G of G. In particular, the pull-back of g is left invariant on the universal cover G (and g is locally homogeneous on M). Moreover, any compact parallelizable 3-manifold admits a holomorphic Riemannian metric of constant sectional curvature. The metric is ﬂat exactly when G is solvable. Proof. Consider X1 , X2 , . . . , Xn the fundamental vector ﬁelds corresponding to the locally free G-action on M. Let g be a holomorphic Riemannian metric on M and denote also by g the associated complex symmetric bilinear form. Then g(Xi , Xj ) is a holomorphic function on M and thus constant, for all 1 ≤ i, j ≤ n. This implies that g comes from a left-invariant holomorphic Riemannian metric on G.

40 / s. dumitrescu

Assume now G is a connected simply connected complex unimodular Lie group of dimension 3. We have only four such Lie groups: C3 , the complex Heisenberg group, the complex SOL group, and SL(2, C) [35]. Note that the group SOL is the complexiﬁcation of the afﬁne isometry group of the Minkowski plane R1,1 or equivalently the isometry group of C2 endowed with its ﬂat holomorphic Riemannian metric. We begin with the case G = C3 . We have seen that C3 admits a ﬂat translation-invariant holomorphic Riemannian metric. The isometry group of this metric is O(3, C) C3 . We recall here that O(3, C) and SL(2, C) are locally isomorphic. Consider now the case G = SL(2, C). The Killing form of the Lie algebra sl(2, C) is a nondegenerate complex quadratic form that endows SL(2, C) with a left-invariant holomorphic Riemannian metric of constant sectional curvature. Since the Killing quadratic form is invariant by the adjoint representation, the isometry group also contains all right translations. In fact, the connected component of the isometry group is SL(2, C) × SL(2, C) acting by left and right translation. On the other hand, \SL(2, C) doesn’t admit ﬂat torsion-free afﬁne connections (or ﬂat holomorphic Riemannian metrics) [18]. It is an easy exercice to exhibit in the isometry group O(3, C) C3 of the ﬂat holomorphic Riemannian space copies of the complex Heisenberg group and of the complex SOL group, which act simply transitively [18]. Thus the ﬂat holomorphic Riemannian space also admits models that are given by leftinvariant metrics on the Heisenberg group and on the SOL group. One can get explicit expression of these holomorphic Riemannian metrics by complexiﬁcation of ﬂat left-invariant Lorentz metrics on the real Heisenberg and SOL groups [51, 52]. On the other hand, there exists no 3-dimensional solvable subgroups in SL(2, C) × SL(2, C) acting with an open orbit on SL(2, C). It follows that the compact quotients of 3-dimensional solvable groups don’t admit holomorphic Riemannian metrics of nonzero constant sectional curvature.

Ghys Nonstandard Examples

As above, for any cocompact lattice in SL(2, C), the quotient M = \SL(2, C) admits a holomorphic Riemannian metric of nonzero constant sectional curvature. It is convenient to consider M as a quotient of S3 = O(4, C)/O(3, C) = SL(2, C) × SL(2, C)/SL(2, C) by , seen as a subgroup of SL(2, C) × SL(2, C) by the trivial embedding γ ∈ → (γ , 1) ∈ SL(2, C) × SL(2, C).

meromorphic almost rigid geometric structures / 41

New interesting examples of manifolds admitting holomorphic Riemannian metrics of nonzero constant sectional curvature have been constructed in [25] by deformation of this embedding of . Those deformations are constructed choosing a morphism u : → SL(2, C) and considering the embedding γ → (γ , u(γ )). Algebraically, the action is given by: (γ , m) ∈ × SL(2, C) → γ mu(γ −1 ) ∈ SL(2, C). It is proved in [25] that, for u close enough to the trivial morphism, acts properly (and freely) on S3 ( ∼ = SL(2, C)) such that the quotient Mu is a complex compact manifold (covered by SL(2, C)) admitting a holomorphic Riemannian metric of non-zero constant sectional curvature. In general, these examples do not admit parallelizable manifolds as ﬁnite covers. Note that left-invariant holomorphic Riemannian metrics on SL(2, C) that are not right-invariant, in general, will not descend on Mu . Let us notice that despite this systematic study in [25], there are still many open questions regarding these examples (including the question of completeness). A real version of this study is in [39, 26, 53]. Dimension 3

The classiﬁcation of complex compact manifolds admitting holomorphic Riemannian metrics is simple in complex dimension 2 (see Section 5). An important step toward the classiﬁcation in dimension 3 was made in [17] with the following result:

theorem 3.3. Any holomorphic Riemannian metric on a compact connected complex 3-manifold is locally homogeneous. More generally, if a compact connected complex 3-manifold M admits a holomorphic Riemannian metric, then any holomorphic geometric structure of afﬁne type on M is locally homogeneous. Observe that the previous result is trivial in dimension 2, since the sectional curvature is a holomorphic function and thus constant on compact complex surfaces: this implies the local homogeneity [61]. In dimension 3, the sectional curvature will be, in general, a nonconstant meromorphic function on the 2Grassmanian of the holomorphic tangent space with poles at the degenerate planes. Thanks to Theorem 3.3, our manifold M is locally modeled on a (G, G/I)geometry in Thurston’s sense [55], where I is a closed subgroup of the Lie group G such that the G-action on G/I preserves some holomorphic

42 / s. dumitrescu

Riemannian metric (see [21] for details and notice that the local Killing Lie algebra of the holomorphic Riemannian metric is the Lie algebra of G). In this context we have a developing map from the universal cover of M into G/I, which is a local diffeomorphism and which is equivariant with respect to the ˜ by deck transformations and on G/I action of the fundamental group on M via the holonomy morphism ρ : π1 (M) → G [55]. Recall that the (G, G/I)-geometry is called complete if the developing map is a diffeomorphism and, consequently, = ρ(π1 (M)) acts properly on G/I such that M is a compact quotient \G/I. A second step was achieved in [21] where, in a common work with Zeghib, we proved the following result that can be seen, in particular, as a completeness result in the case where G is solvable.

theorem 3.4. Let M be a compact connected complex 3-manifold that admits a (locally homogeneous) holomorphic Riemannian metric g. Then i) If the Killing Lie algebra of g has a nontrivial semisimple part, then it preserves some holomorphic Riemannian metric on M with constant sectional curvature. ii) If the Killing Lie algebra of g is solvable, then, up to a ﬁnite unramiﬁed cover, M is a quotient either of the complex Heisenberg group, or of the complex SOL group by a lattice.

remark 3.5. If g is ﬂat, its Killing Lie algebra corresponds to O(3, C) C3, which has a nontrivial semisimple part. Thus, ﬂat holomorphic Riemannian metrics on complex tori are part of point (i) in Theorem 3.4. The point (ii) of the previous theorem is not only about completeness, but also gives a rigidity result in Bieberbach’s sense [61]: G contains a 3dimensional closed subgroup H (either isomorphic to the complex Heisenberg group, or to the complex SOL group), which acts simply transitively (and so identiﬁes) with G/I and (up to a ﬁnite index) the image of the holonomy morphism lies in H. It follows that, up to a ﬁnite cover, M identiﬁes with \H. Since the Heisenberg and SOL groups admit left-invariant ﬂat holomorphic Riemannian metrics this leads to:

corollary 3.6. If a compact connected complex 3-manifold M admits a holomorphic Riemannian metric, then, up to a ﬁnite unramiﬁed cover, M admits a holomorphic Riemannian metric of constant sectional curvature.

meromorphic almost rigid geometric structures / 43

Theorem 3.4 does not end the story, even in dimension 3, essentially because of remaining completeness questions, and those on the algebraic structure of the fundamental group. It remains to classify the compact complex 3-manifolds endowed with a holomorphic Riemannian metric of constant sectional curvature.

ﬂat case. In this case M admits a (O(3, C) C3 , C3 )-geometry. The challenge remains: 1) Markus conjecture: Is M complete? 2) Auslander conjecture: Assuming M is complete, is solvable? Note that these questions are settled in the setting of (real) ﬂat Lorentz manifolds [13, 24], but unsolved for general (real) pseudo-Riemannian metrics. The real part of the holomorphic Riemannian metric is a (real) pseudo-Riemannian metric of signature (3, 3) for which both previous conjectures are still open.

nonﬂat case. In this case G = SL(2, C) × SL(2, C) and I = SL(2, C) is diagonally embedded in the product. The completeness of this geometry on compact complex manifolds is still an open problem, despite a local result of Ghys [25]. Recall that the real analogous of this problem, that is, the completeness of compact manifolds endowed with Lorentz metrics of negative constant sectional curvature was solved in [37], but the proof cannot generalize to other signatures. Higher Dimension

One interesting problem in differential geometry is to decide whether a given homogeneous space G/I possesses a compact quotient. A more general related question is to decide whether there exists compact manifolds locally modeled on (G, G/I) (see, for instance, [6, 8, 38]). The case I = 1, or more generally I compact, reduces to the classical question of existence of cocompact lattices in Lie groups. For homogeneous spaces of non-Riemannian type (i.e., I noncompact) the problem is much harder. The case Sn = O(n + 1, C)/O(n, C) is a geometric situation where these questions can be tested. It turns out that compact quotients of Sn are known to exist only for n = 1, 3, or 7. We discussed the case n = 3 above, and the existence of a compact quotient of S7 was proved in [38]. Here, we dare ask with [38]:

conjecture 3.7. [38] Sn has no compact quotients, for n = 1, 3, 7.

44 / s. dumitrescu

A stronger version of this question was proved in [6] for Sn , if n has the form 4m + 1, with m ∈ N. Keeping in mind our geometric approach, we generalize the question to manifolds locally modeled on Sn . More exactly:

conjecture 3.8. [21] A compact complex manifold endowed with a holomorphic Riemannian metric of constant nonvanishing curvature is complete. In particular, such a manifold has dimension 3 or 7. 4. Applications to Simply Connected Manifolds Remark ﬁrst that Theorem 3.3 has the following direct consequence:

corollary 4.1. A compact connected simply connected 3-manifold admits no holomorphic Riemannian metrics. Proof. Assume, by contradiction, (M, g) being as in the statement of the corollary. Then Theorem 3.3 implies that g is locally homogeneous. Since M is simply connected, the local Killing ﬁelds of the Killing algebra G extend on all of M: the unique connected simply connected complex Lie group G associated to G acts isometrically and transitively on M. Then M is a homogeneous space G/H. Moreover, up to a double cover, G/H admits a holomorphic volume form vol coming from the holomorphic Riemannian metric. Take X1 , X2 , X3 three global Killing ﬁelds on M that are linearly independent at some point. Since vol(X1 , X2 , X3 ) is a holomorphic function on M, it is a nonzero constant and, consequently, X1 , X2 , X3 are linearly independent on M. Hence Wang’s theorem implies that M is a quotient of a 3 dimensional connected simply connected complex Lie group G1 by a discrete subgroup. Since M is simply connected, this discrete subgroup has to be trivial and M identiﬁes with G1 . But there is no compact simply connected complex Lie group: a contradiction.

theorem 4.2. A compact connected simply connected complex n-manifold without nonconstant meromorphic functions admits no holomorphic unimodular rigid geometric structures. Proof. Assume, by contradiction, that (M, φ) veriﬁes the hypothesis. Since a(M) = 0, Theorem 2.1 implies φ is locally homogeneous on an open dense set U. As M is simply connected, elements in the Killing algebra G extend

meromorphic almost rigid geometric structures / 45

on all of M: the connected simply connected complex Lie group G associated to G acts isometrically on M with an open dense orbit. The open dense orbit U identiﬁes with a homogeneous space G/H, where H is a closed subgroup of G. Consider X1 , X2 , . . . , Xn global Killing ﬁelds on M that are linearly independent at some point of the open orbit U. As before, vol(X1 , X2 , . . . , Xn ) is a nonzero constant, where vol is the holomorphic volume form associated to φ. Thus the Xi give a holomorphic parallelization of TM and Wang’s theorem enables us to conclude as in the previous proof. For nonunimodular rigid geometric structures we have the following less precise:

theorem 4.3. [15] Let M be a compact connected simply connected complex nmanifold without nonconstant meromorphic functions and admitting a holomorphic rigid geometric structure φ. Then M is an equivariant compactiﬁcation of \G, where is a discrete noncocompact subgroup in a complex Lie group G. Proof. Since a(M) = 0, Theorem 2.1 implies φ is locally homogeneous on an open dense set U. As before, the extension property of local Killing ﬁelds implies U is a complex homogeneous space G/H, where G is a connected simply connected complex Lie group and H is a closed subgroup in G. We show now that H is a discrete subgroup of G. Assume by contradiction the Lie algebra of H is nontrivial. Take at any point u ∈ U, the isotropy subalgebra Hu (i.e., the Lie subalgebra of Killing ﬁelds vanishing at u). Remark that Hgu = Ad(g)Hu , for any g ∈ G and u ∈ U, where Ad is the adjoint representation. The map u → Hu is a meromorphic map from M to the Grassmanian of d-dimensional vector spaces in G . But M doesn’t admit any nonconstant meromorphic function and this map has to be constant. It follows that Hu is Ad(G)invariant and H is a normal subgroup of G: a contradiction, since the G-action on M is faithful. Thus G is of dimension n and H identiﬁes to a lattice in G. As M is simply connected, U has to be strictly contained in M and M is an equivariant compactiﬁcation of \G. We don’t know whether such equivariant compactiﬁcations of \G admit equivariant holomorphic rigid geometric structures, but the previous result has the following application.

46 / s. dumitrescu

Recall that a well-known open question asks whether the 6-dimensional real sphere S6 admits complex structures or not. In this context, we have the following:

corollary 4.4. If S6 admits a complex structure M, then M admits no holomorphic rigid geometric structures. Proof. The starting point of the proof is a result of [11] where it is proved that M doesn’t admit nonconstant meromorphic functions. If M supports holomorphic rigid geometric structures, then Theorem 4.3 implies that M is an equivariant compactiﬁcation of a homogeneous space. This is in contradiction with the main theorem of [30]. As for Kaehler manifolds we have proved in [15] the following more precise:

theorem 4.5. Let M be a compact connected Kaehler manifold endowed with a holomorphic unimodular rigid geometric structure φ of afﬁne type. Then, up to a ﬁnite unramiﬁed cover, M is a complex torus (quotient of Cn by a lattice) and φ is translation-invariant. The proof is done in two steps. First we prove that φ is locally homogeneous. Then we use a splitting theorem [5], which asserts that such compact Kaehler manifolds with a holomorphic volume form (Calabi-Yau manifolds) are biholomorphic, up to a ﬁnite cover, to a direct product of a complex torus and a compact simply connected Kaehler manifold with a holomorphic volume form. Starting with φ and using the product structure, we construct a holomorphic unimodular rigid geometric structure on the simply connected factor that is locally homogeneous. We conclude as in the proof of Theorem 4.2 that the simply connected factor is trivial. Recently, we adapted this proof in [19] to all holomorphic Cartan geometries of algebraic type on Calabi-Yau manifolds. A similar result was independently proved in [10]. One can also ﬁnd a classiﬁcation of certain holomorphic G-structures of order one on uniruled projective manifolds in [31].

5. Applications to Complex Surfaces theorem 5.1. Let S be a compact complex surface endowed with a holomorphic unimodular rigid geometric structure φ of afﬁne type. Then the Killing Lie algebra of φ is nontrivial.

meromorphic almost rigid geometric structures / 47

The assumption on φ to be afﬁne is essential:

remark 5.2. If S is a compact complex algebraic surface with trivial canonical bundle (for example, a complex torus or an algebraic K3 surface), the geometric structure given by a holomorphic volume form on S together with a holomorphic embedding of S in a complex projective space doesn’t admit any nontrivial local isometry. However, this geometric structure is not of afﬁne type. Proof. The proof is a simple corollary of Theorems 4.5 and 2.1. Indeed, Theorem 2.1 implies the Killing Lie algebra of φ is trivial only if the algebraic dimension a(S) equals 2. But in this case S is algebraic [4] and thus Kaehler. Then Theorem 4.5 applies and the Killing Lie algebra is transitive on S and hence of dimension at least 2. Recall that a complex surface is called minimal if it does not contain any copy of P 1 (C) with self-intersection −1 (see [4], page 91). Then we have:

theorem 5.3. Let S be a compact minimal complex surface that is not biholomorphic to a nonalgebraic K 3 surface or to a nonafﬁne Hopf surface. Then S admits holomorphic rigid geometric structures. remark 5.4. By deﬁnition, a complex algebraic manifold admits an embedding in a complex projective space, which was seen to be a holomorphic rigid geometric structure (of order zero). Proof. By the previous remark, it remains to consider the case of nonalgebraic complex surfaces. Then the Enriques-Kodaira classiﬁcation shows that, up to a ﬁnite unramiﬁed cover, any minimal nonalgebraic complex compact surface is biholomorphic to one of the following complex surfaces: a complex tori, a Hopf surface, an Inoue surface, a K3 surface, a principal elliptic principal bundle over an elliptic curve, or a principal elliptic bundle over a Riemann surface of genus g ≥ 2, with odd-ﬁrst Betti number (see [4], p. 244). However, it is known that complex tori, Inoue surfaces, afﬁne Hopf surfaces, principal elliptic bundles over elliptic curves, and principal elliptic bundles over a Riemann surface of genus g ≥ 2, with odd-ﬁrst Betti number, admit ﬂat holomorphic afﬁne connections [34, 36, 43, 54, 58, 59].

48 / s. dumitrescu

The classiﬁcation of compact complex surfaces admitting holomorphic afﬁne connections (see [34, 36, 43, 58, 59]) implies the following result:

theorem 5.5. Let S be a compact complex surface admitting a holomorphic unimodular afﬁne connection. Then, up to a ﬁnite unramiﬁed cover, either S is a complex torus and the connection is translation-invariant, or S is an elliptic principal bundle over an elliptic curve and the connection is locally modeled on a translation-invariant connection on a complex torus. Since any complex manifold endowed with a holomorphic Riemannian metric inherits a holomorphic (unimodular) afﬁne (Levi-Civita) connection [40, 41], this easily implies the following [17]:

corollary 5.6. Let S be a complex compact surface admitting a holomorphic Riemannian metric g. Then, up to a ﬁnite unramiﬁed cover, S is a complex torus and g is (ﬂat) translation-invariant. In [20] we classiﬁed the local geometry of all torsion-free holomorphic afﬁne connections ∇ on compact complex surfaces. In particular, we proved that either ∇ is locally homogeneous, locally isomorphic to a translation-invariant connection on C2 , or ∇ is a nonﬂat connection on a principal elliptic bundle over a Riemann surface of genus g ≥ 2 with odd ﬁrst Betti number and the local Killing Lie algebra of ∇ is one-dimensional generated by the fundamental vector ﬁeld of the principal ﬁbration. In all cases, ∇ is projectively ﬂat. As a consequence we proved in [20] the following:

theorem 5.7. Normal holomorphic projective connections on compact complex surfaces are ﬂat. Inoue Surfaces

Recall that Inoue surfaces are compact complex surfaces in the class VII0 , which are not Hopf, and have a vanishing second Betti number [4, 33]. In [34, 36] it is proved that any Inoue surface admits a (unique) ﬂat torsionfree holomorphic afﬁne connection. Here we prove the following:

theorem 5.8. Holomorphic geometric structures on Inoue surfaces are locally homogeneous.

remark 5.9. Inoue surfaces do not admit nonconstant meromorphic functions [4].

meromorphic almost rigid geometric structures / 49

Proof. Let τ be a holomorphic geometric structure on an Inoue surface S. Let ∇0 be a ﬂat torsion-free holomorphic afﬁne connection on S. We prove that the holomorphic rigid geometric structure τ = (τ , ∇0 ) is locally homogeneous. Since a(S) = 0, Theorem 2.1 implies τ is locally homogeneous on some maximal open dense set S \ E, where E is a compact analytic subset of S of positive codimension. We want to show that E is empty. We prove ﬁrst that, up to a double cover of S, the subset E is a smooth submanifold in S (this is always true if S is a ﬁnite set; but here S might have components of complex dimension 1). Assume, by contradiction, that E is not a smooth submanifold in S. Choose p ∈ E a singular point in E. In particular, p is not isolated in E, but p is isolated among the singular points of E. Since Isloc (τ ) preserves E, it has to preserve the set of its singular points and thus it ﬁxes p. Consequently any local Killing ﬁeld deﬁned in the neighborhood of p has to vanish at p. Denote by G the Lie algebra of local Killing ﬁelds in the neighborhood of p. Since G acts transitively on an open set, its dimension is at least 2. Each element of G preserves ∇ and ﬁxes p. In exponential coordinates the G -action in the neighborhood of p is linear. This gives an embedding of G in the Lie algebra of GL(2, C) (the image of the isotropy representation at p). In particular, G is of dimension ≤ 4. Suppose ﬁrst that G is of dimension 2. The corresponding subgroups of GL(2, C) are conjugated either of diagonal matrices, to the group or to one a b 1 m of the following subgroups , with a ∈ C∗ and b ∈ C, , 0 a−1 0 n m n ∗ , with m ∈ C∗ and n ∈ C. with m ∈ C and n ∈ C , or 0 1 In the ﬁrst case, the invariant closed subset E lies, in exponential coordinates, in the union of the two eigen directions. In the last two cases, E locally lies in the invariant line y = 0. In all situations, up to a double cover of S, the analytic set E is smooth: a contradiction. We now settle the case where G is of dimension 3 or 4. Then the image of G by the isotropy representation in p is conjugated in GL(2, C) to one of the following subgroups: SL(2, C), GL(2, C), or the group of inversible uppertriangular matrices. But GL(2, C) and SL(2, C) don’t admit invariant subsets other than p, which will be an isolated point in E: impossible. In the last situation, E locally coincides, as before, with the unique invariant line and it is smooth.

50 / s. dumitrescu

Up to a double cover, E is a holomorphic submanifold in S. If E admits a component of dimension 1, then this component will be a union of closed curves. But Inoue surfaces contain no curves [33]. This proves that E is a ﬁnite set. Assume, by contradiction, that E is not empty and consider p ∈ E. The previous arguments show that the Lie algebra G is isomorphic either to sl(2, C) or to gl(2, C). Assume ﬁrst G = sl(2, C). As before, the local action of G in the neighborhood of p is conjugated to the action of G on Tp S, which coincides with the standard linear action of sl(2, C) on C2 . This action has two orbits: the point p and C2 \ {p}. The stabilizer H in SL(2, C) of a nonzero vector x ∈ Tp S is conjugated

1 b , with b ∈ C. 0 1 Observe that the action of G on G/H preserves a nontrivial holomorphic vector ﬁeld. The expression of this vector ﬁeld in linear coordinates (z1 , z2 ) in the ∂ ∂ neighborhood of p is z1 + z2 . ∂z1 ∂z2 Since S \ E is locally modeled on (G, G/H), this vector ﬁeld is well deﬁned on S \ E. But E is of complex codimension 2 in S and the extension theorem of Hartogs [27] shows that the vector ﬁeld extends to a global nontrivial holomorphic vector ﬁeld X on S. The vector ﬁeld X is G -invariant on S \ E and hence on all of S. Since the isotropy action of SL(2, C) at p doesn’t preserve nontrivial vectors in Tp S C2 , it follows that X (p) = 0. This is impossible, since Inoue surfaces don’t admit singular nontrivial holomorphic vector ﬁelds [33]. The proof is the same in the case G = gl(2, C). to the following one-parameter subgroup of SL(2, C):

Meromorphic Afﬁne Connections

In the following theorem we describe some meromorphic afﬁne connections on non-Kaehler principal elliptic bundles over a compact Riemann surface of genus g ≥ 2.

theorem 5.10. Let S be a principal elliptic bundle over a compact Riemann surface of genus g ≥ 2, with odd-ﬁrst Betti number. i) A meromorphic torsion-free afﬁne connection ∇ on S is invariant by the principal ﬁbration (i.e., the fundamental vector ﬁeld of the principal ﬁbration is a Killing ﬁeld) if and only if the set of its poles intersects only a ﬁnite set of ﬁbers. ii) The space of those previous connections admitting simple poles on a single ﬁber, above a point ξ0 ∈ , is a complex afﬁne space of dimension 5g + 1 and

meromorphic almost rigid geometric structures / 51 2 , where H 0 (K 2 ) its underlying vector space identiﬁes with Hξ00 (K2 ) × M2,ξ ξ0 0 is the vector space of meromorphic quadratic differentials on with a single pole of order at most two at ξ0 , and M2,ξ0 is the vector space of quasimodular forms of weight 2 on with a single simple pole at ξ0 . iii) If ∇ is generic among the meromorphic afﬁne connections that satisfy (ii), then the local Killing Lie algebra of ∇ is generated by the fundamental vector ﬁeld of the principal ﬁbration. iv) The meromorphic afﬁne connections that satisfy (ii) are projectively ﬂat away from the poles.

Recall that a quasimodular form of weight 2 on is a holomorphic function f deﬁned on the upper-half-plane H such that, forsome K ∈C, we have f (ξ ) = aγ bγ f (γ ξ )(cγ ξ + dγ )−2 − K (cγ ξ + dγ )−1 , for all γ = ∈ SL(2, R) in the cγ d γ fundamental group of [3, 62]. If K = 0, then f is a classical modular form. Theorem 9 in [3] shows that the space of quasimodular forms of weight 2 on that admit a simple pole in a single orbit is a complex vector space of dimension g + 1 (i.e., the quotient of the space of such quasimodular forms over those which are modular is 1-dimensional).

remark 5.11. The proof below shows that S admits a (ﬂat) holomorphic afﬁne connection. Theorem 2.1 applies and, since any meromorphic function on S is a pull-back of a meromorphic function on [4], the orbits of the pseudogroup of local isometries of any meromorphic geometric structure on S contain the ﬁbers of S. Proof. i) The projection of the set of poles of ∇ on is a closed analytic set that is either a ﬁnite set of points or all of . If this projection is and ∇ is invariant by the principal ﬁbration, then each point of S is a pole, which is impossible. This proves the easy sense of the implication. We will now describe the space of connections ∇, such that only a ﬁnite set of ﬁbers contain poles of ∇. Up to a ﬁnite unramiﬁed cover and a ﬁnite quotient, S admits holomorphic afﬁne structures (i.e., ﬂat torsion-free holomorphic afﬁne connections) that can be built in the following way [36]: Consider a discrete torsion-free subgroup in PSL(2, R) such that = \H, with H the upper-half plane. Take any holomorphic projective structure on , its developing map τ : H → P 1 (C), and its holonomy morphism ρ : → PSL(2, C). This embedding of into PSL(2, C) lifts to SL(2, C) (this

52 / s. dumitrescu

follows from the fact that orientable (real) closed 3-manifolds have a trivial second Stiefel-Whitney class [47]). Choose such a lift and consider as a subgroup of SL(2, C). Denote by W = C2 \ {0} the C∗ -tautological bundle over P 1 (C). The canonical afﬁne structure of C2 induces a -invariant afﬁne structure on W and hence a -invariant holomorphic afﬁne structure on the pull-back τ ∗ (W ) C∗ × H. The -action on τ ∗ (W ) comes from the action by deck transformation on H and from the ρ-action on W . The previous holomorphic afﬁne structure on τ ∗ (W ) is also invariant by the homotheties in the ﬁbers (which commute with the -action). Consider now Z a lattice in C∗ , which acts by multiplication on the ﬁbers of τ ∗ (W ) and takes the quotient of τ ∗ (W ) by × . The quotient is a principal elliptic bundle over , with ﬁber \C∗ , biholomorphic to S. The afﬁne structure inherited by the universal cover C × H of S is the pull-back of the previous afﬁne structure on C∗ × H by the map C × H → C∗ × H

(z, ξ ) → (ez , ξ ). In the following we will consider the ﬂat torsion-free holomorphic afﬁne connection ∇0 on S given by the case whereτ is the standard embedding of a b H into P 1 (C). In this case, the action of γ = ∈ ⊂ SL(2, R) on the c d universal cover C × H of S is easily seen to be given by [36]: γ (z, ξ ) = (z + log(cξ + d), γ ξ ), ∀(z, ξ ) ∈ C × H, where log is a determination of the logarithm and the γ -action on H comes from the standard action of SL(2, R) on H. The difference ∇ − ∇0 is a meromorphic (2, 1)-tensor ω on S, or, equivalently, a × -invariant mermorphic (2,1)-tensor ω˜ on the universal cover C × H. Moreover, ω˜ is invariant by t(z, ξ ) = (z + 2iπ, ξ ). Then we have ∂ ∂ ∂ + f12 (z, ξ )dz ⊗ dξ ⊗ + f21 (z, ξ )dz ⊗ dz ⊗ ω˜ = f11 (z, ξ )dz ⊗ dz ⊗ ∂z ∂z ∂ξ + f22 (z, ξ )dz ⊗ dξ ⊗

∂ ∂ ∂ + g11 (z, ξ )dξ ⊗ dz ⊗ + g12 (z, ξ )dξ ⊗ dξ ⊗ ∂ξ ∂z ∂z

+ g21 (z, ξ )dξ ⊗ dz ⊗

∂ ∂ + g22 (z, ξ )dξ ⊗ dξ ⊗ , ∂ξ ∂ξ

with fij , gij meromorphic functions on C × H, and fij (·, ξ ), gij (·, ξ ) holomorphic except for ξ lying in the union of a ﬁnite number of -orbits in H.

meromorphic almost rigid geometric structures / 53

Notice that the difference between ∇0 and the standard afﬁne structure of C × H is given by f11 = f22 = g21 = 1, the others fij , gij being trivial (see the straightforward computation in [36]). Since ω˜ is -invariant and t-invariant, the functions fij (·, ξ ) and gij (·, ξ ) descend on an elliptic curve for all ξ ∈ H. They are constant for all ξ ∈ H for which they are holomorphic. It follows that fij (·, ξ ) and gij (·, ξ ) are constant for ξ lying in a open dense subset of H and, consequently, for all ξ ∈ H. It follows that the functions fij and gij depend only on ξ . Consequently, the ﬂow ∂ of ∂z preserves ω. ˜ Since the fundamental generator of the principal ﬁbration ∂ also preserves ∇0 , it is a Killing ﬁeld for ∇. ∂z ii) In the following we consider only torsion-free connections: f12 = g11 and f22 = g21 . The -invariance of ω˜ yields the following equations: 1) f11 (ξ ) = f11 (γ ξ ) − cf21 (γ ξ )(cξ + d) 2) f12 (ξ ) = f12 (γ ξ )(cξ + d)−2 − 2c 2 f21 (γ ξ ) − cf22 (γ ξ )(cξ + d)−1 + 2cf11 (γ ξ )(cξ + d)−1 3) f21 (ξ ) = f21 (γ ξ )(cξ + d)2 4) f22 (ξ ) = 2f21 (γ ξ )c(cξ + d) + f22 (γ ξ ) 5) g12 (ξ ) = g12 (γ ξ )(cξ + d)−4 + c 2 f11 (γ ξ )(cξ + d)−2 + cf12 (γ ξ )(cξ + d)−3 − c 3 f21 (γ ξ )(cξ + d)−1 − c 2 f22 (γ ξ )(cξ + d)−2 − cg22 (γ ξ )(cξ + d)−3 6) g22 (ξ ) = g22 (γ ξ )(cξ + d)−2 + cf22 (γ ξ )(cξ + d)−1 + c 2 f21 (γ ξ ), a b for all γ = ∈ . c d The equation (3) implies that f21 is a meromorphic vector ﬁeld on the compact Riemann surface . Since a single pole of order at most 1 is allowed, we get f21 = 0 as a direct consequence of Riemann-Roch theorem and Serre duality ([27], page 245). The equations (1) and (4) imply then that f11 and f22 are meromorphic functions on with a single pole of order at most 1. If f11 , f22 are not constant, they give a biholomorphism between and the projective line P 1 (C) [27]: a contradiction, since the genus of is ≥ 2. We conclude that f11 and f22 are constants. Then we have 2 ) f12 (ξ ) = f12 (γ ξ )(cξ + d)−2 − cf22 (cξ + d)−1 + 2cf11 (cξ + d)−1 5 ) g12 (ξ ) = g12 (γ ξ )(cξ + d)−4 + c 2 (f11 − f22 )(cξ + d)−2 + c(f12 − g22 )(γ ξ )(cξ + d)−3 6 ) g22 (ξ ) = g22 (γ ξ )(cξ + d)−2 + cf22 (cξ + d)−1 .

54 / s. dumitrescu

It follows from (2 ) and (6 ) that f12 and g22 are quasimodular forms of weight 2 on with a single simple pole. The space of those quasimodular forms is a complex vector space of dimension g + 1 (see [3], theorem 9). Equation (5 ) is equivalent to the -invariance of the quadratic differen − g . It follows that 2g + f − g is a tial w(ξ )dξ 2 , where w = 2g12 + f12 12 22 12 22 meromorphic quadratic differential on with a single pole of order at most 2. It is classically known (as an application of Riemann-Roch theorem and Serre duality) that the space of quadratic differentials with a single pole of order at most 2 is of complex dimension 3g − 1 (see, for example, [27]). ∂ ∂ iii) Let X = a(z, ξ ) ∂z + b(z, ξ ) ∂ξ be a local holomorphic Killing ﬁeld of a generic connection ∇, in the neighborhood of a point where ∇ is holomorphic (a and b are holomorphic local functions on C × H). The equation of the Killing ﬁeld is [X , ∇Y Z] = ∇[X ,Y ] Z + ∇Y [X , Z] for all Y , Z tangents to S. It is enough to verify the equation for (Y , Z) ∂ ∂ ∂ ∂ ∂ ∂ , ∂ξ , ∂z , ∂z , and ∂ξ , ∂ξ . corresponding to ∂z This leads to the following PDE system: 1) azz + (1 + f11 )az + 2f12 bz = 0 2) bzz + (1 + 2f22 − f11 )bz = 0 3) azξ + (f11 − f22 )aξ + g12 bz + g11 bξ + 4) bzξ + (1 + f22 )az + (g22 − f12 )bz = 0

∂f12 b=0 ∂ξ

5) aξ ξ − g12 az + (2f12 − g12 )aξ + 2g12 bξ + 6) bξ ξ + 2(1 + f22 )aξ − g12 bz + g22 bξ +

∂g12 b=0 ∂ξ

∂g22 b = 0. ∂ξ

The general solution of the ﬁrst equation is b = ν(ξ )e−μz + C(ξ ), with ν, C holomorphic functions of ξ and μ = 1 + 2f22 − f11 . We replace bz in the ﬁrst equation and we get μ f12 (ξ )ν(ξ )e−μz + A(ξ )e−(1+f11 )z , az = f11 − f22 with A a holomorphic function of ξ . Then equation (4) leads to 1 + f11 f12 − g22 ν(ξ ) e−μz + (1 + f22 )A(ξ )e−(1+f11 )z = 0. μ −ν (ξ ) + f11 − f22 For a generic ∇, we have f11 = f22 , thus μ = 1 + f11 and the functions are C-linearly independent. This implies (1 + f22 )A(ξ ) = 0 and, since for a generic ∇, f22 = −1, we have A(ξ ) = 0.

e−μz , e−(1+f11 )z

meromorphic almost rigid geometric structures / 55

We also get I) ν (ξ ) = ν(ξ )

1 + f11 f12 − g22 . f11 − f22

Now we check equation (3). We replace the partial derivatives of a and b in (3) and we get the following: f12 f12 μ ν (ξ ) − g12 − ν(ξ ) e−μz + ( f12 C) = 0. f11 − f22 f11 − f22 Since generically μ = 0, the functions e−μz and 1 are linearly independent, which yields to ( f12 C) = 0 and to II) ν (ξ ) = ν(ξ )

1 f12

[( f11 − f22 )g12 − f12 ].

Relations (I) and (II) are compatible, for a generic connection, only if ν = 0. This implies b = C(ξ ), a = B(ξ ). Our PDE system becomes: b + f b = 0 3 ) ( f11 − g21 )a + f12 12 b=0 5 ) a + (2f12 − g12 )a + 2g12 b − g12 6 ) b + 2(1 + 2g21 )a + g22 b + g22 b = 0.

Since (f12 b) = 0, we have a = 0 and thus a is a constant function. Equation b = 0, which, for a generic ∇, is compatible with implies then 2g12 b − g12 f12 b + f12 b = 0 only if b = 0. ∂ It follows that X is a constant multiple of ∂z . iv) The projective connection associated to the afﬁne connection ∇ is given by the following second-order ODE [14]: (5 )

ξ = K 0 (z, ξ ) + K 1 (z, ξ )ξ + K 2 (z, ξ )(ξ )2 + K 3 (z, ξ )(ξ )3 , where K 0 = −f21 , = 0, K 1 , = (1 + f11 ) − 2(1 + f22 ), K 2 , = −(g22 − 2f12 ) and K 3 = g12 (see [14]). Liouville [42], followed by Tresse [56] and Cartan [14], proved that this projective connection is projectively ﬂat if and only if both of the following invariants vanish: 1 2 − Kzz − 3Kξ0ξ − 6K 0 Kz3 − 3K 3 Kz0 + 3K 0 Kξ2 + 3K 2 Kξ0 + K 1 Kz2 L1 = 2Kzξ

− 2K 1 Kξ1 , 2 3 − Kξ1ξ − 3Kzz + 6K 3 Kξ0 + 3K 0 Kξ3 − 3K 3 Kz1 − 3K 1 Kz3 − K 2 Kξ1 L2 = 2Kzξ

+ 2K 2 Kz2 .

56 / s. dumitrescu

Here K 0 = 0, K 1 is a constant function and K 2 , K 3 depend only on ξ . This implies the vanishing of both invariants L1 and L2 . I would like to thank G. Chenevier and G. Dloussky for helpful conversations.

References [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11]

[12] [13] [14] [15] [16] [17] [18] [19]

G. D’Ambra, M. Gromov, Lectures on transformations groups: geometry and dynamics, Surveys in Differential Geometry (Cambridge), (1990), 19–111. A. M. Amores, Vector ﬁelds of a ﬁnite type G-structure, J. Differential Geom., 14(1), (1979), 1–6. N. Azaiez, The ring of quasimodular forms for a cocompact group, J. Number Theory, (2007). W. Barth, K. Hulek, C. Peters, A. Van De Ven, Compact complex surfaces, Ergebnisse der Mathematik, Vol. 4, 2nd Ed., Berlin, Springer, (1995). A. Beauville, Variétés Kähleriennes dont la première classe de Chern est nulle, J. Differential Geom., 18, (1983), 755–782. Y. Benoist, Actions propres sur les espaces homogènes réductifs, Annals of Mathematics, 144, (1996), 315–347. Y. Benoist, Orbites de structures rigides, integrable systems and foliations (Montpellier), Boston, Birkaüser, (1997). Y. Benoist, F. Labourie, Sur les espaces homogènes modèles des variétés compactes, Publ. Math. I.H.E.S., 76, (1992), 99–109. J. Benveniste, D. Fisher, Nonexistence of invariant rigid structures and invariant almost rigid structures, Comm. Anal. Geom., 13(1), (2005), 89–111. I. Biswas, B. McKay, Holomorphic Cartan geometries and Calabi-Yau manifolds, arXiv math. AG/0812.3978, (2010). F. Campana, J-P. Demailly, T. Peternell, The algebraic dimension of compact complex threefolds with vanishing second Betti number, Compositio Math., 112(1), (1998) 77– 91. A. Candel, R. Quiroga-Barranco, Gromov’s centralizer theorem, Geom. Dedicata, 100, (2003), 123–155. Y. Carrière, Autour de la conjecture de L. Markus sur les variétés afﬁnes, Invent. Math., 95, (1989), 615–628. E. Cartan, Sur les variétés à connexion projective, Bull. Soc. Math. France, 52, (1924), 205–241. S. Dumitrescu, Structures géométriques holomorphes sur les variétés complexes compactes, Ann. Scient. Ec. Norm. Sup., 34(4), (2001), 557–571. S. Dumitrescu, Métriques riemanniennes holomorphes en petite dimension, Ann. Instit. Fourier, Grenoble, 51(6), (2001), 1663–1690. S. Dumitrescu, Homogénéité locale pour les métriques riemanniennes holomorphes en dimension 3, Ann. Instit. Fourier, Grenoble, 57(3), (2007), 739–773. S. Dumitrescu, Une caractérisation des variétés complexes compactes parallélisables admettant des structures afﬁnes, C.R. Acad. Sci. Paris, Ser. I, 347, (2009), 1183–1187. S. Dumitrescu, Killing ﬁelds of holomorphic Cartan geometries, to appear in Monatshefte fur Mathematik, (2010).

meromorphic almost rigid geometric structures / 57 [20] S. Dumitrescu, Connexions afﬁnes et projectives sur les surfaces complexes compactes, Mathematische Zeitschrift, 264(2), (2010), 301–316. [21] S. Dumitrescu, A. Zeghib, Global rigidity for holomorphic Riemannian metrics on compact threefolds, Math. Ann., 345(1), (2009), 58–81. [22] R. Feres, Rigid geometric structures and actions of semisimple Lie groups, Rigidité, groupe fondamental et dynamique, Panorama et synthèses, Soc. Math. France, Paris, 13, (2002). [23] C. Frances, Une preuve du théorème de Liouville en géométrie conforme dans le cas analytique, Enseign. Math., 49(2), (2003), 95–100. [24] D. Fried, W. Goldman, Three-dimensional afﬁne christallographic groups, Adv. Math., 47(1), (1983), 1–49. [25] E. Ghys, Déformations des structures complexes sur les espaces homogènes de SL(2, C), J. Reine Angew. Math., 468, (1995), 113–138. [26] W. Goldman, Nonstandard Lorentz space forms, J. Diff. Geom., 21(2), (1985), 301–308. [27] P. Grifﬁths, J. Harris, Principles of algebraic geometry, New York, Wiley-Interscience, (1994). [28] M. Gromov, Rigid transformation groups, Géométrie Différentielle, (D. Bernard et Choquet-Bruhat Ed.), Travaux en cours, Hermann, Paris, 33, (1988), 65–141. [29] R. Gunning, Lectures on Riemann surfaces, Princeton, NJ, Princeton University Press, (1966). [30] A. Huckleberry, S. Kebekus, T. Peternell, Group actions on S6 and complex structures on P 3 (C), Duke Math. J., 102(1), (2000), 101–124. [31] J-M. Hwang, N. Mok, Uniruled projective manifolds with irreducible reductive Gstructure, J. Reine Angew. Math., 490 (1997), 55–64. [32] E. Ince, Ordinary differential equations, New York, Dover Publications, (1944). [33] M. Inoue, On surfaces of class VII0 , Invent. Math., 24, (1974), 269–310. [34] M. Inoue, S. Kobayashi, T. Ochiai, Holomorphic afﬁne connections on compact complex surfaces, J. Fac. Sci. Univ. Tokyo, 27(2), (1980), 247–264. [35] A. Kirilov, Eléments de la théorie des représentations, Moscow, M.I.R., (1974). [36] B. Klingler, Structures afﬁnes et projectives sur les surfaces complexes, Ann. Inst. Fourier, Grenoble, 48(2), (1998), 441–477. [37] B. Klingler, Complétude des variétés Lorentziennes à courbure sectionnelle constante, Math. Ann., 306, (1996), 353–370. [38] T. Kobayashi, T. Yoshino, Compact Clifford-Klein form of symmetric spaces—revisited, Pure Appl. Math. Q., 1(3), (2005), 591–663. [39] R. Kulkarni, F. Raymond, 3-dimensional Lorentz space-forms and Seifert ﬁber spaces, J. Diff. Geom., 21(2), (1985), 231–268. [40] C. Lebrun, Spaces of complex null geodesics in complex-Riemannian geometry, Trans. Amer. Math. Soc., 278, (1983), 209–231. [41] C. Lebrun, H-spaces with a cosmological constant, Proc. Roy. Soc. London, Ser. A, 380(1778), (1982), 171–185. [42] R. Liouville, Sur les invariants de certaines équations différentielles et sur leurs applications, Journal de l’Ecole Polytechnique, 59, (1889), 7–76. [43] K. Maehara, On elliptic surfaces whose ﬁrst Betti numbers are odd, Intl. Symp. on Alg. Geom., Kyoto, (1977), 565–574. [44] L. Markus, Cosmological models in differential geometry, mimeographed notes, Univ. of Minnesota, (1962).

58 / s. dumitrescu [45] B. McKay, Characteristic forms of complex Cartan geometries, arXiv math. DG/ 0704.2555, (2009). [46] J. Milnor, Curvatures of left invariant metrics on Lie groups, Adv. in Math., 21, (1976), 293–329. [47] J. Milnor, J. Stasheff, Characteristic classes, Princeton, Princeton University Press, (1974). [48] K. Nomizu, On local and global existence of Killing vector ﬁelds, Ann. of Math. (2), 72, (1960), 105–120. [49] V. Popov, E. Vinberg Invariant theory, Algebraic Geometry 4, E.M.S., 55, (1991), 123– 280. [50] M. Raghunathan, Discrete subgroups of Lie groups, Berlin, Springer-Verlag, (1972). [51] S. Rahmani, Métriques de Lorentz sur les groups de Lie unimodulaires de dimension 3, J. Geom. Phys., 9, (1992), 295–302. [52] N. Rahmani, S. Rahmani, Lorentzian geometry of the Heisenberg group, Geom. Dedicata, 118, (2006), 133–140. [53] F. Salein, Variétés anti-de Sitter de dimension 3 exotiques, Ann. Inst. Fourier, Grenoble, 50(1), (2000), 257–284. [54] T. Suwa, Compact quotient spaces of C2 by afﬁne transformation groups, J. Diff. Geom. 10, (1975), 239–252. [55] W. Thurston, The geometry and the topology of 3-manifolds, Princeton, Princeton University Press, (1983). [56] A. Tresse, Détermination des invariants ponctuels de l’équation différentielle ordinaire du second ordre y = ω(x, y, y ), Leipzig, 87 S, gr. 8, (1896). [57] K. Ueno, Classiﬁcation theory of algebraic varities and compact complex spaces, Lect. Notes in Math., 439, Berlin, Springer-Verlag, (1975). [58] A. Vitter, Afﬁne structures on compact complex manifolds, Invent. Math., 17, (1972), 231–244. [59] C. Wall, Geometric structures on compact complex analytic surfaces, Topology, 25(2), (1986), 119–153. [60] H. Wang, Complex parallelisable manifolds, Proc. Amer. Math. Soc. 5, (1954), 771–776. [61] J. Wolf, Spaces of constant curvature, New York, McGraw-Hill, (1967). [62] D. Zagier, Lecture, Cours at Collège de France, 2000–2001. [63] A. Zeghib, On Gromov’s theory of rigid transformation groups: a dual approach, Ergodic Th. Dyn. Syst., 20(3), (2000), 935–946. [64] R. Zimmer, On the automorphism group of a compact Lorentz manifold and other geometric manifolds, Invent. Math., 83, (1986), 411–426. [65] R. Zimmer, Actions of semisimple groups and discrete subgroups, Proc. Internat. Cong. Math., Berkeley, CA, (1986), 1247–1258. [66] R. Zimmer, Automorphisms groups of geometric manifolds, Proc. Symp. Pure Math., 54(3), (1993), 693–710. [67] R. Zimmer, Ergodic theory and semisimple Lie groups, Boston, Birkhaüser, (1984).

3

renato feres and emily ronshausen

HARMONIC FUNCTIONS OVER GROUP ACTIONS to bob zimmer on his

60 th

birthday

Abstract Let X be a compact space, a countable group of homeomorphisms of X, and μ a probability measure on . From this data one naturally deﬁnes a random walk on X and a related notion of continuous harmonic function. The question we wish to investigate is whether all continuous harmonic functions on X are -invariant. If this is the case, we say that the system satisﬁes the topological Liouville property. We show that if μ is a symmetric probability measure on an arbitrary and X = S1 or the interval [0, 1], then the Liouville property always holds. In addition, when the Poisson boundary of (, μ) can be identiﬁed with the circle, we give a general construction of a non-Liouville action of on S2 , generalizing an example from [FZ2].

1. Introduction A useful idea in the study of group actions without invariant measures is to introduce and study a random Z-action associated to a choice of probability measure on the acting group. To ﬁx notations, let denote a group of transformations of a space X and μ a probability measure on . Then μ deﬁnes a random walk on , which induces a random Z-action on X . See, for example, [Fu] where this point of view is well represented. It is often assumed in this context that the -space admits a μ-stationary measure (also called μ-harmonic measure—see deﬁnition below) of full support, endowing the random Z-action with an invariant measure. This brings along the ergodic theory apparatus that relies on the existence of invariant measures. Often, however, the -space has large regions with possibly interesting dynamics that are not detected by any harmonic measure for a given μ. If this is the case, one can hope to gain some insight into the system represented by (X , , μ) by investigating the space of μ-harmonic functions on X . As pointed out in Proposition 2.4 below, continuous μ-harmonic functions on compact -spaces are -invariant on the support of any μ-harmonic probability measure, so nontrivial (i.e., non--invariant) such functions may yield 59

60 / r. feres and e. ronshausen

information about the regions that harmonic measures do not detect. This is the perspective motivating this paper. We have chosen here to consider compact -spaces and continuous functions, although one may also try to pursue the subject on measurable -spaces. The most basic question to ask about (X , , μ) in the present context is whether it satisﬁes a topological Liouville property, i.e., whether all continuous μ-harmonic functions are -invariant. The main result of the paper is that the Liouville property holds for actions of countable groups on 1-dimensional spaces when μ is a symmetric probability measure.

theorem 1.1. Let be a countable group acting on X by homeomorphisms, where X is either the circle S1 or the interval [0, 1]. Let μ be a symmetric probability measure on . Then the Liouville property holds for (X , , μ). We give at the end of the paper an example of -space for which the Liouville property does not hold. More speciﬁcally, we show that if the Poisson boundary of (, μ) can be identiﬁed with the circle S1 , then it is possible to construct a -action on the two-sphere so that (S2 , , μ) admits continuous μ-harmonic functions that are nonconstant on all orbits in the complement of a pair of ﬁxed points. The action can be chosen to be ergodic with respect to the smooth measure class on S2 , whereas the random walk is transient in the complement of the two ﬁxed points. A general recipe for constructing non-Liouville actions is also proposed. The proof of Theorem 1.1 parallels the proof of the main theorem of [FFP]. That paper is about the Liouville property for codimension-one foliated Brownian motion and harmonic functions for the Laplace-Beltrami operator associated to a Riemannian metric on leaves, rather than discrete group actions. Although the two situations cannot be directly compared, our assumptions on and μ are, in a sense, much more general than related assumptions on that paper. The example of a non-Liouville action on S2 generalizes a similar construction given in [FZ2] in the context of foliations with a Riemannian metric on leaves. See, also, [FZ1] for these issues in the setting of holomorphic functions on foliations. The paper [DK] has some deep results about harmonic functions on foliated spaces in the spirit of the present discussion. This paper was partly written during a visit by the ﬁrst author to the ParisSud 11 University, at Orsay. He wishes to thank the group of Topology and Dynamics for their hospitality and the university for its generous support. He also wishes to thank Vadim Kaimanovich and Bertrand Deroin for helpful and

harmonic functions over group actions / 61

illuminating discussions. In addition, we wish to express our thanks to the anonymous referee for several helpful comments.

2. Harmonic Functions on -spaces In this section we collect some basic properties and general remarks concerning harmonic functions. Let be a countable (inﬁnite) group, μ a probability measure on , and X a compact topological space on which acts by homeomorphisms. It will always be assumed that μ is non-degenerate, that is, the support of μ generates as a semigroup. In particular, μ(γ ) < 1 for all γ ∈ . A continuous function f : X → R is said to be μ-harmonic if, for all x ∈ X , f (x) =

f (γ (x))μ(γ ).

γ ∈

Let H(X , , μ) denote the set of μ-harmonic functions on X and C(X ) the subset consisting of continuous -invariant functions. If H(X , , μ) = C(X ) , we say that (X , , μ) satisﬁes the Liouville property. We note that if f ∈ H(X , , μ) and x ∈ X , then f˜x (γ ) := f (γ −1 (x)) is a harmonic function on in the following sense: f˜x (γ ) =

f˜x (γ η)μ(η−1 ).

η∈

ˆ Under the conventions of [Fu], f˜x is harmonic relative to the right-μ-random walk on , where μ(γ ˆ ) = μ(γ −1 ). Thus, if (, μ) ˆ itself has the Liouville property, that is, if bounded harmonic functions on are constant, then necessarily (X , , μ) has the Liouville property for all X . Therefore, the present subject is uninteresting if the Poisson boundary of (, μ) ˆ reduces to a single point. It is known, for example, that if is amenable, then there exists a symmetric (μ = μ) ˆ nondegenerate μ for which the Poisson boundary is trivial. A good example of a group to keep in mind, on the other hand, is the fundamental group of a closed surface of genus at least 2, represented as isometries of the Poincaré disc. It has a nontrivial boundary for many interesting probability measures. For example, a result due to Furstenberg implies that on this group there exists a probability measure μ relative to which the boundary of coincides with the geometric boundary, S1 , of the disc [Fur].

62 / r. feres and e. ronshausen

Let = P − I be the Laplace operator on the set C(X ) of real-valued continu ous functions on X , where (Pf )(x) := μ(γ )f (γ (x)) is the averaging operator and I is the identity. Clearly, f is harmonic iff f = 0. More generally, f satisfying f ≥ 0 (respectively, f ≤ 0) is said to be superharmonic (respectively, subharmonic).

proposition 2.1. (maximum principle) If a superharmonic f ∈ C(X ) is bounded above by C and f (x) = C for an x ∈ X , then f is constant on the -orbit of x. Proof. This trivially follows from μ being nondegenerate and the remark 0≤

( f (γ (x)) − f (x))μ(γ ) ≤

γ ∈

( f (γ (x)) − C)μ(γ ) ≤ 0,

γ ∈

so f (γ (x)) = C for all γ in the support of μ.

corollary 2.2. If the closure of every -orbit contains a unique minimal set, then the Liouville property holds. Proof. Let X0 be the closure of an orbit. The maximum and minimum values in X0 of a harmonic function are attained on orbits where the function is constant. Since X0 has a single minimal set, these two values must agree. It is useful to introduce a (directed) metric dx0 on the orbit x0 as follows. Given x, y in x0 , we specify a path joining x to y by a sequence α = (γ0 , γ1 , . . . , γn−1 ), μ(γi ) > 0, such that y0 = x, yn = y, and yi+1 = γi yi for i = 0, . . . , n − 1. The length of α is deﬁned as L(α) =

n−1

− ln μ(γi ) ∈ (0, ∞].

i=0

We allow x to be joined to itself by the trivial path, deﬁned by n = 0, α = ∅, and L(α) = 0. Now let dx0 (x, y) = inf L(α), where the inﬁmum is taken over all n and all (γ0 , . . . , γn−1 ) describing a path from x to y.

proposition 2.3. (harnack inequality) Let f be a continuous, positive, subharmonic function on X and x0 ∈ X arbitrary. Then for all x, y ∈ x0 ,

harmonic functions over group actions / 63

f (y) ≤ edx0 (x,y) . f (x) Proof. As f is subharmonic, f (x) ≥ γ ∈ μ(γ )f (γ x) ≥ μ(η)f (ηx) for all η ∈ , so f (ηx)/f (x) ≤ e− ln μ(η) . Multiplying such inequalities along any path α from x to y yields the claim. Given (X , , μ) as above, a measure ν on X is called harmonic if ν is quasi-invariant and μ ∗ ν = ν, where the convolution μ ∗ ν is the image of the product measure μ ⊗ ν under the action map × X → X . It is easily shown that μ ∗ ν = νP, where the averaging operator P acts on ν by duality as νP =

μ(γ )γ∗ ν.

γ ∈

The following useful remark is due, in the context of foliated Brownian motion, to Garnett [Gar]. The proof, which is sketched below for completeness, is adapted from Paulin [Pa].

proposition 2.4. Let ν be a harmonic probability measure on (X , , μ). Then every h in H(X , , μ) is -invariant on the support of ν. Proof. It may be assumed that h is non-negative. Let c be a non-negative constant and deﬁne hc (x) as the minimum of {h(x), c}. Note that Phc ≤ hc . Since ν is harmonic, hc dν = hc d(μ ∗ ν) = Phc dν, therefore (Phc − hc )dν = 0. As the functions involved are continuous, it follows that Phc = hc on the support of ν. Applying the maximum principle to hc |supp(ν) , we conclude that h −1 ([c, ∞)) ∩ supp(ν) is -invariant for all c. This implies that h is -invariant on supp(ν). Let C denote the largest -invariant subset of X such that h|C is -invariant for all h ∈ H(X , , μ). Clearly, C is closed. It is also nonempty since, by the maximum principle, it contains every minimal set in X . Thus if M denotes the closure of the union of all the minimal sets of X relative to the action, then M ⊆ C . Due to Proposition 2.4, the set S , deﬁned as the closure of the union of supports of all harmonic measures, is also contained in C . In fact, it is easily seen that M ⊆ S ⊆ C . It will be explained shortly that the random walk associated to (X , , μ) (deﬁned next) converges to C with probability 1.

64 / r. feres and e. ronshausen

3. Random Walk on (X, , μ) Let (X , , μ) be, as above, a compact -space where is a countable group and μ is a probability measure on . The measure induces a Markov transition kernel on X by setting p(x, y) = 0 if y ∈ / x and μ(γ ), p(x, y) = y=γ x

where the sum is over all γ ∈ such that γ x = y. Let Xn denote the random walk on X with transition probabilities p and initial state x0 . For the set C , deﬁned at the end of the previous section, we have the following:

proposition 3.1. The random walk Xn on X \ C is transient. The proposition means that for all x0 ∈ X \ C and every open U containing C , there exists a random integer N, almost surely ﬁnite, such that Xn ∈ U for all n ≥ N. Prior to proving this fact we need to review some information about the Poisson boundary of . Our main source is Furman [Fu]. The boundary of can be described as a compact Hausdorff -space B endowed with a probability measure of full support, ν, satisfying the following properties: 1) ν is harmonic relative to (, μ); ˆ ˆ 2) For almost every sample path γ¯ = (γ1 , γ2 , . . .) of a (right) μ-random walk on , a limit probability νγ¯ = limn→∞ (γ1 . . . γn )∗ ν exists and is equal to a Dirac measure δγ∞ supported on a random point, η∞ , of B; 3) Let H ∞ (, μ) ˆ denote the space of bounded harmonic functions on relative to μ. ˆ Then the map F : L∞ (B, ν) → H ∞ (, μ) ˆ given by F(φ)(γ ) = φ dγ ν is an isometric bijection such that, for h = F(φ), ∗ B φ(η∞ ) = lim h(γ1 . . . γn ). n→∞

It is observed in [Fu], remark 2.16(i), that if is a discrete group, then, for the topological description of B given there (denoted B in Furman’s article), L∞ (B, ν) coincides with the continuous functions on B. In this case, B is nonmetrizable if has a nontrivial Poisson boundary. However, measured theoretically, the Poisson boundary is always a Lebesgue space.

lemma 3.2. Let ηn = γ1 . . . γn be a right-μ-random ˆ walk starting at the unit element e ∈ . Let h ∈ H ∞ (, μ) ˆ and C a positive constant. Then, for a.e. sample

harmonic functions over group actions / 65

path ηn and every sequence ξn ∈ such that de (ηn , ξn ) < C, the sequences h(ηn ) and h(ξn ) converge to the same value. (Here de is the directed metric used in Proposition 2.3.) Proof. Deﬁne νγ = γ∗ ν. From the above properties of the Poisson boundary and Proposition 2.3, it follows that

dνηn dνξn , max dνξn dνηn

≤ eC

for all n. But since νηn converges to the Dirac measure δη∞ , the sequence νξn must converge to the same measure. We now return to Proposition 3.1. Let h ∈ H(X , , μ). As pointed out above, for each x ∈ X \ C we obtain a harmonic function h˜ x ∈ H ∞ (, μ). ˆ By Lemma 3.2, for each C > 0, for a.e. sample path (x0 , x1 , . . .) of the random walk on X , and for every sequence (y0 , y1 , . . .) such that dx (xi , yi ) < C for all i, both sequences h˜ x (xi ) and h˜ x (yi ) converge and have the same limit. Therefore, as C is arbitrary and h is continuous, h must be constant on the orbit of every limit point of xi . But this is to say that all limit points must lie in C . This concludes the proof of Proposition 3.1. One further simple remark, concerning induced random walks for subgroups, is needed. Let 0 be a ﬁnite-index subgroup of . Let N γ denote the ﬁrst time, n ≥ 1, at which random walk on (, μ) ˆ starting at γ reaches 0 . Then N γ is a Markov time, almost surely ﬁnite. Now deﬁne a probability measure on 0 by μˆ 0 (η) = Prob(γN e = η), that is, μˆ 0 (η) is the probability that random walk, (γ0 , γ1 , . . . ), on , starting at the unit element, will ﬁrst return to 0 at η. General properties of Markov times and martingales imply that the restriction to 0 of a function in H ∞ (, μ) ˆ is harmonic relative to μˆ 0 .

lemma 3.3. Let 0 have ﬁnite index in and μ0 the image of the induced measure μˆ 0 under group inverse. Then H(X , , μ) ⊆ H(X , 0 , μ0 ). In particular, if (X , 0 , μ0 ) satisﬁes the Liouville property, then so does (X , , μ). Proof. This is due to the above remarks about μ0 and the relationship between harmonic functions h on (X , , μ) and h˜ x on (, μ) ˆ or (0 , μˆ 0 ). (See the beginning of Section 2.)

66 / r. feres and e. ronshausen

4. Actions on the Circle We now consider actions of discrete (countable) groups on the circle. The following lemma is a well-known group-actions version of Poincaré’s classiﬁcation of circle homeomorphisms.

lemma 4.1. Let be a countable group of homeomorphisms of S1 . Then one of the following holds: 1) The action is minimal; 2) The action is not minimal and there is a unique minimal set; 3) There is a ﬁnite orbit. Proof. It sufﬁces to prove the lemma for a subgroup of ﬁnite index. Thus we may assume that is orientation preserving. Further assume that the action is not minimal and contains no ﬁnite orbit. Let Z be a minimal set and U an arbitrary connected component of S1 \ Z. This minimal set is unique if the orbit of every point in U can be shown to accumulate on Z. Now, U and γ U either are disjoint or agree for each γ ∈ . If the latter, then the endpoints of U are ﬁxed by γ . As the subgroup of ﬁxing the endpoints of U has inﬁnite index, there are inﬁnitely many disjoint intervals of the form γ U, so we can choose a sequence γn U decreasing to 0 in length. Therefore, the orbit of every point in U must limit on Z. By Corollary 2.2, having a unique minimal set implies the Liouville property. Thus, in order to prove Theorem 1.1, it sufﬁces to assume that the action on S1 contains a ﬁnite orbit. By Lemma 3.3, it can be assumed that has a ﬁxed point in S1 and preserves orientation. This reduces the proof of Theorem 1.1 to showing the Liouville property for orientation-preserving actions on the interval [0, 1].

5. Actions on the Interval We now restrict attention to systems ([0, 1], , μ) for which the -action is orientation preserving. Clearly, then, the only ﬁnite orbits are ﬁxed points, and since the union of the ﬁxed points is a closed subset of [0, 1], we may further restrict attention to intervals without interior ﬁnite orbits.

lemma 5.1. If acts by orientation-preserving homeomorphisms of [0,1] without interior ﬁnite orbits, the orbit of every x ∈ (0, 1) must limit on both 0 and 1.

harmonic functions over group actions / 67

Proof. The supremum and inﬁmum of x are easily seen to be ﬁxed points.

lemma 5.2. Suppose that the Liouville property does not hold for ([0, 1], , μ) and the -action is orientation preserving without interior ﬁnite orbits. Then, 1) There is a unique continuous harmonic f not -invariant such that f (0) = 0, f (1) = 1; 2) f (x) is the probability that random walk on ([0, 1], , μ) starting at x converges to 1; 3) f is not constant on any interior orbit; 4) f is increasing. Proof. Let g be a continuous, harmonic, non--invariant function on [0, 1]. By Lemma 5.1 and the maximum principle, g(0) = g(1), and g(x) lies in the open interval with endpoints g(0), g(1) for each x ∈ (0, 1). Lemma 5.1 also implies that g cannot be constant on any interior orbit, so the set C of Proposition 3.1 coincides with {0, 1}. By composing g with an appropriate afﬁne function of R, one obtains f such that f (i) = i for i = 0, 1 and 0 < f (x) < 1 on interior points. Uniqueness is due to the maximum principle. By Proposition 3.1, the random walk Xnx on [0, 1] starting at x must converge to a random point X∞ ∈ {0, 1} with probability 1. As f is harmonic, the expected value E[ f (Xnx )] is equal to f (x) and limn→∞ f (Xn ) exists almost surely. By continuity, f (Xnx ) converges to either 0 or 1. Therefore, f (x) = lim E[ f (Xnx )] = f (0)Prob(Xnx → 0) + f (1)Prob(Xnx → 1) n→∞

= Prob(Xnx → 1). Finally, given any x1 < x2 and a sample path (γ1 , γ2 , . . .) on for the right-μˆ random walk, the corresponding sample paths for the random walks on [0, 1] satisfy Xnx1 < Xnx2 . Therefore, the probability that Xnx2 converges to 1 is at least as great as the probability that Xnx1 converges to 1. This shows that f (x) is increasing. We can now show how the assumption of a non--invariant harmonic function on [0, 1] leads to a contradiction. Let f be as in Lemma 5.2 and deﬁne a probability measure ν on [0, 1] by extending the deﬁnition ν((a, b]) = f (b) − f (a)

68 / r. feres and e. ronshausen

to the Lebesgue measurable subsets of the interval. At this point we make the further assumption that the measure μ on is symmetric. Then, μ ∗ ν((a, b]) = μ(γ )ν (γ −1 a, γ −1 b] γ ∈

=

μ(γ ) f (γ −1 b) − f (γ −1 a)

γ ∈

=

μ(γ )f (γ b) −

γ ∈

μ(γ )f (γ a)

γ ∈

= f (b) − f (a) = ν((a, b]). This remark, which was shown to us by B. Deroin, simpliﬁes a similar but somewhat more involved argument of an earlier version of this paper. The same argument is used in the proof of proposition 5.7 of [DKN]. From this we obtain the following lemma.

lemma 5.3. If μ is symmetric, ν is a harmonic probability measure. We can now conclude the proof of Theorem 1.1. Since by Proposition 2.4 any function in H(X , , μ) must be -invariant on the support of a harmonic probability measure, and since the above f is not constant on any interior orbit of the -action on [0, 1], we arrive at a contradiction. Therefore, a non-invariant, continuous harmonic function cannot exist.

6. Examples We now give a class of examples of -spaces to illustrate the way in which the Liouville property can fail to hold. We begin with a general remark that suggests a recipe for constructing examples. Let μ be a probability measure on the countable (discrete) group and let μ, ˆ as before, be the image of μ under group inverse. The unit ball in H ∞ (, μ) ˆ (the latter equipped with the supremum norm) is a compact space, which we denote by X0 . Note that acts on X0 by homeomorphisms under the deﬁnition (γ , φ) → γ · φ, where (γ · φ)(η) = φ(γ −1 η). The -space X0 has a tautological continuous μ-harmonic function: f (φ) = φ(e). This is, in fact, μ-harmonic since

harmonic functions over group actions / 69

γ ∈

f (γ · φ)μ(γ ) =

γ ∈

φ(γ −1 )μ(γ ) =

φ(γ )μ(γ ˆ ) = φ(e) = f (φ).

γ ∈

Furthermore, since f (γ φ) = φ(γ −1 ), f is not constant on · φ if φ itself is not a constant function. This simple remark suggests the following approach to ﬁnding examples of non-Liouville actions on a -space S. Suppose we can somehow construct a continuous, -equivariant map : S → X0 . We express equivariance by (γ s) = γ (s) and write (s) = φs . If does not map S entirely into the space of constant functions, then f ◦ deﬁnes a continuous, μ-harmonic function on S that is not -invariant. Thus, what we have shown in Theorem 1.1 amounts to the following:

proposition 6.1. If μ is a symmetric probability measure on , then every -equivariant continuous : S1 → X0 maps into the constant functions. We now construct a continuous -equivariant map : S2 → X0 whose image is not contained in the space of constant functions. By the above remark, this yields a non-Liouville action on S2 . The example generalizes one given in [FZ2]. Let and μˆ be such that the Poisson boundary of (, μ) ˆ can be identiﬁed 1 with the circle S . For example, as already noted, if is a uniform group of isometries of the Poincaré disc, it is possible to ﬁnd a symmetric μ for which the Poisson boundary of the group coincides with the geometric boundary of the disc [Fur]. Let ν be a harmonic measure on S1 as in the discussion immediately after Proposition 3.1. The following general fact was pointed out to us by Kaimanovich.

lemma 6.2. Let μˆ be a nondegenerate probability measure on such that a Poisson boundary (B, ν) of (, μ) ˆ is nontrivial. Then the μ-harmonic ˆ measure ν has no atoms. Proof. Suppose for a contradiction that ν does have atoms, and let b ∈ B be such that ν(b) ≥ ν(b ) for all b ∈ B. Since ν is μ-harmonic ˆ and μˆ is nondegenerate, the maximum principle applied to γ → (γ∗ ν)(b) implies ν(b) = ν(γ b) for all γ ∈ . On the other hand, if (γ1 , γ2 , . . .) is a random walk on (, μ), ˆ then (γ1 . . . γn )∗ ν converges to the Dirac measure δg∞ , where g∞ is the random point on the boundary to which the random walk converges. Writing gn = γ1 . . . γn , then 0 < ν(b) = (gn )∗ ν(b) → δg∞ (b). This is possible only if sample paths converge to b almost surely, contradicting the assumption that the Poisson boundary is nontrivial.

70 / r. feres and e. ronshausen

We now construct a nontrivial -equivariant continuous map : S2 → X0 . Let x = (z, θ) in the cylinder S1 × [0, 2π ] represent the interval Ix ⊂ S1 with endpoints z and zeiθ . Naturally, γ Ix = Iγ x . The circle boundaries S1 × {0} and S1 × {2π } are invariant under and correspond to intervals of length 0 or 2π . The quotient (S1 × [0, 2π ])/ ∼ obtained by collapsing the boundary circles to points is homeomorphic to S2 and the -action on the cylinder deﬁnes an action on the quotient by homeomorphisms ﬁxing two points, denoted N = (S1 × {2π})/ ∼ and S = (S1 × {0})/ ∼. This deﬁnes S2 as a topological -space. Let ν be a harmonic probability measure on S1 , which is granted since S1 is a Poisson boundary of (, μ). ˆ Deﬁne (x) = φx , where φx (γ ) = γ∗ ν(Ix ). Note that is well deﬁned since ν(Ix ) is 0 on all intervals of length 0, and 1 on all intervals of length 2π. A simple calculation shows that φx is μ-harmonic ˆ for 2 each x ∈ S and that is -equivariant. Furthermore, is continuous due to the fact that ν has no atoms. Since the Poisson boundary is nontrivial, does not map into the set of constant functions. This shows that (S2 , , μ) is non-Liouville. In this example, the only -orbits in S2 on which (x) is a constant function are the ﬁxed points N, S, so C = {N, S}, where C is the set deﬁned in Proposition 3.1. The same argument used in Lemma 5.2 shows that g = f ◦ can be composed with an afﬁne transformation of R to ensure g(S) = 0 and g(N) = 1, after which g(x) is the probability that random walk on (S2 , , μ) starting at x ∈ S2 converges to N. When (S1 , ν) coincides with the boundary of the hyperbolic disc, where is a uniform lattice in PSL(2, R), lemma 4.2 of [FZ2] can be used to show that the just-constructed action on S2 is ergodic relative to the smooth measure class on S2 . This dynamical property of the action contrasts with the simple behavior of the random walk, which converges to N or S with probability 1. It is interesting to regard the induced random walk on the space X0 itself, for a general (, μ). ˆ Let I = [−1, 1] ⊂ X0 represent the subspace of constant functions in X0 . Then the random walk on (X0 , , μ) converges toward I, that is, the random walk on X0 \ I is transient. On the other hand, the -action on X0 can be dynamically very complicated. For example, it is shown in [FZ2] for subgroups of PSL(2, R) that the Z-action on X0 induced by a parabolic or hyperbolic element is a chaotic dynamical system.

harmonic functions over group actions / 71

References [DK]

B. Deroin and V. Kleptsyn. Random conformal dynamical systems, GAFA 17 (2007), 1043–1105. [DKN] B. Deroin, V. Klepstyn, A. Navas. Sur la dynamique unidimensionnelle en régularité intermédiaire, Acta Mathematica–Stockholm 199 (2007), no. 2, 199–262. [FFP] S. Fenley, R. Feres, K. Parwani. Harmonic functions on R-covered foliations, Ergodic Theory Dynam. Systems 29 (2009), no. 4, 1141–1161. [FZ1] R. Feres and A. Zhegib. Leafwise holomorphic functions, Proc. Amer. Math. Soc. 131 (2003), no. 6, 1717–1725. [FZ2] R. Feres and A. Zhegib. Dynamics on the space of harmonic functions and the foliated Liouville problem, Ergodic Theory Dynam. Systems 25 (2005), no. 2, 503–516. [Fu] A. Furman. Random walks on groups and random transformations, Handbook of dynamical systems, Vol. 1A, chap. 12. Hasselblatt and Katok, North-Holland, 2002, 931–1014. [Fur] H. Furstenberg. Random walks and discrete subgroups of Lie groups, 1971 Advances in Probability and Related Topics, Vol. 1. New York, Dekker, 1971, 1–63. [Gar] L. Garnett. Foliations, the ergodic theorem and Brownian motion, J. Funct. Anal. 51 (1983), 285–311. [Pa] F. Paulin. Analyse harmonique des relations d’équivalence mesurées discrètes, Markov Proc. Rel. Fields. 5 (1999), 163–200.

4

david fisher

GROUPS ACTING ON MANIFOLDS: AROUND THE ZIMMER PROGRAM to robert zimmer on the occasion of his

60 th

birthday

Abstract This paper is a survey on the Zimmer program. In its broadest form, this program seeks an understanding of actions of large groups on compact manifolds. The goals of this survey are (1) to put in context the original questions and conjectures of Zimmer and Gromov that motivated the program, (2) to indicate the current state of the art on as many of these conjectures and questions as possible, and (3) to indicate a wide variety of open problems and directions of research.

Contents

1 2

3

4

Prologue Some Remarks on Biases and Omissions A Brief Digression: Some Examples of Groups and Actions 2.1 Semisimple Groups and Their Lattices 2.2 Some Actions of Groups and Lattices 2.3 Induced Actions Pre-history 3.1 Local and Global Rigidity of Homomorphisms into Finite Dimensional Groups 3.2 Strong and Super-rigidity 3.3 Harmonic Map Approaches to Rigidity 3.4 A Remark on Cocycle Super-rigidity 3.5 Margulis’s Normal Subgroup Theorem Low-Dimensional Actions: Conjectures and Results 4.1 Dimension 1 4.2 Dimension 2

74 78 79 80 83 85 86 86 88 90 90 90 91 92 93

Author partially supported by NSF grants DMS-0541917 and 0643546, a fellowship from the Radcliffe Institute for Advanced Studies and visiting positions at École Polytechnique, Palaiseau and Université Paris-Nord, Villateneuse.

72

groups acting on manifolds / 73

4.3 4.4 4.5 4.6

5

6

7

8

9 10

Dimension 3 Analytic Actions in Low Dimensions Zimmer’s Full Conjecture, Partial Results Some Approaches to the Conjectures 4.6.1 Discrete Spectrum of Actions 4.6.2 Effective Invariant Metrics 4.7 Some Related Questions and Conjectures on Countable Subgroups of Diff (M) 4.7.1 Groups with Property (T ) and Generic Groups 4.7.2 Universal Lattices 4.7.3 Irreducible Actions of Products 4.7.4 Torsion Groups and Homeo (M) Cocycle Super-rigidity and Immediate Consequences 5.1 Zimmer’s Cocycle Super-rigidity Theorem and Generalizations 5.2 First Applications to Group Actions and the Problem of Regularity Geometric Actions and Rigid Geometric Structures 6.1 Rigid Structures and Structures of Finite Type 6.2 Rigid Structures and Representations of Fundamental Groups 6.3 Afﬁne Actions Close to the Critical Dimension 6.4 Zimmer’s Borel Density Theorem and Generalizations 6.5 Actions Preserving a Complex Structure 6.6 Conjectures and Questions Topological Super-rigidity: Regularity Results from Hyperbolic Dynamics and Applications 7.1 Stability in Hyperbolic Dynamics. 7.2 Uniformly Hyperbolic Actions on Tori 7.3 Rigidity Results for Uniformly Hyperbolic Actions 7.4 Conjectures and Questions on Uniformly Hyperbolic Actions Representations of Fundamental Groups and Arithmetic Quotients 8.1 Linear Images of Fundamental Groups 8.2 Arithmetic Quotients 8.3 Open Questions Exotic Volume Preserving Actions Non-Volume Preserving Actions

96 96 98 100 100 101 103 103 104 104 105 106 106 107 108 109 112 113 114 116 116 118 118 121 123 124 125 126 128 128 129 133

74 / d. fisher

10.1 Stuck’s Examples 10.2 Weinberger’s Examples 10.3 Work of Nevo-Zimmer 11 Groups That Do Act on Manifolds 11.1 Thompson’s Groups 11.2 Highly Distorted Subgroups of Diff (S2 ) 11.3 Topological Construction of Actions on High Dimensional Spheres 12 Rigidity for Other Classes of Acting Groups 12.1 Lattices in Semi-direct Products 12.2 Universal Lattices 12.3 SO(1,n) and SU(1,n) and Their Lattices 12.4 Lattices in Other Locally Compact Groups 12.5 Automorphism Groups of Free and Surface Groups 13 Properties of Subgroups of Diff (M) 13.1 Jordan’s Theorem 13.2 Burnside Problem 13.3 Tits’s Alternative References

133 133 134 136 136 138 138 139 139 140 140 141 141 142 142 143 143 144

1. Prologue Traditionally, the study of dynamical systems is concerned with actions of R or Z on manifolds, that is, with ﬂows and diffeomorphisms. It is natural to consider instead dynamical systems deﬁned by actions of larger discrete or continuous groups. For non-discrete groups, the Hilbert-Smith conjecture is relevant, since the conjecture states that non-discrete, locally compact totally disconnected groups do not act continuously by homeomorphisms on manifolds. For smooth actions known results sufﬁce to rule out actions of totally disconnected groups; the Hilbert-Smith conjecture has been proven for Hölder diffeomorphisms [155, 207]. For inﬁnite discrete groups, the whole universe is open. One might consider the “generalized Zimmer program” to be the study of homomorphisms ρ : → Diff (M) where is a ﬁnitely generated group and M is a compact manifold. In this survey much emphasis will be on a program proposed by Zimmer. Here one considers a very special class of both Lie groups and discrete groups, namely semisimple Lie groups of higher real rank and their lattices.

groups acting on manifolds / 75

Zimmer’s program is motivated by several of Zimmer’s own theorems and observations, many of which we will discuss below. But in broadest strokes, the motivation is simpler. Given a group whose linear and unitary representations are very rigid or constrained, might it also be true that the group’s representations into Diff ∞ (M) are also very rigid or constrained at least when M is a compact manifold? For the higher rank lattices considered by Zimmer, Margulis’s superrigidity theorems classiﬁed ﬁnite dimensional representations and the groups enjoy property (T ) of Kazhdan, which makes unitary representations quite rigid. This motivation also stems from an analogy between semisimple Lie groups and diffeomorphism groups. When M is a compact manifold, not only is Diff ∞ (M) an inﬁnite dimensional Lie group, but its connected component is simple. Simplicity of the connected component of Diff ∞ (M) was proven by Thurston using results of Epstein and Herman [227, 50, 116]. Herman had used Epstein’s work to see that the connected component of Diff ∞ (Tn ) is simple and Thurston’s proof of the general case uses this. See also further work on the topic by Banyaga and Mather [8, 167, 166, 168, 169], as well as Banyaga’s book [9]. A major motivation for Zimmer’s program came from his cocycle superrigidity theorem, discussed in detail below in Section 5. One can think of a homomorphism ρ : → Diff (M) as deﬁning a virtual homomorphism from to GL( dim (M), R). The notion of virtual homomorphisms, now more commonly referred to in our context as cocycles over group actions, was introduced by Mackey [154]. In all of Zimmer’s early papers, it was always assumed that we had a homomorphism ρ : → Diff ∞ (M, ω) where ω is a volume form on M. A major motivation for this is that the cocycle superrigidity theorem referred to in the last paragraph only applies to cocycles over measure preserving actions. That this hypothesis is necessary was conﬁrmed by examples of Stuck, who showed that no rigidity could be hoped for unless one had an invariant volume form or assumed that M was very low dimensional [226]. Recent work of Nevo and Zimmer does explore the non-volume preserving case and is discussed below, along with Stuck’s examples and some others due to Weinberger in Section 10. Let G be a semisimple real Lie group, all of whose simple factors are of real rank at least 2. Let in G be a lattice. The simplest question asked by Zimmer was: Can one classify smooth volume preserving actions of on compact manifolds? At the time of this writing, the answer to this question is still unclear. There certainly are a wider collection of actions than Zimmer may have initially suspected and it is also clear that the moduli space of such

76 / d. fisher

actions is not discrete; see [12, 78, 128] or Section 9 below. But there are still few enough examples known that a classiﬁcation remains plausible. And if one assumes some additional conditions on the actions, then plausible conjectures and striking results abound. We now discuss four paradigmatic conjectures. To make this introduction accessible, these conjectures are all special cases of more general conjectures stated later in the text. In particular, we state all the conjectures for lattices in SL(n, R) when n > 2. The reader not familiar with higher rank lattices can consider the examples of ﬁnite index subgroups of SL(n, Z), with n > 2, rather than for general lattices in SL(n, R). We warn the reader that the algebraic structure of SL(n, Z) and its ﬁnite index subgroups makes many results easier for these groups than for other lattices in SL(n, R). The conjectures we state concern, respectively, (1) classiﬁcation of low dimensional actions, (2) classiﬁcation of geometric actions, (3) classiﬁcation of uniformly hyperbolic actions, and (4) the topology of manifolds admitting volume preserving actions. A motivating conjecture for much recent research is the following:

conjecture 1.1. (zimmer’s conjecture) For any n > 2, any homomorphism ρ : SL(n, Z) → Diff (M) has ﬁnite image if dim (M) < n − 1. The same for any lattice in SL(nR). The dimension bound in the conjecture is clearly sharp, as SL(n, R) and all of its subgroups act on Pn−1 . The conjecture is a special case of a conjecture of Zimmer’s that concerns actions of higher-rank lattices on low dimensional manifolds that we state as Conjecture 4.12 below. Recently, much attention has focused on these conjectures concerning low dimensional actions. In our second, geometric, setting (a special case of ) a major motivating conjecture is the following:

conjecture 1.2. (affine actions) Let < SL(n, R) be a lattice. Then there is a classiﬁcation of actions of on compact manifolds that preserve both a volume form and an afﬁne connection. All such actions are algebraically deﬁned in a sense to be made precise below. (See Deﬁnition 2.4 below.) The easiest example of an algebraically deﬁned action is the action of SL(n, Z) (or any of its subgroups) on Tn . We formulate our deﬁnition of algebraically deﬁned action to include all trivial actions, all isometric actions, and all (skew) products of other actions with these. The conjecture is a special case

groups acting on manifolds / 77

of a conjecture stated below as Conjecture 6.15. In addition to considering more general acting groups, we will consider more general invariant geometric structures, not just afﬁne connections. A remark worth making is that, for the geometric structures we consider, the automorphism group is always a ﬁnite dimensional Lie group. The question of when the full automorphism group of a geometric structure is large is well studied from other points of view; see particularly [135]. However, this question is generally most approachable when the large subgroup is connected and much less is known about discrete subgroups. In particular, geometric approaches to this problem tend to use information about the connected component of the automorphism group and give much less information about the group of components particularly if the connected component is trivial. One is also interested in the possibility of classifying actions under strong dynamical hypotheses. The following conjecture is motivated by the work of Feres-Labourie [61] and Goetze-Spatzier [105, 106] and is similar to conjectures stated in [108, 120]:

conjecture 1.3. Let < SL(n, R) be a lattice. Then there is a classiﬁcation of actions of on a compact manifold M that preserve both a volume form and where one element γ ∈ acts as an Anosov diffeomorphism. All such actions are algebraically deﬁned in a sense to be made precise below. (Again see Deﬁnition 2.4.) In the setting of this particular conjecture, a proof of an older conjecture of Franks concerning Anosov diffeomorphisms would imply that any manifold M as in the conjecture was homeomorphic to an infranilmanifold on which γ is conjugate by a homeomorphism to a standard afﬁne Anosov map. Even assuming Franks’s conjecture, Conjecture 1.3 is open. One can make a more general conjecture by only assuming that γ has some uniformly partially hyperbolic behavior. Various versions of this are discussed in §7; see particularly Conjecture 7.9. We end this introduction by stating a topological conjecture about all manifold admitting smooth volume preserving actions of a simple higher rank algebraic group. Very special cases of this conjecture are known and the conjecture is plausible in light of existing examples. More precise variants and a version for lattice actions will be stated below in §8.

conjecture 1.4. Let G be a simple Lie group of real rank at least 2. Assume G has a faithful action preserving volume on a compact manifold M. Then π1 (M)

78 / d. fisher

has a ﬁnite index subgroup such that surjects onto an arithmetic lattice in a Lie group H where H locally contains G. The conjecture says, more or less, that admitting a G action forces the fundamental group of M to be large. Passage to a ﬁnite index subgroup and a quotient is necessary; see Section 9 and [84] for more discussion. The conjecture might be considered the analogue for group actions of Margulis’s arithmeticity theorem. This survey is organized on the following lines. We begin in §2 by describing in some detail the kinds of groups we consider and examples of their actions on compact manifolds. In §3 we digress with a prehistory of motivating results from rigidity theory. Then in §4, we discuss conjectures and theorems concerning actions on “low dimensional” manifolds. In this section, there are a number of related results and conjectures concerning groups not covered by Zimmer’s original conjectures. Here low-dimensional is in two senses (1) compared to the group as in Conjecture 1.1, and (2) absolutely small, as in dimension being between 1 and 4. This discussion is simpliﬁed by the fact that many theorems conclude with the nonexistence of actions or at least with all actions factoring through ﬁnite quotients. In Section 5 we further describe Zimmer’s motivations for his conjectures by discussing the cocycle superrigidity theorem and some of its consequences for smooth group actions. In Sections 6 and 7, we discuss, respectively, geometric and dynamical conditions under which a simple classiﬁcation might be possible. In Section 8, we discuss another approach to classifying G and actions using topology and representations of fundamental groups in order to produce algebraically deﬁned quotients of actions. Then in Section 9, we describe the known “exotic examples” of the acting groups to consider. This construction reveals some necessary complexity of a high-dimensional classiﬁcation. We then describe known results and examples of actions not preserving volume in Section 10 and some rather surprising group actions on manifolds in Section 11. Finally we end the survey with a collection of questions about the algebraic and geometric structure of ﬁnitely generated subgroups of Diff (M). Some remarks on biases and omissions. Like any survey of this kind, this work is informed by its author’s biases and experiences. There is an additional bias in that this paper emphasizes developments that are close to Zimmer’s own work and conjectures. In particular, the study of rigidity of group actions often focuses on the low dimensional setting where all group actions are conjectured

groups acting on manifolds / 79

to be ﬁnite or trivial. While Zimmer did substantial work in this setting, he also proved many results and made many conjectures in more general settings where any potential classiﬁcation of group actions is necessarily, due to the existence of examples, more complicated. Another omission is that almost nothing will be said here about local rigidity of groups actions, since the author has recently written another survey on that topic [77]. While that survey could already use updating, that update will appear elsewhere. For other surveys of Zimmer’s program and rigidity of large group actions the reader is referred to [68, 144, 258]. The forthcoming book by Zimmer and Witte Morris [258] is a particularly useful introduction to ideas and techniques in Zimmer’s own work. Also of interest are (1) a brief survey of rigidity theory by Spatzier with a more geometric focus [223], (2) an older survey also by Spatzier, with a somewhat broader scope [222], (3) a recent problem list by Margulis on rigidity theory with a focus on measure rigidity [164], and (4) a more recent survey by Lindenstrauss focused on recent developments in measure rigidity [147]. Both of the last two mentioned surveys are particularly oriented toward connections between rigidity and number theory, which are not mentioned at all in this survey. Finally, while all mistakes and errors in this survey are the sole responsibility of its author, I would like to thank many people whose comments led to improvements. These include Danny Calegari, Renato Feres, Etienne Ghys, Daniel Groves, Steve Hurder, Jean-Francois Lafont, Karin Melnick, Nicolas Monod, Andrés Navas, Leonid Polterovich, Pierre Py, Yehuda Shalom, Ralf Spatzier, Dave Witte Morris, and an anonymous referee.

2. A Brief Digression: Some Examples of Groups and Actions In this section we brieﬂy describe some of the groups that will play important roles in the results discussed here. The reader already familiar with semisimple Lie groups and their lattices may want to skip to the Subsection 2.2 where we give descriptions of group actions. The following convention is in force throughout this paper. For deﬁnitions of relevant terms the reader is referred to the following subsections. Convention In this article we will have occasion to refer to three overlapping classes of lattices in Lie groups that are slightly different. Let G be a semisimple Lie group and < G a lattice. We call a higher rank lattice if all simple factors of G have real rank at least 2. We call a lattice with (T ) if all

80 / d. fisher

simple factors of G have property (T ). Lastly, we call an irreducible higher rank lattice if G has real rank at least 2 and is irreducible. 2.1. Semisimple Groups and Their Lattices

By a simple Lie group, we mean a connected Lie group all of whose normal subgroups are discrete, though we make the additional convention that R and S1 are not simple. By a semisimple Lie group we mean the quotient of a product of simple Lie groups by some subgroup of the product of their centers. Note that with our conventions, the center of a simple Lie group is discrete and is in fact the maximal normal subgroup. There is an elaborate structure theory of semisimple Lie groups and the groups are completely classiﬁed; see [115] or [134] for details. Here we merely describe some examples, all of which are matrix groups. All connected semisimple Lie groups are discrete central extensions of matrix groups, so the reader will lose very little by always thinking of matrix groups. 1) The groups SL(n, R), SL(n, C), and SL(n, H) of n by n matrices of determinant 1 over the real numbers, the complex numbers, or the quaternions. 2) The group SP(2n, R) of 2n by 2n matrices of determinant 1 that preserve a real symplectic form on R2n . 3) The groups SO(p, q), SU(p, q), and SP(p, q) of matrices that preserve inner products of signature (p, q) where the inner product is real linear on Rp+q , Hermitian on Cp+q , or quaternionic Hermitian on Hp+q , respectively. Let G be a semisimple Lie group that is a subgroup of GL(n, R). We say that G has real rank k if G has a k-dimensional abelian subgroup that is conjugate to a subgroup of the real diagonal matrices and no k + 1-dimensional abelian subgroups with the same property. The groups in (1) have rank n − 1, the groups in (2) have rank n, and the groups in (3) have rank min (p, q). Since this article focuses primarily on ﬁnitely generated groups, we are more interested in discrete subgroups of Lie groups than in the Lie groups themselves. A discrete subgroup in a Lie group G is called a lattice if G/ has ﬁnite Haar measure. The lattice is called cocompact or uniform if G/ is compact and non-uniform or not cocompact otherwise. If G = G1 ×· · ·×Gn is a product, then we say a lattice < G is irreducible if its projection to each Gi is dense. It is more typical in the literature to insist that projections to all factors are dense, but this deﬁnition is more practical for our purposes. More generally, we make the same deﬁnition for an almost direct product, by which we mean a direct product G modulo some subgroup of the center Z(G). Lattices

groups acting on manifolds / 81

in semisimple Lie groups can always be constructed by arithmetic methods; see [19] and also [175] for more discussion. In fact, one of the most important results in the theory of semisimple Lie groups is that if G is a semisimple Lie group without compact factors, then all irreducible lattices in G are arithmetic unless G is locally isomorphic to SO(1, n) or SU(1, n). For G of real rank at least 2, this is Margulis’s arithmeticity theorem, which he deduced from his superrigidity theorems [159, 202, 162]. For non-uniform lattices, Margulis had an earlier proof that does not use the superrigidity theorems; see [158, 160]. This earlier proof depends on the study of dynamics of unipotent elements on the space G/ , and particularly on what is now known as the “non-divergence of unipotent ﬂows.” Special cases of the super-rigidity theorems were then proven for Sp(1, n) and F4−20 by Corlette [40] and Gromov-Schoen [113], which sufﬁced to imply the statement on arithmeticity given above. As we will be almost exclusively concerned with arithmetic lattices, we do not give examples of non-arithmetic lattices here, but refer the reader to [162] and [175] for more discussion. A formal deﬁnition of arithmeticity, at least when G is algebraic, is

definition 2.1. Let G be a semisimple algebraic Lie group and < G a lattice. Then is arithmetic if there exists a semisimple algebraic Lie group H such that 1) there is a homomorphism π : H 0 → G with compact kernel, 2) there is a rational structure on H such that the projection of the integer points of H to G are commensurable to , that is, π (H(Z))∩ is of ﬁnite index in both H(Z) and . We now give some examples of arithmetic lattices. The simplest is to take the integer points in a simple (or semisimple) group G that is a matrix group, for example, SL(n, Z) or Sp(n, Z). This exact construction always yields lattices, but also always yields non-uniform lattices. In fact, the lattices one can construct in this way have very special properties because they will contain many unipotent matrices. If a lattice is cocompact, it will necessarily contain no unipotent matrices. The standard trick for understanding the structure of lattices in G that become integral points after passing to a compact extension is called change of base. For much more discussion see [162, 175, 245]. We give one example to illustrate the process. Let G = SO(m, n), which we view as the set of matrices in SL(n + m, R) which preserve the inner product ⎞ ⎛ n+m m √ vi w i + ⎝ vi w i ⎠ v, w = − 2 i=1

i=m+1

82 / d. fisher

where vi and wi are the ith components of v and w. This form, and therefore √ G, are deﬁned over the ﬁeld Q( 2) that has a Galois conjugation σ deﬁned √ √ √ by σ ( 2) = − 2. If we look at the points = G(Z[ 2]), we can deﬁne an embedding of in SO(m, n) × SO(m + n) by taking γ to (γ , σ (γ )). It is straightforward to check that this embedding is discrete. In fact, this embeds in H = SO(m, n) × SO(m + n) as integral points for the rational structure on H √ where the rational points are exactly the points (M, σ (M)) where M∈G(Q( 2)). This makes a lattice in H and it is easy to see that projects to a lattice in G, since G is cocompact in H. What is somewhat harder to verify is that is cocompact in H, for which we refer the reader to the list of references above. Similar constructions are possible with SU(m, n) or SP(m, n) in place of SO(m, n) and also with more simple factors and ﬁelds with more Galois automorphisms. There are also a number of other constructions of arithmetic lattices using division algebras. See [194, 175] for a comprehensive treatment. We end this section by deﬁning a key property of many semisimple groups and their lattices. This is property (T ) of Kazhdan, and was introduced by Kazhdan in [131] in order to prove that non-uniform lattices in higher rank semisimple Lie groups are ﬁnitely generated and have ﬁnite abelianization. It has played a fundamental role in many subsequent developments. We do not give Kazhdan’s original deﬁnition, but one which was shown to be equivalent by the work of Delorme [44] and Guichardet [114].

definition 2.2. A locally compact group has property (T ) of Kazhdan if H 1 (, π ) = 0 for every continuous unitary representation π of on a Hilbert space. This is equivalent to saying that any continuous isometric action of on a Hilbert space has a ﬁxed point. remarks 2.3. 1) Kazhdan’s deﬁnition is that the trivial representation is isolated in the Fell topology on the unitary dual of . 2) If a continuous group G has property (T ), so does any lattice in G. This result was proved in [131]. 3) Any semisimple Lie group has property (T ) if and only if it has no simple factors locally isomorphic to SO(1, n) or SU(1, n). For a discussion of this fact and attributions, see [43]. For groups with all simple factors of real rank at least 3, this is proven in [131].

groups acting on manifolds / 83

4) No noncompact amenable group, and in particular no noncompact abelian group, has property (T ). An easy averaging argument shows that all compact groups have property (T ). Groups with property (T ) play an important role in many areas of mathematics and computer science. 2.2. Some Actions of Groups and Lattices

Here we deﬁne and give examples of a general class of actions. A major impetus in Zimmer’s work is determining optimal conditions for actions to lie in this class. The class we describe is slightly more general than the class Zimmer termed “standard actions” in, for example, [249]. Let H be a Lie group and L < H a closed subgroup. Then a diffeomorphism f of H/L is called afﬁne if there is a diffeomorphism f˜ of H such that f ([h]) = f˜ (h) where f˜ = A◦τh with A an automorphism of H with A(L) = L and τh is left translation by some h in H. Two obvious classes of afﬁne diffeomorphisms are left translations on any homogeneous space and linear automorphisms of tori, or more generally, automorphisms of nilmanifolds. A group action is called afﬁne if every element of the group acts by an afﬁne diffeomorphism. It is easy to check that the full group of afﬁne diffeomorphisms Aff (H/L) is a ﬁnite dimensional Lie group and an afﬁne action of a group D is a homomorphism π : D → Aff (H/L). The structure of Aff (H/L) is surprisingly complicated in general; it is a quotient of a subgroup of the group Aut (H) H where Aut (H) is a group of automorphisms of H. For a more detailed discussion of this relationship, see [81, section 6]. While it is not always the case that any afﬁne action of a group D on H/L can be described by a homomorphism π : D → Aut (H) H, this is true for two important special cases: 1) D is a connected semisimple Lie group and L is a cocompact lattice in H, 2) D is a lattice in a semisimple Lie group G where G has no compact factors and no simple factors locally isomorphic to SO(1, n) or SU(1, n), and L is a cocompact lattice in H. These facts are [81, theorems 6.4 and 6.5] where afﬁne actions as in (1) and (2) above are classiﬁed. The most obvious examples of afﬁne actions of large groups are of the following forms, which are frequently referred to as standard actions: 1) Actions of groups by automorphisms of nilmanifolds. That is, let N be a simply connected nilpotent group, < N a lattice (which is necessarily

84 / d. fisher

cocompact), and assume a ﬁnitely generated group acts by automorphisms of N preserving . The most obvious examples of this are when N = Rn , = Zn , and < SL(n, Z), in which case we have a linear action of on Tn . 2) Actions by left translations. That is, let H be a Lie group and < H a cocompact lattice and < H some subgroup. Then acts on H/ by left translations. Note that in this case need not be discrete. 3) Actions by isometries. Here K is a compact group that acts by isometries on some compact manifold M and < K is a subgroup. Note that here is either discrete or a discrete extension of a compact group. We now brieﬂy deﬁne a few more general classes of actions, which we need to formulate most of the conjectures in this paper. We ﬁrst ﬁx some notations. Let A and D be topological groups, and B < A a closed subgroup. Let ρ : D × A/B → A/B be a continuous afﬁne action.

definition 2.4. 1) Let A, B, D, and ρ be as above. Let C be a compact group of afﬁne diffeomorphisms of A/B that commute with the D action. We call the action of D on C\A/B a generalized afﬁne action. 2) Let A, B, D, and ρ be as in (1) above. Let M be a compact Riemannian manifold and ι : D × A/B → Isom (M) a C 1 cocycle. We call the resulting skew product D action on A/B × M a quasi-afﬁne action. If C and D are as in (2), and we have a smooth cocycle α : D × C\ A/B → Isom (M), then we call the resulting skew product action of D on C\A/B × M a generalized quasi-afﬁne action. Many of the conjectures stated in this paper will end with the conclusion that all actions satisfying certain hypotheses are generalized quasi-afﬁne actions. It is not entirely clear that generalized quasi-afﬁne actions of higher rank groups and lattices are much more general than generalized afﬁne actions. The following discussion is somewhat technical and might be skipped on ﬁrst reading. One can always take a product of a generalized afﬁne action with a trivial action on any manifold to obtain a generalized quasi-afﬁne action. One can also do variants on the following. Let H be a semisimple Lie group and < H a cocompact lattice. Let π : → K be any homomorphism of into a compact Lie group. Let ρ be a generalized afﬁne action of G on C\H/ and let M be

groups acting on manifolds / 85

a compact manifold on which K acts. Then there is a generalized quasi-afﬁne action of G on (C\H × M)/.

question 2.5. Is every generalized quasi-afﬁne action of a higher rank simple Lie group of the type just described? The question amounts to asking for an understanding of compact group valued cocycles over quasi-afﬁne actions of higher rank simple Lie groups. I leave it to the interested reader to formulate the analogous question for lattice actions. For some work in this direction, see the paper of Witte Morris and Zimmer [176]. 2.3. Induced Actions

I end this section by describing brieﬂy the standard construction of an induced or suspended action. This notion can be seen as a generalization of the construction of a ﬂow under a function or as an analogue of the more algebraic notion of inducing a representation. Given a group H, a (usually closed) subgroup L, and an action ρ of L on a space X , we can form the space (H × X )/L where L acts on H × X by h · (l, x) = (lh −1 , ρ(h)x). This space now has a natural H action by left multiplication on the ﬁrst coordinate. Many properties of the L action on X can be studied more easily in terms of properties of the H action on (H × X )/L. This construction is particularly useful when L is a lattice in H. This notion suggests the following principle:

principle 2.6. Let be a cocompact lattice in a Lie group G. To classify actions on compact manifolds it sufﬁces to classify G actions on compact manifolds. The principle is a bit subtle to implement in practice, since we clearly need a sufﬁciently detailed classiﬁcation of G actions to be able to tell which ones arise as induction of actions. While it is a bit more technical to state and probably more difﬁcult to use, there is an analogous prinicple for noncocompact lattices. Here one needs to classify G actions on manifolds that are not compact but where the G action preserves a ﬁnite volume. In fact, one needs only to study such actions on manifolds that are ﬁber bundles over G/ with compact ﬁbers. The lemma begs the question as to whether or not one should simply always study G actions. While in many settings this is useful, it is not always. In particular, many known results about actions require hypotheses on the

86 / d. fisher

action where there is no useful way of rephrasing the property as a property of the induced action. Or, perhaps more awkwardly, require assumptions on the induced action that cannot be rephrased in terms of hypotheses on the original action. We will illustrate the difﬁculties in employing Principle 2.6 at several points in this paper. A case where the implications of the principle are particularly clear is a negative result concerning actions of SO(1, n). We will make clear by the proof what we mean by

theorem 2.7. Let G = SO(1, n). Then one cannot classify actions of G on compact manifolds. Proof. For every n there is at least one lattice < G that admits homomorphisms onto non-abelian free groups; see, for example, [150]. And therefore also onto Z. Thus we can take any action of Fn or Z and induce to a G action. It is relatively easy to show that if the induced actions are isomorphic, then so are the actions they are induced from; see, for example, [78]. A classiﬁcation of Z actions would amount to a classiﬁcation of diffeomorphisms and a classiﬁcation of Fn actions would involve classifying all n-tuples of diffeomorphisms. As there is no reasonable classiﬁcation of diffeomorphisms, there is also no reasonable classiﬁcation of n-tuples of diffeomorphisms. We remark that essentially the same theorem holds for actions of SU(1, n) where homomorphisms to Z exist for certain lattices [132]. Much less is known about free quotients of lattices in SU(1, n); see [148] for one example. For a surprising local rigidity result for some lattices in SU(1, n) see [72, theorem 1.3].

3. Pre-history 3.1. Local and Global Rigidity of Homomorphisms into Finite Dimensional Groups

The earliest work on rigidity theory is a series of works by Calabi-Vesentini, Selberg, Calabi, and Weil that resulted in the following:

theorem 3.1. Let G be a semisimple Lie group and assume that G is not locally isomorphic to SL(2, R). Let < G be an irreducible cocompact lattice, then the deﬁning embedding of in G is locally rigid; that is, any embedding ρ close to the deﬁning embedding is conjugate to the deﬁning embedding by a small element of G.

groups acting on manifolds / 87

remark 3.2. 1) If G = SL(2, R), the theorem is false and there is a large, well-studied space of deformation of in G, known as the Teichmueller space. 2) There is an analogue of this theorem for lattices that are not cocompact. This result was proven later and has a more complicated history that we omit here. In this case it is also necessary to exclude G locally isomorphic to SL(2, C). This theorem was originally proven in special cases by Calabi, Calabi and Vesentini, and Selberg. In particular, Selberg gives a proof for cocompact lattices in SL(n, R) for n ≥ 3 in [212], Calabi-Vesentini give a proof when the associated symmetric space X = G/K is Kähler in [29], and Calabi gives a proof for G = SO(1, n) where n ≥ 3 in [28]. Shortly afterward, Weil gave a complete proof of Theorem 3.1 in [231, 233]. In all of the original proofs, the ﬁrst step was to show that any perturbation of was discrete and therefore a cocompact lattice. This is shown in special cases in [28, 29, 212] and proven in a somewhat broader context than Theorem 3.1 in [232]. The different proofs of cases of Theorem 3.1 are also interesting in that there are two fundamentally different sets of techniques employed and this dichotomy continues to play a role in the history of rigidity. Selberg’s proof essentially combines algebraic facts with a study of the dynamics of iterates of matrices. He makes systematic use of the existence of singular directions, or Weyl chamber walls, in maximal diagonalizable subgroups of SL(n, R). Exploiting these singular directions is essential to much later work on rigidity, both of lattices in higher-rank groups and of actions of abelian groups. It seems possible to generalize Selberg’s proof to the case of G an R-split semisimple Lie group with rank at least 2. Selberg’s proof, which depended on asymptotics at inﬁnity of iterates of matrices, inspired Mostow’s explicit use of boundaries in his proof of strong rigidity [179]. Mostow’s work in turn provided inspiration for the use of boundaries in later work of Margulis, Zimmer, and others on rigidity properties of higher-rank groups. The proofs of Calabi, Calabi-Vesentini, and Weil involve studying variations of geometric structures on the associated locally symmetric space. The techniques are analytic and use a variational argument to show that all variations of the geometric structure are trivial. This work is a precursor to much work in geometric analysis studying variations of geometric structures and also informs later work on proving rigidity/vanishing of harmonic forms and

88 / d. fisher

maps. The dichotomy between approaches based on algebra/dynamics and approaches that are in the spirit of geometric analysis continues through much of the history of rigidity and the history of rigidity of group actions in particular. Shortly after completing this work, Weil discovered a new criterion for local rigidity [233]. In the context of Theorem 3.1, this allows one to avoid the step of showing that a perturbation of remains discrete. In addition, this result opened the way for understanding local rigidity of more general representations of discrete groups than the deﬁning representation.

theorem 3.3. Let be a ﬁnitely generated group, G a Lie group, and π : → G a homomorphism. Then π is locally rigid if H 1 (, g) = 0. Here g is the Lie algebra of G and acts on g by AdG ◦π . Weil’s proof of this result uses only the implicit function theorem and elementary properties of the Lie group exponential map. The same theorem is true if G is an algebraic group over any local ﬁeld of characteristic zero. In [233], Weil remarks that if < G is a cocompact lattice and G satisﬁes the hypothesis of Theorem 3.1, then the vanishing of H 1 (, g) can be deduced from the computations in [231]. The vanishing of H 1 (, g) is proven explicitly by Matsushima and Murakami in [170]. Motivated by Weil’s work and other work of Matsushima, conditions for vanishing of H 1 (, g) were then studied by many authors. See particularly [170] and [203]. The results in these papers imply local rigidity of many linear representations of lattices. 3.2. Strong and Super-rigidity

In a major and surprising development, it turns out that in many instances, local rigidity is just the tip of the iceberg and that much stronger rigidity phenomena exist. We now discuss major developments from the 1960s and 1970s. The ﬁrst remarkable result in this direction is Mostow’s rigidity theorem; see [177, 178] and references there. Given G as in Theorem 3.1, and two irreducible cocompact lattices 1 and 2 in G, Mostow proves that any isomorphism from 1 to 2 extends to an isomorphism of G with itself. Combined with the principal theorem of [232], which shows that a perturbation of a lattice is again a lattice, this gives a remarkable and different proof of Theorem 3.1, and Mostow was motivated by the desire for a “more geometric understanding” of Theorem 3.1 [178]. Mostow’s theorem is in fact a good deal stronger,

groups acting on manifolds / 89

and controls not only homomorphisms → G near the deﬁning homomorphism, but any homomorphism into any other simple Lie group G where the image is a lattice. As mentioned above, Mostow’s approach was partially inspired by Selberg’s proof of certain cases of Theorem 3.1 [179]. A key step in Mostow’s proof is the construction of a continuous map between the geometric boundaries of the symmetric spaces associated to G and G . Boundary maps continue to play a key role in many developments in rigidity theory. A new proof of Mostow rigidity, at least for Gi of real rank 1, was provided by Besson, Courtois, and Gallot. Their approach is quite different and has had many other applications concerning rigidity in geometry and dynamics; see, for example, [14, 15, 39]. The next remarkable result in this direction is Margulis’s superrigidity theorem. Margulis proved this theorem as a tool to prove arithmeticity of irreducible uniform lattices in groups of real rank at least 2. For irreducible lattices in semisimple Lie groups of real rank at least 2, the superrigidity theorems classiﬁes all ﬁnite dimensional linear representations. Margulis’s theorem holds for irreducible lattices in semisimple Lie groups of real rank at least 2. Given a lattice < G where G is simply connected, one precise statement of some of Margulis’s results is to say that any linear representation σ of almost extends to a linear representation of G. By this we mean that there is a linear representation σ˜ of G and a bounded image representation σ¯ of such that σ (γ ) = σ˜ (γ )σ¯ (γ ) for all γ in G. Margulis’s theorems also give an essentially complete description of the representations σ¯ , up to some issues concerning ﬁnite image representations. The proof here is partially inspired by Mostow’s work: a key step is the construction of a measurable “boundary map.” However, the methods for producing the boundary map in this case are very dynamical. Margulis’s original proof used Oseledec’s multiplicative ergodic theorem. Later proofs were given by both Furstenberg and Margulis using the theory of group boundaries as developed by Furstenberg from his study of random walks on groups [93, 94]. Furstenberg’s probabilistic version of boundary theory has had a profound inﬂuence on many subsequent developments in rigidity theory. For more discussion of Margulis’s superrigidity theorem, see [159, 161, 162, 245]. Margulis’s theorem, by classifying all linear representations, and not just ones with constrained images, leads one to believe that one might be able to classify all homomorphisms to other interesting classes of topological groups. Zimmer’s program is just one aspect of this theory, in other directions, many authors have studied homomorphisms to isometry groups of non-positively curved (and more general) metric spaces; see, for example, [24, 98, 172].

90 / d. fisher 3.3. Harmonic Map Approaches to Rigidity

In the 1990s, ﬁrst Corlette and then Gromov-Schoen showed that one could prove major cases of Margulis’s superrigidity theorem also for lattices in Sp(1, n) and F4−20 [40, 113]. Corlette considered the case of representations over Archimedean ﬁelds and Gromov and Schoen proved results over other local ﬁelds. These proofs used harmonic maps and proceed in three steps: ﬁrst showing a harmonic map exists, second showing it is smooth, and third using certain special Bochner-type formulas to show that the harmonic mapping must be a local isometry. Combined with the earlier work of Matsushima and Murakami, [169], and Raghunathan [199] leads to a complete classiﬁcation of linear representations for lattices in these groups [170, 203]. It is worth noting that the use of harmonic map techniques in rigidity theory had been pioneered by Siu, who used them to prove generalizations of Mostow rigidity for certain classes of Kähler manifolds [220]. There is also much later work on applying harmonic map techniques to reprove cases of Margulis’s superrigidity theorem by Jost-Yau [122] and Mok-Siu-Yeung [171] among others. We remark in passing that the general problem of existence of harmonic maps for non-compact, ﬁnite volume locally symmetric spaces has not been solved in general. Results of Saper [210] and Jost-Zuo [123] allow one to prove superrigidity for fundamental groups of many such manifolds, but only while assuming arithmeticity. The use of harmonic maps in rigidity was inspired by the use of variational techniques and harmonic forms and functions in work on local rigidity and vanishing of cohomology groups. The original suggestion to use harmonic maps in this setting appears to go back to Calabi [219]. 3.4. A Remark on Cocycle Super-rigidity

An important impetus for the study of rigidity of groups acting on manifolds was Zimmer’s proof of his cocycle superrigidity theorem. We discuss this important result below in Section 5 where we also indicate some of its applications to group actions. This theorem is a generalization of Margulis’s superrigidity theorem to the class of virtual homomorphisms corresponding to cocycles over measure preserving group actions. 3.5. Margulis’s Normal Subgroup Theorem

I end this section by mentioning another result of Margulis’s that has had tremendous importance in results concerning group actions on manifolds. This is the normal subgroups theorem, which says that any normal subgroup in a higher-rank lattice is either ﬁnite or of ﬁnite index; see, for example, [162, 245] for more on the proof. The proof precedes by a remarkable strategy.

groups acting on manifolds / 91

Let N be a normal subgroup of ; we show /N is ﬁnite by showing that it is amenable and has property (T ). This strategy has been applied in other contexts and is a major tool in the construction of simple groups with good geometric properties; see [7, 25, 26, 37]. The proof that /N is amenable already involves one step that might rightly be called a theorem about rigidity of group actions. Margulis shows that if G is a semisimple group of higher rank, P is a minimal parabolic, and is a lattice then any measurable space X that is a measurable quotient of the action on G/P is necessarily of the form G/Q , where Q is a parabolic subgroup containing P. The proof of this result, sometimes called the projective factors theorem, plays a fundamental role in the work of Nevo and Zimmer on non-volume preserving actions. See Section 10.3 for more discussion. It is also worth noting that Dani has proven a topological analogue of Margulis’s result on quotients [42]. That is, he has proven that continuous quotients of the action on G/P are all actions on G/Q . The usual use of Margulis’s normal subgroup theorem in studying rigidity of group actions is usually quite straightforward. If one wants to prove that a group satisfying the normal subgroups theorem acts ﬁnitely, it sufﬁces to ﬁnd one inﬁnite order element that acts trivially.

4. Low-Dimensional Actions: Conjectures and Results We begin this section by discussing results, in particular, very low dimensions; namely dimensions 1, 2, and 3. Before we begin this discussion, we recall a result of Thurston that is often used to show that low-dimensional actions are trivial [228].

theorem 4.1. (thurston stability) Assume is a ﬁnitely generated group with ﬁnite abelianization. Let act on a manifold M by C 1 diffeomorphisms, ﬁxing a point p and with trivial derivative at p. Then the action is trivial in a neighborhood of p. In particular, if M is connected, the action is trivial. The main point of Theorem 4.1 is that to show an action is trivial, it often sufﬁces to ﬁnd a ﬁxed point. This is because, for the groups we consider, there are essentially no non-trivial low dimensional linear representations and therefore the derivative at a ﬁxed point is trivial. More precisely, the groups usually only have ﬁnite image low dimensional linear representations, which allows one to see that the action is trivial on a subgroup of ﬁnite index.

92 / d. fisher 4.1. Dimension 1

The most dramatic results obtained in the Zimmer program concern a question ﬁrst brought into focus by Dave Witte Morris in [236]: can higher rank lattices act on the circle? In fact, the paper [236] is more directly concerned with actions on the line R. A detailed survey of results in this direction is contained in the paper by Witte Morris in this volume, so we do not repeat that discussion here. We merely state a conjecture and a question.

conjecture 4.2. Let be a higher rank lattice. Then any continuous action on S1 is ﬁnite. This conjecture is well known and ﬁrst appeared in print in [101]. By the results in [236] and [146], the case of non-cocompact lattices is almost known. The paper [236] does the case of higher Q rank. The later work of Lifschitz and Witte Morris reduces the general case to the case of quasisplit lattices in SL(3, R) and SL(3, C). It follows from work of Ghys [99, 101, 102] or BurgerMonod [22, 23] that one can assume the action ﬁxes a point, so the question is equivalent to asking if the groups act on the line. We remark that actions of these groups on the circle, which ﬁx a point and are C 1 , are easily seen to be ﬁnite by using Theorem 4.1 and the fact that all the groups in question have ﬁnite ﬁrst homology. An interesting approach might be to study the induced action of on the space of left orders on ; for ideas about this approach, we refer the reader to work of Navas [182] and Witte Morris [174]. Perhaps the following should also be a conjecture, but here we only ask it as a question.

question 4.3. Let be a discrete group with property (T ). Does admit an inﬁnite action by C 1 diffeomorphisms on S1 ? Does admit an inﬁnite action by homeomorphisms on S1 ? By a result of Navas, the answer to the above question is no if C 1 diffeomorphisms are replaced by C k diffeomorphisms for any k > 32 [183]. A result noticed by the author and Margulis [82] and contained in a paper of Bader, Furman, Gelander, and Monod [6] allows one to adapt this proof for values of k slightly less than 32 . By Thurston’s Theorem 4.1, in the C 1 case it sufﬁces to ﬁnd a ﬁxed point for the action. We add a remark here pointed out to the author by Navas, that perhaps justiﬁes only calling Question 4.3 a question. In [184], Navas extends his results from [183] to groups with relative property (T ). This means that no

groups acting on manifolds / 93

such group acts on the circle by C k diffeomorphisms where k > 32 . However, if we let be the semi-direct product of a ﬁnite index free subgroup of SL(2, Z) and Z2 , this group is left orderable, and so acts on S1 ; see, for example, [176]. This is the prototypical example of a group with relative property (T ). This leaves open the possibility that groups with property (T ) would behave in a similar manner and admit continuous actions on the circle and the line. Other possible candidate for a group with property (T ) acting on the circle by homeomorphisms are the more general variants of Thompson’s group F constructed by Stein in [225]. The proofs that F does not have (T ) do not apply to these groups [58]. A closely related question is the following, suggested to the author by Andrés Navas along with examples in the last paragraph.

question 4.4. Is there a group that is bi-orderable and does not admit a proper isometric action on a Hilbert space? That is, it does not have the Haagerup property? For much more information on the fascinating topic of groups of diffeomorphisms of the circle see both the survey by Ghys [102], the more recent book by Navas [181], and the article by Witte Morris in this volume [173]. 4.2. Dimension 2

Already in dimension 2 much less is known. There are some results in the volume preserving setting. In particular, we have

theorem 4.5. Let be a non-uniform irreducible higher rank lattice. Then any volume-preserving action on a closed orientable surface other than S2 is ﬁnite. If has Q-rank at least 1, then the same holds for actions on S2 . This theorem was proven by Polterovich [195] for all surfaces but the sphere and shortly afterward proven by Franks and Handel for all surfaces [87, 86]. It is worth noting that the proofs use entirely different ideas. Polterovich’s proof belongs very clearly to symplectic geometry and also implies some results for actions in higher dimensions. Franks and Handel use a theory of normal forms for C 1 surface diffeomorphisms that they develop in analogy with the Thurston theory of normal forms for surface homeomorphisms with ﬁnite ﬁxed sets. The proof of Franks and Handel can be adapted to a setting where one assumes much less regularity of the invariant measure [88].

94 / d. fisher

There are some major reductions in the proofs that are similar, which we now describe. The ﬁrst of these should be useful for studying actions of cocompact lattices as well. Namely, one can assume that the homomorphism ρ : → Diff (S) deﬁning the action takes values in the connected component. This follows from the fact that any ρ : → MCG(S) is ﬁnite, which can now be deduced from a variety of results [17, 22, 53, 124]. This result holds not only for the groups considered in Theorem 4.5, but also for cocompact lattices and even for lattices in SP(1, n) by results of Sai-Kee Yeung [239]. It may be possible to show something similar for all groups with property (T ) using recent results of Andersen showing that the mapping class group does not have property (T ) [2]. It is unrealistic to expect a simple analogue of this result for homomorphisms to Diff (M)/ Diff (M)0 for general manifolds; see Section 4.3 below for a discussion of dimension 3. Also, in the setting of Theorem 4.5, Margulis’s normal subgroup theorem implies that it sufﬁces to show that a single inﬁnite order element of acts trivially. The proofs of Franks-Handel and Polterovich then use the existence of distortion elements in , a fact established by Lubotzky, Mozes, and Raghunathan in [151]. The main result in both cases shows that Diff (S, ω) does not contain exponentially distorted elements and the proofs of this fact are completely different. An interesting result of Calegari and Freedman shows that this is not true of Diff (S2 ) and that Diff (S2 ) contains subgroups with elements of arbitrarily large distortion [30]. These examples are discussed in more detail in Section 9. Polterovich’s methods also allow him to see that there are no exponentially distorted elements in Diff (M, ω)0 for certain symplectic manifolds (M, ω). Again, this yields some partial results toward Zimmer’s conjecture. On the other hand, Franks and Handel are able to work with a Borel measure μ with some properties and show that any map ρ : → Diff (S, μ) is ﬁnite. We remark here that a variant on this is due to Zimmer, when there is an invariant measure supported on a ﬁnite set.

theorem 4.6. Let be a group with property (T ) acting on a compact surface S by C 1 diffeomorphisms. Assume has a periodic orbit on Sj , then the action is ﬁnite. The proof is quite simple. First, pass to a ﬁnite index subgroup that ﬁxes a point x. Look at the derivative representation dρx at the ﬁxed point. Since has property (T ) and SL(2, R) has the Haagerup property, it is easy to prove that

groups acting on manifolds / 95

the image of dρx is bounded. Bounded subgroups of SL(2, R) are all virtually abelian and this implies that the image of dρx is ﬁnite. Passing to a subgroup of ﬁnite index, one has a ﬁxed point where the derivative action is trivial of a group that has no cohomology in the trivial representation. One now applies Thurston Theorem 4.1. The difﬁculty in combining Theorem 4.6 with ideas from the work of Franks and Handel is that while Franks and Handel can show that individual surface diffeomorphisms have large sets of periodic orbits, their techniques do not easily yield periodic orbits for the entire large group action. We remark here that there is another approach to showing that lattices have no volume preserving actions on surfaces. This approach is similar to the proof that lattices have no C 1 actions on S1 via bounded cohomology [22, 23, 99]. That Diff (S, ω)0 admits many interesting quasimorphisms, follows from the work of Entov-Polterovich [49], Gambaudo-Ghys [96], and Py [198, 200, 199]. By results of Burger and Monod [41], for any higher-rank lattice and any homomorphism ρ : → Diff (S, ω)0 , the image is in the kernel of all of these quasi-morphisms. (To make this statement meaningful and the kernel well deﬁned, one needs to take the homogeneous versions of the quasimorphisms.) What remains to be done is to extract useful dynamical information from this fact; see the article by Py in this volume for more discussion [198]. In our context, the work of Entov-Polterovich mentioned above really only constructs a single quasi-morphism on Diff (S2 , ω)0 . However, this particular quasimorphism is very nice in that it is Lipschitz in metric known as Hofer’s metric on Diff (S2 , ω)0 . This construction does apply more generally in higher dimensions and indicates connections between quasimorphisms and the geometry of Diff (M, ω)0 , for ω a symplectic form, that are beyond the scope of this survey. For an introduction to this fascinating topic, we refer the reader to [195]. Motivated by the above discussion, we recall the following conjecture of Zimmer.

conjecture 4.7. Let be a group with property (T ), then any volume preserving smooth action on surface is ﬁnite. Here the words “volume preserving” or at least “measure preserving” are quite necessary. As SL(3, R) acts on S2 , so does any lattice in SL(3, R) or any irreducible lattice in G where G has SL(3, R) as a factor. The following question seems reasonable; I believe I ﬁrst learned it from Leonid Polterovich.

96 / d. fisher

question 4.8. Let be a higher rank lattice (or even just a group with property (T )) and assume acts by diffeomorphisms on a surface S. Is it true that either (1) the action is ﬁnite or (2) the surface is S2 and the action is smoothly conjugate to an action deﬁned by some embedding i : → SL(3, R) and the projective action of SL(3, R) on S2 ? This question seems quite far beyond existing technology. 4.3. Dimension 3

We now discuss brieﬂy some work of Farb and Shalen that constrains actions by homeomorphisms in dimension 3 [55]. This work makes strong use of the geometry of 3 manifolds, but uses very little about higher rank lattices and is quite soft. A special case of their results is the following:

theorem 4.9. Let M be an irreducible 3-manifold and be a higher rank lattice. Assume acts on M by homeomorphisms so that the action on homology is nontrivial. Then M is homeomorphic to T3 , < SL(3, Z) with ﬁnite index, and the action on H 1 (M) is the standard action on Z 3 . Farb and Shalen actually prove a variant of Theorem 4.9 for an arbitrary 3 manifold admitting a homologically inﬁnite action of a higher rank lattice. This can be considered as a signiﬁcant step toward understanding when, for a higher-rank lattice and M 3 a closed 3 manifold, ρ : → Diff (M 3 ) must have image in Diff (M 3 )0 . Unlike the results discussed above for dimension 2, the answer is not simply “always.” This three dimensional result uses a great deal of the known structure of 3 manifolds, though it does not use the full geometrization conjecture proven by Perelman, but only the Haken case due to Thurston. The same sort of result in higher dimensions seems quite out of reach. A sample question is the following. Here we let be a higher rank lattice and M a compact manifold.

question 4.10. Under what conditions on the topology of M do we know that a homomorphism ρ : → Diff (M) has image in Diff (M)0 ? 4.4. Analytic Actions in Low Dimensions

We ﬁrst mention a direction pursued by Ghys that is in a similar spirit to the Zimmer program, and that has interesting consequences for that program. Recall that the Zassenhaus lemma shows that any discrete linear group generated by small enough elements is nilpotent. The main point of Ghys’s

groups acting on manifolds / 97

article [100] is to attempt to generalize this result for subgroups of Diff ω (M). While the result is not actually true in that context, Ghys does prove some intriguing variants that yield some corollaries for analytic actions of large groups. For instance, he proves that SL(n, Z) for n > 3 admits no analytic action on the 2 sphere. We remark that the attempt to prove the Zassenhaus lemma for diffeomorphism groups suggests that one can attempt to generalize other facts about linear groups to the category of diffeomorphism groups. We discuss several questions in this direction, mainly due to Ghys, in Section 13. For the rest of this subsection we discuss a different approach of Farb and Shalen for showing that real analytic actions of large groups are ﬁnite. This method is pursued in [54, 56, 57]. We begin by giving a cartoon of the main idea. Given any action of a group , an element γ ∈, and the centralizer Z(γ ), it is immediate that Z(γ ) acts on the set of γ ﬁxed points. If the action is analytic, then the ﬁxed sets are analytic and so have good structure and are “reasonably close” to being submanifolds. If one further assumes that all normal subgroups of have ﬁnite index, then this essentially allows one to bootstrap results about Z(γ ) not acting on manifolds of dimension at most n − 1 to facts about not having actions on manifolds of dimension n provided one can show that the ﬁxed set for γ is not empty. This is not true, as analytic sets are not actually manifolds, but the idea can be implemented using the actual structure of analytic sets in a way that yields many results. For example, we have

theorem 4.11. Let M be a real analytic 4 manifold with zero Euler characteristic, then any real analytic, volume preserving action of any ﬁnite index subgroup in SL(n, Z) for n ≥ 7 is trivial. This particular theorem also requires a result of Rebelo [205], which concerns ﬁxed sets for actions of nilpotent groups on T2 and uses the ideas of [100]. The techniques of Farb and Shalen can also be used to prove that certain cocompact higher rank lattices have no real analytic actions on surfaces of genus at least 1. For this result one needs only that the lattice contains an element γ whose centralizer already contains a higher rank lattice. In the paper [54] a more technical condition is required, but this can be removed using the results of Ghys and Burger-Monod on actions of lattices on the circle. (This simpliﬁcation was ﬁrst pointed out to the author by Farb.)

98 / d. fisher

The point is that γ has ﬁxed points for topological reasons and the set of these ﬁxed points contains either (1) a Z(γ ) invariant circle or (2) a Z(γ ) invariant point. Case (1) reduces to case (2) after passing to a ﬁnite index subgroup via the results on circle actions. Case (2) is dealt with by the proof of Theorem 4.6. 4.5. Zimmer’s Full Conjecture, Partial Results

We now state the full form of Zimmer’s conjecture. In fact we generalize it slightly to include all lattices with property (T ). Throughout this subsection G will be a semisimple Lie group with property (T ) and will be a lattice in G. We deﬁne two numerical invariants of these groups. First, for any group F, let d(F) be the lowest dimension in which F admits an inﬁnite image linear representation. We note that the superrigidity theorems imply that d() = d(G) when is a lattice in G and either G is a semisimple group property (T ) or is irreducible higher rank. The second number, n(G) is the lowest dimension of a homogeneous space K /C for a compact group K on which a lattice in G can act via a homomorphism ρ : → K . In Zimmer’s work, n(G) is deﬁned differently, in a way that makes clear that there is a bound on n(G) that does not depend on the choice of . Namely, using the superrigidity theorems, it can be shown that n satisﬁes n(n + 1) ≥ min{dimC G where G is a simple factor of G}. 2 The more fashionable variant of Zimmer’s conjecture is the following, ﬁrst made explicitly by Farb and Shalen for higher rank lattices in [54].

conjecture 4.12. Let be a lattice as above. Let b = min{n, d} and let M be a manifold of dimension less than b − 1, then any action on M is trivial. It is particularly bold to state this conjecture including lattices in Sp(1, n) and F4−20 . This is usually avoided because such lattices have abundant nonvolume preserving actions on certain types of highly regular fractals. I digress brieﬂy to explain why I believe the conjecture is plausible in that case by discussing how the actions on fractals arise. Namely such a lattice , being a hyperbolic group, has many inﬁnite proper quotients that are also hyperbolic groups [111]. If is a quotient of by an inﬁnite index, inﬁnite normal subgroup N, and is hyperbolic, then the boundary ∂ is a space with interesting dynamics and a good (Ahlfors regular) quasi-invariant measure class. However, ∂ is only a manifold when it is a sphere. It seems highly

groups acting on manifolds / 99

unlikely that this is ever the case for these groups and even more unlikely that this is ever the case with a smooth boundary action. The only known way to build a hyperbolic group that acts smoothly on its boundary is to have the group be the fundamental group of a compact negatively curved manifold M with smoothly varying horospheres. If smooth is taken to mean C ∞ , this then implies the manifold is locally symmetric [14] and then superrigidity results make it impossible for this to occur with π1 (M) a quotient of a lattice by an inﬁnite index inﬁnite normal subgroup. If smooth only means C 1 , then even in this context, no result known rules out M having fundamental group a quotient of a lattice in Sp(1, n) or F4−20 . All results one can prove using harmonic map techniques in this context only rule out M with non-positive complexiﬁed sectional curvature. We make the following conjecture, which is stronger than what is needed for Conjecture 4.12.

conjecture 4.13. Let be a cocompact lattice in Sp(1, n) or F4−20 and let be a quotient of that is Gromov hyperbolic. If ∂ is a sphere, then the kernel of the quotient map is ﬁnite and ∂ = ∂. For background on hyperbolic groups and their boundaries from a point of view relevant to this conjecture, see [133]. As a cautionary note, we point the reader to Subsection 11.3 where we recall a construction of Farrell and Lafont that shows that any Gromov hyperbolic group has an action by homeomorphisms on a sphere. The version of Zimmer’s conjecture that Zimmer made in [249] and [250] was only for volume preserving actions. Here we break it down somewhat explicitly to clarify the role of d and n.

conjecture 4.14. (zimmer) Let be a lattice as above and assume acts smoothly on a compact manifold M preserving a volume form. Then if dim (M) < d, the action is isometric. If, in addition, dim (M) < n or if is non-uniform, then the action is ﬁnite. Some ﬁrst remarks are in order. The cocycle super-rigidity theorems (discussed below) imply that, when the conditions of the conjecture hold, there is always a measurable invariant Riemannian metric. Also, the ﬁniteness under the conditions in the second half of the conjecture follow from cases of Margulis’s superrigidity theorem as soon as one knows that the action preserves a smooth Riemannian metric. So from one point of view, the conjecture is really about the regularity of the invariant metric.

100 / d. fisher

We should also mention that the conjecture is proven under several additional hypotheses by Zimmer around the time he made the conjecture. The ﬁrst example is the following:

theorem 4.15. Conjecture 4.14 holds provided the action also preserves a rigid geometric structure, for example, a torsion-free afﬁne connection or a pseudoRiemannian metric. This is proven in [246] for structures of ﬁnite type in the sense of Elie Cartan; see also [250]. The fact that roughly the same proof applies for rigid structures in the sense of Gromov was remarked in [85]. The point is simply that the isometry group of a rigid structure acts properly on some higher order frame bundle and that the existence of the measurable metric implies that has bounded orbits on all frame bundles as soon as has property (T ). This immediately implies that must be contained in a compact subgroup of the isometry group of the structure. Another easy version of the conjecture follows from the proof of Theorem 4.6. That is

theorem 4.16. Let G be a semisimple Lie group with property (T ) and < G a lattice. Let act on a compact manifold M by C 1 diffeomorphisms where dim (M) < d(G). If has a periodic orbit on S, then the action is ﬁnite. A more difﬁcult theorem of Zimmer shows that Conjecture 4.14 holds when the action is distal in a sense deﬁned in [247]. 4.6. Some Approaches to the Conjectures

4.6.1. discrete spectrum of actions. A measure preserving action of a group D on a ﬁnite measure space (X , μ) is said to have discrete spectrum if L2 (X , μ) splits as the sum of ﬁnite dimensional D invariant subspaces. This is a strong condition that is (quite formally) the opposite of weak mixing; for a detailed discussion see [92]. It is a theorem of Mackey (generalizing earlier results of Halmos and Von Neumann for D abelian) that an ergodic discrete spectrum D action is measurably isomorphic to one described by a dense embedding of D into a compact group K and considering a K action on a homogeneous K -space. The following remarkable result of Zimmer from [253] is perhaps the strongest evidence for Conjecture 4.14. This result is little known and has only recently been applied by other authors; see [83, 91].

groups acting on manifolds / 101

theorem 4.17. Let be a group with property (T ) acting by smooth, volume preserving diffeomorphisms on a compact manifold M. Assume in addition that preserves a measurable invariant metric. Then the action has discrete spectrum. This immediately implies that no counterexample to Conjecture 4.14 can be weak mixing or even admit a weak mixing measurable factor. The proof of the theorem involves constructing ﬁnite dimensional subspaces of L2 (M, ω) that are invariant. If enough of these subspaces could be shown to be spanned by smooth functions, one would have a proof of Conjecture 4.14. Here by “enough” we simply mean that it sufﬁces to have a collection of ﬁnite dimensional D invariant subspaces that separate points in M. These functions would then specify a D equivariant smooth embedding of M into RN for some large value of N. To construct ﬁnite dimensional invariant subspaces of L2 (M), Zimmer uses an approach similar to the proof of the Peter-Weyl theorem. Namely, he constructs invariant kernels on L2 (M × M) that are used to deﬁne selfadjoint, compact operators on L2 (M). The eigenspaces of these operators are then ﬁnite dimensional, invariant subspaces of L2 (M). The kernels should be thought of as functions of the distance to the diagonal in M × M. The main difﬁculty here is that for these to be invariant by “distance” we need to mean something deﬁned in terms of the measurable metric instead of a smooth one. The construction of the kernels in this setting is quite technical and we refer readers to the original paper. It would be interesting to try to combine the information garnered from this theorem with other approaches to Zimmer’s conjectures.

4.6.2. effective invariant metrics. We discuss here brieﬂy another approach to Zimmer’s conjecture, due to the author, which seems promising. We begin by brieﬂy recalling the construction of the space of “L2 metrics” on a manifold M. Given a volume form ω on M, we can consider the space of all (smooth) Riemannian metrics on M whose associated volume form is ω. This is the space of smooth sections of a bundle P → M. The ﬁber of P is X = SL(n, R)/SO(n). The bundle P is an associated bundle to the SL(n, R) sub-bundle of the frame bundle of M deﬁned by ω. The space X carries a natural SL(n, R)-invariant Riemannian metric of non-positive curvature; we denote its associated distance function by dX . This induces a natural notion of distance on the space of metrics, given by d(g1 , g2 )2 = M dX (g1 (m), g2 (m))2 dω. The completion of the sections with respect to the metric d will be denoted L2 (M, ω, X ); it is commonly referred to as the space of L2 metrics on M and its

102 / d. fisher

elements will be called L2 metrics on M. That this space is CAT (0) follows easily from the fact that X is CAT (0). For more discussion of X and its structure as a Hilbert manifold, see, for example, [79]. It is easy to check that a volume preserving action on M deﬁnes an isometric action on L2 (M, ω, X ). Given a generating set S for and a metric g in L2 (M, ω, X ), we write disp (g) = maxγ ∈S d(γ g, g). Given a group acting smoothly on M preserving ω, this gives an isometric action on L2 (M, ω, X ) that preserves the subset of smooth metrics. Let S be a generating set for . We deﬁne an operator P : L2 (M, ω, X ) → L2 (M, ω, X )

by taking a metric g to the barycenter of the measure S δ(γ g) . The ﬁrst observation is a consequence of the (standard, ﬁnite dimensional) implicit function theorem.

lemma 4.18. If g is a smooth metric, then Pg is also smooth. Moreover, we have the following two results.

theorem 4.19. Let M be a surface and a group with property (T ) and ﬁnite generating set S. Then there exists 0 < C < 1 such that the operator P satisﬁes: 1) disp (Pg) < C disp (g), and 2) for any g, the limn (P n g) exists and is invariant. A proof of this theorem can be given by using the standard construction of a negative deﬁnite kernel on H2 to produce a negative deﬁnite kernel on L2 (S, μ, H2 ). The theorem is then proved by transferring the ﬁrst property from the resulting action on a Hilbert space. The second property is an obvious consequence of the ﬁrst and completeness of the space L2 (M, ω, X ).

theorem 4.20. Let G be a semisimple Lie group all of whose simple factors have property (T ) and < G a lattice. Let M be a compact manifold such that dim (M) < d(G). Then the operator P on L2 (M, ω, X ) satisﬁes the conclusions of Theorem 4.19. This theorem is proven from results in [80] using convexity of the distance function on L2 (M, ω, X ). For cocompact lattices a proof can be given using results in [138] instead. The problem now reduces to estimating the behavior of the derivatives of n P g for some initial smooth g. This expression clearly involves derivatives of random products of elements of , that is, derivatives of elements of weighted by measures that are convolution powers of equidistributed measure on S. The main cause for optimism is that the fact that disp (P n g) is small

groups acting on manifolds / 103

immediately implies that the ﬁrst derivative of any γ ∈S must be small when measured at that point in L2 (M, ω, X ). One can then try to use estimates on compositions of diffeomorphisms and convexity of derivatives to control derivatives of P n g. The key difﬁculty is that the initial estimate on the ﬁrst derivative of γ applied to P n g is only small in an L2 sense. 4.7. Some Related Questions and Conjectures on Countable Subgroups of Diff (M)

In this subsection, we discuss related conjectures and results on countable subgroups of Diff (M). All of these are motivated by the belief that countable subgroups of Diff (M) are quite special, though considerably less special than, say, linear groups. We defer positive constructions of non-linear subgroups of Diff (M) to Section 11 and a discussion of possible algebraic and geometric properties shared by all ﬁnitely generated subgroups of Diff (M) to Section 13. Here we concentrate on groups that do not act on manifolds, either by theorems or conjecturally.

4.7.1. groups with property (T ) and generic groups. We begin by focusing on actions of groups with property (T ). For a ﬁnitely generated group , we recall that d() is the smallest dimension in which admits an inﬁnite image linear representation. We then make the following conjecture: conjecture 4.21. Let be a group with property (T ) acting smoothly on a compact manifold M, preserving volume. Then if dim (M) < d(), the action preserves a smooth Riemannian metric. For many groups with property (T ), one can produce a measurable invariant metric; see [83]. In fact, in [83], Silberman and the author prove that there are many groups with property (T ) with no volume preserving actions on compact manifolds. Key steps include ﬁnding the invariant measurable metric, applying Zimmer’s theorem on discrete spectrum from Section 4.6, and producing groups with no ﬁnite quotients and so no linear representations at all. A result of Furman announced in [90] provides some further evidence for the conjecture. This result is analogous to Proposition 5.1 and shows that any action of a group with property (T ) either leaves invariant a measurable metric or has positive random entropy. We refer the reader to [90] for more discussion. While the proof of this result is not contained in [90], it is possible to reconstruct it from results there and others in [91]. Conjecture 4.21 and the work in [83] are motivated in part by the following conjecture of Gromov:

104 / d. fisher

conjecture 4.22. There exists a model for random groups in which a “generic” random group admits no smooth actions on compact manifolds. It seems quite likely that the conjecture could be true for random groups in the density model with density more than 13 . These groups have property (T ); see Ollivier’s book [193] for a discussion on random groups. For the conjecture to be literally true would require that a random hyperbolic group have no ﬁnite quotients and it is a well-known question to determine whether there are any hyperbolic groups that are not residually ﬁnite, let alone generic ones. If one is satisﬁed by saying the generic random group has only ﬁnite smooth actions on compact manifolds, one can avoid this well-known open question.

4.7.2. universal lattices. Recently, Shalom has proven that the groups SL(n, Z[X ]) have property (T ) when n > 3. (As well as some more general results.) Shalom refers to these groups as universal lattices. An interesting and approachable question is question 4.23. Let be a ﬁnite index subgroup in SL(n, Z[X ]) and let acting by diffeomorphisms on a compact manifold M preserving volume. If dim (M) < n, is there a measurable invariant metric? If one gives a positive answer to this question, one is then clearly interested in whether or not the metric can be chosen to be smooth. It is possible that this is easier for these “larger” groups than for the lattices originally considered by Zimmer. One can ask a number of variants of this question including trying to prove a full cocycle super-rigidity theorem for these groups; see below. The question just asked is particularly appealing as it can be viewed as a ﬁxed point problem for the action on the space of metrics (see Section 4.6 for a deﬁnition). In fact, one knows one has the invariant metric for the action of any conjugate of SL(n, Z) and only needs to show that there is a consistent choice of invariant metrics over all conjugates. One might try to mimic the approach from [216], though a difﬁculty clearly arises at the point where Shalom applies a scaling limit construction. Scaling limits of L2 (M, ω, X ) can be described using nonstandard analysis, but are quite complicated objects and not usually isomorphic to the original space.

4.7.3. irreducible actions of products. Another interesting variant on Zimmer’s conjecture is introduced in [91]. In that paper Furman and Monod study obstructions to irreducible actions of product groups. A measure

groups acting on manifolds / 105

preserving action of a product 1 × 2 on a measure space (X , μ) is said to be irreducible if both 1 and 2 act ergodically on X . Furman and Monod produce many obstructions to irreducible actions; for example, one can prove from their results that

theorem 4.24. Let = 1 × 2 × 3 . Assume that 1 has property (T ) and no unbounded linear representations. Then there are no irreducible, volume preserving actions on compact manifolds. The same is true for = 1 × 2 if 1 is as above and 2 is solvable. A key step in the proof of this result is to use Zimmer’s result on actions with discrete spectrum to show that the action is measurably conjugate to an irreducible isometric action. In particular, the second conclusion follows since no compact group can contain simultaneously a dense (T ) group and a dense solvable group. This motivates the following conjecture.

conjecture 4.25. Let = 1 × 2 . Assume that 1 has property (T ) and no unbounded linear representations and that 2 is amenable. Then there are no irreducible, volume preserving actions on compact manifolds. One might be tempted to prove this by showing that no compact group contains both a dense ﬁnitely generated amenable group and a dense Kazhdan group. This is, however, not true. The question was raised by Lubotzky in [149] but has recently been resolved in the negative by Kassabov [127]. In the context of irreducible actions, asking that groups have property (T ) is perhaps too strong. In fact, there are already relatively few known irreducible actions of Z2 = Z × Z! If one element of Z2 acts as an Anosov diffeomorphism, then it is conjectured by Katok and Spatzier that all actions are algebraic [130]. Even in more general settings where one assumes non-uniform hyperbolicity, there are now hints that a classiﬁcation might be possible, though clearly more complicated than in the Anosov case; see [125] and references there. In a sense, that paper indicates that the “exotic examples” that might arise in this context may be no worse than those that arise for actions of higher rank lattices.

4.7.4. torsion groups and homeo (m). A completely different and wellstudied aspect of the theory of transformation groups of compact manifolds is the study of ﬁnite subgroups of Homeo (M). It has recently been noted by many authors that this study has applications to the study of “large subgroups” of Homeo (M). Namely, one can produce many ﬁnitely generated

106 / d. fisher

and even ﬁnitely presented groups that have no non-trivial homomorphisms to Homeo (M) for any compact M. This is discussed in, for example, [20, 83, 234]. As far as I know, this observation was ﬁrst made by Ghys as a remark in the introduction to [100]. In all cases, the main trick is to construct inﬁnite, ﬁnitely generated groups that contain inﬁnitely many conjugacy classes of ﬁnite subgroups, usually just copies of (Z/pZ)k for all k > 1. A new method of constructing such groups was recently introduced by Chatterji and Kassabov [38]. As far as the author knows, even if one ﬁxes M in advance, this is the only existing method for producing groups with no non-trivial (or even no inﬁnite image) homomorphisms to Homeo (M) unless M = S1 . For M = S1 we refer back to Subsection 4.1 and to [173].

5. Cocycle Super-rigidity and Immediate Consequences 5.1. Zimmer’s Cocycle Super-rigidity Theorem and Generalizations

A main impetus for studying rigidity of group actions on manifolds came from Zimmer’s theorem on superrigidity for cocycles. This theorem and its proof were strongly motivated by Margulis’s work. In fact, Margulis’s theorem reduces to the special case of Zimmer’s theorem for a certain cocycle α : G × G/ → . In order to avoid technicalities, we describe only a special case of this result, essentially avoiding boundedness and integrability assumptions on cocycles that are automatically fulﬁlled in any context arising from a continuous action on a compact manifold. Let M be a compact manifold, H a matrix group, and P an H bundle over M. For readers not familiar with bundle theory, the results are interesting even in the case where P = M × H. Now let a group act on P continuously by bundle automorphisms, that is, such that there is a action on M for which the projection from P to M is equivariant. Further assume that the action on M is measure preserving and ergodic. The cocycle superrigidity theorem says that if is a lattice in a simply connected, semisimple Lie group G all of whose simple factors are noncompact and have property (T ), then there is a measurable map s : M → H, a representation π : G → H, a compact subgroup K < H that commutes with π (G), and a measurable map × M → K such that 5.1

γ ·s(m) = k(m, γ )π (γ )s(γ ·m).

It is easy to check from this equation that the map K satisﬁes the equation that makes it into a cocycle over the action of . One should view s as providing coordinates on P in which the action is almost a product. For more

groups acting on manifolds / 107

discussion of this theorem, particularly in the case where all simple factors of G have higher rank, the reader should see any of [65, 66, 81, 95, 245]. (The version stated here is only proven in [81]; previous proofs all yielded somewhat more complicated statements that require passing to ﬁnite ergodic extensions of the action.) For the case of G with simple factors of the form Sp(1, n) and F4−20 , the results follow from work of the author and Hitchman [80], building on earlier results of Korevaar-Schoen [138, 139, 140] and CorletteZimmer [41]. As a sample application, let M = Tn and let P be the frame bundle of M, that is, the space of frames in the tangent bundle of M. Since Tn is parallelizable, we have P = Tn × GL(n, R). The cocycle super-rigidity theorem then says that “up to compact noise” the derivative of any measure preserving action on Tn looks measurably like a constant linear map. In fact, the cocycle superrigidity theorems apply more generally to continuous actions on any principal bundle P over M with ﬁber H, an algebraic group, and in this context produces a measurable section s : M → P satisfying equation (5.1). So in fact, cocycle superrigidity implies that for any action preserving a ﬁnite measure on any manifold the derivative cocycle looks measurably like a constant cocycle, up to compact noise. That cocycle superrigidity provides information about actions of groups on manifolds through the derivative cocycle was ﬁrst observed by Furstenberg in [95]. Zimmer originally proved cocycle superrigidity in order to study orbit equivalence of group actions. For recent surveys of subsequent developments concerning orbit equivalence rigidity and other forms of superrigidity for cocycles, see [89, 197, 215]. 5.2. First Applications to Group Actions and the Problem of Regularity

The following result, ﬁrst observed by Furstenberg, is an immediate consequence of cocycle superrigidity.

proposition 5.1. Let G be a semisimple Lie group with no compact factors and with property (T ) of Kazhdan and let < G be a lattice. Assume G or acts by volume preserving diffeomorphisms on a compact manifold M. Then there is a linear representation π : G → GL( dim (M), R) such that the Lyapunov exponents of g∈G or are exactly the absolute values of the eigenvalues of π (g). This proposition is most striking to those familiar with classical dynamics where the problem of estimating, let alone computing, Lyapunov exponents is quite difﬁcult. Another immediate consequence of cocycle superrigidity is the following:

108 / d. fisher

proposition 5.2. Let G or be as above, acting by smooth diffeomorphisms on a compact manifold M preserving volume. If every element of G or acts with zero entropy, then there is a measurable invariant metric on M. In particular, if G admits no nontrivial representations of dim (M), then there is a measurable invariant metric on M. As mentioned above in Subsection 4.5, this reduces Conjecture 4.14 to the following technical conjecture.

conjecture 5.3. Let G or acting on M be as above. If G or preserves a measurable Riemannian metric, then they preserve a smooth-invariant Riemannian metric. We recall that ﬁrst evidence toward this conjecture is Theorem 4.15. We remark that in the proof of that theorem, it is not the case that the measurable metric from Proposition 5.2 is shown to be smooth, but instead it is shown that the image of in Diff (M) lies in a compact subgroup. In other work, Zimmer explicitly improves regularity of the invariant metric; see particularly [247]. In general the problem that arises immediately from cocycle superrigidity in any context is understanding the regularity of the straightening section σ . This question has been studied from many points of view, but still relatively little is known in general. For certain examples of actions of higher rank lattices on compact manifolds, discussed below in Section 9, the σ that straightens the derivative cocycle cannot be made smooth on all of M. It is possible that for volume preserving actions on manifolds and the derivative cocycle, σ can always be chosen to be smooth on a dense open set of full measure. We will return to the theme of regularity of the straightening section in Section 7. First we turn to more geometric contexts in which the output of cocycle superrigidity is also often used but usually more indirectly.

6. Geometric Actions and Rigid Geometric Structures In this section, we discuss the role of rigid geometric structures in the study of actions of large groups. The notion of rigid geometric structure was introduced by Gromov, partially in reaction to Zimmer’s work on large group actions. The ﬁrst subsection of this section recalls the deﬁnition of rigid geometric structure, gives some examples, and explains the relation of Gromov’s

groups acting on manifolds / 109

rigid geometric structures to other notions introduced by Cartan. Subsection 6.2 recalls Gromov’s initial results relating actions of simple groups preserving rigid geometric structures on M to representations of the fundamental group of M, and extensions of these results to lattice actions due to Zimmer and the author. Subsection 6.3 concerns a different topic, namely the rigidity of connection preserving actions of a lattice , particularly on manifolds of dimension not much larger than d(G) as deﬁned in Section 4. Subsection 6.4 recalls some obstructions to actions preserving geometric structures, particularly a result known as Zimmer’s geometric Borel density theorem and some recent related results of Bader, Frances, and Melnick. We discuss actions preserving a complex structure in Subsection 6.5 and end with a subsection 6.6 on questions concerning geometric actions. 6.1. Rigid Structures and Structures of Finite Type

In this subsection we recall the formal deﬁnition of a rigid geometric structure. Since the deﬁnition is somewhat technical, some readers may prefer to skip it and read on keeping in mind examples rather than the general notion. The basic examples of rigid geometric structures are Riemannian metrics, pseudoRiemannian metrics, conformal structures deﬁned by either type of metric, and afﬁne or projective connections. Basic examples of geometric structures that are not rigid are a volume form or symplectic structure. An intermediate type of structure that exhibits some rigidity but that is not literally rigid in the sense discussed here is a complex structure. If N is a manifold, we denote the kth order frame bundle of N by F k (N), and by J s,k (N) the bundle of k-jets at 0 of maps from Rs to N. If N and N are two manifolds, and f : N → N is a map between them, then the k-jet j k ( f ) induces a map J s,k N → J s,k N for all s. We let Dk (N) be the bundle whose ﬁber Dpk at a point p consists of the set of k-jets at p of germs of diffeomorphisms of N ﬁxing p. We abbreviate D0k (Rn ) by Dnk or simply Dk ; this is a real algebraic group. For concreteness, one can represent each element uniquely, in terms of standard coordinates (ξ1 , . . . , ξn ) on Rn , in the form (P1 (ξ1 , . . . , ξn ), . . . , Pn (ξ1 , . . . , ξn )) where P1 , P2 , . . . , Pn are polynomials of degree ≤ k. We denote the vector space of such polynomial maps of degree ≤ k by Pn,k . The group Dnk has a natural action on F k (N), where n is the dimension of N. Suppose we are given an algebraic action of Dnk on a smooth algebraic variety Z. Then following Gromov [112], we make the following deﬁnition:

110 / d. fisher

definition 6.1. 1) An A-structure on N (of order k, of type Z) is a smooth map φ : F k (N) → Z equivariant for the Dnk actions. 2) With notation as above, the rth prolongation of φ, denoted φ r , is the map φ r : F k+r (N) → J n,r (Z) deﬁned by φ r = j r (φ) ◦ ιr+k where ιr+k : k k k+r n,r k k n,r k F (N) → J (F (N)) is the natural inclusion and j (h) : J (F (N)) → J n,r (Z) is as before; this is an A-structure of type J n,r (Z) and order k + r. Equivalently, an A-structure of type Z and order k is a smooth section of the associated bundle F k (N) ×Dk Z over N. Note that an A-structure on N deﬁnes by restriction an A-structure φ|U on any open set U ⊂ N.

remark 6.2. A-structures were introduced in [112]; a good introduction to the subject, with many examples, can be found in [10]. A comprehensive and accessible discussion of the results of [112] concerning actions of simple Lie groups can be found in [67]. Note that if N and N are n-manifolds, and h : N → N is a diffeomorphism, then h induces a bundle map j k (h) : F k (N) → F k (N ).

definition 6.3. 1) If φ : F k (N) → Z, φ : F k (N ) → Z are A-structures, a diffeomorphism h : N → N is an isometry from φ to φ if φ ◦ j k (h) = φ. 2) A local isometry of φ is a diffeomorphism h : U1 → U2 , for open sets U1 , U2 ⊂ N, which is an isometry from φ|U1 to φ|U2 . For p ∈ M denote by Isloc p (φ) the pseudogroup of local isometries of φ ﬁxing p, and, for l ≥ k, we denote by Islp (φ) the set of elements jpl (h) ∈ Dpl such that jpl (φ ◦ jpk (h)) = φ l−k , where both sides are considered as maps F k+l (N) → J l−k (Z). Note that Islp (φ) is a group, and there is a natural homomorphism rpl;m : Islp (φ) → Ism p (φ) for m < l; in general, it is neither injective nor surjective.

definition 6.4. The structure φ is called k-rigid if for every point p, the map rpk+1;k is injective. A ﬁrst remark worth making is that an afﬁne connection is a rigid geometric structure and that any generalized quasi-afﬁne action on a manifold of the form K \H/ × M with discrete preserves an afﬁne connection and

groups acting on manifolds / 111

therefore also a torsion free afﬁne connection. In order to provide some more examples, we recall the following lemma of Gromov.

lemma 6.5. Let V be an algebraic variety and G a group acting algebraically on V . For every k, there is a tautological G invariant geometric structure of order k on V , given by ω : P k (V ) → P k (V )/G. This structure is rigid if and only if the action of G on P k (V ) is free and proper. The conclusion in the ﬁrst sentence is obvious. The second sentence is proven in section 0.4, pages 69–70, of [112].

examples 1) The action of G = SLn (R) on Rn is algebraic. So is the action of G on the manifold N1 obtained by blowing up the origin. The reader can easily verify that the action of G on P 2 (N1 ) is free and proper. 2) We can compactify N1 by N2 by viewing the complement of the blow up as a subset of the projective space P n . Another description of the same action, which may make the rigid structure more visible to the naked eye, is as follows: SLn+1 (R) acts on P n . Let G be SLn (R) < SLn+1 (R) as block diagonal matrices with blocks of size n and 1 and a 1 × 1 block equal to 1. Then G acts on P n ﬁxing a point p. We can obtain N2 by blowing up the ﬁxed point p. The G actions on both P n and N2 are algebraic and again the reader can verify that the action is free and proper on P 2 (N2 ). 3) In the construction from (2) above, there is an action of a group H where H = SLn (R) Rn and G = SLn (R) < SLn (R) Rn < SLn+1 (R). The H action ﬁxes the point p and so also acts on N2 algebraically. However, over the exceptional divisor, the action is never free on any frame bundle, since the subgroup Rn acts trivially to all orders at the exceptional divisor. The behavior in example (3) above illustrates the fact that existence of invariant rigid structures is more complicated for algebraic groups that are not semisimple; see the discussion in section 0.4.C. of [112].

definition 6.6. A rigid geometric structure is called a ﬁnite type geometric structure if V is a homogeneous Dnk space. This is by no means the original deﬁnition that is due to Cartan and predates Gromov’s notion of a rigid geometric structure by several decades. It

112 / d. fisher

is equivalent to Cartan’s deﬁnition of a ﬁnite type structure by the work of Candel and Quiroga-Barranco [31, 32, 201]. Candel and Quiroga-Barranco also give a development of rigid geometric structures that more closely parallels the older notion of a structure of ﬁnite type as presented in, for example, [135]. All standard examples of rigid geometric structures are structures of ﬁnite type. However, example (2) above is not a structure of ﬁnite type. The notion of structure of ﬁnite type was ﬁrst given in terms of prolongations of Lie algebras and so yields a criterion that is, in principle, computable. The work of Candel and Quiroga-Barranco extends this computable nature to general rigid geometric structures. 6.2. Rigid Structures and Representations of Fundamental Groups

In this subsection, we restrict our attention to actions of simple Lie groups G and lattices < G. Many of the results of this section extend to semisimple G, though the formulations become more complicated. A major impetus for Gromov’s introduction of rigid geometric structures is the following theorem from [112].

theorem 6.7. Let G be a simple noncompact Lie group and M a compact real analytic manifold. Assume G acts on M preserving an analytic rigid geometric structure ω and a volume form ν. Further assume the action is ergodic. Then there is a linear representation ρ : π1 (M) → GL(n, R) such that the Zariski closure of ρ(π1 (M)) contains a group locally isomorphic to G. We remark that, by the Tits’s alternative, the theorem immediately implies that actions of the type described cannot occur on manifolds with amenable fundamental group or even on manifolds whose fundamental groups do not contain a free group on two generators [229]. Expository accounts of the proof can be found in [67, 255, 258]. The representation ρ is actually deﬁned on ˜ Killing ﬁelds of the lift ω˜ of ω to M. This is a worthwhile moment to indicate the weakness of Principle 2.6. If we start with a cocompact lattice < G and an action of on a compact manifold M satisfying the assumptions of Theorem 6.7, we can induce the action to a G action on the compact manifold N = (G × M)/ . However, in this setting, there is an obvious representation of π1 (N) satisfying the conclusion of Theorem 6.7. This is because there is a surjection π1 (N) → . In fact for most cocompact lattices < G there are homomorphisms σ : → K where K is compact and simply connected so that we can have act on K by

groups acting on manifolds / 113

left translation and obtain examples where π1 (N) = . For G of higher rank, the following theorem of Zimmer and the author shows that this is the only obstruction to a variant of Theorem 6.7 for lattices.

theorem 6.8. Let < G be a lattice, where G is a simple group and R − rank(G) ≥ 2. Suppose acts analytically and ergodically on a compact manifold M preserving a unimodular rigid geometric structure. Then either 1) the action is isometric and M = K /C where K is a compact Lie group and the action is by right translation via ρ : → K , a dense image homomorphism, or 2) there exists an inﬁnite image linear representation σ : π1 (M) → GLn R, such that the algebraic automorphism group of the Zariski closure of σ (π1 (M)) contains a group locally isomorphic to G. The proof of this theorem makes fundamental use of a notion that does not occur in its statement: the entropy of the action. A key observation is the following formula for the entropy of the restriction of the induced action to : h(G×M)/ (γ ) = hG/ (γ ) + hM (γ ). We then use the fact that the last term on the right hand side is 0 if and only if the action preserves a measurable Riemannian metric as already explained in Proposition 5.1. In this setting, it then follows from Theorem 4.15 that the action is isometric and the other conclusions in (1) are simple consequences of ergodicity of the action. When hM (γ ) > 0 for some , we use relations between the entropy of the action and the Gromov representation discovered by Zimmer [257]. We discuss some of these ideas below in Section 8. 6.3. Afﬁne Actions Close to the Critical Dimension

In this section we brieﬂy describe a direction pursued by Feres, Goetze, Zeghib, and Zimmer that classify connection preserving actions of lattices in dimensions close to, but not less than, the critical dimension d. A representative result is the following:

theorem 6.9. Let n > 2, G = SL(n, R) and < G a lattice. Let M be a compact manifold with dim (M) ≤ n + 1 = d(G) + 1. Assume that acts on M preserving a volume form and an afﬁne connection. Then either

114 / d. fisher

1) dim (M) < n and the action is isometric, 2) dim (M) = n and either the action is isometric or, upon passing to ﬁnite covers and ﬁnite index subgroups, smoothly conjugate to the standard SL(n, Z) action on Tn , or 3) dim (M) = n + 1 and either the action is isometric or, upon passing to ﬁnite covers and ﬁnite index subgroups, smoothly conjugate to the action of SL(n, Z) on Tn+1 = Tn × T where the action on the second factor is trivial. Variants of this theorem under more restrictive hypotheses were obtained by Feres [63], Goetze [104], and Zimmer [248]. The theorem as stated is due to Zeghib [243]. The cocycle superrigidity theorem is used in all of the proofs, mainly to force vanishing of curvature and torsion tensors of the associated connection. One would expect similar results for a more general class of acting groups and also for a wider range of dimensions. Namely, if we let d2 (G) be the dimension of the second non-trivial representation of G, we would expect a similar result to (3) for any afﬁne, volume preserving action on a manifold M with d(G) < d2 (G). With the further assumption that the connection is Riemannian (and still only for lattices in SL(n, R)) this is proven by Zeghib in [242]. Further related results are contained in other papers of Feres [60, 64]. No results of this kind are known for dim (M) ≥ d2 (G). A major difﬁculty arises as soon as one has dim (M) ≥ d2 (G), namely that more complicated examples can arise, including afﬁne actions on nilmanifolds and left translation actions on spaces of the form G/. All the results mentioned in this subsection depend on the particular fact that ﬂat Riemannian manifolds have ﬁnite covers that are tori. 6.4. Zimmer’s Borel Density Theorem and Generalizations

An obvious ﬁrst question concerning G actions is the structure of the stabilizer subgroups. In this direction we have

theorem 6.10. Let G be a simple Lie group acting essentially faithfully on a compact manifold M preserving a volume form, then the stabilizer of almost every point is discrete. As an application of this result, one can prove the following:

theorem 6.11. Let G be a simple Lie group acting essentially faithfully on a compact manifold, preserving a volume form and a homogeneous geometric structure with structure group H. Then there is an inclusion g → h.

groups acting on manifolds / 115

In particular, the theorem provides an obstruction to a higher rank simple Lie group having a volume preserving action on a compact Lorentz manifold. For more discussion of these theorems, see [258]. This observation led to a major series of works studying automorphism groups of Lorentz manifolds; see, for example, [1, 141, 240, 241]. We remark here that Theorem 6.11 does not require that the homogeneous structure be rigid. More recently some closely related phenomena have been discovered in a joint work of Bader, Frances, and Melnick [5]. Their work uses yet another notion of a geometric structure, that of a Cartan geometry. We will not deﬁne this notion rigorously here; it sufﬁces to note that any rigid homogeneous geometric structure deﬁnes a Cartan geometry, as discussed in [5, introduction]. The converse is not completely clear in general, but most classical examples of rigid geometric structures can also be realized as Cartan geometries. A Cartan geometry is essentially a way of saying that a manifold is inﬁnitesimally modeled on some homogeneous space G/P. To recapture the notion of a Riemannian connection, G = O(n) Rn and P = O(n), to recapture the notion of an afﬁne connection G = Gl(n, R) Rn and P = Gl(n, R). Cartan connections come with naturally deﬁned curvatures that vanish if and only if the manifold is locally modeled on G/P. We refer the reader to the book [217] for a general discussion of Cartan geometries and their use in differential geometry. Given a connected linear group L, we deﬁne its real rank, rk(L) as before to be the dimension of a maximal R-diagonalizable subgroup of L, and let n(L) denote the maximal nilpotence degree of a connected nilpotent subgroup. One of the main results of [5] is

theorem 6.12. If a group L acts by automorphisms of a compact Cartan geometry modeled on G/P, then 1) rk(AdL0 ) ≤ rk(Adg P), 2) n(AdL0 ) ≤ n(Adg P). The main point is that L is not assumed either simple or connected. This theorem is deduced from an embedding theorem similar in ﬂavor to Theorem 6.10. In addition, when G is a simple group and P is a maximal parabolic subgroup, Bader, Frances, and Melnick prove rigidity results classifying all possible actions when the rank bound in Theorem 6.12 is achieved. This classiﬁcation essentially says that all examples are algebraic; see [5] for a detailed discussion.

116 / d. fisher

An earlier paper by Feres and Lampe also explored applications of Cartan geometries to rigidity and dynamical conditions for ﬂatness of Cartan geometries [62]. 6.5. Actions Preserving a Complex Structure

We mention here one recent result by Cantat and a few active related directions of research. In [250], Zimmer asked whether one had restrictions on low dimensional holomorphic actions of lattices. The answer was obtained in [36] and is

theorem 6.13. Let G be a connected simple Lie group of real rank at least 2 and suppose < G is a lattice. If admits a faithful action by automorphisms on a compact Kähler manifold M, then the rank of G is at least the complex dimension of M. The proof of the theorem depends primarily on results concerning Aut (M), particularly recent results on holomorphic actions of abelian groups by Dinh and Sibony [47]. It is worth noting that while the theorem does depend on Margulis’s superrigidity theorem, it does not depend on Zimmer’s cocycle superrigidity theorem. The condition that the action is faithful can be weakened considerably using Margulis’s normal subgroups theorem. More recently, Cantat [35], Déserti [46], and others have begun a program of studying large subgroups of automorphism groups of complex manifolds. In particular, Cantat has proven an analogue of Conjecture 4.7 in the context of birational actions on complex surfaces.

theorem 6.14. Let S be a compact Kähler surface and G an inﬁnite, countable group of birational transformations of S. If G has property (T ), then there is a birational map j : S → P2 (C) that conjugates G to a subgroup of Aut (P2 (C)). 6.6. Conjectures and Questions

A major open problem, generalizing Conjecture 1.2, is the following:

conjecture 6.15. (gromov-zimmer) Let D be a semisimple Lie group with all factors having property (T ) or a lattice in such a Lie group. Then any D action on a compact manifold preserving a rigid geometric structure and a volume form is generalized quasi-afﬁne. A word is required on the attribution. Both Gromov [112] and Zimmer [249] made various less precise conjectures concerning the classiﬁcation of actions

groups acting on manifolds / 117

as in the conjecture. The exact statement of the correct conjecture was muddy for several years while it was not known if the Katok-Lewis and Benveniste type examples admitted invariant rigid geometric structures. See Section 9 for more discussion of these examples. In the context of the results of [13], where Benveniste and the author prove that those actions do not preserve rigid geometric structures, this version of the classiﬁcation seems quite plausible. It is perhaps more plausible if one assumes the action is ergodic or that the geometric structure is homogeneous. A possibly easier question that is relevant is the following:

question 6.16. Let M be a compact manifold equipped with a homogeneous rigid geometric structure ω. Assume Aut (M, ω) is ergodic or has a dense orbit. Is M locally homogeneous? Gromov’s theorem on the open-dense implies that a dense open set in M is locally homogeneous even if ω is not homogeneous. The question is whether this homogeneous structure extends to all of M if ω is homogeneous. If ω is not homogeneous, the examples following Lemma 6.5 show that M need not be homogeneous. It seems possible to approach the case of Conjecture 6.15 where ω is homogeneous and D acts ergodically by answering Question 6.16 positively. At this point, one is left with the problem of classifying actions of higher rank groups and lattices on locally homogeneous manifolds. Induction easily reduces one to considering G actions. The techniques Gromov uses to produce the locally homogeneous structure gives slightly more precise information in this setting: one is left trying to classify homogeneous manifolds modeled on H/L where G acts via an inclusion in H centralizing L. In fact this reduces one to problems about locally homogeneous manifolds studied by Zimmer and collaborators; see particularly [142, 143, 145, 256]. For a more general survey of locally homogeneous spaces, see also [136] and [137]. The following question concerns a possible connection between preserving a rigid geometric structure and having uniformly partially hyperbolic dynamics.

question 6.17. Let D be any group acting on a compact manifold M preserving a volume form and a rigid geometric structure. Is it true that if some element d of D has positive Lyapunov exponents, then d is uniformly partially hyperbolic?

118 / d. fisher

This question is of particular interest for us when D is a semisimple Lie group or a lattice, but would be interesting to resolve in general. The main reason to believe that the answer might be yes is that the action of D on the space of frames of M is proper. Having a proper action on a tangent bundle minus the zero section implies that an action is Anosov; see Mañé’s article for a proof [156].

7. Topological Super-rigidity: Regularity Results from Hyperbolic Dynamics and Applications A major area of research in the Zimmer program has been the application of hyperbolic dynamics. This area might be described by the maxim: in the presence of hyperbolic dynamics, the straightening section from cocycle superrigidity is often more regular. Some major successes in this direction are the work of Goetze-Spatzier and Feres-Labourie. In the context of hyperbolic dynamical approaches, there are two main settings. In the ﬁrst, one considers actions of higher rank lattices on a particular class of compact manifolds and a second in which one makes no assumption on the topology of the manifold acted upon. In this section we discuss a few such results, after ﬁrst recalling some facts about the stability of hyperbolic dynamical systems in Subsection 7.1. We then discuss best-known rigidity results for actions on tori in Subsection 7.2 and after this we discuss the work of Goetze-Spatzier and Feres-Labourie in the more general context in Subsection 7.3. The term “topological superrigidity” for this area of research was coined by Zimmer, whose early unpublished notes on the topic dramatically inﬂuenced research in the area [244]. This aspect of the Zimmer program has also given rise to a study of rigidity properties for other uniformly hyperbolic actions of large groups, most particularly higher rank abelian groups. See, for example, [126, 130, 209]. The following subsection recalls basic notions from hyperbolic dynamics that are needed in this section. 7.1. Stability in Hyperbolic Dynamics

A diffeomorphism f of a manifold X is said to be Anosov if there exists a continuous f invariant splitting of the tangent bundle TX = Efu ⊕ Efs and constants a > 1 and C, C > 0 such that for every x∈X , 1) Df n (v u )≥Can v u for all v u ∈Efu (x), and 2) Df n (v s )≤C a−n v s for all v s ∈Efs (x).

groups acting on manifolds / 119

Note that the constants C and C depend on the choice of metric, and that a metric can always be chosen so that C = C = 1. There is an analogous notion for a ﬂow ft , where TX = T O ⊕ Efut ⊕ Efst where T O is the tangent space to the ﬂow direction and vectors in Efut (respectively, Efst ) are uniformly expanded (respectively, uniformly contracted) by the ﬂow. This notion was introduced by Anosov [3] and named after Anosov by Smale [221], who popularized the notion in the United States. One of the earliest results in the subject is Anosov’s proof that Anosov diffeomorphisms are structurally stable, that is, that any C 1 perturbation of an Anosov diffeomorphism is conjugate back to the original diffeomorphism by a homeomorphism. There is an analogous result for ﬂows, though this requires that one introduce a notion of time change that we will not consider here. Since Anosov also showed that C 2 volume preserving Anosov ﬂows and diffeomorphisms are ergodic, structural stability implies the existence of an open set of “chaotic” dynamical systems. The notion of an Anosov diffeomorphism has had many interesting generalizations, for example: Axiom A diffeomorphisms, non-uniformly hyperbolic diffeomorphisms, and diffeomorphisms admitting a dominated splitting. A notion that has been particularly useful in the study of rigidity of group actions is the notion of a partially hyperbolic diffeomorphism as introduced by Hirsch, Pugh, and Shub [177]. Under strong enough hypotheses, these diffeomorphisms have a weaker stability property similar to structural stability. More or less, the diffeomorphisms are hyperbolic relative to some foliation, and any nearby action is hyperbolic to some nearby foliation. To describe more precisely the class of diffeomorphisms considered and the stability property they enjoy, we require some deﬁnitions. The use of the word “foliation” varies with context. Here a foliation by C k leaves will be a continuous foliation whose leaves are C k injectively immersed submanifolds that vary continuously in the C k topology in the transverse direction. To specify transverse regularity we will say that a foliation is transversely C r . A foliation by C k leaves that is tranversely C k is called simply a C k foliation. (Note our language does not agree with that in the reference [118].) Given an automorphism f of a vector bundle E → X and constants a > b ≥ 1, we say f is (a, b)-partially hyperbolic or simply partially hyperbolic if there is a metric on E, a constant and C ≥ 1, and a continuous f invariant, nontrivial splitting E = Efu ⊕ Efc ⊕ Efs such that for every x in X : 1) f n (v u )≥Can v u for all v u ∈Efu (x),

2) f n (v s )≤C −1 a−n v s for all v s ∈Efs (x), and

3) C −1 b−n v 0 < f n (v 0 )≤Cbn v 0 for all v 0 ∈Efc (x) and all integers n.

120 / d. fisher

A C 1 diffeomorphism f of a manifold X is (a, b)-partially hyperbolic if the derivative action Df is (a, b)-partially hyperbolic on TX . We remark that for any partially hyperbolic diffeomorphism, there always exists an adapted metric for which C = 1. Note that Efc is called the central distribution of f , Efu is called the unstable distribution of f , and Efs the stable distribution of f . Integrability of various distributions for partially hyperbolic dynamical systems is the subject of much research. The stable and unstable distributions are always tangent to invariant foliations that we call the stable and unstable foliations and denote by Wfs and Wfu , respectively. If the central distribution is tangent to an f invariant foliation, we call that foliation a central foliation and denote it by Wfc . If there is a unique foliation tangent to the central distribution, we call the central distribution uniquely integrable. For smooth distributions unique integrability is a consequence of integrability, but the central distribution is usually not smooth. If the central distribution of an (a, b)-partially hyperbolic diffeomorphism f is tangent to an invariant foliation Wfc , then we say f is r-normally hyperbolic to Wfc for any r such that a > br . This is a special case of the deﬁnition of r-normally hyperbolic given in [118]. Before stating a version of one of the main results of [118], we need one more deﬁnition. Given a group G, a manifold X , two foliations F and F of X , and two actions ρ and ρ of G on X , such that ρ preserves F and ρ preserves F , following [118] we call ρ and ρ leaf conjugate if there is a homeomorphism h of X such that 1) h(F) = F, and 2) for every leaf L of F and every g∈G, we have h(ρ(g)L) = ρ (g)h(L). The map h is then referred to as a leaf conjugacy between (X , F, ρ) and (X , F , ρ ). This essentially means that the actions are conjugate modulo the central foliations. We state a special case of some of the results of Hirsch-Pugh-Shub on perturbations of partially hyperbolic actions of Z; see [118]. There are also analogous deﬁnitions and results for ﬂows. As these are less important in the study of rigidity, we do not discuss them here.

theorem 7.1. Let f be an (a, b)-partially hyperbolic C k diffeomorphism of a compact manifold M that is k-normally hyperbolic to a C k central foliation Wfc . Then for any δ > 0, if f is a C k diffeomorphism of M that is sufﬁciently C 1 close to f , we have the following:

groups acting on manifolds / 121

1) f is (a , b )-partially hyperbolic, where |a − a | < δ and |b − b | < δ, and the splitting TM = Efu ⊕ Efc ⊕ Efs for f is C 0 close to the splitting for f, 2) there exist f invariant foliations by C k leaves Wfc tangent to Efc , which is close in the natural topology on foliations by C k leaves to Wfc , 3) there exists a (nonunique) homeomorphism h of M with h(Wfc ) = Wfc , and h is C k along leaves of Wfc , furthermore h can be chosen to be C 0 small and C k small along leaves of Wfc , and 4) the homeomorphism h is a leaf conjugacy between the actions (M, Wfc , f ) and (M, Wfc , f ). Conclusion (1) is easy and probably older than [118]. One motivation for Theorem 7.1 is to study stability of dynamical properties of partially hyperbolic diffeomorphisms. See the survey [27] by Burns, Pugh, Shub, and Wilkinson for more discussion of that and related issues. 7.2. Uniformly Hyperbolic Actions on Tori

Many works have been written considering local rigidity of actions with some afﬁne, quasi-afﬁne, and generalized quasi-afﬁne actions with hyperbolic behavior. For a discussion of this, we refer to [77]. Here we only discuss results that prove some sort of global rigidity of groups acting on manifolds. The ﬁrst such results were contained in papers of Katok-Lewis [128] and Katok-LewisZimmer [129]. As these are now special cases of later more general results, we do not discuss them in detail here. In this section we discuss results that provide only continuous conjugacies to standard actions. This is primarily because these results are less technical and easier to state. In this context, one can improve regularity of the conjugacy given certain technical dynamical hypotheses on certain dynamical foliations.

definition 7.2. An action of a group on a manifold M is weakly hyperbolic if there exist elements γ1 , . . . , γk each of which is partially hyperbolic such that the sum of the stable sub-bundles of the γi spans the tangent bundle

to M at every point, that is, i Eγi = TM. To discuss the relevant results we need a related topological notion that captures hyperbolicity at the level of fundamental group. This was introduced in [84] by Whyte and the author. If a group acts on a manifold with torsion free nilpotent fundamental group and the action lifts to the universal cover, then

122 / d. fisher

the action of on π1 (M) gives rise to an action of on the Malcev completion N of π1 (M), which is a nilpotent Lie group. This yields a representation of on the Lie algebra n.

definition 7.3. We say an action of on a manifold M with nilpotent fundamental group is π1 -hyperbolic if for the resulting representation on n, we have ﬁnitely many elements γ1 , . . . , γk such that the sum of their eigenspaces with eigenvalue of modulus less than 1 is all of n. One can make this deﬁnition more general by considering M where π1 (M) has a equivariant nilpotent quotient; see [84]. We now discuss results that follow by combining the work of Margulis-Qian [165] with the later work of Schmidt [211] and Fisher-Hitchman [79].

theorem 7.4. Let M = N/ be a compact nilmanifold and let be a lattice in a semisimple Lie group with property (T ). Assume acts on M such that the action lifts to the universal cover and is π1 (M) hyperbolic, then the action is continuously semi-conjugate to an afﬁne action. This theorem was proven by Margulis and Qian, who noted that if the action contained an Anosov element, then the conjugacy could be taken to be a homeomorphism. In [84], the author and Whyte point out that this theorem extends easily to the case of any compact manifold with torsion free nilpotent fundamental group. In [84], we also discuss extensions to manifolds with fundamental group with a quotient that is nilpotent. Margulis and Qian asked whether the assumption of π1 hyperbolicity could be replaced by the assumption that the action on M was weakly hyperbolic. In the case of actions of Kazhdan groups on tori, Schmidt proved that weak hyperbolicity implies π1 hyperbolicity, yielding:

theorem 7.5. Let M = Tn be a compact torus and let be a lattice in a semisimple Lie group with property (T ). Any weakly hyperbolic action M and that lifts to the universal cover is continuously semi-conjugate to an afﬁne action. Remarks: 1) The contribution of Fisher-Hitchman in both theorems is just in extending cocycle superrigidity to a wider class of groups, as discussed above.

groups acting on manifolds / 123

2) The assumption that the action lifts to the universal cover of M is often vacuous because of results concerning cohomology of higher rank lattices. In particular, it is vacuous for cocompact lattices in simple Lie groups of real rank at least 3. It remains an interesting, open question to take this result and prove that the semiconjugacy is always a conjugacy and is also always a smooth diffeomorphism. 7.3. Rigidity Results for Uniformly Hyperbolic Actions

We begin by discussing some work of Goetze and Spatzier. To avoid technicalities we discuss only some of their results. We begin with the following deﬁnition:

definition 7.6. Let ρ : Zk × M → M be an action and γ1 , . . . , γl be a collection of elements that generate for Zk . We call ρ a Cartan action if 1) each ρ(γi ) is an Anosov diffeomorphism, 2) each ρ(γi ) has 1 dimensional strongest stable foliation, and 3) the strongest stable foliations of the ρ(γi ) are pairwise transverse and span the tangent space to the manifold. It is worth noting that Cartan actions are very special in three ways. First, we assume that a large number of elements in the acting group are Anosov diffeomorphisms; second, we assume that each of these has 1 dimensional strongest stable foliation; and lastly, we assume these 1 dimensional directions span the tangent space. All aspects of these assumptions are used in the following theorem. Reproving it even assuming 2 dimensional strongest stable foliations would require new ideas.

theorem 7.7. Let G be a semisimple Lie group with all simple factors of real rank at least 2 and in G a lattice. Then any volume preserving Cartan action of is smoothly conjugate to an afﬁne action on an infranilmanifold. This is slightly different than the statement in Goetze-Spatzier, where they pass to a ﬁnite cover and a ﬁnite index subgroup. It is not too hard to prove this statement from theirs. The proof spans the two papers [105] and [106]. The ﬁrst paper [105] proves, in a somewhat more general context, that the π-simple section arising in the cocycle superrigidity theorem is in fact Hölder continuous. The second paper [105] makes use of the resulting Hölder

124 / d. fisher

Riemannian metric in conjunction with ideas arising in other work of Katok and Spatzier to produce a smooth homogeneous structure on the manifold. The work of Feres-Labourie differs from other work on rigidity of actions with hyperbolic properties in that it does not make any assumptions concerning the existence of invariant measures. Here we state only some consequences of their results, without giving the exact form of cocycle superrigidity that is their main result.

theorem 7.8. Let be a lattice in SL(n, R) for n ≥ 3 and assume acts smoothly on a compact manifold M of dimension n. Further assume that for the induced action N = (G × M)/ we have 1) every R-semisimple 1-parameter subgroup of G acts transitively on N, and 2) some element g in G is uniformly partially hyperbolic with Es ⊕Ew containing the tangent space to M at any point, then M is a torus and the action on M is a standard afﬁne action. These hypotheses are somewhat technical and essentially ensure that one can apply the topological version of cocycle superrigidity proven in [61]. The proof also uses a deep result of Benoist and Labourie classifying Anosov diffeomorphisms with smooth stable and unstable foliations [11]. The nature of the hypotheses of Theorem 7.8 make an earlier remark clear. Ideally one would only have hypotheses on the action, but here we require hypotheses on the induced action instead. It is not clear how to reformulate these hypotheses on the induced action as hypotheses on the original action. Another consequence of the work of Feres and Labourie is a criterion for promoting invariance of rigid geometric structures. More precisely they give a criterion for a G action to preserve a rigid geometric structure on a dense open subset of a manifold M provided a certain type of subgroup preserves a rigid geometric structure on M. 7.4. Conjectures and Questions on Uniformly Hyperbolic Actions

We begin with a very general variant of Conjecture 1.3.

conjecture 7.9. Let G be a semisimple Lie group all of whose simple factors have property (T ), let < G be a lattice. Assume G or acts smoothly on a compact manifold M preserving volume such that some element g in the acting group is nontrivially uniformly partially hyperbolic. Then the action is generalized quasi-afﬁne.

groups acting on manifolds / 125

There are several weaker variants on this conjecture, where, for example, one assumes the action is volume preserving or weakly hyperbolic. Even the following much weaker variant seems difﬁcult:

conjecture 7.10. Let M = Tn and as in Conjecture 7.9. Assume acts on Tn weakly hyperbolicly and preserving a smooth measure. Then the action is afﬁne. This conjecture amounts to conjecturing that the semiconjugacy in Theorem 7.5 is a diffeomorphism. In the special case where some element of is Anosov, the semiconjugacy is at least a homeomorphism. If this is true and enough dynamical foliations are 1 and 2 dimensional and has higher rank, one can then deduce smoothness of the conjugacy from the work of Rodriguez-Hertz on rigidity of actions of abelian groups [209]. Work in progress by the author, Kalinin, and Spatzier seems likely to provide a similar result when contains many commuting Anosov diffeomorphisms without any assumptions on dimensions of foliations. Finally, we recall an intriguing question from [84] that arises in this context.

question 7.11. Let M be a compact manifold with π1 (M) = Zn and assume < SL(n, Z) has ﬁnite index. Let act on M ﬁxing a point so that the resulting action on π1 (M) is given by the standard representation of SL(n, Z) on Zn . Is it true that dim (M) ≥ n? The question is open even if the action on M is assumed to be smooth. The results of [84] imply that there is a continuous map from M to Tn that is equivariant for the standard action on M. Since the image of M is closed and invariant, it is easy to check that it is all of Tn . So the question amounts to one about the existence of equivariant “space ﬁlling curves,” where curve is taken in the generalized sense of continuous map from a lower-dimensional manifold. In another context there are equivariant space ﬁlling curves, but they seem quite special. They arise as surface group equivariant maps from the circle to S2 and come from three manifolds that ﬁber over the circle; see [34].

8. Representations of Fundamental Groups and Arithmetic Quotients In this section we discuss some results and questions related to topological approaches to classifying actions; particularly, some related to Conjecture 1.4. In Subsection 8.2, we also discuss related results and questions concerning maximal generalized afﬁne quotients of actions.

126 / d. fisher 8.1. Linear Images of Fundamental Groups

This section is fundamentally concerned with the question:

question 8.1. Let G be a semisimple Lie group all of whose factors are not compact and have property (T ). Assume G acts by homeomorphisms on a manifold M preserving a measure. Can we classify linear representations of π1 (M)? Similarly for actions of < G a lattice on a manifold M. We remark that in this context, it is possible that π1 (M) has no inﬁnite image linear representations; see discussion in [84] and Section 9 below. In all known examples where this occurs there is an “obvious” inﬁnite image linear representation on some ﬁnite index subgroup. Also, as ﬁrst observed by Zimmer [251] for actions of Lie groups, under mild conditions on the action, representations of the fundamental group become severely restricted. Further work in this direction was done by Zimmer in conjunction with Spatzier [224] and later Lubotzky [152, 153]. Analogous results for lattices are surprisingly difﬁcult in this context and constitute the authors dissertation [73, 74]. We recall a deﬁnition from [251]:

definition 8.2. Let D be a Lie group and assume that D acts on a compact manifold M preserving a ﬁnite measure μ and ergodically. We call the action ˜ on any ﬁnite cover of M is ergodic. engaging if the action of D There is a slightly more technical deﬁnition of engaging for non-ergodic actions that says there is no loss of ergodicity on passing to ﬁnite covers. That is that the ergodic decomposition of μ and its lifts to ﬁnite covers are canonically identiﬁed by the covering map. There are also two variants of this notion totally engaging and topologically engaging; see for example, [258] for more discussion. It is worth noting that for ergodic D actions on M, the action on any ﬁnite cover has at most ﬁnitely many ergodic components, in fact at most the degree of the cover many ergodic components. We remark here that any generalized afﬁne action of a Lie group is engaging if it is ergodic. The actions constructed below in Section 9 are not in general engaging.

theorem 8.3. Let G be a simple Lie group of real rank at least 2. Assume G acts by homeomorphisms on a compact manifold M, preserving a ﬁnite measure μ and engaging. Assume σ : π1 (M) → GL(n, R) is an inﬁnite image linear representation. Then σ (π1 (M)) contains an arithmetic group HZ where HR contains a group locally isomorphic to G.

groups acting on manifolds / 127

For an expository account of the proof of this theorem and a more detailed discussion of engaging conditions for Lie groups, we refer the reader to [258]. The extension of this theorem to lattice actions is non-trivial and is in fact the author’s dissertation. Even the deﬁnition of engaging requires modiﬁcation, since it is not at all clear that a discrete group action lifts to the universal cover.

definition 8.4. Let D be a discrete group and assume that D acts on a compact manifold M preserving a ﬁnite measure μ and ergodically. We call the action engaging if for every 1) ﬁnite index subgroup D in D, 2) ﬁnite cover M of M, and 3) lift of the D action to M , the action of D of M is ergodic. The deﬁnition does immediately imply that every ﬁnite index subgroup D of D acts ergodically on M. In [74], a deﬁnition is given that does not require the D action to be ergodic, but even in that context the ergodic decomposition for D is assumed to be the same as that for D. Deﬁnition 8.4 is rigged to guarantee the following lemma:

lemma 8.5. Let G be a Lie group and < G a lattice. Assume acts on a manifold M preserving a ﬁnite measure and engaging, then the induced G action on (G × M)/ is engaging. We remark that with our deﬁnitions here, the lemma only makes sense for cocompact, but this is not an essential difﬁculty. We can now state a ﬁrst result for lattice actions. Let = π1 ((G × M)/ ).

theorem 8.6. Let G be a simple Lie group of real rank at least 2 and < G a lattice. Assume acts by homeomorphisms on a compact manifold M, preserving a ﬁnite measure μ and engaging. Assume σ : → GL(n, R) is a linear representation whose restriction to π1 (M) has inﬁnite image. Then σ (π1 (M)) contains an arithmetic group HZ where Aut (HR ) contains a group locally isomorphic to G. This theorem is proven by inducing actions, applying Theorem 8.3, and analyzing the resulting output carefully. One would like to assume σ a priori only deﬁned on π1 (M), but there seems no obvious way to extend such a

128 / d. fisher

representation to the linear representation of required by Theorem 8.3 without a priori information on . This is yet another example of the difﬁculties in using induction to study lattice actions. As a consequence of Theorem 8.6, we discover that at least σ () splits as a semidirect product of and σ (π1 (M)). But this is not clear a priori and not clear a posteriori for . 8.2. Arithmetic Quotients

In the context of Theorems 8.3 and 8.6, one can obtain much greater dynamical information concerning the relation of HZ and the dynamics of the action on M. In particular, there is a compact subgroup C < HR and a measurable equivariant map φ : M → C\HR /HZ , which we refer to as a measurable arithmetic quotient. The papers [73] and [152] prove that there is always a canonical maximal quotient of this kind for any action of G or on any compact manifold, essentially by using Ratner’s theorem to prove that every pair of arithmetic quotients are dominated by a common, larger arithmetic quotient. Earlier results of Zimmer also obtained arithmetic quotients, but only under the assumption that there was an inﬁnite image linear representation with discrete image [254]. The results we mention from [74] and [153] then show that there are “lower bounds” on the size of this arithmetic quotient, provided that the action is engaging in terms of the linear representations of the fundamental group of the manifold. In particular, one obtains arithmetic quotients where HR and HZ are essentially determined by σ (π1 (M)). In fact, σ (π1 (M)) contains HZ and is contained in HQ . The book [258] provides a good description of how to produce arithmetic quotients for G actions and the article [75] provides an exposition of the relevant constructions for actions. Under even stronger hypotheses, the papers [85] and [257] imply that the arithmetic quotient related to the “Gromov representation” discussed above in Subsection 6.2 has the same entropy as the original action. This means that, in a sense, the arithmetic quotient captures most of the dynamics of the original action. 8.3. Open Questions

As promised in the introduction, the analogue of Conjecture 1.4 for lattice actions follows.

conjecture 8.7. Let be a lattice in a semisimple group G with all simple factors of real rank at least 2. Assume acts faithfully, preserving volume on a compact manifold M. Further assume the action is not isometric. Then π1 (M) has

groups acting on manifolds / 129

a ﬁnite index subgroup such that surjects onto an arithmetic lattice in a Lie group H where Aut (H) locally contains G. The following questions about arithmetic quotients are natural.

question 8.8. Let G be a simple Lie group of real rank at least 2 and < G a lattice. Assume that G or act faithfully on a compact manifold M, preserving a smooth volume. 1) Is there a non-trivial measurable arithmetic quotient? 2) Can we take the quotient map φ smooth on an open dense set? Due to a construction in [84], one cannot expect that every M admitting a volume preserving G or action has an arithmetic quotient that is even globally continuous or has π1 (M) admitting an inﬁnite image linear representation. However, the difﬁculties created in those examples all vanish on passage to a ﬁnite cover.

question 8.9. Let G be a simple Lie group of real rank at least 2 and < G a lattice. Assume that G or act smoothly on a compact manifold M, preserving a smooth volume. 1) Is there a ﬁnite cover of M of M such that π1 (M ) admits an inﬁnite image linear representation? 2) Can we ﬁnd a ﬁnite cover M of M, a lift of the action to M (on a subgroup of ﬁnite index) and an arithmetic quotient where the quotient map φ is continuous and smooth on an open dense set? The examples discussed in the next section imply that φ is at best Hölder continuous globally. It is not clear whether the questions and conjectures we have just discussed are any less reasonable for SP(1, n), F4−20 and their lattices.

9. Exotic Volume Preserving Actions In this section, we discuss what is known about what are typically called “exotic actions.” These are the only known smooth volume preserving actions of higher rank lattices and Lie groups that are not generalized afﬁne algebraic. These actions make it clear that a clean classiﬁcation of volume preserving actions is out of reach. In particular, these actions have continuous moduli and provide counterexamples to any naive conjectures of the form “the moduli

130 / d. fisher

space of actions of some lattice on any compact manifold M are countable.” In particular, I explain examples of actions of either or G that have large continuous moduli of deformations as well as manifolds where these moduli have multiple connected components. Essentially all of the examples given here derive from the simple construction of “blowing up” a point or a closed orbit, which was introduced to this subject in [128]. The further developments after that result are all further elaborations on one basic construction. The idea is to use the “blow up” construction to introduce distinguished closed invariant sets that can be varied in some manner to produce deformations of the action. The “blow up” construction is a classical tool from algebraic geometry that takes a manifold N and a point p and constructs from it a new manifold N by replacing p by the space of directions at p. Let RP l be the l dimensional projective space. To blow up a point, we take the product of N × RP dim (N) and then ﬁnd a submanifold where the projection to N is a diffeomorphism off of p and the ﬁber of the projection over p is RP dim (N) . For detailed discussion of this construction we refer the reader to any reasonable book on algebraic geometry. The easiest example to consider is to take the action of SL(n, Z), or any subgroup < SL(n, Z) on the torus Tn and blow up the ﬁxed point, in this case the equivalence class of the origin in Rn . Call the resulting manifold M. Provided is large enough, for example, Zariski dense in SL(n, R), this action of does not preserve the measure deﬁned by any volume form on M. A clever construction introduced in [128] shows that one can alter the standard blowing up procedure in order to produce a one parameter family of SL(n, Z) actions on M, only one of which preserves a volume form. This immediately shows that this action on M admits perturbations, since it cannot be conjugate to the nearby, nonvolume preserving actions. Essentially, one constructs different differentiable structures on M that are diffeomorphic but not equivariantly diffeomorphic. After noticing this construction, one can proceed to build more complicated examples by passing to a subgroup of ﬁnite index, and then blowing up several ﬁxed points. One can also glue together the “blown up” ﬁxed points to obtain an action on a manifold with more complicated topology. In particular, one can achieve a fundamental group that is an essentially arbitrary free product with amalgamation or HNN extension of the fundamental group of the original manifold over the fundamental group of (blown-up) orbits. In the context described in more detail below, of blowing up along closed orbits instead of points, it is not hard to do the blowing up and gluing in a way that guarantees that there are no linear representations of the fundamental group of the “exotic example.” To prove nonexistence of linear representations, one

groups acting on manifolds / 131

chooses examples where all groups involved are higher rank lattices and have very constrained linear representation theory. See [84, 128] for a discussion of the topological complications one can introduce. While these actions do not preserve a rigid geometric structure, they do preserve a slightly more general object, an almost rigid structure introduced by Benveniste and the author in [13] and described below. In [12] it is observed that a similar construction can be used for the action of a simple group G by left translations on a homogeneous space H/ where H is a Lie group containing G and < H is a cocompact lattice. Here we use a slightly more involved construction from algebraic geometry, and “blow up” the directions normal to a closed submanifold. That is, we replace some closed submanifold N in H/ by the projective normal bundle to N. In all cases we consider here, this normal bundle is trivial and so is just N × RP l where l = dim (H) − dim (N). Benveniste used his construction to produce more interesting perturbations of actions of higher rank simple Lie group G or a lattice in G. In particular, he produced volume preserving actions that admit volume preserving perturbations. He does this by choosing G < H such that not only are there closed G orbits but so that the centralizer Z = ZH (G) of G in H has nontrivial connected component. If we take a closed G orbit N, then any translate zN for z in Z is also closed and so we have a continuum of closed G orbits. Benveniste shows that if we choose two closed orbits N and zN to blow up and glue, and then vary z in a small open set, the resulting actions can only be conjugate for a countable set of choices of z. This construction is further elaborated in [78]. Benveniste’s construction is not optimal in several senses, nor is his proof of rigidity. In [78], I give a modiﬁcation of the construction that produces non-conjugate actions for every choice of z in a small enough neighborhood. By blowing up and gluing more pairs of closed orbits, this allows me to produce actions where the space of deformations contains a submanifold of arbitrarily high, ﬁnite dimension. Further, Benveniste’s proof that the deformation are nontrivial is quite involved and applies only to higher rank groups. In [78], I give a different proof of nontriviality of the deformations using consequences of Ratner’s theorem [204] due to Shah [213] and Witte Morris [237]. This shows that the construction produces nontrivial perturbations for any semisimple G and any lattice in G. As mentioned above, in [13] we show that none of these actions preserve any rigid geometric structure in the sense of Gromov but that they do preserve a slightly more complicated object that we call an almost rigid structure. Both

132 / d. fisher

rigid and almost rigid structures are easiest to deﬁne in dimension 1. In this context a rigid structure is a non-vanishing vector ﬁeld and an almost rigid structure is a vector ﬁeld vanishing at isolated points and to ﬁnite degree. We continue to use the notation of subsection 6.1 in order to give the precise deﬁnition of almost rigid geometric structure.

definition 9.1. An A-structure φ is called ( j, k)-almost rigid (or just almost rigid) if for every point p, rpk,k−1 is injective on the subgroup r k+j,k (Isk+j ) ⊂ Isk . Thus k-rigid structures are the (0, k)-almost rigid structures.

basic example. Let V be an n-dimensional manifold. Let X1 , . . . , Xn be a 2 collection of vector ﬁelds on M. This deﬁnes an A-structure ψ of type Rn on M. If X1 , . . ., Xn span the tangent space of V at every point, then the structure is rigid in the sense of Gromov. Suppose instead that there exists a point p in V and X1 ∧ . . . ∧ Xn vanishes to order ≤ j at p in V . Then ψ is a ( j, 1)almost rigid structure. Indeed, let p ∈ M, and let (x1 , . . . , xn ) be coordinates ∂ around p. Suppose that in terms of these coordinates, Xl = am . Suppose l ∂xj j+1

j+1,1

that f ∈ Isp . We must show that rp coordinate functions of f . Then f ∈ 9.1

j+1 Isp

alk − am k

( f ) is trivial. Let ( f 1 , . . . , f n ) be the implies that

∂f l ∂x m

vanishes to order j + 1 at p for all k and l. Let (bkl ) be the matrix so that bkm alm = ∂f l det (asr )δkl . Multiplying expression (9.1) by (bkl ), we see that det (asr ) δkl − ∂x k vanishes to order j + 1. But since by assumption det (asr ) vanishes to order ≤ j, j+1,1 this implies that (∂f l /∂x k )(p) = δkl , so rp ( f ) is the identity, as required. If confused by the notation, the interested reader may ﬁnd it enlightening to work out the basic example in the trivial case n = 1. Similar arguments can be given to show that frames that degenerate to subframes are also almost rigid, provided the order of vanishing of the form deﬁning the frame is always ﬁnite.

question 9.2. Does any smooth (or analytic) action of a higher rank lattice admit a smooth (analytic) almost rigid structure in the sense of [13]? More generally does such an action admit a smooth (analytic) rigid geometric structure on an open dense set of full measure?

groups acting on manifolds / 133

This question is, in a sense, related to the discussion above about regularity of the straightening section in cocycle superrigidity. In essence, cocycle superrigidity provides one with a measurable invariant connection and what one wants to know is whether one can improve the measurable connection to a smooth geometric structure with some degeneracy on a small set. The examples described in this section show, among other things, that one cannot expect the straightening section to be smooth in general, though one might hope it is smooth in the complement of a closed submanifold of positive codimension or at least on an open dense set of full measure. We remark that there are other possible notions of almost rigid structures. See the article in this volume by Dumitrescu for a detailed discussion of a different useful notion in the context of complex analytic manifolds [48]. Dumitrescu’s notion is strictly weaker than the one presented here.

10. Non-Volume Preserving Actions This section describes what is known for non-volume preserving actions. Subsections 10.1 and 10.2 describe examples which show that a classiﬁcation in this setting is not in any sense possible. Subsection 10.3 describes some recent work of Nevo and Zimmer that proves surprisingly strong rigidity results in special settings. 10.1. Stuck’s Examples

The following observations are from Stuck’s paper [226]. Let G be any semisimple group. Let P be a minimal parabolic subgroup. Then there is a homomorphism P → R. As in the proof of Theorem 2.7, one can take any R action, view it is a P action, and induce to a G action. If we take an R action on a manifold M, then the induced action takes place on (G × M)/P. We remark that the G action here is not volume preserving, simply because the G action on G/P is proximal. The same is true of the restriction to any action when is a lattice in G. This construction shows that classifying G actions on all compact manifolds implicitly involves classifying all vector ﬁelds on compact manifolds. It is relatively easy to convince oneself that there is no reasonable sense in which the moduli space of vector ﬁelds can be understood up to smooth conjugacy. 10.2. Weinberger’s Examples

This is a variant on the Katok-Lewis examples obtained by blowing up a point and is similar to a blowing up construction common in foliation theory. The

134 / d. fisher

idea is that one takes an action of a subgroup of SL(n, Z) on M = Tn , removes ¯ and then a ﬁxed or periodic point p, retracts onto a manifold with boundary M, n glues in a copy of the R compactiﬁed at inﬁnity by the projective space of rays. It is relatively easy to check that the resulting space admits a continuous action and even that there are many invariant measures for the action, but no invariant volume. One can also modify this construction by doing the same construction at multiple ﬁxed or periodic points simultaneously and by doing ¯ more complicated gluings on the resulting M. This construction is discussed in [55] and a variant for abelian group actions is discussed in [125]. In the abelian case it is possible to smooth the action, but this does not seem to be the case for actions of higher rank lattices. As far as I can tell, there is no obstruction to repeating this construction for closed orbits as in the case of the algebro-geometric blow up, but this does not seem to be written formally anywhere in the literature. Also, a recent construction of Hurder shows that one can iterate this construction inﬁnitely many times, taking retracts in smaller and smaller neighborhoods of periodic points of higher and higher orders. The resulting object is a kind of fractal admitting an SL(n, Z) action [119]. 10.3. Work of Nevo-Zimmer

In this subsection, we describe some work of Nevo and Zimmer from the sequence of papers [187, 188, 191, 186]. For more detailed discussion see the survey by Nevo and Zimmer [189] as well as the following two articles for related results [190, 192]. Given a group G acting on a space X and a measure μ on G, we call a measure ν on X stationary if μ ∗ ν = ν. We will only consider the case where the group generated by the support μ is G; such measures are often called admissible. This is a natural generalization of the notion of an invariant measure. If G is an amenable group, any action of G on a compact metric space admits an invariant measure. If G is not amenable, invariant measures need not exist, but stationary measures always do. We begin with the following cautionary example:

example 10.1. Let G = SL(n, R) acting on Rn+1 by the standard linear action on the ﬁrst n coordinates and the trivial action on the last. Then the corresponding G action on P(Rn+1 ) has the property that any stationary measure is supported either on the subspace P(Rn ) given by the ﬁrst n coordinates or on the subspace P(R) given by zeroing the ﬁrst n coordinates.

groups acting on manifolds / 135

The proof of this assertion is an easy exercise. The set P(R) is a collection of ﬁxed points, so clearly admits invariant measures. The orbit of any point

in P(Rn+1 )\(P(Rn ) P(R)) is SL(n, R)/(SL(n, R) Rn ). It is straightforward to check that no stationary measures can be supported on unions of sets of this kind. This fact should not be a surprise as it generalizes the fact that invariant measures for amenable group actions are often supported on minimal sets. To state the results of Nevo and Zimmer, we need a slightly stronger notion of admissibility. We say a measure μ on a locally compact group G is strongly admissible if the support of μ generates G and μ∗k is absolutely continuous with respect to Haar measure on G for some positive k. In the papers of Nevo and Zimmer, this stronger notion is called admissible.

theorem 10.2. Let X be a compact G space where G is a semisimple Lie group with all factors of real rank at least 2. Then for any admissible measure μ on G and any μ-stationary measure ν on X , we have either 1) ν is G invariant, or 2) there is a nontrivial measurable quotient of the G space (X , ν) that is of the form (G/Q , Lebesgue) where Q ⊂ G is a parabolic subgroup. The quotient space G/Q is called a projective quotient in the work of Nevo and Zimmer. Theorem 10.2 is most interesting for us for minimal actions, where the measure ν is necessarily supported on all of X and the quotient therefore reﬂects the action on X and not some smaller set. Example 10.1 indicates the reason to be concerned, since there the action on the larger projective space is not detected by any stationary measure. We remark here that Feres and Ronshausen have introduced some interesting ideas for studying group actions on sets not contained in the support of any stationary measure [69]. Similar ideas are developed in a somewhat different context by Deroin, Kleptsyn, and Navas [45]. In [186], Nevo and Zimmer show that for smooth non-measure preserving actions on compact manifolds that also preserve a rigid geometric structure, one can sometimes prove the existence of a projective factor where the factor map is smooth on an open dense set. We also want to mention the main result of Stuck’s paper [226], which we have already cited repeatedly for the fact that non-volume preserving actions of higher rank groups on compact manifolds cannot be classiﬁed.

theorem 10.3. Let G be a semisimple Lie group with ﬁnite center. Assume G acts minimally by homeomorphisms on a compact manifold M. Then either the action

136 / d. fisher

is locally free or the action is induced from a minimal action by homeomorphisms of a proper parabolic subgroup of G on a manifold N. Stuck’s theorem is proven by studying the Gauss map from M to the Grassman variety of subspaces of g deﬁned by taking a point to the Lie algebra of its stabilizer. This technique also plays an important role in the work of Nevo and Zimmer.

11. Groups That Do Act on Manifolds Much of the work discussed so far begs an obvious question: “are there many interesting subgroups of Diff (M) for a general M?” So far we have only seen “large” subgroups of Diff (M) that arise in a geometric fashion, from the pres˜ In this section, ence of a connected Lie group in either Diff (M) or Diff (M). we describe two classes of examples that make it clear that other phenomena exist. The following problem, however, seems open:

problem 11.1. For a compact manifold M with a volume form ω construct a subgroup < Diff (M, ω) such that has no linear representations. An example that is not often considered in this context is that Aut (Fn ) acts on the space Hom (Fn , K ) for K any compact Lie group. Since Hom (Fn , K ) K n this deﬁnes an action of Aut (Fn ) on a manifold. This action clearly preserves the Haar measure on K n ; see [107]. This action is not very well studied; we only know of [76, 97]. Similar constructions are possible and better known with mapping class groups, although in that case one obtains a representation variety that is not usually a manifold. Since Aut (Fn ) has no faithful linear representations, this yields an example of truly “non-linear” action of a large group. This action is still very special and one expects many other examples of non-linear actions. We now describe two constructions that yield many examples if we drop the assumption that the action preserves volume. 11.1. Thompson’s Groups

Richard Thompson introduced a remarkable family of groups, now referred to as Thompson’s groups. These come in various ﬂavors and have been studied from several points of view; see, for example, [33, 117]. For our purposes, the most important of these groups are the one’s typically denoted T . One description of this group is the collection of piecewise linear diffeomorphisms of the circle where the break points and slopes are all dyadic rationals. (One can

groups acting on manifolds / 137

replace the implicit 2 here with other primes and obtain similar, but different, groups.) We record here two important facts about this group T of piecewise linear homeomorphisms of S1 .

theorem 11.2. (thompson, see [33]) The group T is simple. theorem 11.3. (ghys-sergiescu [103]) The deﬁning piecewise linear action of the group T on S1 is conjugate by a homeomorphism to a smooth action. These two facts together provide us with a rather remarkable class of examples of groups that act on manifolds. As ﬁnitely generated linear groups are always residually ﬁnite, the group T has no linear representations whatsoever. A simpler variant of Problem 11.1 is

problem 11.4. Does T admit a volume preserving action on a compact manifold? It is easy to see that compactness is essential in this question. We can construct a smooth action of T on S1 × R simply by taking the Ghys-Sergiescu action on S1 and acting on R by the inverse of the derivative cocycle. Replacing the derivative cocycle with the Jacobian cocycle, this procedure quite generally converts non-volume preserving actions on compact manifolds to volume preserving one’s on non-compact manifolds, but we know of no real application of this in the present context. Another variant of Problem 11.4 is

problem 11.5. Given a compact manifold M and a volume form ω does Diff (M, ω) contain a ﬁnitely generated, inﬁnite discrete simple group? This question is reasonable for any degree of regularity on the diffeomorphisms. In Ghys’s survey on groups acting on the circle, he points out that Thompson’s group can be realized as piecewise SL(2, Z) homeomorphisms of the circle [102]. In [70], Whyte and the author point out that the group of piecewise SL(n, Z) maps on either the torus or the real projective space is quite large. The following are natural questions; see [70] for more discussion.

question 11.6. Are there interesting ﬁnitely generated or ﬁnitely presented subgroups of piecewise SL(n, Z) maps on Tn or Pn−1 ? Can any such group that is not a subgroup of SL(n, Z) be made to act smoothly?

138 / d. fisher 11.2. Highly Distorted Subgroups of Diff (S2 )

In [30], Calegari and Freedman construct a very interesting class of subgroups of Diff ∞ (S2 ). Very roughly, they prove:

theorem 11.7. There is a ﬁnitely generated subgroup G of Diff ∞ (S2 ) that contains a rotation r as an arbitrarily distorted element. Here by arbitrarily distorted we mean that we can choose the group G so that the function f (n) = r n G grows more slowly than any function we choose. It is well known that for linear groups, the function f (n) is at worst a logarithm, so this theorem immediately implies that we can ﬁnd G with no faithful linear representations. This also answered a question raised by Franks and Handel in [88]. More recently, Avila has constructed similar examples in Diff ∞ (S1 ) [4]. This answers a question raised in [30] where such subgroups were constructed in Diff 1 (S1 ). We are naturally led to the following questions. We say a diffeomorphism has full support if the complement of the ﬁxed set is dense.

question 11.8. For which compact manifolds M does Diff ∞ (M) contain arbitrarily distorted elements of full support? The same question for Diff ω (M)? The same question for Diff ∞ (M, ν) where ν is a volume form on M? For the second two questions, we can drop the hypothesis of full support. The second and third questions here seem quite difﬁcult and the answer could conceivably be “none.” The only examples where anything is known in the volume preserving setting are compact surfaces of genus at least 1, where no element is more than quadratically distorted by a result of Polterovich [196]. However, this result depends heavily on the fact that in dimension 2, preserving a volume form is the same as preserving a symplectic structure. 11.3. Topological Construction of Actions on High Dimensional Spheres

In this subsection, we recall a construction due to Farrell and Lafont that allows one to construct actions of a large class of groups on closed disks, and so by doubling, to construct actions on spheres [59]. The construction yields actions on very high-dimensional disks and spheres and is only known to produce actions by homeomorphisms.

groups acting on manifolds / 139

The class of groups involved is the set of groups that admit an EZ-structure. This notion is a modiﬁcation of an earlier notion due to Bestvina of a Zstructure on a group [16]. We do not recall the precise deﬁnition here, but refer the reader to the introduction to [59] and remark here that both torsion free Gromov hyperbolic groups and CAT (0)-groups admit EZ-structures. The result that concerns us is

theorem 11.9. Given a group with an EZ-structure, there is an action of by homeomorphisms on a closed disk. In fact, Farrell and Lafont give a fair amount of information concerning their actions that are quite different from the actions we are usually concerned with here. In particular, the action is properly discontinuous off a closed subset of the boundary of the disk. So from a dynamical viewpoint carries the “interesting part” of the action, for example, it is the support of any stationary measure. Farrell and Lafont point out an analogy between their construction and the action of a Kleinian group on the boundary of hyperbolic space. An interesting general question is

question 11.10. When can the Farrell-Lafont action on a disk or sphere be chosen smooth? 12. Rigidity for Other Classes of Acting Groups In this section, we collect some results and questions concerning actions of other classes of groups. In almost all cases, little is known. 12.1. Lattices in Semi-direct Products

While it is not reasonable to expect classiﬁcation results for arbitrary actions of all lattices in all Lie groups, there are natural broader classes to consider. To pick a reasonable class, a ﬁrst guess is to try to exclude Lie groups whose lattices have homomorphisms onto Z or larger free groups. There are many such groups. For example, the groups Sl(n, Z) Zn that are lattices in Sl(n, R) Rn have property (T ) as soon as n > 2. As it turns out, a reasonable setting is to consider perfect Lie groups with no compact factors or factors locally isomorphic to SO(1, n) or SU(1, n). Any such Lie group will have property (T ) and therefore so will its lattices. Many examples of such lattices are described in [230]. Some ﬁrst rigidity results for these groups are contained in Zimmer’s paper [252]. The relevant full generalization of the cocycle superrigidity

140 / d. fisher

theorem is contained in [238]. In this context it seems that there are probably many rigidity theorems concerning actions of these groups already implicit in the literature, following from the results in [238] and existing arguments. 12.2. Universal Lattices

As mentioned above in Subsection 4.7, the groups SL(n, Z[X ]) for n > 2 have property (T ) by a result of Shalom [216]. His proof also works with larger collections of variables and some other arithmetic groups. The following is an interesting problem.

problem 12.1. Prove cocycle superrigidity for universal lattices. For clarity, we indicate that we expect that all linear representations and cocycles (up to “compact noise”) will be described in terms of representations of the ambient groups, for example, SL(n, R), and will be determined by specifying a numerical value of X . Some partial results toward superrigidity are known by Farb [52] and Shenﬁeld [218]. The problem of completely classifying linear representations in this context does seem to be open. 12.3. SO(1,n) and SU(1,n) and Their Lattices

It is conjectured that all lattices in SO(1, n) and SU(1, n) admit ﬁnite-index subgroups with surjections to Z; see [18]. The conjecture is usually attributed to Thurston and sometimes to Borel for the case of SU(1, n). If this is true, it immediately implies that actions of those lattices can never be classiﬁed. There are still some interesting results concerning actions of these lattices. In [214], Shalom places restrictions on the possible actions of SO(1, n) and SU(1, n) in terms of the fundamental group of the manifold acted upon. This work is similar in spirit to the work of Lubotzky and Zimmer described in Section 8, but requires more restrictive hypotheses. In [78], the author exhibits large moduli spaces of ergodic afﬁne algebraic actions that are constructed for certain lattices in SO(1, n). These moduli spaces are, however, all ﬁnite dimensional. In [71], I construct an inﬁnite-dimensional moduli of deformations of an isometric action of SO(1, n). Both of these constructions rely on a notion of bending introduced by Johnson and Millson in the ﬁnite dimensional setting [121]. I do not formulate any precise questions or conjectures in this direction as I am not sure what phenomenon to expect.

groups acting on manifolds / 141

12.4. Lattices in Other Locally Compact Groups

Much recent work has focused on developing a theory of lattices in locally compact groups other than Lie groups. This theory is fully developed for algebraic groups over other local ﬁelds. Though we did not mention it here, some of Zimmer’s own conjectures were made in the context of S-arithmetic groups, that is, lattices in products of real and p-adic Lie groups. The following conjecture is natural in this context and does not seem to be stated in the literature.

conjecture 12.2. Let G be semisimple algebraic group deﬁned over a ﬁeld k of positive characteristic and < G a lattice. Further assume that all simple factors of G have k-rank at least 2. Then any action on a compact manifold factors through a ﬁnite quotient. The existence of a measurable invariant metric in this context should be something one can deduce from the cocycle superrigidity theorems, though it is not clear that the correct form of these theorems is known or in the literature. There is also a growing interest in lattices in locally compact groups that are not algebraic. We remark here that Kac-Moody lattices typically admit no nontrivial homomorphisms even to Homeo (M); see [83] for a discussion. The only other interesting class of lattices known to the author is the lattices in the isometry group of a product of two trees constructed by Burger and Mozes [26, 25]. These groups are inﬁnite simple ﬁnitely presented groups.

problem 12.3. Do the Burger-Mozes lattices admit any nontrivial homomorphisms to Diff (M) when M is a compact manifold? That these groups do act by homeomorphisms on high dimensional spheres by the construction discussed in Subsection 11.3 was pointed out to the author by Lafont. 12.5. Automorphism Groups of Free and Surface Groups

We brieﬂy mention a last set of natural questions. There is a long-standing analogy between higher rank lattices and two other classes of groups. These are mapping class groups of surfaces and the outer automorphism group of the free group. See [21] for a detailed discussion of this analogy. In the context of this article, this raises the following question. Here we denote the mapping class group of a surface by MCG() and the outer automorphism group of Fn by Out (Fn ).

142 / d. fisher

question 12.4. Assume has genus at least 2 or that n > 2. Does MCG() or Out (Fn ) admit a faithful action on a compact manifold? A faithful action by smooth, volume preserving diffeomorphisms? By not assuming that M is connected, we are implicitly asking the same question about all ﬁnite-index subgroups of MCG() and Out(Fn ). Recall from Section 11 that Aut (Fn ) does admit a volume preserving action on a compact manifold. This makes the question above particularly intriguing.

13. Properties of Subgroups of Diff (M) As remarked in Subsection 4.4, in the paper [100], Ghys attempts to reprove the classical Zassenhaus lemma for linear groups for groups of analytic diffeomorphisms. While the full strength of the Zassenhaus lemma does not hold in this setting, many interesting results do follow. This immediately leaves one wondering to what extent other properties of linear groups might hold for diffeomorphism groups or at least what analogues of many theorems might be true. This direction of research was initiated by Ghys and most of the questions below are due to him. It is worth noting that some properties of ﬁnitely generated linear groups, like residual ﬁniteness, do not appear to have reasonable analogues in the setting of diffeomorphism groups. For residual ﬁniteness, the obvious example is the Thompson group discussed above, which is simple. 13.1. Jordan’s Theorem

For linear groups, there is a classical (and not too difﬁcult) result known as Jordan’s theorem. This says that for any ﬁnite subgroup of GL(n, C), there is a subgroup of index at most c(n) that is abelian. One cannot expect better than this, as S1 has ﬁnite subgroups of arbitrarily large order and is a subgroup of GL(n, C). For proofs of Jordan’s theorem as well as the theorems on linear groups mentioned in the next subsection, we refer the reader to, for example, [208].

question 13.1. Given a compact manifold M and a ﬁnite subgroup F of Diff (M), is there a constant c(M) such that F has an abelian subgroup of index c(M)? As above, one cannot expect more than this simply because one can construct actions of S1 on M and, therefore, ﬁnite abelian subgroups of Diff (M) of arbitrarily large order.

groups acting on manifolds / 143

For this question, it may be more natural to ask about ﬁnite groups of homeomorphisms and not assume any differentiability of the maps. Using the results in, for example, [157], one can show that at most ﬁnitely many simple ﬁnite groups act on a given compact manifold. To be clear, one can show this using the classiﬁcation of ﬁnite simple groups. It would be most interesting to resolve Question 13.1 without reference to the classiﬁcation. A recent preprint of Mundet i Rieri provides some evidence for a positive answer to the question [180]. 13.2. Burnside Problem

A group is called periodic if all of its elements have ﬁnite order. We say a periodic group G has bounded exponent if every element has order at most m. In 1905 Burnside proved that ﬁnitely generated linear groups of bounded exponent are ﬁnite and in 1911 Schur proved that ﬁnitely generated periodic linear groups are ﬁnite. For a general ﬁnitely generated group, this is not true, counterexamples to the Burnside conjecture were constructed in a sequence of works by many authors, including Novikov, Golod, Shafarevich, and Ol’shanskii. We refer the reader to the Website: http : //www − groups.dcs.st − and.ac.uk/ history/HistTopics/Burnside_problem.html for a detailed discussion of the history. This page also discusses the restricted Burnside conjecture resolved by Zelmanov. In this context, the following questions seem natural:

question 13.2. Are there inﬁnite, ﬁnitely generated periodic groups of diffeomorphisms of a compact manifold M? Are there inﬁnite, ﬁnitely generated bounded exponent groups of diffeomorphisms of a compact manifold M? Some ﬁrst results in this direction are contained in the paper of Rebelo and Silva [206]. 13.3. Tits’s alternative

In this section, we ask a sequence of questions related to a famous theorem of Tits’ and more recent variants on it [229].

theorem 13.3. Let be a ﬁnitely generated linear group. Then either contains a free subgroup on two generators or is virtually solvable. The following conjecture of Ghys is a reasonable alternative to this for groups of diffeomorphisms.

144 / d. fisher

conjecture 13.4. Let M be a compact manifold and a ﬁnitely generated group of smooth diffeomorphisms of M. Then either contains a free group on two generators or preserves some measure on M. The best evidence for this conjecture to date is a theorem of Margulis’ that proves the conjecture for M = S1 [163]. Ghys has also asked if the more exact analogue of Tits’s theorem might be true for analytic diffeomorphisms. A closely related question is whether there exist groups of intermediate growth inside of diffeomorphism groups. The growth of a group is the rate of growth of the balls in the Cayley graph. For linear groups, this is either polynomial or exponential. In general, there are groups of intermediate growth, that is, subexponential but not polynomial. The ﬁrst examples of these were discovered by Grigorchuk [109]. Some of these examples were shown by Grigorchuk and Maki to act by homeomorphisms on the interval [110]. More recently, it was shown by Navas that these examples do have an action by C 1 diffeomorphisms on the interval [185]. In the opposite direction, Navas showed that these phenomena disappear if one considers instead C 1+α diffeomorphisms for any positive α. In that setting, Navas shows that any ﬁnitely generated group of diffeomorphisms of the interval with subexponential growth must be nilpotent. It would be interesting to understand what can occur in higher dimensions. A recent related line of research for linear groups concerns uniform exponential growth. A ﬁnitely generated group has uniform exponential growth if the number of elements in a ball of radius r in a Cayley graph for grows at least as λr for some λ > 1 that does not depend on the choice of generators. For linear groups, exponential growth implies uniform exponential growth by a theorem of Eskin, Mozes, and Oh [51]. There are examples of groups having exponential but not uniform exponential growth due to Wilson [235]. This raises the following question. question 13.5. Let M be a compact manifold and a ﬁnitely generated subgroup of Diff (M). If has exponential growth, does have uniform exponential growth? The question seems most likely to have a positive answer if one further assumes that is nonamenable.

