### November 27, 2006

2000 Mathematics Subject Classification. Primary 20F65; Secondary 20E08, 20F67.
<ph f="cmbx">A note on quotients of word hyperbolic groups with Property (T) or Property (FA)</ph>

### Yves de Cornulier

• Abstract. Every discrete group with Kazhdan's Property (T) (resp. Property (FA)) is a quotient of a torsion-free, word hyperbolic group with Property (T) (resp. (FA)).
Recall ([HV, [Ser) that a countable discrete group $G$  has Property (T) (resp. (FA)) if every isometric action of $G$  on an affine Hilbert space (resp. on a simplicial tree) has a fixed point (or, equivalently, has bounded orbits).
As immediate consequences of the definitions, Property (T) and (FA) are inherited by quotients and by extensions.
By a result of Alperin and Watatani [HV,Chap.6, Property (T) implies Property (FA).
The converse is well-known to be false: using [Ser,Corollaire2,p.90, the affine Coxeter group $〈x,y,z|{x}^{2}={y}^{2}={z}^{2}=\left(xy{\right)}^{3}=\left(xz{\right)}^{3}=\left(yz{\right)}^{3}=1〉$  has Property (FA). On the other hand, it has a subgroup of finite index isomorphic to ${\mathbf{Z}}^{2}$  , hence without Property (T). Therefore it does not have Property (T), since Property (T) is known to be inherited by subgroups of finite index (see [BHV,Section2.6for a direct proof from the definition given here).
However, Property (FA) and (T) share some features. For instance [Ser,Théorème15p.81, a countable group $G$  has Property (FA) if and only if it satisfies the three following conditions:
(i) $G$  is finitely generated, (ii) $G$  does not map onto $\mathbf{Z}$  , (iii) $G$  does not decompose as a nontrivial amalgam.
Shalom [Sha,Theorem6.2has proved a similar characterization for Property (T): a countable group $G$  has Property (T) if it satisfies Conditions (i), (ii), and (iii'), where (i) and (ii) as above, and with (iii') defined as: $\text{Cor}\left(G\right)=\left\{{1}_{G}\right\}$  , where the cortex $\text{Cor}\left(G\right)$  of $G$  is defined as the set of (isomorphism classes of ) irreducible unitary representation which cannot be Hausdorff separated from the trivial representation ${1}_{G}$  for the Fell topology (see [HV,Chap.1).
Shalom [Sha,Theorem6.7has proved the following remarkable result for Property (T).
Theorem 1 (Shalom, 2000). For every group $G$  with Property (T), there exists a finitely presented group ${G}_{0}$  with Property (T) which maps onto $G$  .
In other words, this means that, given a finite generating subset for $G$  , only finitely many relations suffice to imply Property (T). This can be interpreted in the topology of marked groups [Chaas: Property (T) is an open property. Relying on ideas of V. Lafforgue, we prove that a similar result holds for Property (FA).
Theorem 2. For every group $G$  with Property (FA), there exists a finitely presented group ${G}_{0}$  with Property (FA) which maps onto $G$  .
We refer to [GHfor the notion, due to Gromov, of word hyperbolicity. We only recall that a word hyperbolic group is a finitely generated group whose Cayley graph satisfies a certain condition meaning that, at large scale, it is negatively curved. We only mention here that word hyperbolic groups are necessarily finitely presented, and that word hyperbolicity is a fundamental notion in combinatorial group theory as in geometric topology.
It was asked [Wo,Question16whether every group with Property (T) is quotient of a group with Property (T) with finiteness conditions stronger than finite presentation. We give an answer here by showing that we can impose word hyperbolicity.
Theorem 3. For every group $G$  with Property (T) (resp. (FA)), there exists a torsion-free word hyperbolic group ${G}_{0}$  with Property (T) (resp. (FA)) which maps onto $G$  .
Theorem  3 is proved by combining Theorem  1 (for Property (T)) and Theorem  2 (for Property (FA)) with the following remarkable result of Ollivier and Wise [OW. Since it involves some technical definitions, we do not quote it in full generality.
Theorem 4 (Ollivier and Wise, 2005). To every finitely presented group $Q$  , we can associate a short exact sequence $1\to N\to G\to Q\to 1$  such that
• (1) $G$  it is torsion-free, word hyperbolic,
• (2) $N$  is 2-generated and has property (T).
Corollary 5. For every finitely presented group $Q$  with Property (T) (resp. Property (FA)), there exists a torsion-free word-hyperbolic group $G$  mapping onto $Q$  with finitely generated kernel.
Proof : Apply Theorem  4 to $Q$  , so that $G$  lies in an extension $1\to N\to G\to Q\to 1$  , where $N$  has Property (T) and $Q$  has Property (T) (resp. (FA)). If $Q$  has Property (T), then, since Property (T) is stable under extensions, so has $G$  . On the other hand, since Property (T) implies Property (FA), $N$  has Property (FA), so that the same arguments imply that if $Q$  has Property (FA), so does $G$  . $■$
Remark 6. In the case of Property (T), Corollary  5 answers a question at the end of [OW.
Proof of Theorem  3 : Let $G$  be a countable group with Property (T) (resp. (FA)). By Theorem  1 (resp. Theorem  2 ), there exists a finitely presented group $Q$  with Property (T) (resp. (FA)) mapping onto $G$  , and by Corollary  5 , there exists a finitely presented group ${G}_{0}$  with Property (T) (resp. (FA)) mapping onto $Q$  , so that ${G}_{0}$  maps onto $G$  . $■$
