Attracting edge and strongly edge reinforced walks (Q2456028)

Let \(\mathcal{G}\) be a connected unoriented graph of bounded degree. A process of one-edge-long jumps (called traversals of the respective edges between the adjacent vertices) is an edge-reinforced random walk on \(\mathcal{G}\) when the probability of a jump is subordinate to a weight function \(W(k)>0, k\in \mathbb{N}\), counting the number of previous traversals of that edge. Thus arising discrete time Markov chains were studied in various cases. Under the condition (H0) \(\sum_{k}W(k)^{-1}<\infty\), \textit{B. Davies} [Probab. Theory Rel. Fields 84, 203--229 (1990; Zbl 0665.60077)] proved for \(\mathcal{G}=\mathbb{Z}\) that almost surely the walk visits only some random points \(i,i+1\) at all large times. \textit{T. Sellke} [Technical Report 94--26, Purdue Univ.] extended this result to any \(\mathbb{Z}^d\), \(d>1\), and conjectured that there exists a.s. some random attracting edge, traversed at all large times, for general \(\mathcal{G}\) of bounded degree. Using the techniques developed in separate papers, the authors prove the conjecture for a class of weights meeting (H0) and an additional hypothesis (H1). Consequently it suffices to add the above condition on \(W\).