References [1] [2]

Scot Adams, Dynamics on Lorentz manifolds, World Scientiﬁc Publishing Co., River Edge, NJ, 2001. MR1875067 (2003c:53101) Jurgen Andersen, Mapping class groups do not have Kazhdan’s property (T ), preprint 2007.

groups acting on manifolds / 145 [3]

[4] [5]

[6]

[7] [8]

[9]

[10]

[11]

[12] [13]

[14]

[15]

[16] [17] [18]

[19] [20] [21]

D. V. Anosov, Geodesic ﬂows on closed Riemann manifolds with negative curvature, Proceedings of the Steklov Institute of Mathematics, No. 90 (1967). Translated from the Russian by S. Feder, American Mathematical Society, Providence, RI, 1969. MR0242194 (39 #3527) Artur Avila, Distortion elements in Diff∞ (R/Z), preprint 2008. Uri Bader, Charles Frances, and Karin Melnick, An embedding theorem for automorphism groups of Cartan geometries, (English summary). Geom. Funct. Anal. 19 (2009), no. 2, 333–355. MR254240 Uri Bader, Alex Furman, Tsachik Gelander, and Nicolas Monod, Property (T) and rigidity for actions on Banach spaces, Acta Math. 198 (2007), no. 1, 57–105. MR2316269 (2008g:22007) Uri Bader and Yehuda Shalom, Factor and normal subgroup theorems for lattices in products of groups, Invent. Math. 163 (2006), no. 2, 415–454. MR2207022 (2006m:22017) Augustin Banyaga, Sur la structure du groupe des difféomorphismes qui préservent une forme symplectique, Comment. Math. Helv. 53 (1978), no. 2, 174–227. MR490874 (80c:58005) ———, The structure of classical diffeomorphism groups, Mathematics and its Applications, vol. 400, Kluwer Academic Publishers Group, Dordrecht, 1997. MR1445290 (98h:22024) Yves Benoist, Orbites des structures rigides (d’après M. Gromov), Integrable systems and foliations/Feuilletages et systèmes intégrables (Montpellier, 1995), Progr. Math., vol. 145, Birkhäuser, Boston, 1997, pp. 1–17. MR1432904 (98c:58126) Yves Benoist and François Labourie, Sur les difféomorphismes d’Anosov afﬁnes à feuilletages stable et instable différentiables, Invent. Math. 111 (1993), no. 2, 285–308. MR11 98811 (94d:58114) E. J. Benveniste, University of Chicago, Ph.D. thesis, unpublished, 1996. E. Jerome Benveniste and David Fisher, Nonexistence of invariant rigid structures and invariant almost rigid structures, Comm. Anal. Geom. 13 (2005), no. 1, 89–111. MR2154667 (2006f:53056) G. Besson, G. Courtois, and S. Gallot, Entropies et rigidités des espaces localement symétriques de courbure strictement négative, Geom. Funct. Anal. 5 (1995), no. 5, 731–799. MR1354289 (96i:58136) Gérard Besson, Gilles Courtois, and Sylvestre Gallot, Minimal entropy and Mostow’s rigidity theorems, Ergodic Theory Dynam. Systems 16 (1996), no. 4, 623–649. MR140 6425 (97e:58177) Mladen Bestvina, Local homology properties of boundaries of groups, Michigan Math. J. 43 (1996), no. 1, 123–139. MR1381603 (97a:57022) Mladen Bestvina and Koji Fujiwara, Bounded cohomology of subgroups of mapping class groups, Geom. Topol. 6 (2002), 69–89 (electronic). MR1914565 (2003f:57003) A. Borel and N. Wallach, Continuous cohomology, discrete subgroups, and representations of reductive groups, second ed., Mathematical Surveys and Monographs, Vol. 67, American Mathematical Society, Providence, RI, 2000. MR1721403 (2000j:22015) Armand Borel, Compact Clifford-Klein forms of symmetric spaces, Topology 2 (1963), 111–122. MR0146301 (26 #3823) Martin Bridson and Karen Vogtmann, Actions of automorphism groups of free groups on homology spheres and acyclic manifolds, to appear Comm. Math. Helv. Martin R. Bridson and Karen Vogtmann, Automorphism groups of free groups, surface groups and free Abelian groups, Problems on mapping class groups and related topics,

146 / d. fisher

[22] [23] [24]

[25]

[26] [27]

[28]

[29] [30]

[31] [32]

[33] [34] [35] [36] [37]

[38] [39]

[40]

Proc. Sympos. Pure Math., Vol. 74, American Mathematical Society, Providence, RI, 2006, pp. 301–316. MR2264548 (2008g:20091) M. Burger and N. Monod, Continuous bounded cohomology and applications to rigidity theory, Geom. Funct. Anal. 12 (2002), no. 2, 219–280. MR1911660 (2003d:53065a) Marc Burger, An extension criterion for lattice actions on the circle, in this volume. ———, Rigidity properties of group actions on CAT(0)-spaces, Proceedings of the International Congress of Mathematicians, Vol. 1, 2 (Zürich, 1994) (Basel), Birkhäuser, 1995, pp. 761–769. MR1403976 (97j:20033) Marc Burger and Shahar Mozes, Groups acting on trees: from local to global structure, Inst. Hautes Études Sci. Publ. Math. (2000), no. 92, 113–150 (2001). MR1839488 (2002i:20041) ———, Lattices in product of trees, Inst. Hautes Études Sci. Publ. Math. (2000), no. 92, 151–194 (2001). MR1839489 (2002i:20042) Keith Burns, Charles Pugh, Michael Shub, and Amie Wilkinson, Recent results about stable ergodicity, Smooth ergodic theory and its applications (Seattle, WA, 1999), Proc. Sympos. Pure Math., Vol. 69, American Mathematical Society, Providence, RI, 2001, pp. 327–366. MR1858538 (2002m:37042) Eugenio Calabi, On compact, Riemannian manifolds with constant curvature. I, Proc. Sympos. Pure Math., Vol. III, American Mathematical Society, Providence, RI, 1961, pp. 155–180. MR0133787 (24 #A3612) Eugenio Calabi and Edoardo Vesentini, On compact, locally symmetric Kähler manifolds, Ann. of Math. (2) 71 (1960), 472–507. MR0111058 (22 #1922b) Danny Calegari and Michael H. Freedman, Distortion in transformation groups, Geom. Topol. 10 (2006), 267–293 (electronic), with an appendix by Yves de Cornulier. MR220 7794 (2007b:37048) A. Candel and R. Quiroga-Barranco, Gromov’s centralizer theorem, Geom. Dedicata 100 (2003), 123–155. MR2011119 (2004m:53075) Alberto Candel and Raul Quiroga-Barranco, Parallelisms, prolongations of Lie algebras and rigid geometric structures, Manuscripta Math. 114 (2004), no. 3, 335–350. MR2076451 (2005f:53056) J. W. Cannon, W. J. Floyd, and W. R. Parry, Introductory notes on Richard Thompson’s groups, Enseign. Math. (2) 42 (1996), no. 3-4, 215–256. MR1426438 (98g:20058) James W. Cannon and William P. Thurston, Group invariant Peano curves, Geom. Topol. 11 (2007), 1315–1355. MR2326947 (2008i:57016) Serge Cantat, Groupes de transformations birationnelles du plan, preprint. ———, Version kählérienne d’une conjecture de Robert J. Zimmer, Ann. Sci. École Norm. Sup. (4) 37 (2004), no. 5, 759–768. MR2103473 (2006b:22010) Pierre-Emmanuel Caprace and Bertrand Rémy, Simplicité abstraite des groupes de KacMoody non afﬁnes, C. R. Math. Acad. Sci. Paris 342 (2006), no. 8, 539–544. MR2217912 (2006k:20102) Indira Chatterji and Martin Kassobov, New examples of ﬁnitely presented groups with strong ﬁxed point properties,J. Topol. Anal. 1 (2009), no. 1, 1–12. MR2649346 Christopher Connell and Benson Farb, Some recent applications of the barycenter method in geometry, Topology and geometry of manifolds (Athens, GA, 2001), Proc. Sympos. Pure Math., Vol. 71, American Mathematical Society, Providence, RI, 2003, pp. 19–50. MR2024628 (2005e:53058) Kevin Corlette, Archimedean superrigidity and hyperbolic geometry, Ann. of Math. (2) 135 (1992), no. 1, 165–182. MR1147961 (92m:57048)

groups acting on manifolds / 147 [41] [42] [43]

[44]

[45]

[46] [47]

[48] [49] [50] [51] [52] [53] [54] [55] [56] [57]

[58] [59] [60] [61] [62]

Kevin Corlette and Robert J. Zimmer, Superrigidity for cocycles and hyperbolic geometry, Internat. J. Math. 5 (1994), no. 3, 273–290. MR1274120 (95g:58055) S. G. Dani, Continuous equivariant images of lattice-actions on boundaries, Ann. of Math. (2) 119 (1984), no. 1, 111–119. MR736562 (85i:22009) Pierre de la Harpe and Alain Valette, La propriété (T ) de Kazhdan pour les groupes localement compacts (avec un appendice de Marc Burger), Astérisque (1989), no. 175, 158, with an appendix by M. Burger. MR1023471 (90m:22001) Patrick Delorme, 1-cohomologie des représentations unitaires des groupes de Lie semi-simples et résolubles. Produits tensoriels continus de représentations, Bull. Soc. Math. France 105 (1977), no. 3, 281–336. MR0578893 (58 #28272) Bertrand Deroin, Victor Kleptsyn, and Andrés Navas, Sur la dynamique unidimensionnelle en régularité intermédiaire, Acta Math. 199 (2007), no. 2, 199–262. MR23 58052 Julie Déserti, Groupe de Cremona et dynamique complexe: une approche de la conjecture de Zimmer, Int. Math. Res. Not. (2006), Art. ID 71701, 27. MR2233717 (2007d:22013) Tien-Cuong Dinh and Nessim Sibony, Groupes commutatifs d’automorphismes d’une variété kählérienne compacte, Duke Math. J. 123 (2004), no. 2, 311–328. MR2066940 (2005g:32020) Sorin Dumitrescu, Meromorphic almost rigid geometric structures, in this volume. Michael Entov and Leonid Polterovich, Calabi quasimorphism and quantum homology, Int. Math. Res. Not. (2003), no. 30, 1635–1676. MR1979584 (2004e:53131) D. B. A. Epstein, The simplicity of certain groups of homeomorphisms, Compositio Math. 22 (1970), 165–173. MR0267589 (42 #2491) Alex Eskin, Shahar Mozes, and Hee Oh, On uniform exponential growth for linear groups, Invent. Math. 160 (2005), no. 1, 1–30. MR2129706 (2006a:20081) Benson Farb, Group actions and Helly’s theorem,Adv. Math. 222 (2009), no. 5, 1574– 1588. MR2555905 Benson Farb and Howard Masur, Superrigidity and mapping class groups, Topology 37 (1998), no. 6, 1169–1176. MR1632912 (99f:57017) Benson Farb and Peter Shalen, Real-analytic actions of lattices, Invent. Math. 135 (1999), no. 2, 273–296. MR1666834 (2000c:22017) ———, Lattice actions, 3-manifolds and homology, Topology 39 (2000), no. 3, 573–587. MR1746910 (2001b:57041) ———, Groups of real-analytic diffeomorphisms of the circle, Ergodic Theory Dynam. Systems 22 (2002), no. 3, 835–844. MR1908556 (2003e:37030) Benson Farb and Peter B. Shalen, Real-analytic, volume-preserving actions of lattices on 4-manifolds, C. R. Math. Acad. Sci. Paris 334 (2002), no. 11, 1011–1014. MR1913726 (2003e:57055) Daniel S. Farley, Proper isometric actions of Thompson’s groups on Hilbert space, Int. Math. Res. Not. (2003), no. 45, 2409–2414. MR2006480 (2004k:22005) F. T. Farrell and J.-F. Lafont, EZ-structures and topological applications, Comment. Math. Helv. 80 (2005), no. 1, 103–121. MR2130569 (2006b:57022) R. Feres, Afﬁne actions of higher rank lattices, Geom. Funct. Anal. 3 (1993), no. 4, 370–394. MR1223436 (96d:22013) R. Feres and F. Labourie, Topological superrigidity and Anosov actions of lattices, Ann. Sci. École Norm. Sup. (4) 31 (1998), no. 5, 599–629. MR1643954 (99k:58112) R. Feres and P. Lampe, Cartan geometries and dynamics, Geom. Dedicata 80 (2000), no. 1–3, 29–41. MR1762497 (2001i:53059)

148 / d. fisher [63] [64] [65]

[66] [67]

[68]

[69] [70] [71] [72] [73] [74] [75]

[76] [77]

[78] [79] [80] [81]

[82] [83] [84]

Renato Feres, Connection preserving actions of lattices in SLn R, Israel J. Math. 79 (1992), no. 1, 1–21. MR1195250 (94a:58039) ———, Actions of discrete linear groups and Zimmer’s conjecture, J. Differential Geom. 42 (1995), no. 3, 554–576. MR1367402 (97a:22016) ———, Dynamical systems and semisimple groups: an introduction, Cambridge Tracts in Mathematics, Vol. 126, Cambridge University Press, Cambridge, 1998. MR1670703 (2001d:22001) ———, An introduction to cocycle super-rigidity, Rigidity in dynamics and geometry (Cambridge, 2000), Springer, Berlin, 2002, pp. 99–134. MR1919397 (2003g:37041) ———, Rigid geometric structures and actions of semisimple Lie groups, Rigidité, groupe fondamental et dynamique, Panor. Synthèses, Vol. 13, Soc. Math. France, Paris, 2002, pp. 121–167. MR1993149 (2004m:53076) Renato Feres and Anatole Katok, Ergodic theory and dynamics of G-spaces (with special emphasis on rigidity phenomena), Handbook of dynamical systems, Vol. 1A, NorthHolland, Amsterdam, 2002, pp. 665–763. MR1928526 (2003j:37005) Renato Feres and Emily Ronshausen, Harmonic functions over group actions, in this volume. D. Fisher and K. Whyte, When is a group action determined by its orbit structure? Geom. Funct. Anal. 13 (2003), no. 6, 1189–1200. MR2033836 (2004k:37045) David Fisher, Cohomology of arithmetic groups and bending group actions, in preparation. ———, First cohomology and local rigidity of group actions, in preparation. ———, A canonical arithmetic quotient for actions of lattices in simple groups, Israel J. Math. 124 (2001), 143–155. MR1856509 (2002f:22016) ———, On the arithmetic structure of lattice actions on compact spaces, Ergodic Theory Dynam. Systems 22 (2002), no. 4, 1141–1168. MR1926279 (2004j:37004) ———, Rigid geometric structures and representations of fundamental groups, Rigidity in dynamics and geometry (Cambridge, 2000), Springer, Berlin, 2002, pp. 135–147. MR1919398 (2003g:22007) ———, Out(Fn ) and the spectral gap conjecture, Int. Math. Res. Not. (2006), Art. ID 26028, 9. MR2250018 (2007f:22006) ———, Local rigidity of group actions: past, present, future, Dynamics, ergodic theory, and geometry, Math. Sci. Res. Inst. Publ., Vol. 54, Cambridge University Press, Cambridge, 2007, pp. 45–97. MR2369442 ———, Deformations of group actions, Trans. Amer. Math. Soc. 360 (2008), no. 1, 491–505 (electronic). MR2342012 David Fisher and Theron Hitchman, Harmonic maps with inﬁnite dimensional targets and cocycle superrigidity, in preparation. ———, Cocycle superrigidity and harmonic maps with inﬁnite-dimensional targets, Int. Math. Res. Not. (2006), 72405, 1–19. MR2211160 David Fisher and G. A. Margulis, Local rigidity for cocycles, Surveys in differential geometry, Vol. VIII (Boston, MA, 2002), Surv. Differ. Geom., VIII, International Press, Somerville, MA, 2003, pp. 191–234. MR2039990 (2004m:22032) David Fisher and Gregory Margulis, Almost isometric actions, property (T ), and local rigidity, Invent. Math. 162 (2005), no. 1, 19–80. MR2198325 David Fisher and Lior Silberman, Groups not acting on manifolds, Int. Math. Res. Not. (2008). David Fisher and Kevin Whyte, Continuous quotients for lattice actions on compact spaces, Geom. Dedicata 87 (2001), no. 1–3, 181–189. MR1866848 (2002j:57070)

groups acting on manifolds / 149 [85] [86] [87] [88] [89] [90]

[91] [92]

[93] [94]

[95]

[96]

[97] [98]

[99]

[100] [101] [102] [103] [104] [105]

David Fisher and Robert J. Zimmer, Geometric lattice actions, entropy and fundamental groups, Comment. Math. Helv. 77 (2002), no. 2, 326–338. MR1915044 (2003e:57056) John Franks and Michael Handel, Area preserving group actions on surfaces, Geom. Topol. 7 (2003), 757–771 (electronic). MR2026546 (2004j:37042) ———, Periodic points of Hamiltonian surface diffeomorphisms, Geom. Topol. 7 (2003), 713–756 (electronic). MR2026545 (2004j:37101) ———, Distortion elements in group actions on surfaces, Duke Math. J. 131 (2006), no. 3, 441–468. MR2219247 (2007c:37018) Alex Furman, A survey of measured group theory, in this volume. ———, Random walks on groups and random transformations, Handbook of dynamical systems, Vol. 1A, North-Holland, Amsterdam, 2002, pp. 931–1014. MR1928529 (2003j:60065) Alex Furman and Nicolas Monod, Product groups acting on manifolds, preprint 2007. H. Furstenberg, Recurrence in ergodic theory and combinatorial number theory, Princeton University Press, Princeton, NJ, 1981, M. B. Porter Lectures. MR603625 (82j: 28010) Harry Furstenberg, A Poisson formula for semi-simple Lie groups, Ann. of Math. (2) 77 (1963), 335–386. MR0146298 (26 #3820) ———, Boundaries of Lie groups and discrete subgroups, Actes du Congrès International des Mathématiciens (Nice, 1970), Tome 2, Gauthier-Villars, Paris, 1971, pp. 301–306. MR0430160 (55 #3167) ———, Rigidity and cocycles for ergodic actions of semisimple Lie groups (after G. A. Margulis and R. Zimmer), Bourbaki Seminar, Vol. 1979/80, Lecture Notes in Math., vol. 842, Springer, Berlin, 1981, pp. 273–292. MR636529 (83j:22003) Jean-Marc Gambaudo and Étienne Ghys, Commutators and diffeomorphisms of surfaces, Ergodic Theory Dynam. Systems 24 (2004), no. 5, 1591–1617. MR2104597 (2006d:37071) Tsachik Gelander, On deformtions of Fn in compact Lie groups, Israel J. Math. 167 (2008), 15–26. MR2448015 (2009j:22011) Tsachik Gelander, Anders Karlsson, and Gregory A. Margulis, Superrigidity, generalized harmonic maps and uniformly convex spaces, Geom. Funct. Anal. 17 (2008), no. 5, 1524– 1550. MR2377496 Étienne Ghys, Groupes d’homéomorphismes du cercle et cohomologie bornée, The Lefschetz centennial conference, Part III (Mexico City, 1984), Contemp. Math., vol. 58, American Mathematical Society, Providence, RI, 1987, pp. 81–106. MR893858 (88m:58024) ———, Sur les groupes engendrés par des difféomorphismes proches de l’identité, Bol. Soc. Brasil. Mat. (N.S.) 24 (1993), no. 2, 137–178. MR1254981 (95f:58017) ———, Actions de réseaux sur le cercle, Invent. Math. 137 (1999), no. 1, 199–231. MR1703323 (2000j:22014) ———, Groups acting on the circle, Enseign. Math. (2) 47 (2001), no. 3–4, 329–407. MR1876932 (2003a:37032) Étienne Ghys and Vlad Sergiescu, Sur un groupe remarquable de difféomorphismes du cercle, Comment. Math. Helv. 62 (1987), no. 2, 185–239. MR896095 (90c:57035) Edward R. Goetze, Connection preserving actions of connected and discrete Lie groups, J. Differential Geom. 40 (1994), no. 3, 595–620. MR1305982 (95m:57052) Edward R. Goetze and Ralf J. Spatzier, On Livˇsic’s theorem, superrigidity, and Anosov actions of semisimple Lie groups, Duke Math. J. 88 (1997), no. 1, 1–27. MR1448015 (98d:58134)

150 / d. fisher [106] ———, Smooth classiﬁcation of Cartan actions of higher rank semisimple Lie groups and their lattices, Ann. of Math. (2) 150 (1999), no. 3, 743–773. MR1740993 (2001c:37029) [107] William M. Goldman, An ergodic action of the outer automorphism group of a free group, Geom. Funct. Anal. 17 (2007), no. 3, 793–805. MR2346275 (2008g:57001) [108] Alexander Gorodnik, Open problems in dynamics and related ﬁelds, J. Mod. Dyn. 1 (2007), no. 1, 1–35. MR2261070 (2007f:37001) [109] R. I. Grigorchuk, Degrees of growth of ﬁnitely generated groups and the theory of invariant means, Izv. Akad. Nauk SSSR Ser. Mat. 48 (1984), no. 5, 939–985. MR764305 (86h:20041) [110] R. I. Grigorchuk and A. Maki, On a group of intermediate growth that acts on a line by homeomorphisms, Mat. Zametki 53 (1993), no. 2, 46–63. MR1220809 (94c:20008) [111] M. Gromov, Hyperbolic groups, Essays in group theory, Math. Sci. Res. Inst. Publ., Vol. 8, Springer, New York, 1987, pp. 75–263. MR919829 (89e:20070) [112] Michael Gromov, Rigid transformations groups, Géométrie différentielle (Paris, 1986), Travaux en Cours, Vol. 33, Hermann, Paris, 1988, pp. 65–139. MR955852 (90d:58173) [113] Mikhail Gromov and Richard Schoen, Harmonic maps into singular spaces and p-adic superrigidity for lattices in groups of rank one, Inst. Hautes Études Sci. Publ. Math. (1992), no. 76, 165–246. MR1215595 (94e:58032) [114] Alain Guichardet, Sur la cohomologie des groupes topologiques. II, Bull. Sci. Math. (2) 96 (1972), 305–332. MR0340464 (49 #5219) [115] Sigurdur Helgason, Differential geometry, Lie groups, and symmetric spaces, Graduate Studies in Mathematics, Vol. 34, American Mathematical Society, Providence, RI, 2001, Corrected reprint of the 1978 original. MR1834454 (2002b:53081) [116] Michael-Robert Herman, Simplicité du groupe des difféomorphismes de classe C ∞ , isotopes à l’identité, du tore de dimension n, C. R. Acad. Sci. Paris Sér. A-B 273 (1971), A232–A234. MR0287585 (44 #4788) [117] Graham Higman, Finitely presented inﬁnite simple groups, Department of Pure Mathematics, Department of Mathematics, I.A.S. Australian National University, Canberra, 1974, Notes on Pure Mathematics, no. 8 (1974). MR0376874 (51 #13049) [118] M. W. Hirsch, C. C. Pugh, and M. Shub, Invariant manifolds, Lecture Notes in Mathematics, Vol. 583, Springer-Verlag, Berlin, 1977. MR0501173 (58 #18595) [119] Steve Hurder, personal communication 2007. [120] Steven Hurder, A survey of rigidity theory for Anosov actions, Differential topology, foliations, and group actions (Rio de Janeiro, 1992), Contemp. Math., Vol. 161, American Mathematical Society, Providence, RI, 1994, pp. 143–173. MR1271833 (95b: 58112) [121] Dennis Johnson and John J. Millson, Deformation spaces associated to compact hyperbolic manifolds, Discrete groups in geometry and analysis (New Haven, CT, 1984), Progr. Math., Vol. 67, Birkhäuser, Boston, 1987, pp. 48–106. MR900823 (88j:22010) [122] Jürgen Jost and Shing-Tung Yau, Harmonic maps and superrigidity, Differential geometry: partial differential equations on manifolds (Los Angeles, CA, 1990), Proc. Sympos. Pure Math., Vol. 54, American Mathematical Society, Providence, RI, 1993, pp. 245– 280. MR1216587 (94m:58060) [123] Jürgen Jost and Kang Zuo, Harmonic maps of inﬁnite energy and rigidity results for representations of fundamental groups of quasiprojective varieties, J. Differential Geom. 47 (1997), no. 3, 469–503. MR1617644 (99a:58046) [124] Vadim A. Kaimanovich and Howard Masur, The Poisson boundary of the mapping class group, Invent. Math. 125 (1996), no. 2, 221–264. MR1395719 (97m:32033)

groups acting on manifolds / 151 [125] Boris Kalinin, Anatole Katok, and Federico Rodriguez-Hertz, Non-uniform measure rigidity, preprint 2009. [126] Boris Kalinin and Ralf Spatzier, On the classiﬁcation of Cartan actions, Geom. Funct. Anal. 17 (2007), no. 2, 468–490. MR2322492 (2008i:37054) [127] Martin Kassabov, personal communication, 2008. [128] A. Katok and J. Lewis, Global rigidity results for lattice actions on tori and new examples of volume-preserving actions, Israel J. Math. 93 (1996), 253–280. MR1380646 (96k: 22021) [129] A. Katok, J. Lewis, and R. Zimmer, Cocycle superrigidity and rigidity for lattice actions on tori, Topology 35 (1996), no. 1, 27–38. MR1367273 (97e:22009) [130] A. Katok and R. J. Spatzier, Differential rigidity of Anosov actions of higher rank Abelian groups and algebraic lattice actions, Tr. Mat. Inst. Steklova 216 (1997), no. Din. Sist. i Smezhnye Vopr., 292–319. MR1632177 (99i:58118) [131] D. A. Každan, On the connection of the dual space of a group with the structure of its closed subgroups, Funkcional. Anal. i Priložen. 1 (1967), 71–74. MR0209390 (35 #288) [132] David Kazhdan, Some applications of the Weil representation, J. Analyse Mat. 32 (1977), 235–248. MR0492089 (58 #11243) [133] Bruce Kleiner, The asymptotic geometry of negatively curved spaces: uniformization, geometrization and rigidity, International Congress of Mathematicians. Vol. II, European Mathematical Society, Zürich, 2006, pp. 743–768. MR2275621 (2007k:53054) [134] Anthony W. Knapp, Lie groups beyond an introduction, Progress in Mathematics, vol. 140, Birkhäuser, Boston, 1996. MR1399083 (98b:22002) [135] Shoshichi Kobayashi, Transformation groups in differential geometry, Classics in Mathematics, Springer-Verlag, Berlin, 1995, Reprint of the 1972 edition. MR1336823 (96c:53040) [136] Toshiyuki Kobayashi, Discontinuous groups for non-Riemannian homogeneous spaces, Mathematics unlimited—2001 and beyond, Springer, Berlin, 2001, pp. 723–747. MR1852186 (2002f:53086) [137] Toshiyuki Kobayashi and Taro Yoshino, Compact Clifford-Klein forms of symmetric spaces—revisited, Pure Appl. Math. Q. 1 (2005), no. 3, part 2, 591–663. MR2201328 (2007h:22013) [138] Nicholas Korevaar and Richard Schoen, Global existence theorems for harmonic maps: ﬁnite rank spaces and approaches to rigidity of group actions, preprint 1999. [139] Nicholas J. Korevaar and Richard M. Schoen, Sobolev spaces and harmonic maps for metric space targets, Comm. Anal. Geom. 1 (1993), no. 3–4, 561–659. MR1266480 (95b:58043) [140] ———, Global existence theorems for harmonic maps to non-locally compact spaces, Comm. Anal. Geom. 5 (1997), no. 2, 333–387. MR1483983 (99b:58061) [141] Nadine Kowalsky, Noncompact simple automorphism groups of Lorentz manifolds and other geometric manifolds, Ann. of Math. (2) 144 (1996), no. 3, 611–640. MR1426887 (98g:57059) [142] F. Labourie, Quelques résultats récents sur les espaces localement homogènes compacts, Manifolds and geometry (Pisa, 1993), Sympos. Math., XXXVI, Cambridge University Press, Cambridge, 1996, pp. 267–283. MR1410076 (97h:53055) [143] F. Labourie, S. Mozes, and R. J. Zimmer, On manifolds locally modelled on nonRiemannian homogeneous spaces, Geom. Funct. Anal. 5 (1995), no. 6, 955–965. MR136 1517 (97j:53053)

152 / d. fisher [144] François Labourie, Large groups actions on manifolds, Proceedings of the International Congress of Mathematicians, Vol. II (Berlin, 1998), no. Extra Vol. II, 1998, pp. 371–380 (electronic). MR1648087 (99k:53069) [145] François Labourie and Robert J. Zimmer, On the non-existence of cocompact lattices for SL(n)/SL(m), Math. Res. Lett. 2 (1995), no. 1, 75–77. MR1312978 (96d:22014) [146] Lucy Lifschitz and Dave Witte Morris, Bounded generation and lattices that cannot act on the line, Pure Appl. Math. Q. 4 (2008), no. 1, part 2, 99–126. MR2405997 (2009m: 22106) [147] Elon Lindenstrauss, Rigidity of multiparameter actions, Israel J. Math. 149 (2005), 199– 226, Probability in mathematics. MR2191215 (2006j:37007) [148] Ron Livne, On certain covers of the universal elliptic curve, unpublished Harvard PhD thesis. [149] Alexander Lubotzky, Discrete groups, expanding graphs and invariant measures, Progress in Mathematics, Vol. 125, Birkhäuser Verlag, Basel, 1994, with an appendix by Jonathan D. Rogawski. MR1308046 (96g:22018) [150] ———, Free quotients and the ﬁrst Betti number of some hyperbolic manifolds, Transform. Groups 1 (1996), no. 1–2, 71–82. MR1390750 (97d:57016) [151] Alexander Lubotzky, Shahar Mozes, and M. S. Raghunathan, The word and Riemannian metrics on lattices of semisimple groups, Inst. Hautes Études Sci. Publ. Math. (2000), no. 91, 5–53 (2001). MR1828742 (2002e:22011) [152] Alexander Lubotzky and Robert J. Zimmer, A canonical arithmetic quotient for simple Lie group actions, Lie groups and ergodic theory (Mumbai, 1996), Tata Inst. Fund. Res. Stud. Math., Vol. 14, Tata Institute of Fundamental Research, Bombay, 1998, pp. 131–142. MR1699362 (2000m:22013) [153] ———, Arithmetic structure of fundamental groups and actions of semisimple Lie groups, Topology 40 (2001), no. 4, 851–869. MR1851566 (2002f:22017) [154] George W. Mackey, Ergodic theory and virtual groups, Math. Ann. 166 (1966), 187–207. MR0201562 (34 #1444) ˘ [155] Iozhe Maleshich, The Hilbert-Smith conjecture for Hölder actions, Uspekhi Mat. Nauk 52 (1997), no. 2(314), 173–174. MR1480156 (99d:57026) [156] Ricardo Mañé, Quasi-Anosov diffeomorphisms and hyperbolic manifolds, Trans. Amer. Math. Soc. 229 (1977), 351–370. MR0482849 (58 #2894) [157] L. N. Mann and J. C. Su, Actions of elementary p-groups on manifolds, Trans. Amer. Math. Soc. 106 (1963), 115–126. MR0143840 (26 #1390) [158] G. A. Margulis, Arithmetic properties of discrete subgroups, Uspehi Mat. Nauk 29 (1974), no. 1 (175), 49–98. MR0463353 (57 #3306a) [159] ———, Discrete groups of motions of manifolds of nonpositive curvature, Proceedings of the International Congress of Mathematicians (Vancouver, BC, 1974), Vol. 2, Canadian Mathematical Congress, Montreal, Quetec, 1975, pp. 21–34. MR0492072 (58 #11226) [160] ———, Non-uniform lattices in semisimple algebraic groups, Lie groups and their representations (Proc. Summer School on Group Representations of the Bolyai János Math. Soc., Budapest, 1971), Halsted, New York, 1975, pp. 371–553. MR0422499 (54 #10486) [161] ———, Arithmeticity of the irreducible lattices in the semisimple groups of rank greater than 1, Invent. Math. 76 (1984), no. 1, 93–120. MR739627 (85j:22021) [162] ———, Discrete subgroups of semisimple Lie groups, Ergebnisse der Mathematik und ihrer Grenzgebiete (3) [Results in Mathematics and Related Areas (3)], Vol. 17, Springer-Verlag, Berlin, 1991. MR1090825 (92h:22021)

groups acting on manifolds / 153 [163] Gregory Margulis, Free subgroups of the homeomorphism group of the circle, C. R. Acad. Sci. Paris Sér. I Math. 331 (2000), no. 9, 669–674. MR1797749 (2002b:37034) [164] ———, Problems and conjectures in rigidity theory, Mathematics: frontiers and perspectives, American Mathematical Society, Providence, RI, 2000, pp. 161–174. MR1754775 (2001d:22008) [165] Gregory A. Margulis and Nantian Qian, Rigidity of weakly hyperbolic actions of higher real rank semisimple Lie groups and their lattices, Ergodic Theory Dynam. Systems 21 (2001), no. 1, 121–164. MR1826664 (2003a:22019) [166] John N. Mather, Commutators of diffeomorphisms, Comment. Math. Helv. 49 (1974), 512–528. MR0356129 (50 #8600) [167] ———, Simplicity of certain groups of diffeomorphisms, Bull. Amer. Math. Soc. 80 (1974), 271–273. MR0339268 (49 #4028) [168] ———, Commutators of diffeomorphisms. II, Comment. Math. Helv. 50 (1975), 33–40. MR0375382 (51 #11576) [169] ———, Foliations and local homology of groups of diffeomorphisms, Proceedings of the International Congress of Mathematicians (Vancouver, BC, 1974), Vol. 2, Canadian Mathematical Congress, Montreal, Quebec, 1975, pp. 35–37. MR0431203 (55 #4205) [170] Yozô Matsushima and Shingo Murakami, On vector bundle valued harmonic forms and automorphic forms on symmetric riemannian manifolds, Ann. of Math. (2) 78 (1963), 365–416. MR0153028 (27 #2997) [171] Ngaiming Mok, Yum Tong Siu, and Sai-Kee Yeung, Geometric superrigidity, Invent. Math. 113 (1993), no. 1, 57–83. MR1223224 (94h:53079) [172] Nicolas Monod, Superrigidity for irreducible lattices and geometric splitting, J. Amer. Math. Soc. 19 (2006), no. 4, 781–814 (electronic). MR2219304 (2007b:22025) [173] Dave Witte Morris, Can lattices in SL(n, R) act on the circle? in this volume. [174] ———, Amenable groups that act on the line, Algebr. Geom. Topol. 6 (2006), 2509–2518. MR2286034 (2008c:20078) [175] ———, An introduction to arithmetic groups, book in preprint form 2007. [176] Dave Witte Morris and Robert J. Zimmer, Ergodic actions of semisimple Lie groups on compact principal bundles, Geom. Dedicata 106 (2004), 11–27. MR2079831 (2005e: 22016) [177] G. D. Mostow, Quasi-conformal mappings in n-space and the rigidity of hyperbolic space forms, Inst. Hautes Études Sci. Publ. Math. (1968), no. 34, 53–104. MR0236383 (38 #4679) [178] ———, Strong rigidity of locally symmetric spaces, Princeton University Press, Princeton, NJ, 1973, Annals of Mathematics Studies, No. 78. MR0385004 (52 #5874) [179] ———, personal communication, 2000. [180] I. Mundet Rieri, Jordan’s theorem for the diffeomorphism group of some manifolds, preprint 2008. [181] Andrés Navas, Groups of circle diffeomorphisms, book in preprint form, 2008. [182] ———, On the dynamics of (left) orderable groups, preprint 2007. [183] ———, Actions de groupes de Kazhdan sur le cercle, Ann. Sci. École Norm. Sup. (4) 35 (2002), no. 5, 749–758. MR1951442 (2003j:58013) [184] ———, Quelques nouveaux phénomènes de rang 1 pour les groupes de difféomorphismes du cercle, Comment. Math. Helv. 80 (2005), no. 2, 355–375. MR2142246 (2006j:57003) [185] ———, Growth of groups and diffeomorphisms of the interval, Geom. Funct. Anal. 18 (2008), no. 3, 988–1028. MR2439001

154 / d. fisher [186] Amos Nevo and Robert J. Zimmer, Invariant rigid geometric structures and smooth projective factors, Geom. Funct. Anal. 19 (2009), no. 2, 520–535. MR2545248 (2010h: 22029) [187] ———, Homogenous projective factors for actions of semi-simple Lie groups, Invent. Math. 138 (1999), no. 2, 229–252. MR1720183 (2000h:22006) [188] ———, Rigidity of Furstenberg entropy for semisimple Lie group actions, Ann. Sci. École Norm. Sup. (4) 33 (2000), no. 3, 321–343. MR1775184 (2001k:22009) [189] ———, Actions of semisimple Lie groups with stationary measure, Rigidity in dynamics and geometry (Cambridge, 2000), Springer, Berlin, 2002, pp. 321–343. MR1919409 (2003j:22029) [190] ———, A generalization of the intermediate factors theorem, J. Anal. Math. 86 (2002), 93–104. MR1894478 (2003f:22019) [191] ———, A structure theorem for actions of semisimple Lie groups, Ann. of Math. (2) 156 (2002), no. 2, 565–594. MR1933077 (2003i:22024) [192] ———, Entropy of stationary measures and bounded tangential de-Rham cohomology of semisimple Lie group actions, Geom. Dedicata 115 (2005), 181–199. MR2180047 (2006k: 22008) [193] Yann Ollivier, A January 2005 invitation to random groups, Ensaios Matemáticos [Mathematical Surveys], Vol. 10, Sociedade Brasileira de Matemática, Rio de Janeiro, 2005. MR2205306 [194] Vladimir Platonov and Andrei Rapinchuk, Algebraic groups and number theory, Pure and Applied Mathematics, Vol. 139, Academic Press, Boston, 1994, Translated from the 1991 Russian original by Rachel Rowen. MR1278263 (95b:11039) [195] Leonid Polterovich, The geometry of the group of symplectic diffeomorphisms, Lectures in Mathematics ETH Zürich, Birkhäuser Verlag, Basel, 2001. MR1826128 (2002g: 53157) [196] ———, Growth of maps, distortion in groups and symplectic geometry, Invent. Math. 150 (2002), no. 3, 655–686. MR1946555 (2003i:53126) [197] Sorin Popa, Deformation and rigidity for group actions and von Neumann algebras, International Congress of Mathematicians. Vol. I, European Mathematical Society, Zürich, 2007, pp. 445–477. MR2334200 (2008k:46186) [198] Pierre Py, Some remarks on area-preserving actions of lattices, in this volume. [199] ———, Quasi-morphismes de Calabi et graphe de Reeb sur le tore, C. R. Math. Acad. Sci. Paris 343 (2006), no. 5, 323–328. MR2253051 (2007e:53116) [200] ———, Quasi-morphismes et invariant de Calabi, Ann. Sci. École Norm. Sup. (4) 39 (2006), no. 1, 177–195. MR2224660 (2007f:53116) [201] R. Quiroga-Barranco and A. Candel, Rigid and ﬁnite type geometric structures, Geom. Dedicata 106 (2004), 123–143. MR2079838 (2005d:53068) [202] M. Raghunathan, Diskretnye podgruppy grupp Li, Izdat. “Mir,” Moscow, 1977, Translated from the English by O. V. Švarcman, Edited by È. B. Vinberg, with a supplement “Arithmeticity of irreducible lattices in semisimple groups of rank greater than 1” by G. A. Margulis. MR0507236 (58 #22394b) [203] M. S. Raghunathan, On the ﬁrst cohomology of discrete subgroups of semisimple Lie groups, Amer. J. Math. 87 (1965), 103–139. MR0173730 (30 #3940) [204] Marina Ratner, On Raghunathan’s measure conjecture, Ann. of Math. (2) 134 (1991), no. 3, 545–607. MR1135878 (93a:22009) [205] Julio C. Rebelo, On nilpotent groups of real analytic diffeomorphisms of the torus, C. R. Acad. Sci. Paris Sér. I Math. 331 (2000), no. 4, 317–322. MR1787192 (2001m:22041)

groups acting on manifolds / 155 [206] Julio C. Rebelo and Ana L. Silva, On the Burnside problem in Diff (M), Discrete Contin. Dyn. Syst. 17 (2007), no. 2, 423–439. MR2257443 (2007j:53107) ˘ cepin, A proof of the Hilbert-Smith conjecture for actions by [207] Du˘san Repov˘s and Evgenij S˘ Lipschitz maps, Math. Ann. 308 (1997), no. 2, 361–364. MR1464908 (99c:57066) [208] Derek J. S. Robinson, A course in the theory of groups, second ed., Graduate Texts in Mathematics, Vol. 80, Springer-Verlag, New York, 1996. MR1357169 (96f:20001) [209] Federico Rodriguez Hertz, Global rigidity of certain Abelian actions by toral automorphisms, J. Mod. Dyn. 1 (2007), no. 3, 425–442. MR2318497 (2008f:37063) [210] Leslie Saper, Tilings and ﬁnite energy retractions of locally symmetric spaces, Comment. Math. Helv. 72 (1997), no. 2, 167–202. MR1470087 (99a:22019) [211] B. Schmidt, Weakly hyperbolic actions of Kazhdan groups on tori, Geom. Funct. Anal. 16 (2006), no. 5, 1139–1156. MR2276535 (2008b:37045) [212] Atle Selberg, On discontinuous groups in higher-dimensional symmetric spaces, Contributions to function theory (Internat. Colloq. Function Theory, Bombay, 1960), Tata Institute of Fundamental Research, Bombay, 1960, pp. 147–164. MR0130324 (24 #A188) [213] Nimish A. Shah, Invariant measures and orbit closures on homogeneous spaces for actions of subgroups generated by unipotent elements, Lie groups and ergodic theory (Mumbai, 1996), Tata Inst. Fund. Res. Stud. Math., Vol. 14, Tata Institute of Fundamental Research, Bombay, 1998, pp. 229–271. MR1699367 (2001a:22012) [214] Yehuda Shalom, Rigidity, unitary representations of semisimple groups, and fundamental groups of manifolds with rank one transformation group, Ann. of Math. (2) 152 (2000), no. 1, 113–182. MR1792293 (2001m:22022) [215] ———, Measurable group theory, European Congress of Mathematics, European Mathematical Society, Zürich, 2005, pp. 391–423. MR2185757 (2006k:37007) [216] ———, The algebraization of Kazhdan’s property (T), International Congress of Mathematicians. Vol. II, European Mathematical Society, Zürich, 2006, pp. 1283–1310. MR2275645 (2008a:22003) [217] R. W. Sharpe, Differential geometry, Graduate Texts in Mathematics, Vol. 166, SpringerVerlag, New York, 1997, Cartan’s generalization of Klein’s Erlangen program, with a foreword by S. S. Chern. MR1453120 (98m:53033) [218] Daniel Shenﬁeld, Semisimple representations of SL(n, Z[X1 , . . . , Xk ]), to appear in Groups, Geometry and Dynamics. [219] Y. T. Siu, Remarks in talk at Margulis’ 60th birthday conference, February, 2006. [220] Yum Tong Siu, The complex-analyticity of harmonic maps and the strong rigidity of compact Kähler manifolds, Ann. of Math. (2) 112 (1980), no. 1, 73–111. MR584075 (81j:53061) [221] S. Smale, Differentiable dynamical systems, Bull. Amer. Math. Soc. 73 (1967), 747–817. MR0228014 (37 #3598) [222] R. J. Spatzier, Harmonic analysis in rigidity theory, Ergodic theory and its connections with harmonic analysis (Alexandria, 1993), London Math. Soc. Lecture Note Ser., Vol. 205, Cambridge University Press, Cambridge, 1995, pp. 153–205. MR1325698 (96c:22019) [223] ———, An invitation to rigidity theory, Modern dynamical systems and applications, Cambridge University Press, Cambridge, 2004, pp. 211–231. MR2090772 (2006a:53041) [224] Ralf J. Spatzier and Robert J. Zimmer, Fundamental groups of negatively curved manifolds and actions of semisimple groups, Topology 30 (1991), no. 4, 591–601. MR1133874 (92m:57047)

156 / d. fisher [225] Melanie Stein, Groups of piecewise linear homeomorphisms, Trans. Amer. Math. Soc. 332 (1992), no. 2, 477–514. MR1094555 (92k:20075) [226] Garrett Stuck, Minimal actions of semisimple groups, Ergodic Theory Dynam. Systems 16 (1996), no. 4, 821–831. MR1406436 (98a:57046) [227] William Thurston, Foliations and groups of diffeomorphisms, Bull. Amer. Math. Soc. 80 (1974), 304–307. MR0339267 (49 #4027) [228] William P. Thurston, A generalization of the Reeb stability theorem, Topology 13 (1974), 347–352. MR0356087 (50 #8558) [229] J. Tits, Free subgroups in linear groups, J. Algebra 20 (1972), 250–270. MR0286898 (44 #4105) [230] Alain Valette, Group pairs with property (T), from arithmetic lattices, Geom. Dedicata 112 (2005), 183–196. MR2163898 (2006d:22014) [231] André Weil, On discrete subgroups of Lie groups, Ann. of Math. (2) 72 (1960), 369–384. MR0137792 (25 #1241) [232] ———, On discrete subgroups of Lie groups. II, Ann. of Math. (2) 75 (1962), 578–602. MR0137793 (25 #1242) [233] ———, Remarks on the cohomology of groups, Ann. of Math. (2) 80 (1964), 149–157. MR0169956 (30 #199) [234] Shmuel Weinberger, Some remarks inspired by the C 0 Zimmer program, in this volume. [235] John S. Wilson, On exponential growth and uniformly exponential growth for groups, Invent. Math. 155 (2004), no. 2, 287–303. MR2031429 (2004k:20085) [236] Dave Witte, Arithmetic groups of higher Q-rank cannot act on 1-manifolds, Proc. Amer. Math. Soc. 122 (1994), no. 2, 333–340. MR1198459 (95a:22014) [237] ———, Measurable quotients of unipotent translations on homogeneous spaces, Trans. Amer. Math. Soc. 345 (1994), no. 2, 577–594. MR1181187 (95a:22005) [238] ———, Cocycle superrigidity for ergodic actions of non-semisimple Lie groups, Lie groups and ergodic theory (Mumbai, 1996), Tata Inst. Fund. Res. Stud. Math., Vol. 14, Tata Institute of Fundamental Research, Bombay, 1998, pp. 367–386. MR1699372 (2000i: 22008) [239] Sai-Kee Yeung, Representations of semisimple lattices in mapping class groups, Int. Math. Res. Not. (2003), no. 31, 1677–1686. MR1981481 (2004d:53047) [240] A. Zeghib, Isometry groups and geodesic foliations of Lorentz manifolds. I. Foundations of Lorentz dynamics, Geom. Funct. Anal. 9 (1999), no. 4, 775–822. MR1719606 (2001g: 53059) [241] ———, Isometry groups and geodesic foliations of Lorentz manifolds. II. Geometry of analytic Lorentz manifolds with large isometry groups, Geom. Funct. Anal. 9 (1999), no. 4, 823–854. MR1719610 (2001g:53060) [242] Abdelghani Zeghib, Le groupe afﬁne d’une variété riemannienne compacte, Comm. Anal. Geom. 5 (1997), no. 1, 199–211. MR1456311 (98g:53065) [243] ———, Sur les actions afﬁnes des groupes discrets, Ann. Inst. Fourier (Grenoble) 47 (1997), no. 2, 641–685. MR1450429 (98d:57068) [244] Robert J. Zimmer, Topological superrigidity, unpublished notes 1998. [245] ———, Ergodic theory and semisimple groups, Monographs in Mathematics, Vol. 81, Birkhäuser Verlag, Basel, 1984. MR776417 (86j:22014) [246] ———, Actions of lattices in semisimple groups preserving a G-structure of ﬁnite type, Ergodic Theory Dynam. Systems 5 (1985), no. 2, 301–306. MR796757 (87g:22011) [247] ———, Lattices in semisimple groups and distal geometric structures, Invent. Math. 80 (1985), no. 1, 123–137. MR784532 (86i:57056)

groups acting on manifolds / 157 [248] ———, On connection-preserving actions of discrete linear groups, Ergodic Theory Dynam. Systems 6 (1986), no. 4, 639–644. MR873437 (88g:57045) [249] ———, Actions of semisimple groups and discrete subgroups, Proceedings of the International Congress of Mathematicians, Vol. 1, 2 (Berkeley, CA, 1986), American Mathematical Society, Providence, RI, 1987, pp. 1247–1258. MR934329 (89j:22024) [250] ———, Lattices in semisimple groups and invariant geometric structures on compact manifolds, Discrete groups in geometry and analysis (New Haven, CT, 1984), Progr. Math., Vol. 67, Birkhäuser, Boston, 1987, pp. 152–210. MR900826 (88i:22025) [251] ———, Representations of fundamental groups of manifolds with a semisimple transformation group, J. Amer. Math. Soc. 2 (1989), no. 2, 201–213. MR973308 (90i:22021) [252] ———, On the algebraic hull of an automorphism group of a principal bundle, Comment. Math. Helv. 65 (1990), no. 3, 375–387. MR1069815 (92f:57050) [253] ———, Spectrum, entropy, and geometric structures for smooth actions of Kazhdan groups, Israel J. Math. 75 (1991), no. 1, 65–80. MR1147291 (93i:22014) [254] ———, Superrigidity, Ratner’s theorem, and fundamental groups, Israel J. Math. 74 (1991), no. 2–3, 199–207. MR1135234 (93b:22019) [255] ———, Automorphism groups and fundamental groups of geometric manifolds, Differential geometry: Riemannian geometry (Los Angeles, CA, 1990), Proc. Sympos. Pure Math., Vol. 54, American Mathematical Society, Providence, RI, 1993, pp. 693–710. MR1216656 (95a:58017) [256] ———, Discrete groups and non-Riemannian homogeneous spaces, J. Amer. Math. Soc. 7 (1994), no. 1, 159–168. MR1207014 (94e:22021) [257] ———, Entropy and arithmetic quotients for simple automorphism groups of geometric manifolds, Geom. Dedicata 107 (2004), 47–56. MR2110753 (2005i:22011) [258] Robert J. Zimmer and Dave Witte Morris, Ergodic theory, groups, and geometry, CBMS Regional Conference Series in Mathematics, 109. Published for the Conference Board of the Mathematical Sciences, Washington, DC, by the American Mathematical Society, Providence, RI, 2008. MR2457556 (2009k:37059)

5

dave witte morris

CAN LATTICES IN SL (n, R) ACT ON THE CIRCLE? to robert j. zimmer on his

60 th

birthday

1. Overview The following theorem is a simple example of the Zimmer program’s principle that large groups should not be able to act on small manifolds. (For a description of the Zimmer program, see D. Fisher’s survey article [Fis] elsewhere in this volume.) Unless stated otherwise, we assume actions are continuous, but we require no additional regularity.

theorem 1.1. (witte) Let be a ﬁnite-index subgroup of SL (n, Z), with n ≥ 3. Then has no faithful action on the circle S1 . remark 1.2. 1) The proof of this theorem is not at all difﬁcult, and will be given in Section 3, after the result is translated to a more algebraic form in Section 2. The proof illustrates the use of calculations with unipotent elements, which is a standard technique in the theory of arithmetic groups. 2) The assumption that n ≥ 3 is essential. Indeed, some ﬁnite-index subgroup of SL (2, Z) is a free group, which has countless actions on the circle (some faithful and some not). A group as in Theorem 1.1 is a lattice in SL (n, R). It is an open question whether the theorem generalizes to the other lattices:

conjecture 1.3. Let be a lattice in SL (n, R), with n ≥ 3. Then has no faithful action on S1 . Note that if this conjecture is true, then every action of on S1 or R has a nontrivial kernel. Since the kernel is a normal subgroup, and the Margulis 158

can lattices in sl (n, R) act on the circle? / 159

normal subgroup theorem tells us that lattices in SL (n, R) are “almost simple” (if n ≥ 3), it follows that the kernel is a ﬁnite-index subgroup of . Therefore, every -orbit is ﬁnite. Such reasoning shows that the conjecture can be restated as follows:

conjecture 1.3 . Let be a lattice in SL (n, R), with n ≥ 3. Whenever acts on S1 , every orbit is ﬁnite. There is considerable evidence for the above conjectures, including the following important theorem:

theorem 1.4. (ghys, burger-monod) If is any lattice in SL (n, R), with n ≥ 3, then every action of on S1 has at least one ﬁnite orbit. remark 1.5. The proof of É. Ghys utilizes amenability (or the Furstenberg boundary); see Sections 6 and 7. The proof of M. Burger and N. Monod is based on a vanishing theorem for bounded cohomology; see Section 8. The above conjectures are about continuous actions, but they can be weakened by considering only differentiable actions. It will be explained in Section 4 that combining the Ghys-Burger-Monod theorem with the well-known Reeb-Thurston stability theorem establishes these weaker conjectures:

corollary 1.6. (ghys, burger-monod) If is any lattice in SL (n, R), with n ≥ 3, then: 1) has no faithful C 1 action on S1 , and 2) whenever acts on S1 by C 1 diffeomorphisms, every orbit is ﬁnite. For n ≥ 3, it is well known that every lattice in SL (n, R) has Kazhdan’s property (T ) (see Deﬁnition 5.2), and the following result shows that if we strengthen the differentiability hypothesis slightly, then the conclusion is true for all groups that have that property. This is a very broad class of groups (including many groups that are not even linear), so one would expect a much stronger result to hold for the special case of lattices in SL (n, R). Thus, this theorem constitutes good evidence for Conjecture 1.3. The proof will be presented in Section 5; it is both elegant and elementary.

theorem 1.7. (navas) If is any inﬁnite, discrete group with Kazhdan’s property (T ), then has no faithful C 2 action on the circle.

160 / d. w. morris

Bounded generation provides another approach to proving Theorem 1.1 and some other cases of Conjecture 1.3. This strategy is explained in Section 9. In spite of the above results, Conjecture 1.3 remains completely open for cocompact lattices:

problem 1.8. Find a cocompact lattice in SL (n, R), for some n, such that no ﬁnite-index subgroup of has a faithful action on S1 . remark 1.9. A ﬁnal section of the paper (§10) brieﬂy discusses the generalization of Conjecture 1.3 to lattices in other semisimple Lie groups and two other topics: lattices that do act on the circle, and actions on trees. notes for §1. The survey of É. Ghys [Gh3] and the forthcoming book of A. Navas [Na2] are excellent introductions to the study of group actions on the circle. Versions of Conjecture 1.3 were discussed in conversation as early as 1990, but the ﬁrst published appearance may be in [Gh2, p. 200]. Expositions of the Margulis normal subgroup theorem can be found in [Ma2, chap. 4] and [Zi2, chap. 8]. Regarding the equivalence of Conjectures 1.3 and 1.3 , see [Uom, thm. 1] for a proof that if every -orbit on a connected manifold is ﬁnite, then the kernel of the action has ﬁnite index. acknowledgments: Preparation of this paper was partially supported by research grants from the Natural Sciences and Engineering Research Council of Canada and the U. S. National Science Foundation. I am grateful to M. Burger, É. Ghys, N. Monod, A. Navas, and an anonymous referee, for helpful comments on a preliminary version of the manuscript. 2. Algebraic Reformulation of the Conjecture In this section, we explain that Conjecture 1.3 can be reformulated in a purely algebraic form (see Proposition 2.8). The proof has two main parts: having an action on S1 is almost the same as having an action on R, and having an action on R is essentially the same as being left orderable.

definition 2.1. is left orderable if there is a left-invariant total order on . More precisely, there is an order relation ≺ on , such that 1) ≺ is a total order (that is, for all a, b ∈ , either a ≺ b or a b, or a = b); and

can lattices in sl (n, R) act on the circle? / 161

2) ≺ is invariant under multiplication on the left (that is, for all a, b, c ∈ , if a ≺ b, then ca ≺ cb).

exercise 2.2. Show is left orderable if and only if there exists a subset P of , such that 1) = P {e} P −1 (disjoint union), where P −1 = { a−1 | a ∈ P }; and 2) P is closed under multiplication (that is, ab ∈ P, for all a, b ∈ P). Thus, being left orderable is a property of the internal algebraic structure of . [Hint: A subset P as above is the “positive cone” of an order ≺. Given P, deﬁne a ≺ b if b−1 a ∈ P. Given ≺, deﬁne P = { a ∈ | a e }. ]

remark 2.3. The group is said to be right orderable if there is a rightinvariant total order on . The following exercise shows that the choice between “left orderable" and “right orderable" is entirely a matter of personal preference. exercise 2.4. Show that is left orderable if and only if is right orderable. [Hint: Deﬁne x ≺≺ y if x −1 ≺ y −1 . Alternatively, the conclusion can be derived from Exercise 2.2 and its analogue for right-orderable groups. ]

With this terminology, we can state the following algebraic conjecture, which will be seen to be equivalent to Conjecture 1.3.

conjecture 2.5. Let be a lattice in SL (n, R), with n ≥ 3. Then is not left orderable. As a tool for showing that Conjecture 2.5 is equivalent to Conjecture 1.3, let us give a geometric interpretation of being left orderable.

lemma 2.6. A countable group is left orderable if and only if there is an orientation-preserving, faithful action of on R. Proof. (⇐) For a, b ∈ , deﬁne a≺b

if

a(0) < b(0).

162 / d. w. morris

It is easy to see that ≺ is transitive and antisymmetric, so it deﬁnes a partial order on . For any c ∈ , the function x → c(x) is a strictly increasing function on R (because the action is orientation preserving). Hence, if a(0) < b(0), then c(a(0)) < c(b(0)). Therefore, ≺ is left-invariant (as required in Deﬁnition 2.1(2)). However, the relation ≺ may not be a total order, because it could happen that a(0) = b(0), even though a = b. It is not difﬁcult to modify the deﬁnition to deal with this technical point (see Exercise 2.7). (⇒) We leave this as an exercise for the reader.1

exercise 2.7. Complete the proof of Lemma 2.6(⇐), by deﬁning a left-invariant total order on . [Hint: Make a list q1 , q2 , . . . of the rational numbers and deﬁne a ≺ b if a(qn ) < b(qn ), where n is minimal with a(qn ) = b(qn ). ]

proposition 2.8. Conjecture 1.3 and 2.5 are equivalent. Proof. We show there exists a counterexample to Conjecture 1.3 if and only if there exists a counterexample to Conjecture 2.5. (⇐) Suppose is a counterexample to Conjecture 2.5, so is left orderable. Then Lemma 2.6 tells us that has a faithful action on R. This implies that acts faithfully on the one-point compactiﬁcation of R, which is homeomorphic to S1 . So is a counterexample to Conjecture 1.3. (⇒) Suppose is a counterexample to Conjecture 1.3, so has a faithful action on S1 . From the Ghys-Burger-Monod theorem (1.4), we know that the action has a ﬁnite orbit. Therefore, some subgroup of ﬁnite index in has a ﬁxed point p. Then acts faithfully on the complement S1 {p}, which is homeomorphic to R. Let be the subgroup of consisting of the elements that act by orientation-preserving homeomorphisms (so is either , or a subgroup of index 2 in ). Then Lemma 2.6 tells us that is left orderable. Furthermore, is a lattice in SL (n, R) (because it has ﬁnite index in ). Therefore, is a counterexample to Conjecture 2.5. 1. Hint: If the order relation (, ≺ ) is dense (that is, if a ≺ b ⇒ ∃c ∈ , a ≺ c ≺ b), then it is order-isomorphic to (Q, < ). The action of on (, ≺ ) by left multiplication extends to a continuous action of on the Dedekind completion, which is homeomorphic to R.

can lattices in sl (n, R) act on the circle? / 163

warning 2.9. The proof does not show that the assertions of the two conjectures are equivalent for each individual lattice . Rather, if one of the conjectures is valid for all lattices , then the other conjecture is also valid for all . More precisely, if one of the conjectures is valid for every ﬁnite-index subgroup of , then the other conjecture is also valid for . remark 2.10. In order to prove that Conjecture 2.5 implies Conjecture 1.3, we appealed to Theorem 1.4, which is quite deep. A stronger algebraic conjecture (that no central extension of is left orderable) can easily be shown to imply Conjecture 1.3: exercise 2.11. A central extension of is a group , such that /Z ∼ = , for some subgroup Z of the center of . Show that if has an orientationpreserving, faithful action on S1 , then some central extension of is left orderable. [Hint: Because R is the universal cover of S1 , every homeomorphism of S1 lifts to a homeomorphism of R, and the lift is unique modulo an element of the fundamental group Z. Let be the subgroup of Homeo+ (R) consisting of all of the possible lifts of all of the elements of , and show that is a central extension that is left orderable. ]

notes for §2. The material in this section is well known. See [KM] for a treatment of the algebraic theory of left-ordered groups. Informative discussions of orderings, actions on the line, and related topics appear in [Gh3, §6.5] and [Na2, §2.2.3–2.2.6]. 3. SL(n, Z ) Has no Faithful Action on the Circle In this section, we prove Theorem 1.1 by exploiting the interaction between some obvious nilpotent subgroups of . The crucial ingredients are Lemma 3.6 and the fact that the theorem can be restated in the following algebraic form (cf. Proposition 2.8).

theorem 3.1. (witte) If is a ﬁnite-index subgroup of SL (n, Z), with n ≥ 3, then is not left orderable. remark 3.2. It is easy to see that the group SL (n, Z) itself is not left orderable, because it has elements of ﬁnite order. But there are subgroups of ﬁnite index that do not have any elements of ﬁnite order, and we will show that they, too, are not left orderable.

164 / d. w. morris

definition 3.3. Suppose ≺ is a left-invariant order relation on . For elements a and b of , we say a is inﬁnitely smaller than b (denoted a b) if either ak ≺ b for all k ∈ Z or ak ≺ b−1 for all k ∈ Z. Notice that the relation is transitive. notation 3.4. The commutator a−1 b−1 ab of elements a and b of a group is denoted [a, b]. It is straightforward to check that a commutes with b if and only if [a, b] = e (the identity element of ). lemma 3.5. Let a and b be elements of . If [a, b] commutes with both a and b, then [bk , am ] = [a, b]−km for all k, m ∈ Z. Proof. Exercise (or see [Gor, Lems. 2.2.2(i) and 2.2.4(iii), pp. 19–20]).

lemma 3.6. (ault, rhemtulla) Suppose a, b, z are nonidentity elements of a left-ordered group H, with [a, b] = zk for some nonzero k ∈ Z, and [a, z] = [b, z] = e. Then either z a or z b. Proof. Assume, for simplicity, that a, b, z ≥ e and k > 0. (All other cases can be reduced to this one by replacing some or all of a, b, z with their inverses, and/or interchanging a with b.) We may assume z a, so a ≺ zp for some p ∈ Z+ . Similarly, we may assume b ≺ zq for some q ∈ Z+ . Then (using the left-invariance of ≺) we have e ≺ a−1 zp ,

e ≺ b−1 zq ,

e ≺ a,

e ≺ b,

e ≺ z.

Hence, for all m ∈ Z+ , we have e ≺ (b−1 zq )m (a−1 zp )m bm am = b−m a−m bm am z(p+q)m

(z commutes with a and b)

= [bm , am ]z(p+q)m = z−km z(p+q)m 2

= z−km

([a, b] = zk and see (lemma 3.5))

2 +(p+q)m

= znegative

(if m is sufﬁciently large)

≺ e. This is a contradiction.

can lattices in sl (n, R) act on the circle? / 165

Proof of Theorem 3.1. Suppose is left orderable. ( This will lead to a contradiction.) Because subgroups of left-orderable groups are left orderable, we may assume n = 3; that is, has ﬁnite index in SL (3, Z). Hence, there is some k ∈ Z+ such that the six matrices ⎡ ⎤ ⎡ ⎤ ⎡ ⎤ 1 k 0 1 0 k 1 0 0 ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ a1 = ⎣0 1 0⎦ , a2 = ⎣0 1 0⎦ , a3 = ⎣0 1 k⎦ , 0 0 1 0 0 1 0 0 1 3.7

⎡

1 ⎢ a4 = ⎣k 0

0 1 0

⎤ ⎡ 0 1 ⎥ ⎢ 0⎦ , a5 = ⎣0 1 k

0 1 0

⎤ ⎡ 0 1 ⎥ ⎢ 0 ⎦ , a 6 = ⎣0 1 0

0 1 k

⎤ 0 ⎥ 0⎦ 1

all belong to . A straightforward check shows that [ai , ai+1 ] = e and [ai−1 , ai+1 ] = (ai )±k for i = 1, . . . , 6, with subscripts read modulo 6. Thus, Lemma 3.6 asserts 3.8

either ai ai−1 or ai ai+1 .

In particular, we must have either a1 a6 or a1 a2 . Assume for deﬁniteness that a1 a2 . ( The other case is very similar.) For each i, Lemma 3.6 implies that if ai−1 ai , then ai ai+1 . Since a1 a2 , we conclude by induction that a1 a2 a3 a4 a5 a6 a1 . Thus a1 a1 , a contradiction.

For the reader acquainted with root systems, let us give a more conceptual presentation of the proof of Theorem 3.1. Alternate Version of the Proof of Theorem 3.1. The root system of SL (3, Z) is of type A2 , pictured in Figure 1. For i = 1, . . . , 6, let Ui be the root space of corresponding to the root αi . (Note that Ui ∼ = Z is cyclic. In the notation of the above proof, the element ai belongs to Ui .) Deﬁne αi αj if there exists u ∈ Uj , such that Ui ≺ u (that is, v ≺ u, for all v ∈ Ui ). It is obvious, from Figure 1, that αi−1 + αi+1 = αi , and that αi + αi+1 is not a root, so e = [Ui−1 , Ui+1 ] ⊂ Ui

and

[Ui , Ui+1 ] = e.

166 / d. w. morris a1 a6

a2

a5

a3 a4

Fig. 1. The root system of SL (3, Z) (type A2 ).

Hence, Lemma 3.6 implies that either αi αi−1 or αi αi+1 . Arguing as in the above proof, we conclude that α1 α2 α3 α4 α5 α6 α1 . Thus α1 α1 , a contradiction.

notes for §3. Theorem 3.1 is due to D. Witte [Wit]. (An exposition of the proof also appears in [Gh3, thm. 7.2].) Lemma 3.6 is a special case of a theorem of J. C. Ault [Aul] and A. H. Rhemtulla [Rhe]: if is any nontrivial, ﬁnitely generated, left-ordered nilpotent group, then there is some a ∈ , such that [, ] ≺ a. An elementary proof that SL (n, Z) has no nontrivial actions on the circle can be found in [BV], but the argument uses the existence of elements of ﬁnite order, so it does not apply to ﬁnite-index subgroups of SL (n, Z). 4. The Reeb-Thurston Stability Theorem The Ghys-Burger-Monod theorem (1.4) tells us that if is a lattice in G = SL (3, R), then any action of on S1 has a ﬁnite orbit; in other words, some ﬁnite-index subgroup of has a ﬁxed point. In order to deduce Corollary 1.6 from this it sufﬁces to show that if the action is by (orientation-preserving) C 1 diffeomorphisms, then acts trivially. This triviality of is immediate from the following result:

proposition 4.1. (reeb-thurston stability theorem) Suppose • is a ﬁnitely generated subgroup of Diff 1 (S1 ), + • has a ﬁxed point, and • the abelianization /[, ] is ﬁnite.

Then = {e} is trivial.

can lattices in sl (n, R) act on the circle? / 167

A differentiable action on S1 that has a ﬁxed point can be transformed into a differentiable action on the unit interval [0, 1] (cf. pf. of Proposition 2.8(⇒)). Thus, Proposition 4.1 can be reformulated as follows:

proposition 4.1 . (reeb-thurston stability theorem) Suppose • I = [0, 1] is the unit interval, • is a ﬁnitely generated subgroup of Diff 1 (I), and + • the abelianization /[, ] is ﬁnite.

Then = {e} is trivial.

Proof in a special case. Deﬁne σ : → R+ by σ (λ) = λ (0). From the Chain Rule, we see that σ is a (multiplicative) homomorphism. Because /[, ] is ﬁnite and R+ is abelian, this implies that σ () is ﬁnite. However, R+ has no nontrivial, ﬁnite subgroups, so this implies that σ () is trivial; therefore λ (0) = 1, for all λ ∈ . For simplicity, let us assume, henceforth, that each element of is real analytic, rather than merely C 1 . (We remark that there is no need to assume is ﬁnitely generated in the real-analytic case. The proof for C 1 actions is similar, but requires some additional effort, and does use the hypothesis that is ﬁnitely generated.) Thus, each element λ of can be expressed as a power series in a neighborhood of 0: λ(x) = x + aλ,2 x 2 + aλ,3 x 3 + · · · . ( There is no constant term, because λ(0) = 0; the coefﬁcient of x is 1, because λ (0) = 1.) Now suppose is nontrivial. (This will lead to a contradiction.) Then there exist n and λ, such that aλ,n = 0.

4.2

We may assume n is minimal; hence λ(x) = x + aλ,n x n + aλ,n+1 x n+1 + · · · for all λ ∈ . Then, for λ, γ ∈ , we have

168 / d. w. morris

x + aλγ ,n x n + O(x n+1 ) = (λγ )(x) = λ(γ (x)) = γ (x) + aλ,n (γ (x))n + O(x n+1 ) = (x + aγ ,n x n + O(x n+1 )) + aλ,n (x + aγ ,n x n + O(x n+1 ))n + O(x n+1 ) = x + (aγ ,n + aλ,n )x n + O(x n+1 ). Thus the map τ : → R, deﬁned by τ (λ) = aλ,n , is an additive homomorphism. Because /[, ] is ﬁnite, but R has no nontrivial ﬁnite subgroups, this implies that τ () is trivial; therefore aλ,n = 0 for every λ ∈ . This contradicts Equation (4.2).

remark 4.3 on the general case Suppose is nontrivial. Then there is no harm in assuming that acts nontrivially on every neighborhood of 0. Hence, letting 0 be a ﬁnite generating set for , we may choose λ0 ∈ 0 and a sequence xn → 0+ , such that, for all n, we have λ0 (xn ) = xn and |λ(xn ) − xn | ≤ |λ0 (xn ) − xn | for all λ ∈ 0 . By passing to a subsequence of {xn }, we may assume τ (λ) = lim

n→∞

λ(xn ) − xn λ0 (xn ) − xn

exists for all λ ∈ 0 . Because λ (0) = 1 and λ is C 1 , it can be shown that τ (λγ ) = τ (λ) + τ (γ ). Therefore τ : → R is a (nontrivial) homomorphism. This is a contradiction.

remark 4.4. (zimmer) Here is the outline of a nice proof of Proposition 4.1 , under the additional assumption that has Kazhdan’s property (T ). 1) It sufﬁces to show that the ﬁxed points of form a dense subset of I; thus, we may assume that 0 and 1 are the only points that are ﬁxed by . 2) We have γ (0) = 1 for all γ ∈ .

can lattices in sl (n, R) act on the circle? / 169

3) Deﬁne a unitary representation of on L2 (I) by f γ (t) = f (γ t) |γ (t)|1/2 . 4) For any γ ∈ and any δ > 0, if f is the characteristic function of a sufﬁciently small neighborhood of 0, then f γ − f < δ f . Hence, this unitary representation has almost-invariant vectors. 5) Because has Kazhdan’s property (T ), this implies there are ﬁxed vectors: there exists f ∈ L2 (I) {0}, such that f γ = f for all γ ∈ . 6) Every point in the essential support of f is a ﬁxed point of . 7) This is a contradiction.

notes for §4. Proposition 4.1 was proved by W. Thurston [Thu] in a more general form that also applies to actions on manifolds of higher dimension. (It generalizes a theorem of G. Reeb.) See [RS] or [Sch] for details of the proof sketched in Remark 4.3. The proof outlined in Remark 4.4 is due to R. J. Zimmer; details appear in [WZ, §5, pp. 108–109]. 5. Smooth Actions of Kazhdan Groups on the Circle In this section, we prove Theorem 1.7. First, let us recall one of the many equivalent deﬁnitions of Kazhdan’s property (T ).

notation 5.1. For any real Hilbert space H, we use Isom (H) to denote the isometry group of H. (We remark that each isometry of H is the composition of a translation with a norm-preserving linear transformation.) definition 5.2. We say that a discrete group has Kazhdan’s property ( T ) if, for every homomorphism ρ : → Isom (H), where H is a real Hilbert space, there exists v ∈ H, such that ρ(g)v = v, for every g ∈ . In short, to say that has Kazhdan’s property (T ) means that every isometric action of on any Hilbert space has a ﬁxed point. The importance of this notion for our purposes stems from the following result, whose proof we omit.

theorem 5.3. (kazhdan) If n ≥ 3, then every lattice in SL (n, R) has Kazhdan’s property (T ).

170 / d. w. morris

exercise 5.4. Show that if is an inﬁnite group with Kazhdan’s property (T ), then is not abelian. [Hint: Every group with Kazhdan’s property (T ) is ﬁnitely generated, and every ﬁnitely generated abelian group is either ﬁnite or has a quotient isomorphic to Z. ]

We now turn to the proof of Theorem 1.7. To simplify notation, we may think of S1 as [−π/2, π/2]. In particular, for any diffeomorphism of S1 and any x ∈ S1 , the derivative g (x) is a well-deﬁned real number.

definition 5.5. • Let F (S1 × S1 ) be the vector space of measurable functions on S1 × S1

(with two functions being identiﬁed if they are equal almost everywhere). • Deﬁne an action of Diff 2 (S1 ) on F (S1 × S1 ) by

F g (x, y) = F(g(x), g(y)) |g (x)|1/2 |g (y)|1/2 for F ∈ F (S1 × S1 ) and g ∈ Diff 2 (S1 ).

1/2 2 • Let F2 = be the L2 -norm of F; note that S1 S1 F(x, y) dx dy F2 = ∞ if F ∈ / L2 (S1 × S1 ). Note that F gh = (F g )h

and

F g 2 = F2

for F ∈ F (S1 × S1 ) and g, h ∈ Diff 2 (S1 ).

notation 5.6. • Choose a positive function f on S1 , such that

— f has a 1/x singularity at the point 0 of S1 , and — f is C ∞ everywhere else; that is, identifying S1 with [−π/2, π/2],we have 5.7

f (x) −

1 ∈ C ∞ (S1 ). |x|

For example, one may take f (x) = | cot x |. • Now deﬁne

(x, y) = f (x − y) on S1 × S1 . Because of the 1/x singularity of f , it is easy to see that ∈ / L2 (S1 × S1 ). For any g ∈ Diff 2 (S1 ), the following calculation shows that the singularity of g cancels the singularity of .

can lattices in sl (n, R) act on the circle? / 171

lemma 5.8. The difference g − is a bounded function on S1 × S1 , for any g ∈ Diff 2 (S1 ). Proof. From Equation (5.7), we see that there is no harm in working with the function 0 (x, y) = 1/|x − y|, instead of . Also, in order to reduce the number of absolute-value signs, let us assume g ≥ 0 everywhere. | g (x, y) − (x, y)| g

≈ | 0 (x, y) − 0 (x, y)| = g (x)1/2 g (y)1/2 0 (g(x), g(y)) − 0 (x, y) 1/2 1/2 g (x) g (y) 1 − = |g(x) − g(y)| |x − y| 1/2 1/2 g (x) g (y) 1 = − |g (t)(x − y)| |x − y| =

|g (x)1/2 g (y)1/2 − g (t)| g (t) |x − y|

|g (x)g (y) − g (t)2 | = 1/2 1/2 (g (x) g (y) + g (t))g (t)|x − y| =O

|g (x)g (y) − g (t)2 | |x − y|

∃t ∈ (x, y), by Mean Value thm.

multiply by conjugate of numerator

g diffeomorphism . ⇒ g is never 0

Now, we wish to show that the numerator is bounded by a constant multiple of the denominator. |g (x)g (y) − g (t)2 | ≤ g (x)|g (y) − g (t)| + g (t)|g (x) − g (t)| = g (x) |g (u)| |y − t| + g (t) |g (v)| |x − t| ≤ g (x) |g (u)| |x − y| + g (t) |g (v)| |x − y| = O(|x − y|)

( Triangle inequality)

Mean Value thm. applied to g (because t ∈ (x, y))

g , g continuous, . so bounded

We will also use the following classical fact:

172 / d. w. morris

Fig. 2. A rectangle Rn (shaded) in S1 × S1 .

lemma 5.9. (hölder, 1901) Every nonabelian group of homeomorphisms of S1 contains at least one nonidentity element that has a ﬁxed point. Proof of Theorem 1.7. Suppose is a discrete group with Kazhdan’s property (T ), and we have a faithful C 2 action of on the circle S1 . From Lemma 5.8, we see that 5.10

( + L2 (S1 × S1 ))g = + L2 (S1 × S1 ) for all g ∈ Diff 2 (S1 ).

Thus, acts (by isometries) on the afﬁne Hilbert space + L2 (S1 × S1 ). Because has Kazhdan’s property, we conclude that has a ﬁxed point F in + L2 (S1 × S1 ): F g = F for all g ∈ . / Because ∈ / L2 (S1 × S1 ) and F − ∈ L2 (S1 × S1 ), it is obvious that F ∈ 2 1 1 L (S × S ). Now, deﬁne a measure μ on S1 × S1 by μ = F 2 dx dy. Then 1) μ is a -invariant measure on S1 × S1 (because F is -invariant); and 2) if R = (a1 , b1 ) × (a2 , b2 ) is a rectangle in S1 × S1 , then, from the 1/x singularity of along the diagonal, we see that ⎧ ⎨∞ if R intersects the diagonal; μ(R) = ⎩ﬁnite if R is away from the diagonal. For example, in Figure 2, the rectangle (a, b) × (b, c) has inﬁnite measure because it touches the diagonal at (b, b), but the shaded rectangle Rk has ﬁnite measure, because it does not touch the diagonal. Because is not abelian, Lemma 5.9 tells us we may choose g ∈ , such that g has a ﬁxed point. Then, by passing to a triple cover of S1 , we obtain an

can lattices in sl (n, R) act on the circle? / 173

action of (a ﬁnite extension of ) in which g has at least three ﬁxed points: g(a) = a,

g(b) = b,

g(c) = c.

By perhaps replacing a and b with different ﬁxed points, we may assume, for x ∈ (a, b), that lim g k (x) =

⎧ ⎨b

as k → ∞

⎩a

as k → −∞.

For each k ∈ Z, deﬁne Rk = (a, g k (x)) × (b, c) (see Figure 2). Then g(Rk ) = Rk+1 , so μ(Rk ) = μ(Rk+1 ). Therefore 0=

∞

(μ(Rk+1 ) − μ(Rk ))

k=−∞

⎛

= μ⎝

∞

⎞ (Rk+1 Rk )⎠

k=−∞

= μ((a, b) × (b, c)). However, the rectangle (a, b) × (b, c) touches the diagonal at the point (b, b), so it has inﬁnite measure. This is a contradiction.

remark 5.11. In the proof of Theorem 1.7, the assumption that the elements of are C 2 was used only to establish Equation (5.10). For this, it is not necessary to show that g − is bounded (as in Lemma 5.8), but only that g − ∈ L2 (S1 × S1 ). The calculations in the proof of Lemma 5.8 show that this holds under the weaker hypothesis that g ∈ C 3/2+ , for any > 0. In fact, A. Navas observed that by using recent results that group with Kazhdan’s property (T ) also have ﬁxed points in certain Lp spaces with p = 2, it can be shown that C 2 can be replaced with C 3/2 in the statement of Theorem 1.7.

174 / d. w. morris

This leads to the following well-known question:

problem 5.12. Can an inﬁnite (discrete) group with Kazhdan’s property (T ) have a faithful C 0 action on S1 ? notes for §5. Theorem 1.7 was ﬁrst proved in [Na1], but the idea to use Lemma 5.8 came from earlier work of A. Pressley and G. Segal [PS], and A. Reznikov [Rez, chap. 2]. Proofs also appear in [BHV, §2.9] and [Na2, §5.2]. The condition in Deﬁnition 5.2 was introduced by J.–P. Serre, and it is not at all obvious that it is equivalent to the original deﬁnition of property (T ) that was given by D. Kazhdan. For a discussion of this, and much more, the standard reference on Kazhdan’s property (T ) is [BHV]. See [Gh3, thm. 6.10] or [Na2, §2.2.4] for a proof of Lemma 5.9. Implications of Kazhdan’s property (T ) for ﬁxed points of actions on Lp spaces (and other Banach spaces) are discussed in [BFGM]. 6. Ghys's Proof That Actions Have a Finite Orbit In this section, we present Ghys’s proof of Theorem 1.4, modulo some facts that will be proved in Section 7. To get started, let us show that it sufﬁces to ﬁnd a -invariant measure on the circle.

definition 6.1. A measure μ on a measure space X is a probability measure if μ(X ) = 1.

lemma 6.2. If • is a discrete group, such that the abelianization /[, ] is ﬁnite, • acts on S1 by orientation-preserving homeomorphisms, and • there is a -invariant probability measure μ on S1 ,

then has at least one ﬁnite orbit on S1 . Proof. We consider two cases. Case 1. Assume μ has at least one atom p. Let p = { gp | g ∈ } be the orbit of p. Because μ is -invariant, we have μ(gp) = μ(p), for every g ∈ . Therefore μ(p) = #(p) μ(p).

can lattices in sl (n, R) act on the circle? / 175

Since μ is a probability measure, we know that μ(p) < ∞, so we conclude that p is a ﬁnite set. That is, the orbit of p is ﬁnite. Case 2. Assume μ has no atoms. To simplify the proof, let us assume that the support of μ is all of S1 . (In other words, every nonempty open subset of S1 has positive measure.) For x, y ∈ S1 , deﬁne d(x, y) = μ([x, y]),

6.3

where [x, y] is a path from x to y, and, for a given x and y, we choose the path [x, y] to minimize μ([x, y]). It is easy to see that d is a metric on S1 . Up to isometry, there is a unique metric on S1 , so we may assume that d is the usual arc-length metric. Because μ is -invariant, we know that d is -invariant, so acts by rotations of the circle. There is no harm in assuming that the action is faithful, so we conclude that is abelian. But every abelian quotient of is ﬁnite, so we conclude that is ﬁnite. Hence, every orbit is ﬁnite.

exercise 6.4. Complete the proof of Lemma theorem 6.2, by eliminating the assumption that the support of μ is all of S1 in Case 2. [Hint: Modify the above proof to show that every orbit in the support of μ is ﬁnite. ]

To simplify the notation, we assume henceforth that n = 3.

notation 6.5. Let • G = SL (3, R), • be a lattice in G, and

⎡ ⎤ ∗ ∗ ∗ ⎥ • P=⎢ ∗ ∗⎦ ⊂ G. ⎣ ∗

We remark that P is a minimal parabolic subgroup of G. Ghys’s proof is based on the following key fact that will be established in §7A by using the fact that the group P is “amenable.”

notation 6.6. Prob (X ) denotes the set of all Radon probability measures on X (where X is any compact, Hausdorff space). This is a closed, convex subset of the unit ball in C(X )∗ , with the weak∗ topology, so Prob (X ) has a natural topology that makes it a compact Hausdorff space.

176 / d. w. morris

proposition 6.7. (furstenberg) If acts continuously on any compact metric space X , then there is a -equivariant, Borel measurable map ψ : G/P → Prob (X ). If ψ were invariant, rather than equivariant, the following fundamental theorem would immediately imply that ψ is constant (a.e.). ( This theorem will be proved in §7B.)

theorem 6.8. (moore ergodicity theorem) If H is any closed, noncompact subgroup of G, then the action of on G/H is ergodic: by deﬁnition, this means that every -invariant, measurable function on G/H is constant (a.e.). The homogeneous space G/P plays a major role in Ghys’s proof, because of its appearance in Proposition 6.7. We will now present a geometric interpretation of this space that is very helpful. Recall that a ﬂag in R3 consists of a pair (, π), where • is a 1-dimensional vector subspace of R3 (a line), and • π is a 2-dimensional vector subspace of R3 (a plane) that contains .

The group G = SL (3, R) acts transitively on the set of ﬂags, and P is the stabilizer of the standard ﬂag F0 = (0 , π0 ), where 0 = (∗, 0, 0) and π0 = (∗, ∗, 0). Therefore:

proposition 6.9. G/P can be identiﬁed with the space F of all ﬂags, by identifying gP with the ﬂag gF0 . Proof of Theorem 1.4. Suppose acts by homeomorphisms on the circle S1 . From Proposition 6.7, we know there is a -equivariant, measurable map ψ : G/P → Prob (S1 ). It will sufﬁce to show that ψ is constant (a.e.), for then • the essential range of ψ consists of a single point μ ∈ Prob (S1 ), and • the essential range of ψ is -invariant, because ψ is -equivariant.

So μ is -invariant, and then Lemma 6.2 implies has a ﬁnite orbit on S1 . There are two basic cases to consider: either the measure ψ(x) consists entirely of atoms, or ψ(x) has no atoms. (Recall that an atom of a measure μ is

can lattices in sl (n, R) act on the circle? / 177

a point p, such that μ({p}) = 0.) It is also possible that ψ(x) consists partly of atoms, and partly of nonatoms, but Corollary 7.5 below tells us that we need only consider the two extreme cases. Case 1. Assume ψ(x) has no atoms, for a.e. x ∈ G/P. • Let Prob0 (S1 ) = { μ ∈ Prob (S1 ) | μ has no atoms }. • By assumption, we know ψ : G/P → Prob0 (S1 ), so we may deﬁne

ψ 2 : (G/P)2 → (Prob0 (S1 ))2

by

ψ 2 (x, y) = (ψ(x), ψ(y)).

Then ψ 2 , like ψ, is measurable and -equivariant. • Deﬁne d : (Prob0 (S1 ))2 → R by

d(μ1 , μ2 ) = sup |μ1 ( J) − μ2 ( J)|, J

where J ranges over all intervals (that is, over all connected subsets of S1 ). Since acts by homeomorphisms, and any homeomorphism of S1 maps intervals to intervals, it is easy to see that d is -invariant. We claim that 6.10

d is continuous,

with respect to the usual weak∗ topology that Prob0 (S1 ) inherits from being a subset of Prob (S1 ). To see this, we note that if μ ∈ Prob0 (S1 ) and > 0, then, because μ has no atoms, we may partition S1 into ﬁnitely many intervals J1 , . . . , Jk , such that μ( Ji ) < for all i. By approximating the characteristic functions of these intervals from above and below, we may construct continuous functions f1 , . . . , f2k and some δ > 0, such that, for ν ∈ Prob (S1 ),

if |ν( fi ) − μ( fi )| < δ for every i, then |ν( Ji ) − μ( Ji )| < for every i. n Then, for any interval J in S1 , we have |ν( J) − μ( J)| < 2 , so d(ν, μ) < 2 . This completes the proof of (6.10). Because ψ 2 is measurable and d is continuous, we know that the composition d ◦ ψ 2 is measurable. In addition, because ψ 2 is -equivariant and d is -invariant, we also know that d ◦ ψ 2 is -invariant. (Note: we are saying invariant, not just equivariant.) From the Moore ergodicity theorem (6.8), we know that is ergodic on (G/P)2 (see Corollary 7.6), so we conclude that d ◦ ψ 2 is constant (a.e.); say d(ψ 2 (x, y)) = c,

for a.e. x, y ∈ G/P.

178 / d. w. morris

We wish to show that c = 0, for then it is clear that ψ(x) = ψ( y) for a.e. x, y ∈ G/P, so ψ is constant (a.e.). It is obvious that d(ψ 2 (x, x)) = 0 for every x ∈ G/P. If ψ were constant everywhere, rather than merely almost everywhere, then it would follow immediately that c = 0. Unfortunately, the diagonal {(x, x)} is a set of measure 0 in (G/P)2 , so a little bit of additional argument is required. By Lusin’s theorem, ψ is continuous on a set K of positive measure. Since the composition of continuous functions is continuous, we conclude that d ◦ ψ 2 is continuous on K × K . So, by continuity, it is constant on all of K × K , not merely almost all. Since d(ψ 2 (x, x)) = 0, this implies d(ψ 2 (x, y)) = 0 for all x, y ∈ K . Since K is a set of positive measure, this implies c = 0, as desired. Case 2. Assume ψ(x) consists entirely of atoms, for a.e. x ∈ G/P. To simplify the notation, without losing the main ideas, let us assume that ψ(x) consists of a single atom, for every x ∈ G/P. Thus, we may think of ψ as a -equivariant, measurable map ψ : G/P → S1 . Surprisingly, even with the simplifying assumption, the argument here seems to be more difﬁcult than in Case 1. The idea is to obtain a contradiction from the -equivariance of ψ, by contrasting two fundamental observations: • Homeo+ (S1 ) is not triply transitive on S1 : if x, y, and z are distinct,

then no orientation-preserving homeomorphism of S1 can map the triple (x, y, z) to (y, x, z)—they have opposite orientations under the circular order on S1 . • The action of GL (2, R) on the projective line RP 1 = R ∪ {∞} by linearfractional transformations ax + b a b if g = g(x) = cx + d c d is triply transitive: if (x1 , y1 , z1 ) and (x2 , y2 , z2 ) are two ordered triples of distinct elements of R ∪ {∞}, then there is some g ∈ GL (2, R) with g(x1 ) = x2 , g(y1 ) = y2 , and g(z1 ) = z2 . To illustrate, let us give an easy proof that is not quite correct; the actual proof is a modiﬁed version of this. Deﬁne ψ 3 : (G/P)3 → (S1 )3

by

ψ 3 (x, y, z) = (ψ(x), ψ(y), ψ(z)).

can lattices in sl (n, R) act on the circle? / 179

Then ψ 3 is -equivariant, so X + = { (x, y, z) ∈ (G/P)3 | (ψ(x), ψ(y), ψ(z)) is positively oriented } is a -invariant, measurable subset of (G/P)3 . Let us assume that is ergodic on (G/P)3 . (Unfortunately, this assumption is false, so it is where the proof breaks down.) Then X + must be (almost) all of (G/P)3 ; thus, (ψ(x), ψ(y), ψ(z)) is positively oriented, for (almost) every x, y, z ∈ G/P. But this is nonsense: either (ψ(x), ψ(y), ψ(z)) or (ψ(y), ψ(x), ψ(z)) is negatively oriented, so there are many negatively oriented triples. To salvage the above faulty proof, we replace (G/P)3 with a subset X , on which does act ergodically. Let ⎡ ⎤ ∗ ∗ ∗ ⎥ • Q =⎢ ⎣∗ ∗ ∗⎦ ⊂ G, and 0 0 ∗ x Q = x Q = x Q, 1 2 3 3 • X = (x1 , x2 , x3 ) ∈ (G/P) . x1 , x2 , x3 distinct Note that X is a submanifold of (G/P)3 . (If this is not obvious, it follows from the fact, proven below, that X is a single G-orbit in (G/P)3 .) Assume, for the moment, that is ergodic on X (with respect to any (hence, every) Lebesgue measure on the manifold X ). Then the above proof, with X in the place of (G/P)3 , implies that X + = { (x, y, z) ∈ X | (ψ(x), ψ(y), ψ(z)) is positively oriented } is a set of measure 0. Then it is not difﬁcult to see that ψ is constant on X (a.e.). Hence ψ is right Q -invariant (a.e.): for each q ∈ Q , we have ψ(xqP) = ψ(xP) for a.e. x ∈ G/P. By a similar argument, we see that ψ is right Q -invariant (a.e.), where ⎡ ⎤ ∗ ∗ ∗ ⎢ ⎥ Q = ⎣0 ∗ ∗⎦ ⊂ G. 0 ∗ ∗ Because Q and Q , taken together, generate all of G, it then follows that ψ is right G-invariant (a.e.). Hence, ψ is constant (a.e.), as desired. All that remains is to show that is ergodic on X . By the Moore ergodicity theorem (6.8), we need only show that 1) G is transitive on X , and 2) the stabilizer of some element of X is not compact.

180 / d. w. morris

These facts are perhaps easiest to establish from a geometric perspective. The (parabolic) subgroup Q is the stabilizer of the plane π0 = (∗, ∗, 0). Hence, QF0 is the set of all ﬂags (, π) with π = π0 . Therefore, under the identiﬁcation of G/P with F , we have 3 π1 = π2 = π3 , . 6.11 X = ((1 , π1 ), (2 , π2 ), (3 , π3 )) ∈ F 1 , 2 , 3 distinct 1) Let us show that G is transitive on X . Given ((1 , π), (2 , π), (3 , π)), ((1 , π ), (2 , π ), (3 , π )) ∈ X , it sufﬁces to show that there exists g ∈ G, such that 6.12

g((1 , π), (2 , π), (3 , π)) = ((1 , π ), (2 , π ), (3 , π )).

Because G is transitive on the set of 2-dimensional subspaces, we may assume π = π = π0 = R2 . Then, because GL (2, R) is triply transitive on RP 1 , there exists T ∈ GL (2, R), such that T (1 , 2 , 3 ) = (1 , 2 , 3 ). Letting ⎡ ⎢ g =⎣ 0

T 0

⎤ 0 ⎥ 0 ⎦ 1 det T

yields Equation (6.12). 2) Let us show that the stabilizer of some element of X is not compact. Let ⎧⎡ ⎫ ⎤ ⎪ ⎪ 0 ⎨ a 0 ⎬ ⎢ ⎥ A = ⎣0 a 0 ⎦ a ∈ R× . ⎪ ⎪ ⎩ ⎭ 0 0 1/a2 Then A is a closed, noncompact subgroup of G. Furthermore, every element of A acts as a scalar on π0 , so every element of A ﬁxes every 1-dimensional vector subspace of π0 . Thus, if ((1 , π), (2 , π), (3 , π)) ∈ X with π = π0 , then A is contained in the stabilizer of this element of X , so the stabilizer is not compact.

can lattices in sl (n, R) act on the circle? / 181

notes for §6. Ghys’s proof ﬁrst appeared in [Gh2]. See [Gh3, §7.3] for an exposition. 7. Additional Ingredients of Ghys's Proof 7A. Amenability and an Equivariant Map

A group is amenable if its action on every compact, convex set has a ﬁxed point. More precisely:

definition 7.1. A Lie group G is amenable if, for every continuous action of G by linear operators on a locally convex topological vector space V , and every nonempty, compact, convex G-invariant subset C of V , the group G has a ﬁxed point in C. example 7.2. 1) If T is any continuous linear operator on V and v is any element of V such that {T n v} is bounded, then every accumulation point of the sequence vn =

1 (Tv + T 2 v + · · · + T n v) n

is a ﬁxed point for T . This implies that cyclic groups are amenable. 2) A generalization of this argument shows that all abelian groups are amenable; this statement is a version of the classical Kakutani-Markov ﬁxed-point theorem. 3) It is not difﬁcult to see that if N is a normal subgroup of G such that N and G/N are both amenable, then G is amenable. 4) Combining the preceding two observations implies that solvable groups are amenable. 5) In particular, the group P of Notation 6.5 is amenable. Proof of Proposition 6.7. Since Prob (X ) is a compact, convex set, a version of the Banach-Alaoglu theorem tells us that L∞ (G; Prob (X )) is compact and convex in a natural weak∗ topology. Let ! " L∞ (G; Prob (X )) = ψ ∈ L∞ (G; Prob (X )) ψ is -equivariant (a.e.) . This is a closed subset of L∞ (G; Prob (X )), so it is compact. It is also convex and nonempty. To say ψ is -equivariant (a.e.) means, for each γ ∈ , that ψ(γ x) = γ · ψ(x) for a.e. x ∈ G; so G acts on L∞ (G; Prob (X )) by translation on the right. Hence, the subgroup P acts on L∞ (G; Prob (X )).

182 / d. w. morris

Since P is amenable (see Example 7.2(5)), it must have a ﬁxed point ψ0 in the compact, convex set L∞ (G; Prob (X )). Then ψ0 is invariant (a.e.) under translation on the right by elements of P, so it factors through (a.e.) to a welldeﬁned map ψ : G/P → Prob (X ). Because ψ0 is -equivariant, it is immediate that ψ is -equivariant. 7B. Moore Ergodicity Theorem

We will obtain the Moore ergodicity theorem (6.8) as an easy consequence of the following result in representation theory:

theorem 7.3. (decay of matrix coefficients) If • G = SL (n, R), • π is a unitary representation of G on a Hilbert space H, such that no non-

zero vector is ﬁxed by π(G); and • {gj } is a sequence of elements of G, such that gj → ∞,

then π(gj )φ | ψ → 0, for every φ, ψ ∈ H. Proof. By passing to a subsequence, we may assume π(gj ) converges weakly to some operator E; that is, π(gj )φ | ψ → Eφ | ψ for every φ, ψ ∈ H. We wish to show ker E = H. Let U = { u ∈ G | gj ugj−1 → e } and U − = { v ∈ G | gj−1 vgj → e }. For u ∈ U, we have Eπ (u)φ | ψ = limπ (gj u)φ | ψ = limπ(gj ugj−1 )π (gj )φ | ψ = limπ(gj )φ | ψ = Eφ | ψ,

can lattices in sl (n, R) act on the circle? / 183

so Eπ (u) = E. Therefore, letting HU be the space of U-invariant vectors in H, we have (HU )⊥ ⊂ ker E. We have E ∗ φ | ψ = φ | Eψ = limφ | π(gj )ψ = limπ (gj−1 )φ | ψ, so the same argument, with E ∗ in the place of E and gj−1 in the place of gj , shows that −

(HU )⊥ ⊂ ker E ∗ . Assume, for simplicity, that each gj is a positive-deﬁnite diagonal matrix: ⎡ ⎤ aj ⎢ ⎥ gj = ⎣ with aj , bj , cj > 0. bj ⎦ cj (It is not difﬁcult to eliminate this hypothesis, by using the Cartan decomposition G = KAK , but that is not necessary for Ghys’s proof.) Then the subgroup generated by {π(gj )} is commutative. Because π is unitary, this means that π (gj ) commutes with both π(gk ) and π (gk )∗ = π (gk−1 ) for every j and k. Therefore, the limit E commutes with its adjoint (that is, E is normal): we have E ∗ E = EE ∗ . Hence Eφ2 = Eφ | Eφ = (E ∗ E)φ | φ = (EE ∗ )φ | φ = E ∗ φ | E ∗ φ = E ∗ φ2 , so ker E = ker E ∗ . Thus, ker E = ker E + ker E ∗ −

⊃ (HU )⊥ + (HU )⊥ −

= (HU ∩ HU )⊥ −

= (HU,U )⊥ . By passing to a subsequence, and then permuting the basis vectors of R3 , we may assume aj ≥ bj ≥ cj .

184 / d. w. morris

Since gj → ∞, we have

aj bj , lim max b j cj j→∞

= ∞.

For deﬁniteness, let us assume lim sup j→∞

so

⎡

aj bj

and

⎤

1

⎢ U=⎣ ∗

0 such that |γ n | ≥ εn (n ∈ N). This allows one to exclude the existence of nontrivial actions of nonuniform lattices thanks to the presence of unipotent elements; see [34]. But it also has different applications: for instance, any ﬁnitely generated nilpotent subgroup of Ham(, ω) is in fact abelian. Quasimorphisms and Bounded Cohomology

Recall that a quasi morphism on a group is a map φ : → R for which there exists a constant C > 0 such that |φ(xy) − φ(x) − φ(y)| ≤ C for any x, y ∈ . The quasi morphism φ is called homogeneous if moreover it satisﬁes φ(x n ) = nφ(x) (n ∈ Z, x ∈ ); that is, φ is a true homomorphism when restricted to cyclic subgroups of . For any quasimorphism φ, the limit φ(x n ) n→∞ n

φh (x) = lim

exists. The map φh is the unique homogeneous quasimorphism such that φ − φh is bounded. We will denote by QM(, R) the vector space of all quasimorphisms on , and by QMh (, R) the subspace of homogeneous quasimorphisms. For more details on quasi morphisms and bounded cohomology see [4, 6] as well as Gromov’s seminal paper [30]. We will return to the link between quasi morphisms and bounded cohomology in the next section. According to a result by Burger and Monod (see [7] or [8]), any homogeneous quasimorphism on a higher-rank lattice as above is a homomorphism, and hence is identically zero since is a Kazhdan group. On the other hand, the group of Hamiltonian diffeomorphisms of a closed surface (which is simple and hence admits no nontrivial homomorphism to R [3]) admits many quasimorphisms: Gambaudo and Ghys [25] proved that the vector space QMh (Ham(, ω), R) is inﬁnite-dimensional. Their proof consists in constructing explicitly a family of linearly independent homogeneous quasimorphisms. More constructions of quasimorphisms on groups of Hamiltonian diffeomorphisms can be found in the work of Entov and Polterovich [13] and of the author [40]. It is now well known (see [29]) that one could try to use this contrast between the lattice and groups of Hamiltonian diffeomorphisms of surfaces to attack the problem of the existence of nontrivial homomorphisms ρ : → Ham(, ω). According to the discussion above, for any homogeneous

212 / p. py

quasimorphism φ : Ham(, ω) → R the invariant φ ◦ ρ : → R vanishes identically. This should give some constraints on the homomorphism ρ. This circle of ideas is nicely discussed in Ghys’s survey [29]. Here, we will explain how one can use more subtly the vanishing results of Burger and Monod in bounded cohomology to obtain much more precise constraints on the homomorphism ρ. Namely, we will use the vanishing of certain bounded cohomology groups with unitary coefﬁcients. Note that the use of bounded cohomology for the study of group actions is not new. Bounded cohomology already appears in the study of groups acting on the circle; see [7, 28].

2. Vanishing Results and Consequences We begin with a brief reminder on the second bounded cohomology group of a discrete group , with unitary coefﬁcients (see for instance [36] for more details). We consider a unitary representation π of , that is, a homomorphism from to the group of unitary operators of a Hilbert space H . We will write ||v|| for the norm of a vector v ∈ H . All the unitary representations we will consider are obtained in the following way: the group is acting by measurepreserving transformations on a probability space (X , μ) and we consider the representation π of on the Hilbert space H = L2 (X , μ) of square integrable functions on X , deﬁned by π(γ )( f ) = f ◦ γ −1 (f ∈ L2 (X , μ)). We will say that a map c : j → H is bounded if the quantity |c|∞,j :=

sup

||c(γ1 , . . . , γj )||

(γ1 ,...,γj )∈j

is ﬁnite. The space Z 2 (, π) of 2-cocycles on with values in H is the space of maps c : 2 → H such that: π(γ1 )(c(γ2 , γ3 )) − c(γ1 γ2 , γ3 ) + c(γ1 , γ2 γ3 ) − c(γ1 , γ2 ) = 0. We will denote by Zb2 (, π) ⊂ Z 2 (, π) the subspace of bounded 2-cocycles. In the same way, the space B1 (, π) ⊂ Z 2 (, π) of coboundaries consists of the maps c : 2 → H that satisfy the equation c(γ1 , γ2 ) = π (γ1 )(v(γ2 )) + v(γ1 ) − v(γ1 γ2 ) for some map v : → H . The subspace of coboundaries c for which the map v in the equation above can be chosen bounded will be denoted by Bb1 (, π). The second bounded cohomology group of with coefﬁcients in H is the quotient Hb2 (, π) = Zb2 (, π)/Bb1 (, π),

some remarks on area-preserving actions of lattices / 213

and the second usual cohomology group of with coefﬁcients in H is the quotient H 2 (, π) = Z 2 (, π)/B1 (, π). Of course, there is a natural map from Hb2 (, π) to H 2 (, π) that sends the class of a bounded cocycle to its usual cohomology class. We will denote by EHb2 (, π) the kernel of this map. In the following proposition C (, H ) is the space of all maps from to H , Cb (, H ) the subspace of those maps that are bounded and d1 u is the coboundary of a map u ∈ C (, H ): d1 u(γ1 , γ2 ) = π (γ1 )(u(γ2 )) + u(γ1 ) − u(γ1 γ2 ).

proposition 2.1. The space EHb2 (, π) is isomorphic to {u ∈ C (, H ), |d1 u|∞,2 < ∞}/ Z 1 (, π) + Cb (, H ) . Proof. If u ∈ C (, H ) is such that |d1 u|∞,2 < ∞, the coboundary d1 u of u is a bounded 2-cocycle, hence deﬁnes a class [d1 u] ∈ Hb2 (, π) that is obviously trivial in usual cohomology. We have thus constructed a map u → [d1 u] from {u ∈ C (, H ), |d1 u|∞,2 < ∞} to EHb2 (, π) that is (by definition) surjective. Let us examine its kernel. The class [d1 u] vanishes in bounded cohomology precisely if there exists a bounded map v : → H such that d1 u = d1 v. In that case w = u − v is a 1-cocycle and we have u = w + v ∈ Z 1 (, π) + Cb (, H ). When H = R, with the trivial representation π0 of , the previous proposition simply proves the following classical fact [4, 6]: the group EHb2 (, π0 ) is isomorphic to the quotient QM(, R)/ Hom(, R) ⊕ Cb (, R) . Note that this last quotient is also isomorphic to the space QMh (, R)/Hom (, R). Let us come back to the case of a higher-rank lattice , as in the introduction. In that case, we have already said that the group EHb2 (, π0 ) is trivial, according to a result of Burger and Monod [7, 8]. In fact, their result is much more general: they prove that the group EHb2 (, π ) vanishes, for any unitary representation π of on a Hilbert space H (the result even holds for more general Banach spaces, but we will not really discuss this here). This can be seen as a strengthening of property (T): the group has the property that for any unitary representation π, the group H 1 (, π ) as well as the group EHb2 (, π ) vanishes. Following Monod [36], we will say that a discrete group with this property has property (TT). We now discuss the dynamical consequences of property (TT).

214 / p. py

Suppose that (X , μ) is a measure-preserving action of on a probability space. We will say that a map u : → L2 (X , μ) is a quasicocycle if there exists a constant C > 0 such that |u(γ1 γ2 ) − π(γ1 )(u(γ2 )) − u(γ1 )| ≤ C almost everywhere, for all γ1 , γ2 ∈ . We will give some examples of quasicocycles deﬁned on groups of Hamiltonian diffeomorphisms of surfaces in the next section. According to the subadditive ergodic theorem (see [32] for instance), if u is a quasicocycle and γ ∈ , the sequence of functions u(γ n ) n converges almost everywhere to a function u (γ ). The reader should observe that since the action of on X is measure preserving, the map u(γ )dμ γ → φ(γ ) = X

is a quasimorphism on . It is not difﬁcult to check that the map γ → φu (γ ) = u (γ )dμ is the unique homogeneous quasimorphism at a bounded distance X from φ. Note that all the quasimorphisms deﬁned by Gambaudo and Ghys [25] and by the author [40] are obtained in this way: by integration of a quasicocycle.

remark 1 The quasimorphisms constructed by Entov and Polterovich [13] are deﬁned in a completely different way; see [39] for a survey. It is not a priori clear how to relate them (in a nontrivial way) to some bounded cohomology class with nontrivial coefﬁcients. remark 2 A weaker deﬁnition of quasicocycle already appears in the litterature; see [44]. There, a map u : → L2 (X , μ) is called a quasicocycle if |d1 u|∞,2 =

sup (γ1 ,γ2

)∈2

|d1 u(γ1 , γ2 )|L2

is ﬁnite. In our deﬁnition, we require the stronger condition that sup (γ1 ,γ2 u(γ n )

|d1 u(γ1 , γ2 )|L∞

)∈2

is ﬁnite, to ensure that ( n )n ≥0 converges almost everywhere for all γ . The notion of quasicocycle is also related to the notion of rough action (see [36], chapter V). We now see the advantage of using the full result of Burger and Monod: if is a group with property (TT) acting on X and u is a quasicocycle, the vanishing result with trivial coefﬁcients tells us that for any γ ∈ the integral

some remarks on area-preserving actions of lattices / 215

φu (γ ) =

u (γ )dμ X

is zero, while the full vanishing result tells us that the function u (γ ) is zero almost everywhere, as the following proposition shows.

proposition 2.2. Suppose has property (TT ). Then, for any measurepreserving action (X , μ) on a probability space and for any quasicocycle u : → L2 (X , μ) we have u (γ ) = 0, almost everywhere ( for all γ ∈ ). Proof. If u is a quasicocycle, the map d1 u is a bounded 2-cocycle. Since the group EHb2 (, π) is trivial, there exists, according to the previous proposition, a bounded map v : → L2 (X , μ) and a 1-cocycle w : → L2 (X , μ) such that u = v + w. We write D := supγ ∈ ||v(γ )||. Since has property (T), there exists a map ϕ ∈ L2 (X , μ) such that w(γ ) = ϕ ◦ γ − ϕ. We obtain: u(γ n ) ϕ ◦ γ n − ϕ v(γ n ) = + . n n n We already know that the sequence The function

u(γ n ) n

converges almost everywhere to u (γ ).

ϕ ◦γn −ϕ 1

= (ϕ ◦ γ − ϕ) ◦ γ j n n n−1 j=0

converges almost everywhere by Birkhoff ’s theorem; it is a classical lemma v(γ n ) in ergodic theory that its limit is 0 almost everywhere. Finally, n converges n v(γ ) almost everywhere to u (γ ). Since the norm of n in L2 (X , μ) is less than D n,

v(γ nk )

there exists a subsequence n that converges almost everywhere to 0. k Hence u (γ ) = 0 almost everywhere.

3. Some Examples We now describe some examples of concrete quasicocycles on groups of Hamiltonian diffeomorphisms of surfaces, taken from [17, 20, 24, 25, 33, 40]. These functions are constructed in the spirit of Schwartzman’s classical asymptotic cycle [42] or of Arnold’s asymptotic Hopf invariant; see [1]. The interested reader will ﬁnd many more examples in the work of Gambaudo and Ghys [25].

216 / p. py

Since we are dealing with diffeomorphisms that are isotopic to the identity, we will always make use of isotopies ( ft )t∈[0,1] from the identity 1l = f0 to a given diffeomorphism f = f1 and will consider various (kinds of) rotation numbers associated to the orbit ( ft (x)) of a point x or to a pair of orbits ( ft (x)), ( ft (y)) (x = y). To ensure that the number we deﬁne does not depend on the choice of the isotopy ( ft )t∈[0,1] but only on f we will often appeal to some results on the topology of the groups Ham(, ω) or Diff 0 (, ω); see [12].

example 1 [17, 24] We ﬁrst give an example of a 1-cocycle deﬁned on the group Diff c (D2 , ω) of area-preserving diffeomorphisms of the disc D2 = {(x, y) ∈ R 2 , |x|2 + |y|2 ≤ 1}, which coincide with the identity near the boundary. This group is torsion free (see [39]). Although there is no nontrivial homomorphism from a Kazhdan group to Diff c (D2 , ω) (thanks to Thurston’s stability theorem), it is worth studying this ﬁrst example. Consider a diffeomorphism f ∈ Diff c (D2 , ω) and choose an isotopy ( ft ) from f0 = 1l to f1 = f . For any two distinct points x, y in the disc, consider the nonzero vector ft (x) − ft (y) ∈ R 2 . When t goes from 0 to 1 the argument e2iπu(t) of this vector varies of a quantity anglef (x, y) := u(1) − u(0). This deﬁnes a continuous function anglef : D2 × D2 − → R, where = {(x, x), x ∈ D2 } is the diagonal. One easily establishes that this function does not depend on the choice of the isotopy ( ft )t∈[0,1] but only on f , and that it is bounded; see [24]. If f , g ∈ Diff c (D2 , ω) and ( ft ) and (gt ) are two isotopies from the identity to f and g, respectively, we can consider the isotopy (ht ) = (gt ) ∗ ( ft g) from the identity to fg. It allows us to establish the relation: anglefg (x, y) = angleg (x, y) + anglef (g(x), g(y)). Hence the map f → anglef −1 deﬁnes a 1-cocycle on the group Diff c (D2 , ω) with values in the space L2 (D2 × D2 , R). Problem. Suppose ⊂ Diff c (D2 , ω) is a ﬁnitely generated group. Can we give a condition on that ensures that the class of the cocycle angle is nonzero in H 1 (, L2 (D2 × D2 , μ2 ))? The following proposition shows that the class of the cocycle f → anglef −1 is always nonzero in the space H 1 (, C 0 (D2 × D2 − )), unless is trivial (here C 0 (D2 × D2 − ) is the space of continuous functions on D2 × D2 − ).

proposition 3.1. Fix a diffeomorphism f ∈ Diff c (D2 , ω). Assume that there exists a continuous function ϕ : D2 × D2 − → R such that anglef = ϕ ◦ f − ϕ. Then f is the identity.

some remarks on area-preserving actions of lattices / 217

Proof. We assume that f is not the identity. Then, it is enough to ﬁnd an integer n and two ﬁxed points x, y of f n such that anglef n (x, y) = 0. Indeed the relation anglef = ϕ ◦ f − ϕ implies anglef n = ϕ ◦ f n − ϕ. If ϕ is continuous, this forces the equality anglef n (x, y) = 0 for any pair (x, y) of distinct ﬁxed points of f n . We will see at the end of this section (see Lemma 3.2 and the paragraph following it) that if f is distinct from the identity there exists a ﬁxed point x0 and a point y0 = x0 such that the number (x0 , y0 ) := lim angle f

1

n→∞ n

anglef n (x0 , y0 )

exists and is nonzero. Consider now the compact annulus A obtained from the disc by blowing up the ﬁxed point x0 thanks to the action of the differential dfx0 . The diffeomorphism f naturally extends to a homeomorphism of A, still denoted f . Let F be the lift of f to the universal cover A of the annulus that pointwise ﬁxes the component of the boundary of A corresponding to the boundary of the disc. If z is in the interior of A (identiﬁed with the interior of the disc, minus the point x0 ), the limit lim

1

n→∞ n

anglef n (x0 , z),

if it exists, is the rotation number ρF (z) (determined by the lift F) of the point z in the annulus. We know that • the point y0 has a nonzero rotation number, and • points close to the boundary have zero rotation number.

Now, let us recall a result of Franks (this is corollary 2.4 of [19]). Let g : A → A be a homeomorphism isotopic to the identity and G be a lift of g to A. If x is a point of A we will denote by ρG (x) the rotation number of x determined by G, if it exists. Let K ⊂ A be a chain transitive compact invariant set for g. See [19] for the notion of chain transitivity; this is a very weak form of recurrence. Assume that there exist two points x1 and x2 in K whose rotation numbers ρG (x1 ) and ρG (x2 ) exist and satisfy ρG (x1 ) < ρG (x2 ). Then, for any rational p p number q ∈]ρG (x1 ), ρG (x2 )[, g has a periodic point with rotation number q . Let us apply the above result to f . Since f is area preserving, the chain transitivity hypothesis will be easily satisﬁed. Almost every point of the interior of A is recurrent, hence every point of A is nonwandering. In particular every point of A is chain recurrent. The compact set K = A is connected, f -invariant, and all its points are chain recurrent, hence it is chain transitive (this is proposition 1.2 in [19]). Franks’s result applies with K = A. The existence of the point y0 and of points with zero rotation number implies that there exists a periodic

218 / p. py

point y1 with nonzero rotation number. If the period of y1 is n, this exactly means that the quantity anglef n (x0 , y1 ) is nonzero.

example 2 [20, 33] If f is a Hamiltonian diffeomorphism of the torus, we have seen in the introduction that for any lift F : R 2 → R 2 of f , one has T2 (F(x) − x)dμ(x) ∈ Z2 . We can therefore choose a particular lift f∗ for f : the unique one for which T2 ( f∗ (x) − x)dμ(x) = 0. Since the map R2 x

→ R2 → f∗ (x) − x

is invariant under integral translations, there exists a map vf : T2 → R 2 such that vf (p(x)) = f∗ (x) − x (where p : R 2 → T2 is the natural projection). From the relation ( f ◦ g)∗ = f∗ ◦ g∗ , one deduces the cocycle relation vf ◦g = vg + vf ◦ g. The map f → vf −1 ∈ L2 (T2 , R 2 ) is therefore a 1-cocycle. Suppose now that ⊂ Ham(T2 , ω) is a ﬁnitely generated subgroup on which the previous cocycle is cohomologically trivial: there exists a measurable map ϕ : T2 → R 2 such that vf = ϕ ◦ f − ϕ ( f ∈ ). Recall ﬁrst that for almost every point x in the torus, we have lim inf ϕ( f n (x)) − ϕ(x) = 0. n

This is an easy consequence of Poincaré’s recurrence theorem and Lusin’s theorem. But we also have the following relation: f∗n (x) − x = ϕ( f n (p(x))) − ϕ(p(x)) (x ∈ R 2 ). Hence, we get that almost every point of the plane is recurrent for the dynamics of the diffeomorphism f∗ (for all f ∈ ). In fact the function ψ : R2 x

→ R2 → x − ϕ(p(x))

is invariant under the lifted action of on the plane (through the diffeomorphisms f∗ ). The domain ψ −1 ([0, 1[2 ) ⊂ R 2 , as well as its translates by integral vectors, has ﬁnite measure and is invariant under . Therefore the vanishing of the cohomology class of the cocycle f → vf −1 gives some constraints on the (measurable) dynamics of .

example 3 [25] We assume here that the genus of is greater than 1. Endow with a hyperbolic metric and identify its universal cover with the Poincaré disc . For each 1-form η on , we will deﬁne a quasicocycle uη with values in the space of continuous functions on . This quasicocycle is a true cocycle if η is closed, and a coboundary if η is exact.

some remarks on area-preserving actions of lattices / 219

If f : → is a Hamiltonian diffeomorphism, we will name f∗ : → the unique lift of f, which commutes with the action of the fundamental group of on . If x ∈ , one can consider the unique geodesic arc α( x ) between x and f∗ ( x ). Let η denote the lift of the 1-form η to . The map → R η x → α(x ) x ∈ is any lift of x, the descends to a function uη ( f , ·) on . If x ∈ and quantity uη ( fg, x) − uη (g, x) − uη ( f , g(x)) equals the integral of the 2-form d η on the geodesic triangle of with vertices x ), f∗ ◦ g∗ ( x ). Since the 2-form d η is bounded (it is invariant by the action x , g∗ ( of the fundamental group of ) this quantity is bounded by the norm of d η times the area of a hyperbolic triangle. We obtain uη ( fg, x) − uη (g, x) − uη ( f , g(x)) ≤ π · |dη|∞ . Hence the map f → uη ( f −1 , ·) is a quasicocycle. The reader will ﬁnd an alternative description of this quasicocycle in [39].

example 4 [25] We now give an example of a quasicocycle deﬁned on the group Diff 0 (T2 , ω). We will make use of the group structure of the torus. We ﬁx a point x∗ in T2 − {0} and choose a homogeneous quasimorphism φ : π1 (T2 − {0}, x∗ ) → R (there are many since π1 (T2 − {0}, x∗ ) is a free group; see [6] as well as [14] for generalizations). For each point v ∈ T2 − {0} we ﬁx a path (αv (t))t∈[0,1] from x∗ to v in T2 − {0} whose length is bounded for a Riemannian metric deﬁned on the compact surface with boundary T2 − {0} obtained by blowing up the origin on the torus. If ( ft )t∈[0,1] is an area-preserving isotopy on T2 and x and y two distinct points on T2 , we can consider the curve ft (x) − ft (y) ∈ T2 − {0}. We close it to form a loop α( f , x, y) = αx−y ∗ ( ft (x) − ft (y)) ∗ αf1 (x)−f1 (y) . Then, we deﬁne a function u(φ, f , ·, ·) on T2 × T2 − by u(φ, f , x, y) = φ(α( f , x, y)). One can easily check that this function does not depend on the choice of the isotopy ( ft )t∈[0,1] but only on f , is measurable, and bounded; see [25]. From the relation α( fg, x, y) = α(g, x, y) ∗ α( f , g(x), g(y))

220 / p. py

and the fact that φ is a quasimorphism, we deduce |u(φ, fg, x, y) − u(φ, g, x, y) − u(φ, f , g(x), g(y))| ≤ C. This means precisely that the map f → u(φ, f −1 , ·, ·) ∈ L2 (T2 × T2 , μ ⊗ μ) is a quasicocycle.

example 5 [40] Consider the 2-sphere, with an area form ω of total area 2. Let (S2 ) be the space of oriented tangent lines to S2 and p : (S2 ) → S2 the canonical projection. There is a natural action of the circle R/Z on (S2 ): if ∈ (S2 ) is a tangent line at x ∈ S2 , e2iπs · is the tangent line at x whose angle with is 2πs. Let X be the (periodic) vector ﬁeld that generates this action of R/Z on (S2 ). We can ﬁnd a 1-form α on (S2 ), invariant by rotations, such that α(X ) = 1 and dα = p∗ ω. This implies that α is a contact form whose Reeb vector ﬁeld is X . Let f be an area-preserving diffeomorphism of S2 and ( ft ) a Hamiltonian isotopy from the identity to f , generated by the Hamiltonian (Ht ). Let Xt be the symplectic gradient of Ht . The isotopy ( ft ) induces an isotopy on (S2 ) through the action of the differential dft : (S2 ) → (S2 ) of ft . The key point of our last example is that there exists a second isotopy on (S2 ) that lifts ( ft ). It is constructed as t be the horizontal lift of Xt to (S2 ), that is, the vector ﬁeld on follows. Let X (S2 ) deﬁned by the equations t ) = Xt . t ) = 0 and p∗ (X α(X t − (Ht ◦ p)X generates an isotopy θ( ft ) : The time-dependent vector ﬁeld X 2 2 (S ) → (S ) that preserves α and whose isotopy class depends only on the isotopy class of ( ft ); see [2]. We can now measure a kind of rotation number associated to any tangent line ∈ (S2 ). Indeed for all t, θ ( ft )() and dft () are tangent lines at the point ft (p()) ∈ S2 . Hence, we can write dft () = e2iπu(t) · θ ( ft )() where e2iπ u(t) is the angle between θ( ft )() and dft (). The quantity w( f , ) = u(1) − u(0) depends only on f and . Then w( f , ·) : (S2 ) → R is a continuous map that satisﬁes w( fg, ) = w(g, ) + w( f , dg()). This is not yet exactly what we need: since the action ( f , ) → df () of the group Ham(S2 , ω) on (S2 ) does not have an invariant measure we cannot consider the function w( f, ·) as a vector in some unitary representation of Ham(, ω). Nevertheless, we can prove the following inequality; see [40]: if 1

some remarks on area-preserving actions of lattices / 221

and 2 are two tangent lines at the same point x = p(1 ) = p(2 ) ∈ S2 , then |w( f , 1 ) − w( f , 2 )| ≤ 10. Thus, if we deﬁne w( f , x) = maxp()=x w( f , ), w( f , ·) is a bounded measurable function on S2 and we have |w( fg, x) − w(g, x) − w( f , g(x))| ≤ 30. The map f → w( f −1 , ·) is therefore a quasicocycle. Motivated by these examples and by the vanishing results of the previous section, we can raise the following question. Question 1

If f : → is a Hamiltonian diffeomorphism of a closed surface, distinct from the identity, can we ﬁnd a quasicocycle u deﬁned on the group Ham(, ω), with values in the space L2 (, μ), or L2 ( × , μ ⊗ μ), or even L2 ( n , μn ) for some integer n, such that the function u ( f ) is not zero almost everywhere? In other words, can we ﬁnd a set of positive measure (in or × . . .) of points for which some kind of rotation number or asymptotic linking number is nonzero? A positive answer to this question would yield a positive answer to Zimmer’s conjecture. Note, however, that we should certainly add some hypothesis on f to get a positive answer. For instance, if f is a rotation of the sphere, u( f ) vanishes for all known quasicocycles. It is probably possible to construct more subtle counterexamples. Assume for instance that we can construct a Hamiltonian diffeomorphism f with the following two properties: f is weakly mixing, and for some algebraic reason, φ( f ) = 0 for any homogeneous quasimorphism deﬁned on the group of Hamiltonian diffeomorphisms. In that case, for any quasicocycle u with values in L2 ( n , μn ) one has u ( f ) = 0 almost everyn where. Indeed, since the diagonal action of f on is ergodic for any n, one has u ( f ) = φu ( f ) = 0 almost everywhere on n . Note, however, that this kind of counterexample cannot appear inside the image ρ() of a higher-rank lattice, since all the diffeomorphisms in the group ρ() have discrete spectrum. We now give two small hints that tend to show that a nontrivial Hamiltonian diffeomorphism of a compact surface contains somewhere a set of positive measure of points that “rotate around each other." Consider ﬁrst a diffeomorphism f ∈ Diff c (D2 , ω), distinct from the identity. We write (x, y) = lim angle f

1

n→∞ n

anglef n (x, y)

222 / p. py (x, y) when this limit exists. Note that, according to Birkhoff’s theorem, angle f 2 2 exists for μ ⊗ μ-almost every point (x, y) ∈ D × D , and if x0 is a ﬁxed point (y, x0 ) exists for μ-almost every point y ∈ D2 . A result of Viterbo of f , angle f [45] implies that there exists a ﬁxed point x0 ∈ D2 for f and a Borel set B ⊂ D2 of positive measure such that (y, x0 ) = 0, angle f

for all y ∈ B. We now explain this fact. Recall ﬁrst that one can deﬁne the symplectic action A(x0 ) of a ﬁxed point x0 of f in the following way. If ( ft )t∈[0,1] is a Hamiltonian isotopy from the identity to f , generated by a Hamiltonian (Ht ) one deﬁnes 1 A(x0 ) = λ+ Ht ( ft (x))dt, γ

0

where γ is the loop ( ft (x0 ))t∈[0,1] and λ is a primitive of the area form ω on D2 . This number does not depend on the choice of the isotopy ( ft ) from the identity to f . The next lemma, which I learned from Patrice Le Calvez, relates the symplectic action of a ﬁxed point to the cocycle anglef .

lemma 3.2. For any ﬁxed point x0 of f one has (y, x0 )dμ(y). A(x0 ) = anglef (y, x0 )dμ(y) = angle f D2

D2

Proof. To compute the action A(x0 ), we can choose an isotopy ( ft )t∈[0,1] such that ft (x0 ) = x0 for all t. We denote by Xt the vector ﬁeld that generates the isotopy and Ht the corresponding compactly supported Hamiltonian. Let dθ be a closed 1-form on the annulus D2 − {x0 } whose integral over a generator of the ﬁrst homology of the annulus equals 1. The integral of the function 1 anglef (·, x0 ) over D2 equals D2 0 dθ(Xt )dt ω. Using the fact that the 3-form dθ ∧ ω is 0 we get dθ(Xt ) ω = dθ ∧ ω(Xt , ·) = −d(Ht dθ). Choose now a small disc Dε around x0 . We have 1 lim D2 −Dε 0 dθ(Xt )dt ω D2 anglef (y, x0 )dμ(y) = ε→0 1 = lim ∂Dε 0 Ht dt dθ . ε→0

Here we have used Stokes’s theorem and the fact that Ht vanishes near the boundary of D2 . The last limit in the equality above coincides with the action

1

some remarks on area-preserving actions of lattices / 223

A(x0 ) = 0 Ht (x0 )dt. We have therefore proved the ﬁrst equality of the lemma. The second equality follows from Birkhoff’s theorem.

Now, using symplectic methods, Viterbo shows that if f is distinct from the identity, there exists a ﬁxed point x0 with nonzero action. Thanks to the interpretation of the action of x0 as the average linking number around x0 , we is nonzero on the deduce the existence of the set B. We get that the map angle f 2 2 set B × {x0 } ⊂ D × D . However, from our point of view, it would be more natural to obtain a set of positive measure in D2 × D2 for the product measure. Second, we consider the simplest examples of Hamiltonian diffeomorphisms: integrable ones. For instance, let us take a smooth function H : t : T2 → T2 . Assume T2 → R and consider the associated Hamiltonian ﬂow ϕH that H is nonconstant. Then, we will show that there is a quasicocycle u on 1 ) is Ham(T2 , ω) with values in L2 (T2 , R 2 ) or in L2 (T2 × T2 , R) such that u (ϕH not zero almost everywhere (for the measure μ on T2 in the ﬁrst case, for the measure μ ⊗ μ in the second one). As H is nonconstant there is an embedding of the annulus i : A = S1 × [0, 1] → T2 t reads ϕ t (i(θ , s)) = i(θ + tϑ(s), s) for a nonvanishing on which the ﬂow ϕH H function ϑ : [0, 1] → R ∗ . To prove this fact, one just considers a connected t component C of a regular level of H. In a neighborhood of C all orbits of ϕH are periodic; hence C is contained in an annulus that is invariant under the t and in which each orbit is periodic. From this one easily deduces ﬂow ϕH the existence of the embedding i. We write ˆi : R × [0, 1] → R 2 for a lift of the map i. Assume ﬁrst that the annulus A is essential in T2 . In that case there exists a nonzero vector w ∈ Z2 such that ˆi(θ + k, s) = ˆi(θ , s) + kw. Consequently, if vϕ 1 is the cocycle from H Example 2, we have vϕHn (i(θ , s)) → ϑ(s)w = 0. n→∞ n

Assume now that the annulus A is inessential. We will make use of the quasicocycle from Example 4. We can assume that the curve i(S1 × {0}) bounds an embedded disc D that is disjoint from i(S1 ×]0, 1]) (up to changing the parameter s by 1 − s). Assume that s2 > s1 and consider the trajectories of t (i(θ , s )) − the two points i(θ1 , s1 ) and i(θ2 , s2 ). The family of curves (ϕH 2 2 t ϕH (i(θ1 , s1 )))t∈[0,n] winds around 0 in the disc D ∪ i(A) (see Figure 1). The loops n , i(θ , s ), i(θ , s )) are therefore almost equal to a power of the commutator α(ϕH 2 2 1 1 [a, b], where a and b are the standard generators of the group π1 (T2 − {0}, x∗ ).

224 / p. py

Fig. 1. Linking of points in an inessential annulus on the torus.

More precisely, we can write n α(ϕH , i(θ2 , s2 ), i(θ1 , s1 )) = αn ∗ [a, b][nϑ(s2 )] ∗ βn

where αn and βn stay in a ﬁxed ﬁnite subset of π1 (T2 − {0}, x∗ ) and [nϑ(s2 )] is n , i(θ , s ), i(θ , s )) − the integer part of nϑ(s2 ). From this we deduce that u(φ, ϕH 2 2 1 1 [nϑ(s )] 2 φ([a, b] ) is bounded independently of n. Therefore, for any homogeneous quasimorphism φ on π1 (T2 − {0}, x∗ ) we have n , i(θ , s ), i(θ , s )) u(φ, ϕH 2 2 1 1 → ϑ(s2 )φ([a, b]). n To conclude we now choose a quasimorphism φ such that φ([a, b]) is nonzero and observe that the set {(i(θ1 , s1 ), i(θ2 , s2 )), s2 > s1 } ⊂ T2 × T2 has positive μ ⊗ μ-measure. Hence we have proved:

theorem Let ρ : → Ham(T2 , ω) be a homomorphism, where is a higherrank lattice, as above. If a diffeomorphism f ∈ ρ() is contained in a Hamiltonian ﬂow, then f is the identity. In fact, this result could have been deduced directly from the existence of Zimmer’s invariant Riemannian metric. Indeed, in the situation above, one can always ﬁnd an embedding i such that the function ϑ is not constant. This implies that for an open set of points x the sequence of differentials n (x)) (dϕH n≥0 tends to inﬁnity. This contradicts the existence of Zimmer’s metric. Yet, the above proof was presented in order to illustrate the possible use of the quasicocycles that we described. Also, it is not difﬁcult to establish a similar result on a surface of genus greater than 1, or on the sphere if we assume that the ﬂow is not conjugated to a 1-parameter group of rotations; see [41]. 4. Final Remarks We have already mentioned that, according to a result of Zimmer, the group ρ() preserves (almost everywhere) a measurable Riemannian metric. In the

some remarks on area-preserving actions of lattices / 225

case where the surface is the torus T2 , we can slightly improve this fact: the group ρ() preserves a pair (X 1 , X 2 ) of linearly independent measurable vector ﬁelds. If g is a Hamiltonian diffeomorphism of T2 , let us consider a Hamiltonian isotopy (gt ) from the identity to g. We identify the differential dgt (x) of the diffeomorphism gt at a point x with a 2 × 2 matrix. Hence, the path

2 (R) in the universal cover of of matrices dgt (x) deﬁnes an element dg(x) ∈ SL SL2 (R), which depends only on g and x. The map Ham(T2 ) × T2 (g, x)

2 (R) → SL → dg(x)

2 (R) to SL2 (R), the map (g, x) → is a cocycle. If p denotes the projection from SL dg(x) = p(dg(x)) is the usual derivative cocycle. Suppose → Ham(T2 , ω) is a Kazhdan group and consider the restriction of the previous cocycle to . According to Zimmer [47, 49], there exists a measurable map ϕ : T2 → SL2 (R) such that ϕ(γ · x)−1 ◦ dγ (x) ◦ ϕ(x) ∈ SO(2),

2 (R) → R the translation almost everywhere (for γ ∈ ). We denote by τ : SL

number associated to the action of SL2 (R) on the line (see [28]) and by T

2 (R). Let

2 (R) be a measurable the generator of the center of SL ϕ : T2 → SL lift of ϕ chosen in such a way that the map x → τ ( ϕ (x)) is bounded. This

2 (R), k ∈ Z). The is always possible since one has τ (a · T k ) = τ (a) + k (a ∈ SL

(x) ◦ map (γ , x) → ϕ (γ · x)−1 ◦ dγ ϕ (x) is a cocycle whose image is contained

2 (R), which is canonically isomorphic in the inverse image of SO(2) in SL to R.

(x) ◦ lemma 4.1. For each γ ∈ the map x → sγ (x) := ϕ (γ · x)−1 ◦ dγ ϕ (x) ∈ R 1 is bounded. Hence the cocycle γ → sγ determines a class in H (, L2 (T2 , R)).

2 (R) is the Proof. Since the restriction of τ to the inverse image of SO(2) in SL identity, it is enough to show that the map x → τ (sγ (x)) is bounded. Since the two maps x → τ ( ϕ (x)) and x → τ ( ϕ (γ · x)−1 ) are bounded thanks to the choice of the lift ϕ and since τ is a quasimorphism, it is enough to check

(x)) is bounded. But this is obvious since this is a that the map x → τ (dγ continuous map from T2 to R. From the lemma, we deduce that there exists a measurable (in fact L2 ) map ψ : T2 → R such that sγ (x) = ψ(γ · x) − ψ(x). Coming back to SL2 (R), we obtain

226 / p. py

ϕ(γ · x)−1 ◦ dγ (x) ◦ ϕ(x) = e2iπ (ψ(γ (x))−ψ(x)) , almost everywhere, where e2iπ u stands for the rotation of angle 2πu in SL2 (R). This tells us that the derivative cocycle is cohomologous to the constant cocycle (equal to the identity), which is equivalent to the existence of the vector ﬁelds X 1 and X 2 . Although the results we have discussed here give more evidence toward Zimmer’s conjecture, it seems difﬁcult to exploit them. In particular, the quest for a set of positive measure of points that “rotates" under the action of a diffeomorphism seems delicate. Note also that we have simply used global cohomological properties of the lattice to give some constraints on the dynamics of each single diffeomorphism in the group ρ() but I was not able to investigate the dynamics of the whole group . Rather than studying measure theoretical properties, it might be useful to study the topological dynamics of the action. In this direction, the work of Calegari [9] contains many promising ideas.

acknowledgments. This work is part of my thesis realized at École Normale Supérieure de Lyon. I would like to thank Étienne Ghys for all the discussions we had around the problems discussed in this text and for his warm encouragement. I would also like to thank François Béguin, Patrice Le Calvez, Frédéric Le Roux, and Leonid Polterovich for the numerous conversations we had around Zimmer’s conjecture. This text was completed while I was enjoying the hospitality of Tel-Aviv University. This work was also partly supported by the French Agence Nationale de la Recherche. Finally, I would like to thank the referee for his or her constructive comments on this text. References [1] [2]

[3] [4] [5]

V. I. Arnold and B. A. Khesin, Topological methods in hydrodynamics, Applied Mathematical Sciences 125, Springer-Verlag, New York, (1998). A. Banyaga, The group of diffeomorphisms preserving a regular contact form, Topology and algebra (Proc. Colloq., Eidgenöss. Tech. Hochsch., Zurich, 1977), 47–53, Monograph. Enseign. Math. 26, Univ. Genève (1978). A. Banyaga, The structure of classical diffeomorphism groups, Mathematics and its applications 400, Kluwer Academic Publishers Group, Dordrecht, (1997). C. Bavard, Longueur stable des commutateurs, Enseign. Math. (2) 37, No. 1–2 (1991), 109–150. A. Borel, Compact Clifford-Klein forms of symmetric spaces, Topology 2 (1963), 111–122.

some remarks on area-preserving actions of lattices / 227 [6]

[7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20]

[21] [22] [23] [24] [25] [26] [27] [28]

R. Brooks, Some remarks on bounded cohomology, Riemann Surfaces and Related Topics: Proceedings of the 1978 Stony Brook Conference (State Univ. New York, Stony Brook, NY, 1978), 53–63, Ann. of Math. Stud. 97, Princeton University Press, Princeton, NJ, (1981). M. Burger and N. Monod, Bounded cohomology of lattices in higher rank Lie groups, J. Eur. Math. Soc. (JEMS) 1, No. 2 (1999), 199–235. M. Burger and N. Monod, Continuous bounded cohomology and applications to rigidity theory, Geom. Funct. Anal. 12, No. 2 (2002), 219–280. D. Calegari, Circular groups, planar groups, and the Euler class, Proceedings of the Casson Fest, Geom. Topol. Monogr. 7, Geom. Topol. Publ., Coventry (2004), 431–491. S. Cantat, Version kählérienne d’une conjecture de Robert J. Zimmer, Ann. Sci. École Norm. Sup. (4) 37, No. 5 (2004), 759–768. J. Déserti, Groupe de Cremona et dynamique complexe: une approche de la conjecture de Zimmer, Int. Math. Res. Not. (2006). C. J. Earle and J. Eells, A ﬁbre bundle description of Teichmüller theory, J. Diff. Geometry 3, (1969), 19–43. M. Entov and L. Polterovich, Calabi quasimorphism and quantum homology, Int. Math. Res. Not., No. 30 (2003), 1635–1676. D. Epstein and K. Fujiwara, The second bounded cohomology of word-hyperbolic groups, Topology 36, No. 6 (1997), 1275–1289. B. Farb and H. Masur, Superrigidity and mapping class groups, Topology 37, No. 6 (1998), 1169–1176. B. Farb and P. Shalen, Real-analytic actions of lattices, Invent. Math. 135, No. 2 (1999), 273–296. A. ÊFathi, Transformations et homéomorphismes préservant la mesure. Systèmes dynamiques minimaux, Thèse, Orsay, (1980). D. Fisher, Talk at the conference “Geometry, Rigidity, and Group Actions," in honor of R. J. Zimmer’s 60th birthday, University of Chicago (September 2007). J. Franks, Recurrence and ﬁxed points of surface homeomorphisms, Ergodic Theory Dynam. Systems 8∗ , Charles Conley Memorial Issue (1988), 99–107. J. Franks, Rotation vectors for surface diffeomorphisms, Proceedings of the International Congress of Mathematicians, Vol.1, 2, (Zürich, 1994), Birkhäuser, Basel (1995), 1179– 1186. J. Franks and M. Handel, Periodic points of Hamiltonian surface diffeomorphisms, Geom. Topol. 7, (2003), 713–756. J. Franks and M. Handel, Area preserving group actions on surfaces, Geom. Topol. 7, (2003), 757–771. J. Franks and M. Handel, Distortion elements in group actions on surfaces, Duke Math. J. 131, No. 3 (2006), 441–468. J.-M. Gambaudo and É. Ghys, Enlacements asymptotiques, Topology 36, No. 6 (1997), 1355–1379. J.-M. Gambaudo and É. Ghys, Commutators and diffeomorphisms of surfaces, Ergodic Theory Dynam. Systems 24, No. 5 (2004), 1591–1617. É. Ghys, Sur les groupes engendrés par des difféomorphismes proches de l’identité, Bol. Soc. Brasil. Mat. (N.S.) 24, No. 2 (1993), 137–178. É. Ghys, Actions de réseaux sur le cercle, Invent. Math. 137, No. 1 (1999), 199–231. É. Ghys, Groups acting on the circle, Enseign. Math. (2) 47, No. 3–4 (2001), 329–407.

228 / p. py [29] É. Ghys, Knots and dynamics, International Congress of Mathematicians, Vol. I, Eur. Math. Soc., Zürich (2007), 247–277. [30] M. Gromov, Volume and bounded cohomology, Inst. Hautes Études Sci. Publ. Math. 56 (1983), 5–99. [31] P. de la Harpe and A. Valette, La propriété (T ) de Kazhdan pour les groupes localement compacts (avec un appendice de Marc Burger), Astérisque No. 175 (1989). [32] J. F. C. Kingman, Subadditive processes, École d’Été de Probabilités de Saint-Flour, V 1975, Lecture Notes in Math. 539, Springer, Berlin, (1976), 167–223. [33] P. Le Calvez, Identity isotopies on surfaces, Dynamique des difféomorphismes conservatifs des surfaces: un point de vue topologique, Panor. Synthèse 21, Soc. Math. France, Paris (2006), 105–143. [34] A. Lubotzky, S. Mozes, and M. S. Raghunathan, The word and Riemannian metrics on lattices of semisimple groups, Inst. Hautes Études Sci. Publ. Math., No. 91 (2000), 5–53. [35] D. McDuff and D. Salamon, Introduction to symplectic topology, Second edition, Oxford Mathematical Monographs, Clarendon Press, Oxford University Press, New York, (1998). [36] N. Monod, Continuous bounded cohomology of locally compact groups, Lecture Notes in Mathematics 1758, Springer-Verlag, Berlin, (2001). [37] A. Navas, Actions de groupes de Kazhdan sur le cercle, Ann. Sci. École Norm. Sup. (4) 35, No. 5 (2002), 749–758. [38] L. Polterovich, Growth of maps, distortion in groups and symplectic geometry, Invent. Math. 150, No. 3 (2002), 655–686. [39] L. Polterovich, Floer homology, dynamics and groups, Morse theoretic methods in nonlinear analysis and in symplectic topology, NATO Sci. Ser. II Math. Phys. Chem. 217, Springer, Dordrecht (2006), 417–438. [40] P. Py, Quasi-morphismes et invariant de Calabi, Ann. Sci. École Norm. Sup. (4) 39, No. 1 (2006), 177–195. [41] P. Py, Quasi-morphismes et difféomorphismes hamiltoniens, Ph.D thesis, École Normale Supérieure de Lyon (2008), available at http://tel.archives-ouvertes.fr/ tel- 00263607/fr/. [42] S. Schwartzman, Asymptotic cycles, Ann. of Maths (2) 66, (1957), 270–284. [43] S. Smale, Diffeomorphisms of the 2-sphere, Proc. Amer. Math. Soc. 10 (1959), 621–626. [44] A. Thom, Low degree bounded cohomology and L2 -invariants for negatively curved groups, Groups, Geometry and Dynamics 3 (2009), 343–358. [45] C. Viterbo, Symplectic topology as the geometry of generating functions, Math. Ann. 292, No. 4 (1992), 685–710. [46] D. Witte, Arithmetic groups of higher Q-rank cannot act on 1-manifolds, Proc. Amer. Math. Soc. 122, No. 2 (1994), 333–340. [47] R. J. Zimmer, Kazhdan groups acting on compact manifolds, Invent. Math. 75, No. 3 (1984), 425–436. [48] R. J. Zimmer, Actions of semisimple groups and discrete subgroups, Proceedings of the International Congress of Mathematicians, Vol. 1, 2 (Berkeley, CA., 1986), 1247–1258, Amer. Math. Soc., Providence, RI (1987). [49] R. J. Zimmer, Spectrum, entropy, and geometric structures for smooth actions of Kazhdan groups, Israel J. Math. 75, No. 1 (1991), 65–80.

7

raul quiroga-barranco

ISOMETRIC ACTIONS OF SIMPLE GROUPS AND TRANSVERSE STRUCTURES: THE INTEGRABLE NORMAL CASE with gratitude to robert j. zimmer on the occasion of his 60 th birthday Abstract For actions with a dense orbit of a connected noncompact simple Lie group G, we obtain some global rigidity results when the actions preserve certain geometric structures. In particular, we prove that for a G-action to be equivalent to 1 on a space of the form (G × K\H)/ , it is necessary and sufﬁcient for the G-action to preserve a pseudo-Riemannian metric and a transverse Riemannian metric to the orbits. A similar result proves that the G-actions on spaces of the form (G × H)/ are characterized by preserving transverse parallelisms. By relating these techniques to the notion of the algebraic hull of an action, we obtain inﬁnitesimal Lie algebra structures on certain geometric manifolds acted upon by G.

1. Introduction In the rest of this work, let G be a connected noncompact simple Lie group with Lie algebra g and M a smooth connected manifold acted upon smoothly by G. There are several examples of such actions that preserve a ﬁnite volume. Some of the most interesting are obtained from Lie group homomorphisms G → L where L is a connected Lie group that admits a lattice . The relevant G-action is then given on the double coset K \L/ , where K is some compact subgroup that centralizes G. Robert Zimmer proposed in [15] to study the ﬁnite measure-preserving G-actions on M and determine to what extent these are given by such double cosets. To understand how to tackle Zimmer’s program, it has been proved very useful to consider G-actions preserving a suitable geometric structure (see [2] and [18]). This is a natural condition since the above double coset examples This work was supported by SNI-Mexico and the grants Conacyt 44620 and Concyteg 07-02-K662-091.

229

230 / r. quiroga-barranco

carry a G-invariant pseudo-Riemannian metric when L is semisimple. Such metric comes from the bi-invariant metric on L obtained from the Killing form of its Lie algebra. The properties of the G-orbits of ﬁnite measure-preserving G-actions are reasonably well known. For example, such G-actions are known to be everywhere locally free when they preserve suitable pseudo-Riemannian metrics (see [2], [13], and [14]). Also in such case, the metric restricted to the orbits can be described precisely (see Lemma 2.6). However, there is still a lack of knowledge of the properties of a manifold, acted upon by G, in the transverse direction to the orbits. In this work, we propose to study the G-actions on M by emphasizing the need to understand the properties of the transverse to the G-orbits. For this, we will be dealing with G-actions that have a dense orbit and preserve a ﬁnite-volume pseudo-Riemannian metric. As observed above, in this case the results from [13] show that the G-orbits deﬁne a foliation, which from now on we will denote with O. By assuming that the orbits are nondegenerate for the pseudo-Riemannian metric on M, we can consider the normal bundle T O⊥ to the orbits as realizing the transverse direction in M. In this respect, we obtain the following structure theorem for G-actions on M when this normal bundle is integrable.

theorem 1.1. Suppose that the G-action on M has a dense orbit and preserves a ﬁnite-volume complete pseudo-Riemannian metric. If the G-orbits in M are nondegenerate and the normal bundle to the orbits T O⊥ is integrable, then there exist: → M to which the G-action lifts, 1) an isometric ﬁnite covering map M and 2) a simply connected complete pseudo-Riemannian manifold N, 3) a discrete subgroup ⊂ G × Iso(N), is G-equivariantly isometric to (G × N)/ . such that M We observe that in the double coset examples K \L/ as above with L semisimple and with the metric coming from the Killing form, the G-orbits are always nondegenerate. Besides that, we prove that for a general G-action on M, the orbits are always nondegenerate when dim (M) < 2 dim (G) (see Lemma 2.7). Also, by developing criteria for the normal bundle T O⊥ to be integrable, we obtain some results where the conclusion of Theorem 1.1 holds. For example, Corollary 3.1 ensures such conclusion when G has high enough real rank. Also, Corollary 3.2 does the same for manifolds M whose dimensions have a suitable bound in terms of G. In this last result we can even dispense with the assumption of having nondegenerate orbits by applying Lemma 2.7.

isometric actions of simple groups and transverse structures / 231

Without assuming in Theorem 1.1 that the G-action has a dense orbit it is still possible to draw some conclusions. More speciﬁcally, if we assume the rest of the hypotheses, a description of the universal covering space of the manifold as a warped product is obtained. This sort of result already appears in [3]. For a G-action on M preserving a pseudo-Riemannian metric, in [10] we considered a certain geometric condition between the metrics on G and M (the former for a bi-invariant metric) from which we concluded that M is, up to a ﬁnite covering space, a double coset of the form (G × K \H)/ . In other words, a double coset as above where G appears as a factor in L. One of the steps used in [10] to achieve this was to prove that the normal bundle T O⊥ is Riemannian. Considering the relevance we are giving to the transverse to the orbits, it is natural to determine the properties of the G-actions that preserve a transverse Riemannian structure. In this context, we obtain the following result, which proves that, up to a ﬁnite covering, the double cosets of the form (G × K \H)/ are characterized as those isometric G-actions that preserve a transverse Riemannian structure on the foliation O. We recall that a semisimple Lie group is isotypic if the complexiﬁcation of its Lie algebra is isomorphic to a sum of identical simple complex ideals.

theorem 1.2. If G is a connected noncompact simple Lie group acting faithfully on a compact manifold M, then the following conditions are equivalent: 1) There is a G-equivariant ﬁnite covering map (G × K \H)/ → M where H is a connected Lie group with a compact subgroup K and ⊂ G × H is a discrete cocompact subgroup such that G is dense in G × H. → M for which the G-action on M lifts to a 2) There is a ﬁnite covering map M with a dense orbit and preserving G-action on M • a pseudo-Riemannian metric for which the orbits are nondegenerate,

and • a transverse Riemannian structure for the foliation O by G-orbits.

Furthermore, if G has ﬁnite center and real rank at least 2, then we can assume in (1) of the above equivalence that G × H is semisimple isotypic with ﬁnite center and that is an irreducible lattice. Based on the previous result, we prove in Theorem 5.1 that the G-actions on M preserving a Lorentzian metric are, up to a ﬁnite covering, those given by double cosets (G × K \H)/ with G locally isomorphic to SL(2, R). We observe that this result improves a similar one found in [2] (see also the introduction of [4]).

232 / r. quiroga-barranco

Continuing with our study of transverse structures, we next consider isometric G-actions preserving a transverse parallelism for the foliation O. We prove that, up to a ﬁnite covering, such actions characterize the double cosets of the form (G × H)/ .

theorem 1.3. If G is a connected noncompact simple Lie group acting faithfully on a compact manifold M, then the following conditions are equivalent: 1) There is a G-equivariant ﬁnite covering map (G × H)/ → M where H is a connected Lie group and ⊂ G × H is a discrete cocompact subgroup such that G is dense in G × H. → M for which the G-action on M lifts to a 2)There is a ﬁnite covering map M G-action on M with a dense orbit and preserving • a pseudo-Riemannian metric for which the orbits are nondegenerate, and • a transverse parallelism for the foliation O by G-orbits. Furthermore, if G has ﬁnite center and real rank at least 2, then we can assume in (1) of the above equivalence that G × H is semisimple isotypic with ﬁnite center and that is an irreducible lattice. The notion of the algebraic hull of an action on a bundle, introduced by Zimmer, is a fundamental tool to understand G-actions as they relate to geometric structures. We recall that the algebraic hull is the smallest algebraic subgroup for which there is a measurable G-invariant reduction of the bundle being acted upon (see [17]). For our setup, where we are interested in the transverse to the orbits, it is then natural to consider the algebraic hull of the G-action on the bundle L(T O⊥ ) for an isometric G-action with nondegenerate orbits. Since a G-action of this sort preserves a pseudo-Riemannian metric on T O⊥ , the algebraic hull for L(T O⊥ ) is in this case a subgroup of O(p, q), for some (p, q). The next result shows that for weakly irreducible manifolds and when the algebraic hull for L(T O⊥ ) is the largest possible for this setup, the manifold being acted upon has inﬁnitesimally at some point the structure of a speciﬁc Lie algebra that contains g. By the last claim in the statement, such Lie algebra structure is nontrivially linked to the geometry of the manifold. For the explicit description of the Lie algebra structure obtained in this result we refer to Lemma 7.1 and Theorem 7.2. We recall that a pseudo-Riemannian manifold is weakly irreducible if the tangent space at some (and hence any) point has no proper nondegenerate subspaces invariant under the restricted holonomy group at that point. Also recall that every simple Lie group with a bi-invariant metric is weakly irreducible.

isometric actions of simple groups and transverse structures / 233

theorem 1.4. Suppose that G has ﬁnite center and real rank at least 2, and that the G-action on M preserves a ﬁnite-volume complete pseudo-Riemannian metric. Also assume that G acts ergodically on M and that the foliation by G-orbits is nondegenerate. Denote with L the algebraic hull for the G-action on the bundle L(T O⊥ ) and with l its Lie algebra. In particular, there is an embedding of Lie algebras l → so(p, q), where (p, q) is the signature of the metric of M restricted to T O⊥ . If this embedding is surjective and M is weakly irreducible, then the following holds: 1) G is locally isomorphic to SO0 (p, q) and dim (M) = (p + q)(p + q + 1)/2, and 2)for some x ∈ M the tangent space Tx M admits a Lie algebra structure isomorphic to either so(p, q + 1) or so(p + 1, q) such that Tx O is a Lie subalgebra isomorphic to g. Furthermore, with respect to the representation of Lemma 2.1, there is a Lie algebra of local Killing ﬁelds vanishing at x, which is isomorphic to g and acts nontrivially on Tx M by derivations of the Lie algebra structure given in (2). With the results developed in this work, we try to show the importance of considering transverse geometric structures to understand the actions of noncompact simple Lie groups. In fact, we believe that some form of the following conjecture could provide a geometric characterization of the double coset examples of G-actions mentioned above.

conjecture 1.5. Consider the double coset G-spaces of the form K \L/ , where L is a semisimple Lie group with an irreducible lattice and a compact subgroup K ; the G-action is then induced by a nontrivial homomorphism G → L whose image centralizes K . Then, a G-action on a manifold M is equivalent to such a double coset G-action for some L, , K if and only if the G-action on M: • has a dense orbit, • preserves a pseudo-Riemannian metric, and • preserves a transverse geometric structure to the orbits suitably related to the

geometry of GK \L. Note that for L = G × H and K a compact subgroup of H, we obtain quotients GK \L = K \H and G\L = H, which naturally carry a Riemannian metric and a parallelism, respectively. Thus Theorems 1.2 and 1.3 verify the conjecture in these cases. One of the main tools used to obtain our results is Gromov’s machinery on geometric G-actions (see [1], [2], [10], and [18]). Such machinery ensures the

234 / r. quiroga-barranco

existence of large families of local Killing ﬁelds. We develop these techniques in Section 2 emphasizing the fact that the Killing ﬁelds thus obtained yield g-module structures on the tangent spaces to M; in this respect, our main result is Proposition 2.3, which is due to Gromov [2] in the analytic/compact case (see also [18]) and was extended to the smooth/ﬁnite volume case in [1] and [10]. In Section 3, we prove that the latter impose restrictions strong enough to guarantee the integrability of the normal bundle under suitable conditions (Corollaries 3.1 to 3.3). We observe that Theorem 1.1 and its consequences obtained in Section 3 are in fact extensions of results obtained in [2]: theorem 5.3.E on page 129 of [2] states, under the assumptions of our Corollary 3.1, that M has a covering G-equivariantly diffeomorphic to G × N for some N. Note, however, that the result in [2] yields only a topological covering and does not further describe the covering group. Theorem 1.2 is obtained in Section 4 by applying the main results and arguments from [10]. As mentioned above, this yields in Section 5 our characterization of Lorentzian manifolds acted upon with a dense orbit by a simple noncompact Lie group. In Section 6 we establish the characterization of Gspaces of the form (G × H)/ provided by Theorem 1.3; for this, one of the main ingredients is given by Theorem 1.1. We observe that the arguments used in Section 6 are based on those found in [11]. To obtain Theorem 1.4 in Section 7, an application of Theorem 1.1 is used again. But now, the notion of the algebraic hull and the deep result about it found in [17] are also fundamental. Ultimately, this is somewhat to be expected, since a computation of the algebraic hull for frame bundles over M is in fact an important step to build Gromov’s machinery for geometric G-actions on M. This is evident in our proof of Proposition 2.3. We want to observe that the conclusion of Theorem 1.4 can be paraphrased by saying that the manifold M has inﬁnitesimally the structure of either SO(p, q + 1) or SO(p + 1, q). It remains the question as to whether or not M can be related more precisely to either of these groups. This problem will be pursued elsewhere (see [7]) by requiring some additional conditions. The author wishes to thank Uri Bader and Shmuel Weinberger for some fruitful conversations.

2. Killing Fields on Manifolds with a Simple Group of Isometries By theorem 4.17 from [13], if the G-action on M has a dense orbit and preserves a ﬁnite-volume pseudo-Riemannian metric, then the action is locally free and so the orbits deﬁne a foliation that we have agreed to denote with O. In this

isometric actions of simple groups and transverse structures / 235

case, it is well known that the bundle T O tangent to the foliation O is a trivial vector bundle isomorphic to M × g, under the isomorphism given by M × g → TO (x, X ) → Xx∗ . For every x ∈ M, this induces an isomorphism between the ﬁber Tx O and g, which we will refer to as the natural isomorphism. Furthermore, if we consider on M × g the product G-action, where the G-action on g is the adjoint one, then such isomorphism is G-equivariant. More precisely, we have dg(X ∗ ) = Ad(g)(X )∗ for every g ∈ G and X ∈ g. Note that for X in the Lie algebra of a group acting on a manifold, we denote with X ∗ the vector ﬁeld on the manifold whose 1-parameter group of diffeomorphisms is given by ( exp (tX ))t through the action on the manifold. For any given pseudo-Riemannian manifold N, we will denote with Kill(N, x) the Lie algebra of germs at x of local Killing vector ﬁelds deﬁned in a neighborhood of x, and with Kill0 (N, x) we will denote the Lie subalgebra consisting of those germs that vanish at x. The following result is a consequence of the Jacobi identity and the fact that the Lie derivative of a metric with respect to its Killing ﬁelds vanishes. In the rest of this work, for a vector space W with a nondegenerate symmetric bilinear form, we will denote with so(W ) the Lie algebra of linear maps on W that are skew-symmetric with respect to the bilinear form.

lemma 2.1. Let N be a pseudo-Riemannian manifold and x ∈ N. Then, the map λx : Kill0 (N, x) → so(Tx N) λx (Z)(v) = [Z, V ]x , where V is any vector ﬁeld such that Vx = v, is a well-deﬁned homomorphism of Lie algebras. From now on, for a given point x of a pseudo-Riemannian manifold, the map λx will denote the homomorphism from the previous lemma. For the proof of our next result we will present some facts about inﬁnitesimal automorphisms and Killing ﬁelds, and we refer to [1] for further details. The tangent bundle of order k of a manifold N is the smooth bundle T (k) N (k) whose ﬁber at x is the space Tx N of (k − 1)-jets at x of vector ﬁelds of N. For

236 / r. quiroga-barranco

every (local) diffeomorphism ϕ : N1 → N2 mapping x1 to x2 we have a linear isomorphism: (k)

(k)

Tx1 N1 → Tx2 N2 jxk−1 (X ) → jxk−1 (dϕ(X )) 1 2 that depends only on the jet jxk1 (ϕ). For N1 = N2 = N, x1 = x2 = x this yields (k)

the group Dx (N) of k-jets at x of local diffeomorphisms ﬁxing x, whose group structure is induced by the composition of maps. In the case when N = Rn and x = 0, we will denote this group with GL(k) (n). We also recall that the k-th order frame bundle over N is the smooth bundle L(k) (N) that consists of the k-jets at 0 of local diffeomorphisms Rn → N; any such a jet j0k (ϕ) thus (k)

(k)

deﬁnes a linear isomorphism T0 Rn → Tϕ(0) N. The structure group of L(k) (N) (k)

is GL(k) (n). With respect to vector ﬁelds, we denote with Dx (N) the space of k-jets at x of vector ﬁelds vanishing at x, and we use the special notation gl(k) (n) when N = Rn and x = 0. If N carries a pseudo-Riemannian metric, for every (k) x ∈ N, we will denote with Autk (N, x) the subgroup of Dx (N) consisting of those k-jets that preserve the metric up to order k at x. Correspondingly, for vector ﬁelds we denote with Killk (N, x) the space of k-jets at x of vector ﬁelds on N that preserve the metric up to order k at x; the subspace of those k-jets whose 0-jet vanishes is denoted with Killk0 (N, x). The next result provides a natural (k) representation of Dx (N) from which the Lie algebras of this group and of (k) Autk (N, x) are described in terms of Dx (N) and Killk0 (N, x), respectively; the proof is elementary, but it is detailed in sections 2 and 4 of [1].

lemma 2.2. For a smooth manifold N and any given point x ∈ N the following properties hold for every k ≥ 1: 1) The map (k)

(k)

x : Dx (N) → GL(Tx N) x ( jxk (ϕ))( jxk−1 (X )) = jxk−1 (dϕ(X )) is a Lie group monomorphism. 2)The assignment [ jxk (X ), jxk (Y )]k = −jxk ([X , Y ]) yields a well-deﬁned Lie alge(k) bra structure on Dx (N). 3)The map (k)

(k)

θx : Dx (N) → gl(Tx N) θx ( jxk (X ))( jxk−1 (Y )) = −jxk−1 ([X , Y ])

isometric actions of simple groups and transverse structures / 237 (k)

is a Lie algebra monomorphism for the Lie algebra structure on Dx (N) given (k) (k) by [·, ·]k . Furthermore, θx (Dx (N)) = Lie(x (Dx (N))). 4)If N has a pseudo-Riemannian metric, then we have θx (Killk0 (N, x)) = Lie(x (Autk (N, x))). In particular, with respect to the homomorphisms x and θx , the Lie algebra of Autk (N, x) is realized by Killk0 (N, x) with the Lie algebra structure given by [·, ·]k . The following result is due to Gromov in the analytic case (see [2]) and it was extended to the smooth case in [10]. We present here a fairly detailed proof based on the results from [1]. In congruence with our notation for M, in the for the in M rest of this work we will use O to denote the foliation by G-orbits lifted G-action on every covering space M of M; this will be so for any covering whether ﬁnite or not.

proposition 2.3. Suppose that the G-action on M has a dense orbit and preserves a ﬁnite-volume pseudo-Riemannian metric. Then, there exists a dense such that for every x ∈ S the following properties are satisﬁed. subset S ⊂ M x), which is an 1) There is a homomorphism of Lie algebras ρx : g → Kill(M, isomorphism onto its image ρx (g) = g(x). x); that is, every element of g(x) vanishes at x. 2) g(x) ⊂ Kill0 (M, 3)For every X , Y ∈ g we have [ρx (X ), Y ∗ ] = [X , Y ]∗ = −[X ∗ , Y ∗ ] in a neighborhood of x. In particular, the elements in g(x) and their corresponding local ﬂows preserve both O and T O⊥ in a neighborhood of x. induces a 4)The homomorphism of Lie algebras λx ◦ ρx : g → so(Tx M) g-module structure on Tx M for which the subspaces Tx O and Tx O⊥ are g-submodules. Proof. Since the proof builds on the notions and results found in [1], we will mostly follow its notation. We will be careful to deﬁne the objects considered but we refer to [1] for further details. For every k ≥ 1, let us denote with σ k : L(k) (M) → Qk the GL(k) (n)equivariant map that deﬁnes the k-th order extension of the geometric structure deﬁned by the pseudo-Riemannian metric on M. Consider the set G k = {jxk−1 (X ∗ ) : X ∈ g, x ∈ M},

238 / r. quiroga-barranco

which, by the local freeness of the G-action, is a smooth subbundle of T (k) M. In fact, we have G = T O, and as with this bundle there is a trivialization given by M × g → Gk (x, X ) → jxk−1 (X ∗ ). The corresponding trivialization of the frame bundle of G k is given by M × GL(g) → L(G k ) (x, A) → Ax , where Ax (X ) = jxk−1 ((AX )∗ ). Note that we have taken g as the standard ﬁber of the bundle G k . (k) Choose a subspace G0 of T0 Rn isomorphic to g. We will ﬁx such subspace as well as an isomorphism with g through which we will identify these two spaces. Let us now consider L(k) (M, G k ) = {α ∈ L(k) (M) : α(G0 ) = Gxk if α ∈ L(k) (M)x }, which is a smooth reduction of L(k) (M) to the subgroup of GL(k) (n) that preserves the subspace G0 ; we will denote such subgroup with GL(k) (n, G0 ). Recall from the remarks preceding Lemma 2.2 that for every j0k (ϕ) ∈ L(k) (M) we obtain a linear isomorphism (k)

(k)

T0 Rn → Tϕ(0) M k−1 j0k−1 (X ) → jϕ(0) (dϕ(X )).

In particular, if we let fk : L(k) (M, G k ) → L(G k ) j0k (ϕ) → j0k (ϕ)|G0 , then, by the identiﬁcation between G0 and g, we can consider fk as a welldeﬁned smooth principal bundle morphism that covers the identity. The associated homomorphism of structure groups for fk is given by πk : GL(k) (n, G0 ) → GL(g) j0k (ϕ) → j0k (ϕ)|G0 , which is clearly surjective. Note that we have used again our identiﬁcation between g and G0 .

isometric actions of simple groups and transverse structures / 239

Fix μ an arbitrary ergodic component for the G-action on M for the pseudoRiemannian volume. Then, there is a measurable reduction P of L(k) (M, G k ) so that σ k (P) is (μ-a.e. over M) a single point q0 ∈ Qk . Furthermore, the structure group of P is the subgroup of GL(k) (n, G0 ) that stabilizes q0 , and in particular it is algebraic. This claim is a consequence of the fact that the GL(k) (n, G0 )-action on Qk is algebraic, which in turn follows from the fact that a pseudo-Riemannian metric is a geometric structure of algebraic type; we refer to section 4 and the proof of proposition 8.4 of [1] for further details. On the other hand, since πk is a surjection and fk is G-equivariant, by proposition 8.2 of [1], there exist reductions Q1 and Q 2 of L(k) (M, G k ) and Z

L(G k ), respectively, to subgroups L1 ⊂ GL(k) (n, G0 ) and Ad(G) ⊂ GL(g), such that fk (Q1 ) ⊂ Q 2 (μ-a.e. over M) and such that πk (L1 ) is a ﬁnite-index subgroup Z

of Ad(G) . Here L1 can be chosen to be the algebraic hull of L(k) (M, G k ) for the G-action on M with respect to the ergodic measure μ. This claim uses the Z

well-known fact that Ad(G) is the algebraic hull of M × GL(g) for the product Z

action. It is not difﬁcult to see that this can be chosen so that Q 2 = M × Ad(G) (μ-a.e. over M) with respect to the above identiﬁcation M × GL(g) ∼ = L(G k ). We can also assume that Q1 ⊂ P, μ-a.e. over M, because the reduction Q1 is the smallest one to an algebraic subgroup. The above discussion ensures that for μ-a.e. x ∈ M, we have the following relations: Z

fk ((Q1 )x ) ⊂ (Q2 )x = {x} × Ad(G) (Q1 )x ⊂ (P)x ⊂ L(k) (M, G k )x σ k ((P)x ) = {q0 }.

Let us now ﬁx a point x such that these conditions hold. Choose αx ∈ (Q1 )x Z

and let fk (αx ) = (x, kx ) where kx ∈ Ad(G) . Since πk is surjective, there exists kx ∈ GL(k) (n, G0 ) such that πk ( kx ) = kx . In particular, by the πk -equivariance of fk we have fk (αx k−1 ) = (x, e). We also have by the same reason x −1 −1 −1 k−1 fk (αx g x ) = fk (αx kx kx g kx ) = (x, kx πk (g)kx ),

for every g ∈ L1 . Also, the inclusion (Q1 )x ⊂ L(k) (M, G k )x implies that, for −1 every g ∈ L1 , the k-jets of diffeomorphisms αx k−1 x , αx g kx considered as linear (k) n (k) k isomorphisms T0 R → Tx M map G0 onto Gx . Hence, from the deﬁnition −1 −1 is a k-jet of local diffeomorof fk it follows that αx gαx−1 = (αx g k−1 x )(αx kx ) (k) (k) phism of M at x whose associated isomorphism Tx M → Tx M maps Gxk onto

240 / r. quiroga-barranco

itself by the assignment ∗ jxk−1 (X ∗ ) → jxk−1 ((kx πk (g)k−1 x X ) ),

for which we have used the above trivialization M × GL(g) ∼ = L(G k ). Since Z

πk (L1 ) has ﬁnite index in Ad(G) it contains the identity component Ad(G), and Z

because kx ∈ Ad(G) the group kx πk (L1 )k−1 x also contains Ad(G). It follows (k) that αx L1 αx−1 is a subgroup of Dx (M) for which the homomorphism from Lemma 2.2(1) induces a homomorphism Hx : αx L1 αx−1 → GL(Gxk ) αx gαx−1 → x (αx gαx−1 )|G k , x

whose image contains Ad(G) ⊂ GL(g) with respect to the identiﬁcation between g and Gxk given by the above isomorphism M × g ∼ = G k . This implies that the induced Lie algebra homomorphism hx : Lie(αx L1 αx−1 ) → gl(Gxk ) has image ad(g), again with respect to the referred identiﬁcation between g and Gxk . On the other hand, we have for every g ∈ L1 σ k ((αx gαx−1 )αx ) = σ k (αx g) = σ k (αx ) because σ k ((Q1 )x ) ⊂ σ k ((P)x ) = {q0 } is a single point. But this identity proves that every such k-jet αx gαx−1 preserves the pseudo-Riemannian metric up order k (see [1]). In other words, αx L1 αx−1 is a subgroup of Aut k (M, x) and by Lemma 2.2 we also have that Lie(αx L1 αx−1 ) is a Lie subalgebra of Killk0 (M, x). From the above remarks, it follows that there is a Lie algebra homomorphism ρ xk : g → Killk0 (M, x) such that (∗ )

θx ( ρxk (X ))(jxk−1 (Y ∗ )) = jxk−1 ([X , Y ]∗ )

for every X , Y ∈ g.

For k ﬁxed, the existence of the homomorphism ρ xk has been established for μ-a.e. x ∈ M, where μ is an arbitrary ergodic component of the pseudoRiemannian volume of M. Thus, for k ﬁxed, it follows that the homomorphism ρ xk exists for every x ∈ Sk , where Sk is some subset of M, which is conull with respect to the pseudo-Riemannian volume of M. Finally, if we let S0 = ∩∞ S , k=1 k then S0 is conull with respect to the pseudo-Riemannian volume and for every x ∈ S0 and every k ≥ 1 there exists a homomorphism ρ xk : g → Killk0 (M, x) satisfying (∗ ).

isometric actions of simple groups and transverse structures / 241

In [6] the notion of k-regular point for a metric in a manifold is introduced. Such regular points satisfy two key properties relevant to our discussion. First, the set of regular points U of M is an open dense subset. Second, for x ∈ U there is some integer k(x) ≥ 1 so that, for k ≥ k(x), every element of Killk0 (M, x) extends uniquely to an element of Kill0 (M, x). The ﬁrst property is found in [6] and the second one is proved in [1], both just using smoothness. Note that the results in [6] are stated for Riemannian manifolds but, as remarked in [1], the ones we consider here apply without change to general pseudo-Riemannian manifolds. The upshot of these remarks is that for every x ∈ U, there is some k(x) ≥ 1 so that the map Jxk : Kill0 (M, x) → Killk0 (M, x) X → jxk (X ), is a linear isomorphism for every k ≥ k(x). Note that in this case, for the usual brackets in Kill0 (M, x) and the brackets [·, ·]k in Killk0 (M, x) considered above, the map Jxk is a Lie algebra anti-isomorphism. For S0 and U as above, consider the dense subset S = S0 ∩ U ⊂ M. Next choose x ∈ S and k ≥ max (k(x), 2). Then, the map Jxk is a Lie algebra anti-isomorphism, and there exists a Lie algebra homomorphism ρ xk : g → k Kill0 (M, x) satisfying (*). If we let ρx = −(Jxk )−1 ◦ ρ xk : g → Kill0 (M, x), then ρx deﬁnes a Lie algebra homomorphism such that jxk−1 ([ρx (X ), Y ∗ ]) = jxk−1 ([X , Y ]∗ ), for every X , Y ∈ g. For this, we have used (*) and the deﬁnition of θx from Lemma 2.2. Since k − 1 ≥ 1 and because germs of Killing ﬁelds are determined by any jet of order at least 1, we conclude that, at our chosen point x, ρx satisﬁes (from our statement) (1), (2), and the identity in (3) in a neighborhood of x replaced with M. The identity in (3) now proves that every element with M of g(x) preserves the tangent bundle to O in a neighborhood of x: that is, the corresponding Lie derivatives map sections of T O into sections of T O. By proposition 2.2 of [5] we conclude that the local ﬂows of the elements of g(x) preserve O as well in a neighborhood of x. Since the elements of g(x) are Killing ﬁelds, we conclude that they (and their local ﬂows) also preserve the normal bundle T O⊥ in a neighborhood of x. This completes the proof replaced with M in of our statement for the dense subset S ⊂ M and for M (1)–(4). Finally, this yields the statement for M for the dense subset, which is the inverse image of S under the covering map since such map is a local isometry.

242 / r. quiroga-barranco

remark 2.4. The conclusions of Proposition 2.3 hold without change for with M in (1)–(4). In fact, our proof some dense subset S ⊂ M by replacing M ﬁrst obtains the required Killing ﬁelds on M, which are then translated into corresponding ones on M. In this work, we will be dealing with and interested in the case where the G-orbits in M are nondegenerate submanifolds with respect to the pseudo are nondegenerate as well Riemannian metric. In this case, the G-orbits on M = T O ⊕ T O⊥ . When this holds, and we have a direct sum decomposition T M that is given, at every x ∈ M, by the we can consider the g-valued 1-form ω on M ∼ composition Tx M → Tx O = g, where the ﬁrst map is the natural projection and the second map is the natural isomorphism. From this, we then deﬁne the g-valued 2-form given by = dω|∧2 T O⊥ . The following result will provide us with a criterion, in terms of , for the normal bundle T O⊥ to be integrable. At the same time, we relate with the g-module structures from Proposition 2.3. This result is very well known and it is essentially contained in [2], [3], and [10], but we include its proof here for the sake of completeness.

lemma 2.5. Let G, M, and S be as in Proposition 2.3. If we assume that the G-orbits are nondegenerate, then → g and x : ∧2 Tx O⊥ → g are both 1) for every x ∈ S, the maps ωx : Tx M homomorphisms of g-modules, for the g-module structures from Proposition 2.3, and 2) the normal bundle T O⊥ is integrable if and only if = 0. Proof. In the rest of the proof, let X ∈ g and x ∈ S be ﬁxed but arbitrarily given. let Z , Z ⊥ be its T O and T O⊥ components, For Z a vector ﬁeld over M, respectively. Since ρx (X ) is a Killing ﬁeld preserving O and T O⊥ it follows that [ρx (X ), Z] = [ρx (X ), Z ], [ρx (X ), Z]⊥ = [ρx (X ), Z ⊥ ]. Denote with α : Tx O → g the inverse map of X → Xx∗ . Then we have ωx (X · Zx ) = ωx ([ρx (X ), Z]x ) = α([ρx (X ), Z ]x ) = α([ρx (X ), ω(Z)∗ ]x )

isometric actions of simple groups and transverse structures / 243

= α([X , ω(Z)]∗x ) = [X , ωx (Z)] = X · ωx (Zx ), thus showing that ωx is a homomorphism of g-modules. Here we used in the second and third identities the deﬁnition of ω, and in the fourth identity the formula from Proposition 2.3(3); the rest follows from the deﬁnition of the g-module structures involved. Next, observe that for every pair of sections Z1 , Z2 of T O⊥ we have

(Z1 ∧ Z2 ) = Z1 (ω(Z2 )) − Z2 (ω(Z1 )) − ω([Z1 , Z2 ]) = −ω([Z1 , Z2 ]), which clearly implies (2). Now let u, v ∈ Tx O⊥ be given and choose U, V sections of T O⊥ extending them, respectively. Hence, using that ω is a homomorphism of g-modules, the Jacobi identity and the above expression relating and ω, we obtain

x (X · (u ∧ v)) = x ((X · u) ∧ v) + x (u ∧ (X · v)) = x ([ρx (X ), U] ∧ V ) + x (U ∧ [ρx (X ), V ]) = −ωx ([[ρx (X ), U], V ]) − ωx ([U, [ρx (X ), V ]]) = −ωx ([ρx (X ), [U, V ]]) = −ωx (X · [U, V ]x ) = −[X , ωx ([U, V ])] = [X , x (U ∧ V )] = X · x (u ∧ v), thus showing that x is a homomorphism of g-modules. Note that we have used that both [ρx (X ), U], [ρx (X ), V ] are sections of T O⊥ . The following result allows us to relate the metric T O coming from M to suitable metrics on G. The proof presented here is due to Gromov (see [2]) and provides our ﬁrst application of Proposition 2.3.

lemma 2.6. Suppose that the G-action on M has a dense orbit and preserves a ﬁnite-volume pseudo-Riemannian metric. Then, for every x ∈ M and with respect to the natural isomorphism g ∼ = Tx O, the metric of M restricted to Tx O deﬁnes an Ad(G)-invariant symmetric bilinear form on g independent of the point x.

244 / r. quiroga-barranco

Proof. With the above-mentioned trivialization of T O, the metric h on M restricted to the orbits and pulled back to M × g yields a map ψ : M → g∗ ⊗ g∗ x → Bx where Bx (X , Y ) = hx (Xx∗ , Yx∗ ). By Remark 2.4, there is a dense subset S ⊂ M so that the conclusions of Proposition 2.3 are satisﬁed for every x ∈ S with the tangent spaces and Killing replaced by those of M. Hence, for every x ∈ S the inner product ﬁelds of M hx is preserved, in the sense of Proposition 2.3(4), by the Killing vector ﬁelds that belong to g(x). In particular, for every x ∈ S and X , Y , Z ∈ g we have hx ([ρx (X ), Y ∗ ]x , Zx∗ ) = −hx (Yx∗ , [ρx (X ), Z ∗ ]x ), which, by Proposition 2.3(3) yields hx ([X , Y ]∗x , Zx∗ ) = −hx (Yx∗ , [X , Z]∗x ). In other words, for every x ∈ S we have Bx ([X , Y ], Z) = −Bx (Y , [X , Z]), for all X , Y , Z ∈ g. This implies that ψ(x) = Bx is an Ad(G)-invariant form on g for every x ∈ S. By the density of S in M, we conclude that the image of ψ lies in the set of Ad(G)-invariant forms. On the other hand, at every x ∈ M and for g ∈ G, X , Y ∈ g we have ∗ ∗ , Ygx ) ψ(gx)(X , Y ) = hgx (Xgx −1 ∗ −1 ∗ = hx (dggx (Xgx ), dggx (Ygx ))

= hx (Ad(g −1 )(X )∗x , Ad(g −1 )(Y )∗x ) = ψ(x)(Ad(g −1 )(X ), Ad(g −1 )(Y )), which shows that ψ is G-equivariant. Note that we used in the second identity that G preserves the metric, and in the third identity we used the remarks at the beginning of this section. The G-equivariance of ψ and the fact that its image lies in G-ﬁxed points implies that ψ is G-invariant. Then, the result follows from the existence of a dense G-orbit. As a simple application of the previous result, we prove the nondegeneracy of the orbits when the manifold acted upon has a suitably bounded dimension.

isometric actions of simple groups and transverse structures / 245

lemma 2.7. Suppose that the G-action on M has a dense orbit and preserves a ﬁnite-volume pseudo-Riemannian metric. If dim (M) < 2 dim (G), then the Gorbits are nondegenerate with respect to the metric on M. Proof. By Lemma 2.6, for every x ∈ M the metric restricted to Tx O corresponds to an Ad(G)-invariant form in g. The kernel of such a form is an ideal and so the metric hx restricted to Tx O is either nondegenerate or zero. Suppose that hx is zero when restricted to Tx O for some x ∈ M. Then, Tx O lies in the null cone of Tx M for the metric hx . Hence, for (m1 , m2 ) the signature of M, we have dim (G) = dim (Tx O) ≤ min (m1 , m2 ). And this implies 2 dim (G) ≤ m1 + m2 = dim (M), which is impossible.

3. Proof of Theorem 1.1 and Some Consequences We start this section by proving Theorem 1.1. Proof of Theorem 1.1. Assuming that T O⊥ is integrable, let F be the induced foliation. We will ﬁrst prove that F is totally geodesic, that is, its leaves are totally geodesic submanifolds of M. We will denote with h the metric on M preserved by G. First note that, if Y , Z are local sections of T O⊥ that preserve the foliation, then we have for every X ∈ g: X ∗ (h(Y , Z)) = h([X ∗ , Y ], Z) + h(Y , [X ∗ , Z]) = 0, because our choices imply that [X ∗ , Y ] and [X ∗ , Z] are sections of T O. In particular, for every Y , Z as above the function h(Y , Z) is constant along the G-orbits. In the notation of [5], we conclude that h is a bundle-like metric for the foliation O. Hence, by the remarks on page 79 of [5] it follows that h induces a transverse metric to the foliation O. By the construction of such transverse metric from h and the arguments in the proof of proposition 3.2 of [5], it is easy to conclude that the foliation O is given by pseudo-Riemannian submersions that deﬁne the transverse metric. More precisely, at every point in M there is an open subset U of M and a pseudo-Riemannian submersion π : U → B such that the ﬁbers of π deﬁne the foliation O restricted to U. We observe that the results of [5] are stated for Riemannian metrics, but those that we use here extend without change to arbitrary pseudo-Riemannian metrics. We will now use the properties of the structural equations from [8] for a pseudo-Riemannian submersion π : U → B as above. Again, the results in [8] are stated for Riemannian submersions, but the ones that we will use are easily seen to hold for pseudo-Riemannian submersions as well. For π as

246 / r. quiroga-barranco

above, let A be the associated fundamental tensor deﬁned in [8]. In particular, by the deﬁnition of A, the second fundamental form for the leaves of F is given by AX Y , for X , Y tangent to F . But by lemma 2 of [8] we have for X , Y tangent to F the identity AX Y = 12 [X , Y ] , where Z denotes the projection of Z onto T O. Hence, the integrability of T O⊥ to F shows that A vanishes on vector ﬁelds tangent to F , thus showing that the leaves of F are totally geodesic. Choose a leaf N of F . Then, one can prove fairly easily that every geodesic in M that is tangent at some point to N remains in N for every value of the parameter of the geodesic; this uses the fact that N is a maximal integral submanifold of T O⊥ and that the leaves of F are totally geodesic. Hence, the completenes of M implies that of N. For our chosen leaf N of F , consider the G-action map restricted to G × N. This deﬁnes a smooth map ϕ : G × N → M, which is G-equivariant. By Lemma 2.6, it follows easily that ϕ is a local isometry for G × N endowed with the product metric where G carries a suitable bi-invariant metric. In particular, G × N is complete and so we conclude from corollary 29 on page 202 from [9] that ϕ is an isometric covering map. Hence, the universal covering map of M ×N → M and the G-action ×N is the left is given by ϕ:G on M lifted to G action on the ﬁrst factor. × Iso(N), that is, that every element in We now claim that π1 (M) ⊂ Iso(G) To see this, let γ ∈ π1 (M) be π1 (M) preserves the factors in the product G × N. ×N given with γ = (γ1 , γ2 ) its component decomposition. Observe that in G we have T(g,x) O = Tg G,

T(g,x) O⊥ = Tx N,

× N. Since γ commutes with the G-action, for every (g, x) ∈ G it preserves ⊥ both T O and T O and so dγ (u) = dγ1 (u) + dγ2 (u) ∈ T O dγ (v) = dγ1 (v) + dγ2 (v) ∈ T O⊥ , and v ∈ T N. We conclude that dγ2 (T O) = 0 and dγ1 (T O⊥ ) = for every u ∈ T G and γ2 is independent of G. This 0, which implies that γ1 is independent of N yields our claim about π1 (M). carries a bi-invariant metric, by the results On the other hand, since G from section 4 of [10] we know that the connected component of the identity is given by Iso0 (G) = L(G)R( G) (the left and right translations) and of Iso(G) Hence, the group = π1 (M) ∩ that it is a ﬁnite index subgroup of Iso(G). × Iso(N)) has ﬁnite index in π1 (M), and so the induced map (G × (Iso0 (G)

isometric actions of simple groups and transverse structures / 247

N)/ → M is a ﬁnite covering. Moreover, every γ ∈ can be written as γ = and γ2 ∈ Iso(N), and since such γ commutes (Lg1 Rg2 , γ2 ), where g1 , g2 ∈ G with the G-action we conclude that (g1 gg2 , γ2 (x)) = (Lg1 Rg2 , γ2 )(g(e, x)) = g((Lg1 Rg2 , γ2 )(e, x)) = (gg1 g2 , γ2 (x)) and x ∈ N, which implies g1 ∈ Z(G). Hence, Lg1 = Rg1 and for every g ∈ G × Iso(N), thus showing that ⊂ R(G) × Iso(N). then γ ∈ R(G) Also note that the covering map G × N → M realizes π1 (G) ⊂ , which → (G × N)/. → (G × induces a covering map G × N We claim that G × N N)/ is a normal covering map. For this we need to check that π1 (G) is a normal subgroup of under the inclusion z → (Rz , e). But by the above and γ2 ∈ remarks, every γ ∈ can be written as γ = (Rg1 , γ2 ), where g1 ∈ G from which we obtain Iso(N), γ (Rz , e)γ −1 = (Rg1 Rz Rg −1 , γ2 γ2−1 ) = (Rz , e) 1

It follows that the group of deck transformations since π1 (G) is central in G. for G × N → (G × N)/ is given by the group = /π1 (G) and that we also = (G × N)/. have (G × N)/ = G × Iso(N) is a From the above, we conclude that ⊂ R(G) × Iso(N) group of deck transformations of G × N → M that induces a G-equivariant → M that satisﬁes the properties required to ﬁnite covering map (G × N)/ obtain Theorem 1.1. The integrability of the normal bundle can be ensured for suitable relations between the Lie group G and the geometry of the manifold on which it acts, thus providing the following results. In what follows, the signature of G, as a pseudo-Riemannian manifold, is always considered with respect to a biinvariant metric. Note that if (n1 , n2 ) is the signature of some bi-invariant metric on G, then the signature of any other bi-invariant metric is either (n1 , n2 ) or (n2 , n1 ).

corollary 3.1. Suppose that the G-action on M has a dense orbit and preserves a ﬁnite-volume complete pseudo-Riemannian metric. If the G-orbits are nondegenerate and either one of the following holds, 1) there is no Lie algebra embedding of g into so(Tx O⊥ ) for every x ∈ M, or 2)for n0 = min (n1 , n2 ) and m0 = min (m1 , m2 ), where (n1 , n2 ) and (m1 , m2 ) are the signatures of G and M, respectively, we have rankR (g) > m0 − n0 , then the conclusion of Theorem 1.1 holds.

248 / r. quiroga-barranco

given as in Proposition 2.3. Proof. Let us consider a subset S ⊂ M First suppose that condition (1) is satisﬁed. Then, for every x ∈ S, the g-module structure on Tx O⊥ given by Proposition 2.3(4) is trivial. By Lemma 2.5(1), being a homomorphism of g-modules, the map x is trivial for every x ∈ S. Since S is dense, we conclude that = 0 and so T O⊥ is integrable by Lemma 2.5(2). Hence, Theorem 1.1 can be applied. Let us now assume that (2) holds. Note that by Lemma 2.6, the signature of T O is either (n1 , n2 ) or (n2 , n1 ). If we let (k1 , k2 ) be the signature of T O⊥ , then it is easily seen that min (k1 , k2 ) ≤ m0 − n0 . Since the real rank of so(Tx O⊥ ) is precisely min (k1 , k2 ) the result in this case follows from the ﬁrst part.

corollary 3.2. Suppose that the G-action on M has a dense orbit and preserves a ﬁnite-volume complete pseudo-Riemannian metric. Let n be the dimension of the smallest g-module V such that ∧2 V contains a g-submodule isomorphic to g. If dim (M) < dim (G) + n, then the conclusion of Theorem 1.1 holds. Proof. First observe that the Lie brackets in g deﬁne a surjective homomorphism of g-modules from ∧2 g onto g. Hence, ∧2 g contains a submodule isomorphic to g, thus implying that n ≤ dim (g). Then, Lemma 2.7 shows that the G-orbits in M are nondegenerate with respect to the metric of M. be a subset given as in Proposition 2.3. From our hypotheLet S ⊂ M ses and since the G-orbits are nondegenerate, we have dim (Tx O⊥ ) < n. In particular, for every x ∈ S and for the g-module structure from Proposition 2.3(4), ∧2 Tx O⊥ does not contain a g-submodule isomorphic to g. Hence, Lemma 2.5(1) implies that x = 0 for every x ∈ S and so that = 0. By Lemma 2.5(2) the bundle T O⊥ is integrable and so Theorem 1.1 can be applied. On a compact manifold with Riemannian normal bundle, we can obtain the conclusions of Theorem 1.1 without having to a priori assume that the manifold is complete and that the normal bundle is integrable.

corollary 3.3. Suppose that the G-action on M has a dense orbit and preserves a pseudo-Riemannian metric. If M is compact, the G-orbits are nondegenerate and the normal bundle T O⊥ is Riemannian, then the conclusions of Theorem 1.1 is Riemannian homogeneous and that hold. Moreover, we can assume that N ⊂ G × Iso0 (N).

isometric actions of simple groups and transverse structures / 249

Proof. The proof is a reﬁnement of that of Theorem 1.1, so we will follow the notation of the latter. First observe that the integrability of T O⊥ follows from the proof of Corol In particular, we have lary 3.1(1) since so(Tx O⊥ ) is compact for every x ∈ M. the hypotheses of Theorem 1.1 except for the completeness of M. In this respect, it is easy to check that the compactness of M and the fact that T O⊥ is Riemannian imply that the geodesics in M perpendicular to the G-orbits are complete. This completeness is enough for the rest of the arguments in the proof of Theorem 1.1 to apply. has Finally, we observe that the existence of a dense G-orbit implies that N a dense orbit by its local isometries and, being Riemannian, we conclude that it is homogeneous. The latter follows from the inﬁnitesimal characterization of homogeneous Riemannian manifolds obtained in [12], and the fact that the orthogonal group is compact for deﬁnite metrics; we refer to [10] for further details. But for a homogeneous Riemannian manifold the group of isometries has ﬁnitely many connected components, and so we can intersect with to obtain the last claim after passing to a ﬁnite covering. G × Iso0 (N)

4. Manifolds with a Transverse Riemannian Structure: Proof of Theorem 1.2 In this section we will characterize those actions that preserve a Riemannian structure transverse to the G-orbits. We start with the following basic result relating transverse geometric structures for the foliation by G-orbits with geometric structures on the normal bundle to such orbits.

lemma 4.1. Suppose that the G-action on M has a dense orbit and preserves a ﬁnite-volume pseudo-Riemannian metric. Also assume that the G-orbits are nondegenerate. If H is a subgroup of GL(k, R), where k = dim (M) − dim (G), then there is a one-to-one correspondence between the G-invariant H-reductions of L(TM/T O) and the G-invariant H-reductions of L(T O⊥ ). In particular, every transverse Hstructure for the foliation O by G-orbits induces a G-invariant H-reduction of L(T O⊥ ). Proof. From the decomposition TM = T O ⊕ T O⊥ we obtain a natural Gequivariant isomorphism TM/T O → T O⊥ , which clearly yields the ﬁrst claim. Next, we recall that a transverse H-structure to the foliation O is given by a reduction P of L(TM/T O), which is invariant under the local ﬂows of vectors

250 / r. quiroga-barranco

ﬁelds tangent to the foliation O. In particular, P is invariant under the G-action on L(TM/T O) thus showing the last claim. The following result is obtained by applying the main theorems from [10].

proposition 4.2. Suppose that the G-action on M has a dense orbit and preserves a pseudo-Riemannian metric. Also assume that M is compact. If the normal bundle to the orbits T O⊥ is Riemannian, then there exist → M, 1) a ﬁnite covering map M 2)a connected Lie group H with a compact subgroup K , and 3)a discrete cocompact subgroup ⊂ G × H such that G is dense in G × H, so that M is G-equivariantly diffeomorphic for which the G-action on M lifts to M to (G × K \H)/ . Furthermore, if G has ﬁnite center and real rank at least 2, then we can assume that G × H is a ﬁnite-center isotypic semisimple Lie group and is an irreducible lattice. Proof. Let us denote with (m1 , m2 ) and (n1 , n2 ) the signatures of M and G, respectively. Also, let us denote m0 = min (m1 , m2 ) and n0 = min (n1 , n2 ). Observe that since T O⊥ is Riemannian, the bundle T O is nondegenerate. Hence, by Lemma 2.6 the signature of T O is either (n1 , n2 ) or (n2 , n1 ); this is because the signature of every Ad(G)-invariant metric on g is the signature of either the Killing form or its negative. Without loss of generality, we can assume that the signature of T O is precisely (n1 , n2 ). Theorems A and B from [10] provide precisely our required conclusions when n0 = m0 holds, except for explicitly stating the property that G is dense in G × H. The latter is obtained as follows. The conclusion of theorem A from [10] ensures the existence of a G-invariant ergodic smooth measure, which in turn implies the existence of a dense G-orbit in (G × H)/ . From this, it is easily seen that we necessarily have the density of the G-orbit of (e, e) in (G × H)/ . Since the map G × H → (G × H)/ is a covering, we conclude that the inverse image under this covering of such orbit, which is G, is dense in G × H. We now consider the possibilities for n0 . We will now assume that the signatures are given in the form (+, −). If n0 = n2 , then the fact that T O⊥ is Riemannian implies that (m1 , m2 ) = (n1 + k, n2 ), where k is the rank of T O⊥ . In particular, n0 = m0 in this case and the result follows from [10]. In case n0 = n1 , we can replace the metric on T O⊥ by its negative to obtain a new G-invariant metric for which the signature of M is (m1 , m2 ) = (n1 , n2 + k),

isometric actions of simple groups and transverse structures / 251

where k is again the rank of T O⊥ . Hence, for this new metric n0 = m0 and so the result follows also from [10]. We now proceed to prove Theorem 1.2. = (G × K \H)/ Proof of Theorem 1.2. Assume that (1) holds. Let us take M and verify that it satisﬁes (2). Note that, with respect to the quotient map G × H → (G × H)/ , the set G projects onto the G-orbit of the class of the identity. Hence, there is a dense G-orbit in M. Next, endow G × H with the product metric given by a bi-invariant metric on G and a Riemannian metric on H, which is left K -invariant and right H-invariant. The latter exists because K is compact. This induces a pseudo-Riemannian metric on G × K \H, which is left G-invariant and right -invariant. Note that the G-orbits are nondegenerate with normal bundle given by the tangent bundle to the factor K \H. Furthermore, by construction the projection G × K \H → K \H is a pseudo-Riemannian submersion and so it deﬁnes a transverse Riemannian structure for the foliation by G-orbits. Since the left G-action and the -action commute with each other, the transverse Riemannian structure on G × K \H induces a corresponding one on the double coset (G × K \H)/ , which is G-invariant. This provides the geometric structures required by (2). has a dense GLet us now assume that (2) holds, so that in particular M orbit and carries the indicated geometric structures. Since the G-orbits in M = TO ⊕ are nondegenerate then we have an orthogonal decomposition T M ⊥ T O which is invariant under G. The existence of a transverse Riemannian and Lemma 4.1 imply the existence structure for the foliation by G-orbits in M of a G-invariant Riemannian metric on T O⊥ . Hence, if we replace with such we obtain a new Riemannian metric the metric on T O⊥ induced from M, G-invariant pseudo-Riemannian metric on M for which T O⊥ is Riemannian. satisﬁes the hypotheses of Proposition 4.2, and Hence, the G-action on M and thus of M, with the properties the latter provides a ﬁnite covering of M, required by (1). Finally, let us assume that G has ﬁnite center and real rank at least 2. First observe that (1) replaced by the conditions in the last part of the statement still implies (2). The only nontrivial property to check is the existence of a dense G-orbit in (G × K \H)/ for an irreducible lattice. But in this case, G is dense in G × H (see for example lemma 6.3 from [10]), which yields the

252 / r. quiroga-barranco

required dense orbit. On the other hand, that (2) implies (1) with the properties from the last part of the statement is a consequence of Proposition 4.2.

5. Actions on Lorentzian Manifolds In this section, we present a characterization of the G-actions on M preserving a Lorentzian metric.

theorem 5.1. Let G be a connected noncompact simple Lie group acting faithfully on a compact manifold M. Then the following conditions are equivalent: 1) The group G is locally isomorphic to SL(2, R) and there is a G-equivariant ﬁnite covering map (G × K \H)/ → M where H is a connected Lie group with a compact subgroup K and ⊂ G × H is a discrete cocompact subgroup such that G is dense in G × H. → M for which the G-action on M lifts to a 2)There is a ﬁnite covering map M with a dense orbit and preserving a Lorentzian metric. G-action on M = (G × K \H)/ Proof. Let us assume (1) and consider the ﬁnite covering M → M. Endow G × H with the product metric given by the bi-invariant metric coming from the Killing form in g and a left K -invariant and right H-invariant Riemannian metric on H. Then consider the induced G-invariant metric on Since G is locally isomorphic to SL(2, R), such metric on M is Lorentzian. M. As in the proof of Theorem 1.2, the density of G implies the existence of a This proves (2). dense G-orbit in M. Let us now assume that (2) holds. By Lemma 2.6, for every x ∈ M, the metric in Tx O corresponds to an Ad(G)-invariant symmetric bilinear form on g. Such form is either 0 or nondegenerate. The former case implies the existence of a null tangent subspace of dimension at least 3, which is impossible; in particular, the G-orbits are nondegenerate. Since G is noncompact and simple, we conclude the existence of an Ad(G)-invariant form on g, which is Lorentzian. But sl(2, R) is the only simple Lie algebra admitting such a form, which implies that G is locally isomorphic to SL(2, R). We also conclude that T O⊥ is Riemannian and so the rest of the claims in (1) follow from Proposition 4.2.

6. Manifolds with a Transverse Parallelism: Proof of Theorem 1.3 The following result is an immediate consequence of the properties of Lie foliations. A proof can be found in [10].

isometric actions of simple groups and transverse structures / 253

lemma 6.1. Let X be a compact manifold carrying a foliation with a transverse Lie structure. If the foliation has a dense leaf, then the lifted foliation to any ﬁnite covering space of X has a dense leaf as well. We now obtain the next result, which describes actions preserving a metric and a transverse parallelism. Its proof is based on some of the arguments found in [11].

proposition 6.2. Suppose that the G-action on M has a dense orbit and preserves a pseudo-Riemannian metric. Also assume that M is compact. If the foliation by G-orbits is nondegenerate (with respect to the pseudo-Riemannian metric) and carries a transverse parallelism, then there exist → M, 1) a ﬁnite covering map M 2)a connected Lie group H, and 3)a discrete cocompact subgroup ⊂ G × H such that G is dense in G × H, so that M is G-equivariantly diffeomorphic for which the G-action on M lifts to M to (G × H)/ . Furthermore, if G has ﬁnite center and real rank at least 2, then we can assume that G × H is a ﬁnite center isotypic semisimple Lie group and is an irreducible lattice. Proof. By Lemma 4.1, the transverse parallelism to the G-orbits yields a Ginvariant trivialization of L(T O⊥ ). Hence, there is a family of G-invariant sections of T O⊥ , say X1 , . . . , Xk , that deﬁnes a basis of T O⊥ on every ﬁber. Let us consider the G-invariant Riemannian metric on T O⊥ for which these vector ﬁelds are orthonormal at every point. Because of the orthogonal decomposition TM = T O ⊕ T O⊥ , if we replace the metric on T O⊥ induced from M with the G-invariant Riemannian metric thus deﬁned from the parallelism, then we obtain a pseudo-Riemannian metric h on M, which is G-invariant, deﬁnes the same orthogonal complement T O⊥ to the orbits and such that this orthogonal complement is Riemannian. In the rest of this proof we will consider M endowed with this new metric h. Hence, by Corollary 3.3 there and a discrete is a simply connected homogeneous Riemannian manifold N for which there is a G-equivariant ﬁnite cocompact subgroup ⊂ G × Iso0 (N) covering map = (G × N)/ → M. M Note that by the proof of Corollary 3.3, the normal bundle T O⊥ is integrable. In particular, the vector ﬁelds Xi given above satisfy:

254 / r. quiroga-barranco

[Xi , Xj ] =

k

fijr Xr

r=1

for some smooth functions fijr deﬁned on M. The G-invariance of the parallelism then implies that the functions fijr are G-invariant and so constant because of the existence of a dense G-orbit. We conclude that the linear span over R of the vector ﬁelds X1 , . . . , Xk is a Lie algebra. On the other hand, the fact that the vector ﬁelds Xi are preserved by the G-action is easily seen to imply that such ﬁelds are foliate in the notation of [5]. Hence, the parallelism X1 , . . . , Xk deﬁnes a transverse Lie structure for the foliation by G-orbits in M. Clearly, this transverse Lie structure induces a Let H be a simply connected Lie group that models corresponding one on M. and consider a corresponding development the transverse Lie structure on M D : G × N → H. In particular, D is a submersion whose ﬁbers have con × {x}. Hence, we nected components given precisely by subsets of the form G conclude that D{e}×N : N → H deﬁnes a local diffeomorphism. is the By the proofs of Corollary 3.3 and Theorem 1.1, the manifold N (isometric) universal covering space of a leaf N for the foliation in M deﬁned 1 , . . . , X k be the pull-backs to N of the restrictions by T O⊥ . Hence, if we let X (which we will of the ﬁelds X1 , . . . , Xk to N, then the induced ﬁelds on G × N × N. denote with the same symbols) deﬁne the transverse Lie structure on G Furthermore, from the previous construction of the metric h, the metric on is given by the condition of the ﬁelds X 1 , . . . , X k being orthonormal at every N point. In particular, if we let h be the Lie algebra of H, then there is a basis i into v1 , . . . , vk of h such that the development D maps the vector ﬁeld X vi , for every i = 1, . . . , k. Without loss of generality we will assume that the transverse H-structures are modeled by taking H with its right translations, and so transverse Lie parallelisms are modeled on right-invariant vector ﬁelds. With this convention, each vi is considered as a right-invariant vector ﬁeld on H. From the above remarks, if we endow H with the right-invariant Riemannian metric for which the vector ﬁelds v1 , . . . , vk are orthonormal at every → H is a local isometry. By corollary 29 on page 202 point, then D{e}×N :N → H is an isomof [9] and since N is complete, we conclude that D{e}×N :N a Lie group structure with respect to which it is etry and so it induces in N with an isomorphism. Hence, we can replace the Riemannian manifold N H carrying the above right invariant Riemannian metric and assume that the × H → H is a development for the transverse Lie strucnatural projection G the space G × N covers M, ture on M. Moreover, by the deﬁnition of M, and so we can assume that the natural projection π : G × H → H is also a

isometric actions of simple groups and transverse structures / 255

with a corresponding development for the transverse Lie structure on M holonomy representation ρ : → H. = G × Iso0 (H), if we choose γ ∈ , then we can Since ⊂ G × Iso0 (N) write γ = (Rγ1 , γ2 ), where γ1 ∈ G and γ2 ∈ Iso0 (H). And so, the ρ -equivariance of π yields for every (g, x) ∈ G × H: ρ (γ ) = xy γ2 (x) = π (gγ1 , γ2 (x)) = π((g, x)γ ) = π (g, x) where y ∈ H. It follows that γ2 is given by a right translation by an element in = (G × H)/ → M is a H, and so we have ⊂ G × R(H) = G × H. Since M has a dense G-orbit and so ﬁnite covering, by Lemma 6.1 we conclude that M G is dense in G × H as in the proof of Proposition 4.2. This concludes the proof of the ﬁrst part of the statement. For the last claim, if G has ﬁnite center and real rank at least 2, then H is semisimple by the main results from [16]. Also, by modding out by a suitable central group we can assume that H has ﬁnite center (see [10]). The arguments at the end of Section 6 from [10] also prove that is an irreducible lattice. Finally, recall that an irreducible lattice in a semisimple Lie group can only exist if the group is isotypic. We can now prove our characterization of actions with a transverse parallelism. Proof of Theorem 1.3. First, let us assume that (1) holds and consider the = (G × H)/ → M. As in the proof of Theorem 1.2 we ﬁnite covering M conclude that M has a dense G-orbit. Note that the right-invariant vector ﬁelds on H deﬁne both a transverse Lie structure on G × H for the foliation given by the factor G, and a right H-invariant Riemannian metric on H. If we consider the product pseudo-Riemannian metric on G × H using a bi-invariant metric on G, then we also obtain a pseudo-Riemannian metric for which the foliation given by the factor G is nondegenerate. Both of these geometric structures on G × H are left G-invariant and right -invariant and so descend to thus establishing (2). corresponding G-invariant geometric structures on M yields (1). The last If we now assume (2), then Proposition 6.2 applied to M claim is also a consequence of Proposition 6.2.

7. Proof of Theorem 1.4 Let us consider G and M as in Proposition 2.3 with S ⊂ M a dense subset provided by this result and Remark 2.4. With the notation of Proposition 2.3, the representation λx ◦ ρx leaves invariant the subspace Tx O⊥ thus deﬁning its

256 / r. quiroga-barranco g-module structure. This induces a homomorphism g → so(Tx O⊥ ) obtained by restricting λx ◦ ρx to the submodule Tx O⊥ . By composition with such

homomorphism, we can consider x obtained from Lemma 2.5 as a map ∧2 Tx O⊥ → so(Tx O⊥ ). For a vector space W with inner product ·, ·, we will say that a bilinear map T : W × W → gl(W ) is of curvature type if it satisﬁes the following conditions for every x, y, z, v, w ∈ W : 1) 2) 3) 4)

T (x, y) = −T (x, y), T (x, y)v, w = − v, T (x, y)w , T (x, y)z + T (y, z)x + T (z, x)y = 0, and T (x, y)v, w = T (v, w)x, y .

Note that (1) and (2) together are equivalent to T inducing a map ∧2 W → so(W ). Also, by the proof of proposition 36 on page 75 of [9] we know that (1), (2), and (3) together imply (4). With the above notation, we have the following result.

lemma 7.1. Let G, M and S ⊂ M be as in Proposition 2.3 and Remark 2.4. For every x ∈ S, deﬁne the bilinear operation [·, ·]0 in Tx M by the assignments: • [X ∗ , Y ∗ ]0 = [X , Y ]∗ , for every X ∗ , Y ∗ ∈ Tx O , (X , Y ∈ g), x x x x x • [X ∗ , v]0 = −[v, X ∗ ]0 = X (v) for every X ∗ ∈ Tx O , v ∈ Tx O ⊥ , (X ∈ g), and x x x • [v1 , v2 ]0 = x (v1 , v2 )∗ for every v1 , v2 ∈ Tx O ⊥ . x

Then, [·, ·]0 yields a Lie algebra structure on Tx M if and only if x is of curvature type when considered as a map ∧2 Tx O⊥ → so(Tx O⊥ ). In this case, the representation of g in Tx M, given by Proposition 2.3(4), preserves the Lie algebra structure of Tx M. Proof. Note that by the above remarks, x is of curvature type if and only if it satisﬁes the above condition (3). Also observe that [·, ·]0 always deﬁnes a skew-symmetric bilinear form in Tx M. Hence, for the ﬁrst claim, we need to show that [·, ·]0 satisﬁes the Jacobi identity if and only if x satisﬁes (3). For the Jacobi identity to hold we only need to verify the following cases. • u1 , u2 , u3 ∈ Tx O . Note that [·, ·]0 maps Tx O × Tx O into Tx O in such a

way that it deﬁnes a skew-symmetric operation that corresponds to the Lie brackets of g under the natural isomorphism g → Tx O given by X → Xx∗ . Hence, the Jacobi identity always holds in this case. • u1 , u2 ∈ Tx O and u3 ∈ Tx O ⊥ . In this case we can write u1 = X ∗ and x u2 = Yx∗ for some X , Y ∈ g. Then, by the deﬁnition of [·, ·]0

isometric actions of simple groups and transverse structures / 257

[[Xx∗ , Yx∗ ]0 , u3 ]0 = [[X , Y ]∗x , u3 ]0 = [X , Y ](u3 ) = X (Y (u3 )) − Y (X (u3 )) = X ([Yx∗ , u3 ]0 ) − Y ([Xx∗ , u3 ]0 ) = [Xx∗ , [Yx∗ , u3 ]0 ]0 − [Yx∗ , [Xx∗ , u3 ]0 ]0 , which proves that the Jacobi identity holds in this case in general. • u1 ∈ Tx O and u2 , u3 ∈ Tx O ⊥ . We can now choose X ∈ g such that u1 =

Xx∗ . Hence, using from Lemma 2.5 the fact that x is a homomorphism of g-modules, we obtain [Xx∗ , [u2 , u3 ]0 ]0 = [Xx∗ , x (u2 , u3 )∗x ]0 = [X , x (u2 , u3 )]∗x = x (X (u2 ), u3 )∗x + x (u2 , X (u3 ))∗x = x ([Xx∗ , u2 ]0 , u3 )∗x + x (u2 , [Xx∗ , u3 ]0 )∗x = [[Xx∗ , u2 ]0 , u3 ]0 + [u2 , [Xx∗ , u3 ]0 ]0 , which yields again the Jacobi identity without extra conditions.

• u1 , u2 , u3 ∈ Tx O ⊥ . The deﬁnition of [·, ·]0 yields now:

[[u1 , u2 ]0 , u3 ]0 = [ x (u1 , u2 )∗x , u3 ]0 = x (u1 , u2 )(u3 ), and so the Jacobi identity is satisﬁed in this case exactly when x satisﬁes condition (3). The above proves the equivalence in the statement, and so it remains to obtain the last claim. For this we need to show that X ([u1 , u2 ]0 ) = [X (u1 ), u2 ]0 + [u1 , X (u2 )]0 , for every X ∈ g and u1 , u2 ∈ Tx M. This is now dealt with through the following cases, which just basically apply the deﬁnitions involved and properties already considered. • u1 , u2 ∈ Tx O . Then, we can write u1 = Y ∗ , u2 = Z ∗ for some Y , Z ∈ g x x

and X ([Yx∗ , Zx∗ ]0 ) = X ([Y , Z]∗x ) = [ρx (X ), [Y , Z]∗ ]x = [X , [Y , Z]]∗x = [[X , Y ], Z]∗x + [Y , [X , Z]]∗x = [[X , Y ]∗x , Zx∗ ]0 + [Yx∗ , [X , Z]∗x ]0 = [[ρx (X ), Y ∗ ]x , Zx∗ ]0 + [Yx∗ , [ρx (X ), Z ∗ ]x ]0 = [X (Yx∗ ), Zx∗ ]0 + [Yx∗ , X (Zx∗ )]0 .

258 / r. quiroga-barranco • u1 ∈ Tx O and u2 ∈ Tx O ⊥ . Now we can write u1 = Y ∗ for Y ∈ g and x

X ([Yx∗ , u2 ]0 ) = X (Y (u2 )) = [X , Y ](u2 ) + Y (X (u2 )) = [[X , Y ]∗x , u2 ]0 + [Yx∗ , X (u2 )]0 = [[ρx (X ), Y ∗ ]x , u2 ]0 + [Yx∗ , X (u2 )]0 = [X (Yx∗ ), u2 ]0 + [Yx∗ , X (u2 )]0 . • u1 , u2 ∈ Tx O ⊥ . We now have

X ([u1 , u2 ]0 ) = X ( x (u1 , u2 )∗x ) = [ρx (X ), x (u1 , u2 )∗ ]x = [X , x (u1 , u2 )]∗x = x (X (u1 ), u2 )∗x + x (u1 , X (u2 ))∗x = [X (u1 ), u2 ]0 + [u1 , X (u2 )]0 .

The following result will allow us to prove Theorem 1.4. It also provides an explicit description of the Lie algebra structure considered in Theorem 1.4.

theorem 7.2. Suppose that G has ﬁnite center and real rank at least 2, and that the G-action on M preserves a ﬁnite-volume complete pseudo-Riemannian metric. Also assume that G acts ergodically on M and that the foliation by G-orbits is nondegenerate. Denote with L the algebraic hull for the G-action on the bundle L(T O⊥ ) and with l its Lie algebra. In particular, there is an embedding of Lie algebras l → so( p, q), where ( p, q) is the signature of the metric of M restricted to T O⊥ . If this embedding is surjective, then one of the following occurs: 1) The conclusion of Theorem 1.1 holds, or 2)G is locally isomorphic to SO0 ( p, q), dim (M) = ( p + q)( p + q + 1)/2, for some x ∈ M the bilinear map x is nonzero and of curvature type and the Lie algebra structure on Tx M obtained from Lemma 7.1 is isomorphic to either so( p, q + 1) or so( p + 1, q). Proof. Since the G-orbits are nondegenerate, G preserves a pseudoRiemannian metric on T O⊥ . Hence, the algebraic hull of L(T O⊥ ) for the G-action can be embedded into the structure group O( p, q) for such metric, where ( p, q) is the signature of T O⊥ . This yields the ﬁrst claim. Let us now assume that the induced embedding l → so( p, q) is surjective. This implies that L is a ﬁnite-index subgroup of O( p, q). By section 3 of [17], we can write L = ZS where S is semisimple without compact factors, Z is compact

isometric actions of simple groups and transverse structures / 259

and centralizes S, and the product is almost direct. In particular, we have a direct product l = z × s. Note that for p + q ≤ 2 the conclusion of Theorem 1.1 holds by Corollary 3.1(1). Hence, we can assume from now on that p + q ≥ 3. Then, the only cases in which so( p, q) is not simple is for so(4) ∼ = so(3) × so(3) ∼ and so(2, 2) = sl(2, R) × sl(2, R). And so, one of the following holds: • s = 0 and z • z = 0 and s

∼ = so( p, q) is compact, or ∼ = so( p, q) is noncompact.

In the ﬁrst case, we conclude that so(Tx O⊥ ) is compact for every x ∈ M, thus implying that the conclusion of Theorem 1.1 follows by Corollary 3.1(1). In particular, we can assume that l = s ∼ = so( p, q). Also, by the arguments in section 3 of [17], there is a surjection g → s of Lie algebras, which implies that g∼ = so( p, q). Hence, G is locally isomorphic to SO0 ( p, q) and because of the decomposition TM = T O ⊕ T O⊥ we also have dim (M) = dim (SO0 ( p, q)) + dim (Tx O⊥ ) = dim (SO0 ( p, q)) + p + q = ( p + q)( p + q + 1)/2. On the other hand, for S ⊂ M given as in Proposition 2.3, if |S = 0, then the conclusion of Theorem 1.1 holds by Lemma 2.5(2). Then, we can choose x ∈ S such that x = 0. In particular, by Lemma 2.5(1) the g-module Tx O⊥ is a nontrivial one. If we consider the representation g → so(Tx O⊥ ) that deﬁnes the g-module structure of Tx O⊥ from Proposition 2.3, then the above remarks show that this representation is an isomorphism. We conclude that Tx O⊥ is a g-module isomorphic to Rp,q with respect to the isomorphism g → so(Tx O⊥ ) thus obtained. For ·, ·p,q the metric in Rp,q preserved by so( p, q) we have the following: Claim: For every c ∈ R, the map Tc : ∧2 Rp,q → so( p, q) given by Tc (u ∧ v) = c v, ·p,q u − c u, ·p,q v, where u, v ∈ Rp,q is a well-deﬁned homomorphism of so( p, q)-modules. Moreover, these maps exhaust all the so( p, q)-module homomorphisms ∧2 Rp,q → so( p, q). The only nontrivial part of the claim is the last statement, which we will now prove. Let T : ∧2 Rp,q → so( p, q) be a homomorphism of so( p, q)-modules. Consider the map T ◦ T1−1 , which is a homomorphism of so( p, q)-modules so( p, q) → so( p, q). Hence, for n = p + q and by complexifying, the map T ◦ T1−1 yields a homomorphism T : so(n, C) → so(n, C) of so(n, C)-modules −1 such that T |so( p,q) = T ◦ T1 . In particular, T (so( p, q)) ⊂ so( p, q) and so T commutes with the conjugation σ of so(n, C) whose ﬁxed point set is so( p, q), that is, T ◦σ = σ ◦ T . Note that since g ∼ = so( p, q) is simple with real rank at least 2, the Lie algebra so(n, C) is simple as well. Hence, the irreducibility of

260 / r. quiroga-barranco so(n, C) as so(n, C)-module implies, by Schur’s Lemma, that there is a com-

plex number c such that T = cIdso(n,C) . But then, the condition T ◦σ = σ ◦ T implies that c is real and so T = Tc . By the Claim and Lemma 2.5(1), with respect to the above-established isomorphisms g → so(Tx O⊥ ) and Tx O⊥ ∼ = Rp,q , we conclude that the map x corresponds to a homomorphism Tc for some real c = 0. It is straightforward to check that Tc is of curvature type; in fact, it yields the curvature of the pseudo-Riemannian manifolds of constant sectional curvature c and signature ( p, q) (see corollary 43 on page 80 of [9]). Hence, Lemma 7.1 applies and it provides a Lie algebra structure on Tx M. It remains to show that such structure is isomorphic to either so( p, q + 1) or so( p + 1, q). By our identiﬁcations, the Lie algebra structure on Tx M is isomorphic to the one obtained on so( p, q) ⊕ Rp,q from the map Tc with the Lie brackets [·, ·]c deﬁned with formulas similar to those in Lemma 7.1 replacing x with Tc . Also, it is straightforward to show that the map given by (so( p, q) ⊕ Rp,q , [·, ·]c ) → so(Rp+q+1 , Ip,q (c)) X cu X +u → , u t Ip,q 0 is an isomorphism of Lie algebras, where Ip,q Ip,q (c) = 0

0 . c

Since so(Rp+q+1 , Ip,q (c)) is clearly isomorphic to either so( p, q + 1) or so( p + 1, q) this yields (2) from our statement. We can now complete the following proof. Proof of Theorem 1.4. Since the hypotheses of Theorem 7.2 are a subset of those of Theorem 1.4, the conclusions of the former hold. Since M is weakly irreducible its universal covering space cannot split isometrically and so part (2) of Theorem 7.2 necessarily holds at some x ∈ M. By the deﬁnition of the Lie algebra structure in Tx M from Lemma 7.1 we conclude that (1) and (2) of Theorem 1.4 are satisﬁed. Finally we observe that, by the proof of Theorem 7.2, we can assume that x ∈ S where S ⊂ M is given by Proposition 2.3. Consider the Lie algebra g(x) of local Killing ﬁelds provided by Proposition 2.3. Then, the last claim of Lemma 7.1 and the deﬁnition of the representations involved implies the last claim of Theorem 1.4 for the Lie algebra g(x).

isometric actions of simple groups and transverse structures / 261

References [1] [2]

[3] [4] [5]

[6] [7] [8] [9] [10] [11] [12] [13] [14] [15]

[16] [17] [18]

A. Candel and R. Quiroga-Barranco, Gromov’s centralizer theorem, Geom. Dedicata 100 (2003), 123–155. M. Gromov, Rigid transformations groups, in Géométrie différentielle, Colloque Géométrie et Physique de 1986 en l’honneur de André Lichnerowicz (D. Bernard and Y. Choquet-Bruhat, eds.), Hermann, Paris, 1988, 65–139. F. G. Hernández-Zamora, Isometric splitting for actions of simple Lie groups on pseudoRiemannian manifolds, Geom. Dedicata 109 (2004), 147–163. N. Kowalsky, Noncompact simple automorphism groups of Lorentz manifolds and other geometric manifolds, Ann. of Math. (2) 144 (1996), no. 3, 611–640. P. Molino, Riemannian foliations, translated from the French by Grant Cairns, with appendices by Cairns, Y. Carrire, É. Ghys, E. Salem, and V. Sergiescu, Progress in Mathematics, 73, Birkhäuser Boston, Boston, MA, 1988. K. Nomizu, On local and global existence of Killing vector ﬁelds, Ann. of Math. 72 (1960), 105–120. G. Olafsson and R. Quiroga-Barranco, Isometric actions of SO0 ( p, q) on low dimensional weakly irreducible pseudo-Riemannian manifolds, in preparation. B. O’Neill, The fundamental equations of a submersion, Michigan Math. J. 13 (1966), 459–469. B. O’Neill, Semi-Riemannian geometry. With applications to relativity, Pure and Applied Mathematics, 103. Academic Press, New York, 1983. R. Quiroga-Barranco, Isometric actions of simple Lie groups on pseudo-Riemannian manifolds, Ann. of Math. (2) 164 (2006), no. 6, 941–969. R. Quiroga-Barranco, Arithmeticity of totally geodesic Lie foliations with locally symmetric leaves, Asian J. Math. 12 (2008), no. 3, 289–297. I. M. Singer, Inﬁnitesimally homogeneous spaces, Comm. Pure Appl. Math. 13 (1960), 685–697. J. Szaro, Isotropy of semisimple group actions on manifolds with geometric structure, Amer. J. Math. 120 (1998), 129–158. A. Zeghib, On afﬁne actions of Lie groups, Math. Z. 227 (1998), no. 2, 245–262. R. J. Zimmer, Actions of semisimple groups and discrete subgroups, Proceedings of the International Congress of Mathematicians, Vol. 1, 2 (Berkeley, CA, 1986), 1247–1258, American Mathematical Society, Providence, RI, 1987. R. J. Zimmer, Arithmeticity of holonomy groups of Lie foliations, J. Amer. Math. Soc. 1 (1988), no. 1, 35–58. R. J. Zimmer, On the algebraic hull of an automorphism group of a principal bundle, Comment. Math. Helv. 65 (1990), no. 3, 375–387. R. J. Zimmer, Automorphism groups and fundamental groups of geometric manifolds, in Differential Geometry: Riemannian Geometry (Los Angeles, CA, 1990), 693–710, Proc. Sympos. Pure Math. 54, Part 3, American Mathematical Society, Providence, RI, 1993.

8

shmuel weinberger

SOME REMARKS INSPIRED BY THE C0 ZIMMER PROGRAM to bob zimmer, with friendship, gratitude, and admiration

In the mid-1980s, when I had just arrived in Chicago, Bob Zimmer was lecturing on his recent work showing how perturbations of isometric actions of Kazhdan groups were dynamically simple (preserve a measurable Riemannian metric) [Zi1]. The heady mixture of the beauty of the results, the audacity of the vision (this is really 0% of what should be true.. . .1 ), and Bob’s charisma and charm attracted me so that I still cannot but help think about the problem of what large discrete groups of Cat-isomorphisms of a manifold can conceivably look like. Alas, my ardor has not been well rewarded in this pursuit. In this paper I would like to make some comments and raise some questions about the case of Cat = Top, that is, the group of homeomorphisms of manifolds. Some of these remarks are rather old, but I will give some details if I am not aware of a discussion elsewhere—while a few are new (at least to me). I hope that this paper spurs more exploration of the fascinating C0 aspect of this nexus of problems. Although this note is short, it has a bit of structure. The beginning is about groups that don’t act, the middle is about lattices and hyperbolic groups and quite tame actions, and the end has an ergodic aspect and relates directly to a paper of Lashof and Zimmer [LZ]. There is also an appendix about deforming homotopy actions to actions at the cost of stabilization. Besides Bob Zimmer, who is the obvious inspiration for this paper, I would also like to thank Kevin Whyte, and the 3Fs: Benson Farb, Steve Ferry, and David Fisher for many conversations that have strongly inﬂuenced my perspective. Courtney Thatcher’s thesis on cyclic group actions on products of spheres has some overlap with the perspective in Section 4. 1. Maybe now we are above the 0% mark, at least in this direction, thanks to the wonderful papers [Be] and [FiM].

262

some remarks inspired by the c0 zimmer program / 263

1. Some Groups That Don't Act Proposition 1. The inﬁnite special linear group SL(Z) does not act topologically, nontrivially, on any compact manifold, or indeed on any manifold whose homology with coefﬁcients in a ﬁeld of positive characteristic is ﬁnitely generated. [Zi2] only mentions the compact case. The more general case is proven simultaneously (and is necessary for the proof, because one must analyze complements of ﬁxed sets as well as ﬁxed sets in the coming argument), and is motivated by the recent paper [ABJLMS], as is Proposition 2 below (see also [FiS] that was written very recently and has some quite deep further observations about the volume-preserving case). Proof. The proof is based on Smith theory considerations. A p-group acting on a mod p-homology manifold has ﬁxed set that is also a mod p-homology manifold of lower dimension. Moreover, the sum of the Betti numbers of the ﬁxed sets can be at most as large for the ﬁxed set as for the original manifold. As a formal consequence, by considerations of the open strata, if (Z/p)k is acting effectively, for k large enough, there must be a very large (Z/p)1 acting freely on some mod p-homology manifold with bounded Betti numbers. However, if l is large enough, this action will automatically have a large kernel for its action on homology (GLr doesn’t have that much p-torsion). Consequently, one can compute the homology of the quotient space by means of the usual Serre spectral sequence. Now if the rank of this kernel is greater than the homology of the space, one sees that the quotient is inﬁnite-dimensional. Putting this all together, we get elements of order p that are acting trivially on M. Since SL(Z) is the normal closure of such elements, we see that the action is actually trivial. Proposition 2. There are ﬁnitely presented groups that do not act nontrivially on any compact manifold. Proof. The idea is to produce ﬁnitely presented groups that contain high-rank ﬁnite subgroups and that are normally generated by enough elements of order p. Many examples are constructed in [ABJLMS]; it might be worth noting that old ideas related to the unsolvability of the triviality problem (a la Rabin) can be used for this. We follow the exposition in [We1].Suppose that π is an effectively acting group with arbitrarily large elementary p-groups generated by speciﬁc words we know. We can make inﬁnitely many HNN extensions, each

264 / s. weinberger

one conjugating one of these elements to another, so that they are now all conjugate in a larger group, called π . We denote by g one of these elements of order p. Now, Higman embed π into a group π . By taking an amalgamated free product (π ∗ Z)∗ Z∗ ZA where A is a knot group (say the trefoil group for concreteness), Z∗ Z is a free subgroup of A that contains a meridian as its ﬁrst generator, and Z∗ Z is embedded into (π ∗ Z) so that the ﬁrst generator goes to [g, t] in the obvious notation. We thus have arranged that any homomorphism that sends any of the elements of order p in our list to e kills the whole group. Whether one prefers Proposition 1 or 2 is a matter of taste. Of course these both beg the question of whether there is a torsion-free group that does not act continuously on any manifold. We note: Proposition 3. Any countable group of ﬁnite homological dimension does have a continuous action on some sphere. Proof. If Bπ is a ﬁnite-dimensional complex, then it can be thickened to a manifold. The universal cover of this manifold is contractible, and if not a Euclidean space, its product with R is (by a well-known theorem of Stallings). This manifold has a free π action. Its one point compactiﬁcation is a sphere, and it has a π action with a single ﬁxed point. This action is rather different than the usual ones we think of coming from symmetric spaces and lattices. It has no dynamics! A smooth version of Proposition 3 seems unlikely (but I have no ideas about how to prove such a thing). On the other hand, the hypothesis is far from necessary. Every manifold (of dimension >0) has an action of a free abelian group of inﬁnite rank. On the other hand, presumably the obvious versions of congruence subgroups of SL(Z) should not act, and we know nothing about this. Problem: What groups are discrete subgroups of Homeo(Dn , rel ∂)? Needless to say, they are torsion free (and they are subgroups of Homeo(M) for any m-manifold for m ≥ n).

2. Actions of Lattices Of course, Zimmer actually suggests that SLn (Z) shouldn’t act on any manifold of dimension < n–1 (except via ﬁnite quotients). In dimension n, the action

some remarks inspired by the c0 zimmer program / 265

on Tn should be rigid. Indeed, the beautiful work of Hurder shows that, with enough smoothness, that action is inﬁnitesimally rigid. It seems worth pointing out that in C0 there are many deformations possible. One deformed action is explained in [FS1], but the process is one that can be repeated inﬁnitely often producing uncountably many new and different actions. The technique here is “insertion” a la [KL]. Whenever one has a C1 action with a ﬁxed point (or indeed, ﬁnite orbits, or indeed, an invariant submanifold) one can glue in a tubular nighborhood acted on via the differential, and extend the action to the complement. Essentially one uses the diffeomorphism of the complement of a disk with the complement of a point inside M – D = M – p. In the neighborhood one can alternatively glue in the cone of the action on the sphere at inﬁnity. These actions are quite different from each other or the original action, as one can see by examining orbits of points. Now, if one wishes, one can insert many times, producing many rings around the ﬁxed point. Also, one can make some of these rings “thick” by coning for a little while before doing a more standard linear insertion. Clearly, this builds uncountably many actions that differ only around a ﬁxed point. (They are distinguished by one another by looking at the limit points of orbits. Note that for compact groups, the action given by insertion is topologically equivalent to the original action.) But these “bull’s-eye” structures can be inserted at will at smaller and smaller scales around any of the ﬁnite orbits associated to the rational points on the torus. As a result, we have beautiful moduli of “leopard spots,” each of which is itself variable according to the bull’s-eye patterns. Thus we have seen a much more beautiful structure than is asserted in the following: Proposition 4. There are uncountably many nontopologically-conjugate SLn (Z) actions on Tn . They can all be deformed to the usual action. These deformations are 1-parameter families of actions that start at the “exotic” action at time 0, and end at the standard action at time 1. And all of these actions map equivariantly to the standard afﬁne action. These actions are just “modiﬁcations” of arithmetic actions. In its most vague form, the Zimmer program asserts that, in some sense,2 all actions of high-rank lattices are built up out of such actions. Thus, Proposition 4 is just par for the course. 2. It is part of the program to ﬁgure out the correct sense.

266 / s. weinberger

There are other deformations that arise from decomposition space theory (aka “Bing topology”); I will give just one example and then move on. Proposition 5. Let π be a countable group with a dense embedding in SO(n). Then SO(n) acts on Sn+k with quotient D1+k . If one considers the induced action over D1+k / ∼ where ∼ is a cell-like semicontinuous decomposition (these exist in profusion if k > 1), the total space is still Sn+k , but the new action of π is never equivalent to any of the other ones. The equivalence of total spaces follows readily from Edwards’s characterization of manifolds and the CE approximation theorem (of Siebenmann, which is a consequence of Edwards’s theorem, as well), see [Da] and [We2] for such deformations. The actions are thus distinguished by the maximal Hausdorff quotients of their orbit spaces, which will be nonmanifold ANR homology manifolds. Part of the Zimmer program, although it is far from the usual part where the actions are ergodic, seems to me to include considerations of free, and more generally, proper discontinuous actions. Of course for these, in the geometric setting, the basic theorem is Mostow rigidity—and superrigidity can be easily viewed as a description of the inﬁnite covolume geometric actions that follow from the action. The C0 part of this portion of the Zimmer program is then an extension of the well-known Borel conjecture. On the topological side there is the fabulous and celebrated work of Farrell and Jones [FJ] that tells one that in the cocompact case, there are only the most obvious actions. For uniform lattices, it then follows that if one is interested in actions on Euclidean spaces, the dimension must be strictly larger than the homogeneous space, but there are interesting remarks that should be made about the proper discontinuous actions that exist in (1) the nonuniform case, (2) in dimensions larger than the original action, and (3) when there is torsion. Regarding the ﬁrst of these, it is clear that spirit of the Zimmer program demands that for any aspherical manifold with fundamental group the dimension should be at least that of G/K where G is the Lie group in which sits. It is the Q-rank that governs the difference between this dimension and the cohomological dimension of dim(G/K)–cd()=Q–rk(). When this is 1, it is not hard to give an ad hoc proof that there isn’t a geometric realization (the fundamental group of the end is a subgroup, which is a Poincaré duality group of dim=dim(G/K)–1 and of inﬁnite index). However, for higher Q-rank, I had been unable to settle it for a number of years, and was very happy when the papers [BKK] and [BF] came out and beautifully conﬁrmed the Zimmer predictions that lower-dimensional examples do not exist.

some remarks inspired by the c0 zimmer program / 267

Problem: In the minimal dimension (i.e., dim(G/K)), is there any useful way to parametrize the proper homotopy types of aspherical manifolds with fundamental group ? First of all, there is an issue of the fundamental groups at inﬁnity. It is very easy to build in any nonuniform case (of dim > 2) uncountably many modiﬁcations of the standard actions that are distinguished by the fundamental group system of the complements of compact sets (viewed as a prosystem of groups). This is a straightforward modiﬁcation of constructions of contractible manifolds. However, such actions have proper comparison maps to the standard action, and are concordant to it (i.e., there is a degeneration of each of these to the standard action). Much more interesting is the observation that Kevin Whyte made (when he was my student) that the manifolds with fundamental group (F2)k obtained by taking various products of thrice-punctured spheres and punctured tori are all different from each other even up to proper type. It would be quite interesting to get a systematic understanding in some sort of homological fashion. Here the problem should be much more doable when the Q-rank is large. Perhaps the answer is quadratic in part of the group homology when Qrank > 12 dim(G/K), cubic when the fraction is around 13 dim(G/K), and so on by analogy to the work of Goodwillie, Klein, and Weiss [GKW]. Nevertheless, this is quite unclear to me at the moment and the discussion below of case (2) suggests that it could be much simpler than that. If we assume that we have the correct proper homotopy type, then one might have a hope that the proper homotopy equivalence is properly homotopic to a homeomorphism, but this is rarely the case. As Stanley Chang and I noticed, when Q-rank > 2, there is always a ﬁnite cover where this fails [ChW].3 Our obstruction is of exponent 2 and our examples are all virtually standard. On the other hand, whenever H4i (; R) = 0 for some i, surgery arguments (e.g., like the ones in that paper that make use, of course, of the Borel-Serre picture of arithmetic manifolds [BoS]4 ) produce (inﬁnitely many) manifolds in the given proper homotopy type distinguished by pi, the i-th Pontrjagin class. These examples will not go away by passing to a ﬁnite-index subgroup. (In characteristic 0, cohomology can only get bigger on passing to a ﬁnite cover.) And, many examples then follow from known results (e.g., see Borel [Bo2]). 3. There is a good heuristic for proper rigidity when Q-rank=2. It is true when Q-rank2 boils down to the fundamental group condition at inﬁnity. (In [ChW] this is interpreted and veriﬁed using Margulis arithmeticity for all, even nonarithmetic, lattices.)

268 / s. weinberger

The situation of being above the cd in the uniform case (or the nonuniform case) is very similar to the issues involved in the proper nonuniform discussion we just had. However, in the uniform case we can be much more explicit about what occurs if, say, the dimension supercedes the cd by > 2 and we assume that the fundamental group system is trivial (i.e., equivalent to the constant system ). In that case, it is not hard to show that the manifold is the interior of a manifold with boundary, which by the theorem of Browder-Casson-HaeﬂigerSullivan-Wall (see [We3]) will automatically be a topological block bundle over K\G/ . The proper homotopy types are then parametrized by the homotopy classes of maps [K\G/ : BAut(Sc−1 )] where c is the difference of dimensions. Rationally this is isomorphic to a group cohomology. For c even, the invariant is simply the Euler class of this bundle. When c is odd, a relatively straightforward calculation shows that it lies in H2(c−1) (K\G/ ; Q) and all elements are realized.5 Note: as here c is the analogue of the Q-rank in the previous discussion, I currently hope that the answers might be simpler than one thinks in the nonuniform case. Including the results of classiﬁcation6 one gets: Summary Proposition 6: The actions of a uniform lattice on K\G × R c that have appropriate fundamental group systems at ∞ (in their quotient) → [K\G/:BTopc ] if c > 2. Rationally7 this can be calculated as a sum of group cohomology groups Ht (K\G/; Q) ⊕H4i (K\G/ ; Q) where t = c, if c is even, and t = 2 (c–1) if c is odd, and the sum is overall positive i. When there is torsion in the situation is much more complicated in several respects. Even in the situation of cocompact proper discontinuous actions there can be different failures of rigidity. Some examples appear in [CK]—and are based on failures of excision in algebraic and hermitian Ktheory, that is, the functors Nil and UNil. When nonzero, these groups tend to be inﬁnitely generated torsion groups.

5. The interpretation of this, not as well known as it should be, invariant is as follows: every odd-dimensional bundle (rationally) has a section that splits off a lower-dimensional bundle. While the Euler class of this complementary bundle is not uniquely determined, its square is. 6. See the discussion after Proposition 7 below. 7. Since there are no group structures here at least on the left, it is worth clarifying here that the map from left to right is ﬁnite to 1 and has image that contains a lattice in the target vector space.

some remarks inspired by the c0 zimmer program / 269

Rather different examples, based on a connection to embedding theory and somewhat technical arguments involving stratiﬁed space theory, can be found in [We3, We4]8 for actions of crystallographic groups. These were extended in [Shi] to many other lattices. They can be inﬁnitely generated nontorsion, for example, detected by analogues of the Alexander polynomial. However, all of these examples (e.g., the algebraic and the embedding theoretic) are virtually trivial in the sense that they become standard on passing to a sufﬁciently large cover that keeps all of the torsion. (When one passes to a cover, only some of the coefﬁcients of the polynomial survive, so any particular example will be killed on passing to a suitable ﬁnite cover.) This can be proved by a geometric argument related to [SW] (the equivariant version of the α-approximation theorem [ChF]) when there are no ﬁxed sets of subgroups that are included in one another in codimension < 3.9 It is worth noting that the above cohomological discussion about the proper analysis for nonuniform lattices can be married to the discussion of the role of the singularities (and embeddings) that come from torsion. In any case, sufﬁce it to say that the analogue of the ⊕H4i (K\G/; Q) part of the previous theorem is a sum of similar terms, one for each stratum that has G-rank > 2, and the form of the term is an equivariant K-group. As a result, nonuniform lattices of Q-rank > 2 with torsion almost never have proper rigidity. In any case, all of these examples certainly can be viewed as “obtained from some procedure applied to the arithmetic examples”—although the exact procedures involved are somewhat involved and might be hard to visualize.

3. Leaving the Aspherical Setting One of the audacious aspects of the Zimmer program is that the manifolds studied are general compact manifolds. Here is a theorem of this sort from [FW]. Theorem: If M = X /π is a compact manifold so that π = π 1 (M) is torsion free and has no normal abelian subgroups, and the isometry group Iso(X ) is not

8. We are just now getting to the point where we can give complete classiﬁcations of some proper cocompact actions that are not rigid. However, it is my feeling that the picture of this subject is currently too complicated to be able to give a survey here. 9. When there are codimension 2 situations, one can use counterexamples to the “Smith conjecture” on unknottedness of ﬁxed sets of cyclic group actions on the sphere, (e.g., [Gi]) to build knot theoretic failures of rigidity that do not die in any ﬁnite cover.

270 / s. weinberger

a compact extension of π , then π is a lattice in a semisimple Lie group and there is a Riemannian ﬁber bundle M → K\G/π . If π has torsion, one has to allow the possibility of “orbiﬁbering” over an orbifold. Note that all the ﬁbers are isometric here, and that therefore the structural group of the bundle is compact. Moreover, the representations π → Iso(F) control the possible M’s. (Taking a representation with inﬁnite image gives an example where no intermediate covers are Riemannian products. Of course many of them are differentiable products.) The following result, sketched in chapter 12 of [We3], frames our discussion. Proposition 7. Suppose that M= X /π is a compact manifold whose fundamental group is a torsion-free lattice in a Lie group and that H∗ (X) is ﬁnitely generated. Then, unless dimM = 4, there is a map M → K\G/π so that the inverse image of every open ball is homeomorphic to X . So from a slightly blurred perspective this map is like a ﬁber bundle map: rather than controlling the inverse image of each point, we control the inverse image of each little open ball. The relevant M’s that occur here, though, are much richer than the ones that occur in the previous theorem. Rather than homomorphisms from π to a Lie group, the objects that occur in this proposition are much more algebraic topological. For example, when X = Sk × R n , if one restricts attention just to the homotopically trivial ﬁbrations, then the relevant classiﬁcation is (independent of k > 110 ) the homotopy classes of maps [K\G/π : G/Top], where the space G/Top was analyzed completely by SullivanKirby-Siebenmann; rationally it is ⊕H4i (K\G/π ; Q). Incidentally, even when π has nontrivial homomorphisms to O(k + 1) they never give rise to nontrivial elements in this cohomology.11 . These actions of π on Sk × R n are never “Lie theoretic.” Conjecture 7 : If M is a manifold with torsion-free fundamental group M = X /π (π = π1 M) so that H∗ (X ) is ﬁnitely generated, then there is an aspherical

10. This is due to Gc /Topc stability as explained in [We3]. 11. These cohomology classes are essentially Pontrjagin classes (that can be very high-dimensional in Top despite the low dimensionality of the “ﬁber”); the vanishing for representations is immediate from the Chern-Weil description of characteristic classes.

some remarks inspired by the c0 zimmer program / 271

homology manifold N with fundamental group π, and for any such N, there is a map f: M →N so that for any small open set O ⊂ N, f−1 (O) < O × X . There are several things that need to be explained about this conjecture. The ﬁrst is that the maps allowed here, when N is a manifold, are exactly the maps produced in the previous theorem; moreover, despite allowing more general N’s than K\G/π in the conjecture, by Quinn’s work on resolution of homology manifolds [Q], the conjecture is a consequence of the theorem. Proposition 8. The conjecture is true for π ⊂ GLn (R) discretely embedded and of ﬁnite type12 (but not necessarily a lattice) and for hyperbolic groups, assuming that cd(π ) > 4. This follows the same lines as the previous proposition, aside from a few points. The ﬁrst is that one must exclude, say, nonuniform lattices. More precisely, if π is the fundamental group of an aspherical manifold, it must satisfy Poincaré duality. That Poincaré duality follows, assuming ﬁniteness of K(π , 1), from a manifold’s having a ﬁnite universal cover as shown in [BW1] (as is the coarseness of satisfying Poincaré duality among such groups)13 . Now, the following is an immediate consequence of [BFMW1]: Proposition 9. If π is a Poincaré duality group of dimension > 4 and satisﬁes the Borel conjecture (see e.g., [FRR]14 ), then there is an aspherical homology manifold with fundamental group π.

12. I believe that if B π has some inﬁnite skeleton, then there cannot be a manifold such as our M, but I haven’t excluded this yet. 13. Nonuniform lattices never satisfy Poincaré duality. They satisfy Bieri-Eckmann duality with an inﬁnitely generated dualizing module. 14. The Borel conjecture required here is the following standard generalization of the most commonly stated one: (ignoring orientation issues for simplicity) suppose that M is an aspherical manifold and f:W → M is a proper homotopy equivalence that is a homeomorphism outside of a compact set, then f is homotopic through such maps to a homeomorphism. For a group, we demand that this hold for all aspherical M with that fundamental group. This extends the usual Borel conjecture from fundamental groups of compact aspherical manifolds to all countable groups of ﬁnite cohomological dimensional, and in the case of fundamental groups of compact aspherical manifolds is equivalent to it if one considers M and M × tori simultaneously. (In other words a counterexample to the extended Borel for some bundle over M will translate into a counterexample to ordinary Borel for M × some torus.)

272 / s. weinberger

Thus the existence of the aspherical homology manifold follows from the results of [FJ] and [BL]. I will soon outline some evidence for not believing that it is possible to improve the existence result to be a manifold. In any case: Proposition 10. Among torsion-free hyperbolic groups, being the fundamental group of a closed aspherical manifold is a coarse condition; it follows from the Gromov boundary being a sphere. The idea for this appears in a discussion of the case of homotopy tori in [BFMW1]. One glues the boundary sphere onto the universal cover to obtain a homology manifold with boundary, and then relates the local index of the interior to that of the boundary. Details will appear in a forthcoming paper with Barthels and Lueck [BLW] that discusses more generally the role of the boundary. It is also worth noting that the conjecture (afﬁrmed in Proposition 8) includes the Borel conjecture in its usual formulation. If M and N are both aspherical manifolds with fundamental group π , there is then a CE map (see [Da]) M → N, which according to [Si] is automatically a uniform limit of homeomorphisms. The conjecture thus accomplishes the liberation of the Borel conjecture from the setting of aspherical manifolds. Finally, the map comes directly from the argument in [BFMW2]. The underlying homotopy theoretic assumption enables one to set up a controlled surgery problem, which is solved by that technology. More precisely, it produces a DDP15 homology manifold controlled homotopy equivalent to the original manifold that “approximately ﬁbers over N.” However, Edwards’s characterization theorem [Da], Quinn’s resolution theory [Q], and the αapproximation theorem [ChF] then apply to show that the homology manifold is homeomorphic to the original manifold M. The situation for groups with torsion is not nearly as pleasant: there are many more sources of obstruction—however, it is consideration of these that leads me to believe that the homology manifolds arising in the conjecture cannot be replaced by manifolds. The naive suggestion would be that π should act proper discontinuously on a contractible (homology) manifold C and that there should be an equivariant

15. DDP stands for “Disjoint Disks Property”; it demands of a space Z that any pair of maps of D2 → Z can be approximated by a disjoint pair. According to Edwards [Da] a resolvable homology manifold of dimension > 4 with this property is a manifold.

some remarks inspired by the c0 zimmer program / 273

map X → C that has the properties of the conjecture. This is completely deﬂated by the failure of the Nielsen conjecture [BW2]. In any case, one would suspect that C actually should have more properties,for example, that the ﬁxed set of every ﬁnite subgroup is also contractible, like K\G in the lattice case. However, even then there are some additional obstructions related to Nil and UNil that would have to vanish. If one is willing to cross M with a manifold with zero Euler characteristic, then one can get rid of the Nil obstructions and then, if π is a lattice with only odd torsion, in rank 1, then one can afﬁrm this conjecture—presumably it is only a matter of time—until the general case will follow. However, for general groups π , even when C exists, nothing implies that the action is locally smoothable. It is very possible for action near the ﬁxed set to be modeled on nonlinearizable (say homotopy linear) actions of a ﬁnite group on the sphere. (These are constructed in [BW1].) This seems to be entirely parallel to having homology manifolds that are locally homotopy spheres (=DDP) but not actually being “locally linear,” that is, resolvable. Remark: In [BFMW1] there were a number of conjectures made about the geometric topology of DDP homology manifolds. Despite the years that have passed almost no progress has been made. The reason for this is probably because they have never been “seen”: they are constructed as GromovHausdorff limits of polyhedra that are themselves constructed using fairly high-power machinery. If one of these arose as a boundary of a hyperbolic group, it would naturally have many self-homeomorphisms, and other additional structures that could lead an optimist to hope that it would be an importat step toward understanding the nonresolvable homology manifolds.

Appendix: Actions on M×Rn This brief appendix gives a bit more information about the possible monodromies possible for the manifolds just discussed (in the torsion-free case). But we shall phrase the result in terms of the Zimmer program. Here Cat is any of the geometric categories Top, PL, or Diff and Aut(?) is the space of self-homotopy equivalences of the space?. Proposition A1. Suppose that is a countable group of ﬁnite cohomological dimension and we have a homomorphism ρ : → π0 Aut(M). Then there is a Cat action of on M×Rn for some n iff there is a lift of Bρ to BAut(M, t) where (M, t) is the component of the space of maps M → BCat that contains

274 / s. weinberger

the stable tangent bundle of M, and Aut denotes the automorphism group. Moreover, the action can be taken free. Note: Note that if the original ρ : → π0 Aut(M) does not preserve the stable tangent bundle, then the automorphism space is empty and the proposition is vacuous. If M is stably parallelizable, then the condition is equivalent to just solving the lifting problem: BAut(M) ↓ Bρ : B → Bπ0 Aut(M) This propositon is the (un)natural marriage of Cooke’s obstruction theory for lifting homotopy actions to actions [Co] combined with Mazur’s theory of stable differential topology [Ma] where one uses ﬁnite dimensionality of B to get a manifold structure when “assembling.” The proof is quite straightforward and is left to the reader. (In the parallelizable case, one simply argues that the action produced by Cooke gives rise to a ﬁnite-dimensional CW complex with a map π1 → whose induced cover is homotopy equivalent to M. “Thickening” this complex to be a parallelizable manifold gives one whose cover is M × Rn according to Mazur.) Remark: On the other hand, for ﬁnite groups this approach seems doomed to failure. The theorem does give rise to obstructions: for example, the Nielson realization problem was disproved for nilmanifolds via this obstruction; see [RS]. However, for trivial ρ clearly it is impossible to ever accomplish the task achieved by the proposition, that is, constructing a free action. There is some literature on this problem for ﬁnite groups, but it is hemmed in by natural and strong hypotheses. Note that according to Sullivan [Su] and Wilkerson [Wi] when M is a compact (or even ﬁnite-type) simply connected manifold, π0 Aut(M) is an arithmetic group. In practice the lifting problem has its subtle aspects, and I hope to devote a future paper to some examples of it. The actions produced by the method are always free and proper discontinuous. As a result, the n here can be rather larger than the cd. (Think about the case of a product of free groups [BKK] and M = a point.) Much more in the spirit of the Zimmer program would be to produce actions with small values of n. With some trepidation, I would like to suggest:

some remarks inspired by the c0 zimmer program / 275

Conjecture: If M is a compact manifold of dimension less than that of the smallest representation of an irreducible lattice in a Lie group G of rank > 1, and n is dense in the torus, then this element is unique and is the unique continuous conjugating map. Note that g can be recovered by its eigenvalues on L2 (T). Corollary. If G is a compact Lie group, then any g can be conjugated to an element of a maximal torus. Translation by g is conjugate measure theoretically to h iff there is an element of GLn (Z) taking g to h (for the usual action of GLn (Z) as automorphisms of the torus). Much more interesting is the situation of continuous conjugacy. Corollary. (G, g) is topologically conjugate to (G, h) iff G/c1() ≈ G/cl() by an isomorphism that pulls back the principle cl = cl bundles.