Question 7. 1) In Theorem  4 , can $G$  be chosen, in addition, residually finite? In [Wis, a similar result is proved, $G$  being torsion-free, word hyperbolic, residually finite, and $N$  finitely generated, but never has Property (T).
2) Let $G$  be a word hyperbolic group (maybe torsion-free), and $H$  a quotient of $G$  generated by $r$  elements. Does there exist an intermediate quotient which is both word hyperbolic and generated by $r$  elements? (The analog statement with “word hyperbolic” replaced by “finitely presented” is immediate.) The motivation is that, in Theorem  3 , we would like to have ${G}_{0}$  generated by no more elements than $G$  . Theorem  4 only tells us that if $G$  is $r$  -generated, then ${G}_{0}$  can be chosen $\left(r+2\right)$  -generated.
We finally turn to the proof of Theorem  2 . The proof relies on ideas of V. Lafforgue (personal communication), who proves the following result.
Theorem (V. Lafforgue). Let $G$  be a group generated by a finite symmetric subset $S$  .
Suppose that $G$  does not have Property (T). Then there exists an isometric action of $G$  on an affine Hilbert space $\mathcal{ℋ}$  , and $v\in \mathcal{ℋ}$  , such that ${sup}_{g\in S}\parallel v-gv{\parallel }^{2}=1$  , and, for all $n$  , ${sup}_{g\in {S}^{n}}\parallel v-gv{\parallel }^{2}\ge n/2$  .
We borrow some of his arguments; the case of trees being much more simpler than that of Hilbert spaces since distances are integer-valued.
Recall that a length function on a group $G$  is a function $L:G\to {\mathbf{R}}_{+}$  such that $L\left(1\right)=0$  , $L\left(g\right)=L\left({g}^{-1}\right)$  , and $L\left(gh\right)\le L\left(g\right)+L\left(h\right)$  for all $g,h\in G$  .
If $G$  acts isometrically on a metric space $X$  , then, for every $x\in X$  , the function $g↦d\left(x,gx\right)$  is a length function. It is well-known and easy that every length function $L$  arises this way.
A tree length function on $G$  is an integer-valued length function satisfying
 $\begin{array}{c}L\left({g}^{-1}h\right)+L\left({k}^{-1}l\right)\le max\left(L\left({g}^{-1}k\right)+L\left({h}^{-1}l\right),L\left({g}^{-1}l\right)+L\left({h}^{-1}k\right)\right)\text{for all}g,h,k,l\in G,\end{array}$ (1)
 $\begin{array}{c}\text{and}L\left({g}^{-1}h\right)+L\left({h}^{-1}k\right)+L\left({k}^{-1}g\right)\in 2\mathbf{N}\text{for all}g,h,k\in G.\end{array}$ (2)
If $G$  acts isometrically on a tree $T$  , then, for every vertex $v\in T$  , the function $g↦d\left(v,gv\right)$  is a tree length function. It is well-known [Chithat every tree length function $L$  arises this way. In particular, $G$  has Property (FA) if and only if every tree length function on $G$  is bounded.
Lemma 8. Let $G$  be a group generated by a symmetric finite set $S$  , and $L$  an unbounded tree length function on $G$  . Then, for all $n\in \mathbf{N}$  , ${sup}_{g\in {S}^{n}}L\left(g\right)\ge n/2$  .
Proof : Let $T$  be a tree on which $G$  acts, ${x}_{0}$  a vertex of $T$  so that $L\left(g\right)=d\left({x}_{0},g{x}_{0}\right)$  for all $g\in G$  .
Let ${B}^{\prime }\left({c}_{n},{r}_{n}\right)$  be the ball of minimal radius containing ${S}^{n}{x}_{0}$  [BrH,Chap.II,Corollary2.8(1).
Observe that ${c}_{n}$  is necessarily either a vertex, either the middle of an edge.
We claim that, for all $n$  , ${r}_{n}\ne {r}_{n+1}$  . Indeed, suppose that ${r}_{n}={r}_{n+1}$  . Then ${S}^{n}{x}_{0}\subset {S}^{n+1}{x}_{0}\subset {B}^{\prime }\left({c}_{n+1},{r}_{n}\right)$  , so that, by uniqueness of the minimal ball, ${c}_{n}={c}_{n+1}$  . On the other hand, if $s\in S$  , then ${s}^{-1}{S}^{n}{x}_{0}\subset {S}^{n+1}{x}_{0}\subset {B}^{\prime }\left({c}_{n+1},{r}_{n}\right)$  , so that ${S}^{n}{x}_{0}\subset {B}^{\prime }\left(s{c}_{n+1},{r}_{n}\right)$  . Again by uniqueness, ${c}_{n}=s{c}_{n+1}$  . It follows that ${c}_{n+1}=s{c}_{n+1}$  , so that ${c}_{n+1}$  is a fixed point for the action. This contradicts the fact that $L$  is unbounded.
Now observe that ${r}_{n}\in \left(1/2\right)\mathbf{N}$  for all $n$  , and $\left({r}_{n}\right)$  is an increasing sequence. It follows that ${r}_{n}\ge n/2$  for all $n$  . Since ${S}^{n}{x}_{0}$  is contained in the ball ${B}^{\prime }\left({x}_{0},{sup}_{g\in {S}^{n}}L\left(g\right)\right)$  , the lemma follows. $■$  Proof of Theorem  2 . Since Property (FA) implies finite generation, we can write $G=F/N$  , where $F$  is, a free group of finite rank, generated by a symmetric finite set $S$  , and $N$  is a normal subgroup of $F$  . Write $N=\cup {N}_{k}$  , with ${N}_{k}◃F$  finitely generated as a normal subgroup of $F$  , and ${N}_{k}\subset {N}_{k+1}$  for all $k$  .
Suppose that, for all $k$  , $F/{N}_{k}$  does not have Property (FA). Let ${L}_{k}$  be an unbounded tree length function on $F/{N}_{k}$  , and view it as a tree length function on $F$  vanishing on ${N}_{k}$  . By Lemma  8 , ${sup}_{{S}^{n}}{L}_{k}\ge n/2$  for all $n$  . By a standard diagonal argument, we can extract a subsequence $\left({L}_{{k}_{i}}\right)$  converging to a function $L$  , so that ${sup}_{{S}^{n}}L\ge n/2$  for all $n$  . Then $L=0$  on $N$  , hence defines an unbounded tree function on $G/N$  . It follows that $G/N$  does not have Property (FA). $■$  Acknowledgments. I am indebted to Vincent Lafforgue for telling me about his results, and for allowing me to use some of his arguments in the present context. I also thank Yann Ollivier and Alain Valette for their comments.
References

