Hyperfinite von Neumann algebras and Poisson boundaries of time dependent random walks (Q1822735)

A right group-invariant Markov random walk on a locally compact group G is given by a sequence \((\sigma_ n)\) of probability measures on G. Let \((X_ n)\) be the canonical sequence of G-valued independent random variables with distributions \((\sigma_ n)\) on the probability space \((G^ N,\sigma)\), where \(\sigma =\prod^{\infty}_{n=1}\sigma_ n\), defined by \(X_ n((g_ j))=g_ n\). Let \(\sigma_ 0\) be a fixed probability distribution on G, the initial distribution, and on the probability space \(\Omega =(G\times G^ n,\sigma_ 0\times \sigma)\) define a Markov chain \((Y_ n)\) of random variables inductively by \(Y_ 0((g_ j))=g_ 0\), \(Y_{n+1}=Y_ nX_{n+1}\), \(n=0,1,2,... \). The group G acts on \(\Omega\) by \(g(g_ j)=(gg_ j)\). The asymptotic algebra \({\mathcal A}_{\Omega}\) of the random walk is defined by \(\cap^{\infty}_{n=1}\sigma (Y_ j,j\geq n)\) where \(\sigma (Y_ k,k\geq n)\) is the abelian \(W^*\)-subalgebra of \(L^{\infty}(\Omega)\) consisting of functions measurable with respect to the \(\sigma\)-algebra generated by \(Y_ j\), \(j\geq n\). The Poisson boundary (B,\(\mu)\) of the random walk is that G-space such that \({\mathcal A}_{\Omega}\) is isomorphic to \(L^{\infty}(B,\mu).\) The main result of the paper shows that the action of G on (B,\(\mu)\) is amenable and approximately transitive. Amenability is taken to mean that there exists a positive mapping P: \(L^{\infty}(B\times g)\to L^{\infty}(B)\) such that P is the identity on \(L^{\infty}(B)\) and, for all h in G, f in \(L^{\infty}(B\times G)\), and F in \(L^{\infty}(B)\), \(P(hf)=h(Pf)\) where \(hf(b,g)=f(bh,h^{-1}g)\) and \(hF(b)=F(bh).\) The action is approximately transitive in the sense that for each \(n<\infty\) and finite measures \(\nu_ 1,\nu_ 2,...,\nu_ n<\mu\), and \(\epsilon >0\), there exists a finite measure \(\nu <\mu,g_ 1,g_ 2,...,g_ m\in G\), and \(\lambda_{jk}\geq 0\), \(k=1,2,...,m\) such that \(\| \nu_ j-\sum^{m}_{k=1}\lambda_{jk}g_ k\nu \| \leq \epsilon\), \(j=1,2,...,n\). The authors go on to prove that when \(G={\mathbb{R}}\) or \({\mathbb{Z}}\) the converse is true. The method of proof involves various operator algebraic techniques and the observation that there is naturally associated to an eigenvalue list of a factor M which is the infinite tensor product of Type I factors a group-invariant time- dependent Markov random walk on R the Poisson boundary of which is the flow of weights for M.