16. Of course, many nonergodic actions—even with continuous deformation—appear in earlier sections.

276 / s. weinberger

As a special case, consider G = SU(n) and g generic, so that < g > is dense in T the maximal torus. Then the maps associated to g and h, SU(n)/T → BT classify the principle bundles differ by the automorphism of GLn (Z); in other words, these are the same bundles, but their structure as principle bundles is different. So we want to know what kinds of automorphisms of G/T are possible. The automorphisms of the cohomology algebra of this space was much studied and for G = SU(n) the only possibility is the Weyl group combined with complex conjugation (see, e.g., [EL]). Thus we obtain and generically there is rigidity: Corollary. For generic g ∈ SU(n), translation by g is only topologically conjugate to translation by elements conguate in SU(n) to g or its complex conjugate. Much more interesting is the situation for nongeneric elements. We will see that these are frequently topologically conjugate. We shall focus on the extreme cases: where cl is ﬁnite or a circle.17 First of all I want to show some conjugacies that are “soft,” that is, follow from general principles with no calculations.18 We will consider Zk where k is a product of a large number of primes. For each prime separately, we have the action of the Weyl group moving the generator to another element of the maximal torus. However, if there are r prime factors, one has something like (#Weyl)r−1 associated representations of Zk modulo action of the Weyl element on the whole maximal torus. Proposition 12.19 These “Weyl-mixed” translations are all topologically conjugate for G = SU(n), n > 2. Proof. For simplicity of notation let us consider the case of two factors. Thus we are considering G/(P×Q) versus G/(P ×Q ) where P and Q have relative

17. In general is a product Zr × T for some torus. 18. On the other hand, they have the usual shocking feeling that follows from applying obviously discontinuous p-adic constructions on standardly manifold theoretic objects: in other words, they are reminiscent of the use of Frobenius to get actions on etale homotopy types that arose in the proof of the Adams conjecture. 19. This proposition is true for all simply connected compact Lie groups other than SU(2); Indeed, I suspect it is true smoothly and for arbitrary compact Lie groups of rank > 1 and perhaps even more general abelian subgroups of G.

some remarks inspired by the c0 zimmer program / 277

orders, and P and P are conjugate in G, as are Q and Q . First of the quotients are homotopy equivalent20 by localization theory [HMR]. To build an equivalence between (simple) spaces,21 it sufﬁces to build rationally compatible equivalences between their localizations at each prime. At P, G/P → G/(P×Q) is an equivalence, and G/P ≈ G/P by a conjugacy that is homotopic to the identity as a map of G (it is induced by an element of G, which is connected). At Q, the argument is the same. At other primes, it’s even easier; both quotients are equivalent to G. These are all visibly rationally compatible. In fact they are simple homotopy equivalent. As n > 2, the Lie group we are considering has rank >1. Therefore one can use an extra circle from the maximal torus to compute that the Reidemeister torsion of these spaces vanish. As the Whitehead groups of cyclic groups are torsion free [BMS], the Whitehead torsion is detected by Reidemeister torsion and we get its vanishing as well. Thus we can consider G/(P ×Q ) ∈ S(G/(P×Q)), where S denotes the structure set of surgery theory. We shall analyze this in terms of the “old-fashioned surgery exact sequence” of Sullivan [SU]. The ﬁrst obstruction we have is the normal invariant of this homotopy equivalence ⊂∈ [G/(P×Q) : F/Cat]. However, by passing to P and Q covers, this map clearly vanishes (on those covers, it’s homotopic to a diffeomorphism), so the map is nullhomotopic. As a result G/(P×Q) and G/(P ×Q ) are normally cobordant. To complete the proof, we have to analyze a ﬁnal surgery obstruction. If dim(G) is even, the relevant surgery group vanishes.22 However, if dim(G) is odd, then there are additional obstructions, the most prominent being the ρ–invariant. This is the G-signature of any free G-manifold bounding our manifold. If S is an S1 in T-disjoint from P×Q and P ×Q , we can use it compute the ρ–invariants of G/(P×Q) and G/(P ×Q ) simultaneously. Note that G bounds the mapping cylinder of G → G/S, and that both ﬁnite groups act homologically trivially on this manifold with boundary. Thus for both, the ρ–invariant is the signature of this manifold 20. All homotopy equivalences are assumed to preserve identiﬁcation of fundamental groups. 21. A space is simple if its fundamental group is abelian and it acts trivially on higher homotopy. Localization theory works well for nilpotent, and hence, simple spaces (see [HMR]). 22. If P and Q are odd. Otherwise there can be a Z2 that is a codimension 1 arf invariant. This element always acts trivially on topological structure sets as is well known (but can be nontrivial smoothly). In many cases this obstruction can be seen to be trivial even smoothly, e.g., if there is a ﬁxed circle for all of the elements of the Weyl group used at P and Q, using a variant of the trick used above and the “numerical Levine formula” of [CW, Dv, Pa].

278 / s. weinberger

with boundary (almost always 0) × the trivial representation. In all cases, these manifolds cannot be distinguished by their ρ–invariants, completing the argument.

Remark: It is clear that one can apply the same method to some positive dimensional topologically cyclic subgroups of the maximal torus—although one will be compelled to only mix using subgroups of the Weyl group. This, though, would give an amusing example of the use of “soft” topological methods from homotopy theory together with surgery giving rise to topological conjugacies of dynamical systems that do have some recurrence. In order to make further progress (i.e., to show that the quotient of G under the relation of topological conjugacy has trivial hausdorﬁﬁcation) we have to study what happens for individual primes. We shall facilitate matters greatly by assuming that p is a large prime compared to the dimension. Moreover, the calculations are suggestive of what occurs for the case of a single circle. Consider now Zp ⊂ T as a diagonal matrix in SU(n), denoted by (a1 , a2 , . . . , an ) where the a’s are integers mod p. We have ai = 0 since our torus is in SU(n). The Chern classes are, of course, the symmetric functions of the ai ’s. Proposition 13. For G = SU(n), two G/Zp ’s have the same homotopy type (for p large w.r.t. n) iff their ﬁrst nonzero Chern classes agree. Moreover, they then have the same homeomoprhism (=diffeomorphism) type. There is a ﬁbration G/Zp → BZp → BSU(n). The map BZp → BSU(n) factors through BZp → BT, where T is the maximal torus and it is easy enough to understand. BSU(n) through dimension = dim(G) at a large prime can be thought of as a product of Eilenberg-MacLane spaces determined by the Chern classes. So BSU(n) → K(Z, 2i) is an isomorphism. The Chern classes thus determine the Postnikov decomposition of G/Zp . The r-th k-invariant is obtained by pulling back the r + 1st Chern class from BZp . However, once a Chern class is nonzero, the cohomology from BZp pulls back trivially in higher dimensions, and G/Zp looks like that Postnikov piece (at p) × with a product of the remaining odd-dimensional spheres. This proves the proposition by the classiﬁcation of spaces via Postnikov towers (and the fact that two ﬁnitedimensional spaces are homotopy equvalent iff their Postnikov towers agree through their dimensions).

some remarks inspired by the c0 zimmer program / 279

A similar method computes their Pontrjagin classes in terms of the Chern class of the representation ρ pulled back to BZp . As a result, only one nonzero Pontrjagin class enters. If our prime is sufﬁciently large, F/Cat can be caught in terms of Pontrjagin classes and we’ve computed the normal invariant. The rest of the surgery exact sequence is computed as before. Also, if p is sufﬁciently large, Top/O has no homotopy groups in the relevant range, so all of these manifolds are diffeomophic because they are homeomorphic and their universal covers are diffeomorphic. Corollary. By choosing p large it is possible to get topological conjugacy classes of elements as dense in G as we would like. So the largest Hausdorff quotient of G/∼is a point. Now let us consider the same analysis applied to BS1 in place of BZp . Again we have ai = 0, but now as an equation in Z. Now consider the Second Chern class = 12 (02 − a2i ). So there are only ﬁnitely many circles in the maximal torus that give rise to bundles with the same Second Chern class. I have not checked the following but it seems reasonable in light of the previous discussion to believe the following: Problem: Is it true that if G is irreducible, then the map from G-conjugacy classes to topological conjugacy of the translations is ﬁnite to 1? The ﬁrst case where one sees inﬁnite indeterminacy is SU(2) × SU(2). Here the maximal torus is of dimension 2. The quotient by any circle is topologically S2 × S3 . This follows very easily from old work of Smale [Sm]. Smale showed that any simply connected spin 5-manifold with H2 E Z is diffeomorphic to S2 × S3 . That the quotient has these properties is easy enough to see from the Gysin sequence associated to the cover G/S → G/T. (S is the circle, and T is the maximal 2-torus). The classifying map for the circle bundle G → G/S is a generator of H2 , and these are equivalent under diffeomorphism. So all of these circle actions are the same, and thus we have an analysis of the topological conjugacy in this way. Remark: As Lashof and Zimmer [LZ] have observed, whenever the compact group G has lattice subgroup, examples where G/A E G/B for nonconjugate subgroups A and B give rise to different ergodic lattice actions on the same

280 / s. weinberger

manifold. Thus, the observations and methods of this section shed a bit of light on these phenomena.

References [ABJLMS] [BL] [BLW] [BMS]

[Be] [BF] [BKK] [BM] [BW1]

[BW2] [Bo1] [Bo2]

[BoS]

[BFMW1] [BFMW2] [CW] [ChF] [ChW]

G. Arzhantseva, M. Bridson, T. Janszkiewicz, I. Leary, A. Minyasin, J. Swiatkowski, Inﬁnite groups with ﬁxed point properties (preprint). A. Barthels and W. Lueck, The Borel conjecture for Cat(0) and hyperbolic groups (preprint). A. Barthels, W. Lueck, and S. Weinberger, Aspherical manifolds with hyperbolic fundamental groups (to appear). H. Bass, J. Milnor, and J.-P. Serre, Solution of the congruence subgroup problem for SLn , n ≥ 3 and Sp2n , n ≥ 2. Inst. Hautes Études Sci. Publ. Math. no. 33 (1967), 59–137. J. Benveniste, Rigidity of isometric lattice actions on compact Riemannian manifolds. Geom. Funct. Anal. 10 (2000), no. 3, 516–542. M. Bestvina and M. Feign, Proper actions of lattices on contractible manifolds. Invent. Math. 150 (2002), no. 2, 237–256. M. Bestvina, M. Kapovich, and B. Kleiner, Van Kampen’s embedding obstruction for discrete groups. Invent. Math. 150 (2002), no. 2, 219–235. M. Bestvina and G. Mess, The boundary of negatively curved groups. J. Amer. Math. Soc. 4 (1991), no. 3, 469–481. J. Block and S. Weinberger, Large scale homology theories and geometry. Geometric topology (Athens, GA, 1993), 522–569, AMS/IP Stud. Adv. Math., 2.1, American Mathematical Society, Providence, RI, 1997. J. Block and S. Weinberger, On the generalized Nielsen realization problem. Comm. Math. Helv. 83 (2008), 21–33. A. Borel, Seminar on transformation groups. Princeton University Press, Princeton, NJ, 1960. A. Borel, Introduction to the cohomology of arithmetic groups. Lie groups and automorphic forms, 51–86, AMS/IP Stud. Adv. Math., 37, American Mathematical Society, Providence, RI, 2006. A. Borel and J.-P. Serre. Corners and arithmetic groups. Avec un appendice: Arrondissement des variétés à coins, par A. Douady et L. Hérault. Comment. Math. Helv. 48 (1973), 436–491. J. Bryant, S. Ferry, W. Mio, and S. Weinberger, Topology of homology manifolds. Ann. of Math. (2) 143 (1996), no. 3, 435–467. J. Bryant, S. Ferry, W. Mio, and S. Weinberger, Desingularizing homology manifolds. Geom. Topol. 11 (2007), 1289–1314. S. Cappell and S. Weinberger, Which H-spaces are manifolds? I. Topology 27 (1988), no. 4, 377–386. T. Chapman and S. Ferry, Approximating homotopy equivalences by homeomorphisms. Amer. J. Math. 101 (1979), no. 3, 583–607. S. Chang and S. Weinberger, Topological nonrigidity of nonuniform lattices. Comm. Pure Appl. Math. 60 (2007), no. 2, 282–290.

some remarks inspired by the c0 zimmer program / 281 [CK]

[Co] [Da] [Dv] [EL] [FS1] [FS2] [FW] [FJ] [Fe] [FRR]

[FiM] [FiS] [Gi] [GKW]

[HMR]

[H] [KL] [LZ] [Ma] [Mi]

F. Connolly and T. Kosniewski, Examples of lack of rigidity in crystallographic groups. Algebraic topology Poznan 1989, 139–145, Lecture Notes in Math., 1474, Springer, Berlin, 1991. G. Cooke, Replacing homotopy actions by topological actions. Trans. Amer. Math. Soc. 237 (1978), 391–406. R. Daverman, Decompositions of manifolds, Reprint of the 1986 original. AMS Chelsea Publishing, Providence, RI, 2007. J. Davis, Evaluation of odd-dimensional surgery obstructions with ﬁnite fundamental group. Topology 27 (1988), no. 2, 179–204. J. Ewing and A. Liulevicius, Homotopy rigidity of linear actions on homogeneous spaces. J. Pure Appl. Algebra 18 (1980), no. 3, 259–267. B. Farb and P. Shalen, Lattice actions, 3-manifolds and homology. Topology 39 (2000), no. 3, 573–587. B. Farb and P. Shalen, Real-analytic actions of lattices. Invent. Math. 135 (1999), no. 2, 273–296. B. Farb and S. Weinberger, Isometries, rigidity, and universal covers, Ann. of Math. 168 (2008), 915–940. T. Farrell and L. Jones, Rigidity for aspherical manifolds with π1 ⊂ GLm (R). Asian J. Math. 2 (1998), no. 2, 215–262. S. Ferry, Homotoping ε–maps to homeomorphisms, Amer. J. Math. 101 (1979), no. 3, 567–582. S. Ferry, A. Ranicki, and J. Rosenberg, A history and survey of the Novikov conjecture. Novikov conjectures, index theorems and rigidity, Vol. 1 (Oberwolfach, 1993), 7–66, London Math. Soc. Lecture Note Ser., 226, Cambridge University Press, Cambridge, 1995. D. Fisher and G. Margulis, Almost isometric actions, property (T), and local rigidity. Invent. Math. 162 (2005), no. 1, 19–80. D. Fisher and L. Silberman, Groups not acting on manifolds, IMRN 2008 (to appear). C. Giffen, The generalized Smith conjecture. Amer. J. Math. 88 (1966), 187–198. T. Goodwillie, J. Klein, and M. Weiss, Spaces of smooth embeddings, disjunction and surgery. Surveys on surgery theory, Vol. 2, 221–284, Ann. of Math. Stud., 149, Princeton University Press, Princeton, NJ, 2001. P. Hilton, G. Mislin, and J. Roitberg, Localization of nilpotent groups and spaces. North-Holland Mathematics Studies, No. 15. Notas de Matemática, No. 55. North-Holland Publishing Co., Amsterdam-Oxford; American Elsevier Publishing Co., New York, 1975. S. Hurder, Rigidity for Anosov actions of higher rank lattices. Ann. of Math. (2) 135 (1992), no. 2, 361–410. A. Katok and J. Lewis, Global rigidity results for lattice actions on tori and new examples of volume-preserving actions. Israel J. Math. 93 (1996), 253–280. R. Lashof and R. Zimmer, Manifolds with inﬁnitely many actions of an arithmetic group. Illinois J. Math. 34 (1990), no. 4, 765–768. B. Mazur, Differential topology from the point of view of simple homotopy theory. Inst. Hautes Études Sci. Publ. Math. no. 15 (1963), 93 pp. J. Milnor, Whitehead torsion, Bull. Amer. Math. Soc. 72 (1966), 358–426.

282 / s. weinberger [Pa]

[Q] [RS] [Shi] [Si] [Sm] [SW] [Su] [We1] [We2]

[We3] [We4]

[Wi] [Zi1] [Zi2]

W. Pardon, The exact sequence of a localization for Witt groups. II. Numerical invariants of odd-dimensional surgery obstructions. Paciﬁc J. Math. 102 (1982), no. 1, 123–170. F. Quinn, An obstruction to the resolution of homology manifolds. Michigan Math. J. 34 (1987), no. 2, 285–291. F. Raymond and L. Scott, Failure of Nielsen’s theorem in higher dimensions. Arch. Math. (Basel) 29 (1977), no. 6, 643–654. N. Shirokova, thesis (University of Chicago, 1998). L. Siebenmann, Approximating cellular maps by homeomorphisms. Topology 11 (1972), 271–294. S. Smale, On the structure of 5-manifolds, Ann. of Math. (2) 75 (1962), 38–46. M. Steinberger and J. West, Approximation by equivariant homeomorphisms. Trans. Amer. Math. Soc. 302 (1987), no. 1, 297–317. D. Sullivan, Inﬁnitesimal computations in topology. Publ. Math. d’IHES 47 (1977), 269–331. S. Weinberger, Computers, rigidity, and moduli: The large scale fractal geometry of Riemannian moduli space. Princeton University Press, Princeton, NJ, 2004. S. Weinberger, Continuous versus discrete symmetry, Geometry and topology (Athens, GA, 1985), 319–323, Lecture Notes in Pure and Appl. Math., 105. Dekker, New York, 1987. S. Weinberger, The topological classiﬁcation of stratiﬁed spaces. Chicago Lectures in Mathematics. University of Chicago Press, Chicago, 1994. S. Weinberger, Nonlinear averaging, embeddings, and group actions. Tel Aviv Topology Conference: Rothenberg Festschrift (1998), 307–314, Contemp. Math., 231. American Mathematical Society, Providence, RI, 1999. C. W. Wilkerson, Applications of minimal simplicial groups. Topology 15 (1976), 111–130. R. Zimmer, Lattices in semisimple groups and distal geometric structures. Invent. Math. 80 (1985), no. 1, 123–137. R. Zimmer, Lattices in semisimple groups and invariant geometric structures on compact manifolds. Discrete groups in geometry and analysis (New Haven, CT, 1984), 152–210, Progr. Math., 67. Birkhäuser Boston, Boston, 1987.

PART 2

Analytic, Ergodic, and Measurable Group Theory

9

michael g. cowling

CALCULUS ON NILPOTENT LIE GROUPS

to my friend bob Abstract Mostow’s rigidity theorem states that if two locally symmetric spaces \G/K (where G/K is a symmetric space of the noncompact type, and is a uniform lattice in the semisimple group G) are homeomorphic, then they are diffeomorphic; the case where G = SL(2, R) has to be excluded. Mostow’s proof treats the rank 1 case and the higher-rank case differently: the former analyzes quasiconformal mappings, while the latter appeals to a theorem of Tits on mappings of spherical buildings. This note discusses a more analytical approach to the higher-rank case, which aims to unify the rank 1 and higher-rank cases, and links to ideas of Gromov and Pansu. Only the case of SL(n, R) is discussed here.

1. Introduction A key step in the proof of Mostow’s rigidity theorem [6] is to identify maps of the boundary space G/P that arise as -equivariant boundary maps of quasiisometries, and to show that all these maps are of the form xP → gxP, for some g in G (here P is a minimal parabolic subgroup of G). In the rank 1 case, the boundary maps to be considered are all “contact” mappings, and in the cases of the groups SO(1, n) and SU(1, n), the ergodicity of is also needed, but for the groups Sp(1, n) and the exceptional group F4,20 , P. Pansu [8] showed that the contact condition alone implies that the maps all arise from the action of G. Later, the differential geometer K. Yamaguchi [10] showed that, with a few exceptions, sufﬁciently smooth “contact” mappings of the boundary G/P all come from the action of the group G. However, the smoothness results needed to apply the machinery of differential geometry are not natural in the context of Mostow’s theorem. We show how the weaker condition of “Pansu differentiability” (deﬁned later), similar to the condition considered by Mostow in the rank 1 case, implies that contact mappings come from the action of G It is a pleasure to thank the anonymous referee of this chapter for his or her useful and constructive criticism.

285

286 / m. g. cowling

(with a few exceptions); for simplicity, here we consider only the case where G is SL(n, R) and n is at least 4. We intend to use this theory elsewhere to give another proof of Mostow’s theorem. If a mapping of a stratiﬁed nilpotent Lie group N is “Pansu differentiable,” then its derivative is a strata-preserving homomorphism of the Lie group or of the Lie algebra (depending on the deﬁnition used); if the mapping is invertible, then the mapping is an automorphism. We characterize the strata-preserving automorphisms of the Lie algebra utn(n, R) of the group of upper-triangular unipotent matrices UTU(n, R), which is the Iwasawa N group for the group SL(n, R). We ﬁrst of all consider the case where n = 4, and then treat the case of larger n by induction. Using this characterization, we show that mappings that are differentiable everywhere, or almost everywhere and satisfying a simple ergodicity condition, arise out of the group action. To state our theorems, we need to deﬁne the “reﬂection in the ant-diagonal” φ : SL(n, R) → SL(n, R), by the formula φ(g)ij = gn+1−j,n+1−i

∀i, j ∈ {1, . . . , n}.

It is clear that this is an automorphism of SL(n, R) that preserves UTU(n, R). Indeed, it is the automorphism that arises from the obvious nontrivial automorphism of the Dynkin diagram of the root system An−1 . Other automorphisms of SL(n, R) that preserve UTU(n, R) include conjugations by diagonal matrices. We write the group of these automorphisms as Ad (MA). We will prove the following results.

theorem 1. Suppose that N is the group of n × n unipotent upper-triangular real matrices, where n ≥ 4, and that f : N → N is Pansu differentiable at p, with a Pansu differentiable local inverse at f ( p). Then the Pansu derivative df ( p) lies in Ad (MA) or φ Ad (MA). theorem 2. Suppose that N is the group of n × n unipotent upper-triangular real matrices, where n ≥ 3, that the continuous mapping f : N → N is Pansu differentiable almost everywhere, and that df is in Ad (MA) almost everywhere. Then f is an afﬁne map, that is, the composition of a group automorphism (that commutes with dilations) and a translation. Theorem 2 has a number of interesting features. First, only pointwise differentiability is assumed. Most theorems in analysis require more than this. For instance, the Fundamental Theorem of Calculus, that is, for a function u : [a, b] → R,

calculus on nilpotent lie groups / 287

u(b) − u(a) =

b

u (x) dx,

a

requires u to be integrable. The Cauchy-Goursat theorem in complex analysis is an exception. There are interesting parallels between the Cauchy-Goursat theorem and Theorem 2: in both cases differentiability and algebraic structure of the derivative leads to strong conclusions. Next, local hypotheses give global conclusions. In complex analysis, the exponential map is differentiable and has a differentiable local inverse, but it is not one to one. Finally, this theorem also has a local version, when f is deﬁned in an open set. In this case, f comes from the action of GL(n, R) (augmented by the map φ).

2. The Pansu Derivative A Lie algebra n is said to be stratiﬁed of step s if there are nontrivial subspaces n1 , n2 , …, ns such that 1

n = n1 ⊕ n2 ⊕ · · · ⊕ ns

[n1 , nj ] = nj+1 ,

where ns+1 := {0}. A stratiﬁed algebra is graded, that is, [nk , nj ] ⊆ nk+j , and nilpotent, because [n, [n, . . . [n, n] . . . ]] = {0}, where there are s + 1 Lie brackets. This follows by induction. Stratiﬁed Lie algebras admit dilations δt , for t in R+ , deﬁned by requiring that δt X := t j X for all X in nj and extending δt to n by linearity. Clearly δt [X , Y ] = [δt X , δt Y ]

∀X ∈ nj ,

∀Y ∈ nk

because n is graded; hence δt is a Lie algebra automorphism. An example of a stratiﬁed Lie algebra is the Lie algebra utn(n, R) of n × n upper-triangular unipotent matrices. Denote by Ei,j the n × n matrix whose entries are all 0 except for the i, jth entry, which is 1. Deﬁne nk to be span{Ei,i+k : 1 ≤ i ≤ n − k} (in other words, the kth layer of n is the sum of all the root spaces where the root is the sum of exactly k simple roots). It is easy to check that the stratiﬁcation conditions 1 hold, and that utn(n, R) is of step n − 1. A stratiﬁed Lie group is a connected, simply connected, necessarily nilpotent Lie group N whose Lie algebra n is stratiﬁed. These occur in areas of mathematics including sub-Riemannian geometry, subelliptic differential operators, and nonholonomic mechanics.

288 / m. g. cowling

The exponential map exp is a bijection from a stratiﬁed Lie algebra n to the corresponding stratiﬁed Lie group N, and the dilations δt may be extended to N by conjugating by the exponential map: δt p = exp(δt log ( p))

∀p ∈ N.

A map f : Rm → Rn is (totally) differentiable at a point p in Rm if and only if lim

t→0+

f ( p + th) − f ( p) t

exists, uniformly on compacta in h; if the limit exists, then we write it as Df ( p)(h). The map Df ( p) : Rm → Rn is linear. It “is” the matrix of partial derivatives of the components of f . Suppose that N 1 and N 2 are stratiﬁed Lie groups. A map f : N 1 → N 2 is Pansu differentiable at a point p in N 1 if and only if lim δt−1 [ f ( p)−1 f ( pδt h)]

t→0+

exists, uniformly on compacta in h; if the limit exists, then we write it as Df ( p)(h). The map Df ( p) : N 1 → N 2 is a group homomorphism that commutes with dilations. It is often more convenient to work with a linearized version of the derivative; we deﬁne df ( p)X := log(DF( p) exp (X ))

∀X ∈ n1 .

The map df ( p) is a Lie algebra homomorphism from n1 to n2 that commutes with dilations, or equivalently, respects the stratiﬁcations of the algebras.

3. Proof of Theorem 1 In this section, we prove Theorem 1, that is, we characterize the stratapreserving automorphisms of utnn (R). Write N for the group UTUn (R) of all upper-triangular unipotent n × n real matrices, whose Lie algebra n is the set utnn (R) of all upper-triangular nilpotent n × n real matrices. 3.1. The Case When n = 3

It is easy to check that there are automorphisms of the algebra utn(3, R) that preserve both strata, but act as arbitrary elements of GL(2, R) on the ﬁrst stratum. Thus Pansu differentiability imposes no restrictions on the derivative, other than the strata-preserving condition and an evident determinant condition, and Theorem 1 fails.

calculus on nilpotent lie groups / 289

A. Korànyi and H.M. Reimann [3, 4] developed a theory of quasiconformal mappings on the Heisenberg group UTU3 (R); these quasiconformal mappings are Pansu differentiable, and the space thereof is inﬁnite-dimensional. Thus Pansu differentiability in the case of UTU3 (R) is not a very restrictive condition. However, Theorem 2 does hold in this case, as we shall see; indeed, the restriction that the Pansu derivative lie in Ad (MA) almost everywhere is very strong, not only for groups of upper-triangular matrices, but also for more general Iwasawa N groups. For instance, for SO(n, 1), where n ≥ 3, the hypotheses of Theorem 2 lead us to consider 1-quasiregular mappings, which form quite a limited class! 3.2. The Case When n = 4

When we take n to be 4, elements of n and of N are of the form ⎛ ⎞ ⎛ ⎞ 0 u x z 1 u x z ⎜0 0 v y ⎟ ⎜0 1 v y ⎟ ⎜ ⎟ ⎜ ⎟ 2 and ⎜ ⎟ ⎜ ⎟, ⎝0 0 0 w ⎠ ⎝0 0 1 w ⎠ 0 0 0 0 0 0 0 1 respectively, where u, v, . . . , z are real numbers. For this example, the dilations are given by ⎞ ⎛ ⎞ ⎛ 1 u x z 1 tu t 2 x t 3 z ⎜0 1 v y ⎟ ⎜0 1 tv t 2 y ⎟ ⎟ ⎜ ⎟ ⎜ δt ⎜ ⎟. ⎟=⎜ ⎝0 0 1 w ⎠ ⎝0 0 1 tw ⎠ 0 0 0 1 0 0 0 1 Let {U, V , W , X , Y , Z} be the basis of utn4 (R) such that the left-hand matrix above is equal to uU + vV + wW + xX + yY + zZ.

3

In formula 2, the group element shown is not the exponential of the algebra element shown. It is easy to check that [U, V ] = X ,

[V , W ] = Y ,

[U, Y ] = Z,

and

[X , W ] = Z,

and all other (undetermined) commutators are zero. Take a map f of UTU4 (R), locally invertible near p. Its Pansu derivative is a strata-respecting automorphism of utn4 (R). We may therefore represent df ( p) in the basis {U, V , W , X , Y , Z} introduced in formula 3 by a 6 × 6 matrix:

290 / m. g. cowling

⎛

a ⎜ ⎜d ⎜ ⎜g ⎜ ⎜0 ⎜ ⎜ ⎝0 0

b e h 0 0 0

c f i 0 0 0

0 0 0 j l 0

0 0 0 k m 0

⎞ 0 ⎟ 0⎟ ⎟ 0⎟ ⎟. 0⎟ ⎟ ⎟ 0⎠ n

It is easy to check that span{U, W } is the unique 2-dimensional abelian subalgebra of n1 , and that span{V } is the centralizer of n2 in n1 . Since df ( p) preserves these subalgebras, we may represent df ( p) by ⎛ ⎞ a 0 c 0 0 0 ⎜ ⎟ ⎜0 e 0 0 0 0 ⎟ ⎜ ⎟ ⎜g 0 i 0 0 0⎟ ⎜ ⎟ ⎜0 0 0 j k 0 ⎟ . ⎜ ⎟ ⎜ ⎟ ⎝0 0 0 l m 0 ⎠ 0 0 0 0 0 n The commutation relation [U, V ] = X implies that [TU, TV ] = TX , whence (using other relations as needed) aeX − egY = [aU + gW , eV ] = jX + lY , and hence j = ae and l = −ge. The relation [V , W ] = Y implies that k = −ce and m = ie analogously. Similarly, the relations [U, X ] = 0, [U, Y ] = Z, [W , Y ] = 0, and [W , X ] = Z yield the equations 0 = al − gj

n = am − gk

0 = ik − cm

n = ij − cl.

All these equations together also imply that aeg = 0

cei = 0

n = aei + cge.

It follows that we may represent df ( p) by a matrix, as shown below. All the entries indicated by letters have to be nonzero for the matrix to be invertible; they may depend on the point p. ⎞ ⎛ ⎞ ⎛ a 0 0 0 0 0 0 0 c 0 0 0 ⎟ ⎜ ⎟ ⎜ 0 0 0⎟ ⎜0 e 0 0 0 0 ⎟ ⎜0 e 0 ⎟ ⎜ ⎟ ⎜ ⎜0 0 i 0 0 0 ⎟ ⎜ 0 0 0⎟ ⎟. ⎜ ⎟ or ⎜g 0 0 ⎜0 0 0 ae 0 0 ⎟ ⎜0 0 0 0 −ce 0 ⎟ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ 0 0⎠ ⎝0 0 0 0 ei 0 ⎠ ⎝0 0 0 −eg 0 0 0 0 cge 0 0 0 0 0 aei 0

calculus on nilpotent lie groups / 291

3.3. The Case When n > 4

Suppose that T is a strata-preserving automorphism of n. It is easy to verify that X in n1 centralizes nn−2 if and only if X is in s, the span of the Ei,i+1 where 2 ≤ i ≤ n − 2. Consequently, T maps s to s. Suppose that TE1,2 = aE1,2 + S + bEn−1,n ,

4

where a and b are real numbers and S ∈ s. We write alg (X , Y , . . . , ) for the Lie algebra generated by X , Y , …. If any argument of alg is a set, then of course we mean the Lie algebra generated by the other arguments and the elements of this set. It is straightforward to check that dim alg (E1,2 , s) = (n − 1)(n − 2)/2, while dim alg (aE1,2 + S + bEn−1,n , s) is equal to ⎧ ⎪ if neither of a or b is 0 ⎪ ⎨n(n − 1)/2 − 1 ⎪ ⎪ ⎩

(n − 1)(n − 2)/2

if exactly one of a or b is 0

(n − 2)(n − 3)/2

if both of a or b is 0.

A strata-preserving automorphism preserves the dimensions of algebras generated by sets, and so exactly one of a and b is nonzero. By composing with φ, if necessary we may suppose that b is zero and a is not. Now T preserves RE1,2 + s, and hence preserves the algebra that this generates, which is isomorphic to utn(n − 1, R). By induction, we may suppose that this restricted version of T lies in Ad (MA), and in particular that S = 0 in formula 4. A similar argument then shows that T also maps En−1,n to a multiple of itself, and from this it follows that T lies in Ad (MA), as enunciated.

4. The Proof of Theorem 2 Theorem 2 includes the hypothesis that df ( p) is in Ad (MA) almost everywhere. This hypothesis is necessary, because we cannot say much about a map of N if its derivative changes from being in Ad (MA) to being in φ Ad (MA) in some possibly uncontrolled way. Suppose that f is Pansu differentiable everywhere. A theorem of Darboux says that the derivative of a function f that is differentiable in an interval has the intermediate value property: if f (x) = a and f (y) = b, and if c lies between a and b, then there exists z between x and y such that f (z) = c. Alternatively, we may say that f maps connected sets to connected sets. In this formulation, Darboux’s theorem has been generalized to Rn and beyond—see, for instance, J. Malý [5]. In our case, the set of possible derivatives df ( p) is the union of two disjoint open sets Ad (MA) and φ Ad (MA). This means that df ( p) is always of

292 / m. g. cowling

one type or always of the other. By composing with the group automorphism φ if necessary, we may suppose that df ( p) is always diagonal. Alternatively, if we knew that f was differentiable almost everywhere and that a group acted ergodically, preserving the set of points p where df ( p) ∈ Ad (MA) and the set of points p where df ( p) ∈ φMA, then we could deduce that df was of one type almost everywhere. In any case, we have stated Theorem 2 with the hypotheses that f is continuous and differentiable almost everywhere, and that df ( p) is in Ad (MA). For any small neighborhood of a point p in N, the set exp (REi,i+1 ) is sent into a set ϒ exp (REi,i+1 ), where ϒ is a neighborhood of f ( p); by shrinking , we may deduce that cosets of exp (REi,i+1 ) are mapped into cosets of exp (REi,i+1 ). Such a map is afﬁne, that is, it is composed of a (left) translation and a group automorphism. This is proved in the forthcoming PhD thesis of Rupert McCallum; a special case is considered in the article [1]. We give an analytic proof of the result here that conveys less geometric information than McCallum’s proof. It should be pointed out that McCallum also considers maps deﬁned on connected open subsets of a stratiﬁed group N, and the proof given here does not (obviously) extend to this case. We ﬁrst recall that it is shown in [1] that a map f of the group UTU3 (R) that sends cosets of exp (RE1,2 ) to cosets of exp (RE1,2 ) and cosets of exp (RE2,3 ) to cosets of exp (RE2,3 ) is afﬁne. Now we consider the case where n = 4. For notational convenience, we again use the basis {U, V , W , X , Y , Z} of utn4 (R) such that ⎛ ⎞ 0 u x z ⎜0 0 v y ⎟ ⎜ ⎟ ⎜ ⎟ = uU + vV + wW + xX + yY + zZ. ⎝0 0 0 w ⎠ 0 0 0 0 We consider four copies of UTU3 (R) in UTU4 (R); their Lie algebras are spanned by {U, V , X }, by {V , W , Y }, by {X , W , Z} and {U, Y , Z}. We know already that f preserves cosets of exp (RU), exp (RV ), and exp (RW ), and show successively that f preserves cosets of exp (RX ), exp (RY ), and exp (RZ). In addition, we show that various partial derivatives are constant on larger and larger sets, in order to conclude that the derivative of f itself is constant. First, f sends cosets of exp (RU) into cosets of exp (RU), and cosets of exp (RV ) into cosets of exp (RV ), and exp (RU) and exp (RV ) generate exp (RU + RV + RX ), which is isomorphic to UTU(3, R), so f sends the coset p exp (RU + RV + RX ) into the coset f ( p) exp (RU + RV + RX ). By translating by p and f ( p), we may temporarily suppose that p and f ( p) are both the

calculus on nilpotent lie groups / 293

identity. By the result for the special case of SL(3, R) in [1], we deduce that f is afﬁne on exp (RU + RV + RX ); by translating back, we deduce that f is afﬁne on p exp (RU + RV + RX ); in particular, the derivative of f , restricted to this coset, is constant, and f maps cosets of exp (RX ) into cosets of exp (RX ). Next, we repeat this argument starting with cosets of exp (RV ) and of exp (RW ), to show that f is afﬁne on p exp (RV + RW + RY ); in particular, the derivative of f , restricted to this coset, is constant, and f maps cosets of exp (RY ) into cosets of exp (RY ). Third, we repeat the argument starting with cosets of exp (RX ) and of exp (RW ), to show that f is afﬁne on p exp (RX + RW + RZ); in particular, the derivative of f , restricted to this coset, is constant. Finally, we repeat the argument, starting with cosets of exp (RU) and of exp (RY ), to show that f is afﬁne on p exp (RU + RY + RZ); in particular, the derivative of f , restricted to this coset, is constant. Recall that the derivative df ( p) of f at p may be represented as the matrix ⎛ ⎞ a 0 0 0 0 0 ⎜ ⎟ ⎜0 e 0 0 0 0 ⎟ ⎜ ⎟ ⎜0 0 i 0 0 0 ⎟ ⎜ ⎟ ⎜0 0 0 ae 0 0 ⎟ , ⎜ ⎟ ⎜ ⎟ ⎝0 0 0 0 ei 0 ⎠ 0 0 0 0 0 aei where the entries a, e, . . ., are functions of p. Our four arguments above have shown that these entries are constant along certain cosets of certain subgroups, and more precisely they show that 5

˜ = V˜ a = Ue ˜ = V˜ e = 0, Ua

6

˜ e = V˜ i = W ˜ i = 0, V˜ e = W

7

˜ (ae) = X˜ i = W ˜ i = 0, and X˜ (ae) = W

8

˜ = Y˜ (ei) = Ua ˜ = Y˜ (ai) = 0, Ua

where T˜ stands for the left-invariant vector ﬁeld associated to the lie algebra element T (T being any of U, V , and so on). ˜ = V˜ e = W ˜ e = 0. This implies that From Equations 5 and 6, we see that Ue ˜ a+ e is constant on the whole group. Further, from Equation 7, we see that eW ˜ ˜ aW e = 0, whence from Equation 6, W a = 0, and combined with Equation ˜ = V˜ a = W ˜ a = 0, whence a is constant on the whole 5, this implies that Ua group. Similarly, i is constant on the whole group, and this implies that f is afﬁne.

294 / m. g. cowling

The case when n ≥ 5 is proved similarly. There are many more copies of UTU3 inside UTUn , but we can still apply the result of [1] to each of these, and build up to the desired conclusion in much the same way as in the case when n = 4.

5. Rigidity for Stratified Groups One of the key ideas of Pansu [8] is that most stratiﬁed groups are rigid, in the sense that the dimension of the space of differentiable maps thereon is ﬁnite (even when we consider maps deﬁned on an open subset). As mentioned above, most Iwasawa N groups are rigid in this sense. The exceptions are the N groups coming from SO(n, 1), SU(n, 1), SL(3, F), and Sp(2, F), where F is either R or C. At the present time, the published proofs of this fact known to this author involve prolongation and use of the Borel-Weil theorem in one case [10] and a careful study of contact ﬂows in the other [2], and require more smoothness than just differentiability. In a paper in preparation, P. Ciatti and the author of this paper classify the strata-preserving automorphisms of all the Iwasawa N groups, by simpler, hands-on methods. The groups listed above are precisely those where the automorphism group is not a ﬁnite extension of Ad (MA) (where KAN and MAN are an Iwasawa decomposition of the associated semisimple group G and a minimal parabolic subgroup of G). If we include the stronger hypothesis that the automorphisms preserve the root spaces, then the two rank 1 families are the only exceptions. I believe that this result, combined with the methods of this chapter, will lead to another proof of rigidity, which treats the rank 1 and higher-rank cases similarly. There is interesting related work going on at the moment on differentiable maps of more general stratiﬁed groups. For instance, it has been shown recently that differentiable maps of free nilpotent Lie groups that are of step at least 4, or of step 3 with at least three generators, are all afﬁne [9]. General criteria for rigidity are also appearing [7].

References [1]

ˇ A. Cap, M.G. Cowling, F. De Mari, M.G. Eastwood, and R. McCallum, The Heisenberg group, SL(3, R) and rigidity, pp. 41–52 in: Harmonic analysis, group representations, automorphic forms and invariant theory in honor of Roger E. Howe, Lecture Notes Series, Institute for Mathematical Sciences, National University of Singapore, Vol. 12. World Scientiﬁc, Singapore, 2007.

calculus on nilpotent lie groups / 295 [2]

M.G. Cowling, F. De Mari, A. Koráányi, and H.M. Reimann, Contact and conformal maps in parabolic geometry. I, Geom. Dedic. 111 (2005), 65–86. [3] A. Korányi and H.M. Reimann, Quasiconformal mappings on the Heisenberg group, Invent. Math. 80 (1985), 309–338. [4] A. Korányi and H.M. Reimann, Foundations of the theory of quasiconformal mappings on the Heisenberg group, Adv. Math. 111 (1995), 1–87. [5] J. Malý, “The Darboux property for gradients”, Real Anal. Exchange 22 (1996–1997), 167–173. [6] G.D. Mostow, Strong rigidity of locally symmetric spaces, Annals of Math. Studies, Vol. 78. Princeton University Press, Princeton, NJ 1973. [7] A. Ottazzi, A sufﬁcient condition for nonrigidity of Carnot groups, Math. Zeits. 259 (2008), 617–629. [8] P. Pansu, Métriques de Carnot-Caratheodory et quasiisométries des espaces symétriques de rang un, Annals of Math. 129 (1989), 1–60. [9] B. Warhurst, Tanaka prolungation of free Lie algebras, Geom. Dedic. 130 (2007), 59–69. [10] K. Yamaguchi, Differential systems associated with simple graded Lie algebras, Adv. Studies in Pure Math. 22 (1993), 413–494.

10

alex furman

A SURVEY OF MEASURED GROUP THEORY

to robert zimmer on the occasion of his

60 th

birthday

Abstract Measured group theory is an area of research that studies inﬁnite groups using measure-theoretic tools, and studies the restrictions that group structure imposes on ergodic-theoretic properties of their actions. This paper is a survey of recent developments focused on the notion of measure equivalence between groups, and orbit equivalence between group actions.

Contents

1

Introduction Disclaimer Acknowledgments Organization of the Paper Preliminary Discussion and Remarks 2.1 Lattices and Other Countable Groups 2.2 Orbit Equivalence in Ergodic Theory 2.3 Further Comments on QI, ME, and Related Topics Measure Equivalence Between Groups 3.1 Measure Equivalence Invariants 3.2 Orbit/Measure Equivalence Rigidity 3.3 How Many Orbit Structures Does a Given Group Have? Measured Equivalence Relations 4.1 Basic Deﬁnitions 4.2 Invariants of Equivalence Relations 4.3 Rigidity of Equivalence Relations Techniques 5.1 Superrigidity in Semisimple Lie Groups 5.2 Superrigidity for Product Groups 5.3 Strong Rigidity for Cocycles

2

3

4

5

Supported by NSF Grant DMS 0604611, and BSF Grant 2004345

296

297 298 298 298 299 299 303 306 308 309 321 325 327 327 331 344 348 348 351 353

a survey of measured group theory / 297

5.4 Cocycle Superrigid Actions 5.5 Constructing Representations 5.6 Local Rigidity for Measurable Cocycles 5.7 Cohomology of cocycles 5.8 Proofs of Some Results Appendix A. Cocycles A.1 The Canonical Class of a Lattice, (Co)induction A.2 OE-Cocycles A.3 ME-Cocycles References

354 355 357 358 362 365 366 367 367 368

1. Introduction This survey concerns an area of mathematics that studies inﬁnite countable groups using measure-theoretic tools, and studies ergodic theory of group actions, emphasizing the impact of group structure on the actions. Measured group theory is a particularly ﬁtting title as it suggests an analogy with geometric group theory. The origins of measured group theory go back to the seminal paper of Robert Zimmer [139], which established a deep connection between questions on orbit equivalence in ergodic theory to Margulis’s celebrated superrigidity theorem for lattices in semisimple groups. The notion of amenable actions, introduced by Zimmer in an earlier work [138], became an indispensable tool in the ﬁeld. Zimmer continued to study orbit structures of actions of large groups in [32, 41, 140–144, 146, 147] and [135]. The monograph [146] had a particularly big impact on both ergodic theorists and people studying big groups, as well as researchers in other areas, such as operator algebras and descriptive set theory.1 In recent years several new layers of results have been added to what we called measured group theory, and this paper aims to give an overview of the current state of the subject. Such a goal is unattainable—any survey is doomed to be partial, biased, and outdated before it appears. Nevertheless, we shall try our best, hoping to encourage further interest in this topic. The reader is also referred to Gaboriau’s paper [58], which contains a very nice overview of some of the results discussed here, and to Shalom’s survey [133], which is even closer to the present paper (hence the similarity of the titles). The monographs by Kechris and Miller [81] and the forthcoming one [80] by Kechris include topics 1. Zimmer’s cocycle superrigidity proved in [139] plays a central role in another area of research, vigorously pursued by Zimmer and others, concerning actions of large groups on manifolds. David Fisher surveys this direction in [42] in this volume.

298 / a. furman

in descriptive set theory related to measured group theory. Readers interested in connections to von Neumann algebras are referred to Vaes’s [136], Popa’s [114], and references therein. The scope of this paper is restricted to interaction of inﬁnite groups with ergodic theory, leaving out the connections to the theory of von Neumann algebras and descriptive set theory. When possible, we try to indicate proofs or ideas of proofs for the stated results. In particular, we chose to include a proof of one cocycle superrigidity theorem (Theorem 5.21), which enables a self-contained presentation of a number of important results: a very rigid equivalence relation (Theorem 4.19) with trivial fundamental group and outer automorphism group (Theorem 4.15), an equivalence relation that cannot be generated by an essentially free action of any group (§4.3.1).

disclaimer. As usual, the quoted results are often presented not in the full possible generality, so the reader should consult the original papers for full details. The responsibility for inaccuracies, misquotes, and other ﬂaws lies solely with the author of these notes. acknowledgments: I would like to express my deep appreciation to Bob Zimmer for his singular contribution to the subject. I would also like to thank Miklos Abert, Aurélien Alvarez, Uri Bader, Damien Gaboriau, Alexander Kechris, Sorin Popa, Yehuda Shalom, and the referee for the corrections and comments on the earlier drafts of the paper. organization of the paper. The paper is organized as follows: the next section is devoted to a general introduction that emphasizes the relations among measure equivalence, quasi-isometry, and orbit equivalence in ergodic theory. One may choose to skip most of this, but read Deﬁnition 2.1 and the following remarks. Section 3 concerns groups considered up to measure equivalence. Section 4 focuses on the notion of equivalence relations with orbit relations as a prime (but not only) example. In both of these sections we consider separately the invariants of the studied objects (groups and relations) and rigidity results, which pertain to questions of classiﬁcation. Section 5 describes the main techniques used in these theories (mostly for rigidity): a discussion of superrigidity phenomena and some of the ad hoc tools used in the subject; generalities on cocycles appear in Appendix A.

a survey of measured group theory / 299

2. Preliminary Discussion and Remarks This section contains an introduction to measure equivalence and related topics and contains a discussion of this framework. Readers familiar with the subject (especially Deﬁnition 2.1 and the following remarks) may skip to the next section in the ﬁrst reading. There are two natural entry points to Measured group theory, corresponding to the ergodic-theoretic and group-theoretic perspectives. Let us start from the latter. 2.1. Lattices and Other Countable Groups

When should two inﬁnite discrete groups be viewed as closely related? Isomorphism of abstract groups is an obvious, maybe trivial, answer. The next degree of closeness would be commensurability: two groups are commensurable if they contain isomorphic subgroups of ﬁnite index. This relation might be relaxed a bit further, by allowing to pass to a quotient modulo ﬁnite normal subgroups. The algebraic notion of being commensurable, modulo ﬁnite kernels, can be vastly generalized in two directions: measure equivalence (measured group theory) and quasi-isometry (geometric group theory). The key notion discussed in this paper is that of measure equivalence of groups, introduced by Gromov in [66, 0.5.E].

definition 2.1. Two inﬁnite discrete countable groups , are measure ME equivalent (abbreviated as ME, and denoted ∼ ) if there exists an inﬁnite measure space (, m) with a measurable, measure-preserving action of × , so that both actions (, m) and (, m) admit ﬁnite-measure fundamental domains Y , X ⊂ : γY = λX . = γ ∈

λ∈

The space (, m) is called a (, )-coupling or ME-coupling. The index of to in is the ratio of the measures of the fundamental domains m(X ) meas(/) = . [ : ] = m(Y ) meas(/ ) We shall motivate this deﬁnition after making a few immediate comments. a) The index [ : ] is well deﬁned—it does not depend on the choice of the fundamental domains X , Y for /, / respectively, because their measures are determined by the group actions on (, m). However,

300 / a. furman

a given pair (, ) might have ME-couplings with different indices (the ∗ corresponding to possible set {[ : ] } is a coset of a subgroup of R+ indices [ : ] of self--couplings. Here it makes sense to focus on ergodic couplings only). b) Any ME-coupling can be decomposed into an integral over a probability space of ergodic ME-couplings, that is, ones for which the × -action is ergodic. c) measure equivalence is indeed an equivalence relation between groups: for any countable the action of × on = with the counting measure m by (γ1 , γ2 ) : γ → γ1 γ γ2−1 provides the trivial self-ME-coupling, giving reﬂexivity; symmetry is obvious from the deﬁnition;2 while transitivity follows from the following construction of composition, or fusion, of ME-couplings. If (, m) is a (1 , 2 ) coupling and ( , m ) is a (2 , 3 ) coupling, then the quotient = × 2 of × under the diagonal action γ2 : (ω, ω ) → (γ2 ω, γ2−1 ω ) inherits a measure m = m × 2 m so that ( , m ) becomes a (1 , 3 ) coupling structure. The indices satisfy: [1 : 3 ] = [1 : 2 ] · [2 : 3 ] . d) The notion of ME can be extended to the broader class of all unimodular locally compact second countable groups: a ME-coupling of G and H is a measure space (, m) with measure space isomorphisms i : (G, mG ) × (Y , ν) ∼ = (, m),

j : (H, mH ) × (X , μ) ∼ = (, m)

with (X , μ), (Y , ν) being ﬁnite measure spaces, so that the actions G (, m), H (, m) given by g : i(g , y) → i(gg , y), h : j(h , x) → j(hh , x) commute. The index is deﬁned by [G : H] = μ(X )/ν(Y ). e) Measure equivalence between countable groups can be viewed as a category, whose objects are countable groups and morphisms between, say and , are possible (, ) couplings. Composition of morphisms is the operation of composition of ME-couplings as in (c). The trivial ME-coupling (, m ) is nothing but the identity of the object . It is also useful to consider quotient maps : (1 , m1 ) → (2 , m2 ) between (, )-couplings (these are 2-morphisms in the category), which are assumed to be × nonsingular maps, that is, ∗ [m1 ] ∼ m2 . Since 2. One should formally distinguish between (, m) as a (, ) coupling, and the same space with the same actions as a (, ) coupling; hereafter we shall denote the latter by ˇ m). (, ˇ Example 2.2 illustrates the need to do so.

a survey of measured group theory / 301

preimage of a fundamental domain is a fundamental domain, it follows (under the ergodicity assumption) that m1 ( −1 (E)) = c · m2 (E), E ⊂ 2 , where 0 < c < ∞. ME self-couplings of that have the trivial -coupling are especially useful, their cocycles are conjugate to isomorphisms. Similarly, (, )-couplings, which have a discrete coupling as a quotient, correspond to virtual isomorphisms (see Lemma 4.18). f ) Finally, one might relax the deﬁnition of quotients by considering equivariant maps : 1 → 2 between (i , i )-couplings (i , mi ) with respect to homomorphisms 1 → 2 , 1 → 2 with ﬁnite kernels and cokernels. Gromov’s motivation for ME comes from the theory of lattices. Recall that a subgroup of a locally compact second countable (lcsc for short) group G is a lattice if is discrete in G and the quotient space G/ carries a ﬁnite G-invariant Borel regular measure (necessarily unique up to normalization); equivalently, if the -action on G by left (equivalently, right) translations admits a Borel fundamental domain of ﬁnite positive Haar measure. A discrete subgroup < G with G/ being compact is automatically a lattice. Such lattices are called uniform or cocompact; others are nonuniform. The standard example of a nonuniform lattice is = SLn (Z) in G = SLn (R). Recall that a lcsc group that admits a lattice is necessarily unimodular. A common theme in the study of lattices (say in Lie, or algebraic groups over local ﬁelds) is that certain properties of the ambient group are inherited by its lattices. From this perspective it is desirable to have a general framework in which lattices in the same group are considered equivalent. Measure equivalence provides such a framework. ME

example 2.2. If and are lattices in the same lcsc group G, then ∼ ; the group G with the Haar measure mG is a (, ) coupling where (γ , λ) : g → γ gλ−1 . ME

ME

(In fact, ∼ G ∼ if ME is considered in the broader context of unimodular lcsc groups: G × {pt} ∼ = × G/ ). This example also illustrates the fact that ˇ is better related to the original (, )-coupling G the dual (, )-coupling G −1 via g → g rather than the identity map. In geometric group theory the basic notion of equivalence is quasi-isometry QI

(QI). Two metric spaces (Xi , di ), i = 1, 2 are quasi-isometric (notation: X1 ∼ X2 ) if there exist maps f : X1 → X2 , g : X2 → X1 , and constants M, A so that

302 / a. furman

d2 ( f (x), f (x )) < M · d1 (x, x ) + A

(x, x ∈ X1 )

d2 (g(y), g(y )) < M · d2 (y, y ) + A

(y, y ∈ X2 )

d1 (g ◦ f (x), x) < A

(x ∈ X1 )

d2 ( f ◦ g(y), y) < A

(y ∈ X2 ).

Two ﬁnitely generated groups are QI if their Cayley graphs (with respect to some/any ﬁnite sets of generators) are QI as metric spaces. It is easy to see that ﬁnitely generated groups commensurable modulo ﬁnite groups are QI. Gromov observes that QI between groups can be characterized as topological equivalence (TE) deﬁned in the following statement.

theorem 2.3. (gromov [66, theorem 0.2.C2 ]) Two ﬁnitely generated groups and are quasi-isometric iff there exists a locally compact space with a continuous action of × , where both actions and are properly discontinuous and cocompact. The space X in the above statement is called a TE-coupling. Here is an idea for the proof. Given a TE-coupling one obtains a quasi-isometry from any point p ∈ by choosing f : → , g : → so that γ p ∈ f (γ )X and λp ∈ g(λ)Y , where X , Y ⊂ are open sets with compact closures and = γ ∈ γ Y = λ∈ λX . To construct a TE-coupling from a quasi-isometry f : → , consider the pointwise closure of the × -orbit of f in the space of all maps → where acts by precomposition on the domain and by postcomposition on the image. For more details see the guided exercise in [67, p. 98]. A nice instance of QI between groups is a situation where the groups admit a common geometric model. Here a geometric model for a ﬁnitely generated group is a (complete) separable metric space (X , d) with a properly discontinuous and cocompact action of on X by isometries. If X is a common QI

QI

geometric model for 1 and 2 , then 1 ∼ X ∼ 2 . For example, fundamental groups i = π1 (Mi ) of compact locally symmetric manifolds M1 and M2 ˜ 2 = X have X as a common geometric ˜1∼ with the same universal cover M =M model. Notice that the common geometric model X itself does not serve as a TE-coupling because the actions of the two groups do not commute. However, a TE-coupling can be explicitly constructed from the group G = Isom (X , d), which is a locally compact (in fact, compactly generated due to ﬁnite generation assumption on i ) second countable group. Indeed, the isometric actions

a survey of measured group theory / 303

i (X , d) deﬁne homomorphisms i → G with ﬁnite kernels and images being uniform lattices. Moreover, the converse is also true: if 1 , 2 admit homomorphisms with ﬁnite kernels and images being uniform lattices in the same compactly generated second countable group G, then they have a common geometric mode—take G with a (pseudo-)metric arising from an analogue of a word metric using compact sets. Hence all uniform lattices in the same group G are QI to each other. Yet, typically, nonuniform lattices in G are not QI to uniform ones—see Farb’s survey [37] for the QI classiﬁcation for lattices in semisimple Lie groups. To summarize this discussion: the notion of measure equivalence is an equivalence relation between countable groups, an important instance of which is given by groups that can be embedded as lattices (uniform or not) in the same lcsc group. It can be viewed as a measure-theoretic analogue of the equivalence relation of being quasi-isometric (for ﬁnitely generated groups) by taking Gromov’s topological equivalence point of view. An important instance of QI/TE is given by groups that can be embedded as uniform lattices in the same lcsc group. In this situation one has both ME and QI. However, we should emphasize that this is merely an analogy: the notions of QI and ME do not imply each other. 2.2. Orbit Equivalence in Ergodic Theory

Ergodic theory investigates dynamical systems from a measure-theoretic point of view. Hereafter we shall be interested in measurable, measurepreserving group actions on a standard nonatomic probability measure space, and will refer to such actions as probability measure preserving (p.m.p.). It is often convenient to assume the action to be ergodic, that is, to require all measurable -invariant sets to be null or conull (that is, μ(E) = 0 or μ(X \ E) = 0). A basic question in this context concerns possible orbit structures of actions. Equivalence of orbit structures is captured by the following notions of orbit equivalence (the notion of an orbit structure itself is discussed in §4.1).

definition 2.4. Two p.m.p. actions (X , μ) and (Y , ν) are orbit OE equivalent (abbreviated OE, denoted (X , μ) ∼ (Y , ν)) if there exists a measure space isomorphism T : (X , μ) ∼ = (Y , ν) which takes -orbits onto -orbits. More precisely, an orbit equivalence is a Borel isomorphism T : X ∼ = −1 Y between conull subsets X ⊂ X and Y ⊂ Y with T∗ μ(E) = μ(T E) = ν(E), E ⊂ Y and T (.x ∩ X ) = .T (x) ∩ Y for x ∈ X .

304 / a. furman

∼ Y between A weak OE, or stable OE (SOE) is a Borel isomorphism T : X = positive measure subsets X ⊂ X and Y ⊂ Y with T∗ μX = νY , where μX = μ(X )−1 · μ|X , νY = ν(Y )−1 · ν|Y , so that T (.x ∩ X ) = .T (x) ∩ Y for all x ∈ X . The index of such an SOE-map T is μ(Y )/ν(X ).

In the study of orbit structure of dynamical systems in the topological or smooth category, one often looks at such objects as ﬁxed or periodic points/orbits. Despite the important role these notions play in the underlying dynamical system, periodic orbits have zero measure and therefore are invisible from the purely measure-theoretic standpoint. Hence OE in ergodic theory is a study of the global orbit structure. This point of view is consistent with the general philosophy of “noncommutative measure theory,” that is, von Neumann algebras. Speciﬁcally, OE in ergodic theory is closely related to the theory of II1 factors as follows. In the 1940s Murray and von Neumann introduced the so-called “groupmeasure space” construction to provide interesting examples of von Neumann factors:3 given a probability measure preserving (or more generally, nonsingular) group action (X , μ) the associated von Neumann algebra M X is a cross-product of with the abelian algebra L∞ (X , μ), namely the weak closure in bounded operators on L2 ( × X ) of the algebra generated by the operators f (g, x) → f (γ g, γ .x) : γ ∈ and f (g, x) → φ(x)f (g, x) : φ ∈ L∞ (X , μ) . Ergodicity of (X , μ) is equivalent to M X being a factor. It turns out that OE

(for essentially free) OE actions X ∼ Y the associated algebras are isomorphic M X ∼ = MY , with the isomorphism identifying the abelian ∞ subalgebras L (X ) and L∞ (Y ). The converse is also true (one has to specify, in addition, an element in H 1 ( X , T))—see Feldman-Moore [39, 40]. So orbit equivalence of (essentially free p.m.p. group actions) ﬁts into the study of II1 factors M X with a special focus on the so-called Cartan subalgebra given by L∞ (X , μ). We refer the reader to Popa’s 2006 ICM lecture [114] and Vaes’s Seminar Bourbaki paper [136] for some more recent reports on this rapidly developing area. The above-mentioned assumption of essential freeness of an action (X , μ) means that, up to a null set, the action is free; equivalently, for μ-a.e. x ∈ X the stabilizer {γ ∈ : γ .x = x} is trivial. This is a natural assumption, when one wants the acting group to “fully reveal itself” in a.e. orbit of the action. Let us now link the notions of OE and ME. 3. von Neumann algebras whose center consists only of scalars.

a survey of measured group theory / 305

theorem 2.5. Two countable groups and are measure equivalent iff they admit essentially free (ergodic) probability measure-preserving actions (X , μ) and (Y , ν) that are stably orbit equivalent. (SOE) =⇒ (ME) direction is more transparent in the special case of orbit equivalence, that is, index 1. Let α : × X → be the cocycle associated to an orbit equivalence T : (X , μ) → (Y , ν) deﬁned by T (g.x) = α(g, x).T (x) (here freeness of Y is used). Consider (, m) = (X × , μ × m ) with the actions 2.1 g : (x, h) → (gx, α(g, x)h),

h : (x, k) → (x, hk−1 )

(g ∈ , h ∈ ).

Then X × {1} is a common fundamental domain for both actions (note that here freeness of X is used). Of course, the same coupling (, m) can be viewed as (Y × , ν × m ) with the -action deﬁned using β : × Y → given by T −1 (h.y) = β(h, y).T −1 (y). In the more general setting of stable OE one needs to adjust the deﬁnition for the cocycles (see [45]) to carry out a similar construction. Alternative packaging for the (OE) =⇒ (ME) argument uses the language of equivalence relations (see §4.1). Identifying Y with X via T −1 , one views RY and R X as a single relation R. Taking = R equipped with the measure μ˜ (§4.1) consider the actions g : (x, y) → (g.x, y),

h : (x, y) → (x, h.y)

(g ∈ , h ∈ ).

Here the diagonal embedding X → R, x → (x, x), gives the fundamental domain for both actions. (ME) =⇒ (SOE). Given an ergodic (, ) coupling (, m), let X , Y ⊂ be fundamental domains for the , actions; these may be chosen so that m(X ∩ Y ) > 0. The ﬁnite measure-preserving actions 2.2

X ∼ = /,

Y ∼ = /

have weakly isomorphic orbit relations, since they appear as the restrictions to X and Y of the relation R× (of type II∞ ); these restrictions coincide on X ∩ Y . The index of this SOE coincides with the ME-index [ : ] (if [ : ] = 1 one can ﬁnd a common fundamental domain X = Y ). The only remaining issue is that the actions X ∼ = /, Y ∼ = / may not be essential free. This can be ﬁxed (see [56]) by passing to an extension : ¯ m) ¯ ¯ are essentially free. Indeed, (, ¯ → (, m) where / and / ¯ = × Z × W , where Z and W are free probability measuretake preserving actions and let g : (ω, z, w) → (gω, gz, w),

h : (ω, z, w) → (hω, z, hw)

(g ∈ , h ∈ ).

306 / a. furman

remark 2.6. Freeness of actions is mostly used in order to deﬁne the rearrangement cocycles for a (stable) orbit equivalence between actions. However, if SOE comes from a ME-coupling, the well-deﬁned ME-cocycles satisfy the desired rearrangement property (such as T (g.x) = α(g, x).T (x)) and freeness becomes superﬂuous. ¯ → is as above and X¯ , Y¯ denote the preimages of X , Y , then X¯ , Y¯ are If : SOE , fundamental domains and the OE-cocycles X¯ ∼ Y¯ coincide with the ME-cocycles associated with X , Y ⊂ . Another, essentially equivalent, point of view is that ME-coupling deﬁnes a weak isomorphism between the groupoids / and / . In case of free actions these groupoids reduce to their relations groupoids, but in general the information about stabilizers is carried by the ME-cocycles.

2.3. Further Comments on QI, ME, and Related Topics

Let be Gromov’s topological equivalence between and . Then any point x ∈ deﬁnes a quasi-isometry qx : → (see the sketch of proof of Theorem 2.3). In ME the maps α(−, x) : → deﬁned for a.e. x ∈ X play a similar role. However, due to their measure-theoretic nature, such maps are insigniﬁcant taken individually, and are studied as a measured family with the additional structure given by the cocycle equation. Topological and measure equivalences are related to the following interesting notion, introduced by Nicolas Monod in [98] under the appealing term “randomorphisms.” Consider the Polish space of all maps f : → with the product uniform topology, and let [, ] = f : → : f (e ) = e . Then acts on [, ] by g : f (x) → f (xg)f (g)−1 , x ∈ . The basic observation is that homomorphisms → are precisely -ﬁxed points of this action.

definition 2.7. A randomorphism is a -invariant probability measure on [, ]. A measurable cocycle c : × X → over a p.m.p. action (X , μ) deﬁnes a randomorphism by pushing forward the measure μ by the cocycle x → c(−, x). Thus orbit equivalence cocycles (see Appendix A.2) correspond to randomorphisms supported on bijections in [, ]. Also note that the

a survey of measured group theory / 307

natural composition operation for randomorphisms, given by the pushforward of the measures under the natural map [1 , 2 ] × [2 , 3 ] → [1 , 3 ],

( f , g) → g ◦ f

corresponds to the composition of couplings. The viewpoint of topological dynamics of the -action on [, ] may be related to quasi-isometries and topological equivalence. For example, points in [, ] with precompact orbits correspond to Lipschitz embeddings → .

2.3.1. using me for qi Although measure equivalence and quasi-isometry are parallel in many ways, these concepts are different and neither one implies the other. Yet, Yehuda Shalom has shown [132] how one can use ME ideas to study QI of amenable groups. The basic observation is that a topological coupling of amenable groups and carries a × -invariant measure m (coming from a -invariant probability measure on /), which gives a measure equivalence. It can be thought of as an invariant distribution on quasi-isometries → , and can be used to induce unitary representations, cohomology with unitary coefﬁcients, and so on from to . Using such constructions, Shalom [132] was able to obtain a list of new QI invariants in the class of amenable groups, such as (co)homology over Q, ordinary Betti numbers βi () among nilpotent groups, and others. Shalom also studied the notion of uniform embedding (UE) between groups and obtained group invariants, which are monotonic with respect to UE. In [125] Roman Sauer obtains further QI-invariants and UE-monotonic invariants using a combination of QI, ME, and homological methods. In another work [126] Sauer used ME point of view to attack problems of purely topological nature related to the work of Gromov. 2.3.2. p -measure equivalence Let and be ﬁnitely generated groups, equipped with some word metrics | · | , | · | . We say that a (, ) coupling (, m) is p for some 1 ≤ p ≤ ∞ if there exist fundamental domains X , Y ⊂ so that the associated ME-cocycles (see Appendix A.3) α : × X → and β : × Y → satisfy ∀g ∈ :

|α(g, −)| ∈ Lp (X , μ),

∀h ∈ :

|β(h, −)| ∈ Lp (Y , ν).

If an p -ME-coupling exists, say that and are p -ME. Clearly any p ME-coupling is q for all q ≤ p. So 1 -ME is the weakest and ∞ -ME is the most stringent among these relations. One can check that p -ME is an

308 / a. furman

equivalence relation on groups (the p condition is preserved under composition of couplings), so we obtain a hierarchy of p -ME categories with 1 -ME being the weakest (largest classes) and at p = ∞ one arrives at ME + QI. Thus p -ME amounts to measure equivalence with some geometric ﬂavor. The setting of 1 -ME is considered in [13, 14] by Uri Bader, Roman Sauer, and the author to analyze rigidity of the least rigid family of lattices—lattices in SOn,1 (R) Isom (HnR ), n ≥ 3, and fundamental groups of general negatively curved manifolds. It should be noted that examples of non amenable ME groups that are not 1 -ME seem to be rare (surface groups and free groups seem to be the main culprits). In particular, it follows from Shalom’s computations in [131] that for n ≥ 3 all lattices in SOn,1 (R) are mutually 1 -ME. We shall return to invariants and rigidity in the 1 -ME framework in Sections 3.1.8 and 3.2.4.

3. Measure Equivalence Between Groups This section is concerned with the notion of measure equivalence between ME

countable groups ∼ (Deﬁnition 2.1). First recall the following deep result (extending previous work of Dye [33, 34] on some amenable groups, and followed by Connes-Feldman-Weiss [27] concerning all nonsingular actions of all amenable groups).

theorem 3.1. (ornstein-weiss [105]) Any two ergodic probability measure-preserving actions of any two inﬁnite countable amenable groups are orbit equivalent. This result implies that all inﬁnite countable amenable groups are ME; moreover, for any two inﬁnite amenable groups and there exists an ergodic ME-coupling with index [ : ] = 1 (hereafter we shall denote this situaOE

tion by ∼ ). Measure equivalence of all amenable groups shows that many QI-invariants are not ME-invariants; these include growth type, being virtually nilpotent, (virtual) cohomological dimension, ﬁnite generations/presentation, and so on. The following are basic constructions and examples of measure equivalent groups: 1) If and can be embedded as lattices in the same lcsc group, then ME

∼ .

a survey of measured group theory / 309 ME

ME

2) If i ∼ i for i = 1, . . . , n, then 1 × · · · × n ∼ 1 × · · · × n . OE

3) If i ∼ i for i ∈ I (i.e. the groups admit an ergodic ME-coupling with OE

index 1), then ( ∗i∈I i ) ∼ ( ∗i∈I i ).4 For 2 = n, m < ∞ the free groups Fn and Fm are commensurable, and thereME

fore are ME (however, F∞ ∼ F2 ). The measure equivalence class ME(F2≤n 0 is said to weakly contain the trivial representation 1 , denoted 1 ≺ π. The trivial representation 1 is (strongly) contained in π , denoted 1 < π , if there exist non-zero π(G)-invariant vectors, that is, H π() = {0}. Of course 1 < π trivially implies 1 ≺ π. We recall: Amenability: is amenable if the trivial representation is weakly contained in the regular representation ρ : → U(2 ()), ρ(g)f (x) = f (g −1 x). OE

4. The appearance of the sharper condition ∼ in (2) is analogous to the one in the QI QI

context: if groups i and i are bi-Lipschitz, then ∗i∈I i ∼ ∗i∈I i .

310 / a. furman

Property (T): has property (T) (Kazhdan [79]) if for every unitary representation π : 1 ≺ π implies 1 < π . This is equivalent to an existence of a compact K ⊂ and > 0 so that any unitary -representation π with (K , )-almost invariant vectors has nontrivial invariant vectors. For compactly generated groups, another equivalent characterization (Delorme and Guichardet) is that any afﬁne isometric -action on a Hilbert space has a ﬁxed point, that is, if H 1 (, π ) = {0} for any (orthogonal) -representation π . We refer to [17] for the details. (HAP): is a-T-menable (or has Haagerup approximation property) if the following equivalent conditions hold: (i) has a mixing -representation weakly containing the trivial one, or (ii) has a proper afﬁne isometric action on a (real) Hilbert space. The class of inﬁnite a-T-menable groups contains amenable groups, free groups but is disjoint from inﬁnite groups with property (T). See [24] as a reference. Measure equivalence allows us to relate unitary representations of one group to another. More concretely, let (, m) be a (, ) coupling, and π : → U(H ) be a unitary -representation. Denote by H˜ the Hilbert space consisting of equivalence classes (mod null sets) of all measurable, -equivariant maps → H with square-integrable norm over a -fundamental domain:

H˜ = f : → H : f (λx) = π (λ)f (x), f 2 < ∞ mod null sets. /

The action of on such functions by translation of the argument deﬁnes a unitary -representation π˜ : → U(H˜ ). This representation is said to be induced from π : → U(H ) via . (In Example 2.2 this is precisely the usual Mackey induction of a unitary representations, of a lattice to the ambient group, followed by a restriction to another lattice). The ME invariance of the properties above (amenability, property (T), Haagerup approximation property) can be deduced from the following observations. Let (, m) be a (, ) ME-coupling, π : → U(H ) a unitary representation, and π˜ : → U(H˜ ) the corresponding induced representation. Then 1) If π is the regular -representation on H = 2 (), then π˜ on H˜ can be identiﬁed with the -representation on L2 (, m) ∼ = n · 2 (), where 2 n = dim L (/) ∈ {1, 2, . . . , ∞}. 2) If 1 ≺ π, then 1 ≺ π. ˜ 3) If (, m) is × ergodic and π is weakly mixing (i.e., 1 < π ⊗ π ∗ ), then 1 < π. ˜

a survey of measured group theory / 311

4) If (, m) is × ergodic and π is mixing (i.e., for all v ∈ H : π (h)v, v → 0 as h → ∞ in ), then π˜ is a mixing -representation. Combining (1) and (2) we obtain that being amenable is an ME-invariant. The deep result of Ornstein-Weiss [105] and Theorem 2.5 imply that any two inﬁnite countable amenable groups are ME. This gives

corollary 3.2. The measure equivalence class of Z is the class of all inﬁnite countable amenable groups ME(Z) = Amen. Bachir Bekka and Alain Valette [16] showed that if does not have property (T), then it admits a weakly mixing representation π weakly containing the trivial one. By (2) and (3) this implies that property (T) is an ME-invariant (this is the argument in [44, corollary 1.4]; see also Zimmer [146, theorem 9.1.7(b)]). The ME-invariance of amenability and Kazhdan’s property for groups indicates that it should be possible to deﬁne these properties for equivalence relations and then relate them to groups. This was done by Zimmer in [138, 141] and was recently further studied in the context of measured groupoids in [8, 9]. We return to this discussion in §4.2.1. The ME-invariance of a-T-menability follows from (2) and (4); see [24, 76].

3.1.2. cost of groups The notion of the cost of an action/relation was introduced by Levitt [90] and developed by Damien Gaboriau [53, 54, 56]; the monographs [81] and [80] also contain an extensive discussion of this topic. The cost of an essentially free p.m.p. action (X , μ), denoted cost ( X ), is the cost of the corresponding orbit relations cost (R X ) as deﬁned in §4.2.4 (it is the inﬁmum of the weights of generating systems for the groupoid where the “weight” is the sum of the measures of the domain/image sets of the generating system). The cost of an action can be turned into a group invariant/(s) by setting C ∗ () = inf cost ( X ), X

C ∗ () = sup cost ( X ) X

where the inﬁmum/supremum are taken over all essentially free p.m.p. actions of (we drop ergodicity assumption here; in the deﬁnition of C ∗ () essential freeness is also superﬂuous). Groups for which C ∗ () = C ∗ () are said to have ﬁxed price, or prix ﬁxe (abbreviated P.F.). For general groups,

312 / a. furman

Gaboriau deﬁned the cost of a group to be the lower one: C () = C ∗ (). To avoid confusion, we shall use here the notation C ∗ () for general groups, and reserve C () for P.F. groups only.

question 3.3. Do all countable groups have property P.F.? The importance of this question will be illustrated in §4.2.4; for example, a positive answer to an apparently weaker Question 4.11 would have applications to groups theory and 3-manifold (Abert-Nikolov [1]). The properties C ∗ = 1, 1 < C ∗ < ∞, and C ∗ = ∞ are ME-invariants. More precisely: ME theorem 3.4. If ∼ , then C ∗ () − 1 = [ : ] · C ∗ () − 1 for some/any (, )-coupling . We do not know whether the same holds for C ∗ . Note that in [54] this MEinvariance is stated for P.F. groups only. Proof. Let be a (, )-coupling with X = / and Y = / being free, where X , Y ⊂ are -, - fundamental domains. Given any essen¯ = × Z with tially free p.m.p. action Z, consider the (, )-coupling the actions g : (ω, z) → (gω, z),

h : (ω, z) → (hω, hz)

(g ∈ , h ∈ ).

¯ ¯ are stably orbit equivalent with and Y¯ = / The actions X¯ = / index [ : ]¯ = [ : ] = c. Hence (using Theorem 4.7 below) we have c · ( cost (R X¯ ) − 1) = cost (RY¯ ) − 1. While X¯ is a skew product over X , the action Y¯ is the diagonal action on Y¯ = Y × Z. Since Y¯ = Y × Z has Z as a -equivariant quotient, it follows (by considering preimages of any “graphing system”) that cost ( Y¯ ) ≤ cost ( Z). Since Z was arbitrary, we deduce C ∗ () − 1 ≥ c · C ∗ () − 1 . A symmetric argument completes the proof.

theorem 3.5. (gaboriau [53,54,56]) The following classes of groups have P.F.:

a survey of measured group theory / 313 1 1) Any ﬁnite group has C ∗ () = C ∗ () = 1 − || . ∗ 2) Inﬁnite amenable groups have C ∗ () = C () = 1. 3) Free group Fn , 1 ≤ n ≤ ∞, have C ∗ (Fn ) = C ∗ (Fn ) = n. 4) Surface groups = π1 ( g ) where g is a closed orientable surface of genus g ≥ 2 have C ∗ () = C ∗ () = 2g − 1. 5) Amalgamated products = A ∗C B of ﬁnite groups have P.F. with

C ∗ () = C ∗ () = 1 − (

1 1 1 + − ). |A| |B| |C|

1 In particular, C ∗ ( SL2 (Z)) = C ∗ ( SL2 (Z)) = 1 + 12 . 6) Assume 1 , 2 have P.F., then the free product 1 ∗ 2 and more general amalgamated free products = 1 ∗A 2 over an amenable group A, has P.F. with

C (1 ∗ 2 ) = C (1 ) + C (2 ),

C (1 ∗A 2 ) = C (1 ) + C (2 ) − C (A).

7) Products = 1 × 2 of inﬁnite nontorsion groups have C ∗ () = C ∗ () = 1. 8) Finitely generated groups containing an inﬁnite amenable normal subgroup have C ∗ () = C ∗ () = 1. 9) Arithmetic lattices of higher Q-rank (e.g., SLn≥3 (Z)) have C ∗ () = C ∗ () = 1. Note that for an inﬁnite group C ∗ () = 1 iff has P.F. of cost 1. So the content of cases (2), (7), (8), and (9) is that C ∗ () = 1.

question 3.6. Is it true that for all (irreducible) lattices in a (semi)simple Lie group G of higher rank have P.F. of C ∗ () = 1? Is it true that any inﬁnite group with Kazhdan’s property (T) has P.F. with C ∗ () = 1? Item (9) in Theorem 3.5 provides a positive answer to the ﬁrst question for some nonuniform lattices in higher-rank Lie groups, but the proof relies on the internal structure of such lattices (chains of pairwise commuting elements), rather than on its relation to the ambient Lie group G (which also has a lot of commuting elements). Note also that Theorem 3.4 implies that C ∗ () = 1 for all higher-rank lattices. The motivation for the second question is that (2) property (T) implies vanishing of the ﬁrst 2 -Betti number, β1 () = 0; while for inﬁnite groups it was shown by Gaboriau that 3.1

(2)

(2)

β1 () = β1 (R X ) ≤ cost (R X ) − 1.

314 / a. furman

Furthermore, there are no known examples of strict inequality. Lattices in (2) higher-rank semisimple Lie groups without property (T) still satisfy β1 () = 0 (2) (an argument in the spirit of the current discussion is β1 for ME groups are positively proportional by Gaboriau’s theorem 3.8, an irreducible lattice in a product is ME to a product of lattices and products of inﬁnite groups (2) have β1 = 0 by the Küneth formula. Shalom’s [130] provides a completely geometric explanation). To give the ﬂavor of the proofs let us indicate the argument for (8) in Theorem 3.5. Let be a group generated by a ﬁnite set {g1 , . . . , gn } and containing an inﬁnite normal amenable subgroup A and (X , μ) be an essentially free (ergodic) p.m.p. action. Since A is amenable, there is a Z-action on X with RAX = RZX (mod null sets), and we let φ0 : X → X denote the action of the generator of Z. Given > 0 one can ﬁnd a subset E ⊂ X with 0 < μ(E) < so that a∈A aE = φ0n E = X mod null sets (if A-action is ergodic, any positive measure set works; in general, one uses the ergodic decomposition). For i = 1, . . . , n let φi be the restriction of gi to E. Now one easily checks that the normality assumption implies that = {φ0 , φ1 , . . . , φn } generates R X , while cost ( ) = 1 + n. For general (not necessarily P.F.) groups i a version of (6) still holds: C ∗ (1 ∗ 2 ) = C ∗ (1 ) + C ∗ (2 ), C ∗ (1 ∗A 2 ) = C ∗ (1 ) + C ∗ (2 ) − C (A) where A is ﬁnite or, more generally, amenable. Very recently Miklos Abert and Benjamin Weiss [2] showed:

theorem 3.7. (abert-weiss [2]) For any discrete countable group , the highest cost C ∗ () is attained by nontrivial Bernoulli actions (X0 , μ0 ) and their essentially free quotients. Some comments are in order. Kechris [80] introduced the following notion: for probability measure-preserving actions of a ﬁxed group say that (X , μ) weakly contains Y if given any ﬁnite measurable partition Y =

n Yi , a ﬁnite set F ⊂ and an > 0, there is a ﬁnite measurable partition i=1 X = ni=1 Xi so that μ(gXi ∩ Xj ) − ν(gYi ∩ Yj ) < (1 ≤ i, j ≤ n, g ∈ F). The motivation for the terminology is the fact that weak containment of actions implies (but not equivalent to) weak containment of the corresponding unitary

a survey of measured group theory / 315

representations: L2 (Y ) L2 (X ). It is clear that a quotient is (weakly) contained in the larger action. It is also easy to see that the cost of a quotient action is no less than that of the original (because one can lift any graphing from a quotient to the larger action maintaining the cost of the graphing). Kechris [80] proves that this (anti)monotonicity still holds in the context of weak containment of essentially free actions of ﬁnitely generated groups, namely: Y X

=⇒

cost ( Y ) ≥ cost ( X ).

In fact, it follows from the more general fact that cost is upper semicontinuous in the topology of actions. Abert and Weiss prove that Bernoulli actions (and their quotients) are weakly contained in any essentially free action of a group. Thus Theorem 3.7 follows from the monotonicity of the cost.

3.1.3. 2 -betti numbers The 2 -Betti numbers of (coverings of) manifolds were introduced by Atiyah in [11]. Cheeger and Gromov [23] deﬁned 2 -Betti (2) numbers βi () ∈ [0, ∞], i ∈ N, for arbitrary countable group as dimensions (in the sense of Murray von Neumann) of certain homology groups (which are Hilbert -modules). For reference we suggest [35], [92]. Here let us just point out the following facts: (2)

1) If is inﬁnite amenable, then βi () = 0, i ∈ N. (2)

(2)

2) For free groups β1 (Fn ) = n − 1 and βi (Fn ) = 0 for i > 1. (2)

3) For groups with property (T), β1 () = 0. (2) (2) (2) 4) Küneth formula: βk (1 × 2 ) = i+j=k βi (1 ) · βj (2 ). 5) Kazhdan’s conjecture, proved by Lück, states that for residually ﬁnite groups satisfying appropriate ﬁniteness properties (e.g., ﬁnite K (pi, 1)) the 2 -Betti numbers are the stable limit of Betti numbers of ﬁnite-index (2) βi (n ) subgroups normalized by the index: βi () = lim [: where > 1 > n] . . . is a chain of normal subgroups of ﬁnite index. (2) 6) The 2 Euler characteristic χ (2) () = ( − 1)i · βi () coincides with the usual Euler characteristic χ () = ( − 1)i · βi (), provided both are deﬁned, as is the case for fundamental group = π1 (M) of a compact aspherical manifold. 7) According to the Hopf-Singer conjecture the 2 -Betti numbers for a fundamental group = π1 (M), of a compact aspherical manifold M vanish except, possibly, in the middle dimension n. Atiyah’s conjecture states that 2 -Betti numbers are integers.

316 / a. furman

The following remarkable result of Damien Gaboriau states that these intricate numeric invariants of groups are preserved under measure equivalence, after a rescaling by the coupling index. ME

theorem 3.8. (gaboriau [55], [57]) Let ∼ be ME-countable groups. Then (2) (2) βi () = c · βi () (i ∈ N) where c = [ : ] is an/the index of some/any (, )-coupling. In fact, Gaboriau introduced the notion of 2 -Betti numbers for II1 -relations and related them to 2 -Betti numbers of groups in case of the orbit relation for an essentially free p.m.p. action—see more comments in §4.2.5 below. Thus the geometric information encoded in the 2 -Betti numbers for fundamental groups of aspherical manifolds, such as Euler characteristic and sometimes the dimension, pass through measure equivalence. In particular, if lattices i (i = 1, 2) (uniform or not) in SUni ,1 (R) are ME, then n1 = n2 ; the same applies to Spni ,1 (R) and SO2ni ,1 (R). (The higher-rank lattices are covered by stronger rigidity statements—see §3.2.1 below). Furthermore, it follows from Gaboriau’s result that in general the set (2) D(2) () = i ∈ N : 0 < βi () < ∞ is an ME-invariant. Conjecture (7) relates this to the dimension of a manifold M in the case of = π1 (M). One shouldn’t expect dim (M) to be an ME-invariant of π1 (M), as the examples of tori show; note also that for any manifold M one ME

has π1 (M × Tn ) ∼ π1 (M × Tk ). However, among negatively curved manifolds Theorem 3.13 below shows that dim (M) is an invariant of 1 -ME. For closed aspherical manifolds M the dimension dim (M) is a QI-invariant of π1 (M). Pansu proved that the whole set D(2) () is a QI-invariant of . However, positive proportionality of 2 -Betti numbers for ME fails under QI; in fact, there are QI groups whose Euler characteristics have opposite signs. Yet

corollary 3.9. For ME groups and with well-deﬁned Euler characteristic, say fundamental groups of compact manifolds, one has χ() = c · χ (),

where

c = [ : ] ∈ (0, ∞).

In particular, the sign (positive, zero, negative) of the Euler characteristic is an ME-invariant.

a survey of measured group theory / 317

3.1.4. cowling-haagerup -invariant This numeric invariant G , taking values in [1, ∞], is deﬁned for any lcsc group G in terms of norm bounds on unit approximation in the Fourier algebra A(G) (see Cowling and Haagerup [31]). The -invariant coincides for a lcsc group and its lattices. Moreover, OE Cowling and Zimmer [32] proved that 1 ∼ 2 implies 1 = 2 . In fact, their proof implies the invariance under measure equivalence (see [76]). So, is an ME-invariant. Cowling and Haagerup [31] computed the -invariant for simple Lie groups and their lattices: in particular, proving that G = 1 for G SOn,1 (R) and SUn,1 (R), G = 2n − 1 for G Spn,1 (R), and G = 21 for the exceptional rank-1 group G = F4(−20) . One may observe that simple Lie groups split into two classes: (1) SOn,1 (R) and SUn,1 (R) family, and (2) G Spn,1 (R), F4(−20) and higher rank. Groups in the ﬁrst class have haagerup approximation property (HAP, a.k.a. a-Tmenability) and G = 1, while groups in the second class have Kazhdan’s property (T) and G > 1. Cowling conjectured that G = 1 and (HAP) might be equivalent. Recently one implication of this conjecture has been disproved: Cornulier, Stalder, and Valette [30] proved that the wreath product H F2 of a ﬁnite group H by the free group F2 has (HAP), while Ozawa and Popa [108] prove that HF2 > 1. The question whether = 1 implies (HAP) is still open. One may deduce now that if is a lattice in G Spn,1 (R) or in F4(−20) and ME

is a lattice in a simple Lie group H, then ∼ iff G H. Indeed, higherrank H are ruled out by Zimmer’s theorem 3.15; H cannot be in the families SOn,1 (R) and SUn,1 (R) by property (T) or Haagerup property; and within the family of Spn,1 (R) and F4(−20) the -invariant detects G (2 -Betti numbers can also be used for this purpose).

3.1.5. treeability, antitreeability, and ergodic dimension In [4] Scott Adams introduced the notion of treeable equivalence relations (see §4.2.3). Following [81], a group is Treeable: if there exists an essentially free p.m.p. -action with a treeable orbit relation. Strongly treeable: if every essentially free p.m.p. -action gives a treeable orbit relation. Antitreeable: if there are no essentially free p.m.p. -actions with a treeable orbit relation.

318 / a. furman

Amenable groups and free groups are strongly treeable. It seems to be still unknown whether there exist treeable but not strongly treeable groups; in particular, it is not clear whether surface groups (that are treeable) are strongly treeable. The properties of being treeable or antitreeable are ME-invariants. MoreME

over, is treeable iff is amenable (i.e., ∼ F1 = Z), or is ME to either F2 or F∞ (this fact uses Hjorth’s [69]; see [81, theorems 28.2 and 28.5]). Groups with Kazhdan’s property (T) are antitreeable [6]. More generally, it follows from the recent work of Alvarez and Gaboriau [7] that a nonamenable group (2) with β1 () = 0 is antitreeable (in view of Equation (3.1) this also strengthens [54, corollaire VI.22], where Gaboriau showed that a nonamenable with C ∗ () = 1 is antitreeable). A treeing of a relation can be seen as a -invariant assignment of pointed trees with as the set of vertices. One may view the relation acting on this measurable family of pointed trees by moving the marked point. More generally, one might deﬁne actions by relations, or measured groupoids, on ﬁelds of simplicial complexes. Gaboriau deﬁnes (see [57]) the geometric dimension of a relation R to be the smallest possible dimension of such a ﬁeld of contractible simplicial complexes; the ergodic dimension of a group will be the minimal geometric dimension over orbit relations R X of all essentially free p.m.p. -actions. In this terminology R is treeable iff it has geometric dimension 1, and a group is treeable if its ergodic dimension is 1. There is also a notion of an approximate geometric/ergodic dimension [57] describing the dimensions of a sequence of subrelations approximating a given orbit relation.

theorem 3.10. (gaboriau [57]) Ergodic dimension and approximate ergodic dimension are ME-invariants. This notion can be used to obtain some information about ME of lattices in the family of rank-1 groups SOn,1 (R). If i < SOni ,1 (R), i = 1, 2 are lattices ME

and 1 ∼ 2 , then Gaboriau’s result on 2 -Betti numbers shows that if one (2) of ni is even, then n1 = n2 . However, for ni = 2ki + 1 all βi vanish. In this case Gaboriau shows, using the above ergodic dimension, that k1 ≤ k2 ≤ 2k1 or k2 ≤ k1 ≤ 2k2 . OE

OE

3.1.6. free products It was mentioned above that if i ∼ , then ∗i∈I i ∼ OE ∗i∈I i (here ∼ means that the two groups admit an ergodic ME-coupling

a survey of measured group theory / 319

with index 1, and equivalently admit essentially free actions that are orbit equivalent). To what extent does the converse hold? Namely, when can one recognize the free factors on the level of measure equivalence? This problem was extensively studied by Ioana, Peterson, and Popa in [75] where strong rigidity results were obtained for orbit relations under certain assumptions on the actions (see §4.2.8). Here let us formulate a recent result from Alvarez and Gaboriau [7] that can be stated in purely group theoretic terms. In [7] a notion of measurably freely indecomposable groups (MFI) is introduced, and it is shown that this class includes all nonamenable (2) groups with β1 = 0. Thus, inﬁnite property (T) groups, nonamenable direct products, are examples of MFI groups. ME

theorem 3.11. (alvarez-gaboriau [7]) Suppose that ∗ni=1 i ∼ ∗m , j=1 j ME

are two sets of MFI groups with i ∼ i for 1 ≤ i = where {i }ni=1 and {j }m j=1 ME

i ≤ n, and j ∼ j for 1 ≤ j = j ≤ m. Then n = m and, up to a permutation ME

of indices, i ∼ i . Another result from [7] concerning decompositions of equivalence relations as free products of subrelations is discussed in §4.2.8. Let us also mention recent works of Kida [88] and Popa and Vaes [120] that describe extremely strong rigidity properties for certain amalgamated products of various rigid groups.

3.1.7. the classes Creg and C In §3.2.2 below we shall discuss rigidity results obtained by Nicolas Monod and Yehuda Shalom in [101] (see also [99, 100] and jointly with Mineyev [96]). These results involve second-bounded cohomology with unitary coefﬁcients: Hb2 (, π )—a certain vector space associated to a countable group and a unitary representation π : → U(Hπ ). (Some background on bounded cohomology can be found in [101, §3] or [98]; for more details see [21, 97]). Monod and Shalom deﬁne the class Creg of groups characterized by the property that Hb2 (, 2 ()) = {0} and (potentially larger) class C of groups with nonvanishing Hb2 (, π ) for some mixing -representation π . Known examples of groups in Creg ⊂ C include groups admitting “hyperboliclike” actions of the following types (see [100], [96]):

320 / a. furman

i) nonelementary simplicial action on some simplicial tree, proper on the set of edges; ii) nonelementary proper isometric action on some proper CAT(-1) space; and iii) nonelementary proper isometric action on some Gromov-hyperbolic graph of bounded valency. Hence Creg includes free groups, free products of arbitrary countable groups, and free products amalgamated over a ﬁnite group (with the usual exceptions of order 2), fundamental groups of negatively curved manifolds, Gromov hyperbolic groups, and nonelementary subgroups of the above families. Examples of groups not in C include amenable groups, products of at least two inﬁnite groups, lattices in higher-rank simple Lie groups (over any local ﬁeld), and irreducible lattices in products of general compactly generated nonamenable groups (see [101, §7]).

theorem 3.12. (monod-shalom [101]) 1) Membership in Creg or C is an ME-invariant. 2) For direct products = 1 × · · · × n where i ∈ Creg are torsion free, the number of factors and their ME types are ME-invariants. ME

3) For as above, if ∼ , then cannot be written as a product of m > n inﬁnite torsion-free factors.

3.1.8. dimension and simplicial volume (1 -me) Geometric properties are hard to capture with the notion of measure equivalence. The 2 -Betti numbers is an exception, but this invariant beneﬁts from its Hilbert space nature. In [13, 14] Uri Bader, Roman Sauer, and the author consider a restricted version of measure equivalence, namely, 1 -ME (see §2.3.2 for a deﬁnition). Being 1 ME is an equivalence relation between ﬁnitely generated groups, in which any two integrable lattices in the same lcsc group are 1 -ME. All uniform lattices are integrable, and so are all lattices in SOn,1 (R) Isom (HnR ) (see §3.2.4). theorem 3.13. (bader-furman-sauer [14]) Let i = π1 (Mi ) where Mi are closed manifolds that admit a Riemannian metric of negative sectional curvature. Assume that 1 and 2 admit an 1 -ME-coupling . Then dim (M1 ) = dim (M2 )

and

M1 = [2 : 1 ] · M2 ,

where Mi denotes the simplicial volume of Mi .

a survey of measured group theory / 321

The simplicial volume M of a closed manifold M, introduced by Gromov in [65], is the norm of the image of the fundamental class under the comparison 1 map Hn (M) → Hn (M) into the 1 -homology, which is an 1 -completion of the usual homology. This is a homotopy invariant of manifolds. Manifolds carrying a Riemannian metric of negative curvature have M > 0 (Gromov [65]). 3.2. Orbit/Measure Equivalence Rigidity

Let us now turn to measure equivalence rigidity results, that is, classiﬁcation results in the ME category. In the introduction to this section we mentioned that the ME class ME(Z) is precisely all inﬁnite amenable groups. The (distinct) classes ME(F2≤n 0, that is, rule out the torus 1,0 , once-punctured torus 1,1 , and spheres 0,p with p ≤ 4 punctures. theorem 3.20. (kida [86]) Let be a ﬁnite-index subgroup in ( g,p ) with 3g + p − 4 > 0, or in a ﬁnite product of such mapping class groups ni=1 ( g,p ) . ME

Then any group ∼ is commensurable up to ﬁnite kernels to , and ergodic ME-coupling has a discrete (, )-coupling as a quotient. This work (spanning [83, 85, 86]) is a real tour de force. Mapping class groups ( ) are often compared to a lattice in a semisimple Lie group G: the Teichmüller space T ( ) is analogous to the symmetric space G/K , Thurston boundary PML( ) analogous to Furstenberg boundary B(G) = G/P, and the curve complex C( ) to the spherical Tits’s building of G. The MCG has been extensively studied as a geometric object, while Kida’s work provides a new ergodic-theoretic perspective. For example, Kida proves that Thurston boundary PML( ) with the Lebesgue measure class is -boundary in the sense of Burger-Monod for the mapping class group, that is, the action of the latter is amenable and doubly ergodic with unitary coefﬁcients. Properties of the MCG action on PML( ) allow Kida to characterize certain subrelations/subgroupoids arising in self-measure equivalence of a MCG; leading to the proof of a cocycle (strong) rigidity Theorem 5.7, which can be viewed as a groupoid version of Ivanov’s rigidity theorem. This strong rigidity theorem can be used with §5.5 to recognize arbitrary groups ME to an MCG. Note that a mapping class group behaves like a “lattice without ambient Lie group”—all its ME-couplings have discrete quotients. Moreover, Kida’s ME rigidity results extend to products of MCGs without any irreducibility assumptions. From this point of view MCGs are more ME rigid than higherrank lattices, despite the fact that they lack many other rigidity attributes, such as property (T) (see Andersen [10]). Added in proof. Very recently additional extremely strong ME rigidity results were obtained in Kida [88] and Popa and Vaes [120] for certain amalgamated products of higher-rank lattices and also mapping class groups. The latter paper also establishes W ∗ -rigidity.

a survey of measured group theory / 325

3.2.4. hyperbolic lattices and 1 -me Measure equivalence is motivated by the theory of lattices, with ME-couplings generalizing the situation of groups embedded as lattices in the same ambient lcsc group. Thus, in the context of semisimple groups, one wonders whether ME rigidity results would parallel Mostow rigidity; and in particular would apply to (lattices in) all simple groups with the usual exception of SL2 (R) SO2,1 (R) SU1,1 (R). The higher-rank situation (at least that of simple groups) is well understood (§3.2.1). In the rank 1 case (consisting of the families SOn,1 (R), SUm,1 (R), Spk,1 (R), and F4(−20) ) known ME-invariants discussed above (namely: property (T), 2 -Betti numbers, -invariant, ergodic dimension) allow to distinguish lattices among most rank 1 groups. This refers to statements of the form: if i < Gi are ME lattices, then 1 ∼ 2 iff G1 G2 . However, ME classiﬁcation such as in Theorems 3.16, 3.17, and 3.20 are not known for rank 1 cases. The ingredient that is missing in the existing approach is an appropriate cocycle superrigidity theorem.7 In a joint work with Uri Bader and Roman Sauer a cocycle strong rigidity theorem is proved for ME-cocycles for lattices in SOn,1 (R) Isom (HnR ), n ≥ 3, under a certain 1 -assumption (see §2.3.2). It is used to obtain the following: theorem 3.21. (bader-furman-sauer [13]) Let is a lattice in G = Isom (Hn ), n ≥ 3, and is some ﬁnitely generated group 1 -ME to then is a lattice in G modulo a ﬁnite normal subgroup. Moreover, any ergodic (, )coupling has a quotient, which is either discrete, or G = Aut (G), or G0 with the Haar measure. Recently Sorin Popa has introduced a new set of ideas for studying orbit equivalence. These results, rather than relying on rigidity of the acting groups alone, exploit rigidity aspects of groups actions of certain type. We shall discuss them in §Sections 5.4, and 5.6, and 5.7. 3.3. How Many Orbit Structures Does a Given Group Have?

Theorem 3.1 of Ornstein and Weiss [105] implies that for an inﬁnite amenable countable group all ergodic probability measure preserving actions (X , μ) deﬁne the same orbit structure, namely, Ramen . What happens for non-amenable groups ? 7. For Spn,1 (R) and F4(−20) a cocycle superrigidity theorem was proved by Corlette and Zimmer [29] (see also Fisher and Hitchman [43]), but these results require boundness assumptions that preclude them from being used for ME-cocycles.

326 / a. furman

theorem 3.22. (epstein [36], after ioana [73] and gabotiaulyons [60]) Any nonamenable countable group has a continuum of essentially free ergodic probability measure-preserving actions (X , μ), no two of which are stably orbit equivalent. Let us brieﬂy discuss the problem and its solution. Since Card( Aut (X , μ) ) = ℵ = 2ℵ0 there are at most continuum many actions for a ﬁxed countable group . In fact, this upper bound on the cardinality of isomorphism classes of actions is achieved, using the corresponding fact about unitary representations and the Gaussian construction. Hence one might expect at most ℵ-many non-OE actions for any given . OE rigidity results showed that some speciﬁc classes of groups indeed have many mutually non-OE actions; this includes higher-rank lattices [64], products of hyperboliclike groups [101, theorem 1.7], and some other classes of groups [110, 113]). But the general question, regarding an arbitrary nonamenable , remained open. Most invariants of equivalence relations depend on the acting group rather than the action, and thus could not be used to distinguish actions of the ﬁxed group . The notable exception to this metamathematical statement appears for nonamenable groups that do not have property (T). For such groups two non-SOE actions can easily be constructed: (1) a strongly ergodic action (using Schmidt’s [129]), and (2) an ergodic action that is not strongly ergodic (using Connes-Weiss [28]). Taking a product with an essentially free weakly mixing strongly ergodic -actions (e.g., the Bernoulli action (X0 , μ0 ) ) makes the above two actions essentially free and distinct. In [68] Greg Hjorth showed that if has property (T), the set of isomorphism classes of orbit structures for essentially free -actions has cardinality ℵ, by proving that the natural map from the isomorphism classes of essentially free ergodic -actions to the isomorphism classes of -orbit structures is at most countable-to-one. More precisely, the space of -actions producing a ﬁxedorbit structure is equipped with a structure of a Polish space (separability) where any two nearby actions are shown to be conjugate. This is an example of proving rigidity up to countable classes combining separability of the ambient space with a local rigidity phenomenon (stemming from property (T); see §5.6 below). These ideas can be traced back to Connes [26] and Popa [106], and play a central role in the most recent developments—see [114] §4]. The challenge now became to show that other nonamenable groups have inﬁnitely, or even ℵ-many, non-OE essentially free ergodic actions. Damien Gaboriau and Sorin Popa [59] achieved this goal for the quintessential representative of a nonamenable group without property (T), namely for the free

a survey of measured group theory / 327

group F2 . Using a sophisticated rigidity vs. separability argument they showed that within a certain rich family of F2 -actions the map from isomorphism classes of actions to orbit structures is countable-to-one. The rigidity component of the argument was this time provided by Popa’s notion of w-rigid actions such as SL2 (Z) T2 , with the rigidity related to the relative property (T) for the semidirect product SL2 (Z) Z2 viewing Z2 as the Pontryagin dual of T2 . In [73] Adrian Ioana obtained a sweeping result showing that any containing a copy of F2 has ℵ-many mutually non-SOE essentially free actions. The basic idea of the construction being to use a family of non-SOE of F2 /F actions F2 Xt to construct co-induced -actions Xt 2 and pushing the solution of F2 -problem to the analysis of the co-induced actions. The class of groups containing F2 covers “most of” the class of nonamenable groups with few, very hard to obtain, exceptions. The ultimate solution to the problem, covering all nonamenable groups, was shortly obtained by Inessa Epstein [36] using a result by Damien Gaboriau and Russel Lyons [60], who proved that any nonamenable contains an F2 in a sort of measure-theoretical sense. Epstein was able to show that this sort of containment sufﬁces to carry out an analogue of Ioana’s co-induction argument [73] to prove Theorem 3.22. Furthermore, in [74] Ioana, Kechris, and Tsankov, jointly with Epstein, show that for any nonamenable the space of all ergodic-free p.m.p. actions taken up to OE not only has cardinality of the continuum, but is also impossible to classify in a very strong sense. One may also add that most of the general results mentioned above show that within certain families of actions the grouping into SOE-ones has countable classes, therefore giving only implicit families of non-SOE actions. In [71] Ioana provided an explicit list of a continuum of mutually non-SOE actions of F2 .

4. Measured Equivalence Relations 4.1. Basic Deﬁnitions

We start with the notion of countable equivalence relations in the Borel setting. It consists of a standard Borel space (X , X ) (cf. [39] for deﬁnitions) and a Borel subset R ⊂ X × X , an equivalence relation whose equivalence classes R[x] = y ∈ X : (x, y) ∈ R are all countable. To construct such relations choose a countable collection = {φi }i∈I of Borel bijections φi : Ai → Bi between Borel subsets Ai , Bi ∈ X , i ∈ I; and let R be the smallest equivalence relation including the graphs of all φi , i ∈ I. More precisely, (x, y) ∈ R iff there exists a ﬁnite sequence i1 , . . . , ik ∈ I and

328 / a. furman

1 , . . . , k ∈ {−1, 1} so that

y = φi k ◦ · · · ◦ φi22 ◦ φi11 (x). k

We shall say that the family generates the relation R . The particular case of a collection = {φi } of Borel isomorphisms of the whole space X generates a countable group = and R = R X = (x, y) : x = y = (x, γ .x) : x ∈ X , γ ∈ . Feldman and Moore [39] proved that any countable Borel equivalence relation admits a generating set whose elements are deﬁned on all of X ; in other words, any equivalence relation appears as the orbit relation R X of a Borel action X of some countable group (see §4.3.1). Given a countable Borel equivalence relation R the full group [R] is deﬁned by [R] = φ ∈ Aut (X , X ) : ∀x ∈ X : (x, φ(x)) ∈ R . The full pseudogroup [[R]] consists of partially deﬁned Borel isomorphisms ψ : Dom(ψ) →Im(ψ), so that Graph(ψ)= (x, ψ(x)) : x ∈ Dom(ψ) ⊂ R. If R is the orbit relation R X of a group action (X , X ), then any φ ∈ [R] has the following “piecewise -structure”: there exist countable partitions

Ai = X = Bi into Borel sets and elements γi ∈ with γi (Ai ) = Bi so that φ(x) = γi x for x ∈ Ai . Elements ψ of the full pseudogroup [[R ]] have a

similar “piecewise -structure” with Ai = Dom(ψ) and Bi = Im(ψ). Let R be a countable Borel equivalence relation on a standard Borel space (X , X ). A measure μ on (X , X ) is R-invariant (respectively, R-quasiinvariant) if for all φ ∈ [R], φ∗ μ = μ (respectively, φ∗ μ ∼ μ). Note that if = {φi : Ai → Bi } is a generating set for R, then μ is R-invariant iff μ is invariant under each φi , that is, μ(φi−1 (E) ∩ Ai ) = μ(E ∩ Bi ) for all E ∈ X . Similarly, quasi-invariance of a measure can be tested on a generating set. The R-saturation of E ∈ X is R[E] = x ∈ X : ∃y ∈ E, (x, y) ∈ R . A R (quasi-) invariant measure μ is ergodic if R[E] is either μ-null or μ-conull for any E ∈ X . In this section we shall focus on countable Borel equivalence relations R on (X , X ) equipped with an ergodic, invariant, nonatomic, probability measure μ on (X , X ). Such a quadruple (X , X , μ, R) is called type II1 -relation. These are precisely the orbit relations of ergodic measure-preserving actions of countable groups on nonatomic standard probability measure spaces (the nontrivial implication follows from the above-mentioned theorem of Feldman and Moore).

a survey of measured group theory / 329

Given a countable Borel relation R on (X , X ) and an R-quasi-invariant probability measure μ, deﬁne inﬁnite measures μ˜ L , μ˜ R on R by # y : (x, y) ∈ E ∩ R dμ(x), μ˜ L (E) = X

μ˜ R (E) =

X

# x : (x, y) ∈ E ∩ R dμ(y).

These measures are equivalent, and coincide if μ is R-invariant, which is our main focus. In this case we shall denote μ˜ = μ˜ L = μ˜ R .

4.1

Hereafter, saying that some property holds a.e. on R would refer to μ-a.e. ˜ (this makes sense even if μ is only R-quasi-invariant).

remark 4.1. In some situations a Borel equivalence relation R on (X , X ) has only one (nonatomic) invariant probability measure. For example, this is the case for the orbit relation of the standard action of a ﬁnite-index subgroup8 < SLn (Z) on the torus Tn = R n /Zn , or for a lattice in a simple center-free Lie group G acting on H/, where H is a simple Lie group, < H is a lattice, and acts by left translations via an embedding j : G → H with j(G) having trivial centralizer in H. In such situations one may gain understanding of the countable Borel equivalence relation R via the study of the II1 -relation corresponding to the unique R-invariant probability measure. As always in the measure-theoretic setting null sets should be considered negligible. So an isomorphism T between (complete) measure spaces (Xi , Xi , μi ), i = 1, 2, is a Borel isomorphism between μi -conull sets T : X1 → X2 with T∗ (μ1 ) = μ2 . In the context of II1 -relations, we declare two relations (Xi , Xi , μi , Ri ), i = 1, 2 to be isomorphic, if there exists a measure space isomorphism T : (X1 , μ1 ) ∼ = (X2 , μ2 ) so that T × T : (R1 , μ˜ 1 ) → (R2 , μ˜ 2 ) is an isomorphism. In other words, after a restriction to conull sets, T satisﬁes (x, y) ∈ R1

⇐⇒

(T (x), T (y)) ∈ R2 .

Let us also adapt the notions of the full group and the full pseudogroup to the measure-theoretic setting, by passing to a quotient Aut (X , X ) → Aut (X , X , μ) where two Borel isomorphisms φ and φ that agree μ-a.e. are identiﬁed. This allows us to focus on the essential measure-theoretic issues. The following easy but useful Lemma illustrates the advantage of this framework. 8. Or just Zariski dense subgroup; see [18].

330 / a. furman

lemma 4.2. Let (X , X , μ, R) be a II1 -relation. Then for A, B ∈ X one has μ(φ(A)B) = 0 for some φ ∈ [R] iff μ(A) = μ(B). 4.1.1. restriction and weak isomorphisms Equivalence relations admit a natural operation of restriction, sometimes called induction, to a subset: given a relation R on X and a measurable subset A ⊂ X the restriction RA to A is RA = R ∩ (A × A). In the presence of, say, R-invariant, measure μ on (X , X ) the restriction to a subset A ⊂ X with μ(A) > 0 preserves the restricted measure μ|A , deﬁned by μ|A (E) = μ(A ∩ E). If μ is a probability measure, we shall denote by μA the normalized restriction μA = μ(A)−1 · μ|A . It is easy to see that ergodicity is preserved, so a restriction of a II1 -relation (X , μ, R) to a positive measure subset A ⊂ X is a II1 -relation (A, μA , RA ).

remark 4.3. Note that it follows from Lemma 4.2 that the isomorphism class of RA depends only on R and on the size μ(A), so RA may be denoted R t where t = μ(A) is 0 < t ≤ 1. One may also deﬁne R t for t > 1. For an integer k > 1 let R k denote the product of R with the full relation on the ﬁnite set {1, . . . , k}, namely the relation on X × {1, . . . , k} with ((x, i), (y, j)) ∈ R k iff (x, y) ∈ R. So (R k )1/k ∼ = R1 ∼ = R. The deﬁnition of R t can now be extended to all 0 < t < ∞ using an easily veriﬁed formula (R t )s ∼ = R ts . This construction is closely related to the notion of an ampliﬁcation in von Neumann algebras: the Murray von Neumann group-measure space construction MR satisﬁes MR t = (MR )t . The operation of restriction/induction allows one to relax the notion of isomorphism of II1 -relations as follows:

definition 4.4. Two II1 -relations R1 and R2 are weakly isomorphic if × R1 ∼ . Equivalently, there exist positive measurable = R2t for some t ∈ R+ subsets Ai ⊂ Xi with μ2 (A2 ) = t · μ1 (A1 ) and an isomorphism between the restrictions of Ri to Ai . Observe that two ergodic probability measure-preserving actions i (Xi , Xi , μi ) of countable groups are orbit equivalent iff the corresponding orbit relations Ri Xi are isomorphic.

a survey of measured group theory / 331

4.2. Invariants of Equivalence Relations

Let us now discuss in some detail several qualitative and numerical properties of II1 equivalence relations that are preserved under isomorphisms and often preserved or rescaled by the index under weak isomorphisms. We refer to such properties as invariants of equivalence relations. Many of these properties are motivated by properties of groups, and often an orbit relation R X of an essentially free action of a countable group would be a reﬂection of the corresponding property of .

4.2.1. amenability, strong ergodicity, and property (t) Amenability of an equivalence relation can be deﬁned in a number of ways. In [138] Zimmer introduced the notion of amenability for a group action on a space with quasiinvariant measure. This notion plays a central role in the theory. This deﬁnition is parallel to the ﬁxed-point characterization of amenability for groups. For equivalence relation R on (X , X ) with a quasi-invariant measure μ it reads as follows. Let E be a separable Banach space, and c : R → Isom (E) be a measurable 1-cocycle, that is, a measurable (with respect to the weak topology on E) map, satisfying μ-a.e.: ˜ c(x, z) = c(x, y) ◦ c(y, z). Let X x → Qx ⊂ E ∗ be a measurable family of nonempty convex compact subsets of the dual space E ∗ taken with the ∗-topology, so that c(x, y)∗ (Qx ) = Qy . The relation R is amenable if any such family contains a measurable invariant section, that is, a measurable assignment X x → p(x) ∈ Qx , so that a.e.: c(x, y)∗ p(x) = p(y). The (original) deﬁnition of amenability for group actions concerned general cocycles c : G × X → Isom (E) rather than the ones depending only on the orbit relation R X . The language of measured groupoids provides a common framework for both settings (see [9]). Any nonsingular action of an amenable group is amenable, because any cocycle c : × X → Isom (E) can be used to deﬁne an afﬁne -action on the closed convex subset of L∞ (X , E ∗ ) = L1 (X , E)∗ consisting of all measurable sections x → p(x) ∈ Qx ; the ﬁxed-point property of provides the desired c ∗ invariant section. The converse is not true: any (countable or lcsc) group admits essentially free amenable action with a quasi-invariant measure—this is the main use of the notion of amenable actions. However, for essentially free,

332 / a. furman

probability measure-preserving actions, amenability of the II1 -relation R X implies (hence is equivalent to) amenability of . Indeed, given an afﬁne action α on a convex compact Q ⊂ E ∗ , one can take Qx = Q for all x ∈ X and set c(gx, x) = α(g); amenability of R X provides an invariant section p : X → Q whose barycenter q = X p(x) dμ(x) would be an α()-ﬁxed point in Q . Connes, Feldman, and Weiss [27] proved that amenable relations are hyperﬁnite in the sense that they can be viewed as an increasing union of ﬁnite subrelations; they also showed that such a relation can be generated by an action of Z (see also [78] by Kaimanovich for a nice exposition and several other nice characterizations of amenability). It follows that there is only one amenable II1 -relation, which we denote hereafter by Ramen . In [141] Zimmer introduced the notion of property (T) for group actions on measure spaces with quasi-invariant measure. The equivalence relation version can be stated as follows. Let H be a separable Hilbert space and let c : R → U(H ) be a measurable 1-cocycle, that is, c satisﬁes c(x, z) = c(x, y) ◦ c(y, z). Then R has property (T) if any such cocycle for which there exists a sequence vn : X → S(H ) of measurable maps into the unit sphere S(H ) with vn (y) − c(x, y)vn → 0

[μ]-a.e. ˜

admits a measurable map u : X → S(H ) with u(y) = c(x, y)u(x) for μ-a.e. ˜ (x, y) ∈ R. For an essentially free probability measure-preserving action (X , μ) the orbit relation R X has property (T) if and only if the group has Kazhdan’s property (T) (in [141] weak mixing of the action was assumed for the “only if ” implication, but this can be removed as in §3.1.1 relying on Bekka - Valette [16]). In [8] Anantharaman-Delaroche studied the notion of property (T) in the context of general measured groupoids. Let R be a II1 -equivalence relation on (X , μ). A sequence {An } of measurable subsets of X is asymptotically R-invariant, if μ(φ(An )An ) → 0 for every φ ∈ [R]. This is satisﬁed trivially if μ(An ) · (1 − μ(An )) → 0. Relation R is strongly ergodic if any asymptotically R-invariant sequence of sets is trivial in the above sense. (Note that the condition of asymptotic invariance may be checked on elements φi of any generating system of R.) The amenable relation Ramen is not strongly ergodic. If an action (X , μ) has a spectral gap (i.e., does not have almost-invariant vectors) in the Koopman

a survey of measured group theory / 333

representation on L2 (X , μ) ! C, then R X is strongly ergodic. Using the fact that the Koopman representation of a Bernoulli action (X0 , μ0 ) is contained in a multiple of the regular representation ∞ · 2 (), Schmidt [128] characterized nonamenable groups by the property that they admit p.m.p. actions with strongly ergodic orbit relation. If R is not strongly ergodic, then it has an amenable relation as a nonsingular quotient (Jones and Schmidt [77]). Connes and Weiss [28] showed that all p.m.p. actions of a group have strongly ergodic orbit relations if and only if has Kazhdan’s property (T). In this short elegant paper they introduced the idea of Gaussian actions as a way of constructing a p.m.p. action from a given unitary representation. In general, strong ergodicity of the orbit relation R X does not imply a spectral gap for the action (X , μ) ([128], [70]). However, this implication does hold for generalized Bernoulli actions (Kechris and Tsankov [82]), and when the action has an ergodic centralizer (Chifan and Ioana [25, lemma 10]).

4.2.2. fundamental group-index values of self-similarity The term × “fundamental group” of a II1 -relation R refers to a subgroup of R+ deﬁned by × : R∼ F (R) = t ∈ R+ = Rt . Equivalently, for R on (X , μ), the fundamental group F (R) consists of all ratios μ(A)/μ(B) where A, B ⊂ X are positive measure subsets with RA ∼ = RB (here one can take one of the sets to be X without loss of generality). The notion is borrowed from a similarly deﬁned concept of the fundamental group of a von Neumann algebra, introduced by Murray and von Neumann × [104]: F (M) = t ∈ R+ : Mt ∼ = M . However, the connection is not direct: even for group space construction M = L∞ (X ) isomorphisms M ∼ = Mt (or even automorphisms of M) need not respect the Cartan subalgebra L∞ (X ) in general. Since the restriction of the amenable relation Ramen to any positive measure subset A ⊂ X is amenable, it follows × . F (Ramen ) = R+

The same obviously applies to the product of any relation with an amenable one. On another extreme are orbit relations R X of essentially free ergodic action of ICC groups with property (T): for such relations the fundamental group F (R X ) is at most countable (Gefter and Golodets [64, Corollary 1.8]).

334 / a. furman

Many relations have a trivial fundamental group. This includes all II1 relations with a nontrivial numeric invariant that scales under restriction: 1) Relations with 1 < cost (R) < ∞; in particular, orbit relation R X for essentially free actions of Fn , 1 < n < ∞, or surface groups. (2) 2) Relations with some nontrivial 2 -Betti number 0 < βi (R) < ∞ for some i ∈ N; in particular, orbit relation R X for essentially free actions (2) of a group with 0 < βi () < ∞ for some i ∈ N, such as lattices in SO2n,1 (R), SUm,1 (R), and Spk,1 (R). Triviality of the fundamental group often appears as a by-product of rigidity of groups and group actions. For example, F (R X ) = {1} in the following situations: 1) Any (essentially free) action of a lattice in a simple Lie group of higher rank ([64]); 2) Any essentially free action of (ﬁnite-index subgroups of products) mapping class groups ([86]); 3) Actions of = 1 × · · · × n , n ≥ 2, of hyperboliclike groups i where each of them acts ergodically ([101]); and 4) Gdsc -cocycle superrigid actions X such as Bernoulli actions of groups with property (T) ([111, 112, 117]). What are other possibilities for the fundamental group beyond the two × and F (R) = {1}? The most comprehensive extreme cases F (R) = R+ answer (to date) to this question is contained in the following result of S. Popa and S. Vaes (see [118] for further references):

theorem 4.5. (popa-vaes, [118, thm 1.1]) There exists a family S of additive subgroups of R that contains all countable groups, and (uncountable) groups of arbitrary Hausdorff dimension in (0, 1), so that for any F ∈ S and any totally disconnected locally compact unimodular group G there exists uncountably many mutually non-SOE essentially free p.m.p. actions of F∞ whose orbit relations R = RF∞ X have F (R) ∼ = G. = exp (F) and Out (R) ∼ Moreover, in these examples the Murray von Neumann group space factor M = L∞ (X ) has F (M) ∼ = Out (R) = exp (F) and Out (M) ∼ = F (R) ∼ H 1 (R, T), where H 1 (R, T) is the ﬁrst cohomology with coefﬁcients in the 1-torus. 4.2.3. treeability An equivalence relation R is said treeable (Adams [4]) if it admits a generating set = {φi } so that the corresponding (nonoriented)

a survey of measured group theory / 335

graph on a.e. R-class is a tree. Basic examples of treeable relations include: Ramen viewing the amenable II1 -relation as the orbit relation of some/any action of Z = F1 , and more generally, RFn X where Fn X is an essentially free action of the free group Fn , 1 ≤ n ≤ ∞. Any restriction of a treeable relation is treeable, and R is treeable iff R t is. If R1 → R2 is a (weak) injective relation morphism and R2 is treeable, then so is R1 —the idea is to lift a treeing graphing from R2 to R1 piece by piece. This way, one shows that if a group admits an essentially free action Z with treeable RZ , and and admit (S)OE essentially free actions X and Y , then the -action on X × Z, g : (x, z) → (gx, α(g, x)z) via the (S)OE cocycle α : × X → has a treeable orbit structure R X ×Z . Since surface groups = π1 ( g ), g ≥ 2, and F2 are lattices in PSL2 (R), hence ME, the former groups have free actions with treeable orbit relations. Are all orbit relations of free actions of a surface group treeable?

4.2.4. cost The notion of cost for II1 -relations corresponds to the notion of rank for discrete countable groups. The notion of cost was introduced by G. Levitt [90] and extensively studied by D. Gaboriau [53, 54, 60]. definition 4.6. Given a generating system = {φi : Ai → Bi }i∈N for a II1 -equivalence relation R on (X , μ) the cost of the graphing is cost ( ) =

μ(Ai ) =

i

μ(Bi )

i

and the cost of the relation is cost (R) = inf cost ( ) : generates R . A generating system deﬁnes a graph structure on every R-class and cost ( ) is half of the average valency of this graph over the space (X , μ). The cost of a II1 -relation takes values in [1, ∞]. In the deﬁnition of the cost of a relation it is important that the relation is probability measure preserving, but ergodicity is not essential. The broader context includes relations with ﬁnite classes; such relations can have values less than 1. For instance, from the orbit relation of a (nonergodic) probability measure-preserving action of a ﬁnite group (X , μ) one gets cost (R X ) = 1 −

1 . ||

336 / a. furman

If R is the orbit relation of some (not necessarily free) action (X , μ), then cost (R) ≤ rank(), where the latter stands for the minimal number of generators for . Indeed, any generating set {g1 , . . . , gk } for gives a generating system = {γi : X → X }ki=1 for R X . Recall that the amenable II1 -relation Ramen can be generated by (any) action of Z. Hence cost (Ramen ) = 1. The cost behaves nicely with respect to restriction:

theorem 4.7. (gaboriau [54]) For a II1 -relation R: t · ( cost (R t ) − 1) = cost (R) − 1

× (t ∈ R+ ).

The following is a key tool for computations of the cost:

theorem 4.8. (gaboriau [54]) Let R be a treeable equivalence relation, and be a graphing of R giving a tree structure to R-classes. Then cost (R) = cost ( ). Conversely, for a relation R with cost (R) < ∞, if the cost is attained by some graphing , then is a treeing of R. The above result (the ﬁrst part) implies that for any essentially free action Fn (X , μ) one has cost (RFn X ) = n. This allowed Gaboriau to prove the following fact, answering a long-standing question:

corollary 4.9. (gaboriau [53], [54]) If essentially free probability measurepreserving actions of Fn and Fm are orbit equivalent, then n = m. Note that Fn and Fm are commensurable for 2 ≤ n, m < ∞, hence they have essentially free actions that are weakly isomorphic. The index of such n−1 weak isomorphism will necessarily be m−1 , or m−1 n−1 (these free groups have P.F.-ﬁxed price). It should be pointed out that one of the major open problems in the theory of von Neumann algebras is whether it is possible for the factors L(Fn ) and L(Fm ) to be isomorphic for n = m (it is known that either all L(Fn ), 2 ≤ n < ∞, are isomorphic, or all distinct). The following powerful result of Greg Hjorth provides a link from treeable relations back to actions of free groups:

a survey of measured group theory / 337

theorem 4.10. (hjorth [69]) Let R be a treeable equivalence relation with n = cost (R) in {1, 2, . . . , ∞}. Then R can be generated by an essentially free action of Fn . The point of the proof is to show that a relation R that has a treeing graphing with average valency 2n admits a (treeing) graphing with a.e. constant valency 2n. The behavior of the cost under passing to a subrelation of ﬁnite index is quite subtle—the following question is still open (to the best of the author’s knowledge).

question 4.11. (gaboriau) Let be a subgroup of ﬁnite index in , and (X , μ) be an essentially free p.m.p. action. Is it true that the costs of the orbit relations of and are related by the index [ : ] cost (R X ) − 1 = [ : ] · ( cost (R X ) − 1)? In general has at most [ : ]-many ergodic components. The extreme case where the number of -ergodic components is maximal: [ : ] corresponds to (X , μ) being a co-induction from an ergodic -action. In this case the above formula easily holds. The real question lies in the other extreme where is ergodic. Recall that the notion of the cost is analogous to the notion of rank for groups, where rank () = inf n ∈ N : ∃ epimorphism Fn → . Schreier’s theorem states that for n ∈ N any subgroup F < Fn of ﬁnite-index [Fn : F] = k is itself free: F ∼ = Fk(n−1)+1 . This implies that for any ﬁnitely generated and any ﬁnite-index subgroup of < one has rank ( ) − 1 ≤ [ : ] · ( rank () − 1) with equality in the case of free groups. Let > 1 > . . . be a chain of subgroups of ﬁnite index. One deﬁnes the rank gradient (Lackenby [89]) of the chain {n } as the limit of the monotonic (!) sequence: RG (, {n }) = lim

n→∞

rank (n ) − 1 . [ : n ]

It is an intriguing question whether (or when) is it true that RG (, {n }) depends only on and not on a particular chain of ﬁnite-index subgroups. One should, of course, assume that the chains in question have trivial intersection, and one might require the chains to consist of normal subgroups in the original group. In the case of free groups RG is indeed independent of the chain.

338 / a. furman

In [1] Abert and Nikolov prove that the rank gradient of a chain of ﬁniteindex subgroups of is given by the cost of a certain associated ergodic p.m.p. -action. Let us formulate a special case of this relation where the chain {n } consists of normal subgroups n with n = {1}. Let K = lim / n ←− denote the proﬁnite completion corresponding to the chain. The -action by left translations on the compact totally disconnected group K preserves the Haar measure mK and (K , mK ) is a free ergodic p.m.p. action. (Let us point out in passing that this action has a spectral gap, implying strong ergodicity, iff the chain has property (τ ) introduced by Lubotzky and Zimmer [91]).

theorem 4.12. (abert-nikolov [1]) With the above notations: RG (, {n }) = cost (R K ) − 1. One direction (≥) is easy to explain. Let Kn be the closure of n in K . Then Kn is an open normal subgroup of K of index m = [ : n ]. Let 1 = g1 , g2 , . . . , gn ∈ be representatives of n -cosets, and h1 , . . . , hk generators of n with k = rank (n ). Consider the graphing = {φ2 , . . . , φm , ψ1 , . . . , ψk }, where φi : Kn → gi Kn are restrictions of gi (2 ≤ i ≤ m), and ψj : Kn → Kn are restrictions of hj (1 ≤ j ≤ k). These maps are easily seen to generate R K , with the cost of cost ( ) = k · mK (Kn ) + (m − 1) · mK (Kn ) =

k−1 rank (n ) − 1 +1 = 1+ . m [ : n ]

Abert and Nikolov observed that a positive answer to Question 4.11 combined with the above result shows that RG () is independent of the choice of a (normal) chain, and therefore is a numeric invariant associated to any residually ﬁnite ﬁnitely generated group. Surprisingly, this turns out to be related to a problem in the theory of compact hyperbolic 3-manifolds concerning rank versus Heegard genus [89]—see [1] for the connection and further discussions. The above result has an application, independent of Question 4.11. Since amenable groups have P.F. with C = 1, it follows that a ﬁnitely generated, residually ﬁnite amenable group has sublinear rank growth for ﬁnite-index normal subgroups with trivial intersection, that is, RG () = 0 for any such chain.

4.2.5. 2 -betti numbers I have already mentioned the 2 -Betti numbers (2) βi () associated with a discrete group and Gaboriau’s proportionality result

a survey of measured group theory / 339

in Theorem 3.8 for measure equivalence between groups. In fact, rather than relating the 2 -Betti numbers of groups via ME, in [57] Gaboriau (2)

- deﬁnes the notion of 2 -Betti numbers βi (R) for a II1 -equivalence relation R; (2) (2) - proves that βi () = βi (R X ) for essentially free ergodic action (X , μ); and (2) (2) - observes that βi (R t ) = t · βi (R) for any II1 -relation. (2)

(2)

The deﬁnition of βi (R) is inspired by the deﬁnition of βi () by Cheeger and Gromov [23]: it uses R-action (or groupoid action) on pointed contractible simplicial complexes, corresponding complexes of Hilbert modules with Raction, and von Neumann dimension with respect to the algebra MR . In the late 1990s Wolgang Lück developed an algebraic notion of dimension for arbitrary modules over von Neumann algebras, in particular giving an alternative approach to 2 -Betti numbers for groups (see [92]). In [124] Roman Sauer used Lück’s notion of dimension to deﬁne 2 -Betti numbers of equivalence relations, and more general measured groupoids, providing an alternative approach to Gaboriau’s results. In [127] Sauer and Thom develop further homological tools (including a spectral sequence associated to strongly normal subrelations) to study 2 -Betti numbers for groups, relations, and measured groupoids.

4.2.6. outer automorphism group Given an equivalence relation R on (X , μ) deﬁne the corresponding automorphism group as the group of selfisomorphisms: Aut (R) = T ∈ Aut (X , μ) : T × T (R) = R (modulo null sets) . The subgroup Inn (R) of inner automorphisms is Inn (R) = T ∈ Aut (X , μ) : (x, T (x)) ∈ R for a.e. x ∈ X . This is just the full group [R], but the above notation emphasizes the fact that it is normal in Aut (R) and suggests to consider the outer automorphism group Out (R) = Aut (R)/ Inn (R). One might think of Out (R) as the group of all measurable permutations of the R-classes on X . Recall (Lemma 4.2) that Inn (R) is a huge group as it acts transitively on (classes mod null sets of) measurable subsets of any given size

340 / a. furman

in X . Yet the quotient Out (R) might be small (even ﬁnite or trivial), and can sometimes be explicitly computed.

remark 4.13. As an abstract group H = Inn (R) is simple, and its automorphisms come from automorphisms of R; in particular Out (H) = Out (R). Moreover, Dye’s reconstruction theorem states that (the isomorphism type of) R is determined by the structure of Inn (R) as an abstract group (see [80, §I.4] for proofs and further facts). Let us also note that the operation of restriction/ampliﬁcation of the relation does not alter the outer automorphism group (cf. [47, lemma 2.2]): Out (R t ) ∼ = Out (R)

× (t ∈ R+ ).

The group Aut (R) has a natural structure of a Polish group [62, 64]. First, recall that if (Y , ν) is a ﬁnite- or inﬁnite-measure Lebesgue space, then Aut (Y , ν) is a Polish group with respect to the weak topology induced from the weak (=strong) operator topology of the unitary group of L2 (Y , ν). This deﬁnes a Polish topology on Aut (R) when the latter is viewed as acting on the inﬁnite-measure space (R, μ). ˜ 9 However, Inn (R) is not always closed in Aut (R), so the topology on Out (R) might be complicated. Alexander Kechris recently found the following surprising connection:

theorem 4.14. (kechris [80, theorem 8.1]) If Out (R)fails to be a Polish group, then cost (R) = 1. Now assume that R can be presented as the orbit relation of an essentially free action (X , μ), so Aut (R) is the group of self-orbit equivalences of X . The centralizer Aut (X , μ) of in Aut (X , μ) embeds in Aut (R) and if is ICC (i.e., has inﬁnite conjugacy classes), then the quotient map out

Aut (R)−→ Out (R) is injective on Aut (X , μ) (cf. [62, lemma 2.6]). So Out (R) has a copy of Aut (X , μ), and the latter might be very big. For example, in the Bernoulli action (X , μ) = (X0 , μ0 ) , it contains Aut (X0 , μ0 ) acting diagonally on the factors. Yet, if has property (T), then Aut (X , μ) · Inn (R X ) is open in the Polish group Aut (R X ). In this case the image of Aut (X , μ) has ﬁnite or countable index in Out (R X ). This fact was observed by Gefter and Golodets in [64, §2], and can be deduced from Proposition 5.14. 9. This topology coincides with the restriction to Aut (R) of the uniform topology on Aut (X , μ) given by the metric d(T , S) = μ x ∈ X : T (x) = S(x) . On all of Aut (X , μ) the uniform topology is complete but not separable; but its restriction to Aut (R) is separable.

a survey of measured group theory / 341

To get a handle on Out (R X ) one looks at OE-cocycles cT : × X → corresponding to elements T ∈ Aut (R X ). It is not difﬁcult to see that cT is conjugate in to the identity (i.e., cT (g, x) = f (gx)−1 gf (x) for f : X → ) iff T is in Aut (X , μ) · Inn (R). Thus, starting from a group or action X with strong rigidity properties for cocycles, one controls Out (R X ) via Aut (X , μ). This general scheme (somewhat implicitly) is behind the ﬁrst example of an equivalence relation with trivial Out (R) constructed by Gefter [61, 62]. Here is a variant of this construction:

theorem 4.15. Let be a torsion-free group with property (T), K a compact connected Lie group without outer automorphisms, and τ : → K a dense embedding. Let L < K be a closed subgroup and consider the ergodic actions (K , mK ) and (K /L, mK /L ) by left translations. Then Out (R K ) ∼ = K,

Out (R K /L ) ∼ = NK (L)/L.

∼ POn−1 (R) < K to be the stabilizer In particular, taking K = POn (R) and L = n of a line in R , the space K /L is the projective space P n−1 , and we get Out (R P n−1 ) = {1} for any property (T) dense subgroup < POn (R). Such a group exists iff n ≥ 5, Zimmer [145, theorem 7]. The preceding discussion, combined with the cocycles superrigidity Theorem 5.21 below, and an easy observation that Aut (K /L, mK /L ) is naturally isomorphic to NK (L)/L, provide a self contained sketch of the proof of the theorem. In the above statement Out (K ) is assumed to be trivial and to be torsion free just to simplify the statement. However, the assumption that K is connected is essential. Indeed, the dense embedding of = PSLn (Z) in the compact proﬁnite group K = PSLn (Zp ) where p is a prime, gives Out (RPSLn (Z)PSLn (Zp ) ) ∼ = PSLn (Qp ) Z/2Z where the Z/2-extension is given by the transpose g → g tr . The inclusion ⊃ was found in [62], and the equality is proved in [47, theorem 1.6], where many other computations of Out (R X ) are carried out for actions of lattices in higher-rank Lie groups. Finally, we recall that the recent preprint [117] of Popa and Vaes quoted above (Theorem 4.5) shows that an arbitrary totally disconnected lcsc group G can arise as Out (R X ) for an essentially free action of a free group F∞ .

342 / a. furman

4.2.7. cohomology Equivalence relations have groups of cohomology associated to them similar to cohomology of groups. These were introduced by Singer [134] and largely emphasized by Feldman and Moore [40]. Given, say, a type II1 equivalence relation R on (X , μ) consider R (n) = (x0 , . . . , xn ) ∈ X n+1 : (xi , xi+1 ) ∈ R equipped with the inﬁnite Lebesgue measure μ˜ (n) deﬁned by (n) μ˜ (A) = # (x1 , . . . , xn ) : (x0 , . . . , xn ) ∈ R (n) dμ(x0 ). X

(R (0) , μ˜ 0 )

to be (X , μ). Note that (R (1) , μ(1) ) is just (R, μ) ˜ from §4.1. Take Since μ is assumed to be R-invariant, the above formula is invariant under permutations of x0 , . . . , xn . Fix a Polish abelian group A written multiplicatively (usually A = T). The space C n (R, A) of n-cochains consists of equivalence classes (modulo μ˜ (n) -null sets) of measurable maps R (n) → A, linked by the operators dn : C n (R, A) → C n+1 (R, A) dn ( f )(x0 , . . . , xn+1 ) =

n+1

i

f (x0 , . . . , xˆ i , . . . , x0 )(−1) .

i=0

= Ker(dn ) the n-cocycles, and Bn (R) = Im(dn−1 ) the nCall coboundaries; the cohomology groups are deﬁned by H n (R) = Z n (R)/Bn (R). In degree n = 1 the 1-cocycles are measurable maps c : (R, μ) → A such that Z n (R)

c(x, y)c(y, x) = c(x, z) and 1-coboundaries have the form b(x, y) = f (x)/f (y) for some measurable f : X → A. If A is a compact abelian group, such as T, then C 1 (R, A) is a Polish group (with respect to convergence in measure). Being a closed subgroup in C 1 (R, A), the 1-cocycles Z 1 (R, A) form a Polish group. Schmidt [129] showed that B1 (R, A) is closed in Z 1 (R, A) iff R is strongly ergodic. There are only few cases where H 1 (R, T) were computed: C.C. Moore [102] constructed a relation with trivial H 1 (R, T). Gefter [63] considered H 1 (R G , T) for actions of property (T) group densely embedded in a semisimple Lie group G. More recently Popa and Sasyk [116] studied H 1 (R X , T) for property (T) groups with Bernoulli actions (X , μ) = (X0 , μ0 ) . In both cases H 1 (R X , T) is shown to coincide with the group of characters Hom(, T). Higher cohomology groups remain mysterious.

a survey of measured group theory / 343

The fact that A is abelian is essential to the deﬁnition of H n (R, A) for n > 1. However, in degree n = 1 one can deﬁne H 1 (R, ) as a set for a general target group 10 . In fact, this notion is commonly used in this theory under the name of measurable cocycles (see Appendix A and §5.1 below). For the deﬁnition in terms of equivalence relations let Z 1 (R, ) denote the set of all measurable maps (mod μ-null ˜ sets) c : (R, μ) ˜ →

s.t.

c(x, z) = c(x, y)c(y, z)

and let H 1 (R, ) = Z 1 (R, )/ ∼ where the equivalence ∼ between c, c ∈ Z 1 (R, ) is declared if c(x, y) = f (x)−1 c (x, y)f (y) for some measurable f : (X , μ) → . If R = R X is the orbit relation of an essentially free action, then Z 1 (R X , ) coincides with the set of measurable cocycles α : × X → by α(g, x) = c(x, gx). Note that Hom(, )/ maps into H 1 (R, ), via cπ (x, y) = π(g) for the unique g ∈ with x = gy. The point of cocycles superrigidity theorems is to show that under favorable conditions this map is surjective.

4.2.8. free decompositions Group theoretic notions such as free products, amalgamated products, and HNN-extensions can be deﬁned in the context of equivalence relations—see Gaboriau [54, section IV]. For example, a II1 relation R is said to split as a free product of subrelations {Ri }i∈I , denoted R = ∗i∈I Ri , if 1) R is generated by {Ri }i∈I , that is, R is the smallest equivalence relation containing the latter family; and 2) almost every chain x = x0 , . . . , xn = y, where xj−1 = xj , (xj−1 , xj ) ∈ Ri(j) and i(j + 1) = i(j), has x = y. If S is yet another subrelation, one says that R splits as a free product of Ri amalgamated over S , R = ∗S Ri , if in condition (2) one replaces xj−1 = xj by (xj−1 , xj ) ∈ S . The obvious example of the situation above is an essentially free action of a free product of groups 3 = 1 ∗ 2 (respectively amalgamated product 5 = 1 ∗4 2 ) on a probability space (X , μ); in this case the orbit relations Ri = Ri X satisfy R3 = R1 ∗ R2 (respectively R5 = R1 ∗R4 R2 ). Another useful construction (see Ioana, Peterson, Popa [75]) is as follows. Given measure preserving (possibly ergodic) relations R1 , R2 on a probability 10. In order to deﬁne the notion of measurability should have a Borel structure, and better be a Polish group; often it is a discrete countable group, or a Lie group.

344 / a. furman

space (X , μ) for T ∈ Aut (X , μ) consider the relation generated by R1 and T (R2 ). It can be shown that for a residual set of T ∈ Aut (X , μ) the resulting relation is a free product of R1 and T (R2 ). A similar construction can be carried out for amalgamated products. Note that in contrast with the category of groups the isomorphism type of the free product is not determined by the free factors alone. Ioana, Peterson, and Popa [75] obtained strong rigidity results for free and amalgamated products of ergodic measured equivalence relations and II1 -factors with certain rigidity properties. Here let us describe some results obtained by Alvarez and Gaboriau [7], which are easier to state; they may be viewed as an analogue of Bass-Serre theory in the context of equivalence relations. Say that a II1 -relation R is freely indecomposable (FI) if R is not a free product of its subrelations. A group is said to be measurably freely indecomposable (MFI) if all its essentially free action give freely indecomposable orbit relations. A group may fail to be MFI even if it is freely indecomposable in the group theoretic sense (surface groups provide an example). Not surprisingly, groups with property (T) are MFI (cf. Adams Spatzier [6]); but more generally (2)

theorem 4.16. (alvarez-gaboriau [7]) If is nonamenable and β1 () = 0, then is MFI. theorem 4.17. (alvarez-gaboriau [7]) Let I, J be two ﬁnite or countableindex sets, {i }i∈I and {j }j∈J be two families of MFI groups, = ∗i∈I i , = ∗j∈J j , and (X , μ), (Y , ν) be essentially free p.m.p. actions SOE

where each i (X , μ) and j (Y , ν) are ergodic. Assume that X ∼ Y. SOE Then |I| = |J| and there is a bijection θ : I → J so that i X ∼ θ(i) Y .

The assumption that each free factor is ergodic is important here; Alvarez and Gaboriau also give an analysis of the general situation (where this assumption is dropped). 4.3. Rigidity of Equivalence Relations

The close relation between ME and SOE allows us to deduce that certain orbit relations R X remember the acting group and the action (X , μ) up to isomorphism, or up to a virtual isomorphism. This slightly technical concept is described in the following:

a survey of measured group theory / 345

lemma 4.18. Suppose an ergodic ME-coupling (, m) of with corresponds SOE to an SOE between ergodic actions T : (X , μ) ∼ (Y , ν). Then the following are equivalent: 1) There exist short exact sequences 1 → 0 → → 1 → 1,

1 → 0 → → 1 → 1

where 0 and 0 are ﬁnite, a discrete (1 , 1 )-coupling (1 , m1 ), and an equivariant map : (, m) → (1 , m1 ); and 2)There exist isomorphism between ﬁnite-index subgroups 1 > 2 ∼ = 2 < 1 , so that 1 X1 = X / 0 and 1 Y1 = Y /0 are induced from some isomorphic ergodic actions 2 X2 ∼ = 2 Y2 . 3)The SOE (or ME) cocycle × X → is conjugate in to a cocycle whose restriction to some ﬁnite-index subgroup 1 is a homomorphism 1 → (the image is necessarily of ﬁnite index). Let us now state two general forms of relation rigidity. Here is one form

theorem 4.19. Let (X , μ) be an ergodic essentially free action of one of the types below, an arbitrary group, and (Y , ν) as essentially free p.m.p. action whose orbit relation RY is weakly isomorphic to R X . Then is commensurable up to ﬁnite kernels to and the actions X and Y are virtually isomorphic; in particular, the SEO-index is necessarily rational. The list of actions X with this SOE-rigidity property includes: 1) is a lattice in a connected, center-free, simple Lie group G of higher rank, and X has no equivariant quotients of the form G/ where < G is a lattice ([45, theorem A]); 2) = 1 × · · · × n where n ≥ 2, i ∈ Creg , and i (X , μ) are ergodic; in addition assume that (Y , ν) is mildly mixing (Monod-Shalom [101]); and 3) is a ﬁnite-index subgroup in a (product of) mapping class groups as in Theorem 3.20 (Kida [85]). a) For a concrete example for (1)–(3) one might take Bernoulli actions (X0 , μ0 ) . In (1) one might also take SLn (Z) Tn or SLn (Z) SLn (Zp ) with n ≥ 3. In (2) one might look at Fn × Fm acting on a compact Lie group K , for

346 / a. furman

example SO3 (R), by (g, h) : k → gkh −1 where Fn , Fm are embedded densely in K . b) In (1) the assumption that there are no -equivariant quotient maps X → G/ is necessary, since given such a quotient there is a -action on SOE

some (X , μ ) with X ∼ X . The rigidity statement in this case is that this is a complete list of groups and their essentially free actions up to virtual isomorphism ([45, theorem C]). The appearance of these factors has to do with (G, mG ) appearing as a quotient of a (, )-coupling. c) The basic technique for establishing the stated rigidity in cases (1)–(3) is to establish condition (1) in Lemma 4.18. This is done by analyzing a self-ˇ (where X = / and Y = / ) and invoking coupling of the form × an analogue of the construction in §5.5. d) In all cases one can sharpen the results (eliminate the “virtual”) by imposing some benign additional assumptions: rule out torsion in the acting groups, and impose ergodicity for actions of ﬁnite-index subgroups. The second stronger form of relation rigidity refers to rigidity of relation morphisms that are obtained from Gdsc -cocycle superrigid actions discovered by Sorin Popa (see §5.4). We illustrate this framework by the following particular statement (see [113, theorem 0.4] and [48, theorem 1.8]).

theorem 4.20. Let (X , μ) be a mixing Gdsc -cocycle superrigid action, such as: 1) A Bernoulli -action on (X0 , μ0 ) , where has property (T), or = 1 × 2 with 1 nonamenable and 2 being inﬁnite; and 2) K /L where → K is a homomorphism with dense image in a simple compact Lie group K with trivial π1 (K ), L < K is a closed subgroup, and has (T). Let be some group with an ergodic essentially free measure-preserving action (Y , ν),11 X ⊂ X a positive measure subset, and T : X → Y a measurable map with T∗ μ ≺ ν and (x1 , x2 ) ∈ R X ∩ (X × X )

=⇒

(T (x1 ), T (x2 )) ∈ RY .

Then there exists ρ

• an exact sequence 0 −→ −→1 with ﬁnite 0 and 1 < ; and • a 1 -ergodic subset Y1 ⊂ Y with 0 < ν(Y1 ) < ∞; and 11. The space (Y , ν) might be ﬁnite- or inﬁnite-measure Lebesgue space.

a survey of measured group theory / 347 • denoting (X1 , μ1 ) = (X , μ)/ 0 and ν1 = ν(Y1 )−1 · ν|Y , there is an isomor1

phism T1 : (X1 , μ1 ) ∼ = (Y1 , ν1 ) of 1 -actions.

Moreover, μ-a.e. T (x) and T1 (0 x) are in the same -orbit.

4.3.1. a question of feldman and moore Feldman and Moore showed [39] that any countable Borel equivalence relation R can be generated by a Borel action of a countable group. They asked whether one can ﬁnd a free action of some group, so that R-classes would be in one-to-one correspondence with the acting group. This question was answered in the negative by Adams [3]. In the context of measured relations, say of type II1 , the question is whether it is possible to generate R (up to null sets) by an essentially free action of some group. This question was also settled in the negative in [45, theorem D], using the following basic constructions: 1) Start with an essentially free action (X , μ) that is rigid as in Theorem 4.19 or 4.20, and let R = (R X )t with an irrational t. 2) Consider a proper embedding G → H of higher-rank simple Lie groups and choose a lattice < H, say G = SL3 (R) ⊂ H = SL4 (R) with = SL4 (Z). Such actions always admit a Borel cross-section X ⊂ H/ for the G-action, equipped with a holonomy-invariant probability measure μ. Take R on (X , μ) to be the relation of being in the same G-orbit. In case (1) one argues as follows: if some group has an essentially free action (Y , ν) with (R X )t = R ∼ = RY , then the rigidity implies that and are commensurable up to ﬁnite kernel, and X is virtually isomorphic to Y . But this would imply that the index t is rational, contrary to the assumption. This strategy can be carried out in other cases of very rigid actions as in [72, 85, 101, 113]. Theorem 5.21 provides an example of this type R = (R K )t where is a Kazhdan group densely embedded in a compact connected Lie group K . So the reader has a sketch of the full proof for a II1 -relation that cannot be generated by an essentially free action of any group. Example of type (2) was introduced by Zimmer in [147], where it was proved that the relation R on such a cross-section cannot be essentially freely generated by a group , which admits a linear representation with an inﬁnite image. The linearity assumption was removed in [45]. This example is particularly interesting since it cannot be “repaired” by restriction/ampliﬁcation; as any R t can be realized as a cross-section of the same G-ﬂow on H/ .

question 4.21. (vershik) Let R on (X , μ) be a II1 -relation that cannot be generated by an essentially free action of a group; and let (X , μ) be some action

348 / a. furman

producing R. One may assume that the action is faithful, that is, → Aut (X , μ) is an embedding. What can be said about and the structure of the measurable family {x }x∈X of the stabilizers of points in X ? In [119] S. Popa and S. Vaes give an example of a II1 -relation R (which is a restriction of the II∞ -relation RSL5 (Z)R5 to a subset A ⊂ R 5 of positive ﬁnite measure), which has property (T) but cannot be generated by an action (not necessarily free) of any group with property (T).

5. Techniques 5.1. Superrigidity in Semisimple Lie Groups

The term “superrigidity” refers to a number of phenomena originated and inspired by the following celebrated discovery of G. A. Margulis.

theorem 5.1. (margulis [93]) Let G and G be (semi)simple connected real center-free Lie groups without compact factors with rk(G) ≥ 2, < G be an irreducible lattice, and π : → G a homomorphism with π() being Zariski dense in G and not precompact. Then π extends to a (rational) epimorphism π¯ : G → G . The actual result is more general than stated, as it applies to products of semisimple algebraic groups over general local ﬁelds. We refer the reader to the comprehensive monograph (Margulis [95]) for the general statements, proofs, and further results and applications, including the famous arithmeticity theorem. The core of (some of the available) proofs of Margulis’s superrigidity theorem is a combination of the theory of algebraic groups and purely ergodictheoretic arguments. The result applies to uniform and nonuniform lattices alike, and it also covers irreducible lattices in higher-rank Lie groups such as SL2 (R) × SL2 (R). Let us also note that the assumption that π() is not precompact in G is redundant if π () is Zariski dense in a real Lie group G (since compact groups over R are algebraic), but is important in general (cf. SLn (Z) < SLn (Qp ) is Zariski dense but precompact). In [139] R. J. Zimmer has obtained a far-reaching generalization of Margulis’s superrigidity, passing from the context of representations of lattices to the framework of measurable cocycles over probability measure-preserving actions (representations of “virtual subgroups” in Mackey’s terminology). The connection can be brieﬂy summarized as follows: given a transitive action G X = G/ and some topological group H, there is a bijection

a survey of measured group theory / 349

between measurable cocycles G × G/ → H modulo cocycle conjugation and homomorphisms → H modulo conjugation in H H 1 (G G/ , H) ∼ = Hom(, H)/H (see §A.1, and [52, 146]). In this correspondence, a representation π : → H extends to a homomorphism G → H iff the corresponding cocycle π ◦ c : G × G/ → H is conjugate to a homomorphism G → H. Zimmer’s cocycle superrigidity theorem states that under appropriate nondegeneracy assumptions a measurable cocycle over an arbitrary p.m.p. ergodic action G (X , μ) is conjugate to a homomorphism.

theorem 5.2. (zimmer [139], see also [146]) Let G, G be a semisimple Lie group as in Theorem 5.1, in particular rkR (G) ≥ 2, let G (X , μ) be an irreducible probability measure-preserving action and c : G × X → G be a measurable cocycle that is Zariski dense and not compact. Then there exist a (rational) epimorphism π : G → G and a measurable f : X → G so that c(g, x) = f (gx)−1 π (g)f (x). In the above statement irreducibility of G (X , μ) means mere ergodicity if G is a simple group, and ergodicity of the action Gi (X , μ) for each factor Gi in the case of a semisimple group G = ni=1 Gi with n ≥ 2 factors. For a lattice < G = Gi in a semisimple group the transitive action G G/ is irreducible precisely iff is an irreducible lattice in G. The notion of being Zariski dense (respectively not compact) for a cocycle c : G × X → H means that c is not conjugate to a cocycle c f taking values in a proper algebraic (respectively compact) subgroup of H. The setting of cocycles over p.m.p. actions adds a great deal of generality to the superrigidity phenomena. The ﬁrst illustration of this is the fact that once cocycle superrigidity is known for actions of G it passes to actions of lattices in G: given an action (X , μ) of a lattice < G one obtains a G-action on X¯ = G × X by acting on the ﬁrst coordinate (just like the composition operation of ME-coupling in §2.1). A cocycle c : × X → H has a natural lift to c¯ : G × X¯ → H and its cohomology is directly related to that of the original cocycle. So cocycle superrigidity theorems have an almost automatic bootstrap from lcsc groups to their lattices. The induced action G X¯ is ergodic iff X is ergodic; however, irreducibility is more subtle. Yet, if (X , μ) is mixing, then G X¯ is mixing and therefore is irreducible. Theorem 3.14 was the ﬁrst application of Zimmer’s cocycle superrigidity 5.2 (Theorem; see [139]). Indeed, if α : × X → is the rearrangement cocycle

350 / a. furman OE

associated to an Orbit Equivalence T : (X , μ) ∼ (X , μ ) where < G, < G are lattices, then, viewing α as taking values in G , Zimmer observes that α is Zariski dense using a form of Borel’s density theorem and deduces that G ∼ = G (here for simplicity the ambient groups are assumed to be simple, connected, center free, and rkR (G) ≥ 2). Moreover, there is a homomorphism π : → G and f : X → G so that α(γ , x) = f (γ x)π (γ )f (x)−1 with π : → π() < G being an isomorphism of lattices.

remark 5.3. At this point it is not clear whether π () should be (conjugate to) , and even assuming π() = whether f takes values in . In fact, the self orbit equivalence of the action on G/ given by g → g −1 gives a rearrangement cocycle c : × G/ → , which is conjugate to the identity → by a unique map f : G/ → G with f∗ (mG/ ) ≺ mG . However, if π() = and f takes values in , it follows that the original actions (X , μ) and (X , μ ) are isomorphic via the identiﬁcation π : ∼ = . We return to this point below. 5.1.1. superrigidity and me-couplings Zimmer’s cocycle superrigidity theorem applied to OE- or ME-cocycles (see SectionsA.2 and A.3) has a natural interpretation in terms of ME-couplings. Let G be a higher-rank simple Lie group (hereafter implicitly, connected, and center free) and, denote by i : G → Aut (G) the adjoint homomorphism (which is an embedding since G is center free). theorem 5.4. ([44, theorem 4.1]) Let G be a higher-rank simple Lie group, 1 , 2 < G lattices, and (, m) an ergodic (1 , 2 )-coupling. Then there exists a unique measurable map : → Aut (G) so that m-a.e. on (γ1 ω) = i(γ1 ) (ω),

(γ2 ω) = (ω)i(γ2 )−1

(γi ∈ i ).

∼ Ad (G) ≤ G ≤ Moreover, ∗ m is either the Haar measure on a group G = Aut (G) or is atomic, in which case 1 and 2 are commensurable. Sketch of the proof. To construct such a , choose a fundamental domain X ⊂ for the 2 -action and look at the ME-cocycle c : 1 × X → 2 < G. Apply Zimmer’s cocycle superrigidity theorem to ﬁnd π : 1 → G and φ : X → G. Viewing G as a subgroup in Aut (G), one may adjust π and φ : X → Aut (G) by some α ∈ Aut (G), so that π is the isomorphism i : 1 → 1 to get c(γ1 , x) = φ(γ1 .x)−1 i(γ1 )φ(x).

a survey of measured group theory / 351

Deﬁne : → Aut (G) by (γ2 x) = φ(x)i(γ2 )−1 and check that it satisﬁes the required relation. To identify the measure ∗ m on Aut (G) one uses Ratner’s theorem, which provides the classiﬁcation of 1 -ergodic ﬁnite measures on ¯ 2 . G/ Theorem 3.16 is then proved using this fact with 1 = 2 plugged into the construction in §5.5 that describes an unknown group essentially as a lattice in G. Note that there are two distinct cases in Theorem 5.4: either ∗ m is atomic, in which case (, m) has a discrete ME-coupling as a quotient, or ∗ m is a Haar measure on a Lie group. The former case leads to a virtual isomorphism between the groups and the actions (this is case (1) in Theorem 4.19); in the ¯ 2 (which is [45, latter, 1 X ∼ = / 2 has a quotient of the form 1 G/ theorem C]). This dichotomy clariﬁes the situation in Remark 5.3 above. 5.2. Superrigidity for Product Groups

Let us now turn to a brief discussion of Monod-Shalom rigidity (see Sections 3.1.7 and 3.2.2). Consider a special case of Margulis-Zimmer superrigidity results where the target group G has rank 1 (say G = PSL2 (R)), while G has higher rank. The conclusion of the superrigidity Theorem 5.1 (respectively Theorem 5.2) is that either a representation (respectively cocycle) is degenerate,12 or there is an epimorphism π : G → G . The latter case occurs if and only if G is semisimple G = Gi , with one of the factors Gi G , and π : G → G pri

factoring through the projection π : G−→Gi G . In this case the given representation of the lattice extends to π (respectively the cocycle is conjugate to the epimorphism π). This special case of Margulis-Zimmer superrigidity, that is, from higherrank G to rank 1 G , was generalized by a number of authors [5, 6, 22] replacing the assumption that the target group G has rank 1 by more geometric notions, such as G = Isom (X ) where X is a proper CAT(-1) space. In the setting considered by Monod and Shalom the target group is “hyperboliclike” in a very general way, while the source group G rather than being a higher-rank semisimple Lie group, is just a product G = G1 × · · · × Gn of n ≥ 2 arbitrary compactly generated (in fact, just lcsc) groups. The philosophy is that the number n ≥ 2 of direct factors provides enough higher-rank properties for such statements.

12. Here precompact, or contained in a parabilic.

352 / a. furman

theorem 5.5. (monod-shalom [100]) Let G = G1 × · · · × Gn be a product of n ≥ 2 lcsc groups, G (X , μ) an irreducible p.m.p. action, H is a hyperboliclike group, and c : G × X → H is a nonelementary measurable cocycle. Then there is a nonelementary closed subgroup H1 < H, a compact normal subgroup K % H1 , a measurable f : X → H, and a homomorphism ρ : Gi → H1 /K from one of the factors Gi of G, so that the conjugate cocycle c f takes values in H1 pri ρ and G × X → H1 → H1 /K is the homomorphism π : G−→Gi −→H1 /K .

This beautiful theorem is proved using the technology of second-bounded cohomology (developed in [20, 21, 97] and applied in this setting in [96, 100]) with the notions of hyperboliclike and nonelementary interpreted in the context of the class Creg . Suppose = 1 × · · · × n , n ≥ 2, is a product of “hyperboliclike” groups. Let (, m) be a self-ME-coupling of . Consider an ME-cocycle × X → , which can be viewed as a combination of n cocycles c

pri

ci : × X −→ −→i

(i = 1, . . . , n)

and assume that / is an irreducible action. Viewing the source group as a product of n ≥ 2 factors acting irreducibly, and recalling that the target groups i are “hyperboliclike,” Monod and Shalom apply Theorem 5.5. The cocycles arising from ME-coupling turn out to be nonelementary, leading to the conclusion that each cocycle ci is conjugate to a homomorphism ρi : j(i) → i , modulo some reductions and ﬁnite kernels. Since i commute, the conjugations can be performed independently and simultaneously on all the cocycles ci . After some intricate analysis of the map i → j(i), kernels and cokernels of ρi , Monod and Shalom show that in the setting of ME-couplings as above the map, i → j(i) is a permutation and ρi are isomorphisms. Thus the original cocycle c can be conjugate to an automorphism of . This ME-cocycle superrigidity can now be plugged into an analogue of Theorem 5.4 to give a measurable bi--equivariant map → , which can be used as an input to a construction like Theorem 5.13. This allows us to identify unknown groups measure equivalent to = 1 × · · · × n . The only delicate SOE

point is that starting from a X ∼ Y and the corresponding (, )ˇ and apply coupling one needs to look at the self--coupling = × the cocycle superrigidity result to / . In order to guarantee that the latter action is irreducible, Monod and Shalom require X to be irreducible and Y to be mildly mixing. They also show that the assumption on mild mixing is necessary for the result.

a survey of measured group theory / 353

In [12] Uri Bader and the author proposed to study higher-rank superrigidity phenomena using a notion of a (generalized) Weyl group, which works well for higher-rank simple Lie groups, arbitrary products G = G1 × · · · × Gn of n ≥ 2 ˜ 2 groups, which are close relatives to lattices in SL3 (Qp ). factors, and exotic A In particular:

theorem 5.6. (bader-furman [12]) Theorem 5.5 holds for target groups from class Dea . Here Dea is a class of hyperboliclike groups that include many of the examples in Creg . Plugging this into the Monod-Shalom machine one obtains the same results of products of groups in class Dea . 5.3. Strong Rigidity for Cocycles

In the proof of Theorem 5.4 Zimmer’s cocycle superrigidity was applied to a measure equivalence cocycle. This is a rather special class of cocycles (see §A.3). If cocycles are analogous to representations of lattices, then ME-cocycles are analogous to isomorphisms between lattices; in particular, they have an “inverse.” Kida’s work on ME for mapping class groups focuses on rigidity results for such cocycles. We shall not attempt to explain the ingredients used in this work, but will just formulate the main technical result analogous to Theorem 5.4. Let be a subgroup of ﬁnite index in ( g,p ) with 3g + p − 4 > 0, C = C( g,p ) denoting its curve complex, and Aut (C) the group of its automorphisms; this is a countable group commensurable to .

theorem 5.7. (kida [86]) Let (, m) be a self-ME-coupling of . Then there exists a measurable map × -equivariant map : → Aut (C). Returning to the point that ME-cocycles are analogous to isomorphism between lattices, one might wonder whether Theorem 5.4 holds in cases where Mostow rigidity applies, speciﬁcally for G of rank 1 with PSL2 (R) excluded. In [13] this is proved for G Isom (HnR ), n ≥ 3, and a restricted ME.

theorem 5.8. (bader-furman-sauer [13]) Theorem 5.4 applies to 1 ME-couplings of lattices in G = SOn,1 (R), n ≥ 3. The proof of this result uses homological methods (1 and other completions of the usual homology) combined with a version of the Gromov-Thurston proof of Mostow rigidity (for Isom (HnR ), n ≥ 3) adapted to this setting.

354 / a. furman 5.4. Cocycle Superrigid Actions

In all the previous examples the structure of the acting group was the sole source for (super)rigidity. Recently Sorin Popa has developed a number of remarkable cocycle superrigidity results of a completely different nature [109– 115]. These results exhibit an extreme form of cocycle superrigidity and rather than relying only on the properties of the acting group , take advantage of the action (X , μ).

definition 5.9. An action (X , μ) is C -cocycle superrigid, where C is some class of topological groups, if for every ∈ C every measurable cocycle c : X → has the form c(g, x) = f (gx)−1 ρ(g)f (x) for some homomorphism ρ : → and some measurable f : X → . Here we shall focus on the class Gdsc of all countable groups; however, the following results hold for all cocycles taking values in a broader class Uﬁn , which contains Gdsc and Gcpt —separable compact groups. Note that the concept of Gdsc -cocycle superrigidity is unprecedentedly strong: there is no assumption on the cocycle, the assumption on the target group is extremely weak, and the “untwisting” takes place in the same target group.

theorem 5.10. (popa [113]) Let be a group with property (T) and (X , μ) = (X0 , μ0 ) be the Bernoulli action. Then (X , μ) is Gdsc -cocycle superrigid. In fact, the result is stronger: it sufﬁces to assume that has relative property (T) with respect to a w-normal subgroup 0 , and (X , μ) has a relatively weakly mixing extension (X¯ , μ) ¯ that is s-malleable, while ¯ 0 (X , μ) ¯ is weakly mixing. Under these conditions (X , μ) is Uﬁn cocycle superrigid. See [113] and [48] for the relevant deﬁnitions and more details. We indicate the proof (of the special case above) in §5.8. Vaguely speaking, Popa’s approach exploits the tension between certain (local) rigidity provided by the acting group and deformations supplied by the action. In the following remarkable result, Popa further relaxed the property (T) assumption.

theorem 5.11. (popa [115]) Let be a group containing a product 1 × 2 where 1 is nonamenable, 2 is inﬁnite, and 1 × 2 is w-normal in . Then any Bernoulli action (X , μ) is Uﬁn -cocycle superrigid.

a survey of measured group theory / 355

The deformations alluded to above take place for the diagonal -action on the square (X × X , μ × μ). This action is supposed to be ergodic; equivalently, the original action should be weakly mixing and satisfy addition properties. Isometric actions, or staying within the ergodic-theoretic terminology, actions with discrete spectrum provide the opposite type of dynamics. These actions have the form K /L where L < K are compact groups, → K a homomorphism with dense image, and acts by left translations. Totally disconnected K corresponds to proﬁnite completion lim / n with respect to a chain of ←− normal subgroups of ﬁnite index. Isometric actions K /L with proﬁnite K , can be called proﬁnite ergodic actions of —these are precisely inverse limits X = lim Xn of transitive -actions on ﬁnite spaces. Adrian Ioana found ←− the following “virtually Gdsc -cocycle superrigidity” phenomenon for proﬁnite actions of Kazhdan groups.

theorem 5.12. (ioana [72]) Let X = K /L be an ergodic proﬁnite action. Assume that has property (T), or a relative property (T) with respect to a normal subgroup 0 , which acts ergodically on X . Then any measurable cocycle c : X → into a discrete group, is conjugate to a cocycle coming from a ﬁnite quotient X → Xn , that is, c is conjugate to a cocycle induced from a homomorphism n → of a ﬁnite-index subgroup. In §5.8.2 a similar result is proven for all discrete spectrum actions (not necessarily proﬁnite ones). 5.5. Constructing Representations

In Geometric group theory many QI-rigidity results are proved using the following trick. Given a metric space X one declares self-quasi-isometries f , g : X → X to be equivalent if sup d( f (x), g(x)) < ∞. x∈X

Then equivalence classes of quasi-isometries (hereafter q.i.) form a group, denoted QI(X ). This group contains (a quotient of) Isom (X ), which can sometimes be identiﬁed within QI(X ) in coarse geometric terms. If is a group with well-understood QI() and is an unknown group q.i. to , then one gets a homomorphism ρ : → Isom () → QI() ∼ = QI() whose kernel and image can then be analyzed.

356 / a. furman

Facing a similar problem in the measure equivalence category, there is a difﬁculty in deﬁning an analogue for QI(). Let us describe a construction that allows us to analyze the class of all groups ME to a given group from an information about self-ME-couplings of . Let G be a lcsc unimodular group. Let us assume that G has the strong ICC property, by which we mean that the only regular Borel conjugationinvariant probability measure on G is the trivial one, namely the Dirac mass δe at the origin. For countable groups this is equivalent to the condition that all nontrivial conjugacy classes are inﬁnite, that is, the usual ICC property. Connected, (semi) simple Lie groups with trivial center and no compact factors provide other examples of strongly ICC groups. Theorems 5.4, 5.7 and 5.8 are instances where a strongly ICC group G has the property that for any13 ME-self-coupling (, m) of a lattice in G there exists a bi--equivariant measurable map to G, that is, a Borel map : → G satisfying m-a.e.: ((γ1 , γ2 )ω) = γ1 (ω)γ2−1

(γ1 , γ2 ∈ ).

It is not difﬁcult to see that the strong ICC property implies that such a map is also unique. (It should also be pointed out that the existence of such maps for self-couplings of lattices is equivalent to the same property for self-couplings of the lcsc group G itself; but here we shall stay in the framework of countable groups). The following general tool shows how these properties of G can be used to classify all groups ME to a lattice < G; up to ﬁnite kernels these turn out to be lattices in G.

theorem 5.13. (bader-furman-sauer [13]) Let G be a strongly ICC lcsc unimodular group, < G a lattice, and some group ME to and (, m) be ˇ of admits a a (, )-coupling. Assume that the self-ME-coupling = × Borel map : → G, satisfying a.e.: ([γ1 x, γ2 y]) = γ1 · ([x, y]) · γ2−1

(γ1 , γ2 ∈ ).

¯ with K ﬁnite and ¯ being Then there exists a short exact sequence K −→−→ a lattice in G, and a Borel map : → G so that a.e.: ([x, y]) = (x) · (y)−1 ,

(γ z) = γ · (z),

(λz) = (z) · λ¯ −1 .

13. In the case of G = Isom (Hn ) we restrict to all 1 -ME-couplings.

a survey of measured group theory / 357

Moreover, the push-forward of ∗ m is a Radon measure on G-invariant under the maps ¯ g → γ g λ,

¯ (γ ∈ , λ¯ ∈ ).

If G is a (semi)simple Lie group, the last condition on the push-forward measure can be analyzed using Ratner’s theorem (as in Theorem 5.4) to deduce that assuming ergodicity ∗ m is either a Haar measure on G, or on a coset of its ﬁnite-index subgroup, or it is (proportional to) counting measure on a coset ¯ as ﬁnite-index subgroups. of a lattice containing and a conjugate of Theorem 5.13 is a streamlined and improved version of similar statements obtained in [44] for higher-rank lattices, in [101] for products, and in [86] for mapping class groups. 5.6. Local Rigidity for Measurable Cocycles

The rigidity vs. deformations approach to rigidity results developed by Sorin Popa led to a number of striking results in von Neumann algebras and in Ergodic theory (some have been mentioned in §5.4). Let us illustrate the rigidity side of this approach by the following simple purely ergodic-theoretic statement, which is a variant of Hjorth’s [68, lemma 2.5]. Recall that one of the several equivalent forms of property (T) is the following statement: a lcsc group G has (T) if there exists a compact K ⊂ G and > 0 so that for any unitary G-representation π and any (K , )-almost invariant unit vector v there exists a G-invariant unit vector w with v − w < 41 .

proposition 5.14. Let G be a group with property (T) and (K , ), as above. Then for any ergodic probability measure-preserving action G (X , μ), any countable group and any pair of cocycles α, β : G × X → with 2 μ x ∈ X : α(g, x) = β(g, x) > 1 − 2

(∀g ∈ K )

there exists a measurable map f : X → so that β = α f . Moreover, one can assume that 3 μ x : f (x) = e > . 4 Proof. Let X˜ = X × be equipped with the inﬁnite measure μ˜ = μ × m where m stands for the counting measure on . Then G acts on (X˜ , μ) ˜ by g : (x, λ) → (g.x, α(g, x)λβ(g, x)−1 ).

358 / a. furman

This action preserves μ˜ and we denote by π the corresponding unitary Grepresentation on L2 (X˜ , μ). ˜ The characteristic function v = 1X ×{e} satisﬁes 2 (g ∈ ) = 2 v − π (g)v2 = 2 − 2Reπ (g)v, v < 2 − 2 1 − 2 ˜ with and therefore there exists a π(G)-invariant unit vector w ∈ L2 (X˜ , ) 1 2 2 v − w < 4 . Since 1 = w = X λ |w(x, λ)| we may deﬁne (x) = λ : |w(x, λ)| = p(x) p(x) = max |w(x, λ)|, λ

and observe that p(x) and the cardinality k(x) of the ﬁnite set (x) are measurable -invariant functions on (X , μ); hence are a.e. constants p(x) = p ∈ (0, 1], 1 k(x) = k ∈ {1, 2, . . . }. Since 16 > v − w2 ≥ (1 − p)2 we have p > 43 . It follows that k = 1 because 1 = w2 ≥ kp2 . Therefore (x) = {f (x)} for some measurable map f : X → . The π(G)-invariance of w gives π(G)-invariance ˜ of the characteristic function of (x, f (x)) ∈ X : x ∈ X , which is equivalent to f (gx) = α(g, x)f (x)β(g, x)−1 and β = αf . Let A = f −1 ({e}) and a = μ(A). Since λ |w(x, λ)|2 is a G-invariant function it is a.e. constant w2 = 1. Hence for x ∈ / A we have |w(x, e)|2 ≤ 2 2 1 − |w(x, f (x))| = 1 − p , and 5.1

9(1 − a) 1 > v − w2 ≥ a · (1 − p2 ) + (1 − a) · (1 − (1 − p2 )) ≥ (1 − a) · p2 > . 16 16 Thus a = μ x ∈ X : f (x) = e > 89 > 43 , as required. 5.7. Cohomology of Cocycles

Let us ﬁx two groups and . There is no real assumption on , it may be any lcsc group, but we shall impose an assumption on . One might focus on the case where is a countable group (class Gdsc ), but versions of the statements below would apply also to separable compact groups, or groups in a larger class Uﬁn of all Polish groups, which embed in the unitary group of a von Neumann algebra with ﬁnite faithful trace,14 or a potentially even larger class Gbinv of groups with a bi-invariant metric, and the class Galg of connected algebraic groups over local ﬁelds, say of zero characteristic.

14. This class, introduced by Popa, contains both discrete countable groups and separable compact ones.

a survey of measured group theory / 359

Given a (not necessarily free) p.m.p. action (X , μ) let Z 1 (X , ) or Z 1 ( X , ) denote the space of all measurable cocycles c : × X → and by H 1 (X , ), or H 1 ( X , ), the space of equivalence classes of cocycles up to conjugation by measurable maps f : X → . If ∈ Galg , we shall focus on a subset Hss1 (X , ) of (classes of) cocycles whose algebraic hull is connected, semisimple, center free, and has no compact factors. Any -equivariant quotient map π : X → Y deﬁnes a pull-back Z 1 (Y , ) → Z 1 (X , ) by c π (g, x) = c(g, π(x)), which descends to π∗

H 1 (Y , )−→H 1 (X , ). ¯ and j : < give rise to push-forward maps Group inclusions i : < , i∗

¯ H 1 (X , )−→H 1 (X , ),

j∗

H 1 ( X , )−→H 1 ( X , ).

question 5.15. What can be said about these maps of the cohomology ? The discussion here is inspired and informed by Popa’s [113]. In particular, the following statements Propositions 5.16(2), 5.17, and 5.18(1), and Corollary 5.20 are variations on Popa’s original [113, lemma 2.11, proposition 3.5, lemma 3.6, theorem 3.1]. Working with class Gbinv makes the proofs more transparent than in Uﬁn —this was done in [48, §3]. Proposition 5.16(3) for semisimple target is implicit in [45, lemma 3.5]. The full treatment of the statements below, including Theorem 5.19, will appear in [49].

proposition 5.16. Let π : X → Y be a -equivariant quotient map. Then π∗

H 1 (Y , )−→H 1 (X , ) is injective in the following cases: 1) is discrete and torsion free. 2) ∈ Gbinv and π : X → Y is relatively weakly mixing. 3) ∈ Galg and H 1 (−, ) is replaced by Hss1 (−, ). The notion of relative weakly mixing was introduced independently by Zimmer [137] and Furstenberg [51]: a -equivariant map π : X → Y is relatively weakly mixing if the -action on the ﬁbered product X ×Y X is ergodic (or ergodic relatively to Y ); this turns out to be equivalent to the condition that X contains no intermediate isometric extensions of Y .

360 / a. furman

¯ ∈ Gbinv be a closed subgroup, and proposition 5.17. Let i : < (X , μ) some p.m.p. action. Then i∗

¯ H 1 (X , )−→H 1 (X , ) is injective. This useful property fails in the Galg setting: if < G is a lattice in a (semi) simple Lie group and c : × G/ → in the canonical class, then viewed as a cocycle into G > , c is conjugate to the identity embedding ∼ = < G, but as a -valued cocycle it cannot be “untwisted.”

proposition 5.18. Let π : X → Y be a quotient map of ergodic actions, and j : < be a normal, or sub-normal, or w-normal closed subgroup acting ergodically on X . Assume that either 1) ∈ Gbinv and π is relatively weakly mixing, or 2) ∈ Galg and one considers Hss1 (−, ). Then H 1 ( Y , ) is the push-out of the rest of the following diagram: H 1 ( X , )

j∗

O

π∗

H 1 ( Y , )

/ H 1 ( X , ) O π∗

j∗

/ H 1 ( Y , ).

In other words, if the restriction to X of a cocycle c : × X → is conjugate to one descending to × Y → , then c has a conjugate that descends to × X → . The condition < is w-normal (weakly normal), which means that there exists a well-ordered chain i of subgroups starting from and ending with , so that i % i+1 and for limit ordinals j = i<j i (Popa). Let πi : X → Yi is a collection of -equivariant quotient maps. Then X has a unique -equivariant quotient p : X → Z = Yi , which is maximal among all common quotients pi : Yi → Z. Identifying -equivariant quotients with -equivariant complete sub-σ -algebras of X , one has p−1 (Z ) = i πi−1 (Yi ); or in the operator algebra formalism p−1 (L∞ (Z)) = i πi−1 (L∞ (Yi )).

a survey of measured group theory / 361

theorem 5.19. Let πi : X → Yi , 1 ≤ i ≤ n, be a ﬁnite collection of equivariant quotients, and Z = ni=1 Yi . Then H 1 (Z, ) is the push-out of H 1 (Yi , ) under conditions (1)–(3) of Proposition 5.16: 8 ppp p p ppp ppp π1∗

H 1 (Y1 , )

H 1 (X , )

O

πi∗

fNNN π ∗ NNN n NNN NN

· · · H 1 (Yi , ) · · · H 1 (Yn , ) fNNN p∗ O 8 ∗ p1 pp NNN n ppp NNN p pi∗ p p NN ppp H 1 (Z, )

More precisely, if ci : × X → are cocycles (in case (3) assume [ci ] ∈ Hss1 (Yi , )), whose pullbacks ci (g, πi (x)) are conjugate over X , then there exists a unique class [c] ∈ H 1 (Z, ), so that c(g, pi (y)) ∼ ci (g, y) in Z 1 (Yi , ) for all 1 ≤ i ≤ n. The proof of this Theorem relies on Proposition 5.16 and contains it as a special case n = 1. This result can be useful to push cocycles to deeper and deeper quotients; if π : X → Y is a minimal quotient to which a cocycle or a family of cocycles can descend up to conjugacy, then it is the minimal or characteristic quotient for these cocycles: if they descend to any quotient X → Y , then necessarily X → Y → Y . For example if < G is a higher-rank lattice, is a discrete group and c : × X → is an OE (or ME) cocycle, then either c descends to a -action on a ﬁnite set (virtual isomorphism case), or to π X −→G/ with lattice in G, where π is uniquely deﬁned by c (initial OE or ME). An important special (and motivating) case of Theorem 5.19 is that of X = Y × Y where Y is a weakly mixing action. Then the projections πi : X → Yi = Y , i = 1, 2, give Z = Y1 ∧ Y2 = {pt} and H 1 ( {pt}, ) = Hom(, ). So

corollary 5.20. (popa [113, theorem 3.1], see also [48, theorem 3.4]) Let Y be a weakly mixing action and c : × Y → a cocycle into ∈ Gbinv . Let X = Y × Y with the diagonal -action, c1 , c2 : × X → the cocycles ci (g, (y1 , y2 )) = c(g, yi ). If c1 ∼ c2 over X , then there exists homomorphism ρ : → and a measurable f : Y → , so that c(g, y) = f (gy)−1 ρ(g) f (y).

362 / a. furman 5.8. Proofs of Some Results

In this section we shall give relatively self-contained proofs of some of the results mentioned above.

5.8.1. sketch of a proof for popa’s cocycle superrigidity theorem 5.10 First note that without loss of generality the base space (X0 , μ0 ) of the Bernoulli action may be assumed to be nonatomic. Indeed, Proposition 5.16(2) implies that for each of the classes Gdsc ⊂ Uﬁn ⊆ Gbinv the corresponding cocycle superrigidity descends through relatively weakly mixing quotients, and ([0, 1], dx) → (X0 , μ0 ) is such. Given any action (X , μ) consider the diagonal -action on (X × X , μ × μ) and its centralizer Aut (X × X ) in the Polish group Aut (X × X , μ × μ). It always contain the ﬂip F : (x, y) → (y, x). Bernoulli actions X = [0, 1] have the special property (called s-malleability by Popa) that there is a path p : [1, 2] → Aut (X × X ),

with

p1 = Id,

p2 = F.

Indeed, the diagonal component-wise action of Aut ([0, 1] × [0, 1]) on X × X = ([0, 1] × [0, 1]) embeds into Aut (X × X ) and can be used to connect Id to F. Fix a cocycle c : X → . Consider the two lifts to X × X → X : ci : X × X → ,

ci (g, (x1 , x2 )) = c(g, xi ),

(i = 1, 2).

Observe that they are connected by the continuous path of cocycles ct (g, (x, y)) = c1 (g, pt (x, y)), 1 ≤ t ≤ 2. Local rigidity in Proposition 5.14 implies that c1 and c2 are conjugate over X × X , and the proof is completed invoking Corollary 5.20. Under the weaker assumption of relative property (T) with respect to a w-normal subgroup, Popa uses Proposition 5.18.

5.8.2. a cocycle superrigidity theorem We state and prove a cocycle superrigidity theorem, inspired and generalizing Adrian Ioana’s Theorem 5.12. Thus a number of statements (Theorems 4.15 and 4.20(2) and §4.3.1) in this survey get a relatively full treatment. The proof is a good illustration of Popa’s deformation vs. rigidity approach. Recall that an ergodic p.m.p. action (X , μ) is said to have a discrete spectrum if the Koopman -representation on L2 (X , μ) is a Hilbert sum of ﬁnite-dimensional subrepresentations. Mackey proved (generalizing the Halmos–von Neumann theorem for Z, and using Peter-Weyl ideas) that discrete spectrum action is measurably isomorphic to the isometric -action on

a survey of measured group theory / 363

(K /L, mK /L ), g : kL → τ (g)kL, where L < K are compact separable groups and τ : → K is a homomorphism with dense image.

theorem 5.21. (after ioana’s theorem 5.12, [72]) Let (X , μ) be an ergodic p.m.p. action with discrete spectrum. Assume that has property (T), or contains a w-normal subgroup 0 with property (T) acting ergodically on (X , μ). Let be an arbitrary torsion-free discrete countable group and c : × X → be a measurable cocycle. Then there is a ﬁnite-index subgroup 1 < , a 1 -ergodic component X1 ⊂ X (of measure μ(X1 ) = [ : 1 ]−1 ), a homomorphism ρ : 1 → , and a measurable map φ : X → , so that the conjugate cocycle c φ restricted to 1 X1 → , is the homomorphism ρ : 1 → . The cocycle c φ : × X → is induced from ρ. The assumption that is torsion free is not essential; in general, one might need to lift the action to a ﬁnite cover Xˆ 1 → X1 via a ﬁnite group that embeds in . If K is a connected Lie group, then 1 = and X1 = X = K /L. The stated result is deduced from the case where L is trivial, that is, X = K , using Proposition 5.16(1). We shall make this simpliﬁcation and assume has property (T) (the modiﬁcation for the more general case uses an appropriate version of Propositions 5.14 and 5.17). An appropriate modiﬁcation of the result handles compact groups as a possible target group for the cocycle. Proof. The K -action by right translations: t : x → xt −1 commutes with the action on K ; in fact, K is precisely the centralizer of in Aut (K , mK ). This allows us to deform the initial cocycle c : × X → , setting ct (g, x) = c(g, xt −1 )

(t ∈ K ).

Let F ⊂ and > 0 be as in the “local rigidity” Proposition 5.14. Then for some open neighborhood U of e ∈ K for every t ∈ U there is a unique measurable ft : K → with ct (g, x) = c(g, xt −1 ) = ft (gx)c(g, x)ft (x)−1

3 μ x : ft (x) = e > . 4

Suppose t, s ∈ U, and ts ∈ U. Then fts (gx) c(g, x) fts (x)−1 = cts (g, x) = c(g, xs−1 t −1 ) = ft (gxs−1 ) c(g, xs−1 ) ft (xs−1 )−1 = ft (gxs−1 )fs (gx) c(g, x) [ft (xs−1 )fs (x)−1 ]−1 .

364 / a. furman

This can be rewritten as F(gx) = c(g, x) F(x) c(g, x)−1 ,

where

F(x) = fts (x)−1 ft (xs−1 )fs (x).

Since ft , fs , fts takes value e with probability > 3/4, it follows that A = F −1 ({e}) has μ(A) > 0. The equation implies -invariance of A. Thus μ(A) = 1 by ergodicity. Hence, whenever t, s, ts ∈ U 5.2

fts (x) = ft (xs−1 )fs (x).

If K is a totally disconnected group, that is, a proﬁnite completion of as in Ioana’s Theorem 5.12, then U contains an open subgroup K1 < K . In this case one can skip the following paragraph. In general, let V be a symmetric neighborhood of e ∈ K so that V 2 ⊂ U, n and let K1 = ∞ n=1 V . Then K1 is an open (hence, also closed) subgroup of K ; in the connected case K1 = K . We shall extend the family {ft : K → }t∈V to be deﬁned for all t ∈ K1 while satisfying Equation (5.2), using a “cocycle continuation” procedure akin to analytic continuation. For t, t ∈ K1 a V -quasipath pt→t from t to t is a sequence t = t0 , t1 , . . . , tn = t where ti ∈ ti−1 V . Two V quasipaths from t to t are V -close if they are within V -neighborhoods from each other. Two V -quasipaths pt→t and qt→t are V -homotopic if there is a (0) (k) chain pt→t = pt→t , . . . , pt→t = qt→t of V -quasipaths where p(i−1) and p(i) are V -close, 1 ≤ i ≤ k. Iterating Equation (5.2) one may continue the deﬁnition of f· from t to t along a V -quasipath; the continuation being the same for V -close quasipaths, and therefore for V -homotopic quasipaths as well (all from t to t ). The possible ambiguity of this cocycle continuation procedure (V ) is encoded in the homotopy group π1 (K1 ) consisting of equivalence classes of V -quasipaths from e → e modulo V -homotopy. We claim that this group (V ) is ﬁnite. In the case of a connected Lie group K1 , π1 (K1 ) is a quotient of π1 (K1 ) that is ﬁnite since K1 , contains a dense property (T) group and can(V ) not have torus factors. This covers the general case as well since π1 (K1 ) “feels” only ﬁnitely many factors when K1 is written as an inverse limit of connected Lie groups and ﬁnite groups. Considering the continuations of f along (V ) V -quasipaths e → e we get a homomorphism π1 (K1 ) → , which must be trivial since was assumed to be torsion free. Therefore, we obtain a family of measurable maps ft : K1 → indexed by t ∈ K1 and still satisfying Equation (5.2). Let 1 = τ −1 (K1 ). Then the index [ : 1 ] = [K : K1 ] is ﬁnite. We shall focus on the restriction c1 of c to 1 K1 . Note that Equation (5.2) is a cocycle equation for the simply transitive action K1 on itself. It follows by a standard argument that it is a coboundary. Indeed, for a.e. x0 ∈ K1 Equation (5.2) holds

a survey of measured group theory / 365

for a.e. t, s ∈ K1 . In particular, for a.e. t, x ∈ K1 , using s = x −1 x0 , one obtains ftx −1 x0 (x0 ) = ft (x)fx −1 x0 (x0 ). This gives ft (x) = φ(xt −1 )φ(x)−1 ,

where

φ(x) = fx −1 x0 (x0 ).

Equation ct = c ft translates into the fact that the cocycle c φ (g, x) = φ(gx)−1 c(g, x)φ(x) satisﬁes for a.e. x, t c φ (g, xt −1 ) = c φ (g, x). Thus c(g, x) does not depend on the space variable. Hence, it is a homomorphism c φ (g, x) = ρ(g). φ

Finally, the fact that c φ is induced from c1 is straightforward.

Appendix A. Cocycles Let G (X , μ) be a measurable, measure-preserving (sometimes just measure class preserving) action of a topological group G on a standard Lebesgue space (X , μ), and H be a topological group. A Borel measurable map c : G × X → H forms a cocycle if for every g1 , g2 ∈ G for μ-a.e. x ∈ X one has c(g2 g1 , x) = c(g2 , g1 .x) · c(g1 , x). If f : X → H is a measurable map and c : G × X → H is a measurable cocycle, deﬁne the f -conjugate c f of c to be c f (g, x) = f (g.x)−1 c(g, x) f (x). It is straightforward to see that c f is also a cocycle. One says that c and c f are (measurably) conjugate, or cohomologous cocycles. The space of all measurable cocycles × X → is denoted by Z 1 ( X , ) and the space of equivalence classes by H 1 ( X , ). Cocycles that do not depend on the space variable c(g, x) = c(g) are precisely homomorphisms c : G → H. So cocycles may be viewed as generalized homomorphisms. In fact, any group action G (X , μ) deﬁnes a measured groupoid G with G (0) = X , and G (1) = (x, gx) : x ∈ X , g ∈ G (see [9] for the background). In this context cocycles can be viewed as homomorphisms G → H. If π : (X , μ) → (Y , ν) is an equivariant quotient map between -actions (so π∗ μ = ν and π ◦ γ = γ for γ ∈ ), then for any target group any cocycle c : × Y → lifts to c¯ : × X → by

366 / a. furman

c¯ (g, x) = c(g, π (x)). Moreover, if c = c f ∼ c in Z 1 ( Y , ), then the lifts c¯ = c¯ f ◦π ∼ c¯ in π Z 1 ( X , ); so X −→Y induces π

H 1 ( X , )←−H 1 ( Y , ). Note that Hom(, ) is Z 1 ( {pt}, ) and classes of cocycles on × X → cohomologous to homomorphisms is precisely the pull-back of H 1 ( {pt}, ). A.1. The Canonical Class of a Lattice, (Co)induction

Let < G be a lattice in a lcsc group. By deﬁnition the transitive G-action on X = G/ has an invariant Borel regular probability measure μ. Let F ⊂ G be a Borel fundamental domain for the right -action on G (i.e., F is a Borel subset of G set that meets every coset g precisely once). Fundamental domains correspond to Borel cross-section σ : G/ → G of the projection G → G/ . Deﬁne cσ : G × G/ →

by

cσ (g, h) = σ (gh)−1 g σ (h).

Clearly, this is a cocycle (a conjugate of the identity homomorphism G → G); however, cσ takes values in the subgroup of G. This cocycle is associated to a choice of the cross-section σ (equivalently, the choice of the fundamental domain); starting from another Borel cross-section σ : G/ → G results in a cohomologous cocycle: f

cσ = cσ

where

f : G/ →

is deﬁned by

σ (x) = f (x)σ (x).

Let be a lattice in G. Then any action (X , μ) gives rise to the induced G-action (a.k.a. suspension) on X¯ = G × X where G acts on the ﬁrst coordinate. Equivalently, X¯ = G/ × X and g : (g , x) → (gg , c(g, g )x) where c : G × G/ → is in the canonical class. Here the G-invariant ﬁnite measure μ¯ = mG/ × μ is ergodic iff μ is -ergodic. If α : × X → H is a cocycle, the induced cocycle α¯ : G × X¯ → H is given by α(g, ¯ (g , x)) = α(c(g, g ), x). The cohomology of α¯ is the same as that of α (one relates maps F : X¯ → H to f : X → H by f (x) = F(e, x) taking instead of e a generic point in G/ ). In particular, α¯ is cohomologous to a homomorpism π¯ : G → H iff α is cohomologous to a homomorphism → H; see [146] for details. Cocycles appear quite naturally in a number of situations such as (volumepreserving) smooth actions on manifolds, where choosing a measurable trivialization of the tangent bundle, the derivative becomes a matrix-valued

a survey of measured group theory / 367

cocycle. We refer the reader to David Fisher’s survey [42] where this type of cocycle is extensively discussed in the context of Zimmer’s program. Here we shall be interested in a different type of cocycles: “rearrangement” cocycles associated to orbit equivalence, measure equivalence, and so on. as follows. A.2. OE-Cocycles

Let (X , μ) and (Y , ν) be two measurable, measure-preserving ergodic actions on probability spaces, and T : (X , μ) → (Y , ν) be an Orbit Equivalence. Assume that the -action is essentially free, that is, for ν-.a.e y ∈ Y , the stabilizer y = h ∈ : h.y = y is trivial. Then for every g ∈ and μ-a.e. x ∈ X , the points T (g.x), T (x) ∈ Y lie on the same -orbit. Let α(g, x) ∈ denote the (a.e. unique) element of with T (g.x) = α(g, x).T (x). Considering x, g.x, g g.x one checks that α is actually a cocycle α : × X → . We shall refer to such α as the OE-cocycle, or the rearrangement cocycle, corresponding to T . Note that for μ-a.e. x, the map α(−, x) : → is a bijection; it describes how the -names of points x ∈ .x translate into the -names of y ∈ .T (x) under the map T . The inverse map T −1 : (Y , ν) → (X , μ) deﬁnes an OEcocycle β : × Y → , which serves as an “inverse” to α in the sense that a.e.: β(α(g, x), T (X )) = g

(g ∈ ).

A.3. ME-Cocycles

Let (, m) be an ME-coupling of two groups and and let Y , X ⊂ be fundamental domains for , actions, respectively. The natural identiﬁcation / ∼ = X , ω → ω ∩ X , translates the -action on / to X by γ :X

x → gα(g, x)x ∈ X

where α(γ , x) is the unique element in taking γ x ∈ into X ⊂ . It is easy to see that α : × X → is a cocycle with respect to the above -action on X , which we denote by a dot γ · x to distinguish it from the -action on . (If and are lattices in G, then α : × G/ → is the restriction of the canonical cocycle G × G/ → ). Similarly we get a cocycle β : × Y → . So the (, ) ME-coupling and a choice of fundamental domains Y ∼ = / , X ∼ = / deﬁne a pair of cocycles

368 / a. furman

α : × / → ,

A.1

β : × / → .

Changing the fundamental domains amounts to conjugating the cocycles and vice versa.

remark 5.22. One can characterize ME-cocycles among all measurable cocycles α : × X → as discrete ones with ﬁnite covolume. These concepts refer to the following construction: let (X˜ , μ) ˜ = (X × , μ × m ) and let act by g : (x, h) → (g.x, α(g, x)h). Say that the cocycle is discrete and has ﬁnite covolume if the action (X˜ , μ) ˜ admits a ﬁnite-measure fundamental domain. References [1] [2] [3] [4] [5] [6] [7] [8] [9]

[10] [11]

[12] [13] [14]

M. Abert and N. Nikolov, Rank gradient, cost of groups and the rank versus Heegaard genus problem (2007), available at arXiv:0701.361. M. Abert and B. Weiss, Bernoulli actions are weakly contained in any free action, in preparation. S. Adams, An equivalence relation that is not freely generated, Proc. Amer. Math. Soc. 102 (1988), no. 3, 565–566. ———, Trees and amenable equivalence relations, Ergod Th. & Dynam. Sys. 10 (1990), no. 1, 1–14. ———, Reduction of cocycles with hyperbolic targets, Ergod. Th. & Dynam. Sys. 16 (1996), no. 6, 1111–1145. S. R. Adams and R. J. Spatzier, Kazhdan groups, cocycles and trees, Amer. J. Math. 112 (1990), no. 2, 271–287. A. Alvarez and D. Gaboriau, Free products, orbit equivalence and measure equivalence rigidity (2008), available at arXiv:0806.2788. C. Anantharaman-Delaroche, Cohomology of property T groupoids and applications, Ergod. Th. & Dynam. Sys. 25 (2005), no. 4, 977–1013. C. Anantharaman-Delaroche and J. Renault,, Amenable groupoids, Monographies de L’Enseignement Mathématique [Monographs of L’Enseignement Mathématique], vol. 36, L’Enseignement Mathématique, Geneva, 2000. With a foreword by Georges Skandalis and Appendix B by E. Germain. J. E. Andersen, Mapping class groups do not have Kazhdan’s property (T) (2007), available at arXiv:0706.2184. M. F. Atiyah, Elliptic operators, discrete groups and von Neumann algebras, Colloque “Analyse et Topologie” en l’Honneur de Henri Cartan (Orsay, 1974),Soc. Math. France, Paris, 1976, pp. 43–72. Astérisque, No. 32–33. U. Bader and A. Furman, Superrigidity via Weyl groups: hyperbolic-like targets (preprint, 2010). U. Bader, A. Furman, and R. Sauer, Intergrable measure equivalence and rigidity of hyperbolic lattices (2010), available at arXiv:1006.5193. ———, Efﬁcient subdivision in hyperbolic groups and applications (2010), available at arXiv:1003.1562.

a survey of measured group theory / 369 [15] [16] [17] [18]

[19] [20] [21] [22] [23] [24]

[25] [26] [27] [28] [29] [30]

[31] [32] [33] [34] [35] [36]

U. Bader, A. Furman, and A. Shaker, Superrigidity, Weyl groups, and actions on the circle (2006), available at arXiv:math/0605276. M. E. B. Bekka and A. Valette, Kazhdan’s property (T) and amenable representations, Math. Z. 212 (1993), no. 2, 293–299. B. Bekka, P. de la Harpe, and A. Valette, Kazhdan’s property (T ), New Mathematical Monographs, Vol. 11, Cambridge University Press, Cambridge, 2008. J. Bourgain, A. Furman, E. Lindenstrauss, and S. Mozes, Invariant measures and stiffness for non-abelian groups of toral automorphisms, C. R. Math. Acad. Sci. Paris 344 (2007), no. 12, 737–742. M. R. Bridson, M. Tweedale, and H. Wilton, Limit groups, positive-genus towers and measure-equivalence, Ergod. Th. & Dynam. Sys. 27 (2007), no. 3, 703–712. M. Burger and N. Monod, Bounded cohomology of lattices in higher rank Lie groups, J. Eur. Math. Soc. (JEMS), 1, (1999), no. 2, 199–235. ———, Continuous bounded cohomology and applications to rigidity theory, Geom. Funct. Anal. 12 (2002), no. 2, 219–280. M. Burger and S. Mozes, CAT(−1)-spaces, divergence groups and their commensurators, J. Amer. Math. Soc. 9 (1996), no. 1, 57–93. J. Cheeger and M. Gromov, L2 -cohomology and group cohomology, Topology 25 (1986), no. 2, 189–215. P.-A. Cherix, M. Cowling, P. Jolissaint, P. Julg, and A. Valette, Groups with the Haagerup property, Progress in Mathematics, vol. 197 Birkhäuser Verlag, Basel, 2001. Gromov’s a-T-menability. I. Chifan and A. Ioana, Ergodic subequivalence relations induced by a Bernoulli action (2008), available at arXiv:0802.2353. A. Connes, A factor of type II1 with countable fundamental group, J. Operator Theory 4 (1980), no. 1, 151–153. A. Connes, J. Feldman, and B. Weiss, An amenable equivalence relation is generated by a single transformation, Ergod. Th. & Dynam. Sys. 1 (1981), no. 4, 431–450 (1982). A. Connes and B. Weiss, Property T and asymptotically invariant sequences, Israel J. Math. 37 (1980), no. 3, 209–210. K. Corlette and R. J. Zimmer, Superrigidity for cocycles and hyperbolic geometry, Internat. J. Math. 5 (1994), no. 3, 273–290. Y. de Cornulier, Y. Stalder, and A. Valette, Proper actions of lamplighter groups associated with free groups, C. R. Math. Acad. Sci. Paris 346 (2008), no. 3–4, 173–176 (English, with English and French summaries). M. Cowling and U. Haagerup, Completely bounded multipliers of the Fourier algebra of a simple Lie group of real rank one, Invent. Math. 96 (1989), no. 3, 507–549. M. Cowling and R. J. Zimmer, Actions of lattices in Sp(1,n), Ergod. Th. & Dynam. Sys. 9 (1989), no. 2, 221–237. H. A. Dye, On groups of measure preserving transformation. I, Amer. J. Math. 81 (1959), 119–159. ———, On groups of measure preserving transformations. II, Amer. J. Math. 85 (1963), 551–576. B. Eckmann, Introduction to l2 -methods in topology: reduced l2 -homology, harmonic chains, l2 -Betti numbers, Israel J. Math. 117 (2000), 183–219. Notes prepared by Guido Mislin. I. Epstein, Orbit inequivalent actions of non-amenable groups (2007), available at arXiv:0707.4215.

370 / a. furman [37] [38] [39] [40] [41] [42] [43] [44] [45] [46] [47] [48] [49] [50] [51] [52]

[53] [54] [55] [56] [57] [58] [59] [60] [61]

B. Farb, The quasi-isometry classiﬁcation of lattices in semisimple Lie groups, Math. Res. Lett. 4 (1997), no. 5, 705–717. B. Farb and S. Weinberger, The intrinsic asymmetry and inhomogeneity of Teichmuller space (2008), available at arXiv:0804.4428. J. Feldman and C. C. Moore, Ergodic equivalence relations, cohomology, and von Neumann algebras. I, Trans. Amer. Math. Soc. 234 (1977), no. 2, 289–324. ———, Ergodic equivalence relations, cohomology, and von Neumann algebras. II, Trans. Amer. Math. Soc. 234 (1977), no. 2, 325–359. J. Feldman, C. E. Sutherland, and R. J. Zimmer, Subrelations of ergodic equivalence relations, Ergod. Th. & Dynam. Sys. 9 (1989), no. 2, 239–269. D. Fisher, Groups acting on manifolds: around the Zimmer program, in this volume. D. Fisher and T. Hitchman, Cocycle superrigidity and harmonic maps with inﬁnite dimensional targets (2005), available at arXiv:math/0511666v3. A. Furman, Gromov’s measure equivalence and rigidity of higher rank lattices, Ann. of Math. (2), 150 (1999), no. 3, 1059–1081. ———, Orbit equivalence rigidity, Ann. of Math. (2) 150 (1999), no. 3, 1083–1108. ———, Mostow-margulis rigidity with locally compact targets, Geom. Funct. Anal. 11 (2001), no. 1, 30–59. ———, Outer automorphism groups of some ergodic equivalence relations, Comment. Math. Helv. 80 (2005), no. 1, 157–196. ———, On Popa’s cocycle superrigidity theorem, Int. Math. Res. Not. IMRN 19 (2007), Art. ID rnm073, 46. ———, Cohomology of measurable cocycles, in preparation. H. Furstenberg, Poisson boundaries and envelopes of discrete groups, Bull. Amer. Math. Soc. 73 (1967), 350–356. ———, Ergodic behavior of diagonal measures and a theorem of Szemerédi on arithmetic progressions, J. Analyse Math. 31 1977, 204–256. ———, Rigidity and cocycles for ergodic actions of semisimple Lie groups (after G. A. Margulis and R. Zimmer), Bourbaki Seminar, Vol. 1979/80, Lecture Notes in Math., vol. 842, Springer, Berlin, 1981, pp. 273–292. D. Gaboriau, Mercuriale de groupes et de relations, C. R. Acad. Sci. Paris Sér. I Math. 326 (1998), no. 2, 219–222. ———, Coût des relations d’équivalence et des groupes, Invent. Math. 139 (2000), no. 1, 41–98. ———, Sur la (co-)homologie L2 des actions préservant une mesure, C. R. Acad. Sci. Paris Sér. I Math. 330 (2000), no. 5, 365–370. ———, On orbit equivalence of measure preserving actions, Rigidity in dynamics and geometry (Cambridge, 2000), Springer, Berlin, 2002, pp. 167–186. ———, Invariants l 2 de relations d’équivalence et de groupes, Publ. Math. Inst. Hautes Etudes Sci. 95, (2002), 93–150. ———, Examples of groups that are measure equivalent to the free group, Ergod. Th. & Dynam. Sys. 25 (2005), no. 6, 1809–1827. D. Gaboriau and S. Popa, An uncountable family of non-orbit equivalent actions of Fn , J. Amer. Math. Soc. 18 (2005), no. 3, 547–559 (electronic). D. Gaboriau and R. Lyons, A measurable-group-theoretic solution to von Neumann’s problem (2007), available at arXiv:0711.1643. S. L. Gefter, Ergodic equivalence relation without outer automorphisms, Dopov./Dokl. Akad. Nauk Ukraïni, no. 11 (1993), 25–27.

a survey of measured group theory / 371 [62]

[63]

[64] [65] [66] [67] [68] [69] [70] [71] [72] [73] [74] [75] [76] [77] [78] [79] [80] [81] [82] [83] [84] [85]

———, Outer automorphism group of the ergodic equivalence relation generated by translations of dense subgroup of compact group on its homogeneous space, Publ. Res. Inst. Math. Sci. 32 (1996), no. 3, 517–538. ———, On cohomologies of ergodic actions of a T-group on homogeneous spaces of a compact Lie group (Russian), Operators in functional spaces and questions of function theory, Collect. Sci. Works, 1987, pp. 77–83. S. L. Gefter and V. Ya. Golodets, Fundamental groups for ergodic actions and actions with unit fundamental groups, Publ. Res. Inst. Math. Sci. 24 (1988), no. 6, 821–847 (1989). M. Gromov, Volume and bounded cohomology, Inst. Hautes Études Sci. Publ. Math. 56 (1982), 5–99 (1983). ———, Asymptotic invariants of inﬁnite groups, Geometric group theory, Vol. 2 (Sussex, 1991), 1993, pp. 1–295. P. de la Harpe, Topics in geometric group theory, Chicago Lectures in Mathematics, University of Chicago Press, Chicago, 2000. G. Hjorth, A converse to Dye’s theorem, Trans. Amer. Math. Soc. 357 (2005), no. 8, 3083–3103 (electronic). ———, A lemma for cost attained, Ann. Pure Appl. Logic 143 (2006), no. 1-3, 87–102. G. Hjorth and A. S. Kechris, Rigidity theorems for actions of product groups and countable Borel equivalence relations, Mem. Amer. Math. Soc. 177 (2005), 833, i–109. A. Ioana, A construction of non-orbit equivalent actions of F2 (2006), available at arXiv:math/0610452. ———, Cocycle superrigidity for proﬁnite actions of property (T) groups (2008), available at arXiv:0805.2998. ———, Orbit inequivalent actions for groups containing a copy of F2 (2007), available at arXiv:0701.5027. A. Ioana, A. S. Kechris, and T. Tsankov, Subequivalence relations and positive-deﬁnite functions (2008), available at arXiv:0806.0430. A. Ioana, J. Peterson, and S. Popa, Amalgamated free products of weakly rigid factors and calculation of their symmetry groups, Acta Math. 200, (2008), no. 1, 85–153. P. Jolissaint, Approximation properties for measure equivalent groups, 2001. preprint. V. F. R. Jones and K. Schmidt, Asymptotically invariant sequences and approximate ﬁniteness, Amer. J. Math. 109 (1987), no. 1, 91–114. V. A. Kaimanovich, Amenability, hyperﬁniteness, and isoperimetric inequalities, C. R. Acad. Sci. Paris Sér. I Math. 325 (1997), no. 9, 999–1004. D. A. Každan, On the connection of the dual space of a group with the structure of its closed subgroups, Funkcional. Anal. i Priložen. 1 (1967), 71–74 Russian. A. S. Kechris, Global aspects of ergodic group actions, Mathematical surveys and monographs, Vol. 160, American Mathematical Society, 2010. to appear. A. S. Kechris, B. D. Miller, Topics in orbit equivalence, Lecture Notes in Mathematics, Vol. 1852, Springer-Verlag, Berlin, 2004. A. S. Kechris and T. Tsankov, Amenable actions and almost invariant sets, Proc. Amer. Math. Soc. 136 (2008), no. 2, 687–697 (electronic). Y. Kida, The mapping class group from the viewpoint of measure equivalence theory, Mem. Amer. Math. Soc. 196 (2008), no. 916, i–190. ———, Classiﬁcation of the mapping class groups up to measure equivalence, Proc. Japan Acad. Ser. A Math. Sci. 82 (2006), no. 1, 4–7. ———, Orbit equivalence rigidity for ergodic actions of the mapping class group, Geom Dedicata 131 (2008 February), no. 1, 99–109.

372 / a. furman [86] [87] [88] [89] [90] [91] [92]

[93]

[94] [95]

[96] [97] [98] [99] [100] [101] [102]

[103] [104] [105] [106]

———, Measure equivalence rigidity of the mapping class group, arXiv math.GR (2006 July), available at arXiv:0607600. ———, Outer automorphism groups of equivalence relations for mapping class group actions, J. Lond. Math. Soc. (2) 78 (2008), no. 3, 622–638. ———, Rigidity of amalgamated free products in measure equivalence theory (2009), available at arXiv:0902.2888. M. Lackenby, Heegaard splittings, the virtually Haken conjecture and property τ (2002), available at arXiv:0205327. G. Levitt, On the cost of generating an equivalence relation, Ergod. Th. & Dynam. Sys. 15 (1995), no. 6, 1173–1181. A. Lubotzky and R. J. Zimmer, Variants of Kazhdan’s property for subgroups of semisimple groups, Israel J. Math. 66 (1989), no. 1-3, 289–299. W. Lück, L2 -invariants: theory and applications to geometry and K -theory, Ergebnisse der Mathematik und ihrer Grenzgebiete. 3. Folge. A series of modern surveys in mathematics [Results in Mathematics and Related Areas. 3rd Series. A Series of Modern Surveys in Mathematics], Vol. 44, Springer-Verlag, Berlin, 2002. G. A. Margulis, Discrete groups of motions of manifolds of nonpositive curvature, Proceedings of the International Congress of Mathematicians (Vancouver, BC, 1974), Vol. 2, Canad. Math. Congress, Montreal, Que. bec, 1975, pp. 21–34 (Russian). ———, Factor groups of discrete subgroups and measure theory, Funktsional. Anal. i Prilozhen. 12 (1978), no. 4, 64–76 (Russian). ———, Discrete subgroups of semisimple Lie groups, Ergebnisse der Mathematik und ihrer Grenzgebiete (3) [Results in Mathematics and Related Areas (3)], Vol. 17 SpringerVerlag, Berlin, 1991. I. Mineyev, N. Monod, and Y. Shalom, Ideal bicombings for hyperbolic groups and applications, Topology 43 (2004), no. 6, 1319–1344. N. Monod, Continuous bounded cohomology of locally compact groups, Lecture Notes in Mathematics, Vol. 1758, Springer-Verlag, Berlin, 2001. ———, An invitation to bounded cohomology, International Congress of Mathematicians. Vol. II, Eur. Math. Soc., Zürich, 2006, pp. 1183–1211. N. Monod and Y. Shalom, Negative curvature from a cohomological viewpoint and cocycle superrigidity, C. R. Math. Acad. Sci. Paris 337 (2003), no. 10, 635–638. ———, Cocycle superrigidity and bounded cohomology for negatively curved spaces, J. Differential Geom. 67 (2004), no. 3, 395–455. ———, Orbit equivalence rigidity and bounded cohomology, Ann. of Math. (2) 164 (2006), no. 3, 825–878. C. C. Moore, Ergodic theory and von Neumann algebras, Operator algebras and applications, Part 2, Kingston, Ontario, 1980), Proc. Sympos. Pure Math., Vol. 38, Amer. Math. Soc., Providence, RI, 1982, pp. 179–226. G. D. Mostow, Strong rigidity of locally symmetric spaces, Princeton University Press, Princeton, NJ, 1973. Annals of Mathematics Studies, no. 78. F. J. Murray and J. von Neumann, On rings of operators. IV, Ann. of Math. (2) 44 (1943), 716–808. D. S. Ornstein and B. Weiss, Ergodic theory of amenable group actions. I. The Rohlin lemma, Bull. Amer. Math. Soc. (N.S.) 2 (1980), no. 1, 161–164. S. Popa, Correspondences, Prépublication Institul Natinoal Pentru Creatie Stiintica si Tehnica (1986).

a survey of measured group theory / 373 [107] N. Ozawa, A Kurosh-type theorem for type II1 factors, Int. Math. Res. Not., (2006), Art. ID 97560, 21. [108] N. Ozawa and S. Popa, On a class of II1 factors with at most one Cartan subalgebra II (2008), available at arXiv:0807.4270v2. [109] S. Popa, On a class of type II1 factors with Betti numbers invariants, Ann. of Math. (2) 163 (2006), no. 3, 809–899. [110] ———, Some computations of 1-cohomology groups and construction of non-orbit-equivalent actions, J. Inst. Math. Jussieu 5 (2006), no. 2, 309–332. [111] ———, Strong rigidity of II1 factors arising from malleable actions of w-rigid groups. I, Invent. Math. 165 (2006), no. 2, 369–408. [112] ———, Strong rigidity of II1 factors arising from malleable actions of w-rigid groups. II, Invent. Math. 165 (2006), no. 2, 409–451. [113] ———, Cocycle and orbit equivalence superrigidity for malleable actions of w-rigid groups, Invent. Math. 170 (2007), no. 2, 243–295. [114] ———, Deformation and rigidity for group actions and von Neumann algebras, International Congress of Mathematicians. Vol. I, Eur. Math. Soc., Zürich, 2007, pp. 445–477. [115] ———, On the superrigidity of malleable actions with spectral gap, J. Amer. Math. Soc. 21 (2008), no. 4, 981–1000. [116] S. Popa and R. Sasyk, On the cohomology of Bernoulli actions, Ergod. Th. & Dynam. Sys. 27 (2007), no. 1, 241–251. [117] S. Popa and S. Vaes, Strong rigidity of generalized Bernoulli actions and computations of their symmetry groups, Adv. Math. 217 (2008), no. 2, 833–872. [118] ———, Actions of F∞ whose II1 factors and orbit equivalence relations have prescribed fundamental group (2008), available at arXiv:0803.3351. [119] ———, Cocycle and orbit superrigidity for lattices in SL (n, R) acting on homogeneous spaces, in this volume. [120] ———, Group measure space decomposition of II1 factors and W ∗ -superrigidity, available at arXiv:0906.2765. [121] M. Ratner, Interactions between ergodic theory, Lie groups, and number theory, 2, (Zürich, 1994), Birkhäuser, Basel, 1995, pp. 157–182. [122] H. Sako, The class S is an ME invariant, Int. Math. Res. Not. (2009), available at arXiv:0901.3374. rnp025. [123] ———, Measure equivalence rigidity and bi-exactness of groups, J. Funct. Anal. (2009), available at arXiv:0901.3376. to appear. [124] R. Sauer, L2 -Betti numbers of discrete measured groupoids, Internat. J. Algebra Comput. 15 (2005), 5-6, 1169–1188. [125] ———, Homological invariants and quasi-isometry, Geom. Funct. Anal. 16 (2006), no. 2, 476–515. [126] ———, Amenable covers, volume and l 2 -Betti numbers of aspherical manifolds (2006), available at arXiv:math/0605627. [127] R. Sauer and A. Thom, A Hochschild-Serre spectral sequence for extensions of discrete measured groupoids (2007), available at arXiv:0707.0906. [128] K. Schmidt, Asymptotically invariant sequences and an action of SL(2, Z) on the 2-sphere, Israel J. Math. 37 (1980), no. 3, 193–208. [129] ———, Amenability, Kazhdan’s property T , strong ergodicity and invariant means for ergodic group-actions, Ergod. Th. & Dynam. Sys. 1 (1981), no. 2, 223–236.

374 / a. furman [130] Y. Shalom, Rigidity of commensurators and irreducible lattices, Invent. Math. 141 (2000), no. 1, 1–54. [131] ———, Rigidity, unitary representations of semisimple groups, and fundamental groups of manifolds with rank one transformation group, Ann. of Math. (2) 152 (2000), no. 1, 113–182. [132] ———, Harmonic analysis, cohomology, and the large-scale geometry of amenable groups, Acta Math. 192 (2004), no. 2, 119–185. [133] ———, Measurable group theory, European Congress of Mathematics, Eur. Math. Soc., Zürich, 2005, pp. 391–423. [134] I. M. Singer, Automorphisms of ﬁnite factors, Amer. J. Math. 77 (1955), 117–133. [135] G. Stuck and R. J. Zimmer, Stabilizers for ergodic actions of higher rank semisimple groups, Ann. of Math. (2) 139 (1994), no. 3, 723–747. [136] S. Vaes, Rigidity results for Bernoulli actions and their von Neumann algebras (after Sorin Popa), Astérisque 311 (2007), Exp. No. 961, viii, 237–294. Séminaire Bourbaki. Vol. 2005/2006. [137] R. J. Zimmer, Extensions of ergodic group actions, Illinois J. Math. 20 (1976), no. 3, 373–409. [138] ———, Amenable ergodic group actions and an application to Poisson boundaries of random walks, J. Functional Analysis 27 (1978), no. 3, 350–372. [139] ———, Strong rigidity for ergodic actions of semisimple Lie groups, Ann. of Math. (2) 112 (1980), no. 3, 511–529. [140] ———, Orbit equivalence and rigidity of ergodic actions of Lie groups, Ergod. Th. & Dynam. Sys. 1 (1981), no. 2, 237–253. [141] ———, On the cohomology of ergodic actions of semisimple Lie groups and discrete subgroups, Amer. J. Math. 103 (1981), no. 5, 937–951. [142] ———, Ergodic theory, semisimple Lie groups, and foliations by manifolds of negative curvature, Inst. Hautes Études Sci. Publ. Math. 55, (1982), 37–62. [143] ———, Ergodic theory, group representations, and rigidity, Bull. Amer. Math. Soc. (N.S.) 6 (1982), no. 3, 383–416. [144] ———, Ergodic actions of semisimple groups and product relations, Ann. of Math. (2) 118 (1983), no. 1, 9–19. [145] ———, Kazhdan groups acting on compact manifolds, Invent. Math. 75 (1984), no. 3, 425–436. [146] ———, Ergodic theory and semisimple groups, Monographs in Mathematics, Vol. 81, Birkhäuser Verlag, Basel, 1984. [147] ———, Groups generating transversals to semisimple Lie group actions, Israel J. Math. 73 1991, no. 2, 151–159.

11

alessandra iozzi

ON RELATIVE PROPERTY (T)

to bob zimmer on his

60 th

birthday

Abstract We present families of pairs (H A,A) with relative property (T), where H is a locally compact group acting continuously by automorphisms on a locally compact abelian group A. The paper is completely self-contained.

1. Introduction Property (T) for groups and for pairs (also referred to as relative property (T)) was ﬁrst introduced by D. Kazhdan in his thesis [14] to show that lattices in semisimple Lie groups are ﬁnitely generated. His proof that SL(n, R) has property (T) for n ≥ 3 follows from the fact that the pair (SL(2, R) R2 , R2 ) has the relative property (T). Recall that a unitary representation π : G → U (H) of a locally compact group G almost has invariant vectors if for every compact subset K ⊂ G and for every > 0 there exists ξ ∈ H with ξ = 1 such that sup π (g)ξ − ξ < . g∈K

Equivalently, one says that the representation π weakly contains the trivial representation 11G . Then a group G has property (T) if any representation that almost has invariant vectors actually has invariant vectors. A typical example of a group with property (T) is a connected semisimple real Lie group all of whose factors have real rank at least 2, for example SL(n, R) for n ≥ 3, or any lattice therein.

definition 1.1. Let L < G be a subgroup of a locally compact group G. The pair (G, L) has relative property (T) if whenever (π , H) is a continuous unitary representation of G, which almost has invariant vectors; there are nonzero L-invariant vectors. Partially supported by FNS grant PP002-102765.

375

376 / a. iozzi

Property (T) has several diverse applications in representation theory, ergodic theory, operator algebras, lattices in algebraic groups over local ﬁelds, geometric group theory, and the theory of networks. Aside the implicit use of relative property (T) in [14], among its ﬁrst applications there is the construction of an inﬁnite family of expanders [15], or the solution of Ruziewicz problem for Rn , [16], both due to Margulis. Further applications include Gaboriau and Popa’s construction [11] of a noncountable family of free ergodic measure-preserving non-orbit equivalent actions of the free group Fr in r ≥ 2 generators on a standard probability space (see also [21] or [10] for generalizations) or Popa’s construction [20, 22, 23] of a factor of type II1 with trivial fundamental group. Finally, although the fact that a pair (G, L) has relative property (T) is a weaker condition than requiring that the group G itself has property (T), in some situations relative property (T) will sufﬁce; for example, Navas [19] has proven that for the nonexistence of “interesting” C2 -actions on the circle, the relative property (T) of a pair (, 0 ), where 0 is normal in , will sufﬁce. Just to give a ﬂavor of how relative property (T) differs from property (T), although we will not use these facts here, observe that if (G, L) has relative property (T) this does not imply that either G or L have property (T), and moreover, (SL(2, Z) Z2 , Z2 ) has relative property (T) but neither SL(2, Z) nor Z2 have property (T). Furthermore, the fact that (G, L) has relative property (T) does not imply that either G or L is compactly generated, as the example of (SL(3, Z) × , F∞ × {e}), where is any group that is not ﬁnitely generated shows.1 Just as for property (T), several different characterizations of relative property (T) are available, for example, in terms of strongly ergodic actions, in terms of Von Neumann algebras, of positive deﬁnite functions, of isometric actions of Hilbert spaces, and so on; moreover, it is possible to deﬁne relative property (T) for pairs (G, X ) where X is any subset [6]. We refer the reader to [13] and [6] and to the references therein. It goes without saying that if either G or L is a group with property (T), then the pair (G, L) has the relative property (T); nontrivial examples of pairs with relative property (T) have been constructed in [5], in [8] (see Theorem 1.3 and the comments thereafter) or in [25] (see the last paragraph of this introduction). The scope of this note is to present families of pairs (G, L) with relative property (T) whose general form is the semidirect product G = H A of a 1. Note, however, that if (G, L) has relative property (T), there exists a compactly generated subgroup H containing L such that (H, L) has relative property (T); see [6].

on relative property (t) / 377

locally compact group H acting continuously by automorphisms on a locally compact abelian group A and L = A. 1.1.

Let be a discrete group, S a ﬁnite set of prime numbers, Z[S] the ring obtained inverting the primes in S and ρ : → GLN (Z[S]) a homomorphism. We will denote by ρp : → GLN (Qp ) the representation obtained by injecting Z[S] into Qp , where Q∞ := R.

theorem 1.2. Assume that the image ρ() of is Zariski dense in SLN . Then the following are equivalent: i) the pair ( Z[S]N , Z[S]N ) has relative property (T); ii) the Z[]-module Z[S]N is ﬁnitely generated; iii) for every p ∈ S ∪ {∞} the image ρp () < SLN (Qp ) is not bounded; and ∗ iv) there is no ρp∗ ()-invariant probability measure on P((QN p ) ), for p ∈ S ∪ {∞}. We remark that (ii) is in contrast with the fact that, unless S = ∅, the ring Z[S] is not ﬁnitely generated as a Z-module; moreover, it is remarkable that although the relative property (T) is analytic in nature, the equivalent property in (ii) above is purely algebraic. As it will be clear from the proof (see Proposition 4.5) Zariski density is not needed for the equivalence of the assertions (i) and (iv). The motivation to establish results of the type of Theorem 1.2 comes from [8], where the following is shown:

theorem 1.3. (fernós, [8]) Let be a ﬁnitely generated group. Then the following are equivalent: i) there exists a representation ρ : → SLn (R) such that the real points Z ρ() (R) of the Zariski closure of the image of ρ is not an amenable group; and ii)there exists a ﬁnite set of primes S, an integer N ∈ N, and a representation → SLN (Z[S]) such that ( Z[S]N , Z[S]N ) has relative property (T).

example 1.4. 1) According to the Tits alternative, if k is any local ﬁeld of zero characteristic and < GL(n, k) is a ﬁnitely generated subgroup, then either is

378 / a. iozzi

virtually solvable or it contains a free subgroup F2 in two generators that is Zariski dense in . Since = SLN (Z) is not virtually solvable, then there is a Zariski-dense F2 < SLN (Z) and thus (F2 ZN , ZN ) has relative property (T). 2) A topological version of the Tits alternative shown by Breuillard and Gelander [4] asserts that if ki are local ﬁelds of zero characteristic and < GL(n, ki ) is a ﬁnitely generated subgroup, then either contains an open solvable subgroup, or a ﬁnitely generated free subgroup that is dense in . Using this result we show the following:

corollary 1.5. For every natural number N ∈ N and every nonempty ﬁnite set of primes S, there is a ﬁnitely generated free group < SLN (Z[S]) such that the pair ( Z[S]N , Z[S]N ) has relative property (T). 1.2.

Now let ρ : G → GL(V ) be a rational representation deﬁned over Q, where G is a connected, semisimple Q-group. Let ρR : GR → GL(VR ) be the representation on the level of real points. We say that ρR is totally unbounded if for any GR -invariant subspace WR ⊂ VR of positive dimension, the group ρ(GR )|WR < GL(WR ) is unbounded.

theorem 1.6. Let ⊂ VQ be a Z-module of maximal rank invariant under GZ . Let < GZ and assume that is Zariski dense in G. The following are equivalent: i) the pair ( , ) has relative property (T); and ii) the representation ρR : GR → GL(VR ) is totally unbounded. A remark is in order regarding the hypothesis of Zariski density of in G. There is a result of Borel [3, theorem 1] to the extent that if G contains no connected normal Q-subgroups N = {e} such that NR is compact, then GZ is Zariski dense in G; this avoids the situation in which if G = G1 × G2 with Gi Q-groups, then G1,Z × G2,Z = GZ ⊂ GR = G1,R × G2,R with G2,R compact, which would prevent the Zariski density of GZ . So Theorem 1.6 applies to any ﬁnite-index subgroup of GZ , provided GR has no compact factors. If ρR is irreducible with unbounded image, it is trivially totally unbounded and hence the pair ( , ) has relative property (T). The construction of such a representation for a group deﬁned over Q, which is either adjoint or

on relative property (t) / 379

simply connected will be given in Lemma 4.7. As an application we have the following:

corollary 1.7. If G is a connected real algebraic semisimple Lie group without compact factors and < G is an arithmetic lattice, then there is a linear representation → SLN (Z) of a ﬁnite-index subgroup < such that ( ZN , ZN ) has relative property (T). Theorem 1.6 is motivated by [25, theorems 1 and 4], where the same assertion as in Corollary 1.7 is proven, under the additional hypothesis that the group G is absolutely simple. Corollary 1.7 can also be deduced from [8, theorems 3 and 7.1] by observing that such an arithmetic lattice satisﬁes property (F∞ ) in [8]. Just like in [25] the value of the integer N is explicitly given (see the proof of Lemma 4.7). Moreover, if < is of ﬁnite-index k := [, ] and ( ZN , ZN ) has relative property (T), then ( ZkN , ZkN ) has relative property (T), [8, step B in the proof of theorem 7.1].

Acknowledgments: The author thanks Marc Burger for useful conversations during the preparation of this paper and the referee for helpful comments. 2. Generalities We collect here, with proofs, few classical facts about relative property (T). We start by recording here the following observation:

lemma 2.1. Let Hi , Bi be locally compact groups such that Hi acts on Bi via a continuous action Hi → Aut (Bi ), i = 1, 2, and let q : H1 B1 → H2 B2 be a continuous homomorphism such that q(B1 ) = B2 . If (H1 B1 , B1 ) has relative property (T), then also (H2 B2 , B2 ) has relative property (T). lemma 2.2. Let H, A1 , A2 , . . . , An be locally compact groups such that H acts on Aj via a continuous action H → Aut (Aj ). The following are equivalent: i) the pair (H (A1 × · · · × An ), A1 × · · · × An ) has relative property (T); and ii) the pair (H Aj , Aj ) has relative property (T) for j = 1, . . . , n. Proof. The implication (i)⇒(ii) follows from Lemma 2.1 by considering the homomorphisms H (A1 × · · · × An ) → H Aj , while (ii)⇒(i) is in [8, lemma 5.2], where it is used that relative property (T) is a property closed under certain extensions.

380 / a. iozzi

The following corollary is essential in the proof of Proposition 4.5.

corollary 2.3. Let H and M be locally compact groups, < M a cocompact lattice and let ρ : H → Aut (M) be a continuous action by automorphisms preserving . Then the pair (H , ) has relative property (T) if and only if the same holds for the pair (H M, M). Although the above result is all we need, we are going to prove the following more general result from which the corollary can be obtained at once by setting L := H and A := M H M =: G.

proposition 2.4. Let G be a locally compact group, A G a normal subgroup, and L < G a closed subgroup. 1) If (G, A) has relative property (T) and L\G has a ﬁnite G-invariant measure, then the pair (L, L ∩ A) also has relative property (T). 2) If A/L ∩ A is compact with ﬁnite A-invariant measure and (L, L ∩ A) has relative property (T), then (G, A) has relative property (T). In order to prove this, we will use a strengthening of relative property (T). We recall ﬁrst that if π : G → U (H) is a unitary representation of a locally compact group G, we say that a sequence {ξn } ⊂ H of vectors is asymptotically π(G)-invariant if ξn = 1 for all n and for every compact subset K ⊂ G we have that 2.1

lim sup π (k)ξn − ξn = 0 .

n→∞

k∈K

If a group is σ -compact, a representation almost has invariant vectors if and only if it has a sequence of asymptotically invariant vectors. Denote by HG the subspace of H consisting of π (G)-invariant vectors. Then it is well known that if G is a group with property (T) and π : G → U (H) is a unitary representation of G with a sequence {ξn } of asymptotically π(G)-invariant vectors, then lim d(ξn , HG ) = 0,

n→∞

Where d denotes the distance to a subspace. Similarly one has the following:

lemma 2.5. Let A G be a normal subgroup of a locally compact group G such that the pair (G, A) has relative property (T). If π : G → U (H) is a continuous

on relative property (t) / 381

unitary representation and {ξn } ⊂ H is a sequence of asymptotically π (G)-invariant vectors, then lim d(ξn , HA ) = 0 .

n→∞

Proof. Since A is normal in G, the Hilbert subspace HA is π (G)-invariant, so that H = HA ⊕ (HA )⊥

is an orthogonal decomposition into π (G)-invariant subspaces. Let {ξn } ⊂ H be a sequence of asymptotically invariant vectors and let ξn = ζn + ζn be the corresponding orthogonal decomposition, so that showing that limn→∞ d(ξn , HA ) = 0 is equivalent to showing that limn→∞ ζn = 0. Let us assume by contradiction that lim supn→∞ ζn = 0, that is, for some > 0, let {ζn k } ∈ (HA )⊥ be a subsequence such that 2.2

for all k ∈ N we have ζn k ≥ .

Since (HA )⊥ is π(G)-invariant and the orthogonal projection is norm decreasing, we have that for all g ∈ G π (g)ζn k − ζn k ≤ π (g)ξnk − ξnk from which, using Equation 2.2, we obtain that ζnk ζn k 1 − ≤ π (g)ξnk − ξnk . π(g) ζnk ζnk If now K ⊂ G is any compact set, since the sequence {ξn } ⊂ H is asymptotically π(G)-invariant, we have ζnk ζn k 1 lim sup π(g) − ≤ lim sup π (g)ξnk − ξnk = 0 , ζnk ζnk k→∞ k∈K k→∞ k∈K from which we deduce that the sequence {ζn k /ζn k } is also asymptotically π (G)-invariant. Hence π (G)|(HA )⊥ has non-zero π (A)-invariant vectors, which is a contradiction.

382 / a. iozzi

Proof of Proposition 2.4. 1) Let (π, H) be a representation of L that almost has invariant vectors. Since weak containment is preserved under induction, the representation ω := G IndG L (π ) of G induced from π weakly contains the representation IndL (11L ) induced to G from the identity representation of L; the latter in turn has invariant vectors since L\G has ﬁnite G-invariant measure and hence, by transitivity of weak containment, ω almost has invariant vectors. Thus, since (G, A) has relative property (T), the Hilbert space Lω := {f : G → H : f is measurable, f (g) = π ()(f (g)),

for almost every g ∈ G and for all ∈ L f (g)2 dμ(g) < ∞} and L\G

of the representation ω has nonzero A-invariant vectors with respect to the action (ω(g)f )(x) = f (xg) for all g ∈ G and almost all x ∈ G. Let f ∈ Lω be such nonzero A-invariant vector. Then for all a ∈ A ∩ L and almost all x ∈ G we have π (a)(f (x)) = f (ax) = f (x(x −1 ax)) = f (x) , which shows that the space of (L ∩ A)-invariant vectors in H is not trivial and hence shows (1). 2) Let π : G → U (H) be a representation of G and {ξn } ∈ H be a sequence of asymptotically π (G)-invariant vectors.2 Then {ξn } is also asymptotically π (L)invariant and since by hypothesis (L, L ∩ A) has relative property (T) and L ∩ A is normal in L, by Lemma 2.5 there exists a sequence {ζn } ∈ HL∩A such that limn→∞ ξn − ζn = 0. After rescaling if necessary, we may assume that ζn = 1. For all n ∈ N the vectors ηn :=

A/A∩L

π (a)ζn dλ(a) ∈ H ,

where λ is the A-invariant probability measure on A/A ∩ L, are obviously π(A)invariant, and we only need to show that there exists an n ∈ N such that ηn = 0. 2. For the sake of simplicity we assume in the sequel that all locally compact groups are σ -compact. If not, the same arguments apply by replacing sequences with generalized sequences.

on relative property (t) / 383

To see this, let F ⊂ A ⊂ G be a compact fundamental domain for the action of A ∩ L on A. Then for all a ∈ A there exists fa ∈ F and a ∈ A ∩ L such that a = fa a , so that, by π(L ∩ A)-invariance of ζn , ηn − ζn =

A/A∩L

π (a)ζn dλ(a) − ζn

≤

A/A∩L

π (a)ζn − ζn dλ(a)

≤ sup π (fa )ζn − ζn . fa ∈F

Now we observe that {ζn } is asymptotically π (G)-invariant: indeed for all g ∈ G, π (g)ζn − ζn ≤ 2ζn − ξn + π (g)ξn − ξn . It follows that limn→∞ ηn − ζn = 0 and thus in particular ηn = 0 for some n ∈ N.

lemma 2.6. Let A be a semidirect product of discrete groups, with A abelian, and assume that ( A, A) has relative property (T). Then the Z[]-module A is ﬁnitely generated. Proof. Let An ↑ A be an increasing sequence of -invariant subgroups ﬁnitely generated over Z[] and ∪An = A. Consider the A-regular action on

2 ( A/ An ) ;

n≥1

the sequence δe( An ) is asymptotically invariant, hence there is an A-invariant vector in some 2 ( A/ An ); this implies that there is a ﬁnite A-orbit in ( A/ An ) ∼ = A/An and hence |A/An | < +∞.

3. Algebraic Actions and Measure Theory We start by recalling the classical fact that if k is a local ﬁeld and Hk consists of the k-points of an algebraic group deﬁned over k, then any k-algebraic action of Hk on the k-points of a k-variety has orbits, which are locally closed in the Hausdorff topology [2]. Recall, moreover, that an action is almost effective if the intersection of the stabilizers is a ﬁnite group.

384 / a. iozzi

At the heart of what we are doing is the following theorem of Gromov: the proof we recall here is more elementary than the original one and appears in [1, théorème 6.5].

theorem 3.1. (gromov) Let k be a local ﬁeld, H a k-algebraic group, W a k-algebraic variety and H × W → W a k-algebraic action. Let μ ∈ M1 (Wk ) be a probability measure on Wk such that the Zariski closure of its support is W , and assume that the action is almost effective. Then StabHk (μ) := {h ∈ Hk : h∗ μ = μ} is compact. Z

Z

Proof. Observe ﬁrst of all that since supp (μ) ⊂ Wk , then supp (μ) ⊂ Wk , which, together with the hypothesis on the support of μ implies that Wk is Zariski dense in W . This and the fact that the action is almost effective imply that 3.1

StabHk (x) =

x∈Wk

StabHk (x) is ﬁnite .

x∈W

The Noetherian property for the k-algebraic group StabHk (x) implies that there exist x1 , . . . , xn ∈ Wk such that 3.2

n j=1

StabHk (xj ) =

StabHk (x) .

x∈Wk

If we now let H act diagonally on W n and deﬁne O := {p ∈ W n : StabHk (p) is ﬁnite} ,

we deduce immediately from Equations 3.1 and 3.2 that O is not empty. Likewise it is easy to see that O is Zariski open and hence, if ν := μ⊗n , ν(Ok ) > 0. Since, moreover, Ok is Hk -invariant, then we conclude that ν|Ok is a ﬁnite measure left-invariant by StabHk (μ). Since Ok is open in WkN and the Hk -orbits in Wk are locally closed, then also the StabHk (μ)-orbits in Ok are locally closed in the Hausdorff topology. By decomposing ν|Ok if necessary into ergodic components and applying general considerations (see [26] or [12, 7]), we deduce that the ﬁnite StabHk

on relative property (t) / 385

(μ)-invariant measure on Ok is supported on an orbit StabHk (μ) · p0 , with p0 ∈ Ok . But StabHk (μ) · p0 ∼ = StabHk (μ)/ StabHk (p0 ) ∩ StabHk (μ) and, by hypothesis, StabHk (p0 ) is ﬁnite. It follows that StabHk (μ) supports a ﬁnite measure that is invariant by translations and hence is compact. We can draw from this theorem some useful consequences on the structure of the stabilizer of a probability measure in projective space. The point is to be able to deal with measures whose support is not necessarily Zariski dense and with actions that are not necessarily effective. To apply the previous theorem and for further reference, let us set up some notation. If k is a local ﬁeld, let E be a ﬁnite-dimensional vector space deﬁned over k. If μ ∈ M1 (P(Ek )), let us set 3.3

W := supp μZ ⊆ P(E)

to be the projective subvariety deﬁned as the Zariski closure of the support of μ and deﬁne the algebraic subgroups 3.4

N(W ) :={h ∈ PGL(E) : h(W ) = W } I(W ) :={h ∈ PGL(E) : h|W = Id W } ,

where I(W ) is normal in N(W ) and both are deﬁned over k. The above notation will be typically applied when E is either a vector space as in the next corollary or the dual of a vector space (the only difference of course is lying in the action).

corollary 3.2. Let Vk be a ﬁnite-dimensional k-vector space seen as the set of k-points of the corresponding vector space V over an algebraic closure of k. The stabilizer StabPGL(Vk ) (μ) of a probability measure μ ∈ M1 (P(Vk )) on the projective space P(Vk ) has a cocompact normal k-subgroup. More precisely, if W ⊂ P(V ) and I(W ) are deﬁned as in Equations (3.3) and (3.4), respectively, then I(W ) is a ksubgroup of PGL(V ) such that I(W )k is normal in StabPGL(Vk ) (μ) and the quotient StabPGL(Vk ) (μ)/I(W )k is compact. Proof. By construction the group H := N(W )/I(W ) is deﬁned over k and so is the algebraic action H × W → W , which is in addition effective by construction. Since N(W )k /I(W )k is the N(W )k -orbit of the identity in (N(W )/I(W ))k = Hk , it is locally closed in the Hausdorff topology and, being a topological group,

386 / a. iozzi

is also closed; but StabPGL(Vk ) (μ)/I(W )k being closed in N(W )k /I(W )k it is also closed in Hk . Since the quotient StabPGL(Vk ) (μ)/I(W )k is closed and contained in the compact StabHk (μ), it is itself compact.

remark 3.3. Furstenberg’s lemma [9] asserts that either the stabilizer of a probability measure μ on P(Vk ) is compact or the support of the measure is contained in the union of two proper subspaces. Using this lemma Zimmer has shown [26, theorem 3.2.4] (see also [18]) that under the same hypotheses as in Corollary 3.2 StabPGL(Vk ) (μ) has a normal subgroup of ﬁnite index that contains a cocompact normal subgroup consisting of the k-points of a k-algebraic group. 4. Applications to Group Pairs Let Vk be as in §3 a k-vector space identiﬁed with the k-points of a vector space V over an algebraic closure of k, and let Vk∗ be its dual. If μ ∈ M1 (P(Vk∗ )), deﬁne 4.1

Gμ := {h ∈ GL(Vk ) : [h ∗ ] ∈ StabPGL(Vk∗ ) (μ)}

where h ∗ denotes the transpose of h. Then we have

proposition 4.1. Let μ ∈ M1 (P(Vk∗ )). Then the pair (Gμ Vk , Vk ) does not have relative property (T).

