Groups of piecewise projective homeomorphisms Nicolas Monod ∗ ∗EPFL, 1015 Lausanne, Switzerland Submitted to Proceedings of the National Academy of Sciences of the United States of America The group of piecewise projective homeomorphisms of the line pro- PA. We write H(A) = G(A) \ H, which is the stabilizer of 1 vides straightforward torsion-free counter-examples to the so-called in G(A). von Neumann conjecture. The examples are so simple that many additional properties can be established. The main result of this article is the following, which relies on a new method for proving non-amenability. Theorem 1. The group H(A) is non-amenable if A 6= Z. Introduction The next result is a sequacious generalization of the corre- In 1924, Banach and Tarski accomplished a rather paradoxical sponding theorem of Brin{Squier about piecewise affine trans- feat. They proved that a solid ball can be decomposed into formations [7] and we claim no originality. five pieces which are then moved around and reassembled in Theorem 2. The group H does not contain any non-abelian free such a way as to obtain two balls identical with the original subgroup. Thus, H(A) inherits this property for any subring one [6]. This wellnigh miraculous duplication was based on A < R. Hausdorff's 1914 work [18]. Thus already H = H(R) itself is a counter-example to In his 1929 study of the Hausdorff–Banach{Tarski para- the von Neumann conjecture. Writing H(A) as the directed dox, von Neumann introduced the concept of amenable union of its finitely generated subgroups, we deduce: groups [37]. Tarski readily proved that amenability is the only obstruction to paradoxical decompositions [34, 35]. However, Corollary 3. For A 6= Z, the groups H(A) contain finitely gen- the known paradoxes relied more prosaically on the existence erated subgroups that are simultaneously non-amenable and of non-abelian free subgroups. Therefore, the main open prob- without non-abelian free subgroups. lem in the subject remained for half a century to find non- amenable groups without free subgroups. Von Neumann's Further properties The groups H(A) seem to enjoy a number name was apparently attached to it by Day in the 1950s. of additional interesting properties, some of which are weaker The problem was finally solved around 1980: Ol0shanski˘ı forms of amenability. In the last section, we shall prove the proved the non-amenability of the Tarski monsters that he following five propositions (and recall the terminology). Here had constructed [28, 29, 30]; Adyan showed that his work A < R is an arbitrary subring. on Burnside groups yields non-amenability [3, 4]. Finitely Proposition 4. All L2-Betti numbers of H(A) and of G(A) van- presented examples were constructed another twenty years ish. 0 later by Ol shanski˘ı–Sapir [27]. There are several more re- Proposition 5. The group H(A) is inner amenable. cent counter-examples [15, 31, 32]. Proposition 6. The group H is bi-orderable and hence so are all its subgroups. It follows that there is no non-trivial homo- Given any subring A < R, we shall define a group G(A) morphism from any Kazhdan group to H. and a subgroup H(A) < G(A) of piecewise projective trans- 1 Proposition 7. Let E ⊆ P be any subset. Then the subgroup of formations. Those will provide concrete, uncomplicated new H(A) which fixes E pointwise is co-amenable in H(A) unless examples with many additional properties. Perhaps ironically, E is dense (in which case the subgroup is trivial). our short proof of non-amenability ultimately relies on basic free groups of matrices, as in Hausdorff's 1914 paradox, even Proposition 8. If H(A) acts by isometries on any proper though the Tits alternative [36] shows that the examples can- CAT(0) space, then either it fixes a point at infinity or it not be linear themselves. preserves a Euclidean subspace. One can also check that H(A) satisfies no group law and Construction has vanishing properties in bounded cohomology (see below). I saw the pale student of unhallowed arts kneeling beside the thing he had put together. Mary Shelley, Frankenstein Non-amenability (introduction to the 1831 edition) An obvious difference between the actions of PSL2(A) and of 1 Consider the natural action of the group PSL2(R) on H(A) on P is that the latter group fixes 1 whilst the former