Amenability of linear-activity automaton groups (Q1949961)

Automaton groups are groups of automorphisms of the \(m\)-regular rooted tree \(X^*\), for an \(m\)-letter alphabet \(X\), whose elements are automatic transductions. Such a transduction may be described by a finite graph, a choice of initial vertex called the root, and labels in \(X\times X\) on each edge, in such a manner that, for every word \(v=v_1\dots v_n\in X^*\), there exists a unique path starting at the root whose consecutive labels are of the form \((v_1,w_1),\dots,(v_n,w_n)\) for some \(w_i\in X\). The corresponding action on \(X^*\) is defined by mapping \(v\) to \(w=w_1\dots w_n\). An automaton may contain vertices \(e\) such that all their outgoing edges are loops labeled \((x,x)\). They represent the identity transformation, and are called trivial states. The activity growth of an automaton is the function mapping \(n\) to the number of paths of length \(n\) starting at the root that do not end in a trivial state. It is easily seen to be either polynomial or exponential. \textit{S. Sidki} noted in [J. Math. Sci., New York 100, No. 1, 1925--1943 (2000; Zbl 1069.20504)] that, for every \(d\), the set of automatic transductions whose activity growth is bounded by a polynomial of degree \(d\) forms a group \(\text{Pol}_m(d)\), and that \(\text{Pol}_2(0)\) already contains striking examples of finitely generated groups such as infinite torsion groups and groups of intermediate word growth. He also proved in [Geom. Dedicata 108, 193--204 (2004; Zbl 1075.20011)] that \(\text{Pol}_m(d)\) does not contain any non-abelian free groups, and asked whether \(\text{Pol}_m(d)\) is amenable. This was proven in [the reviewer, \textit{V. A. Kaimanovich} and \textit{V. V. Nekrashevych}, Duke Math. J. 154, No. 3, 575--598 (2010; Zbl 1268.20026)] for \(\text{Pol}_m(0)\), and is proven in the present paper for \(\text{Pol}_m(1)\). The method of proof consists in considering finitely generated subgroups of \(\text{Pol}_m(1)\); at the cost of increasing the alphabet size \(m\), such a group can be embedded in a specific subgroup of \(\text{Pol}_m(1)\), a ``mother group'' \(M\). A random walk on \(M\) is then constructed, and it is shown to have trivial asymptotic entropy. This implies amenability of \(M\), and hence of \(\text{Pol}_m(1)\). The authors note that their method of proof may, possibly, be pushed up to \(\text{Pol}_m(2)\), but certainly not further, at least if random walks with finite \(1\)-step distribution are to be considered.