Proof. Let us consider W ⊂ P(V ∗ ) and I(W ) as deﬁned in Equations (3.3) and (3.4) with E = V ∗ and let us deﬁne Iμ := {h ∈ Gμ : [h ∗ ] ∈ I(W )k } = {h ∈ GL(Vk ) : [h ∗ ] ∈ I(W )k } . Let [λ] ∈ supp μ, where λ : Vk → k is a nonzero linear form. Then, by deﬁnition of Iμ , there exists a continuous homomorphism χλ : Iμ → k∗ such that for every h ∈ Iμ 4.2

h ∗ λ = χλ (h)λ.

on relative property (t) / 387

But this is equivalent to saying that the map qλ : Iμ Vk →

k∗ k

(h, v) → (χλ (h), λ(v)) is a homomorphism. Since k is not compact and k∗ k is amenable, then (k∗ k, k) does not have relative property (T) (see, for instance, [8, lemma 8.3]), and since qλ is continuous with qλ (Vk ) = k, then also (Iμ Vk , Vk ) does not have relative property (T) (Lemma 2.1). But Iμ is a normal subgroup of Gμ and Gμ /Iμ is compact by Corollary 3.2, so that Proposition 2.4(1) implies that also (Gμ Vk , Vk ) does not have relative property (T). Our source of examples is the following:

theorem 4.2. Let k be a local ﬁeld, Vk a ﬁnite-dimensional k-vector space endowed with its structure of (additive) locally compact group, H a locally compact group, and ρ : H → GL(Vk ) a continuous representation. Then the following are equivalent: i) the pair (H Vk , Vk ) has relative property (T); and ii) there is no H-invariant probability measure on P(Vk∗ ). The proof of the implication (ii)⇒(i) was proven in [5, proposition 7 and examples] and is recalled below for completeness. The case in which k = R and H is a semisimple connected Lie group was established in [24, proposition 2.3]. The implication (i)⇒(ii) in the general case appears in [6, proposition 3.1.9], but it follows here immediately from the above proposition. In fact, assume that via the contragredient action ρ ∗ of ρ, the group H ﬁxes a measure μ ∈ M1 (P(Vk∗ )). With Gμ as in Equation 4.1, consider the continuous homomorphism H Vk

ρ⊗Id

/ Gμ Vk .

Then Proposition 4.1 and the contrapositive of Lemma 2.1 imply that (H Vk , Vk ) does not have relative property (T). We recall now the proof of the implication (ii)⇒(i). Let ρ : H → GL(Vk ) be a continuous representation and let G := H Vk . By ﬁxing a nontrivial character χ ∈ Homc (k, T) we get the usual isomorphism as locally compact

388 / a. iozzi

k of Vk with the dual V ∗ , via groups of the Pontryagin dual V k k V

Vk∗ −→

λ −→ χ ◦ λ . Given a continuous unitary representation π : G → U (H), we have the spectral measure P : B(Vk∗ ) → P (H) of π|Vk , that is, a map associating to every Borel set B ⊂ Vk∗ an orthogonal projection satisfying certain additional properties. Essential is the fact that for every ξ ∈ H, μξ (B) := P(B)ξ , ξ deﬁnes a bounded positive Radon measure μξ ∈ M+ (Vk∗ ) on Vk∗ and π (v)ξ , ξ = μˆ ξ (v) = χ (λ(v)) dμξ (λ) . Vk∗

In other words, μξ is determined uniquely by the diagonal coefﬁcients on Vk associated to ξ and it is easy to see that P(B) is uniquely determined by the map ξ → μξ (B); furthermore, we have for all h ∈ H and B ∈ B(Vk∗ ) P(ρ(h)∗ B) = π(h)−1 P(B) π(h) .

4.3

Now let ξ ∈ H and μξ ∈ M+ (Vk∗ ) and deﬁne mξ := p∗ (μξ |Vk∗ \{0} ) to be the push- forward of the measure μξ |Vk∗ \{0} under the projection p : Vk∗ \ {0} → P(Vk∗ ). If B ∈ B(P(Vk∗ )) is a Borel subset in P(Vk∗ ) and we set B = p−1 (B), using Equation 4.3 we have mξ (ρ ∗ (h)B) − mξ (B) = P(ρ ∗ (h)B )ξ , ξ − P(B )ξ , ξ = P(B )π (h)ξ , π(h)ξ − P(B )ξ , ξ = Re P(B )(π(h)ξ + ξ ), π (h)ξ − ξ , so that |mξ (ρ ∗ (h)B) − mξ (B))| ≤ 2ξ π(h)ξ − ξ . Now for any subset K ⊂ H, introduce the following quantity α(K , ρ) :=

inf

sup |m(ρ ∗ (h)B) − m(B)|

sup

m∈M1 (P(Vk∗ )) B∈B(P(V ∗ )) h∈K k

on relative property (t) / 389

which is somehow the extent to which “measures are moved by ρ ∗ (h).” Since the total mass of the positive measure mξ is μξ (P(Vk∗ )) = μξ (Vk∗ ) − μξ ({0}) , and since there are no ρ(Vk )-invariant vectors if and only if μξ ({0}) = 0 for every ξ = 1, we obtain the following:

proposition 4.3. Let ρ : G → U (H) be a continuous unitary representation of H Vk with no nonzero Vk -invariant vectors. Then for every compact subset K ⊂ H and for every ξ ∈ H with ξ = 1, we have that max ρ(h)ξ − ξ ≥ h∈K

1 α(K , ρ) . 2

The proof of Theorem 4.2(ii)⇒(i) will be complete if we show that H has no invariant measure on P(Vk∗ ) via ρ if and only if there exists a compact set K ⊂ H with α(K , ρ) > 0; but this is just an exercise. 4.1. Proof of the Results in §1.1

We adopt the same notation as in §1.1, namely, is a discrete group, S a ﬁnite set of primes, Z[S] the ring obtained inverting the primes in S, ρ : → GLN (Z[S]) a representation, and ρp : → GLN (Qp ) is the representation obtained by composing ρ with the embedding Z[S] → Qp , where Q∞ := R. Since the diagonal embedding

Z[S]N → RN × QN p p∈S

realizes Z[S]N as a cocompact lattice, Corollary 2.3 can be translated into the following:

corollary 4.4. The pair ( Z[S]N , Z[S]N ) has relative property (T) if and N N only if the pair ( (RN × p∈S QN p ), R × p∈S Qp ) has relative property (T). From this we deduce the equivalence of (i) and (iv) in Theorem 1.2, which we record here separately as it does not require the hypothesis of Zariski density of ρ() in SLN .

proposition 4.5. With the above hypotheses, the following are equivalent: i) ( Z[S]N , Z[S]N ) has relative property (T); and

390 / a. iozzi

ii)for every p ∈ S ∪ {∞}, there is no ρp∗ ()-invariant probability measure on ∗ P((QN p ) ). Proof. Using Corollary 4.4 and Lemma 2.2 we deduce that the pair ( N Z[S]N , Z[S]N ) has relative property (T) if and only if ( QN p , Qp ) has relative property (T) for every p ∈ S ∪ {∞}. Theorem 4.2 then implies the equivalence with (ii). Proof of Theorem 1.2. The chain of implications that we shall prove is as follows: (iii)⇒(iv)⇒(i)⇒(ii)⇒(iii), where, however, the equivalence (iv)⇔(i) is the content of Proposition 4.5. (iii)⇒(iv). We shall show the contrapositive statement; namely, we shall assume that for some p ∈ S ∪ {∞} there exists a ρp∗ ()-invariant probability measure μ and we shall prove that then the image ρp () ⊂ SLN (Qp ) is bounded. N Let W ⊂ P((Qp )∗ ) and I(W ) be deﬁned as in Equations 3.3 and 3.4 with N

E = (Qp )∗ and Qp an algebraic closure of Qp , and let us consider I (W ) = I(W ) ∩ PSLN . Since I (W ) is normalized by ρ() and ρ() is Zariski dense in SLN , then I (W ) is a normal subgroup of PSLN and hence either trivial or the full group. Obviously it cannot be the full group PSLN since this cannot ﬁx pointwise any nonempty subset in projective space. So since I (W ) is trivial, then by Corollary 3.2 StabPSL(QNp ) (μ) is compact, so that ρp () is bounded. (i)⇒(ii). This follows from Lemma 2.6. (ii)⇒(iii). If p = ∞, then it is easy to see that ρ∞ () cannot be bounded because otherwise it would be conjugate into a maximal compact subgroup that is a real algebraic group, thus contradicting the Zariski density of ρ() in SLN . Let us assume that Z[S]N is ﬁnitely generated as a module over Z[], namely there exist aj ∈ Z[S]N , with 1 ≤ j ≤ r, such that Z[S]N = ρ(Z())a1 + · · · + ρ(Z())ar . For p ∈ S, ni ∈ Z, and γi ∈ , using the ultrametric inequality, we have r −∞. Then, for P a.e. ω, all ε > 0, there exist K = K (ω) and an inﬁnite number of instants n such that Sn (ω) − Sn−k (T k ω) ≥ (α − ε)k for all k, K ≤ k ≤ n. S (ω)

In particular, it follows from subadditivity that lim inf k kk ≥ α. There fore Theorem 2 follows (in the case fd P > −∞) because in that case, both

noncommutative ergodic theorems / 399

sequences Sk and −Sk are subadditive and 1 1 Sn d P = sup Sn d P = fd P = α. inf n n n n On the other hand, lim supk

Sk k

is a T -invariant function which, by sub1 p−1 S (T jk ω). Thus the constant additivity, is not bigger than lim supp pk j=0 k lim supk Skk is not bigger than 1k Sk d P. Theorem 3 follows in the case when α > −∞. To treat the case α = −∞ in both theorems, it sufﬁces to replace Sn by max (Sn , −nM), and to let M go to inﬁnity; see [Kr] for details.

2. Noncommutative Ergodic Theorems Observe that Theorem 1 also holds true for any linear operator U of a Hilbert space assuming U ≤ 1. One can take this a step further and deﬁne for any g ∈ H, φ(g) := Ug + f . Then φ is an isometry (or merely 1-Lipschitz in the case U ≤ 1). Note that φ n (0) =

n−1

U kf .

k=0

Pazy proved in [Pa] that more generally for any map φ : H → H such that φ(x) − φ(y) ≤ x − y , it holds that there is a vector v ∈ H such that 1 n φ (0) → v n in norm. This can be reformulated as follows: there is a unit speed geodesic γ (t) = tv/ v in H such that 1

1 φ n (0) − γ (n v ) = 1 φ n (0) − nv → 0. n n

We call this property ray approximation. It turns out that this generalization of the ergodic theorem still holds for more general group actions than the actions of Z. Let G be a second countable, locally compact, Hausdorff topological semigroup, and consider g : → G a measurable map. We form Zn (ω) := g(ω)g(T ω) . . . g(T n−1 ω) and ask whether Zn converges to inﬁnity with some linear speed. Assume G acts on a metric space (X , d) by 1-Lipschitz transformations. Then for a ﬁxed x0 ∈ X we can deﬁne for g ∈ G, |g| := d(x0 , gx0 ). Clearly, up to a bounded error, |Zn (ω)| does not depend on our choice of x0 . We have

400 / a. karlsson and f. ledrappier

proposition 5. Assume the transformation T is ergodic, and |g|d P < ∞. Then there is a nonnegative number α such that for P-almost every ω, as n → ∞: 1 |Zn (ω)| → α. n The number α is given by 2

α = inf n

1 n

|Zn (ω)|d P.

Proof. It sufﬁces to observe that the sequence |Zn (ω)| satisﬁes the hypotheses of Theorem 3. Our hypothesis says that Z1 < ∞. The subaditivity follows from the 1-Lipschitz property: |Zn+m (ω)| = d(x0 , g(ω) . . . g(T n+m−1 ω)x0 ) ≤ d(x0 , g(ω) . . . g(T m−1 ω)x0 )+ + d(g(ω) . . . g(T m−1 ω)x0 , g(ω) . . . g(T n+m−1 ω)x0 ) ≤ |Zm (ω)| + d(x0 , g(T m ω) . . . g(T n+m−1 ω)x0 ) = |Zm (ω)| + |Zn (T m ω)|. Moreover we see that the limit α is given by inf n

1 n

|Zn (ω)|d P.

When α > 0, Proposition 5 says that the points Zn (ω)x go to inﬁnity with a deﬁnite linear speed. The question arises of the convergence in direction of the points Zn (ω)x. Given Equation (1), we expect that an almost everywhere convergence theorem will say that Zn (ω)x will stay at a sublinear distance of a geodesic. We present several results in that direction depending on different geometric hypotheses on the space X . Assume X is a complete, Busemann nonpositively curved and uniformly convex (e.g., CAT (0) or uniformly convex Banach space) metric space. Then,

theorem 6. [KM] Under these assumptions, there is a constant α ≥ 0 and, for P-almost every ω, a geodesic ray γω such that 1 d(Zn (ω)x0 , γω (nα)) → 0. n We outline the proof (see [KM], section 5, for details). Let a(n, ω) = d(x0 , Zn (ω)x0 ) for each n. Consider a triangle consisting of x0 , Zn (ω)x0 , and Zk (ω)x0 . Note that the side of this triangle has lengths a(n, ω), a(k, ω), and (at most) a(n − k, T k ω). Given ε > 0 (and a.e. ω), for k large it holds that

noncommutative ergodic theorems / 401

a(k, ω) ≤ (α + ε)k. Assume now in addition to k being large that n and k are as in Lemma 4. This implies that the triangle is thin in the sense that Zk (ω)x0 lies close to the geodesic segment [x0 , Zn (ω)x0 ]; more precisely, the distance is at most δ(ε)a(k, ω), where δ depends only on the geometry. Thanks to the geometric assumptions this δ(ε) tends to 0 as ε tends to 0. Selecting ε tending to 0 fast enough we can, by selecting suitable n as in Lemma 4 and some simple geometric arguments, obtain a limiting geodesic. Finally, one has essentially from the construction that as m → ∞, the points Zm (ω)x0 lie at a sublinear distance from this geodesic ray. This note is devoted to the generalization of the ergodic theorem to groups of isometries of a metric space (X , d). We assume that the space (X , d) is proper (closed bounded subsets are compact) and we consider the metric compactiﬁcation of X . Deﬁne, for x ∈ X , the function x (z) on X by x (z) = d(x, z) − d(x, x0 ). The assignment x → x is continuous, injective, and takes values in a relatively compact set of functions for the topology of uniform convergence on compact subsets of X . The metric compactiﬁcation X of X is the closure of X for that topology. The metric boundary ∂X := X \ X is made of Lipschitz continuous functions h on X such that h(x0 ) = 0. Elements of ∂X are called horofunctions. Our main result is the following:

theorem 7. (ergodic theorem for isometries [KL1]) Let T be a measure-preserving transformation of the Lebesgue probability space (, A, P), G a locally compact group acting by isometries on a proper space X , and g : → G a measurable map satisfying |g(ω)|d P(ω) < ∞. Then, for P-almost every ω, there is some hω ∈ ∂X such that 1 1 lim − hω (Zn (ω)x0 ) = lim d(x0 , Zn (ω)x0 ). n→∞ n n

n→∞

For the convenience of the reader, the proof of Theorem 7 is given in Section 5. We explain in Section 3 why the convergence in Theorem 7 is equivalent to the ray approximation under the CAT (0) assumption. Note that by Theorem 7 the former convergence holds for all norms on Rd , but that Theorem 7 does not apply to inﬁnite-dimensional Banach spaces. In this case, one can use Lemma 4 to prove a noncommutative ergodic theorem with linear functionals of norm 1, somewhat analogous to horofunctions. Namely,

402 / a. karlsson and f. ledrappier

theorem 8. [Ka] Let Zn (ω) be an ergodic integrable cocycle of 1-Lipschitz selfmaps of a reﬂexive Banach space. Then for P-almost every ω there is a linear functional fω of norm 1 such that lim

n→∞

1 fω (Zn (ω)0) = α. n

On the other hand, Kohlberg-Neyman [KN] found a counterexample to the norm convergence, or more precisely to (1), for general Banach spaces.

3. Case When X Is a CAT(0) Proper Space When the space (X , d) is a proper CAT (0) metric space, both Theorems 6 and 7 apply. Because it is a direct generalization of the important case when G is a linear group, it is often called the Oseledets theorem. In this section we explain how to recover the ray approximation and other more familiar forms of Oseledets theorem from Theorem 7. Many of the geometric ideas in this section go back to Kaimanovich’s extension of Oseledets theorem to more general semisimple groups [K3]. A metric geodesic space (X , d) is called a CAT (0) space if its geodesic triangles are thinner than in the Euclidean space. Namely, consider four points A, B, C, and D ∈ X , D lying on a length-minimizing geodesic going from B to C. Draw four points A , B , C , and D in the Euclidean plane with AB = A B , BD = B D , DC = D C , and CA = C A . The space is called CAT (0) if for any such conﬁguration, AD ≤ A D . Simply connected Riemannian spaces with nonpositive curvature, locally ﬁnite trees, and Euclidean buildings are proper CAT (0) spaces. If X is a CAT (0) space, then the horofunctions h ∈ ∂X are called Busemann functions, and for any h ∈ ∂X , there is a unique geodesic ray σh (t), t ≥ 0 such that σh (0) = x0 and limt→∞ σh (t) = h. We have

corollary 9. Assume moreover that X is a CAT (0) space and that α > 0. Then, for P-almost every ω, as n goes to ∞, 1 lim d Zn (ω)x0 , σhω (αn) = 0, n n where hω is given by Theorem 7. Proof. Consider a geodesic triangle A = Zn (ω)x0 , B = x0 , Ct = σhω (t), for t very large, and choose D = σhω (nα). We want to estimate the distance AD.

noncommutative ergodic theorems / 403

We have AB = d(Zn (ω)x0 , x0 ) = |Zn (ω)| =: nαn (ω); BCt = t, BD = nα; and Ct A = t + σhω (t) (Zn (ω)x0 ) =: t − nβn (ω) + on (t). For almost every ω, we have • limn αn (ω) = α by Theorem 3; • limn βn (ω) = limn − 1 hω (Zn (ω)x0 ) = α by Theorem 7; and n • for a ﬁxed n, limt→∞ on (t) = hω (Zn (ω)x0 ) − hσ (t) (Zn (ω)x0 ) = 0. hω

Construct the comparison ﬁgure A B Ct D , and let t go to ∞. The point Et of B Ct at the same distance from Ct than A converges to the orthogonal projec of A on B C and satisﬁes B E = nβ − o (t). Therefore, B E = nβ . tion E∞ n n n t t ∞ We have 2 2 ) = n2 (αn2 − βn2 ), (D E∞ ) = n2 (βn − α)2 , (AE∞

and therefore, as n → ∞: lim n

1 2 (A D ) = lim (αn2 − βn2 ) + (βn − α)2 = 0. 2 n n

corollary 10. With the same assumptions, we have, for P-almost every ω, Zn (ω)x0 converges to hω in X . In particular, when α > 0 and X is proper CAT (0), the direction hω given by Theorem 7 is unique. Proof. In the above triangle, the geodesic σn joining x0 to Zn (ω)x0 converges to σhω . Therefore, all the accumulation points of Zn (ω)x0 belong to the set seen from x0 in the direction of hω . By the same proof, all the accumulation points of Zn (ω)x0 belong to the set seen from σhω (K ) in the direction of hω , for all K . As K goes to inﬁnity, the intersection of those sets is reduced to the point hω . In the case when G is a linear group, Corollary 9 is closely related to the well-known

theorem 11. (oseledets multiplicative ergodic theorem [O], 1968) Let T be an ergodic transformation of the probability space (, A, P), and A :

404 / a. karlsson and f. ledrappier

→ GL(d, R) a measurable map such that Then there exist

max{ln ||A||, ln ||A−1 ||}d P < ∞.

• real numbers λ1 ≤ λ2 ≤ · · · ≤ λk ; and • integers mi , i = 1, . . . , k with i mi = d, i λi mi = ln |DetA|d P, • for P-almost every ω, a ﬂag of subspaces of Rd

{0} = Vk+1 (ω) ⊂ Vk (ω) ⊂ · · · ⊂ V1 (ω) = Rd with, for all i, 1 ≤ i ≤ k, DimVi = Vi+1 (ω) if and only if as n goes to ∞, lim

j≥i

mj and a vector v belongs to Vi (ω) \

1 ln ||A(T n−1 ω)A(T n−2 ω) . . . A(ω)v|| = λi . n

Observe that, automatically, the Vi depend measurably on ω and are invariant in the sense that A(ω)Vi (ω) = Vi (T ω). The usual complete form of Oseledets theorem follows by comparing the results of Theorem 11 for (T , A) and for (T −1 , A−1 ◦ T −1 ). Fix ω ∈ , and let ei , i = 1, . . . , d be an orthogonal base of Rd such that e ∈ Vi (ω) as soon as ≤ j≥i mj . Write μ1 ≥ · · · ≥ μd for the exponents λj , each counted with multiplicity mj , and consider A(n) (ω) := A(T n−1 ω)A(T n−2 ω) . . . A(ω) in the base (ei ). To verify the statement of Theorem 11, it sufﬁces to show that for all ε > 0 and for n large enough, (n) |Ai,j (ω)| ≤ en(μi +ε) and | ln |DetA(n) (ω)| − μj | ≤ ε. j

With the notations of Section 2, consider the action by isometries of GL(d, R) on the symmetric space GL(d, R)/O(d, R) with origin x0 = O(d, R) d t 2 and distance |g| = j=1 ( ln τi ) , where τi are the eigenvalues of gg . It is t a CAT (0) geodesic proper space. Set g(ω) = A (ω). The moment hypothe sis |g|d P < ∞ is satisﬁed. We have A(n) (ω) = (Zn (ω))t . If α = 0, then the eigenvalues of Zn Znt grow subexponentially and lim n

1 1 1 ln ||A(n) (ω)v|| = lim ln (||Znt v||2 ) = 0. n 2 n n

In this case m1 = d, λ1 = 0 and Theorem 11 holds. We may assume α > 0, and apply Corollary 9. Geodesics starting from the origin are of the form etH , where H is a nonzero symmetric matrix. Therefore, for P-almost every ω, there is a nonzero symmetric matrix H(ω) such that n1 d(exp(nH(ω)), Zn (ω)) goes to 0 as n → ∞ (the constant α has been incorporated in H). In other words, n1 ln of the norm,

noncommutative ergodic theorems / 405

and of the norm of the inverse, of the matrix exp( − nH(ω))(A(n) (ω))t , go to 0 as n → ∞. We claim that this gives the conclusion of Theorem 11 with λi the eigenvalues of H(ω), mi their respective multiplicities and Vi the sums of the eigenspaces corresponding to eigenvalues smaller than λi . Indeed, we write exp(H(ω)) = K (ω)t K (ω) for K an orthogonal matrix and a diagonal matrix with diagonal entries eμi , and A(n) (ω) = Ln (ω)n (ω)Kn (ω) a Cartan decomposition of A(n) with Ln , Kn orthogonal, n a diagonal matrix with nonin(n) (n) creasing diagonal entries exp(nδ1 (ω)) ≥ · · · ≥ exp(nδd (ω)). The conclusion of Corollary 9 is therefore that, for P-almost every ω, n1 ln of the norm, and of the norm of the inverse, of the matrix n (ω)Kn (ω)K t (ω)exp( − n), go to 0 as n → ∞. It follows that, for such an ω, | ln |DetA(n) (ω)| − j μj | goes to 0 as n goes (n)

to ∞. Furthermore, for n large enough, the entries ki,j (ω) of the matrix Kn (ω)K t (ω) satisfy (n)

|ki,j (ω)| ≤ en(μj −δi (n)

+ε)

.

We have A(n) (ω)ej = Ln (ω)n (ω)Kn (ω)ej = n (ω)Kn (ω)K −1 (ω)fj , where fi is the canonical base of Rd . The components of this vector are (n)

enδi (ω) ki,j (ω). Their absolute values are indeed smaller than en(μj +ε) for n large enough. (n)

4. The Case When X Is a Gromov Hyperbolic Space (in Particular, R). Theorem 7 is due to Kaimanovich using an idea of Delzant when X is a Gromov hyperbolic geodesic space even without the condition that X is a proper space [K2]. As in the CAT (0)-case it is there formulated as Zn lies on sublinear distance of a geodesic ray. From Theorem 7 one gets the following:

corollary 12. Assume moreover that X is a Gromov hyperbolic geodesic space and that α > 0. Then, for P-almost every ω, as n goes to ∞, there is a geodesic ray σω such that 1 lim d(Zn (ω)x0 , σhω (αn)) = 0. n n Proof. Take hω given from Theorem 7. It is known, see [BH, p. 428], that for Gromov hyperbolic geodesic spaces it holds that there is a geodesic ray σω such that σω (x0 ) = 0 and

406 / a. karlsson and f. ledrappier

bω ( · ) = lim d(·, σω (t)) − t t→∞

is a horofunction such that bω ( · ) − hω ( · ) ≤ C for some constant C. This bω therefore clearly satisﬁes the conclusion of Theorem 7. Now we use the notation and setup in the proof of Corollary 9. Consider the triangle ABCt . By δ-hyperbolicity D must lie at most δ away from either AB or ACt . Call the closest point X . By the triangle inequality we must have that αn − δ ≤ XB ≤ αn + δ. If X lies on AB, then it is clear that XA = o(n) and hence AD = o(n). If X lies on ACt , then t − αn − δ ≤ XCt ≤ t − αn + δ. In view of that bω (Zn (ω)) ≈ −αn we again reach the conclusion that X , and hence also D, lie on sublinear distance from A.

corollary 13. With the same assumptions, we have that for P-almost every ω, Zn (ω)x0 converges to the point [σω ] in the hyperbolic boundary ∂hyp X . Proof. Clearly, the Gromov product (Zn (ω), σω (αn)) → ∞ as n → ∞ in view of the previous corollary. In the case when G = R and X = (R, |·| ), Corollaries 12 and 13 yield Theorem 2. Indeed, in this case the drift is

α =

fd P

and ∂ R = {h+ = +∞ (z) = −z, h− = −∞ (z) = z}. It follows from Corollary 13 that the index of hω is T invariant and is therefore almost everywhere constant. In other words, the existence of the hω with the required property amounts to the choice of the right sign: hω = sign{

f

(ω)d P(ω)}∞ .

Then, Corollary 12 says exactly that if the function f is integrable, for P-almost every ω 1 f (ω)d P(ω). Sn (ω) → n

noncommutative ergodic theorems / 407

The above observation is not a new proof of Theorem 2, because Theorem 2 is used in the proof of Theorem 7 (see Section 5). We want to only illustrate the meaning of the metric boundary on the simplest example. Nevertheless, it turns out that modifying the translation-invariant metric on X = R might have interesting consequences. The following discussion comes from [KMo], which in turn was inspired by [LL]. Let D : R≥0 → R≥0 be an increasing function, D(t) → ∞ such that D(0) = 0 and D(t)/t → 0 monotonically. From the inequality 1 1 D(t + s) ≤ D(t) t+s t we get the following subadditivity property: D(t)/t s D(s) ≤ D(t) + D(s). D(t + s) ≤ D(t) + D(t) = D(t) + t D(s)/s From all these properties of D, it follows that (R, D( |·| )) is a proper metric space, and clearly invariant under translations. Now we determine ∂ R with respect to this metric. Without loss of generality we may assume that xn → ∞. We claim that for any z h(z) = lim D(xn − z) − D(xn ) = 0. n→∞

Assume not. Then for some s > 0 and an inﬁnite sequence of t → ∞ that D(t + s) − D(t) > c > 0 (wlog). For such s and t with t large so that D(t)/t < c/s, we have D(t)

D(t) + t s D(t + s) D(t) + c D(t) ≥ ≥ = t+s t+s t+s t but this contradicts that D(t)/t is strictly decreasing. Hence ∂ R = {h ≡ 0}. Applying Theorem 7 in this setting yields a result already obtained by Aaronson with a different argument.

theorem 14. (aaronson [A]) Let f : → R such that Then, for P-almost every ω,

D(

f )dμ < ∞.

1

D Sn (ω) = 0. n→∞ n lim

Proof. It was noted above that ∂(R, D( |·| )) only consisted of h = 0. The conclusion then follows from Theorem 7 since h = 0 forces α = 0.

408 / a. karlsson and f. ledrappier

One can relax the conditions on D: for one thing, one can remove having D(0) = 0. More interestingly, the condition that D(t)/t decreases to 0 can be weakened in the following way.

corollary 15. (aaronson-weiss [A]) Let d(t) be an increasing positive function d(t) → ∞, such that d(t) = o(t), d(t + s) ≤ d(t) + d(s) and

d( f )dμ < ∞ for some function f : → R. Then, for P-almost every ω, lim

n→∞

1

d Sn (ω) = 0. n

Proof. Deﬁne D(t) = sup{d(ut)/u : u ≥ 1}. Note that this satisﬁes all the assumptions made on D in Theorem 14. Moreover, d(t) ≤ D(t) ≤ 2d(t), since if D(t) = d(tu)/u, set n = [u] + 1 and then D(u) ≤ d(nt)/u ≤ nd(t)/u ≤ 2d(t). (See [A], p. 66, for more details.) This shows that Theorem 14 actually holds for d in place of D. In particular, Corollary 15 applies to any metric d(., .) on R where balls grow superlinearly (where d(t) := d(0, t)). From this one obtains as a special case classical results like the one of Marcinkiewicz-Zygmund [MZ] and Sawyer [S]:

corollary 16. Let 0 < p < 1. If f ∈ Lp , then for P-almost every ω lim

n→∞

1 Sn = 0. n1/p

Such moment conditions arise naturally in probability theory. These results are known to be best possible in certain ways (e.g., [S] and [A]). For the iid case the converse also holds [MZ]. Another example is as follows:

corollary 17. If f is log-integrable, then for P-almost every ω lim |Sn |1/n = 1.

n→∞

One can modify the metric on any metric space X in the same way replacing d(x, y) with D(d(x, y)), where D(t) satisﬁes the assumptions for Theorem 14 or,

noncommutative ergodic theorems / 409

more generally, the assumptions for Corollary 15. By estimating a subadditive by an additive cocycle in the obvious way a(n, ω) ≤

n−1

a(1, T k ω),

k=0

and Theorem 14 implies that 1 D(d(Zn x0 , x0 )) → 0 a.e., n under the condition that D(d(g(ω)x0 , x0 )) is integrable.

5. Proof of Theorem 7 We begin by a few observations: ﬁrst, we can extend by continuity the action of G to X , and write, for h ∈ X , g ∈ G: g.h(z) = h(g −1 z) − h(g −1 x0 ). Now deﬁne the skew product action on := × X by T (ω, h) = (T ω, g(ω)−1 ·h). n

Observe that T (ω, h) = (T n ω, (Zn (ω))−1 ·h). Deﬁne the Furstenberg cocycle F(ω, h) by F(ω, h) := −h(g(ω)x0 ). We have 3

F n (ω, h) :=

n−1

i

F(T (ω, h)) = −h(Zn (ω)x0 ).

i=0

Relation (3) is proven by induction on n. We have F 1 (ω, h) := −h(g(ω)x0 ) = −h(Z1 (ω)x0 ) and F n (ω, h) = F n−1 (ω, h) + F(T

n−1

(ω, h))

= −h(Zn−1 (ω)x0 ) − (Zn−1 (ω))−1 ·h(g(T n−1 ω)x0 ) = −h(Zn−1 (ω)x0 ) − h(Zn−1 (ω)g(T n−1 (ω))x0 ) + h(Zn−1 (ω)x0 ) = −h(Zn (ω)x0 ). In particular, for any T invariant measure m on such that the projection on is P, we have Fdm ≤ α because 1 1 F(ω, h)dm(ω, h) = −h(Zn (ω)x0 )dm(ω, h) ≤ |Zn (ω)|d P(ω). n n

410 / a. karlsson and f. ledrappier

There is nothing to prove if α = 0. To prove Theorem 7 in the case α > 0, it sufﬁces to construct a T -invariant measure m on such that the projec tion on is P and such that F(ω, h)dm(ω, h) = α. Indeed, since α is the largest possible value of F, we still have the same equality for almost every ergodic component of m. By the ergodic Theorem 2, the set A of (ω, h) such that − n1 h(Zn (ω)x0 ) = n1 F n (ω, h) → α as n goes to ∞ has full measure. Moreover, observe that if h is not a point in ∂X , − n1 hγ (Zn (ω)x0 ) converges to −α. Since α > 0, this shows that A ⊂ ∂X . We get the conclusion of Theorem 7 by choosing for ω → hω a measurable section of the set A. We ﬁnally construct a measure m with those properties. We deﬁne a measure μn on ; for any measurable function on such that

sup (ω, h) d P(ω) < ∞, h∈X

we set:

×X

(ω, h)dμn (ω, h) =

(ω, Zn (ω)x0 )d P(ω).

The set of measures m on such that the projection on is P is a convex compact subset of L∞ (, P (X )) = (L1 (, C(X )))∗ for the weak* topology. The mapping m → (T )∗ m is afﬁne and continuous. We can take for m any weak* limit point of the sequence: 1 i (T )∗ μn . n n−1

ηn =

i=0

The measure m is T -invariant and, since FL1 (,C(X )) = sup |F(ω, h)|d P(ω) = sup |h(g(ω)x0 )|d P(ω) < +∞, h

h

we may write, using Relation (3) and Formula (2):

1 k→∞ nk

Fdm = lim = lim

k→∞

1 nk

1 n k→∞ k

n k −1

1 k→∞ nk

i=0

F nk (ω, Znk (ω)x0 )d P(ω)

= lim = lim

i

(F ◦ T )dμnk

( − Znk (ω)x0 (Znk (ω)x0 ))d P(ω) |Znk (ω)|d P(ω) = α.

By the above discussion this achieves the proof of Theorem 7.

noncommutative ergodic theorems / 411

Observe that by putting together the discussions in Sections 5 and 3, we obtain a proof of Oseledets Theorem 11. As proofs of Theorem 11 go, this one is in some sense rather close to the original one [O], with the somewhat simplifying use of the geometric ideas from [K3] and invariant measures as in [W].

6. Random Walks In this section we consider a probability ν on a group G and apply the preceeding analysis to the random walk Zn = g0 g1 . . . gn−1 , where the gi are independent with distribution ν. We assume • there is a proper left-invariant metric d on G that generates the topology

of G (when G is second-countable locally compact, such a metric always exists, see [St]); • d(e, g)dν(g) < + ∞ (we say that ν has a ﬁrst moment); and • the closed subgroup generated by the support of ν is the whole G (we say that ν is nondegenerate). Then there is a number (ν) ≥ 0 such that, for almost every sequence {gi }, limn n1 d(e, Zn ) = (ν). In the case when the group G is the group SL2 (R) acting on the hyperbolic plane, (ν) is twice the Lyapunov exponent of the independent product of matrices. In that case it is given by a formula involving the stationary measure on the circle, the Furstenberg-Khasminskii formula ([F1]; this appellation seems to be standard, cf. [Ar]). Seeing again the circle as the geometric boundary of the hyperbolic plane, we extend this formula to our general context:

theorem 18. (furstenberg-khasminskii formula for the linear drift, [KL2]) Let (G, ν) verify all the above assumptions, and let G be the metric compactiﬁcation of (G, d). Then there exists a measure μ on G with the following properties: • μ is stationary for the action of G, that is, μ satisﬁes μ = (g∗ μ)dν(g); and • (ν) = h(g −1 )dμ(h)dν(g). Moreover, if (ν) > 0, then μ is supported on ∂G. Proof. In the proof of Theorem 7, we constructed a measure m on × G. The measure μ can be seen as the projection on G of m, but it turns out that the measure μ can be directly constructed. Let (+ , A+ , P) be the space

412 / a. karlsson and f. ledrappier

of sequences {g0 , g1 , . . . } with product topology, σ -algebra, and measure P = ν ⊗N . For n ≥ 0, let νn be the distribution of Zn (ω) in G. In other words, deﬁne, for any continuous function f on G: fdνn = f (g0 g1 · · · gn−1 )dν(g0 )dν(g1 ) · · · dν(gn−1 ), ν0 = δe . We claim that any weak* limit μ of the measures n1 n−1 i=0 νi satisﬁes the conclusions of Theorem 18. Clearly, the measure μ is stationary: for any continuous function f on G, we have nk −1 1 f (gg0 g1 · · · gi−1 ) k→∞ nk

f (g.h)dμ(h)dν(g) = lim

i=0

× dν(g0 )dν(g1 ) · · · dν(gi−1 )dν(g) nk −1 1 fdνi+1 k→∞ nk

= lim

i=0

=

1 [ k→∞ nk

fdμ + lim

=

fdνnk − f (e)]

fdμ.

In the same way, we get h(g

−1

nk −1 1 )dμ(h)dν(g) = lim [d(Zi (ω), g −1 ) k→∞ nk i=0

− d(Zi (ω), e)]d P(ω)dν(g) 1 d(Znk , e)d P(ω) = lim k→∞ nk = (ν). This shows that the measure μ has the desired properties. Moreover, the measure P × μ on the space + × G is T -invariant. There is a unique T invariant measure m on × G that extends P × μ. The measure m satisﬁes all the properties we needed in the proof of Theorem 7. In particular, if (ν) is positive, μ(∂G) = (P × μ)(+ × ∂G) = m( × ∂G) = 1.

noncommutative ergodic theorems / 413

A bounded measurable f : G → R is ν-harmonic if f (g) = f (gh)dν(h) G

for any g ∈ G. Constant functions are obviously ν-harmonic. If f is a bounded Harmonic function, then f (Zn ) is a bounded Martingale and therefore converges almost surely. We say that (G, ν) satisﬁes the Liouville property (or (G, ν) is Liouville) if the constant functions are the only bounded ν-harmonic functions.

corollary 19. [KL2] Let G be a locally compact group with a left-invariant proper metric and ν be a nondegenerate probability measure on G with ﬁrst moment. Then, if (G, ν) is Liouville, there is a 1-Lipschitz homomorphism T : G → R such that for almost every trajectory Zn of the corresponding random walk, we have 1 lim T (Zn ) = T (g)dν(g) = l(ν). n→∞ n G Proof. The key observation is that if (G, ν) is Liouville and G acts continuously on a compact space Y , then every stationary measure μ is invariant. Indeed, for f ∈ C(Y , R), the function ϕ(g) := fd(g∗ μ) is harmonic and bounded, and therefore constant. In particular, the measure μ from Theorem 18 is invariant, and if we set T (g) := h(g −1 )μ(dh), the mapping T is Lipschitz continuous and is a group homomorphism because we have h(g −1 g −1 )μ(dh) T (g g) = = =

(g .h)(g −1 )μ(dh) +

h(g −1 )μ(dh)

h(g −1 )(g∗ μ)(dh) + T (g )

= T (g) + T (g ), where we used the invariance of μ at the last line. Finally, by the FurstenbergKhasminskii formula, we have (ν) = T (g)dμ(g).

414 / a. karlsson and f. ledrappier

A measure ν on G is called symmetric if it is invariant under the mapping g → g −1 . A measure is centered if every homomorphism of G into R is centered, meaning that the ν-weighted mean value of the image is 0. Every symmetric measure with ﬁrst moment ν is centered, since for any homomorphism T : G → R, the mean value, which is T (g)dν(g) = T (g −1 )dν(g) = − T (g)dν(g), G

G

G

must hence equal 0. By simple contraposition from Corollary 19, we get

corollary 20. [KL2] Let G be a locally compact group with a left-invariant proper metric and ν be a nondegenerate centered probability measure on G with ﬁrst moment. Then, if l(ν) > 0, there exist nonconstant bounded ν-harmonic functions. Corollary 20 was known in particular for ν with ﬁnite support ([Va], [M]) or in the continuous case, for ν with compact support and density [Al]. One case when all probability measures on G are centered is when there is no group homomorphism from G to R. We can apply Corollary 20 to a countable ﬁnitely generated group. Let S be a ﬁnite symmetric generator for G, and endow G with the left-invariant metric d(x, y) = |y −1 x| where |z| is the shortest length of an S-word representing z. We say that G has subexponential growth if limn n1 ln An = 0, where An is the number of elements z of G with |z| ≤ n. Such a group has automatically the Liouville property [Av]. This yields

corollary 21. [KL2] Let G be a ﬁnitely generated group with subexponential growth and H 1 (G, R) = 0. Then for any nondegenerate ν on G with ﬁrst moment, we have (ν) = 0. Observe that conversely, if there exists a nontrivial group homomorphism T from a ﬁnitely generated group into R, then there exists a nondegenerate probability ν on G, with ﬁrst moment and (ν) > 0. Indeed, there exists M such that [−M, M] contains all the images of the elements of the generating set S. We can choose ν carried by all the elements of S with images in [0, M]. Since T is nontrivial, T (g)dν(g) > 0. The measure ν is nondegenerate, has ﬁnite support, and (ν) > 0 since for all g ∈ G, |T (g)| ≤ M|g|.

7. Riemannian Covers In this section we consider a complete connected Riemannian manifold (M, g) with bounded sectional curvatures. In particular, if dM is the Riemannian

noncommutative ergodic theorems / 415

distance on M, (M, d) is a proper space. Associated to the metric is the LaplaceBeltrami operator . A function f is harmonic if f = 0. We say that M is Liouville if all bounded and harmonic functions are constant. Associated to is a diffusion process Bt called Brownian motion. Since the curvature is bounded and M is complete, the Brownian motion is deﬁned for all time. For all x ∈ M, there is a probability Px on C(R+ , M) such that the process Bt given by the t coordinate is a Markov process with generator and B0 = x. We can deﬁne 1 g := lim sup dM (x0 , Bt ), t→∞ t for any x0 ∈ M.

theorem 22. [KL3] Assume that (M, g) is a regular covering of a Riemannian manifold that has ﬁnite Riemannian volume and bounded sectional curvatures. Then M is Liouville if and only if lim

t→∞

1 d(x0 , Bt ) = 0 a.s. t

The “if ” part was proved by Kaimanovich, see [K1], and the converse is clear if the Brownian motion is recurrent on M. The proof of the new implication in Theorem 22 in the transient case uses the Furstenberg-Lyons-Sullivan discretization procedure. Let be the covering group of isometries of M. This discretization consists in the construction of a probability measure ν on , with the following properties: • the restriction f (γ ) := F(γ x0 ) is a one-to-one correspondence between

bounded harmonic functions on M and bounded functions on that satisfy f (γ g)ν(g); f (γ ) = g∈

• if γ1 , . . . , γn are chosen independent and with distribution ν, then

limn→∞ n1 dM (x0 , γ1 . . . γn x0 ) exists. It vanishes a.e. if and only if limt→∞ 1t dM (x0 , Bt ) = 0 a.s.; and • in the case when the Brownian motion is transient, one can choose ν symmetric, that is, such that for all γ in , ν(γ −1 ) = ν(γ ). The ﬁrst property goes back to Furstenberg [F2] and has been systematically developed by Lyons and Sullivan [LS] and Kaimanovich [K5]. The second one was observed in certain situations by Guivarc’h [G] and Ballmann [Ba]. Babillot

416 / a. karlsson and f. ledrappier

observed that the modiﬁed construction of [BL] has the symmetry property. Given the above, proving Theorem 22 mostly reduces to Corollary 20, if we can show that hypotheses of Corollary 20 are satisﬁed. We endow with the metric deﬁned by the metric of M on the orbit x0 . This deﬁnes a left-invariant and proper metric on : bounded sets are ﬁnite because they correspond to pieces of the orbit situated in a ball of ﬁnite volume. The measure ν is nondegenerate because its support is the whole . It is shown in [KL3] that the measure ν has a ﬁrst moment. The proof uses the details of the construction, but the idea is that the distribution of ν is given by choosing some random time and looking at the point γ x0 close to the trajectory of the Brownian motion at that time. Since the curvature is bounded from below, if the expectation of the time is ﬁnite, the expectation of the distance of the Brownian point at that time is ﬁnite as well. It also follows that the rates of escape of the Brownian motion and of the random walk are proportional. Therefore, if the manifold (M, g) is Liouville, then the random walk (G, ν) is Liouville. By Corollary 20, limn→∞ n1 dM (x0 , γ1 . . . γn x0 ) = 0 and therefore, limt→∞ 1t dM (x0 , Bt ) = 0 a.s. There are many results about the Liouville property for Riemannian covers of a compact manifold. Theorem 22 implies that the corresponding statements hold for the rate of escape of the Brownian motion. Guivarc’h [G] showed that if the group is not amenable, then (M, g) is not Liouville, whereas when is polycyclic, (M, g) is Liouville (Kaimanovich [K2]). Lyons and Sullivan ([LS], see also [Er] for a simply connected example) have examples of amenable covers without the Liouville property.

acknowledgments. This survey grew out from lectures given by both authors at the IIIème Cycle Romand de Mathématiques in Les Diablerets in March 2008 and from conversations there. We thank Tatiana SmirnovaNagnibeda and Slava Grigorchuk for their invitation to this friendly and stimulating meeting. References [A] [Al] [Ar] [Av]

J. Aaronson, An Introduction to Inﬁnite Ergodic Theory, AMS Math. Surv. Mon. 50, American Mathematical Society, Providence, RI, 1997. G. Alexopoulos, On the mean distance of random walks on groups, Bull. Sci. Math. 111 (1987), 189–199. L. Arnold, Random Dynamical Systems, Springer Monographs in Mathematics, Springer-Verlag, Berlin, 1998. A. Avez, Entropie des groupes de type ﬁni, C. R. Acad. Sci. Paris, Sér. A-B 275 (1972), A1363–A1366.

noncommutative ergodic theorems / 417 [Ba]

W. Ballmann, On the Dirichlet problem at inﬁnity for manifolds of nonpositive curvature, Forum Math. 1 (1989), no. 2, 201–213. [BL] W. Ballmann and F. Ledrappier, Discretization of positive harmonic functions on Riemannian manifolds and Martin boundary. Actes de la Table Ronde de Géométrie Différentielle (Luminy, 1992), 77–92, Sémin. Congr., 1, Soc. Math. France, Paris, 1996. [BH] M. R. Bridson and A. Haeﬂiger, Metric Spaces of Non-positive Curvature, Grundl. Math. Wiss. 319, Springer-Verlag, Berlin, 1999. [Er] A. Erschler, Liouville property for groups and manifolds, Invent. Math. 155 (2004), no. 1, 55–80. [F1] H. Furstenberg, Non-commuting random products, Trans. Amer. Math. Soc. 108 (1963), 377–428. [F2] H. Furstenberg, Random walks and discrete subgroups of Lie groups, Advances in Probability and Related Topics 1 (1971), 1–63. [G] Y. Guivarc’h, Sur la loi des grands nombres et le rayon spectral d’une marche aléatoire, Astérisque 74 (1980), 47–98. [K1] V. A. Kaimanovich, Brownian motion and harmonic functions on covering manifolds: an entropic approach, Soviet Math. Dokl. 33 (1986), no. 3, 812–816. [K2] V. A. Kaimanovich, Boundaries of random walks on polycyclic groups and the law of large numbers for solvable Lie groups, (Russian) Vestnik Leningrad. Univ. Mat. Mekh. Astronom. (1987), vyp. 4, 93–95, 112. [K3] V. A. Kaimanovich, Lyapunov exponents, symmetric spaces and a multiplicative ergodic theorem for semisimple Lie groups, J. Soviet Math. 47 (1989), 2387–2398. [K4] V. A. Kaimanovich, Poisson boundaries of random walks on discrete solvable groups, in Probability Measures on Groups, X (Oberwolfach, 1990), Plenum, New York, 1991, 205–238. [K5] V. A. Kaimanovich, Discretization of Bounded Harmonic Functions on Riemannian Manifolds and Entropy: Potential Theory (Nagoya, 1990), 213–223, de Gruyter, Berlin, 1992. [K6] V.A. Kaimanovich, The Poisson formula for groups with hyperbolic properties, Ann. Math. 152 (2000), 659–692. [KV] V. Kaimanovich and A. Vershik, Random walks on discrete groups: boundary and entropy, Ann. Prob. 11 (1983), 457–490. [Ka] A. Karlsson, Linear rate of escape and convergence in directions, in Proceedings of a Workshop at the Schrödinger Institute, Vienna 2001, (ed. by V.A. Kaimanovich, in collab. with K. Schmidt and W. Woess), de Gruyter, Berlin, 2004. [KL1] A. Karlsson and F. Ledrappier, On laws of large numbers for random walks, Ann. Prob. 34 (2006), 1693–1706. [KL2] A. Karlsson and F. Ledrappier, Drift and entropy for random walks, Pure Appl Math. Quarterly 3 (2007), 1027–1036. [KL3] A. Karlsson and F. Ledrappier, Propriété de Liouville et vitesse de fuite du mouvement Brownien, C. R. Acad. Sci. Paris, Sér. I, 344 (2007), 685–690. [KM] A. Karlsson and G. Margulis, A multiplicative ergodic theorem and nonpositively curved spaces, Comm. Math. Phys. 208 (1999), 107–123. [KMo] A. Karlsson and N. Monod, Strong law of large numbers with concave moments, unpublished note, 2008. [KN] E. Kohlberg and A. Neyman, Asymptotic behavior of nonexpansive mappings in normed linear spaces, Israel J. Math 4 (1981), 269–275.

418 / a. karlsson and f. ledrappier [Kr] [LL] [L] [LS] [MZ] [M] [O] [Pa] [S] [St] [Va] [Ve] [W]

U. Krengel, Ergodic Theorems, de Gruyter Studies in Mathematics, 6, de Gruyter, Berlin, 1985. F. Ledrappier and S. Lim, A proof of a L1/2 ergodic theorem, unpublished note, 2008. T. Lyons, Instability of the Liouville property for quasi-isometric Riemannian manifolds and reversible Markov chains, J. Diff. Geom 26 (1987), 33–66. T. Lyons and D. Sullivan, Function theory, random paths and covering spaces, J. Differential Geom. 19 (1984), no. 2, 299–323. J. Marcinkiewicz and A. Zygmund, Sur les fonctions indépendantes, Fundam. Math. 29 (1937), 60–90. P. Mathieu, Carne-Varopoulos bounds for centered random walks, Ann. Prob. 34 (2006), 987–1011. V. I. Oseledec, A multiplicative ergodic theorem: Lyapunov characteristic numbers for dynamical systems, Trans. Moscow Math. Soc. 19 (1968), 197–231. A. Pazy, Asymptotic behavior of contractions in Hilbert space, Israel J. Math. 9 (1971), 235–240. S. A. Sawyer, Maximal inequalities of weak type, Ann. of Math. (2) 84 (1966), 157–174. R. A. Struble, Metrics in locally compact groups, Comp. Math. 28 (1974), 217–222. N. Th. Varopoulos, Long range estimates for Markov chains, Bull. Sci. Math. 109 (1985), 225–252. A. Vershik, Dynamic theory of growth in groups: entropy, boundary, examples, Russian Math. Surveys 55 (2000), 667–733. P. Walters, A dynamical proof of the multiplicative ergodic theorem, Trans. Amer. Math. Soc. 335 (1993), 245–257.

13

sorin popa 1 and stefaan vaes 2

COCYCLE AND ORBIT SUPERRIGIDITY FOR LATTICES IN SL (n, R) ACTING ON HOMOGENEOUS SPACES to bob zimmer at the occasion of his

60 th

birthday

Abstract We prove cocycle and orbit equivalence superrigidity for lattices in SL (n, R) acting linearly on Rn , as well as acting projectively on certain ﬂag manifolds, including the real projective space. The proof combines operator algebraic techniques with the property (T) in the sense of Zimmer for the action SL (n, Z) Rn , n ≥ 4. We also show that the restriction of the orbit equivalence relation R( SL (n, Z) Rn ) to a subset of ﬁnite Lebesgue measure provides a II1 equivalence relation with property (T) and yet fundamental group equal to R+ .

1. Introduction and Statement of Main Results Over the last few years, operator algebraic methods were used to prove several orbit equivalence and cocycle superrigidity theorems: for Bernoulli actions of property (T) groups [19], of product groups [18], and for proﬁnite actions of property (T) groups [13]. In this paper, we extend the scope of these methods to a more geometric class of actions, like the natural actions of lattices ⊂ SL (n, R) on the vector space Rn , on the projective space Pn−1 (R), and on certain ﬂag manifolds, all of which can be viewed as SL (n, R)-homogeneous spaces. None of these actions is probability measure preserving. Hence, property (T) of the acting group has to be replaced by Zimmer’s notion of property (T) for a nonsingular action (see [25]), which plays a crucial role in this paper. It is shown that for any lattice ⊂ SL (n, R) the linear action Rn has property (T) if and only if n ≥ 4. We then deduce the following theorem; see Theorem 4.1. 1. Partially supported by NSF Grant DMS-0601082. 2. Partially supported by ERC Starting Grant VNALG-200749 and Research Programme G.0231.07 of the Research Foundation–Flanders (FWO).

419

420 / s. popa and s. vaes

theorem 1.1. Let < SL (n, R) be a lattice and let R be the II1 equivalence relation obtained by restricting the orbit equivalence relation R( Rn ) to a set of Lebesgue measure 1, for some n ≥ 4. Then we have • R has property (T), in the sense of Zimmer, yet the fundamental group of R

equals R+ ; and • Rt cannot be implemented by a free action of a group, ∀t > 0. Also, Rt cannot

be implemented by a (not necessarily free) action of a discrete property (T) group, ∀t > 0. Every orbit equivalence relation R( X ) of a probability measure preserving (p.m.p.) action of a property (T) group, has property (T) in the sense of Zimmer. Theorem 1.1 provides the ﬁrst examples of property (T) equivalence relations that do not arise in this way. We say that a Polish group is of ﬁnite type if it can be realized as the closed subgroup of the unitary group of some II1 factor with separable predual. All countable and all second countable compact groups are Polish groups of ﬁnite type. In [19], the ﬁrst author proved that every 1-cocycle for the Bernoulli action of a property (T) group with values in a Polish group of ﬁnite type is cohomologous to a group morphism. We say that actions with this property are Uﬁn -cocycle superrigid. More precisely:

definition 1.2. The nonsingular action G (X , μ) of the locally compact second countable group G on the standard measure space (X , μ) is called Uﬁn cocycle superrigid if every 1-cocycle for the action G (X , μ) with values in a Polish group of ﬁnite type G is cohomologous to a continuous group morphism G → G. Until now, the only known examples of Uﬁn -cocycle superrigid actions were the Bernoulli and Gaussian actions of property (T) groups (cf. [19]) and, respectively, product groups (cf. [18]), all of which are p.m.p. actions. We exhibit here the ﬁrst examples of nonsingular ergodic actions G (X , μ) that are Uﬁn -cocycle superrigid but for which there exist no invariant probability measure ν ∼ μ on X . As it happens, these examples are of genuine geometric nature.

theorem 1.3. The following actions are Uﬁn -cocycle superrigid. 1) For n ≥ 5 and any lattice in SL (n, R), the linear action Rn . 2) For n ≥ 5 and any ﬁnite index subgroup of SL (n, Z), the afﬁne action Zn Rn .

cocycle and orbit superrigidity for lattices in sl(n, R) / 421

3) For n ≥ 4k + 1, any lattice in SL (n, R), and H any closed subgroup of GL (k, R), the action ⎧ ⎨ ×H if ( − 1, −1) ∈ × H, G Mn,k (R) where G := {±(1,1)} ⎩ × H otherwise, by left-right multiplication on the space Mn,k (R) of n × k matrices equipped with the Lebesgue measure. In [19], the ﬁrst author introduced the notion of malleability for measure preserving actions (X , μ). Roughly speaking, this property requires the existence of a ﬂow on X × X , commuting with the diagonal -action and connecting the identity map to the ﬂip map on X × X . Theorem 0.1 in [19] shows that every weakly mixing, malleable p.m.p. action of a property (T) group is Uﬁn -cocycle superrigid. We prove here a cocycle superrigidity result for inﬁnite measure preserving malleable actions. But, property (T) of the group has to be replaced by property (T) of the diagonal action of X × X . In the case of SL (n, Z) Rn , this forces n ≥ 5. Finally, weak mixing has to be replaced by the ergodicity of the 4-fold diagonal action X × X × X × X , which in the case of SL (n, Z) Rn again holds exactly for n ≥ 5. Using the cocycle superrigidity of SL (n, Z) Rn , we then derive a full classiﬁcation of all 1-cocycles for the action SL (n, Z) Tn = Rn /Zn , with values in a Polish group of ﬁnite type. As such, our Example 5.12 below, complements Zimmer’s celebrated cocycle superrigidity theorem [26]. Thus, Zimmer’s result concerns arbitrary actions SL (n, Z) (X , μ), but makes certain restrictions on the target groups (which must be simple linear algebraic groups) and on the cocycles (Zariski density), while our result treats a very speciﬁc action, but allows rather general target groups and requires no conditions on the cocycles. From the cocycle superrigidity Theorem 1.3, we derive orbit equivalence (OE) superrigidity results for the following concrete actions of lattices in SL (n, R) and PSL (n, R): 1) The linear action Rn . 2) The action Tn = Rn /Zn for a ﬁnite-index subgroup of SL (n, Z). 3) The projective action Pn−1 (R). 4) The natural action X of a lattice in PSL (n, R) on the real ﬂag manifold X of signature (d1 , . . . , dl , n). (Recall that points in X are ﬂags {0} ⊂ V1 ⊂ · · · ⊂ Vl ⊂ Rn where Vi is a vector subspace of Rn with dimension di .)

422 / s. popa and s. vaes

The action in (1) has the Lebesgue measure as inﬁnite invariant measure, while the actions in (3) and (4) do not have ﬁnite or inﬁnite invariant measures. All the actions in (1)–(4) are essentially free and ergodic; see Lemma 5.7 for details. The natural invariant measure class on the ﬂag manifold X can be described as follows. Put dl = k and consider the set Mn,k (R) of n × k matrices of rank k, equipped with the Lebesgue measure. Denote by E = (E1 , . . . , El ) the standard ﬂag of signature (d1 , . . . , dl , n), that is Ei = span{e1 , . . . , edi }, where e1 , . . . , en are the standard basis vectors in Rn . The group GL (k, R) acts on Mn,k (R) by right multiplication. This action is free and proper and Mn,k (R)/H → X : A → (AE1 , . . . , AEl ) is an isomorphism. Here, H = {g ∈ GL (k, R) | gEi ⊂ Ei for all i = 1, . . . , l}. Writing k1 = d1 and ki = di − di−1 for i ≥ 2, the group H can of course be written as ⎛ ⎞ GL (k1 , R) ∗ ··· ∗ ⎜ ⎟ 0 GL (k2 , R) · · · ∗ ⎜ ⎟ ⎜ ⎟. 1.1 H=⎜ .. .. .. .. ⎟ . ⎝ ⎠ . . . 0 0 · · · GL (kl , R) Before stating our OE superrigidity results, recall the following terminology. α

β

definition 1.4. Let (X , μ) and (Y , η) be essentially free, ergodic, nonsingular actions of countable groups on standard measure spaces. • A stable orbit equivalence (SOE) between α and β is a nonsingular iso-

morphism : X0 → Y0 between nonnegligible subsets X0 ⊂ X , Y0 ⊂ Y , such that is an isomorphism between the restricted orbit equivalence relations R( X )|X0 and R( Y )|Y0 . • We say that X is induced from 1 X1 , if 1 is a subgroup of , X1 is a nonnegligible subset of X and g · X1 ∩ X1 is negligible for all g ∈ − 1 . For the linear lattice actions and the quotient action SL (n, Z) Tn , we get the following.

theorem 1.5. Let n ≥ 5 and ⊂ SL (n, R) a lattice. Let (Y , η) be any essentially free, ergodic, nonsingular action of the countable group .

cocycle and orbit superrigidity for lattices in sl(n, R) / 423

1) The actions Rn and Y are SOE if and only if Y is conjugate to an induction of one of the following actions: • Rn itself, and • (only in case −1 ∈ ) the quotient action /{±1} Rn /{±1}.

2) The actions SL (n, Z) Tn and Y are SOE if and only if Y is conjugate to an induction of one of the following actions • SL (n, Z) Tn itself, • SL (n, Z) Zn Rn ,

Z n Rn for some λ ∈ N \ {0, 1}, and λZ λZn • (only in case n is even) one of the actions • SL (n, Z)

PSL (n, Z)

Tn

{±1}

or

PSL (n, Z)

Z n Rn /(2Z)n . 2Z {±1}

Part (2) of the above theorem is a slightly more detailed version, with a very different proof, of a result of Furman [7, corollary B], who shows that for all n ≥ 3, the actions SL (n, Z) Tn and Y are SOE if and only if they are virtually conjugate. Furman’s proof relies on Zimmer’s cocycle superrigidity, while ours is based on the cocycle superrigidity Theorem 1.3. To formulate easily the correct OE superrigidity statements for lattice actions on ﬂag manifolds, make the following observations. The real ﬂag manifold of signature (d1 , . . . , dl , n) has a natural 2l -fold

consisting of oriented ﬂags covering X {0} ⊂ (V1 , ω1 ) ⊂ · · · ⊂ (Vl , ωl ) ⊂ Rn where every Vi is a vector subspace of Rn with an orientation ωi . Clearly,

= Mn,k (R)/H0 , where H0 = {g ∈ H | det(g|E ) > 0 for all i = 1, . . . , l}. In X i the Expression 1.1 above, H0 consists of those matrices A that have on the diagonal Aii ∈ GL (ki , R) with det Aii > 0 for all i.

Denote l = H/H0 and observe that l ∼ = (Z/2Z)⊕l . Then, l acts on X by reversing orientations, but keeping the ﬂags. We denote by −1 ∈ l the multiplication by −1 and observe that −1 = 1 in l if all di are even. Clearly,

/l . X =X

theorem 1.6. Let X be the real ﬂag manifold of signature (d1 , . . . , dl , n). Let

the 2l ⊂ PSL (n, R) be a lattice and assume that n ≥ 4dl + 1. Denote by X fold covering of X consisting of oriented ﬂags, as explained before the theorem. Let (Y , η) be any essentially free, ergodic, nonsingular action of the countable group .

424 / s. popa and s. vaes

The actions X and Y are SOE if and only if Y is conjugate to an induction of one of the actions l

/ for some subgroup < l with − 1 ∈ , or X

× l /

/ for some < l with − 1 ∈ and with • X = {±(1, 1)} {±1} · ⊂ GL (n, R). • ×

example 1.7. If n ≥ 5 and ⊂ PSL (n, R) is a lattice, the action Pn−1 (R) is a special case of the ﬂag manifold action treated in Theorem 1.6. Hence, Pn−1 (R) and Y are SOE if and only if Y is conjugate to an induction of either Pn−1 (R) or its double cover Rn /R+ , where

= {±1} · ⊂ GL (n, R). Finally, combining the work of [9] and the above OE superrigidity results, we classify up to stable orbit equivalence, the lattice actions on Rn and on ﬂag manifolds; see Theorems 6.2 and 6.3. At the same time, we compute the outer automorphism group of the associated orbit equivalence relation.

2. Preliminaries Let R be a countable measured equivalence relation on the standard measure space (X , μ), meaning that R is, on the one hand, a Borel subset of X × X and, on the other hand, an equivalence relation on X with countable equivalence (1) (1) classes, such that the σ -ﬁnite measures μl and μr on R deﬁned by (1) μl (Y ) = #{y ∈ X | (x, y) ∈ Y } dμ(x) , X

(1)

μr (Y ) =

X

#{x ∈ X | (x, y) ∈ Y } dμ(y) , (1)

(1)

are mutually absolutely continuous: μl ∼ μr . One calls R ergodic if every R-saturated measurable subset of X is negligible or conegligible. The measure (1) (2) μ is said to be invariant if μl = μr . A II1 equivalence relation is a countable measured ergodic equivalence relation with invariant probability measure. We now recall Zimmer’s deﬁnition of property (T) for a countable measured equivalence relation R. To this end, ﬁrst deﬁne R(2) = {(x, y, z) ∈ X × X × X | x Ry and y Rz}. Note that R(2) comes equipped with a canonical measure class, induced by the σ -ﬁnite measure (2) #{(y, z) ∈ R | (x, y, z) ∈ Y } dμ(x) . μl (Y ) = X

cocycle and orbit superrigidity for lattices in sl(n, R) / 425 (2)

(2)

Note that one deﬁnes absolutely continuous measures μm , respectively, μr , by integrating over y, respectively, z. But we only make use of the measure (2) class given by μl . • A 1-cocycle of R with values in the unitary group U (K ) of a Hilbert space K

is a Borel map c : R → U (K ) satisfying c(x, z) = c(x, y)c(y, z) for almost all (x, y, z) ∈ R(2) . • Suppose that c : R → U (K ) is a 1-cocycle of R. – An invariant unit vector of c is a Borel map ξ : X → K satisfying ξ (x) = c(x, y)ξ (y) for almost all (x, y) ∈ R and ξ (x) = 1 for almost all x ∈ X . – A sequence of almost invariant unit vectors of c is a sequence of Borel maps ξn : X → K satisfying

ξn (x) − c(x, y)ξn (y) → 0

for almost all

(x, y) ∈ R

and ξn (x) = 1 for all n ∈ N and almost all x ∈ X .

definition 2.1. A countable measured equivalence relation R is said to have property (T) in the sense of Zimmer if the following holds: every 1-cocycle of R with values in the unitary group of a Hilbert space and admitting a sequence of almost invariant unit vectors admits a unit invariant vector. We leave the proof of the following lemma as an exercise.

lemma 2.2. Let R be a countable measured equivalence relation on (X , μ) and Y ⊂ X a measurable subset such that the R-saturation of Y is conegligible. Then, the restriction of R to Y has property (T) if and only if R has property (T). 3. Property (T) for Actions of Locally Compact Groups We recalled above Zimmer’s deﬁnition of property (T) for a II1 equivalence relation. In fact, one can deﬁne property (T) for measured groupoids in general; see [1]. We do not need this generality in this paper, but we do need the concept of property (T) for nonsingular actions of locally compact second countable (l.c.s.c.) groups on measure spaces. For a groupoid approach to this deﬁnition, we refer to [1]. For the convenience of the reader, we gather in this section the necessary concepts and results and present them in an operator algebra framework. All von Neumann algebras are supposed to have separable predual and all locally compact groups are supposed to be second countable.

426 / s. popa and s. vaes

Given a von Neumann algebra M ⊂ B (H) and a Hilbert space K , we make use of the W ∗ -module M ⊗ K , that can be deﬁned as {T ∈ B (H, H ⊗ K ) | Tx = (x ⊗ 1)T , ∀x ∈ M }. By construction, M ⊗ K is an (M ⊗ B (K ))-M-bimodule and carries the strong∗ -topology inherited from B (H, H ⊗ K ). If M is a von Neumann algebra, we equip Aut (M) with the Polish topology making the functions Aut (M) → M∗ : α → ω ◦ α continuous for all ω ∈ M∗ . An action α of an l.c.s.c. group G on a von Neumann algebra M, denoted α G M, is a continuous group morphism α : G → Aut (M). α

• A 1-cocycle of an action G M with values in the unitary group U (K ) of

a Hilbert space K , is a strongly continuous map c : G → U (M ⊗ B (K )) satisfying c(gh) = c(g)(αg ⊗ id)(c(h)) for all g, h ∈ G. Note that by Theorem 3 in [16], it makes no difference to assume only that c is a measurable map, with the previous formula holding for almost all (g, h) ∈ G × G. α • An invariant unit vector of the 1-cocycle c of G M, is an element ξ in the W ∗ -module M ⊗ K satisfying ξ ∗ ξ = 1 and c(g)(αg ⊗ id)(ξ ) = ξ for all g ∈ G. α • A sequence of almost invariant unit vectors of the 1-cocycle c of G M, is a sequence ξn ∈ M ⊗ K satisfying ξn∗ ξn = 1 for all n and c(g)(αg ⊗ id)(ξn ) − ξn → 0 ∗ -strongly, uniformly on compact subsets of G. • The action G M is said to have property (T) if every 1-cocycle with values in the unitary group of a Hilbert space and admitting a sequence of almost invariant unit vectors, admits an invariant unit vector. The relation between this notion of property (T) for an action and Zimmer’s property (T) for a measured equivalence relation is given by the following straightforward lemma.

lemma 3.1. Let (X , μ) be a nonsingular action of the countable group and consider the orbit equivalence relation R( X ). If the action L∞ (X ) has property (T) in the sense above, then the equivalence relation R( X ) has property (T) in the sense of Deﬁnition 2.1. The converse holds when (X , μ) is essentially free. Note that by combining Lemmas 2.2 and 3.1, we obtain the following: if (X , μ) is a nonsingular, essentially free ergodic action and Y ⊂ X is nonnegligible, then the restricted equivalence relation R( X )|Y has property (T) iff the action (X , μ) has property (T). The following result is proven for discrete groups in [25], for weakly mixing actions, and in [6], for arbitrary actions (see also [2, corollary 5.16]). These

cocycle and orbit superrigidity for lattices in sl(n, R) / 427

methods work as well in the locally compact case and for completeness, we give a proof in a von Neumann algebra setup. α

proposition 3.2. Let G M be an action of the l.c.s.c. group G on the von Neumann algebra M. Suppose that τ is a faithful normal tracial state on M, invariant under α. Then, G M has property (T) if and only if the group G has property (T). Combining Proposition 3.2 and Lemma 3.1, it follows that for every p.m.p. action (X , μ) of a countable property (T) group , the orbit equivalence relation R( X ) has property (T) in the sense of Deﬁnition 2.1. Theorem 1.1 provides the ﬁrst examples of property (T) equivalence relations that do not arise in this way. Proof of Proposition 3.2. Suppose ﬁrst that G has property (T). Let c : G → U (M ⊗ B (K )) be a 1-cocycle of G M having ξn ∈ M ⊗ K as a sequence of almost invariant unit vectors. Deﬁne the unitaries ug on L2 (M, τ ) by extending αg . Then, π : G → U ( L2 (M, τ ) ⊗ K ) : π (g) = c(g)(ug ⊗ 1) is a unitary representation of G and we can view ξn as a sequence of almost invariant unit vectors. Since G has property (T), π admits an invariant unit vector. Even more, we ﬁnd a sequence ηn ∈ L2 (M, τ ) ⊗ K of π -invariant vectors satisfying ηn − ξn → 0. Since ξn = 1, it follows that ξn∗ ηn − 1 2 → 0. Hence, the right support projection of ξn∗ ηn converges strongly to 1. A fortiori, the right support projection of ηn converges to 1. We view ηn as a closed operator from L2 (M, τ ) to L2 (M, τ ) ⊗ K . Taking the polar decomposition of ηn , we ﬁnd vn ∈ M ⊗ K satisfying c(g)(αg ⊗ id)(vn ) = vn for all g ∈ G, n ∈ N and such that vn∗ vn is a sequence of projections in M converging strongly to 1. Deﬁne the von Neumann algebra M ⊗ B (K ) M ⊗ K N= . (M ⊗ K )∗ M Deﬁne the action (γg ) of G on N by γg

a e

b c(g)(αg ⊗ id)(a)c(g)∗ = (αg ⊗ id)(e)c(g)∗ f

c(g)(αg ⊗ id)(b) αg ( f )

.

428 / s. popa and s. vaes

Deﬁne p = 01 00 , q = 1 − p, and wn = 00 v0n . Then, wn is a sequence of partial isometries in the ﬁxed-point algebra N G , satisfying wn ∈ pN G q and wn∗ wn → q strongly. It follows that q ≺ p in the von Neumann algebra N G . So, we ﬁnd v ∈ M ⊗ K satisfying c(g)(αg ⊗ id)(v) = v for all g and v ∗ v = 1. Hence, G M has property (T). Suppose conversely that G M has property (T). Let π : G → U (K ) be a strongly continuous unitary representation of G admitting ξn as a sequence of almost invariant unit vectors. By [3, theorem 2.12.9], it is sufﬁcient to prove that π has a nonzero ﬁnite-dimensional π (G)-invariant subspace. Deﬁne c : G → U (M ⊗ B (K )) : c(g) = 1 ⊗ π(g). Obviously, c is a 1-cocycle of G M having 1 ⊗ ξn as a sequence of almost invariant unit vectors. By property (T) of G M, we ﬁnd ξ ∈ M ⊗ K satisfying ξ ∗ ξ = 1 and c(g)(αg ⊗ id)(ξ ) = ξ for all g ∈ G. Denoting again by u : g → ug the representation of G on L2 (M, τ ) obtained by extending αg , we ﬁnd that u ⊗ π admits an invariant unit vector ξ . Identify L2 (M, τ ) ⊗ K with the Hilbert space of Hilbert-Schmidt operators from L2 (M, τ ) to K . Then, T := ξ ξ ∗ is a nonzero trace-class operator on K satisfying π (g)T π(g)∗ = T for all g ∈ G. So, for ε > 0 sufﬁciently small, the spectral projection χ[ε,+∞) (T ) projects onto a nonzero ﬁnite-dimensional π(G)-invariant subspace of K . The following is a slight generalization of [2, theorem 5.3]. When H M is an action, we denote by M H the von Neumann algebra of H-ﬁxed points. α

lemma 3.3. Let G M be an action of the l.c.s.c. group G on the von Neumann algebra M. Let H G be a closed normal subgroup and assume that there is a ∗ -isomorphism θ : M → L∞ (H) ⊗ M H satisfying θ ◦ α = (ρ ⊗ id) ◦ θ for all h h h ∈ H, where ρh denotes the right translation by h on L∞ (H). Then, G M has property (T) if and only if G/H M H has property (T). α

Proof. We say that two 1-cocycles c1 , c2 of G M with values in U (K ) are unitarily equivalent if there exists a unitary v ∈ U (M ⊗ B (K )) satisfying c1 (g) = α vc2 (g)(αg ⊗ id)(v ∗ ) for all g ∈ G. We denote by H1 (G M, U (K )) the set of equivalence classes of 1-cocycles. In the ﬁrst part of the proof, we show that the obvious map : H1 (G/H M H , U (K )) → H1 (G M, U (K )) : (c) = c ◦ π

with

π : G → G/H , is a bijection. In the second part of the proof, we show that this map and its inverse preserve the property of having invariant, respectively, almost

cocycle and orbit superrigidity for lattices in sl(n, R) / 429

invariant, vectors. Both parts together show that property (T) of G M is equivalent with property (T) of G/H M H . It is straightforward to check that is well deﬁned and injective. Suppose that c : G → U (M ⊗ B (K )) is a 1-cocycle of G M. In order to prove that c is in the range of , it sufﬁces to prove that c is unitarily equivalent with c satisfying c (h) = 1 for all h ∈ H. Identify throughout U (M ⊗ B (K )) with U ( L∞ (H) ⊗ M H ⊗ B (K )) and view the latter as measurable functions H → U (M H ⊗ B (K )), modulo equality almost everywhere. By [16, theorem 1], take a measurable map ϕ : H × H → U (M H ⊗ B (K )) such that c(h) = ϕ(·, h) for all h ∈ H. Since c is a cocycle, we ﬁnd that ϕ(k, hg) = ϕ(k, h)ϕ(kh, g)

for almost all (k, h, g) ∈ H × H × H.

By the Fubini theorem, take k0 ∈ H such that for almost all (h, g) ∈ H × H, the previous equality holds for (k0 , h, g). Deﬁne the unitary v ∈ U (M ⊗ B (K )) as the function H → U (M H ⊗ B (K )) given by h → ϕ(k0 , k−1 0 h) and set c (g) = ∗ vc(g)(αg ⊗ id)(v ). By construction, c (g) = 1 for almost all g ∈ H and hence for all g ∈ H by continuity. It is an exercise to check that the 1-cocycle c ∈ H1 (G/H M H , U (K )) has an invariant unit vector if and only if (c) has. Also, a sequence of almost invariant unit vectors for c deﬁnes a sequence of almost invariant unit vectors for (c). Finally, suppose that ξn ∈ M ⊗ K is a sequence of almost invariant unit vectors for (c). In order to conclude the proof of the lemma, it sufﬁces to show that there exists a sequence ηn ∈ M H ⊗ K satisfying ηn∗ ηn = 1 for all n and ξn − ηn → 0 ∗ -strongly. Deﬁne the sets X = {v ∈ M ⊗ K | v ∗ v = 1} and Y = {v ∈ M H ⊗ K | v ∗ v = 1}. Through the isomorphism M ∼ = L∞ (H) ⊗ M H , we identify X with the set of measurable functions from H to Y (modulo equality almost everywhere). Take a bounded metric d0 on Y inducing the strong∗ topology. Let μ be a probability measure on H in the same measure class as the Haar measure. Following [16, p. 5], deﬁne the metric d on X by d(v, w) = d0 (v(h), w(h))dμ(h). H

strong∗

Then, d induces the topology on X . It is easy to check that when vn , wn ∈ X such that d(vn , wn ) is summable, then vn (h) − wn (h) → 0 ∗ -strongly for almost every h ∈ H (see [16, proposition 6]). View Y ⊂ X as constant functions. We have to prove that d(ξn , Y ) → 0. Suppose the contrary. Write H as an increasing union of compact subsets Hn . After passage to a subsequence, we ﬁnd ε > 0 such that d(ξn , Y ) > ε for all n and such that

430 / s. popa and s. vaes

d(ξn , (αg ⊗ id)(ξn )) < 2−n

for all n ∈ N, g ∈ Hn .

It follows that for all g ∈ H, we have d0 (ξn (h), ξn (hg)) → 0

for almost all h ∈ H .

By the Fubini theorem, take h0 ∈ H such that d0 (ξn (h0 ), ξn (h0 g)) → 0 for almost all g ∈ H. It follows that d(ξn , ξn (h0 )) → 0, contradicting the assumption that d(ξn , Y ) > ε for all n. If M is a commutative von Neumann algebra, an action H M satisfying the condition in Lemma 3.3 is called essentially free and proper. More precisely:

definition 3.4. The nonsingular action G (X , μ) of the l.c.s.c. group G on the standard measure space (X , μ) is called essentially free and proper if there exists a measurable map π : X → G such that π(g · x) = gπ(x) for almost all (g, x) ∈ G × X . Equivalently, there exists a ∗ -isomorphism L∞ (X , μ) → L∞ (G) ⊗ L∞ (X , μ)G conjugating the G-action on L∞ (X , μ) with the action (ρg ⊗ id)g∈G , where ρg denotes the right translation by g on L∞ (G). Indeed, suppose that we are given π : X → G as above. Denote by X /G the space of ergodic components of G X and deﬁne the map : X → G × X /G : x → (π (x), x). Then, is invertible with the inverse being well deﬁned and given by (g, x) → g −1 π (x)−1 · x. Also, conjugates the action G X with the action ρg × id.

proposition 3.5. Let G be an l.c.s.c. group with closed subgroups H1 , H2 . Then, H1 L∞ (G /H2 ) has property (T) if and only if H2 L∞ (G /H1 ) has property (T). Proof. Set M = L∞ (G ) and G = H1 × H2 acting by left-right translations on M: (α(g,h) (F))(x) = F(g −1 xh)

for all

F ∈ L∞ (G ), x ∈ G , g ∈ H1 , h ∈ H2 .

We apply Lemma 3.3 to G M and the closed normal subgroups H = Hi , i = 1, 2 of G. By the Effros-Mackey theorem (see e.g., [14, theorem II.12.17]), the quotient map G → G /H admits a Borel lifting and hence, there exists an H-equivariant isomorphism M → L∞ (H) ⊗ M H . So, by Lemma 3.3, property (T) of G M is equivalent with property (T) of H1 L∞ (G /H2 ) as well as with property (T) of H2 L∞ (G /H1 ).

cocycle and orbit superrigidity for lattices in sl(n, R) / 431

4. The Lattice Actions Rn Have Property (T) Recall from [6] that if (X , μ) is a free ergodic p.m.p. action, then property (T) of R( X ) in the sense of Zimmer is equivalent with property (T) of the group (see also [25, proposition 2.4] and Proposition 3.2 above). If is a property (T) group, the fundamental group of R( X ) is countable for any free ergodic p.m.p. action (see [12, corollary 1.8] if, moreover, has inﬁnite conjugacy classes (ICC) and see [13, theorem 5.9] for the general case). But more is true: we proved in [22, theorem 6.1] that the fundamental group of a II1 equivalence relation R on (X , μ) is countable whenever the full group [R] contains a property (T) group that implements an ergodic action on (X , μ). As a result, the following theorem is rather surprising: we obtain a II1 equivalence relation R with property (T) and fundamental group R+ ; hence, none of the Rt can be implemented by a free action of a group and none of the Rt can be implemented by a possibly nonfree action of a property (T) group. The following result covers Theorem 1.1 stated in the introduction.

theorem 4.1. Let < SL (n, R) be a lattice and let R be the II1 equivalence relation obtained by restricting the orbit equivalence relation R( Rn ) to a set of Lebesgue measure 1. If n ≥ 4, the equivalence relation R has property (T) in the sense of Zimmer, but nevertheless F (R) = R+ . In particular, • none of the equivalence relations Rt , t > 0 can be implemented by a free action

of a group, • none of the equivalence relations Rt , t > 0 can be implemented by a possibly

nonfree action of a property (T) group. In fact, Rt does not contain any ergodic subequivalence relation that can be implemented by a property (T) group. Proof. By the remark following Lemma 3.1, proving property (T) of R amounts to proving property (T) for the action L∞ (Rn ). Deﬁne the l.c.s.c. group G = SL (n, R) and set H1 = . Consider the linear action G Rn and set H2 = {A ∈ SL (n, R) | Ae1 = e1 }, where e1 denotes the ﬁrst basis vector of Rn . By construction, the action L∞ (Rn ) can be viewed as H1 L∞ (G /H2 ). Hence, by Proposition 3.5, property (T) for this last action is equivalent with property (T) of H2 L∞ (G /H1 ). This action admits a ﬁnite invariant measure, because H1 is a lattice in G . Moreover, H2 ∼ = SL (n − 1, R) Rn−1 , which has property (T) for n ≥ 4. So, it follows from Proposition 3.2 that H2 L∞ (G /H1 ) has property (T). The action on Rn by multiples of the identity matrix scales the Lebesgue measure and commutes with the action of . Hence, the fundamental group

432 / s. popa and s. vaes

of R equals R+ . The statements about implementing Rt by group actions follow from the discussion preceding the theorem.

remark 4.2. Note that a similar reasoning as in the proofs of Lemmas 3.1 and 3.3, and Proposition 3.2 show that those statements hold true if we replace everywhere “property (T)” by “amenable.” By applying this to the action SL (2, Z) R2 , one recovers a result from [2], showing that the action SL (2, Z) R2 is amenable and hence, L∞ (R2 ) SL (2, Z) is isomorphic to the unique hyperﬁnite II∞ factor, by Connes’s theorem [4]. This also shows that the action SL (3, Z) R3 is not amenable, nor has property (T). remark 4.3. Denote by R the II1 equivalence relation constructed in Theorem 4.1. Since R has property (T), [20, theorem 4.4.1] and [17, corollaire 18] imply that R is ﬁnitely generated. Hence, R has ﬁnite cost and ﬁnite ﬁrst 2 Betti number. Since the fundamental group of R equals R+ , it follows from [10, proposition II.6], respectively [11, corollaire 5.7], that actually cost R = 1 (2) and β1 (R) = 0. 5. Cocycle and OE Superrigidity Theorems 5.1. Proof of Theorem 1.3

We prove in this section the cocycle superrigidity Theorem 1.3 as a consequence of the more general Theorem 5.3 below. We do not know whether SL (n, R) Rn is Uﬁn -cocycle superrigid for n = 3, 4. On the other hand, it is generally believed that whenever one has a Uﬁn cocycle superrigidity result, it actually holds true for more general target groups than the Uﬁn -class. However, the following result, which we’ll prove at the end of this subsection, shows that there even exist locally compact target groups for 1-cocycles for the action SL (n, R) Rn , which cannot be untwisted.

proposition 5.1. Let n ≥ 3. The action SL (n, Z) Rn admits a 1-cocycle with values in SL (n − 1, R) Rn−1 that is not cohomologous to a group morphism. Recall from [19] the following deﬁnition of s-malleability of a measure preserving action.

definition 5.2. Let be an l.c.s.c. group and (X , μ) an action preserving the ﬁnite or inﬁnite measure μ. The action is called s-malleable if there exists

cocycle and orbit superrigidity for lattices in sl(n, R) / 433 • a 1-parameter group (αt )t∈R of measure preserving transformations of

X × X , and • an involutive measure preserving transformation β of X × X ,

such that • αt and β commute with the diagonal action X × X , • α1 (x, y) ∈ {y} × X for almost all (x, y) ∈ X × X , • β(x, y) ∈ {x} × X for almost all (x, y) ∈ X × X , and • αt ◦ β = β ◦ α−t for all t ∈ R.

Theorem 0.1 in [19] says the following. Let (X , μ) be an s-malleable, probability measure preserving action and < a normal subgroup with the relative property (T) such that the restriction of (X , μ) to is weakly mixing. Then every 1-cocycle of (X , μ) with values in a Polish group of ﬁnite type, is cohomologous to a group morphism. Recall here that one of the equivalent formulations of weak mixing for a p.m.p. action G (X , μ) of a countable group G is the ergodicity of the diagonal action G (X × X , μ × μ). If G (X , μ) is a weakly mixing p.m.p. action and G (Y , η) is any ergodic p.m.p. action, then the diagonal action G X × Y is ergodic. In particular, the diagonal action G X ×···×X k times

is ergodic for every k, once it is ergodic for k = 2. For inﬁnite measure preserving actions, things are more complicated and, for instance, the diagonal action SL (n, Z) Rn × · · · × Rn k times

is ergodic if and only if k ≤ n − 1. This partially explains the formulation of the following result.

theorem 5.3. Let (X , μ) be an inﬁnite measure preserving, s-malleable action. Assume that • the diagonal action X × X has property (T), • the 4-fold diagonal action X × X × X × X is ergodic.

Then, (X , μ) is Uﬁn -cocycle superrigid.

434 / s. popa and s. vaes

The proof of Theorem 5.3 follows entirely the setup of the proof of [19, theorem 0.1]. But, one has to be careful at those places in [19] where weak mixing is applied. The proper way to deal with these issues lies in the following lemma distilled from the proof of [8, lemma 3.1].

lemma 5.4. Let (Z, d) be a Polish space with separable complete metric d and (αg )g∈G a continuous action of a Polish group G by homeomorphisms of Z. Assume that d is (αg )g∈G -invariant. Let be an l.c.s.c. group and (X , μ), (Y , η) nonsingular actions. Let F : X × Y → Z be a measurable map satisfying F(g · x, g · y) = αω(g,x) (F(x, y)) for almost all (x, y) ∈ X × Y , g ∈ , where ω : × X → G is some measurable map. Assume the diagonal action X × Y × Y satisﬁes L∞ (X × Y × Y ) = ∞ L (X ) ⊗ 1 ⊗ 1 (which holds in particular if the diagonal action X × Y × Y is ergodic). Then, there exists a measurable H : X → Z with F(x, y) = H(x) for almost all (x, y) ∈ X × Y . Proof. Deﬁne the map G : X × Y × Y → R : G(x, y, z) = d(F(x, y), F(x, z)). Since d is (αg )g∈G -invariant, the map G is invariant under the diagonal -action. By our assumption, G(x, y, z) = G0 (x) for almost all (x, y, z) and some measurable map G0 : X → R+ . We claim that G0 (x) = 0 for almost all x ∈ X . Let δ > 0 and assume that G0 (x) ≥ δ for all x in a nonnegligible subset U of X . Cover Z by a sequence (Bn )n∈N of balls of diameter strictly smaller than δ. Write, for every x ∈ X , Fx : y → F(x, y). By the Fubini theorem, for almost every x ∈ U , we have G(x, y, z) ≥ δ for almost all (y, z) ∈ Y × Y . Hence, for almost every x ∈ U , we have that Fx−1 (Bn ) is negligible for every n, which is absurd. So, G(x, y, z) = 0 for almost all (x, y, z) ∈ X × Y × Y . Again by the Fubini theorem, take z ∈ Y such that d(F(x, y), F(x, z)) = 0 for almost all (x, y) ∈ X × Y . Putting H(x) := F(x, z), we are done. Proof of Theorem 5.3. Let N be a II1 factor and G ⊂ U (N) a closed subgroup. Let ω : × X → G be a 1-cocycle, meaning that for all g, h ∈ , we have ω(gh, x) = ω(g, h · x) ω(h, x)

for almost all x ∈ X .

Deﬁne the following 1-cocycles for the diagonal action X × X .

cocycle and orbit superrigidity for lattices in sl(n, R) / 435

ω0 : × X × X → G : ω0 (g, x, y) = ω(g, x)

and

ωt : × X × X → G : ωt (g, x, y) = ω0 (g, αt (x, y)) . Deﬁne the action (ρg )g∈ of by automorphisms of L∞ (X ) ⊗ N by the formula (ρg −1 (F))(x) = ω(g, x)∗ F(g · x)ω(g, x) . Denote by B the von Neumann subalgebra of (ρg )g∈ -ﬁxed points.

claim. Whenever p is a nonzero projection in B, there exists a measurable function ϕ : X × X → N and a nonzero projection q ∈ B such that q ≤ p and such that for all g ∈ , we have ω(g, x)ϕ(x, y) = ϕ(g · x, g · y)ω(g, y),

ϕ(x, y)ϕ(x, y)∗ = q(x),

ϕ(x, y)∗ ϕ(x, y) = q(y) for almost all (x, y) ∈ X × X .

proof of the claim. Since p is (ρg )g∈ -invariant, the function x → τ (p(x)) is -invariant and hence constantly equal to 0 < λ ≤ 1. Let p0 ∈ N be a projection with τ (p0 ) = λ. It follows that inside L∞ (X ) ⊗ N, the projections p and 1 ⊗ p0 are equivalent. Take a partial isometry v ∈ L∞ (X ) ⊗ N such that v ∗ v = p and vv ∗ = 1 ⊗ p0 . Deﬁne η(g, x) = v(g · x)ω(g, x)v(x)∗ and note that η is a 1-cocycle for X with values in U (p0 Np0 ). Set η0 (g, x, y) = η(g, x) and, for all n ≥ 1, ηn (g, x, y) = η0 (g, α2−n (x, y)). 2 Deﬁne the Hilbert space K = ∞ k=1 p0 L (N)p0 . We deﬁne the following 1-cocycle of X × X with values in the unitary group U (K ) of K : (c(g, x, y)ζ )k = η(g, x) ζk ηk (g, x, y)∗ . Deﬁne the map ξn : X × X → (K )1 by the formula ⎧ ⎨τ (p )−1/2 p if k = n , 0 0 (ξn (x, y))k = ⎩0 if k = n . One checks that ξn is a sequence of almost invariant unit vectors. Since X × X has property (T), we ﬁnd an invariant unit vector, that is, a measurable map ξ : X × X → K with ξ (x, y) = 1 for almost all (x, y) and for all g ∈ , ξ (g · x, g · y) = c(g, x, y) ξ (x, y)

436 / s. popa and s. vaes

almost everywhere. It follows that ξ (g · x, g · y)k = η(g, x) ξ (x, y)k ηk (g, x, y)∗ almost everywhere. In particular, for every k, the function (x, y) → ξ (x, y)k is -invariant and hence, constant. Since ξ (x, y) = 1 for almost all (x, y), we can pick k such that ξ (x, y)k = 0 for almost all (x, y). Taking the polar decomposition of ξ (x, y)k , we ﬁnd a nonzero partial isometry ψ ∈ L∞ (X × X ) ⊗ p0 Np0 satisfying ψ(g · x, g · y) ηk (g, x, y) = η(g, x) ψ(x, y) almost everywhere. We set ϕ0 (x, y) := v(x)∗ ψ(x, y)vk (x, y), where v0 (x, y) = v(x) and vk (x, y) = v0 (α2−k (x, y)). It follows that ω(g, x) ϕ0 (x, y) = ϕ0 (g · x, g · y) ωt0 (g, x, y) where t0 = 2−k . Set r(x, y) = ϕ0 (x, y)ϕ0 (x, y)∗ . It follows that r(g · x, g · y) = ω(g, x) r(x, y) ω(g, x)∗ almost everywhere. By Lemma 5.4, we ﬁnd a projection q ∈ L∞ (X ) ⊗ N such that r(x, y) = q(x) almost everywhere. Then, q is a nonzero projection in B and q ≤ p. Similarly, ϕ0 (x, y)∗ ϕ0 (x, y) = q0 (αt0 (x, y)) where q is a projection in L∞ (X ) ⊗ N and q0 (x, y) = q (x). Set q0 (x, y) = q(x) and qt (x, y) = q0 (αt (x, y)). We now construct ϕ1 : X × X → N such that ϕ1 (x, y)ϕ1 (x, y)∗ = q(x), ϕ1 (x, y)∗ ϕ1 (x, y) = q2t0 (x, y) and ω(g, x) ϕ1 (x, y) = ϕ1 (g · x, g · y) ω2t0 (g, x, y) almost everywhere. Continuing the same procedure k times (remember that t0 = 2−k ), we will have found ϕ = ϕk : X × X → N satisfying ϕ(x, y)ϕ(x, y)∗ = q(x), ϕ(x, y)∗ ϕ(x, y) = q1 (x, y) = q(y) and ω(g, x) ϕ(x, y) = ϕ(g · x, g · y) ω1 (g, x, y) = ϕ(g · x, g · y) ω(g, y) , hence proving the claim. In fact, it sufﬁces to take ϕ1 (x, y) = ϕ0 (x, y)ϕ0 (β(α2t0 (x, y)))∗ and to use that (αt )t∈R is a 1-parameter group, β ◦ αt = α−t ◦ β and β(x, y) ∈ {x} × Y for almost all (x, y). So, the claim above has been proven. Using the claim and a maximality argument, we ﬁnd a measurable function ϕ : X × X → U (N) such that ω(g, x) ϕ(x, y) = ϕ(g · x, g · y) ω(g, y)

cocycle and orbit superrigidity for lattices in sl(n, R) / 437

almost everywhere. Set H(x, y, z) = ϕ(x, y)ϕ(y, z). It follows that H(g · x, g · y, g · z) = ω(g, x) H(x, y, z) ω(g, z)∗ for almost all (x, y, z). By Lemma 5.4 and because the 4-fold diagonal action X × X × X × X is ergodic, H is essentially independent of its second variable. So, we ﬁnd a measurable F : X × X → U (N) such that ϕ(x, y) = F(x, z)ϕ(y, z)∗ for almost every (x, y, z). By the Fubini theorem, take z ∈ X such that the previous formula holds for almost all (x, y). Set ψ(x) = F(x, z)∗ and G(y) = ϕ(y, z)∗ . It follows that ψ(g · x) ω(g, x) ψ(x)∗ = G(g · y) ω(g, y) G(y)∗ almost everywhere. Hence, the left-hand side is independent of x and we have found a group morphism δ : G → U (N) and a measurable map ψ : X → U (N) such that ω(g, x) = ψ(g · x)∗ δ(g) ψ(x) almost everywhere. Consider the quotient Polish space U (N)/G with the induced metric, which is invariant under left multiplication by elements of U (N). Write F : X → U (N)/G : F(x) = ψ(x)G . It follows that F(g · x) = δ(g)F(x) almost everywhere. By Lemma 5.4, F is essentially constant. So, we ﬁnd a unitary u ∈ U (N) and a measurable map w : X → G such that ψ(x) = uw(x) almost everywhere. Replacing δ by u ∗ δ( · )u, it follows that ω(g, x) = w(g · x)∗ δ(g)w(x) almost everywhere. In particular, δ(g) ∈ G and we are done. As a principle (cf. [19, proposition 3.6]), once the restriction of a 1-cocycle ω : G × X → G to a closed subgroup H < G is cohomologous to a group morphism H → G and if H is sufﬁciently normal in G and acts sufﬁciently mixingly on X , the entire 1-cocycle ω is cohomologous to a group morphism G → G . In our setting, we need the following:

lemma 5.5. Let G (X , μ) be a nonsingular action of the l.c.s.c. group G and ω : G × X → G a 1-cocycle with values in the Polish group of ﬁnite type G . Let H < G be a closed subgroup and assume that ω|H is cohomologous to a group morphism H → G . If for every g ∈ G, the diagonal action of the group H ∩ gHg −1 on X × X is ergodic, then ω is cohomologous to a morphism G → G . Proof. We may assume that for every h ∈ H, we have ω(h, x) = δ(h) for almost every x ∈ X , where δ : H → G is a continuous group morphism. Let g ∈ G and put F(x) = ω(g, x). Using the cocycle equation, it follows that for all h ∈ H ∩

438 / s. popa and s. vaes

g −1 Hg, we have F(h · x) = δ(ghg −1 )F(x)δ(h)−1 almost everywhere. By Lemma 5.4, F is essentially constant. So, we have shown that for every g ∈ G, the map x → ω(g, x) is essentially constant. It follows that ω(g, x) = δ(g). After showing the following lemma, we can prove Theorem 1.3 and Proposition 5.1.

lemma 5.6. Let be a lattice in SL (n, R) and consider the linear action of on Rn . Let X (k) := Rn × · · · × Rn k times

be the k-fold diagonal action. Then, X (k) • is ergodic if and only if k ≤ n − 1, and • has property (T) if and only if k ≤ n − 3 or k ≥ n (the latter part being as

interesting as the trivial group {e} having property (T)). Proof. Writing the elements of Rn as column vectors, identify, up to measure 0, X (n) with GL (n, R), with the -action given by left multiplication. The determinant function is invariant and not essentially constant, proving that X (k) is nonergodic for k ≥ n. It also follows that for k ≥ n, the action X (k) is essentially free and proper (see Deﬁnition 3.4). Hence, it has property (T) because the trivial group has property (T). Now let k ≤ n − 1. Denoting by (ei )i=1,...,n the standard basis vectors in Rn , the orbit of (e1 , . . . , ek ) ∈ X (k) under the diagonal SL (n, R)-action has complement of measure 0, so that we can identify X (k) with SL (n, R)/H, where H = {A ∈ SL (n, R) | Aei = ei for all i = 1, . . . , k}. Observe that H ∼ = SL (n − k, R) Mn−k (R), where Mn−k (R) denotes the additive group of (n − k) × (n − k) matrices on which SL (n − k, R) acts by multiplication. By Moore’s ergodicity theorem (see e.g., [26, theorem 2.2.6]), H \ SL (n, R) is ergodic and hence, SL (n, R)/H is ergodic. By Proposition 3.5, property (T) of SL (n, R)/H is equivalent with property (T) of H \ SL (n, R). By Proposition 3.2, the latter is equivalent with the group H having property (T), which is in turn equivalent with n − k ≥ 3.

Proof of Theorem 1.3. Observe that part (1) is a special case of part (3), by taking k = 1 and H = {1}. We start by proving part (3). Set X = Mn,k (R). The action

cocycle and orbit superrigidity for lattices in sl(n, R) / 439

X by left multiplication is s-malleable. It sufﬁces to take αt (A, B) = ( cos (πt/2)A + sin (πt/2)B, − sin (πt/2)A + cos (πt/2)B)

and

β(A, B) = (A, −B) whenever t ∈ R and A, B ∈ Mn,k (R). Moreover, X can be viewed as the k-fold diagonal action Rn × · · · × Rn . By Lemma 5.6 and because n ≥ 4k + 1, the diagonal action X × X has property (T) and the 4-fold diagonal action X × X × X × X is ergodic. So, by Theorem 5.3, X is Uﬁn -cocycle superrigid. Since the diagonal action X × X is ergodic and is a normal subgroup of G, Lemma 5.5 implies that G X is Uﬁn -cocycle superrigid. It remains to prove part (2) of the theorem. By part (1), we already know Rn is Uﬁn -cocycle superrigid. Deﬁne, for every x ∈ Zn , x = {g ∈ | gx = x}. We claim that the diagonal action x Rn × Rn is ergodic for every x ∈ Zn . Once this claim is proven, Lemma 5.5 implies that Zn Rn is Uﬁn -cocycle superrigid. For x = 0, the claim follows from Lemma 5.6. Now let x = 0. Deﬁne the closed subgroup H = {g ∈ SL (n, R) | ge1 = e1 } of SL (n, R). Exactly as in the proof of Lemma 5.6, when n ≥ 4, the diagonal action Rn × Rn of any lattice ⊂ H is ergodic. Take g0 ∈ SL (n, Q) with x = g0 e1 . Since is a ﬁnite index subgroup of SL (n, Z), it follows that g0−1 x g0 contains a ﬁnite index subgroup of {g ∈ SL (n, Z) | ge1 = e1 } and hence, is a lattice in H. So, its diagonal action on Rn × Rn is ergodic. Then, the same is true for the diagonal action of x on Rn × Rn . Proof of Proposition 5.1. Write G = SL (n, R) and = SL (n, Z). As before, identify Rn with G/H, where H ∼ = SL (n − 1, R) Rn−1 . Let θ : G/H → G be a Borel lifting. Deﬁne the 1-cocycle ω : × G/H → H by gθ (xH) = θ(gxH)ω(g, xH) whenever g ∈ and xH ∈ G/H. Assume that ω is cohomologous to a group morphism δ : → H. This means that we can choose the lifting θ such that gθ (xH) = θ (gxH)δ(g) for g ∈ and almost all xH ∈ G/H. The image of any group morphism SL (n, Z) → SL (n − 1, R) Rn−1 is ﬁnite (see [23, theorem 6] for an elementary argument). So, we have found a ﬁnite index subgroup 0 ⊂ SL (n, Z) and a measurable map θ : Rn → SL (n, R) such that θ(gx) = gθ(x) for all g ∈ 0 and almost all x ∈ Rn . It follows that the map (x, y) → θ(x)−1 θ(y) is invariant under the diagonal 0 -action,

440 / s. popa and s. vaes

which is ergodic by Lemma 5.6. Hence, θ is essentially constant, which is a contradiction with the formula θ(gx) = gθ(x). 5.2. Proofs of Theorems 1.5 and 1.6

We will deduce Theorems 1.5 and 1.6 from the general Theorem 5.8 below, dealing with arbitrary actions of the form Mn,k (R)/H, where is a lattice and H < GL (k, R) a closed subgroup. First, observe that these actions are essentially free and ergodic.

lemma 5.7. Let n > k and < SL (n, R) any lattice. Let H ⊂ GL (k, R) be a closed subgroup. If ( − 1, −1) ∈ × H, put 0 = /{±1}, otherwise put = 0 . The action 0 Mn,k (R)/H is essentially free and ergodic. It never admits an invariant probability measure. It admits an inﬁnite invariant measure if and only if H is unimodular and satisﬁes det g = ±1 for all g ∈ H. Proof. By Lemma 5.6, Mn,k (R) is ergodic, because k < n. A fortiori, Mn,k (R)/H is ergodic. Denote V = span{e1 , . . . , ek } ⊂ Rn and deﬁne the closed subgroup H1 < SL (n, R) by H1 = {g ∈ SL (n, R) | gV = V and g|V ∈ H } . We can identify 0 Mn,k (R)/H with SL (n, R)/H1 . From this description, essential freeness follows. Also the statement about invariant measures follows, because H1 is unimodular if and only if H is unimodular and satisﬁes det g = ±1 for all g ∈ H.

theorem 5.8. Let H ⊂ GL (k, R) be a closed subgroup and (Y , η) any essentially free, ergodic, nonsingular action of the countable group . Suppose n ≥ 4k + 1. case −1 ∈ H . Let ⊂ PSL (n, R) be a lattice and put := {±1} · ⊂ GL (n, R). The actions Mn,k (R)/H and Y are SOE if and only if Y is conjugate to an induction of one of the following actions: 1) × H/N Mn,k (R)/N, where N H is an open normal subgroup with −1 ∈ N, or 2)

× H/N Mn,k (R)/N, where N H is an open normal subgroup with {±(1, 1)} −1 ∈ N.

cocycle and orbit superrigidity for lattices in sl(n, R) / 441

case −1 ∈ H . Let ⊂ SL (n, R) be a lattice. The actions Mn,k (R)/H and Y are SOE if and only if Y is conjugate to an induction of one of the following actions: 1) × H/N Mn,k (R)/N, where N H is an open normal subgroup, or H 2)(only when −1 ∈ ) × Mn,k (R)/({±1} · N), where N H is {±1} N an open normal subgroup.

remark 5.9. Given a stable orbit equivalence : Mn,k (R)/H → Y between the actions Mn,k (R)/H and Y , Theorem 5.8 provides a conjugacy between one of the listed actions and 1 Y1 such that Y is induced from 1 Y1 . In fact, moreover, one has (x) ∈ · (x) for almost every x ∈ Mn,k (R). If : X0 → Y0 is a SOE between X and Y , we can (and will tacitly) extend to a countable to 1, measurable : X → Y satisfying (g · x) ∈ · (x) for all g ∈ and almost all x ∈ X . If : X → Y is a SOE between the essentially free actions X and Y , we get a Zimmer 1-cocycle ω for the action X with values in , determined by the formula (g · x) = ω(g, x) · (x) almost everywhere. As a general principle, if the 1-cocycle ω is cohomologous to a group morphism → , the stable orbit equivalence is “essentially” given by a conjugacy of the actions: see [26, proposition 4.2.11], [24, lemma 4.7], and [8, theorem 1.8]. In our framework of non-p.m.p. actions, we again need such a principle: see Lemma 5.10 below. Recall from Deﬁnition 3.4 the notion of an essentially free and proper action. σ

lemma 5.10. Let G be an l.c.s.c. group and G (X , μ) a nonsingular, essentially free, ergodic action. Assume that σ is Uﬁn -cocycle superrigid. Let N G be an open normal subgroup such that the restricted action σ |N is proper. Let (Y , η) be an essentially free, ergodic, nonsingular action. If : X /N → Y is an SOE between the actions G/N X /N and Y , then there exists • a subgroup 1 < and a nonnegligible Y1 ⊂ Y such that Y is induced

from 1 Y1 , and • an open normal subgroup N1 G such that σ |N is proper 1

442 / s. popa and s. vaes

such that the actions G/N1 X /N1 and 1 Y1 are conjugate through the nonsingular isomorphism : X /N1 → Y1 and the group isomorphism δ : G/N1 → 1 . Furthermore, (x) ∈ · (x) for almost all x ∈ X . Proof. Let : X /N → Y be an SOE. By cocycle superrigidity of σ , take a measurable map ϕ : X → such that, writing (x) = ϕ(x)−1 · (x), we have (g · x) = δ(g) · (x) almost everywhere, where δ : G → is a continuous group morphism. Put N1 = Ker δ. So, N1 is an open normal subgroup of G. Then, N ∩ N1 is still open in G and we consider X /(N ∩ N1 ) with the quotient map π : X /(N ∩ N1 ) → X /N. It follows that we can view as a measurable map : X /(N ∩ N1 ) → Y such that (π (x)) ∈ · (x) for almost all x ∈ X /(N ∩ N1 ) and (g · x) = δ(g) · (x) almost everywhere. Using the facts that the countable group N/(N ∩ N1 ) acts freely and properly on X /(N ∩ N1 ), that is countable, and that : X /N → Y is locally a nonsingular isomorphism, it follows that X /(N ∩ N1 ) can be partitioned in a sequence of nonnegligible subsets (Un )n such that for every n, |Un is a nonsingular isomorphism between Un and a nonnegligible subset of Y . But then, for every nontrivial element g ∈ N1 /(N ∩ N1 ) and every n, we conclude that g · Un ∩ Un has measure 0. It follows that N1 /(N ∩ N1 ) acts freely and properly on X /(N ∩ N1 ). So, N1 acts freely and properly on X . Hence, we can form the quotient space X /N1 and view as a measurable map : X /N1 → Y such that (x) ∈ · (x) for almost all x ∈ X and (g · x) = δ(g) · (x) almost everywhere. Now, δ : G/N1 → is an injective group morphism. Still, X /N1 can be partitioned into a sequence of nonnegligible subsets (Un ) such that |Un is a nonsingular isomorphism between Un and Vn ⊂ Y . We claim that Vn ∩ Vm has measure 0 for every n = m. If this is not the case, take W ⊂ Un and W ⊂ Um nonnegligible and a nonsingular isomorphism ρ : W → W such that (ρ(x)) = (x) for x ∈ W . Since is an SOE, ρ(x) ∈ (G/N1 ) · x for almost all x ∈ W . Hence, making W smaller but still nonnegligible, we may assume that ρ(x) = g · x for all x ∈ W and some g ∈ G/N1 . Since W ∩ W has measure 0 and the action of G/N1 on X /N1 is essentially free, we get g = e. But also, δ(g) · (x) = (x) for almost all x ∈ W . This is a contradiction with the injectivity of δ and the essential freeness of Y . This proves the claim and we have found that is a nonsingular isomorphism between X /N1 and a nonnegligible subset Y1 ⊂ Y . Set 1 = δ(G/N1 ). It remains to prove that Y is induced from 1 Y1 . So, let h ∈ and assume that h · Y1 ∩ Y1 is nonnegligible. We have to prove that h ∈ 1 . By our assumption, take W , W ⊂ X /N1 nonnegligible and a nonsingular isomorphism ρ : W → W such that h · (x) = (ρ(x)) for all x ∈ W . Since

cocycle and orbit superrigidity for lattices in sl(n, R) / 443

was an SOE, we can make W smaller but still nonnegligible and assume that ρ(x) = g · x for all x ∈ W and some g ∈ G/N1 . But then, h · (x) = δ(g) · (x) for almost all x ∈ W . Since Y is essentially free, it follows that h = δ(g) ∈ 1 . Proof of Theorem 5.8. We prove only the case −1 ∈ H; the case −1 ∈ H being analogous. By Theorem 1.3, the action of G := ( × H)/{±(1, 1)} on Mn,k (R) is Uﬁn -cocycle superrigid. By Lemma 5.10, we have only to prove that the following subgroups of × H are the only open normal subgroups containing ( − 1, −1) and acting properly on Mn,k (R): • {±1} × N, where N H is an open normal subgroup with −1 ∈ N, and • ({1} × N) ∪ ({−1} × −N), where N H is an open normal subgroup with

−1 ∈ N. × H) be a closed normal subgroup acting properly on Mn,k (R). It So, let N ( is sufﬁcient to prove that N ⊂ {±1} × H. Suppose that N ⊂ {±1} × H and take (g, h) ∈ N with g = ±1. Take k ∈ such that the commutator t := kgk−1 g −1 = ±1. It follows that (t, 1) ∈ N. By Margulis’s normal subgroup theorem [15], we have 0 × {1} ⊂ N for some ﬁnite index subgroup 0 < . By Lemma 5.6, 0 acts ergodically on Mn,k (R), contradicting the properness of N Mn,k (R). Proof of Theorem 1.5. Statement (1) follows immediately from Theorem 5.8. We now prove statement (2). We claim that the following are the only normal subgroups N of SL (n, Z) Zn that act properly on Rn : • N = {e}, • N = λZn for some λ ∈ N \ {0}, and • (only when n is even) N = {±1} λZn for λ ∈ {1, 2}.

Statement (2) of Theorem 1.5 then follows from Lemma 5.10 and Theorem 1.3, where the Uﬁn -cocycle superrigidity of the afﬁne action SL (n, Z) Zn Rn was established. So, let N SL (n, Z) Zn be a normal subgroup acting properly on Rn . Suppose ﬁrst that N ⊂ Zn . Taking the commutator of (g, x) ∈ N with g = 1 and an arbitrary (1, y), y ∈ Zn , it follows that H := N ∩ Zn = {0}. Hence, H is a nonzero, globally SL (n, Z)-invariant subgroup of Zn . So, H = λZn for some λ ∈ N \ {0}. If N ⊂ {±1} Zn , it would follow that N has ﬁnite index in SL (n, Z) Zn , contradicting the properness of N Rn . So, we have shown that in all cases N ⊂ {±1} Zn . It is now straightforward to deduce the above list of possibilities for N.

444 / s. popa and s. vaes

Proof of Theorem 1.6. This theorem is a special case of Theorem 5.8.

5.3. Describing All 1-Cocycles of Quotient Actions

Finally, cocycle superrigidity of G (X , μ) allows us to describe all 1-cocycles for G/N X /N when N G is a closed normal subgroup of G acting essentially freely and properly on X . We start with the following proposition, closely related to [21, lemma 5.3], and illustrate it with two examples. σ

proposition 5.11. Let G be an l.c.s.c. group and G (X , μ) a nonsingular action. Let N G be a closed, normal subgroup such that the restriction σ |N is essentially free and proper. Assume that σ is Uﬁn -cocycle superrigid. Choose a measurable map π : X → N satisfying π(g · x) = gπ(x) for almost all (g, x) ∈ N × X . Denote by g → g and x → x the quotient maps G → G/N, respectively, X → X /N. Then, ω:

G X × → G : ω(g, x) = π (g · x)−1 gπ (x) N N

is a well-deﬁned 1-cocycle. Every 1-cocycle for the action G/N X /N with values in a Polish group of ﬁnite type G is cohomologous to δ ◦ ω for a continuous group morphism δ : G → G . If the diagonal action G X × X is ergodic, δ is uniquely determined up to conjugacy by an element of g ∈ G . Proof. Let : G/N × X /N → G be a 1-cocycle with values in the Polish group of ﬁnite type G . From cocycle superrigidity of σ , let ϕ : X → G be a measurable map and δ : G → G a continuous group morphism such that (g, x) = ϕ(g · x)−1 δ(g)ϕ(x) almost everywhere. Replacing g by hg, h ∈ N, it follows that ϕ(hg · x)−1 δ(h) = ϕ(g · x)−1 and hence, ϕ(h · x) = δ(h)ϕ(x) almost everywhere. So, we can deﬁne (x) = δ(π (x))−1 ϕ(x). By construction, makes cohomologous to δ ◦ ω. The uniqueness of δ follows directly from Lemma 5.4.

example 5.12. Let ⊂ SL (n, Z) be a ﬁnite index subgroup and consider the action Tn . • Choosing a measurable map p : Rn → Zn such that p(x + y) = x + p(y)

for all x ∈ Zn and almost all y ∈ Rn , the formula ω : × Rn /Zn → Zn : ω(g, x) = p(g · x)−1 gp(x) deﬁnes a 1-cocycle for Tn with values in Zn .

cocycle and orbit superrigidity for lattices in sl(n, R) / 445 • Every 1-cocycle with values in a Polish group of ﬁnite type G is coho-

mologous with δ ◦ ω for a group morphism δ : Zn → G , uniquely determined up to conjugacy by an element in G . Note that by Zimmer’s cocycle superrigidity theorem [26, theorem 5.2.5], any Zariski-dense 1-cocycle for any ergodic p.m.p. action SL (n, Z) (X , μ), n ≥ 3, taking values in a connected simple real algebraic noncompact centerfree group, is cohomologous to a group morphism. For the speciﬁc action SL (n, Z) Rn /Zn , n ≥ 5, the previous example provides an explicit description of all 1-cocycles of SL (n, Z) Rn /Zn with values in an arbitrary Polish group of ﬁnite type.

example 5.13. Let X be the real ﬂag manifold of signature (d1 , . . . , dl , n), ⊂ PSL (n, R) a lattice and n ≥ 4dl + 1. We obtain as follows all 1-cocycles for the action X with values in a Polish group of ﬁnite type. • Identify X = Mn,k (R)/H and choose a measurable map p : Mn,k (R) →

H satisfying p(Ag) = p(A)g for almost all A ∈ Mn,k (R), g ∈ H. Let

= {±1} · be the double cover of in GL (n, R) and deﬁne G := ( × H)/{±(1, 1)}. The formula ω : × X → G : ω(g, x) = (g, p(gx)p(x)−1 )

mod {±(1, 1)}

deﬁnes a 1-cocycle with values in G. Here → : g → g and Mn,k (R) → X : x → x are the quotient maps. • Every 1-cocycle with values in a Polish group of ﬁnite type G is cohomologous with δ ◦ ω for a continuous group morphism δ : G → G , uniquely determined up to conjugacy by an element in G .

6. Classification up to Orbit Equivalence Combining the results of [9] with Theorem 5.8, we classify up to stable orbit equivalence, the linear lattice actions Rn , as well as the natural lattice actions on ﬂag manifolds. At the same time, we compute the outer automorphism group of the associated orbit equivalence relations. We start with the following elementary lemma.

lemma 6.1. Let (X , μ) be a nonsingular action of the countable group and assume that the diagonal action X × X is ergodic. Then, X is not induced, that is, if X is induced from 1 X1 , then 1 = and μ(X \ X1 ) = 0.

446 / s. popa and s. vaes

Proof. Assume that X is induced from 1 X1 . So, we ﬁnd a quotient map π : X → / 1 satisfying π (g · x) = gπ (x) almost everywhere and X1 = π −1 (e1 ). Hence, the subset {(x, y) ∈ X × X | π (x) = π(y)} is nonnegligible and -invariant. By the ergodicity of X × X , it follows that π (x) = π (y) for almost all (x, y) ∈ X × X . This means that 1 = and μ(X \ X1 ) = 0. If R is a II1 equivalence relation on (X , μ), we denote by [R] the full group of R consisting of nonsingular automorphisms : X → X satisfying (x, (x)) ∈ R for almost every x ∈ X . Then, [R] is a normal subgroup of the automorphism group Aut (R) of R. The quotient group is denoted by Out (R) and called the outer automorphism group of R. The full pseudogroup of R is denoted by [[R]] and consists of nonsingular partial isomorphisms φ : X0 ⊂ X → X1 ⊂ X satisfying (x, φ(x)) ∈ R for almost every x ∈ X0 . We denote X0 = D(φ) and X1 = R(φ).

theorem 6.2. Let n ≥ 5 and ⊂ SL (n, R) a lattice. Let n ≥ 2 and ⊂ SL (n , R) a lattice. If the nonsingular isomorphism : X1 ⊂ Rn → X1 ⊂ Rn is an SOE between Rn and Rn , then • n = n and there exists A ∈ GL (n, R) such that = AA−1 , and • there exists φ ∈ [[R( Rn )]] with R(φ) = X such that (x) = φ(A(x)) 1

for almost every x ∈ X1 .

In particular, Out (R( Rn )) = NGL (n,R) ()/ .

Proof. Let : X1 ⊂ Rn → X1 ⊂ Rn be an SOE between Rn and Rn . Since for n = 2 the equivalence relation R( Rn ) is hyperﬁnite, we have n ≥ 3 and hence the diagonal action Rn × Rn is ergodic. By Lemma 6.1, Rn is not an induced action. By Theorem 1.5, the action Rn is conjugate with either Rn or, in case n is even, /{±1} Rn /{±1}. By Mostow rigidity, the latter is impossible since ∼ = /{±1}. In the former case, we already conclude n = n and we have found a nonsingular isomorphism : Rn → Rn and a group isomorphism δ : → satisfying • (g · x) = δ(g) · (x), and • (x) ∈ · (x),

for all g ∈ and almost all x ∈ Rn . Denoting by B the transpose of the matrix B, by Mostow rigidity, we ﬁnd A ∈ GL (n, R) such that (a) δ(g) = AgA−1 for all g ∈ , or (b) δ(g) = A(g )−1 A−1 for all g ∈ .

cocycle and orbit superrigidity for lattices in sl(n, R) / 447

Deﬁne the subgroup H ⊂ SL (n, R) consisting of matrices g with ge1 = e1 and identify Rn = SL (n, R)/H. In case (b), we would get a conjugacy between the -actions on SL (n, R)/H and SL (n, R)/H , which is ruled out by [9, theorem D]. In case (a), multiplying A by a nonzero scalar if necessary, [9, theorem D] implies that (x) = Ax for almost all x ∈ Rn . Deﬁning the partial isomorphism φ := ◦ A−1 with R(φ) = X1 , it follows that φ ∈ [[R( Rn )]] and that (x) = φ(A(x)) for almost every x ∈ X1 . Let X be the real ﬂag manifold with signature d := (d1 , . . . , dl , n). Denote d := (n − dl , n − dl−1 , . . . , n − d1 , n) . If X is the real ﬂag manifold with signature d , there is a natural diffeomorphism X → X : x → x satisfying g · x = (g )−1 · x for all x ∈ X , g ∈ SL (n, R).

theorem 6.3. Let ⊂ PSL (n, R) be a lattice and X the real ﬂag manifold with signature d := (d1 , . . . , dl , n). Assume that n ≥ 4dl + 1. Let ⊂ PSL (n , R) be a lattice and X the real ﬂag manifold with signature d := (d1 , . . . , dl , n ). If the nonsingular isomorphism : X1 ⊂ X → X1 ⊂ X is an SOE between X and X , then n = n and there exists A ∈ PGL (n, R), φ ∈ [[R( X )]] with R(φ) = X1 such that either • d = d, = AA−1 , (x) = φ(A(x)) for almost every x ∈ X1 , or • d = d , = A A−1 , (x) = φ(A(x)) for almost every x ∈ X1 .

In particular, Out (R( X )) = NPGL (n,R) ()/ . Proof. Let : X1 ⊂ X → X1 ⊂ X be an SOE between X and X . We ﬁrst prove that the diagonal action X × X is ergodic. If Y denotes the ﬂag manifold of signature (1, 2, . . . , n ), the action X is a quotient of the action Y . So, it sufﬁces to prove ergodicity of Y × Y . Denoting by D ⊂ SL (n , R) the subgroup of diagonal matrices, Y × Y can be identiﬁed with SL (n , R)/D, which follows ergodic because D SL (n , R)/ is ergodic by Moore’s ergodicity theorem. So, by Lemma 6.1, X is not an induced action. Since has trivial center, Theorem 1.6 yields a group isomorphism δ : → and a nonsingular isomorphism : X → X satisfying (g · x) = δ(g) · (x) and (x) ∈ · (x) almost everywhere. By Mostow rigidity, n = n and there exists A ∈ PGL (n, R) such that either δ(g) = AgA−1 or δ(g) = A(g )−1 A−1 for all g ∈ .

448 / s. popa and s. vaes

Set k1 = d1 and ki = di − di−1 for 2 ≤ i ≤ l. Put k = dl and deﬁne the closed subgroups H and H1 of GL (k, R) by ⎫ ⎧⎛ ⎞ A11 ∗ ··· ∗ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎟ ⎪ ⎪ ⎬ ⎨⎜ 0 A · · · ∗ ⎟ ⎜ 22 ⎟ ⎜ Aii ∈ GL (ki , R) and det (Aii ) = 0 , H := ⎜ . . . . ⎟ .. .. .. ⎠ ⎪ ⎪ ⎪ ⎪ ⎝ .. ⎪ ⎪ ⎪ ⎪ ⎭ ⎩ 0 0 · · · All ⎧⎛ A ⎪ ⎪ ⎪⎜ 11 ⎪ ⎨⎜ 0 H1 := ⎜ ⎜ .. ⎪ ⎪ ⎝ . ⎪ ⎪ ⎩ 0

∗ A22 .. . 0

⎫ ⎞ ∗ ⎪ ⎪ ⎪ ⎟ ⎪ ⎬ ∗⎟ ⎟ .. ⎟ Aii ∈ GL (ki , R) and det (Aii ) = ±1⎪ . ⎪ . ⎠ ⎪ ⎪ ⎭ · · · All

··· ··· .. .

We identify H/H1 = Rl+ . From now on, we write X as Mn,k (R)/H. Deﬁne analogously the subgroups H , H1 of GL (k , R) and write X as Mn,k (R)/H . Whenever N H is a closed normal subgroup containing −1, the quotient morphism H → H/N gives rise, as in Example 5.13, to a 1-cocycle ωN for the action X with values in H/N such that the action Mn,k (R)/N can be identiﬁed with the action X × H/N given by g · (x, h) = (g · x, ωN (g, x)h) . Since Mn,k (R)/N is ergodic, the 1-cocycle ωN cannot be cohomologous to a 1-cocycle taking values in a proper closed subgroup of H/N. We similarly deﬁne the 1-cocycles ωN for X . Since δ, conjugate the actions X and X , the map μ(g, x) = ωH1 (δ(g), (x)) deﬁnes a 1-cocycle for X , with values in Rl+ and with the property of not being cohomologous to a 1-cocycle taking values in a proper closed subgroup of Rl+ . We now apply Example 5.13, describing all Uﬁn -valued 1-cocycles for X , and note that Rl+ belongs to Uﬁn . Every group mor phism → Rl+ is trivial and every continuous group morphism H → Rl+ is trivial on H1 . So, we ﬁnd a continuous group morphism ρ : Rl+ → Rl+ such that μ is cohomologous to ρ ◦ ωH1 . Since μ cannot be cohomologous to a 1-cocycle taking values in a proper closed subgroup of Rl+ , it follows that ρ is onto. Altogether, we ﬁnd a closed normal subgroup N H, containing H1 , and a continuous isomorphism ρ : H/N → H /H1 such that μ is cohomologous to ρ ◦ ωN . It follows that there exists a nonsingular isomorphism 1 : Mn,k (R)/N → Mn ,k (R)/H1 satisfying 1 (g · x) = δ(g) · 1 (x) and 1 (x)H = (xH) almost everywhere.

cocycle and orbit superrigidity for lattices in sl(n, R) / 449

Since the action of on Mn ,k (R)/H1 is inﬁnite measure preserving, Lemma 5.7 implies that N = H1 . We saw already that either δ(g) = AgA−1 or δ(g) = A(g )−1 A−1 for all g ∈ . In the former case, [9, theorem D] implies that d = d and that there exists B ∈ H such that 1 (x) = AxB for almost every x ∈ Mn,k (R)/H1 . It follows that (x) = A(x) for almost every x ∈ X . In the latter case, we prove analogously that d = d and (x) = A(x).

7. Implementation by Group Actions In [5, p. 292], the question is raised whether every II1 equivalence relation can be implemented by an essentially free action of a countable group. This question has been settled in the negative in [7, theorem D]. In Proposition 7.1 below, we give examples of II1 equivalence relations R on (X , μ) with the following much stronger property: whenever (Y , η) is a free, nonsingular action and : X → Y is a measurable map satisfying (x) ∈ · (y) for almost all (x, y) ∈ R, then there exists y0 ∈ Y such that (x) ∈ · y0 for almost all x ∈ X . Among other examples, [7, theorem D] proves that the restriction of the orbit equivalence relation SL (n, Z) Rn /Zn to a subset of irrational measure, provides a II1 equivalence relation that cannot be implemented by a free action of a group. By [19, theorem 0.3], if [0, 1] is the Bernoulli action of a property (T) group without ﬁnite normal subgroups, the restriction of its orbit equivalence relation to any subset of measure strictly between 0 and 1 is unimplementable by a free action. But [7, theorem D] also provides examples of II1 equivalence relations R such that none of the ampliﬁcations Rt , t > 0, can be implemented by a free action. These equivalence relations are constructed using the following method. Suppose that G is an l.c.s.c. unimodular group and G (X , μ) an essentially free, properly ergodic p.m.p. action. There exists a Borel set Y ⊂ X , a probability measure η on Y , and a neighborhood U of e in G that we equip with a multiple of the Haar measure, such that U × Y → X : (g, y) → g · y provides a measure preserving isomorphism of U × Y onto a nonnegligible subset of X . The restriction of the orbit equivalence relation of G (X , μ) to Y is a II1 equivalence relation on (Y , η). One calls Y ⊂ X a measurable cross section for G (X , μ). A different choice of measurable cross section yields a stably isomorphic II1 equivalence relation. By Theorem 4.1, the restriction of the orbit equivalence relation of SL (n, Z) Rn to a subset of ﬁnite measure, provides other examples of II1 equivalence relations R such that none of the ﬁnite ampliﬁcations can

450 / s. popa and s. vaes

be implemented by a free action. In fact, one can show that R arises as the measurable cross section for the action of SL (n − 1, R) Rn−1 on SL (n, R)/ SL (n, Z). Nevertheless, this example is not covered by [7, theorem D], since SL (n − 1, R) Rn−1 is not semisimple.

proposition 7.1. Let G be an l.c.s.c. connected, unimodular group with normal closed subgroup G0 having the relative property (T). Let HR be a real Hilbert space and π : G → O (HR ) an orthogonal representation. Assume that π is injective and that the restriction of π to G0 is weakly mixing (i.e., has no ﬁnite-dimensional invariant subspaces). Denote by G (X , μ) the associated Gaussian action (see e.g., [8, section 2.7]). Choose a measurable cross-section X1 ⊂ X and denote by R the associated II1 equivalence relation on X1 . If (Y , η) is a free, nonsingular action and : X1 → Y is a measurable map satisfying (x) ∈ · (y) for almost all (x, y) ∈ R, then there exists y0 ∈ Y such that (x) ∈ · y0 for almost all x ∈ X1 . Proof. Choose a measurable map p : X → X1 such that p(x) ∈ G · x for almost all x ∈ X . Deﬁne the 1-cocycle ω : G × X → such that (p(g · x)) = ω(g, x) · (p(x)) almost everywhere. As observed in [8], theorem 0.1 in [19] applies to G (X , μ). Since G is connected, every group morphism from G to is trivial and we ﬁnd a measurable map ϕ : X → such that ω(g, x) = ϕ(g · x)−1 ϕ(x). So, the map x → ϕ(x) · (p(x)) is G-invariant and hence, essentially constant. We therefore ﬁnd y0 ∈ Y such that (p(x)) ∈ · y0 for almost all x ∈ X . This concludes the proof of the proposition.

References [1] [2] [3] [4] [5] [6] [7] [8]

C. Anantharaman-Delaroche, Cohomology of property T groupoids and applications. Ergod. Th. & Dynam. Sys. 25 (2005), 977–1013. P.L. Aubert, Deux actions de SL (2, Z). In Ergodic theory (Les Plans-sur-Bex, 1980), Monograph. Enseign. Math. 29, Univ. Genève, Geneva, 1981, pp. 39–46. B. Bekka, P. de la Harpe, and A. Valette, Kazhdan’s property (T ). New Mathematical Monographs 11. Cambridge University Press, Cambridge, 2008. A. Connes, Classiﬁcation of injective factors. Ann. Math. 104 (1976), 73–115. J. Feldman and C.C. Moore, Ergodic equivalence relations, cohomology, and von Neumann algebras I, II. Trans. Amer. Math. Soc. 234 (1977), 289–324, 325–359. A. Furman, Gromov’s measure equivalence and rigidity of higher rank lattices. Ann. of Math. 150 (1999), 1059–1081. A. Furman, Orbit equivalence rigidity. Ann. of Math. 150 (1999), 1083–1108. A. Furman, On Popa’s cocycle superrigidity theorem. Int. Math. Res. Not. IMRN 19 (2007), Art. ID rnm073, 46 pp.

cocycle and orbit superrigidity for lattices in sl(n, R) / 451 [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22]

[23]

[24] [25] [26]

A. Furman, Measurable rigidity of actions on inﬁnite measure homogeneous spaces, II. J. Amer. Math. Soc. 21 (2008), 479–512. D. Gaboriau, Coût des relations d’équivalence et des groupes. Invent. Math. 139 (2000), 41–98. D. Gaboriau, Invariants l 2 de relations d’équivalence et de groupes. Publ. Math. Inst. Hautes Études Sci. 95 (2002), 93–150. S.L. Gefter and V.Ya. Golodets, Fundamental groups for ergodic actions and actions with unit fundamental groups. Publ. Res. Inst. Math. Sci. 24 (1988), 821–847. A. Ioana, Cocycle superrigidity for proﬁnite actions of property (T) groups (2008), arXiv:0805.2998. To appear in Duke Math. J. A.S. Kechris, Classical descriptive set theory. Graduate Texts in Mathematics 156. Springer-Verlag, New York, 1995. G.A. Margulis, Discrete subgroups of semisimple Lie groups. Ergebnisse der Mathematik und ihrer Grenzgebiete (3) 17. Springer-Verlag, Berlin, 1991. C.C. Moore, Group extensions and cohomology for locally compact groups. III. Trans. Amer. Math. Soc. 221 (1976), 1–33. M. Pichot, Sur la théorie spectrale des relations d’équivalence mesurées. Journal of the Inst. of Math. Jussieu 6 (2007), 453–500. S. Popa, On the superrigidity of malleable actions with spectral gap. J. Amer. Math. Soc. 21 (2008), 981–1000. S. Popa, Cocycle and orbit equivalence superrigidity for malleable actions of w-rigid groups. Invent. Math. 170 (2007), 243–295. S. Popa, Correspondences, INCREST preprint No 56, 1986 (unpublished). S. Popa and S. Vaes, Strong rigidity of generalized Bernoulli actions and computations of their symmetry groups. Adv. Math. 217 (2008), 833–872. S. Popa and S. Vaes, On the fundamental group of II1 factors and equivalence relations arising from group actions. To appear in Quanta of Maths, Proceedings of the Conference in honor of Alain Connes, Clay Math. Inst., 2010. R. Steinberg, Some consequences of the elementary relations in SLn . In Finite groups— coming of age (Montreal, 1982). Contemp. Math. 45, American Mathematical Society, Providence, RI, 1985, pp. 335–350. S. Vaes, Rigidity results for Bernoulli actions and their von Neumann algebras (after Sorin Popa). Séminaire Bourbaki, exp. no. 961. Astérisque 311 (2007), 237–294. R.J. Zimmer, On the cohomology of ergodic actions of semisimple Lie groups and discrete subgroups. Amer. J. Math. 103 (1981), 937–951. R.J. Zimmer, Ergodic theory and semisimple groups. Birkhäuser, Boston, 1984.

PART 3

Geometric Group Theory

14

emmanuel breuillard

HEIGHTS ON SL2 AND FREE SUBGROUPS dedictated to bob zimmer on the occasion of his

60 th

birthday

Abstract We discuss and survey the strong uniform Tits Alternative [8] and give a complete proof of it in the special case of GL2 (C). The main arithmetic ingredient, the height gap theorem [7], which can be seen as a nonabelian analogue to the Lehmer conjecture, is also given a complete treatment in that case. We then give several applications involving expansion properties of SLd (Z/pZ), a uniform l2 spectral gap, and diophantine properties of subgroups of GLd (C).

Contents

1. Introduction 2. Spectral Radius Lemma for Several Matrices 3. Height, Arithmetic Spectral Radius and Minimal Height 4. Height Gap Theorem 5. Simultaneous Quasisymmetrization over Q 6. Ping-pong 7. Proof of Theorems 1.1 and 1.3 for SL2 (C) 8. Uniform Spectral Gap in 2 and Cogrowth of Subgroups 9. Large Girth References

455 464 468 470 476 478 480 484 486 492

1. Introduction 1.1. Statement of the Main Results

Recall that the Tits alternative [31] asserts that any ﬁnitely generated subgroup of GLd (K ), where K is some ﬁeld, contains a nonabelian free subgroup on two generators unless it is amenable, or equivalently in this case, unless it contains a solvable subgroup of ﬁnite index (i.e., is virtually solvable). In [7] and [8], we showed the following uniform version of Tits’s theorem : The author acknowledges support from the IAS Princeton through NSF Grant DMS0635607.

455

456 / e. breuillard

theorem 1.1. (strong uniform tits alternative [8]) For every d ∈ N there is N = N(d) ∈ N such that if K is any ﬁeld and F a ﬁnite symmetric subset of GLd (K ) containing 1, either F N contains two elements that freely generate a nonabelian free group, or the group generated by F is virtually solvable (i.e., contains a ﬁnite index solvable subgroup). We denote by F n = F · ... · F the product set of n copies of F. The purpose of this text is to discuss some consequences of this result and to give a selfcontained proof of it in the special case of SL2 (hence equivalently GL2 ) and K any ﬁeld of characteristic 0 (or, equivalently, for K = C, since every ﬁnitely generated ﬁeld of characteristic 0 embeds in C). This case is already representative of the general case as it captures the main difﬁculty, namely, treat all number ﬁelds in a uniform way. The proof given in [7] and [8] is an elaboration of the proof for SL2 that we are about to give. Theorem 1.1 improves earlier reﬁnements of Tits’s theorem due to EskinMozes-Oh (see [13]) and to T. Gelander and the author (see [9]). These two papers were concerned with the S-arithmetic version of Theorem 1.1, namely, they proved uniformity of N (the “freeness radius” of F) for sets F with coefﬁcients inside a ﬁxed ﬁnitely generated ring. While in [13] the main concern was to prove uniform exponential growth by constructing generators of a free semigroup in F N , in [9] it was shown that the result of [13] could be pushed to get generators of a free group in F N , where N was depending only on d and on the ring generated by the matrix coefﬁcients of the elements of F. The main contribution of this text and of [7] and [8] is to remove the dependence on the ring of coefﬁcients. As in Tits’s proof or in [9], the proof of Theorem 1.1 can be divided into an arithmetic step, on the one hand, and a geometric step on the other. While in [9] (as well as in Tits’s original theorem) the arithmetic step was the easier one and most of the work lied in showing that a certain geometric conﬁguration (the so-called “ping-pong”) did arise, roles are reversed in the proof of [8] in Theorem 1.1 and the arithmetic step is the harder step, while the geometric step routinely follows Tits’s original proof after a careful check that all estimates are indeed uniform over all local ﬁelds. For SL2 , however, none of the usual difﬁculties of higher rank arise and as will become clear below, this geometric step is even more transparent (in that case or other rank-1 situations this geometric step can also be performed differently by acting directly on the hyperbolic space/tree as has been pointed out by T. Gelander). The proof of Theorem 1.1 is effective in the sense that the constant N can, in principle, be made explicit. Examples due to Grigorchuk and de la Harpe [18]

heights on

SL2

and free subgroups / 457

imply that N(d) must tend to inﬁnity with d. They exhibited a sequence (n )n≥0 (n ≤ GL(n, Z)) of 4-generated subgroups whose rate of exponential growth exponent decays to 1. These groups arise by chopping the usual presentation of the Grigorchuk group after ﬁnitely many relators (see [18]). The arithmetic step in Theorem 1.1 relies on the following result, proved by the author in [7], and for which we will give below a self-contained proof in the special case of SL2 . Before stating it, we need some preliminary deﬁnitions. Let Q be an algebraic closure of Q. In [7] we introduced the arithmetic spectral radius (or normalized height) of F, deﬁned as 1 h(F n ), h(F) = lim n→+∞ n where h is the (absolute) height deﬁned for F a ﬁnite subset of Md (Q) by 1 nv log+ ||F||v h(F) = [K : Q] v∈VK

where log+ = max{0, log}, K is the number ﬁeld generated by the matrix coefﬁcients of F, VK is the set of all places of K , and ||F||v = max{||f ||v , f ∈ F} is the maximal operator norm of f ∈ F, where ||f ||v = maxx=0 ||f (x)||v /||x||v for the standard norm ||x||v induced on Kvd by the standard absolute value | · |v on the completion Kv of K associated to v ∈ VK . We have also set nv = [Kv , Qv ], where Qv is the ﬁeld of p-adic numbers if v|p is ﬁnite and is R if v is inﬁnite. The normalization of the absolute value | · |v is such that |λ|nv v is the modulus of the multiplication by λ on Kv . The quantities h(F) and h(F) are well deﬁned, that is, they are independent of the chosen number ﬁeld. If d = 1 and F is just one element {x}, then h(F) = h(x) is the classical (absolute, logarithmic) Weil height of x (see [3]). Moreover, h(F) is invariant under conjugation by elements from GLd (Q). The main statement is

theorem 1.2. (height gap theorem [7]) There is a positive constant ε = ε(d) > 0 such that if F is a ﬁnite subset of GLd (Q) generating a nonvirtually solvable subgroup , then h(F) > ε. Moreover, if the Zariski closure of is semisimple, then h(F) ≤

inf

g∈GLd (Q)

h(gFg −1 ) ≤ C · h(F)

for some absolute constant C = C(d) > 0.

458 / e. breuillard

This result can be seen as a nonabelian version of the well-known Lehmer conjecture from number theory. This conjecture states that the Weil height h(x) of an algebraic number that is not a root of unity must be bounded below by degε0(x) , where ε0 is an absolute constant. If F consists of a single diagonal matrix diag(x, x −1 ), then h(F) = h(x). When d = 2, that is, for SL2 , the ﬁrst part of Theorem 1.2 has the following geometric interpretation: either there is a ﬁnite place v where F acts without global ﬁxed point on the corresponding Bruhat-Tits tree, or after applying some Galois automorphism of C, the set F, as it acts on the hyperbolic 3space H3 via SL2 (C), moves every point away from itself by a positive absolute constant ε. This is analogous to the Margulis lemma in hyperbolic geometry, according to which if F generates a discrete subgroup of SL2 (C) that is not virtually nilpotent, then every point of H3 is moved away from itself by some element of F by some ﬁxed constant (see [30]). The main purpose of Theorem 1.2 is to yield in F, or a bounded power of F, a nice hyperbolic element, that is, a semisimple matrix whose eigenvalue is of modulus at least 2, say, in some local completion. Finally, letime point out that the proof of Theorem 1.1 yields a free subgroup with the extra property that it is uniformly undistorted in the original subgroup, namely:

theorem 1.3. (uniformly quasi-isometrically embedded free) subgroup [8] There is a constant C = C(d) > 0 such that if K is any ﬁeld and is any non virutally solvable subgroup of GLd (K ) generated by a ﬁnite symmetric subset F (giving rise to a word metric d (·, ·) on ), there exists a free subgroup H of generated by two elements (giving rise to a word metric dH (·, ·)) such that for all h ∈ H 1 · d (1, h) ≤ dH (1, h) ≤ d (1, h). C · |F|C 1.2. Some Consequences for Uniform Growth, Spectral Gap, and Diophantine Properties

We will prove here these corollaries for GLd (C), d ≥ 2.

corollary 1.4. (strong uniform exponential growth) There is ε = ε(d) > 0 such that if F is a ﬁnite subset of GLd (C) containing 1 and generating a nonamenable subgroup, then for all n ≥ 1, |F n | ≥ (1 + ε)n . In particular, ρF = lim

n→+∞

1 log |F n | ≥ log (1 + ε) > 0. n

heights on

SL2

and free subgroups / 459

Let us remark that it may be the case that ρF is bounded away from 0 assuming only that F generates a nonvirtually nilpotent subgroup of GL2 (C). However, we observed in [6] that such an assertion, if true, would imply the Lehmer conjecture about the Mahler measure of algebraic numbers. We also observed there that although every linear solvable group of exponential growth contains a free semigroup, no analog of Theorem 1.1 (the UF property of [1]) holds for solvable groups: namely, one may ﬁnd ﬁnite symmetric sets Fn in GL2 (C) containing 1 and generating a solvable subgroup of exponential growth, such that no pair of elements in (Fn )n may generate a free semigroup. Von Neumann showed that groups containing a free subgroup are nonamenable, that is, have a spectral gap in 2 . The uniformity in Theorem 1.1 implies also a uniformity for the spectral gap (see [29] for this observation). More precisely:

corollary 1.5. (strong uniform spectral gap in 2 ) There is ε = ε(d) > 0 with the following property. If F is a ﬁnite subset of GLd (C) containing the identity and generating a nonamenable subgroup and if is a countable subgroup of GLd (C) containing F and f ∈ 2 (), then there is σ ∈ F such that f (σ −1 x) − f (x)2 ≥ ε · |f (x)|2 x∈

x∈

In particular, if F in GLd (C) is a ﬁnite subset containing the identity and generating a nonamenable subgroup, then for every ﬁnite subset A in GLd (C), we have |FA| ≥ (1 + ε)|A|. In [24] Lubotzky, Phillips, and Sarnak showed that for the compact Lie group G = SU(2), the spectral measure of the “Hecke operators” Tμ = −1 1 acting on L20 (G) is supported on [−mμ , mμ ] where mμ is 1≤i≤k gi + gi 2k the norm of Tμ viewed as an operator on 2 (), being the abstract group generated by the gi ’s. This spectral measure is by deﬁnition the limiting distribution of the eigenvalues of Tμ on the nth-dimensional representation of G. Corollary 1.5 implies that mμ is bounded away from 1 independently of μ as soon as k is ﬁxed and is a nonamenable subgroup of G. In other words, there is ε = ε(k) > 0 such that the proportion of eigenvalues of Tμ lying in [−1, −1 + ε] ∪ [1 − ε, 1] tends to 0 as n tends to inﬁnity. The analogous result for Cayley graphs of SL2 (Z/pZ) is also a direct consequence of Corollary 1.5, that is, the spectral measure of any limit of such Cayley graphs is supported on [−1 + ε, 1 − ε]. It is believed (spectral gap conjecture, see [24] and [28]) that 1

460 / e. breuillard

is never an accumulation point of eigenvalues of Tμ for any given μ (or at least almost any in SU(2)) whose support generates a nonamenable subgroup. Since by Kesten’s theorem mμ is also the exponential rate of decay of the return probability (see [21]), we also have

corollary 1.6. (strong uniform decay of return probability) There is ε = ε(k, d) > 0 with the property that for any nonamenable k-generated subgroup of GLd (C) we have P(Sn = 1) ≤ (1 − ε)n

for all n ≥ 1, where Sn is the simple random walk on . The next corollary gives an upper bound on the cogrowth of subgroups of GLd (C).

corollary 1.7. (co-growth gap) Given m ∈ N, there is n(m) > 0 such that for every n ≥ n(m) the following holds: F = {a1 , ..., am } ⊂ GLd (C) generates a non virtually solvable subgroup, if and only if the number of elements w in the free group Fm of word length n such that w(a1 , ..., am ) = 1 is at most (2m − 1 − mεD )n . Here ε, D > 0 are constants depending on d only. This result can be paraphrased by saying that nonamenable subgroups of GLd (C) are very strongly nonamenable, that is, have few relations. This puts a purely group theoretical restriction on an abstract ﬁnitely generated group given in terms of generators and relations to admit an embedding in GLd (C). The uniformity in Theorem 1.1 allows us to reduce mod p and we obtain a statement giving a lower bound on the girth of subgroups of GLd in positive characteristic:

corollary 1.8. (large girth) Given k, d ≥ 2, there is N, M ∈ N and ε0 , C > 0 such that for every prime p and every ﬁeld K of characteristic p and any ﬁnite subset F with k elements generating a subgroup of GLd (K ) that contains no solvable subgroup of index at most M, then F N contains two elements a, b such that w(a, b) = 1 in GLd (K ) for any nontrivial word w in F2 of length at most f (p) = C · ( log p)ε0 . It was conjectured in [16] that the statement of Corollary 1.8 holds for generating subsets F of SL2 (Fp ) with ε0 = 1.

heights on

SL2

and free subgroups / 461

Theorem 1.3 on the uniform QI-embedding of the free subgroup yields a uniform bound for the distortion of the subgroup of large girth. This in turn gives uniform expansion for subsets of say GL2 (Fp ) lying in a ball of radius ≤ ( log p)ε0 . Namely:

corollary 1.9. (uniform expansion for small sets) Given k, d≥ 2, there is N, M ∈ N and ε0 , C, α, β > 0 such that for every prime p and every ﬁeld K of characteristic p and any ﬁnite symmetric subset F with k elements, containing 1 and generating a subgroup of GLd (K ) that contains no solvable subgroup ε of index at most M, then for any subset A ⊂ F C·( log p) 0 there is s ∈ F such that n ε0 |sA A| ≥ α|A|. In particular μ∗n F (e) ≤ (1 − β) for all n ≤ C · ( log p) , where μF is the uniform probability measure on F. If one could get ε0 = 1 in the above corollary, then applying the argument of Bourgain and Gamburd [4] would give a proof that the family of all Cayley graphs of SL2 (Z/pZ) for varying p but with a ﬁxed number of generators forms an expander family. See Lubotzky’s book [23] for background material on expanders. It was also proved in [16] that a random d-regular Cayley graph of GL2 (Fp ) has girth at least (1 − o(1)) logd−1 (p). Here we obtain ε0 = 2−10 for GL2 , which is quite far. In the same vein, one obtains the following two weak forms of "nonLiouvilleness" for subgroups of GLd (C). Let d be some Riemannian distance on GL2 (C).

corollary 1.10. (short words are not simultaneously very liouville) Given d, there is N0 ∈ N and ε1 > 0 with the following property. For every ﬁnite set F ⊂ GLd (C) generating a nonvirtually solvable subgroup, there is δ0 (F) > 0 such that for every δ ∈ (0, δ0 ) there are two short words a, b ∈ F N O such that d(w(a, b), 1) ≥ δ for every reduced word w in the free group F2 with length (w) at most ( log δ −1 )ε1 . In [19] Kaloshin and Rodnianski proved that for G = SU(2) ≤ SL2 (C) almost every pair (a, b) ∈ G × G satisﬁes d(w(a, b), 1) ≥ exp ( − C(a, b) · (w)2 ) for all w ∈ F2 \{e} and some constant C(a, b) > 0. Besides, it is easy to see that if a, b ∈ GL2 (Q), then the pair (a, b) satisﬁes the stronger diophantine condition d(w(a, b), 1) ≥ exp ( − C(a, b) · (w)). It is conjectured in [28] and [16] that this stronger condition also holds for almost every pair (a, b) ∈ SU(2) with respect to Haar measure.

462 / e. breuillard

Our result also allows us to estimate the number of words of length ≤ n that fall in a shrinking neighborhood of 1 in GLd (C). More precisely,

corollary 1.11. (weak diophantine property) There are τ , ε1 , C > 0 with the following property. For every {a, b} ≤ GLd (C) that generates a nonvirtually solvable subgroup, there is δ0 (a, b) > 0 such that for every δ ∈ (0, δ0 ) and every n ≤ C( log δ −1 )ε1 , the proportion of elements w in the free group F2 of word length n such that d(w(a, b), 1) ≤ δ is at most exp ( − τ n). In [15], Gamburd, Jacobson, and Sarnak showed for G = SU(2) that if a pair (a, b) ∈ G satisﬁes the conclusion of Corollary 1.11 with ε1 = 1 and C > C0 (for some explicit C0 > 0), then (a, b) has a spectral gap on L2 (G). In [5], Bourgain and Gamburd showed that if a pair (a, b) ∈ G satisﬁes the above condition with ε1 = 1 and some C = C(a, b) > 0, perhaps small, then (a, b) has a spectral gap on L2 (G). This latter condition is automatically satisﬁed if (a, b) satisﬁes the stronger diophantine condition above, for instance, if (a, b) ∈ GL2 (Q). Hence these pairs have a spectral gap. It is unknown whether there are (topologically generating) pairs with no spectral gap.

remark 1.12. We emphasize here that all the constants in the above theorems and corollaries can be effectively computed. Only at one point in the proof do we use a compactness argument. This is in the proof of Lemma 2.1(b). However, this statement can be given an effective proof valid in Md (C) (available upon request) although the price to pay is a much lengthier argument. Comment on the proof of Corollaries 1.8 to 1.11: Observe (see Proposition 9.2) that the condition on a ﬁnite subset F = {A, B} of GLd (C) that it generates an amenable (or equivalently virtually solvable) subgroup of GLd (C) is an algebraic one, as is the condition that all short words in A and B satisfy a relation of length at most n. Thus the statement of Theorem 1.1 can be read as a union of countably many assertions of ﬁrst-order logic. According to the “compactness theorem” from model theory, since each assertion holds for C it must also hold for an arbitrary ﬁeld K of sufﬁciently large characteristic (depending on n). This readily gives this existence of some function f (p) going to +∞ with p in Corollary 1.8. To derive the bound ( log p)ε0 , as well as the bounds ( log δ −1 )ε0 in Corollaries 1.10 and 1.11, we use instead a standard version of the effective Nullstellensatz due to Masser and Wustholz (see [25]).

heights on

SL2

and free subgroups / 463

1.3. Outline of the Paper and of the Proof of Theorems 1.2 and 1.1 for SL2

In Section 3 we introduce our main objects, the height and normalized height of a ﬁnite set F of matrices and prove basic properties about them. One of the key properties, the comparison between h and e, relies crucially on Section 2, which is devoted to the proof of a key lemma, the spectral radius lemma. This lemma says in substance that unless F can be conjugated in a bounded part of SL2 , one will ﬁnd a short word with letters in F with a large eigenvalue. In Section 4 we prove the ﬁrst part of Theorem 1.2 (the height gap). The proof makes crucial use of equidistribution properties of algebraic numbers of small height and in particular a result of Zhang (Theorem 4.9) and Bilu’s equidistribution theorem for Galois orbits (Theorem 4.4 below). Using the Eskin-Mozes-Oh escape lemma (see Lemma 4.8 below) we ﬁrst reduce to the 2-generated case, when F = {Id, A, B}. Conjugating if necessary, A can be assumed to be diagonal. By making local estimates at each place, and with the help of Bilu’s theorem, we then show that if h(F) is small, then the heights of b11 , b22 , and b12 b21 are small. But as b11 b22 − b12 b21 = 1, Zhang’s theorem quickly yields to a contradiction if b12 b21 = 0. So b12 b21 = 0 and F is made of upper or lower triangular matrices, that is, F generates a solvable subgroup and we are done. Theorem 1.1 is of purely algebraic nature and we begin its proof by showing that its validity over Q implies its validity over C. In the case of SL2 , one needs to exhibit a place v where one can play ping-pong on the projective line P1 (Kv ) for the local ﬁeld Kv , as in Tits’s proof of his alternative. These “ping-pong players” will be the generators of the desired free subgroup. To achieve this, one needs to be able to conjugate F in SL2 (Kv ) in such a way that three conditions are satisﬁed. First, the norm ||F||v ought to be controlled (up to a ﬁxed power) by the maximal eigenvalue say |λ|v of an element, say A, lying in F (or F N for a bounded N). Second, |λ|v should be large enough, that is, bounded away from 1. And third, at least one element, say B, from F or F N , must send the eigenvectors of A far apart from one another with a distance controlled by some negative power of ||F||v . This criterion for ping-pong is explained in Section 6. In Section 7 we show that a place v with these properties does exist. This is done in two steps. First (Section 5), it is shown that the minimal height h(F) can be almost achieved (up to multiplicative and additive constants) by the ordinary height h(F) after possibly conjugating F inside SL2 (Q). This is the second half of Theorem 1.2: this step uses the estimates needed in the ﬁrst part of Theorem 1.2 (i.e. the proof of the height gap). In a second step (Section 7) the product formula on P1 (Q) is used to show that the distances between

464 / e. breuillard

eigenvectors of A and their images under B are controlled in terms of h(F), and hence h(F), only. This implies the existence of a place v satisfying the ﬁrst and third conditions. Theorem 1.2 ensures that v can be chosen to satisfy the second condition also. Sections 8 and 9 are devoted to the applications.

2. Spectral Radius Lemma for Several Matrices In this section we state and prove the crucial Lemma 2.1. It says that given a ﬁnite set of matrices with coefﬁcients in a local ﬁeld, one may always ﬁnd a short word with letters in that ﬁnite set whose maximal eigenvalue is as large as the minimal norm of the ﬁnite set. Together with Proposition 2.5 it can be interpreted as an analogue for several matrices of the classical spectral radius lemma relating the maximal eigenvalue and the rate of growth of the powers of a matrix. This lemma expresses in a condensed form an idea from a key proposition of the work of Eskin-Mozes-Oh where the concept of an almost algebra was used to essentially achieve the same goal. We emphasize here that getting an equality in part (a) of Lemma 2.1 as opposed to a mere inequality like in part (b) of the same lemma is absolutely crucial in the whole proof and in particular in Theorem 1.2. Let k be a local ﬁeld of characteristic 0. Let · k be the standard norm on 2 k , that is, the canonical Euclidean (respectively, Hermitian) norm if k = R (respectively, C) and the sup norm ((x, y)k = max{|x|k , |y|k }) if k is non Archimedean. We will also denote by · k the operator norm induced on M2 (k) by the standard norm · k on k2 . Let Q be a bounded subset of matrices in M2 (k). We set Q = sup g k k g∈Q

and call it the norm of Q . Let k be an algebraic closure of k. It is well known (see Lang’s Algebra [22]) that the absolute value on k extends to a unique absolute 2

value on k, hence the norm · k also extends in a natural way to k and to M2 (k). This allows us to deﬁne the minimal norm of a bounded subset Q of M2 (k) as Ek (Q ) =

inf x∈GL2 (k)

xQx −1 . k

We will also need to consider the maximal eigenvalue of Q , namely k (Q ) = max{|λ|k , λ ∈ spec(q), q ∈ Q }

heights on

SL2

and free subgroups / 465

where spec(q) denotes the set of eigenvalues (the spectrum) of q in k. Finally, let Rk (Q ) be the spectral radius of Q : 1

Rk (Q ) = lim ||Q n ||kn . n→+∞

These quantities are related to one another. The key property concerning them is given in the following assertion (which also holds in Md (k), k ≥ 2; see [7]).

lemma 2.1. (spectral radius lemma) Let Q be a bounded subset of M2 (k), a) if k is non-Archimedean, then k (Q 2 ) = Ek (Q )2 ; and b) if k is Archimedean, there is a constant c ∈ (0, 1) independent of Q , such that k (Q 2 ) ≥ c 2 · Ek (Q )2 . Proof. We make use of the following easy lemmas.

lemma 2.2. Let L be a ﬁeld and Q a subset of M2 (L) such that Q and Q 2 consist of nilpotent matrices. Then there is a basis (u, v) of L2 such that Qu = 0 and Qv ⊂ Lu. Proof. For any A, B ∈ Q , we have A2 = B2 = (AB)2 = 0. It follows, unless A or B are 0, that ker A = ImA and ker B = ImB. Also if AB = 0, we get ker B = ker (AB) = Im(AB) = ImA, while if AB = 0, then ImB = ker A. So at any case ker A = ImA = ker B = ImB. So we have proved that the kernels and images of nonzero elements of Q coincide and are equal to some line say Lu,. Pick v ∈ L2 \{Lu}, then (u, v) forms the desired basis.

lemma 2.3. Let k be a local ﬁeld with ring of integers Ok and uniformizer π. Let A = (aij ) ∈ M2 (Ok ) such that trace(A) and det (A) belong to (π 2 ) and a11 , a22 , a21 ∈ (π ), while a12 ∈ Ok× . Then a21 ∈ (π 2 ). Proof. We have a12 a21 = a11 a22 − det (A) ∈ (π 2 ) and a12 ∈ Ok× , hence a21 ∈ (π)2 . When k is a non-Archimedean local ﬁeld, if a set Id + Q in SL2 (k) and its square have only eigenvalues very close to 1, then it must ﬁx pointwise the 1-neighborhood of some point in the Bruhat-Tits tree of SL2 (k). This is essentially the content of the next lemma.

466 / e. breuillard

lemma 2.4. (small eigenvalues implies large fixed-point set) Let k be a local ﬁeld with ring of integers Ok and uniformizer π together with an absolute value | · |k , which is (uniquely) extended to an algebraic closure k of k. Let Q be a subset of M2 (Ok ) such that k (Q ) and k (Q 2 ) are both ≤ |π |3k . Then there is T ∈ GL2 (k) such that TQT −1 ⊂ π M2 (Ok ). Proof. We can write Q as the disjoint union Q1 ∪ πQ2 where Q1 does not intersect π M2 (Ok ). Let Q = Q1 ∪ Q2 . Then k (Q ) and k (Q 2 ) are both ≤ |π |k . Hence projecting to M2 (L), where L is the residue ﬁeld L = Ok /(π ), the matrices from Q and Q 2 become nilpotent. According to Lemma 2.2, one may ﬁnd a basis (u, v) of L2 such that Q u = 0 and Q v ⊂ Lu. According to Nakayama’s lemma, this basis is the projection of a basis (u, v) of Ok2 . Up to conjugating by a matrix in GL2 (Ok ), we may assume that (u, v) is the canonical basis of Ok2 . Therefore Q consists of matrices A = (aij ) ∈ M2 (Ok ) with a11 , a22 , a21 ∈ (π ). Moreover, matrices in Q1 satisfy a12 ∈ Ok× and hence by Lemma 2.3, a21 ∈ (π 2 ). But for the matrices coming from π Q 2 we also have a21 ∈ (π 2 ). So we have a21 ∈ (π 2 ) for all matrices in Q . Let T = diag(π , 1) ∈ GL2 (k). Then clearly TQT −1 ⊂ π M2 (Ok ). We go back to the proof of Lemma 2.1. We ﬁrst prove (b). By contradiction, if (Q 2 )

such a c did not exist, then we may ﬁnd a sequence of Qn such that E k(Q n)2 → 0. k

Q

n

We can change Qn into E (Qn n ) and thus obtain a sequence of compact sets in k M2 (k) such that Ek (Qn ) = 1 with k (Qn2 ) → 0 and k (Qn ) → 0. Passing to a limit, we obtain a compact subset Q of M2 (k) such that k (Q 2 ) = k (Q ) = 0 while Ek (Q ) = 1. By Lemma 2.2, we can thus ﬁnd a basis (u, v) where Qu = 0 and Qv ⊂ Lu. But then conjugating Q by a suitable diagonal matrix can shrink the norm of Q as much as we want, hence Ek (Q ) = 0. This is a contradiction. We now prove (a). Let π be a uniformizer for k. Let x = log Ek (Q ) where the log is taken in base |π |−1 . Suppose that k (Q 2 ) < Ek (Q )2 and let ε = k 1

x − 12 log k (Q 2 ) > 0. Then as k (Q ) ≤ k (Q 2 ) 2 , we have x − log k (Q ) ≥ ε > 0. Note that with our choice of normalization, log k (Q 2 ) ∈ 12 Z. Let n ∈ N √ √ be such that 2nε > 3. Let k0 be the extension k(2n π ) where 2n π is some 2nroot of π in k. Since x = log Ek (Q ) = log inf {||gQg −1 ||k , g ∈ GL2 (k)}, we may assume after possibly conjugating Q inside GL2 (k1 ), for some ﬁnite extension 1 k1 of k0 , that y := log ||Q ||k ≤ x + 2n and also that y = min{log ||gQg −1 ||k , g ∈ 1 GL2 (k1 )}. Let π1 be a uniformizer in k1 . Then log |π1 |−1 ≤ 2n and y = log |π1 |m k −m for some m ∈ Z. Let Qy = π1 Q ⊂ M2 (Ok1 ). We get log Ek (Qy ) = x − y ≤ 0 3 and log k (Qy ) ≤ 12 log k (Qy2 ) = x − y − ε ≤ −ε ≤ − 2n ≤ log |π1 |3k . We are thus in a position to apply Lemma 2.4, which implies that Qy , and hence

heights on

SL2

and free subgroups / 467

Q itself, can be further conjugated inside GL2 (k1 ) so as to strictly reduce its norm. But this contradicts the minimality of y.

proposition 2.5. Let Q be a bounded subset of M2 (k). We have 1

1

1

1

Rk (Q ) = lim Ek (Q n ) n = inf Ek (Q n ) n = lim k (Q 2n ) 2n = sup k (Q n ) n . n→+∞

n∈N

n→+∞

n∈N

Moreover, if k is non-Archimedean, Rk (Q ) = Ek (Q ), while if k is Archimedean, then c · Ek (Q ) ≤ Rk (Q ) ≤ Ek (Q ), where c is the constant from Lemma 2.1(b). Proof. We omit the proof: these identities follow either directly from the deﬁnitions or as a straightforward application of Lemma 2.1. Note that some periodicity phenomenon may arise if Id ∈ / Q , namely, it may be that k (Q 2n+1 ) = 1 for all n while k (Q 2n ) tends to inﬁnity (for instance, take for Q a set of symmetries around several points on a given geodesic in the 1 hyperbolic plane). However, if Id ∈ Q , then we do have limn→+∞ k (Q n ) n = Rk (Q ). Note also that if Q belongs to SL2 (k), then Ek (Q ) ≥ Rk (Q ) ≥ k (Q ) ≥ 1 and all three quantities remain unchanged if we add Id to Q . The following lemma explains what happens if these quantities are close or equal to 1.

lemma 2.6. (linear growth of displacement squared) Suppose k is Archimedean (i.e., k = R or C). Then we have for every n ∈ N and every bounded subset Q of SL2 (k) containing Id, √n 1 Ek (Q n ) ≥ Ek (Q ) 4 . Moreover, log Rk (Q ) ≥ c1 · log Ek (Q ) · min{1, log Ek (Q )} for some constant c1 > 0. In particular Ek (Q ) = 1 iff Rk (Q ) = 1. Proof. We use nonpositive curvature of hyperbolic space H3 . For x ∈ H3 set L(Q , x) = maxg∈Q d(gx, x) and L(Q ) = inf x L(Q , x). Fix ε > 0, set n := L(Q n ) = 2 log Ek (Q n ), and let rn be the inﬁmum over x ∈ H3 of the smallest radius of a closed ball containing Q n x. Note ﬁrst that rn ≤ n ≤ 2rn . We now prove (1). Fix ε > 0 and let x, y ∈ H3 be such that Q n+1 x is contained in a ball of radius rn+1 + ε around y. Let q ∈ Q be arbitrary. Since Q contains Id, we have Q n x ⊂ Q n+1 x, and qQ n x lies in the two balls of radius rn+1 + ε centered around qy and around y. By the CAT(0) inequality for the

468 / e. breuillard

median, the intersection of the two balls is contained in the ball B of radius t := (rn+1 + ε)2 − d(qy, y)2 /4 centered around the midpoint m between y and qy. Translating by q−1 , we get that Q n x lies in the ball of radius t centered at q−1 m. In particular rn ≤ t. This means d(qy, y)2 ≤ 4((rn+1 + ε)2 − rn2 ). 2 Since q ∈ Q and ε > 0 were arbitrary, we obtain 21 ≤ 4(rn+1 − rn2 ). Sum2 2 2 ming over n, we get n1 ≤ 4rn ≤ 4n , hence (1). But byLemma 2.1(b), 1

1

k (Q 2n ) ≥ c 2 Ek (Q n )2 , hence Rk (Q ) ≥ k (Q 2n ) 2n ≥ c n Ek (Q ) in n, we obtain the desired bound.

1 4n

. Optimizing

3. Height, Arithmetic Spectral Radius and Minimal Height For any rational prime p (or p = ∞) let us ﬁx an algebraic closure Qp of the ﬁeld of p-adic numbers Qp (if p = ∞, set Q∞ = R and Q∞ = C). We take the standard normalization of the absolute value on Qp (i.e., |p|p = 1p ). It admits a unique extension to Qp , which we denote by | · |p . Let Q be the ﬁeld of all algebraic numbers and K a number ﬁeld. Let VK be the set of equivalence classes of valuations on K . For v ∈ VK let Kv be the corresponding completion. For each v ∈ VK , Kv is a ﬁnite extension of Qp for some prime p. We normalize the absolute value on Kv to be the unique one that extends the standard absolute 1

value on Qp . Namely |x|v = |NKv |Qp (x)|pnv where nv = [Kv : Qp ]. Equivalently Kv has nv different embeddings in Qp and each of them gives rise to the same absolute value on Kv . We identify Kv , the algebraic closure of Kv , with Qp . Let Vf be the set of ﬁnite places and V∞ the set of inﬁnite places. Let F be a ﬁnite subset in M2 (K ). For v ∈ VK , in order not to surcharge notation, we will use the subscript v instead of Kv in the quantities Ev (F) = EKv (F), v (F) = Kv (F), and so on. Recall that if x ∈ K then its (absolute, logarithmic) Weil height is by deﬁnition (see, e.g., [3]) the following quantity: 1 h(x) = nv log+ |x|v . [K : Q] v∈VK

It is well deﬁned (i.e., independent of the choice of K x). Let us similarly deﬁne the height of a matrix f ∈ M2 (K ) by 1 nv log+ ||f ||v h(f ) = [K : Q] v∈VK

and the height of a ﬁnite set F of matrices in M2 (K ) by 1 nv log+ ||F||v , h(F) = [K : Q] v∈VK

heights on

SL2

and free subgroups / 469

where nv = [Kv : Qv ]. We also deﬁne the minimal height of F as 1 2 e(F) = nv log+ Ev (F) [K : Q] v∈VK

and the arithmetic spectral radius (or normalized height) of F as 1 nv log+ Rv (F). h(F) = [K : Q] v∈VK

For any height h, we also set h = h∞ + hf , where h∞ is the inﬁnite part of h (i.e., the part of the sum over the inﬁnite places of K ) and hf is the ﬁnite part of h (i.e., the part of the sum over the ﬁnite places of K ). Note that these heights are well deﬁned independently of the number ﬁeld K such that F ⊂ M2 (K ). The above terminology is justiﬁed by the following facts:

proposition 3.1. (basic properties of heights i) For any ﬁnite set F in M2 (Q), we have 1) h(F) = limn→+∞ n1 h(F n ) = inf n∈N n1 h(F n ), 2) ef (F) = hf (F) and e(F) + log c ≤ h(F) ≤ e(F) where c is the constant in Lemma 2.1 (b), and 3) h(F n ) = n · h(F) and h(F ∪ {Id}) = h(F). Proof. This follows easily from Lemma 2.1 and Proposition 2.5.

We also record the following simple observations:

proposition 3.2. (basic properties of heights ii) We have, for a ﬁnite set F in M2 (Q), 1) e(xFx −1 ) = e(F) if x ∈ GL2 (Q), 2) e(F n ) ≤ n · e(F), and 3) if λ is an eigenvalue of an element of F, then h(λ) ≤ h(F) ≤ e(F). Proof. This is clear.

We can also compare e(F) and h(F) when h(F) is small:

proposition 3.3. (basic properties of heights iii) Let c1 be the constant from Lemma 2.6, then

470 / e. breuillard

c1 h∞ (F) ≥ · e∞ (F) · min{1, e∞ (F)} 4 h(F) is for any ﬁnite subset F in SL2 (Q). In particular e(F) is small as soon as small. Proof. From Lemma 2.6, hv (F) ≥ c1 · ev (F) · min{1, ev (F)} for every v ∈ VK . We + may write e∞ (F) = αe (F) + (1 − α)e− (F) where e+ is the average of the ev greater than 1 and e− the average of the ev smaller than 1. Applying CauchySchwarz, we have h∞ (F) ≥ c1 · (αe+ + (1 − α)(e− )2 ). If αe+ (F) ≥ 12 e∞ (F), then h∞ (F) ≥ c21 e∞ (F), and otherwise (1 − α)e− ≥ e2∞ , hence h∞ (F) ≥ c1 (1 − α) (e− )2 ≥ c41 e2∞ . At any case h∞ (F) ≥ c41 · e∞ (F) · min{1, e∞ (F)}.

4. Height Gap Theorem In this section, we prove Theorem 1.2 from the introduction. First observe that according to Propositions 3.3 and 3.1(2), h(F) is small if and only if e(F) is small. So we may as well replace h(F) by e(F) in Theorem 1.2. We now assume that F = {Id, A, B}, with A semisimple. The general case follows from this as we will show in Lemma 4.7. Since e(F) is invariant under conjugation by any element in GL2 (Q), we may assume that A is diagonal, that is,

λ 0 a b 3 A= , B= . 0 λ−1 c d Let deg (λ) be the degree of λ as an algebraic number over Q. The main part of the argument consists in the following proposition:

proposition 4.1. (small normalized height implies small height of matrix coordinates) For every β > 0 there exists d0 ,η > 0 such that, if F = {A, B} are as in (3) and if e(F) ≤ η and deg (λ) ≥ d0 , then max{h(ad), h(bc)} ≤ β. In order to prove this statement, we are ﬁrst going to give local estimates at each place v, then use Bilu’s equidistribution theorem to show that when these estimates are put together, the error terms give only a negligible contribution to the height. Let K be the number ﬁeld generated by the coefﬁcients of A and B. Let v ∈ VK be a place of K . We set sv = log Ev (F) and δ = λ − λ−1 . We ﬁrst show the following local estimate:

heights on

SL2

and free subgroups / 471

lemma 4.2. (local estimates) For each v ∈ VK we have max{|a|v , |d|v , |bc|v } ≤ Cv e2sv max{1, |δ −1 |v }, where Cv is a constant equal to 1 if v is a ﬁnite place and equal to a number C∞ > 1 (independent of v) if v is inﬁnite. Moreover, there are absolute constants ε0 > 0 and C0 > 0 such that if v is inﬁnite and sv ≤ ε0 , then √ √ sv sv + 2 ). max{|ad|v , |bc|v } ≤ 1 + C0 ( sv + |δ|v |δ|v Proof. In order not to overburden notation in this proof we set sv to be some number arbitrarily close but strictly bigger than log Ev (F) and we can let it tend to log Ev (F) at the end. If v is inﬁnite, then Qv = C and SL2 (C) = KAN where K = SU2 (C), A is the subgroup of diagonal matrices with real positive entries, and N is the subgroup of unipotent complex upper-triangular matrices. As K leaves the norm invariant, there must exist a matrix P ∈ AN such that max{PAP −1 , PBP −1 } ≤ esv . Since P ∈ AN, we may write P =

t y with t > 0 and y ∈ C. Then we have, setting δ = λ − λ−1 , 0 t −1

λ tyδ −1 4 , PAP = 0 λ−1

a + cyt −1 bt 2 + dyt − ayt − cy 2 −1 . PBP = t −2 c −yct −1 + d If v is ﬁnite and Kv is the corresponding completion, with ring of integers Ov and uniformizer π, we have SL2 (Kv ) = Kv Av Nv where Kv = SL2 (Ov ), Av = {diag(π n , π −n ), n ∈ Z} and Nv is the subgroup of unipotent upper-triangular

matrices with coefﬁcients in Kv . Hence, we also get a P ∈ Av Nv satisfying (4) with y ∈ Kv and t = π n for some n ∈ Z. We ﬁrst assume that v is ﬁnite. Recall that the operator norm in SL2 (Kv ) is given by the maximum modulus of each matrix coefﬁcient. Hence we must have t −2 c v ≤ esv and tyδ v ≤ esv . It follows that cyt −1 v ≤ e2sv |δ −1 |v and hence |a|v ≤ max{esv , e2sv |δ −1 |v }. Similarly, |d|v ≤ max{esv , e2sv |δ −1 |v }. Hence |ad|v ≤ max{e2sv , e4sv |δ −1 |2v }. Moreover ad − bc = 1, hence |bc|v ≤ max{1, |ad|v } ≤ max{e2sv , e4sv |δ −1 |2v }. Now we assume that v is inﬁnite. Claim: There is u0 > 0 such that if 0 ≤ u ≤ u0 and B ≤ eu , then √ 5 max{|a − d|, |b + c|} ≤ 2 u,

472 / e. breuillard

√ max{|a|2 + |b|2 , |d|2 + |c|2 } ≤ 1 + 6u + 8 u, and √ √ max{|a|, |b|, |c|, |d|} ≤ 1 + 3u + 4 u ≤ 1 + 5 u.

6 7

To prove this, recall that the operator norm in SL2 (C) satisﬁes tr(B∗ B) = = B 2 + B −2 . Hence |a|2 + ... + |d|2 ≤ 1 + e2u , hence ≤ 4 if u is small enough (say, u ≤ .5). On the other hand, for small u, |a − d|2 + |b + c|2 = |a|2 + ... + |d|2 − 2 ≤ e2u − 1 ≤ 4u. Hence (5). Now √ √ |d| ≤ |a| + 2 u and since |a|, |b| ≤ 2, we get |d|2 ≤ |a|2 + 4u + 8 u and vice versa and similarly for b and c. Hence (6) and (7), and the claim is proved. Let now ε > 0 and assume that sv ≤ ε. From (4) we get |λ|2 + |λ−1 |2 + 2 tyδ ≤ 1 + e2ε , hence tyδ 2 ≤ e2ε − 1 ≤ 4ε if ε is small enough. So tyδ ≤ √ 2 ε. Now since PBP −1 ≤ eε , we have |t −2 c| ≤ 2 as soon as ε ≤ 12 . Hence √ √ √ |yct −1 | ≤ 4|δ|ε and max{|a|, |d|} ≤ 1 + 5 ε + 4|δ|ε . Finally, for some absolute √ √ constant C > 0 |ad| ≤ 1 + C( ε + |δ|ε + |δ|ε 2 ). |a|2 + |b|2 + |c|2 + |d|2

On the other hand, |cy 2 | = |t −2 c(ty)2 | ≤

8ε |δ|2

and |d − a||yt| ≤ 2 max{|a|, |d|} 16ε 2 2 −2 c| ≤ 2√ε, and |bc + |yt| ≤ + |δ| 2 . Also by (5), |bt + (d − a)yt − cy + t √ √ 48ε −2 c|2 ≤ 1 + 14√ε, |t −2 c|2 | ≤ 2|bt 2 + t −2 c| ≤ 4 ε + 48|δ| ε + |δ| 2 , and by (6), |t √ hence up to enlarging the absolute constant C, we also have |bc| ≤ 1 + C( ε + √ ε ε |δ| + |δ|2 ). Without the assumption that sv is small, we can make a coarser estimate: 2 2sv |t −2 c|2 ≤ 1 + e2sv , tyδ ≤ 1 + e2sv , hence cyt −1 ≤ 1+e |δ| and max{|a|, |d|} ≤ √ 1+e2sv 2sv ≤ 4e2sv max{1, 1 } and |ad| ≤ 16e4sv max{1, 1 }. Similarly, |δ| + 1 + e |δ| |δ|2 √ 24 ε |δ|

we compute |bc| ≤ 20e4sv max{1, |δ|12 }.

We now put together the local information obtained above to bound the heights. Let n = [K : Q] and Vf and V∞ the set of ﬁnite and inﬁnite places of K . Set ε0 , C0 , and C∞ the constants obtained in the previous lemma. For A > 0 and x ∈ Q, we set 1 A 8 h∞ (x) = nv · log+ |x|v [K : Q] v∈V∞ ,|x|v ≥A

where the sum is limited to those v ∈ V∞ , for which |x|v ≥ A. We have

lemma 4.3. For some constant C2 (2 ≤ C2 ≤ 2 + (2 log C∞ + 4)/ log 2), we have for all ε1 ∈ (0, 12 ) and all ε ≤ min{ε0 , ε12 } √ ε ε−1 9 max{h(ad), h(bc)} ≤ Cε,ε1 e(F) + 6C0 + 2hf (δ −1 ) + C2 · h∞1 (δ −1 ) ε1 2 log C∞ 2| log ε | + ε 1 and δ = λ − λ−1 . where Cε,ε1 = 4 + ε

heights on

SL2

and free subgroups / 473

Proof. Recall that sv = log Ev (F). If v ∈ V∞ and sv ≥ ε, then according to Lemma 4.2 log+ |ad|v ≤ 2 log C∞ + 4sv + 2 log+ |δ −1 |v hence

1 2 log C∞ 1 nv · log+ |ad|v ≤ 4 + nv sv n ε n v∈V∞ ,sv ≥ε

v∈V∞ ,sv ≥ε

+

2 n

nv · log+ |δ −1 |v .

v∈V∞ ,sv ≥ε

Fix ε1 < 12 . On the other hand, if sv ≤ ε ≤ min{ε0 , ε12 } and |δ|v ≥ ε1 , then √ √ √ s ε sv log+ |ad|v ≤ C0 ( sv + |δ|vv + |δ| 2 ) ≤ 3C0 ε1 and, as nv ≤ 2, v

1 n

nv · log+ |ad|v ≤ 6C0

v∈V∞ ,sv ≤ε,|δ|v ≥ε1

√ ε . ε1

While if sv < ε and |δ|v ≤ ε1 ≤ 12 , then log+ |ad|v ≤ C2 log+ |δ −1 |v for some absolute constant C2 , (2 ≤ C2 ≤ 2 + (2 log C∞ + 4)/ log 2), hence 1 n

nv · log+ |ad|v ≤

v∈V∞ ,sv 0 and ε1 > 0 with the following property. If λ ∈ Q is such that h(λ) ≤ η1 , deg (λ) ≥ d1 , then ε−1

h∞1 (

1 )≤α 1−λ

ε−1

where h∞1 was deﬁned in (8). Proof. Let P ∈ Z[X ] be the minimal polynomial of λ, that is, P(X ) = a X i = an x∈O(λ) (X − x). As P(1) ∈ Z\{0}, log |P(1)| = log |an | + 0≤i≤n i x∈O(λ) log |1 − x| ≥ 0. So

log

|1−x|≤ε1

1 ≤ |1 − x|

Recall (see [22] III.1.) that h(λ) = 11

1 n

log

|1−x|≤ε1

log |1 − x| + log |an |.

|1−x|>ε1

1 n

x∈O(λ) log

1 1 ≤ h(λ) + |1 − x| n

+

|x| + log |an | . Hence

log |1 − x|.

|1−x|>ε1

Consider the function fε1 (z) = 1|z−1|>ε1 log |1 − z|. It is locally bounded on C. By Theorem 4.4, for every ε1 > 0, there must exist d1 , η1 > 0 such 1 that, if h(λ) ≤ η1 and deg (λ) ≥ d1 , then n1 x fε1 (x) − 0 fε1 (e2π iθ )dθ ≤ α3 . On the other hand, we verify that θ → log |1 − e2πiθ | is in L1 (0, 1) and 1 2π iθ 0 log |1 − e |dθ = 0. Hence we can choose ε1 > 0 small enough so that 1 2π iθ )dθ ≤ α3 . Combining these inequalities with (11) and choosing 0 fε1 (e ε−1

η1 ≤ α3 , we get h∞1 ((1 − λ)−1 ) ≤ α.

Combining this with Bilu’s theorem, we get

lemma 4.6. (bounding errors terms via bilu’s theorem ii) For every α > 0 there exists η0 > 0 and A1 > 0 such that for any λ ∈ Q, if h(λ) ≤ η0 and d = deg (λ) > A1 , then hf (

1 ) ≤ 2α. 1−λ

Proof. We apply the product formula to μ = 1 − λ, which takes the form h(μ) = h(μ−1 ), hence hf (μ−1 ) = h∞ (μ) − h∞ (μ−1 ) + hf (μ). But hf (μ) = hf (1 − λ) ≤

heights on

SL2

and free subgroups / 475

hf (λ) ≤ η0 and h∞ (μ) − h∞ (μ−1 ) = [K1:Q] v∈V∞ nv · log |μ|v . Lemma 4.5 shows that the convergence in Equation (10) in Bilu’s theorem not only holds for compactly supported functions on C, but also for functions with logarithmic singularities at 1. In particular, it holds for the function f (z) = 1 log |1 − z|, which is exactly what we need, since 0 f (e2π iθ )dθ = 0. Hence 1 v∈V∞ nv · log |μ|v becomes small. We are done. [K :Q] 1 Proof of Proposition 4.1. Since hf ( λ−λ1 −1 ) ≤ hf (λ) + hf ( 1−λ 2 ) and similarly 1 1 A A A h∞ ( λ−λ−1 ) ≤ h∞ (λ) + h∞ ( 1−λ2 ), it follows from the last two lemmas that we ε−1

β as √ 3 ε 2C1 ε1 ≤ β3

can ﬁnd ε1 > 0, η > 0 and d0 ∈ N so that 2hf (δ −1 ) + C2 · h∞1 (δ −1 ) ≤ soon as h(λ) ≤ e(F) ≤ η and deg (λ) ≥ d0 . Then choose ε so the and ﬁnally take η even smaller so that Cε,ε1 η ≤ we are done.

β 3.

Now apply Lemma 4.3 and

End of the proof of Theorem 1.2:. The following lemma allows us, when proving Theorem 1.2, to assume without loss of generality that F = {1, A, B}, where A and B are two semisimple elements in SL2 (Q) that do not satisfy some prescribed ﬁnite set of algebraic relations. More precisely:

lemma 4.7. For every d1 ∈ N, there exists N(d1 ) ∈ N with the following property. Let F be a ﬁnite subset of SL2 (Q) containing 1 and generating a nonvirtually solvable subgroup, then there exists A, B ∈ F N(d1 ) such that A and B are semisimple, generate a nonvirtually solvable subgroup of SL2 , A is not of order at most d1 , and bc ∈ / 2iπ 4iπ {0, −1, e 3 , e 3 } after we conjugate A and B in the form of Equation (3). Proof. This is a direct application of Lemma 4.8 below applied to = F × F in SL2 × SL2 ≤ GL4 with X = X1 ∪ X2 ∪ X3 ∪ X4 where X1 = {(A, B), A or B has order at most d1 }, X2 = {(A, B), tr(A) or tr(B) is 2}, X3 = {(A, B), A and B generate a virtually solvable subgroup} and X4 the Zariski closure of 2iπ 4iπ {(gAg −1 , gBg −1 ), g ∈ SL2 , A diagonal, bc ∈ {0, −1, e 3 , e 3 }}. For dimension reasons X4 is a proper subvariety of SL2 × SL2 and Propositions 9.2 and 9.1 show that so is X3 .

lemma 4.8. (eskin-mozes-oh “escape from subvarieties”, see [13] and [9]) Let K be a ﬁeld, d ∈ N. For every m ∈ N, there is N ∈ N such that if X a K –algebraic subvariety of GLd (K ) such that the sum of the degrees of the geometrically irreducible components of X is at most m, then for any subset

476 / e. breuillard

⊂ GLd (K ) containing Id and generating a subgroup that is not contained in X (K ), we have N X (K ). Observe (as follows from the irreducibility of cyclotomic polynomials, see [22] VI.3.) that for every d0 ∈ N there is η0 > 0 and d1 > 0 such that if h(λ) < η0 and λ is not a root of 1 of order at most d1 , then deg (λ) ≥ d0 . However, recall the following well-known result (which is also a straightforward corollary of Theorem 4.4),

theorem 4.9. (zhang’s theorem [32]) There exists an absolute constant α0 > 0 such that for any x ∈ Q, we have h(x) + h(1 + x) > α0 unless x ∈ {0, −1, e

2iπ 3

,e

4iπ 3

}.

Let β = α20 where α0 is given by Theorem 4.9. Proposition 4.1 yields d0 > 0 and η = η( α20 ) > 0 such that max{h(ad), h(bc)} ≤ β as soon as e({Id, A, B}) ≤ η and deg (λ) ≥ d0 . By Lemma 4.7, if we have some nice A, B ∈ F N(d1 ) . If e(F) ≤ α

min{η( 20 ),η0 } , N(d1 )

then e({Id, A, B}) ≤ min{η, η0 } and λ is not a root of 1 of order at most d1 . Hence deg (λ) ≥ d0 and by Proposition 4.1, h(ad) + h(bc) ≤ 2β = α0 . 2iπ 4iπ Then according to Theorem 4.9, bc ∈ {0, −1, e 3 , e 3 }, which contradicts our min{η(

α0

),η }

0 2 > 0 is the desired gap. choice of A, B. So N(d1 ) This ends the proof of Theorem 1.2. Finally, observe that Theorem 1.2 combined with Propositions 3.1 to 3.3 implies

proposition 4.10. There exists a constant c0 > 0 such that if F is any ﬁnite subset of SL2 (Q) generating a nonvirtually solvable subgroup, then e(F) ≥ h(F) ≥ c0 · e(F).

5. Simultaneous Quasisymmetrization over Q Here we are going to use our previous height estimates once again to show the following proposition. Observe that the minimal height e(F) coincides with the inﬁmum of h(gFg −1 ) over all adelic points g = (gv )v . The lemma we are about to state essentially means that this inﬁmum is attained (up to additive and multiplicative constants) with a conjugating matrix g lying already in SL2 (Q) as opposed to SL2 (A) (the adelic group).

heights on

SL2

and free subgroups / 477

proposition 5.1. (simultaneous quasisymmetrization) There is an absolute constant C > 0 such that if F is a ﬁnite subset of SL2 (Q) generating a nonvirtually solvable subgroup, then there is an element g ∈ SL2 (Q) such that h(gFg −1 ) ≤ C · e(F) + C When a matrix with real entries is symmetric, then its norm coincides with the modulus of its maximal eigenvalue. Thus the proposition amounts to conjugating F simultaneously (i.e., by a single g ∈ SL2 (Q)) in a "quasisymmetric" position. Proof. As we may replace F by a bounded power of it, Lemma 4.7 above allows us to assume that F contains a semisimple element. Let F = {Id, A, B1 , ..., Bk } with A semisimple. Conjugating by some g ∈ SL2 (Q), we may assume that A is in diagonal form and we write each Bi in the form of Equation (3) with entries ai , bi , ci , and di . Changing F into F 2 if necessary, we may assume that both b1 and c1 are not 0 (otherwise F would be contained in the group of upper- or lower-triangular matrices). We may further conjugate F by the diagonal matrix diag(t, t −1 ), where t ∈ Q is a root of t 4 = c1 /b1 , so as to ensure b1 = c1 . Then h(B1 ) ≤ h(a1 ) + h(d1 ) + 2h(b1 ) + log 2. On the one hand, since a1 d1 − b1 c1 = 1, we have b12 = a1 d1 − 1 and 2h(b1 ) = h(b12 ) ≤ h(a1 d1 ) + log 2 ≤ 2e({A, B}) + log 2C∞ . On the other hand, by Lemma 4.2 applied to {A, Bi }, we have max{|ai |v , |di |v } ≤ Cv e2sv max{1, |δ −1 |v }, for every place v, where δ = λ − λ−1 and sv = sv ({A, Bi }) = log Ev ({A, Bi }). Applying Lemma 4.2 to {A, B1 Bi } we get max{|(B1 Bi )11 |v , |(B1 Bi )22 |v } ≤ Cv e2sv max{1, |δ −1 |v } with sv = sv ({A, B1 Bi }) = log Ev ({A, B1 Bi }). We compute the matrix entry (B1 Bi )11 = a1 ai + b1 ci . We get |ci |v = |[(B1 Bi )11 − a1 ai ]b1−1 |v ≤ Cv e2sv max{1, |δ −1 |v } max{1, |b1−1 |v }. Similarly for |bi |v . Hence, ||F||v ≤ Cv max {|λ|v , |λ−1 |v , |ai |v , |di |v , |bi |v , |ci |v } i=1,...,k

≤ Cv max Ev ({A, B1 , B1 Bi })2 · max{1, |δ −1 |v } max{1, |b1−1 |v }. i=1,...,k

In particular, this means that h(F) ≤ 2 log C∞ + 2e(F 2 ) + h(δ) + h(b1 ) ≤ 7e(F) + 4 log 2C∞ . So we are done.

478 / e. breuillard

corollary 5.2. There exists a constant Cqs > 0 such that if F is as in the proposition, then there is an element g ∈ SL2 (Q) such that h(gFg −1 ) ≤ Cqs · e(F). Proof. It is clear from the combination of the previous proposition and Theorem 1.2.

6. Ping-pong Here we state and prove a ping-pong criterion, which gives a sufﬁcient condition on the ﬁnite set F for it, or a bounded power of it, to contain two free generators of a free subgroup. Let k1 , k2 , k3 ∈ N be three positive integers and let k be a local ﬁeld of characteristic 0 with its standard absolute value. We set Ck = 2 if k is Archimedean, and Ck = 1 if k is non-Archimedean. Let F ⊂ SL2 (k) be a ﬁnite set containing 1 such that k (F k1 ) > Ck ||F||k (see Section 2 for notation, it is important to require a strict inequality here when k is non-Archimedean). Let A ∈ F k1 be such that k (A) = k (F k1 ). Then of course A is semisimple and admits two distinct eigenvectors v + and v − in k2q where kq is either k or some

chicago lectures in mathematics series Editors: Spencer J. Bloch, Peter Constantin, Benson Farb, Norman R. Lebovitz, Carlos Kenig, and J. P. May Other Chicago Lectures in Mathematics titles available from the University of Chicago Press Simplical Objects in Algebraic Topology, by J. Peter May (1967, 1993) Fields and Rings, Second Edition, by Irving Kaplansky (1969, 1972) Lie Algebras and Locally Compact Groups, by Irving Kaplansky (1971) Several Complex Variables, by Raghavan Narasimhan (1971) Torsion-Free Modules, by Eben Matlis (1973) Stable Homotopy and Generalised Homology, by J. F. Adams (1974) Rings with Involution, by I. N. Herstein (1976) Theory of Unitary Group Representation, by George V. Mackey (1976) Inﬁnite-Dimensional Optimization and Convexity, by Ivar Ekeland and Thomas Turnbull (1983) Commutative Semigroup Rings, by Robert Gilmer (1984) Navier-Stokes Equations, by Peter Constantin and Ciprian Foias (1988) Essential Results of Functional Analysis, by Robert J. Zimmer (1990) Fuchsian Groups, by Svetlana Katok (1992) Topological Classiﬁcation of Stratiﬁed Spaces, by Shmuel Weinberger (1994) Unstable Modules over the Steenrod Algebra and Sullivan’s Fixed Point Set Conjecture, by Lionel Schwartz (1994) Geometry of Nonpositively Curved Manifolds, by Patrick B. Eberlein (1996) Lectures on Exceptional Lie Groups, by J. F. Adams (1996) Dimension Theory in Dynamical Systems: Contemporary Views and Applications, by Yakov B. Pesin (1997) A Concise Course in Algebraic Topology, by J. P. May (1999) Harmonic Analysis and Partial Differential Equations: Essays in Honor of Alberto Calder´on, edited by Michael Christ, Carlos Kenig, and Cora Sadosky (1999) Topics in Geometric Group Theory, by Pierre de la Harpe (2000) Exterior Differential Systems and Euler-Lagrange Partial Differential Equations, by Robert Bryant, Phillip Grifﬁths, and Daniel Grossman (2003) Ratner’s Theorems on Unipotent Flows, by Dave Witte Morris (2005)

Geometry, Rigidity, and Group Actions Edited by Benson Farb and David Fisher

the university of chicago press • chicago and london

Benson Farb is professor of mathematics at the University of Chicago. He is the author of Problems on Mapping Class Groups and Related Topics and coauthor of Noncommutative Algebra. David Fisher is professor of mathematics at Indiana University.

The University of Chicago Press, Chicago 60637 The University of Chicago Press, Ltd., London © 2011 by The University of Chicago All rights reserved. Published 2011 Printed in the United States of America 20 19 18 17 16 15 14 13 12 11 12345 ISBN-13: 978-0-226-23788-6 (cloth) ISBN-10: 0-226-23788-5 (cloth) Library of Congress Cataloging-in-Publication Data Geometry, rigidity, and group actions/edited by Benson Farb and David Fisher. p. cm.—(Chicago lectures in mathematics series) Festschrift for Robert Zimmer on the occasion of his 60th birthday. ISBN-13: 978-0-226-23788-6 (cloth : alk. paper) ISBN-10: 0-226-23788-5 (cloth : alk. paper) 1. Rigidity (Geometry) 2. Group actions (Mathematics) 3. Manifolds (Mathematics) I. Zimmer, Robert J., 1947– II. Farb, Benson. III. Fisher, David. IV. Series: Chicago lectures in mathematics series. QA613.G465 2011 516 .11—dc22 2010026631 ∞ The paper used in this publication meets the minimum requirements of the American National Standard for Information Sciences—Permanence of Paper for Printed Library Materials, ANSI Z39.48-1992.

To Bob, mentor and friend

Contents

Preface

ix

PART 1 || Group Actions on Manifolds 1.

An Extension Criterion for Lattice Actions on the Circle 3 Marc Burger

2.

Meromorphic Almost Rigid Geometric Structures 32 Sorin Dumitrescu

3.

Harmonic Functions over Group Actions 59 Renato Feres and Emily Ronshausen

4.

Groups Acting on Manifolds: Around the Zimmer Program 72 David Fisher

5.

Can Lattices in SL (n, R) Act on the Circle? 158 Dave Witte Morris

6.

Some Remarks on Area-Preserving Actions of Lattices 208 Pierre Py

7.

Isometric Actions of Simple Groups and Transverse Structures: The Integrable Normal Case 229 Raul Quiroga-Barranco

8.

Some Remarks Inspired by the C0 Zimmer Program 262 Shmuel Weinberger

PART 2 || Analytic, Ergodic, and Measurable Group Theory 9.

Calculus on Nilpotent Lie Groups 285 Michael G. Cowling

viii / contents

10. A Survey of Measured Group Theory 296 Alex Furman

11. On Relative Property (T) 375 Alessandra Iozzi

12. Noncommutative Ergodic Theorems 396 Anders Karlsson and François Ledrappier

13. Cocycle and Orbit Superrigidity for Lattices in SL (n, R) Acting on Homogeneous Spaces 419 Sorin Popa and Stefaan Vaes

PART 3 || Geometric Group Theory 14. Heights on SL2 and Free Subgroups 455 Emmanuel Breuillard

15. Displacing Representations and Orbit Maps 494 Thomas Delzant, Olivier Guichard, François Labourie, and Shahar Mozes

16. Problems on Automorphism Groups of Nonpositively Curved Polyhedral Complexes and Their Lattices 515 Benson Farb, Chris Hruska, and Anne Thomas

17. The Geometry of Twisted Conjugacy Classes in Wreath Products 561 Jennifer Taback and Peter Wong

PART 4 || Group Actions on Representations Varieties 18. Ergodicity of Mapping Class Group Actions on SU(2)-Character Varieties 591 William M. Goldman and Eugene Z. Xia

19. Dynamics of Aut (Fn ) Actions on Group Presentations and Representations 609 Alexander Lubotzky

List of Contributors

645

Preface

In September 2007, a conference was held at the University of Chicago in honor of the sixtieth birthday of Robert J. Zimmer. The conference was a testament to and a celebration of Zimmer’s continuing and lasting inﬂuence on the ﬁelds of geometry, rigidity, and group actions. The wide variety of papers submitted to this volume are just one indication of that inﬂuence. This volume contains both survey papers and research papers. The common theme throughout concerns group actions. The general setup is that we have a group G, which is either a Lie group or a discrete group, acting on some space X . The examples are quite varied, and include, in (informal) order of decreasing “rigidity”: 1. algebraic actions on varieties 2. isometric actions on metric spaces 3. smooth actions on manifolds 4. measurable actions on measure spaces In each instance the goal is to relate algebraic/combinatorial properties of G to geometric/topological/measure-theoretic properties of X and the G action on X . In many cases, properties of G impose surprising restrictions on the possible choices of X and the possible actions on X . The papers by Goldman-Xia and Lubotzky concentrate on (1). The varieties here are representation and character varieties of certain ﬁnitely generated groups , with the automorphism group G = Aut() acting by algebraic automorphisms. This setup can be viewed as a dynamical system, where one can study its dynamical properties (as in Goldman-Xia’s paper), or as an algebraic system (as in Lubotzky’s paper), where information about G can be gleaned from properties of the action. A number of the papers in this volume are concerned with (2). The paper by Farb-Hruska-Thomas considers the case where X is a nonpositively curved

x / preface

simplicial complex and the group G is a lattice in the simplicial automorphism group of X ; here the goal is to extend the classical theory of lattices in algebraic groups over ﬁelds with a discrete valuation to this more general context. Breuillard’s paper provides a strengthening of the famous Tits alternative, and gives further applications to spectral gaps, group growth, and more. Here the G is an arbitrary, ﬁnitely generated subgroup of GL(2, C), and X is a projective space over various ﬁelds. The paper by Delzant-Guichard-Labourie-Mozes considers the general setup of groups acting isometrically on metric spaces. They relate the notion of a “displacing action” to the geometric property of an orbit being quasi-isometrically embedded. This setup is widely applicable, although the case of S-arithmetic groups is emphasized here. Iozzi considers isometric actions of semidirect products G of groups H with (normal) abelian groups A, acting unitarily on a Hilbert space X . The main goal is to give necessary and sufﬁcient conditions for the pair (G, A) to have relative property (T). Taback-Wong use the large-scale geometry of wreath products to study twisted conjugacy classes in these groups. Cowling’s paper develops some calculus on nilpotent groups with a view toward generalizing Mostow’s rigidity theorem concerning groups acting on symmetric spaces. The general study of smooth (usually volume-preserving) actions of (noncompact) Lie groups and their lattices on smooth manifolds goes by the name of the Zimmer program. The paper by Fisher gives a broad survey of this area, while the paper by Morris surveys the already rich case of actions on the circle. Burger’s paper also concerns actions of various large groups on the circle and provides a uniﬁed approach to many of the major results in this area. The Feres-Ronshausen paper studies general properties of groups acting on manifolds without an invariant volume by studying a certain type of harmonic function related to these actions. The primary applications in this paper also concern groups of diffeomorphisms of the circle. Moving up one dimension, Py’s paper provides a brief introduction to an interesting approach to studying volume-preserving diffeomorphisms of compact surfaces using quasi-morphisms. A major theme in the Zimmer program is the study of group actions that preserve geometric structures, a topic represented here in the papers by Dumitrescu and Quiroga-Barranco. While Zimmer’s focus was primarily on rigidity for smooth actions, Weinberger’s paper points to several interesting rigidity phenomena for actions by homeomorphisms. Another major topic of Zimmer’s research concerned cocycles over ergodic actions and orbit equivalence rigidity. Furman’s survey provides a comprehensive overview of this area. The Popa-Vaes paper illustrates some recent developments deepening connections between orbit equivalence rigidity and

preface / xi

operator algebras, and proves some striking new rigidity results. In a very different direction, Karlsson-Ledrappier write a survey on noncommutative ergodic theorems, an area in which cocycles over ergodic actions have played an interesting and important role. Acknowledgment. We would like thank the contributors to this volume, as well as the referees, who generously gave their time to this project.

PART 1

Group Actions on Manifolds

1

marc burger

AN EXTENSION CRITERION FOR LATTICE ACTIONS ON THE CIRCLE to bob zimmer

1. Introduction Let < G be a lattice in a locally compact second countable group G. The aim of this paper is to establish a necessary and sufﬁcient condition for a -action by homeomorphisms of the circle to extend continuously to G. This condition will be expressed in terms of the real bounded Euler class of this action. Combined with classical vanishing theorems in bounded cohomology, one recovers rigidity results of Ghys, Witte-Zimmer, Navas, and Bader-FurmanShaker in a uniﬁed manner. For a survey of various approaches to the problem of classifying lattice actions on the circle we refer to the paper by David Witte Morris [25] in this volume. Let Homeo+ (S1 ) be the group of orientation-preserving homeomorphisms of the circle and e ∈ H 2 (Homeo+ (S1 ), Z) the Euler class; recall that e corresponds to the central extension deﬁned by the universal covering of Homeo+ (S1 ). The Euler class admits a representing cocycle that is bounded and this deﬁnes a bounded class eb ∈ Hb2 (Homeo+ (S1 ), Z) called the bounded Euler class. The relevance of bounded cohomology to the study of group actions on the circle comes from a result of Ghys [14], namely that the bounded Euler class ρ ∗ (eb ) ∈ Hb2 (, Z) of an action ρ : → Homeo+ (S1 ) determines ρ up to quasiconjugation; a quasiconjugation is a self-map of the circle that is weakly cyclic order preserving and in particular not necessarily continuous; see Section 3 for details. If ebR denotes then the bounded class obtained by considering the bounded cocycle deﬁning eb as real valued, we call the invariant ρ ∗ (ebR ) ∈ Hb2 (, R) the real bounded Euler class of ρ. From this point of view we have the following dichotomy (see Proposition 3.2): (E) ρ ∗ (ebR ) = 0: in this case, ρ is quasiconjugated to an action of by rotations; as far as the extension problem is concerned, it reduces to the 3

4 / m. burger

properties of the restriction map Homc (G, R/Z) → Hom(, R/Z). (NE) ρ ∗ (ebR ) = 0: in this case, ρ is quasiconjugated to a minimal unbounded action; that is, every orbit is dense and the group of homeomorphisms ρ() is not equicontinuous. In the ﬁrst case (E) we call ρ elementary and in the second (NE) nonelementary; nonelementary actions are our main object of study in this paper. Concerning the extension problem, an issue that has to be taken care of is the existence of a nontrivial centralizer of the action under consideration. This is illustrated by the following:

example 1.1. Let < PSL(2, R) be a lattice that is nonuniform and torsion free. Since is a free group we can lift the identity to a homomorphism ρk : → PSL(2, R)k ⊂ Homeo+ (S1 ) into the k-fold cyclic covering of PSL(2, R), and this for every k ≥ 1. In this way we get an action that is minimal, unbounded, but for k ≥ 2 does not extend continuously to PSL(2, R). If is torsion-free cocompact, this construction applies provided k divides the Euler characteristic of , which is always the case for k = 2. Thus given a minimal unbounded action one is led to consider its topological S1 -factors; those are easily classiﬁed and in particular there is, up to conjugation, a unique factor ρsp : → Homeo+ (S1 ) that is strongly proximal (Proposition 3.7). This relies on arguments of Ghys that establish that the centralizer of ρ() is a ﬁnite cyclic group; the strongly proximal quotient is then obtained by passing to the quotient by this cyclic group. Our main result is:

theorem 1.2. Let < G be a lattice in a locally compact, second countable group G and ρ : → Homeo+ (S1 ) be a minimal unbounded action. Then the following are equivalent:

an extension criterion for lattice actions on the circle / 5

1) The real bounded Euler class ρ ∗ (ebR ) of ρ is in the image of the restriction map 2 (G, R) → Hb2 (, R). Hbc

2) The strongly proximal factor ρsp of ρ extends continuously to G. The main ingredient in the proof of Theorem 1.2 is a result that, for any countable group , characterizes the bounded classes in Hb2 (, R) obtained from minimal strongly proximal actions in terms of certain cocycles deﬁned on an appropriate Poisson boundary of ; see Theorem 4.5 in Section 4, where the result is proven in the more general context of locally compact, second countable groups. This leads to results of independent interest concerning the extent to which an action is determined by its real bounded Euler class and to the question of the possible values of its norm. These results are summarized in the following theorem and corollaries.

theorem 1.3. Let be a countable group. 1) For a homomorphism ρ : → Homeo+ (S1 ) we have ρ ∗ (ebR ) ≤

1 2

with equality if and only if ρ is quasiconjugated to a minimal strongly proximal action. 2) If two minimal strongly proximal actions ρ1 , ρ2 are not conjugated, then ρ1∗ (ebR ) − ρ2∗ (ebR ) = 1.

remark 1.4. 1) The ﬁrst assertion in Theorem 1.3 echoes results obtained in [7] concerning tight homomorphisms with values in a Lie group of Hermitian type. 2) Let ESP() ⊂ Hb2 (, R) denote the subset consisting of the real bounded Euler classes of minimal strongly proximal -actions and Z[ESP()] its Z-span. We will show (see Section 5) that the norm takes half-integral values on Z[EPS()]. In this context the following question arises, namely if Z[SP()] denotes the free abelian group on the set of conjugacy classes of minimal strongly proximal actions, can one determine the kernel of the homomorphism

6 / m. burger Z[SP()] → Z[ESP()]

ni [ρi ] → ni ρi (ebR ) and what is its signiﬁcance for the dynamics of -actions on S1 ? The following immediate corollary is another instance of the general principle that groups whose second bounded cohomology is ﬁnite-dimensional exhibit rigidity phenomena; compare, for example, with the case of actions by isometries on Hermitian symmetric spaces (see [4], [6]).

corollary 1.5. Let be a countable group and assume that Hb2 (, R) is ﬁnitedimensional. Then there are, up to conjugation, only ﬁnitely many minimal strongly proximal -actions on S1 . Together with the information concerning centralizers of minimal actions we deduce from Theorem 1.3,

corollary 1.6. 1) For a homomorphism ρ : → Homeo+ (S1 ) we have 1 : k∈N , ρ ∗ (ebR ) ∈ {0} ∪ 2k and this value equals (2k)−1 if and only if ρ is quasiconjugated to a minimal unbounded action whose centralizer is of order k. 2) For minimal unbounded actions ρ1 , ρ2 we have ρ1∗ (ebR ) = ρ2∗ (ebR ) if and only if, up to conjugation, ρ2 (γ ) = h(γ ) ρ1 (γ ), γ ∈ where h is a homomorphism with values in the centralizer of ρ1 (). The extension criterion in Theorem 1.2 leads to rigidity theorems when combined with the following two ingredients, namely vanishing theorems in bounded cohomology and the description of continuous homomorphisms from a locally compact group into Homeo+ (S1 ). Concerning the ﬁrst ingredient, we know that the restriction map 2 (G, R) → H 2 (, R) is a isomorphism in the following cases: Hbc b 1) Products (see [8], [21]): G = G1 × · · · × Gn is a Cartesian product of locally compact second countable groups and has dense projection in every factor Gi .

an extension criterion for lattice actions on the circle / 7

2) Higher-rank Lie groups (see [8]): G = G(k) where G is a connected almost simple k-group, k is a local ﬁeld and rankk G ≥ 2. Since it is elementary to classify continuous homomorphisms from a semisimple Lie group over a local ﬁeld into Homeo+ (S1 ) one obtains by combining (1) and (2),

corollary 1.7. ([15], [27]) Let < G := α∈A Gα (kα ) be an irreducible lattice where kα are local ﬁelds, Gα is a connected, simply connected, almost simple kα -group of positive rank. Assume that the sum of the kα -ranks of Gα is at least 2. For a homomorphism ρ : → Homeo+ (S1 ) one of the following holds up to quasiconjugation: 1) ρ has a ﬁnite orbit. 2) ρ is minimal unbounded and its strongly proximal factor ρsp extends continuously to G factoring via the projection on a factor of the form Gα (kα ) = SL(2, R). Concerning locally compact subgroups of Homeo+ (S1 ), one has, owing to the solution of Hilbert’s ﬁfth problem, a wealth of information and in particular, those that are connected and minimal have a simple classiﬁcation; they are up to conjugation, either Rot the subgroup of rotations or PSL(2, R)k , the k-fold cyclic covering of PSL(2, R) (see [10], [16]). When studying continuous homomorphisms from a locally compact group G into Homeo+ (S1 ) one has to deal with the fact that the image is not necessarily closed. In any case we have,

theorem 1.8. Let G be a locally compact group and π : G → Homeo+ (S1 ) a continuous and minimal action. Then one of the following holds: 1) π is conjugated into the group Rot of rotations and has a dense image in it. 2) π surjects onto PSL(2, R)k for some k ≥ 1, up to conjugation. 3) Ker π is an open subgroup of G. From Theorem 1.8 and the vanishing result for products mentioned above we obtain,

corollary 1.9. ([26], [2]) Let G = G1 × · · · × Gn be a product of locally compact second countable groups and < G a lattice with dense projections on each factor Gi . Assume that ρ : → Homeo+ (S1 ) is minimal unbounded. Then the strongly proximal quotient ρsp extends continuously to G,

8 / m. burger

(ρsp )ext : G → Homeo+ (S1 ) and we have one of the following: 1) Ker(ρsp )ext is open in G. 2) Up to conjugation (ρsp )ext factors via a projection onto some factor Gi followed by a continuous surjection onto PSL(2, R) ⊂ Homeo+ (S1 ). Finally, as shown by Bader, Furman, and Shaker, there is also in the context of actions on the circle a commensurator superrigidity theorem, which we state in a way that is somewhat different, but equivalent to their thm. C in [2].

theorem 1.10. ([2]) Let G be a locally compact second countable group, < G a lattice, and < G a subgroup such that ⊂ ⊂ CommG and is dense in G. Let ρ : → Homeo+ (S1 ) be a homomorphism such that ρ() is minimal and unbounded. Then the strongly proximal quotient ρsp extends continuously to G, (ρsp )ext : G → Homeo+ (S1 ) and we have one of the following: 1) Ker(ρsp )ext is open of inﬁnite index in G. 2) (ρsp )ext surjects onto PSL(2, R) ⊂ Homeo+ (S1 ). Let us make the following comment about the hypothesis of Theorem 1.10. If ρ : → Homeo+ (S1 ) is a homomorphism such that ρ() is unbounded, then either ρ() is minimal or there is an exceptional minimal set K ⊂ S1 (see Section 3) that is easily seen to be ρ()-invariant. Thus ρ is quasiconjugated to an action of for which is minimal and unbounded. The above result follows easily from the extension criterion (Theorem 1.2) and the following general fact concerning bounded cohomology:

theorem 1.11. ([20]) Let < G and < < CommG be as in Theorem 1.10, in particular is dense in G. Then the image of the restriction map Hb2 (, R) → Hb2 (, R) 2 (G, R) in H 2 (, R). coincides with the image of Hbc b

We will leave to the interested reader the exercise of deducing Theorem 1.10 from Theorem 1.11 and refer to [20] for elementary proofs of the isomor-

an extension criterion for lattice actions on the circle / 9

phism results for products and higher-rank groups mentioned above, as well as the proof of Theorem 1.11.

acknowledgments: Thanks to Luis Hernandez for his kind invitation to CIMAT where this work was completed. Thanks to Uri Bader and Alex Furman for sharing with me their result on boundary maps. 2. Boundary Maps In this section we present a general existence and uniqueness result concerning measurable equivariant maps that is due to Bader-Furman and is of general interest in rigidity theory. Here and in the sequel G is a second countable locally compact group, G × M → M is a continuous action on a compact metrisable space M, μ ∈ M1 (G) is a spread-out probability measure on G, and (B, νB ) is a standard Lebesgue G-space such that νB is μ-stationary.

theorem 2.1. ([1]) Assume that 1) the G-action is minimal and strongly proximal. 2) For every sequence (gn )n≥1 in G there exists a subsequence (nk )k≥1 such that M→M m −→ gnk (m) converges pointwise. Then every measurable G-equivariant map ϕ : B → M1 (M) takes values in the subset of Dirac measures.

remark 2.2. When G is discrete countable and M = S1 , a result essentially equivalent to Theorem 2.1 has been obtained by Deroin, Kleptsyn, and Navas (see [9]). We recall here that a G-action is strongly proximal if for every μ ∈ M1 (M) the closure Gμ ⊂ M1 (M) contains a Dirac mass. Proof. The main step consists in showing that the G-space M is μ-proximal in the sense of (2.6), def. 3 in [22] VI: this means that for every μ-stationary

10 / m. burger

measure ν ∈ M1 (M) the map 2.1

φν (ω) := lim (x1 , . . . xn ) ν, ω = (xn )n≥1 n→∞

with values in M1 (M), deﬁned almost everywhere on the inﬁnite product (S, μS ) := ∞ n=1 (G, μ), takes values in the set of Dirac measures; this we proceed to show now. The existence of the limit (2.1) follows from the martingale convergence theorem (prop. 2.4 in [22] VI); in addition one has the following remarkable fact ([13], [17] and lemma 1.33 in [11]): lim (x1 . . . xn g) ν = φν (ω)

2.2

n→∞

for μS -almost every ω ∈ S and λ-almost every g ∈ G, where λ is the left Haar measure on G. Fix now a sequence ω = (xn )n≥1 and a subset E ⊂ G of full λ-measure such that (2.2) holds for ω and all g ∈ E. Passing to a subsequence we may assume that the limit s(m) := lim (x1 . . . xn )(m) n→∞

exists for every m ∈ M. Then the map s : M → M, although in general not continuous, admits points of continuity; in fact, they form a Gδ -subset of M. Let then m ∈ M be a point of continuity of s. Since the G-action is minimal and strongly proximal, there exists a sequence (gn )n≥1 in G such that limn→∞ gn ν = δm . By the continuity of the G-action on M there exists for each n ≥ 1 an open neighborhood Un gn such that whenever (hn )n≥1 ∈ n≥1 Un , limn→∞ hn ν = δm . Since λ(Un ) > 0, ∀n ≥ 1, we may choose hn ∈ Un ∩ E and obtain a sequence (hn )n≥1 such that 2.3

a) lim hn ν = δm n→∞

b) lim (x1 . . . xn hk ) ν = φν (ω), for every k ≥ 1 . n→∞

Let V be a neighborhood of s(m) and W m an open neighborhood such that W ⊂ s−1 (V ). Taking into account that s : M → M is a Borel map we have (shk )ν(V ) = (hk ν)(s−1 (V )) ≥ hk ν(W ), which together with lim hk ν(W ) = 1 (see (2.3 a)) implies that k→∞

2.4

lim (shk )(ν) = δs(m) .

k→∞

For every k ≥ 1 and every continuous function f ∈ C(M) we have

an extension criterion for lattice actions on the circle / 11

(shk )ν(f ) =

M

f (shk x) dν(x)

=

lim f (x1 . . . xn hk x) dν(x)

M n→∞

= φν (ω)( f ) where in the second equality we have used the deﬁnition of s and the continuity of f , while in the third we have used dominated convergence and (2.3 b). Thus shk ν = φν (ω) for every k ≥ 1, which together with (2.4) implies that φν (ω) = δs(m) , and hence since ω could be chosen from a subset of full measure, φν takes values in the set of Dirac masses almost everywhere; thus M is μ-proximal. Let now (B, νB ) and ϕ : B → M1 (M) be as in Theorem 2.1. Then ϕ∗ (νB ) ∈ M1 (M1 (M)) is a μ-stationary probability measure on M1 (M) and since M is μ-proximal, prop. 2.9 in [22] VI implies that the support of ϕ∗ (νB ) is contained in the subset of M1 (M) consisting of Dirac masses. This concludes the proof of Theorem 2.1. In the following proposition we give examples of groups of homeomorphisms with the property in Theorem 2.1 (2).

proposition 2.3. Let (G, M) be one of the following pairs consisting of a compact metric space M and a subgroup G of the group of homeomorphisms of M: 1) M = PVk , G = PGL(Vk ) where Vk is a ﬁnite-dimensional vector space over a local ﬁeld k. 2) M = S1 , G = Homeo+ (S1 ). Then for any sequence (gn )n≥1 there is a subsequence such that (gnk )k≥1 converges pointwise. Question 2.4

Let X be a proper complete CAT(0)-space, M = X (∞) its visual boundary, and G = Isom(X ). Does (G, M) satisfy the conclusion of Proposition 2.3? The ﬁrst statement in Proposition 2.3 follows easily by recurrence on dim Vk using Furstenberg’s lemma [12]. Concerning the second, we may compose gn by an appropriate rotation rn such that rn ◦ gn ﬁxes a point in S1 and, passing to a subsequence we may assume that (rn )n≥1 converges uniformly; Proposition 2.3 (2) will then follow from

12 / m. burger

lemma 2.5. Let fn : [0, 1] → [0, 1], n ≥ 1, be a sequence of monotone increasing maps. Then there exists a subsequence converging pointwise. Proof. We may assume that f (r) := limn→∞ fn (r) exists ∀r ∈ Q ∩ [0, 1]. Then f is monotone increasing on Q ∩ [0, 1] and thus F(x) := sup{f (r) : r ≤ x, r ∈ Q ∩ [0, 1]} is a well-deﬁned function on [0, 1] that is monotone increasing and extends f . The set D ⊂ [0, 1] of discontinuities of F is at most countable; passing to a subsequence we may assume then that limn→∞ fn (d) exists for every d ∈ D. Let now t ∈ / D be a point of continuity of F. For every t , t ∈ Q with t ≤ t ≤ t we have F(t ) ≤ ξ ≤ F(t ) where ξ is any accumulation point of the sequence (fn (t))n≥1 . Letting t , t approach t and using the continuity of F we get ξ = F(t) and hence limn→∞ fn (t) = F(t).

3. Elementary Properties of Actions on the Circle In this section we establish the basic dichotomy concerning actions on the circle in terms of their real bounded Euler class, announced in the introduction, and show the uniqueness of the strongly proximal quotient. Along the way we will take the opportunity to recall some well-known facts that will be used later on. Let S1 = Z\R and consider the central extension i

p

+ 1 0 → Z → Homeo+ Z (R) → Homeo (S ) → (e)

where Homeo+ Z (R) is the group of increasing homeomorphisms of R commuting with integer translations and i(n) = T n , where T (x) = x + 1, x ∈ R. A section of p is obtained by associating to every f ∈ Homeo+ (S1 ) the unique lift f : R → R satisfying f (0) ∈ [0, 1). Then we have 3.1

fg T c(f ,g) = f g, f , g ∈ Homeo+ (S1 )

and c(f , g) ∈ {0, 1} is an inhomogeneous 2-cocycle called the Euler cocycle whose class e ∈ H 2 (Homeo+ (S1 ), Z) is called the Euler class. Considering c as a Z-valued, respectively R-valued bounded cocycle leads respectively to the bounded Euler class eb ∈ Hb2 (Homeo+ (S1 ), Z) and the real bounded Euler class ebR ∈ Hb2 (Homeo+ (S1 ), R). Given a group and a homomorphism ρ : → Homeo+ (S1 ) we obtain accordingly three invariants associated to ρ, namely, the Euler class ρ ∗ (e) ∈ H 2 (, Z), the bounded Euler class ρ ∗ (eb ) ∈ Hb2 (, Z), and the real bounded Euler class ρ ∗ (ebR ) ∈ Hb2 (, R). The information

an extension criterion for lattice actions on the circle / 13

contained in ρ ∗ (e) is of purely algebraic nature; ρ ∗ (e) = 0 if and only if ρ lifts to a homomorphism with values in Homeo+ Z (R), which for example is the case when is a free group. Concerning the bounded Euler class we have on the other hand,

theorem 3.1. ([14]) Two actions ρ0 , ρ1 are quasiconjugated if and only if ρ0∗ (eb ) = ρ1∗ (eb ). There are several equivalent ways to look at quasiconjugations. Ghys’s original deﬁnition is as follows: a quasiconjugation is a map h : S1 → S1 induced by a monotone increasing map h : R → R commuting with integer translations; observe that a quasiconjugation is not necessarily continuous. Two actions ρ0 , ρ1 are then quasiconjugated if there is such a map h with 3.2

hρ0 (γ ) = ρ1 (γ ) h, ∀γ ∈ .

Quasiconjugation is an equivalence relation. Observe that if we set (h)+ (x) = limy>x h(y), (h)− (x) = limy<x h(y), then (h)+ and (h)− induce respectively a right continuous h+ and a left continuous h− quasiconjugacy and if (3.2) holds, then we have h± ρ0 (γ ) = ρ1 (γ ) h± , ∀γ ∈ as well. An intrinsic way to deﬁne quasiconjugation rests on the consideration of the orientation cocycle o : (S1 )3 → {−1, 0, 1} deﬁned as follows: ⎧ 1 if x, y, z are positively oriented, ⎪ ⎪ ⎨ 3.3 o(x, y, z) = −1 if x, y, z are negatively oriented, ⎪ ⎪ ⎩ 0 if x, y, z are not pairwise distinct. Then a map h : S1 → S1 is a quasiconjugacy if and only if o(h(x), h(y), h(z)) ≥ 0 whenever o(x, y, z) ≥ 0 . A third characterization is obtained as follows. Deﬁne for x, y ∈ S1 the open interval (x, y) := {z ∈ S1 : o(x, z, y) = 1} and [x, y], (x, y], [x, y) in the obvious way. Given a point b ∈ S1 and a probability measure μ ∈ M1 (S1 ), the map 3.4

hb,μ (x) := μ([b, x)) mod Z

is a left continuous quasiconjugacy and every such quasiconjugacy is obtained this way for a unique pair (b, μ). The behavior under composition with homeomorphisms is particularly simple,

14 / m. burger

hb,μ (ϕ −1 (x)) = hϕ(b),ϕ∗ (μ) (x), ϕ ∈ Homeo+ (S1 ) .

3.5

A very useful tool in the study of group actions on S1 is the following trichotomy for ρ : → Homeo+ (S1 ), see ([18], chap. IV-3)

3.6

1)

There is a ﬁnite orbit and all ﬁnite orbits have the same cardinality.

2)

The action is minimal.

3)

There is a unique proper minimal invariant closed subset F ⊂ S1 , which is a Cantor set. +

In the ﬁrst case ρ is quasiconjugated to an action ρ0 by rotations; a quasiconjugacy is given by hb,μ where μ is the uniform measure along a ﬁnite orbit. Then ρ0 () is ﬁnite cyclic. In the second case ρ() is either bounded (equicontinuous), preserves hence a probability measure μ necessarily with full support and no atoms, and can be conjugated into the subgroup of rotations by the homeomorphism hb,μ ; or ρ() is unbounded. In the third case one can collapse each connected component of S1 \F to a point and obtain this way a continuous quasiconjugacy between ρ and a minimal action. In fact such a quasiconjugacy is given by hb,μ where μ is a probability measure without atoms, and with support, precisely the Cantor set F. With this at hand and denoting as usual rot : Homeo+ (S1 ) → Z\R, the rotation number function, we have

proposition 3.2. Let ρ : → Homeo+ (S1 ) be a homomorphism. 1) ρ ∗ (ebR ) = 0 if and only if ρ is quasiconjugated to a subgroup of the group of rotations. When this is the case, → Z\R γ −→ rot ρ(γ ) is a homomorphism and ρ is quasiconjugated to the action by rotations deﬁned by it. 2) ρ ∗ (ebR ) = 0 if and only if ρ is quasiconjugated to a minimal unbounded action.

an extension criterion for lattice actions on the circle / 15

Proof. 1) Assume ρ ∗ (ebR ) = 0. Then there exists β : → R bounded with dβ(γ , η) := β(γ ) − β(γ η) + β(η) = c(ρ(γ ), s(η)),

∀γ , η ∈ .

Since c( , ) takes values in Z, the map f : → Z\R, γ −→ f (γ ) = β(γ ) mod Z, is a homomorphism. Let ρ0 (γ )(x) := x + f (γ ), x ∈ Z\R be the corresponding action by rotations. Denoting by { } the fractional part of a real number, we then have ρ0 (γ ) (x) = x + {β(x)}, ∀γ ∈ and hence c(ρ0 (γ ), ρ0 (η)) = d({β})(γ , η), ∀γ , η ∈ . This implies for all γ , η ∈ : c(ρ0 (γ ), ρ0 (η)) + d([β])(γ , η) = c(ρ(γ ), ρ(η)) where [β] : → Z is the bounded function associating to γ the integral part [β(γ )] of β(γ ). As a result ρ0∗ (eb ) = ρ ∗ (eb ) and by Ghys’s theorem (Theorem 3.1) ρ and ρ0 are quasiconjugated. Since the rotation number is an invariant of quasiconjugation we have f (γ ) = rot ρ(γ ), ∀γ ∈ and hence γ −→ rot ρ(γ ) is a homomorphism. Conversely, if ρ is quasiconjugated to a homomorphism ρ0 : → Rot, then ρ ∗ (eb ) = ρ0∗ (eb ) and hence ρ ∗ (ebR ) = ρ0∗ (ebR ). But since Rot is compact, ebR |Rot = 0, which implies 0 = ρ0∗ (ebR ) = ρ ∗ (ebR ). 2) If ρ ∗ (ebR ) = 0 then it follows from 1) that ρ is not quasiconjugated into Rot and hence by the trichotomy (3.6) and the discussion following it, ρ is quasiconjugated to a minimal unbounded action. Since bounded cohomology with R-coefﬁcients vanishes for amenable groups we conclude that

corollary 3.3. If is amenable then for any homomorphism ρ : → Homeo+ (S1 ), the map → Z\R γ −→ rot ρ(γ ) is a homomorphism. A simpler proof of the last corollary consists in observing that the rotation number with respect to a -invariant measure coincides with Poincaré’s rotation number and is clearly a group homomorphism. Now we turn to a closer study of minimal unbounded actions. In this context the following deﬁnition will be useful:

16 / m. burger

definition 3.4. A nontrivial S1 -factor of an action ρ : → Homeo+ (S1 ) is a pair (ϕ, ρ0 ) consisting of a nonconstant continuous map ϕ : S1 → S1 and an action ρ0 : → Homeo+ (S1 ) with ϕρ(γ ) = ρ0 (γ )ϕ, ∀γ ∈ . Denoting by ZH+ (L) the centralizer in Homeo+ (S1 ) of a subgroup L we have

theorem 3.5. ([23], [16]) Let ρ : → Homeo+ (S1 ) be minimal unbounded. Then the centralizer Cρ := ZH+ (ρ()) of ρ() in Homeo+ (S1 ) is ﬁnite cyclic and the S1 -factor (ϕ, ρ0 ), where ϕ : S1 → Cρ \S1 is the quotient map and ρ0 the quotient action, is strongly proximal. For minimal circle actions one observes that strong proximality is equivalent to the property that every proper interval can be contracted; we recall that a subset F ⊂ S1 can be contracted if there exists a sequence (γn )n≥1 such that limn→∞ diam(ρ(γn )F) = 0. The idea of the proof of Theorem 3.5 is then as follows: Since ρ is minimal unbounded, in particular not equicontinuous, every point x ∈ S1 has a neighborhood that can be contracted. For every x ∈ S1 the set {I = [x, y)|I can be contracted} is then totally ordered by inclusion, nonempty, and hence contains a unique maximal element that we denote [x, θ(x)). Then one veriﬁes that θ : S1 → S1 is a homeomorphism that is periodic and Cρ = ZH+ (ρ()) = θ . By construction, every proper interval in the quotient Cρ \S1 can be contracted. We record the following immediate

corollary 3.6. Let ρ : → Homeo+ (S1 ) be minimal unbounded. Then ρ is strongly proximal if and only if ZH+ (ρ()) = (e). We now establish the uniqueness of a proximal S1 -factor:

proposition 3.7. Let ρ : → Homeo+ (S1 ) be minimal unbounded and let (ϕ, ρ0 ) be a nontrivial S1 -factor. If ρ0 is strongly proximal, then ϕ is up to conjugation the quotient map by ZH+ (ρ()). First we will need

lemma 3.8. If ρ is minimal and (ρ0 , ϕ) is a nontrivial S1 -factor, then deg ϕ = 0 and ϕ is surjective. Proof. Assume that deg ϕ = 0 and let

ϕ : S1 → R be a lift of ϕ. Then there is for every γ ∈ a c(γ ) ∈ Z such that ϕ.

ϕ ρ(γ ) = T c(γ ) ρ0 (γ )

an extension criterion for lattice actions on the circle / 17

Let M = sup

ϕ (x) and x0 ∈ S1 with M =

ϕ (x0 ). Then x∈S1

ϕ (x) = sup

ϕ (ρ(γ )x)

ϕ (x0 ) = sup

x

x

= c(γ ) + ρ0 (γ ) ( sup

ϕ (x)) x

ϕ (x0 )), = c(γ ) + ρ0 (γ ) (

which implies that ϕ(x0 ) ∈ S1 is ﬁxed by ρ0 (γ ), ∀γ ∈ . The nonempty closed subset ϕ −1 (ϕ(x0 )) being ρ()-invariant equals S1 and thus ϕ is constant. Proof of Proposition 3.7. Let θ be a generator of ZH+ (ρ()) constructed as indicated in the discussion following Theorem 3.5. We claim that ϕθ = ϕ. For x ∈ S1 we consider the closed connected subset Ix := ϕ([x, θ (x)]) and, for a ﬁxed x, distinguish two cases: a) Ix = S1 : then either ϕ(θ(x)) = ϕ(x), which implies by equivariance that ϕθ and ϕ coincide on ρ() · x and hence everywhere, or there is y ∈ (x, θ (x)) with ϕ(x) = ϕ(y) and ϕ([x, y]) = S1 . But by the construction of θ, [x, y] can be contracted, which by continuity of ϕ implies the same for ϕ([x, y]) = S1 ; this is absurd. b) Ix is a proper subset of S1 : Observe that for all γ ∈ , ρ0 (γ )Ix = Iρ(γ )x . Since Ix is a proper interval and ρ0 is strongly proximal we can choose a sequence (γn )n≥1 such that diam (ρ0 (γn )Ix ) tends to zero for n → ∞ and (ρ(γn )x)n≥1 converges to some point y, which by continuity of ϕ implies that ϕ(y) = ϕθ(y). Again by using equivariance and minimality we get ϕ = ϕθ. Thus the claim is established and therefore by passing to the quotient by ZH+ (ρ()) we may assume that ρ is strongly proximal as well. We already

know (Lemma 3.8) that deg ϕ = 0 and ϕ is surjective. Assume then that there exists s = t in S1 with ϕ(s) = ϕ(t). Since deg ϕ = 0 we have either ϕ([s, t]) = S1 or ϕ([t, s]) = S1 . Assume without loss of generality that the former occurs. Since [s, t] is a proper interval in S1 it can be contracted by ρ, which clearly implies a contradiction. Thus ϕ is injective as well and hence a homeomorphism.

4. Strongly Proximal Actions, Boundary Maps, and Cocycles In this section we study continuous group actions of a locally compact group G on the circle using boundary maps. The main goal is Theorem 4.5, which

18 / m. burger

gives a description of the space of conjugacy classes of minimal strongly proximal actions in terms of an explicit space of cocycles on appropriate Poisson boundaries of G. 4.1. Cohomological Preliminaries

For a nondiscrete group some care must be taken when deﬁning the analogues of the various Euler classes in the continuous context. Let G be locally compact, • (G, A) denotes the second countable, and let A be either Z or R. Then Hbc cohomology deﬁned via A-valued bounded Borel cochains on G; in the case A = R these cohomology groups coincide with those obtained by taking the subcomplex of bounded continuous cochains on G; for more details see [5], §2.3. Given a continuous action ρ : G → Homeo+ (S1 ) the function (g, h) −→ c(ρ(g), ρ(h)) is a bounded Borel (inhomogeneous) 2-cocycle on G and leads 2 (G, Z) and ρ ∗ (eb ) ∈ H 2 (G, R), which are respectively to classes ρ ∗ (eb ) ∈ Hbc R bc the bounded Euler class and the real bounded Euler class of the continuous action. Recall that if (B, νB ) is a standard Lebesgue G-space that is doubly ergodic and amenable, there is a canonical isometric isomorphism; see [24] thm. 7.5.3, 4.1

2 ∞ 3 Hbc (G, R) Z Lalt (B , R)G

where the right-hand side is the space of measurable, essentially bounded, alternating G-invariant cocycles on B3 .

proposition 4.1. Let ρ : G → Homeo+ (S1 ) be a continuous action and assume that there exists a measurable G-equivariant map ϕ : B → S1 . Then under the isomorphism in (4.1) the class ρ ∗ (ebR ) corresponds to the cocycle B3 → R 1 2

(x, y, z) −→ − o(ϕ(x), ϕ(y), ϕ(z)) . If in addition the essential image of ϕ contains at least three points, we have ρ ∗ (ebR ) =

1 . 2

We will need the following explicit relationship between the Euler and orientation cocycle; see for instance [19] Lemma 2.1.

lemma 4.2. For every f , g ∈ Homeo+ (S1 ) we have

an extension criterion for lattice actions on the circle / 19

2c( f , g) = − o(˙o, f (˙o), f g(˙o)) + 1 + (δo˙ ( f g(˙o)) − δo˙ ( f (˙o)) − δo˙ (g(˙o))), where o˙ ∈ Z\R. Proof of Proposition 4.1. This is a standard argument using [3]. Since G acts ∞ ((S1 )•+1 , R), d) of bounded alternatcontinuously on S1 , the G-complex (Balt 1 ing Borel cochains on S is a strong resolution of the trivial G-module R (see [3] Prop. 2.1). Then, owing to the properness of the G-action on G and the amenability of the G-action on B, the two complexes of G-modules ∞ (G•+1 , R), d) and (L∞ (B•+1 , R), d) are strong resolutions of R by relatively (Lalt alt injective G-Banach modules [24] section 7.5. Then there is a morphism of G-complexes 4.2

∞ •+1 ∞ •+1 c • : Lalt (G , R) → Lalt (B , R)

extending the identity and any two such are G-equivariantly homotopic; the canonical map induced in cohomology is then an isometric isomorphism that in degree two gives the one mentioned in (4.1). Consider the morphisms of G-resolutions ∞ 1 n+1 ∞ n+1 (G , R) ρ (n) : Balt (S ) , R) → Lalt

and ∞ ∞ n+1 (B , R) ϕ (n) : Balt ((S1 )n+1 , R) → Lalt

deﬁned respectively for α : (S1 )n+1 → R by ρ (n) (α)(g0 , . . . , gn ) := α(ρ(g0 )˙o, . . . , ρ(gn ) o˙ ) and ϕ (n) (α)(x0 , . . . , xn ) := α(ϕ(x0 ), . . . , ϕ(xn )) . Then c (•) ρ (•) and ϕ (•) are both morphisms of G-complexes extending the iden∞ (B•+1 , R), d) is a resolution by relatively injective tity and therefore, since (Lalt G-modules, induce the same map in cohomology. Now we specialize this ∞ (S•+1 )) be the class deﬁned by the orito degree n = 2. Let [o] ∈ H 2 (Balt entation cocycle. Then we have by Lemma 4.2: ρ (2) ([o]) = 2ρ ∗ (ebR ) and in addition that under the isometric isomorphism c (2) , ρ (2) ([o]) is represented by (x, y, z) −→ o(ϕ(x), ϕ(y), ϕ(z)). In particular 2ρ ∗ (ebR ) is the essential supremum of the latter cocycle.