the projective line P1 = P1(R). We endow P1 with its R- does not. The next proposition shows that this is the only topology making it a topological circle. We denote by G the difference as far as the orbit structure is concerned. group of all homeomorphisms of P1 which are piecewise in 1 PSL2(R), each piece being an interval of P , with finitely many pieces. We let H < G be the subgroup fixing the point Reserved for Publication Footnotes 1 2 P1 corresponding to the first basis vector of R2. Thus H is left-orderable since it acts faithfully on the topological line P1 n f1g, preserving orientations. It follows in particular that H is torsion-free. 1 Given a subring A < R, we denote by PA ⊆ P the collec- tion of all fixed points of all hyperbolic elements of PSL2(A). This set is PSL2(A)-invariant and is countable if A is so. We define G(A) to be the subgroup of G given by all elements that are piecewise in PSL2(A) with all interval endpoints in www.pnas.org/cgi/doi/10.1073/pnas.0709640104 PNAS Issue Date Volume Issue Number 1{5 Proposition 9. Let A < R be any subring and let p 2 P1 nf1g. An a.e. free action of a countable group is amenable in Zim- Then mer's sense [40, 4.3] if and only if the associated relation is amenable; see [2, Thm. A]. PSL2(A) · p ⊆ f1g [ H(A) · p: Proof of Theorem 1. Let A 6= Z be a subring of R. Then Thus, the equivalence relations induced by the actions of A contains a countable subring A0 < A which is dense in R. PSL (A) and of H(A) on P1 coincide when restricted to 2 Since H(A0) is a subgroup of H(A), we can assume that A P1 n f1g. itself is countable dense. Now H(A) is a countable group and Γ := PSL (A) is a countable dense subgroup of PSL (R). Proof. We need to show that given g 2 PSL (A) with 2 2 2 It is proved in Th´eor`eme 3 of [10] that the equivalence gp 6= 1, there is an element h 2 H(A) such that hp = gp. relation on PSL (R) induced by the multiplication action of We assume g1 6= 1 since otherwise h = g will do. Equiva- 2 Γ is non-amenable; see also Remarks 10 and 11 below. Equiv- lently, we need an element q 2 G(A) fixing gp and such that alently, the Γ-action on PSL (R) is non-amenable. Viewing q1 = g1, writing h = q−1g. It suffices to find a hyperbolic 2 P1 as a homogeneous space of PSL (R), it follows that the element q 2 PSL (A) with q 1 = g1 and whose fixed points 2 0 2 0 Γ-action on P1 is non-amenable. Indeed, amenability is pre- ξ 2 P1 separate gp from both 1 and g1, see Figure 1. In- ± served under extensions, see [39, 2.4] or [2, Cor. C]. This deed, we can then define q to be the identity on the component action is a.e. free since any non-trivial element has at most of P1 n fξ g containing gp, and define q to coincide with q ± 0 two fixed points. Thus the relation induced by Γ on P1 on the other component. is non-amenable. Restricting to P1 n f1g, we deduce from Proposition 9 that the relation induced by the H(A)-action is g1 also non-amenable. (Amenability is preserved under restric- tion [20, 9.3], but here f1g is a null-set anyway.) Thus H(A) ξ+ r is a non-amenable group. r 1 ξ r − Remark 10. We recall from [10] that the non-amenability of the r Γ-relation on PSL2(R) is a general consequence of the exis- gp tence of a non-discrete non-abelian free subgroup of Γ. Thus Fig. 1. The desired configuration of ξ± the main point of our appeal to [10] is the existence of this r non-discrete free subgroup, but this is much easier to prove directly in the present case of Γ = PSL2(A) than for general a b non-discrete non-soluble Γ. Let be a matrix representative of g; thus, c d Remark 11. Here is a direct argumentp avoiding all the above a; b; c; d 2 A and ad − bc = 1. The assumption g1 6= 1 references in the examples of A = Z[ 2] or A = Z[1=`], where 1 implies c 6= 0 and thus we can assume c > 0. Let q0 be ` is prime. We show directly that the Γ-action on P is not a b + ra given by with r 2 A to be determined later; amenable. We consider Γ as a lattice in L := PSL2(R) × c d + rc PSL2(R) in the first case and in L := PSL2(R) × PSL2(Q`) thus q01 = g1. This matrix is hyperbolic as soon as jrj is in the second case, both times in such a way that the Γ-action large enough to ensure that the trace τ = a + d + rc is larger on P1 extends to the L-action factoring through the first fac- than 2 in absolute value.