Regularity of the entropy for random walks on hyperbolic groups (Q378816): Difference between revisions

In previous investigations it was shown for free groups generated by a finite set of generators \(F\) that the entropy \(h_p:=\lim_n - {1/n}\sum_0^n p^{(n)}(x) \ln p^{(n)}(x)\) depends as real analytic function on \(p\in \mathcal{P}(F)\), the set of probabilities with support \(F\). An analogous result holds for the linear drift \(\ell_p:= \lim_n {1/n}\sum_0^n p^{(n)}(x) |x|\), see [\textit{F. Ledrappier}, Groups Geom. Dyn. 6, No. 2, 317--333 (2012; Zbl 1252.60043); \textit{L. A. Gilch}, Electron. J. Probab. 16, 76--105 (2011; Zbl 1225.60126)]. Here, \(p^{(n)}\) denotes \(n\)-times convolution power and \(|\cdot|\) the word metric. On the other hand, for Gromov hyperbolic groups, it is known that \(h_p\) and \(\ell_p\) depend continuously on the probabilities \(p\) with finite first moment. In the paper under review, the author continues and extends the afore mentioned investigations to Gromov hyperbolic groups, showing in Theorem 1.1 that \(p\mapsto h_p\) and \(p\mapsto \ell_p\) are Lipschitz continuous functions on \(\mathcal{P}(F)\). (In fact, concrete examples show that \(C^1\)-dependence cannot be expected for this class of groups.) The organization of the paper follows the ideas of the author's previous investigations in the case of free groups. \(h_p\) and \(\ell_p\) are representable in terms of Martin kernels, of the exit measure \(p^\infty\) of the random walk (supported by the geometric boundary) and of the probabilities \(\alpha_x^V(v)\) that the first visit of \(V\) occurs in \(v\) (for neighbourhoods \(V\) and \(v\in V\)). Regularity properties of these terms are investigated separately, and the proof of the main result, Theorem 1.1, relies thereon.