20 / m. burger

We will see below that the existence of a map as in Proposition 4.1 imposes strong conditions on the type of action considered. Before we turn to this we will need

lemma 4.3. Let ρ0 , ρ1 : G → Homeo+ (S1 ) be continuous actions and ϕ : S1 → S1 a degree k covering such that ρ0 (g) ϕ = ϕ ρ1 (g),

∀g ∈ G .

Then we have the following equality ρ0∗ (eb ) = k ρ1∗ (eb ) 2 (G, Z). in Hbc

Proof. Let p : R → Z\R denote projection and Mk : R → R, x −→ k · x, so that up to conjugation we have p Mk = ϕp. Then we have ∀g ∈ G: p Mk ρ1 (g) = ϕ p ρ1 (g) = ϕ ρ1 (g) p = ρ0 (g) ϕp = ρ0 (g) p Mk = p ρ0 (g) Mk . Therefore there is α(g) ∈ Z such that 4.3

Mk ρ1 (g) = T α(g) ρ0 (g) Mk .

Evaluation at 0 ∈ R gives k ρ1 (g) (0) = α(g) + ρ0 (g) (0), which ﬁrst shows that α : G → Z is a Borel function and furthermore implies that supg∈G |α(g)| ≤ k + 1. Applying (4.3) to products and using the relation deﬁning the Euler cocycle we get k c(ρ1 (g), ρ1 (h)) + α(gh) − α(g) − α(h) = c(ρ0 (g), ρ0 (h)), which proves the lemma.

Now we come to our ﬁrst application.

corollary 4.4. Let μ ∈ M1 (G) be a spread-out probability measure on G and (B, νB ) a standard Lebesgue G-space, where νB is μ-stationary. Assume that B is doubly ergodic and amenable. Given a minimal unbounded action ρ : G → Homeo+ (S1 ) the following are equivalent:

an extension criterion for lattice actions on the circle / 21

1) ρ is strongly proximal. 2) There exists a G-equivariant measurable map ϕ : B → S1 . 3) ρ ∗ (ebR ) = 12 . Proof. (1) =⇒ (2): follows immediately from Theorem 2.1 and Proposition 2.3 (3). (2) =⇒ (3): follows from Proposition 4.1. (3) =⇒ (1): Let (ψ, ρ0 ) be the strongly proximal factor of ρ. Applying the implication (1) =⇒ (3) to ρ0 we get ρ0∗ (ebR ) = 12 , and for k = deg ψ, we get from Lemma 4.3 that ρ0∗ (ebR ) = k · ρ ∗ (ebR ), which implies k = 1 and hence that ρ is strongly proximal. 4.2. Minimal Strongly Proximal Actions and Cocycles

The main result of this section is

theorem 4.5. Let G be locally compact second countable and let μ ∈ M1 (G), (B, νB ) be as in Corollary 4.4. Then there is a bijection between a) The set of conjugacy classes of continuous G-actions on S1 that are minimal and strongly proximal. b) The set of measurable functions, up to equality almost everywhere, ω : B3 → R such that 1) ω is an alternating strict cocycle. 2) ω(gx, gy, gz) = ω(x, y, z) for every g ∈ G and almost every (x, y, z) ∈ B3 . 3) ω takes values in {±1} almost everywhere. This bijection is implemented by the map to which a continuous minimal strongly proximal action ρ : G → Homeo+ (S1 ) associates the cocycle ω(x, y, z) = o(ϕ(x), ϕ(y), ϕ(z)) where ϕ : B → S1 is the map given by Corollary 4.4. The proof of the theorem is divided into two steps. Fix ω : B3 → R satisfying the properties (1), (2), and (3) above. Step 1: We show that there exists a measurable map ϕ : B → S1 satisfying the following properties:

22 / m. burger

a) ϕ∗ (νB ) ∈ M1 (S1 ) has no atoms. b) the essential image Ess Im ϕ of ϕ equals S1 . c) o(ϕ(x), ϕ(y), ϕ(z)) = ω(x, y, z) for almost every (x, y, z) ∈ B3 . Step 2: Given a measurable function ϕ : B → S1 satisfying (a), (b), and (c) we construct a continuous homomorphism πϕ : G → Homeo+ (S1 ) such that ϕ(gx) = πϕ (g)(ϕ(x)), for every g ∈ G and almost every x ∈ B, and conclude that πϕ is minimal strongly proximal. Finally we show that given ϕ, ψ satisfying (a), (b), and (c), then πϕ and πψ are conjugate. Combining Corollary 4.4, Step 1, and Step 2 clearly completes the proof of Theorem 4.5. 4.3. Step 1: The Construction of the Measurable Map

We ﬁx once and for all a measurable map ω : B3 → R satisfying properties (1), (2), and (3) in Theorem 4.5. For every (x, y) ∈ B2 deﬁne I(x, y) = {z ∈ B : ω(x, z, y) = 1}, which is measurable. The next few lemmas are intended to show that the sets I(x, y) behave like intervals on S1 . We will use the notation E ≡ F to indicate that two sets E, F differ by a set of measure zero; similarly E ⊂ F means that E is contained in F up to a set of measure zero.

•

lemma 4.6. 1) I(x, y) ∪ I(y, x) ≡ B for a.e. (x, y) ∈ B2 . 2) I(x, y) ∩ I(y, x) = φ, ∀(x, y) ∈ B2 . 3) ∀g ∈ G and a.e. (x, y) ∈ B2 , g(I(x, y)) ≡ I(gx, gy). 4) For a.e. (x, y) ∈ B2 , 0 < νB (I(x, y)) < 1. Proof. 1) follows from the fact that ω is alternating and ω(x, y, z) ∈ {±1} for a.e. (x, y, z) ∈ B3 . 2) is obvious. 3) follows from the G-invariance of ω and the fact that the G-action on B preserves the measure class of νB . 4) Consider A1 = {(x, y) ∈ B2 : νB (I(x, y)) = 1} A0 = {(x, y) ∈ B2 : νB (I(x, y)) = 0} .

an extension criterion for lattice actions on the circle / 23

Denoting ∧ the map (x, y) → (y, x), we have from (1) that A1 ≡ A0 ; in addition we have that A1 ∩ A0 = φ and both sets are measurable and G-invariant by (3). Thus if νB2 (A1 ) > 0, then νB2 (A0 ) = νB2 ( A0 ) = νB2 (A1 ) > 0, which contradicts 2 the ergodicity of the G-action on B . Thus A1 and A0 are both of measure zero.

lemma 4.7. 1) For (a, c) ∈ B2 and b ∈ I(a, c) we have 4.4

I(a, c) ⊃ I(a, b) ∪ I(b, c)

4.5

I(a, b) ∩ I(b, c) = φ .

2) For a.e. (a, c) ∈ B2 and a.e. b ∈ I(a, c) 4.6

I(a, c) ⊂ I(a, b) ∪ I(b, c) . •

In particular, for a.e. (a, c) ∈ 4.7

B2

and a.e. b ∈ I(a, c)

I(a, c) ≡ I(a, b) ∪ I(b, c) .

Proof. Applying the cocycle identity to a, x, b, c ∈ B we have 4.8

ω(x, b, c) − ω(a, b, c) + ω(a, x, c) − ω(a, x, b) = 0 .

If b ∈ I(a, c) and x ∈ I(a, b), we have ω(a, b, c) = 1, ω(a, x, b) = 1, which with (4.8) implies ω(x, b, c) + ω(a, x, c) = 2 and hence x ∈ I(a, c). If x ∈ I(b, c) then (4.8) gives ω(a, x, c) − ω(a, x, b) = 2 and hence x ∈ I(a, c). This shows the inclusion (4.4). Concerning (4.5), we apply (4.8) to x ∈ I(a, b) ∩ I(b, c) to get a contradiction. Concerning (4.6), assume that b ∈ I(a, c) and x ∈ I(a, c) and apply (4.8) to get 4.9

ω(x, b, c) = ω(a, x, b).

For a.e. (a, b) ∈ B2 and a.e. x ∈ B we have either ω(a, x, b) = 1 and hence x ∈ I(a, b) or ω(a, x, b) = −1, which with (4.9) gives x ∈ I(b, c). Lemma 4.6 (4) and Lemma 4.7 imply then immediately

lemma 4.8. For a.e. (a, x, y) ∈ B3 we have the following dichotomy: 1) x ∈ I(a, y), I(a, y) ≡ I(a, x) ∪ I(x, y) and νB (I(a, x)) < νB ((a, y)). 2) y ∈ I(a, x), I(a, x) ≡ I(a, y) ∪ I(y, x) and νB (I(a, y)) < νB ((a, x)).

24 / m. burger

Given a ∈ B, we deﬁne fa : B → Z\R by fa (x) := νB (I(a, x)) mod Z . Then we have

lemma 4.9. For a.e. a ∈ B we have 1) E = {(x, y) ∈ B2 : fa (x) = fa (y)} is of measure zero. 2) (fa )∗ (νB ) has no atoms. Proof. 1) If (x, y) ∈ E then either νB (I(a, x)) = 0 and νB (I(a, y)) = 1, or νB (I(a, x)) = νB ((a, y)); in both cases these equalities hold only for a set of (a, x, y)’s of measure zero; in the ﬁrst case this follows from Lemma 4.6 (4) and in the second from Lemma 4.8. 2) follows from (1).

lemma 4.10. For a.e. a ∈ B and a.e. (x, y, z) ∈ B3 o( fa (x), fa (y), fa (z)) = ω(x, y, z) . Proof. Observe that the left-hand side gives a measurable alternating cocycle taking values in {−1, 0, 1}; it follows from Lemma 4.9 that it takes values in {−1, 1} almost everywhere. It therefore sufﬁces to show that if the left-hand side equals 1, then so does the right-hand side. If o( fa (x), fa (y), fa (z)) = 1 then up to cyclically permuting the variables we may assume, using the deﬁnition of fa , that νB (I(a, x)) < νB (I(a, y)) < νB (I(a, z)). Together with Lemma 4.8 this implies x ∈ I(a, y) and y ∈ I(a, z) except possibly for a set of (a, x, y, z)’s of measure zero. Using the cocycle identity for ω applied to (a, x, y, z) we get ω(x, y, z) + ω(a, x, z) = 2 and hence ω(x, y, z) = 1. The following lemma will be left to the reader:

lemma 4.11. Let ξ ∈ M1 (S1 ) be a probability measure without atoms and hξ : S1 → S1 x −→ ξ ([˙o, x)) mod Z the associated quasiconjugacy. Then, 1) hξ is continuous. 2) the measure (hξ )∗ (ξ ) has no atoms and its support equals S1 .

an extension criterion for lattice actions on the circle / 25

3) for ξ 3 -almost every (x, y, z) ∈ (S1 )3 we have o(hξ (x), hξ (y), hξ (z)) = o(x, y, z). Combining Lemma 4.9, 4.10, and 4.11 we obtain the following lemma, which completes Step 1:

lemma 4.12. For a.e. a ∈ B, let ξ = (fa )∗ (νB ) and deﬁne ϕa := hξ ◦ fa . Then 1) (ϕa )∗ (νB ) has no atoms. 2) Ess Im ϕa = S1 . 3) o(ϕa (x), ϕa (y), ϕa (z)) = ω(x, y, z) for a.e. (x, y, z) ∈ B3 . 4.4. Step 2: The Construction of the Action

The following result is the basis for our construction:

lemma 4.13. Let ϕ, ψ : B → S1 be measurable maps such that 1) ϕ∗ (νB ) and ψ∗ (νB ) have no atoms. 2) Ess Im ϕ = Ess Im ψ = S1 . 3) o(ϕ(x), ϕ(y), ϕ(z)) = o(ψ(x), ψ(y), ψ(z)) for a.e. (x, y, z) ∈ B3 . Then the essential image F ⊂ S1 × S1 of the map B → S1 × S1 x −→ (ϕ(x), ψ(x)) is the graph of an orientation-preserving homeomorphism h : S1 → S1 and, h(ϕ(x)) = ψ(x) for a.e. x ∈ B . Proof. The proof is in two steps.

claim 1: Given (ξ1 , η1 ), (ξ2 , η2 ), (ξ3 , η3 ) in F such that ξ1 , ξ2 , ξ3 are pairwise distinct and the same holds for η1 , η2 , η3 , we have o(ξ1 , ξ2 , ξ3 ) = o(η1 , η2 , η3 ) . Pick Vi ξi and Wi ηi open intervals such that V1 , V2 , V3 are pairwise disjoint and the same holds for W1 , W2 , W3 . Then Bi = {x ∈ B : ϕ(x) ∈ Vi and ψ(x) ∈ Wi } is of positive measure, i = 1, 2, 3.

26 / m. burger

We then have for a.e. (x, y, z) ∈ B1 × B2 × B3 , o(ξ1 , ξ2 , ξ3 ) = o(ϕ(x1 ), ϕ(x2 ), ϕ(x3 )) = o(ψ(x1 ), ψ(x2 ), ψ(x3 )) = o(η1 , η2 , η3 ) .

claim 2: Let (ξ1 , η1 ), (ξ2 , η2 ) be in F. Then ξ1 = ξ2 iff η1 = η2 . Assume that ξ1 = ξ2 . We will repeatedly use the fact that (ϕ(x), ψ(x)) ∈ F for a.e. x ∈ B .

4.10

Since Ess Im ϕ = S1 and ξ1 = ξ2 the set ϕ −1 ((ξ1 , ξ2 )) is of positive measure; then in virtue of (4.10) and the hypothesis that ψ∗ (νB ) has no atoms, we can pick t ∈ ϕ −1 ((ξ1 , ξ2 )) with (ϕ(t), ψ(t)) ∈ F and ψ(t) ∈ / {η1 , η2 }. Similarly we can ﬁnd s ∈ ϕ −1 ((ξ2 , ξ1 )) such that (ϕ(s), ψ(s)) ∈ F and ψ(s) ∈ / {ψ(t), η1 , η2 }. Then the cocycle identity applied to (η1 , ψ(t), η2 , ψ(s)} gives o(ψ(t), η2 , ψ(s)) − o(η1 , η2 , ψ(s)) + o(η1 , ψ(t), ψ(s)) − o(η1 , ψ(t), η2 ) = 0 . Applying Claim 1 to the ﬁrst and third terms we get o(ϕ(t), ξ2 , ϕ(s)) − o(η1 , η2 , ψ(s)) + o(ξ1 , ϕ(t), ϕ(s)) − o(η1 , ψ(t), η2 ) = 0 and by our choices we have o(ϕ(t), ξ2 , ϕ(s)) = 1 o(ξ1 , ϕ(t), ϕ(s)) = 1, which implies that o(η1 , η2 , ψ(s)) + o(η1 , ψ(t), η2 ) = 2 and hence η1 = η2 . This shows Claim 2 modulo interchanging the roles of ϕ and ψ. Since pri (F) = S1 we get from Claim 2 that F is the graph of a homeomorphism h, which by Claim 1 is orientation preserving. Finally (4.10) says precisely that h(ϕ(x)) = ψ(x) for a.e. x ∈ B and this completes the proof of the lemma. Now we return to our cocycle ω : B3 → R satisfying the hypothesis of Theorem 4.5 and let ϕ : B → S1 be a measurable map as given by Lemma 4.12. Since ω is G-invariant we can apply Lemma 4.13 to ϕ and ϕg, g ∈ G to obtain an orientation-preserving homeomorphism denoted πϕ (g) ∈ Homeo+ (S1 ) satisfying πϕ (g)(ϕ(x)) = ϕ(gx) for a.e. x ∈ B .

an extension criterion for lattice actions on the circle / 27

This equivariance property implies that πϕ : G → Homeo+ (S1 ) is a homomorphism.

lemma 4.14. 1) πϕ : G → Homeo+ (S1 ) is continuous. It is minimal, unbounded, and strongly proximal. 2) Let ϕ, ψ be as in Lemma 4.13, and πϕ , πψ the corresponding homomorphisms. Let h : S1 → S1 be the homeomorphism given by Lemma 4.13. Then, h πϕ (g) = πψ (g) h

∀g ∈ G.

Proof. Assertion 2) is clear and follows from the various equivariance properties and thus we concentrate on 1). For a.e. x ∈ B, the map g −→ πϕ (g)(ϕ(x)) = ϕ(gx) is measurable and hence the homomorphism πϕ : G → Homeo+ (S1 ) is measurable. Since G is locally compact second countable and since Homeo+ (S1 ) is second countable we deduce that πϕ is continuous. Using Proposition 4.1 we see that πϕ∗ (ebR ) = 0 and hence πϕ is nonelementary. Assume that there is an exceptional minimal set K ⊂ S1 . Let J be a connected component if S1 \K ; since Ess Im ϕ = S1 , +

ϕ −1 ( J) is of positive measure but not of full measure in B. Since ∀g ∈ B we have either πϕ (g) J = J or πϕ (g) J ∩ J = φ, we conclude that either gϕ −1 ( J) ≡ ϕ −1 ( J) or gϕ −1 ( J) ∩ ϕ −1 ( J) = φ. Since the G-action on B is ergodic we have −1 −1 −1 −1 1 g∈G gϕ ( J) = ϕ (S \K ) ≡ B and hence there is g0 ∈ B with g0 ϕ ( J) ∩ −1 ϕ ( J) = φ. Thus h∈G h(ϕ −1 ( J) × ϕ −1 ( J)) is of positive measure in B2 , G-invariant, and does not meet g0 ϕ −1 ( J) × ϕ −1 ( J), itself also of positive measure, this contradicts the ergodicity of the G-action on B × B. Thus πϕ is minimal unbounded. Finally it follows from Corollary 4.4 that πϕ is strongly proximal.

5. Proofs of Theorems 1.2 and 1.3, Remark 1.4(2), and Corollaries 1.5 and 1.6 In order to apply the results obtained so far we recall

theorem 5.1. ([21]) Let G be a locally compact second countable group and μ ∈ M1 (G) a symmetric spread-out probability measure on G. Then the G-action on the associated Poisson

28 / m. burger

boundary (B, νB ) is amenable, doubly ergodic; the same properties hold for the action of a lattice < G. Using Theorem 5.1 for = G and applying Corollary 4.4, Theorem 4.5, and Proposition 4.1 implies readily Theorem 1.3. Concerning Remark 1.4 (2), if ρ1 , . . . , ρn are minimal strongly proximal actions and ωi : B3 → {−1, 1}

are the associated cocycle, then for any ni ∈ Z, ni=1 ni ωi takes integral values and hence by the isomorphism in (4.1) and Proposition 4.1 the norm of

n ∗ b i=1 ni ρi (eR ) is half integral. Corollary 1.6 (1) follows from Proposition 3.2, Proposition 4.1, and Lemma 4.3. Corollary 1.6 (2) follows from Theorem 1.3 (2), which implies that ρ1 , ρ2 have modulo conjugation the same strongly proximal quotient. We turn now to the proof of Theorem 1.2. First assume that for the given minimal unbounded action ρ, ρsp extends continuously. Let k = |ZH+ (ρ())|. Then we have that (ρsp )∗ (ebR ) = k · ρ ∗ (ebR ) (Lemma 4.3) and since (ρsp )∗ (ebR ) is in the image of the restriction map, so is ρ ∗ (ebR ). Conversely, apply Theorem 5.1 and let μ ∈ M1 () be a probability measure of full support such that νB is μ -stationary (see [22]). On the space (B, νB ) the restriction map Hb2 (G, R) → Hb2 (, R) is realized by the inclusion 5.1

∞ 3 ∞ 3 Z Lalt (B , R)G → Z Lalt (B , R) .

Now if ρ ∗ (ebR ) is in the image of the restriction map then so is (ρsp )∗ (ebR ). Let ωρsp : B3 → R be the cocycle associated to ρsp ; using (5.1) we get that ωρsp is G-invariant, satisﬁes all hypotheses of Theorem 4.5 and hence corresponds to a continuous minimal strongly proximal homomorphism π : G → Homeo+ (S1 ). Since ωπ | clearly coincides with ωρsp we conclude from Theorem 4.5 that π | and ρsp are conjugate.

6. Locally Compact Groups Acting on S1 In this section we prove Theorem 1.8 using some results from [16] and [10]. In the sequel, Rot denotes as usual the group of rotations of S1 and, for every k ≥ 1, PSL(2, R)k ⊂ Homeo+ (S1 ) is the k-fold cyclic covering group of PSL(2, R). We record the following

an extension criterion for lattice actions on the circle / 29

lemma 6.1. Let H be connected, semisimple, with ﬁnite center and no compact factors, and π : H → Homeo+ (S1 )

6.1

a continuous nontrivial homomorphism. Then there exists k ∈ N such that up to conjugation, π (H) = PSL(2, R)k . For the following lemma we recall that given a subgroup L < Homeo+ (S1 ), NH+ (L) and ZH+ (L) denote respectively its normalizer and centralizer in Homeo+ (S1 ).

lemma 6.2. 1) NH+ (Rot) = ZH+ (Rot) = Rot. 2) NH+ (PSL(2, R)k ) = PSL(2, R)k . 3) ZH+ (PSL(2, R)k ) = Z (PSL(2, R)k ), which is cyclic of order k. Proof. 1) Let g ∈ NH+ (Rot); if g r g −1 = r −1 ∀r ∈ Rot, then g cannot preserve orientation; hence g ∈ ZH+ (Rot). Composing g with a rotation we may assume g(0) = 0, which on applying r ∈ Rot yields g(r(0)) = r(0) and thus g = e. 2) For g ∈ NH+ (PSL(2, R)k ) and Kk < PSL(2, R)k maximal compact subgroup, g −1 Kk g is maximal compact as well and hence there is h ∈ PSL(2, R)k with g Kk g −1 = h Kk h −1 that is h −1 g ∈ NH+ (Kk ); since Kk is conjugate to Rot we deduce from 1) that h −1 g ∈ Kk . 3) follows from (2). Finally we record the following consequence of Hilbert’s ﬁfth problem:

theorem 6.3. ([8] thm. 3.3.3) Let G be a locally compact group and A(G) G its amenable radical. Let Ga := G/A(G). Then 1) (Ga )0 is connected, semisimple, with trivial center and no compact factors. 2) The centralizer L of (Ga )0 in Ga is totally disconnected; we have L ∩ (Ga )0 = (e) and the product L · (Ga )0 is open of ﬁnite index in Ga . Proof of Theorem 1.8. Let π : G → Homeo+ (S1 ) be a continuous minimal action. We consider π |A(G) and apply our trichotomy:

30 / m. burger

Case 1: π(A(G)) has a ﬁnite orbit; then the set of ﬁnite π (A(G))-orbits with ﬁxed cardinality is nonempty closed and π(G)-invariant, hence coincides with S1 . As a result, π (A(G)) is ﬁnite cyclic. Case 2: An exceptional minimal set for π (A(G)) being unique would also be π (G)-invariant contradicting minimality of π(G). Case 3: π(A(G)) is minimal; since A(G) is amenable (π |A(G) )∗ (ebR ) = 0 and hence, modulo conjugating π , we may assume that π(A(G)) is a dense subgroup of Rot. In particular, π (G) ⊂ NH+ (Rot) = Rot by Lemma 6.2. This gives the ﬁrst alternative in Theorem 1.8. Now we return to Case 1 and consider Gπ := G/(A(G) ∩ Ker π ), which is a ﬁnite extension q : Gπ → Ga of Ga . Observe that π factors through Ga . Let Lπ := q−1 (L); then (Gπ )0 is connected, semisimple with ﬁnite center and no compact factors, Lπ is totally disconnected, and Lπ · (Gπ )0 is open of ﬁnite index in Gπ . We then have two cases: 1) π |(Gπ )0 is trivial. Since Homeo+ (S1 ) does not contain small subgroups, π sends some compact open subgroup of Lπ to the identity, which implies that Ker π is open. 2) π|(Gπ )0 is nontrivial and hence by Lemma 6.1, π ((Gπ )0 ) = PSL(2, R)k ; since π (G) normalizes π ((Gπ )0 ) we get π(G) = PSL(2, R)k by Lemma 6.2 (2).

References [1] [2] [3]

[4] [5] [6] [7]

U. Bader, A. Furman: e-mail exchange, private communication, September 9–October 6, 2006. U. Bader, A. Furman, A. Shaker: Superrigidity, Weyl groups and actions on the circle. ArXiv: math/0605276v4. M. Burger, A. Iozzi: Boundary maps in bounded cohomology. Appendix to: “Continuous bounded cohomology and applications to rigidity theory.” Geom. Funct. Anal., 12(2):281–292 (2002). M. Burger, A. Iozzi: Bounded Kähler class rigidity of actions on Hermitian symmetric spaces. Ann. Sci. Ecole Norm. Sup. (4), 37(1):77–103 (2004). M. Burger, A. Iozzi, A. Wienhard: Surface group representations with maximal Toledo invariant. Annals of Math., 172(1):517–566 (2010). M. Burger, A. Iozzi, A. Wienhard: Hermitian symmetric spaces and Kähler rigidity. Transform. Groups, 12(1):5–32 (2007). M. Burger, A. Iozzi, A. Wienhard: Tight homomorphisms and Hermitian symmetric spaces. Geom. Funct. Anal., 19(3):678–721 (2009).

an extension criterion for lattice actions on the circle / 31 [8] [9] [10] [11] [12] [13] [14]

[15] [16] [17]

[18]

[19]

[20] [21] [22] [23] [24] [25] [26] [27]

M. Burger, N. Monod: Continuous bounded cohomology and applications to rigidity theory. Geom. Funct. Anal., 12(2):219–280 (2002). B. Deroin, V. Kleptsyn, A. Navas: Sur la dynamique unidimensionnelle en régularité intermédiaire. Acta Math., 199(2):199–262 (2007). A. Furman: Mostow-Margulis rigidity with locally compact targets. Geom. Funct. Anal., 11(1):30–59 (2001). A. Furman: Random walks on groups and random transformations. Handbook of dynamical systems, Vol. 1A, 931–1014, North-Holland, Amsterdam, 2002. H. Furstenberg: A Poisson formula for semi-simple Lie groups. Ann. of Math. (2), 77:335–386 (1963). H. Furstenberg: Noncommuting random products. Trans. Amer. Math. Soc., 108:377– 428 (1963). E. Ghys: Groupes d’homéomorphismes du cercle et cohomologie bornée. The Lefschetz centennial conference, Part III (Mexico City, 1984), 81–106, Contemp. Math., 58, III, American Mathematical Society, Providence, RI, 1987. E. Ghys: Actions de réseaux sur le cercle. Invent. Math., 137(1):199–231 (1999). E. Ghys: Groups acting on the circle. Enseign. Math. (2), 47(3–4):329–407 (2001). Y. Guivarc’h, A. Raugi: Products of random matrices: convergence theorems. Random matrices and their applications (Brunswick, ME, 1984), 31–54, Contemp. Math., 50, American Mathematical Society, Providence, RI, 1986. G. Hector, U. Hirsch: Introduction to the geometry of foliations. Part B. Foliations of codimension one. Aspects of Mathematics, E3. Friedr. Vieweg & Sohn, Braunschweig, 1983. A. Iozzi: Bounded cohomology, boundary maps, and rigidity of representations into Homeo+ (S1 ) and SU(1, n). Rigidity in dynamics and geometry (Cambridge, 2000), 237– 260, Springer, Berlin, 2002. A. Iozzi: Vanishing theorems in bounded cohomology, in preparation. V. Kaimanovich: Double ergodicity of the Poisson boundary and applications to bounded cohomology. Geom. Funct. Anal., 13(4):852–861 (2003). G. Margulis: Discrete subgroups of semisimple Lie groups. Ergebnisse der Mathematik und ihrer Grenzgebiete (3), Springer-Verlag, Berlin, 1991. G. Margulis: Free subgroups of the homeomorphism group of the circle. C. R. Acad. Sci. Paris Sr. I Math., 331(9): 669–674 (2000). N. Monod: Continuous bounded cohomology of locally compact groups. Lecture Notes in Mathematics, 1758. Springer-Verlag, Berlin, 2001. D. Witte Morris: Can lattices in SL(n, R) act on the circle? This volume. A. Navas: Quelques nouveaux phénomènes de rang 1 pour les groupes de difféomorphismes du cercle. Comment. Math. Helv., 80(2): 355–375 (2005). D. Witte, R. Zimmer: Actions of semisimple Lie groups on circle bundles. Geom. Dedicata, 87(1–3): 91–121 (2001).

2

sorin dumitrescu

MEROMORPHIC ALMOST RIGID GEOMETRIC STRUCTURES dedicated to bob zimmer Abstract We study the local Killing Lie algebra of meromorphic almost rigid geometric structures on complex manifolds. This leads to classiﬁcation results for compact complex manifolds admitting holomorphic rigid geometric structures.

1. Introduction Zimmer and Gromov have conjectured that “big” actions of Lie groups that preserve unimodular rigid geometric structures (for example, a pseudoRiemannian metric or an afﬁne connection together with a volume form) are “essentially classiﬁable” [1, 28, 64, 65, 66, 67]. We try to address here this general question in the framework of the complex geometry. We study holomorphic (and meromorphic) unimodular geometric structures on compact complex manifolds, and we show that in many cases the holomorphic rigidity implies that the situation is classiﬁable (even without assuming any isometric group action). This is a survey paper, but some results are new (see Theorem 2.1, Corollary 2.2, and Theorem 5.10). In the sequel all complex manifolds are supposed to be smooth and connected. Consider a complex n-manifold M and, for all integers r ≥ 1, consider the associated bundle Rr (M) of r-frames, which is a Dr (Cn )-principal bundle over M, with Dr (Cn )-the (algebraic) group of r-jets at the origin of local biholomorphisms of Cn ﬁxing 0 [1, 28, 7, 12, 22, 63]. Let us consider, as in [1, 28], the following:

definition 1.1. A meromorphic geometric structure (of order r) φ on a complex manifold M is a Dr (Cn )-equivariant meromorphic map from Rr (M) to a 1991 Mathematics Subject Classiﬁcation 53B21, 53C56, 53A55. Keywords and phrases: complex manifolds, rigid geometric structures, transitive Killing Lie algebras This work was partially supported by the ANR Grant Symplexe BLAN 06-3-137237.

32

meromorphic almost rigid geometric structures / 33

quasiprojective variety Z endowed with an algebraic action of Dr (Cn ). If Z is an afﬁne variety, we say that φ is of afﬁne type. In holomorphic local coordinates U on M, the geometric structure φ is given by a meromorphic map φU : U → Z. Denote by P ⊂ M the analytic subset of poles of φ and by M ∗ = M \ P the open dense subset of M where φ is holomorphic. If P is the empty set, then φ is a holomorphic geometric structure on M. If φ uniquely determines a holomorphic volume form on M ∗ , then φ is called unimodular.

definition 1.2. The meromorphic geometric structure φ is called almost rigid if, away from an analytic subset of M of positive codimension, φ is (holomorphic and) rigid in Gromov’s sense [1, 28]. A holomorphic geometric structure on M is called rigid if it is rigid in Gromov’s sense on all of M. Notice that my deﬁniton of almost rigidity is not exactly the same as in [9].

examples On a complex connected n-manifold M the following meromorphic geometric structures are almost rigid. • A meromorphic map φ from M to a projective space P N (C), which is an

embedding on some nontrivial open subset of M. In this case M is of algebraic dimension n (see Section 2). If φ is a holomorphic embedding, then φ is rigid. • A family X1 , X2 , . . . , Xn of meromorphic vector ﬁelds such that there exists an open (dense) subset of M where the Xi span the holomorphic tangent bundle TM. Any homogeneous manifold admits a holomorphic family of such vector ﬁelds. They pull-back after the blowup of a point (or a submanifold) in the homogeneous manifold in a meromorphic family of such vector ﬁelds. In the special case where X1 , X2 , . . . , Xn are holomorphic and span TM on all of M, the corresponding geometric structure is called a holomorphic parallelization of the tangent bundle and it is a (unimodular) holomorphic rigid geometric structure of afﬁne type. • A meromorphic section g of the bundle S2 (T ∗ M) of complex quadratic forms such that g is nondegenerate on an open (dense) subset. If g is holomorphic and nondegenerate on M, then g is a holomorphic rigid geometric structure of afﬁne type called holomorphic Riemannian

34 / s. dumitrescu

metric. Up to a double cover of M, a holomorphic Riemannian metric is unimodular. A holomorphic Riemannian metric has nothing to do with the more usual Hermitian metric. It is in fact nothing but the complex version of a (pseudo-)Riemannian metric. Observe that since complex quadratic forms have no signature, there is no distinction here between the Riemannian and pseudo-Riemannian cases. This observation was at the origin of the niceuse by F. Gauß of the complexiﬁcation technic of (analytic) Riemannian metrics on surfaces, in order to ﬁnd conformal coordinates for them. Actually, the complexiﬁcation of analytic Riemannian metrics leading to holomorphic ones is becoming a standard trick (see for instance [23]). In the blow-up process, the pull-back of a holomorphic Riemannian metric will stay nondegenerate away from the exceptional set and will vanish on the exceptional set. More general, if g is a meromorpic section of S2 (T ∗ M) nondegenerate ˜ bimeromorphic to on an open dense set, then any complex manifold M M inherits a similar section g˜ . Moreover, if g is holomorphic, then g˜ is also holomorphic. This comes from the fact that indeterminacy points of a meromorphic map are of codimension at least 2 (see theorem 2.5 in [57]) and from Levi’s extension principle (see [57] for the details). • Meromorphic afﬁne or projective connections or meromorphic conformal structures in dimension ≥ 3. For example, in local holomorphic coordinates (z1 , . . . , zn ) on the manifold M, a meromorphic afﬁne connection ∇ is determined by the meromorphic functions ijk such that ∇

∂ ∂zi

∂ ∂zj

= ijk ∂z∂ , for all i, j, k ∈ {1, . . . , n}. k

Afﬁne connections are geometric structures of afﬁne type, but projective connections or conformal structures are not of afﬁne type. • If φ is a meromorphic almost rigid geometric structure of order r1 and g is a meromorphic geometric structure of order r2 , then we can put together φ and g in some meromorphic almost rigid geometric structure of order max(r1 , r2 ), denoted by (φ, g) [28, 1, 7, 12, 22, 63]. • The s-jet (prolongation) φ (s) of a meromorphic almost rigid geometric structure φ is still a meromorphic almost rigid geometric structure [28, 1, 7, 12, 22, 63]. Recall that local biholomorphisms of M that preserve a meromorphic geometric structure φ are called local isometries. Note that local isometries of (φ, g) are the local isometries of φ, which also preserve g.

meromorphic almost rigid geometric structures / 35

The set of local isometries of a holomorphic rigid geometric structure φ is a Lie pseudogroup Isloc (φ) generated by a Lie algebra of local vector ﬁelds called (local) Killing Lie algebra. If the local Killing Lie algebra is transitive on M, then φ is called locally homogeneous.

2. Local Isometries and Meromorphic Functions The maximal number of algebraically independent meromorphic functions on a complex manifold M is called the algebraic dimension a(M) of M. Recall that a theorem of Siegel proves that a complex n-manifold M admits at most n algebraically independent meromorphic functions [57]. Then a(M) ∈ {0, 1, . . . , n} and for algebraic manifolds a(M) = n. We will say that two points in M are in the same ﬁber of the algebraic reduction of M if any meromorphic function on M takes the same value at the two points. There exists some open dense set in M where the ﬁbers of the algebraic reduction are the ﬁbers of a holomorphic ﬁbration on an algebraic manifold of dimension a(M) and any meromorphic function on M is the pull-back of a meromorphic function on the basis [57]. The following theorem shows that the ﬁbers of the algebraic reduction are in the same orbit of the pseudogroup of local isometries for any meromorphic almost rigid geometric structure on M. This is a meromorphic version of the celebrated Gromov’s open dense orbit theorem [1, 28] (see also [7, 12, 22, 63]).

theorem 2.1. Let M be a connected complex manifold of dimension n, which admits a meromorphic almost rigid geometric structure φ. Then, there exists a nowhere-dense analytic subset S in M, such that M \ S is Isloc (φ)-invariant and the orbits of Isloc (φ) in M \ S are the ﬁbers of a holomorphic ﬁbration of constant rank. The dimension of the ﬁbers is ≥ n − a(M), where a(M) is the algebraic dimension of M. Recall that g = (φ, g) is still almost rigid for any meromorphic (not necessarily almost rigid) geometric structure g on M. This yields to the following:

corollary 2.2. If g is a meromorphic geometric structure on M, then, away from a nowhere dense analytic subset, the Isloc (g)-orbits are of dimension ≥ n − a(M). In particular, if a(M) = 0, then g is locally homogeneous outside a nowhere dense analytic subset of M (the Killing Lie algebra G of g = (φ, g) is transitive on

36 / s. dumitrescu

an open dense set). Moreover, if φ is unimodular and G is unimodular and simply transitive, then g is locally homogeneous on all of the open dense set where φ and g are holomorphic and φ is rigid. In this case, if φ is unimodular holomorphic and rigid and g is holomorphic, then g is locally homogeneous on M. Proof. Let φ be of order r, given by a map φ : Rr (M) → Z. For each positive integer s we consider the s-jet φ (s) of φ. This is a D(r+s) (Cn )-equivariant meromorphic map R(r+s) (M) → Z (s) , where Z (s) is the algebraic variety of the s-jets at the origin of holomorphic maps from Cn to Z. One can ﬁnd the expression of the (algebraic) D(r+s) (Cn )-action on Z (s) in [12, 22]. Since φ is almost rigid, there exists a nowhere dense analytic subset S in M, containing the poles of φ, and a positive integer s such that two points m, m in M\S are in the same orbit of Isloc (φ) if and only if φ (s) sends the ﬁbers of R(r+s) (M) above m and m on the same D(r+s) (Cn )-orbit in Z (s) [28, 1]. Rosenlicht’s theorem (see [49]) shows that there exists a D(r+s) (Cn )-invariant stratiﬁcation Z (s) = Z0 ⊃ · · · ⊃ Zl , such that Zi+1 is Zariski closed in Zi , the quotient of Zi \ Zi+1 by D(r+s) (Cn ) is a complex manifold, and rational D(r+s) (Cn )-invariant functions on Zi separate orbits in Zi \ Zi+1 . Consider the open dense Isloc (φ)-invariant subset U in M \ S , such that (s) φ is of constant rank above U and the image of R(r+s) (M)|U through φ (s) lies in Zi \ Zi+1 but not in Zi+1 . Then the orbits of Isloc (φ) in U are the ﬁbers of a ﬁbration of constant rank (on the quotient of Zi \ Zi+1 by D(r+s) (Cn )). Obviously, U = M \ S, where S is a nowhere dense analytic subset in M. If m and n are two points in U that are not in the same Isloc (φ)-orbit, then the corresponding ﬁbers of R(r+s) (M)|U are sent by φ (s) on two distinct D(r+s) (Cn )orbits in Zi \ Zi+1 . By Rosenlicht’s theorem there exists a D(r+s) (Cn )-invariant rational function F : Zi \ Zi+1 → C, which takes distinct values at these two orbits. The meromorphic function F ◦ φ (s) : R(r+s) (M) → C is D(r+s) (Cn )-invariant and descends in a Isloc (φ)-invariant meromorphic function on M that takes distinct values at m and at n. Consequently, the complex codimension in U of the Isloc (φ)-orbits is ≤ a(M), which ﬁnishes the proof.

meromorphic almost rigid geometric structures / 37

We now deduce the corollary: Proof. It is convenient to put together φ and g in some extra geometric structure g = (φ, g). Now g is a meromorphic almost rigid geometric structure and Theorem 2.1 shows that the complex dimension of a generic Isloc (g )-orbit is ≥ n − a(M). As each Isloc (g )-orbit is contained in a Isloc (g)-orbit, the result follows also for g. If a(M) = 0, the Killing Lie algebra G of g is transitive on a maximal open dense subset U in M. Suppose now that φ is unimodular and also that G is unimodular. The other inclusion being trivial, it is enough to show that U contains the maximal subset of M where φ and g are holomorphic and φ is rigid. Pick up a point m in the previous subset. We want to show that m is in U. Since g = (φ, g) is holomorphic and rigid in the neighborhood of m, it follows that m admits an open neighborhood Um in M such that any local holomorphic Killing ﬁeld of g deﬁned on a connected open subset in Um extends on all of Um [48, 2]. Since G acts transitively on U, choose local linearly independent Killing ﬁelds X1 , . . . , Xn on a connected open set included in U ∩ Um . As φ is unimodular, it determines a holomorphic volume form vol on Um (if necessary we restrict to a smaller Um ), which is preserved by Isloc (g ). But Isloc (g ) acts transitively on U ∩ Um and G is supposed to be unimodular. This implies that the function vol(X1 , . . . , Xn ) is G -invariant and, consequently, a (nonzero) constant on U ∩ Um . On the other hand, X1 , . . . , Xn extend in some holomorphic Killing ﬁelds X˜ 1 , . . . , X˜ n deﬁned on all of Um . The holomorphic function vol(X˜ 1 , . . . , X˜ n ) is a nonzero constant on Um : in particular, X˜ 1 (m), . . . , X˜ n (m) are linearly independent. We proved that G acts transitively in the neighborhood of m and thus m ∈ U.

3. Classification Results Parallelizations of the Tangent Bundle

We begin with the classiﬁcation of compact complex manifolds that admit a holomorphic parallelization of the tangent bundle [60].

theorem 3.1. (wang) Let M be a compact connected complex manifold with a holomorphic parallelization of the tangent bundle. Then M is a quotient \G, where G is a connected simply connected complex Lie group and is a uniform lattice in G. Moreover, M is Kaehler if and only if G is abelian (and M is a complex torus).

38 / s. dumitrescu

Proof. Let n be the complex dimension of M and consider X1 , X2 , . . . , Xn global holomorphic vector ﬁelds on M that span TM. Then, for all 1 ≤ i, j ≤ n, we have ij

ij

ij

[Xi , Xj ] = f1 X1 + f2 X2 + · · · + fn Xn , ij

with fk holomorphic functions on M. Since M is compact, these functions have to be constant and, consequently, X1 , X2 , . . . , Xn generate a n-dimensional Lie algebra G that acts simply transitively on M. By Lie’s theorem there exists a unique connected simply connected complex Lie group G corresponding to G . In particular, the holomorphic parallelization is locally homogeneous, locally modeled on the parallelization given by translation-invariant vector ﬁelds on the Lie group G. Since M is compact, the Xi are complete and they deﬁne a holomorphic simply transitive action of G on M. Hence M is a quotient of G by a cocompact discrete subgroup in G. Assume now M is Kaehler. Then, any holomorphic form on M has to be closed [27]. Consider ω1 , ω2 , . . . , ωn the dual basis with respect to X1 , X2 , . . . , Xn . The one forms ωi are holomorphic and translation-invariant on G (they descend on \G). The Lie-Cartan formula dωi (Xj , Xk ) = −ωi ([Xj , Xk ]), shows that the one forms ωi are all closed if and only if G is abelian and thus M is a complex torus. Holomorphic Riemannian Metrics

As in the real case, a holomorphic Riemannian metric on M gives rise to a covariant differential calculus, that is, a Levi-Civita (holomorphic) afﬁne connection, and to geometric features: curvature tensors and geodesic (complex) curves [40, 41]. Locally, a holomorphic Riemannian metric has the expression gij (z) dzi dzj , where (gij (z)) is a complex inversible symmetric matrix depending holomorphically on z. The standard example is that of the global ﬂat holomorphic Riemannian metric dz21 + dz22 + · · · + dz2n on Cn . This metric is translationinvariant and thus descends to any quotient of Cn by a lattice. Hence, complex tori possess (ﬂat) holomorphic Riemannian metrics. This is, however, a very special situation since, contrary to real case, only few compact complex manifolds admit holomorphic Riemannian metrics. In fact, Yau’s proof of the Calabi conjecture shows that, up to ﬁnite unramiﬁed covers, complex tori

meromorphic almost rigid geometric structures / 39

are the only compact Kaehler manifolds admitting holomorphic Riemannian metrics [34]. However, very interesting examples, constructed by Ghys in [25], do exist on 3-dimensional complex non-Kaehler manifolds and deserve classiﬁcation. Notice that parallelizable manifolds admit holomorphic Riemannian metrics coming from left-invariant holomorphic Riemannian metrics on G (which can be constructed by left translating any complex nondegenerate quadratic form deﬁned on the Lie algebra G ). Ghys’s examples of 3-dimensional compact complex manifolds endowed with holomorphic Riemannian metrics are obtained by deformation of the complex structure on parallelizable manifolds \SL(2, C) [25]. They are nonstandard, meaning they do not admit parallelizable manifolds as ﬁnite unramiﬁed covers. Those nonstandard examples will be described later on. A ﬁrst obstruction to the existence of a holomorphic Riemannian metric on a compact complex manifold is the vanishing of its ﬁrst Chern class. Indeed, a holomorphic Riemannian metric on M provides an isomorphism between TM and T ∗ M. In particular, the canonical bundle K is isomorphic to the anticanonical bundle K −1 and thus K 2 is trivial. This means that the ﬁrst Chern class of M vanishes and, up to a double unramiﬁed cover, M possesses a holomorphic volume form. The following proposition describes holomorphic Riemannian metrics on parallelizable manifolds:

proposition 3.2. Let M = \G a compact parallelizable manifold, with G a simply connected complex Lie group and a uniform lattice in G. Then, any holomorphic Riemannian metric g on M comes from a nondegenerate complex quadratic form on the Lie algebra G of G. In particular, the pull-back of g is left invariant on the universal cover G (and g is locally homogeneous on M). Moreover, any compact parallelizable 3-manifold admits a holomorphic Riemannian metric of constant sectional curvature. The metric is ﬂat exactly when G is solvable. Proof. Consider X1 , X2 , . . . , Xn the fundamental vector ﬁelds corresponding to the locally free G-action on M. Let g be a holomorphic Riemannian metric on M and denote also by g the associated complex symmetric bilinear form. Then g(Xi , Xj ) is a holomorphic function on M and thus constant, for all 1 ≤ i, j ≤ n. This implies that g comes from a left-invariant holomorphic Riemannian metric on G.

40 / s. dumitrescu

Assume now G is a connected simply connected complex unimodular Lie group of dimension 3. We have only four such Lie groups: C3 , the complex Heisenberg group, the complex SOL group, and SL(2, C) [35]. Note that the group SOL is the complexiﬁcation of the afﬁne isometry group of the Minkowski plane R1,1 or equivalently the isometry group of C2 endowed with its ﬂat holomorphic Riemannian metric. We begin with the case G = C3 . We have seen that C3 admits a ﬂat translation-invariant holomorphic Riemannian metric. The isometry group of this metric is O(3, C) C3 . We recall here that O(3, C) and SL(2, C) are locally isomorphic. Consider now the case G = SL(2, C). The Killing form of the Lie algebra sl(2, C) is a nondegenerate complex quadratic form that endows SL(2, C) with a left-invariant holomorphic Riemannian metric of constant sectional curvature. Since the Killing quadratic form is invariant by the adjoint representation, the isometry group also contains all right translations. In fact, the connected component of the isometry group is SL(2, C) × SL(2, C) acting by left and right translation. On the other hand, \SL(2, C) doesn’t admit ﬂat torsion-free afﬁne connections (or ﬂat holomorphic Riemannian metrics) [18]. It is an easy exercice to exhibit in the isometry group O(3, C) C3 of the ﬂat holomorphic Riemannian space copies of the complex Heisenberg group and of the complex SOL group, which act simply transitively [18]. Thus the ﬂat holomorphic Riemannian space also admits models that are given by leftinvariant metrics on the Heisenberg group and on the SOL group. One can get explicit expression of these holomorphic Riemannian metrics by complexiﬁcation of ﬂat left-invariant Lorentz metrics on the real Heisenberg and SOL groups [51, 52]. On the other hand, there exists no 3-dimensional solvable subgroups in SL(2, C) × SL(2, C) acting with an open orbit on SL(2, C). It follows that the compact quotients of 3-dimensional solvable groups don’t admit holomorphic Riemannian metrics of nonzero constant sectional curvature.

Ghys Nonstandard Examples

As above, for any cocompact lattice in SL(2, C), the quotient M = \SL(2, C) admits a holomorphic Riemannian metric of nonzero constant sectional curvature. It is convenient to consider M as a quotient of S3 = O(4, C)/O(3, C) = SL(2, C) × SL(2, C)/SL(2, C) by , seen as a subgroup of SL(2, C) × SL(2, C) by the trivial embedding γ ∈ → (γ , 1) ∈ SL(2, C) × SL(2, C).

meromorphic almost rigid geometric structures / 41

New interesting examples of manifolds admitting holomorphic Riemannian metrics of nonzero constant sectional curvature have been constructed in [25] by deformation of this embedding of . Those deformations are constructed choosing a morphism u : → SL(2, C) and considering the embedding γ → (γ , u(γ )). Algebraically, the action is given by: (γ , m) ∈ × SL(2, C) → γ mu(γ −1 ) ∈ SL(2, C). It is proved in [25] that, for u close enough to the trivial morphism, acts properly (and freely) on S3 ( ∼ = SL(2, C)) such that the quotient Mu is a complex compact manifold (covered by SL(2, C)) admitting a holomorphic Riemannian metric of non-zero constant sectional curvature. In general, these examples do not admit parallelizable manifolds as ﬁnite covers. Note that left-invariant holomorphic Riemannian metrics on SL(2, C) that are not right-invariant, in general, will not descend on Mu . Let us notice that despite this systematic study in [25], there are still many open questions regarding these examples (including the question of completeness). A real version of this study is in [39, 26, 53]. Dimension 3

The classiﬁcation of complex compact manifolds admitting holomorphic Riemannian metrics is simple in complex dimension 2 (see Section 5). An important step toward the classiﬁcation in dimension 3 was made in [17] with the following result:

theorem 3.3. Any holomorphic Riemannian metric on a compact connected complex 3-manifold is locally homogeneous. More generally, if a compact connected complex 3-manifold M admits a holomorphic Riemannian metric, then any holomorphic geometric structure of afﬁne type on M is locally homogeneous. Observe that the previous result is trivial in dimension 2, since the sectional curvature is a holomorphic function and thus constant on compact complex surfaces: this implies the local homogeneity [61]. In dimension 3, the sectional curvature will be, in general, a nonconstant meromorphic function on the 2Grassmanian of the holomorphic tangent space with poles at the degenerate planes. Thanks to Theorem 3.3, our manifold M is locally modeled on a (G, G/I)geometry in Thurston’s sense [55], where I is a closed subgroup of the Lie group G such that the G-action on G/I preserves some holomorphic

42 / s. dumitrescu

Riemannian metric (see [21] for details and notice that the local Killing Lie algebra of the holomorphic Riemannian metric is the Lie algebra of G). In this context we have a developing map from the universal cover of M into G/I, which is a local diffeomorphism and which is equivariant with respect to the ˜ by deck transformations and on G/I action of the fundamental group on M via the holonomy morphism ρ : π1 (M) → G [55]. Recall that the (G, G/I)-geometry is called complete if the developing map is a diffeomorphism and, consequently, = ρ(π1 (M)) acts properly on G/I such that M is a compact quotient \G/I. A second step was achieved in [21] where, in a common work with Zeghib, we proved the following result that can be seen, in particular, as a completeness result in the case where G is solvable.

theorem 3.4. Let M be a compact connected complex 3-manifold that admits a (locally homogeneous) holomorphic Riemannian metric g. Then i) If the Killing Lie algebra of g has a nontrivial semisimple part, then it preserves some holomorphic Riemannian metric on M with constant sectional curvature. ii) If the Killing Lie algebra of g is solvable, then, up to a ﬁnite unramiﬁed cover, M is a quotient either of the complex Heisenberg group, or of the complex SOL group by a lattice.

remark 3.5. If g is ﬂat, its Killing Lie algebra corresponds to O(3, C) C3, which has a nontrivial semisimple part. Thus, ﬂat holomorphic Riemannian metrics on complex tori are part of point (i) in Theorem 3.4. The point (ii) of the previous theorem is not only about completeness, but also gives a rigidity result in Bieberbach’s sense [61]: G contains a 3dimensional closed subgroup H (either isomorphic to the complex Heisenberg group, or to the complex SOL group), which acts simply transitively (and so identiﬁes) with G/I and (up to a ﬁnite index) the image of the holonomy morphism lies in H. It follows that, up to a ﬁnite cover, M identiﬁes with \H. Since the Heisenberg and SOL groups admit left-invariant ﬂat holomorphic Riemannian metrics this leads to:

corollary 3.6. If a compact connected complex 3-manifold M admits a holomorphic Riemannian metric, then, up to a ﬁnite unramiﬁed cover, M admits a holomorphic Riemannian metric of constant sectional curvature.

meromorphic almost rigid geometric structures / 43

Theorem 3.4 does not end the story, even in dimension 3, essentially because of remaining completeness questions, and those on the algebraic structure of the fundamental group. It remains to classify the compact complex 3-manifolds endowed with a holomorphic Riemannian metric of constant sectional curvature.

ﬂat case. In this case M admits a (O(3, C) C3 , C3 )-geometry. The challenge remains: 1) Markus conjecture: Is M complete? 2) Auslander conjecture: Assuming M is complete, is solvable? Note that these questions are settled in the setting of (real) ﬂat Lorentz manifolds [13, 24], but unsolved for general (real) pseudo-Riemannian metrics. The real part of the holomorphic Riemannian metric is a (real) pseudo-Riemannian metric of signature (3, 3) for which both previous conjectures are still open.

nonﬂat case. In this case G = SL(2, C) × SL(2, C) and I = SL(2, C) is diagonally embedded in the product. The completeness of this geometry on compact complex manifolds is still an open problem, despite a local result of Ghys [25]. Recall that the real analogous of this problem, that is, the completeness of compact manifolds endowed with Lorentz metrics of negative constant sectional curvature was solved in [37], but the proof cannot generalize to other signatures. Higher Dimension

One interesting problem in differential geometry is to decide whether a given homogeneous space G/I possesses a compact quotient. A more general related question is to decide whether there exists compact manifolds locally modeled on (G, G/I) (see, for instance, [6, 8, 38]). The case I = 1, or more generally I compact, reduces to the classical question of existence of cocompact lattices in Lie groups. For homogeneous spaces of non-Riemannian type (i.e., I noncompact) the problem is much harder. The case Sn = O(n + 1, C)/O(n, C) is a geometric situation where these questions can be tested. It turns out that compact quotients of Sn are known to exist only for n = 1, 3, or 7. We discussed the case n = 3 above, and the existence of a compact quotient of S7 was proved in [38]. Here, we dare ask with [38]:

conjecture 3.7. [38] Sn has no compact quotients, for n = 1, 3, 7.

44 / s. dumitrescu

A stronger version of this question was proved in [6] for Sn , if n has the form 4m + 1, with m ∈ N. Keeping in mind our geometric approach, we generalize the question to manifolds locally modeled on Sn . More exactly:

conjecture 3.8. [21] A compact complex manifold endowed with a holomorphic Riemannian metric of constant nonvanishing curvature is complete. In particular, such a manifold has dimension 3 or 7. 4. Applications to Simply Connected Manifolds Remark ﬁrst that Theorem 3.3 has the following direct consequence:

corollary 4.1. A compact connected simply connected 3-manifold admits no holomorphic Riemannian metrics. Proof. Assume, by contradiction, (M, g) being as in the statement of the corollary. Then Theorem 3.3 implies that g is locally homogeneous. Since M is simply connected, the local Killing ﬁelds of the Killing algebra G extend on all of M: the unique connected simply connected complex Lie group G associated to G acts isometrically and transitively on M. Then M is a homogeneous space G/H. Moreover, up to a double cover, G/H admits a holomorphic volume form vol coming from the holomorphic Riemannian metric. Take X1 , X2 , X3 three global Killing ﬁelds on M that are linearly independent at some point. Since vol(X1 , X2 , X3 ) is a holomorphic function on M, it is a nonzero constant and, consequently, X1 , X2 , X3 are linearly independent on M. Hence Wang’s theorem implies that M is a quotient of a 3 dimensional connected simply connected complex Lie group G1 by a discrete subgroup. Since M is simply connected, this discrete subgroup has to be trivial and M identiﬁes with G1 . But there is no compact simply connected complex Lie group: a contradiction.

theorem 4.2. A compact connected simply connected complex n-manifold without nonconstant meromorphic functions admits no holomorphic unimodular rigid geometric structures. Proof. Assume, by contradiction, that (M, φ) veriﬁes the hypothesis. Since a(M) = 0, Theorem 2.1 implies φ is locally homogeneous on an open dense set U. As M is simply connected, elements in the Killing algebra G extend

meromorphic almost rigid geometric structures / 45

on all of M: the connected simply connected complex Lie group G associated to G acts isometrically on M with an open dense orbit. The open dense orbit U identiﬁes with a homogeneous space G/H, where H is a closed subgroup of G. Consider X1 , X2 , . . . , Xn global Killing ﬁelds on M that are linearly independent at some point of the open orbit U. As before, vol(X1 , X2 , . . . , Xn ) is a nonzero constant, where vol is the holomorphic volume form associated to φ. Thus the Xi give a holomorphic parallelization of TM and Wang’s theorem enables us to conclude as in the previous proof. For nonunimodular rigid geometric structures we have the following less precise:

theorem 4.3. [15] Let M be a compact connected simply connected complex nmanifold without nonconstant meromorphic functions and admitting a holomorphic rigid geometric structure φ. Then M is an equivariant compactiﬁcation of \G, where is a discrete noncocompact subgroup in a complex Lie group G. Proof. Since a(M) = 0, Theorem 2.1 implies φ is locally homogeneous on an open dense set U. As before, the extension property of local Killing ﬁelds implies U is a complex homogeneous space G/H, where G is a connected simply connected complex Lie group and H is a closed subgroup in G. We show now that H is a discrete subgroup of G. Assume by contradiction the Lie algebra of H is nontrivial. Take at any point u ∈ U, the isotropy subalgebra Hu (i.e., the Lie subalgebra of Killing ﬁelds vanishing at u). Remark that Hgu = Ad(g)Hu , for any g ∈ G and u ∈ U, where Ad is the adjoint representation. The map u → Hu is a meromorphic map from M to the Grassmanian of d-dimensional vector spaces in G . But M doesn’t admit any nonconstant meromorphic function and this map has to be constant. It follows that Hu is Ad(G)invariant and H is a normal subgroup of G: a contradiction, since the G-action on M is faithful. Thus G is of dimension n and H identiﬁes to a lattice in G. As M is simply connected, U has to be strictly contained in M and M is an equivariant compactiﬁcation of \G. We don’t know whether such equivariant compactiﬁcations of \G admit equivariant holomorphic rigid geometric structures, but the previous result has the following application.

46 / s. dumitrescu

Recall that a well-known open question asks whether the 6-dimensional real sphere S6 admits complex structures or not. In this context, we have the following:

corollary 4.4. If S6 admits a complex structure M, then M admits no holomorphic rigid geometric structures. Proof. The starting point of the proof is a result of [11] where it is proved that M doesn’t admit nonconstant meromorphic functions. If M supports holomorphic rigid geometric structures, then Theorem 4.3 implies that M is an equivariant compactiﬁcation of a homogeneous space. This is in contradiction with the main theorem of [30]. As for Kaehler manifolds we have proved in [15] the following more precise:

theorem 4.5. Let M be a compact connected Kaehler manifold endowed with a holomorphic unimodular rigid geometric structure φ of afﬁne type. Then, up to a ﬁnite unramiﬁed cover, M is a complex torus (quotient of Cn by a lattice) and φ is translation-invariant. The proof is done in two steps. First we prove that φ is locally homogeneous. Then we use a splitting theorem [5], which asserts that such compact Kaehler manifolds with a holomorphic volume form (Calabi-Yau manifolds) are biholomorphic, up to a ﬁnite cover, to a direct product of a complex torus and a compact simply connected Kaehler manifold with a holomorphic volume form. Starting with φ and using the product structure, we construct a holomorphic unimodular rigid geometric structure on the simply connected factor that is locally homogeneous. We conclude as in the proof of Theorem 4.2 that the simply connected factor is trivial. Recently, we adapted this proof in [19] to all holomorphic Cartan geometries of algebraic type on Calabi-Yau manifolds. A similar result was independently proved in [10]. One can also ﬁnd a classiﬁcation of certain holomorphic G-structures of order one on uniruled projective manifolds in [31].

5. Applications to Complex Surfaces theorem 5.1. Let S be a compact complex surface endowed with a holomorphic unimodular rigid geometric structure φ of afﬁne type. Then the Killing Lie algebra of φ is nontrivial.

meromorphic almost rigid geometric structures / 47

The assumption on φ to be afﬁne is essential:

remark 5.2. If S is a compact complex algebraic surface with trivial canonical bundle (for example, a complex torus or an algebraic K3 surface), the geometric structure given by a holomorphic volume form on S together with a holomorphic embedding of S in a complex projective space doesn’t admit any nontrivial local isometry. However, this geometric structure is not of afﬁne type. Proof. The proof is a simple corollary of Theorems 4.5 and 2.1. Indeed, Theorem 2.1 implies the Killing Lie algebra of φ is trivial only if the algebraic dimension a(S) equals 2. But in this case S is algebraic [4] and thus Kaehler. Then Theorem 4.5 applies and the Killing Lie algebra is transitive on S and hence of dimension at least 2. Recall that a complex surface is called minimal if it does not contain any copy of P 1 (C) with self-intersection −1 (see [4], page 91). Then we have:

theorem 5.3. Let S be a compact minimal complex surface that is not biholomorphic to a nonalgebraic K 3 surface or to a nonafﬁne Hopf surface. Then S admits holomorphic rigid geometric structures. remark 5.4. By deﬁnition, a complex algebraic manifold admits an embedding in a complex projective space, which was seen to be a holomorphic rigid geometric structure (of order zero). Proof. By the previous remark, it remains to consider the case of nonalgebraic complex surfaces. Then the Enriques-Kodaira classiﬁcation shows that, up to a ﬁnite unramiﬁed cover, any minimal nonalgebraic complex compact surface is biholomorphic to one of the following complex surfaces: a complex tori, a Hopf surface, an Inoue surface, a K3 surface, a principal elliptic principal bundle over an elliptic curve, or a principal elliptic bundle over a Riemann surface of genus g ≥ 2, with odd-ﬁrst Betti number (see [4], p. 244). However, it is known that complex tori, Inoue surfaces, afﬁne Hopf surfaces, principal elliptic bundles over elliptic curves, and principal elliptic bundles over a Riemann surface of genus g ≥ 2, with odd-ﬁrst Betti number, admit ﬂat holomorphic afﬁne connections [34, 36, 43, 54, 58, 59].

48 / s. dumitrescu

The classiﬁcation of compact complex surfaces admitting holomorphic afﬁne connections (see [34, 36, 43, 58, 59]) implies the following result:

theorem 5.5. Let S be a compact complex surface admitting a holomorphic unimodular afﬁne connection. Then, up to a ﬁnite unramiﬁed cover, either S is a complex torus and the connection is translation-invariant, or S is an elliptic principal bundle over an elliptic curve and the connection is locally modeled on a translation-invariant connection on a complex torus. Since any complex manifold endowed with a holomorphic Riemannian metric inherits a holomorphic (unimodular) afﬁne (Levi-Civita) connection [40, 41], this easily implies the following [17]:

corollary 5.6. Let S be a complex compact surface admitting a holomorphic Riemannian metric g. Then, up to a ﬁnite unramiﬁed cover, S is a complex torus and g is (ﬂat) translation-invariant. In [20] we classiﬁed the local geometry of all torsion-free holomorphic afﬁne connections ∇ on compact complex surfaces. In particular, we proved that either ∇ is locally homogeneous, locally isomorphic to a translation-invariant connection on C2 , or ∇ is a nonﬂat connection on a principal elliptic bundle over a Riemann surface of genus g ≥ 2 with odd ﬁrst Betti number and the local Killing Lie algebra of ∇ is one-dimensional generated by the fundamental vector ﬁeld of the principal ﬁbration. In all cases, ∇ is projectively ﬂat. As a consequence we proved in [20] the following:

theorem 5.7. Normal holomorphic projective connections on compact complex surfaces are ﬂat. Inoue Surfaces

Recall that Inoue surfaces are compact complex surfaces in the class VII0 , which are not Hopf, and have a vanishing second Betti number [4, 33]. In [34, 36] it is proved that any Inoue surface admits a (unique) ﬂat torsionfree holomorphic afﬁne connection. Here we prove the following:

theorem 5.8. Holomorphic geometric structures on Inoue surfaces are locally homogeneous.

remark 5.9. Inoue surfaces do not admit nonconstant meromorphic functions [4].

meromorphic almost rigid geometric structures / 49

Proof. Let τ be a holomorphic geometric structure on an Inoue surface S. Let ∇0 be a ﬂat torsion-free holomorphic afﬁne connection on S. We prove that the holomorphic rigid geometric structure τ = (τ , ∇0 ) is locally homogeneous. Since a(S) = 0, Theorem 2.1 implies τ is locally homogeneous on some maximal open dense set S \ E, where E is a compact analytic subset of S of positive codimension. We want to show that E is empty. We prove ﬁrst that, up to a double cover of S, the subset E is a smooth submanifold in S (this is always true if S is a ﬁnite set; but here S might have components of complex dimension 1). Assume, by contradiction, that E is not a smooth submanifold in S. Choose p ∈ E a singular point in E. In particular, p is not isolated in E, but p is isolated among the singular points of E. Since Isloc (τ ) preserves E, it has to preserve the set of its singular points and thus it ﬁxes p. Consequently any local Killing ﬁeld deﬁned in the neighborhood of p has to vanish at p. Denote by G the Lie algebra of local Killing ﬁelds in the neighborhood of p. Since G acts transitively on an open set, its dimension is at least 2. Each element of G preserves ∇ and ﬁxes p. In exponential coordinates the G -action in the neighborhood of p is linear. This gives an embedding of G in the Lie algebra of GL(2, C) (the image of the isotropy representation at p). In particular, G is of dimension ≤ 4. Suppose ﬁrst that G is of dimension 2. The corresponding subgroups of GL(2, C) are conjugated either of diagonal matrices, to the group or to one a b 1 m of the following subgroups , with a ∈ C∗ and b ∈ C, , 0 a−1 0 n m n ∗ , with m ∈ C∗ and n ∈ C. with m ∈ C and n ∈ C , or 0 1 In the ﬁrst case, the invariant closed subset E lies, in exponential coordinates, in the union of the two eigen directions. In the last two cases, E locally lies in the invariant line y = 0. In all situations, up to a double cover of S, the analytic set E is smooth: a contradiction. We now settle the case where G is of dimension 3 or 4. Then the image of G by the isotropy representation in p is conjugated in GL(2, C) to one of the following subgroups: SL(2, C), GL(2, C), or the group of inversible uppertriangular matrices. But GL(2, C) and SL(2, C) don’t admit invariant subsets other than p, which will be an isolated point in E: impossible. In the last situation, E locally coincides, as before, with the unique invariant line and it is smooth.

50 / s. dumitrescu

Up to a double cover, E is a holomorphic submanifold in S. If E admits a component of dimension 1, then this component will be a union of closed curves. But Inoue surfaces contain no curves [33]. This proves that E is a ﬁnite set. Assume, by contradiction, that E is not empty and consider p ∈ E. The previous arguments show that the Lie algebra G is isomorphic either to sl(2, C) or to gl(2, C). Assume ﬁrst G = sl(2, C). As before, the local action of G in the neighborhood of p is conjugated to the action of G on Tp S, which coincides with the standard linear action of sl(2, C) on C2 . This action has two orbits: the point p and C2 \ {p}. The stabilizer H in SL(2, C) of a nonzero vector x ∈ Tp S is conjugated

1 b , with b ∈ C. 0 1 Observe that the action of G on G/H preserves a nontrivial holomorphic vector ﬁeld. The expression of this vector ﬁeld in linear coordinates (z1 , z2 ) in the ∂ ∂ neighborhood of p is z1 + z2 . ∂z1 ∂z2 Since S \ E is locally modeled on (G, G/H), this vector ﬁeld is well deﬁned on S \ E. But E is of complex codimension 2 in S and the extension theorem of Hartogs [27] shows that the vector ﬁeld extends to a global nontrivial holomorphic vector ﬁeld X on S. The vector ﬁeld X is G -invariant on S \ E and hence on all of S. Since the isotropy action of SL(2, C) at p doesn’t preserve nontrivial vectors in Tp S C2 , it follows that X (p) = 0. This is impossible, since Inoue surfaces don’t admit singular nontrivial holomorphic vector ﬁelds [33]. The proof is the same in the case G = gl(2, C). to the following one-parameter subgroup of SL(2, C):

Meromorphic Afﬁne Connections

In the following theorem we describe some meromorphic afﬁne connections on non-Kaehler principal elliptic bundles over a compact Riemann surface of genus g ≥ 2.

theorem 5.10. Let S be a principal elliptic bundle over a compact Riemann surface of genus g ≥ 2, with odd-ﬁrst Betti number. i) A meromorphic torsion-free afﬁne connection ∇ on S is invariant by the principal ﬁbration (i.e., the fundamental vector ﬁeld of the principal ﬁbration is a Killing ﬁeld) if and only if the set of its poles intersects only a ﬁnite set of ﬁbers. ii) The space of those previous connections admitting simple poles on a single ﬁber, above a point ξ0 ∈ , is a complex afﬁne space of dimension 5g + 1 and

meromorphic almost rigid geometric structures / 51 2 , where H 0 (K 2 ) its underlying vector space identiﬁes with Hξ00 (K2 ) × M2,ξ ξ0 0 is the vector space of meromorphic quadratic differentials on with a single pole of order at most two at ξ0 , and M2,ξ0 is the vector space of quasimodular forms of weight 2 on with a single simple pole at ξ0 . iii) If ∇ is generic among the meromorphic afﬁne connections that satisfy (ii), then the local Killing Lie algebra of ∇ is generated by the fundamental vector ﬁeld of the principal ﬁbration. iv) The meromorphic afﬁne connections that satisfy (ii) are projectively ﬂat away from the poles.

Recall that a quasimodular form of weight 2 on is a holomorphic function f deﬁned on the upper-half-plane H such that, forsome K ∈C, we have f (ξ ) = aγ bγ f (γ ξ )(cγ ξ + dγ )−2 − K (cγ ξ + dγ )−1 , for all γ = ∈ SL(2, R) in the cγ d γ fundamental group of [3, 62]. If K = 0, then f is a classical modular form. Theorem 9 in [3] shows that the space of quasimodular forms of weight 2 on that admit a simple pole in a single orbit is a complex vector space of dimension g + 1 (i.e., the quotient of the space of such quasimodular forms over those which are modular is 1-dimensional).

remark 5.11. The proof below shows that S admits a (ﬂat) holomorphic afﬁne connection. Theorem 2.1 applies and, since any meromorphic function on S is a pull-back of a meromorphic function on [4], the orbits of the pseudogroup of local isometries of any meromorphic geometric structure on S contain the ﬁbers of S. Proof. i) The projection of the set of poles of ∇ on is a closed analytic set that is either a ﬁnite set of points or all of . If this projection is and ∇ is invariant by the principal ﬁbration, then each point of S is a pole, which is impossible. This proves the easy sense of the implication. We will now describe the space of connections ∇, such that only a ﬁnite set of ﬁbers contain poles of ∇. Up to a ﬁnite unramiﬁed cover and a ﬁnite quotient, S admits holomorphic afﬁne structures (i.e., ﬂat torsion-free holomorphic afﬁne connections) that can be built in the following way [36]: Consider a discrete torsion-free subgroup in PSL(2, R) such that = \H, with H the upper-half plane. Take any holomorphic projective structure on , its developing map τ : H → P 1 (C), and its holonomy morphism ρ : → PSL(2, C). This embedding of into PSL(2, C) lifts to SL(2, C) (this

52 / s. dumitrescu

follows from the fact that orientable (real) closed 3-manifolds have a trivial second Stiefel-Whitney class [47]). Choose such a lift and consider as a subgroup of SL(2, C). Denote by W = C2 \ {0} the C∗ -tautological bundle over P 1 (C). The canonical afﬁne structure of C2 induces a -invariant afﬁne structure on W and hence a -invariant holomorphic afﬁne structure on the pull-back τ ∗ (W ) C∗ × H. The -action on τ ∗ (W ) comes from the action by deck transformation on H and from the ρ-action on W . The previous holomorphic afﬁne structure on τ ∗ (W ) is also invariant by the homotheties in the ﬁbers (which commute with the -action). Consider now Z a lattice in C∗ , which acts by multiplication on the ﬁbers of τ ∗ (W ) and takes the quotient of τ ∗ (W ) by × . The quotient is a principal elliptic bundle over , with ﬁber \C∗ , biholomorphic to S. The afﬁne structure inherited by the universal cover C × H of S is the pull-back of the previous afﬁne structure on C∗ × H by the map C × H → C∗ × H

(z, ξ ) → (ez , ξ ). In the following we will consider the ﬂat torsion-free holomorphic afﬁne connection ∇0 on S given by the case whereτ is the standard embedding of a b H into P 1 (C). In this case, the action of γ = ∈ ⊂ SL(2, R) on the c d universal cover C × H of S is easily seen to be given by [36]: γ (z, ξ ) = (z + log(cξ + d), γ ξ ), ∀(z, ξ ) ∈ C × H, where log is a determination of the logarithm and the γ -action on H comes from the standard action of SL(2, R) on H. The difference ∇ − ∇0 is a meromorphic (2, 1)-tensor ω on S, or, equivalently, a × -invariant mermorphic (2,1)-tensor ω˜ on the universal cover C × H. Moreover, ω˜ is invariant by t(z, ξ ) = (z + 2iπ, ξ ). Then we have ∂ ∂ ∂ + f12 (z, ξ )dz ⊗ dξ ⊗ + f21 (z, ξ )dz ⊗ dz ⊗ ω˜ = f11 (z, ξ )dz ⊗ dz ⊗ ∂z ∂z ∂ξ + f22 (z, ξ )dz ⊗ dξ ⊗

∂ ∂ ∂ + g11 (z, ξ )dξ ⊗ dz ⊗ + g12 (z, ξ )dξ ⊗ dξ ⊗ ∂ξ ∂z ∂z

+ g21 (z, ξ )dξ ⊗ dz ⊗

∂ ∂ + g22 (z, ξ )dξ ⊗ dξ ⊗ , ∂ξ ∂ξ

with fij , gij meromorphic functions on C × H, and fij (·, ξ ), gij (·, ξ ) holomorphic except for ξ lying in the union of a ﬁnite number of -orbits in H.

meromorphic almost rigid geometric structures / 53

Notice that the difference between ∇0 and the standard afﬁne structure of C × H is given by f11 = f22 = g21 = 1, the others fij , gij being trivial (see the straightforward computation in [36]). Since ω˜ is -invariant and t-invariant, the functions fij (·, ξ ) and gij (·, ξ ) descend on an elliptic curve for all ξ ∈ H. They are constant for all ξ ∈ H for which they are holomorphic. It follows that fij (·, ξ ) and gij (·, ξ ) are constant for ξ lying in a open dense subset of H and, consequently, for all ξ ∈ H. It follows that the functions fij and gij depend only on ξ . Consequently, the ﬂow ∂ of ∂z preserves ω. ˜ Since the fundamental generator of the principal ﬁbration ∂ also preserves ∇0 , it is a Killing ﬁeld for ∇. ∂z ii) In the following we consider only torsion-free connections: f12 = g11 and f22 = g21 . The -invariance of ω˜ yields the following equations: 1) f11 (ξ ) = f11 (γ ξ ) − cf21 (γ ξ )(cξ + d) 2) f12 (ξ ) = f12 (γ ξ )(cξ + d)−2 − 2c 2 f21 (γ ξ ) − cf22 (γ ξ )(cξ + d)−1 + 2cf11 (γ ξ )(cξ + d)−1 3) f21 (ξ ) = f21 (γ ξ )(cξ + d)2 4) f22 (ξ ) = 2f21 (γ ξ )c(cξ + d) + f22 (γ ξ ) 5) g12 (ξ ) = g12 (γ ξ )(cξ + d)−4 + c 2 f11 (γ ξ )(cξ + d)−2 + cf12 (γ ξ )(cξ + d)−3 − c 3 f21 (γ ξ )(cξ + d)−1 − c 2 f22 (γ ξ )(cξ + d)−2 − cg22 (γ ξ )(cξ + d)−3 6) g22 (ξ ) = g22 (γ ξ )(cξ + d)−2 + cf22 (γ ξ )(cξ + d)−1 + c 2 f21 (γ ξ ), a b for all γ = ∈ . c d The equation (3) implies that f21 is a meromorphic vector ﬁeld on the compact Riemann surface . Since a single pole of order at most 1 is allowed, we get f21 = 0 as a direct consequence of Riemann-Roch theorem and Serre duality ([27], page 245). The equations (1) and (4) imply then that f11 and f22 are meromorphic functions on with a single pole of order at most 1. If f11 , f22 are not constant, they give a biholomorphism between and the projective line P 1 (C) [27]: a contradiction, since the genus of is ≥ 2. We conclude that f11 and f22 are constants. Then we have 2 ) f12 (ξ ) = f12 (γ ξ )(cξ + d)−2 − cf22 (cξ + d)−1 + 2cf11 (cξ + d)−1 5 ) g12 (ξ ) = g12 (γ ξ )(cξ + d)−4 + c 2 (f11 − f22 )(cξ + d)−2 + c(f12 − g22 )(γ ξ )(cξ + d)−3 6 ) g22 (ξ ) = g22 (γ ξ )(cξ + d)−2 + cf22 (cξ + d)−1 .

54 / s. dumitrescu

It follows from (2 ) and (6 ) that f12 and g22 are quasimodular forms of weight 2 on with a single simple pole. The space of those quasimodular forms is a complex vector space of dimension g + 1 (see [3], theorem 9). Equation (5 ) is equivalent to the -invariance of the quadratic differen − g . It follows that 2g + f − g is a tial w(ξ )dξ 2 , where w = 2g12 + f12 12 22 12 22 meromorphic quadratic differential on with a single pole of order at most 2. It is classically known (as an application of Riemann-Roch theorem and Serre duality) that the space of quadratic differentials with a single pole of order at most 2 is of complex dimension 3g − 1 (see, for example, [27]). ∂ ∂ iii) Let X = a(z, ξ ) ∂z + b(z, ξ ) ∂ξ be a local holomorphic Killing ﬁeld of a generic connection ∇, in the neighborhood of a point where ∇ is holomorphic (a and b are holomorphic local functions on C × H). The equation of the Killing ﬁeld is [X , ∇Y Z] = ∇[X ,Y ] Z + ∇Y [X , Z] for all Y , Z tangents to S. It is enough to verify the equation for (Y , Z) ∂ ∂ ∂ ∂ ∂ ∂ , ∂ξ , ∂z , ∂z , and ∂ξ , ∂ξ . corresponding to ∂z This leads to the following PDE system: 1) azz + (1 + f11 )az + 2f12 bz = 0 2) bzz + (1 + 2f22 − f11 )bz = 0 3) azξ + (f11 − f22 )aξ + g12 bz + g11 bξ + 4) bzξ + (1 + f22 )az + (g22 − f12 )bz = 0

∂f12 b=0 ∂ξ

5) aξ ξ − g12 az + (2f12 − g12 )aξ + 2g12 bξ + 6) bξ ξ + 2(1 + f22 )aξ − g12 bz + g22 bξ +

∂g12 b=0 ∂ξ

∂g22 b = 0. ∂ξ

The general solution of the ﬁrst equation is b = ν(ξ )e−μz + C(ξ ), with ν, C holomorphic functions of ξ and μ = 1 + 2f22 − f11 . We replace bz in the ﬁrst equation and we get μ f12 (ξ )ν(ξ )e−μz + A(ξ )e−(1+f11 )z , az = f11 − f22 with A a holomorphic function of ξ . Then equation (4) leads to 1 + f11 f12 − g22 ν(ξ ) e−μz + (1 + f22 )A(ξ )e−(1+f11 )z = 0. μ −ν (ξ ) + f11 − f22 For a generic ∇, we have f11 = f22 , thus μ = 1 + f11 and the functions are C-linearly independent. This implies (1 + f22 )A(ξ ) = 0 and, since for a generic ∇, f22 = −1, we have A(ξ ) = 0.

e−μz , e−(1+f11 )z

meromorphic almost rigid geometric structures / 55

We also get I) ν (ξ ) = ν(ξ )

1 + f11 f12 − g22 . f11 − f22

Now we check equation (3). We replace the partial derivatives of a and b in (3) and we get the following: f12 f12 μ ν (ξ ) − g12 − ν(ξ ) e−μz + ( f12 C) = 0. f11 − f22 f11 − f22 Since generically μ = 0, the functions e−μz and 1 are linearly independent, which yields to ( f12 C) = 0 and to II) ν (ξ ) = ν(ξ )

1 f12

[( f11 − f22 )g12 − f12 ].

Relations (I) and (II) are compatible, for a generic connection, only if ν = 0. This implies b = C(ξ ), a = B(ξ ). Our PDE system becomes: b + f b = 0 3 ) ( f11 − g21 )a + f12 12 b=0 5 ) a + (2f12 − g12 )a + 2g12 b − g12 6 ) b + 2(1 + 2g21 )a + g22 b + g22 b = 0.

Since (f12 b) = 0, we have a = 0 and thus a is a constant function. Equation b = 0, which, for a generic ∇, is compatible with implies then 2g12 b − g12 f12 b + f12 b = 0 only if b = 0. ∂ It follows that X is a constant multiple of ∂z . iv) The projective connection associated to the afﬁne connection ∇ is given by the following second-order ODE [14]: (5 )

ξ = K 0 (z, ξ ) + K 1 (z, ξ )ξ + K 2 (z, ξ )(ξ )2 + K 3 (z, ξ )(ξ )3 , where K 0 = −f21 , = 0, K 1 , = (1 + f11 ) − 2(1 + f22 ), K 2 , = −(g22 − 2f12 ) and K 3 = g12 (see [14]). Liouville [42], followed by Tresse [56] and Cartan [14], proved that this projective connection is projectively ﬂat if and only if both of the following invariants vanish: 1 2 − Kzz − 3Kξ0ξ − 6K 0 Kz3 − 3K 3 Kz0 + 3K 0 Kξ2 + 3K 2 Kξ0 + K 1 Kz2 L1 = 2Kzξ

− 2K 1 Kξ1 , 2 3 − Kξ1ξ − 3Kzz + 6K 3 Kξ0 + 3K 0 Kξ3 − 3K 3 Kz1 − 3K 1 Kz3 − K 2 Kξ1 L2 = 2Kzξ

+ 2K 2 Kz2 .

56 / s. dumitrescu

Here K 0 = 0, K 1 is a constant function and K 2 , K 3 depend only on ξ . This implies the vanishing of both invariants L1 and L2 . I would like to thank G. Chenevier and G. Dloussky for helpful conversations.

References [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11]

[12] [13] [14] [15] [16] [17] [18] [19]

G. D’Ambra, M. Gromov, Lectures on transformations groups: geometry and dynamics, Surveys in Differential Geometry (Cambridge), (1990), 19–111. A. M. Amores, Vector ﬁelds of a ﬁnite type G-structure, J. Differential Geom., 14(1), (1979), 1–6. N. Azaiez, The ring of quasimodular forms for a cocompact group, J. Number Theory, (2007). W. Barth, K. Hulek, C. Peters, A. Van De Ven, Compact complex surfaces, Ergebnisse der Mathematik, Vol. 4, 2nd Ed., Berlin, Springer, (1995). A. Beauville, Variétés Kähleriennes dont la première classe de Chern est nulle, J. Differential Geom., 18, (1983), 755–782. Y. Benoist, Actions propres sur les espaces homogènes réductifs, Annals of Mathematics, 144, (1996), 315–347. Y. Benoist, Orbites de structures rigides, integrable systems and foliations (Montpellier), Boston, Birkaüser, (1997). Y. Benoist, F. Labourie, Sur les espaces homogènes modèles des variétés compactes, Publ. Math. I.H.E.S., 76, (1992), 99–109. J. Benveniste, D. Fisher, Nonexistence of invariant rigid structures and invariant almost rigid structures, Comm. Anal. Geom., 13(1), (2005), 89–111. I. Biswas, B. McKay, Holomorphic Cartan geometries and Calabi-Yau manifolds, arXiv math. AG/0812.3978, (2010). F. Campana, J-P. Demailly, T. Peternell, The algebraic dimension of compact complex threefolds with vanishing second Betti number, Compositio Math., 112(1), (1998) 77– 91. A. Candel, R. Quiroga-Barranco, Gromov’s centralizer theorem, Geom. Dedicata, 100, (2003), 123–155. Y. Carrière, Autour de la conjecture de L. Markus sur les variétés afﬁnes, Invent. Math., 95, (1989), 615–628. E. Cartan, Sur les variétés à connexion projective, Bull. Soc. Math. France, 52, (1924), 205–241. S. Dumitrescu, Structures géométriques holomorphes sur les variétés complexes compactes, Ann. Scient. Ec. Norm. Sup., 34(4), (2001), 557–571. S. Dumitrescu, Métriques riemanniennes holomorphes en petite dimension, Ann. Instit. Fourier, Grenoble, 51(6), (2001), 1663–1690. S. Dumitrescu, Homogénéité locale pour les métriques riemanniennes holomorphes en dimension 3, Ann. Instit. Fourier, Grenoble, 57(3), (2007), 739–773. S. Dumitrescu, Une caractérisation des variétés complexes compactes parallélisables admettant des structures afﬁnes, C.R. Acad. Sci. Paris, Ser. I, 347, (2009), 1183–1187. S. Dumitrescu, Killing ﬁelds of holomorphic Cartan geometries, to appear in Monatshefte fur Mathematik, (2010).

meromorphic almost rigid geometric structures / 57 [20] S. Dumitrescu, Connexions afﬁnes et projectives sur les surfaces complexes compactes, Mathematische Zeitschrift, 264(2), (2010), 301–316. [21] S. Dumitrescu, A. Zeghib, Global rigidity for holomorphic Riemannian metrics on compact threefolds, Math. Ann., 345(1), (2009), 58–81. [22] R. Feres, Rigid geometric structures and actions of semisimple Lie groups, Rigidité, groupe fondamental et dynamique, Panorama et synthèses, Soc. Math. France, Paris, 13, (2002). [23] C. Frances, Une preuve du théorème de Liouville en géométrie conforme dans le cas analytique, Enseign. Math., 49(2), (2003), 95–100. [24] D. Fried, W. Goldman, Three-dimensional afﬁne christallographic groups, Adv. Math., 47(1), (1983), 1–49. [25] E. Ghys, Déformations des structures complexes sur les espaces homogènes de SL(2, C), J. Reine Angew. Math., 468, (1995), 113–138. [26] W. Goldman, Nonstandard Lorentz space forms, J. Diff. Geom., 21(2), (1985), 301–308. [27] P. Grifﬁths, J. Harris, Principles of algebraic geometry, New York, Wiley-Interscience, (1994). [28] M. Gromov, Rigid transformation groups, Géométrie Différentielle, (D. Bernard et Choquet-Bruhat Ed.), Travaux en cours, Hermann, Paris, 33, (1988), 65–141. [29] R. Gunning, Lectures on Riemann surfaces, Princeton, NJ, Princeton University Press, (1966). [30] A. Huckleberry, S. Kebekus, T. Peternell, Group actions on S6 and complex structures on P 3 (C), Duke Math. J., 102(1), (2000), 101–124. [31] J-M. Hwang, N. Mok, Uniruled projective manifolds with irreducible reductive Gstructure, J. Reine Angew. Math., 490 (1997), 55–64. [32] E. Ince, Ordinary differential equations, New York, Dover Publications, (1944). [33] M. Inoue, On surfaces of class VII0 , Invent. Math., 24, (1974), 269–310. [34] M. Inoue, S. Kobayashi, T. Ochiai, Holomorphic afﬁne connections on compact complex surfaces, J. Fac. Sci. Univ. Tokyo, 27(2), (1980), 247–264. [35] A. Kirilov, Eléments de la théorie des représentations, Moscow, M.I.R., (1974). [36] B. Klingler, Structures afﬁnes et projectives sur les surfaces complexes, Ann. Inst. Fourier, Grenoble, 48(2), (1998), 441–477. [37] B. Klingler, Complétude des variétés Lorentziennes à courbure sectionnelle constante, Math. Ann., 306, (1996), 353–370. [38] T. Kobayashi, T. Yoshino, Compact Clifford-Klein form of symmetric spaces—revisited, Pure Appl. Math. Q., 1(3), (2005), 591–663. [39] R. Kulkarni, F. Raymond, 3-dimensional Lorentz space-forms and Seifert ﬁber spaces, J. Diff. Geom., 21(2), (1985), 231–268. [40] C. Lebrun, Spaces of complex null geodesics in complex-Riemannian geometry, Trans. Amer. Math. Soc., 278, (1983), 209–231. [41] C. Lebrun, H-spaces with a cosmological constant, Proc. Roy. Soc. London, Ser. A, 380(1778), (1982), 171–185. [42] R. Liouville, Sur les invariants de certaines équations différentielles et sur leurs applications, Journal de l’Ecole Polytechnique, 59, (1889), 7–76. [43] K. Maehara, On elliptic surfaces whose ﬁrst Betti numbers are odd, Intl. Symp. on Alg. Geom., Kyoto, (1977), 565–574. [44] L. Markus, Cosmological models in differential geometry, mimeographed notes, Univ. of Minnesota, (1962).

58 / s. dumitrescu [45] B. McKay, Characteristic forms of complex Cartan geometries, arXiv math. DG/ 0704.2555, (2009). [46] J. Milnor, Curvatures of left invariant metrics on Lie groups, Adv. in Math., 21, (1976), 293–329. [47] J. Milnor, J. Stasheff, Characteristic classes, Princeton, Princeton University Press, (1974). [48] K. Nomizu, On local and global existence of Killing vector ﬁelds, Ann. of Math. (2), 72, (1960), 105–120. [49] V. Popov, E. Vinberg Invariant theory, Algebraic Geometry 4, E.M.S., 55, (1991), 123– 280. [50] M. Raghunathan, Discrete subgroups of Lie groups, Berlin, Springer-Verlag, (1972). [51] S. Rahmani, Métriques de Lorentz sur les groups de Lie unimodulaires de dimension 3, J. Geom. Phys., 9, (1992), 295–302. [52] N. Rahmani, S. Rahmani, Lorentzian geometry of the Heisenberg group, Geom. Dedicata, 118, (2006), 133–140. [53] F. Salein, Variétés anti-de Sitter de dimension 3 exotiques, Ann. Inst. Fourier, Grenoble, 50(1), (2000), 257–284. [54] T. Suwa, Compact quotient spaces of C2 by afﬁne transformation groups, J. Diff. Geom. 10, (1975), 239–252. [55] W. Thurston, The geometry and the topology of 3-manifolds, Princeton, Princeton University Press, (1983). [56] A. Tresse, Détermination des invariants ponctuels de l’équation différentielle ordinaire du second ordre y = ω(x, y, y ), Leipzig, 87 S, gr. 8, (1896). [57] K. Ueno, Classiﬁcation theory of algebraic varities and compact complex spaces, Lect. Notes in Math., 439, Berlin, Springer-Verlag, (1975). [58] A. Vitter, Afﬁne structures on compact complex manifolds, Invent. Math., 17, (1972), 231–244. [59] C. Wall, Geometric structures on compact complex analytic surfaces, Topology, 25(2), (1986), 119–153. [60] H. Wang, Complex parallelisable manifolds, Proc. Amer. Math. Soc. 5, (1954), 771–776. [61] J. Wolf, Spaces of constant curvature, New York, McGraw-Hill, (1967). [62] D. Zagier, Lecture, Cours at Collège de France, 2000–2001. [63] A. Zeghib, On Gromov’s theory of rigid transformation groups: a dual approach, Ergodic Th. Dyn. Syst., 20(3), (2000), 935–946. [64] R. Zimmer, On the automorphism group of a compact Lorentz manifold and other geometric manifolds, Invent. Math., 83, (1986), 411–426. [65] R. Zimmer, Actions of semisimple groups and discrete subgroups, Proc. Internat. Cong. Math., Berkeley, CA, (1986), 1247–1258. [66] R. Zimmer, Automorphisms groups of geometric manifolds, Proc. Symp. Pure Math., 54(3), (1993), 693–710. [67] R. Zimmer, Ergodic theory and semisimple Lie groups, Boston, Birkhaüser, (1984).

3

renato feres and emily ronshausen

HARMONIC FUNCTIONS OVER GROUP ACTIONS to bob zimmer on his

60 th

birthday

Abstract Let X be a compact space, a countable group of homeomorphisms of X, and μ a probability measure on . From this data one naturally deﬁnes a random walk on X and a related notion of continuous harmonic function. The question we wish to investigate is whether all continuous harmonic functions on X are -invariant. If this is the case, we say that the system satisﬁes the topological Liouville property. We show that if μ is a symmetric probability measure on an arbitrary and X = S1 or the interval [0, 1], then the Liouville property always holds. In addition, when the Poisson boundary of (, μ) can be identiﬁed with the circle, we give a general construction of a non-Liouville action of on S2 , generalizing an example from [FZ2].

1. Introduction A useful idea in the study of group actions without invariant measures is to introduce and study a random Z-action associated to a choice of probability measure on the acting group. To ﬁx notations, let denote a group of transformations of a space X and μ a probability measure on . Then μ deﬁnes a random walk on , which induces a random Z-action on X . See, for example, [Fu] where this point of view is well represented. It is often assumed in this context that the -space admits a μ-stationary measure (also called μ-harmonic measure—see deﬁnition below) of full support, endowing the random Z-action with an invariant measure. This brings along the ergodic theory apparatus that relies on the existence of invariant measures. Often, however, the -space has large regions with possibly interesting dynamics that are not detected by any harmonic measure for a given μ. If this is the case, one can hope to gain some insight into the system represented by (X , , μ) by investigating the space of μ-harmonic functions on X . As pointed out in Proposition 2.4 below, continuous μ-harmonic functions on compact -spaces are -invariant on the support of any μ-harmonic probability measure, so nontrivial (i.e., non--invariant) such functions may yield 59

60 / r. feres and e. ronshausen

information about the regions that harmonic measures do not detect. This is the perspective motivating this paper. We have chosen here to consider compact -spaces and continuous functions, although one may also try to pursue the subject on measurable -spaces. The most basic question to ask about (X , , μ) in the present context is whether it satisﬁes a topological Liouville property, i.e., whether all continuous μ-harmonic functions are -invariant. The main result of the paper is that the Liouville property holds for actions of countable groups on 1-dimensional spaces when μ is a symmetric probability measure.

theorem 1.1. Let be a countable group acting on X by homeomorphisms, where X is either the circle S1 or the interval [0, 1]. Let μ be a symmetric probability measure on . Then the Liouville property holds for (X , , μ). We give at the end of the paper an example of -space for which the Liouville property does not hold. More speciﬁcally, we show that if the Poisson boundary of (, μ) can be identiﬁed with the circle S1 , then it is possible to construct a -action on the two-sphere so that (S2 , , μ) admits continuous μ-harmonic functions that are nonconstant on all orbits in the complement of a pair of ﬁxed points. The action can be chosen to be ergodic with respect to the smooth measure class on S2 , whereas the random walk is transient in the complement of the two ﬁxed points. A general recipe for constructing non-Liouville actions is also proposed. The proof of Theorem 1.1 parallels the proof of the main theorem of [FFP]. That paper is about the Liouville property for codimension-one foliated Brownian motion and harmonic functions for the Laplace-Beltrami operator associated to a Riemannian metric on leaves, rather than discrete group actions. Although the two situations cannot be directly compared, our assumptions on and μ are, in a sense, much more general than related assumptions on that paper. The example of a non-Liouville action on S2 generalizes a similar construction given in [FZ2] in the context of foliations with a Riemannian metric on leaves. See, also, [FZ1] for these issues in the setting of holomorphic functions on foliations. The paper [DK] has some deep results about harmonic functions on foliated spaces in the spirit of the present discussion. This paper was partly written during a visit by the ﬁrst author to the ParisSud 11 University, at Orsay. He wishes to thank the group of Topology and Dynamics for their hospitality and the university for its generous support. He also wishes to thank Vadim Kaimanovich and Bertrand Deroin for helpful and

harmonic functions over group actions / 61

illuminating discussions. In addition, we wish to express our thanks to the anonymous referee for several helpful comments.

2. Harmonic Functions on -spaces In this section we collect some basic properties and general remarks concerning harmonic functions. Let be a countable (inﬁnite) group, μ a probability measure on , and X a compact topological space on which acts by homeomorphisms. It will always be assumed that μ is non-degenerate, that is, the support of μ generates as a semigroup. In particular, μ(γ ) < 1 for all γ ∈ . A continuous function f : X → R is said to be μ-harmonic if, for all x ∈ X , f (x) =

f (γ (x))μ(γ ).

γ ∈

Let H(X , , μ) denote the set of μ-harmonic functions on X and C(X ) the subset consisting of continuous -invariant functions. If H(X , , μ) = C(X ) , we say that (X , , μ) satisﬁes the Liouville property. We note that if f ∈ H(X , , μ) and x ∈ X , then f˜x (γ ) := f (γ −1 (x)) is a harmonic function on in the following sense: f˜x (γ ) =

f˜x (γ η)μ(η−1 ).

η∈

ˆ Under the conventions of [Fu], f˜x is harmonic relative to the right-μ-random walk on , where μ(γ ˆ ) = μ(γ −1 ). Thus, if (, μ) ˆ itself has the Liouville property, that is, if bounded harmonic functions on are constant, then necessarily (X , , μ) has the Liouville property for all X . Therefore, the present subject is uninteresting if the Poisson boundary of (, μ) ˆ reduces to a single point. It is known, for example, that if is amenable, then there exists a symmetric (μ = μ) ˆ nondegenerate μ for which the Poisson boundary is trivial. A good example of a group to keep in mind, on the other hand, is the fundamental group of a closed surface of genus at least 2, represented as isometries of the Poincaré disc. It has a nontrivial boundary for many interesting probability measures. For example, a result due to Furstenberg implies that on this group there exists a probability measure μ relative to which the boundary of coincides with the geometric boundary, S1 , of the disc [Fur].

62 / r. feres and e. ronshausen

Let = P − I be the Laplace operator on the set C(X ) of real-valued continu ous functions on X , where (Pf )(x) := μ(γ )f (γ (x)) is the averaging operator and I is the identity. Clearly, f is harmonic iff f = 0. More generally, f satisfying f ≥ 0 (respectively, f ≤ 0) is said to be superharmonic (respectively, subharmonic).

proposition 2.1. (maximum principle) If a superharmonic f ∈ C(X ) is bounded above by C and f (x) = C for an x ∈ X , then f is constant on the -orbit of x. Proof. This trivially follows from μ being nondegenerate and the remark 0≤

( f (γ (x)) − f (x))μ(γ ) ≤

γ ∈

( f (γ (x)) − C)μ(γ ) ≤ 0,

γ ∈

so f (γ (x)) = C for all γ in the support of μ.

corollary 2.2. If the closure of every -orbit contains a unique minimal set, then the Liouville property holds. Proof. Let X0 be the closure of an orbit. The maximum and minimum values in X0 of a harmonic function are attained on orbits where the function is constant. Since X0 has a single minimal set, these two values must agree. It is useful to introduce a (directed) metric dx0 on the orbit x0 as follows. Given x, y in x0 , we specify a path joining x to y by a sequence α = (γ0 , γ1 , . . . , γn−1 ), μ(γi ) > 0, such that y0 = x, yn = y, and yi+1 = γi yi for i = 0, . . . , n − 1. The length of α is deﬁned as L(α) =

n−1

− ln μ(γi ) ∈ (0, ∞].

i=0

We allow x to be joined to itself by the trivial path, deﬁned by n = 0, α = ∅, and L(α) = 0. Now let dx0 (x, y) = inf L(α), where the inﬁmum is taken over all n and all (γ0 , . . . , γn−1 ) describing a path from x to y.

proposition 2.3. (harnack inequality) Let f be a continuous, positive, subharmonic function on X and x0 ∈ X arbitrary. Then for all x, y ∈ x0 ,

harmonic functions over group actions / 63

f (y) ≤ edx0 (x,y) . f (x) Proof. As f is subharmonic, f (x) ≥ γ ∈ μ(γ )f (γ x) ≥ μ(η)f (ηx) for all η ∈ , so f (ηx)/f (x) ≤ e− ln μ(η) . Multiplying such inequalities along any path α from x to y yields the claim. Given (X , , μ) as above, a measure ν on X is called harmonic if ν is quasi-invariant and μ ∗ ν = ν, where the convolution μ ∗ ν is the image of the product measure μ ⊗ ν under the action map × X → X . It is easily shown that μ ∗ ν = νP, where the averaging operator P acts on ν by duality as νP =

μ(γ )γ∗ ν.

γ ∈

The following useful remark is due, in the context of foliated Brownian motion, to Garnett [Gar]. The proof, which is sketched below for completeness, is adapted from Paulin [Pa].

proposition 2.4. Let ν be a harmonic probability measure on (X , , μ). Then every h in H(X , , μ) is -invariant on the support of ν. Proof. It may be assumed that h is non-negative. Let c be a non-negative constant and deﬁne hc (x) as the minimum of {h(x), c}. Note that Phc ≤ hc . Since ν is harmonic, hc dν = hc d(μ ∗ ν) = Phc dν, therefore (Phc − hc )dν = 0. As the functions involved are continuous, it follows that Phc = hc on the support of ν. Applying the maximum principle to hc |supp(ν) , we conclude that h −1 ([c, ∞)) ∩ supp(ν) is -invariant for all c. This implies that h is -invariant on supp(ν). Let C denote the largest -invariant subset of X such that h|C is -invariant for all h ∈ H(X , , μ). Clearly, C is closed. It is also nonempty since, by the maximum principle, it contains every minimal set in X . Thus if M denotes the closure of the union of all the minimal sets of X relative to the action, then M ⊆ C . Due to Proposition 2.4, the set S , deﬁned as the closure of the union of supports of all harmonic measures, is also contained in C . In fact, it is easily seen that M ⊆ S ⊆ C . It will be explained shortly that the random walk associated to (X , , μ) (deﬁned next) converges to C with probability 1.

64 / r. feres and e. ronshausen

3. Random Walk on (X, , μ) Let (X , , μ) be, as above, a compact -space where is a countable group and μ is a probability measure on . The measure induces a Markov transition kernel on X by setting p(x, y) = 0 if y ∈ / x and μ(γ ), p(x, y) = y=γ x

where the sum is over all γ ∈ such that γ x = y. Let Xn denote the random walk on X with transition probabilities p and initial state x0 . For the set C , deﬁned at the end of the previous section, we have the following:

proposition 3.1. The random walk Xn on X \ C is transient. The proposition means that for all x0 ∈ X \ C and every open U containing C , there exists a random integer N, almost surely ﬁnite, such that Xn ∈ U for all n ≥ N. Prior to proving this fact we need to review some information about the Poisson boundary of . Our main source is Furman [Fu]. The boundary of can be described as a compact Hausdorff -space B endowed with a probability measure of full support, ν, satisfying the following properties: 1) ν is harmonic relative to (, μ); ˆ ˆ 2) For almost every sample path γ¯ = (γ1 , γ2 , . . .) of a (right) μ-random walk on , a limit probability νγ¯ = limn→∞ (γ1 . . . γn )∗ ν exists and is equal to a Dirac measure δγ∞ supported on a random point, η∞ , of B; 3) Let H ∞ (, μ) ˆ denote the space of bounded harmonic functions on relative to μ. ˆ Then the map F : L∞ (B, ν) → H ∞ (, μ) ˆ given by F(φ)(γ ) = φ dγ ν is an isometric bijection such that, for h = F(φ), ∗ B φ(η∞ ) = lim h(γ1 . . . γn ). n→∞

It is observed in [Fu], remark 2.16(i), that if is a discrete group, then, for the topological description of B given there (denoted B in Furman’s article), L∞ (B, ν) coincides with the continuous functions on B. In this case, B is nonmetrizable if has a nontrivial Poisson boundary. However, measured theoretically, the Poisson boundary is always a Lebesgue space.

lemma 3.2. Let ηn = γ1 . . . γn be a right-μ-random ˆ walk starting at the unit element e ∈ . Let h ∈ H ∞ (, μ) ˆ and C a positive constant. Then, for a.e. sample

harmonic functions over group actions / 65

path ηn and every sequence ξn ∈ such that de (ηn , ξn ) < C, the sequences h(ηn ) and h(ξn ) converge to the same value. (Here de is the directed metric used in Proposition 2.3.) Proof. Deﬁne νγ = γ∗ ν. From the above properties of the Poisson boundary and Proposition 2.3, it follows that

dνηn dνξn , max dνξn dνηn

≤ eC

for all n. But since νηn converges to the Dirac measure δη∞ , the sequence νξn must converge to the same measure. We now return to Proposition 3.1. Let h ∈ H(X , , μ). As pointed out above, for each x ∈ X \ C we obtain a harmonic function h˜ x ∈ H ∞ (, μ). ˆ By Lemma 3.2, for each C > 0, for a.e. sample path (x0 , x1 , . . .) of the random walk on X , and for every sequence (y0 , y1 , . . .) such that dx (xi , yi ) < C for all i, both sequences h˜ x (xi ) and h˜ x (yi ) converge and have the same limit. Therefore, as C is arbitrary and h is continuous, h must be constant on the orbit of every limit point of xi . But this is to say that all limit points must lie in C . This concludes the proof of Proposition 3.1. One further simple remark, concerning induced random walks for subgroups, is needed. Let 0 be a ﬁnite-index subgroup of . Let N γ denote the ﬁrst time, n ≥ 1, at which random walk on (, μ) ˆ starting at γ reaches 0 . Then N γ is a Markov time, almost surely ﬁnite. Now deﬁne a probability measure on 0 by μˆ 0 (η) = Prob(γN e = η), that is, μˆ 0 (η) is the probability that random walk, (γ0 , γ1 , . . . ), on , starting at the unit element, will ﬁrst return to 0 at η. General properties of Markov times and martingales imply that the restriction to 0 of a function in H ∞ (, μ) ˆ is harmonic relative to μˆ 0 .

lemma 3.3. Let 0 have ﬁnite index in and μ0 the image of the induced measure μˆ 0 under group inverse. Then H(X , , μ) ⊆ H(X , 0 , μ0 ). In particular, if (X , 0 , μ0 ) satisﬁes the Liouville property, then so does (X , , μ). Proof. This is due to the above remarks about μ0 and the relationship between harmonic functions h on (X , , μ) and h˜ x on (, μ) ˆ or (0 , μˆ 0 ). (See the beginning of Section 2.)

66 / r. feres and e. ronshausen

4. Actions on the Circle We now consider actions of discrete (countable) groups on the circle. The following lemma is a well-known group-actions version of Poincaré’s classiﬁcation of circle homeomorphisms.

lemma 4.1. Let be a countable group of homeomorphisms of S1 . Then one of the following holds: 1) The action is minimal; 2) The action is not minimal and there is a unique minimal set; 3) There is a ﬁnite orbit. Proof. It sufﬁces to prove the lemma for a subgroup of ﬁnite index. Thus we may assume that is orientation preserving. Further assume that the action is not minimal and contains no ﬁnite orbit. Let Z be a minimal set and U an arbitrary connected component of S1 \ Z. This minimal set is unique if the orbit of every point in U can be shown to accumulate on Z. Now, U and γ U either are disjoint or agree for each γ ∈ . If the latter, then the endpoints of U are ﬁxed by γ . As the subgroup of ﬁxing the endpoints of U has inﬁnite index, there are inﬁnitely many disjoint intervals of the form γ U, so we can choose a sequence γn U decreasing to 0 in length. Therefore, the orbit of every point in U must limit on Z. By Corollary 2.2, having a unique minimal set implies the Liouville property. Thus, in order to prove Theorem 1.1, it sufﬁces to assume that the action on S1 contains a ﬁnite orbit. By Lemma 3.3, it can be assumed that has a ﬁxed point in S1 and preserves orientation. This reduces the proof of Theorem 1.1 to showing the Liouville property for orientation-preserving actions on the interval [0, 1].

5. Actions on the Interval We now restrict attention to systems ([0, 1], , μ) for which the -action is orientation preserving. Clearly, then, the only ﬁnite orbits are ﬁxed points, and since the union of the ﬁxed points is a closed subset of [0, 1], we may further restrict attention to intervals without interior ﬁnite orbits.

lemma 5.1. If acts by orientation-preserving homeomorphisms of [0,1] without interior ﬁnite orbits, the orbit of every x ∈ (0, 1) must limit on both 0 and 1.

harmonic functions over group actions / 67

Proof. The supremum and inﬁmum of x are easily seen to be ﬁxed points.

lemma 5.2. Suppose that the Liouville property does not hold for ([0, 1], , μ) and the -action is orientation preserving without interior ﬁnite orbits. Then, 1) There is a unique continuous harmonic f not -invariant such that f (0) = 0, f (1) = 1; 2) f (x) is the probability that random walk on ([0, 1], , μ) starting at x converges to 1; 3) f is not constant on any interior orbit; 4) f is increasing. Proof. Let g be a continuous, harmonic, non--invariant function on [0, 1]. By Lemma 5.1 and the maximum principle, g(0) = g(1), and g(x) lies in the open interval with endpoints g(0), g(1) for each x ∈ (0, 1). Lemma 5.1 also implies that g cannot be constant on any interior orbit, so the set C of Proposition 3.1 coincides with {0, 1}. By composing g with an appropriate afﬁne function of R, one obtains f such that f (i) = i for i = 0, 1 and 0 < f (x) < 1 on interior points. Uniqueness is due to the maximum principle. By Proposition 3.1, the random walk Xnx on [0, 1] starting at x must converge to a random point X∞ ∈ {0, 1} with probability 1. As f is harmonic, the expected value E[ f (Xnx )] is equal to f (x) and limn→∞ f (Xn ) exists almost surely. By continuity, f (Xnx ) converges to either 0 or 1. Therefore, f (x) = lim E[ f (Xnx )] = f (0)Prob(Xnx → 0) + f (1)Prob(Xnx → 1) n→∞

= Prob(Xnx → 1). Finally, given any x1 < x2 and a sample path (γ1 , γ2 , . . .) on for the right-μˆ random walk, the corresponding sample paths for the random walks on [0, 1] satisfy Xnx1 < Xnx2 . Therefore, the probability that Xnx2 converges to 1 is at least as great as the probability that Xnx1 converges to 1. This shows that f (x) is increasing. We can now show how the assumption of a non--invariant harmonic function on [0, 1] leads to a contradiction. Let f be as in Lemma 5.2 and deﬁne a probability measure ν on [0, 1] by extending the deﬁnition ν((a, b]) = f (b) − f (a)

68 / r. feres and e. ronshausen

to the Lebesgue measurable subsets of the interval. At this point we make the further assumption that the measure μ on is symmetric. Then, μ ∗ ν((a, b]) = μ(γ )ν (γ −1 a, γ −1 b] γ ∈

=

μ(γ ) f (γ −1 b) − f (γ −1 a)

γ ∈

=

μ(γ )f (γ b) −

γ ∈

μ(γ )f (γ a)

γ ∈

= f (b) − f (a) = ν((a, b]). This remark, which was shown to us by B. Deroin, simpliﬁes a similar but somewhat more involved argument of an earlier version of this paper. The same argument is used in the proof of proposition 5.7 of [DKN]. From this we obtain the following lemma.

lemma 5.3. If μ is symmetric, ν is a harmonic probability measure. We can now conclude the proof of Theorem 1.1. Since by Proposition 2.4 any function in H(X , , μ) must be -invariant on the support of a harmonic probability measure, and since the above f is not constant on any interior orbit of the -action on [0, 1], we arrive at a contradiction. Therefore, a non-invariant, continuous harmonic function cannot exist.

6. Examples We now give a class of examples of -spaces to illustrate the way in which the Liouville property can fail to hold. We begin with a general remark that suggests a recipe for constructing examples. Let μ be a probability measure on the countable (discrete) group and let μ, ˆ as before, be the image of μ under group inverse. The unit ball in H ∞ (, μ) ˆ (the latter equipped with the supremum norm) is a compact space, which we denote by X0 . Note that acts on X0 by homeomorphisms under the deﬁnition (γ , φ) → γ · φ, where (γ · φ)(η) = φ(γ −1 η). The -space X0 has a tautological continuous μ-harmonic function: f (φ) = φ(e). This is, in fact, μ-harmonic since

harmonic functions over group actions / 69

γ ∈

f (γ · φ)μ(γ ) =

γ ∈

φ(γ −1 )μ(γ ) =

φ(γ )μ(γ ˆ ) = φ(e) = f (φ).

γ ∈

Furthermore, since f (γ φ) = φ(γ −1 ), f is not constant on · φ if φ itself is not a constant function. This simple remark suggests the following approach to ﬁnding examples of non-Liouville actions on a -space S. Suppose we can somehow construct a continuous, -equivariant map : S → X0 . We express equivariance by (γ s) = γ (s) and write (s) = φs . If does not map S entirely into the space of constant functions, then f ◦ deﬁnes a continuous, μ-harmonic function on S that is not -invariant. Thus, what we have shown in Theorem 1.1 amounts to the following:

proposition 6.1. If μ is a symmetric probability measure on , then every -equivariant continuous : S1 → X0 maps into the constant functions. We now construct a continuous -equivariant map : S2 → X0 whose image is not contained in the space of constant functions. By the above remark, this yields a non-Liouville action on S2 . The example generalizes one given in [FZ2]. Let and μˆ be such that the Poisson boundary of (, μ) ˆ can be identiﬁed 1 with the circle S . For example, as already noted, if is a uniform group of isometries of the Poincaré disc, it is possible to ﬁnd a symmetric μ for which the Poisson boundary of the group coincides with the geometric boundary of the disc [Fur]. Let ν be a harmonic measure on S1 as in the discussion immediately after Proposition 3.1. The following general fact was pointed out to us by Kaimanovich.

lemma 6.2. Let μˆ be a nondegenerate probability measure on such that a Poisson boundary (B, ν) of (, μ) ˆ is nontrivial. Then the μ-harmonic ˆ measure ν has no atoms. Proof. Suppose for a contradiction that ν does have atoms, and let b ∈ B be such that ν(b) ≥ ν(b ) for all b ∈ B. Since ν is μ-harmonic ˆ and μˆ is nondegenerate, the maximum principle applied to γ → (γ∗ ν)(b) implies ν(b) = ν(γ b) for all γ ∈ . On the other hand, if (γ1 , γ2 , . . .) is a random walk on (, μ), ˆ then (γ1 . . . γn )∗ ν converges to the Dirac measure δg∞ , where g∞ is the random point on the boundary to which the random walk converges. Writing gn = γ1 . . . γn , then 0 < ν(b) = (gn )∗ ν(b) → δg∞ (b). This is possible only if sample paths converge to b almost surely, contradicting the assumption that the Poisson boundary is nontrivial.

70 / r. feres and e. ronshausen

We now construct a nontrivial -equivariant continuous map : S2 → X0 . Let x = (z, θ) in the cylinder S1 × [0, 2π ] represent the interval Ix ⊂ S1 with endpoints z and zeiθ . Naturally, γ Ix = Iγ x . The circle boundaries S1 × {0} and S1 × {2π } are invariant under and correspond to intervals of length 0 or 2π . The quotient (S1 × [0, 2π ])/ ∼ obtained by collapsing the boundary circles to points is homeomorphic to S2 and the -action on the cylinder deﬁnes an action on the quotient by homeomorphisms ﬁxing two points, denoted N = (S1 × {2π})/ ∼ and S = (S1 × {0})/ ∼. This deﬁnes S2 as a topological -space. Let ν be a harmonic probability measure on S1 , which is granted since S1 is a Poisson boundary of (, μ). ˆ Deﬁne (x) = φx , where φx (γ ) = γ∗ ν(Ix ). Note that is well deﬁned since ν(Ix ) is 0 on all intervals of length 0, and 1 on all intervals of length 2π. A simple calculation shows that φx is μ-harmonic ˆ for 2 each x ∈ S and that is -equivariant. Furthermore, is continuous due to the fact that ν has no atoms. Since the Poisson boundary is nontrivial, does not map into the set of constant functions. This shows that (S2 , , μ) is non-Liouville. In this example, the only -orbits in S2 on which (x) is a constant function are the ﬁxed points N, S, so C = {N, S}, where C is the set deﬁned in Proposition 3.1. The same argument used in Lemma 5.2 shows that g = f ◦ can be composed with an afﬁne transformation of R to ensure g(S) = 0 and g(N) = 1, after which g(x) is the probability that random walk on (S2 , , μ) starting at x ∈ S2 converges to N. When (S1 , ν) coincides with the boundary of the hyperbolic disc, where is a uniform lattice in PSL(2, R), lemma 4.2 of [FZ2] can be used to show that the just-constructed action on S2 is ergodic relative to the smooth measure class on S2 . This dynamical property of the action contrasts with the simple behavior of the random walk, which converges to N or S with probability 1. It is interesting to regard the induced random walk on the space X0 itself, for a general (, μ). ˆ Let I = [−1, 1] ⊂ X0 represent the subspace of constant functions in X0 . Then the random walk on (X0 , , μ) converges toward I, that is, the random walk on X0 \ I is transient. On the other hand, the -action on X0 can be dynamically very complicated. For example, it is shown in [FZ2] for subgroups of PSL(2, R) that the Z-action on X0 induced by a parabolic or hyperbolic element is a chaotic dynamical system.

harmonic functions over group actions / 71

References [DK]

B. Deroin and V. Kleptsyn. Random conformal dynamical systems, GAFA 17 (2007), 1043–1105. [DKN] B. Deroin, V. Klepstyn, A. Navas. Sur la dynamique unidimensionnelle en régularité intermédiaire, Acta Mathematica–Stockholm 199 (2007), no. 2, 199–262. [FFP] S. Fenley, R. Feres, K. Parwani. Harmonic functions on R-covered foliations, Ergodic Theory Dynam. Systems 29 (2009), no. 4, 1141–1161. [FZ1] R. Feres and A. Zhegib. Leafwise holomorphic functions, Proc. Amer. Math. Soc. 131 (2003), no. 6, 1717–1725. [FZ2] R. Feres and A. Zhegib. Dynamics on the space of harmonic functions and the foliated Liouville problem, Ergodic Theory Dynam. Systems 25 (2005), no. 2, 503–516. [Fu] A. Furman. Random walks on groups and random transformations, Handbook of dynamical systems, Vol. 1A, chap. 12. Hasselblatt and Katok, North-Holland, 2002, 931–1014. [Fur] H. Furstenberg. Random walks and discrete subgroups of Lie groups, 1971 Advances in Probability and Related Topics, Vol. 1. New York, Dekker, 1971, 1–63. [Gar] L. Garnett. Foliations, the ergodic theorem and Brownian motion, J. Funct. Anal. 51 (1983), 285–311. [Pa] F. Paulin. Analyse harmonique des relations d’équivalence mesurées discrètes, Markov Proc. Rel. Fields. 5 (1999), 163–200.

4

david fisher

GROUPS ACTING ON MANIFOLDS: AROUND THE ZIMMER PROGRAM to robert zimmer on the occasion of his

60 th

birthday

Abstract This paper is a survey on the Zimmer program. In its broadest form, this program seeks an understanding of actions of large groups on compact manifolds. The goals of this survey are (1) to put in context the original questions and conjectures of Zimmer and Gromov that motivated the program, (2) to indicate the current state of the art on as many of these conjectures and questions as possible, and (3) to indicate a wide variety of open problems and directions of research.

Contents

1 2

3

4

Prologue Some Remarks on Biases and Omissions A Brief Digression: Some Examples of Groups and Actions 2.1 Semisimple Groups and Their Lattices 2.2 Some Actions of Groups and Lattices 2.3 Induced Actions Pre-history 3.1 Local and Global Rigidity of Homomorphisms into Finite Dimensional Groups 3.2 Strong and Super-rigidity 3.3 Harmonic Map Approaches to Rigidity 3.4 A Remark on Cocycle Super-rigidity 3.5 Margulis’s Normal Subgroup Theorem Low-Dimensional Actions: Conjectures and Results 4.1 Dimension 1 4.2 Dimension 2

74 78 79 80 83 85 86 86 88 90 90 90 91 92 93

Author partially supported by NSF grants DMS-0541917 and 0643546, a fellowship from the Radcliffe Institute for Advanced Studies and visiting positions at École Polytechnique, Palaiseau and Université Paris-Nord, Villateneuse.

72

groups acting on manifolds / 73

4.3 4.4 4.5 4.6

5

6

7

8

9 10

Dimension 3 Analytic Actions in Low Dimensions Zimmer’s Full Conjecture, Partial Results Some Approaches to the Conjectures 4.6.1 Discrete Spectrum of Actions 4.6.2 Effective Invariant Metrics 4.7 Some Related Questions and Conjectures on Countable Subgroups of Diff (M) 4.7.1 Groups with Property (T ) and Generic Groups 4.7.2 Universal Lattices 4.7.3 Irreducible Actions of Products 4.7.4 Torsion Groups and Homeo (M) Cocycle Super-rigidity and Immediate Consequences 5.1 Zimmer’s Cocycle Super-rigidity Theorem and Generalizations 5.2 First Applications to Group Actions and the Problem of Regularity Geometric Actions and Rigid Geometric Structures 6.1 Rigid Structures and Structures of Finite Type 6.2 Rigid Structures and Representations of Fundamental Groups 6.3 Afﬁne Actions Close to the Critical Dimension 6.4 Zimmer’s Borel Density Theorem and Generalizations 6.5 Actions Preserving a Complex Structure 6.6 Conjectures and Questions Topological Super-rigidity: Regularity Results from Hyperbolic Dynamics and Applications 7.1 Stability in Hyperbolic Dynamics. 7.2 Uniformly Hyperbolic Actions on Tori 7.3 Rigidity Results for Uniformly Hyperbolic Actions 7.4 Conjectures and Questions on Uniformly Hyperbolic Actions Representations of Fundamental Groups and Arithmetic Quotients 8.1 Linear Images of Fundamental Groups 8.2 Arithmetic Quotients 8.3 Open Questions Exotic Volume Preserving Actions Non-Volume Preserving Actions

96 96 98 100 100 101 103 103 104 104 105 106 106 107 108 109 112 113 114 116 116 118 118 121 123 124 125 126 128 128 129 133

74 / d. fisher

10.1 Stuck’s Examples 10.2 Weinberger’s Examples 10.3 Work of Nevo-Zimmer 11 Groups That Do Act on Manifolds 11.1 Thompson’s Groups 11.2 Highly Distorted Subgroups of Diff (S2 ) 11.3 Topological Construction of Actions on High Dimensional Spheres 12 Rigidity for Other Classes of Acting Groups 12.1 Lattices in Semi-direct Products 12.2 Universal Lattices 12.3 SO(1,n) and SU(1,n) and Their Lattices 12.4 Lattices in Other Locally Compact Groups 12.5 Automorphism Groups of Free and Surface Groups 13 Properties of Subgroups of Diff (M) 13.1 Jordan’s Theorem 13.2 Burnside Problem 13.3 Tits’s Alternative References

133 133 134 136 136 138 138 139 139 140 140 141 141 142 142 143 143 144

1. Prologue Traditionally, the study of dynamical systems is concerned with actions of R or Z on manifolds, that is, with ﬂows and diffeomorphisms. It is natural to consider instead dynamical systems deﬁned by actions of larger discrete or continuous groups. For non-discrete groups, the Hilbert-Smith conjecture is relevant, since the conjecture states that non-discrete, locally compact totally disconnected groups do not act continuously by homeomorphisms on manifolds. For smooth actions known results sufﬁce to rule out actions of totally disconnected groups; the Hilbert-Smith conjecture has been proven for Hölder diffeomorphisms [155, 207]. For inﬁnite discrete groups, the whole universe is open. One might consider the “generalized Zimmer program” to be the study of homomorphisms ρ : → Diff (M) where is a ﬁnitely generated group and M is a compact manifold. In this survey much emphasis will be on a program proposed by Zimmer. Here one considers a very special class of both Lie groups and discrete groups, namely semisimple Lie groups of higher real rank and their lattices.

groups acting on manifolds / 75

Zimmer’s program is motivated by several of Zimmer’s own theorems and observations, many of which we will discuss below. But in broadest strokes, the motivation is simpler. Given a group whose linear and unitary representations are very rigid or constrained, might it also be true that the group’s representations into Diff ∞ (M) are also very rigid or constrained at least when M is a compact manifold? For the higher rank lattices considered by Zimmer, Margulis’s superrigidity theorems classiﬁed ﬁnite dimensional representations and the groups enjoy property (T ) of Kazhdan, which makes unitary representations quite rigid. This motivation also stems from an analogy between semisimple Lie groups and diffeomorphism groups. When M is a compact manifold, not only is Diff ∞ (M) an inﬁnite dimensional Lie group, but its connected component is simple. Simplicity of the connected component of Diff ∞ (M) was proven by Thurston using results of Epstein and Herman [227, 50, 116]. Herman had used Epstein’s work to see that the connected component of Diff ∞ (Tn ) is simple and Thurston’s proof of the general case uses this. See also further work on the topic by Banyaga and Mather [8, 167, 166, 168, 169], as well as Banyaga’s book [9]. A major motivation for Zimmer’s program came from his cocycle superrigidity theorem, discussed in detail below in Section 5. One can think of a homomorphism ρ : → Diff (M) as deﬁning a virtual homomorphism from to GL( dim (M), R). The notion of virtual homomorphisms, now more commonly referred to in our context as cocycles over group actions, was introduced by Mackey [154]. In all of Zimmer’s early papers, it was always assumed that we had a homomorphism ρ : → Diff ∞ (M, ω) where ω is a volume form on M. A major motivation for this is that the cocycle superrigidity theorem referred to in the last paragraph only applies to cocycles over measure preserving actions. That this hypothesis is necessary was conﬁrmed by examples of Stuck, who showed that no rigidity could be hoped for unless one had an invariant volume form or assumed that M was very low dimensional [226]. Recent work of Nevo and Zimmer does explore the non-volume preserving case and is discussed below, along with Stuck’s examples and some others due to Weinberger in Section 10. Let G be a semisimple real Lie group, all of whose simple factors are of real rank at least 2. Let in G be a lattice. The simplest question asked by Zimmer was: Can one classify smooth volume preserving actions of on compact manifolds? At the time of this writing, the answer to this question is still unclear. There certainly are a wider collection of actions than Zimmer may have initially suspected and it is also clear that the moduli space of such

76 / d. fisher

actions is not discrete; see [12, 78, 128] or Section 9 below. But there are still few enough examples known that a classiﬁcation remains plausible. And if one assumes some additional conditions on the actions, then plausible conjectures and striking results abound. We now discuss four paradigmatic conjectures. To make this introduction accessible, these conjectures are all special cases of more general conjectures stated later in the text. In particular, we state all the conjectures for lattices in SL(n, R) when n > 2. The reader not familiar with higher rank lattices can consider the examples of ﬁnite index subgroups of SL(n, Z), with n > 2, rather than for general lattices in SL(n, R). We warn the reader that the algebraic structure of SL(n, Z) and its ﬁnite index subgroups makes many results easier for these groups than for other lattices in SL(n, R). The conjectures we state concern, respectively, (1) classiﬁcation of low dimensional actions, (2) classiﬁcation of geometric actions, (3) classiﬁcation of uniformly hyperbolic actions, and (4) the topology of manifolds admitting volume preserving actions. A motivating conjecture for much recent research is the following:

conjecture 1.1. (zimmer’s conjecture) For any n > 2, any homomorphism ρ : SL(n, Z) → Diff (M) has ﬁnite image if dim (M) < n − 1. The same for any lattice in SL(nR). The dimension bound in the conjecture is clearly sharp, as SL(n, R) and all of its subgroups act on Pn−1 . The conjecture is a special case of a conjecture of Zimmer’s that concerns actions of higher-rank lattices on low dimensional manifolds that we state as Conjecture 4.12 below. Recently, much attention has focused on these conjectures concerning low dimensional actions. In our second, geometric, setting (a special case of ) a major motivating conjecture is the following:

conjecture 1.2. (affine actions) Let < SL(n, R) be a lattice. Then there is a classiﬁcation of actions of on compact manifolds that preserve both a volume form and an afﬁne connection. All such actions are algebraically deﬁned in a sense to be made precise below. (See Deﬁnition 2.4 below.) The easiest example of an algebraically deﬁned action is the action of SL(n, Z) (or any of its subgroups) on Tn . We formulate our deﬁnition of algebraically deﬁned action to include all trivial actions, all isometric actions, and all (skew) products of other actions with these. The conjecture is a special case

groups acting on manifolds / 77

of a conjecture stated below as Conjecture 6.15. In addition to considering more general acting groups, we will consider more general invariant geometric structures, not just afﬁne connections. A remark worth making is that, for the geometric structures we consider, the automorphism group is always a ﬁnite dimensional Lie group. The question of when the full automorphism group of a geometric structure is large is well studied from other points of view; see particularly [135]. However, this question is generally most approachable when the large subgroup is connected and much less is known about discrete subgroups. In particular, geometric approaches to this problem tend to use information about the connected component of the automorphism group and give much less information about the group of components particularly if the connected component is trivial. One is also interested in the possibility of classifying actions under strong dynamical hypotheses. The following conjecture is motivated by the work of Feres-Labourie [61] and Goetze-Spatzier [105, 106] and is similar to conjectures stated in [108, 120]:

conjecture 1.3. Let < SL(n, R) be a lattice. Then there is a classiﬁcation of actions of on a compact manifold M that preserve both a volume form and where one element γ ∈ acts as an Anosov diffeomorphism. All such actions are algebraically deﬁned in a sense to be made precise below. (Again see Deﬁnition 2.4.) In the setting of this particular conjecture, a proof of an older conjecture of Franks concerning Anosov diffeomorphisms would imply that any manifold M as in the conjecture was homeomorphic to an infranilmanifold on which γ is conjugate by a homeomorphism to a standard afﬁne Anosov map. Even assuming Franks’s conjecture, Conjecture 1.3 is open. One can make a more general conjecture by only assuming that γ has some uniformly partially hyperbolic behavior. Various versions of this are discussed in §7; see particularly Conjecture 7.9. We end this introduction by stating a topological conjecture about all manifold admitting smooth volume preserving actions of a simple higher rank algebraic group. Very special cases of this conjecture are known and the conjecture is plausible in light of existing examples. More precise variants and a version for lattice actions will be stated below in §8.

conjecture 1.4. Let G be a simple Lie group of real rank at least 2. Assume G has a faithful action preserving volume on a compact manifold M. Then π1 (M)

78 / d. fisher

has a ﬁnite index subgroup such that surjects onto an arithmetic lattice in a Lie group H where H locally contains G. The conjecture says, more or less, that admitting a G action forces the fundamental group of M to be large. Passage to a ﬁnite index subgroup and a quotient is necessary; see Section 9 and [84] for more discussion. The conjecture might be considered the analogue for group actions of Margulis’s arithmeticity theorem. This survey is organized on the following lines. We begin in §2 by describing in some detail the kinds of groups we consider and examples of their actions on compact manifolds. In §3 we digress with a prehistory of motivating results from rigidity theory. Then in §4, we discuss conjectures and theorems concerning actions on “low dimensional” manifolds. In this section, there are a number of related results and conjectures concerning groups not covered by Zimmer’s original conjectures. Here low-dimensional is in two senses (1) compared to the group as in Conjecture 1.1, and (2) absolutely small, as in dimension being between 1 and 4. This discussion is simpliﬁed by the fact that many theorems conclude with the nonexistence of actions or at least with all actions factoring through ﬁnite quotients. In Section 5 we further describe Zimmer’s motivations for his conjectures by discussing the cocycle superrigidity theorem and some of its consequences for smooth group actions. In Sections 6 and 7, we discuss, respectively, geometric and dynamical conditions under which a simple classiﬁcation might be possible. In Section 8, we discuss another approach to classifying G and actions using topology and representations of fundamental groups in order to produce algebraically deﬁned quotients of actions. Then in Section 9, we describe the known “exotic examples” of the acting groups to consider. This construction reveals some necessary complexity of a high-dimensional classiﬁcation. We then describe known results and examples of actions not preserving volume in Section 10 and some rather surprising group actions on manifolds in Section 11. Finally we end the survey with a collection of questions about the algebraic and geometric structure of ﬁnitely generated subgroups of Diff (M). Some remarks on biases and omissions. Like any survey of this kind, this work is informed by its author’s biases and experiences. There is an additional bias in that this paper emphasizes developments that are close to Zimmer’s own work and conjectures. In particular, the study of rigidity of group actions often focuses on the low dimensional setting where all group actions are conjectured

groups acting on manifolds / 79

to be ﬁnite or trivial. While Zimmer did substantial work in this setting, he also proved many results and made many conjectures in more general settings where any potential classiﬁcation of group actions is necessarily, due to the existence of examples, more complicated. Another omission is that almost nothing will be said here about local rigidity of groups actions, since the author has recently written another survey on that topic [77]. While that survey could already use updating, that update will appear elsewhere. For other surveys of Zimmer’s program and rigidity of large group actions the reader is referred to [68, 144, 258]. The forthcoming book by Zimmer and Witte Morris [258] is a particularly useful introduction to ideas and techniques in Zimmer’s own work. Also of interest are (1) a brief survey of rigidity theory by Spatzier with a more geometric focus [223], (2) an older survey also by Spatzier, with a somewhat broader scope [222], (3) a recent problem list by Margulis on rigidity theory with a focus on measure rigidity [164], and (4) a more recent survey by Lindenstrauss focused on recent developments in measure rigidity [147]. Both of the last two mentioned surveys are particularly oriented toward connections between rigidity and number theory, which are not mentioned at all in this survey. Finally, while all mistakes and errors in this survey are the sole responsibility of its author, I would like to thank many people whose comments led to improvements. These include Danny Calegari, Renato Feres, Etienne Ghys, Daniel Groves, Steve Hurder, Jean-Francois Lafont, Karin Melnick, Nicolas Monod, Andrés Navas, Leonid Polterovich, Pierre Py, Yehuda Shalom, Ralf Spatzier, Dave Witte Morris, and an anonymous referee.

2. A Brief Digression: Some Examples of Groups and Actions In this section we brieﬂy describe some of the groups that will play important roles in the results discussed here. The reader already familiar with semisimple Lie groups and their lattices may want to skip to the Subsection 2.2 where we give descriptions of group actions. The following convention is in force throughout this paper. For deﬁnitions of relevant terms the reader is referred to the following subsections. Convention In this article we will have occasion to refer to three overlapping classes of lattices in Lie groups that are slightly different. Let G be a semisimple Lie group and < G a lattice. We call a higher rank lattice if all simple factors of G have real rank at least 2. We call a lattice with (T ) if all

80 / d. fisher

simple factors of G have property (T ). Lastly, we call an irreducible higher rank lattice if G has real rank at least 2 and is irreducible. 2.1. Semisimple Groups and Their Lattices

By a simple Lie group, we mean a connected Lie group all of whose normal subgroups are discrete, though we make the additional convention that R and S1 are not simple. By a semisimple Lie group we mean the quotient of a product of simple Lie groups by some subgroup of the product of their centers. Note that with our conventions, the center of a simple Lie group is discrete and is in fact the maximal normal subgroup. There is an elaborate structure theory of semisimple Lie groups and the groups are completely classiﬁed; see [115] or [134] for details. Here we merely describe some examples, all of which are matrix groups. All connected semisimple Lie groups are discrete central extensions of matrix groups, so the reader will lose very little by always thinking of matrix groups. 1) The groups SL(n, R), SL(n, C), and SL(n, H) of n by n matrices of determinant 1 over the real numbers, the complex numbers, or the quaternions. 2) The group SP(2n, R) of 2n by 2n matrices of determinant 1 that preserve a real symplectic form on R2n . 3) The groups SO(p, q), SU(p, q), and SP(p, q) of matrices that preserve inner products of signature (p, q) where the inner product is real linear on Rp+q , Hermitian on Cp+q , or quaternionic Hermitian on Hp+q , respectively. Let G be a semisimple Lie group that is a subgroup of GL(n, R). We say that G has real rank k if G has a k-dimensional abelian subgroup that is conjugate to a subgroup of the real diagonal matrices and no k + 1-dimensional abelian subgroups with the same property. The groups in (1) have rank n − 1, the groups in (2) have rank n, and the groups in (3) have rank min (p, q). Since this article focuses primarily on ﬁnitely generated groups, we are more interested in discrete subgroups of Lie groups than in the Lie groups themselves. A discrete subgroup in a Lie group G is called a lattice if G/ has ﬁnite Haar measure. The lattice is called cocompact or uniform if G/ is compact and non-uniform or not cocompact otherwise. If G = G1 ×· · ·×Gn is a product, then we say a lattice < G is irreducible if its projection to each Gi is dense. It is more typical in the literature to insist that projections to all factors are dense, but this deﬁnition is more practical for our purposes. More generally, we make the same deﬁnition for an almost direct product, by which we mean a direct product G modulo some subgroup of the center Z(G). Lattices

groups acting on manifolds / 81

in semisimple Lie groups can always be constructed by arithmetic methods; see [19] and also [175] for more discussion. In fact, one of the most important results in the theory of semisimple Lie groups is that if G is a semisimple Lie group without compact factors, then all irreducible lattices in G are arithmetic unless G is locally isomorphic to SO(1, n) or SU(1, n). For G of real rank at least 2, this is Margulis’s arithmeticity theorem, which he deduced from his superrigidity theorems [159, 202, 162]. For non-uniform lattices, Margulis had an earlier proof that does not use the superrigidity theorems; see [158, 160]. This earlier proof depends on the study of dynamics of unipotent elements on the space G/ , and particularly on what is now known as the “non-divergence of unipotent ﬂows.” Special cases of the super-rigidity theorems were then proven for Sp(1, n) and F4−20 by Corlette [40] and Gromov-Schoen [113], which sufﬁced to imply the statement on arithmeticity given above. As we will be almost exclusively concerned with arithmetic lattices, we do not give examples of non-arithmetic lattices here, but refer the reader to [162] and [175] for more discussion. A formal deﬁnition of arithmeticity, at least when G is algebraic, is

definition 2.1. Let G be a semisimple algebraic Lie group and < G a lattice. Then is arithmetic if there exists a semisimple algebraic Lie group H such that 1) there is a homomorphism π : H 0 → G with compact kernel, 2) there is a rational structure on H such that the projection of the integer points of H to G are commensurable to , that is, π (H(Z))∩ is of ﬁnite index in both H(Z) and . We now give some examples of arithmetic lattices. The simplest is to take the integer points in a simple (or semisimple) group G that is a matrix group, for example, SL(n, Z) or Sp(n, Z). This exact construction always yields lattices, but also always yields non-uniform lattices. In fact, the lattices one can construct in this way have very special properties because they will contain many unipotent matrices. If a lattice is cocompact, it will necessarily contain no unipotent matrices. The standard trick for understanding the structure of lattices in G that become integral points after passing to a compact extension is called change of base. For much more discussion see [162, 175, 245]. We give one example to illustrate the process. Let G = SO(m, n), which we view as the set of matrices in SL(n + m, R) which preserve the inner product ⎞ ⎛ n+m m √ vi w i + ⎝ vi w i ⎠ v, w = − 2 i=1

i=m+1

82 / d. fisher

where vi and wi are the ith components of v and w. This form, and therefore √ G, are deﬁned over the ﬁeld Q( 2) that has a Galois conjugation σ deﬁned √ √ √ by σ ( 2) = − 2. If we look at the points = G(Z[ 2]), we can deﬁne an embedding of in SO(m, n) × SO(m + n) by taking γ to (γ , σ (γ )). It is straightforward to check that this embedding is discrete. In fact, this embeds in H = SO(m, n) × SO(m + n) as integral points for the rational structure on H √ where the rational points are exactly the points (M, σ (M)) where M∈G(Q( 2)). This makes a lattice in H and it is easy to see that projects to a lattice in G, since G is cocompact in H. What is somewhat harder to verify is that is cocompact in H, for which we refer the reader to the list of references above. Similar constructions are possible with SU(m, n) or SP(m, n) in place of SO(m, n) and also with more simple factors and ﬁelds with more Galois automorphisms. There are also a number of other constructions of arithmetic lattices using division algebras. See [194, 175] for a comprehensive treatment. We end this section by deﬁning a key property of many semisimple groups and their lattices. This is property (T ) of Kazhdan, and was introduced by Kazhdan in [131] in order to prove that non-uniform lattices in higher rank semisimple Lie groups are ﬁnitely generated and have ﬁnite abelianization. It has played a fundamental role in many subsequent developments. We do not give Kazhdan’s original deﬁnition, but one which was shown to be equivalent by the work of Delorme [44] and Guichardet [114].

definition 2.2. A locally compact group has property (T ) of Kazhdan if H 1 (, π ) = 0 for every continuous unitary representation π of on a Hilbert space. This is equivalent to saying that any continuous isometric action of on a Hilbert space has a ﬁxed point. remarks 2.3. 1) Kazhdan’s deﬁnition is that the trivial representation is isolated in the Fell topology on the unitary dual of . 2) If a continuous group G has property (T ), so does any lattice in G. This result was proved in [131]. 3) Any semisimple Lie group has property (T ) if and only if it has no simple factors locally isomorphic to SO(1, n) or SU(1, n). For a discussion of this fact and attributions, see [43]. For groups with all simple factors of real rank at least 3, this is proven in [131].

groups acting on manifolds / 83

4) No noncompact amenable group, and in particular no noncompact abelian group, has property (T ). An easy averaging argument shows that all compact groups have property (T ). Groups with property (T ) play an important role in many areas of mathematics and computer science. 2.2. Some Actions of Groups and Lattices

Here we deﬁne and give examples of a general class of actions. A major impetus in Zimmer’s work is determining optimal conditions for actions to lie in this class. The class we describe is slightly more general than the class Zimmer termed “standard actions” in, for example, [249]. Let H be a Lie group and L < H a closed subgroup. Then a diffeomorphism f of H/L is called afﬁne if there is a diffeomorphism f˜ of H such that f ([h]) = f˜ (h) where f˜ = A◦τh with A an automorphism of H with A(L) = L and τh is left translation by some h in H. Two obvious classes of afﬁne diffeomorphisms are left translations on any homogeneous space and linear automorphisms of tori, or more generally, automorphisms of nilmanifolds. A group action is called afﬁne if every element of the group acts by an afﬁne diffeomorphism. It is easy to check that the full group of afﬁne diffeomorphisms Aff (H/L) is a ﬁnite dimensional Lie group and an afﬁne action of a group D is a homomorphism π : D → Aff (H/L). The structure of Aff (H/L) is surprisingly complicated in general; it is a quotient of a subgroup of the group Aut (H) H where Aut (H) is a group of automorphisms of H. For a more detailed discussion of this relationship, see [81, section 6]. While it is not always the case that any afﬁne action of a group D on H/L can be described by a homomorphism π : D → Aut (H) H, this is true for two important special cases: 1) D is a connected semisimple Lie group and L is a cocompact lattice in H, 2) D is a lattice in a semisimple Lie group G where G has no compact factors and no simple factors locally isomorphic to SO(1, n) or SU(1, n), and L is a cocompact lattice in H. These facts are [81, theorems 6.4 and 6.5] where afﬁne actions as in (1) and (2) above are classiﬁed. The most obvious examples of afﬁne actions of large groups are of the following forms, which are frequently referred to as standard actions: 1) Actions of groups by automorphisms of nilmanifolds. That is, let N be a simply connected nilpotent group, < N a lattice (which is necessarily

84 / d. fisher

cocompact), and assume a ﬁnitely generated group acts by automorphisms of N preserving . The most obvious examples of this are when N = Rn , = Zn , and < SL(n, Z), in which case we have a linear action of on Tn . 2) Actions by left translations. That is, let H be a Lie group and < H a cocompact lattice and < H some subgroup. Then acts on H/ by left translations. Note that in this case need not be discrete. 3) Actions by isometries. Here K is a compact group that acts by isometries on some compact manifold M and < K is a subgroup. Note that here is either discrete or a discrete extension of a compact group. We now brieﬂy deﬁne a few more general classes of actions, which we need to formulate most of the conjectures in this paper. We ﬁrst ﬁx some notations. Let A and D be topological groups, and B < A a closed subgroup. Let ρ : D × A/B → A/B be a continuous afﬁne action.

definition 2.4. 1) Let A, B, D, and ρ be as above. Let C be a compact group of afﬁne diffeomorphisms of A/B that commute with the D action. We call the action of D on C\A/B a generalized afﬁne action. 2) Let A, B, D, and ρ be as in (1) above. Let M be a compact Riemannian manifold and ι : D × A/B → Isom (M) a C 1 cocycle. We call the resulting skew product D action on A/B × M a quasi-afﬁne action. If C and D are as in (2), and we have a smooth cocycle α : D × C\ A/B → Isom (M), then we call the resulting skew product action of D on C\A/B × M a generalized quasi-afﬁne action. Many of the conjectures stated in this paper will end with the conclusion that all actions satisfying certain hypotheses are generalized quasi-afﬁne actions. It is not entirely clear that generalized quasi-afﬁne actions of higher rank groups and lattices are much more general than generalized afﬁne actions. The following discussion is somewhat technical and might be skipped on ﬁrst reading. One can always take a product of a generalized afﬁne action with a trivial action on any manifold to obtain a generalized quasi-afﬁne action. One can also do variants on the following. Let H be a semisimple Lie group and < H a cocompact lattice. Let π : → K be any homomorphism of into a compact Lie group. Let ρ be a generalized afﬁne action of G on C\H/ and let M be

groups acting on manifolds / 85

a compact manifold on which K acts. Then there is a generalized quasi-afﬁne action of G on (C\H × M)/.

question 2.5. Is every generalized quasi-afﬁne action of a higher rank simple Lie group of the type just described? The question amounts to asking for an understanding of compact group valued cocycles over quasi-afﬁne actions of higher rank simple Lie groups. I leave it to the interested reader to formulate the analogous question for lattice actions. For some work in this direction, see the paper of Witte Morris and Zimmer [176]. 2.3. Induced Actions

I end this section by describing brieﬂy the standard construction of an induced or suspended action. This notion can be seen as a generalization of the construction of a ﬂow under a function or as an analogue of the more algebraic notion of inducing a representation. Given a group H, a (usually closed) subgroup L, and an action ρ of L on a space X , we can form the space (H × X )/L where L acts on H × X by h · (l, x) = (lh −1 , ρ(h)x). This space now has a natural H action by left multiplication on the ﬁrst coordinate. Many properties of the L action on X can be studied more easily in terms of properties of the H action on (H × X )/L. This construction is particularly useful when L is a lattice in H. This notion suggests the following principle:

principle 2.6. Let be a cocompact lattice in a Lie group G. To classify actions on compact manifolds it sufﬁces to classify G actions on compact manifolds. The principle is a bit subtle to implement in practice, since we clearly need a sufﬁciently detailed classiﬁcation of G actions to be able to tell which ones arise as induction of actions. While it is a bit more technical to state and probably more difﬁcult to use, there is an analogous prinicple for noncocompact lattices. Here one needs to classify G actions on manifolds that are not compact but where the G action preserves a ﬁnite volume. In fact, one needs only to study such actions on manifolds that are ﬁber bundles over G/ with compact ﬁbers. The lemma begs the question as to whether or not one should simply always study G actions. While in many settings this is useful, it is not always. In particular, many known results about actions require hypotheses on the

86 / d. fisher

action where there is no useful way of rephrasing the property as a property of the induced action. Or, perhaps more awkwardly, require assumptions on the induced action that cannot be rephrased in terms of hypotheses on the original action. We will illustrate the difﬁculties in employing Principle 2.6 at several points in this paper. A case where the implications of the principle are particularly clear is a negative result concerning actions of SO(1, n). We will make clear by the proof what we mean by

theorem 2.7. Let G = SO(1, n). Then one cannot classify actions of G on compact manifolds. Proof. For every n there is at least one lattice < G that admits homomorphisms onto non-abelian free groups; see, for example, [150]. And therefore also onto Z. Thus we can take any action of Fn or Z and induce to a G action. It is relatively easy to show that if the induced actions are isomorphic, then so are the actions they are induced from; see, for example, [78]. A classiﬁcation of Z actions would amount to a classiﬁcation of diffeomorphisms and a classiﬁcation of Fn actions would involve classifying all n-tuples of diffeomorphisms. As there is no reasonable classiﬁcation of diffeomorphisms, there is also no reasonable classiﬁcation of n-tuples of diffeomorphisms. We remark that essentially the same theorem holds for actions of SU(1, n) where homomorphisms to Z exist for certain lattices [132]. Much less is known about free quotients of lattices in SU(1, n); see [148] for one example. For a surprising local rigidity result for some lattices in SU(1, n) see [72, theorem 1.3].

3. Pre-history 3.1. Local and Global Rigidity of Homomorphisms into Finite Dimensional Groups

The earliest work on rigidity theory is a series of works by Calabi-Vesentini, Selberg, Calabi, and Weil that resulted in the following:

theorem 3.1. Let G be a semisimple Lie group and assume that G is not locally isomorphic to SL(2, R). Let < G be an irreducible cocompact lattice, then the deﬁning embedding of in G is locally rigid; that is, any embedding ρ close to the deﬁning embedding is conjugate to the deﬁning embedding by a small element of G.

groups acting on manifolds / 87

remark 3.2. 1) If G = SL(2, R), the theorem is false and there is a large, well-studied space of deformation of in G, known as the Teichmueller space. 2) There is an analogue of this theorem for lattices that are not cocompact. This result was proven later and has a more complicated history that we omit here. In this case it is also necessary to exclude G locally isomorphic to SL(2, C). This theorem was originally proven in special cases by Calabi, Calabi and Vesentini, and Selberg. In particular, Selberg gives a proof for cocompact lattices in SL(n, R) for n ≥ 3 in [212], Calabi-Vesentini give a proof when the associated symmetric space X = G/K is Kähler in [29], and Calabi gives a proof for G = SO(1, n) where n ≥ 3 in [28]. Shortly afterward, Weil gave a complete proof of Theorem 3.1 in [231, 233]. In all of the original proofs, the ﬁrst step was to show that any perturbation of was discrete and therefore a cocompact lattice. This is shown in special cases in [28, 29, 212] and proven in a somewhat broader context than Theorem 3.1 in [232]. The different proofs of cases of Theorem 3.1 are also interesting in that there are two fundamentally different sets of techniques employed and this dichotomy continues to play a role in the history of rigidity. Selberg’s proof essentially combines algebraic facts with a study of the dynamics of iterates of matrices. He makes systematic use of the existence of singular directions, or Weyl chamber walls, in maximal diagonalizable subgroups of SL(n, R). Exploiting these singular directions is essential to much later work on rigidity, both of lattices in higher-rank groups and of actions of abelian groups. It seems possible to generalize Selberg’s proof to the case of G an R-split semisimple Lie group with rank at least 2. Selberg’s proof, which depended on asymptotics at inﬁnity of iterates of matrices, inspired Mostow’s explicit use of boundaries in his proof of strong rigidity [179]. Mostow’s work in turn provided inspiration for the use of boundaries in later work of Margulis, Zimmer, and others on rigidity properties of higher-rank groups. The proofs of Calabi, Calabi-Vesentini, and Weil involve studying variations of geometric structures on the associated locally symmetric space. The techniques are analytic and use a variational argument to show that all variations of the geometric structure are trivial. This work is a precursor to much work in geometric analysis studying variations of geometric structures and also informs later work on proving rigidity/vanishing of harmonic forms and

88 / d. fisher

maps. The dichotomy between approaches based on algebra/dynamics and approaches that are in the spirit of geometric analysis continues through much of the history of rigidity and the history of rigidity of group actions in particular. Shortly after completing this work, Weil discovered a new criterion for local rigidity [233]. In the context of Theorem 3.1, this allows one to avoid the step of showing that a perturbation of remains discrete. In addition, this result opened the way for understanding local rigidity of more general representations of discrete groups than the deﬁning representation.

theorem 3.3. Let be a ﬁnitely generated group, G a Lie group, and π : → G a homomorphism. Then π is locally rigid if H 1 (, g) = 0. Here g is the Lie algebra of G and acts on g by AdG ◦π . Weil’s proof of this result uses only the implicit function theorem and elementary properties of the Lie group exponential map. The same theorem is true if G is an algebraic group over any local ﬁeld of characteristic zero. In [233], Weil remarks that if < G is a cocompact lattice and G satisﬁes the hypothesis of Theorem 3.1, then the vanishing of H 1 (, g) can be deduced from the computations in [231]. The vanishing of H 1 (, g) is proven explicitly by Matsushima and Murakami in [170]. Motivated by Weil’s work and other work of Matsushima, conditions for vanishing of H 1 (, g) were then studied by many authors. See particularly [170] and [203]. The results in these papers imply local rigidity of many linear representations of lattices. 3.2. Strong and Super-rigidity

In a major and surprising development, it turns out that in many instances, local rigidity is just the tip of the iceberg and that much stronger rigidity phenomena exist. We now discuss major developments from the 1960s and 1970s. The ﬁrst remarkable result in this direction is Mostow’s rigidity theorem; see [177, 178] and references there. Given G as in Theorem 3.1, and two irreducible cocompact lattices 1 and 2 in G, Mostow proves that any isomorphism from 1 to 2 extends to an isomorphism of G with itself. Combined with the principal theorem of [232], which shows that a perturbation of a lattice is again a lattice, this gives a remarkable and different proof of Theorem 3.1, and Mostow was motivated by the desire for a “more geometric understanding” of Theorem 3.1 [178]. Mostow’s theorem is in fact a good deal stronger,

groups acting on manifolds / 89

and controls not only homomorphisms → G near the deﬁning homomorphism, but any homomorphism into any other simple Lie group G where the image is a lattice. As mentioned above, Mostow’s approach was partially inspired by Selberg’s proof of certain cases of Theorem 3.1 [179]. A key step in Mostow’s proof is the construction of a continuous map between the geometric boundaries of the symmetric spaces associated to G and G . Boundary maps continue to play a key role in many developments in rigidity theory. A new proof of Mostow rigidity, at least for Gi of real rank 1, was provided by Besson, Courtois, and Gallot. Their approach is quite different and has had many other applications concerning rigidity in geometry and dynamics; see, for example, [14, 15, 39]. The next remarkable result in this direction is Margulis’s superrigidity theorem. Margulis proved this theorem as a tool to prove arithmeticity of irreducible uniform lattices in groups of real rank at least 2. For irreducible lattices in semisimple Lie groups of real rank at least 2, the superrigidity theorems classiﬁes all ﬁnite dimensional linear representations. Margulis’s theorem holds for irreducible lattices in semisimple Lie groups of real rank at least 2. Given a lattice < G where G is simply connected, one precise statement of some of Margulis’s results is to say that any linear representation σ of almost extends to a linear representation of G. By this we mean that there is a linear representation σ˜ of G and a bounded image representation σ¯ of such that σ (γ ) = σ˜ (γ )σ¯ (γ ) for all γ in G. Margulis’s theorems also give an essentially complete description of the representations σ¯ , up to some issues concerning ﬁnite image representations. The proof here is partially inspired by Mostow’s work: a key step is the construction of a measurable “boundary map.” However, the methods for producing the boundary map in this case are very dynamical. Margulis’s original proof used Oseledec’s multiplicative ergodic theorem. Later proofs were given by both Furstenberg and Margulis using the theory of group boundaries as developed by Furstenberg from his study of random walks on groups [93, 94]. Furstenberg’s probabilistic version of boundary theory has had a profound inﬂuence on many subsequent developments in rigidity theory. For more discussion of Margulis’s superrigidity theorem, see [159, 161, 162, 245]. Margulis’s theorem, by classifying all linear representations, and not just ones with constrained images, leads one to believe that one might be able to classify all homomorphisms to other interesting classes of topological groups. Zimmer’s program is just one aspect of this theory, in other directions, many authors have studied homomorphisms to isometry groups of non-positively curved (and more general) metric spaces; see, for example, [24, 98, 172].

90 / d. fisher 3.3. Harmonic Map Approaches to Rigidity

In the 1990s, ﬁrst Corlette and then Gromov-Schoen showed that one could prove major cases of Margulis’s superrigidity theorem also for lattices in Sp(1, n) and F4−20 [40, 113]. Corlette considered the case of representations over Archimedean ﬁelds and Gromov and Schoen proved results over other local ﬁelds. These proofs used harmonic maps and proceed in three steps: ﬁrst showing a harmonic map exists, second showing it is smooth, and third using certain special Bochner-type formulas to show that the harmonic mapping must be a local isometry. Combined with the earlier work of Matsushima and Murakami, [169], and Raghunathan [199] leads to a complete classiﬁcation of linear representations for lattices in these groups [170, 203]. It is worth noting that the use of harmonic map techniques in rigidity theory had been pioneered by Siu, who used them to prove generalizations of Mostow rigidity for certain classes of Kähler manifolds [220]. There is also much later work on applying harmonic map techniques to reprove cases of Margulis’s superrigidity theorem by Jost-Yau [122] and Mok-Siu-Yeung [171] among others. We remark in passing that the general problem of existence of harmonic maps for non-compact, ﬁnite volume locally symmetric spaces has not been solved in general. Results of Saper [210] and Jost-Zuo [123] allow one to prove superrigidity for fundamental groups of many such manifolds, but only while assuming arithmeticity. The use of harmonic maps in rigidity was inspired by the use of variational techniques and harmonic forms and functions in work on local rigidity and vanishing of cohomology groups. The original suggestion to use harmonic maps in this setting appears to go back to Calabi [219]. 3.4. A Remark on Cocycle Super-rigidity

An important impetus for the study of rigidity of groups acting on manifolds was Zimmer’s proof of his cocycle superrigidity theorem. We discuss this important result below in Section 5 where we also indicate some of its applications to group actions. This theorem is a generalization of Margulis’s superrigidity theorem to the class of virtual homomorphisms corresponding to cocycles over measure preserving group actions. 3.5. Margulis’s Normal Subgroup Theorem

I end this section by mentioning another result of Margulis’s that has had tremendous importance in results concerning group actions on manifolds. This is the normal subgroups theorem, which says that any normal subgroup in a higher-rank lattice is either ﬁnite or of ﬁnite index; see, for example, [162, 245] for more on the proof. The proof precedes by a remarkable strategy.

groups acting on manifolds / 91

Let N be a normal subgroup of ; we show /N is ﬁnite by showing that it is amenable and has property (T ). This strategy has been applied in other contexts and is a major tool in the construction of simple groups with good geometric properties; see [7, 25, 26, 37]. The proof that /N is amenable already involves one step that might rightly be called a theorem about rigidity of group actions. Margulis shows that if G is a semisimple group of higher rank, P is a minimal parabolic, and is a lattice then any measurable space X that is a measurable quotient of the action on G/P is necessarily of the form G/Q , where Q is a parabolic subgroup containing P. The proof of this result, sometimes called the projective factors theorem, plays a fundamental role in the work of Nevo and Zimmer on non-volume preserving actions. See Section 10.3 for more discussion. It is also worth noting that Dani has proven a topological analogue of Margulis’s result on quotients [42]. That is, he has proven that continuous quotients of the action on G/P are all actions on G/Q . The usual use of Margulis’s normal subgroup theorem in studying rigidity of group actions is usually quite straightforward. If one wants to prove that a group satisfying the normal subgroups theorem acts ﬁnitely, it sufﬁces to ﬁnd one inﬁnite order element that acts trivially.

4. Low-Dimensional Actions: Conjectures and Results We begin this section by discussing results, in particular, very low dimensions; namely dimensions 1, 2, and 3. Before we begin this discussion, we recall a result of Thurston that is often used to show that low-dimensional actions are trivial [228].

theorem 4.1. (thurston stability) Assume is a ﬁnitely generated group with ﬁnite abelianization. Let act on a manifold M by C 1 diffeomorphisms, ﬁxing a point p and with trivial derivative at p. Then the action is trivial in a neighborhood of p. In particular, if M is connected, the action is trivial. The main point of Theorem 4.1 is that to show an action is trivial, it often sufﬁces to ﬁnd a ﬁxed point. This is because, for the groups we consider, there are essentially no non-trivial low dimensional linear representations and therefore the derivative at a ﬁxed point is trivial. More precisely, the groups usually only have ﬁnite image low dimensional linear representations, which allows one to see that the action is trivial on a subgroup of ﬁnite index.

92 / d. fisher 4.1. Dimension 1

The most dramatic results obtained in the Zimmer program concern a question ﬁrst brought into focus by Dave Witte Morris in [236]: can higher rank lattices act on the circle? In fact, the paper [236] is more directly concerned with actions on the line R. A detailed survey of results in this direction is contained in the paper by Witte Morris in this volume, so we do not repeat that discussion here. We merely state a conjecture and a question.

conjecture 4.2. Let be a higher rank lattice. Then any continuous action on S1 is ﬁnite. This conjecture is well known and ﬁrst appeared in print in [101]. By the results in [236] and [146], the case of non-cocompact lattices is almost known. The paper [236] does the case of higher Q rank. The later work of Lifschitz and Witte Morris reduces the general case to the case of quasisplit lattices in SL(3, R) and SL(3, C). It follows from work of Ghys [99, 101, 102] or BurgerMonod [22, 23] that one can assume the action ﬁxes a point, so the question is equivalent to asking if the groups act on the line. We remark that actions of these groups on the circle, which ﬁx a point and are C 1 , are easily seen to be ﬁnite by using Theorem 4.1 and the fact that all the groups in question have ﬁnite ﬁrst homology. An interesting approach might be to study the induced action of on the space of left orders on ; for ideas about this approach, we refer the reader to work of Navas [182] and Witte Morris [174]. Perhaps the following should also be a conjecture, but here we only ask it as a question.

question 4.3. Let be a discrete group with property (T ). Does admit an inﬁnite action by C 1 diffeomorphisms on S1 ? Does admit an inﬁnite action by homeomorphisms on S1 ? By a result of Navas, the answer to the above question is no if C 1 diffeomorphisms are replaced by C k diffeomorphisms for any k > 32 [183]. A result noticed by the author and Margulis [82] and contained in a paper of Bader, Furman, Gelander, and Monod [6] allows one to adapt this proof for values of k slightly less than 32 . By Thurston’s Theorem 4.1, in the C 1 case it sufﬁces to ﬁnd a ﬁxed point for the action. We add a remark here pointed out to the author by Navas, that perhaps justiﬁes only calling Question 4.3 a question. In [184], Navas extends his results from [183] to groups with relative property (T ). This means that no

groups acting on manifolds / 93

such group acts on the circle by C k diffeomorphisms where k > 32 . However, if we let be the semi-direct product of a ﬁnite index free subgroup of SL(2, Z) and Z2 , this group is left orderable, and so acts on S1 ; see, for example, [176]. This is the prototypical example of a group with relative property (T ). This leaves open the possibility that groups with property (T ) would behave in a similar manner and admit continuous actions on the circle and the line. Other possible candidate for a group with property (T ) acting on the circle by homeomorphisms are the more general variants of Thompson’s group F constructed by Stein in [225]. The proofs that F does not have (T ) do not apply to these groups [58]. A closely related question is the following, suggested to the author by Andrés Navas along with examples in the last paragraph.

question 4.4. Is there a group that is bi-orderable and does not admit a proper isometric action on a Hilbert space? That is, it does not have the Haagerup property? For much more information on the fascinating topic of groups of diffeomorphisms of the circle see both the survey by Ghys [102], the more recent book by Navas [181], and the article by Witte Morris in this volume [173]. 4.2. Dimension 2

Already in dimension 2 much less is known. There are some results in the volume preserving setting. In particular, we have

theorem 4.5. Let be a non-uniform irreducible higher rank lattice. Then any volume-preserving action on a closed orientable surface other than S2 is ﬁnite. If has Q-rank at least 1, then the same holds for actions on S2 . This theorem was proven by Polterovich [195] for all surfaces but the sphere and shortly afterward proven by Franks and Handel for all surfaces [87, 86]. It is worth noting that the proofs use entirely different ideas. Polterovich’s proof belongs very clearly to symplectic geometry and also implies some results for actions in higher dimensions. Franks and Handel use a theory of normal forms for C 1 surface diffeomorphisms that they develop in analogy with the Thurston theory of normal forms for surface homeomorphisms with ﬁnite ﬁxed sets. The proof of Franks and Handel can be adapted to a setting where one assumes much less regularity of the invariant measure [88].

94 / d. fisher

There are some major reductions in the proofs that are similar, which we now describe. The ﬁrst of these should be useful for studying actions of cocompact lattices as well. Namely, one can assume that the homomorphism ρ : → Diff (S) deﬁning the action takes values in the connected component. This follows from the fact that any ρ : → MCG(S) is ﬁnite, which can now be deduced from a variety of results [17, 22, 53, 124]. This result holds not only for the groups considered in Theorem 4.5, but also for cocompact lattices and even for lattices in SP(1, n) by results of Sai-Kee Yeung [239]. It may be possible to show something similar for all groups with property (T ) using recent results of Andersen showing that the mapping class group does not have property (T ) [2]. It is unrealistic to expect a simple analogue of this result for homomorphisms to Diff (M)/ Diff (M)0 for general manifolds; see Section 4.3 below for a discussion of dimension 3. Also, in the setting of Theorem 4.5, Margulis’s normal subgroup theorem implies that it sufﬁces to show that a single inﬁnite order element of acts trivially. The proofs of Franks-Handel and Polterovich then use the existence of distortion elements in , a fact established by Lubotzky, Mozes, and Raghunathan in [151]. The main result in both cases shows that Diff (S, ω) does not contain exponentially distorted elements and the proofs of this fact are completely different. An interesting result of Calegari and Freedman shows that this is not true of Diff (S2 ) and that Diff (S2 ) contains subgroups with elements of arbitrarily large distortion [30]. These examples are discussed in more detail in Section 9. Polterovich’s methods also allow him to see that there are no exponentially distorted elements in Diff (M, ω)0 for certain symplectic manifolds (M, ω). Again, this yields some partial results toward Zimmer’s conjecture. On the other hand, Franks and Handel are able to work with a Borel measure μ with some properties and show that any map ρ : → Diff (S, μ) is ﬁnite. We remark here that a variant on this is due to Zimmer, when there is an invariant measure supported on a ﬁnite set.

theorem 4.6. Let be a group with property (T ) acting on a compact surface S by C 1 diffeomorphisms. Assume has a periodic orbit on Sj , then the action is ﬁnite. The proof is quite simple. First, pass to a ﬁnite index subgroup that ﬁxes a point x. Look at the derivative representation dρx at the ﬁxed point. Since has property (T ) and SL(2, R) has the Haagerup property, it is easy to prove that

groups acting on manifolds / 95

the image of dρx is bounded. Bounded subgroups of SL(2, R) are all virtually abelian and this implies that the image of dρx is ﬁnite. Passing to a subgroup of ﬁnite index, one has a ﬁxed point where the derivative action is trivial of a group that has no cohomology in the trivial representation. One now applies Thurston Theorem 4.1. The difﬁculty in combining Theorem 4.6 with ideas from the work of Franks and Handel is that while Franks and Handel can show that individual surface diffeomorphisms have large sets of periodic orbits, their techniques do not easily yield periodic orbits for the entire large group action. We remark here that there is another approach to showing that lattices have no volume preserving actions on surfaces. This approach is similar to the proof that lattices have no C 1 actions on S1 via bounded cohomology [22, 23, 99]. That Diff (S, ω)0 admits many interesting quasimorphisms, follows from the work of Entov-Polterovich [49], Gambaudo-Ghys [96], and Py [198, 200, 199]. By results of Burger and Monod [41], for any higher-rank lattice and any homomorphism ρ : → Diff (S, ω)0 , the image is in the kernel of all of these quasi-morphisms. (To make this statement meaningful and the kernel well deﬁned, one needs to take the homogeneous versions of the quasimorphisms.) What remains to be done is to extract useful dynamical information from this fact; see the article by Py in this volume for more discussion [198]. In our context, the work of Entov-Polterovich mentioned above really only constructs a single quasi-morphism on Diff (S2 , ω)0 . However, this particular quasimorphism is very nice in that it is Lipschitz in metric known as Hofer’s metric on Diff (S2 , ω)0 . This construction does apply more generally in higher dimensions and indicates connections between quasimorphisms and the geometry of Diff (M, ω)0 , for ω a symplectic form, that are beyond the scope of this survey. For an introduction to this fascinating topic, we refer the reader to [195]. Motivated by the above discussion, we recall the following conjecture of Zimmer.

conjecture 4.7. Let be a group with property (T ), then any volume preserving smooth action on surface is ﬁnite. Here the words “volume preserving” or at least “measure preserving” are quite necessary. As SL(3, R) acts on S2 , so does any lattice in SL(3, R) or any irreducible lattice in G where G has SL(3, R) as a factor. The following question seems reasonable; I believe I ﬁrst learned it from Leonid Polterovich.

96 / d. fisher

question 4.8. Let be a higher rank lattice (or even just a group with property (T )) and assume acts by diffeomorphisms on a surface S. Is it true that either (1) the action is ﬁnite or (2) the surface is S2 and the action is smoothly conjugate to an action deﬁned by some embedding i : → SL(3, R) and the projective action of SL(3, R) on S2 ? This question seems quite far beyond existing technology. 4.3. Dimension 3

We now discuss brieﬂy some work of Farb and Shalen that constrains actions by homeomorphisms in dimension 3 [55]. This work makes strong use of the geometry of 3 manifolds, but uses very little about higher rank lattices and is quite soft. A special case of their results is the following:

theorem 4.9. Let M be an irreducible 3-manifold and be a higher rank lattice. Assume acts on M by homeomorphisms so that the action on homology is nontrivial. Then M is homeomorphic to T3 , < SL(3, Z) with ﬁnite index, and the action on H 1 (M) is the standard action on Z 3 . Farb and Shalen actually prove a variant of Theorem 4.9 for an arbitrary 3 manifold admitting a homologically inﬁnite action of a higher rank lattice. This can be considered as a signiﬁcant step toward understanding when, for a higher-rank lattice and M 3 a closed 3 manifold, ρ : → Diff (M 3 ) must have image in Diff (M 3 )0 . Unlike the results discussed above for dimension 2, the answer is not simply “always.” This three dimensional result uses a great deal of the known structure of 3 manifolds, though it does not use the full geometrization conjecture proven by Perelman, but only the Haken case due to Thurston. The same sort of result in higher dimensions seems quite out of reach. A sample question is the following. Here we let be a higher rank lattice and M a compact manifold.

question 4.10. Under what conditions on the topology of M do we know that a homomorphism ρ : → Diff (M) has image in Diff (M)0 ? 4.4. Analytic Actions in Low Dimensions

We ﬁrst mention a direction pursued by Ghys that is in a similar spirit to the Zimmer program, and that has interesting consequences for that program. Recall that the Zassenhaus lemma shows that any discrete linear group generated by small enough elements is nilpotent. The main point of Ghys’s

groups acting on manifolds / 97

article [100] is to attempt to generalize this result for subgroups of Diff ω (M). While the result is not actually true in that context, Ghys does prove some intriguing variants that yield some corollaries for analytic actions of large groups. For instance, he proves that SL(n, Z) for n > 3 admits no analytic action on the 2 sphere. We remark that the attempt to prove the Zassenhaus lemma for diffeomorphism groups suggests that one can attempt to generalize other facts about linear groups to the category of diffeomorphism groups. We discuss several questions in this direction, mainly due to Ghys, in Section 13. For the rest of this subsection we discuss a different approach of Farb and Shalen for showing that real analytic actions of large groups are ﬁnite. This method is pursued in [54, 56, 57]. We begin by giving a cartoon of the main idea. Given any action of a group , an element γ ∈, and the centralizer Z(γ ), it is immediate that Z(γ ) acts on the set of γ ﬁxed points. If the action is analytic, then the ﬁxed sets are analytic and so have good structure and are “reasonably close” to being submanifolds. If one further assumes that all normal subgroups of have ﬁnite index, then this essentially allows one to bootstrap results about Z(γ ) not acting on manifolds of dimension at most n − 1 to facts about not having actions on manifolds of dimension n provided one can show that the ﬁxed set for γ is not empty. This is not true, as analytic sets are not actually manifolds, but the idea can be implemented using the actual structure of analytic sets in a way that yields many results. For example, we have

theorem 4.11. Let M be a real analytic 4 manifold with zero Euler characteristic, then any real analytic, volume preserving action of any ﬁnite index subgroup in SL(n, Z) for n ≥ 7 is trivial. This particular theorem also requires a result of Rebelo [205], which concerns ﬁxed sets for actions of nilpotent groups on T2 and uses the ideas of [100]. The techniques of Farb and Shalen can also be used to prove that certain cocompact higher rank lattices have no real analytic actions on surfaces of genus at least 1. For this result one needs only that the lattice contains an element γ whose centralizer already contains a higher rank lattice. In the paper [54] a more technical condition is required, but this can be removed using the results of Ghys and Burger-Monod on actions of lattices on the circle. (This simpliﬁcation was ﬁrst pointed out to the author by Farb.)

98 / d. fisher

The point is that γ has ﬁxed points for topological reasons and the set of these ﬁxed points contains either (1) a Z(γ ) invariant circle or (2) a Z(γ ) invariant point. Case (1) reduces to case (2) after passing to a ﬁnite index subgroup via the results on circle actions. Case (2) is dealt with by the proof of Theorem 4.6. 4.5. Zimmer’s Full Conjecture, Partial Results

We now state the full form of Zimmer’s conjecture. In fact we generalize it slightly to include all lattices with property (T ). Throughout this subsection G will be a semisimple Lie group with property (T ) and will be a lattice in G. We deﬁne two numerical invariants of these groups. First, for any group F, let d(F) be the lowest dimension in which F admits an inﬁnite image linear representation. We note that the superrigidity theorems imply that d() = d(G) when is a lattice in G and either G is a semisimple group property (T ) or is irreducible higher rank. The second number, n(G) is the lowest dimension of a homogeneous space K /C for a compact group K on which a lattice in G can act via a homomorphism ρ : → K . In Zimmer’s work, n(G) is deﬁned differently, in a way that makes clear that there is a bound on n(G) that does not depend on the choice of . Namely, using the superrigidity theorems, it can be shown that n satisﬁes n(n + 1) ≥ min{dimC G where G is a simple factor of G}. 2 The more fashionable variant of Zimmer’s conjecture is the following, ﬁrst made explicitly by Farb and Shalen for higher rank lattices in [54].

conjecture 4.12. Let be a lattice as above. Let b = min{n, d} and let M be a manifold of dimension less than b − 1, then any action on M is trivial. It is particularly bold to state this conjecture including lattices in Sp(1, n) and F4−20 . This is usually avoided because such lattices have abundant nonvolume preserving actions on certain types of highly regular fractals. I digress brieﬂy to explain why I believe the conjecture is plausible in that case by discussing how the actions on fractals arise. Namely such a lattice , being a hyperbolic group, has many inﬁnite proper quotients that are also hyperbolic groups [111]. If is a quotient of by an inﬁnite index, inﬁnite normal subgroup N, and is hyperbolic, then the boundary ∂ is a space with interesting dynamics and a good (Ahlfors regular) quasi-invariant measure class. However, ∂ is only a manifold when it is a sphere. It seems highly

groups acting on manifolds / 99

unlikely that this is ever the case for these groups and even more unlikely that this is ever the case with a smooth boundary action. The only known way to build a hyperbolic group that acts smoothly on its boundary is to have the group be the fundamental group of a compact negatively curved manifold M with smoothly varying horospheres. If smooth is taken to mean C ∞ , this then implies the manifold is locally symmetric [14] and then superrigidity results make it impossible for this to occur with π1 (M) a quotient of a lattice by an inﬁnite index inﬁnite normal subgroup. If smooth only means C 1 , then even in this context, no result known rules out M having fundamental group a quotient of a lattice in Sp(1, n) or F4−20 . All results one can prove using harmonic map techniques in this context only rule out M with non-positive complexiﬁed sectional curvature. We make the following conjecture, which is stronger than what is needed for Conjecture 4.12.

conjecture 4.13. Let be a cocompact lattice in Sp(1, n) or F4−20 and let be a quotient of that is Gromov hyperbolic. If ∂ is a sphere, then the kernel of the quotient map is ﬁnite and ∂ = ∂. For background on hyperbolic groups and their boundaries from a point of view relevant to this conjecture, see [133]. As a cautionary note, we point the reader to Subsection 11.3 where we recall a construction of Farrell and Lafont that shows that any Gromov hyperbolic group has an action by homeomorphisms on a sphere. The version of Zimmer’s conjecture that Zimmer made in [249] and [250] was only for volume preserving actions. Here we break it down somewhat explicitly to clarify the role of d and n.

conjecture 4.14. (zimmer) Let be a lattice as above and assume acts smoothly on a compact manifold M preserving a volume form. Then if dim (M) < d, the action is isometric. If, in addition, dim (M) < n or if is non-uniform, then the action is ﬁnite. Some ﬁrst remarks are in order. The cocycle super-rigidity theorems (discussed below) imply that, when the conditions of the conjecture hold, there is always a measurable invariant Riemannian metric. Also, the ﬁniteness under the conditions in the second half of the conjecture follow from cases of Margulis’s superrigidity theorem as soon as one knows that the action preserves a smooth Riemannian metric. So from one point of view, the conjecture is really about the regularity of the invariant metric.

100 / d. fisher

We should also mention that the conjecture is proven under several additional hypotheses by Zimmer around the time he made the conjecture. The ﬁrst example is the following:

theorem 4.15. Conjecture 4.14 holds provided the action also preserves a rigid geometric structure, for example, a torsion-free afﬁne connection or a pseudoRiemannian metric. This is proven in [246] for structures of ﬁnite type in the sense of Elie Cartan; see also [250]. The fact that roughly the same proof applies for rigid structures in the sense of Gromov was remarked in [85]. The point is simply that the isometry group of a rigid structure acts properly on some higher order frame bundle and that the existence of the measurable metric implies that has bounded orbits on all frame bundles as soon as has property (T ). This immediately implies that must be contained in a compact subgroup of the isometry group of the structure. Another easy version of the conjecture follows from the proof of Theorem 4.6. That is

theorem 4.16. Let G be a semisimple Lie group with property (T ) and < G a lattice. Let act on a compact manifold M by C 1 diffeomorphisms where dim (M) < d(G). If has a periodic orbit on S, then the action is ﬁnite. A more difﬁcult theorem of Zimmer shows that Conjecture 4.14 holds when the action is distal in a sense deﬁned in [247]. 4.6. Some Approaches to the Conjectures

4.6.1. discrete spectrum of actions. A measure preserving action of a group D on a ﬁnite measure space (X , μ) is said to have discrete spectrum if L2 (X , μ) splits as the sum of ﬁnite dimensional D invariant subspaces. This is a strong condition that is (quite formally) the opposite of weak mixing; for a detailed discussion see [92]. It is a theorem of Mackey (generalizing earlier results of Halmos and Von Neumann for D abelian) that an ergodic discrete spectrum D action is measurably isomorphic to one described by a dense embedding of D into a compact group K and considering a K action on a homogeneous K -space. The following remarkable result of Zimmer from [253] is perhaps the strongest evidence for Conjecture 4.14. This result is little known and has only recently been applied by other authors; see [83, 91].

groups acting on manifolds / 101

theorem 4.17. Let be a group with property (T ) acting by smooth, volume preserving diffeomorphisms on a compact manifold M. Assume in addition that preserves a measurable invariant metric. Then the action has discrete spectrum. This immediately implies that no counterexample to Conjecture 4.14 can be weak mixing or even admit a weak mixing measurable factor. The proof of the theorem involves constructing ﬁnite dimensional subspaces of L2 (M, ω) that are invariant. If enough of these subspaces could be shown to be spanned by smooth functions, one would have a proof of Conjecture 4.14. Here by “enough” we simply mean that it sufﬁces to have a collection of ﬁnite dimensional D invariant subspaces that separate points in M. These functions would then specify a D equivariant smooth embedding of M into RN for some large value of N. To construct ﬁnite dimensional invariant subspaces of L2 (M), Zimmer uses an approach similar to the proof of the Peter-Weyl theorem. Namely, he constructs invariant kernels on L2 (M × M) that are used to deﬁne selfadjoint, compact operators on L2 (M). The eigenspaces of these operators are then ﬁnite dimensional, invariant subspaces of L2 (M). The kernels should be thought of as functions of the distance to the diagonal in M × M. The main difﬁculty here is that for these to be invariant by “distance” we need to mean something deﬁned in terms of the measurable metric instead of a smooth one. The construction of the kernels in this setting is quite technical and we refer readers to the original paper. It would be interesting to try to combine the information garnered from this theorem with other approaches to Zimmer’s conjectures.

4.6.2. effective invariant metrics. We discuss here brieﬂy another approach to Zimmer’s conjecture, due to the author, which seems promising. We begin by brieﬂy recalling the construction of the space of “L2 metrics” on a manifold M. Given a volume form ω on M, we can consider the space of all (smooth) Riemannian metrics on M whose associated volume form is ω. This is the space of smooth sections of a bundle P → M. The ﬁber of P is X = SL(n, R)/SO(n). The bundle P is an associated bundle to the SL(n, R) sub-bundle of the frame bundle of M deﬁned by ω. The space X carries a natural SL(n, R)-invariant Riemannian metric of non-positive curvature; we denote its associated distance function by dX . This induces a natural notion of distance on the space of metrics, given by d(g1 , g2 )2 = M dX (g1 (m), g2 (m))2 dω. The completion of the sections with respect to the metric d will be denoted L2 (M, ω, X ); it is commonly referred to as the space of L2 metrics on M and its

102 / d. fisher

elements will be called L2 metrics on M. That this space is CAT (0) follows easily from the fact that X is CAT (0). For more discussion of X and its structure as a Hilbert manifold, see, for example, [79]. It is easy to check that a volume preserving action on M deﬁnes an isometric action on L2 (M, ω, X ). Given a generating set S for and a metric g in L2 (M, ω, X ), we write disp (g) = maxγ ∈S d(γ g, g). Given a group acting smoothly on M preserving ω, this gives an isometric action on L2 (M, ω, X ) that preserves the subset of smooth metrics. Let S be a generating set for . We deﬁne an operator P : L2 (M, ω, X ) → L2 (M, ω, X )

by taking a metric g to the barycenter of the measure S δ(γ g) . The ﬁrst observation is a consequence of the (standard, ﬁnite dimensional) implicit function theorem.

lemma 4.18. If g is a smooth metric, then Pg is also smooth. Moreover, we have the following two results.

theorem 4.19. Let M be a surface and a group with property (T ) and ﬁnite generating set S. Then there exists 0 < C < 1 such that the operator P satisﬁes: 1) disp (Pg) < C disp (g), and 2) for any g, the limn (P n g) exists and is invariant. A proof of this theorem can be given by using the standard construction of a negative deﬁnite kernel on H2 to produce a negative deﬁnite kernel on L2 (S, μ, H2 ). The theorem is then proved by transferring the ﬁrst property from the resulting action on a Hilbert space. The second property is an obvious consequence of the ﬁrst and completeness of the space L2 (M, ω, X ).

theorem 4.20. Let G be a semisimple Lie group all of whose simple factors have property (T ) and < G a lattice. Let M be a compact manifold such that dim (M) < d(G). Then the operator P on L2 (M, ω, X ) satisﬁes the conclusions of Theorem 4.19. This theorem is proven from results in [80] using convexity of the distance function on L2 (M, ω, X ). For cocompact lattices a proof can be given using results in [138] instead. The problem now reduces to estimating the behavior of the derivatives of n P g for some initial smooth g. This expression clearly involves derivatives of random products of elements of , that is, derivatives of elements of weighted by measures that are convolution powers of equidistributed measure on S. The main cause for optimism is that the fact that disp (P n g) is small

groups acting on manifolds / 103

immediately implies that the ﬁrst derivative of any γ ∈S must be small when measured at that point in L2 (M, ω, X ). One can then try to use estimates on compositions of diffeomorphisms and convexity of derivatives to control derivatives of P n g. The key difﬁculty is that the initial estimate on the ﬁrst derivative of γ applied to P n g is only small in an L2 sense. 4.7. Some Related Questions and Conjectures on Countable Subgroups of Diff (M)

In this subsection, we discuss related conjectures and results on countable subgroups of Diff (M). All of these are motivated by the belief that countable subgroups of Diff (M) are quite special, though considerably less special than, say, linear groups. We defer positive constructions of non-linear subgroups of Diff (M) to Section 11 and a discussion of possible algebraic and geometric properties shared by all ﬁnitely generated subgroups of Diff (M) to Section 13. Here we concentrate on groups that do not act on manifolds, either by theorems or conjecturally.

4.7.1. groups with property (T ) and generic groups. We begin by focusing on actions of groups with property (T ). For a ﬁnitely generated group , we recall that d() is the smallest dimension in which admits an inﬁnite image linear representation. We then make the following conjecture: conjecture 4.21. Let be a group with property (T ) acting smoothly on a compact manifold M, preserving volume. Then if dim (M) < d(), the action preserves a smooth Riemannian metric. For many groups with property (T ), one can produce a measurable invariant metric; see [83]. In fact, in [83], Silberman and the author prove that there are many groups with property (T ) with no volume preserving actions on compact manifolds. Key steps include ﬁnding the invariant measurable metric, applying Zimmer’s theorem on discrete spectrum from Section 4.6, and producing groups with no ﬁnite quotients and so no linear representations at all. A result of Furman announced in [90] provides some further evidence for the conjecture. This result is analogous to Proposition 5.1 and shows that any action of a group with property (T ) either leaves invariant a measurable metric or has positive random entropy. We refer the reader to [90] for more discussion. While the proof of this result is not contained in [90], it is possible to reconstruct it from results there and others in [91]. Conjecture 4.21 and the work in [83] are motivated in part by the following conjecture of Gromov:

104 / d. fisher

conjecture 4.22. There exists a model for random groups in which a “generic” random group admits no smooth actions on compact manifolds. It seems quite likely that the conjecture could be true for random groups in the density model with density more than 13 . These groups have property (T ); see Ollivier’s book [193] for a discussion on random groups. For the conjecture to be literally true would require that a random hyperbolic group have no ﬁnite quotients and it is a well-known question to determine whether there are any hyperbolic groups that are not residually ﬁnite, let alone generic ones. If one is satisﬁed by saying the generic random group has only ﬁnite smooth actions on compact manifolds, one can avoid this well-known open question.

4.7.2. universal lattices. Recently, Shalom has proven that the groups SL(n, Z[X ]) have property (T ) when n > 3. (As well as some more general results.) Shalom refers to these groups as universal lattices. An interesting and approachable question is question 4.23. Let be a ﬁnite index subgroup in SL(n, Z[X ]) and let acting by diffeomorphisms on a compact manifold M preserving volume. If dim (M) < n, is there a measurable invariant metric? If one gives a positive answer to this question, one is then clearly interested in whether or not the metric can be chosen to be smooth. It is possible that this is easier for these “larger” groups than for the lattices originally considered by Zimmer. One can ask a number of variants of this question including trying to prove a full cocycle super-rigidity theorem for these groups; see below. The question just asked is particularly appealing as it can be viewed as a ﬁxed point problem for the action on the space of metrics (see Section 4.6 for a deﬁnition). In fact, one knows one has the invariant metric for the action of any conjugate of SL(n, Z) and only needs to show that there is a consistent choice of invariant metrics over all conjugates. One might try to mimic the approach from [216], though a difﬁculty clearly arises at the point where Shalom applies a scaling limit construction. Scaling limits of L2 (M, ω, X ) can be described using nonstandard analysis, but are quite complicated objects and not usually isomorphic to the original space.

4.7.3. irreducible actions of products. Another interesting variant on Zimmer’s conjecture is introduced in [91]. In that paper Furman and Monod study obstructions to irreducible actions of product groups. A measure

groups acting on manifolds / 105

preserving action of a product 1 × 2 on a measure space (X , μ) is said to be irreducible if both 1 and 2 act ergodically on X . Furman and Monod produce many obstructions to irreducible actions; for example, one can prove from their results that

theorem 4.24. Let = 1 × 2 × 3 . Assume that 1 has property (T ) and no unbounded linear representations. Then there are no irreducible, volume preserving actions on compact manifolds. The same is true for = 1 × 2 if 1 is as above and 2 is solvable. A key step in the proof of this result is to use Zimmer’s result on actions with discrete spectrum to show that the action is measurably conjugate to an irreducible isometric action. In particular, the second conclusion follows since no compact group can contain simultaneously a dense (T ) group and a dense solvable group. This motivates the following conjecture.

conjecture 4.25. Let = 1 × 2 . Assume that 1 has property (T ) and no unbounded linear representations and that 2 is amenable. Then there are no irreducible, volume preserving actions on compact manifolds. One might be tempted to prove this by showing that no compact group contains both a dense ﬁnitely generated amenable group and a dense Kazhdan group. This is, however, not true. The question was raised by Lubotzky in [149] but has recently been resolved in the negative by Kassabov [127]. In the context of irreducible actions, asking that groups have property (T ) is perhaps too strong. In fact, there are already relatively few known irreducible actions of Z2 = Z × Z! If one element of Z2 acts as an Anosov diffeomorphism, then it is conjectured by Katok and Spatzier that all actions are algebraic [130]. Even in more general settings where one assumes non-uniform hyperbolicity, there are now hints that a classiﬁcation might be possible, though clearly more complicated than in the Anosov case; see [125] and references there. In a sense, that paper indicates that the “exotic examples” that might arise in this context may be no worse than those that arise for actions of higher rank lattices.

4.7.4. torsion groups and homeo (m). A completely different and wellstudied aspect of the theory of transformation groups of compact manifolds is the study of ﬁnite subgroups of Homeo (M). It has recently been noted by many authors that this study has applications to the study of “large subgroups” of Homeo (M). Namely, one can produce many ﬁnitely generated

106 / d. fisher

and even ﬁnitely presented groups that have no non-trivial homomorphisms to Homeo (M) for any compact M. This is discussed in, for example, [20, 83, 234]. As far as I know, this observation was ﬁrst made by Ghys as a remark in the introduction to [100]. In all cases, the main trick is to construct inﬁnite, ﬁnitely generated groups that contain inﬁnitely many conjugacy classes of ﬁnite subgroups, usually just copies of (Z/pZ)k for all k > 1. A new method of constructing such groups was recently introduced by Chatterji and Kassabov [38]. As far as the author knows, even if one ﬁxes M in advance, this is the only existing method for producing groups with no non-trivial (or even no inﬁnite image) homomorphisms to Homeo (M) unless M = S1 . For M = S1 we refer back to Subsection 4.1 and to [173].

5. Cocycle Super-rigidity and Immediate Consequences 5.1. Zimmer’s Cocycle Super-rigidity Theorem and Generalizations

A main impetus for studying rigidity of group actions on manifolds came from Zimmer’s theorem on superrigidity for cocycles. This theorem and its proof were strongly motivated by Margulis’s work. In fact, Margulis’s theorem reduces to the special case of Zimmer’s theorem for a certain cocycle α : G × G/ → . In order to avoid technicalities, we describe only a special case of this result, essentially avoiding boundedness and integrability assumptions on cocycles that are automatically fulﬁlled in any context arising from a continuous action on a compact manifold. Let M be a compact manifold, H a matrix group, and P an H bundle over M. For readers not familiar with bundle theory, the results are interesting even in the case where P = M × H. Now let a group act on P continuously by bundle automorphisms, that is, such that there is a action on M for which the projection from P to M is equivariant. Further assume that the action on M is measure preserving and ergodic. The cocycle superrigidity theorem says that if is a lattice in a simply connected, semisimple Lie group G all of whose simple factors are noncompact and have property (T ), then there is a measurable map s : M → H, a representation π : G → H, a compact subgroup K < H that commutes with π (G), and a measurable map × M → K such that 5.1

γ ·s(m) = k(m, γ )π (γ )s(γ ·m).

It is easy to check from this equation that the map K satisﬁes the equation that makes it into a cocycle over the action of . One should view s as providing coordinates on P in which the action is almost a product. For more

groups acting on manifolds / 107

discussion of this theorem, particularly in the case where all simple factors of G have higher rank, the reader should see any of [65, 66, 81, 95, 245]. (The version stated here is only proven in [81]; previous proofs all yielded somewhat more complicated statements that require passing to ﬁnite ergodic extensions of the action.) For the case of G with simple factors of the form Sp(1, n) and F4−20 , the results follow from work of the author and Hitchman [80], building on earlier results of Korevaar-Schoen [138, 139, 140] and CorletteZimmer [41]. As a sample application, let M = Tn and let P be the frame bundle of M, that is, the space of frames in the tangent bundle of M. Since Tn is parallelizable, we have P = Tn × GL(n, R). The cocycle super-rigidity theorem then says that “up to compact noise” the derivative of any measure preserving action on Tn looks measurably like a constant linear map. In fact, the cocycle superrigidity theorems apply more generally to continuous actions on any principal bundle P over M with ﬁber H, an algebraic group, and in this context produces a measurable section s : M → P satisfying equation (5.1). So in fact, cocycle superrigidity implies that for any action preserving a ﬁnite measure on any manifold the derivative cocycle looks measurably like a constant cocycle, up to compact noise. That cocycle superrigidity provides information about actions of groups on manifolds through the derivative cocycle was ﬁrst observed by Furstenberg in [95]. Zimmer originally proved cocycle superrigidity in order to study orbit equivalence of group actions. For recent surveys of subsequent developments concerning orbit equivalence rigidity and other forms of superrigidity for cocycles, see [89, 197, 215]. 5.2. First Applications to Group Actions and the Problem of Regularity

The following result, ﬁrst observed by Furstenberg, is an immediate consequence of cocycle superrigidity.

proposition 5.1. Let G be a semisimple Lie group with no compact factors and with property (T ) of Kazhdan and let < G be a lattice. Assume G or acts by volume preserving diffeomorphisms on a compact manifold M. Then there is a linear representation π : G → GL( dim (M), R) such that the Lyapunov exponents of g∈G or are exactly the absolute values of the eigenvalues of π (g). This proposition is most striking to those familiar with classical dynamics where the problem of estimating, let alone computing, Lyapunov exponents is quite difﬁcult. Another immediate consequence of cocycle superrigidity is the following:

108 / d. fisher

proposition 5.2. Let G or be as above, acting by smooth diffeomorphisms on a compact manifold M preserving volume. If every element of G or acts with zero entropy, then there is a measurable invariant metric on M. In particular, if G admits no nontrivial representations of dim (M), then there is a measurable invariant metric on M. As mentioned above in Subsection 4.5, this reduces Conjecture 4.14 to the following technical conjecture.

conjecture 5.3. Let G or acting on M be as above. If G or preserves a measurable Riemannian metric, then they preserve a smooth-invariant Riemannian metric. We recall that ﬁrst evidence toward this conjecture is Theorem 4.15. We remark that in the proof of that theorem, it is not the case that the measurable metric from Proposition 5.2 is shown to be smooth, but instead it is shown that the image of in Diff (M) lies in a compact subgroup. In other work, Zimmer explicitly improves regularity of the invariant metric; see particularly [247]. In general the problem that arises immediately from cocycle superrigidity in any context is understanding the regularity of the straightening section σ . This question has been studied from many points of view, but still relatively little is known in general. For certain examples of actions of higher rank lattices on compact manifolds, discussed below in Section 9, the σ that straightens the derivative cocycle cannot be made smooth on all of M. It is possible that for volume preserving actions on manifolds and the derivative cocycle, σ can always be chosen to be smooth on a dense open set of full measure. We will return to the theme of regularity of the straightening section in Section 7. First we turn to more geometric contexts in which the output of cocycle superrigidity is also often used but usually more indirectly.

6. Geometric Actions and Rigid Geometric Structures In this section, we discuss the role of rigid geometric structures in the study of actions of large groups. The notion of rigid geometric structure was introduced by Gromov, partially in reaction to Zimmer’s work on large group actions. The ﬁrst subsection of this section recalls the deﬁnition of rigid geometric structure, gives some examples, and explains the relation of Gromov’s

groups acting on manifolds / 109

rigid geometric structures to other notions introduced by Cartan. Subsection 6.2 recalls Gromov’s initial results relating actions of simple groups preserving rigid geometric structures on M to representations of the fundamental group of M, and extensions of these results to lattice actions due to Zimmer and the author. Subsection 6.3 concerns a different topic, namely the rigidity of connection preserving actions of a lattice , particularly on manifolds of dimension not much larger than d(G) as deﬁned in Section 4. Subsection 6.4 recalls some obstructions to actions preserving geometric structures, particularly a result known as Zimmer’s geometric Borel density theorem and some recent related results of Bader, Frances, and Melnick. We discuss actions preserving a complex structure in Subsection 6.5 and end with a subsection 6.6 on questions concerning geometric actions. 6.1. Rigid Structures and Structures of Finite Type

In this subsection we recall the formal deﬁnition of a rigid geometric structure. Since the deﬁnition is somewhat technical, some readers may prefer to skip it and read on keeping in mind examples rather than the general notion. The basic examples of rigid geometric structures are Riemannian metrics, pseudoRiemannian metrics, conformal structures deﬁned by either type of metric, and afﬁne or projective connections. Basic examples of geometric structures that are not rigid are a volume form or symplectic structure. An intermediate type of structure that exhibits some rigidity but that is not literally rigid in the sense discussed here is a complex structure. If N is a manifold, we denote the kth order frame bundle of N by F k (N), and by J s,k (N) the bundle of k-jets at 0 of maps from Rs to N. If N and N are two manifolds, and f : N → N is a map between them, then the k-jet j k ( f ) induces a map J s,k N → J s,k N for all s. We let Dk (N) be the bundle whose ﬁber Dpk at a point p consists of the set of k-jets at p of germs of diffeomorphisms of N ﬁxing p. We abbreviate D0k (Rn ) by Dnk or simply Dk ; this is a real algebraic group. For concreteness, one can represent each element uniquely, in terms of standard coordinates (ξ1 , . . . , ξn ) on Rn , in the form (P1 (ξ1 , . . . , ξn ), . . . , Pn (ξ1 , . . . , ξn )) where P1 , P2 , . . . , Pn are polynomials of degree ≤ k. We denote the vector space of such polynomial maps of degree ≤ k by Pn,k . The group Dnk has a natural action on F k (N), where n is the dimension of N. Suppose we are given an algebraic action of Dnk on a smooth algebraic variety Z. Then following Gromov [112], we make the following deﬁnition:

110 / d. fisher

definition 6.1. 1) An A-structure on N (of order k, of type Z) is a smooth map φ : F k (N) → Z equivariant for the Dnk actions. 2) With notation as above, the rth prolongation of φ, denoted φ r , is the map φ r : F k+r (N) → J n,r (Z) deﬁned by φ r = j r (φ) ◦ ιr+k where ιr+k : k k k+r n,r k k n,r k F (N) → J (F (N)) is the natural inclusion and j (h) : J (F (N)) → J n,r (Z) is as before; this is an A-structure of type J n,r (Z) and order k + r. Equivalently, an A-structure of type Z and order k is a smooth section of the associated bundle F k (N) ×Dk Z over N. Note that an A-structure on N deﬁnes by restriction an A-structure φ|U on any open set U ⊂ N.

remark 6.2. A-structures were introduced in [112]; a good introduction to the subject, with many examples, can be found in [10]. A comprehensive and accessible discussion of the results of [112] concerning actions of simple Lie groups can be found in [67]. Note that if N and N are n-manifolds, and h : N → N is a diffeomorphism, then h induces a bundle map j k (h) : F k (N) → F k (N ).

definition 6.3. 1) If φ : F k (N) → Z, φ : F k (N ) → Z are A-structures, a diffeomorphism h : N → N is an isometry from φ to φ if φ ◦ j k (h) = φ. 2) A local isometry of φ is a diffeomorphism h : U1 → U2 , for open sets U1 , U2 ⊂ N, which is an isometry from φ|U1 to φ|U2 . For p ∈ M denote by Isloc p (φ) the pseudogroup of local isometries of φ ﬁxing p, and, for l ≥ k, we denote by Islp (φ) the set of elements jpl (h) ∈ Dpl such that jpl (φ ◦ jpk (h)) = φ l−k , where both sides are considered as maps F k+l (N) → J l−k (Z). Note that Islp (φ) is a group, and there is a natural homomorphism rpl;m : Islp (φ) → Ism p (φ) for m < l; in general, it is neither injective nor surjective.

definition 6.4. The structure φ is called k-rigid if for every point p, the map rpk+1;k is injective. A ﬁrst remark worth making is that an afﬁne connection is a rigid geometric structure and that any generalized quasi-afﬁne action on a manifold of the form K \H/ × M with discrete preserves an afﬁne connection and

groups acting on manifolds / 111

therefore also a torsion free afﬁne connection. In order to provide some more examples, we recall the following lemma of Gromov.

lemma 6.5. Let V be an algebraic variety and G a group acting algebraically on V . For every k, there is a tautological G invariant geometric structure of order k on V , given by ω : P k (V ) → P k (V )/G. This structure is rigid if and only if the action of G on P k (V ) is free and proper. The conclusion in the ﬁrst sentence is obvious. The second sentence is proven in section 0.4, pages 69–70, of [112].

examples 1) The action of G = SLn (R) on Rn is algebraic. So is the action of G on the manifold N1 obtained by blowing up the origin. The reader can easily verify that the action of G on P 2 (N1 ) is free and proper. 2) We can compactify N1 by N2 by viewing the complement of the blow up as a subset of the projective space P n . Another description of the same action, which may make the rigid structure more visible to the naked eye, is as follows: SLn+1 (R) acts on P n . Let G be SLn (R) < SLn+1 (R) as block diagonal matrices with blocks of size n and 1 and a 1 × 1 block equal to 1. Then G acts on P n ﬁxing a point p. We can obtain N2 by blowing up the ﬁxed point p. The G actions on both P n and N2 are algebraic and again the reader can verify that the action is free and proper on P 2 (N2 ). 3) In the construction from (2) above, there is an action of a group H where H = SLn (R) Rn and G = SLn (R) < SLn (R) Rn < SLn+1 (R). The H action ﬁxes the point p and so also acts on N2 algebraically. However, over the exceptional divisor, the action is never free on any frame bundle, since the subgroup Rn acts trivially to all orders at the exceptional divisor. The behavior in example (3) above illustrates the fact that existence of invariant rigid structures is more complicated for algebraic groups that are not semisimple; see the discussion in section 0.4.C. of [112].

definition 6.6. A rigid geometric structure is called a ﬁnite type geometric structure if V is a homogeneous Dnk space. This is by no means the original deﬁnition that is due to Cartan and predates Gromov’s notion of a rigid geometric structure by several decades. It

112 / d. fisher

is equivalent to Cartan’s deﬁnition of a ﬁnite type structure by the work of Candel and Quiroga-Barranco [31, 32, 201]. Candel and Quiroga-Barranco also give a development of rigid geometric structures that more closely parallels the older notion of a structure of ﬁnite type as presented in, for example, [135]. All standard examples of rigid geometric structures are structures of ﬁnite type. However, example (2) above is not a structure of ﬁnite type. The notion of structure of ﬁnite type was ﬁrst given in terms of prolongations of Lie algebras and so yields a criterion that is, in principle, computable. The work of Candel and Quiroga-Barranco extends this computable nature to general rigid geometric structures. 6.2. Rigid Structures and Representations of Fundamental Groups

In this subsection, we restrict our attention to actions of simple Lie groups G and lattices < G. Many of the results of this section extend to semisimple G, though the formulations become more complicated. A major impetus for Gromov’s introduction of rigid geometric structures is the following theorem from [112].

theorem 6.7. Let G be a simple noncompact Lie group and M a compact real analytic manifold. Assume G acts on M preserving an analytic rigid geometric structure ω and a volume form ν. Further assume the action is ergodic. Then there is a linear representation ρ : π1 (M) → GL(n, R) such that the Zariski closure of ρ(π1 (M)) contains a group locally isomorphic to G. We remark that, by the Tits’s alternative, the theorem immediately implies that actions of the type described cannot occur on manifolds with amenable fundamental group or even on manifolds whose fundamental groups do not contain a free group on two generators [229]. Expository accounts of the proof can be found in [67, 255, 258]. The representation ρ is actually deﬁned on ˜ Killing ﬁelds of the lift ω˜ of ω to M. This is a worthwhile moment to indicate the weakness of Principle 2.6. If we start with a cocompact lattice < G and an action of on a compact manifold M satisfying the assumptions of Theorem 6.7, we can induce the action to a G action on the compact manifold N = (G × M)/ . However, in this setting, there is an obvious representation of π1 (N) satisfying the conclusion of Theorem 6.7. This is because there is a surjection π1 (N) → . In fact for most cocompact lattices < G there are homomorphisms σ : → K where K is compact and simply connected so that we can have act on K by

groups acting on manifolds / 113

left translation and obtain examples where π1 (N) = . For G of higher rank, the following theorem of Zimmer and the author shows that this is the only obstruction to a variant of Theorem 6.7 for lattices.

theorem 6.8. Let < G be a lattice, where G is a simple group and R − rank(G) ≥ 2. Suppose acts analytically and ergodically on a compact manifold M preserving a unimodular rigid geometric structure. Then either 1) the action is isometric and M = K /C where K is a compact Lie group and the action is by right translation via ρ : → K , a dense image homomorphism, or 2) there exists an inﬁnite image linear representation σ : π1 (M) → GLn R, such that the algebraic automorphism group of the Zariski closure of σ (π1 (M)) contains a group locally isomorphic to G. The proof of this theorem makes fundamental use of a notion that does not occur in its statement: the entropy of the action. A key observation is the following formula for the entropy of the restriction of the induced action to : h(G×M)/ (γ ) = hG/ (γ ) + hM (γ ). We then use the fact that the last term on the right hand side is 0 if and only if the action preserves a measurable Riemannian metric as already explained in Proposition 5.1. In this setting, it then follows from Theorem 4.15 that the action is isometric and the other conclusions in (1) are simple consequences of ergodicity of the action. When hM (γ ) > 0 for some , we use relations between the entropy of the action and the Gromov representation discovered by Zimmer [257]. We discuss some of these ideas below in Section 8. 6.3. Afﬁne Actions Close to the Critical Dimension

In this section we brieﬂy describe a direction pursued by Feres, Goetze, Zeghib, and Zimmer that classify connection preserving actions of lattices in dimensions close to, but not less than, the critical dimension d. A representative result is the following:

theorem 6.9. Let n > 2, G = SL(n, R) and < G a lattice. Let M be a compact manifold with dim (M) ≤ n + 1 = d(G) + 1. Assume that acts on M preserving a volume form and an afﬁne connection. Then either

114 / d. fisher

1) dim (M) < n and the action is isometric, 2) dim (M) = n and either the action is isometric or, upon passing to ﬁnite covers and ﬁnite index subgroups, smoothly conjugate to the standard SL(n, Z) action on Tn , or 3) dim (M) = n + 1 and either the action is isometric or, upon passing to ﬁnite covers and ﬁnite index subgroups, smoothly conjugate to the action of SL(n, Z) on Tn+1 = Tn × T where the action on the second factor is trivial. Variants of this theorem under more restrictive hypotheses were obtained by Feres [63], Goetze [104], and Zimmer [248]. The theorem as stated is due to Zeghib [243]. The cocycle superrigidity theorem is used in all of the proofs, mainly to force vanishing of curvature and torsion tensors of the associated connection. One would expect similar results for a more general class of acting groups and also for a wider range of dimensions. Namely, if we let d2 (G) be the dimension of the second non-trivial representation of G, we would expect a similar result to (3) for any afﬁne, volume preserving action on a manifold M with d(G) < d2 (G). With the further assumption that the connection is Riemannian (and still only for lattices in SL(n, R)) this is proven by Zeghib in [242]. Further related results are contained in other papers of Feres [60, 64]. No results of this kind are known for dim (M) ≥ d2 (G). A major difﬁculty arises as soon as one has dim (M) ≥ d2 (G), namely that more complicated examples can arise, including afﬁne actions on nilmanifolds and left translation actions on spaces of the form G/. All the results mentioned in this subsection depend on the particular fact that ﬂat Riemannian manifolds have ﬁnite covers that are tori. 6.4. Zimmer’s Borel Density Theorem and Generalizations

An obvious ﬁrst question concerning G actions is the structure of the stabilizer subgroups. In this direction we have

theorem 6.10. Let G be a simple Lie group acting essentially faithfully on a compact manifold M preserving a volume form, then the stabilizer of almost every point is discrete. As an application of this result, one can prove the following:

theorem 6.11. Let G be a simple Lie group acting essentially faithfully on a compact manifold, preserving a volume form and a homogeneous geometric structure with structure group H. Then there is an inclusion g → h.

groups acting on manifolds / 115

In particular, the theorem provides an obstruction to a higher rank simple Lie group having a volume preserving action on a compact Lorentz manifold. For more discussion of these theorems, see [258]. This observation led to a major series of works studying automorphism groups of Lorentz manifolds; see, for example, [1, 141, 240, 241]. We remark here that Theorem 6.11 does not require that the homogeneous structure be rigid. More recently some closely related phenomena have been discovered in a joint work of Bader, Frances, and Melnick [5]. Their work uses yet another notion of a geometric structure, that of a Cartan geometry. We will not deﬁne this notion rigorously here; it sufﬁces to note that any rigid homogeneous geometric structure deﬁnes a Cartan geometry, as discussed in [5, introduction]. The converse is not completely clear in general, but most classical examples of rigid geometric structures can also be realized as Cartan geometries. A Cartan geometry is essentially a way of saying that a manifold is inﬁnitesimally modeled on some homogeneous space G/P. To recapture the notion of a Riemannian connection, G = O(n) Rn and P = O(n), to recapture the notion of an afﬁne connection G = Gl(n, R) Rn and P = Gl(n, R). Cartan connections come with naturally deﬁned curvatures that vanish if and only if the manifold is locally modeled on G/P. We refer the reader to the book [217] for a general discussion of Cartan geometries and their use in differential geometry. Given a connected linear group L, we deﬁne its real rank, rk(L) as before to be the dimension of a maximal R-diagonalizable subgroup of L, and let n(L) denote the maximal nilpotence degree of a connected nilpotent subgroup. One of the main results of [5] is

theorem 6.12. If a group L acts by automorphisms of a compact Cartan geometry modeled on G/P, then 1) rk(AdL0 ) ≤ rk(Adg P), 2) n(AdL0 ) ≤ n(Adg P). The main point is that L is not assumed either simple or connected. This theorem is deduced from an embedding theorem similar in ﬂavor to Theorem 6.10. In addition, when G is a simple group and P is a maximal parabolic subgroup, Bader, Frances, and Melnick prove rigidity results classifying all possible actions when the rank bound in Theorem 6.12 is achieved. This classiﬁcation essentially says that all examples are algebraic; see [5] for a detailed discussion.

116 / d. fisher

An earlier paper by Feres and Lampe also explored applications of Cartan geometries to rigidity and dynamical conditions for ﬂatness of Cartan geometries [62]. 6.5. Actions Preserving a Complex Structure

We mention here one recent result by Cantat and a few active related directions of research. In [250], Zimmer asked whether one had restrictions on low dimensional holomorphic actions of lattices. The answer was obtained in [36] and is

theorem 6.13. Let G be a connected simple Lie group of real rank at least 2 and suppose < G is a lattice. If admits a faithful action by automorphisms on a compact Kähler manifold M, then the rank of G is at least the complex dimension of M. The proof of the theorem depends primarily on results concerning Aut (M), particularly recent results on holomorphic actions of abelian groups by Dinh and Sibony [47]. It is worth noting that while the theorem does depend on Margulis’s superrigidity theorem, it does not depend on Zimmer’s cocycle superrigidity theorem. The condition that the action is faithful can be weakened considerably using Margulis’s normal subgroups theorem. More recently, Cantat [35], Déserti [46], and others have begun a program of studying large subgroups of automorphism groups of complex manifolds. In particular, Cantat has proven an analogue of Conjecture 4.7 in the context of birational actions on complex surfaces.

theorem 6.14. Let S be a compact Kähler surface and G an inﬁnite, countable group of birational transformations of S. If G has property (T ), then there is a birational map j : S → P2 (C) that conjugates G to a subgroup of Aut (P2 (C)). 6.6. Conjectures and Questions

A major open problem, generalizing Conjecture 1.2, is the following:

conjecture 6.15. (gromov-zimmer) Let D be a semisimple Lie group with all factors having property (T ) or a lattice in such a Lie group. Then any D action on a compact manifold preserving a rigid geometric structure and a volume form is generalized quasi-afﬁne. A word is required on the attribution. Both Gromov [112] and Zimmer [249] made various less precise conjectures concerning the classiﬁcation of actions

groups acting on manifolds / 117

as in the conjecture. The exact statement of the correct conjecture was muddy for several years while it was not known if the Katok-Lewis and Benveniste type examples admitted invariant rigid geometric structures. See Section 9 for more discussion of these examples. In the context of the results of [13], where Benveniste and the author prove that those actions do not preserve rigid geometric structures, this version of the classiﬁcation seems quite plausible. It is perhaps more plausible if one assumes the action is ergodic or that the geometric structure is homogeneous. A possibly easier question that is relevant is the following:

question 6.16. Let M be a compact manifold equipped with a homogeneous rigid geometric structure ω. Assume Aut (M, ω) is ergodic or has a dense orbit. Is M locally homogeneous? Gromov’s theorem on the open-dense implies that a dense open set in M is locally homogeneous even if ω is not homogeneous. The question is whether this homogeneous structure extends to all of M if ω is homogeneous. If ω is not homogeneous, the examples following Lemma 6.5 show that M need not be homogeneous. It seems possible to approach the case of Conjecture 6.15 where ω is homogeneous and D acts ergodically by answering Question 6.16 positively. At this point, one is left with the problem of classifying actions of higher rank groups and lattices on locally homogeneous manifolds. Induction easily reduces one to considering G actions. The techniques Gromov uses to produce the locally homogeneous structure gives slightly more precise information in this setting: one is left trying to classify homogeneous manifolds modeled on H/L where G acts via an inclusion in H centralizing L. In fact this reduces one to problems about locally homogeneous manifolds studied by Zimmer and collaborators; see particularly [142, 143, 145, 256]. For a more general survey of locally homogeneous spaces, see also [136] and [137]. The following question concerns a possible connection between preserving a rigid geometric structure and having uniformly partially hyperbolic dynamics.

question 6.17. Let D be any group acting on a compact manifold M preserving a volume form and a rigid geometric structure. Is it true that if some element d of D has positive Lyapunov exponents, then d is uniformly partially hyperbolic?

118 / d. fisher

This question is of particular interest for us when D is a semisimple Lie group or a lattice, but would be interesting to resolve in general. The main reason to believe that the answer might be yes is that the action of D on the space of frames of M is proper. Having a proper action on a tangent bundle minus the zero section implies that an action is Anosov; see Mañé’s article for a proof [156].

7. Topological Super-rigidity: Regularity Results from Hyperbolic Dynamics and Applications A major area of research in the Zimmer program has been the application of hyperbolic dynamics. This area might be described by the maxim: in the presence of hyperbolic dynamics, the straightening section from cocycle superrigidity is often more regular. Some major successes in this direction are the work of Goetze-Spatzier and Feres-Labourie. In the context of hyperbolic dynamical approaches, there are two main settings. In the ﬁrst, one considers actions of higher rank lattices on a particular class of compact manifolds and a second in which one makes no assumption on the topology of the manifold acted upon. In this section we discuss a few such results, after ﬁrst recalling some facts about the stability of hyperbolic dynamical systems in Subsection 7.1. We then discuss best-known rigidity results for actions on tori in Subsection 7.2 and after this we discuss the work of Goetze-Spatzier and Feres-Labourie in the more general context in Subsection 7.3. The term “topological superrigidity” for this area of research was coined by Zimmer, whose early unpublished notes on the topic dramatically inﬂuenced research in the area [244]. This aspect of the Zimmer program has also given rise to a study of rigidity properties for other uniformly hyperbolic actions of large groups, most particularly higher rank abelian groups. See, for example, [126, 130, 209]. The following subsection recalls basic notions from hyperbolic dynamics that are needed in this section. 7.1. Stability in Hyperbolic Dynamics

A diffeomorphism f of a manifold X is said to be Anosov if there exists a continuous f invariant splitting of the tangent bundle TX = Efu ⊕ Efs and constants a > 1 and C, C > 0 such that for every x∈X , 1) Df n (v u )≥Can v u for all v u ∈Efu (x), and 2) Df n (v s )≤C a−n v s for all v s ∈Efs (x).

groups acting on manifolds / 119

Note that the constants C and C depend on the choice of metric, and that a metric can always be chosen so that C = C = 1. There is an analogous notion for a ﬂow ft , where TX = T O ⊕ Efut ⊕ Efst where T O is the tangent space to the ﬂow direction and vectors in Efut (respectively, Efst ) are uniformly expanded (respectively, uniformly contracted) by the ﬂow. This notion was introduced by Anosov [3] and named after Anosov by Smale [221], who popularized the notion in the United States. One of the earliest results in the subject is Anosov’s proof that Anosov diffeomorphisms are structurally stable, that is, that any C 1 perturbation of an Anosov diffeomorphism is conjugate back to the original diffeomorphism by a homeomorphism. There is an analogous result for ﬂows, though this requires that one introduce a notion of time change that we will not consider here. Since Anosov also showed that C 2 volume preserving Anosov ﬂows and diffeomorphisms are ergodic, structural stability implies the existence of an open set of “chaotic” dynamical systems. The notion of an Anosov diffeomorphism has had many interesting generalizations, for example: Axiom A diffeomorphisms, non-uniformly hyperbolic diffeomorphisms, and diffeomorphisms admitting a dominated splitting. A notion that has been particularly useful in the study of rigidity of group actions is the notion of a partially hyperbolic diffeomorphism as introduced by Hirsch, Pugh, and Shub [177]. Under strong enough hypotheses, these diffeomorphisms have a weaker stability property similar to structural stability. More or less, the diffeomorphisms are hyperbolic relative to some foliation, and any nearby action is hyperbolic to some nearby foliation. To describe more precisely the class of diffeomorphisms considered and the stability property they enjoy, we require some deﬁnitions. The use of the word “foliation” varies with context. Here a foliation by C k leaves will be a continuous foliation whose leaves are C k injectively immersed submanifolds that vary continuously in the C k topology in the transverse direction. To specify transverse regularity we will say that a foliation is transversely C r . A foliation by C k leaves that is tranversely C k is called simply a C k foliation. (Note our language does not agree with that in the reference [118].) Given an automorphism f of a vector bundle E → X and constants a > b ≥ 1, we say f is (a, b)-partially hyperbolic or simply partially hyperbolic if there is a metric on E, a constant and C ≥ 1, and a continuous f invariant, nontrivial splitting E = Efu ⊕ Efc ⊕ Efs such that for every x in X : 1) f n (v u )≥Can v u for all v u ∈Efu (x),

2) f n (v s )≤C −1 a−n v s for all v s ∈Efs (x), and

3) C −1 b−n v 0 < f n (v 0 )≤Cbn v 0 for all v 0 ∈Efc (x) and all integers n.

120 / d. fisher

A C 1 diffeomorphism f of a manifold X is (a, b)-partially hyperbolic if the derivative action Df is (a, b)-partially hyperbolic on TX . We remark that for any partially hyperbolic diffeomorphism, there always exists an adapted metric for which C = 1. Note that Efc is called the central distribution of f , Efu is called the unstable distribution of f , and Efs the stable distribution of f . Integrability of various distributions for partially hyperbolic dynamical systems is the subject of much research. The stable and unstable distributions are always tangent to invariant foliations that we call the stable and unstable foliations and denote by Wfs and Wfu , respectively. If the central distribution is tangent to an f invariant foliation, we call that foliation a central foliation and denote it by Wfc . If there is a unique foliation tangent to the central distribution, we call the central distribution uniquely integrable. For smooth distributions unique integrability is a consequence of integrability, but the central distribution is usually not smooth. If the central distribution of an (a, b)-partially hyperbolic diffeomorphism f is tangent to an invariant foliation Wfc , then we say f is r-normally hyperbolic to Wfc for any r such that a > br . This is a special case of the deﬁnition of r-normally hyperbolic given in [118]. Before stating a version of one of the main results of [118], we need one more deﬁnition. Given a group G, a manifold X , two foliations F and F of X , and two actions ρ and ρ of G on X , such that ρ preserves F and ρ preserves F , following [118] we call ρ and ρ leaf conjugate if there is a homeomorphism h of X such that 1) h(F) = F, and 2) for every leaf L of F and every g∈G, we have h(ρ(g)L) = ρ (g)h(L). The map h is then referred to as a leaf conjugacy between (X , F, ρ) and (X , F , ρ ). This essentially means that the actions are conjugate modulo the central foliations. We state a special case of some of the results of Hirsch-Pugh-Shub on perturbations of partially hyperbolic actions of Z; see [118]. There are also analogous deﬁnitions and results for ﬂows. As these are less important in the study of rigidity, we do not discuss them here.

theorem 7.1. Let f be an (a, b)-partially hyperbolic C k diffeomorphism of a compact manifold M that is k-normally hyperbolic to a C k central foliation Wfc . Then for any δ > 0, if f is a C k diffeomorphism of M that is sufﬁciently C 1 close to f , we have the following:

groups acting on manifolds / 121

1) f is (a , b )-partially hyperbolic, where |a − a | < δ and |b − b | < δ, and the splitting TM = Efu ⊕ Efc ⊕ Efs for f is C 0 close to the splitting for f, 2) there exist f invariant foliations by C k leaves Wfc tangent to Efc , which is close in the natural topology on foliations by C k leaves to Wfc , 3) there exists a (nonunique) homeomorphism h of M with h(Wfc ) = Wfc , and h is C k along leaves of Wfc , furthermore h can be chosen to be C 0 small and C k small along leaves of Wfc , and 4) the homeomorphism h is a leaf conjugacy between the actions (M, Wfc , f ) and (M, Wfc , f ). Conclusion (1) is easy and probably older than [118]. One motivation for Theorem 7.1 is to study stability of dynamical properties of partially hyperbolic diffeomorphisms. See the survey [27] by Burns, Pugh, Shub, and Wilkinson for more discussion of that and related issues. 7.2. Uniformly Hyperbolic Actions on Tori

Many works have been written considering local rigidity of actions with some afﬁne, quasi-afﬁne, and generalized quasi-afﬁne actions with hyperbolic behavior. For a discussion of this, we refer to [77]. Here we only discuss results that prove some sort of global rigidity of groups acting on manifolds. The ﬁrst such results were contained in papers of Katok-Lewis [128] and Katok-LewisZimmer [129]. As these are now special cases of later more general results, we do not discuss them in detail here. In this section we discuss results that provide only continuous conjugacies to standard actions. This is primarily because these results are less technical and easier to state. In this context, one can improve regularity of the conjugacy given certain technical dynamical hypotheses on certain dynamical foliations.

definition 7.2. An action of a group on a manifold M is weakly hyperbolic if there exist elements γ1 , . . . , γk each of which is partially hyperbolic such that the sum of the stable sub-bundles of the γi spans the tangent bundle

to M at every point, that is, i Eγi = TM. To discuss the relevant results we need a related topological notion that captures hyperbolicity at the level of fundamental group. This was introduced in [84] by Whyte and the author. If a group acts on a manifold with torsion free nilpotent fundamental group and the action lifts to the universal cover, then

122 / d. fisher

the action of on π1 (M) gives rise to an action of on the Malcev completion N of π1 (M), which is a nilpotent Lie group. This yields a representation of on the Lie algebra n.

definition 7.3. We say an action of on a manifold M with nilpotent fundamental group is π1 -hyperbolic if for the resulting representation on n, we have ﬁnitely many elements γ1 , . . . , γk such that the sum of their eigenspaces with eigenvalue of modulus less than 1 is all of n. One can make this deﬁnition more general by considering M where π1 (M) has a equivariant nilpotent quotient; see [84]. We now discuss results that follow by combining the work of Margulis-Qian [165] with the later work of Schmidt [211] and Fisher-Hitchman [79].

theorem 7.4. Let M = N/ be a compact nilmanifold and let be a lattice in a semisimple Lie group with property (T ). Assume acts on M such that the action lifts to the universal cover and is π1 (M) hyperbolic, then the action is continuously semi-conjugate to an afﬁne action. This theorem was proven by Margulis and Qian, who noted that if the action contained an Anosov element, then the conjugacy could be taken to be a homeomorphism. In [84], the author and Whyte point out that this theorem extends easily to the case of any compact manifold with torsion free nilpotent fundamental group. In [84], we also discuss extensions to manifolds with fundamental group with a quotient that is nilpotent. Margulis and Qian asked whether the assumption of π1 hyperbolicity could be replaced by the assumption that the action on M was weakly hyperbolic. In the case of actions of Kazhdan groups on tori, Schmidt proved that weak hyperbolicity implies π1 hyperbolicity, yielding:

theorem 7.5. Let M = Tn be a compact torus and let be a lattice in a semisimple Lie group with property (T ). Any weakly hyperbolic action M and that lifts to the universal cover is continuously semi-conjugate to an afﬁne action. Remarks: 1) The contribution of Fisher-Hitchman in both theorems is just in extending cocycle superrigidity to a wider class of groups, as discussed above.

groups acting on manifolds / 123

2) The assumption that the action lifts to the universal cover of M is often vacuous because of results concerning cohomology of higher rank lattices. In particular, it is vacuous for cocompact lattices in simple Lie groups of real rank at least 3. It remains an interesting, open question to take this result and prove that the semiconjugacy is always a conjugacy and is also always a smooth diffeomorphism. 7.3. Rigidity Results for Uniformly Hyperbolic Actions

We begin by discussing some work of Goetze and Spatzier. To avoid technicalities we discuss only some of their results. We begin with the following deﬁnition:

definition 7.6. Let ρ : Zk × M → M be an action and γ1 , . . . , γl be a collection of elements that generate for Zk . We call ρ a Cartan action if 1) each ρ(γi ) is an Anosov diffeomorphism, 2) each ρ(γi ) has 1 dimensional strongest stable foliation, and 3) the strongest stable foliations of the ρ(γi ) are pairwise transverse and span the tangent space to the manifold. It is worth noting that Cartan actions are very special in three ways. First, we assume that a large number of elements in the acting group are Anosov diffeomorphisms; second, we assume that each of these has 1 dimensional strongest stable foliation; and lastly, we assume these 1 dimensional directions span the tangent space. All aspects of these assumptions are used in the following theorem. Reproving it even assuming 2 dimensional strongest stable foliations would require new ideas.

theorem 7.7. Let G be a semisimple Lie group with all simple factors of real rank at least 2 and in G a lattice. Then any volume preserving Cartan action of is smoothly conjugate to an afﬁne action on an infranilmanifold. This is slightly different than the statement in Goetze-Spatzier, where they pass to a ﬁnite cover and a ﬁnite index subgroup. It is not too hard to prove this statement from theirs. The proof spans the two papers [105] and [106]. The ﬁrst paper [105] proves, in a somewhat more general context, that the π-simple section arising in the cocycle superrigidity theorem is in fact Hölder continuous. The second paper [105] makes use of the resulting Hölder

124 / d. fisher

Riemannian metric in conjunction with ideas arising in other work of Katok and Spatzier to produce a smooth homogeneous structure on the manifold. The work of Feres-Labourie differs from other work on rigidity of actions with hyperbolic properties in that it does not make any assumptions concerning the existence of invariant measures. Here we state only some consequences of their results, without giving the exact form of cocycle superrigidity that is their main result.

theorem 7.8. Let be a lattice in SL(n, R) for n ≥ 3 and assume acts smoothly on a compact manifold M of dimension n. Further assume that for the induced action N = (G × M)/ we have 1) every R-semisimple 1-parameter subgroup of G acts transitively on N, and 2) some element g in G is uniformly partially hyperbolic with Es ⊕Ew containing the tangent space to M at any point, then M is a torus and the action on M is a standard afﬁne action. These hypotheses are somewhat technical and essentially ensure that one can apply the topological version of cocycle superrigidity proven in [61]. The proof also uses a deep result of Benoist and Labourie classifying Anosov diffeomorphisms with smooth stable and unstable foliations [11]. The nature of the hypotheses of Theorem 7.8 make an earlier remark clear. Ideally one would only have hypotheses on the action, but here we require hypotheses on the induced action instead. It is not clear how to reformulate these hypotheses on the induced action as hypotheses on the original action. Another consequence of the work of Feres and Labourie is a criterion for promoting invariance of rigid geometric structures. More precisely they give a criterion for a G action to preserve a rigid geometric structure on a dense open subset of a manifold M provided a certain type of subgroup preserves a rigid geometric structure on M. 7.4. Conjectures and Questions on Uniformly Hyperbolic Actions

We begin with a very general variant of Conjecture 1.3.

conjecture 7.9. Let G be a semisimple Lie group all of whose simple factors have property (T ), let < G be a lattice. Assume G or acts smoothly on a compact manifold M preserving volume such that some element g in the acting group is nontrivially uniformly partially hyperbolic. Then the action is generalized quasi-afﬁne.

groups acting on manifolds / 125

There are several weaker variants on this conjecture, where, for example, one assumes the action is volume preserving or weakly hyperbolic. Even the following much weaker variant seems difﬁcult:

conjecture 7.10. Let M = Tn and as in Conjecture 7.9. Assume acts on Tn weakly hyperbolicly and preserving a smooth measure. Then the action is afﬁne. This conjecture amounts to conjecturing that the semiconjugacy in Theorem 7.5 is a diffeomorphism. In the special case where some element of is Anosov, the semiconjugacy is at least a homeomorphism. If this is true and enough dynamical foliations are 1 and 2 dimensional and has higher rank, one can then deduce smoothness of the conjugacy from the work of Rodriguez-Hertz on rigidity of actions of abelian groups [209]. Work in progress by the author, Kalinin, and Spatzier seems likely to provide a similar result when contains many commuting Anosov diffeomorphisms without any assumptions on dimensions of foliations. Finally, we recall an intriguing question from [84] that arises in this context.

question 7.11. Let M be a compact manifold with π1 (M) = Zn and assume < SL(n, Z) has ﬁnite index. Let act on M ﬁxing a point so that the resulting action on π1 (M) is given by the standard representation of SL(n, Z) on Zn . Is it true that dim (M) ≥ n? The question is open even if the action on M is assumed to be smooth. The results of [84] imply that there is a continuous map from M to Tn that is equivariant for the standard action on M. Since the image of M is closed and invariant, it is easy to check that it is all of Tn . So the question amounts to one about the existence of equivariant “space ﬁlling curves,” where curve is taken in the generalized sense of continuous map from a lower-dimensional manifold. In another context there are equivariant space ﬁlling curves, but they seem quite special. They arise as surface group equivariant maps from the circle to S2 and come from three manifolds that ﬁber over the circle; see [34].

8. Representations of Fundamental Groups and Arithmetic Quotients In this section we discuss some results and questions related to topological approaches to classifying actions; particularly, some related to Conjecture 1.4. In Subsection 8.2, we also discuss related results and questions concerning maximal generalized afﬁne quotients of actions.

126 / d. fisher 8.1. Linear Images of Fundamental Groups

This section is fundamentally concerned with the question:

question 8.1. Let G be a semisimple Lie group all of whose factors are not compact and have property (T ). Assume G acts by homeomorphisms on a manifold M preserving a measure. Can we classify linear representations of π1 (M)? Similarly for actions of < G a lattice on a manifold M. We remark that in this context, it is possible that π1 (M) has no inﬁnite image linear representations; see discussion in [84] and Section 9 below. In all known examples where this occurs there is an “obvious” inﬁnite image linear representation on some ﬁnite index subgroup. Also, as ﬁrst observed by Zimmer [251] for actions of Lie groups, under mild conditions on the action, representations of the fundamental group become severely restricted. Further work in this direction was done by Zimmer in conjunction with Spatzier [224] and later Lubotzky [152, 153]. Analogous results for lattices are surprisingly difﬁcult in this context and constitute the authors dissertation [73, 74]. We recall a deﬁnition from [251]:

definition 8.2. Let D be a Lie group and assume that D acts on a compact manifold M preserving a ﬁnite measure μ and ergodically. We call the action ˜ on any ﬁnite cover of M is ergodic. engaging if the action of D There is a slightly more technical deﬁnition of engaging for non-ergodic actions that says there is no loss of ergodicity on passing to ﬁnite covers. That is that the ergodic decomposition of μ and its lifts to ﬁnite covers are canonically identiﬁed by the covering map. There are also two variants of this notion totally engaging and topologically engaging; see for example, [258] for more discussion. It is worth noting that for ergodic D actions on M, the action on any ﬁnite cover has at most ﬁnitely many ergodic components, in fact at most the degree of the cover many ergodic components. We remark here that any generalized afﬁne action of a Lie group is engaging if it is ergodic. The actions constructed below in Section 9 are not in general engaging.

theorem 8.3. Let G be a simple Lie group of real rank at least 2. Assume G acts by homeomorphisms on a compact manifold M, preserving a ﬁnite measure μ and engaging. Assume σ : π1 (M) → GL(n, R) is an inﬁnite image linear representation. Then σ (π1 (M)) contains an arithmetic group HZ where HR contains a group locally isomorphic to G.

groups acting on manifolds / 127

For an expository account of the proof of this theorem and a more detailed discussion of engaging conditions for Lie groups, we refer the reader to [258]. The extension of this theorem to lattice actions is non-trivial and is in fact the author’s dissertation. Even the deﬁnition of engaging requires modiﬁcation, since it is not at all clear that a discrete group action lifts to the universal cover.

definition 8.4. Let D be a discrete group and assume that D acts on a compact manifold M preserving a ﬁnite measure μ and ergodically. We call the action engaging if for every 1) ﬁnite index subgroup D in D, 2) ﬁnite cover M of M, and 3) lift of the D action to M , the action of D of M is ergodic. The deﬁnition does immediately imply that every ﬁnite index subgroup D of D acts ergodically on M. In [74], a deﬁnition is given that does not require the D action to be ergodic, but even in that context the ergodic decomposition for D is assumed to be the same as that for D. Deﬁnition 8.4 is rigged to guarantee the following lemma:

lemma 8.5. Let G be a Lie group and < G a lattice. Assume acts on a manifold M preserving a ﬁnite measure and engaging, then the induced G action on (G × M)/ is engaging. We remark that with our deﬁnitions here, the lemma only makes sense for cocompact, but this is not an essential difﬁculty. We can now state a ﬁrst result for lattice actions. Let = π1 ((G × M)/ ).

theorem 8.6. Let G be a simple Lie group of real rank at least 2 and < G a lattice. Assume acts by homeomorphisms on a compact manifold M, preserving a ﬁnite measure μ and engaging. Assume σ : → GL(n, R) is a linear representation whose restriction to π1 (M) has inﬁnite image. Then σ (π1 (M)) contains an arithmetic group HZ where Aut (HR ) contains a group locally isomorphic to G. This theorem is proven by inducing actions, applying Theorem 8.3, and analyzing the resulting output carefully. One would like to assume σ a priori only deﬁned on π1 (M), but there seems no obvious way to extend such a

128 / d. fisher

representation to the linear representation of required by Theorem 8.3 without a priori information on . This is yet another example of the difﬁculties in using induction to study lattice actions. As a consequence of Theorem 8.6, we discover that at least σ () splits as a semidirect product of and σ (π1 (M)). But this is not clear a priori and not clear a posteriori for . 8.2. Arithmetic Quotients

In the context of Theorems 8.3 and 8.6, one can obtain much greater dynamical information concerning the relation of HZ and the dynamics of the action on M. In particular, there is a compact subgroup C < HR and a measurable equivariant map φ : M → C\HR /HZ , which we refer to as a measurable arithmetic quotient. The papers [73] and [152] prove that there is always a canonical maximal quotient of this kind for any action of G or on any compact manifold, essentially by using Ratner’s theorem to prove that every pair of arithmetic quotients are dominated by a common, larger arithmetic quotient. Earlier results of Zimmer also obtained arithmetic quotients, but only under the assumption that there was an inﬁnite image linear representation with discrete image [254]. The results we mention from [74] and [153] then show that there are “lower bounds” on the size of this arithmetic quotient, provided that the action is engaging in terms of the linear representations of the fundamental group of the manifold. In particular, one obtains arithmetic quotients where HR and HZ are essentially determined by σ (π1 (M)). In fact, σ (π1 (M)) contains HZ and is contained in HQ . The book [258] provides a good description of how to produce arithmetic quotients for G actions and the article [75] provides an exposition of the relevant constructions for actions. Under even stronger hypotheses, the papers [85] and [257] imply that the arithmetic quotient related to the “Gromov representation” discussed above in Subsection 6.2 has the same entropy as the original action. This means that, in a sense, the arithmetic quotient captures most of the dynamics of the original action. 8.3. Open Questions

As promised in the introduction, the analogue of Conjecture 1.4 for lattice actions follows.

conjecture 8.7. Let be a lattice in a semisimple group G with all simple factors of real rank at least 2. Assume acts faithfully, preserving volume on a compact manifold M. Further assume the action is not isometric. Then π1 (M) has

groups acting on manifolds / 129

a ﬁnite index subgroup such that surjects onto an arithmetic lattice in a Lie group H where Aut (H) locally contains G. The following questions about arithmetic quotients are natural.

question 8.8. Let G be a simple Lie group of real rank at least 2 and < G a lattice. Assume that G or act faithfully on a compact manifold M, preserving a smooth volume. 1) Is there a non-trivial measurable arithmetic quotient? 2) Can we take the quotient map φ smooth on an open dense set? Due to a construction in [84], one cannot expect that every M admitting a volume preserving G or action has an arithmetic quotient that is even globally continuous or has π1 (M) admitting an inﬁnite image linear representation. However, the difﬁculties created in those examples all vanish on passage to a ﬁnite cover.

question 8.9. Let G be a simple Lie group of real rank at least 2 and < G a lattice. Assume that G or act smoothly on a compact manifold M, preserving a smooth volume. 1) Is there a ﬁnite cover of M of M such that π1 (M ) admits an inﬁnite image linear representation? 2) Can we ﬁnd a ﬁnite cover M of M, a lift of the action to M (on a subgroup of ﬁnite index) and an arithmetic quotient where the quotient map φ is continuous and smooth on an open dense set? The examples discussed in the next section imply that φ is at best Hölder continuous globally. It is not clear whether the questions and conjectures we have just discussed are any less reasonable for SP(1, n), F4−20 and their lattices.