1. Bachir Bekka, Pierre de la Harpe, Alain Valette. “Kazhdan's Property (T)”, forthcoming book, available at , 2004. ☻ open access ✓
2. Martin R. Bridson, André Haefliger. “Metric spaces of non-positive curvature”. Springer-Verlag, Berlin, 1999.
3. Christophe Champetier. L'espace des groupes de type fini, Topology 39(4) 657-680, 2000.
4. I. M. Chiswell. Abstract length functions in groups. Math. Proc. Camb. Phil. Soc. 80, 451-463, 1976.
5. Étienne Ghys, Pierre de la Harpe. “Sur les groupes hyperboliques, d'après Mikhael Gromov”. Progress in Mathematics, 83. Birkhäuser, 1990.
6. Pierre de la Harpe, Alain Valette. “La propriété (T) de Kazhdan pour les groupes localement compacts”, Astérisque 175, SMF, 1989.
7. Yann Ollivier, Daniel Wise. Kazhdan groups with infinite outer automorphism group. To appear in Trans. A.M.S., 2005.
8. Jean-Pierre Serre. “Arbres, amalgames, ${\text{SL}}_{2}$  ”. Astérisque 46, SMF, 1977.
9. Yehuda Shalom. Rigidity of commensurators and irreducible lattices. Inv. Math. 141(1):1-54, 2000.
10. Daniel T. Wise. A residually finite version of Rips's construction. Bull. London Math. Soc. 35(1): 23-29, 2003.
11. Report of the workshop Geometrization of Kazhdan's Property (T) (organizers: B. Bekka, P. de la Harpe, A. Valette; 2001). Unpublished; currently available at . ☻ open access ✓

Yves de Cornulier École Polytechnique Fédérale de Lausanne (EPFL) Institut de Géométrie, Algèbre et Topologie (IGAT) CH-1015 Lausanne, Switzerland E-mail: decornul@clipper.ens.fr