Relative expanders (Q2345929)

The main result of this paper is a construction of a sequence of graphs with uniformly bounded (actually constant) degrees which on the one hand are not uniformly coarsely embeddable into a Hilbert space, and on the other hand do not contain any family of expanders weakly. Recall that a sequence \(\{X_n\}_{n=1}^\infty\) of graphs with uniformly bounded degrees is said to contain a family \(\{Y_n\}_{n=1}^\infty\) of expanders weakly if there are maps \(f_n:Y_n\to X_n\) (with some abuse of notation we use \(X_n,Y_n\) to denote the vertex sets of \(X_n,Y_n\)) satisfying the following conditions: (1) the Lipschitz constants of \(\{f_n\}\) are uniformly bounded; (2) \(\lim_{n\to\infty}\max_{y\in Y_n}|f_n^{-1}(f_n(y))|/|Y_n|=0\). By uniformly coarse embeddability of a sequence we mean a sequence of embeddings satisfying the standard inequalities where the lower and upper control functions are the same for the whole sequence. The problem about the existence of such a sequence of graphs arises naturally in connection with results of \textit{G. Kasparov} and \textit{G.-L. Yu} [Adv. Math. 206, No. 1, 1--56 (2006; Zbl 1102.19003)]. It seems that the first paper where this problem was explicitly mentioned is [the reviewer, Topol. Proc. 33, 163--183 (2009; Zbl 1179.54042)]; see also Chapter~7 and Section 11.2 in [the reviewer, Metric embeddings. Bilipschitz and coarse embeddings into Banach spaces. Berlin: de Gruyter (2013; Zbl 1279.46001)] for related discussions. The construction of the paper is very interesting and we enclose a brief outline of it. The construction is based on Kazhdan's property (T) for pairs, also known as relative property (T); this is the reason why the authors call the constructed graphs relative expanders. The authors start by recalling the fact that the pair \((\mathrm{SL}(2,\mathbb{Z})\ltimes\mathbb{Z}^2, \mathbb{Z}^2)\) has property (T). This result goes back to \textit{D. A. Kazhdan} [Funct. Anal. Appl. 1, 63--65 (1967); translation from Funkts. Anal. Prilozh. 1, No. 1, 71--74 (1967; Zbl 0168.27602)] and \textit{G. A. Margulis} [Ergodic Theory Dyn. Syst. 2, 383--396 (1982; Zbl 0532.28012)]. Then they apply results of \textit{M. Burger} [J. Reine Angew. Math. 413, 36--67 (1991; Zbl 0704.22009)] to get that property (T) still holds if we replace the first element of the pair twice. The first time we replace it by its subgroup \(Q\ltimes\mathbb{Z}^2\), where \(Q\) consists of matrices whose reduction modulo \(2\) is the identity (this replacement is needed in some of the further group-theoretic constructions). The second time we replace \(Q\ltimes\mathbb{Z}^2\) by \(\mathbb{F}_3\ltimes\mathbb{Z}^2\); here we use the fact that the group \(Q\) is generated by \(3\) elements and thus there is a natural quotient of \(\mathbb{F}_3\) (the free group with three generators) onto \(Q\). Writing \(\mathbb{F}_3\ltimes\mathbb{Z}^2\) we mean the action of \(\mathbb{F}_3\) on \(\mathbb{Z}^2\) determined by this quotient and the standard action of \(Q\) on \(\mathbb{Z}^2\). A relative version of the result of \textit{P. Delorme} [Bull. Soc. Math. Fr. 105, 281--336 (1977; Zbl 0404.22006)] implies that the pair \((\mathbb{F}_3\ltimes\mathbb{Z}^2, \mathbb{Z}^2)\) has the relative property (FH), defined as follows: Let \(H\) be a closed subgroup of a topological group \(G\). The pair \((G,H)\) has relative property (FH) if every affine isometric action of \(G\) on a real Hilbert space has an \(H\)-fixed point. After that, the authors make the following important observation: Let \(G\) be a finitely generated group, \(S\) be a finite generating set in \(G\), and suppose that \((G,Y)\) has relative property (FH). Then there exists \(C>0\) such that for every finite quotient \(Q\) of \(G\), every function \(f\) from \(Q\) to a Hilbert space satisfies the following ``relative Poincaré'' inequality: For every \(y\in Y\), \[ \sum_{g\in Q}||f(g\bar y)-f(g)||^2\leq C\sum_{g\in Q,~s\in S}||f(g\bar s)-f(g)||^2, \] where \(\bar y\) and \(\bar s\) denote the natural images of \(y\) and \(s\) in \(Q\). It is easy to see that this ``relative Poincaré'' inequality implies lack of uniformly coarse embeddability into a Hilbert space of any sequence of finite quotients \(\{Q_n\}\) of \(G\), endowed with the metrics corresponding to the natural images of \(S\) in \(Q_n\), for which images of \(Y\) do not have uniformly bounded diameters. To get such a sequence \(\{Q_n\}\), which additionally does not contain a sequence of expanders weakly, the authors prove the following: (1) Semidirect products of sequences of finite groups which are uniformly coarsely embeddable into a Hilbert space cannot contain expanders weakly (one has to be more specific about the metric in this statement, see Proposition 2 in the paper). (2) Some of the finite quotients \(\{Q_n\}\) of the group \(\mathbb{F}_3\ltimes\mathbb{Z}^2\) described above can be presented as semidirect products of groups which are uniformly coarsely embeddable into a Hilbert space. This is done using a very interesting and important free-group-related construction of \textit{G. Arzhantseva} et al. [Geom. Funct. Anal. 22, No. 1, 22--36 (2012; Zbl 1275.46013)]. The authors also give a different construction (based on generalized wreath products) as well as study coarse embeddability of the obtained spaces into \(L_p\), \(2<p<\infty\). In the reviewer's opinion, this piece of research contains many interesting ideas which deserve further exploration. The authors suggest some interesting directions for further study as open problems.