9. Exotic Volume Preserving Actions In this section, we discuss what is known about what are typically called “exotic actions.” These are the only known smooth volume preserving actions of higher rank lattices and Lie groups that are not generalized afﬁne algebraic. These actions make it clear that a clean classiﬁcation of volume preserving actions is out of reach. In particular, these actions have continuous moduli and provide counterexamples to any naive conjectures of the form “the moduli

130 / d. fisher

space of actions of some lattice on any compact manifold M are countable.” In particular, I explain examples of actions of either or G that have large continuous moduli of deformations as well as manifolds where these moduli have multiple connected components. Essentially all of the examples given here derive from the simple construction of “blowing up” a point or a closed orbit, which was introduced to this subject in [128]. The further developments after that result are all further elaborations on one basic construction. The idea is to use the “blow up” construction to introduce distinguished closed invariant sets that can be varied in some manner to produce deformations of the action. The “blow up” construction is a classical tool from algebraic geometry that takes a manifold N and a point p and constructs from it a new manifold N by replacing p by the space of directions at p. Let RP l be the l dimensional projective space. To blow up a point, we take the product of N × RP dim (N) and then ﬁnd a submanifold where the projection to N is a diffeomorphism off of p and the ﬁber of the projection over p is RP dim (N) . For detailed discussion of this construction we refer the reader to any reasonable book on algebraic geometry. The easiest example to consider is to take the action of SL(n, Z), or any subgroup < SL(n, Z) on the torus Tn and blow up the ﬁxed point, in this case the equivalence class of the origin in Rn . Call the resulting manifold M. Provided is large enough, for example, Zariski dense in SL(n, R), this action of does not preserve the measure deﬁned by any volume form on M. A clever construction introduced in [128] shows that one can alter the standard blowing up procedure in order to produce a one parameter family of SL(n, Z) actions on M, only one of which preserves a volume form. This immediately shows that this action on M admits perturbations, since it cannot be conjugate to the nearby, nonvolume preserving actions. Essentially, one constructs different differentiable structures on M that are diffeomorphic but not equivariantly diffeomorphic. After noticing this construction, one can proceed to build more complicated examples by passing to a subgroup of ﬁnite index, and then blowing up several ﬁxed points. One can also glue together the “blown up” ﬁxed points to obtain an action on a manifold with more complicated topology. In particular, one can achieve a fundamental group that is an essentially arbitrary free product with amalgamation or HNN extension of the fundamental group of the original manifold over the fundamental group of (blown-up) orbits. In the context described in more detail below, of blowing up along closed orbits instead of points, it is not hard to do the blowing up and gluing in a way that guarantees that there are no linear representations of the fundamental group of the “exotic example.” To prove nonexistence of linear representations, one

groups acting on manifolds / 131

chooses examples where all groups involved are higher rank lattices and have very constrained linear representation theory. See [84, 128] for a discussion of the topological complications one can introduce. While these actions do not preserve a rigid geometric structure, they do preserve a slightly more general object, an almost rigid structure introduced by Benveniste and the author in [13] and described below. In [12] it is observed that a similar construction can be used for the action of a simple group G by left translations on a homogeneous space H/ where H is a Lie group containing G and < H is a cocompact lattice. Here we use a slightly more involved construction from algebraic geometry, and “blow up” the directions normal to a closed submanifold. That is, we replace some closed submanifold N in H/ by the projective normal bundle to N. In all cases we consider here, this normal bundle is trivial and so is just N × RP l where l = dim (H) − dim (N). Benveniste used his construction to produce more interesting perturbations of actions of higher rank simple Lie group G or a lattice in G. In particular, he produced volume preserving actions that admit volume preserving perturbations. He does this by choosing G < H such that not only are there closed G orbits but so that the centralizer Z = ZH (G) of G in H has nontrivial connected component. If we take a closed G orbit N, then any translate zN for z in Z is also closed and so we have a continuum of closed G orbits. Benveniste shows that if we choose two closed orbits N and zN to blow up and glue, and then vary z in a small open set, the resulting actions can only be conjugate for a countable set of choices of z. This construction is further elaborated in [78]. Benveniste’s construction is not optimal in several senses, nor is his proof of rigidity. In [78], I give a modiﬁcation of the construction that produces non-conjugate actions for every choice of z in a small enough neighborhood. By blowing up and gluing more pairs of closed orbits, this allows me to produce actions where the space of deformations contains a submanifold of arbitrarily high, ﬁnite dimension. Further, Benveniste’s proof that the deformation are nontrivial is quite involved and applies only to higher rank groups. In [78], I give a different proof of nontriviality of the deformations using consequences of Ratner’s theorem [204] due to Shah [213] and Witte Morris [237]. This shows that the construction produces nontrivial perturbations for any semisimple G and any lattice in G. As mentioned above, in [13] we show that none of these actions preserve any rigid geometric structure in the sense of Gromov but that they do preserve a slightly more complicated object that we call an almost rigid structure. Both

132 / d. fisher

rigid and almost rigid structures are easiest to deﬁne in dimension 1. In this context a rigid structure is a non-vanishing vector ﬁeld and an almost rigid structure is a vector ﬁeld vanishing at isolated points and to ﬁnite degree. We continue to use the notation of subsection 6.1 in order to give the precise deﬁnition of almost rigid geometric structure.

definition 9.1. An A-structure φ is called ( j, k)-almost rigid (or just almost rigid) if for every point p, rpk,k−1 is injective on the subgroup r k+j,k (Isk+j ) ⊂ Isk . Thus k-rigid structures are the (0, k)-almost rigid structures.

basic example. Let V be an n-dimensional manifold. Let X1 , . . . , Xn be a 2 collection of vector ﬁelds on M. This deﬁnes an A-structure ψ of type Rn on M. If X1 , . . ., Xn span the tangent space of V at every point, then the structure is rigid in the sense of Gromov. Suppose instead that there exists a point p in V and X1 ∧ . . . ∧ Xn vanishes to order ≤ j at p in V . Then ψ is a ( j, 1)almost rigid structure. Indeed, let p ∈ M, and let (x1 , . . . , xn ) be coordinates ∂ around p. Suppose that in terms of these coordinates, Xl = am . Suppose l ∂xj j+1

j+1,1

that f ∈ Isp . We must show that rp coordinate functions of f . Then f ∈ 9.1

j+1 Isp

alk − am k

( f ) is trivial. Let ( f 1 , . . . , f n ) be the implies that

∂f l ∂x m

vanishes to order j + 1 at p for all k and l. Let (bkl ) be the matrix so that bkm alm = ∂f l det (asr )δkl . Multiplying expression (9.1) by (bkl ), we see that det (asr ) δkl − ∂x k vanishes to order j + 1. But since by assumption det (asr ) vanishes to order ≤ j, j+1,1 this implies that (∂f l /∂x k )(p) = δkl , so rp ( f ) is the identity, as required. If confused by the notation, the interested reader may ﬁnd it enlightening to work out the basic example in the trivial case n = 1. Similar arguments can be given to show that frames that degenerate to subframes are also almost rigid, provided the order of vanishing of the form deﬁning the frame is always ﬁnite.

question 9.2. Does any smooth (or analytic) action of a higher rank lattice admit a smooth (analytic) almost rigid structure in the sense of [13]? More generally does such an action admit a smooth (analytic) rigid geometric structure on an open dense set of full measure?

groups acting on manifolds / 133

This question is, in a sense, related to the discussion above about regularity of the straightening section in cocycle superrigidity. In essence, cocycle superrigidity provides one with a measurable invariant connection and what one wants to know is whether one can improve the measurable connection to a smooth geometric structure with some degeneracy on a small set. The examples described in this section show, among other things, that one cannot expect the straightening section to be smooth in general, though one might hope it is smooth in the complement of a closed submanifold of positive codimension or at least on an open dense set of full measure. We remark that there are other possible notions of almost rigid structures. See the article in this volume by Dumitrescu for a detailed discussion of a different useful notion in the context of complex analytic manifolds [48]. Dumitrescu’s notion is strictly weaker than the one presented here.

10. Non-Volume Preserving Actions This section describes what is known for non-volume preserving actions. Subsections 10.1 and 10.2 describe examples which show that a classiﬁcation in this setting is not in any sense possible. Subsection 10.3 describes some recent work of Nevo and Zimmer that proves surprisingly strong rigidity results in special settings. 10.1. Stuck’s Examples

The following observations are from Stuck’s paper [226]. Let G be any semisimple group. Let P be a minimal parabolic subgroup. Then there is a homomorphism P → R. As in the proof of Theorem 2.7, one can take any R action, view it is a P action, and induce to a G action. If we take an R action on a manifold M, then the induced action takes place on (G × M)/P. We remark that the G action here is not volume preserving, simply because the G action on G/P is proximal. The same is true of the restriction to any action when is a lattice in G. This construction shows that classifying G actions on all compact manifolds implicitly involves classifying all vector ﬁelds on compact manifolds. It is relatively easy to convince oneself that there is no reasonable sense in which the moduli space of vector ﬁelds can be understood up to smooth conjugacy. 10.2. Weinberger’s Examples

This is a variant on the Katok-Lewis examples obtained by blowing up a point and is similar to a blowing up construction common in foliation theory. The

134 / d. fisher

idea is that one takes an action of a subgroup of SL(n, Z) on M = Tn , removes ¯ and then a ﬁxed or periodic point p, retracts onto a manifold with boundary M, n glues in a copy of the R compactiﬁed at inﬁnity by the projective space of rays. It is relatively easy to check that the resulting space admits a continuous action and even that there are many invariant measures for the action, but no invariant volume. One can also modify this construction by doing the same construction at multiple ﬁxed or periodic points simultaneously and by doing ¯ more complicated gluings on the resulting M. This construction is discussed in [55] and a variant for abelian group actions is discussed in [125]. In the abelian case it is possible to smooth the action, but this does not seem to be the case for actions of higher rank lattices. As far as I can tell, there is no obstruction to repeating this construction for closed orbits as in the case of the algebro-geometric blow up, but this does not seem to be written formally anywhere in the literature. Also, a recent construction of Hurder shows that one can iterate this construction inﬁnitely many times, taking retracts in smaller and smaller neighborhoods of periodic points of higher and higher orders. The resulting object is a kind of fractal admitting an SL(n, Z) action [119]. 10.3. Work of Nevo-Zimmer

In this subsection, we describe some work of Nevo and Zimmer from the sequence of papers [187, 188, 191, 186]. For more detailed discussion see the survey by Nevo and Zimmer [189] as well as the following two articles for related results [190, 192]. Given a group G acting on a space X and a measure μ on G, we call a measure ν on X stationary if μ ∗ ν = ν. We will only consider the case where the group generated by the support μ is G; such measures are often called admissible. This is a natural generalization of the notion of an invariant measure. If G is an amenable group, any action of G on a compact metric space admits an invariant measure. If G is not amenable, invariant measures need not exist, but stationary measures always do. We begin with the following cautionary example:

example 10.1. Let G = SL(n, R) acting on Rn+1 by the standard linear action on the ﬁrst n coordinates and the trivial action on the last. Then the corresponding G action on P(Rn+1 ) has the property that any stationary measure is supported either on the subspace P(Rn ) given by the ﬁrst n coordinates or on the subspace P(R) given by zeroing the ﬁrst n coordinates.

groups acting on manifolds / 135

The proof of this assertion is an easy exercise. The set P(R) is a collection of ﬁxed points, so clearly admits invariant measures. The orbit of any point

in P(Rn+1 )\(P(Rn ) P(R)) is SL(n, R)/(SL(n, R) Rn ). It is straightforward to check that no stationary measures can be supported on unions of sets of this kind. This fact should not be a surprise as it generalizes the fact that invariant measures for amenable group actions are often supported on minimal sets. To state the results of Nevo and Zimmer, we need a slightly stronger notion of admissibility. We say a measure μ on a locally compact group G is strongly admissible if the support of μ generates G and μ∗k is absolutely continuous with respect to Haar measure on G for some positive k. In the papers of Nevo and Zimmer, this stronger notion is called admissible.

theorem 10.2. Let X be a compact G space where G is a semisimple Lie group with all factors of real rank at least 2. Then for any admissible measure μ on G and any μ-stationary measure ν on X , we have either 1) ν is G invariant, or 2) there is a nontrivial measurable quotient of the G space (X , ν) that is of the form (G/Q , Lebesgue) where Q ⊂ G is a parabolic subgroup. The quotient space G/Q is called a projective quotient in the work of Nevo and Zimmer. Theorem 10.2 is most interesting for us for minimal actions, where the measure ν is necessarily supported on all of X and the quotient therefore reﬂects the action on X and not some smaller set. Example 10.1 indicates the reason to be concerned, since there the action on the larger projective space is not detected by any stationary measure. We remark here that Feres and Ronshausen have introduced some interesting ideas for studying group actions on sets not contained in the support of any stationary measure [69]. Similar ideas are developed in a somewhat different context by Deroin, Kleptsyn, and Navas [45]. In [186], Nevo and Zimmer show that for smooth non-measure preserving actions on compact manifolds that also preserve a rigid geometric structure, one can sometimes prove the existence of a projective factor where the factor map is smooth on an open dense set. We also want to mention the main result of Stuck’s paper [226], which we have already cited repeatedly for the fact that non-volume preserving actions of higher rank groups on compact manifolds cannot be classiﬁed.

theorem 10.3. Let G be a semisimple Lie group with ﬁnite center. Assume G acts minimally by homeomorphisms on a compact manifold M. Then either the action

136 / d. fisher

is locally free or the action is induced from a minimal action by homeomorphisms of a proper parabolic subgroup of G on a manifold N. Stuck’s theorem is proven by studying the Gauss map from M to the Grassman variety of subspaces of g deﬁned by taking a point to the Lie algebra of its stabilizer. This technique also plays an important role in the work of Nevo and Zimmer.

11. Groups That Do Act on Manifolds Much of the work discussed so far begs an obvious question: “are there many interesting subgroups of Diff (M) for a general M?” So far we have only seen “large” subgroups of Diff (M) that arise in a geometric fashion, from the pres˜ In this section, ence of a connected Lie group in either Diff (M) or Diff (M). we describe two classes of examples that make it clear that other phenomena exist. The following problem, however, seems open:

problem 11.1. For a compact manifold M with a volume form ω construct a subgroup < Diff (M, ω) such that has no linear representations. An example that is not often considered in this context is that Aut (Fn ) acts on the space Hom (Fn , K ) for K any compact Lie group. Since Hom (Fn , K ) K n this deﬁnes an action of Aut (Fn ) on a manifold. This action clearly preserves the Haar measure on K n ; see [107]. This action is not very well studied; we only know of [76, 97]. Similar constructions are possible and better known with mapping class groups, although in that case one obtains a representation variety that is not usually a manifold. Since Aut (Fn ) has no faithful linear representations, this yields an example of truly “non-linear” action of a large group. This action is still very special and one expects many other examples of non-linear actions. We now describe two constructions that yield many examples if we drop the assumption that the action preserves volume. 11.1. Thompson’s Groups

Richard Thompson introduced a remarkable family of groups, now referred to as Thompson’s groups. These come in various ﬂavors and have been studied from several points of view; see, for example, [33, 117]. For our purposes, the most important of these groups are the one’s typically denoted T . One description of this group is the collection of piecewise linear diffeomorphisms of the circle where the break points and slopes are all dyadic rationals. (One can

groups acting on manifolds / 137

replace the implicit 2 here with other primes and obtain similar, but different, groups.) We record here two important facts about this group T of piecewise linear homeomorphisms of S1 .

theorem 11.2. (thompson, see [33]) The group T is simple. theorem 11.3. (ghys-sergiescu [103]) The deﬁning piecewise linear action of the group T on S1 is conjugate by a homeomorphism to a smooth action. These two facts together provide us with a rather remarkable class of examples of groups that act on manifolds. As ﬁnitely generated linear groups are always residually ﬁnite, the group T has no linear representations whatsoever. A simpler variant of Problem 11.1 is

problem 11.4. Does T admit a volume preserving action on a compact manifold? It is easy to see that compactness is essential in this question. We can construct a smooth action of T on S1 × R simply by taking the Ghys-Sergiescu action on S1 and acting on R by the inverse of the derivative cocycle. Replacing the derivative cocycle with the Jacobian cocycle, this procedure quite generally converts non-volume preserving actions on compact manifolds to volume preserving one’s on non-compact manifolds, but we know of no real application of this in the present context. Another variant of Problem 11.4 is

problem 11.5. Given a compact manifold M and a volume form ω does Diff (M, ω) contain a ﬁnitely generated, inﬁnite discrete simple group? This question is reasonable for any degree of regularity on the diffeomorphisms. In Ghys’s survey on groups acting on the circle, he points out that Thompson’s group can be realized as piecewise SL(2, Z) homeomorphisms of the circle [102]. In [70], Whyte and the author point out that the group of piecewise SL(n, Z) maps on either the torus or the real projective space is quite large. The following are natural questions; see [70] for more discussion.

question 11.6. Are there interesting ﬁnitely generated or ﬁnitely presented subgroups of piecewise SL(n, Z) maps on Tn or Pn−1 ? Can any such group that is not a subgroup of SL(n, Z) be made to act smoothly?

138 / d. fisher 11.2. Highly Distorted Subgroups of Diff (S2 )

In [30], Calegari and Freedman construct a very interesting class of subgroups of Diff ∞ (S2 ). Very roughly, they prove:

theorem 11.7. There is a ﬁnitely generated subgroup G of Diff ∞ (S2 ) that contains a rotation r as an arbitrarily distorted element. Here by arbitrarily distorted we mean that we can choose the group G so that the function f (n) = r n G grows more slowly than any function we choose. It is well known that for linear groups, the function f (n) is at worst a logarithm, so this theorem immediately implies that we can ﬁnd G with no faithful linear representations. This also answered a question raised by Franks and Handel in [88]. More recently, Avila has constructed similar examples in Diff ∞ (S1 ) [4]. This answers a question raised in [30] where such subgroups were constructed in Diff 1 (S1 ). We are naturally led to the following questions. We say a diffeomorphism has full support if the complement of the ﬁxed set is dense.

question 11.8. For which compact manifolds M does Diff ∞ (M) contain arbitrarily distorted elements of full support? The same question for Diff ω (M)? The same question for Diff ∞ (M, ν) where ν is a volume form on M? For the second two questions, we can drop the hypothesis of full support. The second and third questions here seem quite difﬁcult and the answer could conceivably be “none.” The only examples where anything is known in the volume preserving setting are compact surfaces of genus at least 1, where no element is more than quadratically distorted by a result of Polterovich [196]. However, this result depends heavily on the fact that in dimension 2, preserving a volume form is the same as preserving a symplectic structure. 11.3. Topological Construction of Actions on High Dimensional Spheres

In this subsection, we recall a construction due to Farrell and Lafont that allows one to construct actions of a large class of groups on closed disks, and so by doubling, to construct actions on spheres [59]. The construction yields actions on very high-dimensional disks and spheres and is only known to produce actions by homeomorphisms.

groups acting on manifolds / 139

The class of groups involved is the set of groups that admit an EZ-structure. This notion is a modiﬁcation of an earlier notion due to Bestvina of a Zstructure on a group [16]. We do not recall the precise deﬁnition here, but refer the reader to the introduction to [59] and remark here that both torsion free Gromov hyperbolic groups and CAT (0)-groups admit EZ-structures. The result that concerns us is

theorem 11.9. Given a group with an EZ-structure, there is an action of by homeomorphisms on a closed disk. In fact, Farrell and Lafont give a fair amount of information concerning their actions that are quite different from the actions we are usually concerned with here. In particular, the action is properly discontinuous off a closed subset of the boundary of the disk. So from a dynamical viewpoint carries the “interesting part” of the action, for example, it is the support of any stationary measure. Farrell and Lafont point out an analogy between their construction and the action of a Kleinian group on the boundary of hyperbolic space. An interesting general question is

question 11.10. When can the Farrell-Lafont action on a disk or sphere be chosen smooth? 12. Rigidity for Other Classes of Acting Groups In this section, we collect some results and questions concerning actions of other classes of groups. In almost all cases, little is known. 12.1. Lattices in Semi-direct Products

While it is not reasonable to expect classiﬁcation results for arbitrary actions of all lattices in all Lie groups, there are natural broader classes to consider. To pick a reasonable class, a ﬁrst guess is to try to exclude Lie groups whose lattices have homomorphisms onto Z or larger free groups. There are many such groups. For example, the groups Sl(n, Z) Zn that are lattices in Sl(n, R) Rn have property (T ) as soon as n > 2. As it turns out, a reasonable setting is to consider perfect Lie groups with no compact factors or factors locally isomorphic to SO(1, n) or SU(1, n). Any such Lie group will have property (T ) and therefore so will its lattices. Many examples of such lattices are described in [230]. Some ﬁrst rigidity results for these groups are contained in Zimmer’s paper [252]. The relevant full generalization of the cocycle superrigidity

140 / d. fisher

theorem is contained in [238]. In this context it seems that there are probably many rigidity theorems concerning actions of these groups already implicit in the literature, following from the results in [238] and existing arguments. 12.2. Universal Lattices

As mentioned above in Subsection 4.7, the groups SL(n, Z[X ]) for n > 2 have property (T ) by a result of Shalom [216]. His proof also works with larger collections of variables and some other arithmetic groups. The following is an interesting problem.

problem 12.1. Prove cocycle superrigidity for universal lattices. For clarity, we indicate that we expect that all linear representations and cocycles (up to “compact noise”) will be described in terms of representations of the ambient groups, for example, SL(n, R), and will be determined by specifying a numerical value of X . Some partial results toward superrigidity are known by Farb [52] and Shenﬁeld [218]. The problem of completely classifying linear representations in this context does seem to be open. 12.3. SO(1,n) and SU(1,n) and Their Lattices

It is conjectured that all lattices in SO(1, n) and SU(1, n) admit ﬁnite-index subgroups with surjections to Z; see [18]. The conjecture is usually attributed to Thurston and sometimes to Borel for the case of SU(1, n). If this is true, it immediately implies that actions of those lattices can never be classiﬁed. There are still some interesting results concerning actions of these lattices. In [214], Shalom places restrictions on the possible actions of SO(1, n) and SU(1, n) in terms of the fundamental group of the manifold acted upon. This work is similar in spirit to the work of Lubotzky and Zimmer described in Section 8, but requires more restrictive hypotheses. In [78], the author exhibits large moduli spaces of ergodic afﬁne algebraic actions that are constructed for certain lattices in SO(1, n). These moduli spaces are, however, all ﬁnite dimensional. In [71], I construct an inﬁnite-dimensional moduli of deformations of an isometric action of SO(1, n). Both of these constructions rely on a notion of bending introduced by Johnson and Millson in the ﬁnite dimensional setting [121]. I do not formulate any precise questions or conjectures in this direction as I am not sure what phenomenon to expect.

groups acting on manifolds / 141

12.4. Lattices in Other Locally Compact Groups

Much recent work has focused on developing a theory of lattices in locally compact groups other than Lie groups. This theory is fully developed for algebraic groups over other local ﬁelds. Though we did not mention it here, some of Zimmer’s own conjectures were made in the context of S-arithmetic groups, that is, lattices in products of real and p-adic Lie groups. The following conjecture is natural in this context and does not seem to be stated in the literature.

conjecture 12.2. Let G be semisimple algebraic group deﬁned over a ﬁeld k of positive characteristic and < G a lattice. Further assume that all simple factors of G have k-rank at least 2. Then any action on a compact manifold factors through a ﬁnite quotient. The existence of a measurable invariant metric in this context should be something one can deduce from the cocycle superrigidity theorems, though it is not clear that the correct form of these theorems is known or in the literature. There is also a growing interest in lattices in locally compact groups that are not algebraic. We remark here that Kac-Moody lattices typically admit no nontrivial homomorphisms even to Homeo (M); see [83] for a discussion. The only other interesting class of lattices known to the author is the lattices in the isometry group of a product of two trees constructed by Burger and Mozes [26, 25]. These groups are inﬁnite simple ﬁnitely presented groups.

problem 12.3. Do the Burger-Mozes lattices admit any nontrivial homomorphisms to Diff (M) when M is a compact manifold? That these groups do act by homeomorphisms on high dimensional spheres by the construction discussed in Subsection 11.3 was pointed out to the author by Lafont. 12.5. Automorphism Groups of Free and Surface Groups

We brieﬂy mention a last set of natural questions. There is a long-standing analogy between higher rank lattices and two other classes of groups. These are mapping class groups of surfaces and the outer automorphism group of the free group. See [21] for a detailed discussion of this analogy. In the context of this article, this raises the following question. Here we denote the mapping class group of a surface by MCG() and the outer automorphism group of Fn by Out (Fn ).

142 / d. fisher

question 12.4. Assume has genus at least 2 or that n > 2. Does MCG() or Out (Fn ) admit a faithful action on a compact manifold? A faithful action by smooth, volume preserving diffeomorphisms? By not assuming that M is connected, we are implicitly asking the same question about all ﬁnite-index subgroups of MCG() and Out(Fn ). Recall from Section 11 that Aut (Fn ) does admit a volume preserving action on a compact manifold. This makes the question above particularly intriguing.

13. Properties of Subgroups of Diff (M) As remarked in Subsection 4.4, in the paper [100], Ghys attempts to reprove the classical Zassenhaus lemma for linear groups for groups of analytic diffeomorphisms. While the full strength of the Zassenhaus lemma does not hold in this setting, many interesting results do follow. This immediately leaves one wondering to what extent other properties of linear groups might hold for diffeomorphism groups or at least what analogues of many theorems might be true. This direction of research was initiated by Ghys and most of the questions below are due to him. It is worth noting that some properties of ﬁnitely generated linear groups, like residual ﬁniteness, do not appear to have reasonable analogues in the setting of diffeomorphism groups. For residual ﬁniteness, the obvious example is the Thompson group discussed above, which is simple. 13.1. Jordan’s Theorem

For linear groups, there is a classical (and not too difﬁcult) result known as Jordan’s theorem. This says that for any ﬁnite subgroup of GL(n, C), there is a subgroup of index at most c(n) that is abelian. One cannot expect better than this, as S1 has ﬁnite subgroups of arbitrarily large order and is a subgroup of GL(n, C). For proofs of Jordan’s theorem as well as the theorems on linear groups mentioned in the next subsection, we refer the reader to, for example, [208].

question 13.1. Given a compact manifold M and a ﬁnite subgroup F of Diff (M), is there a constant c(M) such that F has an abelian subgroup of index c(M)? As above, one cannot expect more than this simply because one can construct actions of S1 on M and, therefore, ﬁnite abelian subgroups of Diff (M) of arbitrarily large order.

groups acting on manifolds / 143

For this question, it may be more natural to ask about ﬁnite groups of homeomorphisms and not assume any differentiability of the maps. Using the results in, for example, [157], one can show that at most ﬁnitely many simple ﬁnite groups act on a given compact manifold. To be clear, one can show this using the classiﬁcation of ﬁnite simple groups. It would be most interesting to resolve Question 13.1 without reference to the classiﬁcation. A recent preprint of Mundet i Rieri provides some evidence for a positive answer to the question [180]. 13.2. Burnside Problem

A group is called periodic if all of its elements have ﬁnite order. We say a periodic group G has bounded exponent if every element has order at most m. In 1905 Burnside proved that ﬁnitely generated linear groups of bounded exponent are ﬁnite and in 1911 Schur proved that ﬁnitely generated periodic linear groups are ﬁnite. For a general ﬁnitely generated group, this is not true, counterexamples to the Burnside conjecture were constructed in a sequence of works by many authors, including Novikov, Golod, Shafarevich, and Ol’shanskii. We refer the reader to the Website: http : //www − groups.dcs.st − and.ac.uk/ history/HistTopics/Burnside_problem.html for a detailed discussion of the history. This page also discusses the restricted Burnside conjecture resolved by Zelmanov. In this context, the following questions seem natural:

question 13.2. Are there inﬁnite, ﬁnitely generated periodic groups of diffeomorphisms of a compact manifold M? Are there inﬁnite, ﬁnitely generated bounded exponent groups of diffeomorphisms of a compact manifold M? Some ﬁrst results in this direction are contained in the paper of Rebelo and Silva [206]. 13.3. Tits’s alternative

In this section, we ask a sequence of questions related to a famous theorem of Tits’ and more recent variants on it [229].

theorem 13.3. Let be a ﬁnitely generated linear group. Then either contains a free subgroup on two generators or is virtually solvable. The following conjecture of Ghys is a reasonable alternative to this for groups of diffeomorphisms.

144 / d. fisher

conjecture 13.4. Let M be a compact manifold and a ﬁnitely generated group of smooth diffeomorphisms of M. Then either contains a free group on two generators or preserves some measure on M. The best evidence for this conjecture to date is a theorem of Margulis’ that proves the conjecture for M = S1 [163]. Ghys has also asked if the more exact analogue of Tits’s theorem might be true for analytic diffeomorphisms. A closely related question is whether there exist groups of intermediate growth inside of diffeomorphism groups. The growth of a group is the rate of growth of the balls in the Cayley graph. For linear groups, this is either polynomial or exponential. In general, there are groups of intermediate growth, that is, subexponential but not polynomial. The ﬁrst examples of these were discovered by Grigorchuk [109]. Some of these examples were shown by Grigorchuk and Maki to act by homeomorphisms on the interval [110]. More recently, it was shown by Navas that these examples do have an action by C 1 diffeomorphisms on the interval [185]. In the opposite direction, Navas showed that these phenomena disappear if one considers instead C 1+α diffeomorphisms for any positive α. In that setting, Navas shows that any ﬁnitely generated group of diffeomorphisms of the interval with subexponential growth must be nilpotent. It would be interesting to understand what can occur in higher dimensions. A recent related line of research for linear groups concerns uniform exponential growth. A ﬁnitely generated group has uniform exponential growth if the number of elements in a ball of radius r in a Cayley graph for grows at least as λr for some λ > 1 that does not depend on the choice of generators. For linear groups, exponential growth implies uniform exponential growth by a theorem of Eskin, Mozes, and Oh [51]. There are examples of groups having exponential but not uniform exponential growth due to Wilson [235]. This raises the following question. question 13.5. Let M be a compact manifold and a ﬁnitely generated subgroup of Diff (M). If has exponential growth, does have uniform exponential growth? The question seems most likely to have a positive answer if one further assumes that is nonamenable.

References [1] [2]

Scot Adams, Dynamics on Lorentz manifolds, World Scientiﬁc Publishing Co., River Edge, NJ, 2001. MR1875067 (2003c:53101) Jurgen Andersen, Mapping class groups do not have Kazhdan’s property (T ), preprint 2007.

groups acting on manifolds / 145 [3]

[4] [5]

[6]

[7] [8]

[9]

[10]

[11]

[12] [13]

[14]

[15]

[16] [17] [18]

[19] [20] [21]

D. V. Anosov, Geodesic ﬂows on closed Riemann manifolds with negative curvature, Proceedings of the Steklov Institute of Mathematics, No. 90 (1967). Translated from the Russian by S. Feder, American Mathematical Society, Providence, RI, 1969. MR0242194 (39 #3527) Artur Avila, Distortion elements in Diff∞ (R/Z), preprint 2008. Uri Bader, Charles Frances, and Karin Melnick, An embedding theorem for automorphism groups of Cartan geometries, (English summary). Geom. Funct. Anal. 19 (2009), no. 2, 333–355. MR254240 Uri Bader, Alex Furman, Tsachik Gelander, and Nicolas Monod, Property (T) and rigidity for actions on Banach spaces, Acta Math. 198 (2007), no. 1, 57–105. MR2316269 (2008g:22007) Uri Bader and Yehuda Shalom, Factor and normal subgroup theorems for lattices in products of groups, Invent. Math. 163 (2006), no. 2, 415–454. MR2207022 (2006m:22017) Augustin Banyaga, Sur la structure du groupe des difféomorphismes qui préservent une forme symplectique, Comment. Math. Helv. 53 (1978), no. 2, 174–227. MR490874 (80c:58005) ———, The structure of classical diffeomorphism groups, Mathematics and its Applications, vol. 400, Kluwer Academic Publishers Group, Dordrecht, 1997. MR1445290 (98h:22024) Yves Benoist, Orbites des structures rigides (d’après M. Gromov), Integrable systems and foliations/Feuilletages et systèmes intégrables (Montpellier, 1995), Progr. Math., vol. 145, Birkhäuser, Boston, 1997, pp. 1–17. MR1432904 (98c:58126) Yves Benoist and François Labourie, Sur les difféomorphismes d’Anosov afﬁnes à feuilletages stable et instable différentiables, Invent. Math. 111 (1993), no. 2, 285–308. MR11 98811 (94d:58114) E. J. Benveniste, University of Chicago, Ph.D. thesis, unpublished, 1996. E. Jerome Benveniste and David Fisher, Nonexistence of invariant rigid structures and invariant almost rigid structures, Comm. Anal. Geom. 13 (2005), no. 1, 89–111. MR2154667 (2006f:53056) G. Besson, G. Courtois, and S. Gallot, Entropies et rigidités des espaces localement symétriques de courbure strictement négative, Geom. Funct. Anal. 5 (1995), no. 5, 731–799. MR1354289 (96i:58136) Gérard Besson, Gilles Courtois, and Sylvestre Gallot, Minimal entropy and Mostow’s rigidity theorems, Ergodic Theory Dynam. Systems 16 (1996), no. 4, 623–649. MR140 6425 (97e:58177) Mladen Bestvina, Local homology properties of boundaries of groups, Michigan Math. J. 43 (1996), no. 1, 123–139. MR1381603 (97a:57022) Mladen Bestvina and Koji Fujiwara, Bounded cohomology of subgroups of mapping class groups, Geom. Topol. 6 (2002), 69–89 (electronic). MR1914565 (2003f:57003) A. Borel and N. Wallach, Continuous cohomology, discrete subgroups, and representations of reductive groups, second ed., Mathematical Surveys and Monographs, Vol. 67, American Mathematical Society, Providence, RI, 2000. MR1721403 (2000j:22015) Armand Borel, Compact Clifford-Klein forms of symmetric spaces, Topology 2 (1963), 111–122. MR0146301 (26 #3823) Martin Bridson and Karen Vogtmann, Actions of automorphism groups of free groups on homology spheres and acyclic manifolds, to appear Comm. Math. Helv. Martin R. Bridson and Karen Vogtmann, Automorphism groups of free groups, surface groups and free Abelian groups, Problems on mapping class groups and related topics,

146 / d. fisher

[22] [23] [24]

[25]

[26] [27]

[28]

[29] [30]

[31] [32]

[33] [34] [35] [36] [37]

[38] [39]

[40]

Proc. Sympos. Pure Math., Vol. 74, American Mathematical Society, Providence, RI, 2006, pp. 301–316. MR2264548 (2008g:20091) M. Burger and N. Monod, Continuous bounded cohomology and applications to rigidity theory, Geom. Funct. Anal. 12 (2002), no. 2, 219–280. MR1911660 (2003d:53065a) Marc Burger, An extension criterion for lattice actions on the circle, in this volume. ———, Rigidity properties of group actions on CAT(0)-spaces, Proceedings of the International Congress of Mathematicians, Vol. 1, 2 (Zürich, 1994) (Basel), Birkhäuser, 1995, pp. 761–769. MR1403976 (97j:20033) Marc Burger and Shahar Mozes, Groups acting on trees: from local to global structure, Inst. Hautes Études Sci. Publ. Math. (2000), no. 92, 113–150 (2001). MR1839488 (2002i:20041) ———, Lattices in product of trees, Inst. Hautes Études Sci. Publ. Math. (2000), no. 92, 151–194 (2001). MR1839489 (2002i:20042) Keith Burns, Charles Pugh, Michael Shub, and Amie Wilkinson, Recent results about stable ergodicity, Smooth ergodic theory and its applications (Seattle, WA, 1999), Proc. Sympos. Pure Math., Vol. 69, American Mathematical Society, Providence, RI, 2001, pp. 327–366. MR1858538 (2002m:37042) Eugenio Calabi, On compact, Riemannian manifolds with constant curvature. I, Proc. Sympos. Pure Math., Vol. III, American Mathematical Society, Providence, RI, 1961, pp. 155–180. MR0133787 (24 #A3612) Eugenio Calabi and Edoardo Vesentini, On compact, locally symmetric Kähler manifolds, Ann. of Math. (2) 71 (1960), 472–507. MR0111058 (22 #1922b) Danny Calegari and Michael H. Freedman, Distortion in transformation groups, Geom. Topol. 10 (2006), 267–293 (electronic), with an appendix by Yves de Cornulier. MR220 7794 (2007b:37048) A. Candel and R. Quiroga-Barranco, Gromov’s centralizer theorem, Geom. Dedicata 100 (2003), 123–155. MR2011119 (2004m:53075) Alberto Candel and Raul Quiroga-Barranco, Parallelisms, prolongations of Lie algebras and rigid geometric structures, Manuscripta Math. 114 (2004), no. 3, 335–350. MR2076451 (2005f:53056) J. W. Cannon, W. J. Floyd, and W. R. Parry, Introductory notes on Richard Thompson’s groups, Enseign. Math. (2) 42 (1996), no. 3-4, 215–256. MR1426438 (98g:20058) James W. Cannon and William P. Thurston, Group invariant Peano curves, Geom. Topol. 11 (2007), 1315–1355. MR2326947 (2008i:57016) Serge Cantat, Groupes de transformations birationnelles du plan, preprint. ———, Version kählérienne d’une conjecture de Robert J. Zimmer, Ann. Sci. École Norm. Sup. (4) 37 (2004), no. 5, 759–768. MR2103473 (2006b:22010) Pierre-Emmanuel Caprace and Bertrand Rémy, Simplicité abstraite des groupes de KacMoody non afﬁnes, C. R. Math. Acad. Sci. Paris 342 (2006), no. 8, 539–544. MR2217912 (2006k:20102) Indira Chatterji and Martin Kassobov, New examples of ﬁnitely presented groups with strong ﬁxed point properties,J. Topol. Anal. 1 (2009), no. 1, 1–12. MR2649346 Christopher Connell and Benson Farb, Some recent applications of the barycenter method in geometry, Topology and geometry of manifolds (Athens, GA, 2001), Proc. Sympos. Pure Math., Vol. 71, American Mathematical Society, Providence, RI, 2003, pp. 19–50. MR2024628 (2005e:53058) Kevin Corlette, Archimedean superrigidity and hyperbolic geometry, Ann. of Math. (2) 135 (1992), no. 1, 165–182. MR1147961 (92m:57048)

groups acting on manifolds / 147 [41] [42] [43]

[44]

[45]

[46] [47]

[48] [49] [50] [51] [52] [53] [54] [55] [56] [57]

[58] [59] [60] [61] [62]

Kevin Corlette and Robert J. Zimmer, Superrigidity for cocycles and hyperbolic geometry, Internat. J. Math. 5 (1994), no. 3, 273–290. MR1274120 (95g:58055) S. G. Dani, Continuous equivariant images of lattice-actions on boundaries, Ann. of Math. (2) 119 (1984), no. 1, 111–119. MR736562 (85i:22009) Pierre de la Harpe and Alain Valette, La propriété (T ) de Kazhdan pour les groupes localement compacts (avec un appendice de Marc Burger), Astérisque (1989), no. 175, 158, with an appendix by M. Burger. MR1023471 (90m:22001) Patrick Delorme, 1-cohomologie des représentations unitaires des groupes de Lie semi-simples et résolubles. Produits tensoriels continus de représentations, Bull. Soc. Math. France 105 (1977), no. 3, 281–336. MR0578893 (58 #28272) Bertrand Deroin, Victor Kleptsyn, and Andrés Navas, Sur la dynamique unidimensionnelle en régularité intermédiaire, Acta Math. 199 (2007), no. 2, 199–262. MR23 58052 Julie Déserti, Groupe de Cremona et dynamique complexe: une approche de la conjecture de Zimmer, Int. Math. Res. Not. (2006), Art. ID 71701, 27. MR2233717 (2007d:22013) Tien-Cuong Dinh and Nessim Sibony, Groupes commutatifs d’automorphismes d’une variété kählérienne compacte, Duke Math. J. 123 (2004), no. 2, 311–328. MR2066940 (2005g:32020) Sorin Dumitrescu, Meromorphic almost rigid geometric structures, in this volume. Michael Entov and Leonid Polterovich, Calabi quasimorphism and quantum homology, Int. Math. Res. Not. (2003), no. 30, 1635–1676. MR1979584 (2004e:53131) D. B. A. Epstein, The simplicity of certain groups of homeomorphisms, Compositio Math. 22 (1970), 165–173. MR0267589 (42 #2491) Alex Eskin, Shahar Mozes, and Hee Oh, On uniform exponential growth for linear groups, Invent. Math. 160 (2005), no. 1, 1–30. MR2129706 (2006a:20081) Benson Farb, Group actions and Helly’s theorem,Adv. Math. 222 (2009), no. 5, 1574– 1588. MR2555905 Benson Farb and Howard Masur, Superrigidity and mapping class groups, Topology 37 (1998), no. 6, 1169–1176. MR1632912 (99f:57017) Benson Farb and Peter Shalen, Real-analytic actions of lattices, Invent. Math. 135 (1999), no. 2, 273–296. MR1666834 (2000c:22017) ———, Lattice actions, 3-manifolds and homology, Topology 39 (2000), no. 3, 573–587. MR1746910 (2001b:57041) ———, Groups of real-analytic diffeomorphisms of the circle, Ergodic Theory Dynam. Systems 22 (2002), no. 3, 835–844. MR1908556 (2003e:37030) Benson Farb and Peter B. Shalen, Real-analytic, volume-preserving actions of lattices on 4-manifolds, C. R. Math. Acad. Sci. Paris 334 (2002), no. 11, 1011–1014. MR1913726 (2003e:57055) Daniel S. Farley, Proper isometric actions of Thompson’s groups on Hilbert space, Int. Math. Res. Not. (2003), no. 45, 2409–2414. MR2006480 (2004k:22005) F. T. Farrell and J.-F. Lafont, EZ-structures and topological applications, Comment. Math. Helv. 80 (2005), no. 1, 103–121. MR2130569 (2006b:57022) R. Feres, Afﬁne actions of higher rank lattices, Geom. Funct. Anal. 3 (1993), no. 4, 370–394. MR1223436 (96d:22013) R. Feres and F. Labourie, Topological superrigidity and Anosov actions of lattices, Ann. Sci. École Norm. Sup. (4) 31 (1998), no. 5, 599–629. MR1643954 (99k:58112) R. Feres and P. Lampe, Cartan geometries and dynamics, Geom. Dedicata 80 (2000), no. 1–3, 29–41. MR1762497 (2001i:53059)

148 / d. fisher [63] [64] [65]

[66] [67]

[68]

[69] [70] [71] [72] [73] [74] [75]

[76] [77]

[78] [79] [80] [81]

[82] [83] [84]

Renato Feres, Connection preserving actions of lattices in SLn R, Israel J. Math. 79 (1992), no. 1, 1–21. MR1195250 (94a:58039) ———, Actions of discrete linear groups and Zimmer’s conjecture, J. Differential Geom. 42 (1995), no. 3, 554–576. MR1367402 (97a:22016) ———, Dynamical systems and semisimple groups: an introduction, Cambridge Tracts in Mathematics, Vol. 126, Cambridge University Press, Cambridge, 1998. MR1670703 (2001d:22001) ———, An introduction to cocycle super-rigidity, Rigidity in dynamics and geometry (Cambridge, 2000), Springer, Berlin, 2002, pp. 99–134. MR1919397 (2003g:37041) ———, Rigid geometric structures and actions of semisimple Lie groups, Rigidité, groupe fondamental et dynamique, Panor. Synthèses, Vol. 13, Soc. Math. France, Paris, 2002, pp. 121–167. MR1993149 (2004m:53076) Renato Feres and Anatole Katok, Ergodic theory and dynamics of G-spaces (with special emphasis on rigidity phenomena), Handbook of dynamical systems, Vol. 1A, NorthHolland, Amsterdam, 2002, pp. 665–763. MR1928526 (2003j:37005) Renato Feres and Emily Ronshausen, Harmonic functions over group actions, in this volume. D. Fisher and K. Whyte, When is a group action determined by its orbit structure? Geom. Funct. Anal. 13 (2003), no. 6, 1189–1200. MR2033836 (2004k:37045) David Fisher, Cohomology of arithmetic groups and bending group actions, in preparation. ———, First cohomology and local rigidity of group actions, in preparation. ———, A canonical arithmetic quotient for actions of lattices in simple groups, Israel J. Math. 124 (2001), 143–155. MR1856509 (2002f:22016) ———, On the arithmetic structure of lattice actions on compact spaces, Ergodic Theory Dynam. Systems 22 (2002), no. 4, 1141–1168. MR1926279 (2004j:37004) ———, Rigid geometric structures and representations of fundamental groups, Rigidity in dynamics and geometry (Cambridge, 2000), Springer, Berlin, 2002, pp. 135–147. MR1919398 (2003g:22007) ———, Out(Fn ) and the spectral gap conjecture, Int. Math. Res. Not. (2006), Art. ID 26028, 9. MR2250018 (2007f:22006) ———, Local rigidity of group actions: past, present, future, Dynamics, ergodic theory, and geometry, Math. Sci. Res. Inst. Publ., Vol. 54, Cambridge University Press, Cambridge, 2007, pp. 45–97. MR2369442 ———, Deformations of group actions, Trans. Amer. Math. Soc. 360 (2008), no. 1, 491–505 (electronic). MR2342012 David Fisher and Theron Hitchman, Harmonic maps with inﬁnite dimensional targets and cocycle superrigidity, in preparation. ———, Cocycle superrigidity and harmonic maps with inﬁnite-dimensional targets, Int. Math. Res. Not. (2006), 72405, 1–19. MR2211160 David Fisher and G. A. Margulis, Local rigidity for cocycles, Surveys in differential geometry, Vol. VIII (Boston, MA, 2002), Surv. Differ. Geom., VIII, International Press, Somerville, MA, 2003, pp. 191–234. MR2039990 (2004m:22032) David Fisher and Gregory Margulis, Almost isometric actions, property (T ), and local rigidity, Invent. Math. 162 (2005), no. 1, 19–80. MR2198325 David Fisher and Lior Silberman, Groups not acting on manifolds, Int. Math. Res. Not. (2008). David Fisher and Kevin Whyte, Continuous quotients for lattice actions on compact spaces, Geom. Dedicata 87 (2001), no. 1–3, 181–189. MR1866848 (2002j:57070)

groups acting on manifolds / 149 [85] [86] [87] [88] [89] [90]

[91] [92]

[93] [94]

[95]

[96]

[97] [98]

[99]

[100] [101] [102] [103] [104] [105]

David Fisher and Robert J. Zimmer, Geometric lattice actions, entropy and fundamental groups, Comment. Math. Helv. 77 (2002), no. 2, 326–338. MR1915044 (2003e:57056) John Franks and Michael Handel, Area preserving group actions on surfaces, Geom. Topol. 7 (2003), 757–771 (electronic). MR2026546 (2004j:37042) ———, Periodic points of Hamiltonian surface diffeomorphisms, Geom. Topol. 7 (2003), 713–756 (electronic). MR2026545 (2004j:37101) ———, Distortion elements in group actions on surfaces, Duke Math. J. 131 (2006), no. 3, 441–468. MR2219247 (2007c:37018) Alex Furman, A survey of measured group theory, in this volume. ———, Random walks on groups and random transformations, Handbook of dynamical systems, Vol. 1A, North-Holland, Amsterdam, 2002, pp. 931–1014. MR1928529 (2003j:60065) Alex Furman and Nicolas Monod, Product groups acting on manifolds, preprint 2007. H. Furstenberg, Recurrence in ergodic theory and combinatorial number theory, Princeton University Press, Princeton, NJ, 1981, M. B. Porter Lectures. MR603625 (82j: 28010) Harry Furstenberg, A Poisson formula for semi-simple Lie groups, Ann. of Math. (2) 77 (1963), 335–386. MR0146298 (26 #3820) ———, Boundaries of Lie groups and discrete subgroups, Actes du Congrès International des Mathématiciens (Nice, 1970), Tome 2, Gauthier-Villars, Paris, 1971, pp. 301–306. MR0430160 (55 #3167) ———, Rigidity and cocycles for ergodic actions of semisimple Lie groups (after G. A. Margulis and R. Zimmer), Bourbaki Seminar, Vol. 1979/80, Lecture Notes in Math., vol. 842, Springer, Berlin, 1981, pp. 273–292. MR636529 (83j:22003) Jean-Marc Gambaudo and Étienne Ghys, Commutators and diffeomorphisms of surfaces, Ergodic Theory Dynam. Systems 24 (2004), no. 5, 1591–1617. MR2104597 (2006d:37071) Tsachik Gelander, On deformtions of Fn in compact Lie groups, Israel J. Math. 167 (2008), 15–26. MR2448015 (2009j:22011) Tsachik Gelander, Anders Karlsson, and Gregory A. Margulis, Superrigidity, generalized harmonic maps and uniformly convex spaces, Geom. Funct. Anal. 17 (2008), no. 5, 1524– 1550. MR2377496 Étienne Ghys, Groupes d’homéomorphismes du cercle et cohomologie bornée, The Lefschetz centennial conference, Part III (Mexico City, 1984), Contemp. Math., vol. 58, American Mathematical Society, Providence, RI, 1987, pp. 81–106. MR893858 (88m:58024) ———, Sur les groupes engendrés par des difféomorphismes proches de l’identité, Bol. Soc. Brasil. Mat. (N.S.) 24 (1993), no. 2, 137–178. MR1254981 (95f:58017) ———, Actions de réseaux sur le cercle, Invent. Math. 137 (1999), no. 1, 199–231. MR1703323 (2000j:22014) ———, Groups acting on the circle, Enseign. Math. (2) 47 (2001), no. 3–4, 329–407. MR1876932 (2003a:37032) Étienne Ghys and Vlad Sergiescu, Sur un groupe remarquable de difféomorphismes du cercle, Comment. Math. Helv. 62 (1987), no. 2, 185–239. MR896095 (90c:57035) Edward R. Goetze, Connection preserving actions of connected and discrete Lie groups, J. Differential Geom. 40 (1994), no. 3, 595–620. MR1305982 (95m:57052) Edward R. Goetze and Ralf J. Spatzier, On Livˇsic’s theorem, superrigidity, and Anosov actions of semisimple Lie groups, Duke Math. J. 88 (1997), no. 1, 1–27. MR1448015 (98d:58134)

150 / d. fisher [106] ———, Smooth classiﬁcation of Cartan actions of higher rank semisimple Lie groups and their lattices, Ann. of Math. (2) 150 (1999), no. 3, 743–773. MR1740993 (2001c:37029) [107] William M. Goldman, An ergodic action of the outer automorphism group of a free group, Geom. Funct. Anal. 17 (2007), no. 3, 793–805. MR2346275 (2008g:57001) [108] Alexander Gorodnik, Open problems in dynamics and related ﬁelds, J. Mod. Dyn. 1 (2007), no. 1, 1–35. MR2261070 (2007f:37001) [109] R. I. Grigorchuk, Degrees of growth of ﬁnitely generated groups and the theory of invariant means, Izv. Akad. Nauk SSSR Ser. Mat. 48 (1984), no. 5, 939–985. MR764305 (86h:20041) [110] R. I. Grigorchuk and A. Maki, On a group of intermediate growth that acts on a line by homeomorphisms, Mat. Zametki 53 (1993), no. 2, 46–63. MR1220809 (94c:20008) [111] M. Gromov, Hyperbolic groups, Essays in group theory, Math. Sci. Res. Inst. Publ., Vol. 8, Springer, New York, 1987, pp. 75–263. MR919829 (89e:20070) [112] Michael Gromov, Rigid transformations groups, Géométrie différentielle (Paris, 1986), Travaux en Cours, Vol. 33, Hermann, Paris, 1988, pp. 65–139. MR955852 (90d:58173) [113] Mikhail Gromov and Richard Schoen, Harmonic maps into singular spaces and p-adic superrigidity for lattices in groups of rank one, Inst. Hautes Études Sci. Publ. Math. (1992), no. 76, 165–246. MR1215595 (94e:58032) [114] Alain Guichardet, Sur la cohomologie des groupes topologiques. II, Bull. Sci. Math. (2) 96 (1972), 305–332. MR0340464 (49 #5219) [115] Sigurdur Helgason, Differential geometry, Lie groups, and symmetric spaces, Graduate Studies in Mathematics, Vol. 34, American Mathematical Society, Providence, RI, 2001, Corrected reprint of the 1978 original. MR1834454 (2002b:53081) [116] Michael-Robert Herman, Simplicité du groupe des difféomorphismes de classe C ∞ , isotopes à l’identité, du tore de dimension n, C. R. Acad. Sci. Paris Sér. A-B 273 (1971), A232–A234. MR0287585 (44 #4788) [117] Graham Higman, Finitely presented inﬁnite simple groups, Department of Pure Mathematics, Department of Mathematics, I.A.S. Australian National University, Canberra, 1974, Notes on Pure Mathematics, no. 8 (1974). MR0376874 (51 #13049) [118] M. W. Hirsch, C. C. Pugh, and M. Shub, Invariant manifolds, Lecture Notes in Mathematics, Vol. 583, Springer-Verlag, Berlin, 1977. MR0501173 (58 #18595) [119] Steve Hurder, personal communication 2007. [120] Steven Hurder, A survey of rigidity theory for Anosov actions, Differential topology, foliations, and group actions (Rio de Janeiro, 1992), Contemp. Math., Vol. 161, American Mathematical Society, Providence, RI, 1994, pp. 143–173. MR1271833 (95b: 58112) [121] Dennis Johnson and John J. Millson, Deformation spaces associated to compact hyperbolic manifolds, Discrete groups in geometry and analysis (New Haven, CT, 1984), Progr. Math., Vol. 67, Birkhäuser, Boston, 1987, pp. 48–106. MR900823 (88j:22010) [122] Jürgen Jost and Shing-Tung Yau, Harmonic maps and superrigidity, Differential geometry: partial differential equations on manifolds (Los Angeles, CA, 1990), Proc. Sympos. Pure Math., Vol. 54, American Mathematical Society, Providence, RI, 1993, pp. 245– 280. MR1216587 (94m:58060) [123] Jürgen Jost and Kang Zuo, Harmonic maps of inﬁnite energy and rigidity results for representations of fundamental groups of quasiprojective varieties, J. Differential Geom. 47 (1997), no. 3, 469–503. MR1617644 (99a:58046) [124] Vadim A. Kaimanovich and Howard Masur, The Poisson boundary of the mapping class group, Invent. Math. 125 (1996), no. 2, 221–264. MR1395719 (97m:32033)

groups acting on manifolds / 151 [125] Boris Kalinin, Anatole Katok, and Federico Rodriguez-Hertz, Non-uniform measure rigidity, preprint 2009. [126] Boris Kalinin and Ralf Spatzier, On the classiﬁcation of Cartan actions, Geom. Funct. Anal. 17 (2007), no. 2, 468–490. MR2322492 (2008i:37054) [127] Martin Kassabov, personal communication, 2008. [128] A. Katok and J. Lewis, Global rigidity results for lattice actions on tori and new examples of volume-preserving actions, Israel J. Math. 93 (1996), 253–280. MR1380646 (96k: 22021) [129] A. Katok, J. Lewis, and R. Zimmer, Cocycle superrigidity and rigidity for lattice actions on tori, Topology 35 (1996), no. 1, 27–38. MR1367273 (97e:22009) [130] A. Katok and R. J. Spatzier, Differential rigidity of Anosov actions of higher rank Abelian groups and algebraic lattice actions, Tr. Mat. Inst. Steklova 216 (1997), no. Din. Sist. i Smezhnye Vopr., 292–319. MR1632177 (99i:58118) [131] D. A. Každan, On the connection of the dual space of a group with the structure of its closed subgroups, Funkcional. Anal. i Priložen. 1 (1967), 71–74. MR0209390 (35 #288) [132] David Kazhdan, Some applications of the Weil representation, J. Analyse Mat. 32 (1977), 235–248. MR0492089 (58 #11243) [133] Bruce Kleiner, The asymptotic geometry of negatively curved spaces: uniformization, geometrization and rigidity, International Congress of Mathematicians. Vol. II, European Mathematical Society, Zürich, 2006, pp. 743–768. MR2275621 (2007k:53054) [134] Anthony W. Knapp, Lie groups beyond an introduction, Progress in Mathematics, vol. 140, Birkhäuser, Boston, 1996. MR1399083 (98b:22002) [135] Shoshichi Kobayashi, Transformation groups in differential geometry, Classics in Mathematics, Springer-Verlag, Berlin, 1995, Reprint of the 1972 edition. MR1336823 (96c:53040) [136] Toshiyuki Kobayashi, Discontinuous groups for non-Riemannian homogeneous spaces, Mathematics unlimited—2001 and beyond, Springer, Berlin, 2001, pp. 723–747. MR1852186 (2002f:53086) [137] Toshiyuki Kobayashi and Taro Yoshino, Compact Clifford-Klein forms of symmetric spaces—revisited, Pure Appl. Math. Q. 1 (2005), no. 3, part 2, 591–663. MR2201328 (2007h:22013) [138] Nicholas Korevaar and Richard Schoen, Global existence theorems for harmonic maps: ﬁnite rank spaces and approaches to rigidity of group actions, preprint 1999. [139] Nicholas J. Korevaar and Richard M. Schoen, Sobolev spaces and harmonic maps for metric space targets, Comm. Anal. Geom. 1 (1993), no. 3–4, 561–659. MR1266480 (95b:58043) [140] ———, Global existence theorems for harmonic maps to non-locally compact spaces, Comm. Anal. Geom. 5 (1997), no. 2, 333–387. MR1483983 (99b:58061) [141] Nadine Kowalsky, Noncompact simple automorphism groups of Lorentz manifolds and other geometric manifolds, Ann. of Math. (2) 144 (1996), no. 3, 611–640. MR1426887 (98g:57059) [142] F. Labourie, Quelques résultats récents sur les espaces localement homogènes compacts, Manifolds and geometry (Pisa, 1993), Sympos. Math., XXXVI, Cambridge University Press, Cambridge, 1996, pp. 267–283. MR1410076 (97h:53055) [143] F. Labourie, S. Mozes, and R. J. Zimmer, On manifolds locally modelled on nonRiemannian homogeneous spaces, Geom. Funct. Anal. 5 (1995), no. 6, 955–965. MR136 1517 (97j:53053)

152 / d. fisher [144] François Labourie, Large groups actions on manifolds, Proceedings of the International Congress of Mathematicians, Vol. II (Berlin, 1998), no. Extra Vol. II, 1998, pp. 371–380 (electronic). MR1648087 (99k:53069) [145] François Labourie and Robert J. Zimmer, On the non-existence of cocompact lattices for SL(n)/SL(m), Math. Res. Lett. 2 (1995), no. 1, 75–77. MR1312978 (96d:22014) [146] Lucy Lifschitz and Dave Witte Morris, Bounded generation and lattices that cannot act on the line, Pure Appl. Math. Q. 4 (2008), no. 1, part 2, 99–126. MR2405997 (2009m: 22106) [147] Elon Lindenstrauss, Rigidity of multiparameter actions, Israel J. Math. 149 (2005), 199– 226, Probability in mathematics. MR2191215 (2006j:37007) [148] Ron Livne, On certain covers of the universal elliptic curve, unpublished Harvard PhD thesis. [149] Alexander Lubotzky, Discrete groups, expanding graphs and invariant measures, Progress in Mathematics, Vol. 125, Birkhäuser Verlag, Basel, 1994, with an appendix by Jonathan D. Rogawski. MR1308046 (96g:22018) [150] ———, Free quotients and the ﬁrst Betti number of some hyperbolic manifolds, Transform. Groups 1 (1996), no. 1–2, 71–82. MR1390750 (97d:57016) [151] Alexander Lubotzky, Shahar Mozes, and M. S. Raghunathan, The word and Riemannian metrics on lattices of semisimple groups, Inst. Hautes Études Sci. Publ. Math. (2000), no. 91, 5–53 (2001). MR1828742 (2002e:22011) [152] Alexander Lubotzky and Robert J. Zimmer, A canonical arithmetic quotient for simple Lie group actions, Lie groups and ergodic theory (Mumbai, 1996), Tata Inst. Fund. Res. Stud. Math., Vol. 14, Tata Institute of Fundamental Research, Bombay, 1998, pp. 131–142. MR1699362 (2000m:22013) [153] ———, Arithmetic structure of fundamental groups and actions of semisimple Lie groups, Topology 40 (2001), no. 4, 851–869. MR1851566 (2002f:22017) [154] George W. Mackey, Ergodic theory and virtual groups, Math. Ann. 166 (1966), 187–207. MR0201562 (34 #1444) ˘ [155] Iozhe Maleshich, The Hilbert-Smith conjecture for Hölder actions, Uspekhi Mat. Nauk 52 (1997), no. 2(314), 173–174. MR1480156 (99d:57026) [156] Ricardo Mañé, Quasi-Anosov diffeomorphisms and hyperbolic manifolds, Trans. Amer. Math. Soc. 229 (1977), 351–370. MR0482849 (58 #2894) [157] L. N. Mann and J. C. Su, Actions of elementary p-groups on manifolds, Trans. Amer. Math. Soc. 106 (1963), 115–126. MR0143840 (26 #1390) [158] G. A. Margulis, Arithmetic properties of discrete subgroups, Uspehi Mat. Nauk 29 (1974), no. 1 (175), 49–98. MR0463353 (57 #3306a) [159] ———, Discrete groups of motions of manifolds of nonpositive curvature, Proceedings of the International Congress of Mathematicians (Vancouver, BC, 1974), Vol. 2, Canadian Mathematical Congress, Montreal, Quetec, 1975, pp. 21–34. MR0492072 (58 #11226) [160] ———, Non-uniform lattices in semisimple algebraic groups, Lie groups and their representations (Proc. Summer School on Group Representations of the Bolyai János Math. Soc., Budapest, 1971), Halsted, New York, 1975, pp. 371–553. MR0422499 (54 #10486) [161] ———, Arithmeticity of the irreducible lattices in the semisimple groups of rank greater than 1, Invent. Math. 76 (1984), no. 1, 93–120. MR739627 (85j:22021) [162] ———, Discrete subgroups of semisimple Lie groups, Ergebnisse der Mathematik und ihrer Grenzgebiete (3) [Results in Mathematics and Related Areas (3)], Vol. 17, Springer-Verlag, Berlin, 1991. MR1090825 (92h:22021)

groups acting on manifolds / 153 [163] Gregory Margulis, Free subgroups of the homeomorphism group of the circle, C. R. Acad. Sci. Paris Sér. I Math. 331 (2000), no. 9, 669–674. MR1797749 (2002b:37034) [164] ———, Problems and conjectures in rigidity theory, Mathematics: frontiers and perspectives, American Mathematical Society, Providence, RI, 2000, pp. 161–174. MR1754775 (2001d:22008) [165] Gregory A. Margulis and Nantian Qian, Rigidity of weakly hyperbolic actions of higher real rank semisimple Lie groups and their lattices, Ergodic Theory Dynam. Systems 21 (2001), no. 1, 121–164. MR1826664 (2003a:22019) [166] John N. Mather, Commutators of diffeomorphisms, Comment. Math. Helv. 49 (1974), 512–528. MR0356129 (50 #8600) [167] ———, Simplicity of certain groups of diffeomorphisms, Bull. Amer. Math. Soc. 80 (1974), 271–273. MR0339268 (49 #4028) [168] ———, Commutators of diffeomorphisms. II, Comment. Math. Helv. 50 (1975), 33–40. MR0375382 (51 #11576) [169] ———, Foliations and local homology of groups of diffeomorphisms, Proceedings of the International Congress of Mathematicians (Vancouver, BC, 1974), Vol. 2, Canadian Mathematical Congress, Montreal, Quebec, 1975, pp. 35–37. MR0431203 (55 #4205) [170] Yozô Matsushima and Shingo Murakami, On vector bundle valued harmonic forms and automorphic forms on symmetric riemannian manifolds, Ann. of Math. (2) 78 (1963), 365–416. MR0153028 (27 #2997) [171] Ngaiming Mok, Yum Tong Siu, and Sai-Kee Yeung, Geometric superrigidity, Invent. Math. 113 (1993), no. 1, 57–83. MR1223224 (94h:53079) [172] Nicolas Monod, Superrigidity for irreducible lattices and geometric splitting, J. Amer. Math. Soc. 19 (2006), no. 4, 781–814 (electronic). MR2219304 (2007b:22025) [173] Dave Witte Morris, Can lattices in SL(n, R) act on the circle? in this volume. [174] ———, Amenable groups that act on the line, Algebr. Geom. Topol. 6 (2006), 2509–2518. MR2286034 (2008c:20078) [175] ———, An introduction to arithmetic groups, book in preprint form 2007. [176] Dave Witte Morris and Robert J. Zimmer, Ergodic actions of semisimple Lie groups on compact principal bundles, Geom. Dedicata 106 (2004), 11–27. MR2079831 (2005e: 22016) [177] G. D. Mostow, Quasi-conformal mappings in n-space and the rigidity of hyperbolic space forms, Inst. Hautes Études Sci. Publ. Math. (1968), no. 34, 53–104. MR0236383 (38 #4679) [178] ———, Strong rigidity of locally symmetric spaces, Princeton University Press, Princeton, NJ, 1973, Annals of Mathematics Studies, No. 78. MR0385004 (52 #5874) [179] ———, personal communication, 2000. [180] I. Mundet Rieri, Jordan’s theorem for the diffeomorphism group of some manifolds, preprint 2008. [181] Andrés Navas, Groups of circle diffeomorphisms, book in preprint form, 2008. [182] ———, On the dynamics of (left) orderable groups, preprint 2007. [183] ———, Actions de groupes de Kazhdan sur le cercle, Ann. Sci. École Norm. Sup. (4) 35 (2002), no. 5, 749–758. MR1951442 (2003j:58013) [184] ———, Quelques nouveaux phénomènes de rang 1 pour les groupes de difféomorphismes du cercle, Comment. Math. Helv. 80 (2005), no. 2, 355–375. MR2142246 (2006j:57003) [185] ———, Growth of groups and diffeomorphisms of the interval, Geom. Funct. Anal. 18 (2008), no. 3, 988–1028. MR2439001

154 / d. fisher [186] Amos Nevo and Robert J. Zimmer, Invariant rigid geometric structures and smooth projective factors, Geom. Funct. Anal. 19 (2009), no. 2, 520–535. MR2545248 (2010h: 22029) [187] ———, Homogenous projective factors for actions of semi-simple Lie groups, Invent. Math. 138 (1999), no. 2, 229–252. MR1720183 (2000h:22006) [188] ———, Rigidity of Furstenberg entropy for semisimple Lie group actions, Ann. Sci. École Norm. Sup. (4) 33 (2000), no. 3, 321–343. MR1775184 (2001k:22009) [189] ———, Actions of semisimple Lie groups with stationary measure, Rigidity in dynamics and geometry (Cambridge, 2000), Springer, Berlin, 2002, pp. 321–343. MR1919409 (2003j:22029) [190] ———, A generalization of the intermediate factors theorem, J. Anal. Math. 86 (2002), 93–104. MR1894478 (2003f:22019) [191] ———, A structure theorem for actions of semisimple Lie groups, Ann. of Math. (2) 156 (2002), no. 2, 565–594. MR1933077 (2003i:22024) [192] ———, Entropy of stationary measures and bounded tangential de-Rham cohomology of semisimple Lie group actions, Geom. Dedicata 115 (2005), 181–199. MR2180047 (2006k: 22008) [193] Yann Ollivier, A January 2005 invitation to random groups, Ensaios Matemáticos [Mathematical Surveys], Vol. 10, Sociedade Brasileira de Matemática, Rio de Janeiro, 2005. MR2205306 [194] Vladimir Platonov and Andrei Rapinchuk, Algebraic groups and number theory, Pure and Applied Mathematics, Vol. 139, Academic Press, Boston, 1994, Translated from the 1991 Russian original by Rachel Rowen. MR1278263 (95b:11039) [195] Leonid Polterovich, The geometry of the group of symplectic diffeomorphisms, Lectures in Mathematics ETH Zürich, Birkhäuser Verlag, Basel, 2001. MR1826128 (2002g: 53157) [196] ———, Growth of maps, distortion in groups and symplectic geometry, Invent. Math. 150 (2002), no. 3, 655–686. MR1946555 (2003i:53126) [197] Sorin Popa, Deformation and rigidity for group actions and von Neumann algebras, International Congress of Mathematicians. Vol. I, European Mathematical Society, Zürich, 2007, pp. 445–477. MR2334200 (2008k:46186) [198] Pierre Py, Some remarks on area-preserving actions of lattices, in this volume. [199] ———, Quasi-morphismes de Calabi et graphe de Reeb sur le tore, C. R. Math. Acad. Sci. Paris 343 (2006), no. 5, 323–328. MR2253051 (2007e:53116) [200] ———, Quasi-morphismes et invariant de Calabi, Ann. Sci. École Norm. Sup. (4) 39 (2006), no. 1, 177–195. MR2224660 (2007f:53116) [201] R. Quiroga-Barranco and A. Candel, Rigid and ﬁnite type geometric structures, Geom. Dedicata 106 (2004), 123–143. MR2079838 (2005d:53068) [202] M. Raghunathan, Diskretnye podgruppy grupp Li, Izdat. “Mir,” Moscow, 1977, Translated from the English by O. V. Švarcman, Edited by È. B. Vinberg, with a supplement “Arithmeticity of irreducible lattices in semisimple groups of rank greater than 1” by G. A. Margulis. MR0507236 (58 #22394b) [203] M. S. Raghunathan, On the ﬁrst cohomology of discrete subgroups of semisimple Lie groups, Amer. J. Math. 87 (1965), 103–139. MR0173730 (30 #3940) [204] Marina Ratner, On Raghunathan’s measure conjecture, Ann. of Math. (2) 134 (1991), no. 3, 545–607. MR1135878 (93a:22009) [205] Julio C. Rebelo, On nilpotent groups of real analytic diffeomorphisms of the torus, C. R. Acad. Sci. Paris Sér. I Math. 331 (2000), no. 4, 317–322. MR1787192 (2001m:22041)

groups acting on manifolds / 155 [206] Julio C. Rebelo and Ana L. Silva, On the Burnside problem in Diff (M), Discrete Contin. Dyn. Syst. 17 (2007), no. 2, 423–439. MR2257443 (2007j:53107) ˘ cepin, A proof of the Hilbert-Smith conjecture for actions by [207] Du˘san Repov˘s and Evgenij S˘ Lipschitz maps, Math. Ann. 308 (1997), no. 2, 361–364. MR1464908 (99c:57066) [208] Derek J. S. Robinson, A course in the theory of groups, second ed., Graduate Texts in Mathematics, Vol. 80, Springer-Verlag, New York, 1996. MR1357169 (96f:20001) [209] Federico Rodriguez Hertz, Global rigidity of certain Abelian actions by toral automorphisms, J. Mod. Dyn. 1 (2007), no. 3, 425–442. MR2318497 (2008f:37063) [210] Leslie Saper, Tilings and ﬁnite energy retractions of locally symmetric spaces, Comment. Math. Helv. 72 (1997), no. 2, 167–202. MR1470087 (99a:22019) [211] B. Schmidt, Weakly hyperbolic actions of Kazhdan groups on tori, Geom. Funct. Anal. 16 (2006), no. 5, 1139–1156. MR2276535 (2008b:37045) [212] Atle Selberg, On discontinuous groups in higher-dimensional symmetric spaces, Contributions to function theory (Internat. Colloq. Function Theory, Bombay, 1960), Tata Institute of Fundamental Research, Bombay, 1960, pp. 147–164. MR0130324 (24 #A188) [213] Nimish A. Shah, Invariant measures and orbit closures on homogeneous spaces for actions of subgroups generated by unipotent elements, Lie groups and ergodic theory (Mumbai, 1996), Tata Inst. Fund. Res. Stud. Math., Vol. 14, Tata Institute of Fundamental Research, Bombay, 1998, pp. 229–271. MR1699367 (2001a:22012) [214] Yehuda Shalom, Rigidity, unitary representations of semisimple groups, and fundamental groups of manifolds with rank one transformation group, Ann. of Math. (2) 152 (2000), no. 1, 113–182. MR1792293 (2001m:22022) [215] ———, Measurable group theory, European Congress of Mathematics, European Mathematical Society, Zürich, 2005, pp. 391–423. MR2185757 (2006k:37007) [216] ———, The algebraization of Kazhdan’s property (T), International Congress of Mathematicians. Vol. II, European Mathematical Society, Zürich, 2006, pp. 1283–1310. MR2275645 (2008a:22003) [217] R. W. Sharpe, Differential geometry, Graduate Texts in Mathematics, Vol. 166, SpringerVerlag, New York, 1997, Cartan’s generalization of Klein’s Erlangen program, with a foreword by S. S. Chern. MR1453120 (98m:53033) [218] Daniel Shenﬁeld, Semisimple representations of SL(n, Z[X1 , . . . , Xk ]), to appear in Groups, Geometry and Dynamics. [219] Y. T. Siu, Remarks in talk at Margulis’ 60th birthday conference, February, 2006. [220] Yum Tong Siu, The complex-analyticity of harmonic maps and the strong rigidity of compact Kähler manifolds, Ann. of Math. (2) 112 (1980), no. 1, 73–111. MR584075 (81j:53061) [221] S. Smale, Differentiable dynamical systems, Bull. Amer. Math. Soc. 73 (1967), 747–817. MR0228014 (37 #3598) [222] R. J. Spatzier, Harmonic analysis in rigidity theory, Ergodic theory and its connections with harmonic analysis (Alexandria, 1993), London Math. Soc. Lecture Note Ser., Vol. 205, Cambridge University Press, Cambridge, 1995, pp. 153–205. MR1325698 (96c:22019) [223] ———, An invitation to rigidity theory, Modern dynamical systems and applications, Cambridge University Press, Cambridge, 2004, pp. 211–231. MR2090772 (2006a:53041) [224] Ralf J. Spatzier and Robert J. Zimmer, Fundamental groups of negatively curved manifolds and actions of semisimple groups, Topology 30 (1991), no. 4, 591–601. MR1133874 (92m:57047)

156 / d. fisher [225] Melanie Stein, Groups of piecewise linear homeomorphisms, Trans. Amer. Math. Soc. 332 (1992), no. 2, 477–514. MR1094555 (92k:20075) [226] Garrett Stuck, Minimal actions of semisimple groups, Ergodic Theory Dynam. Systems 16 (1996), no. 4, 821–831. MR1406436 (98a:57046) [227] William Thurston, Foliations and groups of diffeomorphisms, Bull. Amer. Math. Soc. 80 (1974), 304–307. MR0339267 (49 #4027) [228] William P. Thurston, A generalization of the Reeb stability theorem, Topology 13 (1974), 347–352. MR0356087 (50 #8558) [229] J. Tits, Free subgroups in linear groups, J. Algebra 20 (1972), 250–270. MR0286898 (44 #4105) [230] Alain Valette, Group pairs with property (T), from arithmetic lattices, Geom. Dedicata 112 (2005), 183–196. MR2163898 (2006d:22014) [231] André Weil, On discrete subgroups of Lie groups, Ann. of Math. (2) 72 (1960), 369–384. MR0137792 (25 #1241) [232] ———, On discrete subgroups of Lie groups. II, Ann. of Math. (2) 75 (1962), 578–602. MR0137793 (25 #1242) [233] ———, Remarks on the cohomology of groups, Ann. of Math. (2) 80 (1964), 149–157. MR0169956 (30 #199) [234] Shmuel Weinberger, Some remarks inspired by the C 0 Zimmer program, in this volume. [235] John S. Wilson, On exponential growth and uniformly exponential growth for groups, Invent. Math. 155 (2004), no. 2, 287–303. MR2031429 (2004k:20085) [236] Dave Witte, Arithmetic groups of higher Q-rank cannot act on 1-manifolds, Proc. Amer. Math. Soc. 122 (1994), no. 2, 333–340. MR1198459 (95a:22014) [237] ———, Measurable quotients of unipotent translations on homogeneous spaces, Trans. Amer. Math. Soc. 345 (1994), no. 2, 577–594. MR1181187 (95a:22005) [238] ———, Cocycle superrigidity for ergodic actions of non-semisimple Lie groups, Lie groups and ergodic theory (Mumbai, 1996), Tata Inst. Fund. Res. Stud. Math., Vol. 14, Tata Institute of Fundamental Research, Bombay, 1998, pp. 367–386. MR1699372 (2000i: 22008) [239] Sai-Kee Yeung, Representations of semisimple lattices in mapping class groups, Int. Math. Res. Not. (2003), no. 31, 1677–1686. MR1981481 (2004d:53047) [240] A. Zeghib, Isometry groups and geodesic foliations of Lorentz manifolds. I. Foundations of Lorentz dynamics, Geom. Funct. Anal. 9 (1999), no. 4, 775–822. MR1719606 (2001g: 53059) [241] ———, Isometry groups and geodesic foliations of Lorentz manifolds. II. Geometry of analytic Lorentz manifolds with large isometry groups, Geom. Funct. Anal. 9 (1999), no. 4, 823–854. MR1719610 (2001g:53060) [242] Abdelghani Zeghib, Le groupe afﬁne d’une variété riemannienne compacte, Comm. Anal. Geom. 5 (1997), no. 1, 199–211. MR1456311 (98g:53065) [243] ———, Sur les actions afﬁnes des groupes discrets, Ann. Inst. Fourier (Grenoble) 47 (1997), no. 2, 641–685. MR1450429 (98d:57068) [244] Robert J. Zimmer, Topological superrigidity, unpublished notes 1998. [245] ———, Ergodic theory and semisimple groups, Monographs in Mathematics, Vol. 81, Birkhäuser Verlag, Basel, 1984. MR776417 (86j:22014) [246] ———, Actions of lattices in semisimple groups preserving a G-structure of ﬁnite type, Ergodic Theory Dynam. Systems 5 (1985), no. 2, 301–306. MR796757 (87g:22011) [247] ———, Lattices in semisimple groups and distal geometric structures, Invent. Math. 80 (1985), no. 1, 123–137. MR784532 (86i:57056)

groups acting on manifolds / 157 [248] ———, On connection-preserving actions of discrete linear groups, Ergodic Theory Dynam. Systems 6 (1986), no. 4, 639–644. MR873437 (88g:57045) [249] ———, Actions of semisimple groups and discrete subgroups, Proceedings of the International Congress of Mathematicians, Vol. 1, 2 (Berkeley, CA, 1986), American Mathematical Society, Providence, RI, 1987, pp. 1247–1258. MR934329 (89j:22024) [250] ———, Lattices in semisimple groups and invariant geometric structures on compact manifolds, Discrete groups in geometry and analysis (New Haven, CT, 1984), Progr. Math., Vol. 67, Birkhäuser, Boston, 1987, pp. 152–210. MR900826 (88i:22025) [251] ———, Representations of fundamental groups of manifolds with a semisimple transformation group, J. Amer. Math. Soc. 2 (1989), no. 2, 201–213. MR973308 (90i:22021) [252] ———, On the algebraic hull of an automorphism group of a principal bundle, Comment. Math. Helv. 65 (1990), no. 3, 375–387. MR1069815 (92f:57050) [253] ———, Spectrum, entropy, and geometric structures for smooth actions of Kazhdan groups, Israel J. Math. 75 (1991), no. 1, 65–80. MR1147291 (93i:22014) [254] ———, Superrigidity, Ratner’s theorem, and fundamental groups, Israel J. Math. 74 (1991), no. 2–3, 199–207. MR1135234 (93b:22019) [255] ———, Automorphism groups and fundamental groups of geometric manifolds, Differential geometry: Riemannian geometry (Los Angeles, CA, 1990), Proc. Sympos. Pure Math., Vol. 54, American Mathematical Society, Providence, RI, 1993, pp. 693–710. MR1216656 (95a:58017) [256] ———, Discrete groups and non-Riemannian homogeneous spaces, J. Amer. Math. Soc. 7 (1994), no. 1, 159–168. MR1207014 (94e:22021) [257] ———, Entropy and arithmetic quotients for simple automorphism groups of geometric manifolds, Geom. Dedicata 107 (2004), 47–56. MR2110753 (2005i:22011) [258] Robert J. Zimmer and Dave Witte Morris, Ergodic theory, groups, and geometry, CBMS Regional Conference Series in Mathematics, 109. Published for the Conference Board of the Mathematical Sciences, Washington, DC, by the American Mathematical Society, Providence, RI, 2008. MR2457556 (2009k:37059)

5

dave witte morris

CAN LATTICES IN SL (n, R) ACT ON THE CIRCLE? to robert j. zimmer on his

60 th

birthday

1. Overview The following theorem is a simple example of the Zimmer program’s principle that large groups should not be able to act on small manifolds. (For a description of the Zimmer program, see D. Fisher’s survey article [Fis] elsewhere in this volume.) Unless stated otherwise, we assume actions are continuous, but we require no additional regularity.

theorem 1.1. (witte) Let be a ﬁnite-index subgroup of SL (n, Z), with n ≥ 3. Then has no faithful action on the circle S1 . remark 1.2. 1) The proof of this theorem is not at all difﬁcult, and will be given in Section 3, after the result is translated to a more algebraic form in Section 2. The proof illustrates the use of calculations with unipotent elements, which is a standard technique in the theory of arithmetic groups. 2) The assumption that n ≥ 3 is essential. Indeed, some ﬁnite-index subgroup of SL (2, Z) is a free group, which has countless actions on the circle (some faithful and some not). A group as in Theorem 1.1 is a lattice in SL (n, R). It is an open question whether the theorem generalizes to the other lattices:

conjecture 1.3. Let be a lattice in SL (n, R), with n ≥ 3. Then has no faithful action on S1 . Note that if this conjecture is true, then every action of on S1 or R has a nontrivial kernel. Since the kernel is a normal subgroup, and the Margulis 158

can lattices in sl (n, R) act on the circle? / 159

normal subgroup theorem tells us that lattices in SL (n, R) are “almost simple” (if n ≥ 3), it follows that the kernel is a ﬁnite-index subgroup of . Therefore, every -orbit is ﬁnite. Such reasoning shows that the conjecture can be restated as follows:

conjecture 1.3 . Let be a lattice in SL (n, R), with n ≥ 3. Whenever acts on S1 , every orbit is ﬁnite. There is considerable evidence for the above conjectures, including the following important theorem:

theorem 1.4. (ghys, burger-monod) If is any lattice in SL (n, R), with n ≥ 3, then every action of on S1 has at least one ﬁnite orbit. remark 1.5. The proof of É. Ghys utilizes amenability (or the Furstenberg boundary); see Sections 6 and 7. The proof of M. Burger and N. Monod is based on a vanishing theorem for bounded cohomology; see Section 8. The above conjectures are about continuous actions, but they can be weakened by considering only differentiable actions. It will be explained in Section 4 that combining the Ghys-Burger-Monod theorem with the well-known Reeb-Thurston stability theorem establishes these weaker conjectures:

corollary 1.6. (ghys, burger-monod) If is any lattice in SL (n, R), with n ≥ 3, then: 1) has no faithful C 1 action on S1 , and 2) whenever acts on S1 by C 1 diffeomorphisms, every orbit is ﬁnite. For n ≥ 3, it is well known that every lattice in SL (n, R) has Kazhdan’s property (T ) (see Deﬁnition 5.2), and the following result shows that if we strengthen the differentiability hypothesis slightly, then the conclusion is true for all groups that have that property. This is a very broad class of groups (including many groups that are not even linear), so one would expect a much stronger result to hold for the special case of lattices in SL (n, R). Thus, this theorem constitutes good evidence for Conjecture 1.3. The proof will be presented in Section 5; it is both elegant and elementary.

theorem 1.7. (navas) If is any inﬁnite, discrete group with Kazhdan’s property (T ), then has no faithful C 2 action on the circle.

160 / d. w. morris

Bounded generation provides another approach to proving Theorem 1.1 and some other cases of Conjecture 1.3. This strategy is explained in Section 9. In spite of the above results, Conjecture 1.3 remains completely open for cocompact lattices:

problem 1.8. Find a cocompact lattice in SL (n, R), for some n, such that no ﬁnite-index subgroup of has a faithful action on S1 . remark 1.9. A ﬁnal section of the paper (§10) brieﬂy discusses the generalization of Conjecture 1.3 to lattices in other semisimple Lie groups and two other topics: lattices that do act on the circle, and actions on trees. notes for §1. The survey of É. Ghys [Gh3] and the forthcoming book of A. Navas [Na2] are excellent introductions to the study of group actions on the circle. Versions of Conjecture 1.3 were discussed in conversation as early as 1990, but the ﬁrst published appearance may be in [Gh2, p. 200]. Expositions of the Margulis normal subgroup theorem can be found in [Ma2, chap. 4] and [Zi2, chap. 8]. Regarding the equivalence of Conjectures 1.3 and 1.3 , see [Uom, thm. 1] for a proof that if every -orbit on a connected manifold is ﬁnite, then the kernel of the action has ﬁnite index. acknowledgments: Preparation of this paper was partially supported by research grants from the Natural Sciences and Engineering Research Council of Canada and the U. S. National Science Foundation. I am grateful to M. Burger, É. Ghys, N. Monod, A. Navas, and an anonymous referee, for helpful comments on a preliminary version of the manuscript. 2. Algebraic Reformulation of the Conjecture In this section, we explain that Conjecture 1.3 can be reformulated in a purely algebraic form (see Proposition 2.8). The proof has two main parts: having an action on S1 is almost the same as having an action on R, and having an action on R is essentially the same as being left orderable.

definition 2.1. is left orderable if there is a left-invariant total order on . More precisely, there is an order relation ≺ on , such that 1) ≺ is a total order (that is, for all a, b ∈ , either a ≺ b or a b, or a = b); and

can lattices in sl (n, R) act on the circle? / 161

2) ≺ is invariant under multiplication on the left (that is, for all a, b, c ∈ , if a ≺ b, then ca ≺ cb).

exercise 2.2. Show is left orderable if and only if there exists a subset P of , such that 1) = P {e} P −1 (disjoint union), where P −1 = { a−1 | a ∈ P }; and 2) P is closed under multiplication (that is, ab ∈ P, for all a, b ∈ P). Thus, being left orderable is a property of the internal algebraic structure of . [Hint: A subset P as above is the “positive cone” of an order ≺. Given P, deﬁne a ≺ b if b−1 a ∈ P. Given ≺, deﬁne P = { a ∈ | a e }. ]

remark 2.3. The group is said to be right orderable if there is a rightinvariant total order on . The following exercise shows that the choice between “left orderable" and “right orderable" is entirely a matter of personal preference. exercise 2.4. Show that is left orderable if and only if is right orderable. [Hint: Deﬁne x ≺≺ y if x −1 ≺ y −1 . Alternatively, the conclusion can be derived from Exercise 2.2 and its analogue for right-orderable groups. ]

With this terminology, we can state the following algebraic conjecture, which will be seen to be equivalent to Conjecture 1.3.

conjecture 2.5. Let be a lattice in SL (n, R), with n ≥ 3. Then is not left orderable. As a tool for showing that Conjecture 2.5 is equivalent to Conjecture 1.3, let us give a geometric interpretation of being left orderable.

lemma 2.6. A countable group is left orderable if and only if there is an orientation-preserving, faithful action of on R. Proof. (⇐) For a, b ∈ , deﬁne a≺b

if

a(0) < b(0).

162 / d. w. morris

It is easy to see that ≺ is transitive and antisymmetric, so it deﬁnes a partial order on . For any c ∈ , the function x → c(x) is a strictly increasing function on R (because the action is orientation preserving). Hence, if a(0) < b(0), then c(a(0)) < c(b(0)). Therefore, ≺ is left-invariant (as required in Deﬁnition 2.1(2)). However, the relation ≺ may not be a total order, because it could happen that a(0) = b(0), even though a = b. It is not difﬁcult to modify the deﬁnition to deal with this technical point (see Exercise 2.7). (⇒) We leave this as an exercise for the reader.1

exercise 2.7. Complete the proof of Lemma 2.6(⇐), by deﬁning a left-invariant total order on . [Hint: Make a list q1 , q2 , . . . of the rational numbers and deﬁne a ≺ b if a(qn ) < b(qn ), where n is minimal with a(qn ) = b(qn ). ]

proposition 2.8. Conjecture 1.3 and 2.5 are equivalent. Proof. We show there exists a counterexample to Conjecture 1.3 if and only if there exists a counterexample to Conjecture 2.5. (⇐) Suppose is a counterexample to Conjecture 2.5, so is left orderable. Then Lemma 2.6 tells us that has a faithful action on R. This implies that acts faithfully on the one-point compactiﬁcation of R, which is homeomorphic to S1 . So is a counterexample to Conjecture 1.3. (⇒) Suppose is a counterexample to Conjecture 1.3, so has a faithful action on S1 . From the Ghys-Burger-Monod theorem (1.4), we know that the action has a ﬁnite orbit. Therefore, some subgroup of ﬁnite index in has a ﬁxed point p. Then acts faithfully on the complement S1 {p}, which is homeomorphic to R. Let be the subgroup of consisting of the elements that act by orientation-preserving homeomorphisms (so is either , or a subgroup of index 2 in ). Then Lemma 2.6 tells us that is left orderable. Furthermore, is a lattice in SL (n, R) (because it has ﬁnite index in ). Therefore, is a counterexample to Conjecture 2.5. 1. Hint: If the order relation (, ≺ ) is dense (that is, if a ≺ b ⇒ ∃c ∈ , a ≺ c ≺ b), then it is order-isomorphic to (Q, < ). The action of on (, ≺ ) by left multiplication extends to a continuous action of on the Dedekind completion, which is homeomorphic to R.

can lattices in sl (n, R) act on the circle? / 163

warning 2.9. The proof does not show that the assertions of the two conjectures are equivalent for each individual lattice . Rather, if one of the conjectures is valid for all lattices , then the other conjecture is also valid for all . More precisely, if one of the conjectures is valid for every ﬁnite-index subgroup of , then the other conjecture is also valid for . remark 2.10. In order to prove that Conjecture 2.5 implies Conjecture 1.3, we appealed to Theorem 1.4, which is quite deep. A stronger algebraic conjecture (that no central extension of is left orderable) can easily be shown to imply Conjecture 1.3: exercise 2.11. A central extension of is a group , such that /Z ∼ = , for some subgroup Z of the center of . Show that if has an orientationpreserving, faithful action on S1 , then some central extension of is left orderable. [Hint: Because R is the universal cover of S1 , every homeomorphism of S1 lifts to a homeomorphism of R, and the lift is unique modulo an element of the fundamental group Z. Let be the subgroup of Homeo+ (R) consisting of all of the possible lifts of all of the elements of , and show that is a central extension that is left orderable. ]

notes for §2. The material in this section is well known. See [KM] for a treatment of the algebraic theory of left-ordered groups. Informative discussions of orderings, actions on the line, and related topics appear in [Gh3, §6.5] and [Na2, §2.2.3–2.2.6]. 3. SL(n, Z ) Has no Faithful Action on the Circle In this section, we prove Theorem 1.1 by exploiting the interaction between some obvious nilpotent subgroups of . The crucial ingredients are Lemma 3.6 and the fact that the theorem can be restated in the following algebraic form (cf. Proposition 2.8).

theorem 3.1. (witte) If is a ﬁnite-index subgroup of SL (n, Z), with n ≥ 3, then is not left orderable. remark 3.2. It is easy to see that the group SL (n, Z) itself is not left orderable, because it has elements of ﬁnite order. But there are subgroups of ﬁnite index that do not have any elements of ﬁnite order, and we will show that they, too, are not left orderable.

164 / d. w. morris

definition 3.3. Suppose ≺ is a left-invariant order relation on . For elements a and b of , we say a is inﬁnitely smaller than b (denoted a b) if either ak ≺ b for all k ∈ Z or ak ≺ b−1 for all k ∈ Z. Notice that the relation is transitive. notation 3.4. The commutator a−1 b−1 ab of elements a and b of a group is denoted [a, b]. It is straightforward to check that a commutes with b if and only if [a, b] = e (the identity element of ). lemma 3.5. Let a and b be elements of . If [a, b] commutes with both a and b, then [bk , am ] = [a, b]−km for all k, m ∈ Z. Proof. Exercise (or see [Gor, Lems. 2.2.2(i) and 2.2.4(iii), pp. 19–20]).

lemma 3.6. (ault, rhemtulla) Suppose a, b, z are nonidentity elements of a left-ordered group H, with [a, b] = zk for some nonzero k ∈ Z, and [a, z] = [b, z] = e. Then either z a or z b. Proof. Assume, for simplicity, that a, b, z ≥ e and k > 0. (All other cases can be reduced to this one by replacing some or all of a, b, z with their inverses, and/or interchanging a with b.) We may assume z a, so a ≺ zp for some p ∈ Z+ . Similarly, we may assume b ≺ zq for some q ∈ Z+ . Then (using the left-invariance of ≺) we have e ≺ a−1 zp ,

e ≺ b−1 zq ,

e ≺ a,

e ≺ b,

e ≺ z.

Hence, for all m ∈ Z+ , we have e ≺ (b−1 zq )m (a−1 zp )m bm am = b−m a−m bm am z(p+q)m

(z commutes with a and b)

= [bm , am ]z(p+q)m = z−km z(p+q)m 2

= z−km

([a, b] = zk and see (lemma 3.5))

2 +(p+q)m

= znegative

(if m is sufﬁciently large)

≺ e. This is a contradiction.

can lattices in sl (n, R) act on the circle? / 165

Proof of Theorem 3.1. Suppose is left orderable. ( This will lead to a contradiction.) Because subgroups of left-orderable groups are left orderable, we may assume n = 3; that is, has ﬁnite index in SL (3, Z). Hence, there is some k ∈ Z+ such that the six matrices ⎡ ⎤ ⎡ ⎤ ⎡ ⎤ 1 k 0 1 0 k 1 0 0 ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ a1 = ⎣0 1 0⎦ , a2 = ⎣0 1 0⎦ , a3 = ⎣0 1 k⎦ , 0 0 1 0 0 1 0 0 1 3.7

⎡

1 ⎢ a4 = ⎣k 0

0 1 0

⎤ ⎡ 0 1 ⎥ ⎢ 0⎦ , a5 = ⎣0 1 k

0 1 0

⎤ ⎡ 0 1 ⎥ ⎢ 0 ⎦ , a 6 = ⎣0 1 0

0 1 k

⎤ 0 ⎥ 0⎦ 1

all belong to . A straightforward check shows that [ai , ai+1 ] = e and [ai−1 , ai+1 ] = (ai )±k for i = 1, . . . , 6, with subscripts read modulo 6. Thus, Lemma 3.6 asserts 3.8

either ai ai−1 or ai ai+1 .

In particular, we must have either a1 a6 or a1 a2 . Assume for deﬁniteness that a1 a2 . ( The other case is very similar.) For each i, Lemma 3.6 implies that if ai−1 ai , then ai ai+1 . Since a1 a2 , we conclude by induction that a1 a2 a3 a4 a5 a6 a1 . Thus a1 a1 , a contradiction.

For the reader acquainted with root systems, let us give a more conceptual presentation of the proof of Theorem 3.1. Alternate Version of the Proof of Theorem 3.1. The root system of SL (3, Z) is of type A2 , pictured in Figure 1. For i = 1, . . . , 6, let Ui be the root space of corresponding to the root αi . (Note that Ui ∼ = Z is cyclic. In the notation of the above proof, the element ai belongs to Ui .) Deﬁne αi αj if there exists u ∈ Uj , such that Ui ≺ u (that is, v ≺ u, for all v ∈ Ui ). It is obvious, from Figure 1, that αi−1 + αi+1 = αi , and that αi + αi+1 is not a root, so e = [Ui−1 , Ui+1 ] ⊂ Ui

and

[Ui , Ui+1 ] = e.

166 / d. w. morris a1 a6

a2

a5

a3 a4

Fig. 1. The root system of SL (3, Z) (type A2 ).

Hence, Lemma 3.6 implies that either αi αi−1 or αi αi+1 . Arguing as in the above proof, we conclude that α1 α2 α3 α4 α5 α6 α1 . Thus α1 α1 , a contradiction.

notes for §3. Theorem 3.1 is due to D. Witte [Wit]. (An exposition of the proof also appears in [Gh3, thm. 7.2].) Lemma 3.6 is a special case of a theorem of J. C. Ault [Aul] and A. H. Rhemtulla [Rhe]: if is any nontrivial, ﬁnitely generated, left-ordered nilpotent group, then there is some a ∈ , such that [, ] ≺ a. An elementary proof that SL (n, Z) has no nontrivial actions on the circle can be found in [BV], but the argument uses the existence of elements of ﬁnite order, so it does not apply to ﬁnite-index subgroups of SL (n, Z). 4. The Reeb-Thurston Stability Theorem The Ghys-Burger-Monod theorem (1.4) tells us that if is a lattice in G = SL (3, R), then any action of on S1 has a ﬁnite orbit; in other words, some ﬁnite-index subgroup of has a ﬁxed point. In order to deduce Corollary 1.6 from this it sufﬁces to show that if the action is by (orientation-preserving) C 1 diffeomorphisms, then acts trivially. This triviality of is immediate from the following result:

proposition 4.1. (reeb-thurston stability theorem) Suppose • is a ﬁnitely generated subgroup of Diff 1 (S1 ), + • has a ﬁxed point, and • the abelianization /[, ] is ﬁnite.

Then = {e} is trivial.

can lattices in sl (n, R) act on the circle? / 167

A differentiable action on S1 that has a ﬁxed point can be transformed into a differentiable action on the unit interval [0, 1] (cf. pf. of Proposition 2.8(⇒)). Thus, Proposition 4.1 can be reformulated as follows:

proposition 4.1 . (reeb-thurston stability theorem) Suppose • I = [0, 1] is the unit interval, • is a ﬁnitely generated subgroup of Diff 1 (I), and + • the abelianization /[, ] is ﬁnite.

Then = {e} is trivial.

Proof in a special case. Deﬁne σ : → R+ by σ (λ) = λ (0). From the Chain Rule, we see that σ is a (multiplicative) homomorphism. Because /[, ] is ﬁnite and R+ is abelian, this implies that σ () is ﬁnite. However, R+ has no nontrivial, ﬁnite subgroups, so this implies that σ () is trivial; therefore λ (0) = 1, for all λ ∈ . For simplicity, let us assume, henceforth, that each element of is real analytic, rather than merely C 1 . (We remark that there is no need to assume is ﬁnitely generated in the real-analytic case. The proof for C 1 actions is similar, but requires some additional effort, and does use the hypothesis that is ﬁnitely generated.) Thus, each element λ of can be expressed as a power series in a neighborhood of 0: λ(x) = x + aλ,2 x 2 + aλ,3 x 3 + · · · . ( There is no constant term, because λ(0) = 0; the coefﬁcient of x is 1, because λ (0) = 1.) Now suppose is nontrivial. (This will lead to a contradiction.) Then there exist n and λ, such that aλ,n = 0.

4.2

We may assume n is minimal; hence λ(x) = x + aλ,n x n + aλ,n+1 x n+1 + · · · for all λ ∈ . Then, for λ, γ ∈ , we have

168 / d. w. morris

x + aλγ ,n x n + O(x n+1 ) = (λγ )(x) = λ(γ (x)) = γ (x) + aλ,n (γ (x))n + O(x n+1 ) = (x + aγ ,n x n + O(x n+1 )) + aλ,n (x + aγ ,n x n + O(x n+1 ))n + O(x n+1 ) = x + (aγ ,n + aλ,n )x n + O(x n+1 ). Thus the map τ : → R, deﬁned by τ (λ) = aλ,n , is an additive homomorphism. Because /[, ] is ﬁnite, but R has no nontrivial ﬁnite subgroups, this implies that τ () is trivial; therefore aλ,n = 0 for every λ ∈ . This contradicts Equation (4.2).

remark 4.3 on the general case Suppose is nontrivial. Then there is no harm in assuming that acts nontrivially on every neighborhood of 0. Hence, letting 0 be a ﬁnite generating set for , we may choose λ0 ∈ 0 and a sequence xn → 0+ , such that, for all n, we have λ0 (xn ) = xn and |λ(xn ) − xn | ≤ |λ0 (xn ) − xn | for all λ ∈ 0 . By passing to a subsequence of {xn }, we may assume τ (λ) = lim

n→∞

λ(xn ) − xn λ0 (xn ) − xn

exists for all λ ∈ 0 . Because λ (0) = 1 and λ is C 1 , it can be shown that τ (λγ ) = τ (λ) + τ (γ ). Therefore τ : → R is a (nontrivial) homomorphism. This is a contradiction.

remark 4.4. (zimmer) Here is the outline of a nice proof of Proposition 4.1 , under the additional assumption that has Kazhdan’s property (T ). 1) It sufﬁces to show that the ﬁxed points of form a dense subset of I; thus, we may assume that 0 and 1 are the only points that are ﬁxed by . 2) We have γ (0) = 1 for all γ ∈ .

can lattices in sl (n, R) act on the circle? / 169

3) Deﬁne a unitary representation of on L2 (I) by f γ (t) = f (γ t) |γ (t)|1/2 . 4) For any γ ∈ and any δ > 0, if f is the characteristic function of a sufﬁciently small neighborhood of 0, then f γ − f < δ f . Hence, this unitary representation has almost-invariant vectors. 5) Because has Kazhdan’s property (T ), this implies there are ﬁxed vectors: there exists f ∈ L2 (I) {0}, such that f γ = f for all γ ∈ . 6) Every point in the essential support of f is a ﬁxed point of . 7) This is a contradiction.

notes for §4. Proposition 4.1 was proved by W. Thurston [Thu] in a more general form that also applies to actions on manifolds of higher dimension. (It generalizes a theorem of G. Reeb.) See [RS] or [Sch] for details of the proof sketched in Remark 4.3. The proof outlined in Remark 4.4 is due to R. J. Zimmer; details appear in [WZ, §5, pp. 108–109]. 5. Smooth Actions of Kazhdan Groups on the Circle In this section, we prove Theorem 1.7. First, let us recall one of the many equivalent deﬁnitions of Kazhdan’s property (T ).

notation 5.1. For any real Hilbert space H, we use Isom (H) to denote the isometry group of H. (We remark that each isometry of H is the composition of a translation with a norm-preserving linear transformation.) definition 5.2. We say that a discrete group has Kazhdan’s property ( T ) if, for every homomorphism ρ : → Isom (H), where H is a real Hilbert space, there exists v ∈ H, such that ρ(g)v = v, for every g ∈ . In short, to say that has Kazhdan’s property (T ) means that every isometric action of on any Hilbert space has a ﬁxed point. The importance of this notion for our purposes stems from the following result, whose proof we omit.

theorem 5.3. (kazhdan) If n ≥ 3, then every lattice in SL (n, R) has Kazhdan’s property (T ).

170 / d. w. morris

exercise 5.4. Show that if is an inﬁnite group with Kazhdan’s property (T ), then is not abelian. [Hint: Every group with Kazhdan’s property (T ) is ﬁnitely generated, and every ﬁnitely generated abelian group is either ﬁnite or has a quotient isomorphic to Z. ]

We now turn to the proof of Theorem 1.7. To simplify notation, we may think of S1 as [−π/2, π/2]. In particular, for any diffeomorphism of S1 and any x ∈ S1 , the derivative g (x) is a well-deﬁned real number.

definition 5.5. • Let F (S1 × S1 ) be the vector space of measurable functions on S1 × S1

(with two functions being identiﬁed if they are equal almost everywhere). • Deﬁne an action of Diff 2 (S1 ) on F (S1 × S1 ) by

F g (x, y) = F(g(x), g(y)) |g (x)|1/2 |g (y)|1/2 for F ∈ F (S1 × S1 ) and g ∈ Diff 2 (S1 ).

1/2 2 • Let F2 = be the L2 -norm of F; note that S1 S1 F(x, y) dx dy F2 = ∞ if F ∈ / L2 (S1 × S1 ). Note that F gh = (F g )h

and

F g 2 = F2

for F ∈ F (S1 × S1 ) and g, h ∈ Diff 2 (S1 ).

notation 5.6. • Choose a positive function f on S1 , such that

— f has a 1/x singularity at the point 0 of S1 , and — f is C ∞ everywhere else; that is, identifying S1 with [−π/2, π/2],we have 5.7

f (x) −

1 ∈ C ∞ (S1 ). |x|

For example, one may take f (x) = | cot x |. • Now deﬁne

(x, y) = f (x − y) on S1 × S1 . Because of the 1/x singularity of f , it is easy to see that ∈ / L2 (S1 × S1 ). For any g ∈ Diff 2 (S1 ), the following calculation shows that the singularity of g cancels the singularity of .

can lattices in sl (n, R) act on the circle? / 171

lemma 5.8. The difference g − is a bounded function on S1 × S1 , for any g ∈ Diff 2 (S1 ). Proof. From Equation (5.7), we see that there is no harm in working with the function 0 (x, y) = 1/|x − y|, instead of . Also, in order to reduce the number of absolute-value signs, let us assume g ≥ 0 everywhere. | g (x, y) − (x, y)| g

≈ | 0 (x, y) − 0 (x, y)| = g (x)1/2 g (y)1/2 0 (g(x), g(y)) − 0 (x, y) 1/2 1/2 g (x) g (y) 1 − = |g(x) − g(y)| |x − y| 1/2 1/2 g (x) g (y) 1 = − |g (t)(x − y)| |x − y| =

|g (x)1/2 g (y)1/2 − g (t)| g (t) |x − y|

|g (x)g (y) − g (t)2 | = 1/2 1/2 (g (x) g (y) + g (t))g (t)|x − y| =O

|g (x)g (y) − g (t)2 | |x − y|

∃t ∈ (x, y), by Mean Value thm.

multiply by conjugate of numerator

g diffeomorphism . ⇒ g is never 0

Now, we wish to show that the numerator is bounded by a constant multiple of the denominator. |g (x)g (y) − g (t)2 | ≤ g (x)|g (y) − g (t)| + g (t)|g (x) − g (t)| = g (x) |g (u)| |y − t| + g (t) |g (v)| |x − t| ≤ g (x) |g (u)| |x − y| + g (t) |g (v)| |x − y| = O(|x − y|)

( Triangle inequality)

Mean Value thm. applied to g (because t ∈ (x, y))

g , g continuous, . so bounded

We will also use the following classical fact:

172 / d. w. morris

Fig. 2. A rectangle Rn (shaded) in S1 × S1 .

lemma 5.9. (hölder, 1901) Every nonabelian group of homeomorphisms of S1 contains at least one nonidentity element that has a ﬁxed point. Proof of Theorem 1.7. Suppose is a discrete group with Kazhdan’s property (T ), and we have a faithful C 2 action of on the circle S1 . From Lemma 5.8, we see that 5.10

( + L2 (S1 × S1 ))g = + L2 (S1 × S1 ) for all g ∈ Diff 2 (S1 ).

Thus, acts (by isometries) on the afﬁne Hilbert space + L2 (S1 × S1 ). Because has Kazhdan’s property, we conclude that has a ﬁxed point F in + L2 (S1 × S1 ): F g = F for all g ∈ . / Because ∈ / L2 (S1 × S1 ) and F − ∈ L2 (S1 × S1 ), it is obvious that F ∈ 2 1 1 L (S × S ). Now, deﬁne a measure μ on S1 × S1 by μ = F 2 dx dy. Then 1) μ is a -invariant measure on S1 × S1 (because F is -invariant); and 2) if R = (a1 , b1 ) × (a2 , b2 ) is a rectangle in S1 × S1 , then, from the 1/x singularity of along the diagonal, we see that ⎧ ⎨∞ if R intersects the diagonal; μ(R) = ⎩ﬁnite if R is away from the diagonal. For example, in Figure 2, the rectangle (a, b) × (b, c) has inﬁnite measure because it touches the diagonal at (b, b), but the shaded rectangle Rk has ﬁnite measure, because it does not touch the diagonal. Because is not abelian, Lemma 5.9 tells us we may choose g ∈ , such that g has a ﬁxed point. Then, by passing to a triple cover of S1 , we obtain an

can lattices in sl (n, R) act on the circle? / 173

action of (a ﬁnite extension of ) in which g has at least three ﬁxed points: g(a) = a,

g(b) = b,

g(c) = c.

By perhaps replacing a and b with different ﬁxed points, we may assume, for x ∈ (a, b), that lim g k (x) =

⎧ ⎨b

as k → ∞

⎩a

as k → −∞.

For each k ∈ Z, deﬁne Rk = (a, g k (x)) × (b, c) (see Figure 2). Then g(Rk ) = Rk+1 , so μ(Rk ) = μ(Rk+1 ). Therefore 0=

∞

(μ(Rk+1 ) − μ(Rk ))

k=−∞

⎛

= μ⎝

∞

⎞ (Rk+1 Rk )⎠

k=−∞

= μ((a, b) × (b, c)). However, the rectangle (a, b) × (b, c) touches the diagonal at the point (b, b), so it has inﬁnite measure. This is a contradiction.

remark 5.11. In the proof of Theorem 1.7, the assumption that the elements of are C 2 was used only to establish Equation (5.10). For this, it is not necessary to show that g − is bounded (as in Lemma 5.8), but only that g − ∈ L2 (S1 × S1 ). The calculations in the proof of Lemma 5.8 show that this holds under the weaker hypothesis that g ∈ C 3/2+ , for any > 0. In fact, A. Navas observed that by using recent results that group with Kazhdan’s property (T ) also have ﬁxed points in certain Lp spaces with p = 2, it can be shown that C 2 can be replaced with C 3/2 in the statement of Theorem 1.7.

174 / d. w. morris

This leads to the following well-known question:

problem 5.12. Can an inﬁnite (discrete) group with Kazhdan’s property (T ) have a faithful C 0 action on S1 ? notes for §5. Theorem 1.7 was ﬁrst proved in [Na1], but the idea to use Lemma 5.8 came from earlier work of A. Pressley and G. Segal [PS], and A. Reznikov [Rez, chap. 2]. Proofs also appear in [BHV, §2.9] and [Na2, §5.2]. The condition in Deﬁnition 5.2 was introduced by J.–P. Serre, and it is not at all obvious that it is equivalent to the original deﬁnition of property (T ) that was given by D. Kazhdan. For a discussion of this, and much more, the standard reference on Kazhdan’s property (T ) is [BHV]. See [Gh3, thm. 6.10] or [Na2, §2.2.4] for a proof of Lemma 5.9. Implications of Kazhdan’s property (T ) for ﬁxed points of actions on Lp spaces (and other Banach spaces) are discussed in [BFGM]. 6. Ghys's Proof That Actions Have a Finite Orbit In this section, we present Ghys’s proof of Theorem 1.4, modulo some facts that will be proved in Section 7. To get started, let us show that it sufﬁces to ﬁnd a -invariant measure on the circle.

definition 6.1. A measure μ on a measure space X is a probability measure if μ(X ) = 1.

lemma 6.2. If • is a discrete group, such that the abelianization /[, ] is ﬁnite, • acts on S1 by orientation-preserving homeomorphisms, and • there is a -invariant probability measure μ on S1 ,

then has at least one ﬁnite orbit on S1 . Proof. We consider two cases. Case 1. Assume μ has at least one atom p. Let p = { gp | g ∈ } be the orbit of p. Because μ is -invariant, we have μ(gp) = μ(p), for every g ∈ . Therefore μ(p) = #(p) μ(p).

can lattices in sl (n, R) act on the circle? / 175

Since μ is a probability measure, we know that μ(p) < ∞, so we conclude that p is a ﬁnite set. That is, the orbit of p is ﬁnite. Case 2. Assume μ has no atoms. To simplify the proof, let us assume that the support of μ is all of S1 . (In other words, every nonempty open subset of S1 has positive measure.) For x, y ∈ S1 , deﬁne d(x, y) = μ([x, y]),

6.3

where [x, y] is a path from x to y, and, for a given x and y, we choose the path [x, y] to minimize μ([x, y]). It is easy to see that d is a metric on S1 . Up to isometry, there is a unique metric on S1 , so we may assume that d is the usual arc-length metric. Because μ is -invariant, we know that d is -invariant, so acts by rotations of the circle. There is no harm in assuming that the action is faithful, so we conclude that is abelian. But every abelian quotient of is ﬁnite, so we conclude that is ﬁnite. Hence, every orbit is ﬁnite.

exercise 6.4. Complete the proof of Lemma theorem 6.2, by eliminating the assumption that the support of μ is all of S1 in Case 2. [Hint: Modify the above proof to show that every orbit in the support of μ is ﬁnite. ]

To simplify the notation, we assume henceforth that n = 3.

notation 6.5. Let • G = SL (3, R), • be a lattice in G, and

⎡ ⎤ ∗ ∗ ∗ ⎥ • P=⎢ ∗ ∗⎦ ⊂ G. ⎣ ∗

We remark that P is a minimal parabolic subgroup of G. Ghys’s proof is based on the following key fact that will be established in §7A by using the fact that the group P is “amenable.”

notation 6.6. Prob (X ) denotes the set of all Radon probability measures on X (where X is any compact, Hausdorff space). This is a closed, convex subset of the unit ball in C(X )∗ , with the weak∗ topology, so Prob (X ) has a natural topology that makes it a compact Hausdorff space.

176 / d. w. morris

proposition 6.7. (furstenberg) If acts continuously on any compact metric space X , then there is a -equivariant, Borel measurable map ψ : G/P → Prob (X ). If ψ were invariant, rather than equivariant, the following fundamental theorem would immediately imply that ψ is constant (a.e.). ( This theorem will be proved in §7B.)

theorem 6.8. (moore ergodicity theorem) If H is any closed, noncompact subgroup of G, then the action of on G/H is ergodic: by deﬁnition, this means that every -invariant, measurable function on G/H is constant (a.e.). The homogeneous space G/P plays a major role in Ghys’s proof, because of its appearance in Proposition 6.7. We will now present a geometric interpretation of this space that is very helpful. Recall that a ﬂag in R3 consists of a pair (, π), where • is a 1-dimensional vector subspace of R3 (a line), and • π is a 2-dimensional vector subspace of R3 (a plane) that contains .

The group G = SL (3, R) acts transitively on the set of ﬂags, and P is the stabilizer of the standard ﬂag F0 = (0 , π0 ), where 0 = (∗, 0, 0) and π0 = (∗, ∗, 0). Therefore:

proposition 6.9. G/P can be identiﬁed with the space F of all ﬂags, by identifying gP with the ﬂag gF0 . Proof of Theorem 1.4. Suppose acts by homeomorphisms on the circle S1 . From Proposition 6.7, we know there is a -equivariant, measurable map ψ : G/P → Prob (S1 ). It will sufﬁce to show that ψ is constant (a.e.), for then • the essential range of ψ consists of a single point μ ∈ Prob (S1 ), and • the essential range of ψ is -invariant, because ψ is -equivariant.

So μ is -invariant, and then Lemma 6.2 implies has a ﬁnite orbit on S1 . There are two basic cases to consider: either the measure ψ(x) consists entirely of atoms, or ψ(x) has no atoms. (Recall that an atom of a measure μ is

can lattices in sl (n, R) act on the circle? / 177

a point p, such that μ({p}) = 0.) It is also possible that ψ(x) consists partly of atoms, and partly of nonatoms, but Corollary 7.5 below tells us that we need only consider the two extreme cases. Case 1. Assume ψ(x) has no atoms, for a.e. x ∈ G/P. • Let Prob0 (S1 ) = { μ ∈ Prob (S1 ) | μ has no atoms }. • By assumption, we know ψ : G/P → Prob0 (S1 ), so we may deﬁne

ψ 2 : (G/P)2 → (Prob0 (S1 ))2

by

ψ 2 (x, y) = (ψ(x), ψ(y)).

Then ψ 2 , like ψ, is measurable and -equivariant. • Deﬁne d : (Prob0 (S1 ))2 → R by

d(μ1 , μ2 ) = sup |μ1 ( J) − μ2 ( J)|, J

where J ranges over all intervals (that is, over all connected subsets of S1 ). Since acts by homeomorphisms, and any homeomorphism of S1 maps intervals to intervals, it is easy to see that d is -invariant. We claim that 6.10

d is continuous,

with respect to the usual weak∗ topology that Prob0 (S1 ) inherits from being a subset of Prob (S1 ). To see this, we note that if μ ∈ Prob0 (S1 ) and > 0, then, because μ has no atoms, we may partition S1 into ﬁnitely many intervals J1 , . . . , Jk , such that μ( Ji ) < for all i. By approximating the characteristic functions of these intervals from above and below, we may construct continuous functions f1 , . . . , f2k and some δ > 0, such that, for ν ∈ Prob (S1 ),

if |ν( fi ) − μ( fi )| < δ for every i, then |ν( Ji ) − μ( Ji )| < for every i. n Then, for any interval J in S1 , we have |ν( J) − μ( J)| < 2 , so d(ν, μ) < 2 . This completes the proof of (6.10). Because ψ 2 is measurable and d is continuous, we know that the composition d ◦ ψ 2 is measurable. In addition, because ψ 2 is -equivariant and d is -invariant, we also know that d ◦ ψ 2 is -invariant. (Note: we are saying invariant, not just equivariant.) From the Moore ergodicity theorem (6.8), we know that is ergodic on (G/P)2 (see Corollary 7.6), so we conclude that d ◦ ψ 2 is constant (a.e.); say d(ψ 2 (x, y)) = c,

for a.e. x, y ∈ G/P.

178 / d. w. morris

We wish to show that c = 0, for then it is clear that ψ(x) = ψ( y) for a.e. x, y ∈ G/P, so ψ is constant (a.e.). It is obvious that d(ψ 2 (x, x)) = 0 for every x ∈ G/P. If ψ were constant everywhere, rather than merely almost everywhere, then it would follow immediately that c = 0. Unfortunately, the diagonal {(x, x)} is a set of measure 0 in (G/P)2 , so a little bit of additional argument is required. By Lusin’s theorem, ψ is continuous on a set K of positive measure. Since the composition of continuous functions is continuous, we conclude that d ◦ ψ 2 is continuous on K × K . So, by continuity, it is constant on all of K × K , not merely almost all. Since d(ψ 2 (x, x)) = 0, this implies d(ψ 2 (x, y)) = 0 for all x, y ∈ K . Since K is a set of positive measure, this implies c = 0, as desired. Case 2. Assume ψ(x) consists entirely of atoms, for a.e. x ∈ G/P. To simplify the notation, without losing the main ideas, let us assume that ψ(x) consists of a single atom, for every x ∈ G/P. Thus, we may think of ψ as a -equivariant, measurable map ψ : G/P → S1 . Surprisingly, even with the simplifying assumption, the argument here seems to be more difﬁcult than in Case 1. The idea is to obtain a contradiction from the -equivariance of ψ, by contrasting two fundamental observations: • Homeo+ (S1 ) is not triply transitive on S1 : if x, y, and z are distinct,

then no orientation-preserving homeomorphism of S1 can map the triple (x, y, z) to (y, x, z)—they have opposite orientations under the circular order on S1 . • The action of GL (2, R) on the projective line RP 1 = R ∪ {∞} by linearfractional transformations ax + b a b if g = g(x) = cx + d c d is triply transitive: if (x1 , y1 , z1 ) and (x2 , y2 , z2 ) are two ordered triples of distinct elements of R ∪ {∞}, then there is some g ∈ GL (2, R) with g(x1 ) = x2 , g(y1 ) = y2 , and g(z1 ) = z2 . To illustrate, let us give an easy proof that is not quite correct; the actual proof is a modiﬁed version of this. Deﬁne ψ 3 : (G/P)3 → (S1 )3

by

ψ 3 (x, y, z) = (ψ(x), ψ(y), ψ(z)).

can lattices in sl (n, R) act on the circle? / 179

Then ψ 3 is -equivariant, so X + = { (x, y, z) ∈ (G/P)3 | (ψ(x), ψ(y), ψ(z)) is positively oriented } is a -invariant, measurable subset of (G/P)3 . Let us assume that is ergodic on (G/P)3 . (Unfortunately, this assumption is false, so it is where the proof breaks down.) Then X + must be (almost) all of (G/P)3 ; thus, (ψ(x), ψ(y), ψ(z)) is positively oriented, for (almost) every x, y, z ∈ G/P. But this is nonsense: either (ψ(x), ψ(y), ψ(z)) or (ψ(y), ψ(x), ψ(z)) is negatively oriented, so there are many negatively oriented triples. To salvage the above faulty proof, we replace (G/P)3 with a subset X , on which does act ergodically. Let ⎡ ⎤ ∗ ∗ ∗ ⎥ • Q =⎢ ⎣∗ ∗ ∗⎦ ⊂ G, and 0 0 ∗ x Q = x Q = x Q, 1 2 3 3 • X = (x1 , x2 , x3 ) ∈ (G/P) . x1 , x2 , x3 distinct Note that X is a submanifold of (G/P)3 . (If this is not obvious, it follows from the fact, proven below, that X is a single G-orbit in (G/P)3 .) Assume, for the moment, that is ergodic on X (with respect to any (hence, every) Lebesgue measure on the manifold X ). Then the above proof, with X in the place of (G/P)3 , implies that X + = { (x, y, z) ∈ X | (ψ(x), ψ(y), ψ(z)) is positively oriented } is a set of measure 0. Then it is not difﬁcult to see that ψ is constant on X (a.e.). Hence ψ is right Q -invariant (a.e.): for each q ∈ Q , we have ψ(xqP) = ψ(xP) for a.e. x ∈ G/P. By a similar argument, we see that ψ is right Q -invariant (a.e.), where ⎡ ⎤ ∗ ∗ ∗ ⎢ ⎥ Q = ⎣0 ∗ ∗⎦ ⊂ G. 0 ∗ ∗ Because Q and Q , taken together, generate all of G, it then follows that ψ is right G-invariant (a.e.). Hence, ψ is constant (a.e.), as desired. All that remains is to show that is ergodic on X . By the Moore ergodicity theorem (6.8), we need only show that 1) G is transitive on X , and 2) the stabilizer of some element of X is not compact.

180 / d. w. morris

These facts are perhaps easiest to establish from a geometric perspective. The (parabolic) subgroup Q is the stabilizer of the plane π0 = (∗, ∗, 0). Hence, QF0 is the set of all ﬂags (, π) with π = π0 . Therefore, under the identiﬁcation of G/P with F , we have 3 π1 = π2 = π3 , . 6.11 X = ((1 , π1 ), (2 , π2 ), (3 , π3 )) ∈ F 1 , 2 , 3 distinct 1) Let us show that G is transitive on X . Given ((1 , π), (2 , π), (3 , π)), ((1 , π ), (2 , π ), (3 , π )) ∈ X , it sufﬁces to show that there exists g ∈ G, such that 6.12

g((1 , π), (2 , π), (3 , π)) = ((1 , π ), (2 , π ), (3 , π )).

Because G is transitive on the set of 2-dimensional subspaces, we may assume π = π = π0 = R2 . Then, because GL (2, R) is triply transitive on RP 1 , there exists T ∈ GL (2, R), such that T (1 , 2 , 3 ) = (1 , 2 , 3 ). Letting ⎡ ⎢ g =⎣ 0

T 0

⎤ 0 ⎥ 0 ⎦ 1 det T

yields Equation (6.12). 2) Let us show that the stabilizer of some element of X is not compact. Let ⎧⎡ ⎫ ⎤ ⎪ ⎪ 0 ⎨ a 0 ⎬ ⎢ ⎥ A = ⎣0 a 0 ⎦ a ∈ R× . ⎪ ⎪ ⎩ ⎭ 0 0 1/a2 Then A is a closed, noncompact subgroup of G. Furthermore, every element of A acts as a scalar on π0 , so every element of A ﬁxes every 1-dimensional vector subspace of π0 . Thus, if ((1 , π), (2 , π), (3 , π)) ∈ X with π = π0 , then A is contained in the stabilizer of this element of X , so the stabilizer is not compact.

can lattices in sl (n, R) act on the circle? / 181

notes for §6. Ghys’s proof ﬁrst appeared in [Gh2]. See [Gh3, §7.3] for an exposition. 7. Additional Ingredients of Ghys's Proof 7A. Amenability and an Equivariant Map

A group is amenable if its action on every compact, convex set has a ﬁxed point. More precisely:

definition 7.1. A Lie group G is amenable if, for every continuous action of G by linear operators on a locally convex topological vector space V , and every nonempty, compact, convex G-invariant subset C of V , the group G has a ﬁxed point in C. example 7.2. 1) If T is any continuous linear operator on V and v is any element of V such that {T n v} is bounded, then every accumulation point of the sequence vn =

1 (Tv + T 2 v + · · · + T n v) n

is a ﬁxed point for T . This implies that cyclic groups are amenable. 2) A generalization of this argument shows that all abelian groups are amenable; this statement is a version of the classical Kakutani-Markov ﬁxed-point theorem. 3) It is not difﬁcult to see that if N is a normal subgroup of G such that N and G/N are both amenable, then G is amenable. 4) Combining the preceding two observations implies that solvable groups are amenable. 5) In particular, the group P of Notation 6.5 is amenable. Proof of Proposition 6.7. Since Prob (X ) is a compact, convex set, a version of the Banach-Alaoglu theorem tells us that L∞ (G; Prob (X )) is compact and convex in a natural weak∗ topology. Let ! " L∞ (G; Prob (X )) = ψ ∈ L∞ (G; Prob (X )) ψ is -equivariant (a.e.) . This is a closed subset of L∞ (G; Prob (X )), so it is compact. It is also convex and nonempty. To say ψ is -equivariant (a.e.) means, for each γ ∈ , that ψ(γ x) = γ · ψ(x) for a.e. x ∈ G; so G acts on L∞ (G; Prob (X )) by translation on the right. Hence, the subgroup P acts on L∞ (G; Prob (X )).

182 / d. w. morris

Since P is amenable (see Example 7.2(5)), it must have a ﬁxed point ψ0 in the compact, convex set L∞ (G; Prob (X )). Then ψ0 is invariant (a.e.) under translation on the right by elements of P, so it factors through (a.e.) to a welldeﬁned map ψ : G/P → Prob (X ). Because ψ0 is -equivariant, it is immediate that ψ is -equivariant. 7B. Moore Ergodicity Theorem

We will obtain the Moore ergodicity theorem (6.8) as an easy consequence of the following result in representation theory:

theorem 7.3. (decay of matrix coefficients) If • G = SL (n, R), • π is a unitary representation of G on a Hilbert space H, such that no non-

zero vector is ﬁxed by π(G); and • {gj } is a sequence of elements of G, such that gj → ∞,

then π(gj )φ | ψ → 0, for every φ, ψ ∈ H. Proof. By passing to a subsequence, we may assume π(gj ) converges weakly to some operator E; that is, π(gj )φ | ψ → Eφ | ψ for every φ, ψ ∈ H. We wish to show ker E = H. Let U = { u ∈ G | gj ugj−1 → e } and U − = { v ∈ G | gj−1 vgj → e }. For u ∈ U, we have Eπ (u)φ | ψ = limπ (gj u)φ | ψ = limπ(gj ugj−1 )π (gj )φ | ψ = limπ(gj )φ | ψ = Eφ | ψ,

can lattices in sl (n, R) act on the circle? / 183

so Eπ (u) = E. Therefore, letting HU be the space of U-invariant vectors in H, we have (HU )⊥ ⊂ ker E. We have E ∗ φ | ψ = φ | Eψ = limφ | π(gj )ψ = limπ (gj−1 )φ | ψ, so the same argument, with E ∗ in the place of E and gj−1 in the place of gj , shows that −

(HU )⊥ ⊂ ker E ∗ . Assume, for simplicity, that each gj is a positive-deﬁnite diagonal matrix: ⎡ ⎤ aj ⎢ ⎥ gj = ⎣ with aj , bj , cj > 0. bj ⎦ cj (It is not difﬁcult to eliminate this hypothesis, by using the Cartan decomposition G = KAK , but that is not necessary for Ghys’s proof.) Then the subgroup generated by {π(gj )} is commutative. Because π is unitary, this means that π (gj ) commutes with both π(gk ) and π (gk )∗ = π (gk−1 ) for every j and k. Therefore, the limit E commutes with its adjoint (that is, E is normal): we have E ∗ E = EE ∗ . Hence Eφ2 = Eφ | Eφ = (E ∗ E)φ | φ = (EE ∗ )φ | φ = E ∗ φ | E ∗ φ = E ∗ φ2 , so ker E = ker E ∗ . Thus, ker E = ker E + ker E ∗ −

⊃ (HU )⊥ + (HU )⊥ −

= (HU ∩ HU )⊥ −

= (HU,U )⊥ . By passing to a subsequence, and then permuting the basis vectors of R3 , we may assume aj ≥ bj ≥ cj .

184 / d. w. morris

Since gj → ∞, we have

aj bj , lim max b j cj j→∞

= ∞.

For deﬁniteness, let us assume lim sup j→∞

so

⎡

aj bj

and

⎤

1

⎢ U=⎣ ∗

0 such that |γ n | ≥ εn (n ∈ N). This allows one to exclude the existence of nontrivial actions of nonuniform lattices thanks to the presence of unipotent elements; see [34]. But it also has different applications: for instance, any ﬁnitely generated nilpotent subgroup of Ham(, ω) is in fact abelian. Quasimorphisms and Bounded Cohomology

Recall that a quasi morphism on a group is a map φ : → R for which there exists a constant C > 0 such that |φ(xy) − φ(x) − φ(y)| ≤ C for any x, y ∈ . The quasi morphism φ is called homogeneous if moreover it satisﬁes φ(x n ) = nφ(x) (n ∈ Z, x ∈ ); that is, φ is a true homomorphism when restricted to cyclic subgroups of . For any quasimorphism φ, the limit φ(x n ) n→∞ n

φh (x) = lim

exists. The map φh is the unique homogeneous quasimorphism such that φ − φh is bounded. We will denote by QM(, R) the vector space of all quasimorphisms on , and by QMh (, R) the subspace of homogeneous quasimorphisms. For more details on quasi morphisms and bounded cohomology see [4, 6] as well as Gromov’s seminal paper [30]. We will return to the link between quasi morphisms and bounded cohomology in the next section. According to a result by Burger and Monod (see [7] or [8]), any homogeneous quasimorphism on a higher-rank lattice as above is a homomorphism, and hence is identically zero since is a Kazhdan group. On the other hand, the group of Hamiltonian diffeomorphisms of a closed surface (which is simple and hence admits no nontrivial homomorphism to R [3]) admits many quasimorphisms: Gambaudo and Ghys [25] proved that the vector space QMh (Ham(, ω), R) is inﬁnite-dimensional. Their proof consists in constructing explicitly a family of linearly independent homogeneous quasimorphisms. More constructions of quasimorphisms on groups of Hamiltonian diffeomorphisms can be found in the work of Entov and Polterovich [13] and of the author [40]. It is now well known (see [29]) that one could try to use this contrast between the lattice and groups of Hamiltonian diffeomorphisms of surfaces to attack the problem of the existence of nontrivial homomorphisms ρ : → Ham(, ω). According to the discussion above, for any homogeneous

212 / p. py

quasimorphism φ : Ham(, ω) → R the invariant φ ◦ ρ : → R vanishes identically. This should give some constraints on the homomorphism ρ. This circle of ideas is nicely discussed in Ghys’s survey [29]. Here, we will explain how one can use more subtly the vanishing results of Burger and Monod in bounded cohomology to obtain much more precise constraints on the homomorphism ρ. Namely, we will use the vanishing of certain bounded cohomology groups with unitary coefﬁcients. Note that the use of bounded cohomology for the study of group actions is not new. Bounded cohomology already appears in the study of groups acting on the circle; see [7, 28].

2. Vanishing Results and Consequences We begin with a brief reminder on the second bounded cohomology group of a discrete group , with unitary coefﬁcients (see for instance [36] for more details). We consider a unitary representation π of , that is, a homomorphism from to the group of unitary operators of a Hilbert space H . We will write ||v|| for the norm of a vector v ∈ H . All the unitary representations we will consider are obtained in the following way: the group is acting by measurepreserving transformations on a probability space (X , μ) and we consider the representation π of on the Hilbert space H = L2 (X , μ) of square integrable functions on X , deﬁned by π(γ )( f ) = f ◦ γ −1 (f ∈ L2 (X , μ)). We will say that a map c : j → H is bounded if the quantity |c|∞,j :=

sup

||c(γ1 , . . . , γj )||

(γ1 ,...,γj )∈j

is ﬁnite. The space Z 2 (, π) of 2-cocycles on with values in H is the space of maps c : 2 → H such that: π(γ1 )(c(γ2 , γ3 )) − c(γ1 γ2 , γ3 ) + c(γ1 , γ2 γ3 ) − c(γ1 , γ2 ) = 0. We will denote by Zb2 (, π) ⊂ Z 2 (, π) the subspace of bounded 2-cocycles. In the same way, the space B1 (, π) ⊂ Z 2 (, π) of coboundaries consists of the maps c : 2 → H that satisfy the equation c(γ1 , γ2 ) = π (γ1 )(v(γ2 )) + v(γ1 ) − v(γ1 γ2 ) for some map v : → H . The subspace of coboundaries c for which the map v in the equation above can be chosen bounded will be denoted by Bb1 (, π). The second bounded cohomology group of with coefﬁcients in H is the quotient Hb2 (, π) = Zb2 (, π)/Bb1 (, π),

some remarks on area-preserving actions of lattices / 213

and the second usual cohomology group of with coefﬁcients in H is the quotient H 2 (, π) = Z 2 (, π)/B1 (, π). Of course, there is a natural map from Hb2 (, π) to H 2 (, π) that sends the class of a bounded cocycle to its usual cohomology class. We will denote by EHb2 (, π) the kernel of this map. In the following proposition C (, H ) is the space of all maps from to H , Cb (, H ) the subspace of those maps that are bounded and d1 u is the coboundary of a map u ∈ C (, H ): d1 u(γ1 , γ2 ) = π (γ1 )(u(γ2 )) + u(γ1 ) − u(γ1 γ2 ).

proposition 2.1. The space EHb2 (, π) is isomorphic to {u ∈ C (, H ), |d1 u|∞,2 < ∞}/ Z 1 (, π) + Cb (, H ) . Proof. If u ∈ C (, H ) is such that |d1 u|∞,2 < ∞, the coboundary d1 u of u is a bounded 2-cocycle, hence deﬁnes a class [d1 u] ∈ Hb2 (, π) that is obviously trivial in usual cohomology. We have thus constructed a map u → [d1 u] from {u ∈ C (, H ), |d1 u|∞,2 < ∞} to EHb2 (, π) that is (by definition) surjective. Let us examine its kernel. The class [d1 u] vanishes in bounded cohomology precisely if there exists a bounded map v : → H such that d1 u = d1 v. In that case w = u − v is a 1-cocycle and we have u = w + v ∈ Z 1 (, π) + Cb (, H ). When H = R, with the trivial representation π0 of , the previous proposition simply proves the following classical fact [4, 6]: the group EHb2 (, π0 ) is isomorphic to the quotient QM(, R)/ Hom(, R) ⊕ Cb (, R) . Note that this last quotient is also isomorphic to the space QMh (, R)/Hom (, R). Let us come back to the case of a higher-rank lattice , as in the introduction. In that case, we have already said that the group EHb2 (, π0 ) is trivial, according to a result of Burger and Monod [7, 8]. In fact, their result is much more general: they prove that the group EHb2 (, π ) vanishes, for any unitary representation π of on a Hilbert space H (the result even holds for more general Banach spaces, but we will not really discuss this here). This can be seen as a strengthening of property (T): the group has the property that for any unitary representation π, the group H 1 (, π ) as well as the group EHb2 (, π ) vanishes. Following Monod [36], we will say that a discrete group with this property has property (TT). We now discuss the dynamical consequences of property (TT).

214 / p. py

Suppose that (X , μ) is a measure-preserving action of on a probability space. We will say that a map u : → L2 (X , μ) is a quasicocycle if there exists a constant C > 0 such that |u(γ1 γ2 ) − π(γ1 )(u(γ2 )) − u(γ1 )| ≤ C almost everywhere, for all γ1 , γ2 ∈ . We will give some examples of quasicocycles deﬁned on groups of Hamiltonian diffeomorphisms of surfaces in the next section. According to the subadditive ergodic theorem (see [32] for instance), if u is a quasicocycle and γ ∈ , the sequence of functions u(γ n ) n converges almost everywhere to a function u (γ ). The reader should observe that since the action of on X is measure preserving, the map u(γ )dμ γ → φ(γ ) = X

is a quasimorphism on . It is not difﬁcult to check that the map γ → φu (γ ) = u (γ )dμ is the unique homogeneous quasimorphism at a bounded distance X from φ. Note that all the quasimorphisms deﬁned by Gambaudo and Ghys [25] and by the author [40] are obtained in this way: by integration of a quasicocycle.

remark 1 The quasimorphisms constructed by Entov and Polterovich [13] are deﬁned in a completely different way; see [39] for a survey. It is not a priori clear how to relate them (in a nontrivial way) to some bounded cohomology class with nontrivial coefﬁcients. remark 2 A weaker deﬁnition of quasicocycle already appears in the litterature; see [44]. There, a map u : → L2 (X , μ) is called a quasicocycle if |d1 u|∞,2 =

sup (γ1 ,γ2

)∈2

|d1 u(γ1 , γ2 )|L2

is ﬁnite. In our deﬁnition, we require the stronger condition that sup (γ1 ,γ2 u(γ n )

|d1 u(γ1 , γ2 )|L∞

)∈2

is ﬁnite, to ensure that ( n )n ≥0 converges almost everywhere for all γ . The notion of quasicocycle is also related to the notion of rough action (see [36], chapter V). We now see the advantage of using the full result of Burger and Monod: if is a group with property (TT) acting on X and u is a quasicocycle, the vanishing result with trivial coefﬁcients tells us that for any γ ∈ the integral

some remarks on area-preserving actions of lattices / 215

φu (γ ) =

u (γ )dμ X

is zero, while the full vanishing result tells us that the function u (γ ) is zero almost everywhere, as the following proposition shows.

proposition 2.2. Suppose has property (TT ). Then, for any measurepreserving action (X , μ) on a probability space and for any quasicocycle u : → L2 (X , μ) we have u (γ ) = 0, almost everywhere ( for all γ ∈ ). Proof. If u is a quasicocycle, the map d1 u is a bounded 2-cocycle. Since the group EHb2 (, π) is trivial, there exists, according to the previous proposition, a bounded map v : → L2 (X , μ) and a 1-cocycle w : → L2 (X , μ) such that u = v + w. We write D := supγ ∈ ||v(γ )||. Since has property (T), there exists a map ϕ ∈ L2 (X , μ) such that w(γ ) = ϕ ◦ γ − ϕ. We obtain: u(γ n ) ϕ ◦ γ n − ϕ v(γ n ) = + . n n n We already know that the sequence The function

u(γ n ) n

converges almost everywhere to u (γ ).

ϕ ◦γn −ϕ 1

= (ϕ ◦ γ − ϕ) ◦ γ j n n n−1 j=0

converges almost everywhere by Birkhoff ’s theorem; it is a classical lemma v(γ n ) in ergodic theory that its limit is 0 almost everywhere. Finally, n converges n v(γ ) almost everywhere to u (γ ). Since the norm of n in L2 (X , μ) is less than D n,

v(γ nk )

there exists a subsequence n that converges almost everywhere to 0. k Hence u (γ ) = 0 almost everywhere.

3. Some Examples We now describe some examples of concrete quasicocycles on groups of Hamiltonian diffeomorphisms of surfaces, taken from [17, 20, 24, 25, 33, 40]. These functions are constructed in the spirit of Schwartzman’s classical asymptotic cycle [42] or of Arnold’s asymptotic Hopf invariant; see [1]. The interested reader will ﬁnd many more examples in the work of Gambaudo and Ghys [25].

216 / p. py

Since we are dealing with diffeomorphisms that are isotopic to the identity, we will always make use of isotopies ( ft )t∈[0,1] from the identity 1l = f0 to a given diffeomorphism f = f1 and will consider various (kinds of) rotation numbers associated to the orbit ( ft (x)) of a point x or to a pair of orbits ( ft (x)), ( ft (y)) (x = y). To ensure that the number we deﬁne does not depend on the choice of the isotopy ( ft )t∈[0,1] but only on f we will often appeal to some results on the topology of the groups Ham(, ω) or Diff 0 (, ω); see [12].

example 1 [17, 24] We ﬁrst give an example of a 1-cocycle deﬁned on the group Diff c (D2 , ω) of area-preserving diffeomorphisms of the disc D2 = {(x, y) ∈ R 2 , |x|2 + |y|2 ≤ 1}, which coincide with the identity near the boundary. This group is torsion free (see [39]). Although there is no nontrivial homomorphism from a Kazhdan group to Diff c (D2 , ω) (thanks to Thurston’s stability theorem), it is worth studying this ﬁrst example. Consider a diffeomorphism f ∈ Diff c (D2 , ω) and choose an isotopy ( ft ) from f0 = 1l to f1 = f . For any two distinct points x, y in the disc, consider the nonzero vector ft (x) − ft (y) ∈ R 2 . When t goes from 0 to 1 the argument e2iπu(t) of this vector varies of a quantity anglef (x, y) := u(1) − u(0). This deﬁnes a continuous function anglef : D2 × D2 − → R, where = {(x, x), x ∈ D2 } is the diagonal. One easily establishes that this function does not depend on the choice of the isotopy ( ft )t∈[0,1] but only on f , and that it is bounded; see [24]. If f , g ∈ Diff c (D2 , ω) and ( ft ) and (gt ) are two isotopies from the identity to f and g, respectively, we can consider the isotopy (ht ) = (gt ) ∗ ( ft g) from the identity to fg. It allows us to establish the relation: anglefg (x, y) = angleg (x, y) + anglef (g(x), g(y)). Hence the map f → anglef −1 deﬁnes a 1-cocycle on the group Diff c (D2 , ω) with values in the space L2 (D2 × D2 , R). Problem. Suppose ⊂ Diff c (D2 , ω) is a ﬁnitely generated group. Can we give a condition on that ensures that the class of the cocycle angle is nonzero in H 1 (, L2 (D2 × D2 , μ2 ))? The following proposition shows that the class of the cocycle f → anglef −1 is always nonzero in the space H 1 (, C 0 (D2 × D2 − )), unless is trivial (here C 0 (D2 × D2 − ) is the space of continuous functions on D2 × D2 − ).

proposition 3.1. Fix a diffeomorphism f ∈ Diff c (D2 , ω). Assume that there exists a continuous function ϕ : D2 × D2 − → R such that anglef = ϕ ◦ f − ϕ. Then f is the identity.

some remarks on area-preserving actions of lattices / 217

Proof. We assume that f is not the identity. Then, it is enough to ﬁnd an integer n and two ﬁxed points x, y of f n such that anglef n (x, y) = 0. Indeed the relation anglef = ϕ ◦ f − ϕ implies anglef n = ϕ ◦ f n − ϕ. If ϕ is continuous, this forces the equality anglef n (x, y) = 0 for any pair (x, y) of distinct ﬁxed points of f n . We will see at the end of this section (see Lemma 3.2 and the paragraph following it) that if f is distinct from the identity there exists a ﬁxed point x0 and a point y0 = x0 such that the number (x0 , y0 ) := lim angle f

1

n→∞ n

anglef n (x0 , y0 )

exists and is nonzero. Consider now the compact annulus A obtained from the disc by blowing up the ﬁxed point x0 thanks to the action of the differential dfx0 . The diffeomorphism f naturally extends to a homeomorphism of A, still denoted f . Let F be the lift of f to the universal cover A of the annulus that pointwise ﬁxes the component of the boundary of A corresponding to the boundary of the disc. If z is in the interior of A (identiﬁed with the interior of the disc, minus the point x0 ), the limit lim

1

n→∞ n

anglef n (x0 , z),

if it exists, is the rotation number ρF (z) (determined by the lift F) of the point z in the annulus. We know that • the point y0 has a nonzero rotation number, and • points close to the boundary have zero rotation number.

Now, let us recall a result of Franks (this is corollary 2.4 of [19]). Let g : A → A be a homeomorphism isotopic to the identity and G be a lift of g to A. If x is a point of A we will denote by ρG (x) the rotation number of x determined by G, if it exists. Let K ⊂ A be a chain transitive compact invariant set for g. See [19] for the notion of chain transitivity; this is a very weak form of recurrence. Assume that there exist two points x1 and x2 in K whose rotation numbers ρG (x1 ) and ρG (x2 ) exist and satisfy ρG (x1 ) < ρG (x2 ). Then, for any rational p p number q ∈]ρG (x1 ), ρG (x2 )[, g has a periodic point with rotation number q . Let us apply the above result to f . Since f is area preserving, the chain transitivity hypothesis will be easily satisﬁed. Almost every point of the interior of A is recurrent, hence every point of A is nonwandering. In particular every point of A is chain recurrent. The compact set K = A is connected, f -invariant, and all its points are chain recurrent, hence it is chain transitive (this is proposition 1.2 in [19]). Franks’s result applies with K = A. The existence of the point y0 and of points with zero rotation number implies that there exists a periodic

218 / p. py

point y1 with nonzero rotation number. If the period of y1 is n, this exactly means that the quantity anglef n (x0 , y1 ) is nonzero.

example 2 [20, 33] If f is a Hamiltonian diffeomorphism of the torus, we have seen in the introduction that for any lift F : R 2 → R 2 of f , one has T2 (F(x) − x)dμ(x) ∈ Z2 . We can therefore choose a particular lift f∗ for f : the unique one for which T2 ( f∗ (x) − x)dμ(x) = 0. Since the map R2 x

→ R2 → f∗ (x) − x

is invariant under integral translations, there exists a map vf : T2 → R 2 such that vf (p(x)) = f∗ (x) − x (where p : R 2 → T2 is the natural projection). From the relation ( f ◦ g)∗ = f∗ ◦ g∗ , one deduces the cocycle relation vf ◦g = vg + vf ◦ g. The map f → vf −1 ∈ L2 (T2 , R 2 ) is therefore a 1-cocycle. Suppose now that ⊂ Ham(T2 , ω) is a ﬁnitely generated subgroup on which the previous cocycle is cohomologically trivial: there exists a measurable map ϕ : T2 → R 2 such that vf = ϕ ◦ f − ϕ ( f ∈ ). Recall ﬁrst that for almost every point x in the torus, we have lim inf ϕ( f n (x)) − ϕ(x) = 0. n

This is an easy consequence of Poincaré’s recurrence theorem and Lusin’s theorem. But we also have the following relation: f∗n (x) − x = ϕ( f n (p(x))) − ϕ(p(x)) (x ∈ R 2 ). Hence, we get that almost every point of the plane is recurrent for the dynamics of the diffeomorphism f∗ (for all f ∈ ). In fact the function ψ : R2 x

→ R2 → x − ϕ(p(x))

is invariant under the lifted action of on the plane (through the diffeomorphisms f∗ ). The domain ψ −1 ([0, 1[2 ) ⊂ R 2 , as well as its translates by integral vectors, has ﬁnite measure and is invariant under . Therefore the vanishing of the cohomology class of the cocycle f → vf −1 gives some constraints on the (measurable) dynamics of .

example 3 [25] We assume here that the genus of is greater than 1. Endow with a hyperbolic metric and identify its universal cover with the Poincaré disc . For each 1-form η on , we will deﬁne a quasicocycle uη with values in the space of continuous functions on . This quasicocycle is a true cocycle if η is closed, and a coboundary if η is exact.

some remarks on area-preserving actions of lattices / 219

If f : → is a Hamiltonian diffeomorphism, we will name f∗ : → the unique lift of f, which commutes with the action of the fundamental group of on . If x ∈ , one can consider the unique geodesic arc α( x ) between x and f∗ ( x ). Let η denote the lift of the 1-form η to . The map → R η x → α(x ) x ∈ is any lift of x, the descends to a function uη ( f , ·) on . If x ∈ and quantity uη ( fg, x) − uη (g, x) − uη ( f , g(x)) equals the integral of the 2-form d η on the geodesic triangle of with vertices x ), f∗ ◦ g∗ ( x ). Since the 2-form d η is bounded (it is invariant by the action x , g∗ ( of the fundamental group of ) this quantity is bounded by the norm of d η times the area of a hyperbolic triangle. We obtain uη ( fg, x) − uη (g, x) − uη ( f , g(x)) ≤ π · |dη|∞ . Hence the map f → uη ( f −1 , ·) is a quasicocycle. The reader will ﬁnd an alternative description of this quasicocycle in [39].

example 4 [25] We now give an example of a quasicocycle deﬁned on the group Diff 0 (T2 , ω). We will make use of the group structure of the torus. We ﬁx a point x∗ in T2 − {0} and choose a homogeneous quasimorphism φ : π1 (T2 − {0}, x∗ ) → R (there are many since π1 (T2 − {0}, x∗ ) is a free group; see [6] as well as [14] for generalizations). For each point v ∈ T2 − {0} we ﬁx a path (αv (t))t∈[0,1] from x∗ to v in T2 − {0} whose length is bounded for a Riemannian metric deﬁned on the compact surface with boundary T2 − {0} obtained by blowing up the origin on the torus. If ( ft )t∈[0,1] is an area-preserving isotopy on T2 and x and y two distinct points on T2 , we can consider the curve ft (x) − ft (y) ∈ T2 − {0}. We close it to form a loop α( f , x, y) = αx−y ∗ ( ft (x) − ft (y)) ∗ αf1 (x)−f1 (y) . Then, we deﬁne a function u(φ, f , ·, ·) on T2 × T2 − by u(φ, f , x, y) = φ(α( f , x, y)). One can easily check that this function does not depend on the choice of the isotopy ( ft )t∈[0,1] but only on f , is measurable, and bounded; see [25]. From the relation α( fg, x, y) = α(g, x, y) ∗ α( f , g(x), g(y))

220 / p. py

and the fact that φ is a quasimorphism, we deduce |u(φ, fg, x, y) − u(φ, g, x, y) − u(φ, f , g(x), g(y))| ≤ C. This means precisely that the map f → u(φ, f −1 , ·, ·) ∈ L2 (T2 × T2 , μ ⊗ μ) is a quasicocycle.

example 5 [40] Consider the 2-sphere, with an area form ω of total area 2. Let (S2 ) be the space of oriented tangent lines to S2 and p : (S2 ) → S2 the canonical projection. There is a natural action of the circle R/Z on (S2 ): if ∈ (S2 ) is a tangent line at x ∈ S2 , e2iπs · is the tangent line at x whose angle with is 2πs. Let X be the (periodic) vector ﬁeld that generates this action of R/Z on (S2 ). We can ﬁnd a 1-form α on (S2 ), invariant by rotations, such that α(X ) = 1 and dα = p∗ ω. This implies that α is a contact form whose Reeb vector ﬁeld is X . Let f be an area-preserving diffeomorphism of S2 and ( ft ) a Hamiltonian isotopy from the identity to f , generated by the Hamiltonian (Ht ). Let Xt be the symplectic gradient of Ht . The isotopy ( ft ) induces an isotopy on (S2 ) through the action of the differential dft : (S2 ) → (S2 ) of ft . The key point of our last example is that there exists a second isotopy on (S2 ) that lifts ( ft ). It is constructed as t be the horizontal lift of Xt to (S2 ), that is, the vector ﬁeld on follows. Let X (S2 ) deﬁned by the equations t ) = Xt . t ) = 0 and p∗ (X α(X t − (Ht ◦ p)X generates an isotopy θ( ft ) : The time-dependent vector ﬁeld X 2 2 (S ) → (S ) that preserves α and whose isotopy class depends only on the isotopy class of ( ft ); see [2]. We can now measure a kind of rotation number associated to any tangent line ∈ (S2 ). Indeed for all t, θ ( ft )() and dft () are tangent lines at the point ft (p()) ∈ S2 . Hence, we can write dft () = e2iπu(t) · θ ( ft )() where e2iπ u(t) is the angle between θ( ft )() and dft (). The quantity w( f , ) = u(1) − u(0) depends only on f and . Then w( f , ·) : (S2 ) → R is a continuous map that satisﬁes w( fg, ) = w(g, ) + w( f , dg()). This is not yet exactly what we need: since the action ( f , ) → df () of the group Ham(S2 , ω) on (S2 ) does not have an invariant measure we cannot consider the function w( f, ·) as a vector in some unitary representation of Ham(, ω). Nevertheless, we can prove the following inequality; see [40]: if 1

some remarks on area-preserving actions of lattices / 221

and 2 are two tangent lines at the same point x = p(1 ) = p(2 ) ∈ S2 , then |w( f , 1 ) − w( f , 2 )| ≤ 10. Thus, if we deﬁne w( f , x) = maxp()=x w( f , ), w( f , ·) is a bounded measurable function on S2 and we have |w( fg, x) − w(g, x) − w( f , g(x))| ≤ 30. The map f → w( f −1 , ·) is therefore a quasicocycle. Motivated by these examples and by the vanishing results of the previous section, we can raise the following question. Question 1

If f : → is a Hamiltonian diffeomorphism of a closed surface, distinct from the identity, can we ﬁnd a quasicocycle u deﬁned on the group Ham(, ω), with values in the space L2 (, μ), or L2 ( × , μ ⊗ μ), or even L2 ( n , μn ) for some integer n, such that the function u ( f ) is not zero almost everywhere? In other words, can we ﬁnd a set of positive measure (in or × . . .) of points for which some kind of rotation number or asymptotic linking number is nonzero? A positive answer to this question would yield a positive answer to Zimmer’s conjecture. Note, however, that we should certainly add some hypothesis on f to get a positive answer. For instance, if f is a rotation of the sphere, u( f ) vanishes for all known quasicocycles. It is probably possible to construct more subtle counterexamples. Assume for instance that we can construct a Hamiltonian diffeomorphism f with the following two properties: f is weakly mixing, and for some algebraic reason, φ( f ) = 0 for any homogeneous quasimorphism deﬁned on the group of Hamiltonian diffeomorphisms. In that case, for any quasicocycle u with values in L2 ( n , μn ) one has u ( f ) = 0 almost everyn where. Indeed, since the diagonal action of f on is ergodic for any n, one has u ( f ) = φu ( f ) = 0 almost everywhere on n . Note, however, that this kind of counterexample cannot appear inside the image ρ() of a higher-rank lattice, since all the diffeomorphisms in the group ρ() have discrete spectrum. We now give two small hints that tend to show that a nontrivial Hamiltonian diffeomorphism of a compact surface contains somewhere a set of positive measure of points that “rotate around each other." Consider ﬁrst a diffeomorphism f ∈ Diff c (D2 , ω), distinct from the identity. We write (x, y) = lim angle f

1

n→∞ n

anglef n (x, y)

222 / p. py (x, y) when this limit exists. Note that, according to Birkhoff’s theorem, angle f 2 2 exists for μ ⊗ μ-almost every point (x, y) ∈ D × D , and if x0 is a ﬁxed point (y, x0 ) exists for μ-almost every point y ∈ D2 . A result of Viterbo of f , angle f [45] implies that there exists a ﬁxed point x0 ∈ D2 for f and a Borel set B ⊂ D2 of positive measure such that (y, x0 ) = 0, angle f

for all y ∈ B. We now explain this fact. Recall ﬁrst that one can deﬁne the symplectic action A(x0 ) of a ﬁxed point x0 of f in the following way. If ( ft )t∈[0,1] is a Hamiltonian isotopy from the identity to f , generated by a Hamiltonian (Ht ) one deﬁnes 1 A(x0 ) = λ+ Ht ( ft (x))dt, γ

0

where γ is the loop ( ft (x0 ))t∈[0,1] and λ is a primitive of the area form ω on D2 . This number does not depend on the choice of the isotopy ( ft ) from the identity to f . The next lemma, which I learned from Patrice Le Calvez, relates the symplectic action of a ﬁxed point to the cocycle anglef .

lemma 3.2. For any ﬁxed point x0 of f one has (y, x0 )dμ(y). A(x0 ) = anglef (y, x0 )dμ(y) = angle f D2

D2

Proof. To compute the action A(x0 ), we can choose an isotopy ( ft )t∈[0,1] such that ft (x0 ) = x0 for all t. We denote by Xt the vector ﬁeld that generates the isotopy and Ht the corresponding compactly supported Hamiltonian. Let dθ be a closed 1-form on the annulus D2 − {x0 } whose integral over a generator of the ﬁrst homology of the annulus equals 1. The integral of the function 1 anglef (·, x0 ) over D2 equals D2 0 dθ(Xt )dt ω. Using the fact that the 3-form dθ ∧ ω is 0 we get dθ(Xt ) ω = dθ ∧ ω(Xt , ·) = −d(Ht dθ). Choose now a small disc Dε around x0 . We have 1 lim D2 −Dε 0 dθ(Xt )dt ω D2 anglef (y, x0 )dμ(y) = ε→0 1 = lim ∂Dε 0 Ht dt dθ . ε→0

Here we have used Stokes’s theorem and the fact that Ht vanishes near the boundary of D2 . The last limit in the equality above coincides with the action

1

some remarks on area-preserving actions of lattices / 223

A(x0 ) = 0 Ht (x0 )dt. We have therefore proved the ﬁrst equality of the lemma. The second equality follows from Birkhoff’s theorem.

Now, using symplectic methods, Viterbo shows that if f is distinct from the identity, there exists a ﬁxed point x0 with nonzero action. Thanks to the interpretation of the action of x0 as the average linking number around x0 , we is nonzero on the deduce the existence of the set B. We get that the map angle f 2 2 set B × {x0 } ⊂ D × D . However, from our point of view, it would be more natural to obtain a set of positive measure in D2 × D2 for the product measure. Second, we consider the simplest examples of Hamiltonian diffeomorphisms: integrable ones. For instance, let us take a smooth function H : t : T2 → T2 . Assume T2 → R and consider the associated Hamiltonian ﬂow ϕH that H is nonconstant. Then, we will show that there is a quasicocycle u on 1 ) is Ham(T2 , ω) with values in L2 (T2 , R 2 ) or in L2 (T2 × T2 , R) such that u (ϕH not zero almost everywhere (for the measure μ on T2 in the ﬁrst case, for the measure μ ⊗ μ in the second one). As H is nonconstant there is an embedding of the annulus i : A = S1 × [0, 1] → T2 t reads ϕ t (i(θ , s)) = i(θ + tϑ(s), s) for a nonvanishing on which the ﬂow ϕH H function ϑ : [0, 1] → R ∗ . To prove this fact, one just considers a connected t component C of a regular level of H. In a neighborhood of C all orbits of ϕH are periodic; hence C is contained in an annulus that is invariant under the t and in which each orbit is periodic. From this one easily deduces ﬂow ϕH the existence of the embedding i. We write ˆi : R × [0, 1] → R 2 for a lift of the map i. Assume ﬁrst that the annulus A is essential in T2 . In that case there exists a nonzero vector w ∈ Z2 such that ˆi(θ + k, s) = ˆi(θ , s) + kw. Consequently, if vϕ 1 is the cocycle from H Example 2, we have vϕHn (i(θ , s)) → ϑ(s)w = 0. n→∞ n

Assume now that the annulus A is inessential. We will make use of the quasicocycle from Example 4. We can assume that the curve i(S1 × {0}) bounds an embedded disc D that is disjoint from i(S1 ×]0, 1]) (up to changing the parameter s by 1 − s). Assume that s2 > s1 and consider the trajectories of t (i(θ , s )) − the two points i(θ1 , s1 ) and i(θ2 , s2 ). The family of curves (ϕH 2 2 t ϕH (i(θ1 , s1 )))t∈[0,n] winds around 0 in the disc D ∪ i(A) (see Figure 1). The loops n , i(θ , s ), i(θ , s )) are therefore almost equal to a power of the commutator α(ϕH 2 2 1 1 [a, b], where a and b are the standard generators of the group π1 (T2 − {0}, x∗ ).

224 / p. py

Fig. 1. Linking of points in an inessential annulus on the torus.

More precisely, we can write n α(ϕH , i(θ2 , s2 ), i(θ1 , s1 )) = αn ∗ [a, b][nϑ(s2 )] ∗ βn

where αn and βn stay in a ﬁxed ﬁnite subset of π1 (T2 − {0}, x∗ ) and [nϑ(s2 )] is n , i(θ , s ), i(θ , s )) − the integer part of nϑ(s2 ). From this we deduce that u(φ, ϕH 2 2 1 1 [nϑ(s )] 2 φ([a, b] ) is bounded independently of n. Therefore, for any homogeneous quasimorphism φ on π1 (T2 − {0}, x∗ ) we have n , i(θ , s ), i(θ , s )) u(φ, ϕH 2 2 1 1 → ϑ(s2 )φ([a, b]). n To conclude we now choose a quasimorphism φ such that φ([a, b]) is nonzero and observe that the set {(i(θ1 , s1 ), i(θ2 , s2 )), s2 > s1 } ⊂ T2 × T2 has positive μ ⊗ μ-measure. Hence we have proved:

theorem Let ρ : → Ham(T2 , ω) be a homomorphism, where is a higherrank lattice, as above. If a diffeomorphism f ∈ ρ() is contained in a Hamiltonian ﬂow, then f is the identity. In fact, this result could have been deduced directly from the existence of Zimmer’s invariant Riemannian metric. Indeed, in the situation above, one can always ﬁnd an embedding i such that the function ϑ is not constant. This implies that for an open set of points x the sequence of differentials n (x)) (dϕH n≥0 tends to inﬁnity. This contradicts the existence of Zimmer’s metric. Yet, the above proof was presented in order to illustrate the possible use of the quasicocycles that we described. Also, it is not difﬁcult to establish a similar result on a surface of genus greater than 1, or on the sphere if we assume that the ﬂow is not conjugated to a 1-parameter group of rotations; see [41]. 4. Final Remarks We have already mentioned that, according to a result of Zimmer, the group ρ() preserves (almost everywhere) a measurable Riemannian metric. In the

some remarks on area-preserving actions of lattices / 225

case where the surface is the torus T2 , we can slightly improve this fact: the group ρ() preserves a pair (X 1 , X 2 ) of linearly independent measurable vector ﬁelds. If g is a Hamiltonian diffeomorphism of T2 , let us consider a Hamiltonian isotopy (gt ) from the identity to g. We identify the differential dgt (x) of the diffeomorphism gt at a point x with a 2 × 2 matrix. Hence, the path

2 (R) in the universal cover of of matrices dgt (x) deﬁnes an element dg(x) ∈ SL SL2 (R), which depends only on g and x. The map Ham(T2 ) × T2 (g, x)

2 (R) → SL → dg(x)

2 (R) to SL2 (R), the map (g, x) → is a cocycle. If p denotes the projection from SL dg(x) = p(dg(x)) is the usual derivative cocycle. Suppose → Ham(T2 , ω) is a Kazhdan group and consider the restriction of the previous cocycle to . According to Zimmer [47, 49], there exists a measurable map ϕ : T2 → SL2 (R) such that ϕ(γ · x)−1 ◦ dγ (x) ◦ ϕ(x) ∈ SO(2),

2 (R) → R the translation almost everywhere (for γ ∈ ). We denote by τ : SL

number associated to the action of SL2 (R) on the line (see [28]) and by T

2 (R). Let

2 (R) be a measurable the generator of the center of SL ϕ : T2 → SL lift of ϕ chosen in such a way that the map x → τ ( ϕ (x)) is bounded. This

2 (R), k ∈ Z). The is always possible since one has τ (a · T k ) = τ (a) + k (a ∈ SL

(x) ◦ map (γ , x) → ϕ (γ · x)−1 ◦ dγ ϕ (x) is a cocycle whose image is contained

2 (R), which is canonically isomorphic in the inverse image of SO(2) in SL to R.

(x) ◦ lemma 4.1. For each γ ∈ the map x → sγ (x) := ϕ (γ · x)−1 ◦ dγ ϕ (x) ∈ R 1 is bounded. Hence the cocycle γ → sγ determines a class in H (, L2 (T2 , R)).

2 (R) is the Proof. Since the restriction of τ to the inverse image of SO(2) in SL identity, it is enough to show that the map x → τ (sγ (x)) is bounded. Since the two maps x → τ ( ϕ (x)) and x → τ ( ϕ (γ · x)−1 ) are bounded thanks to the choice of the lift ϕ and since τ is a quasimorphism, it is enough to check

(x)) is bounded. But this is obvious since this is a that the map x → τ (dγ continuous map from T2 to R. From the lemma, we deduce that there exists a measurable (in fact L2 ) map ψ : T2 → R such that sγ (x) = ψ(γ · x) − ψ(x). Coming back to SL2 (R), we obtain

226 / p. py

ϕ(γ · x)−1 ◦ dγ (x) ◦ ϕ(x) = e2iπ (ψ(γ (x))−ψ(x)) , almost everywhere, where e2iπ u stands for the rotation of angle 2πu in SL2 (R). This tells us that the derivative cocycle is cohomologous to the constant cocycle (equal to the identity), which is equivalent to the existence of the vector ﬁelds X 1 and X 2 . Although the results we have discussed here give more evidence toward Zimmer’s conjecture, it seems difﬁcult to exploit them. In particular, the quest for a set of positive measure of points that “rotates" under the action of a diffeomorphism seems delicate. Note also that we have simply used global cohomological properties of the lattice to give some constraints on the dynamics of each single diffeomorphism in the group ρ() but I was not able to investigate the dynamics of the whole group . Rather than studying measure theoretical properties, it might be useful to study the topological dynamics of the action. In this direction, the work of Calegari [9] contains many promising ideas.

acknowledgments. This work is part of my thesis realized at École Normale Supérieure de Lyon. I would like to thank Étienne Ghys for all the discussions we had around the problems discussed in this text and for his warm encouragement. I would also like to thank François Béguin, Patrice Le Calvez, Frédéric Le Roux, and Leonid Polterovich for the numerous conversations we had around Zimmer’s conjecture. This text was completed while I was enjoying the hospitality of Tel-Aviv University. This work was also partly supported by the French Agence Nationale de la Recherche. Finally, I would like to thank the referee for his or her constructive comments on this text. References [1] [2]

[3] [4] [5]

V. I. Arnold and B. A. Khesin, Topological methods in hydrodynamics, Applied Mathematical Sciences 125, Springer-Verlag, New York, (1998). A. Banyaga, The group of diffeomorphisms preserving a regular contact form, Topology and algebra (Proc. Colloq., Eidgenöss. Tech. Hochsch., Zurich, 1977), 47–53, Monograph. Enseign. Math. 26, Univ. Genève (1978). A. Banyaga, The structure of classical diffeomorphism groups, Mathematics and its applications 400, Kluwer Academic Publishers Group, Dordrecht, (1997). C. Bavard, Longueur stable des commutateurs, Enseign. Math. (2) 37, No. 1–2 (1991), 109–150. A. Borel, Compact Clifford-Klein forms of symmetric spaces, Topology 2 (1963), 111–122.

some remarks on area-preserving actions of lattices / 227 [6]

[7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20]

[21] [22] [23] [24] [25] [26] [27] [28]

R. Brooks, Some remarks on bounded cohomology, Riemann Surfaces and Related Topics: Proceedings of the 1978 Stony Brook Conference (State Univ. New York, Stony Brook, NY, 1978), 53–63, Ann. of Math. Stud. 97, Princeton University Press, Princeton, NJ, (1981). M. Burger and N. Monod, Bounded cohomology of lattices in higher rank Lie groups, J. Eur. Math. Soc. (JEMS) 1, No. 2 (1999), 199–235. M. Burger and N. Monod, Continuous bounded cohomology and applications to rigidity theory, Geom. Funct. Anal. 12, No. 2 (2002), 219–280. D. Calegari, Circular groups, planar groups, and the Euler class, Proceedings of the Casson Fest, Geom. Topol. Monogr. 7, Geom. Topol. Publ., Coventry (2004), 431–491. S. Cantat, Version kählérienne d’une conjecture de Robert J. Zimmer, Ann. Sci. École Norm. Sup. (4) 37, No. 5 (2004), 759–768. J. Déserti, Groupe de Cremona et dynamique complexe: une approche de la conjecture de Zimmer, Int. Math. Res. Not. (2006). C. J. Earle and J. Eells, A ﬁbre bundle description of Teichmüller theory, J. Diff. Geometry 3, (1969), 19–43. M. Entov and L. Polterovich, Calabi quasimorphism and quantum homology, Int. Math. Res. Not., No. 30 (2003), 1635–1676. D. Epstein and K. Fujiwara, The second bounded cohomology of word-hyperbolic groups, Topology 36, No. 6 (1997), 1275–1289. B. Farb and H. Masur, Superrigidity and mapping class groups, Topology 37, No. 6 (1998), 1169–1176. B. Farb and P. Shalen, Real-analytic actions of lattices, Invent. Math. 135, No. 2 (1999), 273–296. A. ÊFathi, Transformations et homéomorphismes préservant la mesure. Systèmes dynamiques minimaux, Thèse, Orsay, (1980). D. Fisher, Talk at the conference “Geometry, Rigidity, and Group Actions," in honor of R. J. Zimmer’s 60th birthday, University of Chicago (September 2007). J. Franks, Recurrence and ﬁxed points of surface homeomorphisms, Ergodic Theory Dynam. Systems 8∗ , Charles Conley Memorial Issue (1988), 99–107. J. Franks, Rotation vectors for surface diffeomorphisms, Proceedings of the International Congress of Mathematicians, Vol.1, 2, (Zürich, 1994), Birkhäuser, Basel (1995), 1179– 1186. J. Franks and M. Handel, Periodic points of Hamiltonian surface diffeomorphisms, Geom. Topol. 7, (2003), 713–756. J. Franks and M. Handel, Area preserving group actions on surfaces, Geom. Topol. 7, (2003), 757–771. J. Franks and M. Handel, Distortion elements in group actions on surfaces, Duke Math. J. 131, No. 3 (2006), 441–468. J.-M. Gambaudo and É. Ghys, Enlacements asymptotiques, Topology 36, No. 6 (1997), 1355–1379. J.-M. Gambaudo and É. Ghys, Commutators and diffeomorphisms of surfaces, Ergodic Theory Dynam. Systems 24, No. 5 (2004), 1591–1617. É. Ghys, Sur les groupes engendrés par des difféomorphismes proches de l’identité, Bol. Soc. Brasil. Mat. (N.S.) 24, No. 2 (1993), 137–178. É. Ghys, Actions de réseaux sur le cercle, Invent. Math. 137, No. 1 (1999), 199–231. É. Ghys, Groups acting on the circle, Enseign. Math. (2) 47, No. 3–4 (2001), 329–407.

228 / p. py [29] É. Ghys, Knots and dynamics, International Congress of Mathematicians, Vol. I, Eur. Math. Soc., Zürich (2007), 247–277. [30] M. Gromov, Volume and bounded cohomology, Inst. Hautes Études Sci. Publ. Math. 56 (1983), 5–99. [31] P. de la Harpe and A. Valette, La propriété (T ) de Kazhdan pour les groupes localement compacts (avec un appendice de Marc Burger), Astérisque No. 175 (1989). [32] J. F. C. Kingman, Subadditive processes, École d’Été de Probabilités de Saint-Flour, V 1975, Lecture Notes in Math. 539, Springer, Berlin, (1976), 167–223. [33] P. Le Calvez, Identity isotopies on surfaces, Dynamique des difféomorphismes conservatifs des surfaces: un point de vue topologique, Panor. Synthèse 21, Soc. Math. France, Paris (2006), 105–143. [34] A. Lubotzky, S. Mozes, and M. S. Raghunathan, The word and Riemannian metrics on lattices of semisimple groups, Inst. Hautes Études Sci. Publ. Math., No. 91 (2000), 5–53. [35] D. McDuff and D. Salamon, Introduction to symplectic topology, Second edition, Oxford Mathematical Monographs, Clarendon Press, Oxford University Press, New York, (1998). [36] N. Monod, Continuous bounded cohomology of locally compact groups, Lecture Notes in Mathematics 1758, Springer-Verlag, Berlin, (2001). [37] A. Navas, Actions de groupes de Kazhdan sur le cercle, Ann. Sci. École Norm. Sup. (4) 35, No. 5 (2002), 749–758. [38] L. Polterovich, Growth of maps, distortion in groups and symplectic geometry, Invent. Math. 150, No. 3 (2002), 655–686. [39] L. Polterovich, Floer homology, dynamics and groups, Morse theoretic methods in nonlinear analysis and in symplectic topology, NATO Sci. Ser. II Math. Phys. Chem. 217, Springer, Dordrecht (2006), 417–438. [40] P. Py, Quasi-morphismes et invariant de Calabi, Ann. Sci. École Norm. Sup. (4) 39, No. 1 (2006), 177–195. [41] P. Py, Quasi-morphismes et difféomorphismes hamiltoniens, Ph.D thesis, École Normale Supérieure de Lyon (2008), available at http://tel.archives-ouvertes.fr/ tel- 00263607/fr/. [42] S. Schwartzman, Asymptotic cycles, Ann. of Maths (2) 66, (1957), 270–284. [43] S. Smale, Diffeomorphisms of the 2-sphere, Proc. Amer. Math. Soc. 10 (1959), 621–626. [44] A. Thom, Low degree bounded cohomology and L2 -invariants for negatively curved groups, Groups, Geometry and Dynamics 3 (2009), 343–358. [45] C. Viterbo, Symplectic topology as the geometry of generating functions, Math. Ann. 292, No. 4 (1992), 685–710. [46] D. Witte, Arithmetic groups of higher Q-rank cannot act on 1-manifolds, Proc. Amer. Math. Soc. 122, No. 2 (1994), 333–340. [47] R. J. Zimmer, Kazhdan groups acting on compact manifolds, Invent. Math. 75, No. 3 (1984), 425–436. [48] R. J. Zimmer, Actions of semisimple groups and discrete subgroups, Proceedings of the International Congress of Mathematicians, Vol. 1, 2 (Berkeley, CA., 1986), 1247–1258, Amer. Math. Soc., Providence, RI (1987). [49] R. J. Zimmer, Spectrum, entropy, and geometric structures for smooth actions of Kazhdan groups, Israel J. Math. 75, No. 1 (1991), 65–80.

7

raul quiroga-barranco

ISOMETRIC ACTIONS OF SIMPLE GROUPS AND TRANSVERSE STRUCTURES: THE INTEGRABLE NORMAL CASE with gratitude to robert j. zimmer on the occasion of his 60 th birthday Abstract For actions with a dense orbit of a connected noncompact simple Lie group G, we obtain some global rigidity results when the actions preserve certain geometric structures. In particular, we prove that for a G-action to be equivalent to 1 on a space of the form (G × K\H)/ , it is necessary and sufﬁcient for the G-action to preserve a pseudo-Riemannian metric and a transverse Riemannian metric to the orbits. A similar result proves that the G-actions on spaces of the form (G × H)/ are characterized by preserving transverse parallelisms. By relating these techniques to the notion of the algebraic hull of an action, we obtain inﬁnitesimal Lie algebra structures on certain geometric manifolds acted upon by G.

1. Introduction In the rest of this work, let G be a connected noncompact simple Lie group with Lie algebra g and M a smooth connected manifold acted upon smoothly by G. There are several examples of such actions that preserve a ﬁnite volume. Some of the most interesting are obtained from Lie group homomorphisms G → L where L is a connected Lie group that admits a lattice . The relevant G-action is then given on the double coset K \L/ , where K is some compact subgroup that centralizes G. Robert Zimmer proposed in [15] to study the ﬁnite measure-preserving G-actions on M and determine to what extent these are given by such double cosets. To understand how to tackle Zimmer’s program, it has been proved very useful to consider G-actions preserving a suitable geometric structure (see [2] and [18]). This is a natural condition since the above double coset examples This work was supported by SNI-Mexico and the grants Conacyt 44620 and Concyteg 07-02-K662-091.

229

230 / r. quiroga-barranco

carry a G-invariant pseudo-Riemannian metric when L is semisimple. Such metric comes from the bi-invariant metric on L obtained from the Killing form of its Lie algebra. The properties of the G-orbits of ﬁnite measure-preserving G-actions are reasonably well known. For example, such G-actions are known to be everywhere locally free when they preserve suitable pseudo-Riemannian metrics (see [2], [13], and [14]). Also in such case, the metric restricted to the orbits can be described precisely (see Lemma 2.6). However, there is still a lack of knowledge of the properties of a manifold, acted upon by G, in the transverse direction to the orbits. In this work, we propose to study the G-actions on M by emphasizing the need to understand the properties of the transverse to the G-orbits. For this, we will be dealing with G-actions that have a dense orbit and preserve a ﬁnite-volume pseudo-Riemannian metric. As observed above, in this case the results from [13] show that the G-orbits deﬁne a foliation, which from now on we will denote with O. By assuming that the orbits are nondegenerate for the pseudo-Riemannian metric on M, we can consider the normal bundle T O⊥ to the orbits as realizing the transverse direction in M. In this respect, we obtain the following structure theorem for G-actions on M when this normal bundle is integrable.

theorem 1.1. Suppose that the G-action on M has a dense orbit and preserves a ﬁnite-volume complete pseudo-Riemannian metric. If the G-orbits in M are nondegenerate and the normal bundle to the orbits T O⊥ is integrable, then there exist: → M to which the G-action lifts, 1) an isometric ﬁnite covering map M and 2) a simply connected complete pseudo-Riemannian manifold N, 3) a discrete subgroup ⊂ G × Iso(N), is G-equivariantly isometric to (G × N)/ . such that M We observe that in the double coset examples K \L/ as above with L semisimple and with the metric coming from the Killing form, the G-orbits are always nondegenerate. Besides that, we prove that for a general G-action on M, the orbits are always nondegenerate when dim (M) < 2 dim (G) (see Lemma 2.7). Also, by developing criteria for the normal bundle T O⊥ to be integrable, we obtain some results where the conclusion of Theorem 1.1 holds. For example, Corollary 3.1 ensures such conclusion when G has high enough real rank. Also, Corollary 3.2 does the same for manifolds M whose dimensions have a suitable bound in terms of G. In this last result we can even dispense with the assumption of having nondegenerate orbits by applying Lemma 2.7.

isometric actions of simple groups and transverse structures / 231

Without assuming in Theorem 1.1 that the G-action has a dense orbit it is still possible to draw some conclusions. More speciﬁcally, if we assume the rest of the hypotheses, a description of the universal covering space of the manifold as a warped product is obtained. This sort of result already appears in [3]. For a G-action on M preserving a pseudo-Riemannian metric, in [10] we considered a certain geometric condition between the metrics on G and M (the former for a bi-invariant metric) from which we concluded that M is, up to a ﬁnite covering space, a double coset of the form (G × K \H)/ . In other words, a double coset as above where G appears as a factor in L. One of the steps used in [10] to achieve this was to prove that the normal bundle T O⊥ is Riemannian. Considering the relevance we are giving to the transverse to the orbits, it is natural to determine the properties of the G-actions that preserve a transverse Riemannian structure. In this context, we obtain the following result, which proves that, up to a ﬁnite covering, the double cosets of the form (G × K \H)/ are characterized as those isometric G-actions that preserve a transverse Riemannian structure on the foliation O. We recall that a semisimple Lie group is isotypic if the complexiﬁcation of its Lie algebra is isomorphic to a sum of identical simple complex ideals.

theorem 1.2. If G is a connected noncompact simple Lie group acting faithfully on a compact manifold M, then the following conditions are equivalent: 1) There is a G-equivariant ﬁnite covering map (G × K \H)/ → M where H is a connected Lie group with a compact subgroup K and ⊂ G × H is a discrete cocompact subgroup such that G is dense in G × H. → M for which the G-action on M lifts to a 2) There is a ﬁnite covering map M with a dense orbit and preserving G-action on M • a pseudo-Riemannian metric for which the orbits are nondegenerate,

and • a transverse Riemannian structure for the foliation O by G-orbits.

Furthermore, if G has ﬁnite center and real rank at least 2, then we can assume in (1) of the above equivalence that G × H is semisimple isotypic with ﬁnite center and that is an irreducible lattice. Based on the previous result, we prove in Theorem 5.1 that the G-actions on M preserving a Lorentzian metric are, up to a ﬁnite covering, those given by double cosets (G × K \H)/ with G locally isomorphic to SL(2, R). We observe that this result improves a similar one found in [2] (see also the introduction of [4]).

232 / r. quiroga-barranco

Continuing with our study of transverse structures, we next consider isometric G-actions preserving a transverse parallelism for the foliation O. We prove that, up to a ﬁnite covering, such actions characterize the double cosets of the form (G × H)/ .

theorem 1.3. If G is a connected noncompact simple Lie group acting faithfully on a compact manifold M, then the following conditions are equivalent: 1) There is a G-equivariant ﬁnite covering map (G × H)/ → M where H is a connected Lie group and ⊂ G × H is a discrete cocompact subgroup such that G is dense in G × H. → M for which the G-action on M lifts to a 2)There is a ﬁnite covering map M G-action on M with a dense orbit and preserving • a pseudo-Riemannian metric for which the orbits are nondegenerate, and • a transverse parallelism for the foliation O by G-orbits. Furthermore, if G has ﬁnite center and real rank at least 2, then we can assume in (1) of the above equivalence that G × H is semisimple isotypic with ﬁnite center and that is an irreducible lattice. The notion of the algebraic hull of an action on a bundle, introduced by Zimmer, is a fundamental tool to understand G-actions as they relate to geometric structures. We recall that the algebraic hull is the smallest algebraic subgroup for which there is a measurable G-invariant reduction of the bundle being acted upon (see [17]). For our setup, where we are interested in the transverse to the orbits, it is then natural to consider the algebraic hull of the G-action on the bundle L(T O⊥ ) for an isometric G-action with nondegenerate orbits. Since a G-action of this sort preserves a pseudo-Riemannian metric on T O⊥ , the algebraic hull for L(T O⊥ ) is in this case a subgroup of O(p, q), for some (p, q). The next result shows that for weakly irreducible manifolds and when the algebraic hull for L(T O⊥ ) is the largest possible for this setup, the manifold being acted upon has inﬁnitesimally at some point the structure of a speciﬁc Lie algebra that contains g. By the last claim in the statement, such Lie algebra structure is nontrivially linked to the geometry of the manifold. For the explicit description of the Lie algebra structure obtained in this result we refer to Lemma 7.1 and Theorem 7.2. We recall that a pseudo-Riemannian manifold is weakly irreducible if the tangent space at some (and hence any) point has no proper nondegenerate subspaces invariant under the restricted holonomy group at that point. Also recall that every simple Lie group with a bi-invariant metric is weakly irreducible.

isometric actions of simple groups and transverse structures / 233

theorem 1.4. Suppose that G has ﬁnite center and real rank at least 2, and that the G-action on M preserves a ﬁnite-volume complete pseudo-Riemannian metric. Also assume that G acts ergodically on M and that the foliation by G-orbits is nondegenerate. Denote with L the algebraic hull for the G-action on the bundle L(T O⊥ ) and with l its Lie algebra. In particular, there is an embedding of Lie algebras l → so(p, q), where (p, q) is the signature of the metric of M restricted to T O⊥ . If this embedding is surjective and M is weakly irreducible, then the following holds: 1) G is locally isomorphic to SO0 (p, q) and dim (M) = (p + q)(p + q + 1)/2, and 2)for some x ∈ M the tangent space Tx M admits a Lie algebra structure isomorphic to either so(p, q + 1) or so(p + 1, q) such that Tx O is a Lie subalgebra isomorphic to g. Furthermore, with respect to the representation of Lemma 2.1, there is a Lie algebra of local Killing ﬁelds vanishing at x, which is isomorphic to g and acts nontrivially on Tx M by derivations of the Lie algebra structure given in (2). With the results developed in this work, we try to show the importance of considering transverse geometric structures to understand the actions of noncompact simple Lie groups. In fact, we believe that some form of the following conjecture could provide a geometric characterization of the double coset examples of G-actions mentioned above.

conjecture 1.5. Consider the double coset G-spaces of the form K \L/ , where L is a semisimple Lie group with an irreducible lattice and a compact subgroup K ; the G-action is then induced by a nontrivial homomorphism G → L whose image centralizes K . Then, a G-action on a manifold M is equivalent to such a double coset G-action for some L, , K if and only if the G-action on M: • has a dense orbit, • preserves a pseudo-Riemannian metric, and • preserves a transverse geometric structure to the orbits suitably related to the

geometry of GK \L. Note that for L = G × H and K a compact subgroup of H, we obtain quotients GK \L = K \H and G\L = H, which naturally carry a Riemannian metric and a parallelism, respectively. Thus Theorems 1.2 and 1.3 verify the conjecture in these cases. One of the main tools used to obtain our results is Gromov’s machinery on geometric G-actions (see [1], [2], [10], and [18]). Such machinery ensures the

234 / r. quiroga-barranco

existence of large families of local Killing ﬁelds. We develop these techniques in Section 2 emphasizing the fact that the Killing ﬁelds thus obtained yield g-module structures on the tangent spaces to M; in this respect, our main result is Proposition 2.3, which is due to Gromov [2] in the analytic/compact case (see also [18]) and was extended to the smooth/ﬁnite volume case in [1] and [10]. In Section 3, we prove that the latter impose restrictions strong enough to guarantee the integrability of the normal bundle under suitable conditions (Corollaries 3.1 to 3.3). We observe that Theorem 1.1 and its consequences obtained in Section 3 are in fact extensions of results obtained in [2]: theorem 5.3.E on page 129 of [2] states, under the assumptions of our Corollary 3.1, that M has a covering G-equivariantly diffeomorphic to G × N for some N. Note, however, that the result in [2] yields only a topological covering and does not further describe the covering group. Theorem 1.2 is obtained in Section 4 by applying the main results and arguments from [10]. As mentioned above, this yields in Section 5 our characterization of Lorentzian manifolds acted upon with a dense orbit by a simple noncompact Lie group. In Section 6 we establish the characterization of Gspaces of the form (G × H)/ provided by Theorem 1.3; for this, one of the main ingredients is given by Theorem 1.1. We observe that the arguments used in Section 6 are based on those found in [11]. To obtain Theorem 1.4 in Section 7, an application of Theorem 1.1 is used again. But now, the notion of the algebraic hull and the deep result about it found in [17] are also fundamental. Ultimately, this is somewhat to be expected, since a computation of the algebraic hull for frame bundles over M is in fact an important step to build Gromov’s machinery for geometric G-actions on M. This is evident in our proof of Proposition 2.3. We want to observe that the conclusion of Theorem 1.4 can be paraphrased by saying that the manifold M has inﬁnitesimally the structure of either SO(p, q + 1) or SO(p + 1, q). It remains the question as to whether or not M can be related more precisely to either of these groups. This problem will be pursued elsewhere (see [7]) by requiring some additional conditions. The author wishes to thank Uri Bader and Shmuel Weinberger for some fruitful conversations.

2. Killing Fields on Manifolds with a Simple Group of Isometries By theorem 4.17 from [13], if the G-action on M has a dense orbit and preserves a ﬁnite-volume pseudo-Riemannian metric, then the action is locally free and so the orbits deﬁne a foliation that we have agreed to denote with O. In this

isometric actions of simple groups and transverse structures / 235

case, it is well known that the bundle T O tangent to the foliation O is a trivial vector bundle isomorphic to M × g, under the isomorphism given by M × g → TO (x, X ) → Xx∗ . For every x ∈ M, this induces an isomorphism between the ﬁber Tx O and g, which we will refer to as the natural isomorphism. Furthermore, if we consider on M × g the product G-action, where the G-action on g is the adjoint one, then such isomorphism is G-equivariant. More precisely, we have dg(X ∗ ) = Ad(g)(X )∗ for every g ∈ G and X ∈ g. Note that for X in the Lie algebra of a group acting on a manifold, we denote with X ∗ the vector ﬁeld on the manifold whose 1-parameter group of diffeomorphisms is given by ( exp (tX ))t through the action on the manifold. For any given pseudo-Riemannian manifold N, we will denote with Kill(N, x) the Lie algebra of germs at x of local Killing vector ﬁelds deﬁned in a neighborhood of x, and with Kill0 (N, x) we will denote the Lie subalgebra consisting of those germs that vanish at x. The following result is a consequence of the Jacobi identity and the fact that the Lie derivative of a metric with respect to its Killing ﬁelds vanishes. In the rest of this work, for a vector space W with a nondegenerate symmetric bilinear form, we will denote with so(W ) the Lie algebra of linear maps on W that are skew-symmetric with respect to the bilinear form.

lemma 2.1. Let N be a pseudo-Riemannian manifold and x ∈ N. Then, the map λx : Kill0 (N, x) → so(Tx N) λx (Z)(v) = [Z, V ]x , where V is any vector ﬁeld such that Vx = v, is a well-deﬁned homomorphism of Lie algebras. From now on, for a given point x of a pseudo-Riemannian manifold, the map λx will denote the homomorphism from the previous lemma. For the proof of our next result we will present some facts about inﬁnitesimal automorphisms and Killing ﬁelds, and we refer to [1] for further details. The tangent bundle of order k of a manifold N is the smooth bundle T (k) N (k) whose ﬁber at x is the space Tx N of (k − 1)-jets at x of vector ﬁelds of N. For

236 / r. quiroga-barranco

every (local) diffeomorphism ϕ : N1 → N2 mapping x1 to x2 we have a linear isomorphism: (k)

(k)

Tx1 N1 → Tx2 N2 jxk−1 (X ) → jxk−1 (dϕ(X )) 1 2 that depends only on the jet jxk1 (ϕ). For N1 = N2 = N, x1 = x2 = x this yields (k)

the group Dx (N) of k-jets at x of local diffeomorphisms ﬁxing x, whose group structure is induced by the composition of maps. In the case when N = Rn and x = 0, we will denote this group with GL(k) (n). We also recall that the k-th order frame bundle over N is the smooth bundle L(k) (N) that consists of the k-jets at 0 of local diffeomorphisms Rn → N; any such a jet j0k (ϕ) thus (k)

(k)

deﬁnes a linear isomorphism T0 Rn → Tϕ(0) N. The structure group of L(k) (N) (k)

is GL(k) (n). With respect to vector ﬁelds, we denote with Dx (N) the space of k-jets at x of vector ﬁelds vanishing at x, and we use the special notation gl(k) (n) when N = Rn and x = 0. If N carries a pseudo-Riemannian metric, for every (k) x ∈ N, we will denote with Autk (N, x) the subgroup of Dx (N) consisting of those k-jets that preserve the metric up to order k at x. Correspondingly, for vector ﬁelds we denote with Killk (N, x) the space of k-jets at x of vector ﬁelds on N that preserve the metric up to order k at x; the subspace of those k-jets whose 0-jet vanishes is denoted with Killk0 (N, x). The next result provides a natural (k) representation of Dx (N) from which the Lie algebras of this group and of (k) Autk (N, x) are described in terms of Dx (N) and Killk0 (N, x), respectively; the proof is elementary, but it is detailed in sections 2 and 4 of [1].

lemma 2.2. For a smooth manifold N and any given point x ∈ N the following properties hold for every k ≥ 1: 1) The map (k)

(k)

x : Dx (N) → GL(Tx N) x ( jxk (ϕ))( jxk−1 (X )) = jxk−1 (dϕ(X )) is a Lie group monomorphism. 2)The assignment [ jxk (X ), jxk (Y )]k = −jxk ([X , Y ]) yields a well-deﬁned Lie alge(k) bra structure on Dx (N). 3)The map (k)

(k)

θx : Dx (N) → gl(Tx N) θx ( jxk (X ))( jxk−1 (Y )) = −jxk−1 ([X , Y ])

isometric actions of simple groups and transverse structures / 237 (k)

is a Lie algebra monomorphism for the Lie algebra structure on Dx (N) given (k) (k) by [·, ·]k . Furthermore, θx (Dx (N)) = Lie(x (Dx (N))). 4)If N has a pseudo-Riemannian metric, then we have θx (Killk0 (N, x)) = Lie(x (Autk (N, x))). In particular, with respect to the homomorphisms x and θx , the Lie algebra of Autk (N, x) is realized by Killk0 (N, x) with the Lie algebra structure given by [·, ·]k . The following result is due to Gromov in the analytic case (see [2]) and it was extended to the smooth case in [10]. We present here a fairly detailed proof based on the results from [1]. In congruence with our notation for M, in the for the in M rest of this work we will use O to denote the foliation by G-orbits lifted G-action on every covering space M of M; this will be so for any covering whether ﬁnite or not.

proposition 2.3. Suppose that the G-action on M has a dense orbit and preserves a ﬁnite-volume pseudo-Riemannian metric. Then, there exists a dense such that for every x ∈ S the following properties are satisﬁed. subset S ⊂ M x), which is an 1) There is a homomorphism of Lie algebras ρx : g → Kill(M, isomorphism onto its image ρx (g) = g(x). x); that is, every element of g(x) vanishes at x. 2) g(x) ⊂ Kill0 (M, 3)For every X , Y ∈ g we have [ρx (X ), Y ∗ ] = [X , Y ]∗ = −[X ∗ , Y ∗ ] in a neighborhood of x. In particular, the elements in g(x) and their corresponding local ﬂows preserve both O and T O⊥ in a neighborhood of x. induces a 4)The homomorphism of Lie algebras λx ◦ ρx : g → so(Tx M) g-module structure on Tx M for which the subspaces Tx O and Tx O⊥ are g-submodules. Proof. Since the proof builds on the notions and results found in [1], we will mostly follow its notation. We will be careful to deﬁne the objects considered but we refer to [1] for further details. For every k ≥ 1, let us denote with σ k : L(k) (M) → Qk the GL(k) (n)equivariant map that deﬁnes the k-th order extension of the geometric structure deﬁned by the pseudo-Riemannian metric on M. Consider the set G k = {jxk−1 (X ∗ ) : X ∈ g, x ∈ M},

238 / r. quiroga-barranco

which, by the local freeness of the G-action, is a smooth subbundle of T (k) M. In fact, we have G = T O, and as with this bundle there is a trivialization given by M × g → Gk (x, X ) → jxk−1 (X ∗ ). The corresponding trivialization of the frame bundle of G k is given by M × GL(g) → L(G k ) (x, A) → Ax , where Ax (X ) = jxk−1 ((AX )∗ ). Note that we have taken g as the standard ﬁber of the bundle G k . (k) Choose a subspace G0 of T0 Rn isomorphic to g. We will ﬁx such subspace as well as an isomorphism with g through which we will identify these two spaces. Let us now consider L(k) (M, G k ) = {α ∈ L(k) (M) : α(G0 ) = Gxk if α ∈ L(k) (M)x }, which is a smooth reduction of L(k) (M) to the subgroup of GL(k) (n) that preserves the subspace G0 ; we will denote such subgroup with GL(k) (n, G0 ). Recall from the remarks preceding Lemma 2.2 that for every j0k (ϕ) ∈ L(k) (M) we obtain a linear isomorphism (k)

(k)

T0 Rn → Tϕ(0) M k−1 j0k−1 (X ) → jϕ(0) (dϕ(X )).

In particular, if we let fk : L(k) (M, G k ) → L(G k ) j0k (ϕ) → j0k (ϕ)|G0 , then, by the identiﬁcation between G0 and g, we can consider fk as a welldeﬁned smooth principal bundle morphism that covers the identity. The associated homomorphism of structure groups for fk is given by πk : GL(k) (n, G0 ) → GL(g) j0k (ϕ) → j0k (ϕ)|G0 , which is clearly surjective. Note that we have used again our identiﬁcation between g and G0 .

isometric actions of simple groups and transverse structures / 239

Fix μ an arbitrary ergodic component for the G-action on M for the pseudoRiemannian volume. Then, there is a measurable reduction P of L(k) (M, G k ) so that σ k (P) is (μ-a.e. over M) a single point q0 ∈ Qk . Furthermore, the structure group of P is the subgroup of GL(k) (n, G0 ) that stabilizes q0 , and in particular it is algebraic. This claim is a consequence of the fact that the GL(k) (n, G0 )-action on Qk is algebraic, which in turn follows from the fact that a pseudo-Riemannian metric is a geometric structure of algebraic type; we refer to section 4 and the proof of proposition 8.4 of [1] for further details. On the other hand, since πk is a surjection and fk is G-equivariant, by proposition 8.2 of [1], there exist reductions Q1 and Q 2 of L(k) (M, G k ) and Z

L(G k ), respectively, to subgroups L1 ⊂ GL(k) (n, G0 ) and Ad(G) ⊂ GL(g), such that fk (Q1 ) ⊂ Q 2 (μ-a.e. over M) and such that πk (L1 ) is a ﬁnite-index subgroup Z

of Ad(G) . Here L1 can be chosen to be the algebraic hull of L(k) (M, G k ) for the G-action on M with respect to the ergodic measure μ. This claim uses the Z

well-known fact that Ad(G) is the algebraic hull of M × GL(g) for the product Z

action. It is not difﬁcult to see that this can be chosen so that Q 2 = M × Ad(G) (μ-a.e. over M) with respect to the above identiﬁcation M × GL(g) ∼ = L(G k ). We can also assume that Q1 ⊂ P, μ-a.e. over M, because the reduction Q1 is the smallest one to an algebraic subgroup. The above discussion ensures that for μ-a.e. x ∈ M, we have the following relations: Z

fk ((Q1 )x ) ⊂ (Q2 )x = {x} × Ad(G) (Q1 )x ⊂ (P)x ⊂ L(k) (M, G k )x σ k ((P)x ) = {q0 }.

Let us now ﬁx a point x such that these conditions hold. Choose αx ∈ (Q1 )x Z

and let fk (αx ) = (x, kx ) where kx ∈ Ad(G) . Since πk is surjective, there exists kx ∈ GL(k) (n, G0 ) such that πk ( kx ) = kx . In particular, by the πk -equivariance of fk we have fk (αx k−1 ) = (x, e). We also have by the same reason x −1 −1 −1 k−1 fk (αx g x ) = fk (αx kx kx g kx ) = (x, kx πk (g)kx ),

for every g ∈ L1 . Also, the inclusion (Q1 )x ⊂ L(k) (M, G k )x implies that, for −1 every g ∈ L1 , the k-jets of diffeomorphisms αx k−1 x , αx g kx considered as linear (k) n (k) k isomorphisms T0 R → Tx M map G0 onto Gx . Hence, from the deﬁnition −1 −1 is a k-jet of local diffeomorof fk it follows that αx gαx−1 = (αx g k−1 x )(αx kx ) (k) (k) phism of M at x whose associated isomorphism Tx M → Tx M maps Gxk onto

240 / r. quiroga-barranco

itself by the assignment ∗ jxk−1 (X ∗ ) → jxk−1 ((kx πk (g)k−1 x X ) ),

for which we have used the above trivialization M × GL(g) ∼ = L(G k ). Since Z

πk (L1 ) has ﬁnite index in Ad(G) it contains the identity component Ad(G), and Z

because kx ∈ Ad(G) the group kx πk (L1 )k−1 x also contains Ad(G). It follows (k) that αx L1 αx−1 is a subgroup of Dx (M) for which the homomorphism from Lemma 2.2(1) induces a homomorphism Hx : αx L1 αx−1 → GL(Gxk ) αx gαx−1 → x (αx gαx−1 )|G k , x

whose image contains Ad(G) ⊂ GL(g) with respect to the identiﬁcation between g and Gxk given by the above isomorphism M × g ∼ = G k . This implies that the induced Lie algebra homomorphism hx : Lie(αx L1 αx−1 ) → gl(Gxk ) has image ad(g), again with respect to the referred identiﬁcation between g and Gxk . On the other hand, we have for every g ∈ L1 σ k ((αx gαx−1 )αx ) = σ k (αx g) = σ k (αx ) because σ k ((Q1 )x ) ⊂ σ k ((P)x ) = {q0 } is a single point. But this identity proves that every such k-jet αx gαx−1 preserves the pseudo-Riemannian metric up order k (see [1]). In other words, αx L1 αx−1 is a subgroup of Aut k (M, x) and by Lemma 2.2 we also have that Lie(αx L1 αx−1 ) is a Lie subalgebra of Killk0 (M, x). From the above remarks, it follows that there is a Lie algebra homomorphism ρ xk : g → Killk0 (M, x) such that (∗ )

θx ( ρxk (X ))(jxk−1 (Y ∗ )) = jxk−1 ([X , Y ]∗ )

for every X , Y ∈ g.

For k ﬁxed, the existence of the homomorphism ρ xk has been established for μ-a.e. x ∈ M, where μ is an arbitrary ergodic component of the pseudoRiemannian volume of M. Thus, for k ﬁxed, it follows that the homomorphism ρ xk exists for every x ∈ Sk , where Sk is some subset of M, which is conull with respect to the pseudo-Riemannian volume of M. Finally, if we let S0 = ∩∞ S , k=1 k then S0 is conull with respect to the pseudo-Riemannian volume and for every x ∈ S0 and every k ≥ 1 there exists a homomorphism ρ xk : g → Killk0 (M, x) satisfying (∗ ).

isometric actions of simple groups and transverse structures / 241

In [6] the notion of k-regular point for a metric in a manifold is introduced. Such regular points satisfy two key properties relevant to our discussion. First, the set of regular points U of M is an open dense subset. Second, for x ∈ U there is some integer k(x) ≥ 1 so that, for k ≥ k(x), every element of Killk0 (M, x) extends uniquely to an element of Kill0 (M, x). The ﬁrst property is found in [6] and the second one is proved in [1], both just using smoothness. Note that the results in [6] are stated for Riemannian manifolds but, as remarked in [1], the ones we consider here apply without change to general pseudo-Riemannian manifolds. The upshot of these remarks is that for every x ∈ U, there is some k(x) ≥ 1 so that the map Jxk : Kill0 (M, x) → Killk0 (M, x) X → jxk (X ), is a linear isomorphism for every k ≥ k(x). Note that in this case, for the usual brackets in Kill0 (M, x) and the brackets [·, ·]k in Killk0 (M, x) considered above, the map Jxk is a Lie algebra anti-isomorphism. For S0 and U as above, consider the dense subset S = S0 ∩ U ⊂ M. Next choose x ∈ S and k ≥ max (k(x), 2). Then, the map Jxk is a Lie algebra anti-isomorphism, and there exists a Lie algebra homomorphism ρ xk : g → k Kill0 (M, x) satisfying (*). If we let ρx = −(Jxk )−1 ◦ ρ xk : g → Kill0 (M, x), then ρx deﬁnes a Lie algebra homomorphism such that jxk−1 ([ρx (X ), Y ∗ ]) = jxk−1 ([X , Y ]∗ ), for every X , Y ∈ g. For this, we have used (*) and the deﬁnition of θx from Lemma 2.2. Since k − 1 ≥ 1 and because germs of Killing ﬁelds are determined by any jet of order at least 1, we conclude that, at our chosen point x, ρx satisﬁes (from our statement) (1), (2), and the identity in (3) in a neighborhood of x replaced with M. The identity in (3) now proves that every element with M of g(x) preserves the tangent bundle to O in a neighborhood of x: that is, the corresponding Lie derivatives map sections of T O into sections of T O. By proposition 2.2 of [5] we conclude that the local ﬂows of the elements of g(x) preserve O as well in a neighborhood of x. Since the elements of g(x) are Killing ﬁelds, we conclude that they (and their local ﬂows) also preserve the normal bundle T O⊥ in a neighborhood of x. This completes the proof replaced with M in of our statement for the dense subset S ⊂ M and for M (1)–(4). Finally, this yields the statement for M for the dense subset, which is the inverse image of S under the covering map since such map is a local isometry.

242 / r. quiroga-barranco

remark 2.4. The conclusions of Proposition 2.3 hold without change for with M in (1)–(4). In fact, our proof some dense subset S ⊂ M by replacing M ﬁrst obtains the required Killing ﬁelds on M, which are then translated into corresponding ones on M. In this work, we will be dealing with and interested in the case where the G-orbits in M are nondegenerate submanifolds with respect to the pseudo are nondegenerate as well Riemannian metric. In this case, the G-orbits on M = T O ⊕ T O⊥ . When this holds, and we have a direct sum decomposition T M that is given, at every x ∈ M, by the we can consider the g-valued 1-form ω on M ∼ composition Tx M → Tx O = g, where the ﬁrst map is the natural projection and the second map is the natural isomorphism. From this, we then deﬁne the g-valued 2-form given by = dω|∧2 T O⊥ . The following result will provide us with a criterion, in terms of , for the normal bundle T O⊥ to be integrable. At the same time, we relate with the g-module structures from Proposition 2.3. This result is very well known and it is essentially contained in [2], [3], and [10], but we include its proof here for the sake of completeness.

lemma 2.5. Let G, M, and S be as in Proposition 2.3. If we assume that the G-orbits are nondegenerate, then → g and x : ∧2 Tx O⊥ → g are both 1) for every x ∈ S, the maps ωx : Tx M homomorphisms of g-modules, for the g-module structures from Proposition 2.3, and 2) the normal bundle T O⊥ is integrable if and only if = 0. Proof. In the rest of the proof, let X ∈ g and x ∈ S be ﬁxed but arbitrarily given. let Z , Z ⊥ be its T O and T O⊥ components, For Z a vector ﬁeld over M, respectively. Since ρx (X ) is a Killing ﬁeld preserving O and T O⊥ it follows that [ρx (X ), Z] = [ρx (X ), Z ], [ρx (X ), Z]⊥ = [ρx (X ), Z ⊥ ]. Denote with α : Tx O → g the inverse map of X → Xx∗ . Then we have ωx (X · Zx ) = ωx ([ρx (X ), Z]x ) = α([ρx (X ), Z ]x ) = α([ρx (X ), ω(Z)∗ ]x )

isometric actions of simple groups and transverse structures / 243

= α([X , ω(Z)]∗x ) = [X , ωx (Z)] = X · ωx (Zx ), thus showing that ωx is a homomorphism of g-modules. Here we used in the second and third identities the deﬁnition of ω, and in the fourth identity the formula from Proposition 2.3(3); the rest follows from the deﬁnition of the g-module structures involved. Next, observe that for every pair of sections Z1 , Z2 of T O⊥ we have

(Z1 ∧ Z2 ) = Z1 (ω(Z2 )) − Z2 (ω(Z1 )) − ω([Z1 , Z2 ]) = −ω([Z1 , Z2 ]), which clearly implies (2). Now let u, v ∈ Tx O⊥ be given and choose U, V sections of T O⊥ extending them, respectively. Hence, using that ω is a homomorphism of g-modules, the Jacobi identity and the above expression relating and ω, we obtain

x (X · (u ∧ v)) = x ((X · u) ∧ v) + x (u ∧ (X · v)) = x ([ρx (X ), U] ∧ V ) + x (U ∧ [ρx (X ), V ]) = −ωx ([[ρx (X ), U], V ]) − ωx ([U, [ρx (X ), V ]]) = −ωx ([ρx (X ), [U, V ]]) = −ωx (X · [U, V ]x ) = −[X , ωx ([U, V ])] = [X , x (U ∧ V )] = X · x (u ∧ v), thus showing that x is a homomorphism of g-modules. Note that we have used that both [ρx (X ), U], [ρx (X ), V ] are sections of T O⊥ . The following result allows us to relate the metric T O coming from M to suitable metrics on G. The proof presented here is due to Gromov (see [2]) and provides our ﬁrst application of Proposition 2.3.

lemma 2.6. Suppose that the G-action on M has a dense orbit and preserves a ﬁnite-volume pseudo-Riemannian metric. Then, for every x ∈ M and with respect to the natural isomorphism g ∼ = Tx O, the metric of M restricted to Tx O deﬁnes an Ad(G)-invariant symmetric bilinear form on g independent of the point x.

244 / r. quiroga-barranco

Proof. With the above-mentioned trivialization of T O, the metric h on M restricted to the orbits and pulled back to M × g yields a map ψ : M → g∗ ⊗ g∗ x → Bx where Bx (X , Y ) = hx (Xx∗ , Yx∗ ). By Remark 2.4, there is a dense subset S ⊂ M so that the conclusions of Proposition 2.3 are satisﬁed for every x ∈ S with the tangent spaces and Killing replaced by those of M. Hence, for every x ∈ S the inner product ﬁelds of M hx is preserved, in the sense of Proposition 2.3(4), by the Killing vector ﬁelds that belong to g(x). In particular, for every x ∈ S and X , Y , Z ∈ g we have hx ([ρx (X ), Y ∗ ]x , Zx∗ ) = −hx (Yx∗ , [ρx (X ), Z ∗ ]x ), which, by Proposition 2.3(3) yields hx ([X , Y ]∗x , Zx∗ ) = −hx (Yx∗ , [X , Z]∗x ). In other words, for every x ∈ S we have Bx ([X , Y ], Z) = −Bx (Y , [X , Z]), for all X , Y , Z ∈ g. This implies that ψ(x) = Bx is an Ad(G)-invariant form on g for every x ∈ S. By the density of S in M, we conclude that the image of ψ lies in the set of Ad(G)-invariant forms. On the other hand, at every x ∈ M and for g ∈ G, X , Y ∈ g we have ∗ ∗ , Ygx ) ψ(gx)(X , Y ) = hgx (Xgx −1 ∗ −1 ∗ = hx (dggx (Xgx ), dggx (Ygx ))

= hx (Ad(g −1 )(X )∗x , Ad(g −1 )(Y )∗x ) = ψ(x)(Ad(g −1 )(X ), Ad(g −1 )(Y )), which shows that ψ is G-equivariant. Note that we used in the second identity that G preserves the metric, and in the third identity we used the remarks at the beginning of this section. The G-equivariance of ψ and the fact that its image lies in G-ﬁxed points implies that ψ is G-invariant. Then, the result follows from the existence of a dense G-orbit. As a simple application of the previous result, we prove the nondegeneracy of the orbits when the manifold acted upon has a suitably bounded dimension.

isometric actions of simple groups and transverse structures / 245

lemma 2.7. Suppose that the G-action on M has a dense orbit and preserves a ﬁnite-volume pseudo-Riemannian metric. If dim (M) < 2 dim (G), then the Gorbits are nondegenerate with respect to the metric on M. Proof. By Lemma 2.6, for every x ∈ M the metric restricted to Tx O corresponds to an Ad(G)-invariant form in g. The kernel of such a form is an ideal and so the metric hx restricted to Tx O is either nondegenerate or zero. Suppose that hx is zero when restricted to Tx O for some x ∈ M. Then, Tx O lies in the null cone of Tx M for the metric hx . Hence, for (m1 , m2 ) the signature of M, we have dim (G) = dim (Tx O) ≤ min (m1 , m2 ). And this implies 2 dim (G) ≤ m1 + m2 = dim (M), which is impossible.

3. Proof of Theorem 1.1 and Some Consequences We start this section by proving Theorem 1.1. Proof of Theorem 1.1. Assuming that T O⊥ is integrable, let F be the induced foliation. We will ﬁrst prove that F is totally geodesic, that is, its leaves are totally geodesic submanifolds of M. We will denote with h the metric on M preserved by G. First note that, if Y , Z are local sections of T O⊥ that preserve the foliation, then we have for every X ∈ g: X ∗ (h(Y , Z)) = h([X ∗ , Y ], Z) + h(Y , [X ∗ , Z]) = 0, because our choices imply that [X ∗ , Y ] and [X ∗ , Z] are sections of T O. In particular, for every Y , Z as above the function h(Y , Z) is constant along the G-orbits. In the notation of [5], we conclude that h is a bundle-like metric for the foliation O. Hence, by the remarks on page 79 of [5] it follows that h induces a transverse metric to the foliation O. By the construction of such transverse metric from h and the arguments in the proof of proposition 3.2 of [5], it is easy to conclude that the foliation O is given by pseudo-Riemannian submersions that deﬁne the transverse metric. More precisely, at every point in M there is an open subset U of M and a pseudo-Riemannian submersion π : U → B such that the ﬁbers of π deﬁne the foliation O restricted to U. We observe that the results of [5] are stated for Riemannian metrics, but those that we use here extend without change to arbitrary pseudo-Riemannian metrics. We will now use the properties of the structural equations from [8] for a pseudo-Riemannian submersion π : U → B as above. Again, the results in [8] are stated for Riemannian submersions, but the ones that we will use are easily seen to hold for pseudo-Riemannian submersions as well. For π as

246 / r. quiroga-barranco

above, let A be the associated fundamental tensor deﬁned in [8]. In particular, by the deﬁnition of A, the second fundamental form for the leaves of F is given by AX Y , for X , Y tangent to F . But by lemma 2 of [8] we have for X , Y tangent to F the identity AX Y = 12 [X , Y ] , where Z denotes the projection of Z onto T O. Hence, the integrability of T O⊥ to F shows that A vanishes on vector ﬁelds tangent to F , thus showing that the leaves of F are totally geodesic. Choose a leaf N of F . Then, one can prove fairly easily that every geodesic in M that is tangent at some point to N remains in N for every value of the parameter of the geodesic; this uses the fact that N is a maximal integral submanifold of T O⊥ and that the leaves of F are totally geodesic. Hence, the completenes of M implies that of N. For our chosen leaf N of F , consider the G-action map restricted to G × N. This deﬁnes a smooth map ϕ : G × N → M, which is G-equivariant. By Lemma 2.6, it follows easily that ϕ is a local isometry for G × N endowed with the product metric where G carries a suitable bi-invariant metric. In particular, G × N is complete and so we conclude from corollary 29 on page 202 from [9] that ϕ is an isometric covering map. Hence, the universal covering map of M ×N → M and the G-action ×N is the left is given by ϕ:G on M lifted to G action on the ﬁrst factor. × Iso(N), that is, that every element in We now claim that π1 (M) ⊂ Iso(G) To see this, let γ ∈ π1 (M) be π1 (M) preserves the factors in the product G × N. ×N given with γ = (γ1 , γ2 ) its component decomposition. Observe that in G we have T(g,x) O = Tg G,

T(g,x) O⊥ = Tx N,

× N. Since γ commutes with the G-action, for every (g, x) ∈ G it preserves ⊥ both T O and T O and so dγ (u) = dγ1 (u) + dγ2 (u) ∈ T O dγ (v) = dγ1 (v) + dγ2 (v) ∈ T O⊥ , and v ∈ T N. We conclude that dγ2 (T O) = 0 and dγ1 (T O⊥ ) = for every u ∈ T G and γ2 is independent of G. This 0, which implies that γ1 is independent of N yields our claim about π1 (M). carries a bi-invariant metric, by the results On the other hand, since G from section 4 of [10] we know that the connected component of the identity is given by Iso0 (G) = L(G)R( G) (the left and right translations) and of Iso(G) Hence, the group = π1 (M) ∩ that it is a ﬁnite index subgroup of Iso(G). × Iso(N)) has ﬁnite index in π1 (M), and so the induced map (G × (Iso0 (G)

isometric actions of simple groups and transverse structures / 247

N)/ → M is a ﬁnite covering. Moreover, every γ ∈ can be written as γ = and γ2 ∈ Iso(N), and since such γ commutes (Lg1 Rg2 , γ2 ), where g1 , g2 ∈ G with the G-action we conclude that (g1 gg2 , γ2 (x)) = (Lg1 Rg2 , γ2 )(g(e, x)) = g((Lg1 Rg2 , γ2 )(e, x)) = (gg1 g2 , γ2 (x)) and x ∈ N, which implies g1 ∈ Z(G). Hence, Lg1 = Rg1 and for every g ∈ G × Iso(N), thus showing that ⊂ R(G) × Iso(N). then γ ∈ R(G) Also note that the covering map G × N → M realizes π1 (G) ⊂ , which → (G × N)/. → (G × induces a covering map G × N We claim that G × N N)/ is a normal covering map. For this we need to check that π1 (G) is a normal subgroup of under the inclusion z → (Rz , e). But by the above and γ2 ∈ remarks, every γ ∈ can be written as γ = (Rg1 , γ2 ), where g1 ∈ G from which we obtain Iso(N), γ (Rz , e)γ −1 = (Rg1 Rz Rg −1 , γ2 γ2−1 ) = (Rz , e) 1

It follows that the group of deck transformations since π1 (G) is central in G. for G × N → (G × N)/ is given by the group = /π1 (G) and that we also = (G × N)/. have (G × N)/ = G × Iso(N) is a From the above, we conclude that ⊂ R(G) × Iso(N) group of deck transformations of G × N → M that induces a G-equivariant → M that satisﬁes the properties required to ﬁnite covering map (G × N)/ obtain Theorem 1.1. The integrability of the normal bundle can be ensured for suitable relations between the Lie group G and the geometry of the manifold on which it acts, thus providing the following results. In what follows, the signature of G, as a pseudo-Riemannian manifold, is always considered with respect to a biinvariant metric. Note that if (n1 , n2 ) is the signature of some bi-invariant metric on G, then the signature of any other bi-invariant metric is either (n1 , n2 ) or (n2 , n1 ).

corollary 3.1. Suppose that the G-action on M has a dense orbit and preserves a ﬁnite-volume complete pseudo-Riemannian metric. If the G-orbits are nondegenerate and either one of the following holds, 1) there is no Lie algebra embedding of g into so(Tx O⊥ ) for every x ∈ M, or 2)for n0 = min (n1 , n2 ) and m0 = min (m1 , m2 ), where (n1 , n2 ) and (m1 , m2 ) are the signatures of G and M, respectively, we have rankR (g) > m0 − n0 , then the conclusion of Theorem 1.1 holds.

248 / r. quiroga-barranco

given as in Proposition 2.3. Proof. Let us consider a subset S ⊂ M First suppose that condition (1) is satisﬁed. Then, for every x ∈ S, the g-module structure on Tx O⊥ given by Proposition 2.3(4) is trivial. By Lemma 2.5(1), being a homomorphism of g-modules, the map x is trivial for every x ∈ S. Since S is dense, we conclude that = 0 and so T O⊥ is integrable by Lemma 2.5(2). Hence, Theorem 1.1 can be applied. Let us now assume that (2) holds. Note that by Lemma 2.6, the signature of T O is either (n1 , n2 ) or (n2 , n1 ). If we let (k1 , k2 ) be the signature of T O⊥ , then it is easily seen that min (k1 , k2 ) ≤ m0 − n0 . Since the real rank of so(Tx O⊥ ) is precisely min (k1 , k2 ) the result in this case follows from the ﬁrst part.

corollary 3.2. Suppose that the G-action on M has a dense orbit and preserves a ﬁnite-volume complete pseudo-Riemannian metric. Let n be the dimension of the smallest g-module V such that ∧2 V contains a g-submodule isomorphic to g. If dim (M) < dim (G) + n, then the conclusion of Theorem 1.1 holds. Proof. First observe that the Lie brackets in g deﬁne a surjective homomorphism of g-modules from ∧2 g onto g. Hence, ∧2 g contains a submodule isomorphic to g, thus implying that n ≤ dim (g). Then, Lemma 2.7 shows that the G-orbits in M are nondegenerate with respect to the metric of M. be a subset given as in Proposition 2.3. From our hypotheLet S ⊂ M ses and since the G-orbits are nondegenerate, we have dim (Tx O⊥ ) < n. In particular, for every x ∈ S and for the g-module structure from Proposition 2.3(4), ∧2 Tx O⊥ does not contain a g-submodule isomorphic to g. Hence, Lemma 2.5(1) implies that x = 0 for every x ∈ S and so that = 0. By Lemma 2.5(2) the bundle T O⊥ is integrable and so Theorem 1.1 can be applied. On a compact manifold with Riemannian normal bundle, we can obtain the conclusions of Theorem 1.1 without having to a priori assume that the manifold is complete and that the normal bundle is integrable.

corollary 3.3. Suppose that the G-action on M has a dense orbit and preserves a pseudo-Riemannian metric. If M is compact, the G-orbits are nondegenerate and the normal bundle T O⊥ is Riemannian, then the conclusions of Theorem 1.1 is Riemannian homogeneous and that hold. Moreover, we can assume that N ⊂ G × Iso0 (N).

isometric actions of simple groups and transverse structures / 249

Proof. The proof is a reﬁnement of that of Theorem 1.1, so we will follow the notation of the latter. First observe that the integrability of T O⊥ follows from the proof of Corol In particular, we have lary 3.1(1) since so(Tx O⊥ ) is compact for every x ∈ M. the hypotheses of Theorem 1.1 except for the completeness of M. In this respect, it is easy to check that the compactness of M and the fact that T O⊥ is Riemannian imply that the geodesics in M perpendicular to the G-orbits are complete. This completeness is enough for the rest of the arguments in the proof of Theorem 1.1 to apply. has Finally, we observe that the existence of a dense G-orbit implies that N a dense orbit by its local isometries and, being Riemannian, we conclude that it is homogeneous. The latter follows from the inﬁnitesimal characterization of homogeneous Riemannian manifolds obtained in [12], and the fact that the orthogonal group is compact for deﬁnite metrics; we refer to [10] for further details. But for a homogeneous Riemannian manifold the group of isometries has ﬁnitely many connected components, and so we can intersect with to obtain the last claim after passing to a ﬁnite covering. G × Iso0 (N)

4. Manifolds with a Transverse Riemannian Structure: Proof of Theorem 1.2 In this section we will characterize those actions that preserve a Riemannian structure transverse to the G-orbits. We start with the following basic result relating transverse geometric structures for the foliation by G-orbits with geometric structures on the normal bundle to such orbits.

lemma 4.1. Suppose that the G-action on M has a dense orbit and preserves a ﬁnite-volume pseudo-Riemannian metric. Also assume that the G-orbits are nondegenerate. If H is a subgroup of GL(k, R), where k = dim (M) − dim (G), then there is a one-to-one correspondence between the G-invariant H-reductions of L(TM/T O) and the G-invariant H-reductions of L(T O⊥ ). In particular, every transverse Hstructure for the foliation O by G-orbits induces a G-invariant H-reduction of L(T O⊥ ). Proof. From the decomposition TM = T O ⊕ T O⊥ we obtain a natural Gequivariant isomorphism TM/T O → T O⊥ , which clearly yields the ﬁrst claim. Next, we recall that a transverse H-structure to the foliation O is given by a reduction P of L(TM/T O), which is invariant under the local ﬂows of vectors

250 / r. quiroga-barranco

ﬁelds tangent to the foliation O. In particular, P is invariant under the G-action on L(TM/T O) thus showing the last claim. The following result is obtained by applying the main theorems from [10].

proposition 4.2. Suppose that the G-action on M has a dense orbit and preserves a pseudo-Riemannian metric. Also assume that M is compact. If the normal bundle to the orbits T O⊥ is Riemannian, then there exist → M, 1) a ﬁnite covering map M 2)a connected Lie group H with a compact subgroup K , and 3)a discrete cocompact subgroup ⊂ G × H such that G is dense in G × H, so that M is G-equivariantly diffeomorphic for which the G-action on M lifts to M to (G × K \H)/ . Furthermore, if G has ﬁnite center and real rank at least 2, then we can assume that G × H is a ﬁnite-center isotypic semisimple Lie group and is an irreducible lattice. Proof. Let us denote with (m1 , m2 ) and (n1 , n2 ) the signatures of M and G, respectively. Also, let us denote m0 = min (m1 , m2 ) and n0 = min (n1 , n2 ). Observe that since T O⊥ is Riemannian, the bundle T O is nondegenerate. Hence, by Lemma 2.6 the signature of T O is either (n1 , n2 ) or (n2 , n1 ); this is because the signature of every Ad(G)-invariant metric on g is the signature of either the Killing form or its negative. Without loss of generality, we can assume that the signature of T O is precisely (n1 , n2 ). Theorems A and B from [10] provide precisely our required conclusions when n0 = m0 holds, except for explicitly stating the property that G is dense in G × H. The latter is obtained as follows. The conclusion of theorem A from [10] ensures the existence of a G-invariant ergodic smooth measure, which in turn implies the existence of a dense G-orbit in (G × H)/ . From this, it is easily seen that we necessarily have the density of the G-orbit of (e, e) in (G × H)/ . Since the map G × H → (G × H)/ is a covering, we conclude that the inverse image under this covering of such orbit, which is G, is dense in G × H. We now consider the possibilities for n0 . We will now assume that the signatures are given in the form (+, −). If n0 = n2 , then the fact that T O⊥ is Riemannian implies that (m1 , m2 ) = (n1 + k, n2 ), where k is the rank of T O⊥ . In particular, n0 = m0 in this case and the result follows from [10]. In case n0 = n1 , we can replace the metric on T O⊥ by its negative to obtain a new G-invariant metric for which the signature of M is (m1 , m2 ) = (n1 , n2 + k),

isometric actions of simple groups and transverse structures / 251

where k is again the rank of T O⊥ . Hence, for this new metric n0 = m0 and so the result follows also from [10]. We now proceed to prove Theorem 1.2. = (G × K \H)/ Proof of Theorem 1.2. Assume that (1) holds. Let us take M and verify that it satisﬁes (2). Note that, with respect to the quotient map G × H → (G × H)/ , the set G projects onto the G-orbit of the class of the identity. Hence, there is a dense G-orbit in M. Next, endow G × H with the product metric given by a bi-invariant metric on G and a Riemannian metric on H, which is left K -invariant and right H-invariant. The latter exists because K is compact. This induces a pseudo-Riemannian metric on G × K \H, which is left G-invariant and right -invariant. Note that the G-orbits are nondegenerate with normal bundle given by the tangent bundle to the factor K \H. Furthermore, by construction the projection G × K \H → K \H is a pseudo-Riemannian submersion and so it deﬁnes a transverse Riemannian structure for the foliation by G-orbits. Since the left G-action and the -action commute with each other, the transverse Riemannian structure on G × K \H induces a corresponding one on the double coset (G × K \H)/ , which is G-invariant. This provides the geometric structures required by (2). has a dense GLet us now assume that (2) holds, so that in particular M orbit and carries the indicated geometric structures. Since the G-orbits in M = TO ⊕ are nondegenerate then we have an orthogonal decomposition T M ⊥ T O which is invariant under G. The existence of a transverse Riemannian and Lemma 4.1 imply the existence structure for the foliation by G-orbits in M of a G-invariant Riemannian metric on T O⊥ . Hence, if we replace with such we obtain a new Riemannian metric the metric on T O⊥ induced from M, G-invariant pseudo-Riemannian metric on M for which T O⊥ is Riemannian. satisﬁes the hypotheses of Proposition 4.2, and Hence, the G-action on M and thus of M, with the properties the latter provides a ﬁnite covering of M, required by (1). Finally, let us assume that G has ﬁnite center and real rank at least 2. First observe that (1) replaced by the conditions in the last part of the statement still implies (2). The only nontrivial property to check is the existence of a dense G-orbit in (G × K \H)/ for an irreducible lattice. But in this case, G is dense in G × H (see for example lemma 6.3 from [10]), which yields the

252 / r. quiroga-barranco

required dense orbit. On the other hand, that (2) implies (1) with the properties from the last part of the statement is a consequence of Proposition 4.2.

5. Actions on Lorentzian Manifolds In this section, we present a characterization of the G-actions on M preserving a Lorentzian metric.

theorem 5.1. Let G be a connected noncompact simple Lie group acting faithfully on a compact manifold M. Then the following conditions are equivalent: 1) The group G is locally isomorphic to SL(2, R) and there is a G-equivariant ﬁnite covering map (G × K \H)/ → M where H is a connected Lie group with a compact subgroup K and ⊂ G × H is a discrete cocompact subgroup such that G is dense in G × H. → M for which the G-action on M lifts to a 2)There is a ﬁnite covering map M with a dense orbit and preserving a Lorentzian metric. G-action on M = (G × K \H)/ Proof. Let us assume (1) and consider the ﬁnite covering M → M. Endow G × H with the product metric given by the bi-invariant metric coming from the Killing form in g and a left K -invariant and right H-invariant Riemannian metric on H. Then consider the induced G-invariant metric on Since G is locally isomorphic to SL(2, R), such metric on M is Lorentzian. M. As in the proof of Theorem 1.2, the density of G implies the existence of a This proves (2). dense G-orbit in M. Let us now assume that (2) holds. By Lemma 2.6, for every x ∈ M, the metric in Tx O corresponds to an Ad(G)-invariant symmetric bilinear form on g. Such form is either 0 or nondegenerate. The former case implies the existence of a null tangent subspace of dimension at least 3, which is impossible; in particular, the G-orbits are nondegenerate. Since G is noncompact and simple, we conclude the existence of an Ad(G)-invariant form on g, which is Lorentzian. But sl(2, R) is the only simple Lie algebra admitting such a form, which implies that G is locally isomorphic to SL(2, R). We also conclude that T O⊥ is Riemannian and so the rest of the claims in (1) follow from Proposition 4.2.

6. Manifolds with a Transverse Parallelism: Proof of Theorem 1.3 The following result is an immediate consequence of the properties of Lie foliations. A proof can be found in [10].

isometric actions of simple groups and transverse structures / 253

lemma 6.1. Let X be a compact manifold carrying a foliation with a transverse Lie structure. If the foliation has a dense leaf, then the lifted foliation to any ﬁnite covering space of X has a dense leaf as well. We now obtain the next result, which describes actions preserving a metric and a transverse parallelism. Its proof is based on some of the arguments found in [11].

proposition 6.2. Suppose that the G-action on M has a dense orbit and preserves a pseudo-Riemannian metric. Also assume that M is compact. If the foliation by G-orbits is nondegenerate (with respect to the pseudo-Riemannian metric) and carries a transverse parallelism, then there exist → M, 1) a ﬁnite covering map M 2)a connected Lie group H, and 3)a discrete cocompact subgroup ⊂ G × H such that G is dense in G × H, so that M is G-equivariantly diffeomorphic for which the G-action on M lifts to M to (G × H)/ . Furthermore, if G has ﬁnite center and real rank at least 2, then we can assume that G × H is a ﬁnite center isotypic semisimple Lie group and is an irreducible lattice. Proof. By Lemma 4.1, the transverse parallelism to the G-orbits yields a Ginvariant trivialization of L(T O⊥ ). Hence, there is a family of G-invariant sections of T O⊥ , say X1 , . . . , Xk , that deﬁnes a basis of T O⊥ on every ﬁber. Let us consider the G-invariant Riemannian metric on T O⊥ for which these vector ﬁelds are orthonormal at every point. Because of the orthogonal decomposition TM = T O ⊕ T O⊥ , if we replace the metric on T O⊥ induced from M with the G-invariant Riemannian metric thus deﬁned from the parallelism, then we obtain a pseudo-Riemannian metric h on M, which is G-invariant, deﬁnes the same orthogonal complement T O⊥ to the orbits and such that this orthogonal complement is Riemannian. In the rest of this proof we will consider M endowed with this new metric h. Hence, by Corollary 3.3 there and a discrete is a simply connected homogeneous Riemannian manifold N for which there is a G-equivariant ﬁnite cocompact subgroup ⊂ G × Iso0 (N) covering map = (G × N)/ → M. M Note that by the proof of Corollary 3.3, the normal bundle T O⊥ is integrable. In particular, the vector ﬁelds Xi given above satisfy:

254 / r. quiroga-barranco

[Xi , Xj ] =

k

fijr Xr

r=1

for some smooth functions fijr deﬁned on M. The G-invariance of the parallelism then implies that the functions fijr are G-invariant and so constant because of the existence of a dense G-orbit. We conclude that the linear span over R of the vector ﬁelds X1 , . . . , Xk is a Lie algebra. On the other hand, the fact that the vector ﬁelds Xi are preserved by the G-action is easily seen to imply that such ﬁelds are foliate in the notation of [5]. Hence, the parallelism X1 , . . . , Xk deﬁnes a transverse Lie structure for the foliation by G-orbits in M. Clearly, this transverse Lie structure induces a Let H be a simply connected Lie group that models corresponding one on M. and consider a corresponding development the transverse Lie structure on M D : G × N → H. In particular, D is a submersion whose ﬁbers have con × {x}. Hence, we nected components given precisely by subsets of the form G conclude that D{e}×N : N → H deﬁnes a local diffeomorphism. is the By the proofs of Corollary 3.3 and Theorem 1.1, the manifold N (isometric) universal covering space of a leaf N for the foliation in M deﬁned 1 , . . . , X k be the pull-backs to N of the restrictions by T O⊥ . Hence, if we let X (which we will of the ﬁelds X1 , . . . , Xk to N, then the induced ﬁelds on G × N × N. denote with the same symbols) deﬁne the transverse Lie structure on G Furthermore, from the previous construction of the metric h, the metric on is given by the condition of the ﬁelds X 1 , . . . , X k being orthonormal at every N point. In particular, if we let h be the Lie algebra of H, then there is a basis i into v1 , . . . , vk of h such that the development D maps the vector ﬁeld X vi , for every i = 1, . . . , k. Without loss of generality we will assume that the transverse H-structures are modeled by taking H with its right translations, and so transverse Lie parallelisms are modeled on right-invariant vector ﬁelds. With this convention, each vi is considered as a right-invariant vector ﬁeld on H. From the above remarks, if we endow H with the right-invariant Riemannian metric for which the vector ﬁelds v1 , . . . , vk are orthonormal at every → H is a local isometry. By corollary 29 on page 202 point, then D{e}×N :N → H is an isomof [9] and since N is complete, we conclude that D{e}×N :N a Lie group structure with respect to which it is etry and so it induces in N with an isomorphism. Hence, we can replace the Riemannian manifold N H carrying the above right invariant Riemannian metric and assume that the × H → H is a development for the transverse Lie strucnatural projection G the space G × N covers M, ture on M. Moreover, by the deﬁnition of M, and so we can assume that the natural projection π : G × H → H is also a

isometric actions of simple groups and transverse structures / 255

with a corresponding development for the transverse Lie structure on M holonomy representation ρ : → H. = G × Iso0 (H), if we choose γ ∈ , then we can Since ⊂ G × Iso0 (N) write γ = (Rγ1 , γ2 ), where γ1 ∈ G and γ2 ∈ Iso0 (H). And so, the ρ -equivariance of π yields for every (g, x) ∈ G × H: ρ (γ ) = xy γ2 (x) = π (gγ1 , γ2 (x)) = π((g, x)γ ) = π (g, x) where y ∈ H. It follows that γ2 is given by a right translation by an element in = (G × H)/ → M is a H, and so we have ⊂ G × R(H) = G × H. Since M has a dense G-orbit and so ﬁnite covering, by Lemma 6.1 we conclude that M G is dense in G × H as in the proof of Proposition 4.2. This concludes the proof of the ﬁrst part of the statement. For the last claim, if G has ﬁnite center and real rank at least 2, then H is semisimple by the main results from [16]. Also, by modding out by a suitable central group we can assume that H has ﬁnite center (see [10]). The arguments at the end of Section 6 from [10] also prove that is an irreducible lattice. Finally, recall that an irreducible lattice in a semisimple Lie group can only exist if the group is isotypic. We can now prove our characterization of actions with a transverse parallelism. Proof of Theorem 1.3. First, let us assume that (1) holds and consider the = (G × H)/ → M. As in the proof of Theorem 1.2 we ﬁnite covering M conclude that M has a dense G-orbit. Note that the right-invariant vector ﬁelds on H deﬁne both a transverse Lie structure on G × H for the foliation given by the factor G, and a right H-invariant Riemannian metric on H. If we consider the product pseudo-Riemannian metric on G × H using a bi-invariant metric on G, then we also obtain a pseudo-Riemannian metric for which the foliation given by the factor G is nondegenerate. Both of these geometric structures on G × H are left G-invariant and right -invariant and so descend to thus establishing (2). corresponding G-invariant geometric structures on M yields (1). The last If we now assume (2), then Proposition 6.2 applied to M claim is also a consequence of Proposition 6.2.

7. Proof of Theorem 1.4 Let us consider G and M as in Proposition 2.3 with S ⊂ M a dense subset provided by this result and Remark 2.4. With the notation of Proposition 2.3, the representation λx ◦ ρx leaves invariant the subspace Tx O⊥ thus deﬁning its

256 / r. quiroga-barranco g-module structure. This induces a homomorphism g → so(Tx O⊥ ) obtained by restricting λx ◦ ρx to the submodule Tx O⊥ . By composition with such

homomorphism, we can consider x obtained from Lemma 2.5 as a map ∧2 Tx O⊥ → so(Tx O⊥ ). For a vector space W with inner product ·, ·, we will say that a bilinear map T : W × W → gl(W ) is of curvature type if it satisﬁes the following conditions for every x, y, z, v, w ∈ W : 1) 2) 3) 4)

T (x, y) = −T (x, y), T (x, y)v, w = − v, T (x, y)w , T (x, y)z + T (y, z)x + T (z, x)y = 0, and T (x, y)v, w = T (v, w)x, y .

