A characterization of Markov chains on infinite graphs by limiting distributions (Q1924569)

An irreducible transient Markov chain \(X=(X_n,P_x)\) valued on V, the set of vertices of a locally finite, infinite and connected graph \(\Gamma =(V,E)\) with E the set of edges, unoriented and without loops, is considered. Assume that the positivity of transition probability \(p(x,y)\) implies \([x,y]\in E.\) Then the limiting variable \(X_{\infty}\) takes value in the Martin boundary \(\Omega\) of the chain X with law \(\nu_x\) under \( P_x\). The following problem is studied: When does \(\{\nu_x:x\in V\}\) uniquely determine \(p(x,y)\)? The main result is: Under the additional regularity assumption of supp\((\nu_x)=\Omega\) for any \( x\in V\), the family of limiting distributions \(\{\nu_x:\) \(x\in V\}\) uniquely determines transition probabilites of X iff all vertices are cutvertices. Applications on random walks on free products of finite groups are also discussed. Examples are provided.