Amenable graphs and the spectral radius of extensions of Markov maps

Abstract

We discuss relations between the amenability of a graph and spectral properties of a random walk driven by a dynamical system. In order to include graphs which are not locally compact, we introduce the concept of amenability of weighted graphs, which generalises the usual notion as the new definition is shown to be equivalent to Foelner's condition. As a first result, we obtain the following generalisation of Kesten's amenability criterion to graphs and non-independent increments: If the random walk is driven by a full-branched Gibbs-Markov map, the graph is amenable with respect to the weight induced by the random walk if and only if the spectral radius of the associated Markov operator is equal to one. By employing inducing schemes, one then obtains criteria for amenability through Markov maps with less regularity. We conclude the paper with the following applications to Schreier graphs. If the random walk is driven by an uniformly expanding map with non-Markovian increments, then, under certain conditions, the Schreier graph is amenable if the probability of a return in time n does not decay exponentially in n. Furthermore, in the context of geometrically finite Kleinian groups, one obtains a version of Brooks's amenability criterion for not necessarily normal subgroups.