Note that (1) and (2) together are equivalent to T inducing a map ∧2 W → so(W ). Also, by the proof of proposition 36 on page 75 of [9] we know that (1), (2), and (3) together imply (4). With the above notation, we have the following result.

lemma 7.1. Let G, M and S ⊂ M be as in Proposition 2.3 and Remark 2.4. For every x ∈ S, deﬁne the bilinear operation [·, ·]0 in Tx M by the assignments: • [X ∗ , Y ∗ ]0 = [X , Y ]∗ , for every X ∗ , Y ∗ ∈ Tx O , (X , Y ∈ g), x x x x x • [X ∗ , v]0 = −[v, X ∗ ]0 = X (v) for every X ∗ ∈ Tx O , v ∈ Tx O ⊥ , (X ∈ g), and x x x • [v1 , v2 ]0 = x (v1 , v2 )∗ for every v1 , v2 ∈ Tx O ⊥ . x

Then, [·, ·]0 yields a Lie algebra structure on Tx M if and only if x is of curvature type when considered as a map ∧2 Tx O⊥ → so(Tx O⊥ ). In this case, the representation of g in Tx M, given by Proposition 2.3(4), preserves the Lie algebra structure of Tx M. Proof. Note that by the above remarks, x is of curvature type if and only if it satisﬁes the above condition (3). Also observe that [·, ·]0 always deﬁnes a skew-symmetric bilinear form in Tx M. Hence, for the ﬁrst claim, we need to show that [·, ·]0 satisﬁes the Jacobi identity if and only if x satisﬁes (3). For the Jacobi identity to hold we only need to verify the following cases. • u1 , u2 , u3 ∈ Tx O . Note that [·, ·]0 maps Tx O × Tx O into Tx O in such a

way that it deﬁnes a skew-symmetric operation that corresponds to the Lie brackets of g under the natural isomorphism g → Tx O given by X → Xx∗ . Hence, the Jacobi identity always holds in this case. • u1 , u2 ∈ Tx O and u3 ∈ Tx O ⊥ . In this case we can write u1 = X ∗ and x u2 = Yx∗ for some X , Y ∈ g. Then, by the deﬁnition of [·, ·]0

isometric actions of simple groups and transverse structures / 257

[[Xx∗ , Yx∗ ]0 , u3 ]0 = [[X , Y ]∗x , u3 ]0 = [X , Y ](u3 ) = X (Y (u3 )) − Y (X (u3 )) = X ([Yx∗ , u3 ]0 ) − Y ([Xx∗ , u3 ]0 ) = [Xx∗ , [Yx∗ , u3 ]0 ]0 − [Yx∗ , [Xx∗ , u3 ]0 ]0 , which proves that the Jacobi identity holds in this case in general. • u1 ∈ Tx O and u2 , u3 ∈ Tx O ⊥ . We can now choose X ∈ g such that u1 =

Xx∗ . Hence, using from Lemma 2.5 the fact that x is a homomorphism of g-modules, we obtain [Xx∗ , [u2 , u3 ]0 ]0 = [Xx∗ , x (u2 , u3 )∗x ]0 = [X , x (u2 , u3 )]∗x = x (X (u2 ), u3 )∗x + x (u2 , X (u3 ))∗x = x ([Xx∗ , u2 ]0 , u3 )∗x + x (u2 , [Xx∗ , u3 ]0 )∗x = [[Xx∗ , u2 ]0 , u3 ]0 + [u2 , [Xx∗ , u3 ]0 ]0 , which yields again the Jacobi identity without extra conditions.

• u1 , u2 , u3 ∈ Tx O ⊥ . The deﬁnition of [·, ·]0 yields now:

[[u1 , u2 ]0 , u3 ]0 = [ x (u1 , u2 )∗x , u3 ]0 = x (u1 , u2 )(u3 ), and so the Jacobi identity is satisﬁed in this case exactly when x satisﬁes condition (3). The above proves the equivalence in the statement, and so it remains to obtain the last claim. For this we need to show that X ([u1 , u2 ]0 ) = [X (u1 ), u2 ]0 + [u1 , X (u2 )]0 , for every X ∈ g and u1 , u2 ∈ Tx M. This is now dealt with through the following cases, which just basically apply the deﬁnitions involved and properties already considered. • u1 , u2 ∈ Tx O . Then, we can write u1 = Y ∗ , u2 = Z ∗ for some Y , Z ∈ g x x

and X ([Yx∗ , Zx∗ ]0 ) = X ([Y , Z]∗x ) = [ρx (X ), [Y , Z]∗ ]x = [X , [Y , Z]]∗x = [[X , Y ], Z]∗x + [Y , [X , Z]]∗x = [[X , Y ]∗x , Zx∗ ]0 + [Yx∗ , [X , Z]∗x ]0 = [[ρx (X ), Y ∗ ]x , Zx∗ ]0 + [Yx∗ , [ρx (X ), Z ∗ ]x ]0 = [X (Yx∗ ), Zx∗ ]0 + [Yx∗ , X (Zx∗ )]0 .

258 / r. quiroga-barranco • u1 ∈ Tx O and u2 ∈ Tx O ⊥ . Now we can write u1 = Y ∗ for Y ∈ g and x

X ([Yx∗ , u2 ]0 ) = X (Y (u2 )) = [X , Y ](u2 ) + Y (X (u2 )) = [[X , Y ]∗x , u2 ]0 + [Yx∗ , X (u2 )]0 = [[ρx (X ), Y ∗ ]x , u2 ]0 + [Yx∗ , X (u2 )]0 = [X (Yx∗ ), u2 ]0 + [Yx∗ , X (u2 )]0 . • u1 , u2 ∈ Tx O ⊥ . We now have

X ([u1 , u2 ]0 ) = X ( x (u1 , u2 )∗x ) = [ρx (X ), x (u1 , u2 )∗ ]x = [X , x (u1 , u2 )]∗x = x (X (u1 ), u2 )∗x + x (u1 , X (u2 ))∗x = [X (u1 ), u2 ]0 + [u1 , X (u2 )]0 .

The following result will allow us to prove Theorem 1.4. It also provides an explicit description of the Lie algebra structure considered in Theorem 1.4.

theorem 7.2. Suppose that G has ﬁnite center and real rank at least 2, and that the G-action on M preserves a ﬁnite-volume complete pseudo-Riemannian metric. Also assume that G acts ergodically on M and that the foliation by G-orbits is nondegenerate. Denote with L the algebraic hull for the G-action on the bundle L(T O⊥ ) and with l its Lie algebra. In particular, there is an embedding of Lie algebras l → so( p, q), where ( p, q) is the signature of the metric of M restricted to T O⊥ . If this embedding is surjective, then one of the following occurs: 1) The conclusion of Theorem 1.1 holds, or 2)G is locally isomorphic to SO0 ( p, q), dim (M) = ( p + q)( p + q + 1)/2, for some x ∈ M the bilinear map x is nonzero and of curvature type and the Lie algebra structure on Tx M obtained from Lemma 7.1 is isomorphic to either so( p, q + 1) or so( p + 1, q). Proof. Since the G-orbits are nondegenerate, G preserves a pseudoRiemannian metric on T O⊥ . Hence, the algebraic hull of L(T O⊥ ) for the G-action can be embedded into the structure group O( p, q) for such metric, where ( p, q) is the signature of T O⊥ . This yields the ﬁrst claim. Let us now assume that the induced embedding l → so( p, q) is surjective. This implies that L is a ﬁnite-index subgroup of O( p, q). By section 3 of [17], we can write L = ZS where S is semisimple without compact factors, Z is compact

isometric actions of simple groups and transverse structures / 259

and centralizes S, and the product is almost direct. In particular, we have a direct product l = z × s. Note that for p + q ≤ 2 the conclusion of Theorem 1.1 holds by Corollary 3.1(1). Hence, we can assume from now on that p + q ≥ 3. Then, the only cases in which so( p, q) is not simple is for so(4) ∼ = so(3) × so(3) ∼ and so(2, 2) = sl(2, R) × sl(2, R). And so, one of the following holds: • s = 0 and z • z = 0 and s

∼ = so( p, q) is compact, or ∼ = so( p, q) is noncompact.

In the ﬁrst case, we conclude that so(Tx O⊥ ) is compact for every x ∈ M, thus implying that the conclusion of Theorem 1.1 follows by Corollary 3.1(1). In particular, we can assume that l = s ∼ = so( p, q). Also, by the arguments in section 3 of [17], there is a surjection g → s of Lie algebras, which implies that g∼ = so( p, q). Hence, G is locally isomorphic to SO0 ( p, q) and because of the decomposition TM = T O ⊕ T O⊥ we also have dim (M) = dim (SO0 ( p, q)) + dim (Tx O⊥ ) = dim (SO0 ( p, q)) + p + q = ( p + q)( p + q + 1)/2. On the other hand, for S ⊂ M given as in Proposition 2.3, if |S = 0, then the conclusion of Theorem 1.1 holds by Lemma 2.5(2). Then, we can choose x ∈ S such that x = 0. In particular, by Lemma 2.5(1) the g-module Tx O⊥ is a nontrivial one. If we consider the representation g → so(Tx O⊥ ) that deﬁnes the g-module structure of Tx O⊥ from Proposition 2.3, then the above remarks show that this representation is an isomorphism. We conclude that Tx O⊥ is a g-module isomorphic to Rp,q with respect to the isomorphism g → so(Tx O⊥ ) thus obtained. For ·, ·p,q the metric in Rp,q preserved by so( p, q) we have the following: Claim: For every c ∈ R, the map Tc : ∧2 Rp,q → so( p, q) given by Tc (u ∧ v) = c v, ·p,q u − c u, ·p,q v, where u, v ∈ Rp,q is a well-deﬁned homomorphism of so( p, q)-modules. Moreover, these maps exhaust all the so( p, q)-module homomorphisms ∧2 Rp,q → so( p, q). The only nontrivial part of the claim is the last statement, which we will now prove. Let T : ∧2 Rp,q → so( p, q) be a homomorphism of so( p, q)-modules. Consider the map T ◦ T1−1 , which is a homomorphism of so( p, q)-modules so( p, q) → so( p, q). Hence, for n = p + q and by complexifying, the map T ◦ T1−1 yields a homomorphism T : so(n, C) → so(n, C) of so(n, C)-modules −1 such that T |so( p,q) = T ◦ T1 . In particular, T (so( p, q)) ⊂ so( p, q) and so T commutes with the conjugation σ of so(n, C) whose ﬁxed point set is so( p, q), that is, T ◦σ = σ ◦ T . Note that since g ∼ = so( p, q) is simple with real rank at least 2, the Lie algebra so(n, C) is simple as well. Hence, the irreducibility of

260 / r. quiroga-barranco so(n, C) as so(n, C)-module implies, by Schur’s Lemma, that there is a com-

plex number c such that T = cIdso(n,C) . But then, the condition T ◦σ = σ ◦ T implies that c is real and so T = Tc . By the Claim and Lemma 2.5(1), with respect to the above-established isomorphisms g → so(Tx O⊥ ) and Tx O⊥ ∼ = Rp,q , we conclude that the map x corresponds to a homomorphism Tc for some real c = 0. It is straightforward to check that Tc is of curvature type; in fact, it yields the curvature of the pseudo-Riemannian manifolds of constant sectional curvature c and signature ( p, q) (see corollary 43 on page 80 of [9]). Hence, Lemma 7.1 applies and it provides a Lie algebra structure on Tx M. It remains to show that such structure is isomorphic to either so( p, q + 1) or so( p + 1, q). By our identiﬁcations, the Lie algebra structure on Tx M is isomorphic to the one obtained on so( p, q) ⊕ Rp,q from the map Tc with the Lie brackets [·, ·]c deﬁned with formulas similar to those in Lemma 7.1 replacing x with Tc . Also, it is straightforward to show that the map given by (so( p, q) ⊕ Rp,q , [·, ·]c ) → so(Rp+q+1 , Ip,q (c)) X cu X +u → , u t Ip,q 0 is an isomorphism of Lie algebras, where Ip,q Ip,q (c) = 0

0 . c

Since so(Rp+q+1 , Ip,q (c)) is clearly isomorphic to either so( p, q + 1) or so( p + 1, q) this yields (2) from our statement. We can now complete the following proof. Proof of Theorem 1.4. Since the hypotheses of Theorem 7.2 are a subset of those of Theorem 1.4, the conclusions of the former hold. Since M is weakly irreducible its universal covering space cannot split isometrically and so part (2) of Theorem 7.2 necessarily holds at some x ∈ M. By the deﬁnition of the Lie algebra structure in Tx M from Lemma 7.1 we conclude that (1) and (2) of Theorem 1.4 are satisﬁed. Finally we observe that, by the proof of Theorem 7.2, we can assume that x ∈ S where S ⊂ M is given by Proposition 2.3. Consider the Lie algebra g(x) of local Killing ﬁelds provided by Proposition 2.3. Then, the last claim of Lemma 7.1 and the deﬁnition of the representations involved implies the last claim of Theorem 1.4 for the Lie algebra g(x).

isometric actions of simple groups and transverse structures / 261

References [1] [2]

[3] [4] [5]

[6] [7] [8] [9] [10] [11] [12] [13] [14] [15]

[16] [17] [18]

A. Candel and R. Quiroga-Barranco, Gromov’s centralizer theorem, Geom. Dedicata 100 (2003), 123–155. M. Gromov, Rigid transformations groups, in Géométrie différentielle, Colloque Géométrie et Physique de 1986 en l’honneur de André Lichnerowicz (D. Bernard and Y. Choquet-Bruhat, eds.), Hermann, Paris, 1988, 65–139. F. G. Hernández-Zamora, Isometric splitting for actions of simple Lie groups on pseudoRiemannian manifolds, Geom. Dedicata 109 (2004), 147–163. N. Kowalsky, Noncompact simple automorphism groups of Lorentz manifolds and other geometric manifolds, Ann. of Math. (2) 144 (1996), no. 3, 611–640. P. Molino, Riemannian foliations, translated from the French by Grant Cairns, with appendices by Cairns, Y. Carrire, É. Ghys, E. Salem, and V. Sergiescu, Progress in Mathematics, 73, Birkhäuser Boston, Boston, MA, 1988. K. Nomizu, On local and global existence of Killing vector ﬁelds, Ann. of Math. 72 (1960), 105–120. G. Olafsson and R. Quiroga-Barranco, Isometric actions of SO0 ( p, q) on low dimensional weakly irreducible pseudo-Riemannian manifolds, in preparation. B. O’Neill, The fundamental equations of a submersion, Michigan Math. J. 13 (1966), 459–469. B. O’Neill, Semi-Riemannian geometry. With applications to relativity, Pure and Applied Mathematics, 103. Academic Press, New York, 1983. R. Quiroga-Barranco, Isometric actions of simple Lie groups on pseudo-Riemannian manifolds, Ann. of Math. (2) 164 (2006), no. 6, 941–969. R. Quiroga-Barranco, Arithmeticity of totally geodesic Lie foliations with locally symmetric leaves, Asian J. Math. 12 (2008), no. 3, 289–297. I. M. Singer, Inﬁnitesimally homogeneous spaces, Comm. Pure Appl. Math. 13 (1960), 685–697. J. Szaro, Isotropy of semisimple group actions on manifolds with geometric structure, Amer. J. Math. 120 (1998), 129–158. A. Zeghib, On afﬁne actions of Lie groups, Math. Z. 227 (1998), no. 2, 245–262. R. J. Zimmer, Actions of semisimple groups and discrete subgroups, Proceedings of the International Congress of Mathematicians, Vol. 1, 2 (Berkeley, CA, 1986), 1247–1258, American Mathematical Society, Providence, RI, 1987. R. J. Zimmer, Arithmeticity of holonomy groups of Lie foliations, J. Amer. Math. Soc. 1 (1988), no. 1, 35–58. R. J. Zimmer, On the algebraic hull of an automorphism group of a principal bundle, Comment. Math. Helv. 65 (1990), no. 3, 375–387. R. J. Zimmer, Automorphism groups and fundamental groups of geometric manifolds, in Differential Geometry: Riemannian Geometry (Los Angeles, CA, 1990), 693–710, Proc. Sympos. Pure Math. 54, Part 3, American Mathematical Society, Providence, RI, 1993.

8

shmuel weinberger

SOME REMARKS INSPIRED BY THE C0 ZIMMER PROGRAM to bob zimmer, with friendship, gratitude, and admiration

In the mid-1980s, when I had just arrived in Chicago, Bob Zimmer was lecturing on his recent work showing how perturbations of isometric actions of Kazhdan groups were dynamically simple (preserve a measurable Riemannian metric) [Zi1]. The heady mixture of the beauty of the results, the audacity of the vision (this is really 0% of what should be true.. . .1 ), and Bob’s charisma and charm attracted me so that I still cannot but help think about the problem of what large discrete groups of Cat-isomorphisms of a manifold can conceivably look like. Alas, my ardor has not been well rewarded in this pursuit. In this paper I would like to make some comments and raise some questions about the case of Cat = Top, that is, the group of homeomorphisms of manifolds. Some of these remarks are rather old, but I will give some details if I am not aware of a discussion elsewhere—while a few are new (at least to me). I hope that this paper spurs more exploration of the fascinating C0 aspect of this nexus of problems. Although this note is short, it has a bit of structure. The beginning is about groups that don’t act, the middle is about lattices and hyperbolic groups and quite tame actions, and the end has an ergodic aspect and relates directly to a paper of Lashof and Zimmer [LZ]. There is also an appendix about deforming homotopy actions to actions at the cost of stabilization. Besides Bob Zimmer, who is the obvious inspiration for this paper, I would also like to thank Kevin Whyte, and the 3Fs: Benson Farb, Steve Ferry, and David Fisher for many conversations that have strongly inﬂuenced my perspective. Courtney Thatcher’s thesis on cyclic group actions on products of spheres has some overlap with the perspective in Section 4. 1. Maybe now we are above the 0% mark, at least in this direction, thanks to the wonderful papers [Be] and [FiM].

262

some remarks inspired by the c0 zimmer program / 263

1. Some Groups That Don't Act Proposition 1. The inﬁnite special linear group SL(Z) does not act topologically, nontrivially, on any compact manifold, or indeed on any manifold whose homology with coefﬁcients in a ﬁeld of positive characteristic is ﬁnitely generated. [Zi2] only mentions the compact case. The more general case is proven simultaneously (and is necessary for the proof, because one must analyze complements of ﬁxed sets as well as ﬁxed sets in the coming argument), and is motivated by the recent paper [ABJLMS], as is Proposition 2 below (see also [FiS] that was written very recently and has some quite deep further observations about the volume-preserving case). Proof. The proof is based on Smith theory considerations. A p-group acting on a mod p-homology manifold has ﬁxed set that is also a mod p-homology manifold of lower dimension. Moreover, the sum of the Betti numbers of the ﬁxed sets can be at most as large for the ﬁxed set as for the original manifold. As a formal consequence, by considerations of the open strata, if (Z/p)k is acting effectively, for k large enough, there must be a very large (Z/p)1 acting freely on some mod p-homology manifold with bounded Betti numbers. However, if l is large enough, this action will automatically have a large kernel for its action on homology (GLr doesn’t have that much p-torsion). Consequently, one can compute the homology of the quotient space by means of the usual Serre spectral sequence. Now if the rank of this kernel is greater than the homology of the space, one sees that the quotient is inﬁnite-dimensional. Putting this all together, we get elements of order p that are acting trivially on M. Since SL(Z) is the normal closure of such elements, we see that the action is actually trivial. Proposition 2. There are ﬁnitely presented groups that do not act nontrivially on any compact manifold. Proof. The idea is to produce ﬁnitely presented groups that contain high-rank ﬁnite subgroups and that are normally generated by enough elements of order p. Many examples are constructed in [ABJLMS]; it might be worth noting that old ideas related to the unsolvability of the triviality problem (a la Rabin) can be used for this. We follow the exposition in [We1].Suppose that π is an effectively acting group with arbitrarily large elementary p-groups generated by speciﬁc words we know. We can make inﬁnitely many HNN extensions, each

264 / s. weinberger

one conjugating one of these elements to another, so that they are now all conjugate in a larger group, called π . We denote by g one of these elements of order p. Now, Higman embed π into a group π . By taking an amalgamated free product (π ∗ Z)∗ Z∗ ZA where A is a knot group (say the trefoil group for concreteness), Z∗ Z is a free subgroup of A that contains a meridian as its ﬁrst generator, and Z∗ Z is embedded into (π ∗ Z) so that the ﬁrst generator goes to [g, t] in the obvious notation. We thus have arranged that any homomorphism that sends any of the elements of order p in our list to e kills the whole group. Whether one prefers Proposition 1 or 2 is a matter of taste. Of course these both beg the question of whether there is a torsion-free group that does not act continuously on any manifold. We note: Proposition 3. Any countable group of ﬁnite homological dimension does have a continuous action on some sphere. Proof. If Bπ is a ﬁnite-dimensional complex, then it can be thickened to a manifold. The universal cover of this manifold is contractible, and if not a Euclidean space, its product with R is (by a well-known theorem of Stallings). This manifold has a free π action. Its one point compactiﬁcation is a sphere, and it has a π action with a single ﬁxed point. This action is rather different than the usual ones we think of coming from symmetric spaces and lattices. It has no dynamics! A smooth version of Proposition 3 seems unlikely (but I have no ideas about how to prove such a thing). On the other hand, the hypothesis is far from necessary. Every manifold (of dimension >0) has an action of a free abelian group of inﬁnite rank. On the other hand, presumably the obvious versions of congruence subgroups of SL(Z) should not act, and we know nothing about this. Problem: What groups are discrete subgroups of Homeo(Dn , rel ∂)? Needless to say, they are torsion free (and they are subgroups of Homeo(M) for any m-manifold for m ≥ n).

2. Actions of Lattices Of course, Zimmer actually suggests that SLn (Z) shouldn’t act on any manifold of dimension < n–1 (except via ﬁnite quotients). In dimension n, the action

some remarks inspired by the c0 zimmer program / 265

on Tn should be rigid. Indeed, the beautiful work of Hurder shows that, with enough smoothness, that action is inﬁnitesimally rigid. It seems worth pointing out that in C0 there are many deformations possible. One deformed action is explained in [FS1], but the process is one that can be repeated inﬁnitely often producing uncountably many new and different actions. The technique here is “insertion” a la [KL]. Whenever one has a C1 action with a ﬁxed point (or indeed, ﬁnite orbits, or indeed, an invariant submanifold) one can glue in a tubular nighborhood acted on via the differential, and extend the action to the complement. Essentially one uses the diffeomorphism of the complement of a disk with the complement of a point inside M – D = M – p. In the neighborhood one can alternatively glue in the cone of the action on the sphere at inﬁnity. These actions are quite different from each other or the original action, as one can see by examining orbits of points. Now, if one wishes, one can insert many times, producing many rings around the ﬁxed point. Also, one can make some of these rings “thick” by coning for a little while before doing a more standard linear insertion. Clearly, this builds uncountably many actions that differ only around a ﬁxed point. (They are distinguished by one another by looking at the limit points of orbits. Note that for compact groups, the action given by insertion is topologically equivalent to the original action.) But these “bull’s-eye” structures can be inserted at will at smaller and smaller scales around any of the ﬁnite orbits associated to the rational points on the torus. As a result, we have beautiful moduli of “leopard spots,” each of which is itself variable according to the bull’s-eye patterns. Thus we have seen a much more beautiful structure than is asserted in the following: Proposition 4. There are uncountably many nontopologically-conjugate SLn (Z) actions on Tn . They can all be deformed to the usual action. These deformations are 1-parameter families of actions that start at the “exotic” action at time 0, and end at the standard action at time 1. And all of these actions map equivariantly to the standard afﬁne action. These actions are just “modiﬁcations” of arithmetic actions. In its most vague form, the Zimmer program asserts that, in some sense,2 all actions of high-rank lattices are built up out of such actions. Thus, Proposition 4 is just par for the course. 2. It is part of the program to ﬁgure out the correct sense.

266 / s. weinberger

There are other deformations that arise from decomposition space theory (aka “Bing topology”); I will give just one example and then move on. Proposition 5. Let π be a countable group with a dense embedding in SO(n). Then SO(n) acts on Sn+k with quotient D1+k . If one considers the induced action over D1+k / ∼ where ∼ is a cell-like semicontinuous decomposition (these exist in profusion if k > 1), the total space is still Sn+k , but the new action of π is never equivalent to any of the other ones. The equivalence of total spaces follows readily from Edwards’s characterization of manifolds and the CE approximation theorem (of Siebenmann, which is a consequence of Edwards’s theorem, as well), see [Da] and [We2] for such deformations. The actions are thus distinguished by the maximal Hausdorff quotients of their orbit spaces, which will be nonmanifold ANR homology manifolds. Part of the Zimmer program, although it is far from the usual part where the actions are ergodic, seems to me to include considerations of free, and more generally, proper discontinuous actions. Of course for these, in the geometric setting, the basic theorem is Mostow rigidity—and superrigidity can be easily viewed as a description of the inﬁnite covolume geometric actions that follow from the action. The C0 part of this portion of the Zimmer program is then an extension of the well-known Borel conjecture. On the topological side there is the fabulous and celebrated work of Farrell and Jones [FJ] that tells one that in the cocompact case, there are only the most obvious actions. For uniform lattices, it then follows that if one is interested in actions on Euclidean spaces, the dimension must be strictly larger than the homogeneous space, but there are interesting remarks that should be made about the proper discontinuous actions that exist in (1) the nonuniform case, (2) in dimensions larger than the original action, and (3) when there is torsion. Regarding the ﬁrst of these, it is clear that spirit of the Zimmer program demands that for any aspherical manifold with fundamental group the dimension should be at least that of G/K where G is the Lie group in which sits. It is the Q-rank that governs the difference between this dimension and the cohomological dimension of dim(G/K)–cd()=Q–rk(). When this is 1, it is not hard to give an ad hoc proof that there isn’t a geometric realization (the fundamental group of the end is a subgroup, which is a Poincaré duality group of dim=dim(G/K)–1 and of inﬁnite index). However, for higher Q-rank, I had been unable to settle it for a number of years, and was very happy when the papers [BKK] and [BF] came out and beautifully conﬁrmed the Zimmer predictions that lower-dimensional examples do not exist.

some remarks inspired by the c0 zimmer program / 267

Problem: In the minimal dimension (i.e., dim(G/K)), is there any useful way to parametrize the proper homotopy types of aspherical manifolds with fundamental group ? First of all, there is an issue of the fundamental groups at inﬁnity. It is very easy to build in any nonuniform case (of dim > 2) uncountably many modiﬁcations of the standard actions that are distinguished by the fundamental group system of the complements of compact sets (viewed as a prosystem of groups). This is a straightforward modiﬁcation of constructions of contractible manifolds. However, such actions have proper comparison maps to the standard action, and are concordant to it (i.e., there is a degeneration of each of these to the standard action). Much more interesting is the observation that Kevin Whyte made (when he was my student) that the manifolds with fundamental group (F2)k obtained by taking various products of thrice-punctured spheres and punctured tori are all different from each other even up to proper type. It would be quite interesting to get a systematic understanding in some sort of homological fashion. Here the problem should be much more doable when the Q-rank is large. Perhaps the answer is quadratic in part of the group homology when Qrank > 12 dim(G/K), cubic when the fraction is around 13 dim(G/K), and so on by analogy to the work of Goodwillie, Klein, and Weiss [GKW]. Nevertheless, this is quite unclear to me at the moment and the discussion below of case (2) suggests that it could be much simpler than that. If we assume that we have the correct proper homotopy type, then one might have a hope that the proper homotopy equivalence is properly homotopic to a homeomorphism, but this is rarely the case. As Stanley Chang and I noticed, when Q-rank > 2, there is always a ﬁnite cover where this fails [ChW].3 Our obstruction is of exponent 2 and our examples are all virtually standard. On the other hand, whenever H4i (; R) = 0 for some i, surgery arguments (e.g., like the ones in that paper that make use, of course, of the Borel-Serre picture of arithmetic manifolds [BoS]4 ) produce (inﬁnitely many) manifolds in the given proper homotopy type distinguished by pi, the i-th Pontrjagin class. These examples will not go away by passing to a ﬁnite-index subgroup. (In characteristic 0, cohomology can only get bigger on passing to a ﬁnite cover.) And, many examples then follow from known results (e.g., see Borel [Bo2]). 3. There is a good heuristic for proper rigidity when Q-rank=2. It is true when Q-rank2 boils down to the fundamental group condition at inﬁnity. (In [ChW] this is interpreted and veriﬁed using Margulis arithmeticity for all, even nonarithmetic, lattices.)

268 / s. weinberger

The situation of being above the cd in the uniform case (or the nonuniform case) is very similar to the issues involved in the proper nonuniform discussion we just had. However, in the uniform case we can be much more explicit about what occurs if, say, the dimension supercedes the cd by > 2 and we assume that the fundamental group system is trivial (i.e., equivalent to the constant system ). In that case, it is not hard to show that the manifold is the interior of a manifold with boundary, which by the theorem of Browder-Casson-HaeﬂigerSullivan-Wall (see [We3]) will automatically be a topological block bundle over K\G/ . The proper homotopy types are then parametrized by the homotopy classes of maps [K\G/ : BAut(Sc−1 )] where c is the difference of dimensions. Rationally this is isomorphic to a group cohomology. For c even, the invariant is simply the Euler class of this bundle. When c is odd, a relatively straightforward calculation shows that it lies in H2(c−1) (K\G/ ; Q) and all elements are realized.5 Note: as here c is the analogue of the Q-rank in the previous discussion, I currently hope that the answers might be simpler than one thinks in the nonuniform case. Including the results of classiﬁcation6 one gets: Summary Proposition 6: The actions of a uniform lattice on K\G × R c that have appropriate fundamental group systems at ∞ (in their quotient) → [K\G/:BTopc ] if c > 2. Rationally7 this can be calculated as a sum of group cohomology groups Ht (K\G/; Q) ⊕H4i (K\G/ ; Q) where t = c, if c is even, and t = 2 (c–1) if c is odd, and the sum is overall positive i. When there is torsion in the situation is much more complicated in several respects. Even in the situation of cocompact proper discontinuous actions there can be different failures of rigidity. Some examples appear in [CK]—and are based on failures of excision in algebraic and hermitian Ktheory, that is, the functors Nil and UNil. When nonzero, these groups tend to be inﬁnitely generated torsion groups.

5. The interpretation of this, not as well known as it should be, invariant is as follows: every odd-dimensional bundle (rationally) has a section that splits off a lower-dimensional bundle. While the Euler class of this complementary bundle is not uniquely determined, its square is. 6. See the discussion after Proposition 7 below. 7. Since there are no group structures here at least on the left, it is worth clarifying here that the map from left to right is ﬁnite to 1 and has image that contains a lattice in the target vector space.

some remarks inspired by the c0 zimmer program / 269

Rather different examples, based on a connection to embedding theory and somewhat technical arguments involving stratiﬁed space theory, can be found in [We3, We4]8 for actions of crystallographic groups. These were extended in [Shi] to many other lattices. They can be inﬁnitely generated nontorsion, for example, detected by analogues of the Alexander polynomial. However, all of these examples (e.g., the algebraic and the embedding theoretic) are virtually trivial in the sense that they become standard on passing to a sufﬁciently large cover that keeps all of the torsion. (When one passes to a cover, only some of the coefﬁcients of the polynomial survive, so any particular example will be killed on passing to a suitable ﬁnite cover.) This can be proved by a geometric argument related to [SW] (the equivariant version of the α-approximation theorem [ChF]) when there are no ﬁxed sets of subgroups that are included in one another in codimension < 3.9 It is worth noting that the above cohomological discussion about the proper analysis for nonuniform lattices can be married to the discussion of the role of the singularities (and embeddings) that come from torsion. In any case, sufﬁce it to say that the analogue of the ⊕H4i (K\G/; Q) part of the previous theorem is a sum of similar terms, one for each stratum that has G-rank > 2, and the form of the term is an equivariant K-group. As a result, nonuniform lattices of Q-rank > 2 with torsion almost never have proper rigidity. In any case, all of these examples certainly can be viewed as “obtained from some procedure applied to the arithmetic examples”—although the exact procedures involved are somewhat involved and might be hard to visualize.

3. Leaving the Aspherical Setting One of the audacious aspects of the Zimmer program is that the manifolds studied are general compact manifolds. Here is a theorem of this sort from [FW]. Theorem: If M = X /π is a compact manifold so that π = π 1 (M) is torsion free and has no normal abelian subgroups, and the isometry group Iso(X ) is not

8. We are just now getting to the point where we can give complete classiﬁcations of some proper cocompact actions that are not rigid. However, it is my feeling that the picture of this subject is currently too complicated to be able to give a survey here. 9. When there are codimension 2 situations, one can use counterexamples to the “Smith conjecture” on unknottedness of ﬁxed sets of cyclic group actions on the sphere, (e.g., [Gi]) to build knot theoretic failures of rigidity that do not die in any ﬁnite cover.

270 / s. weinberger

a compact extension of π , then π is a lattice in a semisimple Lie group and there is a Riemannian ﬁber bundle M → K\G/π . If π has torsion, one has to allow the possibility of “orbiﬁbering” over an orbifold. Note that all the ﬁbers are isometric here, and that therefore the structural group of the bundle is compact. Moreover, the representations π → Iso(F) control the possible M’s. (Taking a representation with inﬁnite image gives an example where no intermediate covers are Riemannian products. Of course many of them are differentiable products.) The following result, sketched in chapter 12 of [We3], frames our discussion. Proposition 7. Suppose that M= X /π is a compact manifold whose fundamental group is a torsion-free lattice in a Lie group and that H∗ (X) is ﬁnitely generated. Then, unless dimM = 4, there is a map M → K\G/π so that the inverse image of every open ball is homeomorphic to X . So from a slightly blurred perspective this map is like a ﬁber bundle map: rather than controlling the inverse image of each point, we control the inverse image of each little open ball. The relevant M’s that occur here, though, are much richer than the ones that occur in the previous theorem. Rather than homomorphisms from π to a Lie group, the objects that occur in this proposition are much more algebraic topological. For example, when X = Sk × R n , if one restricts attention just to the homotopically trivial ﬁbrations, then the relevant classiﬁcation is (independent of k > 110 ) the homotopy classes of maps [K\G/π : G/Top], where the space G/Top was analyzed completely by SullivanKirby-Siebenmann; rationally it is ⊕H4i (K\G/π ; Q). Incidentally, even when π has nontrivial homomorphisms to O(k + 1) they never give rise to nontrivial elements in this cohomology.11 . These actions of π on Sk × R n are never “Lie theoretic.” Conjecture 7 : If M is a manifold with torsion-free fundamental group M = X /π (π = π1 M) so that H∗ (X ) is ﬁnitely generated, then there is an aspherical

10. This is due to Gc /Topc stability as explained in [We3]. 11. These cohomology classes are essentially Pontrjagin classes (that can be very high-dimensional in Top despite the low dimensionality of the “ﬁber”); the vanishing for representations is immediate from the Chern-Weil description of characteristic classes.

some remarks inspired by the c0 zimmer program / 271

homology manifold N with fundamental group π, and for any such N, there is a map f: M →N so that for any small open set O ⊂ N, f−1 (O) < O × X . There are several things that need to be explained about this conjecture. The ﬁrst is that the maps allowed here, when N is a manifold, are exactly the maps produced in the previous theorem; moreover, despite allowing more general N’s than K\G/π in the conjecture, by Quinn’s work on resolution of homology manifolds [Q], the conjecture is a consequence of the theorem. Proposition 8. The conjecture is true for π ⊂ GLn (R) discretely embedded and of ﬁnite type12 (but not necessarily a lattice) and for hyperbolic groups, assuming that cd(π ) > 4. This follows the same lines as the previous proposition, aside from a few points. The ﬁrst is that one must exclude, say, nonuniform lattices. More precisely, if π is the fundamental group of an aspherical manifold, it must satisfy Poincaré duality. That Poincaré duality follows, assuming ﬁniteness of K(π , 1), from a manifold’s having a ﬁnite universal cover as shown in [BW1] (as is the coarseness of satisfying Poincaré duality among such groups)13 . Now, the following is an immediate consequence of [BFMW1]: Proposition 9. If π is a Poincaré duality group of dimension > 4 and satisﬁes the Borel conjecture (see e.g., [FRR]14 ), then there is an aspherical homology manifold with fundamental group π.

12. I believe that if B π has some inﬁnite skeleton, then there cannot be a manifold such as our M, but I haven’t excluded this yet. 13. Nonuniform lattices never satisfy Poincaré duality. They satisfy Bieri-Eckmann duality with an inﬁnitely generated dualizing module. 14. The Borel conjecture required here is the following standard generalization of the most commonly stated one: (ignoring orientation issues for simplicity) suppose that M is an aspherical manifold and f:W → M is a proper homotopy equivalence that is a homeomorphism outside of a compact set, then f is homotopic through such maps to a homeomorphism. For a group, we demand that this hold for all aspherical M with that fundamental group. This extends the usual Borel conjecture from fundamental groups of compact aspherical manifolds to all countable groups of ﬁnite cohomological dimensional, and in the case of fundamental groups of compact aspherical manifolds is equivalent to it if one considers M and M × tori simultaneously. (In other words a counterexample to the extended Borel for some bundle over M will translate into a counterexample to ordinary Borel for M × some torus.)

272 / s. weinberger

Thus the existence of the aspherical homology manifold follows from the results of [FJ] and [BL]. I will soon outline some evidence for not believing that it is possible to improve the existence result to be a manifold. In any case: Proposition 10. Among torsion-free hyperbolic groups, being the fundamental group of a closed aspherical manifold is a coarse condition; it follows from the Gromov boundary being a sphere. The idea for this appears in a discussion of the case of homotopy tori in [BFMW1]. One glues the boundary sphere onto the universal cover to obtain a homology manifold with boundary, and then relates the local index of the interior to that of the boundary. Details will appear in a forthcoming paper with Barthels and Lueck [BLW] that discusses more generally the role of the boundary. It is also worth noting that the conjecture (afﬁrmed in Proposition 8) includes the Borel conjecture in its usual formulation. If M and N are both aspherical manifolds with fundamental group π , there is then a CE map (see [Da]) M → N, which according to [Si] is automatically a uniform limit of homeomorphisms. The conjecture thus accomplishes the liberation of the Borel conjecture from the setting of aspherical manifolds. Finally, the map comes directly from the argument in [BFMW2]. The underlying homotopy theoretic assumption enables one to set up a controlled surgery problem, which is solved by that technology. More precisely, it produces a DDP15 homology manifold controlled homotopy equivalent to the original manifold that “approximately ﬁbers over N.” However, Edwards’s characterization theorem [Da], Quinn’s resolution theory [Q], and the αapproximation theorem [ChF] then apply to show that the homology manifold is homeomorphic to the original manifold M. The situation for groups with torsion is not nearly as pleasant: there are many more sources of obstruction—however, it is consideration of these that leads me to believe that the homology manifolds arising in the conjecture cannot be replaced by manifolds. The naive suggestion would be that π should act proper discontinuously on a contractible (homology) manifold C and that there should be an equivariant

15. DDP stands for “Disjoint Disks Property”; it demands of a space Z that any pair of maps of D2 → Z can be approximated by a disjoint pair. According to Edwards [Da] a resolvable homology manifold of dimension > 4 with this property is a manifold.

some remarks inspired by the c0 zimmer program / 273

map X → C that has the properties of the conjecture. This is completely deﬂated by the failure of the Nielsen conjecture [BW2]. In any case, one would suspect that C actually should have more properties,for example, that the ﬁxed set of every ﬁnite subgroup is also contractible, like K\G in the lattice case. However, even then there are some additional obstructions related to Nil and UNil that would have to vanish. If one is willing to cross M with a manifold with zero Euler characteristic, then one can get rid of the Nil obstructions and then, if π is a lattice with only odd torsion, in rank 1, then one can afﬁrm this conjecture—presumably it is only a matter of time—until the general case will follow. However, for general groups π , even when C exists, nothing implies that the action is locally smoothable. It is very possible for action near the ﬁxed set to be modeled on nonlinearizable (say homotopy linear) actions of a ﬁnite group on the sphere. (These are constructed in [BW1].) This seems to be entirely parallel to having homology manifolds that are locally homotopy spheres (=DDP) but not actually being “locally linear,” that is, resolvable. Remark: In [BFMW1] there were a number of conjectures made about the geometric topology of DDP homology manifolds. Despite the years that have passed almost no progress has been made. The reason for this is probably because they have never been “seen”: they are constructed as GromovHausdorff limits of polyhedra that are themselves constructed using fairly high-power machinery. If one of these arose as a boundary of a hyperbolic group, it would naturally have many self-homeomorphisms, and other additional structures that could lead an optimist to hope that it would be an importat step toward understanding the nonresolvable homology manifolds.

Appendix: Actions on M×Rn This brief appendix gives a bit more information about the possible monodromies possible for the manifolds just discussed (in the torsion-free case). But we shall phrase the result in terms of the Zimmer program. Here Cat is any of the geometric categories Top, PL, or Diff and Aut(?) is the space of self-homotopy equivalences of the space?. Proposition A1. Suppose that is a countable group of ﬁnite cohomological dimension and we have a homomorphism ρ : → π0 Aut(M). Then there is a Cat action of on M×Rn for some n iff there is a lift of Bρ to BAut(M, t) where (M, t) is the component of the space of maps M → BCat that contains

274 / s. weinberger

the stable tangent bundle of M, and Aut denotes the automorphism group. Moreover, the action can be taken free. Note: Note that if the original ρ : → π0 Aut(M) does not preserve the stable tangent bundle, then the automorphism space is empty and the proposition is vacuous. If M is stably parallelizable, then the condition is equivalent to just solving the lifting problem: BAut(M) ↓ Bρ : B → Bπ0 Aut(M) This propositon is the (un)natural marriage of Cooke’s obstruction theory for lifting homotopy actions to actions [Co] combined with Mazur’s theory of stable differential topology [Ma] where one uses ﬁnite dimensionality of B to get a manifold structure when “assembling.” The proof is quite straightforward and is left to the reader. (In the parallelizable case, one simply argues that the action produced by Cooke gives rise to a ﬁnite-dimensional CW complex with a map π1 → whose induced cover is homotopy equivalent to M. “Thickening” this complex to be a parallelizable manifold gives one whose cover is M × Rn according to Mazur.) Remark: On the other hand, for ﬁnite groups this approach seems doomed to failure. The theorem does give rise to obstructions: for example, the Nielson realization problem was disproved for nilmanifolds via this obstruction; see [RS]. However, for trivial ρ clearly it is impossible to ever accomplish the task achieved by the proposition, that is, constructing a free action. There is some literature on this problem for ﬁnite groups, but it is hemmed in by natural and strong hypotheses. Note that according to Sullivan [Su] and Wilkerson [Wi] when M is a compact (or even ﬁnite-type) simply connected manifold, π0 Aut(M) is an arithmetic group. In practice the lifting problem has its subtle aspects, and I hope to devote a future paper to some examples of it. The actions produced by the method are always free and proper discontinuous. As a result, the n here can be rather larger than the cd. (Think about the case of a product of free groups [BKK] and M = a point.) Much more in the spirit of the Zimmer program would be to produce actions with small values of n. With some trepidation, I would like to suggest:

some remarks inspired by the c0 zimmer program / 275

Conjecture: If M is a compact manifold of dimension less than that of the smallest representation of an irreducible lattice in a Lie group G of rank > 1, and n is dense in the torus, then this element is unique and is the unique continuous conjugating map. Note that g can be recovered by its eigenvalues on L2 (T). Corollary. If G is a compact Lie group, then any g can be conjugated to an element of a maximal torus. Translation by g is conjugate measure theoretically to h iff there is an element of GLn (Z) taking g to h (for the usual action of GLn (Z) as automorphisms of the torus). Much more interesting is the situation of continuous conjugacy. Corollary. (G, g) is topologically conjugate to (G, h) iff G/c1() ≈ G/cl() by an isomorphism that pulls back the principle cl = cl bundles.

16. Of course, many nonergodic actions—even with continuous deformation—appear in earlier sections.

276 / s. weinberger

As a special case, consider G = SU(n) and g generic, so that < g > is dense in T the maximal torus. Then the maps associated to g and h, SU(n)/T → BT classify the principle bundles differ by the automorphism of GLn (Z); in other words, these are the same bundles, but their structure as principle bundles is different. So we want to know what kinds of automorphisms of G/T are possible. The automorphisms of the cohomology algebra of this space was much studied and for G = SU(n) the only possibility is the Weyl group combined with complex conjugation (see, e.g., [EL]). Thus we obtain and generically there is rigidity: Corollary. For generic g ∈ SU(n), translation by g is only topologically conjugate to translation by elements conguate in SU(n) to g or its complex conjugate. Much more interesting is the situation for nongeneric elements. We will see that these are frequently topologically conjugate. We shall focus on the extreme cases: where cl is ﬁnite or a circle.17 First of all I want to show some conjugacies that are “soft,” that is, follow from general principles with no calculations.18 We will consider Zk where k is a product of a large number of primes. For each prime separately, we have the action of the Weyl group moving the generator to another element of the maximal torus. However, if there are r prime factors, one has something like (#Weyl)r−1 associated representations of Zk modulo action of the Weyl element on the whole maximal torus. Proposition 12.19 These “Weyl-mixed” translations are all topologically conjugate for G = SU(n), n > 2. Proof. For simplicity of notation let us consider the case of two factors. Thus we are considering G/(P×Q) versus G/(P ×Q ) where P and Q have relative

17. In general is a product Zr × T for some torus. 18. On the other hand, they have the usual shocking feeling that follows from applying obviously discontinuous p-adic constructions on standardly manifold theoretic objects: in other words, they are reminiscent of the use of Frobenius to get actions on etale homotopy types that arose in the proof of the Adams conjecture. 19. This proposition is true for all simply connected compact Lie groups other than SU(2); Indeed, I suspect it is true smoothly and for arbitrary compact Lie groups of rank > 1 and perhaps even more general abelian subgroups of G.

some remarks inspired by the c0 zimmer program / 277

orders, and P and P are conjugate in G, as are Q and Q . First of the quotients are homotopy equivalent20 by localization theory [HMR]. To build an equivalence between (simple) spaces,21 it sufﬁces to build rationally compatible equivalences between their localizations at each prime. At P, G/P → G/(P×Q) is an equivalence, and G/P ≈ G/P by a conjugacy that is homotopic to the identity as a map of G (it is induced by an element of G, which is connected). At Q, the argument is the same. At other primes, it’s even easier; both quotients are equivalent to G. These are all visibly rationally compatible. In fact they are simple homotopy equivalent. As n > 2, the Lie group we are considering has rank >1. Therefore one can use an extra circle from the maximal torus to compute that the Reidemeister torsion of these spaces vanish. As the Whitehead groups of cyclic groups are torsion free [BMS], the Whitehead torsion is detected by Reidemeister torsion and we get its vanishing as well. Thus we can consider G/(P ×Q ) ∈ S(G/(P×Q)), where S denotes the structure set of surgery theory. We shall analyze this in terms of the “old-fashioned surgery exact sequence” of Sullivan [SU]. The ﬁrst obstruction we have is the normal invariant of this homotopy equivalence ⊂∈ [G/(P×Q) : F/Cat]. However, by passing to P and Q covers, this map clearly vanishes (on those covers, it’s homotopic to a diffeomorphism), so the map is nullhomotopic. As a result G/(P×Q) and G/(P ×Q ) are normally cobordant. To complete the proof, we have to analyze a ﬁnal surgery obstruction. If dim(G) is even, the relevant surgery group vanishes.22 However, if dim(G) is odd, then there are additional obstructions, the most prominent being the ρ–invariant. This is the G-signature of any free G-manifold bounding our manifold. If S is an S1 in T-disjoint from P×Q and P ×Q , we can use it compute the ρ–invariants of G/(P×Q) and G/(P ×Q ) simultaneously. Note that G bounds the mapping cylinder of G → G/S, and that both ﬁnite groups act homologically trivially on this manifold with boundary. Thus for both, the ρ–invariant is the signature of this manifold 20. All homotopy equivalences are assumed to preserve identiﬁcation of fundamental groups. 21. A space is simple if its fundamental group is abelian and it acts trivially on higher homotopy. Localization theory works well for nilpotent, and hence, simple spaces (see [HMR]). 22. If P and Q are odd. Otherwise there can be a Z2 that is a codimension 1 arf invariant. This element always acts trivially on topological structure sets as is well known (but can be nontrivial smoothly). In many cases this obstruction can be seen to be trivial even smoothly, e.g., if there is a ﬁxed circle for all of the elements of the Weyl group used at P and Q, using a variant of the trick used above and the “numerical Levine formula” of [CW, Dv, Pa].

278 / s. weinberger

with boundary (almost always 0) × the trivial representation. In all cases, these manifolds cannot be distinguished by their ρ–invariants, completing the argument.

Remark: It is clear that one can apply the same method to some positive dimensional topologically cyclic subgroups of the maximal torus—although one will be compelled to only mix using subgroups of the Weyl group. This, though, would give an amusing example of the use of “soft” topological methods from homotopy theory together with surgery giving rise to topological conjugacies of dynamical systems that do have some recurrence. In order to make further progress (i.e., to show that the quotient of G under the relation of topological conjugacy has trivial hausdorﬁﬁcation) we have to study what happens for individual primes. We shall facilitate matters greatly by assuming that p is a large prime compared to the dimension. Moreover, the calculations are suggestive of what occurs for the case of a single circle. Consider now Zp ⊂ T as a diagonal matrix in SU(n), denoted by (a1 , a2 , . . . , an ) where the a’s are integers mod p. We have ai = 0 since our torus is in SU(n). The Chern classes are, of course, the symmetric functions of the ai ’s. Proposition 13. For G = SU(n), two G/Zp ’s have the same homotopy type (for p large w.r.t. n) iff their ﬁrst nonzero Chern classes agree. Moreover, they then have the same homeomoprhism (=diffeomorphism) type. There is a ﬁbration G/Zp → BZp → BSU(n). The map BZp → BSU(n) factors through BZp → BT, where T is the maximal torus and it is easy enough to understand. BSU(n) through dimension = dim(G) at a large prime can be thought of as a product of Eilenberg-MacLane spaces determined by the Chern classes. So BSU(n) → K(Z, 2i) is an isomorphism. The Chern classes thus determine the Postnikov decomposition of G/Zp . The r-th k-invariant is obtained by pulling back the r + 1st Chern class from BZp . However, once a Chern class is nonzero, the cohomology from BZp pulls back trivially in higher dimensions, and G/Zp looks like that Postnikov piece (at p) × with a product of the remaining odd-dimensional spheres. This proves the proposition by the classiﬁcation of spaces via Postnikov towers (and the fact that two ﬁnitedimensional spaces are homotopy equvalent iff their Postnikov towers agree through their dimensions).

some remarks inspired by the c0 zimmer program / 279

A similar method computes their Pontrjagin classes in terms of the Chern class of the representation ρ pulled back to BZp . As a result, only one nonzero Pontrjagin class enters. If our prime is sufﬁciently large, F/Cat can be caught in terms of Pontrjagin classes and we’ve computed the normal invariant. The rest of the surgery exact sequence is computed as before. Also, if p is sufﬁciently large, Top/O has no homotopy groups in the relevant range, so all of these manifolds are diffeomophic because they are homeomorphic and their universal covers are diffeomorphic. Corollary. By choosing p large it is possible to get topological conjugacy classes of elements as dense in G as we would like. So the largest Hausdorff quotient of G/∼is a point. Now let us consider the same analysis applied to BS1 in place of BZp . Again we have ai = 0, but now as an equation in Z. Now consider the Second Chern class = 12 (02 − a2i ). So there are only ﬁnitely many circles in the maximal torus that give rise to bundles with the same Second Chern class. I have not checked the following but it seems reasonable in light of the previous discussion to believe the following: Problem: Is it true that if G is irreducible, then the map from G-conjugacy classes to topological conjugacy of the translations is ﬁnite to 1? The ﬁrst case where one sees inﬁnite indeterminacy is SU(2) × SU(2). Here the maximal torus is of dimension 2. The quotient by any circle is topologically S2 × S3 . This follows very easily from old work of Smale [Sm]. Smale showed that any simply connected spin 5-manifold with H2 E Z is diffeomorphic to S2 × S3 . That the quotient has these properties is easy enough to see from the Gysin sequence associated to the cover G/S → G/T. (S is the circle, and T is the maximal 2-torus). The classifying map for the circle bundle G → G/S is a generator of H2 , and these are equivalent under diffeomorphism. So all of these circle actions are the same, and thus we have an analysis of the topological conjugacy in this way. Remark: As Lashof and Zimmer [LZ] have observed, whenever the compact group G has lattice subgroup, examples where G/A E G/B for nonconjugate subgroups A and B give rise to different ergodic lattice actions on the same

280 / s. weinberger

manifold. Thus, the observations and methods of this section shed a bit of light on these phenomena.

References [ABJLMS] [BL] [BLW] [BMS]

[Be] [BF] [BKK] [BM] [BW1]

[BW2] [Bo1] [Bo2]

[BoS]

[BFMW1] [BFMW2] [CW] [ChF] [ChW]

G. Arzhantseva, M. Bridson, T. Janszkiewicz, I. Leary, A. Minyasin, J. Swiatkowski, Inﬁnite groups with ﬁxed point properties (preprint). A. Barthels and W. Lueck, The Borel conjecture for Cat(0) and hyperbolic groups (preprint). A. Barthels, W. Lueck, and S. Weinberger, Aspherical manifolds with hyperbolic fundamental groups (to appear). H. Bass, J. Milnor, and J.-P. Serre, Solution of the congruence subgroup problem for SLn , n ≥ 3 and Sp2n , n ≥ 2. Inst. Hautes Études Sci. Publ. Math. no. 33 (1967), 59–137. J. Benveniste, Rigidity of isometric lattice actions on compact Riemannian manifolds. Geom. Funct. Anal. 10 (2000), no. 3, 516–542. M. Bestvina and M. Feign, Proper actions of lattices on contractible manifolds. Invent. Math. 150 (2002), no. 2, 237–256. M. Bestvina, M. Kapovich, and B. Kleiner, Van Kampen’s embedding obstruction for discrete groups. Invent. Math. 150 (2002), no. 2, 219–235. M. Bestvina and G. Mess, The boundary of negatively curved groups. J. Amer. Math. Soc. 4 (1991), no. 3, 469–481. J. Block and S. Weinberger, Large scale homology theories and geometry. Geometric topology (Athens, GA, 1993), 522–569, AMS/IP Stud. Adv. Math., 2.1, American Mathematical Society, Providence, RI, 1997. J. Block and S. Weinberger, On the generalized Nielsen realization problem. Comm. Math. Helv. 83 (2008), 21–33. A. Borel, Seminar on transformation groups. Princeton University Press, Princeton, NJ, 1960. A. Borel, Introduction to the cohomology of arithmetic groups. Lie groups and automorphic forms, 51–86, AMS/IP Stud. Adv. Math., 37, American Mathematical Society, Providence, RI, 2006. A. Borel and J.-P. Serre. Corners and arithmetic groups. Avec un appendice: Arrondissement des variétés à coins, par A. Douady et L. Hérault. Comment. Math. Helv. 48 (1973), 436–491. J. Bryant, S. Ferry, W. Mio, and S. Weinberger, Topology of homology manifolds. Ann. of Math. (2) 143 (1996), no. 3, 435–467. J. Bryant, S. Ferry, W. Mio, and S. Weinberger, Desingularizing homology manifolds. Geom. Topol. 11 (2007), 1289–1314. S. Cappell and S. Weinberger, Which H-spaces are manifolds? I. Topology 27 (1988), no. 4, 377–386. T. Chapman and S. Ferry, Approximating homotopy equivalences by homeomorphisms. Amer. J. Math. 101 (1979), no. 3, 583–607. S. Chang and S. Weinberger, Topological nonrigidity of nonuniform lattices. Comm. Pure Appl. Math. 60 (2007), no. 2, 282–290.

some remarks inspired by the c0 zimmer program / 281 [CK]

[Co] [Da] [Dv] [EL] [FS1] [FS2] [FW] [FJ] [Fe] [FRR]

[FiM] [FiS] [Gi] [GKW]

[HMR]

[H] [KL] [LZ] [Ma] [Mi]

F. Connolly and T. Kosniewski, Examples of lack of rigidity in crystallographic groups. Algebraic topology Poznan 1989, 139–145, Lecture Notes in Math., 1474, Springer, Berlin, 1991. G. Cooke, Replacing homotopy actions by topological actions. Trans. Amer. Math. Soc. 237 (1978), 391–406. R. Daverman, Decompositions of manifolds, Reprint of the 1986 original. AMS Chelsea Publishing, Providence, RI, 2007. J. Davis, Evaluation of odd-dimensional surgery obstructions with ﬁnite fundamental group. Topology 27 (1988), no. 2, 179–204. J. Ewing and A. Liulevicius, Homotopy rigidity of linear actions on homogeneous spaces. J. Pure Appl. Algebra 18 (1980), no. 3, 259–267. B. Farb and P. Shalen, Lattice actions, 3-manifolds and homology. Topology 39 (2000), no. 3, 573–587. B. Farb and P. Shalen, Real-analytic actions of lattices. Invent. Math. 135 (1999), no. 2, 273–296. B. Farb and S. Weinberger, Isometries, rigidity, and universal covers, Ann. of Math. 168 (2008), 915–940. T. Farrell and L. Jones, Rigidity for aspherical manifolds with π1 ⊂ GLm (R). Asian J. Math. 2 (1998), no. 2, 215–262. S. Ferry, Homotoping ε–maps to homeomorphisms, Amer. J. Math. 101 (1979), no. 3, 567–582. S. Ferry, A. Ranicki, and J. Rosenberg, A history and survey of the Novikov conjecture. Novikov conjectures, index theorems and rigidity, Vol. 1 (Oberwolfach, 1993), 7–66, London Math. Soc. Lecture Note Ser., 226, Cambridge University Press, Cambridge, 1995. D. Fisher and G. Margulis, Almost isometric actions, property (T), and local rigidity. Invent. Math. 162 (2005), no. 1, 19–80. D. Fisher and L. Silberman, Groups not acting on manifolds, IMRN 2008 (to appear). C. Giffen, The generalized Smith conjecture. Amer. J. Math. 88 (1966), 187–198. T. Goodwillie, J. Klein, and M. Weiss, Spaces of smooth embeddings, disjunction and surgery. Surveys on surgery theory, Vol. 2, 221–284, Ann. of Math. Stud., 149, Princeton University Press, Princeton, NJ, 2001. P. Hilton, G. Mislin, and J. Roitberg, Localization of nilpotent groups and spaces. North-Holland Mathematics Studies, No. 15. Notas de Matemática, No. 55. North-Holland Publishing Co., Amsterdam-Oxford; American Elsevier Publishing Co., New York, 1975. S. Hurder, Rigidity for Anosov actions of higher rank lattices. Ann. of Math. (2) 135 (1992), no. 2, 361–410. A. Katok and J. Lewis, Global rigidity results for lattice actions on tori and new examples of volume-preserving actions. Israel J. Math. 93 (1996), 253–280. R. Lashof and R. Zimmer, Manifolds with inﬁnitely many actions of an arithmetic group. Illinois J. Math. 34 (1990), no. 4, 765–768. B. Mazur, Differential topology from the point of view of simple homotopy theory. Inst. Hautes Études Sci. Publ. Math. no. 15 (1963), 93 pp. J. Milnor, Whitehead torsion, Bull. Amer. Math. Soc. 72 (1966), 358–426.

282 / s. weinberger [Pa]

[Q] [RS] [Shi] [Si] [Sm] [SW] [Su] [We1] [We2]

[We3] [We4]

[Wi] [Zi1] [Zi2]

W. Pardon, The exact sequence of a localization for Witt groups. II. Numerical invariants of odd-dimensional surgery obstructions. Paciﬁc J. Math. 102 (1982), no. 1, 123–170. F. Quinn, An obstruction to the resolution of homology manifolds. Michigan Math. J. 34 (1987), no. 2, 285–291. F. Raymond and L. Scott, Failure of Nielsen’s theorem in higher dimensions. Arch. Math. (Basel) 29 (1977), no. 6, 643–654. N. Shirokova, thesis (University of Chicago, 1998). L. Siebenmann, Approximating cellular maps by homeomorphisms. Topology 11 (1972), 271–294. S. Smale, On the structure of 5-manifolds, Ann. of Math. (2) 75 (1962), 38–46. M. Steinberger and J. West, Approximation by equivariant homeomorphisms. Trans. Amer. Math. Soc. 302 (1987), no. 1, 297–317. D. Sullivan, Inﬁnitesimal computations in topology. Publ. Math. d’IHES 47 (1977), 269–331. S. Weinberger, Computers, rigidity, and moduli: The large scale fractal geometry of Riemannian moduli space. Princeton University Press, Princeton, NJ, 2004. S. Weinberger, Continuous versus discrete symmetry, Geometry and topology (Athens, GA, 1985), 319–323, Lecture Notes in Pure and Appl. Math., 105. Dekker, New York, 1987. S. Weinberger, The topological classiﬁcation of stratiﬁed spaces. Chicago Lectures in Mathematics. University of Chicago Press, Chicago, 1994. S. Weinberger, Nonlinear averaging, embeddings, and group actions. Tel Aviv Topology Conference: Rothenberg Festschrift (1998), 307–314, Contemp. Math., 231. American Mathematical Society, Providence, RI, 1999. C. W. Wilkerson, Applications of minimal simplicial groups. Topology 15 (1976), 111–130. R. Zimmer, Lattices in semisimple groups and distal geometric structures. Invent. Math. 80 (1985), no. 1, 123–137. R. Zimmer, Lattices in semisimple groups and invariant geometric structures on compact manifolds. Discrete groups in geometry and analysis (New Haven, CT, 1984), 152–210, Progr. Math., 67. Birkhäuser Boston, Boston, 1987.

PART 2

Analytic, Ergodic, and Measurable Group Theory

9

michael g. cowling

CALCULUS ON NILPOTENT LIE GROUPS

to my friend bob Abstract Mostow’s rigidity theorem states that if two locally symmetric spaces \G/K (where G/K is a symmetric space of the noncompact type, and is a uniform lattice in the semisimple group G) are homeomorphic, then they are diffeomorphic; the case where G = SL(2, R) has to be excluded. Mostow’s proof treats the rank 1 case and the higher-rank case differently: the former analyzes quasiconformal mappings, while the latter appeals to a theorem of Tits on mappings of spherical buildings. This note discusses a more analytical approach to the higher-rank case, which aims to unify the rank 1 and higher-rank cases, and links to ideas of Gromov and Pansu. Only the case of SL(n, R) is discussed here.

1. Introduction A key step in the proof of Mostow’s rigidity theorem [6] is to identify maps of the boundary space G/P that arise as -equivariant boundary maps of quasiisometries, and to show that all these maps are of the form xP → gxP, for some g in G (here P is a minimal parabolic subgroup of G). In the rank 1 case, the boundary maps to be considered are all “contact” mappings, and in the cases of the groups SO(1, n) and SU(1, n), the ergodicity of is also needed, but for the groups Sp(1, n) and the exceptional group F4,20 , P. Pansu [8] showed that the contact condition alone implies that the maps all arise from the action of G. Later, the differential geometer K. Yamaguchi [10] showed that, with a few exceptions, sufﬁciently smooth “contact” mappings of the boundary G/P all come from the action of the group G. However, the smoothness results needed to apply the machinery of differential geometry are not natural in the context of Mostow’s theorem. We show how the weaker condition of “Pansu differentiability” (deﬁned later), similar to the condition considered by Mostow in the rank 1 case, implies that contact mappings come from the action of G It is a pleasure to thank the anonymous referee of this chapter for his or her useful and constructive criticism.

285

286 / m. g. cowling

(with a few exceptions); for simplicity, here we consider only the case where G is SL(n, R) and n is at least 4. We intend to use this theory elsewhere to give another proof of Mostow’s theorem. If a mapping of a stratiﬁed nilpotent Lie group N is “Pansu differentiable,” then its derivative is a strata-preserving homomorphism of the Lie group or of the Lie algebra (depending on the deﬁnition used); if the mapping is invertible, then the mapping is an automorphism. We characterize the strata-preserving automorphisms of the Lie algebra utn(n, R) of the group of upper-triangular unipotent matrices UTU(n, R), which is the Iwasawa N group for the group SL(n, R). We ﬁrst of all consider the case where n = 4, and then treat the case of larger n by induction. Using this characterization, we show that mappings that are differentiable everywhere, or almost everywhere and satisfying a simple ergodicity condition, arise out of the group action. To state our theorems, we need to deﬁne the “reﬂection in the ant-diagonal” φ : SL(n, R) → SL(n, R), by the formula φ(g)ij = gn+1−j,n+1−i

∀i, j ∈ {1, . . . , n}.

It is clear that this is an automorphism of SL(n, R) that preserves UTU(n, R). Indeed, it is the automorphism that arises from the obvious nontrivial automorphism of the Dynkin diagram of the root system An−1 . Other automorphisms of SL(n, R) that preserve UTU(n, R) include conjugations by diagonal matrices. We write the group of these automorphisms as Ad (MA). We will prove the following results.

theorem 1. Suppose that N is the group of n × n unipotent upper-triangular real matrices, where n ≥ 4, and that f : N → N is Pansu differentiable at p, with a Pansu differentiable local inverse at f ( p). Then the Pansu derivative df ( p) lies in Ad (MA) or φ Ad (MA). theorem 2. Suppose that N is the group of n × n unipotent upper-triangular real matrices, where n ≥ 3, that the continuous mapping f : N → N is Pansu differentiable almost everywhere, and that df is in Ad (MA) almost everywhere. Then f is an afﬁne map, that is, the composition of a group automorphism (that commutes with dilations) and a translation. Theorem 2 has a number of interesting features. First, only pointwise differentiability is assumed. Most theorems in analysis require more than this. For instance, the Fundamental Theorem of Calculus, that is, for a function u : [a, b] → R,

calculus on nilpotent lie groups / 287

u(b) − u(a) =

b

u (x) dx,

a

requires u to be integrable. The Cauchy-Goursat theorem in complex analysis is an exception. There are interesting parallels between the Cauchy-Goursat theorem and Theorem 2: in both cases differentiability and algebraic structure of the derivative leads to strong conclusions. Next, local hypotheses give global conclusions. In complex analysis, the exponential map is differentiable and has a differentiable local inverse, but it is not one to one. Finally, this theorem also has a local version, when f is deﬁned in an open set. In this case, f comes from the action of GL(n, R) (augmented by the map φ).

2. The Pansu Derivative A Lie algebra n is said to be stratiﬁed of step s if there are nontrivial subspaces n1 , n2 , …, ns such that 1

n = n1 ⊕ n2 ⊕ · · · ⊕ ns

[n1 , nj ] = nj+1 ,

where ns+1 := {0}. A stratiﬁed algebra is graded, that is, [nk , nj ] ⊆ nk+j , and nilpotent, because [n, [n, . . . [n, n] . . . ]] = {0}, where there are s + 1 Lie brackets. This follows by induction. Stratiﬁed Lie algebras admit dilations δt , for t in R+ , deﬁned by requiring that δt X := t j X for all X in nj and extending δt to n by linearity. Clearly δt [X , Y ] = [δt X , δt Y ]

∀X ∈ nj ,

∀Y ∈ nk

because n is graded; hence δt is a Lie algebra automorphism. An example of a stratiﬁed Lie algebra is the Lie algebra utn(n, R) of n × n upper-triangular unipotent matrices. Denote by Ei,j the n × n matrix whose entries are all 0 except for the i, jth entry, which is 1. Deﬁne nk to be span{Ei,i+k : 1 ≤ i ≤ n − k} (in other words, the kth layer of n is the sum of all the root spaces where the root is the sum of exactly k simple roots). It is easy to check that the stratiﬁcation conditions 1 hold, and that utn(n, R) is of step n − 1. A stratiﬁed Lie group is a connected, simply connected, necessarily nilpotent Lie group N whose Lie algebra n is stratiﬁed. These occur in areas of mathematics including sub-Riemannian geometry, subelliptic differential operators, and nonholonomic mechanics.

288 / m. g. cowling

The exponential map exp is a bijection from a stratiﬁed Lie algebra n to the corresponding stratiﬁed Lie group N, and the dilations δt may be extended to N by conjugating by the exponential map: δt p = exp(δt log ( p))

∀p ∈ N.

A map f : Rm → Rn is (totally) differentiable at a point p in Rm if and only if lim

t→0+

f ( p + th) − f ( p) t

exists, uniformly on compacta in h; if the limit exists, then we write it as Df ( p)(h). The map Df ( p) : Rm → Rn is linear. It “is” the matrix of partial derivatives of the components of f . Suppose that N 1 and N 2 are stratiﬁed Lie groups. A map f : N 1 → N 2 is Pansu differentiable at a point p in N 1 if and only if lim δt−1 [ f ( p)−1 f ( pδt h)]

t→0+

exists, uniformly on compacta in h; if the limit exists, then we write it as Df ( p)(h). The map Df ( p) : N 1 → N 2 is a group homomorphism that commutes with dilations. It is often more convenient to work with a linearized version of the derivative; we deﬁne df ( p)X := log(DF( p) exp (X ))

∀X ∈ n1 .

The map df ( p) is a Lie algebra homomorphism from n1 to n2 that commutes with dilations, or equivalently, respects the stratiﬁcations of the algebras.

3. Proof of Theorem 1 In this section, we prove Theorem 1, that is, we characterize the stratapreserving automorphisms of utnn (R). Write N for the group UTUn (R) of all upper-triangular unipotent n × n real matrices, whose Lie algebra n is the set utnn (R) of all upper-triangular nilpotent n × n real matrices. 3.1. The Case When n = 3

It is easy to check that there are automorphisms of the algebra utn(3, R) that preserve both strata, but act as arbitrary elements of GL(2, R) on the ﬁrst stratum. Thus Pansu differentiability imposes no restrictions on the derivative, other than the strata-preserving condition and an evident determinant condition, and Theorem 1 fails.

calculus on nilpotent lie groups / 289

A. Korànyi and H.M. Reimann [3, 4] developed a theory of quasiconformal mappings on the Heisenberg group UTU3 (R); these quasiconformal mappings are Pansu differentiable, and the space thereof is inﬁnite-dimensional. Thus Pansu differentiability in the case of UTU3 (R) is not a very restrictive condition. However, Theorem 2 does hold in this case, as we shall see; indeed, the restriction that the Pansu derivative lie in Ad (MA) almost everywhere is very strong, not only for groups of upper-triangular matrices, but also for more general Iwasawa N groups. For instance, for SO(n, 1), where n ≥ 3, the hypotheses of Theorem 2 lead us to consider 1-quasiregular mappings, which form quite a limited class! 3.2. The Case When n = 4

When we take n to be 4, elements of n and of N are of the form ⎛ ⎞ ⎛ ⎞ 0 u x z 1 u x z ⎜0 0 v y ⎟ ⎜0 1 v y ⎟ ⎜ ⎟ ⎜ ⎟ 2 and ⎜ ⎟ ⎜ ⎟, ⎝0 0 0 w ⎠ ⎝0 0 1 w ⎠ 0 0 0 0 0 0 0 1 respectively, where u, v, . . . , z are real numbers. For this example, the dilations are given by ⎞ ⎛ ⎞ ⎛ 1 u x z 1 tu t 2 x t 3 z ⎜0 1 v y ⎟ ⎜0 1 tv t 2 y ⎟ ⎟ ⎜ ⎟ ⎜ δt ⎜ ⎟. ⎟=⎜ ⎝0 0 1 w ⎠ ⎝0 0 1 tw ⎠ 0 0 0 1 0 0 0 1 Let {U, V , W , X , Y , Z} be the basis of utn4 (R) such that the left-hand matrix above is equal to uU + vV + wW + xX + yY + zZ.

3

In formula 2, the group element shown is not the exponential of the algebra element shown. It is easy to check that [U, V ] = X ,

[V , W ] = Y ,

[U, Y ] = Z,

and

[X , W ] = Z,

and all other (undetermined) commutators are zero. Take a map f of UTU4 (R), locally invertible near p. Its Pansu derivative is a strata-respecting automorphism of utn4 (R). We may therefore represent df ( p) in the basis {U, V , W , X , Y , Z} introduced in formula 3 by a 6 × 6 matrix:

290 / m. g. cowling

⎛

a ⎜ ⎜d ⎜ ⎜g ⎜ ⎜0 ⎜ ⎜ ⎝0 0

b e h 0 0 0

c f i 0 0 0

0 0 0 j l 0

0 0 0 k m 0

⎞ 0 ⎟ 0⎟ ⎟ 0⎟ ⎟. 0⎟ ⎟ ⎟ 0⎠ n

It is easy to check that span{U, W } is the unique 2-dimensional abelian subalgebra of n1 , and that span{V } is the centralizer of n2 in n1 . Since df ( p) preserves these subalgebras, we may represent df ( p) by ⎛ ⎞ a 0 c 0 0 0 ⎜ ⎟ ⎜0 e 0 0 0 0 ⎟ ⎜ ⎟ ⎜g 0 i 0 0 0⎟ ⎜ ⎟ ⎜0 0 0 j k 0 ⎟ . ⎜ ⎟ ⎜ ⎟ ⎝0 0 0 l m 0 ⎠ 0 0 0 0 0 n The commutation relation [U, V ] = X implies that [TU, TV ] = TX , whence (using other relations as needed) aeX − egY = [aU + gW , eV ] = jX + lY , and hence j = ae and l = −ge. The relation [V , W ] = Y implies that k = −ce and m = ie analogously. Similarly, the relations [U, X ] = 0, [U, Y ] = Z, [W , Y ] = 0, and [W , X ] = Z yield the equations 0 = al − gj

n = am − gk

0 = ik − cm

n = ij − cl.

All these equations together also imply that aeg = 0

cei = 0

n = aei + cge.

It follows that we may represent df ( p) by a matrix, as shown below. All the entries indicated by letters have to be nonzero for the matrix to be invertible; they may depend on the point p. ⎞ ⎛ ⎞ ⎛ a 0 0 0 0 0 0 0 c 0 0 0 ⎟ ⎜ ⎟ ⎜ 0 0 0⎟ ⎜0 e 0 0 0 0 ⎟ ⎜0 e 0 ⎟ ⎜ ⎟ ⎜ ⎜0 0 i 0 0 0 ⎟ ⎜ 0 0 0⎟ ⎟. ⎜ ⎟ or ⎜g 0 0 ⎜0 0 0 ae 0 0 ⎟ ⎜0 0 0 0 −ce 0 ⎟ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ 0 0⎠ ⎝0 0 0 0 ei 0 ⎠ ⎝0 0 0 −eg 0 0 0 0 cge 0 0 0 0 0 aei 0

calculus on nilpotent lie groups / 291

3.3. The Case When n > 4

Suppose that T is a strata-preserving automorphism of n. It is easy to verify that X in n1 centralizes nn−2 if and only if X is in s, the span of the Ei,i+1 where 2 ≤ i ≤ n − 2. Consequently, T maps s to s. Suppose that TE1,2 = aE1,2 + S + bEn−1,n ,

4

where a and b are real numbers and S ∈ s. We write alg (X , Y , . . . , ) for the Lie algebra generated by X , Y , …. If any argument of alg is a set, then of course we mean the Lie algebra generated by the other arguments and the elements of this set. It is straightforward to check that dim alg (E1,2 , s) = (n − 1)(n − 2)/2, while dim alg (aE1,2 + S + bEn−1,n , s) is equal to ⎧ ⎪ if neither of a or b is 0 ⎪ ⎨n(n − 1)/2 − 1 ⎪ ⎪ ⎩

(n − 1)(n − 2)/2

if exactly one of a or b is 0

(n − 2)(n − 3)/2

if both of a or b is 0.

A strata-preserving automorphism preserves the dimensions of algebras generated by sets, and so exactly one of a and b is nonzero. By composing with φ, if necessary we may suppose that b is zero and a is not. Now T preserves RE1,2 + s, and hence preserves the algebra that this generates, which is isomorphic to utn(n − 1, R). By induction, we may suppose that this restricted version of T lies in Ad (MA), and in particular that S = 0 in formula 4. A similar argument then shows that T also maps En−1,n to a multiple of itself, and from this it follows that T lies in Ad (MA), as enunciated.

4. The Proof of Theorem 2 Theorem 2 includes the hypothesis that df ( p) is in Ad (MA) almost everywhere. This hypothesis is necessary, because we cannot say much about a map of N if its derivative changes from being in Ad (MA) to being in φ Ad (MA) in some possibly uncontrolled way. Suppose that f is Pansu differentiable everywhere. A theorem of Darboux says that the derivative of a function f that is differentiable in an interval has the intermediate value property: if f (x) = a and f (y) = b, and if c lies between a and b, then there exists z between x and y such that f (z) = c. Alternatively, we may say that f maps connected sets to connected sets. In this formulation, Darboux’s theorem has been generalized to Rn and beyond—see, for instance, J. Malý [5]. In our case, the set of possible derivatives df ( p) is the union of two disjoint open sets Ad (MA) and φ Ad (MA). This means that df ( p) is always of

292 / m. g. cowling

one type or always of the other. By composing with the group automorphism φ if necessary, we may suppose that df ( p) is always diagonal. Alternatively, if we knew that f was differentiable almost everywhere and that a group acted ergodically, preserving the set of points p where df ( p) ∈ Ad (MA) and the set of points p where df ( p) ∈ φMA, then we could deduce that df was of one type almost everywhere. In any case, we have stated Theorem 2 with the hypotheses that f is continuous and differentiable almost everywhere, and that df ( p) is in Ad (MA). For any small neighborhood of a point p in N, the set exp (REi,i+1 ) is sent into a set ϒ exp (REi,i+1 ), where ϒ is a neighborhood of f ( p); by shrinking , we may deduce that cosets of exp (REi,i+1 ) are mapped into cosets of exp (REi,i+1 ). Such a map is afﬁne, that is, it is composed of a (left) translation and a group automorphism. This is proved in the forthcoming PhD thesis of Rupert McCallum; a special case is considered in the article [1]. We give an analytic proof of the result here that conveys less geometric information than McCallum’s proof. It should be pointed out that McCallum also considers maps deﬁned on connected open subsets of a stratiﬁed group N, and the proof given here does not (obviously) extend to this case. We ﬁrst recall that it is shown in [1] that a map f of the group UTU3 (R) that sends cosets of exp (RE1,2 ) to cosets of exp (RE1,2 ) and cosets of exp (RE2,3 ) to cosets of exp (RE2,3 ) is afﬁne. Now we consider the case where n = 4. For notational convenience, we again use the basis {U, V , W , X , Y , Z} of utn4 (R) such that ⎛ ⎞ 0 u x z ⎜0 0 v y ⎟ ⎜ ⎟ ⎜ ⎟ = uU + vV + wW + xX + yY + zZ. ⎝0 0 0 w ⎠ 0 0 0 0 We consider four copies of UTU3 (R) in UTU4 (R); their Lie algebras are spanned by {U, V , X }, by {V , W , Y }, by {X , W , Z} and {U, Y , Z}. We know already that f preserves cosets of exp (RU), exp (RV ), and exp (RW ), and show successively that f preserves cosets of exp (RX ), exp (RY ), and exp (RZ). In addition, we show that various partial derivatives are constant on larger and larger sets, in order to conclude that the derivative of f itself is constant. First, f sends cosets of exp (RU) into cosets of exp (RU), and cosets of exp (RV ) into cosets of exp (RV ), and exp (RU) and exp (RV ) generate exp (RU + RV + RX ), which is isomorphic to UTU(3, R), so f sends the coset p exp (RU + RV + RX ) into the coset f ( p) exp (RU + RV + RX ). By translating by p and f ( p), we may temporarily suppose that p and f ( p) are both the

calculus on nilpotent lie groups / 293

identity. By the result for the special case of SL(3, R) in [1], we deduce that f is afﬁne on exp (RU + RV + RX ); by translating back, we deduce that f is afﬁne on p exp (RU + RV + RX ); in particular, the derivative of f , restricted to this coset, is constant, and f maps cosets of exp (RX ) into cosets of exp (RX ). Next, we repeat this argument starting with cosets of exp (RV ) and of exp (RW ), to show that f is afﬁne on p exp (RV + RW + RY ); in particular, the derivative of f , restricted to this coset, is constant, and f maps cosets of exp (RY ) into cosets of exp (RY ). Third, we repeat the argument starting with cosets of exp (RX ) and of exp (RW ), to show that f is afﬁne on p exp (RX + RW + RZ); in particular, the derivative of f , restricted to this coset, is constant. Finally, we repeat the argument, starting with cosets of exp (RU) and of exp (RY ), to show that f is afﬁne on p exp (RU + RY + RZ); in particular, the derivative of f , restricted to this coset, is constant. Recall that the derivative df ( p) of f at p may be represented as the matrix ⎛ ⎞ a 0 0 0 0 0 ⎜ ⎟ ⎜0 e 0 0 0 0 ⎟ ⎜ ⎟ ⎜0 0 i 0 0 0 ⎟ ⎜ ⎟ ⎜0 0 0 ae 0 0 ⎟ , ⎜ ⎟ ⎜ ⎟ ⎝0 0 0 0 ei 0 ⎠ 0 0 0 0 0 aei where the entries a, e, . . ., are functions of p. Our four arguments above have shown that these entries are constant along certain cosets of certain subgroups, and more precisely they show that 5

˜ = V˜ a = Ue ˜ = V˜ e = 0, Ua

6

˜ e = V˜ i = W ˜ i = 0, V˜ e = W

7

˜ (ae) = X˜ i = W ˜ i = 0, and X˜ (ae) = W

8

˜ = Y˜ (ei) = Ua ˜ = Y˜ (ai) = 0, Ua

where T˜ stands for the left-invariant vector ﬁeld associated to the lie algebra element T (T being any of U, V , and so on). ˜ = V˜ e = W ˜ e = 0. This implies that From Equations 5 and 6, we see that Ue ˜ a+ e is constant on the whole group. Further, from Equation 7, we see that eW ˜ ˜ aW e = 0, whence from Equation 6, W a = 0, and combined with Equation ˜ = V˜ a = W ˜ a = 0, whence a is constant on the whole 5, this implies that Ua group. Similarly, i is constant on the whole group, and this implies that f is afﬁne.

294 / m. g. cowling

The case when n ≥ 5 is proved similarly. There are many more copies of UTU3 inside UTUn , but we can still apply the result of [1] to each of these, and build up to the desired conclusion in much the same way as in the case when n = 4.

5. Rigidity for Stratified Groups One of the key ideas of Pansu [8] is that most stratiﬁed groups are rigid, in the sense that the dimension of the space of differentiable maps thereon is ﬁnite (even when we consider maps deﬁned on an open subset). As mentioned above, most Iwasawa N groups are rigid in this sense. The exceptions are the N groups coming from SO(n, 1), SU(n, 1), SL(3, F), and Sp(2, F), where F is either R or C. At the present time, the published proofs of this fact known to this author involve prolongation and use of the Borel-Weil theorem in one case [10] and a careful study of contact ﬂows in the other [2], and require more smoothness than just differentiability. In a paper in preparation, P. Ciatti and the author of this paper classify the strata-preserving automorphisms of all the Iwasawa N groups, by simpler, hands-on methods. The groups listed above are precisely those where the automorphism group is not a ﬁnite extension of Ad (MA) (where KAN and MAN are an Iwasawa decomposition of the associated semisimple group G and a minimal parabolic subgroup of G). If we include the stronger hypothesis that the automorphisms preserve the root spaces, then the two rank 1 families are the only exceptions. I believe that this result, combined with the methods of this chapter, will lead to another proof of rigidity, which treats the rank 1 and higher-rank cases similarly. There is interesting related work going on at the moment on differentiable maps of more general stratiﬁed groups. For instance, it has been shown recently that differentiable maps of free nilpotent Lie groups that are of step at least 4, or of step 3 with at least three generators, are all afﬁne [9]. General criteria for rigidity are also appearing [7].

References [1]

ˇ A. Cap, M.G. Cowling, F. De Mari, M.G. Eastwood, and R. McCallum, The Heisenberg group, SL(3, R) and rigidity, pp. 41–52 in: Harmonic analysis, group representations, automorphic forms and invariant theory in honor of Roger E. Howe, Lecture Notes Series, Institute for Mathematical Sciences, National University of Singapore, Vol. 12. World Scientiﬁc, Singapore, 2007.

calculus on nilpotent lie groups / 295 [2]

M.G. Cowling, F. De Mari, A. Koráányi, and H.M. Reimann, Contact and conformal maps in parabolic geometry. I, Geom. Dedic. 111 (2005), 65–86. [3] A. Korányi and H.M. Reimann, Quasiconformal mappings on the Heisenberg group, Invent. Math. 80 (1985), 309–338. [4] A. Korányi and H.M. Reimann, Foundations of the theory of quasiconformal mappings on the Heisenberg group, Adv. Math. 111 (1995), 1–87. [5] J. Malý, “The Darboux property for gradients”, Real Anal. Exchange 22 (1996–1997), 167–173. [6] G.D. Mostow, Strong rigidity of locally symmetric spaces, Annals of Math. Studies, Vol. 78. Princeton University Press, Princeton, NJ 1973. [7] A. Ottazzi, A sufﬁcient condition for nonrigidity of Carnot groups, Math. Zeits. 259 (2008), 617–629. [8] P. Pansu, Métriques de Carnot-Caratheodory et quasiisométries des espaces symétriques de rang un, Annals of Math. 129 (1989), 1–60. [9] B. Warhurst, Tanaka prolungation of free Lie algebras, Geom. Dedic. 130 (2007), 59–69. [10] K. Yamaguchi, Differential systems associated with simple graded Lie algebras, Adv. Studies in Pure Math. 22 (1993), 413–494.

10

alex furman

A SURVEY OF MEASURED GROUP THEORY

to robert zimmer on the occasion of his

60 th

birthday

Abstract Measured group theory is an area of research that studies inﬁnite groups using measure-theoretic tools, and studies the restrictions that group structure imposes on ergodic-theoretic properties of their actions. This paper is a survey of recent developments focused on the notion of measure equivalence between groups, and orbit equivalence between group actions.

Contents

1

Introduction Disclaimer Acknowledgments Organization of the Paper Preliminary Discussion and Remarks 2.1 Lattices and Other Countable Groups 2.2 Orbit Equivalence in Ergodic Theory 2.3 Further Comments on QI, ME, and Related Topics Measure Equivalence Between Groups 3.1 Measure Equivalence Invariants 3.2 Orbit/Measure Equivalence Rigidity 3.3 How Many Orbit Structures Does a Given Group Have? Measured Equivalence Relations 4.1 Basic Deﬁnitions 4.2 Invariants of Equivalence Relations 4.3 Rigidity of Equivalence Relations Techniques 5.1 Superrigidity in Semisimple Lie Groups 5.2 Superrigidity for Product Groups 5.3 Strong Rigidity for Cocycles

2

3

4

5

Supported by NSF Grant DMS 0604611, and BSF Grant 2004345

296

297 298 298 298 299 299 303 306 308 309 321 325 327 327 331 344 348 348 351 353

a survey of measured group theory / 297

5.4 Cocycle Superrigid Actions 5.5 Constructing Representations 5.6 Local Rigidity for Measurable Cocycles 5.7 Cohomology of cocycles 5.8 Proofs of Some Results Appendix A. Cocycles A.1 The Canonical Class of a Lattice, (Co)induction A.2 OE-Cocycles A.3 ME-Cocycles References

354 355 357 358 362 365 366 367 367 368

1. Introduction This survey concerns an area of mathematics that studies inﬁnite countable groups using measure-theoretic tools, and studies ergodic theory of group actions, emphasizing the impact of group structure on the actions. Measured group theory is a particularly ﬁtting title as it suggests an analogy with geometric group theory. The origins of measured group theory go back to the seminal paper of Robert Zimmer [139], which established a deep connection between questions on orbit equivalence in ergodic theory to Margulis’s celebrated superrigidity theorem for lattices in semisimple groups. The notion of amenable actions, introduced by Zimmer in an earlier work [138], became an indispensable tool in the ﬁeld. Zimmer continued to study orbit structures of actions of large groups in [32, 41, 140–144, 146, 147] and [135]. The monograph [146] had a particularly big impact on both ergodic theorists and people studying big groups, as well as researchers in other areas, such as operator algebras and descriptive set theory.1 In recent years several new layers of results have been added to what we called measured group theory, and this paper aims to give an overview of the current state of the subject. Such a goal is unattainable—any survey is doomed to be partial, biased, and outdated before it appears. Nevertheless, we shall try our best, hoping to encourage further interest in this topic. The reader is also referred to Gaboriau’s paper [58], which contains a very nice overview of some of the results discussed here, and to Shalom’s survey [133], which is even closer to the present paper (hence the similarity of the titles). The monographs by Kechris and Miller [81] and the forthcoming one [80] by Kechris include topics 1. Zimmer’s cocycle superrigidity proved in [139] plays a central role in another area of research, vigorously pursued by Zimmer and others, concerning actions of large groups on manifolds. David Fisher surveys this direction in [42] in this volume.

298 / a. furman

in descriptive set theory related to measured group theory. Readers interested in connections to von Neumann algebras are referred to Vaes’s [136], Popa’s [114], and references therein. The scope of this paper is restricted to interaction of inﬁnite groups with ergodic theory, leaving out the connections to the theory of von Neumann algebras and descriptive set theory. When possible, we try to indicate proofs or ideas of proofs for the stated results. In particular, we chose to include a proof of one cocycle superrigidity theorem (Theorem 5.21), which enables a self-contained presentation of a number of important results: a very rigid equivalence relation (Theorem 4.19) with trivial fundamental group and outer automorphism group (Theorem 4.15), an equivalence relation that cannot be generated by an essentially free action of any group (§4.3.1).

disclaimer. As usual, the quoted results are often presented not in the full possible generality, so the reader should consult the original papers for full details. The responsibility for inaccuracies, misquotes, and other ﬂaws lies solely with the author of these notes. acknowledgments: I would like to express my deep appreciation to Bob Zimmer for his singular contribution to the subject. I would also like to thank Miklos Abert, Aurélien Alvarez, Uri Bader, Damien Gaboriau, Alexander Kechris, Sorin Popa, Yehuda Shalom, and the referee for the corrections and comments on the earlier drafts of the paper. organization of the paper. The paper is organized as follows: the next section is devoted to a general introduction that emphasizes the relations among measure equivalence, quasi-isometry, and orbit equivalence in ergodic theory. One may choose to skip most of this, but read Deﬁnition 2.1 and the following remarks. Section 3 concerns groups considered up to measure equivalence. Section 4 focuses on the notion of equivalence relations with orbit relations as a prime (but not only) example. In both of these sections we consider separately the invariants of the studied objects (groups and relations) and rigidity results, which pertain to questions of classiﬁcation. Section 5 describes the main techniques used in these theories (mostly for rigidity): a discussion of superrigidity phenomena and some of the ad hoc tools used in the subject; generalities on cocycles appear in Appendix A.

a survey of measured group theory / 299

2. Preliminary Discussion and Remarks This section contains an introduction to measure equivalence and related topics and contains a discussion of this framework. Readers familiar with the subject (especially Deﬁnition 2.1 and the following remarks) may skip to the next section in the ﬁrst reading. There are two natural entry points to Measured group theory, corresponding to the ergodic-theoretic and group-theoretic perspectives. Let us start from the latter. 2.1. Lattices and Other Countable Groups

When should two inﬁnite discrete groups be viewed as closely related? Isomorphism of abstract groups is an obvious, maybe trivial, answer. The next degree of closeness would be commensurability: two groups are commensurable if they contain isomorphic subgroups of ﬁnite index. This relation might be relaxed a bit further, by allowing to pass to a quotient modulo ﬁnite normal subgroups. The algebraic notion of being commensurable, modulo ﬁnite kernels, can be vastly generalized in two directions: measure equivalence (measured group theory) and quasi-isometry (geometric group theory). The key notion discussed in this paper is that of measure equivalence of groups, introduced by Gromov in [66, 0.5.E].

definition 2.1. Two inﬁnite discrete countable groups , are measure ME equivalent (abbreviated as ME, and denoted ∼ ) if there exists an inﬁnite measure space (, m) with a measurable, measure-preserving action of × , so that both actions (, m) and (, m) admit ﬁnite-measure fundamental domains Y , X ⊂ : γY = λX . = γ ∈

λ∈

The space (, m) is called a (, )-coupling or ME-coupling. The index of to in is the ratio of the measures of the fundamental domains m(X ) meas(/) = . [ : ] = m(Y ) meas(/ ) We shall motivate this deﬁnition after making a few immediate comments. a) The index [ : ] is well deﬁned—it does not depend on the choice of the fundamental domains X , Y for /, / respectively, because their measures are determined by the group actions on (, m). However,

300 / a. furman

a given pair (, ) might have ME-couplings with different indices (the ∗ corresponding to possible set {[ : ] } is a coset of a subgroup of R+ indices [ : ] of self--couplings. Here it makes sense to focus on ergodic couplings only). b) Any ME-coupling can be decomposed into an integral over a probability space of ergodic ME-couplings, that is, ones for which the × -action is ergodic. c) measure equivalence is indeed an equivalence relation between groups: for any countable the action of × on = with the counting measure m by (γ1 , γ2 ) : γ → γ1 γ γ2−1 provides the trivial self-ME-coupling, giving reﬂexivity; symmetry is obvious from the deﬁnition;2 while transitivity follows from the following construction of composition, or fusion, of ME-couplings. If (, m) is a (1 , 2 ) coupling and ( , m ) is a (2 , 3 ) coupling, then the quotient = × 2 of × under the diagonal action γ2 : (ω, ω ) → (γ2 ω, γ2−1 ω ) inherits a measure m = m × 2 m so that ( , m ) becomes a (1 , 3 ) coupling structure. The indices satisfy: [1 : 3 ] = [1 : 2 ] · [2 : 3 ] . d) The notion of ME can be extended to the broader class of all unimodular locally compact second countable groups: a ME-coupling of G and H is a measure space (, m) with measure space isomorphisms i : (G, mG ) × (Y , ν) ∼ = (, m),

j : (H, mH ) × (X , μ) ∼ = (, m)

with (X , μ), (Y , ν) being ﬁnite measure spaces, so that the actions G (, m), H (, m) given by g : i(g , y) → i(gg , y), h : j(h , x) → j(hh , x) commute. The index is deﬁned by [G : H] = μ(X )/ν(Y ). e) Measure equivalence between countable groups can be viewed as a category, whose objects are countable groups and morphisms between, say and , are possible (, ) couplings. Composition of morphisms is the operation of composition of ME-couplings as in (c). The trivial ME-coupling (, m ) is nothing but the identity of the object . It is also useful to consider quotient maps : (1 , m1 ) → (2 , m2 ) between (, )-couplings (these are 2-morphisms in the category), which are assumed to be × nonsingular maps, that is, ∗ [m1 ] ∼ m2 . Since 2. One should formally distinguish between (, m) as a (, ) coupling, and the same space with the same actions as a (, ) coupling; hereafter we shall denote the latter by ˇ m). (, ˇ Example 2.2 illustrates the need to do so.

a survey of measured group theory / 301

preimage of a fundamental domain is a fundamental domain, it follows (under the ergodicity assumption) that m1 ( −1 (E)) = c · m2 (E), E ⊂ 2 , where 0 < c < ∞. ME self-couplings of that have the trivial -coupling are especially useful, their cocycles are conjugate to isomorphisms. Similarly, (, )-couplings, which have a discrete coupling as a quotient, correspond to virtual isomorphisms (see Lemma 4.18). f ) Finally, one might relax the deﬁnition of quotients by considering equivariant maps : 1 → 2 between (i , i )-couplings (i , mi ) with respect to homomorphisms 1 → 2 , 1 → 2 with ﬁnite kernels and cokernels. Gromov’s motivation for ME comes from the theory of lattices. Recall that a subgroup of a locally compact second countable (lcsc for short) group G is a lattice if is discrete in G and the quotient space G/ carries a ﬁnite G-invariant Borel regular measure (necessarily unique up to normalization); equivalently, if the -action on G by left (equivalently, right) translations admits a Borel fundamental domain of ﬁnite positive Haar measure. A discrete subgroup < G with G/ being compact is automatically a lattice. Such lattices are called uniform or cocompact; others are nonuniform. The standard example of a nonuniform lattice is = SLn (Z) in G = SLn (R). Recall that a lcsc group that admits a lattice is necessarily unimodular. A common theme in the study of lattices (say in Lie, or algebraic groups over local ﬁelds) is that certain properties of the ambient group are inherited by its lattices. From this perspective it is desirable to have a general framework in which lattices in the same group are considered equivalent. Measure equivalence provides such a framework. ME

example 2.2. If and are lattices in the same lcsc group G, then ∼ ; the group G with the Haar measure mG is a (, ) coupling where (γ , λ) : g → γ gλ−1 . ME

ME

(In fact, ∼ G ∼ if ME is considered in the broader context of unimodular lcsc groups: G × {pt} ∼ = × G/ ). This example also illustrates the fact that ˇ is better related to the original (, )-coupling G the dual (, )-coupling G −1 via g → g rather than the identity map. In geometric group theory the basic notion of equivalence is quasi-isometry QI

(QI). Two metric spaces (Xi , di ), i = 1, 2 are quasi-isometric (notation: X1 ∼ X2 ) if there exist maps f : X1 → X2 , g : X2 → X1 , and constants M, A so that

302 / a. furman

d2 ( f (x), f (x )) < M · d1 (x, x ) + A

(x, x ∈ X1 )

d2 (g(y), g(y )) < M · d2 (y, y ) + A

(y, y ∈ X2 )

d1 (g ◦ f (x), x) < A

(x ∈ X1 )

d2 ( f ◦ g(y), y) < A

(y ∈ X2 ).

Two ﬁnitely generated groups are QI if their Cayley graphs (with respect to some/any ﬁnite sets of generators) are QI as metric spaces. It is easy to see that ﬁnitely generated groups commensurable modulo ﬁnite groups are QI. Gromov observes that QI between groups can be characterized as topological equivalence (TE) deﬁned in the following statement.

theorem 2.3. (gromov [66, theorem 0.2.C2 ]) Two ﬁnitely generated groups and are quasi-isometric iff there exists a locally compact space with a continuous action of × , where both actions and are properly discontinuous and cocompact. The space X in the above statement is called a TE-coupling. Here is an idea for the proof. Given a TE-coupling one obtains a quasi-isometry from any point p ∈ by choosing f : → , g : → so that γ p ∈ f (γ )X and λp ∈ g(λ)Y , where X , Y ⊂ are open sets with compact closures and = γ ∈ γ Y = λ∈ λX . To construct a TE-coupling from a quasi-isometry f : → , consider the pointwise closure of the × -orbit of f in the space of all maps → where acts by precomposition on the domain and by postcomposition on the image. For more details see the guided exercise in [67, p. 98]. A nice instance of QI between groups is a situation where the groups admit a common geometric model. Here a geometric model for a ﬁnitely generated group is a (complete) separable metric space (X , d) with a properly discontinuous and cocompact action of on X by isometries. If X is a common QI

QI

geometric model for 1 and 2 , then 1 ∼ X ∼ 2 . For example, fundamental groups i = π1 (Mi ) of compact locally symmetric manifolds M1 and M2 ˜ 2 = X have X as a common geometric ˜1∼ with the same universal cover M =M model. Notice that the common geometric model X itself does not serve as a TE-coupling because the actions of the two groups do not commute. However, a TE-coupling can be explicitly constructed from the group G = Isom (X , d), which is a locally compact (in fact, compactly generated due to ﬁnite generation assumption on i ) second countable group. Indeed, the isometric actions

a survey of measured group theory / 303

i (X , d) deﬁne homomorphisms i → G with ﬁnite kernels and images being uniform lattices. Moreover, the converse is also true: if 1 , 2 admit homomorphisms with ﬁnite kernels and images being uniform lattices in the same compactly generated second countable group G, then they have a common geometric mode—take G with a (pseudo-)metric arising from an analogue of a word metric using compact sets. Hence all uniform lattices in the same group G are QI to each other. Yet, typically, nonuniform lattices in G are not QI to uniform ones—see Farb’s survey [37] for the QI classiﬁcation for lattices in semisimple Lie groups. To summarize this discussion: the notion of measure equivalence is an equivalence relation between countable groups, an important instance of which is given by groups that can be embedded as lattices (uniform or not) in the same lcsc group. It can be viewed as a measure-theoretic analogue of the equivalence relation of being quasi-isometric (for ﬁnitely generated groups) by taking Gromov’s topological equivalence point of view. An important instance of QI/TE is given by groups that can be embedded as uniform lattices in the same lcsc group. In this situation one has both ME and QI. However, we should emphasize that this is merely an analogy: the notions of QI and ME do not imply each other. 2.2. Orbit Equivalence in Ergodic Theory

Ergodic theory investigates dynamical systems from a measure-theoretic point of view. Hereafter we shall be interested in measurable, measurepreserving group actions on a standard nonatomic probability measure space, and will refer to such actions as probability measure preserving (p.m.p.). It is often convenient to assume the action to be ergodic, that is, to require all measurable -invariant sets to be null or conull (that is, μ(E) = 0 or μ(X \ E) = 0). A basic question in this context concerns possible orbit structures of actions. Equivalence of orbit structures is captured by the following notions of orbit equivalence (the notion of an orbit structure itself is discussed in §4.1).

definition 2.4. Two p.m.p. actions (X , μ) and (Y , ν) are orbit OE equivalent (abbreviated OE, denoted (X , μ) ∼ (Y , ν)) if there exists a measure space isomorphism T : (X , μ) ∼ = (Y , ν) which takes -orbits onto -orbits. More precisely, an orbit equivalence is a Borel isomorphism T : X ∼ = −1 Y between conull subsets X ⊂ X and Y ⊂ Y with T∗ μ(E) = μ(T E) = ν(E), E ⊂ Y and T (.x ∩ X ) = .T (x) ∩ Y for x ∈ X .

304 / a. furman

∼ Y between A weak OE, or stable OE (SOE) is a Borel isomorphism T : X = positive measure subsets X ⊂ X and Y ⊂ Y with T∗ μX = νY , where μX = μ(X )−1 · μ|X , νY = ν(Y )−1 · ν|Y , so that T (.x ∩ X ) = .T (x) ∩ Y for all x ∈ X . The index of such an SOE-map T is μ(Y )/ν(X ).

In the study of orbit structure of dynamical systems in the topological or smooth category, one often looks at such objects as ﬁxed or periodic points/orbits. Despite the important role these notions play in the underlying dynamical system, periodic orbits have zero measure and therefore are invisible from the purely measure-theoretic standpoint. Hence OE in ergodic theory is a study of the global orbit structure. This point of view is consistent with the general philosophy of “noncommutative measure theory,” that is, von Neumann algebras. Speciﬁcally, OE in ergodic theory is closely related to the theory of II1 factors as follows. In the 1940s Murray and von Neumann introduced the so-called “groupmeasure space” construction to provide interesting examples of von Neumann factors:3 given a probability measure preserving (or more generally, nonsingular) group action (X , μ) the associated von Neumann algebra M X is a cross-product of with the abelian algebra L∞ (X , μ), namely the weak closure in bounded operators on L2 ( × X ) of the algebra generated by the operators f (g, x) → f (γ g, γ .x) : γ ∈ and f (g, x) → φ(x)f (g, x) : φ ∈ L∞ (X , μ) . Ergodicity of (X , μ) is equivalent to M X being a factor. It turns out that OE

(for essentially free) OE actions X ∼ Y the associated algebras are isomorphic M X ∼ = MY , with the isomorphism identifying the abelian ∞ subalgebras L (X ) and L∞ (Y ). The converse is also true (one has to specify, in addition, an element in H 1 ( X , T))—see Feldman-Moore [39, 40]. So orbit equivalence of (essentially free p.m.p. group actions) ﬁts into the study of II1 factors M X with a special focus on the so-called Cartan subalgebra given by L∞ (X , μ). We refer the reader to Popa’s 2006 ICM lecture [114] and Vaes’s Seminar Bourbaki paper [136] for some more recent reports on this rapidly developing area. The above-mentioned assumption of essential freeness of an action (X , μ) means that, up to a null set, the action is free; equivalently, for μ-a.e. x ∈ X the stabilizer {γ ∈ : γ .x = x} is trivial. This is a natural assumption, when one wants the acting group to “fully reveal itself” in a.e. orbit of the action. Let us now link the notions of OE and ME. 3. von Neumann algebras whose center consists only of scalars.

a survey of measured group theory / 305

theorem 2.5. Two countable groups and are measure equivalent iff they admit essentially free (ergodic) probability measure-preserving actions (X , μ) and (Y , ν) that are stably orbit equivalent. (SOE) =⇒ (ME) direction is more transparent in the special case of orbit equivalence, that is, index 1. Let α : × X → be the cocycle associated to an orbit equivalence T : (X , μ) → (Y , ν) deﬁned by T (g.x) = α(g, x).T (x) (here freeness of Y is used). Consider (, m) = (X × , μ × m ) with the actions 2.1 g : (x, h) → (gx, α(g, x)h),

h : (x, k) → (x, hk−1 )

(g ∈ , h ∈ ).

Then X × {1} is a common fundamental domain for both actions (note that here freeness of X is used). Of course, the same coupling (, m) can be viewed as (Y × , ν × m ) with the -action deﬁned using β : × Y → given by T −1 (h.y) = β(h, y).T −1 (y). In the more general setting of stable OE one needs to adjust the deﬁnition for the cocycles (see [45]) to carry out a similar construction. Alternative packaging for the (OE) =⇒ (ME) argument uses the language of equivalence relations (see §4.1). Identifying Y with X via T −1 , one views RY and R X as a single relation R. Taking = R equipped with the measure μ˜ (§4.1) consider the actions g : (x, y) → (g.x, y),

h : (x, y) → (x, h.y)

(g ∈ , h ∈ ).

Here the diagonal embedding X → R, x → (x, x), gives the fundamental domain for both actions. (ME) =⇒ (SOE). Given an ergodic (, ) coupling (, m), let X , Y ⊂ be fundamental domains for the , actions; these may be chosen so that m(X ∩ Y ) > 0. The ﬁnite measure-preserving actions 2.2

X ∼ = /,

Y ∼ = /

have weakly isomorphic orbit relations, since they appear as the restrictions to X and Y of the relation R× (of type II∞ ); these restrictions coincide on X ∩ Y . The index of this SOE coincides with the ME-index [ : ] (if [ : ] = 1 one can ﬁnd a common fundamental domain X = Y ). The only remaining issue is that the actions X ∼ = /, Y ∼ = / may not be essential free. This can be ﬁxed (see [56]) by passing to an extension : ¯ m) ¯ ¯ are essentially free. Indeed, (, ¯ → (, m) where / and / ¯ = × Z × W , where Z and W are free probability measuretake preserving actions and let g : (ω, z, w) → (gω, gz, w),

h : (ω, z, w) → (hω, z, hw)

(g ∈ , h ∈ ).

306 / a. furman

remark 2.6. Freeness of actions is mostly used in order to deﬁne the rearrangement cocycles for a (stable) orbit equivalence between actions. However, if SOE comes from a ME-coupling, the well-deﬁned ME-cocycles satisfy the desired rearrangement property (such as T (g.x) = α(g, x).T (x)) and freeness becomes superﬂuous. ¯ → is as above and X¯ , Y¯ denote the preimages of X , Y , then X¯ , Y¯ are If : SOE , fundamental domains and the OE-cocycles X¯ ∼ Y¯ coincide with the ME-cocycles associated with X , Y ⊂ . Another, essentially equivalent, point of view is that ME-coupling deﬁnes a weak isomorphism between the groupoids / and / . In case of free actions these groupoids reduce to their relations groupoids, but in general the information about stabilizers is carried by the ME-cocycles.

2.3. Further Comments on QI, ME, and Related Topics

Let be Gromov’s topological equivalence between and . Then any point x ∈ deﬁnes a quasi-isometry qx : → (see the sketch of proof of Theorem 2.3). In ME the maps α(−, x) : → deﬁned for a.e. x ∈ X play a similar role. However, due to their measure-theoretic nature, such maps are insigniﬁcant taken individually, and are studied as a measured family with the additional structure given by the cocycle equation. Topological and measure equivalences are related to the following interesting notion, introduced by Nicolas Monod in [98] under the appealing term “randomorphisms.” Consider the Polish space of all maps f : → with the product uniform topology, and let [, ] = f : → : f (e ) = e . Then acts on [, ] by g : f (x) → f (xg)f (g)−1 , x ∈ . The basic observation is that homomorphisms → are precisely -ﬁxed points of this action.

definition 2.7. A randomorphism is a -invariant probability measure on [, ]. A measurable cocycle c : × X → over a p.m.p. action (X , μ) deﬁnes a randomorphism by pushing forward the measure μ by the cocycle x → c(−, x). Thus orbit equivalence cocycles (see Appendix A.2) correspond to randomorphisms supported on bijections in [, ]. Also note that the

a survey of measured group theory / 307

natural composition operation for randomorphisms, given by the pushforward of the measures under the natural map [1 , 2 ] × [2 , 3 ] → [1 , 3 ],

( f , g) → g ◦ f

corresponds to the composition of couplings. The viewpoint of topological dynamics of the -action on [, ] may be related to quasi-isometries and topological equivalence. For example, points in [, ] with precompact orbits correspond to Lipschitz embeddings → .

2.3.1. using me for qi Although measure equivalence and quasi-isometry are parallel in many ways, these concepts are different and neither one implies the other. Yet, Yehuda Shalom has shown [132] how one can use ME ideas to study QI of amenable groups. The basic observation is that a topological coupling of amenable groups and carries a × -invariant measure m (coming from a -invariant probability measure on /), which gives a measure equivalence. It can be thought of as an invariant distribution on quasi-isometries → , and can be used to induce unitary representations, cohomology with unitary coefﬁcients, and so on from to . Using such constructions, Shalom [132] was able to obtain a list of new QI invariants in the class of amenable groups, such as (co)homology over Q, ordinary Betti numbers βi () among nilpotent groups, and others. Shalom also studied the notion of uniform embedding (UE) between groups and obtained group invariants, which are monotonic with respect to UE. In [125] Roman Sauer obtains further QI-invariants and UE-monotonic invariants using a combination of QI, ME, and homological methods. In another work [126] Sauer used ME point of view to attack problems of purely topological nature related to the work of Gromov. 2.3.2. p -measure equivalence Let and be ﬁnitely generated groups, equipped with some word metrics | · | , | · | . We say that a (, ) coupling (, m) is p for some 1 ≤ p ≤ ∞ if there exist fundamental domains X , Y ⊂ so that the associated ME-cocycles (see Appendix A.3) α : × X → and β : × Y → satisfy ∀g ∈ :

|α(g, −)| ∈ Lp (X , μ),

∀h ∈ :

|β(h, −)| ∈ Lp (Y , ν).

If an p -ME-coupling exists, say that and are p -ME. Clearly any p ME-coupling is q for all q ≤ p. So 1 -ME is the weakest and ∞ -ME is the most stringent among these relations. One can check that p -ME is an

308 / a. furman

equivalence relation on groups (the p condition is preserved under composition of couplings), so we obtain a hierarchy of p -ME categories with 1 -ME being the weakest (largest classes) and at p = ∞ one arrives at ME + QI. Thus p -ME amounts to measure equivalence with some geometric ﬂavor. The setting of 1 -ME is considered in [13, 14] by Uri Bader, Roman Sauer, and the author to analyze rigidity of the least rigid family of lattices—lattices in SOn,1 (R) Isom (HnR ), n ≥ 3, and fundamental groups of general negatively curved manifolds. It should be noted that examples of non amenable ME groups that are not 1 -ME seem to be rare (surface groups and free groups seem to be the main culprits). In particular, it follows from Shalom’s computations in [131] that for n ≥ 3 all lattices in SOn,1 (R) are mutually 1 -ME. We shall return to invariants and rigidity in the 1 -ME framework in Sections 3.1.8 and 3.2.4.

3. Measure Equivalence Between Groups This section is concerned with the notion of measure equivalence between ME

countable groups ∼ (Deﬁnition 2.1). First recall the following deep result (extending previous work of Dye [33, 34] on some amenable groups, and followed by Connes-Feldman-Weiss [27] concerning all nonsingular actions of all amenable groups).

theorem 3.1. (ornstein-weiss [105]) Any two ergodic probability measure-preserving actions of any two inﬁnite countable amenable groups are orbit equivalent. This result implies that all inﬁnite countable amenable groups are ME; moreover, for any two inﬁnite amenable groups and there exists an ergodic ME-coupling with index [ : ] = 1 (hereafter we shall denote this situaOE

tion by ∼ ). Measure equivalence of all amenable groups shows that many QI-invariants are not ME-invariants; these include growth type, being virtually nilpotent, (virtual) cohomological dimension, ﬁnite generations/presentation, and so on. The following are basic constructions and examples of measure equivalent groups: 1) If and can be embedded as lattices in the same lcsc group, then ME

∼ .

a survey of measured group theory / 309 ME

ME

2) If i ∼ i for i = 1, . . . , n, then 1 × · · · × n ∼ 1 × · · · × n . OE

3) If i ∼ i for i ∈ I (i.e. the groups admit an ergodic ME-coupling with OE

index 1), then ( ∗i∈I i ) ∼ ( ∗i∈I i ).4 For 2 = n, m < ∞ the free groups Fn and Fm are commensurable, and thereME

fore are ME (however, F∞ ∼ F2 ). The measure equivalence class ME(F2≤n 0 is said to weakly contain the trivial representation 1 , denoted 1 ≺ π. The trivial representation 1 is (strongly) contained in π , denoted 1 < π , if there exist non-zero π(G)-invariant vectors, that is, H π() = {0}. Of course 1 < π trivially implies 1 ≺ π. We recall: Amenability: is amenable if the trivial representation is weakly contained in the regular representation ρ : → U(2 ()), ρ(g)f (x) = f (g −1 x). OE

4. The appearance of the sharper condition ∼ in (2) is analogous to the one in the QI QI

context: if groups i and i are bi-Lipschitz, then ∗i∈I i ∼ ∗i∈I i .

310 / a. furman

Property (T): has property (T) (Kazhdan [79]) if for every unitary representation π : 1 ≺ π implies 1 < π . This is equivalent to an existence of a compact K ⊂ and > 0 so that any unitary -representation π with (K , )-almost invariant vectors has nontrivial invariant vectors. For compactly generated groups, another equivalent characterization (Delorme and Guichardet) is that any afﬁne isometric -action on a Hilbert space has a ﬁxed point, that is, if H 1 (, π ) = {0} for any (orthogonal) -representation π . We refer to [17] for the details. (HAP): is a-T-menable (or has Haagerup approximation property) if the following equivalent conditions hold: (i) has a mixing -representation weakly containing the trivial one, or (ii) has a proper afﬁne isometric action on a (real) Hilbert space. The class of inﬁnite a-T-menable groups contains amenable groups, free groups but is disjoint from inﬁnite groups with property (T). See [24] as a reference. Measure equivalence allows us to relate unitary representations of one group to another. More concretely, let (, m) be a (, ) coupling, and π : → U(H ) be a unitary -representation. Denote by H˜ the Hilbert space consisting of equivalence classes (mod null sets) of all measurable, -equivariant maps → H with square-integrable norm over a -fundamental domain:

H˜ = f : → H : f (λx) = π (λ)f (x), f 2 < ∞ mod null sets. /

The action of on such functions by translation of the argument deﬁnes a unitary -representation π˜ : → U(H˜ ). This representation is said to be induced from π : → U(H ) via . (In Example 2.2 this is precisely the usual Mackey induction of a unitary representations, of a lattice to the ambient group, followed by a restriction to another lattice). The ME invariance of the properties above (amenability, property (T), Haagerup approximation property) can be deduced from the following observations. Let (, m) be a (, ) ME-coupling, π : → U(H ) a unitary representation, and π˜ : → U(H˜ ) the corresponding induced representation. Then 1) If π is the regular -representation on H = 2 (), then π˜ on H˜ can be identiﬁed with the -representation on L2 (, m) ∼ = n · 2 (), where 2 n = dim L (/) ∈ {1, 2, . . . , ∞}. 2) If 1 ≺ π, then 1 ≺ π. ˜ 3) If (, m) is × ergodic and π is weakly mixing (i.e., 1 < π ⊗ π ∗ ), then 1 < π. ˜

a survey of measured group theory / 311

4) If (, m) is × ergodic and π is mixing (i.e., for all v ∈ H : π (h)v, v → 0 as h → ∞ in ), then π˜ is a mixing -representation. Combining (1) and (2) we obtain that being amenable is an ME-invariant. The deep result of Ornstein-Weiss [105] and Theorem 2.5 imply that any two inﬁnite countable amenable groups are ME. This gives

corollary 3.2. The measure equivalence class of Z is the class of all inﬁnite countable amenable groups ME(Z) = Amen. Bachir Bekka and Alain Valette [16] showed that if does not have property (T), then it admits a weakly mixing representation π weakly containing the trivial one. By (2) and (3) this implies that property (T) is an ME-invariant (this is the argument in [44, corollary 1.4]; see also Zimmer [146, theorem 9.1.7(b)]). The ME-invariance of amenability and Kazhdan’s property for groups indicates that it should be possible to deﬁne these properties for equivalence relations and then relate them to groups. This was done by Zimmer in [138, 141] and was recently further studied in the context of measured groupoids in [8, 9]. We return to this discussion in §4.2.1. The ME-invariance of a-T-menability follows from (2) and (4); see [24, 76].

3.1.2. cost of groups The notion of the cost of an action/relation was introduced by Levitt [90] and developed by Damien Gaboriau [53, 54, 56]; the monographs [81] and [80] also contain an extensive discussion of this topic. The cost of an essentially free p.m.p. action (X , μ), denoted cost ( X ), is the cost of the corresponding orbit relations cost (R X ) as deﬁned in §4.2.4 (it is the inﬁmum of the weights of generating systems for the groupoid where the “weight” is the sum of the measures of the domain/image sets of the generating system). The cost of an action can be turned into a group invariant/(s) by setting C ∗ () = inf cost ( X ), X

C ∗ () = sup cost ( X ) X

where the inﬁmum/supremum are taken over all essentially free p.m.p. actions of (we drop ergodicity assumption here; in the deﬁnition of C ∗ () essential freeness is also superﬂuous). Groups for which C ∗ () = C ∗ () are said to have ﬁxed price, or prix ﬁxe (abbreviated P.F.). For general groups,

312 / a. furman

Gaboriau deﬁned the cost of a group to be the lower one: C () = C ∗ (). To avoid confusion, we shall use here the notation C ∗ () for general groups, and reserve C () for P.F. groups only.

question 3.3. Do all countable groups have property P.F.? The importance of this question will be illustrated in §4.2.4; for example, a positive answer to an apparently weaker Question 4.11 would have applications to groups theory and 3-manifold (Abert-Nikolov [1]). The properties C ∗ = 1, 1 < C ∗ < ∞, and C ∗ = ∞ are ME-invariants. More precisely: ME theorem 3.4. If ∼ , then C ∗ () − 1 = [ : ] · C ∗ () − 1 for some/any (, )-coupling . We do not know whether the same holds for C ∗ . Note that in [54] this MEinvariance is stated for P.F. groups only. Proof. Let be a (, )-coupling with X = / and Y = / being free, where X , Y ⊂ are -, - fundamental domains. Given any essen¯ = × Z with tially free p.m.p. action Z, consider the (, )-coupling the actions g : (ω, z) → (gω, z),

h : (ω, z) → (hω, hz)

(g ∈ , h ∈ ).

¯ ¯ are stably orbit equivalent with and Y¯ = / The actions X¯ = / index [ : ]¯ = [ : ] = c. Hence (using Theorem 4.7 below) we have c · ( cost (R X¯ ) − 1) = cost (RY¯ ) − 1. While X¯ is a skew product over X , the action Y¯ is the diagonal action on Y¯ = Y × Z. Since Y¯ = Y × Z has Z as a -equivariant quotient, it follows (by considering preimages of any “graphing system”) that cost ( Y¯ ) ≤ cost ( Z). Since Z was arbitrary, we deduce C ∗ () − 1 ≥ c · C ∗ () − 1 . A symmetric argument completes the proof.

theorem 3.5. (gaboriau [53,54,56]) The following classes of groups have P.F.:

a survey of measured group theory / 313 1 1) Any ﬁnite group has C ∗ () = C ∗ () = 1 − || . ∗ 2) Inﬁnite amenable groups have C ∗ () = C () = 1. 3) Free group Fn , 1 ≤ n ≤ ∞, have C ∗ (Fn ) = C ∗ (Fn ) = n. 4) Surface groups = π1 ( g ) where g is a closed orientable surface of genus g ≥ 2 have C ∗ () = C ∗ () = 2g − 1. 5) Amalgamated products = A ∗C B of ﬁnite groups have P.F. with

C ∗ () = C ∗ () = 1 − (

1 1 1 + − ). |A| |B| |C|

1 In particular, C ∗ ( SL2 (Z)) = C ∗ ( SL2 (Z)) = 1 + 12 . 6) Assume 1 , 2 have P.F., then the free product 1 ∗ 2 and more general amalgamated free products = 1 ∗A 2 over an amenable group A, has P.F. with

C (1 ∗ 2 ) = C (1 ) + C (2 ),

C (1 ∗A 2 ) = C (1 ) + C (2 ) − C (A).

7) Products = 1 × 2 of inﬁnite nontorsion groups have C ∗ () = C ∗ () = 1. 8) Finitely generated groups containing an inﬁnite amenable normal subgroup have C ∗ () = C ∗ () = 1. 9) Arithmetic lattices of higher Q-rank (e.g., SLn≥3 (Z)) have C ∗ () = C ∗ () = 1. Note that for an inﬁnite group C ∗ () = 1 iff has P.F. of cost 1. So the content of cases (2), (7), (8), and (9) is that C ∗ () = 1.

question 3.6. Is it true that for all (irreducible) lattices in a (semi)simple Lie group G of higher rank have P.F. of C ∗ () = 1? Is it true that any inﬁnite group with Kazhdan’s property (T) has P.F. with C ∗ () = 1? Item (9) in Theorem 3.5 provides a positive answer to the ﬁrst question for some nonuniform lattices in higher-rank Lie groups, but the proof relies on the internal structure of such lattices (chains of pairwise commuting elements), rather than on its relation to the ambient Lie group G (which also has a lot of commuting elements). Note also that Theorem 3.4 implies that C ∗ () = 1 for all higher-rank lattices. The motivation for the second question is that (2) property (T) implies vanishing of the ﬁrst 2 -Betti number, β1 () = 0; while for inﬁnite groups it was shown by Gaboriau that 3.1

(2)

(2)

β1 () = β1 (R X ) ≤ cost (R X ) − 1.

314 / a. furman

Furthermore, there are no known examples of strict inequality. Lattices in (2) higher-rank semisimple Lie groups without property (T) still satisfy β1 () = 0 (2) (an argument in the spirit of the current discussion is β1 for ME groups are positively proportional by Gaboriau’s theorem 3.8, an irreducible lattice in a product is ME to a product of lattices and products of inﬁnite groups (2) have β1 = 0 by the Küneth formula. Shalom’s [130] provides a completely geometric explanation). To give the ﬂavor of the proofs let us indicate the argument for (8) in Theorem 3.5. Let be a group generated by a ﬁnite set {g1 , . . . , gn } and containing an inﬁnite normal amenable subgroup A and (X , μ) be an essentially free (ergodic) p.m.p. action. Since A is amenable, there is a Z-action on X with RAX = RZX (mod null sets), and we let φ0 : X → X denote the action of the generator of Z. Given > 0 one can ﬁnd a subset E ⊂ X with 0 < μ(E) < so that a∈A aE = φ0n E = X mod null sets (if A-action is ergodic, any positive measure set works; in general, one uses the ergodic decomposition). For i = 1, . . . , n let φi be the restriction of gi to E. Now one easily checks that the normality assumption implies that = {φ0 , φ1 , . . . , φn } generates R X , while cost ( ) = 1 + n. For general (not necessarily P.F.) groups i a version of (6) still holds: C ∗ (1 ∗ 2 ) = C ∗ (1 ) + C ∗ (2 ), C ∗ (1 ∗A 2 ) = C ∗ (1 ) + C ∗ (2 ) − C (A) where A is ﬁnite or, more generally, amenable. Very recently Miklos Abert and Benjamin Weiss [2] showed:

theorem 3.7. (abert-weiss [2]) For any discrete countable group , the highest cost C ∗ () is attained by nontrivial Bernoulli actions (X0 , μ0 ) and their essentially free quotients. Some comments are in order. Kechris [80] introduced the following notion: for probability measure-preserving actions of a ﬁxed group say that (X , μ) weakly contains Y if given any ﬁnite measurable partition Y =

n Yi , a ﬁnite set F ⊂ and an > 0, there is a ﬁnite measurable partition i=1 X = ni=1 Xi so that μ(gXi ∩ Xj ) − ν(gYi ∩ Yj ) < (1 ≤ i, j ≤ n, g ∈ F). The motivation for the terminology is the fact that weak containment of actions implies (but not equivalent to) weak containment of the corresponding unitary

a survey of measured group theory / 315

representations: L2 (Y ) L2 (X ). It is clear that a quotient is (weakly) contained in the larger action. It is also easy to see that the cost of a quotient action is no less than that of the original (because one can lift any graphing from a quotient to the larger action maintaining the cost of the graphing). Kechris [80] proves that this (anti)monotonicity still holds in the context of weak containment of essentially free actions of ﬁnitely generated groups, namely: Y X

=⇒

cost ( Y ) ≥ cost ( X ).

In fact, it follows from the more general fact that cost is upper semicontinuous in the topology of actions. Abert and Weiss prove that Bernoulli actions (and their quotients) are weakly contained in any essentially free action of a group. Thus Theorem 3.7 follows from the monotonicity of the cost.

3.1.3. 2 -betti numbers The 2 -Betti numbers of (coverings of) manifolds were introduced by Atiyah in [11]. Cheeger and Gromov [23] deﬁned 2 -Betti (2) numbers βi () ∈ [0, ∞], i ∈ N, for arbitrary countable group as dimensions (in the sense of Murray von Neumann) of certain homology groups (which are Hilbert -modules). For reference we suggest [35], [92]. Here let us just point out the following facts: (2)

1) If is inﬁnite amenable, then βi () = 0, i ∈ N. (2)

(2)

2) For free groups β1 (Fn ) = n − 1 and βi (Fn ) = 0 for i > 1. (2)

3) For groups with property (T), β1 () = 0. (2) (2) (2) 4) Küneth formula: βk (1 × 2 ) = i+j=k βi (1 ) · βj (2 ). 5) Kazhdan’s conjecture, proved by Lück, states that for residually ﬁnite groups satisfying appropriate ﬁniteness properties (e.g., ﬁnite K (pi, 1)) the 2 -Betti numbers are the stable limit of Betti numbers of ﬁnite-index (2) βi (n ) subgroups normalized by the index: βi () = lim [: where > 1 > n] . . . is a chain of normal subgroups of ﬁnite index. (2) 6) The 2 Euler characteristic χ (2) () = ( − 1)i · βi () coincides with the usual Euler characteristic χ () = ( − 1)i · βi (), provided both are deﬁned, as is the case for fundamental group = π1 (M) of a compact aspherical manifold. 7) According to the Hopf-Singer conjecture the 2 -Betti numbers for a fundamental group = π1 (M), of a compact aspherical manifold M vanish except, possibly, in the middle dimension n. Atiyah’s conjecture states that 2 -Betti numbers are integers.

316 / a. furman

The following remarkable result of Damien Gaboriau states that these intricate numeric invariants of groups are preserved under measure equivalence, after a rescaling by the coupling index. ME

theorem 3.8. (gaboriau [55], [57]) Let ∼ be ME-countable groups. Then (2) (2) βi () = c · βi () (i ∈ N) where c = [ : ] is an/the index of some/any (, )-coupling. In fact, Gaboriau introduced the notion of 2 -Betti numbers for II1 -relations and related them to 2 -Betti numbers of groups in case of the orbit relation for an essentially free p.m.p. action—see more comments in §4.2.5 below. Thus the geometric information encoded in the 2 -Betti numbers for fundamental groups of aspherical manifolds, such as Euler characteristic and sometimes the dimension, pass through measure equivalence. In particular, if lattices i (i = 1, 2) (uniform or not) in SUni ,1 (R) are ME, then n1 = n2 ; the same applies to Spni ,1 (R) and SO2ni ,1 (R). (The higher-rank lattices are covered by stronger rigidity statements—see §3.2.1 below). Furthermore, it follows from Gaboriau’s result that in general the set (2) D(2) () = i ∈ N : 0 < βi () < ∞ is an ME-invariant. Conjecture (7) relates this to the dimension of a manifold M in the case of = π1 (M). One shouldn’t expect dim (M) to be an ME-invariant of π1 (M), as the examples of tori show; note also that for any manifold M one ME

has π1 (M × Tn ) ∼ π1 (M × Tk ). However, among negatively curved manifolds Theorem 3.13 below shows that dim (M) is an invariant of 1 -ME. For closed aspherical manifolds M the dimension dim (M) is a QI-invariant of π1 (M). Pansu proved that the whole set D(2) () is a QI-invariant of . However, positive proportionality of 2 -Betti numbers for ME fails under QI; in fact, there are QI groups whose Euler characteristics have opposite signs. Yet

corollary 3.9. For ME groups and with well-deﬁned Euler characteristic, say fundamental groups of compact manifolds, one has χ() = c · χ (),

where

c = [ : ] ∈ (0, ∞).

In particular, the sign (positive, zero, negative) of the Euler characteristic is an ME-invariant.

a survey of measured group theory / 317

3.1.4. cowling-haagerup -invariant This numeric invariant G , taking values in [1, ∞], is deﬁned for any lcsc group G in terms of norm bounds on unit approximation in the Fourier algebra A(G) (see Cowling and Haagerup [31]). The -invariant coincides for a lcsc group and its lattices. Moreover, OE Cowling and Zimmer [32] proved that 1 ∼ 2 implies 1 = 2 . In fact, their proof implies the invariance under measure equivalence (see [76]). So, is an ME-invariant. Cowling and Haagerup [31] computed the -invariant for simple Lie groups and their lattices: in particular, proving that G = 1 for G SOn,1 (R) and SUn,1 (R), G = 2n − 1 for G Spn,1 (R), and G = 21 for the exceptional rank-1 group G = F4(−20) . One may observe that simple Lie groups split into two classes: (1) SOn,1 (R) and SUn,1 (R) family, and (2) G Spn,1 (R), F4(−20) and higher rank. Groups in the ﬁrst class have haagerup approximation property (HAP, a.k.a. a-Tmenability) and G = 1, while groups in the second class have Kazhdan’s property (T) and G > 1. Cowling conjectured that G = 1 and (HAP) might be equivalent. Recently one implication of this conjecture has been disproved: Cornulier, Stalder, and Valette [30] proved that the wreath product H F2 of a ﬁnite group H by the free group F2 has (HAP), while Ozawa and Popa [108] prove that HF2 > 1. The question whether = 1 implies (HAP) is still open. One may deduce now that if is a lattice in G Spn,1 (R) or in F4(−20) and ME

is a lattice in a simple Lie group H, then ∼ iff G H. Indeed, higherrank H are ruled out by Zimmer’s theorem 3.15; H cannot be in the families SOn,1 (R) and SUn,1 (R) by property (T) or Haagerup property; and within the family of Spn,1 (R) and F4(−20) the -invariant detects G (2 -Betti numbers can also be used for this purpose).

3.1.5. treeability, antitreeability, and ergodic dimension In [4] Scott Adams introduced the notion of treeable equivalence relations (see §4.2.3). Following [81], a group is Treeable: if there exists an essentially free p.m.p. -action with a treeable orbit relation. Strongly treeable: if every essentially free p.m.p. -action gives a treeable orbit relation. Antitreeable: if there are no essentially free p.m.p. -actions with a treeable orbit relation.

318 / a. furman

Amenable groups and free groups are strongly treeable. It seems to be still unknown whether there exist treeable but not strongly treeable groups; in particular, it is not clear whether surface groups (that are treeable) are strongly treeable. The properties of being treeable or antitreeable are ME-invariants. MoreME

over, is treeable iff is amenable (i.e., ∼ F1 = Z), or is ME to either F2 or F∞ (this fact uses Hjorth’s [69]; see [81, theorems 28.2 and 28.5]). Groups with Kazhdan’s property (T) are antitreeable [6]. More generally, it follows from the recent work of Alvarez and Gaboriau [7] that a nonamenable group (2) with β1 () = 0 is antitreeable (in view of Equation (3.1) this also strengthens [54, corollaire VI.22], where Gaboriau showed that a nonamenable with C ∗ () = 1 is antitreeable). A treeing of a relation can be seen as a -invariant assignment of pointed trees with as the set of vertices. One may view the relation acting on this measurable family of pointed trees by moving the marked point. More generally, one might deﬁne actions by relations, or measured groupoids, on ﬁelds of simplicial complexes. Gaboriau deﬁnes (see [57]) the geometric dimension of a relation R to be the smallest possible dimension of such a ﬁeld of contractible simplicial complexes; the ergodic dimension of a group will be the minimal geometric dimension over orbit relations R X of all essentially free p.m.p. -actions. In this terminology R is treeable iff it has geometric dimension 1, and a group is treeable if its ergodic dimension is 1. There is also a notion of an approximate geometric/ergodic dimension [57] describing the dimensions of a sequence of subrelations approximating a given orbit relation.

theorem 3.10. (gaboriau [57]) Ergodic dimension and approximate ergodic dimension are ME-invariants. This notion can be used to obtain some information about ME of lattices in the family of rank-1 groups SOn,1 (R). If i < SOni ,1 (R), i = 1, 2 are lattices ME

and 1 ∼ 2 , then Gaboriau’s result on 2 -Betti numbers shows that if one (2) of ni is even, then n1 = n2 . However, for ni = 2ki + 1 all βi vanish. In this case Gaboriau shows, using the above ergodic dimension, that k1 ≤ k2 ≤ 2k1 or k2 ≤ k1 ≤ 2k2 . OE

OE

3.1.6. free products It was mentioned above that if i ∼ , then ∗i∈I i ∼ OE ∗i∈I i (here ∼ means that the two groups admit an ergodic ME-coupling

a survey of measured group theory / 319

with index 1, and equivalently admit essentially free actions that are orbit equivalent). To what extent does the converse hold? Namely, when can one recognize the free factors on the level of measure equivalence? This problem was extensively studied by Ioana, Peterson, and Popa in [75] where strong rigidity results were obtained for orbit relations under certain assumptions on the actions (see §4.2.8). Here let us formulate a recent result from Alvarez and Gaboriau [7] that can be stated in purely group theoretic terms. In [7] a notion of measurably freely indecomposable groups (MFI) is introduced, and it is shown that this class includes all nonamenable (2) groups with β1 = 0. Thus, inﬁnite property (T) groups, nonamenable direct products, are examples of MFI groups. ME

theorem 3.11. (alvarez-gaboriau [7]) Suppose that ∗ni=1 i ∼ ∗m , j=1 j ME

are two sets of MFI groups with i ∼ i for 1 ≤ i = where {i }ni=1 and {j }m j=1 ME

i ≤ n, and j ∼ j for 1 ≤ j = j ≤ m. Then n = m and, up to a permutation ME

of indices, i ∼ i . Another result from [7] concerning decompositions of equivalence relations as free products of subrelations is discussed in §4.2.8. Let us also mention recent works of Kida [88] and Popa and Vaes [120] that describe extremely strong rigidity properties for certain amalgamated products of various rigid groups.

3.1.7. the classes Creg and C In §3.2.2 below we shall discuss rigidity results obtained by Nicolas Monod and Yehuda Shalom in [101] (see also [99, 100] and jointly with Mineyev [96]). These results involve second-bounded cohomology with unitary coefﬁcients: Hb2 (, π )—a certain vector space associated to a countable group and a unitary representation π : → U(Hπ ). (Some background on bounded cohomology can be found in [101, §3] or [98]; for more details see [21, 97]). Monod and Shalom deﬁne the class Creg of groups characterized by the property that Hb2 (, 2 ()) = {0} and (potentially larger) class C of groups with nonvanishing Hb2 (, π ) for some mixing -representation π . Known examples of groups in Creg ⊂ C include groups admitting “hyperboliclike” actions of the following types (see [100], [96]):

320 / a. furman

i) nonelementary simplicial action on some simplicial tree, proper on the set of edges; ii) nonelementary proper isometric action on some proper CAT(-1) space; and iii) nonelementary proper isometric action on some Gromov-hyperbolic graph of bounded valency. Hence Creg includes free groups, free products of arbitrary countable groups, and free products amalgamated over a ﬁnite group (with the usual exceptions of order 2), fundamental groups of negatively curved manifolds, Gromov hyperbolic groups, and nonelementary subgroups of the above families. Examples of groups not in C include amenable groups, products of at least two inﬁnite groups, lattices in higher-rank simple Lie groups (over any local ﬁeld), and irreducible lattices in products of general compactly generated nonamenable groups (see [101, §7]).

theorem 3.12. (monod-shalom [101]) 1) Membership in Creg or C is an ME-invariant. 2) For direct products = 1 × · · · × n where i ∈ Creg are torsion free, the number of factors and their ME types are ME-invariants. ME

3) For as above, if ∼ , then cannot be written as a product of m > n inﬁnite torsion-free factors.

3.1.8. dimension and simplicial volume (1 -me) Geometric properties are hard to capture with the notion of measure equivalence. The 2 -Betti numbers is an exception, but this invariant beneﬁts from its Hilbert space nature. In [13, 14] Uri Bader, Roman Sauer, and the author consider a restricted version of measure equivalence, namely, 1 -ME (see §2.3.2 for a deﬁnition). Being 1 ME is an equivalence relation between ﬁnitely generated groups, in which any two integrable lattices in the same lcsc group are 1 -ME. All uniform lattices are integrable, and so are all lattices in SOn,1 (R) Isom (HnR ) (see §3.2.4). theorem 3.13. (bader-furman-sauer [14]) Let i = π1 (Mi ) where Mi are closed manifolds that admit a Riemannian metric of negative sectional curvature. Assume that 1 and 2 admit an 1 -ME-coupling . Then dim (M1 ) = dim (M2 )

and

M1 = [2 : 1 ] · M2 ,

where Mi denotes the simplicial volume of Mi .

a survey of measured group theory / 321

The simplicial volume M of a closed manifold M, introduced by Gromov in [65], is the norm of the image of the fundamental class under the comparison 1 map Hn (M) → Hn (M) into the 1 -homology, which is an 1 -completion of the usual homology. This is a homotopy invariant of manifolds. Manifolds carrying a Riemannian metric of negative curvature have M > 0 (Gromov [65]). 3.2. Orbit/Measure Equivalence Rigidity

Let us now turn to measure equivalence rigidity results, that is, classiﬁcation results in the ME category. In the introduction to this section we mentioned that the ME class ME(Z) is precisely all inﬁnite amenable groups. The (distinct) classes ME(F2≤n 0, that is, rule out the torus 1,0 , once-punctured torus 1,1 , and spheres 0,p with p ≤ 4 punctures. theorem 3.20. (kida [86]) Let be a ﬁnite-index subgroup in ( g,p ) with 3g + p − 4 > 0, or in a ﬁnite product of such mapping class groups ni=1 ( g,p ) . ME

Then any group ∼ is commensurable up to ﬁnite kernels to , and ergodic ME-coupling has a discrete (, )-coupling as a quotient. This work (spanning [83, 85, 86]) is a real tour de force. Mapping class groups ( ) are often compared to a lattice in a semisimple Lie group G: the Teichmüller space T ( ) is analogous to the symmetric space G/K , Thurston boundary PML( ) analogous to Furstenberg boundary B(G) = G/P, and the curve complex C( ) to the spherical Tits’s building of G. The MCG has been extensively studied as a geometric object, while Kida’s work provides a new ergodic-theoretic perspective. For example, Kida proves that Thurston boundary PML( ) with the Lebesgue measure class is -boundary in the sense of Burger-Monod for the mapping class group, that is, the action of the latter is amenable and doubly ergodic with unitary coefﬁcients. Properties of the MCG action on PML( ) allow Kida to characterize certain subrelations/subgroupoids arising in self-measure equivalence of a MCG; leading to the proof of a cocycle (strong) rigidity Theorem 5.7, which can be viewed as a groupoid version of Ivanov’s rigidity theorem. This strong rigidity theorem can be used with §5.5 to recognize arbitrary groups ME to an MCG. Note that a mapping class group behaves like a “lattice without ambient Lie group”—all its ME-couplings have discrete quotients. Moreover, Kida’s ME rigidity results extend to products of MCGs without any irreducibility assumptions. From this point of view MCGs are more ME rigid than higherrank lattices, despite the fact that they lack many other rigidity attributes, such as property (T) (see Andersen [10]). Added in proof. Very recently additional extremely strong ME rigidity results were obtained in Kida [88] and Popa and Vaes [120] for certain amalgamated products of higher-rank lattices and also mapping class groups. The latter paper also establishes W ∗ -rigidity.

a survey of measured group theory / 325

3.2.4. hyperbolic lattices and 1 -me Measure equivalence is motivated by the theory of lattices, with ME-couplings generalizing the situation of groups embedded as lattices in the same ambient lcsc group. Thus, in the context of semisimple groups, one wonders whether ME rigidity results would parallel Mostow rigidity; and in particular would apply to (lattices in) all simple groups with the usual exception of SL2 (R) SO2,1 (R) SU1,1 (R). The higher-rank situation (at least that of simple groups) is well understood (§3.2.1). In the rank 1 case (consisting of the families SOn,1 (R), SUm,1 (R), Spk,1 (R), and F4(−20) ) known ME-invariants discussed above (namely: property (T), 2 -Betti numbers, -invariant, ergodic dimension) allow to distinguish lattices among most rank 1 groups. This refers to statements of the form: if i < Gi are ME lattices, then 1 ∼ 2 iff G1 G2 . However, ME classiﬁcation such as in Theorems 3.16, 3.17, and 3.20 are not known for rank 1 cases. The ingredient that is missing in the existing approach is an appropriate cocycle superrigidity theorem.7 In a joint work with Uri Bader and Roman Sauer a cocycle strong rigidity theorem is proved for ME-cocycles for lattices in SOn,1 (R) Isom (HnR ), n ≥ 3, under a certain 1 -assumption (see §2.3.2). It is used to obtain the following: theorem 3.21. (bader-furman-sauer [13]) Let is a lattice in G = Isom (Hn ), n ≥ 3, and is some ﬁnitely generated group 1 -ME to then is a lattice in G modulo a ﬁnite normal subgroup. Moreover, any ergodic (, )coupling has a quotient, which is either discrete, or G = Aut (G), or G0 with the Haar measure. Recently Sorin Popa has introduced a new set of ideas for studying orbit equivalence. These results, rather than relying on rigidity of the acting groups alone, exploit rigidity aspects of groups actions of certain type. We shall discuss them in §Sections 5.4, and 5.6, and 5.7. 3.3. How Many Orbit Structures Does a Given Group Have?

Theorem 3.1 of Ornstein and Weiss [105] implies that for an inﬁnite amenable countable group all ergodic probability measure preserving actions (X , μ) deﬁne the same orbit structure, namely, Ramen . What happens for non-amenable groups ? 7. For Spn,1 (R) and F4(−20) a cocycle superrigidity theorem was proved by Corlette and Zimmer [29] (see also Fisher and Hitchman [43]), but these results require boundness assumptions that preclude them from being used for ME-cocycles.

326 / a. furman

theorem 3.22. (epstein [36], after ioana [73] and gabotiaulyons [60]) Any nonamenable countable group has a continuum of essentially free ergodic probability measure-preserving actions (X , μ), no two of which are stably orbit equivalent. Let us brieﬂy discuss the problem and its solution. Since Card( Aut (X , μ) ) = ℵ = 2ℵ0 there are at most continuum many actions for a ﬁxed countable group . In fact, this upper bound on the cardinality of isomorphism classes of actions is achieved, using the corresponding fact about unitary representations and the Gaussian construction. Hence one might expect at most ℵ-many non-OE actions for any given . OE rigidity results showed that some speciﬁc classes of groups indeed have many mutually non-OE actions; this includes higher-rank lattices [64], products of hyperboliclike groups [101, theorem 1.7], and some other classes of groups [110, 113]). But the general question, regarding an arbitrary nonamenable , remained open. Most invariants of equivalence relations depend on the acting group rather than the action, and thus could not be used to distinguish actions of the ﬁxed group . The notable exception to this metamathematical statement appears for nonamenable groups that do not have property (T). For such groups two non-SOE actions can easily be constructed: (1) a strongly ergodic action (using Schmidt’s [129]), and (2) an ergodic action that is not strongly ergodic (using Connes-Weiss [28]). Taking a product with an essentially free weakly mixing strongly ergodic -actions (e.g., the Bernoulli action (X0 , μ0 ) ) makes the above two actions essentially free and distinct. In [68] Greg Hjorth showed that if has property (T), the set of isomorphism classes of orbit structures for essentially free -actions has cardinality ℵ, by proving that the natural map from the isomorphism classes of essentially free ergodic -actions to the isomorphism classes of -orbit structures is at most countable-to-one. More precisely, the space of -actions producing a ﬁxedorbit structure is equipped with a structure of a Polish space (separability) where any two nearby actions are shown to be conjugate. This is an example of proving rigidity up to countable classes combining separability of the ambient space with a local rigidity phenomenon (stemming from property (T); see §5.6 below). These ideas can be traced back to Connes [26] and Popa [106], and play a central role in the most recent developments—see [114] §4]. The challenge now became to show that other nonamenable groups have inﬁnitely, or even ℵ-many, non-OE essentially free ergodic actions. Damien Gaboriau and Sorin Popa [59] achieved this goal for the quintessential representative of a nonamenable group without property (T), namely for the free

a survey of measured group theory / 327

group F2 . Using a sophisticated rigidity vs. separability argument they showed that within a certain rich family of F2 -actions the map from isomorphism classes of actions to orbit structures is countable-to-one. The rigidity component of the argument was this time provided by Popa’s notion of w-rigid actions such as SL2 (Z) T2 , with the rigidity related to the relative property (T) for the semidirect product SL2 (Z) Z2 viewing Z2 as the Pontryagin dual of T2 . In [73] Adrian Ioana obtained a sweeping result showing that any containing a copy of F2 has ℵ-many mutually non-SOE essentially free actions. The basic idea of the construction being to use a family of non-SOE of F2 /F actions F2 Xt to construct co-induced -actions Xt 2 and pushing the solution of F2 -problem to the analysis of the co-induced actions. The class of groups containing F2 covers “most of” the class of nonamenable groups with few, very hard to obtain, exceptions. The ultimate solution to the problem, covering all nonamenable groups, was shortly obtained by Inessa Epstein [36] using a result by Damien Gaboriau and Russel Lyons [60], who proved that any nonamenable contains an F2 in a sort of measure-theoretical sense. Epstein was able to show that this sort of containment sufﬁces to carry out an analogue of Ioana’s co-induction argument [73] to prove Theorem 3.22. Furthermore, in [74] Ioana, Kechris, and Tsankov, jointly with Epstein, show that for any nonamenable the space of all ergodic-free p.m.p. actions taken up to OE not only has cardinality of the continuum, but is also impossible to classify in a very strong sense. One may also add that most of the general results mentioned above show that within certain families of actions the grouping into SOE-ones has countable classes, therefore giving only implicit families of non-SOE actions. In [71] Ioana provided an explicit list of a continuum of mutually non-SOE actions of F2 .

4. Measured Equivalence Relations 4.1. Basic Deﬁnitions

We start with the notion of countable equivalence relations in the Borel setting. It consists of a standard Borel space (X , X ) (cf. [39] for deﬁnitions) and a Borel subset R ⊂ X × X , an equivalence relation whose equivalence classes R[x] = y ∈ X : (x, y) ∈ R are all countable. To construct such relations choose a countable collection = {φi }i∈I of Borel bijections φi : Ai → Bi between Borel subsets Ai , Bi ∈ X , i ∈ I; and let R be the smallest equivalence relation including the graphs of all φi , i ∈ I. More precisely, (x, y) ∈ R iff there exists a ﬁnite sequence i1 , . . . , ik ∈ I and

328 / a. furman

1 , . . . , k ∈ {−1, 1} so that

y = φi k ◦ · · · ◦ φi22 ◦ φi11 (x). k

We shall say that the family generates the relation R . The particular case of a collection = {φi } of Borel isomorphisms of the whole space X generates a countable group = and R = R X = (x, y) : x = y = (x, γ .x) : x ∈ X , γ ∈ . Feldman and Moore [39] proved that any countable Borel equivalence relation admits a generating set whose elements are deﬁned on all of X ; in other words, any equivalence relation appears as the orbit relation R X of a Borel action X of some countable group (see §4.3.1). Given a countable Borel equivalence relation R the full group [R] is deﬁned by [R] = φ ∈ Aut (X , X ) : ∀x ∈ X : (x, φ(x)) ∈ R . The full pseudogroup [[R]] consists of partially deﬁned Borel isomorphisms ψ : Dom(ψ) →Im(ψ), so that Graph(ψ)= (x, ψ(x)) : x ∈ Dom(ψ) ⊂ R. If R is the orbit relation R X of a group action (X , X ), then any φ ∈ [R] has the following “piecewise -structure”: there exist countable partitions

Ai = X = Bi into Borel sets and elements γi ∈ with γi (Ai ) = Bi so that φ(x) = γi x for x ∈ Ai . Elements ψ of the full pseudogroup [[R ]] have a

similar “piecewise -structure” with Ai = Dom(ψ) and Bi = Im(ψ). Let R be a countable Borel equivalence relation on a standard Borel space (X , X ). A measure μ on (X , X ) is R-invariant (respectively, R-quasiinvariant) if for all φ ∈ [R], φ∗ μ = μ (respectively, φ∗ μ ∼ μ). Note that if = {φi : Ai → Bi } is a generating set for R, then μ is R-invariant iff μ is invariant under each φi , that is, μ(φi−1 (E) ∩ Ai ) = μ(E ∩ Bi ) for all E ∈ X . Similarly, quasi-invariance of a measure can be tested on a generating set. The R-saturation of E ∈ X is R[E] = x ∈ X : ∃y ∈ E, (x, y) ∈ R . A R (quasi-) invariant measure μ is ergodic if R[E] is either μ-null or μ-conull for any E ∈ X . In this section we shall focus on countable Borel equivalence relations R on (X , X ) equipped with an ergodic, invariant, nonatomic, probability measure μ on (X , X ). Such a quadruple (X , X , μ, R) is called type II1 -relation. These are precisely the orbit relations of ergodic measure-preserving actions of countable groups on nonatomic standard probability measure spaces (the nontrivial implication follows from the above-mentioned theorem of Feldman and Moore).

a survey of measured group theory / 329

Given a countable Borel relation R on (X , X ) and an R-quasi-invariant probability measure μ, deﬁne inﬁnite measures μ˜ L , μ˜ R on R by # y : (x, y) ∈ E ∩ R dμ(x), μ˜ L (E) = X

μ˜ R (E) =

X

# x : (x, y) ∈ E ∩ R dμ(y).

These measures are equivalent, and coincide if μ is R-invariant, which is our main focus. In this case we shall denote μ˜ = μ˜ L = μ˜ R .

4.1

Hereafter, saying that some property holds a.e. on R would refer to μ-a.e. ˜ (this makes sense even if μ is only R-quasi-invariant).

remark 4.1. In some situations a Borel equivalence relation R on (X , X ) has only one (nonatomic) invariant probability measure. For example, this is the case for the orbit relation of the standard action of a ﬁnite-index subgroup8 < SLn (Z) on the torus Tn = R n /Zn , or for a lattice in a simple center-free Lie group G acting on H/, where H is a simple Lie group, < H is a lattice, and acts by left translations via an embedding j : G → H with j(G) having trivial centralizer in H. In such situations one may gain understanding of the countable Borel equivalence relation R via the study of the II1 -relation corresponding to the unique R-invariant probability measure. As always in the measure-theoretic setting null sets should be considered negligible. So an isomorphism T between (complete) measure spaces (Xi , Xi , μi ), i = 1, 2, is a Borel isomorphism between μi -conull sets T : X1 → X2 with T∗ (μ1 ) = μ2 . In the context of II1 -relations, we declare two relations (Xi , Xi , μi , Ri ), i = 1, 2 to be isomorphic, if there exists a measure space isomorphism T : (X1 , μ1 ) ∼ = (X2 , μ2 ) so that T × T : (R1 , μ˜ 1 ) → (R2 , μ˜ 2 ) is an isomorphism. In other words, after a restriction to conull sets, T satisﬁes (x, y) ∈ R1

⇐⇒

(T (x), T (y)) ∈ R2 .

Let us also adapt the notions of the full group and the full pseudogroup to the measure-theoretic setting, by passing to a quotient Aut (X , X ) → Aut (X , X , μ) where two Borel isomorphisms φ and φ that agree μ-a.e. are identiﬁed. This allows us to focus on the essential measure-theoretic issues. The following easy but useful Lemma illustrates the advantage of this framework. 8. Or just Zariski dense subgroup; see [18].

330 / a. furman

lemma 4.2. Let (X , X , μ, R) be a II1 -relation. Then for A, B ∈ X one has μ(φ(A)B) = 0 for some φ ∈ [R] iff μ(A) = μ(B). 4.1.1. restriction and weak isomorphisms Equivalence relations admit a natural operation of restriction, sometimes called induction, to a subset: given a relation R on X and a measurable subset A ⊂ X the restriction RA to A is RA = R ∩ (A × A). In the presence of, say, R-invariant, measure μ on (X , X ) the restriction to a subset A ⊂ X with μ(A) > 0 preserves the restricted measure μ|A , deﬁned by μ|A (E) = μ(A ∩ E). If μ is a probability measure, we shall denote by μA the normalized restriction μA = μ(A)−1 · μ|A . It is easy to see that ergodicity is preserved, so a restriction of a II1 -relation (X , μ, R) to a positive measure subset A ⊂ X is a II1 -relation (A, μA , RA ).

remark 4.3. Note that it follows from Lemma 4.2 that the isomorphism class of RA depends only on R and on the size μ(A), so RA may be denoted R t where t = μ(A) is 0 < t ≤ 1. One may also deﬁne R t for t > 1. For an integer k > 1 let R k denote the product of R with the full relation on the ﬁnite set {1, . . . , k}, namely the relation on X × {1, . . . , k} with ((x, i), (y, j)) ∈ R k iff (x, y) ∈ R. So (R k )1/k ∼ = R1 ∼ = R. The deﬁnition of R t can now be extended to all 0 < t < ∞ using an easily veriﬁed formula (R t )s ∼ = R ts . This construction is closely related to the notion of an ampliﬁcation in von Neumann algebras: the Murray von Neumann group-measure space construction MR satisﬁes MR t = (MR )t . The operation of restriction/induction allows one to relax the notion of isomorphism of II1 -relations as follows:

definition 4.4. Two II1 -relations R1 and R2 are weakly isomorphic if × R1 ∼ . Equivalently, there exist positive measurable = R2t for some t ∈ R+ subsets Ai ⊂ Xi with μ2 (A2 ) = t · μ1 (A1 ) and an isomorphism between the restrictions of Ri to Ai . Observe that two ergodic probability measure-preserving actions i (Xi , Xi , μi ) of countable groups are orbit equivalent iff the corresponding orbit relations Ri Xi are isomorphic.

a survey of measured group theory / 331

4.2. Invariants of Equivalence Relations

Let us now discuss in some detail several qualitative and numerical properties of II1 equivalence relations that are preserved under isomorphisms and often preserved or rescaled by the index under weak isomorphisms. We refer to such properties as invariants of equivalence relations. Many of these properties are motivated by properties of groups, and often an orbit relation R X of an essentially free action of a countable group would be a reﬂection of the corresponding property of .

4.2.1. amenability, strong ergodicity, and property (t) Amenability of an equivalence relation can be deﬁned in a number of ways. In [138] Zimmer introduced the notion of amenability for a group action on a space with quasiinvariant measure. This notion plays a central role in the theory. This deﬁnition is parallel to the ﬁxed-point characterization of amenability for groups. For equivalence relation R on (X , X ) with a quasi-invariant measure μ it reads as follows. Let E be a separable Banach space, and c : R → Isom (E) be a measurable 1-cocycle, that is, a measurable (with respect to the weak topology on E) map, satisfying μ-a.e.: ˜ c(x, z) = c(x, y) ◦ c(y, z). Let X x → Qx ⊂ E ∗ be a measurable family of nonempty convex compact subsets of the dual space E ∗ taken with the ∗-topology, so that c(x, y)∗ (Qx ) = Qy . The relation R is amenable if any such family contains a measurable invariant section, that is, a measurable assignment X x → p(x) ∈ Qx , so that a.e.: c(x, y)∗ p(x) = p(y). The (original) deﬁnition of amenability for group actions concerned general cocycles c : G × X → Isom (E) rather than the ones depending only on the orbit relation R X . The language of measured groupoids provides a common framework for both settings (see [9]). Any nonsingular action of an amenable group is amenable, because any cocycle c : × X → Isom (E) can be used to deﬁne an afﬁne -action on the closed convex subset of L∞ (X , E ∗ ) = L1 (X , E)∗ consisting of all measurable sections x → p(x) ∈ Qx ; the ﬁxed-point property of provides the desired c ∗ invariant section. The converse is not true: any (countable or lcsc) group admits essentially free amenable action with a quasi-invariant measure—this is the main use of the notion of amenable actions. However, for essentially free,

332 / a. furman

probability measure-preserving actions, amenability of the II1 -relation R X implies (hence is equivalent to) amenability of . Indeed, given an afﬁne action α on a convex compact Q ⊂ E ∗ , one can take Qx = Q for all x ∈ X and set c(gx, x) = α(g); amenability of R X provides an invariant section p : X → Q whose barycenter q = X p(x) dμ(x) would be an α()-ﬁxed point in Q . Connes, Feldman, and Weiss [27] proved that amenable relations are hyperﬁnite in the sense that they can be viewed as an increasing union of ﬁnite subrelations; they also showed that such a relation can be generated by an action of Z (see also [78] by Kaimanovich for a nice exposition and several other nice characterizations of amenability). It follows that there is only one amenable II1 -relation, which we denote hereafter by Ramen . In [141] Zimmer introduced the notion of property (T) for group actions on measure spaces with quasi-invariant measure. The equivalence relation version can be stated as follows. Let H be a separable Hilbert space and let c : R → U(H ) be a measurable 1-cocycle, that is, c satisﬁes c(x, z) = c(x, y) ◦ c(y, z). Then R has property (T) if any such cocycle for which there exists a sequence vn : X → S(H ) of measurable maps into the unit sphere S(H ) with vn (y) − c(x, y)vn → 0

[μ]-a.e. ˜

admits a measurable map u : X → S(H ) with u(y) = c(x, y)u(x) for μ-a.e. ˜ (x, y) ∈ R. For an essentially free probability measure-preserving action (X , μ) the orbit relation R X has property (T) if and only if the group has Kazhdan’s property (T) (in [141] weak mixing of the action was assumed for the “only if ” implication, but this can be removed as in §3.1.1 relying on Bekka - Valette [16]). In [8] Anantharaman-Delaroche studied the notion of property (T) in the context of general measured groupoids. Let R be a II1 -equivalence relation on (X , μ). A sequence {An } of measurable subsets of X is asymptotically R-invariant, if μ(φ(An )An ) → 0 for every φ ∈ [R]. This is satisﬁed trivially if μ(An ) · (1 − μ(An )) → 0. Relation R is strongly ergodic if any asymptotically R-invariant sequence of sets is trivial in the above sense. (Note that the condition of asymptotic invariance may be checked on elements φi of any generating system of R.) The amenable relation Ramen is not strongly ergodic. If an action (X , μ) has a spectral gap (i.e., does not have almost-invariant vectors) in the Koopman

a survey of measured group theory / 333

representation on L2 (X , μ) ! C, then R X is strongly ergodic. Using the fact that the Koopman representation of a Bernoulli action (X0 , μ0 ) is contained in a multiple of the regular representation ∞ · 2 (), Schmidt [128] characterized nonamenable groups by the property that they admit p.m.p. actions with strongly ergodic orbit relation. If R is not strongly ergodic, then it has an amenable relation as a nonsingular quotient (Jones and Schmidt [77]). Connes and Weiss [28] showed that all p.m.p. actions of a group have strongly ergodic orbit relations if and only if has Kazhdan’s property (T). In this short elegant paper they introduced the idea of Gaussian actions as a way of constructing a p.m.p. action from a given unitary representation. In general, strong ergodicity of the orbit relation R X does not imply a spectral gap for the action (X , μ) ([128], [70]). However, this implication does hold for generalized Bernoulli actions (Kechris and Tsankov [82]), and when the action has an ergodic centralizer (Chifan and Ioana [25, lemma 10]).

4.2.2. fundamental group-index values of self-similarity The term × “fundamental group” of a II1 -relation R refers to a subgroup of R+ deﬁned by × : R∼ F (R) = t ∈ R+ = Rt . Equivalently, for R on (X , μ), the fundamental group F (R) consists of all ratios μ(A)/μ(B) where A, B ⊂ X are positive measure subsets with RA ∼ = RB (here one can take one of the sets to be X without loss of generality). The notion is borrowed from a similarly deﬁned concept of the fundamental group of a von Neumann algebra, introduced by Murray and von Neumann × [104]: F (M) = t ∈ R+ : Mt ∼ = M . However, the connection is not direct: even for group space construction M = L∞ (X ) isomorphisms M ∼ = Mt (or even automorphisms of M) need not respect the Cartan subalgebra L∞ (X ) in general. Since the restriction of the amenable relation Ramen to any positive measure subset A ⊂ X is amenable, it follows × . F (Ramen ) = R+

The same obviously applies to the product of any relation with an amenable one. On another extreme are orbit relations R X of essentially free ergodic action of ICC groups with property (T): for such relations the fundamental group F (R X ) is at most countable (Gefter and Golodets [64, Corollary 1.8]).

334 / a. furman

Many relations have a trivial fundamental group. This includes all II1 relations with a nontrivial numeric invariant that scales under restriction: 1) Relations with 1 < cost (R) < ∞; in particular, orbit relation R X for essentially free actions of Fn , 1 < n < ∞, or surface groups. (2) 2) Relations with some nontrivial 2 -Betti number 0 < βi (R) < ∞ for some i ∈ N; in particular, orbit relation R X for essentially free actions (2) of a group with 0 < βi () < ∞ for some i ∈ N, such as lattices in SO2n,1 (R), SUm,1 (R), and Spk,1 (R). Triviality of the fundamental group often appears as a by-product of rigidity of groups and group actions. For example, F (R X ) = {1} in the following situations: 1) Any (essentially free) action of a lattice in a simple Lie group of higher rank ([64]); 2) Any essentially free action of (ﬁnite-index subgroups of products) mapping class groups ([86]); 3) Actions of = 1 × · · · × n , n ≥ 2, of hyperboliclike groups i where each of them acts ergodically ([101]); and 4) Gdsc -cocycle superrigid actions X such as Bernoulli actions of groups with property (T) ([111, 112, 117]). What are other possibilities for the fundamental group beyond the two × and F (R) = {1}? The most comprehensive extreme cases F (R) = R+ answer (to date) to this question is contained in the following result of S. Popa and S. Vaes (see [118] for further references):

theorem 4.5. (popa-vaes, [118, thm 1.1]) There exists a family S of additive subgroups of R that contains all countable groups, and (uncountable) groups of arbitrary Hausdorff dimension in (0, 1), so that for any F ∈ S and any totally disconnected locally compact unimodular group G there exists uncountably many mutually non-SOE essentially free p.m.p. actions of F∞ whose orbit relations R = RF∞ X have F (R) ∼ = G. = exp (F) and Out (R) ∼ Moreover, in these examples the Murray von Neumann group space factor M = L∞ (X ) has F (M) ∼ = Out (R) = exp (F) and Out (M) ∼ = F (R) ∼ H 1 (R, T), where H 1 (R, T) is the ﬁrst cohomology with coefﬁcients in the 1-torus. 4.2.3. treeability An equivalence relation R is said treeable (Adams [4]) if it admits a generating set = {φi } so that the corresponding (nonoriented)

a survey of measured group theory / 335

graph on a.e. R-class is a tree. Basic examples of treeable relations include: Ramen viewing the amenable II1 -relation as the orbit relation of some/any action of Z = F1 , and more generally, RFn X where Fn X is an essentially free action of the free group Fn , 1 ≤ n ≤ ∞. Any restriction of a treeable relation is treeable, and R is treeable iff R t is. If R1 → R2 is a (weak) injective relation morphism and R2 is treeable, then so is R1 —the idea is to lift a treeing graphing from R2 to R1 piece by piece. This way, one shows that if a group admits an essentially free action Z with treeable RZ , and and admit (S)OE essentially free actions X and Y , then the -action on X × Z, g : (x, z) → (gx, α(g, x)z) via the (S)OE cocycle α : × X → has a treeable orbit structure R X ×Z . Since surface groups = π1 ( g ), g ≥ 2, and F2 are lattices in PSL2 (R), hence ME, the former groups have free actions with treeable orbit relations. Are all orbit relations of free actions of a surface group treeable?

4.2.4. cost The notion of cost for II1 -relations corresponds to the notion of rank for discrete countable groups. The notion of cost was introduced by G. Levitt [90] and extensively studied by D. Gaboriau [53, 54, 60]. definition 4.6. Given a generating system = {φi : Ai → Bi }i∈N for a II1 -equivalence relation R on (X , μ) the cost of the graphing is cost ( ) =

μ(Ai ) =

i

μ(Bi )

i

and the cost of the relation is cost (R) = inf cost ( ) : generates R . A generating system deﬁnes a graph structure on every R-class and cost ( ) is half of the average valency of this graph over the space (X , μ). The cost of a II1 -relation takes values in [1, ∞]. In the deﬁnition of the cost of a relation it is important that the relation is probability measure preserving, but ergodicity is not essential. The broader context includes relations with ﬁnite classes; such relations can have values less than 1. For instance, from the orbit relation of a (nonergodic) probability measure-preserving action of a ﬁnite group (X , μ) one gets cost (R X ) = 1 −

1 . ||

336 / a. furman

If R is the orbit relation of some (not necessarily free) action (X , μ), then cost (R) ≤ rank(), where the latter stands for the minimal number of generators for . Indeed, any generating set {g1 , . . . , gk } for gives a generating system = {γi : X → X }ki=1 for R X . Recall that the amenable II1 -relation Ramen can be generated by (any) action of Z. Hence cost (Ramen ) = 1. The cost behaves nicely with respect to restriction:

theorem 4.7. (gaboriau [54]) For a II1 -relation R: t · ( cost (R t ) − 1) = cost (R) − 1

× (t ∈ R+ ).

The following is a key tool for computations of the cost:

theorem 4.8. (gaboriau [54]) Let R be a treeable equivalence relation, and be a graphing of R giving a tree structure to R-classes. Then cost (R) = cost ( ). Conversely, for a relation R with cost (R) < ∞, if the cost is attained by some graphing , then is a treeing of R. The above result (the ﬁrst part) implies that for any essentially free action Fn (X , μ) one has cost (RFn X ) = n. This allowed Gaboriau to prove the following fact, answering a long-standing question:

corollary 4.9. (gaboriau [53], [54]) If essentially free probability measurepreserving actions of Fn and Fm are orbit equivalent, then n = m. Note that Fn and Fm are commensurable for 2 ≤ n, m < ∞, hence they have essentially free actions that are weakly isomorphic. The index of such n−1 weak isomorphism will necessarily be m−1 , or m−1 n−1 (these free groups have P.F.-ﬁxed price). It should be pointed out that one of the major open problems in the theory of von Neumann algebras is whether it is possible for the factors L(Fn ) and L(Fm ) to be isomorphic for n = m (it is known that either all L(Fn ), 2 ≤ n < ∞, are isomorphic, or all distinct). The following powerful result of Greg Hjorth provides a link from treeable relations back to actions of free groups:

a survey of measured group theory / 337

theorem 4.10. (hjorth [69]) Let R be a treeable equivalence relation with n = cost (R) in {1, 2, . . . , ∞}. Then R can be generated by an essentially free action of Fn . The point of the proof is to show that a relation R that has a treeing graphing with average valency 2n admits a (treeing) graphing with a.e. constant valency 2n. The behavior of the cost under passing to a subrelation of ﬁnite index is quite subtle—the following question is still open (to the best of the author’s knowledge).

question 4.11. (gaboriau) Let be a subgroup of ﬁnite index in , and (X , μ) be an essentially free p.m.p. action. Is it true that the costs of the orbit relations of and are related by the index [ : ] cost (R X ) − 1 = [ : ] · ( cost (R X ) − 1)? In general has at most [ : ]-many ergodic components. The extreme case where the number of -ergodic components is maximal: [ : ] corresponds to (X , μ) being a co-induction from an ergodic -action. In this case the above formula easily holds. The real question lies in the other extreme where is ergodic. Recall that the notion of the cost is analogous to the notion of rank for groups, where rank () = inf n ∈ N : ∃ epimorphism Fn → . Schreier’s theorem states that for n ∈ N any subgroup F < Fn of ﬁnite-index [Fn : F] = k is itself free: F ∼ = Fk(n−1)+1 . This implies that for any ﬁnitely generated and any ﬁnite-index subgroup of < one has rank ( ) − 1 ≤ [ : ] · ( rank () − 1) with equality in the case of free groups. Let > 1 > . . . be a chain of subgroups of ﬁnite index. One deﬁnes the rank gradient (Lackenby [89]) of the chain {n } as the limit of the monotonic (!) sequence: RG (, {n }) = lim

n→∞

rank (n ) − 1 . [ : n ]

It is an intriguing question whether (or when) is it true that RG (, {n }) depends only on and not on a particular chain of ﬁnite-index subgroups. One should, of course, assume that the chains in question have trivial intersection, and one might require the chains to consist of normal subgroups in the original group. In the case of free groups RG is indeed independent of the chain.

338 / a. furman

In [1] Abert and Nikolov prove that the rank gradient of a chain of ﬁniteindex subgroups of is given by the cost of a certain associated ergodic p.m.p. -action. Let us formulate a special case of this relation where the chain {n } consists of normal subgroups n with n = {1}. Let K = lim / n ←− denote the proﬁnite completion corresponding to the chain. The -action by left translations on the compact totally disconnected group K preserves the Haar measure mK and (K , mK ) is a free ergodic p.m.p. action. (Let us point out in passing that this action has a spectral gap, implying strong ergodicity, iff the chain has property (τ ) introduced by Lubotzky and Zimmer [91]).

theorem 4.12. (abert-nikolov [1]) With the above notations: RG (, {n }) = cost (R K ) − 1. One direction (≥) is easy to explain. Let Kn be the closure of n in K . Then Kn is an open normal subgroup of K of index m = [ : n ]. Let 1 = g1 , g2 , . . . , gn ∈ be representatives of n -cosets, and h1 , . . . , hk generators of n with k = rank (n ). Consider the graphing = {φ2 , . . . , φm , ψ1 , . . . , ψk }, where φi : Kn → gi Kn are restrictions of gi (2 ≤ i ≤ m), and ψj : Kn → Kn are restrictions of hj (1 ≤ j ≤ k). These maps are easily seen to generate R K , with the cost of cost ( ) = k · mK (Kn ) + (m − 1) · mK (Kn ) =

k−1 rank (n ) − 1 +1 = 1+ . m [ : n ]

Abert and Nikolov observed that a positive answer to Question 4.11 combined with the above result shows that RG () is independent of the choice of a (normal) chain, and therefore is a numeric invariant associated to any residually ﬁnite ﬁnitely generated group. Surprisingly, this turns out to be related to a problem in the theory of compact hyperbolic 3-manifolds concerning rank versus Heegard genus [89]—see [1] for the connection and further discussions. The above result has an application, independent of Question 4.11. Since amenable groups have P.F. with C = 1, it follows that a ﬁnitely generated, residually ﬁnite amenable group has sublinear rank growth for ﬁnite-index normal subgroups with trivial intersection, that is, RG () = 0 for any such chain.

4.2.5. 2 -betti numbers I have already mentioned the 2 -Betti numbers (2) βi () associated with a discrete group and Gaboriau’s proportionality result

a survey of measured group theory / 339

in Theorem 3.8 for measure equivalence between groups. In fact, rather than relating the 2 -Betti numbers of groups via ME, in [57] Gaboriau (2)

- deﬁnes the notion of 2 -Betti numbers βi (R) for a II1 -equivalence relation R; (2) (2) - proves that βi () = βi (R X ) for essentially free ergodic action (X , μ); and (2) (2) - observes that βi (R t ) = t · βi (R) for any II1 -relation. (2)

(2)

The deﬁnition of βi (R) is inspired by the deﬁnition of βi () by Cheeger and Gromov [23]: it uses R-action (or groupoid action) on pointed contractible simplicial complexes, corresponding complexes of Hilbert modules with Raction, and von Neumann dimension with respect to the algebra MR . In the late 1990s Wolgang Lück developed an algebraic notion of dimension for arbitrary modules over von Neumann algebras, in particular giving an alternative approach to 2 -Betti numbers for groups (see [92]). In [124] Roman Sauer used Lück’s notion of dimension to deﬁne 2 -Betti numbers of equivalence relations, and more general measured groupoids, providing an alternative approach to Gaboriau’s results. In [127] Sauer and Thom develop further homological tools (including a spectral sequence associated to strongly normal subrelations) to study 2 -Betti numbers for groups, relations, and measured groupoids.

4.2.6. outer automorphism group Given an equivalence relation R on (X , μ) deﬁne the corresponding automorphism group as the group of selfisomorphisms: Aut (R) = T ∈ Aut (X , μ) : T × T (R) = R (modulo null sets) . The subgroup Inn (R) of inner automorphisms is Inn (R) = T ∈ Aut (X , μ) : (x, T (x)) ∈ R for a.e. x ∈ X . This is just the full group [R], but the above notation emphasizes the fact that it is normal in Aut (R) and suggests to consider the outer automorphism group Out (R) = Aut (R)/ Inn (R). One might think of Out (R) as the group of all measurable permutations of the R-classes on X . Recall (Lemma 4.2) that Inn (R) is a huge group as it acts transitively on (classes mod null sets of) measurable subsets of any given size

340 / a. furman

in X . Yet the quotient Out (R) might be small (even ﬁnite or trivial), and can sometimes be explicitly computed.

remark 4.13. As an abstract group H = Inn (R) is simple, and its automorphisms come from automorphisms of R; in particular Out (H) = Out (R). Moreover, Dye’s reconstruction theorem states that (the isomorphism type of) R is determined by the structure of Inn (R) as an abstract group (see [80, §I.4] for proofs and further facts). Let us also note that the operation of restriction/ampliﬁcation of the relation does not alter the outer automorphism group (cf. [47, lemma 2.2]): Out (R t ) ∼ = Out (R)

× (t ∈ R+ ).

The group Aut (R) has a natural structure of a Polish group [62, 64]. First, recall that if (Y , ν) is a ﬁnite- or inﬁnite-measure Lebesgue space, then Aut (Y , ν) is a Polish group with respect to the weak topology induced from the weak (=strong) operator topology of the unitary group of L2 (Y , ν). This deﬁnes a Polish topology on Aut (R) when the latter is viewed as acting on the inﬁnite-measure space (R, μ). ˜ 9 However, Inn (R) is not always closed in Aut (R), so the topology on Out (R) might be complicated. Alexander Kechris recently found the following surprising connection:

theorem 4.14. (kechris [80, theorem 8.1]) If Out (R)fails to be a Polish group, then cost (R) = 1. Now assume that R can be presented as the orbit relation of an essentially free action (X , μ), so Aut (R) is the group of self-orbit equivalences of X . The centralizer Aut (X , μ) of in Aut (X , μ) embeds in Aut (R) and if is ICC (i.e., has inﬁnite conjugacy classes), then the quotient map out

Aut (R)−→ Out (R) is injective on Aut (X , μ) (cf. [62, lemma 2.6]). So Out (R) has a copy of Aut (X , μ), and the latter might be very big. For example, in the Bernoulli action (X , μ) = (X0 , μ0 ) , it contains Aut (X0 , μ0 ) acting diagonally on the factors. Yet, if has property (T), then Aut (X , μ) · Inn (R X ) is open in the Polish group Aut (R X ). In this case the image of Aut (X , μ) has ﬁnite or countable index in Out (R X ). This fact was observed by Gefter and Golodets in [64, §2], and can be deduced from Proposition 5.14. 9. This topology coincides with the restriction to Aut (R) of the uniform topology on Aut (X , μ) given by the metric d(T , S) = μ x ∈ X : T (x) = S(x) . On all of Aut (X , μ) the uniform topology is complete but not separable; but its restriction to Aut (R) is separable.

a survey of measured group theory / 341

To get a handle on Out (R X ) one looks at OE-cocycles cT : × X → corresponding to elements T ∈ Aut (R X ). It is not difﬁcult to see that cT is conjugate in to the identity (i.e., cT (g, x) = f (gx)−1 gf (x) for f : X → ) iff T is in Aut (X , μ) · Inn (R). Thus, starting from a group or action X with strong rigidity properties for cocycles, one controls Out (R X ) via Aut (X , μ). This general scheme (somewhat implicitly) is behind the ﬁrst example of an equivalence relation with trivial Out (R) constructed by Gefter [61, 62]. Here is a variant of this construction:

theorem 4.15. Let be a torsion-free group with property (T), K a compact connected Lie group without outer automorphisms, and τ : → K a dense embedding. Let L < K be a closed subgroup and consider the ergodic actions (K , mK ) and (K /L, mK /L ) by left translations. Then Out (R K ) ∼ = K,

Out (R K /L ) ∼ = NK (L)/L.

∼ POn−1 (R) < K to be the stabilizer In particular, taking K = POn (R) and L = n of a line in R , the space K /L is the projective space P n−1 , and we get Out (R P n−1 ) = {1} for any property (T) dense subgroup < POn (R). Such a group exists iff n ≥ 5, Zimmer [145, theorem 7]. The preceding discussion, combined with the cocycles superrigidity Theorem 5.21 below, and an easy observation that Aut (K /L, mK /L ) is naturally isomorphic to NK (L)/L, provide a self contained sketch of the proof of the theorem. In the above statement Out (K ) is assumed to be trivial and to be torsion free just to simplify the statement. However, the assumption that K is connected is essential. Indeed, the dense embedding of = PSLn (Z) in the compact proﬁnite group K = PSLn (Zp ) where p is a prime, gives Out (RPSLn (Z)PSLn (Zp ) ) ∼ = PSLn (Qp ) Z/2Z where the Z/2-extension is given by the transpose g → g tr . The inclusion ⊃ was found in [62], and the equality is proved in [47, theorem 1.6], where many other computations of Out (R X ) are carried out for actions of lattices in higher-rank Lie groups. Finally, we recall that the recent preprint [117] of Popa and Vaes quoted above (Theorem 4.5) shows that an arbitrary totally disconnected lcsc group G can arise as Out (R X ) for an essentially free action of a free group F∞ .

342 / a. furman

4.2.7. cohomology Equivalence relations have groups of cohomology associated to them similar to cohomology of groups. These were introduced by Singer [134] and largely emphasized by Feldman and Moore [40]. Given, say, a type II1 equivalence relation R on (X , μ) consider R (n) = (x0 , . . . , xn ) ∈ X n+1 : (xi , xi+1 ) ∈ R equipped with the inﬁnite Lebesgue measure μ˜ (n) deﬁned by (n) μ˜ (A) = # (x1 , . . . , xn ) : (x0 , . . . , xn ) ∈ R (n) dμ(x0 ). X

(R (0) , μ˜ 0 )

to be (X , μ). Note that (R (1) , μ(1) ) is just (R, μ) ˜ from §4.1. Take Since μ is assumed to be R-invariant, the above formula is invariant under permutations of x0 , . . . , xn . Fix a Polish abelian group A written multiplicatively (usually A = T). The space C n (R, A) of n-cochains consists of equivalence classes (modulo μ˜ (n) -null sets) of measurable maps R (n) → A, linked by the operators dn : C n (R, A) → C n+1 (R, A) dn ( f )(x0 , . . . , xn+1 ) =

n+1

i

f (x0 , . . . , xˆ i , . . . , x0 )(−1) .

i=0

= Ker(dn ) the n-cocycles, and Bn (R) = Im(dn−1 ) the nCall coboundaries; the cohomology groups are deﬁned by H n (R) = Z n (R)/Bn (R). In degree n = 1 the 1-cocycles are measurable maps c : (R, μ) → A such that Z n (R)

c(x, y)c(y, x) = c(x, z) and 1-coboundaries have the form b(x, y) = f (x)/f (y) for some measurable f : X → A. If A is a compact abelian group, such as T, then C 1 (R, A) is a Polish group (with respect to convergence in measure). Being a closed subgroup in C 1 (R, A), the 1-cocycles Z 1 (R, A) form a Polish group. Schmidt [129] showed that B1 (R, A) is closed in Z 1 (R, A) iff R is strongly ergodic. There are only few cases where H 1 (R, T) were computed: C.C. Moore [102] constructed a relation with trivial H 1 (R, T). Gefter [63] considered H 1 (R G , T) for actions of property (T) group densely embedded in a semisimple Lie group G. More recently Popa and Sasyk [116] studied H 1 (R X , T) for property (T) groups with Bernoulli actions (X , μ) = (X0 , μ0 ) . In both cases H 1 (R X , T) is shown to coincide with the group of characters Hom(, T). Higher cohomology groups remain mysterious.

a survey of measured group theory / 343

The fact that A is abelian is essential to the deﬁnition of H n (R, A) for n > 1. However, in degree n = 1 one can deﬁne H 1 (R, ) as a set for a general target group 10 . In fact, this notion is commonly used in this theory under the name of measurable cocycles (see Appendix A and §5.1 below). For the deﬁnition in terms of equivalence relations let Z 1 (R, ) denote the set of all measurable maps (mod μ-null ˜ sets) c : (R, μ) ˜ →

s.t.

c(x, z) = c(x, y)c(y, z)

and let H 1 (R, ) = Z 1 (R, )/ ∼ where the equivalence ∼ between c, c ∈ Z 1 (R, ) is declared if c(x, y) = f (x)−1 c (x, y)f (y) for some measurable f : (X , μ) → . If R = R X is the orbit relation of an essentially free action, then Z 1 (R X , ) coincides with the set of measurable cocycles α : × X → by α(g, x) = c(x, gx). Note that Hom(, )/ maps into H 1 (R, ), via cπ (x, y) = π(g) for the unique g ∈ with x = gy. The point of cocycles superrigidity theorems is to show that under favorable conditions this map is surjective.

4.2.8. free decompositions Group theoretic notions such as free products, amalgamated products, and HNN-extensions can be deﬁned in the context of equivalence relations—see Gaboriau [54, section IV]. For example, a II1 relation R is said to split as a free product of subrelations {Ri }i∈I , denoted R = ∗i∈I Ri , if 1) R is generated by {Ri }i∈I , that is, R is the smallest equivalence relation containing the latter family; and 2) almost every chain x = x0 , . . . , xn = y, where xj−1 = xj , (xj−1 , xj ) ∈ Ri(j) and i(j + 1) = i(j), has x = y. If S is yet another subrelation, one says that R splits as a free product of Ri amalgamated over S , R = ∗S Ri , if in condition (2) one replaces xj−1 = xj by (xj−1 , xj ) ∈ S . The obvious example of the situation above is an essentially free action of a free product of groups 3 = 1 ∗ 2 (respectively amalgamated product 5 = 1 ∗4 2 ) on a probability space (X , μ); in this case the orbit relations Ri = Ri X satisfy R3 = R1 ∗ R2 (respectively R5 = R1 ∗R4 R2 ). Another useful construction (see Ioana, Peterson, Popa [75]) is as follows. Given measure preserving (possibly ergodic) relations R1 , R2 on a probability 10. In order to deﬁne the notion of measurability should have a Borel structure, and better be a Polish group; often it is a discrete countable group, or a Lie group.

344 / a. furman

space (X , μ) for T ∈ Aut (X , μ) consider the relation generated by R1 and T (R2 ). It can be shown that for a residual set of T ∈ Aut (X , μ) the resulting relation is a free product of R1 and T (R2 ). A similar construction can be carried out for amalgamated products. Note that in contrast with the category of groups the isomorphism type of the free product is not determined by the free factors alone. Ioana, Peterson, and Popa [75] obtained strong rigidity results for free and amalgamated products of ergodic measured equivalence relations and II1 -factors with certain rigidity properties. Here let us describe some results obtained by Alvarez and Gaboriau [7], which are easier to state; they may be viewed as an analogue of Bass-Serre theory in the context of equivalence relations. Say that a II1 -relation R is freely indecomposable (FI) if R is not a free product of its subrelations. A group is said to be measurably freely indecomposable (MFI) if all its essentially free action give freely indecomposable orbit relations. A group may fail to be MFI even if it is freely indecomposable in the group theoretic sense (surface groups provide an example). Not surprisingly, groups with property (T) are MFI (cf. Adams Spatzier [6]); but more generally (2)

theorem 4.16. (alvarez-gaboriau [7]) If is nonamenable and β1 () = 0, then is MFI. theorem 4.17. (alvarez-gaboriau [7]) Let I, J be two ﬁnite or countableindex sets, {i }i∈I and {j }j∈J be two families of MFI groups, = ∗i∈I i , = ∗j∈J j , and (X , μ), (Y , ν) be essentially free p.m.p. actions SOE

where each i (X , μ) and j (Y , ν) are ergodic. Assume that X ∼ Y. SOE Then |I| = |J| and there is a bijection θ : I → J so that i X ∼ θ(i) Y .

The assumption that each free factor is ergodic is important here; Alvarez and Gaboriau also give an analysis of the general situation (where this assumption is dropped). 4.3. Rigidity of Equivalence Relations

The close relation between ME and SOE allows us to deduce that certain orbit relations R X remember the acting group and the action (X , μ) up to isomorphism, or up to a virtual isomorphism. This slightly technical concept is described in the following:

a survey of measured group theory / 345

lemma 4.18. Suppose an ergodic ME-coupling (, m) of with corresponds SOE to an SOE between ergodic actions T : (X , μ) ∼ (Y , ν). Then the following are equivalent: 1) There exist short exact sequences 1 → 0 → → 1 → 1,

1 → 0 → → 1 → 1

where 0 and 0 are ﬁnite, a discrete (1 , 1 )-coupling (1 , m1 ), and an equivariant map : (, m) → (1 , m1 ); and 2)There exist isomorphism between ﬁnite-index subgroups 1 > 2 ∼ = 2 < 1 , so that 1 X1 = X / 0 and 1 Y1 = Y /0 are induced from some isomorphic ergodic actions 2 X2 ∼ = 2 Y2 . 3)The SOE (or ME) cocycle × X → is conjugate in to a cocycle whose restriction to some ﬁnite-index subgroup 1 is a homomorphism 1 → (the image is necessarily of ﬁnite index). Let us now state two general forms of relation rigidity. Here is one form

theorem 4.19. Let (X , μ) be an ergodic essentially free action of one of the types below, an arbitrary group, and (Y , ν) as essentially free p.m.p. action whose orbit relation RY is weakly isomorphic to R X . Then is commensurable up to ﬁnite kernels to and the actions X and Y are virtually isomorphic; in particular, the SEO-index is necessarily rational. The list of actions X with this SOE-rigidity property includes: 1) is a lattice in a connected, center-free, simple Lie group G of higher rank, and X has no equivariant quotients of the form G/ where < G is a lattice ([45, theorem A]); 2) = 1 × · · · × n where n ≥ 2, i ∈ Creg , and i (X , μ) are ergodic; in addition assume that (Y , ν) is mildly mixing (Monod-Shalom [101]); and 3) is a ﬁnite-index subgroup in a (product of) mapping class groups as in Theorem 3.20 (Kida [85]). a) For a concrete example for (1)–(3) one might take Bernoulli actions (X0 , μ0 ) . In (1) one might also take SLn (Z) Tn or SLn (Z) SLn (Zp ) with n ≥ 3. In (2) one might look at Fn × Fm acting on a compact Lie group K , for

346 / a. furman

example SO3 (R), by (g, h) : k → gkh −1 where Fn , Fm are embedded densely in K . b) In (1) the assumption that there are no -equivariant quotient maps X → G/ is necessary, since given such a quotient there is a -action on SOE

some (X , μ ) with X ∼ X . The rigidity statement in this case is that this is a complete list of groups and their essentially free actions up to virtual isomorphism ([45, theorem C]). The appearance of these factors has to do with (G, mG ) appearing as a quotient of a (, )-coupling. c) The basic technique for establishing the stated rigidity in cases (1)–(3) is to establish condition (1) in Lemma 4.18. This is done by analyzing a self-ˇ (where X = / and Y = / ) and invoking coupling of the form × an analogue of the construction in §5.5. d) In all cases one can sharpen the results (eliminate the “virtual”) by imposing some benign additional assumptions: rule out torsion in the acting groups, and impose ergodicity for actions of ﬁnite-index subgroups. The second stronger form of relation rigidity refers to rigidity of relation morphisms that are obtained from Gdsc -cocycle superrigid actions discovered by Sorin Popa (see §5.4). We illustrate this framework by the following particular statement (see [113, theorem 0.4] and [48, theorem 1.8]).

theorem 4.20. Let (X , μ) be a mixing Gdsc -cocycle superrigid action, such as: 1) A Bernoulli -action on (X0 , μ0 ) , where has property (T), or = 1 × 2 with 1 nonamenable and 2 being inﬁnite; and 2) K /L where → K is a homomorphism with dense image in a simple compact Lie group K with trivial π1 (K ), L < K is a closed subgroup, and has (T). Let be some group with an ergodic essentially free measure-preserving action (Y , ν),11 X ⊂ X a positive measure subset, and T : X → Y a measurable map with T∗ μ ≺ ν and (x1 , x2 ) ∈ R X ∩ (X × X )

=⇒

(T (x1 ), T (x2 )) ∈ RY .

Then there exists ρ

• an exact sequence 0 −→ −→1 with ﬁnite 0 and 1 < ; and • a 1 -ergodic subset Y1 ⊂ Y with 0 < ν(Y1 ) < ∞; and 11. The space (Y , ν) might be ﬁnite- or inﬁnite-measure Lebesgue space.

a survey of measured group theory / 347 • denoting (X1 , μ1 ) = (X , μ)/ 0 and ν1 = ν(Y1 )−1 · ν|Y , there is an isomor1

phism T1 : (X1 , μ1 ) ∼ = (Y1 , ν1 ) of 1 -actions.

Moreover, μ-a.e. T (x) and T1 (0 x) are in the same -orbit.

4.3.1. a question of feldman and moore Feldman and Moore showed [39] that any countable Borel equivalence relation R can be generated by a Borel action of a countable group. They asked whether one can ﬁnd a free action of some group, so that R-classes would be in one-to-one correspondence with the acting group. This question was answered in the negative by Adams [3]. In the context of measured relations, say of type II1 , the question is whether it is possible to generate R (up to null sets) by an essentially free action of some group. This question was also settled in the negative in [45, theorem D], using the following basic constructions: 1) Start with an essentially free action (X , μ) that is rigid as in Theorem 4.19 or 4.20, and let R = (R X )t with an irrational t. 2) Consider a proper embedding G → H of higher-rank simple Lie groups and choose a lattice < H, say G = SL3 (R) ⊂ H = SL4 (R) with = SL4 (Z). Such actions always admit a Borel cross-section X ⊂ H/ for the G-action, equipped with a holonomy-invariant probability measure μ. Take R on (X , μ) to be the relation of being in the same G-orbit. In case (1) one argues as follows: if some group has an essentially free action (Y , ν) with (R X )t = R ∼ = RY , then the rigidity implies that and are commensurable up to ﬁnite kernel, and X is virtually isomorphic to Y . But this would imply that the index t is rational, contrary to the assumption. This strategy can be carried out in other cases of very rigid actions as in [72, 85, 101, 113]. Theorem 5.21 provides an example of this type R = (R K )t where is a Kazhdan group densely embedded in a compact connected Lie group K . So the reader has a sketch of the full proof for a II1 -relation that cannot be generated by an essentially free action of any group. Example of type (2) was introduced by Zimmer in [147], where it was proved that the relation R on such a cross-section cannot be essentially freely generated by a group , which admits a linear representation with an inﬁnite image. The linearity assumption was removed in [45]. This example is particularly interesting since it cannot be “repaired” by restriction/ampliﬁcation; as any R t can be realized as a cross-section of the same G-ﬂow on H/ .

question 4.21. (vershik) Let R on (X , μ) be a II1 -relation that cannot be generated by an essentially free action of a group; and let (X , μ) be some action

348 / a. furman

producing R. One may assume that the action is faithful, that is, → Aut (X , μ) is an embedding. What can be said about and the structure of the measurable family {x }x∈X of the stabilizers of points in X ? In [119] S. Popa and S. Vaes give an example of a II1 -relation R (which is a restriction of the II∞ -relation RSL5 (Z)R5 to a subset A ⊂ R 5 of positive ﬁnite measure), which has property (T) but cannot be generated by an action (not necessarily free) of any group with property (T).

5. Techniques 5.1. Superrigidity in Semisimple Lie Groups

The term “superrigidity” refers to a number of phenomena originated and inspired by the following celebrated discovery of G. A. Margulis.

theorem 5.1. (margulis [93]) Let G and G be (semi)simple connected real center-free Lie groups without compact factors with rk(G) ≥ 2, < G be an irreducible lattice, and π : → G a homomorphism with π() being Zariski dense in G and not precompact. Then π extends to a (rational) epimorphism π¯ : G → G . The actual result is more general than stated, as it applies to products of semisimple algebraic groups over general local ﬁelds. We refer the reader to the comprehensive monograph (Margulis [95]) for the general statements, proofs, and further results and applications, including the famous arithmeticity theorem. The core of (some of the available) proofs of Margulis’s superrigidity theorem is a combination of the theory of algebraic groups and purely ergodictheoretic arguments. The result applies to uniform and nonuniform lattices alike, and it also covers irreducible lattices in higher-rank Lie groups such as SL2 (R) × SL2 (R). Let us also note that the assumption that π() is not precompact in G is redundant if π () is Zariski dense in a real Lie group G (since compact groups over R are algebraic), but is important in general (cf. SLn (Z) < SLn (Qp ) is Zariski dense but precompact). In [139] R. J. Zimmer has obtained a far-reaching generalization of Margulis’s superrigidity, passing from the context of representations of lattices to the framework of measurable cocycles over probability measure-preserving actions (representations of “virtual subgroups” in Mackey’s terminology). The connection can be brieﬂy summarized as follows: given a transitive action G X = G/ and some topological group H, there is a bijection

a survey of measured group theory / 349

between measurable cocycles G × G/ → H modulo cocycle conjugation and homomorphisms → H modulo conjugation in H H 1 (G G/ , H) ∼ = Hom(, H)/H (see §A.1, and [52, 146]). In this correspondence, a representation π : → H extends to a homomorphism G → H iff the corresponding cocycle π ◦ c : G × G/ → H is conjugate to a homomorphism G → H. Zimmer’s cocycle superrigidity theorem states that under appropriate nondegeneracy assumptions a measurable cocycle over an arbitrary p.m.p. ergodic action G (X , μ) is conjugate to a homomorphism.

theorem 5.2. (zimmer [139], see also [146]) Let G, G be a semisimple Lie group as in Theorem 5.1, in particular rkR (G) ≥ 2, let G (X , μ) be an irreducible probability measure-preserving action and c : G × X → G be a measurable cocycle that is Zariski dense and not compact. Then there exist a (rational) epimorphism π : G → G and a measurable f : X → G so that c(g, x) = f (gx)−1 π (g)f (x). In the above statement irreducibility of G (X , μ) means mere ergodicity if G is a simple group, and ergodicity of the action Gi (X , μ) for each factor Gi in the case of a semisimple group G = ni=1 Gi with n ≥ 2 factors. For a lattice < G = Gi in a semisimple group the transitive action G G/ is irreducible precisely iff is an irreducible lattice in G. The notion of being Zariski dense (respectively not compact) for a cocycle c : G × X → H means that c is not conjugate to a cocycle c f taking values in a proper algebraic (respectively compact) subgroup of H. The setting of cocycles over p.m.p. actions adds a great deal of generality to the superrigidity phenomena. The ﬁrst illustration of this is the fact that once cocycle superrigidity is known for actions of G it passes to actions of lattices in G: given an action (X , μ) of a lattice < G one obtains a G-action on X¯ = G × X by acting on the ﬁrst coordinate (just like the composition operation of ME-coupling in §2.1). A cocycle c : × X → H has a natural lift to c¯ : G × X¯ → H and its cohomology is directly related to that of the original cocycle. So cocycle superrigidity theorems have an almost automatic bootstrap from lcsc groups to their lattices. The induced action G X¯ is ergodic iff X is ergodic; however, irreducibility is more subtle. Yet, if (X , μ) is mixing, then G X¯ is mixing and therefore is irreducible. Theorem 3.14 was the ﬁrst application of Zimmer’s cocycle superrigidity 5.2 (Theorem; see [139]). Indeed, if α : × X → is the rearrangement cocycle

350 / a. furman OE

associated to an Orbit Equivalence T : (X , μ) ∼ (X , μ ) where < G, < G are lattices, then, viewing α as taking values in G , Zimmer observes that α is Zariski dense using a form of Borel’s density theorem and deduces that G ∼ = G (here for simplicity the ambient groups are assumed to be simple, connected, center free, and rkR (G) ≥ 2). Moreover, there is a homomorphism π : → G and f : X → G so that α(γ , x) = f (γ x)π (γ )f (x)−1 with π : → π() < G being an isomorphism of lattices.

remark 5.3. At this point it is not clear whether π () should be (conjugate to) , and even assuming π() = whether f takes values in . In fact, the self orbit equivalence of the action on G/ given by g → g −1 gives a rearrangement cocycle c : × G/ → , which is conjugate to the identity → by a unique map f : G/ → G with f∗ (mG/ ) ≺ mG . However, if π() = and f takes values in , it follows that the original actions (X , μ) and (X , μ ) are isomorphic via the identiﬁcation π : ∼ = . We return to this point below. 5.1.1. superrigidity and me-couplings Zimmer’s cocycle superrigidity theorem applied to OE- or ME-cocycles (see SectionsA.2 and A.3) has a natural interpretation in terms of ME-couplings. Let G be a higher-rank simple Lie group (hereafter implicitly, connected, and center free) and, denote by i : G → Aut (G) the adjoint homomorphism (which is an embedding since G is center free). theorem 5.4. ([44, theorem 4.1]) Let G be a higher-rank simple Lie group, 1 , 2 < G lattices, and (, m) an ergodic (1 , 2 )-coupling. Then there exists a unique measurable map : → Aut (G) so that m-a.e. on (γ1 ω) = i(γ1 ) (ω),

(γ2 ω) = (ω)i(γ2 )−1

(γi ∈ i ).

∼ Ad (G) ≤ G ≤ Moreover, ∗ m is either the Haar measure on a group G = Aut (G) or is atomic, in which case 1 and 2 are commensurable. Sketch of the proof. To construct such a , choose a fundamental domain X ⊂ for the 2 -action and look at the ME-cocycle c : 1 × X → 2 < G. Apply Zimmer’s cocycle superrigidity theorem to ﬁnd π : 1 → G and φ : X → G. Viewing G as a subgroup in Aut (G), one may adjust π and φ : X → Aut (G) by some α ∈ Aut (G), so that π is the isomorphism i : 1 → 1 to get c(γ1 , x) = φ(γ1 .x)−1 i(γ1 )φ(x).

a survey of measured group theory / 351

Deﬁne : → Aut (G) by (γ2 x) = φ(x)i(γ2 )−1 and check that it satisﬁes the required relation. To identify the measure ∗ m on Aut (G) one uses Ratner’s theorem, which provides the classiﬁcation of 1 -ergodic ﬁnite measures on ¯ 2 . G/ Theorem 3.16 is then proved using this fact with 1 = 2 plugged into the construction in §5.5 that describes an unknown group essentially as a lattice in G. Note that there are two distinct cases in Theorem 5.4: either ∗ m is atomic, in which case (, m) has a discrete ME-coupling as a quotient, or ∗ m is a Haar measure on a Lie group. The former case leads to a virtual isomorphism between the groups and the actions (this is case (1) in Theorem 4.19); in the ¯ 2 (which is [45, latter, 1 X ∼ = / 2 has a quotient of the form 1 G/ theorem C]). This dichotomy clariﬁes the situation in Remark 5.3 above. 5.2. Superrigidity for Product Groups

Let us now turn to a brief discussion of Monod-Shalom rigidity (see Sections 3.1.7 and 3.2.2). Consider a special case of Margulis-Zimmer superrigidity results where the target group G has rank 1 (say G = PSL2 (R)), while G has higher rank. The conclusion of the superrigidity Theorem 5.1 (respectively Theorem 5.2) is that either a representation (respectively cocycle) is degenerate,12 or there is an epimorphism π : G → G . The latter case occurs if and only if G is semisimple G = Gi , with one of the factors Gi G , and π : G → G pri

factoring through the projection π : G−→Gi G . In this case the given representation of the lattice extends to π (respectively the cocycle is conjugate to the epimorphism π). This special case of Margulis-Zimmer superrigidity, that is, from higherrank G to rank 1 G , was generalized by a number of authors [5, 6, 22] replacing the assumption that the target group G has rank 1 by more geometric notions, such as G = Isom (X ) where X is a proper CAT(-1) space. In the setting considered by Monod and Shalom the target group is “hyperboliclike” in a very general way, while the source group G rather than being a higher-rank semisimple Lie group, is just a product G = G1 × · · · × Gn of n ≥ 2 arbitrary compactly generated (in fact, just lcsc) groups. The philosophy is that the number n ≥ 2 of direct factors provides enough higher-rank properties for such statements.

12. Here precompact, or contained in a parabilic.

352 / a. furman

theorem 5.5. (monod-shalom [100]) Let G = G1 × · · · × Gn be a product of n ≥ 2 lcsc groups, G (X , μ) an irreducible p.m.p. action, H is a hyperboliclike group, and c : G × X → H is a nonelementary measurable cocycle. Then there is a nonelementary closed subgroup H1 < H, a compact normal subgroup K % H1 , a measurable f : X → H, and a homomorphism ρ : Gi → H1 /K from one of the factors Gi of G, so that the conjugate cocycle c f takes values in H1 pri ρ and G × X → H1 → H1 /K is the homomorphism π : G−→Gi −→H1 /K .

This beautiful theorem is proved using the technology of second-bounded cohomology (developed in [20, 21, 97] and applied in this setting in [96, 100]) with the notions of hyperboliclike and nonelementary interpreted in the context of the class Creg . Suppose = 1 × · · · × n , n ≥ 2, is a product of “hyperboliclike” groups. Let (, m) be a self-ME-coupling of . Consider an ME-cocycle × X → , which can be viewed as a combination of n cocycles c

pri

ci : × X −→ −→i

(i = 1, . . . , n)

and assume that / is an irreducible action. Viewing the source group as a product of n ≥ 2 factors acting irreducibly, and recalling that the target groups i are “hyperboliclike,” Monod and Shalom apply Theorem 5.5. The cocycles arising from ME-coupling turn out to be nonelementary, leading to the conclusion that each cocycle ci is conjugate to a homomorphism ρi : j(i) → i , modulo some reductions and ﬁnite kernels. Since i commute, the conjugations can be performed independently and simultaneously on all the cocycles ci . After some intricate analysis of the map i → j(i), kernels and cokernels of ρi , Monod and Shalom show that in the setting of ME-couplings as above the map, i → j(i) is a permutation and ρi are isomorphisms. Thus the original cocycle c can be conjugate to an automorphism of . This ME-cocycle superrigidity can now be plugged into an analogue of Theorem 5.4 to give a measurable bi--equivariant map → , which can be used as an input to a construction like Theorem 5.13. This allows us to identify unknown groups measure equivalent to = 1 × · · · × n . The only delicate SOE

point is that starting from a X ∼ Y and the corresponding (, )ˇ and apply coupling one needs to look at the self--coupling = × the cocycle superrigidity result to / . In order to guarantee that the latter action is irreducible, Monod and Shalom require X to be irreducible and Y to be mildly mixing. They also show that the assumption on mild mixing is necessary for the result.

a survey of measured group theory / 353

In [12] Uri Bader and the author proposed to study higher-rank superrigidity phenomena using a notion of a (generalized) Weyl group, which works well for higher-rank simple Lie groups, arbitrary products G = G1 × · · · × Gn of n ≥ 2 ˜ 2 groups, which are close relatives to lattices in SL3 (Qp ). factors, and exotic A In particular:

theorem 5.6. (bader-furman [12]) Theorem 5.5 holds for target groups from class Dea . Here Dea is a class of hyperboliclike groups that include many of the examples in Creg . Plugging this into the Monod-Shalom machine one obtains the same results of products of groups in class Dea . 5.3. Strong Rigidity for Cocycles

In the proof of Theorem 5.4 Zimmer’s cocycle superrigidity was applied to a measure equivalence cocycle. This is a rather special class of cocycles (see §A.3). If cocycles are analogous to representations of lattices, then ME-cocycles are analogous to isomorphisms between lattices; in particular, they have an “inverse.” Kida’s work on ME for mapping class groups focuses on rigidity results for such cocycles. We shall not attempt to explain the ingredients used in this work, but will just formulate the main technical result analogous to Theorem 5.4. Let be a subgroup of ﬁnite index in ( g,p ) with 3g + p − 4 > 0, C = C( g,p ) denoting its curve complex, and Aut (C) the group of its automorphisms; this is a countable group commensurable to .

theorem 5.7. (kida [86]) Let (, m) be a self-ME-coupling of . Then there exists a measurable map × -equivariant map : → Aut (C). Returning to the point that ME-cocycles are analogous to isomorphism between lattices, one might wonder whether Theorem 5.4 holds in cases where Mostow rigidity applies, speciﬁcally for G of rank 1 with PSL2 (R) excluded. In [13] this is proved for G Isom (HnR ), n ≥ 3, and a restricted ME.

theorem 5.8. (bader-furman-sauer [13]) Theorem 5.4 applies to 1 ME-couplings of lattices in G = SOn,1 (R), n ≥ 3. The proof of this result uses homological methods (1 and other completions of the usual homology) combined with a version of the Gromov-Thurston proof of Mostow rigidity (for Isom (HnR ), n ≥ 3) adapted to this setting.

354 / a. furman 5.4. Cocycle Superrigid Actions

In all the previous examples the structure of the acting group was the sole source for (super)rigidity. Recently Sorin Popa has developed a number of remarkable cocycle superrigidity results of a completely different nature [109– 115]. These results exhibit an extreme form of cocycle superrigidity and rather than relying only on the properties of the acting group , take advantage of the action (X , μ).

definition 5.9. An action (X , μ) is C -cocycle superrigid, where C is some class of topological groups, if for every ∈ C every measurable cocycle c : X → has the form c(g, x) = f (gx)−1 ρ(g)f (x) for some homomorphism ρ : → and some measurable f : X → . Here we shall focus on the class Gdsc of all countable groups; however, the following results hold for all cocycles taking values in a broader class Uﬁn , which contains Gdsc and Gcpt —separable compact groups. Note that the concept of Gdsc -cocycle superrigidity is unprecedentedly strong: there is no assumption on the cocycle, the assumption on the target group is extremely weak, and the “untwisting” takes place in the same target group.

theorem 5.10. (popa [113]) Let be a group with property (T) and (X , μ) = (X0 , μ0 ) be the Bernoulli action. Then (X , μ) is Gdsc -cocycle superrigid. In fact, the result is stronger: it sufﬁces to assume that has relative property (T) with respect to a w-normal subgroup 0 , and (X , μ) has a relatively weakly mixing extension (X¯ , μ) ¯ that is s-malleable, while ¯ 0 (X , μ) ¯ is weakly mixing. Under these conditions (X , μ) is Uﬁn cocycle superrigid. See [113] and [48] for the relevant deﬁnitions and more details. We indicate the proof (of the special case above) in §5.8. Vaguely speaking, Popa’s approach exploits the tension between certain (local) rigidity provided by the acting group and deformations supplied by the action. In the following remarkable result, Popa further relaxed the property (T) assumption.

theorem 5.11. (popa [115]) Let be a group containing a product 1 × 2 where 1 is nonamenable, 2 is inﬁnite, and 1 × 2 is w-normal in . Then any Bernoulli action (X , μ) is Uﬁn -cocycle superrigid.

a survey of measured group theory / 355

The deformations alluded to above take place for the diagonal -action on the square (X × X , μ × μ). This action is supposed to be ergodic; equivalently, the original action should be weakly mixing and satisfy addition properties. Isometric actions, or staying within the ergodic-theoretic terminology, actions with discrete spectrum provide the opposite type of dynamics. These actions have the form K /L where L < K are compact groups, → K a homomorphism with dense image, and acts by left translations. Totally disconnected K corresponds to proﬁnite completion lim / n with respect to a chain of ←− normal subgroups of ﬁnite index. Isometric actions K /L with proﬁnite K , can be called proﬁnite ergodic actions of —these are precisely inverse limits X = lim Xn of transitive -actions on ﬁnite spaces. Adrian Ioana found ←− the following “virtually Gdsc -cocycle superrigidity” phenomenon for proﬁnite actions of Kazhdan groups.

theorem 5.12. (ioana [72]) Let X = K /L be an ergodic proﬁnite action. Assume that has property (T), or a relative property (T) with respect to a normal subgroup 0 , which acts ergodically on X . Then any measurable cocycle c : X → into a discrete group, is conjugate to a cocycle coming from a ﬁnite quotient X → Xn , that is, c is conjugate to a cocycle induced from a homomorphism n → of a ﬁnite-index subgroup. In §5.8.2 a similar result is proven for all discrete spectrum actions (not necessarily proﬁnite ones). 5.5. Constructing Representations

In Geometric group theory many QI-rigidity results are proved using the following trick. Given a metric space X one declares self-quasi-isometries f , g : X → X to be equivalent if sup d( f (x), g(x)) < ∞. x∈X

Then equivalence classes of quasi-isometries (hereafter q.i.) form a group, denoted QI(X ). This group contains (a quotient of) Isom (X ), which can sometimes be identiﬁed within QI(X ) in coarse geometric terms. If is a group with well-understood QI() and is an unknown group q.i. to , then one gets a homomorphism ρ : → Isom () → QI() ∼ = QI() whose kernel and image can then be analyzed.

356 / a. furman

Facing a similar problem in the measure equivalence category, there is a difﬁculty in deﬁning an analogue for QI(). Let us describe a construction that allows us to analyze the class of all groups ME to a given group from an information about self-ME-couplings of . Let G be a lcsc unimodular group. Let us assume that G has the strong ICC property, by which we mean that the only regular Borel conjugationinvariant probability measure on G is the trivial one, namely the Dirac mass δe at the origin. For countable groups this is equivalent to the condition that all nontrivial conjugacy classes are inﬁnite, that is, the usual ICC property. Connected, (semi) simple Lie groups with trivial center and no compact factors provide other examples of strongly ICC groups. Theorems 5.4, 5.7 and 5.8 are instances where a strongly ICC group G has the property that for any13 ME-self-coupling (, m) of a lattice in G there exists a bi--equivariant measurable map to G, that is, a Borel map : → G satisfying m-a.e.: ((γ1 , γ2 )ω) = γ1 (ω)γ2−1

(γ1 , γ2 ∈ ).

It is not difﬁcult to see that the strong ICC property implies that such a map is also unique. (It should also be pointed out that the existence of such maps for self-couplings of lattices is equivalent to the same property for self-couplings of the lcsc group G itself; but here we shall stay in the framework of countable groups). The following general tool shows how these properties of G can be used to classify all groups ME to a lattice < G; up to ﬁnite kernels these turn out to be lattices in G.

theorem 5.13. (bader-furman-sauer [13]) Let G be a strongly ICC lcsc unimodular group, < G a lattice, and some group ME to and (, m) be ˇ of admits a a (, )-coupling. Assume that the self-ME-coupling = × Borel map : → G, satisfying a.e.: ([γ1 x, γ2 y]) = γ1 · ([x, y]) · γ2−1

(γ1 , γ2 ∈ ).

¯ with K ﬁnite and ¯ being Then there exists a short exact sequence K −→−→ a lattice in G, and a Borel map : → G so that a.e.: ([x, y]) = (x) · (y)−1 ,

(γ z) = γ · (z),

(λz) = (z) · λ¯ −1 .

13. In the case of G = Isom (Hn ) we restrict to all 1 -ME-couplings.

a survey of measured group theory / 357

Moreover, the push-forward of ∗ m is a Radon measure on G-invariant under the maps ¯ g → γ g λ,

¯ (γ ∈ , λ¯ ∈ ).

If G is a (semi)simple Lie group, the last condition on the push-forward measure can be analyzed using Ratner’s theorem (as in Theorem 5.4) to deduce that assuming ergodicity ∗ m is either a Haar measure on G, or on a coset of its ﬁnite-index subgroup, or it is (proportional to) counting measure on a coset ¯ as ﬁnite-index subgroups. of a lattice containing and a conjugate of Theorem 5.13 is a streamlined and improved version of similar statements obtained in [44] for higher-rank lattices, in [101] for products, and in [86] for mapping class groups. 5.6. Local Rigidity for Measurable Cocycles

The rigidity vs. deformations approach to rigidity results developed by Sorin Popa led to a number of striking results in von Neumann algebras and in Ergodic theory (some have been mentioned in §5.4). Let us illustrate the rigidity side of this approach by the following simple purely ergodic-theoretic statement, which is a variant of Hjorth’s [68, lemma 2.5]. Recall that one of the several equivalent forms of property (T) is the following statement: a lcsc group G has (T) if there exists a compact K ⊂ G and > 0 so that for any unitary G-representation π and any (K , )-almost invariant unit vector v there exists a G-invariant unit vector w with v − w < 41 .

proposition 5.14. Let G be a group with property (T) and (K , ), as above. Then for any ergodic probability measure-preserving action G (X , μ), any countable group and any pair of cocycles α, β : G × X → with 2 μ x ∈ X : α(g, x) = β(g, x) > 1 − 2

(∀g ∈ K )

there exists a measurable map f : X → so that β = α f . Moreover, one can assume that 3 μ x : f (x) = e > . 4 Proof. Let X˜ = X × be equipped with the inﬁnite measure μ˜ = μ × m where m stands for the counting measure on . Then G acts on (X˜ , μ) ˜ by g : (x, λ) → (g.x, α(g, x)λβ(g, x)−1 ).

358 / a. furman

This action preserves μ˜ and we denote by π the corresponding unitary Grepresentation on L2 (X˜ , μ). ˜ The characteristic function v = 1X ×{e} satisﬁes 2 (g ∈ ) = 2 v − π (g)v2 = 2 − 2Reπ (g)v, v < 2 − 2 1 − 2 ˜ with and therefore there exists a π(G)-invariant unit vector w ∈ L2 (X˜ , ) 1 2 2 v − w < 4 . Since 1 = w = X λ |w(x, λ)| we may deﬁne (x) = λ : |w(x, λ)| = p(x) p(x) = max |w(x, λ)|, λ

and observe that p(x) and the cardinality k(x) of the ﬁnite set (x) are measurable -invariant functions on (X , μ); hence are a.e. constants p(x) = p ∈ (0, 1], 1 k(x) = k ∈ {1, 2, . . . }. Since 16 > v − w2 ≥ (1 − p)2 we have p > 43 . It follows that k = 1 because 1 = w2 ≥ kp2 . Therefore (x) = {f (x)} for some measurable map f : X → . The π(G)-invariance of w gives π(G)-invariance ˜ of the characteristic function of (x, f (x)) ∈ X : x ∈ X , which is equivalent to f (gx) = α(g, x)f (x)β(g, x)−1 and β = αf . Let A = f −1 ({e}) and a = μ(A). Since λ |w(x, λ)|2 is a G-invariant function it is a.e. constant w2 = 1. Hence for x ∈ / A we have |w(x, e)|2 ≤ 2 2 1 − |w(x, f (x))| = 1 − p , and 5.1

9(1 − a) 1 > v − w2 ≥ a · (1 − p2 ) + (1 − a) · (1 − (1 − p2 )) ≥ (1 − a) · p2 > . 16 16 Thus a = μ x ∈ X : f (x) = e > 89 > 43 , as required. 5.7. Cohomology of Cocycles

Let us ﬁx two groups and . There is no real assumption on , it may be any lcsc group, but we shall impose an assumption on . One might focus on the case where is a countable group (class Gdsc ), but versions of the statements below would apply also to separable compact groups, or groups in a larger class Uﬁn of all Polish groups, which embed in the unitary group of a von Neumann algebra with ﬁnite faithful trace,14 or a potentially even larger class Gbinv of groups with a bi-invariant metric, and the class Galg of connected algebraic groups over local ﬁelds, say of zero characteristic.

14. This class, introduced by Popa, contains both discrete countable groups and separable compact ones.

a survey of measured group theory / 359

Given a (not necessarily free) p.m.p. action (X , μ) let Z 1 (X , ) or Z 1 ( X , ) denote the space of all measurable cocycles c : × X → and by H 1 (X , ), or H 1 ( X , ), the space of equivalence classes of cocycles up to conjugation by measurable maps f : X → . If ∈ Galg , we shall focus on a subset Hss1 (X , ) of (classes of) cocycles whose algebraic hull is connected, semisimple, center free, and has no compact factors. Any -equivariant quotient map π : X → Y deﬁnes a pull-back Z 1 (Y , ) → Z 1 (X , ) by c π (g, x) = c(g, π(x)), which descends to π∗

H 1 (Y , )−→H 1 (X , ). ¯ and j : < give rise to push-forward maps Group inclusions i : < , i∗

¯ H 1 (X , )−→H 1 (X , ),

j∗

H 1 ( X , )−→H 1 ( X , ).

question 5.15. What can be said about these maps of the cohomology ? The discussion here is inspired and informed by Popa’s [113]. In particular, the following statements Propositions 5.16(2), 5.17, and 5.18(1), and Corollary 5.20 are variations on Popa’s original [113, lemma 2.11, proposition 3.5, lemma 3.6, theorem 3.1]. Working with class Gbinv makes the proofs more transparent than in Uﬁn —this was done in [48, §3]. Proposition 5.16(3) for semisimple target is implicit in [45, lemma 3.5]. The full treatment of the statements below, including Theorem 5.19, will appear in [49].

proposition 5.16. Let π : X → Y be a -equivariant quotient map. Then π∗

H 1 (Y , )−→H 1 (X , ) is injective in the following cases: 1) is discrete and torsion free. 2) ∈ Gbinv and π : X → Y is relatively weakly mixing. 3) ∈ Galg and H 1 (−, ) is replaced by Hss1 (−, ). The notion of relative weakly mixing was introduced independently by Zimmer [137] and Furstenberg [51]: a -equivariant map π : X → Y is relatively weakly mixing if the -action on the ﬁbered product X ×Y X is ergodic (or ergodic relatively to Y ); this turns out to be equivalent to the condition that X contains no intermediate isometric extensions of Y .

360 / a. furman

¯ ∈ Gbinv be a closed subgroup, and proposition 5.17. Let i : < (X , μ) some p.m.p. action. Then i∗

¯ H 1 (X , )−→H 1 (X , ) is injective. This useful property fails in the Galg setting: if < G is a lattice in a (semi) simple Lie group and c : × G/ → in the canonical class, then viewed as a cocycle into G > , c is conjugate to the identity embedding ∼ = < G, but as a -valued cocycle it cannot be “untwisted.”

proposition 5.18. Let π : X → Y be a quotient map of ergodic actions, and j : < be a normal, or sub-normal, or w-normal closed subgroup acting ergodically on X . Assume that either 1) ∈ Gbinv and π is relatively weakly mixing, or 2) ∈ Galg and one considers Hss1 (−, ). Then H 1 ( Y , ) is the push-out of the rest of the following diagram: H 1 ( X , )

j∗

O

π∗

H 1 ( Y , )

/ H 1 ( X , ) O π∗

j∗

/ H 1 ( Y , ).

In other words, if the restriction to X of a cocycle c : × X → is conjugate to one descending to × Y → , then c has a conjugate that descends to × X → . The condition < is w-normal (weakly normal), which means that there exists a well-ordered chain i of subgroups starting from and ending with , so that i % i+1 and for limit ordinals j = i<j i (Popa). Let πi : X → Yi is a collection of -equivariant quotient maps. Then X has a unique -equivariant quotient p : X → Z = Yi , which is maximal among all common quotients pi : Yi → Z. Identifying -equivariant quotients with -equivariant complete sub-σ -algebras of X , one has p−1 (Z ) = i πi−1 (Yi ); or in the operator algebra formalism p−1 (L∞ (Z)) = i πi−1 (L∞ (Yi )).

a survey of measured group theory / 361

theorem 5.19. Let πi : X → Yi , 1 ≤ i ≤ n, be a ﬁnite collection of equivariant quotients, and Z = ni=1 Yi . Then H 1 (Z, ) is the push-out of H 1 (Yi , ) under conditions (1)–(3) of Proposition 5.16: 8 ppp p p ppp ppp π1∗

H 1 (Y1 , )

H 1 (X , )

O

πi∗

fNNN π ∗ NNN n NNN NN

· · · H 1 (Yi , ) · · · H 1 (Yn , ) fNNN p∗ O 8 ∗ p1 pp NNN n ppp NNN p pi∗ p p NN ppp H 1 (Z, )

More precisely, if ci : × X → are cocycles (in case (3) assume [ci ] ∈ Hss1 (Yi , )), whose pullbacks ci (g, πi (x)) are conjugate over X , then there exists a unique class [c] ∈ H 1 (Z, ), so that c(g, pi (y)) ∼ ci (g, y) in Z 1 (Yi , ) for all 1 ≤ i ≤ n. The proof of this Theorem relies on Proposition 5.16 and contains it as a special case n = 1. This result can be useful to push cocycles to deeper and deeper quotients; if π : X → Y is a minimal quotient to which a cocycle or a family of cocycles can descend up to conjugacy, then it is the minimal or characteristic quotient for these cocycles: if they descend to any quotient X → Y , then necessarily X → Y → Y . For example if < G is a higher-rank lattice, is a discrete group and c : × X → is an OE (or ME) cocycle, then either c descends to a -action on a ﬁnite set (virtual isomorphism case), or to π X −→G/ with lattice in G, where π is uniquely deﬁned by c (initial OE or ME). An important special (and motivating) case of Theorem 5.19 is that of X = Y × Y where Y is a weakly mixing action. Then the projections πi : X → Yi = Y , i = 1, 2, give Z = Y1 ∧ Y2 = {pt} and H 1 ( {pt}, ) = Hom(, ). So

corollary 5.20. (popa [113, theorem 3.1], see also [48, theorem 3.4]) Let Y be a weakly mixing action and c : × Y → a cocycle into ∈ Gbinv . Let X = Y × Y with the diagonal -action, c1 , c2 : × X → the cocycles ci (g, (y1 , y2 )) = c(g, yi ). If c1 ∼ c2 over X , then there exists homomorphism ρ : → and a measurable f : Y → , so that c(g, y) = f (gy)−1 ρ(g) f (y).

362 / a. furman 5.8. Proofs of Some Results

In this section we shall give relatively self-contained proofs of some of the results mentioned above.

5.8.1. sketch of a proof for popa’s cocycle superrigidity theorem 5.10 First note that without loss of generality the base space (X0 , μ0 ) of the Bernoulli action may be assumed to be nonatomic. Indeed, Proposition 5.16(2) implies that for each of the classes Gdsc ⊂ Uﬁn ⊆ Gbinv the corresponding cocycle superrigidity descends through relatively weakly mixing quotients, and ([0, 1], dx) → (X0 , μ0 ) is such. Given any action (X , μ) consider the diagonal -action on (X × X , μ × μ) and its centralizer Aut (X × X ) in the Polish group Aut (X × X , μ × μ). It always contain the ﬂip F : (x, y) → (y, x). Bernoulli actions X = [0, 1] have the special property (called s-malleability by Popa) that there is a path p : [1, 2] → Aut (X × X ),

with

p1 = Id,

p2 = F.

Indeed, the diagonal component-wise action of Aut ([0, 1] × [0, 1]) on X × X = ([0, 1] × [0, 1]) embeds into Aut (X × X ) and can be used to connect Id to F. Fix a cocycle c : X → . Consider the two lifts to X × X → X : ci : X × X → ,

ci (g, (x1 , x2 )) = c(g, xi ),

(i = 1, 2).

Observe that they are connected by the continuous path of cocycles ct (g, (x, y)) = c1 (g, pt (x, y)), 1 ≤ t ≤ 2. Local rigidity in Proposition 5.14 implies that c1 and c2 are conjugate over X × X , and the proof is completed invoking Corollary 5.20. Under the weaker assumption of relative property (T) with respect to a w-normal subgroup, Popa uses Proposition 5.18.

5.8.2. a cocycle superrigidity theorem We state and prove a cocycle superrigidity theorem, inspired and generalizing Adrian Ioana’s Theorem 5.12. Thus a number of statements (Theorems 4.15 and 4.20(2) and §4.3.1) in this survey get a relatively full treatment. The proof is a good illustration of Popa’s deformation vs. rigidity approach. Recall that an ergodic p.m.p. action (X , μ) is said to have a discrete spectrum if the Koopman -representation on L2 (X , μ) is a Hilbert sum of ﬁnite-dimensional subrepresentations. Mackey proved (generalizing the Halmos–von Neumann theorem for Z, and using Peter-Weyl ideas) that discrete spectrum action is measurably isomorphic to the isometric -action on

a survey of measured group theory / 363

(K /L, mK /L ), g : kL → τ (g)kL, where L < K are compact separable groups and τ : → K is a homomorphism with dense image.

theorem 5.21. (after ioana’s theorem 5.12, [72]) Let (X , μ) be an ergodic p.m.p. action with discrete spectrum. Assume that has property (T), or contains a w-normal subgroup 0 with property (T) acting ergodically on (X , μ). Let be an arbitrary torsion-free discrete countable group and c : × X → be a measurable cocycle. Then there is a ﬁnite-index subgroup 1 < , a 1 -ergodic component X1 ⊂ X (of measure μ(X1 ) = [ : 1 ]−1 ), a homomorphism ρ : 1 → , and a measurable map φ : X → , so that the conjugate cocycle c φ restricted to 1 X1 → , is the homomorphism ρ : 1 → . The cocycle c φ : × X → is induced from ρ. The assumption that is torsion free is not essential; in general, one might need to lift the action to a ﬁnite cover Xˆ 1 → X1 via a ﬁnite group that embeds in . If K is a connected Lie group, then 1 = and X1 = X = K /L. The stated result is deduced from the case where L is trivial, that is, X = K , using Proposition 5.16(1). We shall make this simpliﬁcation and assume has property (T) (the modiﬁcation for the more general case uses an appropriate version of Propositions 5.14 and 5.17). An appropriate modiﬁcation of the result handles compact groups as a possible target group for the cocycle. Proof. The K -action by right translations: t : x → xt −1 commutes with the action on K ; in fact, K is precisely the centralizer of in Aut (K , mK ). This allows us to deform the initial cocycle c : × X → , setting ct (g, x) = c(g, xt −1 )

(t ∈ K ).

Let F ⊂ and > 0 be as in the “local rigidity” Proposition 5.14. Then for some open neighborhood U of e ∈ K for every t ∈ U there is a unique measurable ft : K → with ct (g, x) = c(g, xt −1 ) = ft (gx)c(g, x)ft (x)−1

3 μ x : ft (x) = e > . 4

Suppose t, s ∈ U, and ts ∈ U. Then fts (gx) c(g, x) fts (x)−1 = cts (g, x) = c(g, xs−1 t −1 ) = ft (gxs−1 ) c(g, xs−1 ) ft (xs−1 )−1 = ft (gxs−1 )fs (gx) c(g, x) [ft (xs−1 )fs (x)−1 ]−1 .

364 / a. furman

This can be rewritten as F(gx) = c(g, x) F(x) c(g, x)−1 ,

where

F(x) = fts (x)−1 ft (xs−1 )fs (x).

Since ft , fs , fts takes value e with probability > 3/4, it follows that A = F −1 ({e}) has μ(A) > 0. The equation implies -invariance of A. Thus μ(A) = 1 by ergodicity. Hence, whenever t, s, ts ∈ U 5.2

fts (x) = ft (xs−1 )fs (x).

If K is a totally disconnected group, that is, a proﬁnite completion of as in Ioana’s Theorem 5.12, then U contains an open subgroup K1 < K . In this case one can skip the following paragraph. In general, let V be a symmetric neighborhood of e ∈ K so that V 2 ⊂ U, n and let K1 = ∞ n=1 V . Then K1 is an open (hence, also closed) subgroup of K ; in the connected case K1 = K . We shall extend the family {ft : K → }t∈V to be deﬁned for all t ∈ K1 while satisfying Equation (5.2), using a “cocycle continuation” procedure akin to analytic continuation. For t, t ∈ K1 a V -quasipath pt→t from t to t is a sequence t = t0 , t1 , . . . , tn = t where ti ∈ ti−1 V . Two V quasipaths from t to t are V -close if they are within V -neighborhoods from each other. Two V -quasipaths pt→t and qt→t are V -homotopic if there is a (0) (k) chain pt→t = pt→t , . . . , pt→t = qt→t of V -quasipaths where p(i−1) and p(i) are V -close, 1 ≤ i ≤ k. Iterating Equation (5.2) one may continue the deﬁnition of f· from t to t along a V -quasipath; the continuation being the same for V -close quasipaths, and therefore for V -homotopic quasipaths as well (all from t to t ). The possible ambiguity of this cocycle continuation procedure (V ) is encoded in the homotopy group π1 (K1 ) consisting of equivalence classes of V -quasipaths from e → e modulo V -homotopy. We claim that this group (V ) is ﬁnite. In the case of a connected Lie group K1 , π1 (K1 ) is a quotient of π1 (K1 ) that is ﬁnite since K1 , contains a dense property (T) group and can(V ) not have torus factors. This covers the general case as well since π1 (K1 ) “feels” only ﬁnitely many factors when K1 is written as an inverse limit of connected Lie groups and ﬁnite groups. Considering the continuations of f along (V ) V -quasipaths e → e we get a homomorphism π1 (K1 ) → , which must be trivial since was assumed to be torsion free. Therefore, we obtain a family of measurable maps ft : K1 → indexed by t ∈ K1 and still satisfying Equation (5.2). Let 1 = τ −1 (K1 ). Then the index [ : 1 ] = [K : K1 ] is ﬁnite. We shall focus on the restriction c1 of c to 1 K1 . Note that Equation (5.2) is a cocycle equation for the simply transitive action K1 on itself. It follows by a standard argument that it is a coboundary. Indeed, for a.e. x0 ∈ K1 Equation (5.2) holds

a survey of measured group theory / 365

for a.e. t, s ∈ K1 . In particular, for a.e. t, x ∈ K1 , using s = x −1 x0 , one obtains ftx −1 x0 (x0 ) = ft (x)fx −1 x0 (x0 ). This gives ft (x) = φ(xt −1 )φ(x)−1 ,

where

φ(x) = fx −1 x0 (x0 ).

Equation ct = c ft translates into the fact that the cocycle c φ (g, x) = φ(gx)−1 c(g, x)φ(x) satisﬁes for a.e. x, t c φ (g, xt −1 ) = c φ (g, x). Thus c(g, x) does not depend on the space variable. Hence, it is a homomorphism c φ (g, x) = ρ(g). φ

Finally, the fact that c φ is induced from c1 is straightforward.

Appendix A. Cocycles Let G (X , μ) be a measurable, measure-preserving (sometimes just measure class preserving) action of a topological group G on a standard Lebesgue space (X , μ), and H be a topological group. A Borel measurable map c : G × X → H forms a cocycle if for every g1 , g2 ∈ G for μ-a.e. x ∈ X one has c(g2 g1 , x) = c(g2 , g1 .x) · c(g1 , x). If f : X → H is a measurable map and c : G × X → H is a measurable cocycle, deﬁne the f -conjugate c f of c to be c f (g, x) = f (g.x)−1 c(g, x) f (x). It is straightforward to see that c f is also a cocycle. One says that c and c f are (measurably) conjugate, or cohomologous cocycles. The space of all measurable cocycles × X → is denoted by Z 1 ( X , ) and the space of equivalence classes by H 1 ( X , ). Cocycles that do not depend on the space variable c(g, x) = c(g) are precisely homomorphisms c : G → H. So cocycles may be viewed as generalized homomorphisms. In fact, any group action G (X , μ) deﬁnes a measured groupoid G with G (0) = X , and G (1) = (x, gx) : x ∈ X , g ∈ G (see [9] for the background). In this context cocycles can be viewed as homomorphisms G → H. If π : (X , μ) → (Y , ν) is an equivariant quotient map between -actions (so π∗ μ = ν and π ◦ γ = γ for γ ∈ ), then for any target group any cocycle c : × Y → lifts to c¯ : × X → by

366 / a. furman

c¯ (g, x) = c(g, π (x)). Moreover, if c = c f ∼ c in Z 1 ( Y , ), then the lifts c¯ = c¯ f ◦π ∼ c¯ in π Z 1 ( X , ); so X −→Y induces π

H 1 ( X , )←−H 1 ( Y , ). Note that Hom(, ) is Z 1 ( {pt}, ) and classes of cocycles on × X → cohomologous to homomorphisms is precisely the pull-back of H 1 ( {pt}, ). A.1. The Canonical Class of a Lattice, (Co)induction

Let < G be a lattice in a lcsc group. By deﬁnition the transitive G-action on X = G/ has an invariant Borel regular probability measure μ. Let F ⊂ G be a Borel fundamental domain for the right -action on G (i.e., F is a Borel subset of G set that meets every coset g precisely once). Fundamental domains correspond to Borel cross-section σ : G/ → G of the projection G → G/ . Deﬁne cσ : G × G/ →

by

cσ (g, h) = σ (gh)−1 g σ (h).

Clearly, this is a cocycle (a conjugate of the identity homomorphism G → G); however, cσ takes values in the subgroup of G. This cocycle is associated to a choice of the cross-section σ (equivalently, the choice of the fundamental domain); starting from another Borel cross-section σ : G/ → G results in a cohomologous cocycle: f

cσ = cσ

where

f : G/ →

is deﬁned by

σ (x) = f (x)σ (x).

Let be a lattice in G. Then any action (X , μ) gives rise to the induced G-action (a.k.a. suspension) on X¯ = G × X where G acts on the ﬁrst coordinate. Equivalently, X¯ = G/ × X and g : (g , x) → (gg , c(g, g )x) where c : G × G/ → is in the canonical class. Here the G-invariant ﬁnite measure μ¯ = mG/ × μ is ergodic iff μ is -ergodic. If α : × X → H is a cocycle, the induced cocycle α¯ : G × X¯ → H is given by α(g, ¯ (g , x)) = α(c(g, g ), x). The cohomology of α¯ is the same as that of α (one relates maps F : X¯ → H to f : X → H by f (x) = F(e, x) taking instead of e a generic point in G/ ). In particular, α¯ is cohomologous to a homomorpism π¯ : G → H iff α is cohomologous to a homomorphism → H; see [146] for details. Cocycles appear quite naturally in a number of situations such as (volumepreserving) smooth actions on manifolds, where choosing a measurable trivialization of the tangent bundle, the derivative becomes a matrix-valued

a survey of measured group theory / 367

cocycle. We refer the reader to David Fisher’s survey [42] where this type of cocycle is extensively discussed in the context of Zimmer’s program. Here we shall be interested in a different type of cocycles: “rearrangement” cocycles associated to orbit equivalence, measure equivalence, and so on. as follows. A.2. OE-Cocycles

Let (X , μ) and (Y , ν) be two measurable, measure-preserving ergodic actions on probability spaces, and T : (X , μ) → (Y , ν) be an Orbit Equivalence. Assume that the -action is essentially free, that is, for ν-.a.e y ∈ Y , the stabilizer y = h ∈ : h.y = y is trivial. Then for every g ∈ and μ-a.e. x ∈ X , the points T (g.x), T (x) ∈ Y lie on the same -orbit. Let α(g, x) ∈ denote the (a.e. unique) element of with T (g.x) = α(g, x).T (x). Considering x, g.x, g g.x one checks that α is actually a cocycle α : × X → . We shall refer to such α as the OE-cocycle, or the rearrangement cocycle, corresponding to T . Note that for μ-a.e. x, the map α(−, x) : → is a bijection; it describes how the -names of points x ∈ .x translate into the -names of y ∈ .T (x) under the map T . The inverse map T −1 : (Y , ν) → (X , μ) deﬁnes an OEcocycle β : × Y → , which serves as an “inverse” to α in the sense that a.e.: β(α(g, x), T (X )) = g

(g ∈ ).

A.3. ME-Cocycles

Let (, m) be an ME-coupling of two groups and and let Y , X ⊂ be fundamental domains for , actions, respectively. The natural identiﬁcation / ∼ = X , ω → ω ∩ X , translates the -action on / to X by γ :X

x → gα(g, x)x ∈ X

where α(γ , x) is the unique element in taking γ x ∈ into X ⊂ . It is easy to see that α : × X → is a cocycle with respect to the above -action on X , which we denote by a dot γ · x to distinguish it from the -action on . (If and are lattices in G, then α : × G/ → is the restriction of the canonical cocycle G × G/ → ). Similarly we get a cocycle β : × Y → . So the (, ) ME-coupling and a choice of fundamental domains Y ∼ = / , X ∼ = / deﬁne a pair of cocycles

368 / a. furman

α : × / → ,

A.1

β : × / → .

Changing the fundamental domains amounts to conjugating the cocycles and vice versa.

remark 5.22. One can characterize ME-cocycles among all measurable cocycles α : × X → as discrete ones with ﬁnite covolume. These concepts refer to the following construction: let (X˜ , μ) ˜ = (X × , μ × m ) and let act by g : (x, h) → (g.x, α(g, x)h). Say that the cocycle is discrete and has ﬁnite covolume if the action (X˜ , μ) ˜ admits a ﬁnite-measure fundamental domain. References [1] [2] [3] [4] [5] [6] [7] [8] [9]

[10] [11]

[12] [13] [14]

M. Abert and N. Nikolov, Rank gradient, cost of groups and the rank versus Heegaard genus problem (2007), available at arXiv:0701.361. M. Abert and B. Weiss, Bernoulli actions are weakly contained in any free action, in preparation. S. Adams, An equivalence relation that is not freely generated, Proc. Amer. Math. Soc. 102 (1988), no. 3, 565–566. ———, Trees and amenable equivalence relations, Ergod Th. & Dynam. Sys. 10 (1990), no. 1, 1–14. ———, Reduction of cocycles with hyperbolic targets, Ergod. Th. & Dynam. Sys. 16 (1996), no. 6, 1111–1145. S. R. Adams and R. J. Spatzier, Kazhdan groups, cocycles and trees, Amer. J. Math. 112 (1990), no. 2, 271–287. A. Alvarez and D. Gaboriau, Free products, orbit equivalence and measure equivalence rigidity (2008), available at arXiv:0806.2788. C. Anantharaman-Delaroche, Cohomology of property T groupoids and applications, Ergod. Th. & Dynam. Sys. 25 (2005), no. 4, 977–1013. C. Anantharaman-Delaroche and J. Renault,, Amenable groupoids, Monographies de L’Enseignement Mathématique [Monographs of L’Enseignement Mathématique], vol. 36, L’Enseignement Mathématique, Geneva, 2000. With a foreword by Georges Skandalis and Appendix B by E. Germain. J. E. Andersen, Mapping class groups do not have Kazhdan’s property (T) (2007), available at arXiv:0706.2184. M. F. Atiyah, Elliptic operators, discrete groups and von Neumann algebras, Colloque “Analyse et Topologie” en l’Honneur de Henri Cartan (Orsay, 1974),Soc. Math. France, Paris, 1976, pp. 43–72. Astérisque, No. 32–33. U. Bader and A. Furman, Superrigidity via Weyl groups: hyperbolic-like targets (preprint, 2010). U. Bader, A. Furman, and R. Sauer, Intergrable measure equivalence and rigidity of hyperbolic lattices (2010), available at arXiv:1006.5193. ———, Efﬁcient subdivision in hyperbolic groups and applications (2010), available at arXiv:1003.1562.

a survey of measured group theory / 369 [15] [16] [17] [18]

[19] [20] [21] [22] [23] [24]

[25] [26] [27] [28] [29] [30]

[31] [32] [33] [34] [35] [36]

U. Bader, A. Furman, and A. Shaker, Superrigidity, Weyl groups, and actions on the circle (2006), available at arXiv:math/0605276. M. E. B. Bekka and A. Valette, Kazhdan’s property (T) and amenable representations, Math. Z. 212 (1993), no. 2, 293–299. B. Bekka, P. de la Harpe, and A. Valette, Kazhdan’s property (T ), New Mathematical Monographs, Vol. 11, Cambridge University Press, Cambridge, 2008. J. Bourgain, A. Furman, E. Lindenstrauss, and S. Mozes, Invariant measures and stiffness for non-abelian groups of toral automorphisms, C. R. Math. Acad. Sci. Paris 344 (2007), no. 12, 737–742. M. R. Bridson, M. Tweedale, and H. Wilton, Limit groups, positive-genus towers and measure-equivalence, Ergod. Th. & Dynam. Sys. 27 (2007), no. 3, 703–712. M. Burger and N. Monod, Bounded cohomology of lattices in higher rank Lie groups, J. Eur. Math. Soc. (JEMS), 1, (1999), no. 2, 199–235. ———, Continuous bounded cohomology and applications to rigidity theory, Geom. Funct. Anal. 12 (2002), no. 2, 219–280. M. Burger and S. Mozes, CAT(−1)-spaces, divergence groups and their commensurators, J. Amer. Math. Soc. 9 (1996), no. 1, 57–93. J. Cheeger and M. Gromov, L2 -cohomology and group cohomology, Topology 25 (1986), no. 2, 189–215. P.-A. Cherix, M. Cowling, P. Jolissaint, P. Julg, and A. Valette, Groups with the Haagerup property, Progress in Mathematics, vol. 197 Birkhäuser Verlag, Basel, 2001. Gromov’s a-T-menability. I. Chifan and A. Ioana, Ergodic subequivalence relations induced by a Bernoulli action (2008), available at arXiv:0802.2353. A. Connes, A factor of type II1 with countable fundamental group, J. Operator Theory 4 (1980), no. 1, 151–153. A. Connes, J. Feldman, and B. Weiss, An amenable equivalence relation is generated by a single transformation, Ergod. Th. & Dynam. Sys. 1 (1981), no. 4, 431–450 (1982). A. Connes and B. Weiss, Property T and asymptotically invariant sequences, Israel J. Math. 37 (1980), no. 3, 209–210. K. Corlette and R. J. Zimmer, Superrigidity for cocycles and hyperbolic geometry, Internat. J. Math. 5 (1994), no. 3, 273–290. Y. de Cornulier, Y. Stalder, and A. Valette, Proper actions of lamplighter groups associated with free groups, C. R. Math. Acad. Sci. Paris 346 (2008), no. 3–4, 173–176 (English, with English and French summaries). M. Cowling and U. Haagerup, Completely bounded multipliers of the Fourier algebra of a simple Lie group of real rank one, Invent. Math. 96 (1989), no. 3, 507–549. M. Cowling and R. J. Zimmer, Actions of lattices in Sp(1,n), Ergod. Th. & Dynam. Sys. 9 (1989), no. 2, 221–237. H. A. Dye, On groups of measure preserving transformation. I, Amer. J. Math. 81 (1959), 119–159. ———, On groups of measure preserving transformations. II, Amer. J. Math. 85 (1963), 551–576. B. Eckmann, Introduction to l2 -methods in topology: reduced l2 -homology, harmonic chains, l2 -Betti numbers, Israel J. Math. 117 (2000), 183–219. Notes prepared by Guido Mislin. I. Epstein, Orbit inequivalent actions of non-amenable groups (2007), available at arXiv:0707.4215.

370 / a. furman [37] [38] [39] [40] [41] [42] [43] [44] [45] [46] [47] [48] [49] [50] [51] [52]

[53] [54] [55] [56] [57] [58] [59] [60] [61]

B. Farb, The quasi-isometry classiﬁcation of lattices in semisimple Lie groups, Math. Res. Lett. 4 (1997), no. 5, 705–717. B. Farb and S. Weinberger, The intrinsic asymmetry and inhomogeneity of Teichmuller space (2008), available at arXiv:0804.4428. J. Feldman and C. C. Moore, Ergodic equivalence relations, cohomology, and von Neumann algebras. I, Trans. Amer. Math. Soc. 234 (1977), no. 2, 289–324. ———, Ergodic equivalence relations, cohomology, and von Neumann algebras. II, Trans. Amer. Math. Soc. 234 (1977), no. 2, 325–359. J. Feldman, C. E. Sutherland, and R. J. Zimmer, Subrelations of ergodic equivalence relations, Ergod. Th. & Dynam. Sys. 9 (1989), no. 2, 239–269. D. Fisher, Groups acting on manifolds: around the Zimmer program, in this volume. D. Fisher and T. Hitchman, Cocycle superrigidity and harmonic maps with inﬁnite dimensional targets (2005), available at arXiv:math/0511666v3. A. Furman, Gromov’s measure equivalence and rigidity of higher rank lattices, Ann. of Math. (2), 150 (1999), no. 3, 1059–1081. ———, Orbit equivalence rigidity, Ann. of Math. (2) 150 (1999), no. 3, 1083–1108. ———, Mostow-margulis rigidity with locally compact targets, Geom. Funct. Anal. 11 (2001), no. 1, 30–59. ———, Outer automorphism groups of some ergodic equivalence relations, Comment. Math. Helv. 80 (2005), no. 1, 157–196. ———, On Popa’s cocycle superrigidity theorem, Int. Math. Res. Not. IMRN 19 (2007), Art. ID rnm073, 46. ———, Cohomology of measurable cocycles, in preparation. H. Furstenberg, Poisson boundaries and envelopes of discrete groups, Bull. Amer. Math. Soc. 73 (1967), 350–356. ———, Ergodic behavior of diagonal measures and a theorem of Szemerédi on arithmetic progressions, J. Analyse Math. 31 1977, 204–256. ———, Rigidity and cocycles for ergodic actions of semisimple Lie groups (after G. A. Margulis and R. Zimmer), Bourbaki Seminar, Vol. 1979/80, Lecture Notes in Math., vol. 842, Springer, Berlin, 1981, pp. 273–292. D. Gaboriau, Mercuriale de groupes et de relations, C. R. Acad. Sci. Paris Sér. I Math. 326 (1998), no. 2, 219–222. ———, Coût des relations d’équivalence et des groupes, Invent. Math. 139 (2000), no. 1, 41–98. ———, Sur la (co-)homologie L2 des actions préservant une mesure, C. R. Acad. Sci. Paris Sér. I Math. 330 (2000), no. 5, 365–370. ———, On orbit equivalence of measure preserving actions, Rigidity in dynamics and geometry (Cambridge, 2000), Springer, Berlin, 2002, pp. 167–186. ———, Invariants l 2 de relations d’équivalence et de groupes, Publ. Math. Inst. Hautes Etudes Sci. 95, (2002), 93–150. ———, Examples of groups that are measure equivalent to the free group, Ergod. Th. & Dynam. Sys. 25 (2005), no. 6, 1809–1827. D. Gaboriau and S. Popa, An uncountable family of non-orbit equivalent actions of Fn , J. Amer. Math. Soc. 18 (2005), no. 3, 547–559 (electronic). D. Gaboriau and R. Lyons, A measurable-group-theoretic solution to von Neumann’s problem (2007), available at arXiv:0711.1643. S. L. Gefter, Ergodic equivalence relation without outer automorphisms, Dopov./Dokl. Akad. Nauk Ukraïni, no. 11 (1993), 25–27.

a survey of measured group theory / 371 [62]

[63]

[64] [65] [66] [67] [68] [69] [70] [71] [72] [73] [74] [75] [76] [77] [78] [79] [80] [81] [82] [83] [84] [85]

———, Outer automorphism group of the ergodic equivalence relation generated by translations of dense subgroup of compact group on its homogeneous space, Publ. Res. Inst. Math. Sci. 32 (1996), no. 3, 517–538. ———, On cohomologies of ergodic actions of a T-group on homogeneous spaces of a compact Lie group (Russian), Operators in functional spaces and questions of function theory, Collect. Sci. Works, 1987, pp. 77–83. S. L. Gefter and V. Ya. Golodets, Fundamental groups for ergodic actions and actions with unit fundamental groups, Publ. Res. Inst. Math. Sci. 24 (1988), no. 6, 821–847 (1989). M. Gromov, Volume and bounded cohomology, Inst. Hautes Études Sci. Publ. Math. 56 (1982), 5–99 (1983). ———, Asymptotic invariants of inﬁnite groups, Geometric group theory, Vol. 2 (Sussex, 1991), 1993, pp. 1–295. P. de la Harpe, Topics in geometric group theory, Chicago Lectures in Mathematics, University of Chicago Press, Chicago, 2000. G. Hjorth, A converse to Dye’s theorem, Trans. Amer. Math. Soc. 357 (2005), no. 8, 3083–3103 (electronic). ———, A lemma for cost attained, Ann. Pure Appl. Logic 143 (2006), no. 1-3, 87–102. G. Hjorth and A. S. Kechris, Rigidity theorems for actions of product groups and countable Borel equivalence relations, Mem. Amer. Math. Soc. 177 (2005), 833, i–109. A. Ioana, A construction of non-orbit equivalent actions of F2 (2006), available at arXiv:math/0610452. ———, Cocycle superrigidity for proﬁnite actions of property (T) groups (2008), available at arXiv:0805.2998. ———, Orbit inequivalent actions for groups containing a copy of F2 (2007), available at arXiv:0701.5027. A. Ioana, A. S. Kechris, and T. Tsankov, Subequivalence relations and positive-deﬁnite functions (2008), available at arXiv:0806.0430. A. Ioana, J. Peterson, and S. Popa, Amalgamated free products of weakly rigid factors and calculation of their symmetry groups, Acta Math. 200, (2008), no. 1, 85–153. P. Jolissaint, Approximation properties for measure equivalent groups, 2001. preprint. V. F. R. Jones and K. Schmidt, Asymptotically invariant sequences and approximate ﬁniteness, Amer. J. Math. 109 (1987), no. 1, 91–114. V. A. Kaimanovich, Amenability, hyperﬁniteness, and isoperimetric inequalities, C. R. Acad. Sci. Paris Sér. I Math. 325 (1997), no. 9, 999–1004. D. A. Každan, On the connection of the dual space of a group with the structure of its closed subgroups, Funkcional. Anal. i Priložen. 1 (1967), 71–74 Russian. A. S. Kechris, Global aspects of ergodic group actions, Mathematical surveys and monographs, Vol. 160, American Mathematical Society, 2010. to appear. A. S. Kechris, B. D. Miller, Topics in orbit equivalence, Lecture Notes in Mathematics, Vol. 1852, Springer-Verlag, Berlin, 2004. A. S. Kechris and T. Tsankov, Amenable actions and almost invariant sets, Proc. Amer. Math. Soc. 136 (2008), no. 2, 687–697 (electronic). Y. Kida, The mapping class group from the viewpoint of measure equivalence theory, Mem. Amer. Math. Soc. 196 (2008), no. 916, i–190. ———, Classiﬁcation of the mapping class groups up to measure equivalence, Proc. Japan Acad. Ser. A Math. Sci. 82 (2006), no. 1, 4–7. ———, Orbit equivalence rigidity for ergodic actions of the mapping class group, Geom Dedicata 131 (2008 February), no. 1, 99–109.

372 / a. furman [86] [87] [88] [89] [90] [91] [92]

[93]

[94] [95]

[96] [97] [98] [99] [100] [101] [102]

[103] [104] [105] [106]

———, Measure equivalence rigidity of the mapping class group, arXiv math.GR (2006 July), available at arXiv:0607600. ———, Outer automorphism groups of equivalence relations for mapping class group actions, J. Lond. Math. Soc. (2) 78 (2008), no. 3, 622–638. ———, Rigidity of amalgamated free products in measure equivalence theory (2009), available at arXiv:0902.2888. M. Lackenby, Heegaard splittings, the virtually Haken conjecture and property τ (2002), available at arXiv:0205327. G. Levitt, On the cost of generating an equivalence relation, Ergod. Th. & Dynam. Sys. 15 (1995), no. 6, 1173–1181. A. Lubotzky and R. J. Zimmer, Variants of Kazhdan’s property for subgroups of semisimple groups, Israel J. Math. 66 (1989), no. 1-3, 289–299. W. Lück, L2 -invariants: theory and applications to geometry and K -theory, Ergebnisse der Mathematik und ihrer Grenzgebiete. 3. Folge. A series of modern surveys in mathematics [Results in Mathematics and Related Areas. 3rd Series. A Series of Modern Surveys in Mathematics], Vol. 44, Springer-Verlag, Berlin, 2002. G. A. Margulis, Discrete groups of motions of manifolds of nonpositive curvature, Proceedings of the International Congress of Mathematicians (Vancouver, BC, 1974), Vol. 2, Canad. Math. Congress, Montreal, Que. bec, 1975, pp. 21–34 (Russian). ———, Factor groups of discrete subgroups and measure theory, Funktsional. Anal. i Prilozhen. 12 (1978), no. 4, 64–76 (Russian). ———, Discrete subgroups of semisimple Lie groups, Ergebnisse der Mathematik und ihrer Grenzgebiete (3) [Results in Mathematics and Related Areas (3)], Vol. 17 SpringerVerlag, Berlin, 1991. I. Mineyev, N. Monod, and Y. Shalom, Ideal bicombings for hyperbolic groups and applications, Topology 43 (2004), no. 6, 1319–1344. N. Monod, Continuous bounded cohomology of locally compact groups, Lecture Notes in Mathematics, Vol. 1758, Springer-Verlag, Berlin, 2001. ———, An invitation to bounded cohomology, International Congress of Mathematicians. Vol. II, Eur. Math. Soc., Zürich, 2006, pp. 1183–1211. N. Monod and Y. Shalom, Negative curvature from a cohomological viewpoint and cocycle superrigidity, C. R. Math. Acad. Sci. Paris 337 (2003), no. 10, 635–638. ———, Cocycle superrigidity and bounded cohomology for negatively curved spaces, J. Differential Geom. 67 (2004), no. 3, 395–455. ———, Orbit equivalence rigidity and bounded cohomology, Ann. of Math. (2) 164 (2006), no. 3, 825–878. C. C. Moore, Ergodic theory and von Neumann algebras, Operator algebras and applications, Part 2, Kingston, Ontario, 1980), Proc. Sympos. Pure Math., Vol. 38, Amer. Math. Soc., Providence, RI, 1982, pp. 179–226. G. D. Mostow, Strong rigidity of locally symmetric spaces, Princeton University Press, Princeton, NJ, 1973. Annals of Mathematics Studies, no. 78. F. J. Murray and J. von Neumann, On rings of operators. IV, Ann. of Math. (2) 44 (1943), 716–808. D. S. Ornstein and B. Weiss, Ergodic theory of amenable group actions. I. The Rohlin lemma, Bull. Amer. Math. Soc. (N.S.) 2 (1980), no. 1, 161–164. S. Popa, Correspondences, Prépublication Institul Natinoal Pentru Creatie Stiintica si Tehnica (1986).

a survey of measured group theory / 373 [107] N. Ozawa, A Kurosh-type theorem for type II1 factors, Int. Math. Res. Not., (2006), Art. ID 97560, 21. [108] N. Ozawa and S. Popa, On a class of II1 factors with at most one Cartan subalgebra II (2008), available at arXiv:0807.4270v2. [109] S. Popa, On a class of type II1 factors with Betti numbers invariants, Ann. of Math. (2) 163 (2006), no. 3, 809–899. [110] ———, Some computations of 1-cohomology groups and construction of non-orbit-equivalent actions, J. Inst. Math. Jussieu 5 (2006), no. 2, 309–332. [111] ———, Strong rigidity of II1 factors arising from malleable actions of w-rigid groups. I, Invent. Math. 165 (2006), no. 2, 369–408. [112] ———, Strong rigidity of II1 factors arising from malleable actions of w-rigid groups. II, Invent. Math. 165 (2006), no. 2, 409–451. [113] ———, Cocycle and orbit equivalence superrigidity for malleable actions of w-rigid groups, Invent. Math. 170 (2007), no. 2, 243–295. [114] ———, Deformation and rigidity for group actions and von Neumann algebras, International Congress of Mathematicians. Vol. I, Eur. Math. Soc., Zürich, 2007, pp. 445–477. [115] ———, On the superrigidity of malleable actions with spectral gap, J. Amer. Math. Soc. 21 (2008), no. 4, 981–1000. [116] S. Popa and R. Sasyk, On the cohomology of Bernoulli actions, Ergod. Th. & Dynam. Sys. 27 (2007), no. 1, 241–251. [117] S. Popa and S. Vaes, Strong rigidity of generalized Bernoulli actions and computations of their symmetry groups, Adv. Math. 217 (2008), no. 2, 833–872. [118] ———, Actions of F∞ whose II1 factors and orbit equivalence relations have prescribed fundamental group (2008), available at arXiv:0803.3351. [119] ———, Cocycle and orbit superrigidity for lattices in SL (n, R) acting on homogeneous spaces, in this volume. [120] ———, Group measure space decomposition of II1 factors and W ∗ -superrigidity, available at arXiv:0906.2765. [121] M. Ratner, Interactions between ergodic theory, Lie groups, and number theory, 2, (Zürich, 1994), Birkhäuser, Basel, 1995, pp. 157–182. [122] H. Sako, The class S is an ME invariant, Int. Math. Res. Not. (2009), available at arXiv:0901.3374. rnp025. [123] ———, Measure equivalence rigidity and bi-exactness of groups, J. Funct. Anal. (2009), available at arXiv:0901.3376. to appear. [124] R. Sauer, L2 -Betti numbers of discrete measured groupoids, Internat. J. Algebra Comput. 15 (2005), 5-6, 1169–1188. [125] ———, Homological invariants and quasi-isometry, Geom. Funct. Anal. 16 (2006), no. 2, 476–515. [126] ———, Amenable covers, volume and l 2 -Betti numbers of aspherical manifolds (2006), available at arXiv:math/0605627. [127] R. Sauer and A. Thom, A Hochschild-Serre spectral sequence for extensions of discrete measured groupoids (2007), available at arXiv:0707.0906. [128] K. Schmidt, Asymptotically invariant sequences and an action of SL(2, Z) on the 2-sphere, Israel J. Math. 37 (1980), no. 3, 193–208. [129] ———, Amenability, Kazhdan’s property T , strong ergodicity and invariant means for ergodic group-actions, Ergod. Th. & Dynam. Sys. 1 (1981), no. 2, 223–236.

374 / a. furman [130] Y. Shalom, Rigidity of commensurators and irreducible lattices, Invent. Math. 141 (2000), no. 1, 1–54. [131] ———, Rigidity, unitary representations of semisimple groups, and fundamental groups of manifolds with rank one transformation group, Ann. of Math. (2) 152 (2000), no. 1, 113–182. [132] ———, Harmonic analysis, cohomology, and the large-scale geometry of amenable groups, Acta Math. 192 (2004), no. 2, 119–185. [133] ———, Measurable group theory, European Congress of Mathematics, Eur. Math. Soc., Zürich, 2005, pp. 391–423. [134] I. M. Singer, Automorphisms of ﬁnite factors, Amer. J. Math. 77 (1955), 117–133. [135] G. Stuck and R. J. Zimmer, Stabilizers for ergodic actions of higher rank semisimple groups, Ann. of Math. (2) 139 (1994), no. 3, 723–747. [136] S. Vaes, Rigidity results for Bernoulli actions and their von Neumann algebras (after Sorin Popa), Astérisque 311 (2007), Exp. No. 961, viii, 237–294. Séminaire Bourbaki. Vol. 2005/2006. [137] R. J. Zimmer, Extensions of ergodic group actions, Illinois J. Math. 20 (1976), no. 3, 373–409. [138] ———, Amenable ergodic group actions and an application to Poisson boundaries of random walks, J. Functional Analysis 27 (1978), no. 3, 350–372. [139] ———, Strong rigidity for ergodic actions of semisimple Lie groups, Ann. of Math. (2) 112 (1980), no. 3, 511–529. [140] ———, Orbit equivalence and rigidity of ergodic actions of Lie groups, Ergod. Th. & Dynam. Sys. 1 (1981), no. 2, 237–253. [141] ———, On the cohomology of ergodic actions of semisimple Lie groups and discrete subgroups, Amer. J. Math. 103 (1981), no. 5, 937–951. [142] ———, Ergodic theory, semisimple Lie groups, and foliations by manifolds of negative curvature, Inst. Hautes Études Sci. Publ. Math. 55, (1982), 37–62. [143] ———, Ergodic theory, group representations, and rigidity, Bull. Amer. Math. Soc. (N.S.) 6 (1982), no. 3, 383–416. [144] ———, Ergodic actions of semisimple groups and product relations, Ann. of Math. (2) 118 (1983), no. 1, 9–19. [145] ———, Kazhdan groups acting on compact manifolds, Invent. Math. 75 (1984), no. 3, 425–436. [146] ———, Ergodic theory and semisimple groups, Monographs in Mathematics, Vol. 81, Birkhäuser Verlag, Basel, 1984. [147] ———, Groups generating transversals to semisimple Lie group actions, Israel J. Math. 73 1991, no. 2, 151–159.

11

alessandra iozzi

ON RELATIVE PROPERTY (T)

to bob zimmer on his

60 th

birthday

Abstract We present families of pairs (H A,A) with relative property (T), where H is a locally compact group acting continuously by automorphisms on a locally compact abelian group A. The paper is completely self-contained.

1. Introduction Property (T) for groups and for pairs (also referred to as relative property (T)) was ﬁrst introduced by D. Kazhdan in his thesis [14] to show that lattices in semisimple Lie groups are ﬁnitely generated. His proof that SL(n, R) has property (T) for n ≥ 3 follows from the fact that the pair (SL(2, R) R2 , R2 ) has the relative property (T). Recall that a unitary representation π : G → U (H) of a locally compact group G almost has invariant vectors if for every compact subset K ⊂ G and for every > 0 there exists ξ ∈ H with ξ = 1 such that sup π (g)ξ − ξ < . g∈K

Equivalently, one says that the representation π weakly contains the trivial representation 11G . Then a group G has property (T) if any representation that almost has invariant vectors actually has invariant vectors. A typical example of a group with property (T) is a connected semisimple real Lie group all of whose factors have real rank at least 2, for example SL(n, R) for n ≥ 3, or any lattice therein.

definition 1.1. Let L < G be a subgroup of a locally compact group G. The pair (G, L) has relative property (T) if whenever (π , H) is a continuous unitary representation of G, which almost has invariant vectors; there are nonzero L-invariant vectors. Partially supported by FNS grant PP002-102765.

375

376 / a. iozzi

Property (T) has several diverse applications in representation theory, ergodic theory, operator algebras, lattices in algebraic groups over local ﬁelds, geometric group theory, and the theory of networks. Aside the implicit use of relative property (T) in [14], among its ﬁrst applications there is the construction of an inﬁnite family of expanders [15], or the solution of Ruziewicz problem for Rn , [16], both due to Margulis. Further applications include Gaboriau and Popa’s construction [11] of a noncountable family of free ergodic measure-preserving non-orbit equivalent actions of the free group Fr in r ≥ 2 generators on a standard probability space (see also [21] or [10] for generalizations) or Popa’s construction [20, 22, 23] of a factor of type II1 with trivial fundamental group. Finally, although the fact that a pair (G, L) has relative property (T) is a weaker condition than requiring that the group G itself has property (T), in some situations relative property (T) will sufﬁce; for example, Navas [19] has proven that for the nonexistence of “interesting” C2 -actions on the circle, the relative property (T) of a pair (, 0 ), where 0 is normal in , will sufﬁce. Just to give a ﬂavor of how relative property (T) differs from property (T), although we will not use these facts here, observe that if (G, L) has relative property (T) this does not imply that either G or L have property (T), and moreover, (SL(2, Z) Z2 , Z2 ) has relative property (T) but neither SL(2, Z) nor Z2 have property (T). Furthermore, the fact that (G, L) has relative property (T) does not imply that either G or L is compactly generated, as the example of (SL(3, Z) × , F∞ × {e}), where is any group that is not ﬁnitely generated shows.1 Just as for property (T), several different characterizations of relative property (T) are available, for example, in terms of strongly ergodic actions, in terms of Von Neumann algebras, of positive deﬁnite functions, of isometric actions of Hilbert spaces, and so on; moreover, it is possible to deﬁne relative property (T) for pairs (G, X ) where X is any subset [6]. We refer the reader to [13] and [6] and to the references therein. It goes without saying that if either G or L is a group with property (T), then the pair (G, L) has the relative property (T); nontrivial examples of pairs with relative property (T) have been constructed in [5], in [8] (see Theorem 1.3 and the comments thereafter) or in [25] (see the last paragraph of this introduction). The scope of this note is to present families of pairs (G, L) with relative property (T) whose general form is the semidirect product G = H A of a 1. Note, however, that if (G, L) has relative property (T), there exists a compactly generated subgroup H containing L such that (H, L) has relative property (T); see [6].

on relative property (t) / 377

locally compact group H acting continuously by automorphisms on a locally compact abelian group A and L = A. 1.1.

Let be a discrete group, S a ﬁnite set of prime numbers, Z[S] the ring obtained inverting the primes in S and ρ : → GLN (Z[S]) a homomorphism. We will denote by ρp : → GLN (Qp ) the representation obtained by injecting Z[S] into Qp , where Q∞ := R.

theorem 1.2. Assume that the image ρ() of is Zariski dense in SLN . Then the following are equivalent: i) the pair ( Z[S]N , Z[S]N ) has relative property (T); ii) the Z[]-module Z[S]N is ﬁnitely generated; iii) for every p ∈ S ∪ {∞} the image ρp () < SLN (Qp ) is not bounded; and ∗ iv) there is no ρp∗ ()-invariant probability measure on P((QN p ) ), for p ∈ S ∪ {∞}. We remark that (ii) is in contrast with the fact that, unless S = ∅, the ring Z[S] is not ﬁnitely generated as a Z-module; moreover, it is remarkable that although the relative property (T) is analytic in nature, the equivalent property in (ii) above is purely algebraic. As it will be clear from the proof (see Proposition 4.5) Zariski density is not needed for the equivalence of the assertions (i) and (iv). The motivation to establish results of the type of Theorem 1.2 comes from [8], where the following is shown:

theorem 1.3. (fernós, [8]) Let be a ﬁnitely generated group. Then the following are equivalent: i) there exists a representation ρ : → SLn (R) such that the real points Z ρ() (R) of the Zariski closure of the image of ρ is not an amenable group; and ii)there exists a ﬁnite set of primes S, an integer N ∈ N, and a representation → SLN (Z[S]) such that ( Z[S]N , Z[S]N ) has relative property (T).

example 1.4. 1) According to the Tits alternative, if k is any local ﬁeld of zero characteristic and < GL(n, k) is a ﬁnitely generated subgroup, then either is

378 / a. iozzi

virtually solvable or it contains a free subgroup F2 in two generators that is Zariski dense in . Since = SLN (Z) is not virtually solvable, then there is a Zariski-dense F2 < SLN (Z) and thus (F2 ZN , ZN ) has relative property (T). 2) A topological version of the Tits alternative shown by Breuillard and Gelander [4] asserts that if ki are local ﬁelds of zero characteristic and < GL(n, ki ) is a ﬁnitely generated subgroup, then either contains an open solvable subgroup, or a ﬁnitely generated free subgroup that is dense in . Using this result we show the following:

corollary 1.5. For every natural number N ∈ N and every nonempty ﬁnite set of primes S, there is a ﬁnitely generated free group < SLN (Z[S]) such that the pair ( Z[S]N , Z[S]N ) has relative property (T). 1.2.

Now let ρ : G → GL(V ) be a rational representation deﬁned over Q, where G is a connected, semisimple Q-group. Let ρR : GR → GL(VR ) be the representation on the level of real points. We say that ρR is totally unbounded if for any GR -invariant subspace WR ⊂ VR of positive dimension, the group ρ(GR )|WR < GL(WR ) is unbounded.

theorem 1.6. Let ⊂ VQ be a Z-module of maximal rank invariant under GZ . Let < GZ and assume that is Zariski dense in G. The following are equivalent: i) the pair ( , ) has relative property (T); and ii) the representation ρR : GR → GL(VR ) is totally unbounded. A remark is in order regarding the hypothesis of Zariski density of in G. There is a result of Borel [3, theorem 1] to the extent that if G contains no connected normal Q-subgroups N = {e} such that NR is compact, then GZ is Zariski dense in G; this avoids the situation in which if G = G1 × G2 with Gi Q-groups, then G1,Z × G2,Z = GZ ⊂ GR = G1,R × G2,R with G2,R compact, which would prevent the Zariski density of GZ . So Theorem 1.6 applies to any ﬁnite-index subgroup of GZ , provided GR has no compact factors. If ρR is irreducible with unbounded image, it is trivially totally unbounded and hence the pair ( , ) has relative property (T). The construction of such a representation for a group deﬁned over Q, which is either adjoint or

on relative property (t) / 379

simply connected will be given in Lemma 4.7. As an application we have the following:

corollary 1.7. If G is a connected real algebraic semisimple Lie group without compact factors and < G is an arithmetic lattice, then there is a linear representation → SLN (Z) of a ﬁnite-index subgroup < such that ( ZN , ZN ) has relative property (T). Theorem 1.6 is motivated by [25, theorems 1 and 4], where the same assertion as in Corollary 1.7 is proven, under the additional hypothesis that the group G is absolutely simple. Corollary 1.7 can also be deduced from [8, theorems 3 and 7.1] by observing that such an arithmetic lattice satisﬁes property (F∞ ) in [8]. Just like in [25] the value of the integer N is explicitly given (see the proof of Lemma 4.7). Moreover, if < is of ﬁnite-index k := [, ] and ( ZN , ZN ) has relative property (T), then ( ZkN , ZkN ) has relative property (T), [8, step B in the proof of theorem 7.1].

Acknowledgments: The author thanks Marc Burger for useful conversations during the preparation of this paper and the referee for helpful comments. 2. Generalities We collect here, with proofs, few classical facts about relative property (T). We start by recording here the following observation:

lemma 2.1. Let Hi , Bi be locally compact groups such that Hi acts on Bi via a continuous action Hi → Aut (Bi ), i = 1, 2, and let q : H1 B1 → H2 B2 be a continuous homomorphism such that q(B1 ) = B2 . If (H1 B1 , B1 ) has relative property (T), then also (H2 B2 , B2 ) has relative property (T). lemma 2.2. Let H, A1 , A2 , . . . , An be locally compact groups such that H acts on Aj via a continuous action H → Aut (Aj ). The following are equivalent: i) the pair (H (A1 × · · · × An ), A1 × · · · × An ) has relative property (T); and ii) the pair (H Aj , Aj ) has relative property (T) for j = 1, . . . , n. Proof. The implication (i)⇒(ii) follows from Lemma 2.1 by considering the homomorphisms H (A1 × · · · × An ) → H Aj , while (ii)⇒(i) is in [8, lemma 5.2], where it is used that relative property (T) is a property closed under certain extensions.

380 / a. iozzi

The following corollary is essential in the proof of Proposition 4.5.

corollary 2.3. Let H and M be locally compact groups, < M a cocompact lattice and let ρ : H → Aut (M) be a continuous action by automorphisms preserving . Then the pair (H , ) has relative property (T) if and only if the same holds for the pair (H M, M). Although the above result is all we need, we are going to prove the following more general result from which the corollary can be obtained at once by setting L := H and A := M H M =: G.

proposition 2.4. Let G be a locally compact group, A G a normal subgroup, and L < G a closed subgroup. 1) If (G, A) has relative property (T) and L\G has a ﬁnite G-invariant measure, then the pair (L, L ∩ A) also has relative property (T). 2) If A/L ∩ A is compact with ﬁnite A-invariant measure and (L, L ∩ A) has relative property (T), then (G, A) has relative property (T). In order to prove this, we will use a strengthening of relative property (T). We recall ﬁrst that if π : G → U (H) is a unitary representation of a locally compact group G, we say that a sequence {ξn } ⊂ H of vectors is asymptotically π(G)-invariant if ξn = 1 for all n and for every compact subset K ⊂ G we have that 2.1

lim sup π (k)ξn − ξn = 0 .

n→∞

k∈K

If a group is σ -compact, a representation almost has invariant vectors if and only if it has a sequence of asymptotically invariant vectors. Denote by HG the subspace of H consisting of π (G)-invariant vectors. Then it is well known that if G is a group with property (T) and π : G → U (H) is a unitary representation of G with a sequence {ξn } of asymptotically π(G)-invariant vectors, then lim d(ξn , HG ) = 0,

n→∞

Where d denotes the distance to a subspace. Similarly one has the following:

lemma 2.5. Let A G be a normal subgroup of a locally compact group G such that the pair (G, A) has relative property (T). If π : G → U (H) is a continuous

on relative property (t) / 381

unitary representation and {ξn } ⊂ H is a sequence of asymptotically π (G)-invariant vectors, then lim d(ξn , HA ) = 0 .

n→∞

Proof. Since A is normal in G, the Hilbert subspace HA is π (G)-invariant, so that H = HA ⊕ (HA )⊥

is an orthogonal decomposition into π (G)-invariant subspaces. Let {ξn } ⊂ H be a sequence of asymptotically invariant vectors and let ξn = ζn + ζn be the corresponding orthogonal decomposition, so that showing that limn→∞ d(ξn , HA ) = 0 is equivalent to showing that limn→∞ ζn = 0. Let us assume by contradiction that lim supn→∞ ζn = 0, that is, for some > 0, let {ζn k } ∈ (HA )⊥ be a subsequence such that 2.2

for all k ∈ N we have ζn k ≥ .

Since (HA )⊥ is π(G)-invariant and the orthogonal projection is norm decreasing, we have that for all g ∈ G π (g)ζn k − ζn k ≤ π (g)ξnk − ξnk from which, using Equation 2.2, we obtain that ζnk ζn k 1 − ≤ π (g)ξnk − ξnk . π(g) ζnk ζnk If now K ⊂ G is any compact set, since the sequence {ξn } ⊂ H is asymptotically π(G)-invariant, we have ζnk ζn k 1 lim sup π(g) − ≤ lim sup π (g)ξnk − ξnk = 0 , ζnk ζnk k→∞ k∈K k→∞ k∈K from which we deduce that the sequence {ζn k /ζn k } is also asymptotically π (G)-invariant. Hence π (G)|(HA )⊥ has non-zero π (A)-invariant vectors, which is a contradiction.

382 / a. iozzi

Proof of Proposition 2.4. 1) Let (π, H) be a representation of L that almost has invariant vectors. Since weak containment is preserved under induction, the representation ω := G IndG L (π ) of G induced from π weakly contains the representation IndL (11L ) induced to G from the identity representation of L; the latter in turn has invariant vectors since L\G has ﬁnite G-invariant measure and hence, by transitivity of weak containment, ω almost has invariant vectors. Thus, since (G, A) has relative property (T), the Hilbert space Lω := {f : G → H : f is measurable, f (g) = π ()(f (g)),

for almost every g ∈ G and for all ∈ L f (g)2 dμ(g) < ∞} and L\G

of the representation ω has nonzero A-invariant vectors with respect to the action (ω(g)f )(x) = f (xg) for all g ∈ G and almost all x ∈ G. Let f ∈ Lω be such nonzero A-invariant vector. Then for all a ∈ A ∩ L and almost all x ∈ G we have π (a)(f (x)) = f (ax) = f (x(x −1 ax)) = f (x) , which shows that the space of (L ∩ A)-invariant vectors in H is not trivial and hence shows (1). 2) Let π : G → U (H) be a representation of G and {ξn } ∈ H be a sequence of asymptotically π (G)-invariant vectors.2 Then {ξn } is also asymptotically π (L)invariant and since by hypothesis (L, L ∩ A) has relative property (T) and L ∩ A is normal in L, by Lemma 2.5 there exists a sequence {ζn } ∈ HL∩A such that limn→∞ ξn − ζn = 0. After rescaling if necessary, we may assume that ζn = 1. For all n ∈ N the vectors ηn :=

A/A∩L

π (a)ζn dλ(a) ∈ H ,

where λ is the A-invariant probability measure on A/A ∩ L, are obviously π(A)invariant, and we only need to show that there exists an n ∈ N such that ηn = 0. 2. For the sake of simplicity we assume in the sequel that all locally compact groups are σ -compact. If not, the same arguments apply by replacing sequences with generalized sequences.

on relative property (t) / 383

To see this, let F ⊂ A ⊂ G be a compact fundamental domain for the action of A ∩ L on A. Then for all a ∈ A there exists fa ∈ F and a ∈ A ∩ L such that a = fa a , so that, by π(L ∩ A)-invariance of ζn , ηn − ζn =

A/A∩L

π (a)ζn dλ(a) − ζn

≤

A/A∩L

π (a)ζn − ζn dλ(a)

≤ sup π (fa )ζn − ζn . fa ∈F

Now we observe that {ζn } is asymptotically π (G)-invariant: indeed for all g ∈ G, π (g)ζn − ζn ≤ 2ζn − ξn + π (g)ξn − ξn . It follows that limn→∞ ηn − ζn = 0 and thus in particular ηn = 0 for some n ∈ N.

lemma 2.6. Let A be a semidirect product of discrete groups, with A abelian, and assume that ( A, A) has relative property (T). Then the Z[]-module A is ﬁnitely generated. Proof. Let An ↑ A be an increasing sequence of -invariant subgroups ﬁnitely generated over Z[] and ∪An = A. Consider the A-regular action on

2 ( A/ An ) ;

n≥1

the sequence δe( An ) is asymptotically invariant, hence there is an A-invariant vector in some 2 ( A/ An ); this implies that there is a ﬁnite A-orbit in ( A/ An ) ∼ = A/An and hence |A/An | < +∞.

3. Algebraic Actions and Measure Theory We start by recalling the classical fact that if k is a local ﬁeld and Hk consists of the k-points of an algebraic group deﬁned over k, then any k-algebraic action of Hk on the k-points of a k-variety has orbits, which are locally closed in the Hausdorff topology [2]. Recall, moreover, that an action is almost effective if the intersection of the stabilizers is a ﬁnite group.

384 / a. iozzi

At the heart of what we are doing is the following theorem of Gromov: the proof we recall here is more elementary than the original one and appears in [1, théorème 6.5].

theorem 3.1. (gromov) Let k be a local ﬁeld, H a k-algebraic group, W a k-algebraic variety and H × W → W a k-algebraic action. Let μ ∈ M1 (Wk ) be a probability measure on Wk such that the Zariski closure of its support is W , and assume that the action is almost effective. Then StabHk (μ) := {h ∈ Hk : h∗ μ = μ} is compact. Z

Z

Proof. Observe ﬁrst of all that since supp (μ) ⊂ Wk , then supp (μ) ⊂ Wk , which, together with the hypothesis on the support of μ implies that Wk is Zariski dense in W . This and the fact that the action is almost effective imply that 3.1

StabHk (x) =

x∈Wk

StabHk (x) is ﬁnite .

x∈W

The Noetherian property for the k-algebraic group StabHk (x) implies that there exist x1 , . . . , xn ∈ Wk such that 3.2

n j=1

StabHk (xj ) =

StabHk (x) .

x∈Wk

If we now let H act diagonally on W n and deﬁne O := {p ∈ W n : StabHk (p) is ﬁnite} ,

we deduce immediately from Equations 3.1 and 3.2 that O is not empty. Likewise it is easy to see that O is Zariski open and hence, if ν := μ⊗n , ν(Ok ) > 0. Since, moreover, Ok is Hk -invariant, then we conclude that ν|Ok is a ﬁnite measure left-invariant by StabHk (μ). Since Ok is open in WkN and the Hk -orbits in Wk are locally closed, then also the StabHk (μ)-orbits in Ok are locally closed in the Hausdorff topology. By decomposing ν|Ok if necessary into ergodic components and applying general considerations (see [26] or [12, 7]), we deduce that the ﬁnite StabHk

on relative property (t) / 385

(μ)-invariant measure on Ok is supported on an orbit StabHk (μ) · p0 , with p0 ∈ Ok . But StabHk (μ) · p0 ∼ = StabHk (μ)/ StabHk (p0 ) ∩ StabHk (μ) and, by hypothesis, StabHk (p0 ) is ﬁnite. It follows that StabHk (μ) supports a ﬁnite measure that is invariant by translations and hence is compact. We can draw from this theorem some useful consequences on the structure of the stabilizer of a probability measure in projective space. The point is to be able to deal with measures whose support is not necessarily Zariski dense and with actions that are not necessarily effective. To apply the previous theorem and for further reference, let us set up some notation. If k is a local ﬁeld, let E be a ﬁnite-dimensional vector space deﬁned over k. If μ ∈ M1 (P(Ek )), let us set 3.3

W := supp μZ ⊆ P(E)

to be the projective subvariety deﬁned as the Zariski closure of the support of μ and deﬁne the algebraic subgroups 3.4

N(W ) :={h ∈ PGL(E) : h(W ) = W } I(W ) :={h ∈ PGL(E) : h|W = Id W } ,

where I(W ) is normal in N(W ) and both are deﬁned over k. The above notation will be typically applied when E is either a vector space as in the next corollary or the dual of a vector space (the only difference of course is lying in the action).

corollary 3.2. Let Vk be a ﬁnite-dimensional k-vector space seen as the set of k-points of the corresponding vector space V over an algebraic closure of k. The stabilizer StabPGL(Vk ) (μ) of a probability measure μ ∈ M1 (P(Vk )) on the projective space P(Vk ) has a cocompact normal k-subgroup. More precisely, if W ⊂ P(V ) and I(W ) are deﬁned as in Equations (3.3) and (3.4), respectively, then I(W ) is a ksubgroup of PGL(V ) such that I(W )k is normal in StabPGL(Vk ) (μ) and the quotient StabPGL(Vk ) (μ)/I(W )k is compact. Proof. By construction the group H := N(W )/I(W ) is deﬁned over k and so is the algebraic action H × W → W , which is in addition effective by construction. Since N(W )k /I(W )k is the N(W )k -orbit of the identity in (N(W )/I(W ))k = Hk , it is locally closed in the Hausdorff topology and, being a topological group,

386 / a. iozzi

is also closed; but StabPGL(Vk ) (μ)/I(W )k being closed in N(W )k /I(W )k it is also closed in Hk . Since the quotient StabPGL(Vk ) (μ)/I(W )k is closed and contained in the compact StabHk (μ), it is itself compact.

remark 3.3. Furstenberg’s lemma [9] asserts that either the stabilizer of a probability measure μ on P(Vk ) is compact or the support of the measure is contained in the union of two proper subspaces. Using this lemma Zimmer has shown [26, theorem 3.2.4] (see also [18]) that under the same hypotheses as in Corollary 3.2 StabPGL(Vk ) (μ) has a normal subgroup of ﬁnite index that contains a cocompact normal subgroup consisting of the k-points of a k-algebraic group. 4. Applications to Group Pairs Let Vk be as in §3 a k-vector space identiﬁed with the k-points of a vector space V over an algebraic closure of k, and let Vk∗ be its dual. If μ ∈ M1 (P(Vk∗ )), deﬁne 4.1

Gμ := {h ∈ GL(Vk ) : [h ∗ ] ∈ StabPGL(Vk∗ ) (μ)}

where h ∗ denotes the transpose of h. Then we have

proposition 4.1. Let μ ∈ M1 (P(Vk∗ )). Then the pair (Gμ Vk , Vk ) does not have relative property (T).

Proof. Let us consider W ⊂ P(V ∗ ) and I(W ) as deﬁned in Equations (3.3) and (3.4) with E = V ∗ and let us deﬁne Iμ := {h ∈ Gμ : [h ∗ ] ∈ I(W )k } = {h ∈ GL(Vk ) : [h ∗ ] ∈ I(W )k } . Let [λ] ∈ supp μ, where λ : Vk → k is a nonzero linear form. Then, by deﬁnition of Iμ , there exists a continuous homomorphism χλ : Iμ → k∗ such that for every h ∈ Iμ 4.2

h ∗ λ = χλ (h)λ.

on relative property (t) / 387

But this is equivalent to saying that the map qλ : Iμ Vk →

k∗ k

(h, v) → (χλ (h), λ(v)) is a homomorphism. Since k is not compact and k∗ k is amenable, then (k∗ k, k) does not have relative property (T) (see, for instance, [8, lemma 8.3]), and since qλ is continuous with qλ (Vk ) = k, then also (Iμ Vk , Vk ) does not have relative property (T) (Lemma 2.1). But Iμ is a normal subgroup of Gμ and Gμ /Iμ is compact by Corollary 3.2, so that Proposition 2.4(1) implies that also (Gμ Vk , Vk ) does not have relative property (T). Our source of examples is the following:

theorem 4.2. Let k be a local ﬁeld, Vk a ﬁnite-dimensional k-vector space endowed with its structure of (additive) locally compact group, H a locally compact group, and ρ : H → GL(Vk ) a continuous representation. Then the following are equivalent: i) the pair (H Vk , Vk ) has relative property (T); and ii) there is no H-invariant probability measure on P(Vk∗ ). The proof of the implication (ii)⇒(i) was proven in [5, proposition 7 and examples] and is recalled below for completeness. The case in which k = R and H is a semisimple connected Lie group was established in [24, proposition 2.3]. The implication (i)⇒(ii) in the general case appears in [6, proposition 3.1.9], but it follows here immediately from the above proposition. In fact, assume that via the contragredient action ρ ∗ of ρ, the group H ﬁxes a measure μ ∈ M1 (P(Vk∗ )). With Gμ as in Equation 4.1, consider the continuous homomorphism H Vk

ρ⊗Id

/ Gμ Vk .

Then Proposition 4.1 and the contrapositive of Lemma 2.1 imply that (H Vk , Vk ) does not have relative property (T). We recall now the proof of the implication (ii)⇒(i). Let ρ : H → GL(Vk ) be a continuous representation and let G := H Vk . By ﬁxing a nontrivial character χ ∈ Homc (k, T) we get the usual isomorphism as locally compact

388 / a. iozzi

k of Vk with the dual V ∗ , via groups of the Pontryagin dual V k k V

Vk∗ −→

λ −→ χ ◦ λ . Given a continuous unitary representation π : G → U (H), we have the spectral measure P : B(Vk∗ ) → P (H) of π|Vk , that is, a map associating to every Borel set B ⊂ Vk∗ an orthogonal projection satisfying certain additional properties. Essential is the fact that for every ξ ∈ H, μξ (B) := P(B)ξ , ξ deﬁnes a bounded positive Radon measure μξ ∈ M+ (Vk∗ ) on Vk∗ and π (v)ξ , ξ = μˆ ξ (v) = χ (λ(v)) dμξ (λ) . Vk∗

In other words, μξ is determined uniquely by the diagonal coefﬁcients on Vk associated to ξ and it is easy to see that P(B) is uniquely determined by the map ξ → μξ (B); furthermore, we have for all h ∈ H and B ∈ B(Vk∗ ) P(ρ(h)∗ B) = π(h)−1 P(B) π(h) .

4.3

Now let ξ ∈ H and μξ ∈ M+ (Vk∗ ) and deﬁne mξ := p∗ (μξ |Vk∗ \{0} ) to be the push- forward of the measure μξ |Vk∗ \{0} under the projection p : Vk∗ \ {0} → P(Vk∗ ). If B ∈ B(P(Vk∗ )) is a Borel subset in P(Vk∗ ) and we set B = p−1 (B), using Equation 4.3 we have mξ (ρ ∗ (h)B) − mξ (B) = P(ρ ∗ (h)B )ξ , ξ − P(B )ξ , ξ = P(B )π (h)ξ , π(h)ξ − P(B )ξ , ξ = Re P(B )(π(h)ξ + ξ ), π (h)ξ − ξ , so that |mξ (ρ ∗ (h)B) − mξ (B))| ≤ 2ξ π(h)ξ − ξ . Now for any subset K ⊂ H, introduce the following quantity α(K , ρ) :=

inf

sup |m(ρ ∗ (h)B) − m(B)|

sup

m∈M1 (P(Vk∗ )) B∈B(P(V ∗ )) h∈K k

on relative property (t) / 389

which is somehow the extent to which “measures are moved by ρ ∗ (h).” Since the total mass of the positive measure mξ is μξ (P(Vk∗ )) = μξ (Vk∗ ) − μξ ({0}) , and since there are no ρ(Vk )-invariant vectors if and only if μξ ({0}) = 0 for every ξ = 1, we obtain the following:

proposition 4.3. Let ρ : G → U (H) be a continuous unitary representation of H Vk with no nonzero Vk -invariant vectors. Then for every compact subset K ⊂ H and for every ξ ∈ H with ξ = 1, we have that max ρ(h)ξ − ξ ≥ h∈K

1 α(K , ρ) . 2

The proof of Theorem 4.2(ii)⇒(i) will be complete if we show that H has no invariant measure on P(Vk∗ ) via ρ if and only if there exists a compact set K ⊂ H with α(K , ρ) > 0; but this is just an exercise. 4.1. Proof of the Results in §1.1

We adopt the same notation as in §1.1, namely, is a discrete group, S a ﬁnite set of primes, Z[S] the ring obtained inverting the primes in S, ρ : → GLN (Z[S]) a representation, and ρp : → GLN (Qp ) is the representation obtained by composing ρ with the embedding Z[S] → Qp , where Q∞ := R. Since the diagonal embedding

Z[S]N → RN × QN p p∈S

realizes Z[S]N as a cocompact lattice, Corollary 2.3 can be translated into the following:

corollary 4.4. The pair ( Z[S]N , Z[S]N ) has relative property (T) if and N N only if the pair ( (RN × p∈S QN p ), R × p∈S Qp ) has relative property (T). From this we deduce the equivalence of (i) and (iv) in Theorem 1.2, which we record here separately as it does not require the hypothesis of Zariski density of ρ() in SLN .

proposition 4.5. With the above hypotheses, the following are equivalent: i) ( Z[S]N , Z[S]N ) has relative property (T); and

390 / a. iozzi

ii)for every p ∈ S ∪ {∞}, there is no ρp∗ ()-invariant probability measure on ∗ P((QN p ) ). Proof. Using Corollary 4.4 and Lemma 2.2 we deduce that the pair ( N Z[S]N , Z[S]N ) has relative property (T) if and only if ( QN p , Qp ) has relative property (T) for every p ∈ S ∪ {∞}. Theorem 4.2 then implies the equivalence with (ii). Proof of Theorem 1.2. The chain of implications that we shall prove is as follows: (iii)⇒(iv)⇒(i)⇒(ii)⇒(iii), where, however, the equivalence (iv)⇔(i) is the content of Proposition 4.5. (iii)⇒(iv). We shall show the contrapositive statement; namely, we shall assume that for some p ∈ S ∪ {∞} there exists a ρp∗ ()-invariant probability measure μ and we shall prove that then the image ρp () ⊂ SLN (Qp ) is bounded. N Let W ⊂ P((Qp )∗ ) and I(W ) be deﬁned as in Equations 3.3 and 3.4 with N

E = (Qp )∗ and Qp an algebraic closure of Qp , and let us consider I (W ) = I(W ) ∩ PSLN . Since I (W ) is normalized by ρ() and ρ() is Zariski dense in SLN , then I (W ) is a normal subgroup of PSLN and hence either trivial or the full group. Obviously it cannot be the full group PSLN since this cannot ﬁx pointwise any nonempty subset in projective space. So since I (W ) is trivial, then by Corollary 3.2 StabPSL(QNp ) (μ) is compact, so that ρp () is bounded. (i)⇒(ii). This follows from Lemma 2.6. (ii)⇒(iii). If p = ∞, then it is easy to see that ρ∞ () cannot be bounded because otherwise it would be conjugate into a maximal compact subgroup that is a real algebraic group, thus contradicting the Zariski density of ρ() in SLN . Let us assume that Z[S]N is ﬁnitely generated as a module over Z[], namely there exist aj ∈ Z[S]N , with 1 ≤ j ≤ r, such that Z[S]N = ρ(Z())a1 + · · · + ρ(Z())ar . For p ∈ S, ni ∈ Z, and γi ∈ , using the ultrametric inequality, we have r −∞. Then, for P a.e. ω, all ε > 0, there exist K = K (ω) and an inﬁnite number of instants n such that Sn (ω) − Sn−k (T k ω) ≥ (α − ε)k for all k, K ≤ k ≤ n. S (ω)

In particular, it follows from subadditivity that lim inf k kk ≥ α. There fore Theorem 2 follows (in the case fd P > −∞) because in that case, both

noncommutative ergodic theorems / 399

sequences Sk and −Sk are subadditive and 1 1 Sn d P = sup Sn d P = fd P = α. inf n n n n On the other hand, lim supk

Sk k

is a T -invariant function which, by sub1 p−1 S (T jk ω). Thus the constant additivity, is not bigger than lim supp pk j=0 k lim supk Skk is not bigger than 1k Sk d P. Theorem 3 follows in the case when α > −∞. To treat the case α = −∞ in both theorems, it sufﬁces to replace Sn by max (Sn , −nM), and to let M go to inﬁnity; see [Kr] for details.

2. Noncommutative Ergodic Theorems Observe that Theorem 1 also holds true for any linear operator U of a Hilbert space assuming U ≤ 1. One can take this a step further and deﬁne for any g ∈ H, φ(g) := Ug + f . Then φ is an isometry (or merely 1-Lipschitz in the case U ≤ 1). Note that φ n (0) =

n−1

U kf .

k=0

Pazy proved in [Pa] that more generally for any map φ : H → H such that φ(x) − φ(y) ≤ x − y , it holds that there is a vector v ∈ H such that 1 n φ (0) → v n in norm. This can be reformulated as follows: there is a unit speed geodesic γ (t) = tv/ v in H such that 1

1 φ n (0) − γ (n v ) = 1 φ n (0) − nv → 0. n n

We call this property ray approximation. It turns out that this generalization of the ergodic theorem still holds for more general group actions than the actions of Z. Let G be a second countable, locally compact, Hausdorff topological semigroup, and consider g : → G a measurable map. We form Zn (ω) := g(ω)g(T ω) . . . g(T n−1 ω) and ask whether Zn converges to inﬁnity with some linear speed. Assume G acts on a metric space (X , d) by 1-Lipschitz transformations. Then for a ﬁxed x0 ∈ X we can deﬁne for g ∈ G, |g| := d(x0 , gx0 ). Clearly, up to a bounded error, |Zn (ω)| does not depend on our choice of x0 . We have

400 / a. karlsson and f. ledrappier

proposition 5. Assume the transformation T is ergodic, and |g|d P < ∞. Then there is a nonnegative number α such that for P-almost every ω, as n → ∞: 1 |Zn (ω)| → α. n The number α is given by 2

α = inf n

1 n

|Zn (ω)|d P.

Proof. It sufﬁces to observe that the sequence |Zn (ω)| satisﬁes the hypotheses of Theorem 3. Our hypothesis says that Z1 < ∞. The subaditivity follows from the 1-Lipschitz property: |Zn+m (ω)| = d(x0 , g(ω) . . . g(T n+m−1 ω)x0 ) ≤ d(x0 , g(ω) . . . g(T m−1 ω)x0 )+ + d(g(ω) . . . g(T m−1 ω)x0 , g(ω) . . . g(T n+m−1 ω)x0 ) ≤ |Zm (ω)| + d(x0 , g(T m ω) . . . g(T n+m−1 ω)x0 ) = |Zm (ω)| + |Zn (T m ω)|. Moreover we see that the limit α is given by inf n

1 n

|Zn (ω)|d P.

When α > 0, Proposition 5 says that the points Zn (ω)x go to inﬁnity with a deﬁnite linear speed. The question arises of the convergence in direction of the points Zn (ω)x. Given Equation (1), we expect that an almost everywhere convergence theorem will say that Zn (ω)x will stay at a sublinear distance of a geodesic. We present several results in that direction depending on different geometric hypotheses on the space X . Assume X is a complete, Busemann nonpositively curved and uniformly convex (e.g., CAT (0) or uniformly convex Banach space) metric space. Then,

theorem 6. [KM] Under these assumptions, there is a constant α ≥ 0 and, for P-almost every ω, a geodesic ray γω such that 1 d(Zn (ω)x0 , γω (nα)) → 0. n We outline the proof (see [KM], section 5, for details). Let a(n, ω) = d(x0 , Zn (ω)x0 ) for each n. Consider a triangle consisting of x0 , Zn (ω)x0 , and Zk (ω)x0 . Note that the side of this triangle has lengths a(n, ω), a(k, ω), and (at most) a(n − k, T k ω). Given ε > 0 (and a.e. ω), for k large it holds that

noncommutative ergodic theorems / 401

a(k, ω) ≤ (α + ε)k. Assume now in addition to k being large that n and k are as in Lemma 4. This implies that the triangle is thin in the sense that Zk (ω)x0 lies close to the geodesic segment [x0 , Zn (ω)x0 ]; more precisely, the distance is at most δ(ε)a(k, ω), where δ depends only on the geometry. Thanks to the geometric assumptions this δ(ε) tends to 0 as ε tends to 0. Selecting ε tending to 0 fast enough we can, by selecting suitable n as in Lemma 4 and some simple geometric arguments, obtain a limiting geodesic. Finally, one has essentially from the construction that as m → ∞, the points Zm (ω)x0 lie at a sublinear distance from this geodesic ray. This note is devoted to the generalization of the ergodic theorem to groups of isometries of a metric space (X , d). We assume that the space (X , d) is proper (closed bounded subsets are compact) and we consider the metric compactiﬁcation of X . Deﬁne, for x ∈ X , the function x (z) on X by x (z) = d(x, z) − d(x, x0 ). The assignment x → x is continuous, injective, and takes values in a relatively compact set of functions for the topology of uniform convergence on compact subsets of X . The metric compactiﬁcation X of X is the closure of X for that topology. The metric boundary ∂X := X \ X is made of Lipschitz continuous functions h on X such that h(x0 ) = 0. Elements of ∂X are called horofunctions. Our main result is the following:

theorem 7. (ergodic theorem for isometries [KL1]) Let T be a measure-preserving transformation of the Lebesgue probability space (, A, P), G a locally compact group acting by isometries on a proper space X , and g : → G a measurable map satisfying |g(ω)|d P(ω) < ∞. Then, for P-almost every ω, there is some hω ∈ ∂X such that 1 1 lim − hω (Zn (ω)x0 ) = lim d(x0 , Zn (ω)x0 ). n→∞ n n

n→∞

For the convenience of the reader, the proof of Theorem 7 is given in Section 5. We explain in Section 3 why the convergence in Theorem 7 is equivalent to the ray approximation under the CAT (0) assumption. Note that by Theorem 7 the former convergence holds for all norms on Rd , but that Theorem 7 does not apply to inﬁnite-dimensional Banach spaces. In this case, one can use Lemma 4 to prove a noncommutative ergodic theorem with linear functionals of norm 1, somewhat analogous to horofunctions. Namely,

402 / a. karlsson and f. ledrappier

theorem 8. [Ka] Let Zn (ω) be an ergodic integrable cocycle of 1-Lipschitz selfmaps of a reﬂexive Banach space. Then for P-almost every ω there is a linear functional fω of norm 1 such that lim

n→∞

1 fω (Zn (ω)0) = α. n

On the other hand, Kohlberg-Neyman [KN] found a counterexample to the norm convergence, or more precisely to (1), for general Banach spaces.

3. Case When X Is a CAT(0) Proper Space When the space (X , d) is a proper CAT (0) metric space, both Theorems 6 and 7 apply. Because it is a direct generalization of the important case when G is a linear group, it is often called the Oseledets theorem. In this section we explain how to recover the ray approximation and other more familiar forms of Oseledets theorem from Theorem 7. Many of the geometric ideas in this section go back to Kaimanovich’s extension of Oseledets theorem to more general semisimple groups [K3]. A metric geodesic space (X , d) is called a CAT (0) space if its geodesic triangles are thinner than in the Euclidean space. Namely, consider four points A, B, C, and D ∈ X , D lying on a length-minimizing geodesic going from B to C. Draw four points A , B , C , and D in the Euclidean plane with AB = A B , BD = B D , DC = D C , and CA = C A . The space is called CAT (0) if for any such conﬁguration, AD ≤ A D . Simply connected Riemannian spaces with nonpositive curvature, locally ﬁnite trees, and Euclidean buildings are proper CAT (0) spaces. If X is a CAT (0) space, then the horofunctions h ∈ ∂X are called Busemann functions, and for any h ∈ ∂X , there is a unique geodesic ray σh (t), t ≥ 0 such that σh (0) = x0 and limt→∞ σh (t) = h. We have

corollary 9. Assume moreover that X is a CAT (0) space and that α > 0. Then, for P-almost every ω, as n goes to ∞, 1 lim d Zn (ω)x0 , σhω (αn) = 0, n n where hω is given by Theorem 7. Proof. Consider a geodesic triangle A = Zn (ω)x0 , B = x0 , Ct = σhω (t), for t very large, and choose D = σhω (nα). We want to estimate the distance AD.

noncommutative ergodic theorems / 403

We have AB = d(Zn (ω)x0 , x0 ) = |Zn (ω)| =: nαn (ω); BCt = t, BD = nα; and Ct A = t + σhω (t) (Zn (ω)x0 ) =: t − nβn (ω) + on (t). For almost every ω, we have • limn αn (ω) = α by Theorem 3; • limn βn (ω) = limn − 1 hω (Zn (ω)x0 ) = α by Theorem 7; and n • for a ﬁxed n, limt→∞ on (t) = hω (Zn (ω)x0 ) − hσ (t) (Zn (ω)x0 ) = 0. hω

Construct the comparison ﬁgure A B Ct D , and let t go to ∞. The point Et of B Ct at the same distance from Ct than A converges to the orthogonal projec of A on B C and satisﬁes B E = nβ − o (t). Therefore, B E = nβ . tion E∞ n n n t t ∞ We have 2 2 ) = n2 (αn2 − βn2 ), (D E∞ ) = n2 (βn − α)2 , (AE∞

and therefore, as n → ∞: lim n

1 2 (A D ) = lim (αn2 − βn2 ) + (βn − α)2 = 0. 2 n n

corollary 10. With the same assumptions, we have, for P-almost every ω, Zn (ω)x0 converges to hω in X . In particular, when α > 0 and X is proper CAT (0), the direction hω given by Theorem 7 is unique. Proof. In the above triangle, the geodesic σn joining x0 to Zn (ω)x0 converges to σhω . Therefore, all the accumulation points of Zn (ω)x0 belong to the set seen from x0 in the direction of hω . By the same proof, all the accumulation points of Zn (ω)x0 belong to the set seen from σhω (K ) in the direction of hω , for all K . As K goes to inﬁnity, the intersection of those sets is reduced to the point hω . In the case when G is a linear group, Corollary 9 is closely related to the well-known

theorem 11. (oseledets multiplicative ergodic theorem [O], 1968) Let T be an ergodic transformation of the probability space (, A, P), and A :

404 / a. karlsson and f. ledrappier

→ GL(d, R) a measurable map such that Then there exist

max{ln ||A||, ln ||A−1 ||}d P < ∞.

• real numbers λ1 ≤ λ2 ≤ · · · ≤ λk ; and • integers mi , i = 1, . . . , k with i mi = d, i λi mi = ln |DetA|d P, • for P-almost every ω, a ﬂag of subspaces of Rd

{0} = Vk+1 (ω) ⊂ Vk (ω) ⊂ · · · ⊂ V1 (ω) = Rd with, for all i, 1 ≤ i ≤ k, DimVi = Vi+1 (ω) if and only if as n goes to ∞, lim

j≥i

mj and a vector v belongs to Vi (ω) \

1 ln ||A(T n−1 ω)A(T n−2 ω) . . . A(ω)v|| = λi . n

Observe that, automatically, the Vi depend measurably on ω and are invariant in the sense that A(ω)Vi (ω) = Vi (T ω). The usual complete form of Oseledets theorem follows by comparing the results of Theorem 11 for (T , A) and for (T −1 , A−1 ◦ T −1 ). Fix ω ∈ , and let ei , i = 1, . . . , d be an orthogonal base of Rd such that e ∈ Vi (ω) as soon as ≤ j≥i mj . Write μ1 ≥ · · · ≥ μd for the exponents λj , each counted with multiplicity mj , and consider A(n) (ω) := A(T n−1 ω)A(T n−2 ω) . . . A(ω) in the base (ei ). To verify the statement of Theorem 11, it sufﬁces to show that for all ε > 0 and for n large enough, (n) |Ai,j (ω)| ≤ en(μi +ε) and | ln |DetA(n) (ω)| − μj | ≤ ε. j

With the notations of Section 2, consider the action by isometries of GL(d, R) on the symmetric space GL(d, R)/O(d, R) with origin x0 = O(d, R) d t 2 and distance |g| = j=1 ( ln τi ) , where τi are the eigenvalues of gg . It is t a CAT (0) geodesic proper space. Set g(ω) = A (ω). The moment hypothe sis |g|d P < ∞ is satisﬁed. We have A(n) (ω) = (Zn (ω))t . If α = 0, then the eigenvalues of Zn Znt grow subexponentially and lim n

1 1 1 ln ||A(n) (ω)v|| = lim ln (||Znt v||2 ) = 0. n 2 n n

In this case m1 = d, λ1 = 0 and Theorem 11 holds. We may assume α > 0, and apply Corollary 9. Geodesics starting from the origin are of the form etH , where H is a nonzero symmetric matrix. Therefore, for P-almost every ω, there is a nonzero symmetric matrix H(ω) such that n1 d(exp(nH(ω)), Zn (ω)) goes to 0 as n → ∞ (the constant α has been incorporated in H). In other words, n1 ln of the norm,

noncommutative ergodic theorems / 405

and of the norm of the inverse, of the matrix exp( − nH(ω))(A(n) (ω))t , go to 0 as n → ∞. We claim that this gives the conclusion of Theorem 11 with λi the eigenvalues of H(ω), mi their respective multiplicities and Vi the sums of the eigenspaces corresponding to eigenvalues smaller than λi . Indeed, we write exp(H(ω)) = K (ω)t K (ω) for K an orthogonal matrix and a diagonal matrix with diagonal entries eμi , and A(n) (ω) = Ln (ω)n (ω)Kn (ω) a Cartan decomposition of A(n) with Ln , Kn orthogonal, n a diagonal matrix with nonin(n) (n) creasing diagonal entries exp(nδ1 (ω)) ≥ · · · ≥ exp(nδd (ω)). The conclusion of Corollary 9 is therefore that, for P-almost every ω, n1 ln of the norm, and of the norm of the inverse, of the matrix n (ω)Kn (ω)K t (ω)exp( − n), go to 0 as n → ∞. It follows that, for such an ω, | ln |DetA(n) (ω)| − j μj | goes to 0 as n goes (n)

to ∞. Furthermore, for n large enough, the entries ki,j (ω) of the matrix Kn (ω)K t (ω) satisfy (n)

|ki,j (ω)| ≤ en(μj −δi (n)

+ε)

.

We have A(n) (ω)ej = Ln (ω)n (ω)Kn (ω)ej = n (ω)Kn (ω)K −1 (ω)fj , where fi is the canonical base of Rd . The components of this vector are (n)

enδi (ω) ki,j (ω). Their absolute values are indeed smaller than en(μj +ε) for n large enough. (n)

4. The Case When X Is a Gromov Hyperbolic Space (in Particular, R). Theorem 7 is due to Kaimanovich using an idea of Delzant when X is a Gromov hyperbolic geodesic space even without the condition that X is a proper space [K2]. As in the CAT (0)-case it is there formulated as Zn lies on sublinear distance of a geodesic ray. From Theorem 7 one gets the following:

corollary 12. Assume moreover that X is a Gromov hyperbolic geodesic space and that α > 0. Then, for P-almost every ω, as n goes to ∞, there is a geodesic ray σω such that 1 lim d(Zn (ω)x0 , σhω (αn)) = 0. n n Proof. Take hω given from Theorem 7. It is known, see [BH, p. 428], that for Gromov hyperbolic geodesic spaces it holds that there is a geodesic ray σω such that σω (x0 ) = 0 and

406 / a. karlsson and f. ledrappier

bω ( · ) = lim d(·, σω (t)) − t t→∞

is a horofunction such that bω ( · ) − hω ( · ) ≤ C for some constant C. This bω therefore clearly satisﬁes the conclusion of Theorem 7. Now we use the notation and setup in the proof of Corollary 9. Consider the triangle ABCt . By δ-hyperbolicity D must lie at most δ away from either AB or ACt . Call the closest point X . By the triangle inequality we must have that αn − δ ≤ XB ≤ αn + δ. If X lies on AB, then it is clear that XA = o(n) and hence AD = o(n). If X lies on ACt , then t − αn − δ ≤ XCt ≤ t − αn + δ. In view of that bω (Zn (ω)) ≈ −αn we again reach the conclusion that X , and hence also D, lie on sublinear distance from A.

corollary 13. With the same assumptions, we have that for P-almost every ω, Zn (ω)x0 converges to the point [σω ] in the hyperbolic boundary ∂hyp X . Proof. Clearly, the Gromov product (Zn (ω), σω (αn)) → ∞ as n → ∞ in view of the previous corollary. In the case when G = R and X = (R, |·| ), Corollaries 12 and 13 yield Theorem 2. Indeed, in this case the drift is

α =

fd P

and ∂ R = {h+ = +∞ (z) = −z, h− = −∞ (z) = z}. It follows from Corollary 13 that the index of hω is T invariant and is therefore almost everywhere constant. In other words, the existence of the hω with the required property amounts to the choice of the right sign: hω = sign{

f

(ω)d P(ω)}∞ .

Then, Corollary 12 says exactly that if the function f is integrable, for P-almost every ω 1 f (ω)d P(ω). Sn (ω) → n

noncommutative ergodic theorems / 407

The above observation is not a new proof of Theorem 2, because Theorem 2 is used in the proof of Theorem 7 (see Section 5). We want to only illustrate the meaning of the metric boundary on the simplest example. Nevertheless, it turns out that modifying the translation-invariant metric on X = R might have interesting consequences. The following discussion comes from [KMo], which in turn was inspired by [LL]. Let D : R≥0 → R≥0 be an increasing function, D(t) → ∞ such that D(0) = 0 and D(t)/t → 0 monotonically. From the inequality 1 1 D(t + s) ≤ D(t) t+s t we get the following subadditivity property: D(t)/t s D(s) ≤ D(t) + D(s). D(t + s) ≤ D(t) + D(t) = D(t) + t D(s)/s From all these properties of D, it follows that (R, D( |·| )) is a proper metric space, and clearly invariant under translations. Now we determine ∂ R with respect to this metric. Without loss of generality we may assume that xn → ∞. We claim that for any z h(z) = lim D(xn − z) − D(xn ) = 0. n→∞

Assume not. Then for some s > 0 and an inﬁnite sequence of t → ∞ that D(t + s) − D(t) > c > 0 (wlog). For such s and t with t large so that D(t)/t < c/s, we have D(t)

D(t) + t s D(t + s) D(t) + c D(t) ≥ ≥ = t+s t+s t+s t but this contradicts that D(t)/t is strictly decreasing. Hence ∂ R = {h ≡ 0}. Applying Theorem 7 in this setting yields a result already obtained by Aaronson with a different argument.

theorem 14. (aaronson [A]) Let f : → R such that Then, for P-almost every ω,

D(

f )dμ < ∞.

1

D Sn (ω) = 0. n→∞ n lim

Proof. It was noted above that ∂(R, D( |·| )) only consisted of h = 0. The conclusion then follows from Theorem 7 since h = 0 forces α = 0.

408 / a. karlsson and f. ledrappier

One can relax the conditions on D: for one thing, one can remove having D(0) = 0. More interestingly, the condition that D(t)/t decreases to 0 can be weakened in the following way.

corollary 15. (aaronson-weiss [A]) Let d(t) be an increasing positive function d(t) → ∞, such that d(t) = o(t), d(t + s) ≤ d(t) + d(s) and

d( f )dμ < ∞ for some function f : → R. Then, for P-almost every ω, lim

n→∞

1

d Sn (ω) = 0. n

Proof. Deﬁne D(t) = sup{d(ut)/u : u ≥ 1}. Note that this satisﬁes all the assumptions made on D in Theorem 14. Moreover, d(t) ≤ D(t) ≤ 2d(t), since if D(t) = d(tu)/u, set n = [u] + 1 and then D(u) ≤ d(nt)/u ≤ nd(t)/u ≤ 2d(t). (See [A], p. 66, for more details.) This shows that Theorem 14 actually holds for d in place of D. In particular, Corollary 15 applies to any metric d(., .) on R where balls grow superlinearly (where d(t) := d(0, t)). From this one obtains as a special case classical results like the one of Marcinkiewicz-Zygmund [MZ] and Sawyer [S]:

corollary 16. Let 0 < p < 1. If f ∈ Lp , then for P-almost every ω lim

n→∞

1 Sn = 0. n1/p

Such moment conditions arise naturally in probability theory. These results are known to be best possible in certain ways (e.g., [S] and [A]). For the iid case the converse also holds [MZ]. Another example is as follows:

corollary 17. If f is log-integrable, then for P-almost every ω lim |Sn |1/n = 1.

n→∞

One can modify the metric on any metric space X in the same way replacing d(x, y) with D(d(x, y)), where D(t) satisﬁes the assumptions for Theorem 14 or,

noncommutative ergodic theorems / 409

more generally, the assumptions for Corollary 15. By estimating a subadditive by an additive cocycle in the obvious way a(n, ω) ≤

n−1

a(1, T k ω),

k=0

and Theorem 14 implies that 1 D(d(Zn x0 , x0 )) → 0 a.e., n under the condition that D(d(g(ω)x0 , x0 )) is integrable.

5. Proof of Theorem 7 We begin by a few observations: ﬁrst, we can extend by continuity the action of G to X , and write, for h ∈ X , g ∈ G: g.h(z) = h(g −1 z) − h(g −1 x0 ). Now deﬁne the skew product action on := × X by T (ω, h) = (T ω, g(ω)−1 ·h). n

Observe that T (ω, h) = (T n ω, (Zn (ω))−1 ·h). Deﬁne the Furstenberg cocycle F(ω, h) by F(ω, h) := −h(g(ω)x0 ). We have 3

F n (ω, h) :=

n−1

i

F(T (ω, h)) = −h(Zn (ω)x0 ).

i=0

Relation (3) is proven by induction on n. We have F 1 (ω, h) := −h(g(ω)x0 ) = −h(Z1 (ω)x0 ) and F n (ω, h) = F n−1 (ω, h) + F(T

n−1

(ω, h))

= −h(Zn−1 (ω)x0 ) − (Zn−1 (ω))−1 ·h(g(T n−1 ω)x0 ) = −h(Zn−1 (ω)x0 ) − h(Zn−1 (ω)g(T n−1 (ω))x0 ) + h(Zn−1 (ω)x0 ) = −h(Zn (ω)x0 ). In particular, for any T invariant measure m on such that the projection on is P, we have Fdm ≤ α because 1 1 F(ω, h)dm(ω, h) = −h(Zn (ω)x0 )dm(ω, h) ≤ |Zn (ω)|d P(ω). n n

410 / a. karlsson and f. ledrappier

There is nothing to prove if α = 0. To prove Theorem 7 in the case α > 0, it sufﬁces to construct a T -invariant measure m on such that the projec tion on is P and such that F(ω, h)dm(ω, h) = α. Indeed, since α is the largest possible value of F, we still have the same equality for almost every ergodic component of m. By the ergodic Theorem 2, the set A of (ω, h) such that − n1 h(Zn (ω)x0 ) = n1 F n (ω, h) → α as n goes to ∞ has full measure. Moreover, observe that if h is not a point in ∂X , − n1 hγ (Zn (ω)x0 ) converges to −α. Since α > 0, this shows that A ⊂ ∂X . We get the conclusion of Theorem 7 by choosing for ω → hω a measurable section of the set A. We ﬁnally construct a measure m with those properties. We deﬁne a measure μn on ; for any measurable function on such that

sup (ω, h) d P(ω) < ∞, h∈X

we set:

×X

(ω, h)dμn (ω, h) =

(ω, Zn (ω)x0 )d P(ω).

The set of measures m on such that the projection on is P is a convex compact subset of L∞ (, P (X )) = (L1 (, C(X )))∗ for the weak* topology. The mapping m → (T )∗ m is afﬁne and continuous. We can take for m any weak* limit point of the sequence: 1 i (T )∗ μn . n n−1

ηn =

i=0

The measure m is T -invariant and, since FL1 (,C(X )) = sup |F(ω, h)|d P(ω) = sup |h(g(ω)x0 )|d P(ω) < +∞, h

h

we may write, using Relation (3) and Formula (2):

1 k→∞ nk

Fdm = lim = lim

k→∞

1 nk

1 n k→∞ k

n k −1

1 k→∞ nk

i=0

F nk (ω, Znk (ω)x0 )d P(ω)

= lim = lim

i

(F ◦ T )dμnk

( − Znk (ω)x0 (Znk (ω)x0 ))d P(ω) |Znk (ω)|d P(ω) = α.

By the above discussion this achieves the proof of Theorem 7.

noncommutative ergodic theorems / 411

Observe that by putting together the discussions in Sections 5 and 3, we obtain a proof of Oseledets Theorem 11. As proofs of Theorem 11 go, this one is in some sense rather close to the original one [O], with the somewhat simplifying use of the geometric ideas from [K3] and invariant measures as in [W].

6. Random Walks In this section we consider a probability ν on a group G and apply the preceeding analysis to the random walk Zn = g0 g1 . . . gn−1 , where the gi are independent with distribution ν. We assume • there is a proper left-invariant metric d on G that generates the topology

of G (when G is second-countable locally compact, such a metric always exists, see [St]); • d(e, g)dν(g) < + ∞ (we say that ν has a ﬁrst moment); and • the closed subgroup generated by the support of ν is the whole G (we say that ν is nondegenerate). Then there is a number (ν) ≥ 0 such that, for almost every sequence {gi }, limn n1 d(e, Zn ) = (ν). In the case when the group G is the group SL2 (R) acting on the hyperbolic plane, (ν) is twice the Lyapunov exponent of the independent product of matrices. In that case it is given by a formula involving the stationary measure on the circle, the Furstenberg-Khasminskii formula ([F1]; this appellation seems to be standard, cf. [Ar]). Seeing again the circle as the geometric boundary of the hyperbolic plane, we extend this formula to our general context:

theorem 18. (furstenberg-khasminskii formula for the linear drift, [KL2]) Let (G, ν) verify all the above assumptions, and let G be the metric compactiﬁcation of (G, d). Then there exists a measure μ on G with the following properties: • μ is stationary for the action of G, that is, μ satisﬁes μ = (g∗ μ)dν(g); and • (ν) = h(g −1 )dμ(h)dν(g). Moreover, if (ν) > 0, then μ is supported on ∂G. Proof. In the proof of Theorem 7, we constructed a measure m on × G. The measure μ can be seen as the projection on G of m, but it turns out that the measure μ can be directly constructed. Let (+ , A+ , P) be the space

412 / a. karlsson and f. ledrappier

of sequences {g0 , g1 , . . . } with product topology, σ -algebra, and measure P = ν ⊗N . For n ≥ 0, let νn be the distribution of Zn (ω) in G. In other words, deﬁne, for any continuous function f on G: fdνn = f (g0 g1 · · · gn−1 )dν(g0 )dν(g1 ) · · · dν(gn−1 ), ν0 = δe . We claim that any weak* limit μ of the measures n1 n−1 i=0 νi satisﬁes the conclusions of Theorem 18. Clearly, the measure μ is stationary: for any continuous function f on G, we have nk −1 1 f (gg0 g1 · · · gi−1 ) k→∞ nk

f (g.h)dμ(h)dν(g) = lim

i=0

× dν(g0 )dν(g1 ) · · · dν(gi−1 )dν(g) nk −1 1 fdνi+1 k→∞ nk

= lim

i=0

=

1 [ k→∞ nk

fdμ + lim

=

fdνnk − f (e)]

fdμ.

In the same way, we get h(g

−1

nk −1 1 )dμ(h)dν(g) = lim [d(Zi (ω), g −1 ) k→∞ nk i=0

− d(Zi (ω), e)]d P(ω)dν(g) 1 d(Znk , e)d P(ω) = lim k→∞ nk = (ν). This shows that the measure μ has the desired properties. Moreover, the measure P × μ on the space + × G is T -invariant. There is a unique T invariant measure m on × G that extends P × μ. The measure m satisﬁes all the properties we needed in the proof of Theorem 7. In particular, if (ν) is positive, μ(∂G) = (P × μ)(+ × ∂G) = m( × ∂G) = 1.

noncommutative ergodic theorems / 413

A bounded measurable f : G → R is ν-harmonic if f (g) = f (gh)dν(h) G

for any g ∈ G. Constant functions are obviously ν-harmonic. If f is a bounded Harmonic function, then f (Zn ) is a bounded Martingale and therefore converges almost surely. We say that (G, ν) satisﬁes the Liouville property (or (G, ν) is Liouville) if the constant functions are the only bounded ν-harmonic functions.

corollary 19. [KL2] Let G be a locally compact group with a left-invariant proper metric and ν be a nondegenerate probability measure on G with ﬁrst moment. Then, if (G, ν) is Liouville, there is a 1-Lipschitz homomorphism T : G → R such that for almost every trajectory Zn of the corresponding random walk, we have 1 lim T (Zn ) = T (g)dν(g) = l(ν). n→∞ n G Proof. The key observation is that if (G, ν) is Liouville and G acts continuously on a compact space Y , then every stationary measure μ is invariant. Indeed, for f ∈ C(Y , R), the function ϕ(g) := fd(g∗ μ) is harmonic and bounded, and therefore constant. In particular, the measure μ from Theorem 18 is invariant, and if we set T (g) := h(g −1 )μ(dh), the mapping T is Lipschitz continuous and is a group homomorphism because we have h(g −1 g −1 )μ(dh) T (g g) = = =

(g .h)(g −1 )μ(dh) +

h(g −1 )μ(dh)

h(g −1 )(g∗ μ)(dh) + T (g )

= T (g) + T (g ), where we used the invariance of μ at the last line. Finally, by the FurstenbergKhasminskii formula, we have (ν) = T (g)dμ(g).

414 / a. karlsson and f. ledrappier

A measure ν on G is called symmetric if it is invariant under the mapping g → g −1 . A measure is centered if every homomorphism of G into R is centered, meaning that the ν-weighted mean value of the image is 0. Every symmetric measure with ﬁrst moment ν is centered, since for any homomorphism T : G → R, the mean value, which is T (g)dν(g) = T (g −1 )dν(g) = − T (g)dν(g), G

G

G

must hence equal 0. By simple contraposition from Corollary 19, we get

corollary 20. [KL2] Let G be a locally compact group with a left-invariant proper metric and ν be a nondegenerate centered probability measure on G with ﬁrst moment. Then, if l(ν) > 0, there exist nonconstant bounded ν-harmonic functions. Corollary 20 was known in particular for ν with ﬁnite support ([Va], [M]) or in the continuous case, for ν with compact support and density [Al]. One case when all probability measures on G are centered is when there is no group homomorphism from G to R. We can apply Corollary 20 to a countable ﬁnitely generated group. Let S be a ﬁnite symmetric generator for G, and endow G with the left-invariant metric d(x, y) = |y −1 x| where |z| is the shortest length of an S-word representing z. We say that G has subexponential growth if limn n1 ln An = 0, where An is the number of elements z of G with |z| ≤ n. Such a group has automatically the Liouville property [Av]. This yields

corollary 21. [KL2] Let G be a ﬁnitely generated group with subexponential growth and H 1 (G, R) = 0. Then for any nondegenerate ν on G with ﬁrst moment, we have (ν) = 0. Observe that conversely, if there exists a nontrivial group homomorphism T from a ﬁnitely generated group into R, then there exists a nondegenerate probability ν on G, with ﬁrst moment and (ν) > 0. Indeed, there exists M such that [−M, M] contains all the images of the elements of the generating set S. We can choose ν carried by all the elements of S with images in [0, M]. Since T is nontrivial, T (g)dν(g) > 0. The measure ν is nondegenerate, has ﬁnite support, and (ν) > 0 since for all g ∈ G, |T (g)| ≤ M|g|.

7. Riemannian Covers In this section we consider a complete connected Riemannian manifold (M, g) with bounded sectional curvatures. In particular, if dM is the Riemannian

noncommutative ergodic theorems / 415

distance on M, (M, d) is a proper space. Associated to the metric is the LaplaceBeltrami operator . A function f is harmonic if f = 0. We say that M is Liouville if all bounded and harmonic functions are constant. Associated to is a diffusion process Bt called Brownian motion. Since the curvature is bounded and M is complete, the Brownian motion is deﬁned for all time. For all x ∈ M, there is a probability Px on C(R+ , M) such that the process Bt given by the t coordinate is a Markov process with generator and B0 = x. We can deﬁne 1 g := lim sup dM (x0 , Bt ), t→∞ t for any x0 ∈ M.

theorem 22. [KL3] Assume that (M, g) is a regular covering of a Riemannian manifold that has ﬁnite Riemannian volume and bounded sectional curvatures. Then M is Liouville if and only if lim

t→∞

1 d(x0 , Bt ) = 0 a.s. t

The “if ” part was proved by Kaimanovich, see [K1], and the converse is clear if the Brownian motion is recurrent on M. The proof of the new implication in Theorem 22 in the transient case uses the Furstenberg-Lyons-Sullivan discretization procedure. Let be the covering group of isometries of M. This discretization consists in the construction of a probability measure ν on , with the following properties: • the restriction f (γ ) := F(γ x0 ) is a one-to-one correspondence between

bounded harmonic functions on M and bounded functions on that satisfy f (γ g)ν(g); f (γ ) = g∈

• if γ1 , . . . , γn are chosen independent and with distribution ν, then

limn→∞ n1 dM (x0 , γ1 . . . γn x0 ) exists. It vanishes a.e. if and only if limt→∞ 1t dM (x0 , Bt ) = 0 a.s.; and • in the case when the Brownian motion is transient, one can choose ν symmetric, that is, such that for all γ in , ν(γ −1 ) = ν(γ ). The ﬁrst property goes back to Furstenberg [F2] and has been systematically developed by Lyons and Sullivan [LS] and Kaimanovich [K5]. The second one was observed in certain situations by Guivarc’h [G] and Ballmann [Ba]. Babillot

416 / a. karlsson and f. ledrappier

observed that the modiﬁed construction of [BL] has the symmetry property. Given the above, proving Theorem 22 mostly reduces to Corollary 20, if we can show that hypotheses of Corollary 20 are satisﬁed. We endow with the metric deﬁned by the metric of M on the orbit x0 . This deﬁnes a left-invariant and proper metric on : bounded sets are ﬁnite because they correspond to pieces of the orbit situated in a ball of ﬁnite volume. The measure ν is nondegenerate because its support is the whole . It is shown in [KL3] that the measure ν has a ﬁrst moment. The proof uses the details of the construction, but the idea is that the distribution of ν is given by choosing some random time and looking at the point γ x0 close to the trajectory of the Brownian motion at that time. Since the curvature is bounded from below, if the expectation of the time is ﬁnite, the expectation of the distance of the Brownian point at that time is ﬁnite as well. It also follows that the rates of escape of the Brownian motion and of the random walk are proportional. Therefore, if the manifold (M, g) is Liouville, then the random walk (G, ν) is Liouville. By Corollary 20, limn→∞ n1 dM (x0 , γ1 . . . γn x0 ) = 0 and therefore, limt→∞ 1t dM (x0 , Bt ) = 0 a.s. There are many results about the Liouville property for Riemannian covers of a compact manifold. Theorem 22 implies that the corresponding statements hold for the rate of escape of the Brownian motion. Guivarc’h [G] showed that if the group is not amenable, then (M, g) is not Liouville, whereas when is polycyclic, (M, g) is Liouville (Kaimanovich [K2]). Lyons and Sullivan ([LS], see also [Er] for a simply connected example) have examples of amenable covers without the Liouville property.

acknowledgments. This survey grew out from lectures given by both authors at the IIIème Cycle Romand de Mathématiques in Les Diablerets in March 2008 and from conversations there. We thank Tatiana SmirnovaNagnibeda and Slava Grigorchuk for their invitation to this friendly and stimulating meeting. References [A] [Al] [Ar] [Av]

J. Aaronson, An Introduction to Inﬁnite Ergodic Theory, AMS Math. Surv. Mon. 50, American Mathematical Society, Providence, RI, 1997. G. Alexopoulos, On the mean distance of random walks on groups, Bull. Sci. Math. 111 (1987), 189–199. L. Arnold, Random Dynamical Systems, Springer Monographs in Mathematics, Springer-Verlag, Berlin, 1998. A. Avez, Entropie des groupes de type ﬁni, C. R. Acad. Sci. Paris, Sér. A-B 275 (1972), A1363–A1366.

noncommutative ergodic theorems / 417 [Ba]

W. Ballmann, On the Dirichlet problem at inﬁnity for manifolds of nonpositive curvature, Forum Math. 1 (1989), no. 2, 201–213. [BL] W. Ballmann and F. Ledrappier, Discretization of positive harmonic functions on Riemannian manifolds and Martin boundary. Actes de la Table Ronde de Géométrie Différentielle (Luminy, 1992), 77–92, Sémin. Congr., 1, Soc. Math. France, Paris, 1996. [BH] M. R. Bridson and A. Haeﬂiger, Metric Spaces of Non-positive Curvature, Grundl. Math. Wiss. 319, Springer-Verlag, Berlin, 1999. [Er] A. Erschler, Liouville property for groups and manifolds, Invent. Math. 155 (2004), no. 1, 55–80. [F1] H. Furstenberg, Non-commuting random products, Trans. Amer. Math. Soc. 108 (1963), 377–428. [F2] H. Furstenberg, Random walks and discrete subgroups of Lie groups, Advances in Probability and Related Topics 1 (1971), 1–63. [G] Y. Guivarc’h, Sur la loi des grands nombres et le rayon spectral d’une marche aléatoire, Astérisque 74 (1980), 47–98. [K1] V. A. Kaimanovich, Brownian motion and harmonic functions on covering manifolds: an entropic approach, Soviet Math. Dokl. 33 (1986), no. 3, 812–816. [K2] V. A. Kaimanovich, Boundaries of random walks on polycyclic groups and the law of large numbers for solvable Lie groups, (Russian) Vestnik Leningrad. Univ. Mat. Mekh. Astronom. (1987), vyp. 4, 93–95, 112. [K3] V. A. Kaimanovich, Lyapunov exponents, symmetric spaces and a multiplicative ergodic theorem for semisimple Lie groups, J. Soviet Math. 47 (1989), 2387–2398. [K4] V. A. Kaimanovich, Poisson boundaries of random walks on discrete solvable groups, in Probability Measures on Groups, X (Oberwolfach, 1990), Plenum, New York, 1991, 205–238. [K5] V. A. Kaimanovich, Discretization of Bounded Harmonic Functions on Riemannian Manifolds and Entropy: Potential Theory (Nagoya, 1990), 213–223, de Gruyter, Berlin, 1992. [K6] V.A. Kaimanovich, The Poisson formula for groups with hyperbolic properties, Ann. Math. 152 (2000), 659–692. [KV] V. Kaimanovich and A. Vershik, Random walks on discrete groups: boundary and entropy, Ann. Prob. 11 (1983), 457–490. [Ka] A. Karlsson, Linear rate of escape and convergence in directions, in Proceedings of a Workshop at the Schrödinger Institute, Vienna 2001, (ed. by V.A. Kaimanovich, in collab. with K. Schmidt and W. Woess), de Gruyter, Berlin, 2004. [KL1] A. Karlsson and F. Ledrappier, On laws of large numbers for random walks, Ann. Prob. 34 (2006), 1693–1706. [KL2] A. Karlsson and F. Ledrappier, Drift and entropy for random walks, Pure Appl Math. Quarterly 3 (2007), 1027–1036. [KL3] A. Karlsson and F. Ledrappier, Propriété de Liouville et vitesse de fuite du mouvement Brownien, C. R. Acad. Sci. Paris, Sér. I, 344 (2007), 685–690. [KM] A. Karlsson and G. Margulis, A multiplicative ergodic theorem and nonpositively curved spaces, Comm. Math. Phys. 208 (1999), 107–123. [KMo] A. Karlsson and N. Monod, Strong law of large numbers with concave moments, unpublished note, 2008. [KN] E. Kohlberg and A. Neyman, Asymptotic behavior of nonexpansive mappings in normed linear spaces, Israel J. Math 4 (1981), 269–275.

418 / a. karlsson and f. ledrappier [Kr] [LL] [L] [LS] [MZ] [M] [O] [Pa] [S] [St] [Va] [Ve] [W]

U. Krengel, Ergodic Theorems, de Gruyter Studies in Mathematics, 6, de Gruyter, Berlin, 1985. F. Ledrappier and S. Lim, A proof of a L1/2 ergodic theorem, unpublished note, 2008. T. Lyons, Instability of the Liouville property for quasi-isometric Riemannian manifolds and reversible Markov chains, J. Diff. Geom 26 (1987), 33–66. T. Lyons and D. Sullivan, Function theory, random paths and covering spaces, J. Differential Geom. 19 (1984), no. 2, 299–323. J. Marcinkiewicz and A. Zygmund, Sur les fonctions indépendantes, Fundam. Math. 29 (1937), 60–90. P. Mathieu, Carne-Varopoulos bounds for centered random walks, Ann. Prob. 34 (2006), 987–1011. V. I. Oseledec, A multiplicative ergodic theorem: Lyapunov characteristic numbers for dynamical systems, Trans. Moscow Math. Soc. 19 (1968), 197–231. A. Pazy, Asymptotic behavior of contractions in Hilbert space, Israel J. Math. 9 (1971), 235–240. S. A. Sawyer, Maximal inequalities of weak type, Ann. of Math. (2) 84 (1966), 157–174. R. A. Struble, Metrics in locally compact groups, Comp. Math. 28 (1974), 217–222. N. Th. Varopoulos, Long range estimates for Markov chains, Bull. Sci. Math. 109 (1985), 225–252. A. Vershik, Dynamic theory of growth in groups: entropy, boundary, examples, Russian Math. Surveys 55 (2000), 667–733. P. Walters, A dynamical proof of the multiplicative ergodic theorem, Trans. Amer. Math. Soc. 335 (1993), 245–257.

13

sorin popa 1 and stefaan vaes 2

COCYCLE AND ORBIT SUPERRIGIDITY FOR LATTICES IN SL (n, R) ACTING ON HOMOGENEOUS SPACES to bob zimmer at the occasion of his

60 th

birthday

Abstract We prove cocycle and orbit equivalence superrigidity for lattices in SL (n, R) acting linearly on Rn , as well as acting projectively on certain ﬂag manifolds, including the real projective space. The proof combines operator algebraic techniques with the property (T) in the sense of Zimmer for the action SL (n, Z) Rn , n ≥ 4. We also show that the restriction of the orbit equivalence relation R( SL (n, Z) Rn ) to a subset of ﬁnite Lebesgue measure provides a II1 equivalence relation with property (T) and yet fundamental group equal to R+ .

1. Introduction and Statement of Main Results Over the last few years, operator algebraic methods were used to prove several orbit equivalence and cocycle superrigidity theorems: for Bernoulli actions of property (T) groups [19], of product groups [18], and for proﬁnite actions of property (T) groups [13]. In this paper, we extend the scope of these methods to a more geometric class of actions, like the natural actions of lattices ⊂ SL (n, R) on the vector space Rn , on the projective space Pn−1 (R), and on certain ﬂag manifolds, all of which can be viewed as SL (n, R)-homogeneous spaces. None of these actions is probability measure preserving. Hence, property (T) of the acting group has to be replaced by Zimmer’s notion of property (T) for a nonsingular action (see [25]), which plays a crucial role in this paper. It is shown that for any lattice ⊂ SL (n, R) the linear action Rn has property (T) if and only if n ≥ 4. We then deduce the following theorem; see Theorem 4.1. 1. Partially supported by NSF Grant DMS-0601082. 2. Partially supported by ERC Starting Grant VNALG-200749 and Research Programme G.0231.07 of the Research Foundation–Flanders (FWO).

419

420 / s. popa and s. vaes

theorem 1.1. Let < SL (n, R) be a lattice and let R be the II1 equivalence relation obtained by restricting the orbit equivalence relation R( Rn ) to a set of Lebesgue measure 1, for some n ≥ 4. Then we have • R has property (T), in the sense of Zimmer, yet the fundamental group of R

equals R+ ; and • Rt cannot be implemented by a free action of a group, ∀t > 0. Also, Rt cannot

be implemented by a (not necessarily free) action of a discrete property (T) group, ∀t > 0. Every orbit equivalence relation R( X ) of a probability measure preserving (p.m.p.) action of a property (T) group, has property (T) in the sense of Zimmer. Theorem 1.1 provides the ﬁrst examples of property (T) equivalence relations that do not arise in this way. We say that a Polish group is of ﬁnite type if it can be realized as the closed subgroup of the unitary group of some II1 factor with separable predual. All countable and all second countable compact groups are Polish groups of ﬁnite type. In [19], the ﬁrst author proved that every 1-cocycle for the Bernoulli action of a property (T) group with values in a Polish group of ﬁnite type is cohomologous to a group morphism. We say that actions with this property are Uﬁn -cocycle superrigid. More precisely:

definition 1.2. The nonsingular action G (X , μ) of the locally compact second countable group G on the standard measure space (X , μ) is called Uﬁn cocycle superrigid if every 1-cocycle for the action G (X , μ) with values in a Polish group of ﬁnite type G is cohomologous to a continuous group morphism G → G. Until now, the only known examples of Uﬁn -cocycle superrigid actions were the Bernoulli and Gaussian actions of property (T) groups (cf. [19]) and, respectively, product groups (cf. [18]), all of which are p.m.p. actions. We exhibit here the ﬁrst examples of nonsingular ergodic actions G (X , μ) that are Uﬁn -cocycle superrigid but for which there exist no invariant probability measure ν ∼ μ on X . As it happens, these examples are of genuine geometric nature.

theorem 1.3. The following actions are Uﬁn -cocycle superrigid. 1) For n ≥ 5 and any lattice in SL (n, R), the linear action Rn . 2) For n ≥ 5 and any ﬁnite index subgroup of SL (n, Z), the afﬁne action Zn Rn .

cocycle and orbit superrigidity for lattices in sl(n, R) / 421

3) For n ≥ 4k + 1, any lattice in SL (n, R), and H any closed subgroup of GL (k, R), the action ⎧ ⎨ ×H if ( − 1, −1) ∈ × H, G Mn,k (R) where G := {±(1,1)} ⎩ × H otherwise, by left-right multiplication on the space Mn,k (R) of n × k matrices equipped with the Lebesgue measure. In [19], the ﬁrst author introduced the notion of malleability for measure preserving actions (X , μ). Roughly speaking, this property requires the existence of a ﬂow on X × X , commuting with the diagonal -action and connecting the identity map to the ﬂip map on X × X . Theorem 0.1 in [19] shows that every weakly mixing, malleable p.m.p. action of a property (T) group is Uﬁn -cocycle superrigid. We prove here a cocycle superrigidity result for inﬁnite measure preserving malleable actions. But, property (T) of the group has to be replaced by property (T) of the diagonal action of X × X . In the case of SL (n, Z) Rn , this forces n ≥ 5. Finally, weak mixing has to be replaced by the ergodicity of the 4-fold diagonal action X × X × X × X , which in the case of SL (n, Z) Rn again holds exactly for n ≥ 5. Using the cocycle superrigidity of SL (n, Z) Rn , we then derive a full classiﬁcation of all 1-cocycles for the action SL (n, Z) Tn = Rn /Zn , with values in a Polish group of ﬁnite type. As such, our Example 5.12 below, complements Zimmer’s celebrated cocycle superrigidity theorem [26]. Thus, Zimmer’s result concerns arbitrary actions SL (n, Z) (X , μ), but makes certain restrictions on the target groups (which must be simple linear algebraic groups) and on the cocycles (Zariski density), while our result treats a very speciﬁc action, but allows rather general target groups and requires no conditions on the cocycles. From the cocycle superrigidity Theorem 1.3, we derive orbit equivalence (OE) superrigidity results for the following concrete actions of lattices in SL (n, R) and PSL (n, R): 1) The linear action Rn . 2) The action Tn = Rn /Zn for a ﬁnite-index subgroup of SL (n, Z). 3) The projective action Pn−1 (R). 4) The natural action X of a lattice in PSL (n, R) on the real ﬂag manifold X of signature (d1 , . . . , dl , n). (Recall that points in X are ﬂags {0} ⊂ V1 ⊂ · · · ⊂ Vl ⊂ Rn where Vi is a vector subspace of Rn with dimension di .)

422 / s. popa and s. vaes

The action in (1) has the Lebesgue measure as inﬁnite invariant measure, while the actions in (3) and (4) do not have ﬁnite or inﬁnite invariant measures. All the actions in (1)–(4) are essentially free and ergodic; see Lemma 5.7 for details. The natural invariant measure class on the ﬂag manifold X can be described as follows. Put dl = k and consider the set Mn,k (R) of n × k matrices of rank k, equipped with the Lebesgue measure. Denote by E = (E1 , . . . , El ) the standard ﬂag of signature (d1 , . . . , dl , n), that is Ei = span{e1 , . . . , edi }, where e1 , . . . , en are the standard basis vectors in Rn . The group GL (k, R) acts on Mn,k (R) by right multiplication. This action is free and proper and Mn,k (R)/H → X : A → (AE1 , . . . , AEl ) is an isomorphism. Here, H = {g ∈ GL (k, R) | gEi ⊂ Ei for all i = 1, . . . , l}. Writing k1 = d1 and ki = di − di−1 for i ≥ 2, the group H can of course be written as ⎛ ⎞ GL (k1 , R) ∗ ··· ∗ ⎜ ⎟ 0 GL (k2 , R) · · · ∗ ⎜ ⎟ ⎜ ⎟. 1.1 H=⎜ .. .. .. .. ⎟ . ⎝ ⎠ . . . 0 0 · · · GL (kl , R) Before stating our OE superrigidity results, recall the following terminology. α

β

definition 1.4. Let (X , μ) and (Y , η) be essentially free, ergodic, nonsingular actions of countable groups on standard measure spaces. • A stable orbit equivalence (SOE) between α and β is a nonsingular iso-

morphism : X0 → Y0 between nonnegligible subsets X0 ⊂ X , Y0 ⊂ Y , such that is an isomorphism between the restricted orbit equivalence relations R( X )|X0 and R( Y )|Y0 . • We say that X is induced from 1 X1 , if 1 is a subgroup of , X1 is a nonnegligible subset of X and g · X1 ∩ X1 is negligible for all g ∈ − 1 . For the linear lattice actions and the quotient action SL (n, Z) Tn , we get the following.

theorem 1.5. Let n ≥ 5 and ⊂ SL (n, R) a lattice. Let (Y , η) be any essentially free, ergodic, nonsingular action of the countable group .

cocycle and orbit superrigidity for lattices in sl(n, R) / 423

1) The actions Rn and Y are SOE if and only if Y is conjugate to an induction of one of the following actions: • Rn itself, and • (only in case −1 ∈ ) the quotient action /{±1} Rn /{±1}.

2) The actions SL (n, Z) Tn and Y are SOE if and only if Y is conjugate to an induction of one of the following actions • SL (n, Z) Tn itself, • SL (n, Z) Zn Rn ,

Z n Rn for some λ ∈ N \ {0, 1}, and λZ λZn • (only in case n is even) one of the actions • SL (n, Z)

PSL (n, Z)

Tn

{±1}

or

PSL (n, Z)

Z n Rn /(2Z)n . 2Z {±1}

Part (2) of the above theorem is a slightly more detailed version, with a very different proof, of a result of Furman [7, corollary B], who shows that for all n ≥ 3, the actions SL (n, Z) Tn and Y are SOE if and only if they are virtually conjugate. Furman’s proof relies on Zimmer’s cocycle superrigidity, while ours is based on the cocycle superrigidity Theorem 1.3. To formulate easily the correct OE superrigidity statements for lattice actions on ﬂag manifolds, make the following observations. The real ﬂag manifold of signature (d1 , . . . , dl , n) has a natural 2l -fold

consisting of oriented ﬂags covering X {0} ⊂ (V1 , ω1 ) ⊂ · · · ⊂ (Vl , ωl ) ⊂ Rn where every Vi is a vector subspace of Rn with an orientation ωi . Clearly,

= Mn,k (R)/H0 , where H0 = {g ∈ H | det(g|E ) > 0 for all i = 1, . . . , l}. In X i the Expression 1.1 above, H0 consists of those matrices A that have on the diagonal Aii ∈ GL (ki , R) with det Aii > 0 for all i.

Denote l = H/H0 and observe that l ∼ = (Z/2Z)⊕l . Then, l acts on X by reversing orientations, but keeping the ﬂags. We denote by −1 ∈ l the multiplication by −1 and observe that −1 = 1 in l if all di are even. Clearly,

/l . X =X

theorem 1.6. Let X be the real ﬂag manifold of signature (d1 , . . . , dl , n). Let

the 2l ⊂ PSL (n, R) be a lattice and assume that n ≥ 4dl + 1. Denote by X fold covering of X consisting of oriented ﬂags, as explained before the theorem. Let (Y , η) be any essentially free, ergodic, nonsingular action of the countable group .

424 / s. popa and s. vaes

The actions X and Y are SOE if and only if Y is conjugate to an induction of one of the actions l

/ for some subgroup < l with − 1 ∈ , or X

× l /

/ for some < l with − 1 ∈ and with • X = {±(1, 1)} {±1} · ⊂ GL (n, R). • ×

example 1.7. If n ≥ 5 and ⊂ PSL (n, R) is a lattice, the action Pn−1 (R) is a special case of the ﬂag manifold action treated in Theorem 1.6. Hence, Pn−1 (R) and Y are SOE if and only if Y is conjugate to an induction of either Pn−1 (R) or its double cover Rn /R+ , where

= {±1} · ⊂ GL (n, R). Finally, combining the work of [9] and the above OE superrigidity results, we classify up to stable orbit equivalence, the lattice actions on Rn and on ﬂag manifolds; see Theorems 6.2 and 6.3. At the same time, we compute the outer automorphism group of the associated orbit equivalence relation.

2. Preliminaries Let R be a countable measured equivalence relation on the standard measure space (X , μ), meaning that R is, on the one hand, a Borel subset of X × X and, on the other hand, an equivalence relation on X with countable equivalence (1) (1) classes, such that the σ -ﬁnite measures μl and μr on R deﬁned by (1) μl (Y ) = #{y ∈ X | (x, y) ∈ Y } dμ(x) , X

(1)

μr (Y ) =

X

#{x ∈ X | (x, y) ∈ Y } dμ(y) , (1)

(1)

are mutually absolutely continuous: μl ∼ μr . One calls R ergodic if every R-saturated measurable subset of X is negligible or conegligible. The measure (1) (2) μ is said to be invariant if μl = μr . A II1 equivalence relation is a countable measured ergodic equivalence relation with invariant probability measure. We now recall Zimmer’s deﬁnition of property (T) for a countable measured equivalence relation R. To this end, ﬁrst deﬁne R(2) = {(x, y, z) ∈ X × X × X | x Ry and y Rz}. Note that R(2) comes equipped with a canonical measure class, induced by the σ -ﬁnite measure (2) #{(y, z) ∈ R | (x, y, z) ∈ Y } dμ(x) . μl (Y ) = X

cocycle and orbit superrigidity for lattices in sl(n, R) / 425 (2)

(2)

Note that one deﬁnes absolutely continuous measures μm , respectively, μr , by integrating over y, respectively, z. But we only make use of the measure (2) class given by μl . • A 1-cocycle of R with values in the unitary group U (K ) of a Hilbert space K

is a Borel map c : R → U (K ) satisfying c(x, z) = c(x, y)c(y, z) for almost all (x, y, z) ∈ R(2) . • Suppose that c : R → U (K ) is a 1-cocycle of R. – An invariant unit vector of c is a Borel map ξ : X → K satisfying ξ (x) = c(x, y)ξ (y) for almost all (x, y) ∈ R and ξ (x) = 1 for almost all x ∈ X . – A sequence of almost invariant unit vectors of c is a sequence of Borel maps ξn : X → K satisfying

ξn (x) − c(x, y)ξn (y) → 0

for almost all

(x, y) ∈ R

and ξn (x) = 1 for all n ∈ N and almost all x ∈ X .

definition 2.1. A countable measured equivalence relation R is said to have property (T) in the sense of Zimmer if the following holds: every 1-cocycle of R with values in the unitary group of a Hilbert space and admitting a sequence of almost invariant unit vectors admits a unit invariant vector. We leave the proof of the following lemma as an exercise.

lemma 2.2. Let R be a countable measured equivalence relation on (X , μ) and Y ⊂ X a measurable subset such that the R-saturation of Y is conegligible. Then, the restriction of R to Y has property (T) if and only if R has property (T). 3. Property (T) for Actions of Locally Compact Groups We recalled above Zimmer’s deﬁnition of property (T) for a II1 equivalence relation. In fact, one can deﬁne property (T) for measured groupoids in general; see [1]. We do not need this generality in this paper, but we do need the concept of property (T) for nonsingular actions of locally compact second countable (l.c.s.c.) groups on measure spaces. For a groupoid approach to this deﬁnition, we refer to [1]. For the convenience of the reader, we gather in this section the necessary concepts and results and present them in an operator algebra framework. All von Neumann algebras are supposed to have separable predual and all locally compact groups are supposed to be second countable.

426 / s. popa and s. vaes

Given a von Neumann algebra M ⊂ B (H) and a Hilbert space K , we make use of the W ∗ -module M ⊗ K , that can be deﬁned as {T ∈ B (H, H ⊗ K ) | Tx = (x ⊗ 1)T , ∀x ∈ M }. By construction, M ⊗ K is an (M ⊗ B (K ))-M-bimodule and carries the strong∗ -topology inherited from B (H, H ⊗ K ). If M is a von Neumann algebra, we equip Aut (M) with the Polish topology making the functions Aut (M) → M∗ : α → ω ◦ α continuous for all ω ∈ M∗ . An action α of an l.c.s.c. group G on a von Neumann algebra M, denoted α G M, is a continuous group morphism α : G → Aut (M). α

• A 1-cocycle of an action G M with values in the unitary group U (K ) of

a Hilbert space K , is a strongly continuous map c : G → U (M ⊗ B (K )) satisfying c(gh) = c(g)(αg ⊗ id)(c(h)) for all g, h ∈ G. Note that by Theorem 3 in [16], it makes no difference to assume only that c is a measurable map, with the previous formula holding for almost all (g, h) ∈ G × G. α • An invariant unit vector of the 1-cocycle c of G M, is an element ξ in the W ∗ -module M ⊗ K satisfying ξ ∗ ξ = 1 and c(g)(αg ⊗ id)(ξ ) = ξ for all g ∈ G. α • A sequence of almost invariant unit vectors of the 1-cocycle c of G M, is a sequence ξn ∈ M ⊗ K satisfying ξn∗ ξn = 1 for all n and c(g)(αg ⊗ id)(ξn ) − ξn → 0 ∗ -strongly, uniformly on compact subsets of G. • The action G M is said to have property (T) if every 1-cocycle with values in the unitary group of a Hilbert space and admitting a sequence of almost invariant unit vectors, admits an invariant unit vector. The relation between this notion of property (T) for an action and Zimmer’s property (T) for a measured equivalence relation is given by the following straightforward lemma.

lemma 3.1. Let (X , μ) be a nonsingular action of the countable group and consider the orbit equivalence relation R( X ). If the action L∞ (X ) has property (T) in the sense above, then the equivalence relation R( X ) has property (T) in the sense of Deﬁnition 2.1. The converse holds when (X , μ) is essentially free. Note that by combining Lemmas 2.2 and 3.1, we obtain the following: if (X , μ) is a nonsingular, essentially free ergodic action and Y ⊂ X is nonnegligible, then the restricted equivalence relation R( X )|Y has property (T) iff the action (X , μ) has property (T). The following result is proven for discrete groups in [25], for weakly mixing actions, and in [6], for arbitrary actions (see also [2, corollary 5.16]). These

cocycle and orbit superrigidity for lattices in sl(n, R) / 427

methods work as well in the locally compact case and for completeness, we give a proof in a von Neumann algebra setup. α

proposition 3.2. Let G M be an action of the l.c.s.c. group G on the von Neumann algebra M. Suppose that τ is a faithful normal tracial state on M, invariant under α. Then, G M has property (T) if and only if the group G has property (T). Combining Proposition 3.2 and Lemma 3.1, it follows that for every p.m.p. action (X , μ) of a countable property (T) group , the orbit equivalence relation R( X ) has property (T) in the sense of Deﬁnition 2.1. Theorem 1.1 provides the ﬁrst examples of property (T) equivalence relations that do not arise in this way. Proof of Proposition 3.2. Suppose ﬁrst that G has property (T). Let c : G → U (M ⊗ B (K )) be a 1-cocycle of G M having ξn ∈ M ⊗ K as a sequence of almost invariant unit vectors. Deﬁne the unitaries ug on L2 (M, τ ) by extending αg . Then, π : G → U ( L2 (M, τ ) ⊗ K ) : π (g) = c(g)(ug ⊗ 1) is a unitary representation of G and we can view ξn as a sequence of almost invariant unit vectors. Since G has property (T), π admits an invariant unit vector. Even more, we ﬁnd a sequence ηn ∈ L2 (M, τ ) ⊗ K of π -invariant vectors satisfying ηn − ξn → 0. Since ξn = 1, it follows that ξn∗ ηn − 1 2 → 0. Hence, the right support projection of ξn∗ ηn converges strongly to 1. A fortiori, the right support projection of ηn converges to 1. We view ηn as a closed operator from L2 (M, τ ) to L2 (M, τ ) ⊗ K . Taking the polar decomposition of ηn , we ﬁnd vn ∈ M ⊗ K satisfying c(g)(αg ⊗ id)(vn ) = vn for all g ∈ G, n ∈ N and such that vn∗ vn is a sequence of projections in M converging strongly to 1. Deﬁne the von Neumann algebra M ⊗ B (K ) M ⊗ K N= . (M ⊗ K )∗ M Deﬁne the action (γg ) of G on N by γg

a e

b c(g)(αg ⊗ id)(a)c(g)∗ = (αg ⊗ id)(e)c(g)∗ f

c(g)(αg ⊗ id)(b) αg ( f )

.

428 / s. popa and s. vaes

Deﬁne p = 01 00 , q = 1 − p, and wn = 00 v0n . Then, wn is a sequence of partial isometries in the ﬁxed-point algebra N G , satisfying wn ∈ pN G q and wn∗ wn → q strongly. It follows that q ≺ p in the von Neumann algebra N G . So, we ﬁnd v ∈ M ⊗ K satisfying c(g)(αg ⊗ id)(v) = v for all g and v ∗ v = 1. Hence, G M has property (T). Suppose conversely that G M has property (T). Let π : G → U (K ) be a strongly continuous unitary representation of G admitting ξn as a sequence of almost invariant unit vectors. By [3, theorem 2.12.9], it is sufﬁcient to prove that π has a nonzero ﬁnite-dimensional π (G)-invariant subspace. Deﬁne c : G → U (M ⊗ B (K )) : c(g) = 1 ⊗ π(g). Obviously, c is a 1-cocycle of G M having 1 ⊗ ξn as a sequence of almost invariant unit vectors. By property (T) of G M, we ﬁnd ξ ∈ M ⊗ K satisfying ξ ∗ ξ = 1 and c(g)(αg ⊗ id)(ξ ) = ξ for all g ∈ G. Denoting again by u : g → ug the representation of G on L2 (M, τ ) obtained by extending αg , we ﬁnd that u ⊗ π admits an invariant unit vector ξ . Identify L2 (M, τ ) ⊗ K with the Hilbert space of Hilbert-Schmidt operators from L2 (M, τ ) to K . Then, T := ξ ξ ∗ is a nonzero trace-class operator on K satisfying π (g)T π(g)∗ = T for all g ∈ G. So, for ε > 0 sufﬁciently small, the spectral projection χ[ε,+∞) (T ) projects onto a nonzero ﬁnite-dimensional π(G)-invariant subspace of K . The following is a slight generalization of [2, theorem 5.3]. When H M is an action, we denote by M H the von Neumann algebra of H-ﬁxed points. α

lemma 3.3. Let G M be an action of the l.c.s.c. group G on the von Neumann algebra M. Let H G be a closed normal subgroup and assume that there is a ∗ -isomorphism θ : M → L∞ (H) ⊗ M H satisfying θ ◦ α = (ρ ⊗ id) ◦ θ for all h h h ∈ H, where ρh denotes the right translation by h on L∞ (H). Then, G M has property (T) if and only if G/H M H has property (T). α

Proof. We say that two 1-cocycles c1 , c2 of G M with values in U (K ) are unitarily equivalent if there exists a unitary v ∈ U (M ⊗ B (K )) satisfying c1 (g) = α vc2 (g)(αg ⊗ id)(v ∗ ) for all g ∈ G. We denote by H1 (G M, U (K )) the set of equivalence classes of 1-cocycles. In the ﬁrst part of the proof, we show that the obvious map : H1 (G/H M H , U (K )) → H1 (G M, U (K )) : (c) = c ◦ π

with

π : G → G/H , is a bijection. In the second part of the proof, we show that this map and its inverse preserve the property of having invariant, respectively, almost

cocycle and orbit superrigidity for lattices in sl(n, R) / 429

invariant, vectors. Both parts together show that property (T) of G M is equivalent with property (T) of G/H M H . It is straightforward to check that is well deﬁned and injective. Suppose that c : G → U (M ⊗ B (K )) is a 1-cocycle of G M. In order to prove that c is in the range of , it sufﬁces to prove that c is unitarily equivalent with c satisfying c (h) = 1 for all h ∈ H. Identify throughout U (M ⊗ B (K )) with U ( L∞ (H) ⊗ M H ⊗ B (K )) and view the latter as measurable functions H → U (M H ⊗ B (K )), modulo equality almost everywhere. By [16, theorem 1], take a measurable map ϕ : H × H → U (M H ⊗ B (K )) such that c(h) = ϕ(·, h) for all h ∈ H. Since c is a cocycle, we ﬁnd that ϕ(k, hg) = ϕ(k, h)ϕ(kh, g)

for almost all (k, h, g) ∈ H × H × H.

By the Fubini theorem, take k0 ∈ H such that for almost all (h, g) ∈ H × H, the previous equality holds for (k0 , h, g). Deﬁne the unitary v ∈ U (M ⊗ B (K )) as the function H → U (M H ⊗ B (K )) given by h → ϕ(k0 , k−1 0 h) and set c (g) = ∗ vc(g)(αg ⊗ id)(v ). By construction, c (g) = 1 for almost all g ∈ H and hence for all g ∈ H by continuity. It is an exercise to check that the 1-cocycle c ∈ H1 (G/H M H , U (K )) has an invariant unit vector if and only if (c) has. Also, a sequence of almost invariant unit vectors for c deﬁnes a sequence of almost invariant unit vectors for (c). Finally, suppose that ξn ∈ M ⊗ K is a sequence of almost invariant unit vectors for (c). In order to conclude the proof of the lemma, it sufﬁces to show that there exists a sequence ηn ∈ M H ⊗ K satisfying ηn∗ ηn = 1 for all n and ξn − ηn → 0 ∗ -strongly. Deﬁne the sets X = {v ∈ M ⊗ K | v ∗ v = 1} and Y = {v ∈ M H ⊗ K | v ∗ v = 1}. Through the isomorphism M ∼ = L∞ (H) ⊗ M H , we identify X with the set of measurable functions from H to Y (modulo equality almost everywhere). Take a bounded metric d0 on Y inducing the strong∗ topology. Let μ be a probability measure on H in the same measure class as the Haar measure. Following [16, p. 5], deﬁne the metric d on X by d(v, w) = d0 (v(h), w(h))dμ(h). H

strong∗

Then, d induces the topology on X . It is easy to check that when vn , wn ∈ X such that d(vn , wn ) is summable, then vn (h) − wn (h) → 0 ∗ -strongly for almost every h ∈ H (see [16, proposition 6]). View Y ⊂ X as constant functions. We have to prove that d(ξn , Y ) → 0. Suppose the contrary. Write H as an increasing union of compact subsets Hn . After passage to a subsequence, we ﬁnd ε > 0 such that d(ξn , Y ) > ε for all n and such that

430 / s. popa and s. vaes

d(ξn , (αg ⊗ id)(ξn )) < 2−n

for all n ∈ N, g ∈ Hn .

It follows that for all g ∈ H, we have d0 (ξn (h), ξn (hg)) → 0

for almost all h ∈ H .

By the Fubini theorem, take h0 ∈ H such that d0 (ξn (h0 ), ξn (h0 g)) → 0 for almost all g ∈ H. It follows that d(ξn , ξn (h0 )) → 0, contradicting the assumption that d(ξn , Y ) > ε for all n. If M is a commutative von Neumann algebra, an action H M satisfying the condition in Lemma 3.3 is called essentially free and proper. More precisely:

definition 3.4. The nonsingular action G (X , μ) of the l.c.s.c. group G on the standard measure space (X , μ) is called essentially free and proper if there exists a measurable map π : X → G such that π(g · x) = gπ(x) for almost all (g, x) ∈ G × X . Equivalently, there exists a ∗ -isomorphism L∞ (X , μ) → L∞ (G) ⊗ L∞ (X , μ)G conjugating the G-action on L∞ (X , μ) with the action (ρg ⊗ id)g∈G , where ρg denotes the right translation by g on L∞ (G). Indeed, suppose that we are given π : X → G as above. Denote by X /G the space of ergodic components of G X and deﬁne the map : X → G × X /G : x → (π (x), x). Then, is invertible with the inverse being well deﬁned and given by (g, x) → g −1 π (x)−1 · x. Also, conjugates the action G X with the action ρg × id.

proposition 3.5. Let G be an l.c.s.c. group with closed subgroups H1 , H2 . Then, H1 L∞ (G /H2 ) has property (T) if and only if H2 L∞ (G /H1 ) has property (T). Proof. Set M = L∞ (G ) and G = H1 × H2 acting by left-right translations on M: (α(g,h) (F))(x) = F(g −1 xh)

for all

F ∈ L∞ (G ), x ∈ G , g ∈ H1 , h ∈ H2 .

We apply Lemma 3.3 to G M and the closed normal subgroups H = Hi , i = 1, 2 of G. By the Effros-Mackey theorem (see e.g., [14, theorem II.12.17]), the quotient map G → G /H admits a Borel lifting and hence, there exists an H-equivariant isomorphism M → L∞ (H) ⊗ M H . So, by Lemma 3.3, property (T) of G M is equivalent with property (T) of H1 L∞ (G /H2 ) as well as with property (T) of H2 L∞ (G /H1 ).

cocycle and orbit superrigidity for lattices in sl(n, R) / 431

4. The Lattice Actions Rn Have Property (T) Recall from [6] that if (X , μ) is a free ergodic p.m.p. action, then property (T) of R( X ) in the sense of Zimmer is equivalent with property (T) of the group (see also [25, proposition 2.4] and Proposition 3.2 above). If is a property (T) group, the fundamental group of R( X ) is countable for any free ergodic p.m.p. action (see [12, corollary 1.8] if, moreover, has inﬁnite conjugacy classes (ICC) and see [13, theorem 5.9] for the general case). But more is true: we proved in [22, theorem 6.1] that the fundamental group of a II1 equivalence relation R on (X , μ) is countable whenever the full group [R] contains a property (T) group that implements an ergodic action on (X , μ). As a result, the following theorem is rather surprising: we obtain a II1 equivalence relation R with property (T) and fundamental group R+ ; hence, none of the Rt can be implemented by a free action of a group and none of the Rt can be implemented by a possibly nonfree action of a property (T) group. The following result covers Theorem 1.1 stated in the introduction.

theorem 4.1. Let < SL (n, R) be a lattice and let R be the II1 equivalence relation obtained by restricting the orbit equivalence relation R( Rn ) to a set of Lebesgue measure 1. If n ≥ 4, the equivalence relation R has property (T) in the sense of Zimmer, but nevertheless F (R) = R+ . In particular, • none of the equivalence relations Rt , t > 0 can be implemented by a free action

of a group, • none of the equivalence relations Rt , t > 0 can be implemented by a possibly

nonfree action of a property (T) group. In fact, Rt does not contain any ergodic subequivalence relation that can be implemented by a property (T) group. Proof. By the remark following Lemma 3.1, proving property (T) of R amounts to proving property (T) for the action L∞ (Rn ). Deﬁne the l.c.s.c. group G = SL (n, R) and set H1 = . Consider the linear action G Rn and set H2 = {A ∈ SL (n, R) | Ae1 = e1 }, where e1 denotes the ﬁrst basis vector of Rn . By construction, the action L∞ (Rn ) can be viewed as H1 L∞ (G /H2 ). Hence, by Proposition 3.5, property (T) for this last action is equivalent with property (T) of H2 L∞ (G /H1 ). This action admits a ﬁnite invariant measure, because H1 is a lattice in G . Moreover, H2 ∼ = SL (n − 1, R) Rn−1 , which has property (T) for n ≥ 4. So, it follows from Proposition 3.2 that H2 L∞ (G /H1 ) has property (T). The action on Rn by multiples of the identity matrix scales the Lebesgue measure and commutes with the action of . Hence, the fundamental group

432 / s. popa and s. vaes

of R equals R+ . The statements about implementing Rt by group actions follow from the discussion preceding the theorem.

remark 4.2. Note that a similar reasoning as in the proofs of Lemmas 3.1 and 3.3, and Proposition 3.2 show that those statements hold true if we replace everywhere “property (T)” by “amenable.” By applying this to the action SL (2, Z) R2 , one recovers a result from [2], showing that the action SL (2, Z) R2 is amenable and hence, L∞ (R2 ) SL (2, Z) is isomorphic to the unique hyperﬁnite II∞ factor, by Connes’s theorem [4]. This also shows that the action SL (3, Z) R3 is not amenable, nor has property (T). remark 4.3. Denote by R the II1 equivalence relation constructed in Theorem 4.1. Since R has property (T), [20, theorem 4.4.1] and [17, corollaire 18] imply that R is ﬁnitely generated. Hence, R has ﬁnite cost and ﬁnite ﬁrst 2 Betti number. Since the fundamental group of R equals R+ , it follows from [10, proposition II.6], respectively [11, corollaire 5.7], that actually cost R = 1 (2) and β1 (R) = 0. 5. Cocycle and OE Superrigidity Theorems 5.1. Proof of Theorem 1.3

We prove in this section the cocycle superrigidity Theorem 1.3 as a consequence of the more general Theorem 5.3 below. We do not know whether SL (n, R) Rn is Uﬁn -cocycle superrigid for n = 3, 4. On the other hand, it is generally believed that whenever one has a Uﬁn cocycle superrigidity result, it actually holds true for more general target groups than the Uﬁn -class. However, the following result, which we’ll prove at the end of this subsection, shows that there even exist locally compact target groups for 1-cocycles for the action SL (n, R) Rn , which cannot be untwisted.

proposition 5.1. Let n ≥ 3. The action SL (n, Z) Rn admits a 1-cocycle with values in SL (n − 1, R) Rn−1 that is not cohomologous to a group morphism. Recall from [19] the following deﬁnition of s-malleability of a measure preserving action.

definition 5.2. Let be an l.c.s.c. group and (X , μ) an action preserving the ﬁnite or inﬁnite measure μ. The action is called s-malleable if there exists

cocycle and orbit superrigidity for lattices in sl(n, R) / 433 • a 1-parameter group (αt )t∈R of measure preserving transformations of

X × X , and • an involutive measure preserving transformation β of X × X ,

such that • αt and β commute with the diagonal action X × X , • α1 (x, y) ∈ {y} × X for almost all (x, y) ∈ X × X , • β(x, y) ∈ {x} × X for almost all (x, y) ∈ X × X , and • αt ◦ β = β ◦ α−t for all t ∈ R.

Theorem 0.1 in [19] says the following. Let (X , μ) be an s-malleable, probability measure preserving action and < a normal subgroup with the relative property (T) such that the restriction of (X , μ) to is weakly mixing. Then every 1-cocycle of (X , μ) with values in a Polish group of ﬁnite type, is cohomologous to a group morphism. Recall here that one of the equivalent formulations of weak mixing for a p.m.p. action G (X , μ) of a countable group G is the ergodicity of the diagonal action G (X × X , μ × μ). If G (X , μ) is a weakly mixing p.m.p. action and G (Y , η) is any ergodic p.m.p. action, then the diagonal action G X × Y is ergodic. In particular, the diagonal action G X ×···×X k times

is ergodic for every k, once it is ergodic for k = 2. For inﬁnite measure preserving actions, things are more complicated and, for instance, the diagonal action SL (n, Z) Rn × · · · × Rn k times

is ergodic if and only if k ≤ n − 1. This partially explains the formulation of the following result.

theorem 5.3. Let (X , μ) be an inﬁnite measure preserving, s-malleable action. Assume that • the diagonal action X × X has property (T), • the 4-fold diagonal action X × X × X × X is ergodic.

Then, (X , μ) is Uﬁn -cocycle superrigid.

434 / s. popa and s. vaes

The proof of Theorem 5.3 follows entirely the setup of the proof of [19, theorem 0.1]. But, one has to be careful at those places in [19] where weak mixing is applied. The proper way to deal with these issues lies in the following lemma distilled from the proof of [8, lemma 3.1].

lemma 5.4. Let (Z, d) be a Polish space with separable complete metric d and (αg )g∈G a continuous action of a Polish group G by homeomorphisms of Z. Assume that d is (αg )g∈G -invariant. Let be an l.c.s.c. group and (X , μ), (Y , η) nonsingular actions. Let F : X × Y → Z be a measurable map satisfying F(g · x, g · y) = αω(g,x) (F(x, y)) for almost all (x, y) ∈ X × Y , g ∈ , where ω : × X → G is some measurable map. Assume the diagonal action X × Y × Y satisﬁes L∞ (X × Y × Y ) = ∞ L (X ) ⊗ 1 ⊗ 1 (which holds in particular if the diagonal action X × Y × Y is ergodic). Then, there exists a measurable H : X → Z with F(x, y) = H(x) for almost all (x, y) ∈ X × Y . Proof. Deﬁne the map G : X × Y × Y → R : G(x, y, z) = d(F(x, y), F(x, z)). Since d is (αg )g∈G -invariant, the map G is invariant under the diagonal -action. By our assumption, G(x, y, z) = G0 (x) for almost all (x, y, z) and some measurable map G0 : X → R+ . We claim that G0 (x) = 0 for almost all x ∈ X . Let δ > 0 and assume that G0 (x) ≥ δ for all x in a nonnegligible subset U of X . Cover Z by a sequence (Bn )n∈N of balls of diameter strictly smaller than δ. Write, for every x ∈ X , Fx : y → F(x, y). By the Fubini theorem, for almost every x ∈ U , we have G(x, y, z) ≥ δ for almost all (y, z) ∈ Y × Y . Hence, for almost every x ∈ U , we have that Fx−1 (Bn ) is negligible for every n, which is absurd. So, G(x, y, z) = 0 for almost all (x, y, z) ∈ X × Y × Y . Again by the Fubini theorem, take z ∈ Y such that d(F(x, y), F(x, z)) = 0 for almost all (x, y) ∈ X × Y . Putting H(x) := F(x, z), we are done. Proof of Theorem 5.3. Let N be a II1 factor and G ⊂ U (N) a closed subgroup. Let ω : × X → G be a 1-cocycle, meaning that for all g, h ∈ , we have ω(gh, x) = ω(g, h · x) ω(h, x)

for almost all x ∈ X .

Deﬁne the following 1-cocycles for the diagonal action X × X .

cocycle and orbit superrigidity for lattices in sl(n, R) / 435

ω0 : × X × X → G : ω0 (g, x, y) = ω(g, x)

and

ωt : × X × X → G : ωt (g, x, y) = ω0 (g, αt (x, y)) . Deﬁne the action (ρg )g∈ of by automorphisms of L∞ (X ) ⊗ N by the formula (ρg −1 (F))(x) = ω(g, x)∗ F(g · x)ω(g, x) . Denote by B the von Neumann subalgebra of (ρg )g∈ -ﬁxed points.

claim. Whenever p is a nonzero projection in B, there exists a measurable function ϕ : X × X → N and a nonzero projection q ∈ B such that q ≤ p and such that for all g ∈ , we have ω(g, x)ϕ(x, y) = ϕ(g · x, g · y)ω(g, y),

ϕ(x, y)ϕ(x, y)∗ = q(x),

ϕ(x, y)∗ ϕ(x, y) = q(y) for almost all (x, y) ∈ X × X .

proof of the claim. Since p is (ρg )g∈ -invariant, the function x → τ (p(x)) is -invariant and hence constantly equal to 0 < λ ≤ 1. Let p0 ∈ N be a projection with τ (p0 ) = λ. It follows that inside L∞ (X ) ⊗ N, the projections p and 1 ⊗ p0 are equivalent. Take a partial isometry v ∈ L∞ (X ) ⊗ N such that v ∗ v = p and vv ∗ = 1 ⊗ p0 . Deﬁne η(g, x) = v(g · x)ω(g, x)v(x)∗ and note that η is a 1-cocycle for X with values in U (p0 Np0 ). Set η0 (g, x, y) = η(g, x) and, for all n ≥ 1, ηn (g, x, y) = η0 (g, α2−n (x, y)). 2 Deﬁne the Hilbert space K = ∞ k=1 p0 L (N)p0 . We deﬁne the following 1-cocycle of X × X with values in the unitary group U (K ) of K : (c(g, x, y)ζ )k = η(g, x) ζk ηk (g, x, y)∗ . Deﬁne the map ξn : X × X → (K )1 by the formula ⎧ ⎨τ (p )−1/2 p if k = n , 0 0 (ξn (x, y))k = ⎩0 if k = n . One checks that ξn is a sequence of almost invariant unit vectors. Since X × X has property (T), we ﬁnd an invariant unit vector, that is, a measurable map ξ : X × X → K with ξ (x, y) = 1 for almost all (x, y) and for all g ∈ , ξ (g · x, g · y) = c(g, x, y) ξ (x, y)

436 / s. popa and s. vaes

almost everywhere. It follows that ξ (g · x, g · y)k = η(g, x) ξ (x, y)k ηk (g, x, y)∗ almost everywhere. In particular, for every k, the function (x, y) → ξ (x, y)k is -invariant and hence, constant. Since ξ (x, y) = 1 for almost all (x, y), we can pick k such that ξ (x, y)k = 0 for almost all (x, y). Taking the polar decomposition of ξ (x, y)k , we ﬁnd a nonzero partial isometry ψ ∈ L∞ (X × X ) ⊗ p0 Np0 satisfying ψ(g · x, g · y) ηk (g, x, y) = η(g, x) ψ(x, y) almost everywhere. We set ϕ0 (x, y) := v(x)∗ ψ(x, y)vk (x, y), where v0 (x, y) = v(x) and vk (x, y) = v0 (α2−k (x, y)). It follows that ω(g, x) ϕ0 (x, y) = ϕ0 (g · x, g · y) ωt0 (g, x, y) where t0 = 2−k . Set r(x, y) = ϕ0 (x, y)ϕ0 (x, y)∗ . It follows that r(g · x, g · y) = ω(g, x) r(x, y) ω(g, x)∗ almost everywhere. By Lemma 5.4, we ﬁnd a projection q ∈ L∞ (X ) ⊗ N such that r(x, y) = q(x) almost everywhere. Then, q is a nonzero projection in B and q ≤ p. Similarly, ϕ0 (x, y)∗ ϕ0 (x, y) = q0 (αt0 (x, y)) where q is a projection in L∞ (X ) ⊗ N and q0 (x, y) = q (x). Set q0 (x, y) = q(x) and qt (x, y) = q0 (αt (x, y)). We now construct ϕ1 : X × X → N such that ϕ1 (x, y)ϕ1 (x, y)∗ = q(x), ϕ1 (x, y)∗ ϕ1 (x, y) = q2t0 (x, y) and ω(g, x) ϕ1 (x, y) = ϕ1 (g · x, g · y) ω2t0 (g, x, y) almost everywhere. Continuing the same procedure k times (remember that t0 = 2−k ), we will have found ϕ = ϕk : X × X → N satisfying ϕ(x, y)ϕ(x, y)∗ = q(x), ϕ(x, y)∗ ϕ(x, y) = q1 (x, y) = q(y) and ω(g, x) ϕ(x, y) = ϕ(g · x, g · y) ω1 (g, x, y) = ϕ(g · x, g · y) ω(g, y) , hence proving the claim. In fact, it sufﬁces to take ϕ1 (x, y) = ϕ0 (x, y)ϕ0 (β(α2t0 (x, y)))∗ and to use that (αt )t∈R is a 1-parameter group, β ◦ αt = α−t ◦ β and β(x, y) ∈ {x} × Y for almost all (x, y). So, the claim above has been proven. Using the claim and a maximality argument, we ﬁnd a measurable function ϕ : X × X → U (N) such that ω(g, x) ϕ(x, y) = ϕ(g · x, g · y) ω(g, y)

cocycle and orbit superrigidity for lattices in sl(n, R) / 437

almost everywhere. Set H(x, y, z) = ϕ(x, y)ϕ(y, z). It follows that H(g · x, g · y, g · z) = ω(g, x) H(x, y, z) ω(g, z)∗ for almost all (x, y, z). By Lemma 5.4 and because the 4-fold diagonal action X × X × X × X is ergodic, H is essentially independent of its second variable. So, we ﬁnd a measurable F : X × X → U (N) such that ϕ(x, y) = F(x, z)ϕ(y, z)∗ for almost every (x, y, z). By the Fubini theorem, take z ∈ X such that the previous formula holds for almost all (x, y). Set ψ(x) = F(x, z)∗ and G(y) = ϕ(y, z)∗ . It follows that ψ(g · x) ω(g, x) ψ(x)∗ = G(g · y) ω(g, y) G(y)∗ almost everywhere. Hence, the left-hand side is independent of x and we have found a group morphism δ : G → U (N) and a measurable map ψ : X → U (N) such that ω(g, x) = ψ(g · x)∗ δ(g) ψ(x) almost everywhere. Consider the quotient Polish space U (N)/G with the induced metric, which is invariant under left multiplication by elements of U (N). Write F : X → U (N)/G : F(x) = ψ(x)G . It follows that F(g · x) = δ(g)F(x) almost everywhere. By Lemma 5.4, F is essentially constant. So, we ﬁnd a unitary u ∈ U (N) and a measurable map w : X → G such that ψ(x) = uw(x) almost everywhere. Replacing δ by u ∗ δ( · )u, it follows that ω(g, x) = w(g · x)∗ δ(g)w(x) almost everywhere. In particular, δ(g) ∈ G and we are done. As a principle (cf. [19, proposition 3.6]), once the restriction of a 1-cocycle ω : G × X → G to a closed subgroup H < G is cohomologous to a group morphism H → G and if H is sufﬁciently normal in G and acts sufﬁciently mixingly on X , the entire 1-cocycle ω is cohomologous to a group morphism G → G . In our setting, we need the following:

lemma 5.5. Let G (X , μ) be a nonsingular action of the l.c.s.c. group G and ω : G × X → G a 1-cocycle with values in the Polish group of ﬁnite type G . Let H < G be a closed subgroup and assume that ω|H is cohomologous to a group morphism H → G . If for every g ∈ G, the diagonal action of the group H ∩ gHg −1 on X × X is ergodic, then ω is cohomologous to a morphism G → G . Proof. We may assume that for every h ∈ H, we have ω(h, x) = δ(h) for almost every x ∈ X , where δ : H → G is a continuous group morphism. Let g ∈ G and put F(x) = ω(g, x). Using the cocycle equation, it follows that for all h ∈ H ∩

438 / s. popa and s. vaes

g −1 Hg, we have F(h · x) = δ(ghg −1 )F(x)δ(h)−1 almost everywhere. By Lemma 5.4, F is essentially constant. So, we have shown that for every g ∈ G, the map x → ω(g, x) is essentially constant. It follows that ω(g, x) = δ(g). After showing the following lemma, we can prove Theorem 1.3 and Proposition 5.1.

lemma 5.6. Let be a lattice in SL (n, R) and consider the linear action of on Rn . Let X (k) := Rn × · · · × Rn k times

be the k-fold diagonal action. Then, X (k) • is ergodic if and only if k ≤ n − 1, and • has property (T) if and only if k ≤ n − 3 or k ≥ n (the latter part being as

interesting as the trivial group {e} having property (T)). Proof. Writing the elements of Rn as column vectors, identify, up to measure 0, X (n) with GL (n, R), with the -action given by left multiplication. The determinant function is invariant and not essentially constant, proving that X (k) is nonergodic for k ≥ n. It also follows that for k ≥ n, the action X (k) is essentially free and proper (see Deﬁnition 3.4). Hence, it has property (T) because the trivial group has property (T). Now let k ≤ n − 1. Denoting by (ei )i=1,...,n the standard basis vectors in Rn , the orbit of (e1 , . . . , ek ) ∈ X (k) under the diagonal SL (n, R)-action has complement of measure 0, so that we can identify X (k) with SL (n, R)/H, where H = {A ∈ SL (n, R) | Aei = ei for all i = 1, . . . , k}. Observe that H ∼ = SL (n − k, R) Mn−k (R), where Mn−k (R) denotes the additive group of (n − k) × (n − k) matrices on which SL (n − k, R) acts by multiplication. By Moore’s ergodicity theorem (see e.g., [26, theorem 2.2.6]), H \ SL (n, R) is ergodic and hence, SL (n, R)/H is ergodic. By Proposition 3.5, property (T) of SL (n, R)/H is equivalent with property (T) of H \ SL (n, R). By Proposition 3.2, the latter is equivalent with the group H having property (T), which is in turn equivalent with n − k ≥ 3.

Proof of Theorem 1.3. Observe that part (1) is a special case of part (3), by taking k = 1 and H = {1}. We start by proving part (3). Set X = Mn,k (R). The action

cocycle and orbit superrigidity for lattices in sl(n, R) / 439

X by left multiplication is s-malleable. It sufﬁces to take αt (A, B) = ( cos (πt/2)A + sin (πt/2)B, − sin (πt/2)A + cos (πt/2)B)

and

β(A, B) = (A, −B) whenever t ∈ R and A, B ∈ Mn,k (R). Moreover, X can be viewed as the k-fold diagonal action Rn × · · · × Rn . By Lemma 5.6 and because n ≥ 4k + 1, the diagonal action X × X has property (T) and the 4-fold diagonal action X × X × X × X is ergodic. So, by Theorem 5.3, X is Uﬁn -cocycle superrigid. Since the diagonal action X × X is ergodic and is a normal subgroup of G, Lemma 5.5 implies that G X is Uﬁn -cocycle superrigid. It remains to prove part (2) of the theorem. By part (1), we already know Rn is Uﬁn -cocycle superrigid. Deﬁne, for every x ∈ Zn , x = {g ∈ | gx = x}. We claim that the diagonal action x Rn × Rn is ergodic for every x ∈ Zn . Once this claim is proven, Lemma 5.5 implies that Zn Rn is Uﬁn -cocycle superrigid. For x = 0, the claim follows from Lemma 5.6. Now let x = 0. Deﬁne the closed subgroup H = {g ∈ SL (n, R) | ge1 = e1 } of SL (n, R). Exactly as in the proof of Lemma 5.6, when n ≥ 4, the diagonal action Rn × Rn of any lattice ⊂ H is ergodic. Take g0 ∈ SL (n, Q) with x = g0 e1 . Since is a ﬁnite index subgroup of SL (n, Z), it follows that g0−1 x g0 contains a ﬁnite index subgroup of {g ∈ SL (n, Z) | ge1 = e1 } and hence, is a lattice in H. So, its diagonal action on Rn × Rn is ergodic. Then, the same is true for the diagonal action of x on Rn × Rn . Proof of Proposition 5.1. Write G = SL (n, R) and = SL (n, Z). As before, identify Rn with G/H, where H ∼ = SL (n − 1, R) Rn−1 . Let θ : G/H → G be a Borel lifting. Deﬁne the 1-cocycle ω : × G/H → H by gθ (xH) = θ(gxH)ω(g, xH) whenever g ∈ and xH ∈ G/H. Assume that ω is cohomologous to a group morphism δ : → H. This means that we can choose the lifting θ such that gθ (xH) = θ (gxH)δ(g) for g ∈ and almost all xH ∈ G/H. The image of any group morphism SL (n, Z) → SL (n − 1, R) Rn−1 is ﬁnite (see [23, theorem 6] for an elementary argument). So, we have found a ﬁnite index subgroup 0 ⊂ SL (n, Z) and a measurable map θ : Rn → SL (n, R) such that θ(gx) = gθ(x) for all g ∈ 0 and almost all x ∈ Rn . It follows that the map (x, y) → θ(x)−1 θ(y) is invariant under the diagonal 0 -action,

440 / s. popa and s. vaes

which is ergodic by Lemma 5.6. Hence, θ is essentially constant, which is a contradiction with the formula θ(gx) = gθ(x). 5.2. Proofs of Theorems 1.5 and 1.6

We will deduce Theorems 1.5 and 1.6 from the general Theorem 5.8 below, dealing with arbitrary actions of the form Mn,k (R)/H, where is a lattice and H < GL (k, R) a closed subgroup. First, observe that these actions are essentially free and ergodic.

lemma 5.7. Let n > k and < SL (n, R) any lattice. Let H ⊂ GL (k, R) be a closed subgroup. If ( − 1, −1) ∈ × H, put 0 = /{±1}, otherwise put = 0 . The action 0 Mn,k (R)/H is essentially free and ergodic. It never admits an invariant probability measure. It admits an inﬁnite invariant measure if and only if H is unimodular and satisﬁes det g = ±1 for all g ∈ H. Proof. By Lemma 5.6, Mn,k (R) is ergodic, because k < n. A fortiori, Mn,k (R)/H is ergodic. Denote V = span{e1 , . . . , ek } ⊂ Rn and deﬁne the closed subgroup H1 < SL (n, R) by H1 = {g ∈ SL (n, R) | gV = V and g|V ∈ H } . We can identify 0 Mn,k (R)/H with SL (n, R)/H1 . From this description, essential freeness follows. Also the statement about invariant measures follows, because H1 is unimodular if and only if H is unimodular and satisﬁes det g = ±1 for all g ∈ H.

theorem 5.8. Let H ⊂ GL (k, R) be a closed subgroup and (Y , η) any essentially free, ergodic, nonsingular action of the countable group . Suppose n ≥ 4k + 1. case −1 ∈ H . Let ⊂ PSL (n, R) be a lattice and put := {±1} · ⊂ GL (n, R). The actions Mn,k (R)/H and Y are SOE if and only if Y is conjugate to an induction of one of the following actions: 1) × H/N Mn,k (R)/N, where N H is an open normal subgroup with −1 ∈ N, or 2)

× H/N Mn,k (R)/N, where N H is an open normal subgroup with {±(1, 1)} −1 ∈ N.

cocycle and orbit superrigidity for lattices in sl(n, R) / 441

case −1 ∈ H . Let ⊂ SL (n, R) be a lattice. The actions Mn,k (R)/H and Y are SOE if and only if Y is conjugate to an induction of one of the following actions: 1) × H/N Mn,k (R)/N, where N H is an open normal subgroup, or H 2)(only when −1 ∈ ) × Mn,k (R)/({±1} · N), where N H is {±1} N an open normal subgroup.

remark 5.9. Given a stable orbit equivalence : Mn,k (R)/H → Y between the actions Mn,k (R)/H and Y , Theorem 5.8 provides a conjugacy between one of the listed actions and 1 Y1 such that Y is induced from 1 Y1 . In fact, moreover, one has (x) ∈ · (x) for almost every x ∈ Mn,k (R). If : X0 → Y0 is a SOE between X and Y , we can (and will tacitly) extend to a countable to 1, measurable : X → Y satisfying (g · x) ∈ · (x) for all g ∈ and almost all x ∈ X . If : X → Y is a SOE between the essentially free actions X and Y , we get a Zimmer 1-cocycle ω for the action X with values in , determined by the formula (g · x) = ω(g, x) · (x) almost everywhere. As a general principle, if the 1-cocycle ω is cohomologous to a group morphism → , the stable orbit equivalence is “essentially” given by a conjugacy of the actions: see [26, proposition 4.2.11], [24, lemma 4.7], and [8, theorem 1.8]. In our framework of non-p.m.p. actions, we again need such a principle: see Lemma 5.10 below. Recall from Deﬁnition 3.4 the notion of an essentially free and proper action. σ

lemma 5.10. Let G be an l.c.s.c. group and G (X , μ) a nonsingular, essentially free, ergodic action. Assume that σ is Uﬁn -cocycle superrigid. Let N G be an open normal subgroup such that the restricted action σ |N is proper. Let (Y , η) be an essentially free, ergodic, nonsingular action. If : X /N → Y is an SOE between the actions G/N X /N and Y , then there exists • a subgroup 1 < and a nonnegligible Y1 ⊂ Y such that Y is induced

from 1 Y1 , and • an open normal subgroup N1 G such that σ |N is proper 1

442 / s. popa and s. vaes

such that the actions G/N1 X /N1 and 1 Y1 are conjugate through the nonsingular isomorphism : X /N1 → Y1 and the group isomorphism δ : G/N1 → 1 . Furthermore, (x) ∈ · (x) for almost all x ∈ X . Proof. Let : X /N → Y be an SOE. By cocycle superrigidity of σ , take a measurable map ϕ : X → such that, writing (x) = ϕ(x)−1 · (x), we have (g · x) = δ(g) · (x) almost everywhere, where δ : G → is a continuous group morphism. Put N1 = Ker δ. So, N1 is an open normal subgroup of G. Then, N ∩ N1 is still open in G and we consider X /(N ∩ N1 ) with the quotient map π : X /(N ∩ N1 ) → X /N. It follows that we can view as a measurable map : X /(N ∩ N1 ) → Y such that (π (x)) ∈ · (x) for almost all x ∈ X /(N ∩ N1 ) and (g · x) = δ(g) · (x) almost everywhere. Using the facts that the countable group N/(N ∩ N1 ) acts freely and properly on X /(N ∩ N1 ), that is countable, and that : X /N → Y is locally a nonsingular isomorphism, it follows that X /(N ∩ N1 ) can be partitioned in a sequence of nonnegligible subsets (Un )n such that for every n, |Un is a nonsingular isomorphism between Un and a nonnegligible subset of Y . But then, for every nontrivial element g ∈ N1 /(N ∩ N1 ) and every n, we conclude that g · Un ∩ Un has measure 0. It follows that N1 /(N ∩ N1 ) acts freely and properly on X /(N ∩ N1 ). So, N1 acts freely and properly on X . Hence, we can form the quotient space X /N1 and view as a measurable map : X /N1 → Y such that (x) ∈ · (x) for almost all x ∈ X and (g · x) = δ(g) · (x) almost everywhere. Now, δ : G/N1 → is an injective group morphism. Still, X /N1 can be partitioned into a sequence of nonnegligible subsets (Un ) such that |Un is a nonsingular isomorphism between Un and Vn ⊂ Y . We claim that Vn ∩ Vm has measure 0 for every n = m. If this is not the case, take W ⊂ Un and W ⊂ Um nonnegligible and a nonsingular isomorphism ρ : W → W such that (ρ(x)) = (x) for x ∈ W . Since is an SOE, ρ(x) ∈ (G/N1 ) · x for almost all x ∈ W . Hence, making W smaller but still nonnegligible, we may assume that ρ(x) = g · x for all x ∈ W and some g ∈ G/N1 . Since W ∩ W has measure 0 and the action of G/N1 on X /N1 is essentially free, we get g = e. But also, δ(g) · (x) = (x) for almost all x ∈ W . This is a contradiction with the injectivity of δ and the essential freeness of Y . This proves the claim and we have found that is a nonsingular isomorphism between X /N1 and a nonnegligible subset Y1 ⊂ Y . Set 1 = δ(G/N1 ). It remains to prove that Y is induced from 1 Y1 . So, let h ∈ and assume that h · Y1 ∩ Y1 is nonnegligible. We have to prove that h ∈ 1 . By our assumption, take W , W ⊂ X /N1 nonnegligible and a nonsingular isomorphism ρ : W → W such that h · (x) = (ρ(x)) for all x ∈ W . Since

cocycle and orbit superrigidity for lattices in sl(n, R) / 443

was an SOE, we can make W smaller but still nonnegligible and assume that ρ(x) = g · x for all x ∈ W and some g ∈ G/N1 . But then, h · (x) = δ(g) · (x) for almost all x ∈ W . Since Y is essentially free, it follows that h = δ(g) ∈ 1 . Proof of Theorem 5.8. We prove only the case −1 ∈ H; the case −1 ∈ H being analogous. By Theorem 1.3, the action of G := ( × H)/{±(1, 1)} on Mn,k (R) is Uﬁn -cocycle superrigid. By Lemma 5.10, we have only to prove that the following subgroups of × H are the only open normal subgroups containing ( − 1, −1) and acting properly on Mn,k (R): • {±1} × N, where N H is an open normal subgroup with −1 ∈ N, and • ({1} × N) ∪ ({−1} × −N), where N H is an open normal subgroup with

−1 ∈ N. × H) be a closed normal subgroup acting properly on Mn,k (R). It So, let N ( is sufﬁcient to prove that N ⊂ {±1} × H. Suppose that N ⊂ {±1} × H and take (g, h) ∈ N with g = ±1. Take k ∈ such that the commutator t := kgk−1 g −1 = ±1. It follows that (t, 1) ∈ N. By Margulis’s normal subgroup theorem [15], we have 0 × {1} ⊂ N for some ﬁnite index subgroup 0 < . By Lemma 5.6, 0 acts ergodically on Mn,k (R), contradicting the properness of N Mn,k (R). Proof of Theorem 1.5. Statement (1) follows immediately from Theorem 5.8. We now prove statement (2). We claim that the following are the only normal subgroups N of SL (n, Z) Zn that act properly on Rn : • N = {e}, • N = λZn for some λ ∈ N \ {0}, and • (only when n is even) N = {±1} λZn for λ ∈ {1, 2}.

Statement (2) of Theorem 1.5 then follows from Lemma 5.10 and Theorem 1.3, where the Uﬁn -cocycle superrigidity of the afﬁne action SL (n, Z) Zn Rn was established. So, let N SL (n, Z) Zn be a normal subgroup acting properly on Rn . Suppose ﬁrst that N ⊂ Zn . Taking the commutator of (g, x) ∈ N with g = 1 and an arbitrary (1, y), y ∈ Zn , it follows that H := N ∩ Zn = {0}. Hence, H is a nonzero, globally SL (n, Z)-invariant subgroup of Zn . So, H = λZn for some λ ∈ N \ {0}. If N ⊂ {±1} Zn , it would follow that N has ﬁnite index in SL (n, Z) Zn , contradicting the properness of N Rn . So, we have shown that in all cases N ⊂ {±1} Zn . It is now straightforward to deduce the above list of possibilities for N.

444 / s. popa and s. vaes

Proof of Theorem 1.6. This theorem is a special case of Theorem 5.8.

5.3. Describing All 1-Cocycles of Quotient Actions

Finally, cocycle superrigidity of G (X , μ) allows us to describe all 1-cocycles for G/N X /N when N G is a closed normal subgroup of G acting essentially freely and properly on X . We start with the following proposition, closely related to [21, lemma 5.3], and illustrate it with two examples. σ

proposition 5.11. Let G be an l.c.s.c. group and G (X , μ) a nonsingular action. Let N G be a closed, normal subgroup such that the restriction σ |N is essentially free and proper. Assume that σ is Uﬁn -cocycle superrigid. Choose a measurable map π : X → N satisfying π(g · x) = gπ(x) for almost all (g, x) ∈ N × X . Denote by g → g and x → x the quotient maps G → G/N, respectively, X → X /N. Then, ω:

G X × → G : ω(g, x) = π (g · x)−1 gπ (x) N N

is a well-deﬁned 1-cocycle. Every 1-cocycle for the action G/N X /N with values in a Polish group of ﬁnite type G is cohomologous to δ ◦ ω for a continuous group morphism δ : G → G . If the diagonal action G X × X is ergodic, δ is uniquely determined up to conjugacy by an element of g ∈ G . Proof. Let : G/N × X /N → G be a 1-cocycle with values in the Polish group of ﬁnite type G . From cocycle superrigidity of σ , let ϕ : X → G be a measurable map and δ : G → G a continuous group morphism such that (g, x) = ϕ(g · x)−1 δ(g)ϕ(x) almost everywhere. Replacing g by hg, h ∈ N, it follows that ϕ(hg · x)−1 δ(h) = ϕ(g · x)−1 and hence, ϕ(h · x) = δ(h)ϕ(x) almost everywhere. So, we can deﬁne (x) = δ(π (x))−1 ϕ(x). By construction, makes cohomologous to δ ◦ ω. The uniqueness of δ follows directly from Lemma 5.4.

example 5.12. Let ⊂ SL (n, Z) be a ﬁnite index subgroup and consider the action Tn . • Choosing a measurable map p : Rn → Zn such that p(x + y) = x + p(y)

for all x ∈ Zn and almost all y ∈ Rn , the formula ω : × Rn /Zn → Zn : ω(g, x) = p(g · x)−1 gp(x) deﬁnes a 1-cocycle for Tn with values in Zn .

cocycle and orbit superrigidity for lattices in sl(n, R) / 445 • Every 1-cocycle with values in a Polish group of ﬁnite type G is coho-

mologous with δ ◦ ω for a group morphism δ : Zn → G , uniquely determined up to conjugacy by an element in G . Note that by Zimmer’s cocycle superrigidity theorem [26, theorem 5.2.5], any Zariski-dense 1-cocycle for any ergodic p.m.p. action SL (n, Z) (X , μ), n ≥ 3, taking values in a connected simple real algebraic noncompact centerfree group, is cohomologous to a group morphism. For the speciﬁc action SL (n, Z) Rn /Zn , n ≥ 5, the previous example provides an explicit description of all 1-cocycles of SL (n, Z) Rn /Zn with values in an arbitrary Polish group of ﬁnite type.

example 5.13. Let X be the real ﬂag manifold of signature (d1 , . . . , dl , n), ⊂ PSL (n, R) a lattice and n ≥ 4dl + 1. We obtain as follows all 1-cocycles for the action X with values in a Polish group of ﬁnite type. • Identify X = Mn,k (R)/H and choose a measurable map p : Mn,k (R) →

H satisfying p(Ag) = p(A)g for almost all A ∈ Mn,k (R), g ∈ H. Let

= {±1} · be the double cover of in GL (n, R) and deﬁne G := ( × H)/{±(1, 1)}. The formula ω : × X → G : ω(g, x) = (g, p(gx)p(x)−1 )

mod {±(1, 1)}

deﬁnes a 1-cocycle with values in G. Here → : g → g and Mn,k (R) → X : x → x are the quotient maps. • Every 1-cocycle with values in a Polish group of ﬁnite type G is cohomologous with δ ◦ ω for a continuous group morphism δ : G → G , uniquely determined up to conjugacy by an element in G .

6. Classification up to Orbit Equivalence Combining the results of [9] with Theorem 5.8, we classify up to stable orbit equivalence, the linear lattice actions Rn , as well as the natural lattice actions on ﬂag manifolds. At the same time, we compute the outer automorphism group of the associated orbit equivalence relations. We start with the following elementary lemma.

lemma 6.1. Let (X , μ) be a nonsingular action of the countable group and assume that the diagonal action X × X is ergodic. Then, X is not induced, that is, if X is induced from 1 X1 , then 1 = and μ(X \ X1 ) = 0.

446 / s. popa and s. vaes

Proof. Assume that X is induced from 1 X1 . So, we ﬁnd a quotient map π : X → / 1 satisfying π (g · x) = gπ (x) almost everywhere and X1 = π −1 (e1 ). Hence, the subset {(x, y) ∈ X × X | π (x) = π(y)} is nonnegligible and -invariant. By the ergodicity of X × X , it follows that π (x) = π (y) for almost all (x, y) ∈ X × X . This means that 1 = and μ(X \ X1 ) = 0. If R is a II1 equivalence relation on (X , μ), we denote by [R] the full group of R consisting of nonsingular automorphisms : X → X satisfying (x, (x)) ∈ R for almost every x ∈ X . Then, [R] is a normal subgroup of the automorphism group Aut (R) of R. The quotient group is denoted by Out (R) and called the outer automorphism group of R. The full pseudogroup of R is denoted by [[R]] and consists of nonsingular partial isomorphisms φ : X0 ⊂ X → X1 ⊂ X satisfying (x, φ(x)) ∈ R for almost every x ∈ X0 . We denote X0 = D(φ) and X1 = R(φ).

theorem 6.2. Let n ≥ 5 and ⊂ SL (n, R) a lattice. Let n ≥ 2 and ⊂ SL (n , R) a lattice. If the nonsingular isomorphism : X1 ⊂ Rn → X1 ⊂ Rn is an SOE between Rn and Rn , then • n = n and there exists A ∈ GL (n, R) such that = AA−1 , and • there exists φ ∈ [[R( Rn )]] with R(φ) = X such that (x) = φ(A(x)) 1

for almost every x ∈ X1 .

In particular, Out (R( Rn )) = NGL (n,R) ()/ .

Proof. Let : X1 ⊂ Rn → X1 ⊂ Rn be an SOE between Rn and Rn . Since for n = 2 the equivalence relation R( Rn ) is hyperﬁnite, we have n ≥ 3 and hence the diagonal action Rn × Rn is ergodic. By Lemma 6.1, Rn is not an induced action. By Theorem 1.5, the action Rn is conjugate with either Rn or, in case n is even, /{±1} Rn /{±1}. By Mostow rigidity, the latter is impossible since ∼ = /{±1}. In the former case, we already conclude n = n and we have found a nonsingular isomorphism : Rn → Rn and a group isomorphism δ : → satisfying • (g · x) = δ(g) · (x), and • (x) ∈ · (x),

for all g ∈ and almost all x ∈ Rn . Denoting by B the transpose of the matrix B, by Mostow rigidity, we ﬁnd A ∈ GL (n, R) such that (a) δ(g) = AgA−1 for all g ∈ , or (b) δ(g) = A(g )−1 A−1 for all g ∈ .

cocycle and orbit superrigidity for lattices in sl(n, R) / 447

Deﬁne the subgroup H ⊂ SL (n, R) consisting of matrices g with ge1 = e1 and identify Rn = SL (n, R)/H. In case (b), we would get a conjugacy between the -actions on SL (n, R)/H and SL (n, R)/H , which is ruled out by [9, theorem D]. In case (a), multiplying A by a nonzero scalar if necessary, [9, theorem D] implies that (x) = Ax for almost all x ∈ Rn . Deﬁning the partial isomorphism φ := ◦ A−1 with R(φ) = X1 , it follows that φ ∈ [[R( Rn )]] and that (x) = φ(A(x)) for almost every x ∈ X1 . Let X be the real ﬂag manifold with signature d := (d1 , . . . , dl , n). Denote d := (n − dl , n − dl−1 , . . . , n − d1 , n) . If X is the real ﬂag manifold with signature d , there is a natural diffeomorphism X → X : x → x satisfying g · x = (g )−1 · x for all x ∈ X , g ∈ SL (n, R).

theorem 6.3. Let ⊂ PSL (n, R) be a lattice and X the real ﬂag manifold with signature d := (d1 , . . . , dl , n). Assume that n ≥ 4dl + 1. Let ⊂ PSL (n , R) be a lattice and X the real ﬂag manifold with signature d := (d1 , . . . , dl , n ). If the nonsingular isomorphism : X1 ⊂ X → X1 ⊂ X is an SOE between X and X , then n = n and there exists A ∈ PGL (n, R), φ ∈ [[R( X )]] with R(φ) = X1 such that either • d = d, = AA−1 , (x) = φ(A(x)) for almost every x ∈ X1 , or • d = d , = A A−1 , (x) = φ(A(x)) for almost every x ∈ X1 .

In particular, Out (R( X )) = NPGL (n,R) ()/ . Proof. Let : X1 ⊂ X → X1 ⊂ X be an SOE between X and X . We ﬁrst prove that the diagonal action X × X is ergodic. If Y denotes the ﬂag manifold of signature (1, 2, . . . , n ), the action X is a quotient of the action Y . So, it sufﬁces to prove ergodicity of Y × Y . Denoting by D ⊂ SL (n , R) the subgroup of diagonal matrices, Y × Y can be identiﬁed with SL (n , R)/D, which follows ergodic because D SL (n , R)/ is ergodic by Moore’s ergodicity theorem. So, by Lemma 6.1, X is not an induced action. Since has trivial center, Theorem 1.6 yields a group isomorphism δ : → and a nonsingular isomorphism : X → X satisfying (g · x) = δ(g) · (x) and (x) ∈ · (x) almost everywhere. By Mostow rigidity, n = n and there exists A ∈ PGL (n, R) such that either δ(g) = AgA−1 or δ(g) = A(g )−1 A−1 for all g ∈ .

448 / s. popa and s. vaes

Set k1 = d1 and ki = di − di−1 for 2 ≤ i ≤ l. Put k = dl and deﬁne the closed subgroups H and H1 of GL (k, R) by ⎫ ⎧⎛ ⎞ A11 ∗ ··· ∗ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎟ ⎪ ⎪ ⎬ ⎨⎜ 0 A · · · ∗ ⎟ ⎜ 22 ⎟ ⎜ Aii ∈ GL (ki , R) and det (Aii ) = 0 , H := ⎜ . . . . ⎟ .. .. .. ⎠ ⎪ ⎪ ⎪ ⎪ ⎝ .. ⎪ ⎪ ⎪ ⎪ ⎭ ⎩ 0 0 · · · All ⎧⎛ A ⎪ ⎪ ⎪⎜ 11 ⎪ ⎨⎜ 0 H1 := ⎜ ⎜ .. ⎪ ⎪ ⎝ . ⎪ ⎪ ⎩ 0

∗ A22 .. . 0

⎫ ⎞ ∗ ⎪ ⎪ ⎪ ⎟ ⎪ ⎬ ∗⎟ ⎟ .. ⎟ Aii ∈ GL (ki , R) and det (Aii ) = ±1⎪ . ⎪ . ⎠ ⎪ ⎪ ⎭ · · · All

··· ··· .. .

We identify H/H1 = Rl+ . From now on, we write X as Mn,k (R)/H. Deﬁne analogously the subgroups H , H1 of GL (k , R) and write X as Mn,k (R)/H . Whenever N H is a closed normal subgroup containing −1, the quotient morphism H → H/N gives rise, as in Example 5.13, to a 1-cocycle ωN for the action X with values in H/N such that the action Mn,k (R)/N can be identiﬁed with the action X × H/N given by g · (x, h) = (g · x, ωN (g, x)h) . Since Mn,k (R)/N is ergodic, the 1-cocycle ωN cannot be cohomologous to a 1-cocycle taking values in a proper closed subgroup of H/N. We similarly deﬁne the 1-cocycles ωN for X . Since δ, conjugate the actions X and X , the map μ(g, x) = ωH1 (δ(g), (x)) deﬁnes a 1-cocycle for X , with values in Rl+ and with the property of not being cohomologous to a 1-cocycle taking values in a proper closed subgroup of Rl+ . We now apply Example 5.13, describing all Uﬁn -valued 1-cocycles for X , and note that Rl+ belongs to Uﬁn . Every group mor phism → Rl+ is trivial and every continuous group morphism H → Rl+ is trivial on H1 . So, we ﬁnd a continuous group morphism ρ : Rl+ → Rl+ such that μ is cohomologous to ρ ◦ ωH1 . Since μ cannot be cohomologous to a 1-cocycle taking values in a proper closed subgroup of Rl+ , it follows that ρ is onto. Altogether, we ﬁnd a closed normal subgroup N H, containing H1 , and a continuous isomorphism ρ : H/N → H /H1 such that μ is cohomologous to ρ ◦ ωN . It follows that there exists a nonsingular isomorphism 1 : Mn,k (R)/N → Mn ,k (R)/H1 satisfying 1 (g · x) = δ(g) · 1 (x) and 1 (x)H = (xH) almost everywhere.

cocycle and orbit superrigidity for lattices in sl(n, R) / 449

Since the action of on Mn ,k (R)/H1 is inﬁnite measure preserving, Lemma 5.7 implies that N = H1 . We saw already that either δ(g) = AgA−1 or δ(g) = A(g )−1 A−1 for all g ∈ . In the former case, [9, theorem D] implies that d = d and that there exists B ∈ H such that 1 (x) = AxB for almost every x ∈ Mn,k (R)/H1 . It follows that (x) = A(x) for almost every x ∈ X . In the latter case, we prove analogously that d = d and (x) = A(x).

7. Implementation by Group Actions In [5, p. 292], the question is raised whether every II1 equivalence relation can be implemented by an essentially free action of a countable group. This question has been settled in the negative in [7, theorem D]. In Proposition 7.1 below, we give examples of II1 equivalence relations R on (X , μ) with the following much stronger property: whenever (Y , η) is a free, nonsingular action and : X → Y is a measurable map satisfying (x) ∈ · (y) for almost all (x, y) ∈ R, then there exists y0 ∈ Y such that (x) ∈ · y0 for almost all x ∈ X . Among other examples, [7, theorem D] proves that the restriction of the orbit equivalence relation SL (n, Z) Rn /Zn to a subset of irrational measure, provides a II1 equivalence relation that cannot be implemented by a free action of a group. By [19, theorem 0.3], if [0, 1] is the Bernoulli action of a property (T) group without ﬁnite normal subgroups, the restriction of its orbit equivalence relation to any subset of measure strictly between 0 and 1 is unimplementable by a free action. But [7, theorem D] also provides examples of II1 equivalence relations R such that none of the ampliﬁcations Rt , t > 0, can be implemented by a free action. These equivalence relations are constructed using the following method. Suppose that G is an l.c.s.c. unimodular group and G (X , μ) an essentially free, properly ergodic p.m.p. action. There exists a Borel set Y ⊂ X , a probability measure η on Y , and a neighborhood U of e in G that we equip with a multiple of the Haar measure, such that U × Y → X : (g, y) → g · y provides a measure preserving isomorphism of U × Y onto a nonnegligible subset of X . The restriction of the orbit equivalence relation of G (X , μ) to Y is a II1 equivalence relation on (Y , η). One calls Y ⊂ X a measurable cross section for G (X , μ). A different choice of measurable cross section yields a stably isomorphic II1 equivalence relation. By Theorem 4.1, the restriction of the orbit equivalence relation of SL (n, Z) Rn to a subset of ﬁnite measure, provides other examples of II1 equivalence relations R such that none of the ﬁnite ampliﬁcations can

450 / s. popa and s. vaes

be implemented by a free action. In fact, one can show that R arises as the measurable cross section for the action of SL (n − 1, R) Rn−1 on SL (n, R)/ SL (n, Z). Nevertheless, this example is not covered by [7, theorem D], since SL (n − 1, R) Rn−1 is not semisimple.

proposition 7.1. Let G be an l.c.s.c. connected, unimodular group with normal closed subgroup G0 having the relative property (T). Let HR be a real Hilbert space and π : G → O (HR ) an orthogonal representation. Assume that π is injective and that the restriction of π to G0 is weakly mixing (i.e., has no ﬁnite-dimensional invariant subspaces). Denote by G (X , μ) the associated Gaussian action (see e.g., [8, section 2.7]). Choose a measurable cross-section X1 ⊂ X and denote by R the associated II1 equivalence relation on X1 . If (Y , η) is a free, nonsingular action and : X1 → Y is a measurable map satisfying (x) ∈ · (y) for almost all (x, y) ∈ R, then there exists y0 ∈ Y such that (x) ∈ · y0 for almost all x ∈ X1 . Proof. Choose a measurable map p : X → X1 such that p(x) ∈ G · x for almost all x ∈ X . Deﬁne the 1-cocycle ω : G × X → such that (p(g · x)) = ω(g, x) · (p(x)) almost everywhere. As observed in [8], theorem 0.1 in [19] applies to G (X , μ). Since G is connected, every group morphism from G to is trivial and we ﬁnd a measurable map ϕ : X → such that ω(g, x) = ϕ(g · x)−1 ϕ(x). So, the map x → ϕ(x) · (p(x)) is G-invariant and hence, essentially constant. We therefore ﬁnd y0 ∈ Y such that (p(x)) ∈ · y0 for almost all x ∈ X . This concludes the proof of the proposition.

References [1] [2] [3] [4] [5] [6] [7] [8]

C. Anantharaman-Delaroche, Cohomology of property T groupoids and applications. Ergod. Th. & Dynam. Sys. 25 (2005), 977–1013. P.L. Aubert, Deux actions de SL (2, Z). In Ergodic theory (Les Plans-sur-Bex, 1980), Monograph. Enseign. Math. 29, Univ. Genève, Geneva, 1981, pp. 39–46. B. Bekka, P. de la Harpe, and A. Valette, Kazhdan’s property (T ). New Mathematical Monographs 11. Cambridge University Press, Cambridge, 2008. A. Connes, Classiﬁcation of injective factors. Ann. Math. 104 (1976), 73–115. J. Feldman and C.C. Moore, Ergodic equivalence relations, cohomology, and von Neumann algebras I, II. Trans. Amer. Math. Soc. 234 (1977), 289–324, 325–359. A. Furman, Gromov’s measure equivalence and rigidity of higher rank lattices. Ann. of Math. 150 (1999), 1059–1081. A. Furman, Orbit equivalence rigidity. Ann. of Math. 150 (1999), 1083–1108. A. Furman, On Popa’s cocycle superrigidity theorem. Int. Math. Res. Not. IMRN 19 (2007), Art. ID rnm073, 46 pp.

cocycle and orbit superrigidity for lattices in sl(n, R) / 451 [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22]

[23]

[24] [25] [26]

A. Furman, Measurable rigidity of actions on inﬁnite measure homogeneous spaces, II. J. Amer. Math. Soc. 21 (2008), 479–512. D. Gaboriau, Coût des relations d’équivalence et des groupes. Invent. Math. 139 (2000), 41–98. D. Gaboriau, Invariants l 2 de relations d’équivalence et de groupes. Publ. Math. Inst. Hautes Études Sci. 95 (2002), 93–150. S.L. Gefter and V.Ya. Golodets, Fundamental groups for ergodic actions and actions with unit fundamental groups. Publ. Res. Inst. Math. Sci. 24 (1988), 821–847. A. Ioana, Cocycle superrigidity for proﬁnite actions of property (T) groups (2008), arXiv:0805.2998. To appear in Duke Math. J. A.S. Kechris, Classical descriptive set theory. Graduate Texts in Mathematics 156. Springer-Verlag, New York, 1995. G.A. Margulis, Discrete subgroups of semisimple Lie groups. Ergebnisse der Mathematik und ihrer Grenzgebiete (3) 17. Springer-Verlag, Berlin, 1991. C.C. Moore, Group extensions and cohomology for locally compact groups. III. Trans. Amer. Math. Soc. 221 (1976), 1–33. M. Pichot, Sur la théorie spectrale des relations d’équivalence mesurées. Journal of the Inst. of Math. Jussieu 6 (2007), 453–500. S. Popa, On the superrigidity of malleable actions with spectral gap. J. Amer. Math. Soc. 21 (2008), 981–1000. S. Popa, Cocycle and orbit equivalence superrigidity for malleable actions of w-rigid groups. Invent. Math. 170 (2007), 243–295. S. Popa, Correspondences, INCREST preprint No 56, 1986 (unpublished). S. Popa and S. Vaes, Strong rigidity of generalized Bernoulli actions and computations of their symmetry groups. Adv. Math. 217 (2008), 833–872. S. Popa and S. Vaes, On the fundamental group of II1 factors and equivalence relations arising from group actions. To appear in Quanta of Maths, Proceedings of the Conference in honor of Alain Connes, Clay Math. Inst., 2010. R. Steinberg, Some consequences of the elementary relations in SLn . In Finite groups— coming of age (Montreal, 1982). Contemp. Math. 45, American Mathematical Society, Providence, RI, 1985, pp. 335–350. S. Vaes, Rigidity results for Bernoulli actions and their von Neumann algebras (after Sorin Popa). Séminaire Bourbaki, exp. no. 961. Astérisque 311 (2007), 237–294. R.J. Zimmer, On the cohomology of ergodic actions of semisimple Lie groups and discrete subgroups. Amer. J. Math. 103 (1981), 937–951. R.J. Zimmer, Ergodic theory and semisimple groups. Birkhäuser, Boston, 1984.

PART 3

Geometric Group Theory

14

emmanuel breuillard

HEIGHTS ON SL2 AND FREE SUBGROUPS dedictated to bob zimmer on the occasion of his

60 th

birthday

Abstract We discuss and survey the strong uniform Tits Alternative [8] and give a complete proof of it in the special case of GL2 (C). The main arithmetic ingredient, the height gap theorem [7], which can be seen as a nonabelian analogue to the Lehmer conjecture, is also given a complete treatment in that case. We then give several applications involving expansion properties of SLd (Z/pZ), a uniform l2 spectral gap, and diophantine properties of subgroups of GLd (C).

Contents

1. Introduction 2. Spectral Radius Lemma for Several Matrices 3. Height, Arithmetic Spectral Radius and Minimal Height 4. Height Gap Theorem 5. Simultaneous Quasisymmetrization over Q 6. Ping-pong 7. Proof of Theorems 1.1 and 1.3 for SL2 (C) 8. Uniform Spectral Gap in 2 and Cogrowth of Subgroups 9. Large Girth References

455 464 468 470 476 478 480 484 486 492

1. Introduction 1.1. Statement of the Main Results

Recall that the Tits alternative [31] asserts that any ﬁnitely generated subgroup of GLd (K ), where K is some ﬁeld, contains a nonabelian free subgroup on two generators unless it is amenable, or equivalently in this case, unless it contains a solvable subgroup of ﬁnite index (i.e., is virtually solvable). In [7] and [8], we showed the following uniform version of Tits’s theorem : The author acknowledges support from the IAS Princeton through NSF Grant DMS0635607.

455

456 / e. breuillard

theorem 1.1. (strong uniform tits alternative [8]) For every d ∈ N there is N = N(d) ∈ N such that if K is any ﬁeld and F a ﬁnite symmetric subset of GLd (K ) containing 1, either F N contains two elements that freely generate a nonabelian free group, or the group generated by F is virtually solvable (i.e., contains a ﬁnite index solvable subgroup). We denote by F n = F · ... · F the product set of n copies of F. The purpose of this text is to discuss some consequences of this result and to give a selfcontained proof of it in the special case of SL2 (hence equivalently GL2 ) and K any ﬁeld of characteristic 0 (or, equivalently, for K = C, since every ﬁnitely generated ﬁeld of characteristic 0 embeds in C). This case is already representative of the general case as it captures the main difﬁculty, namely, treat all number ﬁelds in a uniform way. The proof given in [7] and [8] is an elaboration of the proof for SL2 that we are about to give. Theorem 1.1 improves earlier reﬁnements of Tits’s theorem due to EskinMozes-Oh (see [13]) and to T. Gelander and the author (see [9]). These two papers were concerned with the S-arithmetic version of Theorem 1.1, namely, they proved uniformity of N (the “freeness radius” of F) for sets F with coefﬁcients inside a ﬁxed ﬁnitely generated ring. While in [13] the main concern was to prove uniform exponential growth by constructing generators of a free semigroup in F N , in [9] it was shown that the result of [13] could be pushed to get generators of a free group in F N , where N was depending only on d and on the ring generated by the matrix coefﬁcients of the elements of F. The main contribution of this text and of [7] and [8] is to remove the dependence on the ring of coefﬁcients. As in Tits’s proof or in [9], the proof of Theorem 1.1 can be divided into an arithmetic step, on the one hand, and a geometric step on the other. While in [9] (as well as in Tits’s original theorem) the arithmetic step was the easier one and most of the work lied in showing that a certain geometric conﬁguration (the so-called “ping-pong”) did arise, roles are reversed in the proof of [8] in Theorem 1.1 and the arithmetic step is the harder step, while the geometric step routinely follows Tits’s original proof after a careful check that all estimates are indeed uniform over all local ﬁelds. For SL2 , however, none of the usual difﬁculties of higher rank arise and as will become clear below, this geometric step is even more transparent (in that case or other rank-1 situations this geometric step can also be performed differently by acting directly on the hyperbolic space/tree as has been pointed out by T. Gelander). The proof of Theorem 1.1 is effective in the sense that the constant N can, in principle, be made explicit. Examples due to Grigorchuk and de la Harpe [18]

heights on

SL2

and free subgroups / 457

imply that N(d) must tend to inﬁnity with d. They exhibited a sequence (n )n≥0 (n ≤ GL(n, Z)) of 4-generated subgroups whose rate of exponential growth exponent decays to 1. These groups arise by chopping the usual presentation of the Grigorchuk group after ﬁnitely many relators (see [18]). The arithmetic step in Theorem 1.1 relies on the following result, proved by the author in [7], and for which we will give below a self-contained proof in the special case of SL2 . Before stating it, we need some preliminary deﬁnitions. Let Q be an algebraic closure of Q. In [7] we introduced the arithmetic spectral radius (or normalized height) of F, deﬁned as 1 h(F n ), h(F) = lim n→+∞ n where h is the (absolute) height deﬁned for F a ﬁnite subset of Md (Q) by 1 nv log+ ||F||v h(F) = [K : Q] v∈VK

where log+ = max{0, log}, K is the number ﬁeld generated by the matrix coefﬁcients of F, VK is the set of all places of K , and ||F||v = max{||f ||v , f ∈ F} is the maximal operator norm of f ∈ F, where ||f ||v = maxx=0 ||f (x)||v /||x||v for the standard norm ||x||v induced on Kvd by the standard absolute value | · |v on the completion Kv of K associated to v ∈ VK . We have also set nv = [Kv , Qv ], where Qv is the ﬁeld of p-adic numbers if v|p is ﬁnite and is R if v is inﬁnite. The normalization of the absolute value | · |v is such that |λ|nv v is the modulus of the multiplication by λ on Kv . The quantities h(F) and h(F) are well deﬁned, that is, they are independent of the chosen number ﬁeld. If d = 1 and F is just one element {x}, then h(F) = h(x) is the classical (absolute, logarithmic) Weil height of x (see [3]). Moreover, h(F) is invariant under conjugation by elements from GLd (Q). The main statement is

theorem 1.2. (height gap theorem [7]) There is a positive constant ε = ε(d) > 0 such that if F is a ﬁnite subset of GLd (Q) generating a nonvirtually solvable subgroup , then h(F) > ε. Moreover, if the Zariski closure of is semisimple, then h(F) ≤

inf

g∈GLd (Q)

h(gFg −1 ) ≤ C · h(F)

for some absolute constant C = C(d) > 0.

458 / e. breuillard

This result can be seen as a nonabelian version of the well-known Lehmer conjecture from number theory. This conjecture states that the Weil height h(x) of an algebraic number that is not a root of unity must be bounded below by degε0(x) , where ε0 is an absolute constant. If F consists of a single diagonal matrix diag(x, x −1 ), then h(F) = h(x). When d = 2, that is, for SL2 , the ﬁrst part of Theorem 1.2 has the following geometric interpretation: either there is a ﬁnite place v where F acts without global ﬁxed point on the corresponding Bruhat-Tits tree, or after applying some Galois automorphism of C, the set F, as it acts on the hyperbolic 3space H3 via SL2 (C), moves every point away from itself by a positive absolute constant ε. This is analogous to the Margulis lemma in hyperbolic geometry, according to which if F generates a discrete subgroup of SL2 (C) that is not virtually nilpotent, then every point of H3 is moved away from itself by some element of F by some ﬁxed constant (see [30]). The main purpose of Theorem 1.2 is to yield in F, or a bounded power of F, a nice hyperbolic element, that is, a semisimple matrix whose eigenvalue is of modulus at least 2, say, in some local completion. Finally, letime point out that the proof of Theorem 1.1 yields a free subgroup with the extra property that it is uniformly undistorted in the original subgroup, namely:

theorem 1.3. (uniformly quasi-isometrically embedded free) subgroup [8] There is a constant C = C(d) > 0 such that if K is any ﬁeld and is any non virutally solvable subgroup of GLd (K ) generated by a ﬁnite symmetric subset F (giving rise to a word metric d (·, ·) on ), there exists a free subgroup H of generated by two elements (giving rise to a word metric dH (·, ·)) such that for all h ∈ H 1 · d (1, h) ≤ dH (1, h) ≤ d (1, h). C · |F|C 1.2. Some Consequences for Uniform Growth, Spectral Gap, and Diophantine Properties

We will prove here these corollaries for GLd (C), d ≥ 2.

corollary 1.4. (strong uniform exponential growth) There is ε = ε(d) > 0 such that if F is a ﬁnite subset of GLd (C) containing 1 and generating a nonamenable subgroup, then for all n ≥ 1, |F n | ≥ (1 + ε)n . In particular, ρF = lim

n→+∞

1 log |F n | ≥ log (1 + ε) > 0. n

heights on

SL2

and free subgroups / 459

Let us remark that it may be the case that ρF is bounded away from 0 assuming only that F generates a nonvirtually nilpotent subgroup of GL2 (C). However, we observed in [6] that such an assertion, if true, would imply the Lehmer conjecture about the Mahler measure of algebraic numbers. We also observed there that although every linear solvable group of exponential growth contains a free semigroup, no analog of Theorem 1.1 (the UF property of [1]) holds for solvable groups: namely, one may ﬁnd ﬁnite symmetric sets Fn in GL2 (C) containing 1 and generating a solvable subgroup of exponential growth, such that no pair of elements in (Fn )n may generate a free semigroup. Von Neumann showed that groups containing a free subgroup are nonamenable, that is, have a spectral gap in 2 . The uniformity in Theorem 1.1 implies also a uniformity for the spectral gap (see [29] for this observation). More precisely:

corollary 1.5. (strong uniform spectral gap in 2 ) There is ε = ε(d) > 0 with the following property. If F is a ﬁnite subset of GLd (C) containing the identity and generating a nonamenable subgroup and if is a countable subgroup of GLd (C) containing F and f ∈ 2 (), then there is σ ∈ F such that f (σ −1 x) − f (x)2 ≥ ε · |f (x)|2 x∈

x∈

In particular, if F in GLd (C) is a ﬁnite subset containing the identity and generating a nonamenable subgroup, then for every ﬁnite subset A in GLd (C), we have |FA| ≥ (1 + ε)|A|. In [24] Lubotzky, Phillips, and Sarnak showed that for the compact Lie group G = SU(2), the spectral measure of the “Hecke operators” Tμ = −1 1 acting on L20 (G) is supported on [−mμ , mμ ] where mμ is 1≤i≤k gi + gi 2k the norm of Tμ viewed as an operator on 2 (), being the abstract group generated by the gi ’s. This spectral measure is by deﬁnition the limiting distribution of the eigenvalues of Tμ on the nth-dimensional representation of G. Corollary 1.5 implies that mμ is bounded away from 1 independently of μ as soon as k is ﬁxed and is a nonamenable subgroup of G. In other words, there is ε = ε(k) > 0 such that the proportion of eigenvalues of Tμ lying in [−1, −1 + ε] ∪ [1 − ε, 1] tends to 0 as n tends to inﬁnity. The analogous result for Cayley graphs of SL2 (Z/pZ) is also a direct consequence of Corollary 1.5, that is, the spectral measure of any limit of such Cayley graphs is supported on [−1 + ε, 1 − ε]. It is believed (spectral gap conjecture, see [24] and [28]) that 1

460 / e. breuillard

is never an accumulation point of eigenvalues of Tμ for any given μ (or at least almost any in SU(2)) whose support generates a nonamenable subgroup. Since by Kesten’s theorem mμ is also the exponential rate of decay of the return probability (see [21]), we also have

corollary 1.6. (strong uniform decay of return probability) There is ε = ε(k, d) > 0 with the property that for any nonamenable k-generated subgroup of GLd (C) we have P(Sn = 1) ≤ (1 − ε)n

for all n ≥ 1, where Sn is the simple random walk on . The next corollary gives an upper bound on the cogrowth of subgroups of GLd (C).

corollary 1.7. (co-growth gap) Given m ∈ N, there is n(m) > 0 such that for every n ≥ n(m) the following holds: F = {a1 , ..., am } ⊂ GLd (C) generates a non virtually solvable subgroup, if and only if the number of elements w in the free group Fm of word length n such that w(a1 , ..., am ) = 1 is at most (2m − 1 − mεD )n . Here ε, D > 0 are constants depending on d only. This result can be paraphrased by saying that nonamenable subgroups of GLd (C) are very strongly nonamenable, that is, have few relations. This puts a purely group theoretical restriction on an abstract ﬁnitely generated group given in terms of generators and relations to admit an embedding in GLd (C). The uniformity in Theorem 1.1 allows us to reduce mod p and we obtain a statement giving a lower bound on the girth of subgroups of GLd in positive characteristic:

corollary 1.8. (large girth) Given k, d ≥ 2, there is N, M ∈ N and ε0 , C > 0 such that for every prime p and every ﬁeld K of characteristic p and any ﬁnite subset F with k elements generating a subgroup of GLd (K ) that contains no solvable subgroup of index at most M, then F N contains two elements a, b such that w(a, b) = 1 in GLd (K ) for any nontrivial word w in F2 of length at most f (p) = C · ( log p)ε0 . It was conjectured in [16] that the statement of Corollary 1.8 holds for generating subsets F of SL2 (Fp ) with ε0 = 1.

heights on

SL2

and free subgroups / 461

Theorem 1.3 on the uniform QI-embedding of the free subgroup yields a uniform bound for the distortion of the subgroup of large girth. This in turn gives uniform expansion for subsets of say GL2 (Fp ) lying in a ball of radius ≤ ( log p)ε0 . Namely:

corollary 1.9. (uniform expansion for small sets) Given k, d≥ 2, there is N, M ∈ N and ε0 , C, α, β > 0 such that for every prime p and every ﬁeld K of characteristic p and any ﬁnite symmetric subset F with k elements, containing 1 and generating a subgroup of GLd (K ) that contains no solvable subgroup ε of index at most M, then for any subset A ⊂ F C·( log p) 0 there is s ∈ F such that n ε0 |sA A| ≥ α|A|. In particular μ∗n F (e) ≤ (1 − β) for all n ≤ C · ( log p) , where μF is the uniform probability measure on F. If one could get ε0 = 1 in the above corollary, then applying the argument of Bourgain and Gamburd [4] would give a proof that the family of all Cayley graphs of SL2 (Z/pZ) for varying p but with a ﬁxed number of generators forms an expander family. See Lubotzky’s book [23] for background material on expanders. It was also proved in [16] that a random d-regular Cayley graph of GL2 (Fp ) has girth at least (1 − o(1)) logd−1 (p). Here we obtain ε0 = 2−10 for GL2 , which is quite far. In the same vein, one obtains the following two weak forms of "nonLiouvilleness" for subgroups of GLd (C). Let d be some Riemannian distance on GL2 (C).

corollary 1.10. (short words are not simultaneously very liouville) Given d, there is N0 ∈ N and ε1 > 0 with the following property. For every ﬁnite set F ⊂ GLd (C) generating a nonvirtually solvable subgroup, there is δ0 (F) > 0 such that for every δ ∈ (0, δ0 ) there are two short words a, b ∈ F N O such that d(w(a, b), 1) ≥ δ for every reduced word w in the free group F2 with length (w) at most ( log δ −1 )ε1 . In [19] Kaloshin and Rodnianski proved that for G = SU(2) ≤ SL2 (C) almost every pair (a, b) ∈ G × G satisﬁes d(w(a, b), 1) ≥ exp ( − C(a, b) · (w)2 ) for all w ∈ F2 \{e} and some constant C(a, b) > 0. Besides, it is easy to see that if a, b ∈ GL2 (Q), then the pair (a, b) satisﬁes the stronger diophantine condition d(w(a, b), 1) ≥ exp ( − C(a, b) · (w)). It is conjectured in [28] and [16] that this stronger condition also holds for almost every pair (a, b) ∈ SU(2) with respect to Haar measure.

462 / e. breuillard

Our result also allows us to estimate the number of words of length ≤ n that fall in a shrinking neighborhood of 1 in GLd (C). More precisely,

corollary 1.11. (weak diophantine property) There are τ , ε1 , C > 0 with the following property. For every {a, b} ≤ GLd (C) that generates a nonvirtually solvable subgroup, there is δ0 (a, b) > 0 such that for every δ ∈ (0, δ0 ) and every n ≤ C( log δ −1 )ε1 , the proportion of elements w in the free group F2 of word length n such that d(w(a, b), 1) ≤ δ is at most exp ( − τ n). In [15], Gamburd, Jacobson, and Sarnak showed for G = SU(2) that if a pair (a, b) ∈ G satisﬁes the conclusion of Corollary 1.11 with ε1 = 1 and C > C0 (for some explicit C0 > 0), then (a, b) has a spectral gap on L2 (G). In [5], Bourgain and Gamburd showed that if a pair (a, b) ∈ G satisﬁes the above condition with ε1 = 1 and some C = C(a, b) > 0, perhaps small, then (a, b) has a spectral gap on L2 (G). This latter condition is automatically satisﬁed if (a, b) satisﬁes the stronger diophantine condition above, for instance, if (a, b) ∈ GL2 (Q). Hence these pairs have a spectral gap. It is unknown whether there are (topologically generating) pairs with no spectral gap.

remark 1.12. We emphasize here that all the constants in the above theorems and corollaries can be effectively computed. Only at one point in the proof do we use a compactness argument. This is in the proof of Lemma 2.1(b). However, this statement can be given an effective proof valid in Md (C) (available upon request) although the price to pay is a much lengthier argument. Comment on the proof of Corollaries 1.8 to 1.11: Observe (see Proposition 9.2) that the condition on a ﬁnite subset F = {A, B} of GLd (C) that it generates an amenable (or equivalently virtually solvable) subgroup of GLd (C) is an algebraic one, as is the condition that all short words in A and B satisfy a relation of length at most n. Thus the statement of Theorem 1.1 can be read as a union of countably many assertions of ﬁrst-order logic. According to the “compactness theorem” from model theory, since each assertion holds for C it must also hold for an arbitrary ﬁeld K of sufﬁciently large characteristic (depending on n). This readily gives this existence of some function f (p) going to +∞ with p in Corollary 1.8. To derive the bound ( log p)ε0 , as well as the bounds ( log δ −1 )ε0 in Corollaries 1.10 and 1.11, we use instead a standard version of the effective Nullstellensatz due to Masser and Wustholz (see [25]).

heights on

SL2

and free subgroups / 463

1.3. Outline of the Paper and of the Proof of Theorems 1.2 and 1.1 for SL2

In Section 3 we introduce our main objects, the height and normalized height of a ﬁnite set F of matrices and prove basic properties about them. One of the key properties, the comparison between h and e, relies crucially on Section 2, which is devoted to the proof of a key lemma, the spectral radius lemma. This lemma says in substance that unless F can be conjugated in a bounded part of SL2 , one will ﬁnd a short word with letters in F with a large eigenvalue. In Section 4 we prove the ﬁrst part of Theorem 1.2 (the height gap). The proof makes crucial use of equidistribution properties of algebraic numbers of small height and in particular a result of Zhang (Theorem 4.9) and Bilu’s equidistribution theorem for Galois orbits (Theorem 4.4 below). Using the Eskin-Mozes-Oh escape lemma (see Lemma 4.8 below) we ﬁrst reduce to the 2-generated case, when F = {Id, A, B}. Conjugating if necessary, A can be assumed to be diagonal. By making local estimates at each place, and with the help of Bilu’s theorem, we then show that if h(F) is small, then the heights of b11 , b22 , and b12 b21 are small. But as b11 b22 − b12 b21 = 1, Zhang’s theorem quickly yields to a contradiction if b12 b21 = 0. So b12 b21 = 0 and F is made of upper or lower triangular matrices, that is, F generates a solvable subgroup and we are done. Theorem 1.1 is of purely algebraic nature and we begin its proof by showing that its validity over Q implies its validity over C. In the case of SL2 , one needs to exhibit a place v where one can play ping-pong on the projective line P1 (Kv ) for the local ﬁeld Kv , as in Tits’s proof of his alternative. These “ping-pong players” will be the generators of the desired free subgroup. To achieve this, one needs to be able to conjugate F in SL2 (Kv ) in such a way that three conditions are satisﬁed. First, the norm ||F||v ought to be controlled (up to a ﬁxed power) by the maximal eigenvalue say |λ|v of an element, say A, lying in F (or F N for a bounded N). Second, |λ|v should be large enough, that is, bounded away from 1. And third, at least one element, say B, from F or F N , must send the eigenvectors of A far apart from one another with a distance controlled by some negative power of ||F||v . This criterion for ping-pong is explained in Section 6. In Section 7 we show that a place v with these properties does exist. This is done in two steps. First (Section 5), it is shown that the minimal height h(F) can be almost achieved (up to multiplicative and additive constants) by the ordinary height h(F) after possibly conjugating F inside SL2 (Q). This is the second half of Theorem 1.2: this step uses the estimates needed in the ﬁrst part of Theorem 1.2 (i.e. the proof of the height gap). In a second step (Section 7) the product formula on P1 (Q) is used to show that the distances between

464 / e. breuillard

eigenvectors of A and their images under B are controlled in terms of h(F), and hence h(F), only. This implies the existence of a place v satisfying the ﬁrst and third conditions. Theorem 1.2 ensures that v can be chosen to satisfy the second condition also. Sections 8 and 9 are devoted to the applications.

2. Spectral Radius Lemma for Several Matrices In this section we state and prove the crucial Lemma 2.1. It says that given a ﬁnite set of matrices with coefﬁcients in a local ﬁeld, one may always ﬁnd a short word with letters in that ﬁnite set whose maximal eigenvalue is as large as the minimal norm of the ﬁnite set. Together with Proposition 2.5 it can be interpreted as an analogue for several matrices of the classical spectral radius lemma relating the maximal eigenvalue and the rate of growth of the powers of a matrix. This lemma expresses in a condensed form an idea from a key proposition of the work of Eskin-Mozes-Oh where the concept of an almost algebra was used to essentially achieve the same goal. We emphasize here that getting an equality in part (a) of Lemma 2.1 as opposed to a mere inequality like in part (b) of the same lemma is absolutely crucial in the whole proof and in particular in Theorem 1.2. Let k be a local ﬁeld of characteristic 0. Let · k be the standard norm on 2 k , that is, the canonical Euclidean (respectively, Hermitian) norm if k = R (respectively, C) and the sup norm ((x, y)k = max{|x|k , |y|k }) if k is non Archimedean. We will also denote by · k the operator norm induced on M2 (k) by the standard norm · k on k2 . Let Q be a bounded subset of matrices in M2 (k). We set Q = sup g k k g∈Q

and call it the norm of Q . Let k be an algebraic closure of k. It is well known (see Lang’s Algebra [22]) that the absolute value on k extends to a unique absolute 2

value on k, hence the norm · k also extends in a natural way to k and to M2 (k). This allows us to deﬁne the minimal norm of a bounded subset Q of M2 (k) as Ek (Q ) =

inf x∈GL2 (k)

xQx −1 . k

We will also need to consider the maximal eigenvalue of Q , namely k (Q ) = max{|λ|k , λ ∈ spec(q), q ∈ Q }

heights on

SL2

and free subgroups / 465

where spec(q) denotes the set of eigenvalues (the spectrum) of q in k. Finally, let Rk (Q ) be the spectral radius of Q : 1

Rk (Q ) = lim ||Q n ||kn . n→+∞

These quantities are related to one another. The key property concerning them is given in the following assertion (which also holds in Md (k), k ≥ 2; see [7]).

lemma 2.1. (spectral radius lemma) Let Q be a bounded subset of M2 (k), a) if k is non-Archimedean, then k (Q 2 ) = Ek (Q )2 ; and b) if k is Archimedean, there is a constant c ∈ (0, 1) independent of Q , such that k (Q 2 ) ≥ c 2 · Ek (Q )2 . Proof. We make use of the following easy lemmas.

lemma 2.2. Let L be a ﬁeld and Q a subset of M2 (L) such that Q and Q 2 consist of nilpotent matrices. Then there is a basis (u, v) of L2 such that Qu = 0 and Qv ⊂ Lu. Proof. For any A, B ∈ Q , we have A2 = B2 = (AB)2 = 0. It follows, unless A or B are 0, that ker A = ImA and ker B = ImB. Also if AB = 0, we get ker B = ker (AB) = Im(AB) = ImA, while if AB = 0, then ImB = ker A. So at any case ker A = ImA = ker B = ImB. So we have proved that the kernels and images of nonzero elements of Q coincide and are equal to some line say Lu,. Pick v ∈ L2 \{Lu}, then (u, v) forms the desired basis.

lemma 2.3. Let k be a local ﬁeld with ring of integers Ok and uniformizer π. Let A = (aij ) ∈ M2 (Ok ) such that trace(A) and det (A) belong to (π 2 ) and a11 , a22 , a21 ∈ (π ), while a12 ∈ Ok× . Then a21 ∈ (π 2 ). Proof. We have a12 a21 = a11 a22 − det (A) ∈ (π 2 ) and a12 ∈ Ok× , hence a21 ∈ (π)2 . When k is a non-Archimedean local ﬁeld, if a set Id + Q in SL2 (k) and its square have only eigenvalues very close to 1, then it must ﬁx pointwise the 1-neighborhood of some point in the Bruhat-Tits tree of SL2 (k). This is essentially the content of the next lemma.

466 / e. breuillard

lemma 2.4. (small eigenvalues implies large fixed-point set) Let k be a local ﬁeld with ring of integers Ok and uniformizer π together with an absolute value | · |k , which is (uniquely) extended to an algebraic closure k of k. Let Q be a subset of M2 (Ok ) such that k (Q ) and k (Q 2 ) are both ≤ |π |3k . Then there is T ∈ GL2 (k) such that TQT −1 ⊂ π M2 (Ok ). Proof. We can write Q as the disjoint union Q1 ∪ πQ2 where Q1 does not intersect π M2 (Ok ). Let Q = Q1 ∪ Q2 . Then k (Q ) and k (Q 2 ) are both ≤ |π |k . Hence projecting to M2 (L), where L is the residue ﬁeld L = Ok /(π ), the matrices from Q and Q 2 become nilpotent. According to Lemma 2.2, one may ﬁnd a basis (u, v) of L2 such that Q u = 0 and Q v ⊂ Lu. According to Nakayama’s lemma, this basis is the projection of a basis (u, v) of Ok2 . Up to conjugating by a matrix in GL2 (Ok ), we may assume that (u, v) is the canonical basis of Ok2 . Therefore Q consists of matrices A = (aij ) ∈ M2 (Ok ) with a11 , a22 , a21 ∈ (π ). Moreover, matrices in Q1 satisfy a12 ∈ Ok× and hence by Lemma 2.3, a21 ∈ (π 2 ). But for the matrices coming from π Q 2 we also have a21 ∈ (π 2 ). So we have a21 ∈ (π 2 ) for all matrices in Q . Let T = diag(π , 1) ∈ GL2 (k). Then clearly TQT −1 ⊂ π M2 (Ok ). We go back to the proof of Lemma 2.1. We ﬁrst prove (b). By contradiction, if (Q 2 )

such a c did not exist, then we may ﬁnd a sequence of Qn such that E k(Q n)2 → 0. k

Q

n

We can change Qn into E (Qn n ) and thus obtain a sequence of compact sets in k M2 (k) such that Ek (Qn ) = 1 with k (Qn2 ) → 0 and k (Qn ) → 0. Passing to a limit, we obtain a compact subset Q of M2 (k) such that k (Q 2 ) = k (Q ) = 0 while Ek (Q ) = 1. By Lemma 2.2, we can thus ﬁnd a basis (u, v) where Qu = 0 and Qv ⊂ Lu. But then conjugating Q by a suitable diagonal matrix can shrink the norm of Q as much as we want, hence Ek (Q ) = 0. This is a contradiction. We now prove (a). Let π be a uniformizer for k. Let x = log Ek (Q ) where the log is taken in base |π |−1 . Suppose that k (Q 2 ) < Ek (Q )2 and let ε = k 1

x − 12 log k (Q 2 ) > 0. Then as k (Q ) ≤ k (Q 2 ) 2 , we have x − log k (Q ) ≥ ε > 0. Note that with our choice of normalization, log k (Q 2 ) ∈ 12 Z. Let n ∈ N √ √ be such that 2nε > 3. Let k0 be the extension k(2n π ) where 2n π is some 2nroot of π in k. Since x = log Ek (Q ) = log inf {||gQg −1 ||k , g ∈ GL2 (k)}, we may assume after possibly conjugating Q inside GL2 (k1 ), for some ﬁnite extension 1 k1 of k0 , that y := log ||Q ||k ≤ x + 2n and also that y = min{log ||gQg −1 ||k , g ∈ 1 GL2 (k1 )}. Let π1 be a uniformizer in k1 . Then log |π1 |−1 ≤ 2n and y = log |π1 |m k −m for some m ∈ Z. Let Qy = π1 Q ⊂ M2 (Ok1 ). We get log Ek (Qy ) = x − y ≤ 0 3 and log k (Qy ) ≤ 12 log k (Qy2 ) = x − y − ε ≤ −ε ≤ − 2n ≤ log |π1 |3k . We are thus in a position to apply Lemma 2.4, which implies that Qy , and hence

heights on

SL2

and free subgroups / 467

Q itself, can be further conjugated inside GL2 (k1 ) so as to strictly reduce its norm. But this contradicts the minimality of y.

proposition 2.5. Let Q be a bounded subset of M2 (k). We have 1

1

1

1

Rk (Q ) = lim Ek (Q n ) n = inf Ek (Q n ) n = lim k (Q 2n ) 2n = sup k (Q n ) n . n→+∞

n∈N

n→+∞

n∈N

Moreover, if k is non-Archimedean, Rk (Q ) = Ek (Q ), while if k is Archimedean, then c · Ek (Q ) ≤ Rk (Q ) ≤ Ek (Q ), where c is the constant from Lemma 2.1(b). Proof. We omit the proof: these identities follow either directly from the deﬁnitions or as a straightforward application of Lemma 2.1. Note that some periodicity phenomenon may arise if Id ∈ / Q , namely, it may be that k (Q 2n+1 ) = 1 for all n while k (Q 2n ) tends to inﬁnity (for instance, take for Q a set of symmetries around several points on a given geodesic in the 1 hyperbolic plane). However, if Id ∈ Q , then we do have limn→+∞ k (Q n ) n = Rk (Q ). Note also that if Q belongs to SL2 (k), then Ek (Q ) ≥ Rk (Q ) ≥ k (Q ) ≥ 1 and all three quantities remain unchanged if we add Id to Q . The following lemma explains what happens if these quantities are close or equal to 1.

lemma 2.6. (linear growth of displacement squared) Suppose k is Archimedean (i.e., k = R or C). Then we have for every n ∈ N and every bounded subset Q of SL2 (k) containing Id, √n 1 Ek (Q n ) ≥ Ek (Q ) 4 . Moreover, log Rk (Q ) ≥ c1 · log Ek (Q ) · min{1, log Ek (Q )} for some constant c1 > 0. In particular Ek (Q ) = 1 iff Rk (Q ) = 1. Proof. We use nonpositive curvature of hyperbolic space H3 . For x ∈ H3 set L(Q , x) = maxg∈Q d(gx, x) and L(Q ) = inf x L(Q , x). Fix ε > 0, set n := L(Q n ) = 2 log Ek (Q n ), and let rn be the inﬁmum over x ∈ H3 of the smallest radius of a closed ball containing Q n x. Note ﬁrst that rn ≤ n ≤ 2rn . We now prove (1). Fix ε > 0 and let x, y ∈ H3 be such that Q n+1 x is contained in a ball of radius rn+1 + ε around y. Let q ∈ Q be arbitrary. Since Q contains Id, we have Q n x ⊂ Q n+1 x, and qQ n x lies in the two balls of radius rn+1 + ε centered around qy and around y. By the CAT(0) inequality for the

468 / e. breuillard

median, the intersection of the two balls is contained in the ball B of radius t := (rn+1 + ε)2 − d(qy, y)2 /4 centered around the midpoint m between y and qy. Translating by q−1 , we get that Q n x lies in the ball of radius t centered at q−1 m. In particular rn ≤ t. This means d(qy, y)2 ≤ 4((rn+1 + ε)2 − rn2 ). 2 Since q ∈ Q and ε > 0 were arbitrary, we obtain 21 ≤ 4(rn+1 − rn2 ). Sum2 2 2 ming over n, we get n1 ≤ 4rn ≤ 4n , hence (1). But byLemma 2.1(b), 1

1

k (Q 2n ) ≥ c 2 Ek (Q n )2 , hence Rk (Q ) ≥ k (Q 2n ) 2n ≥ c n Ek (Q ) in n, we obtain the desired bound.

1 4n

. Optimizing

3. Height, Arithmetic Spectral Radius and Minimal Height For any rational prime p (or p = ∞) let us ﬁx an algebraic closure Qp of the ﬁeld of p-adic numbers Qp (if p = ∞, set Q∞ = R and Q∞ = C). We take the standard normalization of the absolute value on Qp (i.e., |p|p = 1p ). It admits a unique extension to Qp , which we denote by | · |p . Let Q be the ﬁeld of all algebraic numbers and K a number ﬁeld. Let VK be the set of equivalence classes of valuations on K . For v ∈ VK let Kv be the corresponding completion. For each v ∈ VK , Kv is a ﬁnite extension of Qp for some prime p. We normalize the absolute value on Kv to be the unique one that extends the standard absolute 1

value on Qp . Namely |x|v = |NKv |Qp (x)|pnv where nv = [Kv : Qp ]. Equivalently Kv has nv different embeddings in Qp and each of them gives rise to the same absolute value on Kv . We identify Kv , the algebraic closure of Kv , with Qp . Let Vf be the set of ﬁnite places and V∞ the set of inﬁnite places. Let F be a ﬁnite subset in M2 (K ). For v ∈ VK , in order not to surcharge notation, we will use the subscript v instead of Kv in the quantities Ev (F) = EKv (F), v (F) = Kv (F), and so on. Recall that if x ∈ K then its (absolute, logarithmic) Weil height is by deﬁnition (see, e.g., [3]) the following quantity: 1 h(x) = nv log+ |x|v . [K : Q] v∈VK

It is well deﬁned (i.e., independent of the choice of K x). Let us similarly deﬁne the height of a matrix f ∈ M2 (K ) by 1 nv log+ ||f ||v h(f ) = [K : Q] v∈VK

and the height of a ﬁnite set F of matrices in M2 (K ) by 1 nv log+ ||F||v , h(F) = [K : Q] v∈VK

heights on

SL2

and free subgroups / 469

where nv = [Kv : Qv ]. We also deﬁne the minimal height of F as 1 2 e(F) = nv log+ Ev (F) [K : Q] v∈VK

and the arithmetic spectral radius (or normalized height) of F as 1 nv log+ Rv (F). h(F) = [K : Q] v∈VK

For any height h, we also set h = h∞ + hf , where h∞ is the inﬁnite part of h (i.e., the part of the sum over the inﬁnite places of K ) and hf is the ﬁnite part of h (i.e., the part of the sum over the ﬁnite places of K ). Note that these heights are well deﬁned independently of the number ﬁeld K such that F ⊂ M2 (K ). The above terminology is justiﬁed by the following facts:

proposition 3.1. (basic properties of heights i) For any ﬁnite set F in M2 (Q), we have 1) h(F) = limn→+∞ n1 h(F n ) = inf n∈N n1 h(F n ), 2) ef (F) = hf (F) and e(F) + log c ≤ h(F) ≤ e(F) where c is the constant in Lemma 2.1 (b), and 3) h(F n ) = n · h(F) and h(F ∪ {Id}) = h(F). Proof. This follows easily from Lemma 2.1 and Proposition 2.5.

We also record the following simple observations:

proposition 3.2. (basic properties of heights ii) We have, for a ﬁnite set F in M2 (Q), 1) e(xFx −1 ) = e(F) if x ∈ GL2 (Q), 2) e(F n ) ≤ n · e(F), and 3) if λ is an eigenvalue of an element of F, then h(λ) ≤ h(F) ≤ e(F). Proof. This is clear.

We can also compare e(F) and h(F) when h(F) is small:

proposition 3.3. (basic properties of heights iii) Let c1 be the constant from Lemma 2.6, then

470 / e. breuillard

c1 h∞ (F) ≥ · e∞ (F) · min{1, e∞ (F)} 4 h(F) is for any ﬁnite subset F in SL2 (Q). In particular e(F) is small as soon as small. Proof. From Lemma 2.6, hv (F) ≥ c1 · ev (F) · min{1, ev (F)} for every v ∈ VK . We + may write e∞ (F) = αe (F) + (1 − α)e− (F) where e+ is the average of the ev greater than 1 and e− the average of the ev smaller than 1. Applying CauchySchwarz, we have h∞ (F) ≥ c1 · (αe+ + (1 − α)(e− )2 ). If αe+ (F) ≥ 12 e∞ (F), then h∞ (F) ≥ c21 e∞ (F), and otherwise (1 − α)e− ≥ e2∞ , hence h∞ (F) ≥ c1 (1 − α) (e− )2 ≥ c41 e2∞ . At any case h∞ (F) ≥ c41 · e∞ (F) · min{1, e∞ (F)}.

4. Height Gap Theorem In this section, we prove Theorem 1.2 from the introduction. First observe that according to Propositions 3.3 and 3.1(2), h(F) is small if and only if e(F) is small. So we may as well replace h(F) by e(F) in Theorem 1.2. We now assume that F = {Id, A, B}, with A semisimple. The general case follows from this as we will show in Lemma 4.7. Since e(F) is invariant under conjugation by any element in GL2 (Q), we may assume that A is diagonal, that is,

λ 0 a b 3 A= , B= . 0 λ−1 c d Let deg (λ) be the degree of λ as an algebraic number over Q. The main part of the argument consists in the following proposition:

proposition 4.1. (small normalized height implies small height of matrix coordinates) For every β > 0 there exists d0 ,η > 0 such that, if F = {A, B} are as in (3) and if e(F) ≤ η and deg (λ) ≥ d0 , then max{h(ad), h(bc)} ≤ β. In order to prove this statement, we are ﬁrst going to give local estimates at each place v, then use Bilu’s equidistribution theorem to show that when these estimates are put together, the error terms give only a negligible contribution to the height. Let K be the number ﬁeld generated by the coefﬁcients of A and B. Let v ∈ VK be a place of K . We set sv = log Ev (F) and δ = λ − λ−1 . We ﬁrst show the following local estimate:

heights on

SL2

and free subgroups / 471

lemma 4.2. (local estimates) For each v ∈ VK we have max{|a|v , |d|v , |bc|v } ≤ Cv e2sv max{1, |δ −1 |v }, where Cv is a constant equal to 1 if v is a ﬁnite place and equal to a number C∞ > 1 (independent of v) if v is inﬁnite. Moreover, there are absolute constants ε0 > 0 and C0 > 0 such that if v is inﬁnite and sv ≤ ε0 , then √ √ sv sv + 2 ). max{|ad|v , |bc|v } ≤ 1 + C0 ( sv + |δ|v |δ|v Proof. In order not to overburden notation in this proof we set sv to be some number arbitrarily close but strictly bigger than log Ev (F) and we can let it tend to log Ev (F) at the end. If v is inﬁnite, then Qv = C and SL2 (C) = KAN where K = SU2 (C), A is the subgroup of diagonal matrices with real positive entries, and N is the subgroup of unipotent complex upper-triangular matrices. As K leaves the norm invariant, there must exist a matrix P ∈ AN such that max{PAP −1 , PBP −1 } ≤ esv . Since P ∈ AN, we may write P =

t y with t > 0 and y ∈ C. Then we have, setting δ = λ − λ−1 , 0 t −1

λ tyδ −1 4 , PAP = 0 λ−1

a + cyt −1 bt 2 + dyt − ayt − cy 2 −1 . PBP = t −2 c −yct −1 + d If v is ﬁnite and Kv is the corresponding completion, with ring of integers Ov and uniformizer π, we have SL2 (Kv ) = Kv Av Nv where Kv = SL2 (Ov ), Av = {diag(π n , π −n ), n ∈ Z} and Nv is the subgroup of unipotent upper-triangular

matrices with coefﬁcients in Kv . Hence, we also get a P ∈ Av Nv satisfying (4) with y ∈ Kv and t = π n for some n ∈ Z. We ﬁrst assume that v is ﬁnite. Recall that the operator norm in SL2 (Kv ) is given by the maximum modulus of each matrix coefﬁcient. Hence we must have t −2 c v ≤ esv and tyδ v ≤ esv . It follows that cyt −1 v ≤ e2sv |δ −1 |v and hence |a|v ≤ max{esv , e2sv |δ −1 |v }. Similarly, |d|v ≤ max{esv , e2sv |δ −1 |v }. Hence |ad|v ≤ max{e2sv , e4sv |δ −1 |2v }. Moreover ad − bc = 1, hence |bc|v ≤ max{1, |ad|v } ≤ max{e2sv , e4sv |δ −1 |2v }. Now we assume that v is inﬁnite. Claim: There is u0 > 0 such that if 0 ≤ u ≤ u0 and B ≤ eu , then √ 5 max{|a − d|, |b + c|} ≤ 2 u,

472 / e. breuillard

√ max{|a|2 + |b|2 , |d|2 + |c|2 } ≤ 1 + 6u + 8 u, and √ √ max{|a|, |b|, |c|, |d|} ≤ 1 + 3u + 4 u ≤ 1 + 5 u.

6 7

To prove this, recall that the operator norm in SL2 (C) satisﬁes tr(B∗ B) = = B 2 + B −2 . Hence |a|2 + ... + |d|2 ≤ 1 + e2u , hence ≤ 4 if u is small enough (say, u ≤ .5). On the other hand, for small u, |a − d|2 + |b + c|2 = |a|2 + ... + |d|2 − 2 ≤ e2u − 1 ≤ 4u. Hence (5). Now √ √ |d| ≤ |a| + 2 u and since |a|, |b| ≤ 2, we get |d|2 ≤ |a|2 + 4u + 8 u and vice versa and similarly for b and c. Hence (6) and (7), and the claim is proved. Let now ε > 0 and assume that sv ≤ ε. From (4) we get |λ|2 + |λ−1 |2 + 2 tyδ ≤ 1 + e2ε , hence tyδ 2 ≤ e2ε − 1 ≤ 4ε if ε is small enough. So tyδ ≤ √ 2 ε. Now since PBP −1 ≤ eε , we have |t −2 c| ≤ 2 as soon as ε ≤ 12 . Hence √ √ √ |yct −1 | ≤ 4|δ|ε and max{|a|, |d|} ≤ 1 + 5 ε + 4|δ|ε . Finally, for some absolute √ √ constant C > 0 |ad| ≤ 1 + C( ε + |δ|ε + |δ|ε 2 ). |a|2 + |b|2 + |c|2 + |d|2

On the other hand, |cy 2 | = |t −2 c(ty)2 | ≤

8ε |δ|2

and |d − a||yt| ≤ 2 max{|a|, |d|} 16ε 2 2 −2 c| ≤ 2√ε, and |bc + |yt| ≤ + |δ| 2 . Also by (5), |bt + (d − a)yt − cy + t √ √ 48ε −2 c|2 ≤ 1 + 14√ε, |t −2 c|2 | ≤ 2|bt 2 + t −2 c| ≤ 4 ε + 48|δ| ε + |δ| 2 , and by (6), |t √ hence up to enlarging the absolute constant C, we also have |bc| ≤ 1 + C( ε + √ ε ε |δ| + |δ|2 ). Without the assumption that sv is small, we can make a coarser estimate: 2 2sv |t −2 c|2 ≤ 1 + e2sv , tyδ ≤ 1 + e2sv , hence cyt −1 ≤ 1+e |δ| and max{|a|, |d|} ≤ √ 1+e2sv 2sv ≤ 4e2sv max{1, 1 } and |ad| ≤ 16e4sv max{1, 1 }. Similarly, |δ| + 1 + e |δ| |δ|2 √ 24 ε |δ|

we compute |bc| ≤ 20e4sv max{1, |δ|12 }.

We now put together the local information obtained above to bound the heights. Let n = [K : Q] and Vf and V∞ the set of ﬁnite and inﬁnite places of K . Set ε0 , C0 , and C∞ the constants obtained in the previous lemma. For A > 0 and x ∈ Q, we set 1 A 8 h∞ (x) = nv · log+ |x|v [K : Q] v∈V∞ ,|x|v ≥A

where the sum is limited to those v ∈ V∞ , for which |x|v ≥ A. We have

lemma 4.3. For some constant C2 (2 ≤ C2 ≤ 2 + (2 log C∞ + 4)/ log 2), we have for all ε1 ∈ (0, 12 ) and all ε ≤ min{ε0 , ε12 } √ ε ε−1 9 max{h(ad), h(bc)} ≤ Cε,ε1 e(F) + 6C0 + 2hf (δ −1 ) + C2 · h∞1 (δ −1 ) ε1 2 log C∞ 2| log ε | + ε 1 and δ = λ − λ−1 . where Cε,ε1 = 4 + ε

heights on

SL2

and free subgroups / 473

Proof. Recall that sv = log Ev (F). If v ∈ V∞ and sv ≥ ε, then according to Lemma 4.2 log+ |ad|v ≤ 2 log C∞ + 4sv + 2 log+ |δ −1 |v hence

1 2 log C∞ 1 nv · log+ |ad|v ≤ 4 + nv sv n ε n v∈V∞ ,sv ≥ε

v∈V∞ ,sv ≥ε

+

2 n

nv · log+ |δ −1 |v .

v∈V∞ ,sv ≥ε

Fix ε1 < 12 . On the other hand, if sv ≤ ε ≤ min{ε0 , ε12 } and |δ|v ≥ ε1 , then √ √ √ s ε sv log+ |ad|v ≤ C0 ( sv + |δ|vv + |δ| 2 ) ≤ 3C0 ε1 and, as nv ≤ 2, v

1 n

nv · log+ |ad|v ≤ 6C0

v∈V∞ ,sv ≤ε,|δ|v ≥ε1

√ ε . ε1

While if sv < ε and |δ|v ≤ ε1 ≤ 12 , then log+ |ad|v ≤ C2 log+ |δ −1 |v for some absolute constant C2 , (2 ≤ C2 ≤ 2 + (2 log C∞ + 4)/ log 2), hence 1 n

nv · log+ |ad|v ≤

v∈V∞ ,sv 0 and ε1 > 0 with the following property. If λ ∈ Q is such that h(λ) ≤ η1 , deg (λ) ≥ d1 , then ε−1

h∞1 (

1 )≤α 1−λ

ε−1

where h∞1 was deﬁned in (8). Proof. Let P ∈ Z[X ] be the minimal polynomial of λ, that is, P(X ) = a X i = an x∈O(λ) (X − x). As P(1) ∈ Z\{0}, log |P(1)| = log |an | + 0≤i≤n i x∈O(λ) log |1 − x| ≥ 0. So

log

|1−x|≤ε1

1 ≤ |1 − x|

Recall (see [22] III.1.) that h(λ) = 11

1 n

log

|1−x|≤ε1

log |1 − x| + log |an |.

|1−x|>ε1

1 n

x∈O(λ) log

1 1 ≤ h(λ) + |1 − x| n

+

|x| + log |an | . Hence

log |1 − x|.

|1−x|>ε1

Consider the function fε1 (z) = 1|z−1|>ε1 log |1 − z|. It is locally bounded on C. By Theorem 4.4, for every ε1 > 0, there must exist d1 , η1 > 0 such 1 that, if h(λ) ≤ η1 and deg (λ) ≥ d1 , then n1 x fε1 (x) − 0 fε1 (e2π iθ )dθ ≤ α3 . On the other hand, we verify that θ → log |1 − e2πiθ | is in L1 (0, 1) and 1 2π iθ 0 log |1 − e |dθ = 0. Hence we can choose ε1 > 0 small enough so that 1 2π iθ )dθ ≤ α3 . Combining these inequalities with (11) and choosing 0 fε1 (e ε−1

η1 ≤ α3 , we get h∞1 ((1 − λ)−1 ) ≤ α.

Combining this with Bilu’s theorem, we get

lemma 4.6. (bounding errors terms via bilu’s theorem ii) For every α > 0 there exists η0 > 0 and A1 > 0 such that for any λ ∈ Q, if h(λ) ≤ η0 and d = deg (λ) > A1 , then hf (

1 ) ≤ 2α. 1−λ

Proof. We apply the product formula to μ = 1 − λ, which takes the form h(μ) = h(μ−1 ), hence hf (μ−1 ) = h∞ (μ) − h∞ (μ−1 ) + hf (μ). But hf (μ) = hf (1 − λ) ≤

heights on

SL2

and free subgroups / 475

hf (λ) ≤ η0 and h∞ (μ) − h∞ (μ−1 ) = [K1:Q] v∈V∞ nv · log |μ|v . Lemma 4.5 shows that the convergence in Equation (10) in Bilu’s theorem not only holds for compactly supported functions on C, but also for functions with logarithmic singularities at 1. In particular, it holds for the function f (z) = 1 log |1 − z|, which is exactly what we need, since 0 f (e2π iθ )dθ = 0. Hence 1 v∈V∞ nv · log |μ|v becomes small. We are done. [K :Q] 1 Proof of Proposition 4.1. Since hf ( λ−λ1 −1 ) ≤ hf (λ) + hf ( 1−λ 2 ) and similarly 1 1 A A A h∞ ( λ−λ−1 ) ≤ h∞ (λ) + h∞ ( 1−λ2 ), it follows from the last two lemmas that we ε−1

β as √ 3 ε 2C1 ε1 ≤ β3

can ﬁnd ε1 > 0, η > 0 and d0 ∈ N so that 2hf (δ −1 ) + C2 · h∞1 (δ −1 ) ≤ soon as h(λ) ≤ e(F) ≤ η and deg (λ) ≥ d0 . Then choose ε so the and ﬁnally take η even smaller so that Cε,ε1 η ≤ we are done.

β 3.

Now apply Lemma 4.3 and

End of the proof of Theorem 1.2:. The following lemma allows us, when proving Theorem 1.2, to assume without loss of generality that F = {1, A, B}, where A and B are two semisimple elements in SL2 (Q) that do not satisfy some prescribed ﬁnite set of algebraic relations. More precisely:

lemma 4.7. For every d1 ∈ N, there exists N(d1 ) ∈ N with the following property. Let F be a ﬁnite subset of SL2 (Q) containing 1 and generating a nonvirtually solvable subgroup, then there exists A, B ∈ F N(d1 ) such that A and B are semisimple, generate a nonvirtually solvable subgroup of SL2 , A is not of order at most d1 , and bc ∈ / 2iπ 4iπ {0, −1, e 3 , e 3 } after we conjugate A and B in the form of Equation (3). Proof. This is a direct application of Lemma 4.8 below applied to = F × F in SL2 × SL2 ≤ GL4 with X = X1 ∪ X2 ∪ X3 ∪ X4 where X1 = {(A, B), A or B has order at most d1 }, X2 = {(A, B), tr(A) or tr(B) is 2}, X3 = {(A, B), A and B generate a virtually solvable subgroup} and X4 the Zariski closure of 2iπ 4iπ {(gAg −1 , gBg −1 ), g ∈ SL2 , A diagonal, bc ∈ {0, −1, e 3 , e 3 }}. For dimension reasons X4 is a proper subvariety of SL2 × SL2 and Propositions 9.2 and 9.1 show that so is X3 .

lemma 4.8. (eskin-mozes-oh “escape from subvarieties”, see [13] and [9]) Let K be a ﬁeld, d ∈ N. For every m ∈ N, there is N ∈ N such that if X a K –algebraic subvariety of GLd (K ) such that the sum of the degrees of the geometrically irreducible components of X is at most m, then for any subset

476 / e. breuillard

⊂ GLd (K ) containing Id and generating a subgroup that is not contained in X (K ), we have N X (K ). Observe (as follows from the irreducibility of cyclotomic polynomials, see [22] VI.3.) that for every d0 ∈ N there is η0 > 0 and d1 > 0 such that if h(λ) < η0 and λ is not a root of 1 of order at most d1 , then deg (λ) ≥ d0 . However, recall the following well-known result (which is also a straightforward corollary of Theorem 4.4),

theorem 4.9. (zhang’s theorem [32]) There exists an absolute constant α0 > 0 such that for any x ∈ Q, we have h(x) + h(1 + x) > α0 unless x ∈ {0, −1, e

2iπ 3

,e

4iπ 3

}.

Let β = α20 where α0 is given by Theorem 4.9. Proposition 4.1 yields d0 > 0 and η = η( α20 ) > 0 such that max{h(ad), h(bc)} ≤ β as soon as e({Id, A, B}) ≤ η and deg (λ) ≥ d0 . By Lemma 4.7, if we have some nice A, B ∈ F N(d1 ) . If e(F) ≤ α

min{η( 20 ),η0 } , N(d1 )

then e({Id, A, B}) ≤ min{η, η0 } and λ is not a root of 1 of order at most d1 . Hence deg (λ) ≥ d0 and by Proposition 4.1, h(ad) + h(bc) ≤ 2β = α0 . 2iπ 4iπ Then according to Theorem 4.9, bc ∈ {0, −1, e 3 , e 3 }, which contradicts our min{η(

α0

),η }

0 2 > 0 is the desired gap. choice of A, B. So N(d1 ) This ends the proof of Theorem 1.2. Finally, observe that Theorem 1.2 combined with Propositions 3.1 to 3.3 implies

proposition 4.10. There exists a constant c0 > 0 such that if F is any ﬁnite subset of SL2 (Q) generating a nonvirtually solvable subgroup, then e(F) ≥ h(F) ≥ c0 · e(F).

5. Simultaneous Quasisymmetrization over Q Here we are going to use our previous height estimates once again to show the following proposition. Observe that the minimal height e(F) coincides with the inﬁmum of h(gFg −1 ) over all adelic points g = (gv )v . The lemma we are about to state essentially means that this inﬁmum is attained (up to additive and multiplicative constants) with a conjugating matrix g lying already in SL2 (Q) as opposed to SL2 (A) (the adelic group).

heights on

SL2

and free subgroups / 477

proposition 5.1. (simultaneous quasisymmetrization) There is an absolute constant C > 0 such that if F is a ﬁnite subset of SL2 (Q) generating a nonvirtually solvable subgroup, then there is an element g ∈ SL2 (Q) such that h(gFg −1 ) ≤ C · e(F) + C When a matrix with real entries is symmetric, then its norm coincides with the modulus of its maximal eigenvalue. Thus the proposition amounts to conjugating F simultaneously (i.e., by a single g ∈ SL2 (Q)) in a "quasisymmetric" position. Proof. As we may replace F by a bounded power of it, Lemma 4.7 above allows us to assume that F contains a semisimple element. Let F = {Id, A, B1 , ..., Bk } with A semisimple. Conjugating by some g ∈ SL2 (Q), we may assume that A is in diagonal form and we write each Bi in the form of Equation (3) with entries ai , bi , ci , and di . Changing F into F 2 if necessary, we may assume that both b1 and c1 are not 0 (otherwise F would be contained in the group of upper- or lower-triangular matrices). We may further conjugate F by the diagonal matrix diag(t, t −1 ), where t ∈ Q is a root of t 4 = c1 /b1 , so as to ensure b1 = c1 . Then h(B1 ) ≤ h(a1 ) + h(d1 ) + 2h(b1 ) + log 2. On the one hand, since a1 d1 − b1 c1 = 1, we have b12 = a1 d1 − 1 and 2h(b1 ) = h(b12 ) ≤ h(a1 d1 ) + log 2 ≤ 2e({A, B}) + log 2C∞ . On the other hand, by Lemma 4.2 applied to {A, Bi }, we have max{|ai |v , |di |v } ≤ Cv e2sv max{1, |δ −1 |v }, for every place v, where δ = λ − λ−1 and sv = sv ({A, Bi }) = log Ev ({A, Bi }). Applying Lemma 4.2 to {A, B1 Bi } we get max{|(B1 Bi )11 |v , |(B1 Bi )22 |v } ≤ Cv e2sv max{1, |δ −1 |v } with sv = sv ({A, B1 Bi }) = log Ev ({A, B1 Bi }). We compute the matrix entry (B1 Bi )11 = a1 ai + b1 ci . We get |ci |v = |[(B1 Bi )11 − a1 ai ]b1−1 |v ≤ Cv e2sv max{1, |δ −1 |v } max{1, |b1−1 |v }. Similarly for |bi |v . Hence, ||F||v ≤ Cv max {|λ|v , |λ−1 |v , |ai |v , |di |v , |bi |v , |ci |v } i=1,...,k

≤ Cv max Ev ({A, B1 , B1 Bi })2 · max{1, |δ −1 |v } max{1, |b1−1 |v }. i=1,...,k

In particular, this means that h(F) ≤ 2 log C∞ + 2e(F 2 ) + h(δ) + h(b1 ) ≤ 7e(F) + 4 log 2C∞ . So we are done.

478 / e. breuillard

corollary 5.2. There exists a constant Cqs > 0 such that if F is as in the proposition, then there is an element g ∈ SL2 (Q) such that h(gFg −1 ) ≤ Cqs · e(F). Proof. It is clear from the combination of the previous proposition and Theorem 1.2.

6. Ping-pong Here we state and prove a ping-pong criterion, which gives a sufﬁcient condition on the ﬁnite set F for it, or a bounded power of it, to contain two free generators of a free subgroup. Let k1 , k2 , k3 ∈ N be three positive integers and let k be a local ﬁeld of characteristic 0 with its standard absolute value. We set Ck = 2 if k is Archimedean, and Ck = 1 if k is non-Archimedean. Let F ⊂ SL2 (k) be a ﬁnite set containing 1 such that k (F k1 ) > Ck ||F||k (see Section 2 for notation, it is important to require a strict inequality here when k is non-Archimedean). Let A ∈ F k1 be such that k (A) = k (F k1 ). Then of course A is semisimple and admits two distinct eigenvectors v + and v − in k2q where kq is either k or some