Transience of edge-reinforced random walk (Q2356047)

The authors consider a nonoriented connected locally finite graph without loops, \(G=(V,E,\sim)\), and a random process \((X_n)_{n\in\mathbb{N}}\) defined on some probability space with values in \(V\). This process \((X_n)_{n\geq0}\) is called an edge-reinforced random walk with starting point \(i_0\in V\) and positive weights \((a_e)_{e\in E}\) if \(X_0=i_0\) and, for all \(n\in \mathbb{N}\), \[ \mathbb{P}(X_{n+1}=j|\sigma(X_0,\dots,X_n))=\mathbf{1}_{\{j\sim X_n\}}\frac{Z_n(\{X_n,j\})}{\sum_{k\sim X_n}Z_n(\{X_n,k\})}, \] where \(Z_n(e)\), \(e\in V\), is the sum of the initial weight \(a_e\) and the number of crosses of \(e\) by the process \((X_n)_{n\in\mathbb{N}}\) up to time \(n\). The ERRW process, introduced by \textit{D. Coppersmith} and \textit{P. Diaconis} in [``Random walks with reinforcement'', unpublished manuscript (1986)], was conjectured to exhibit a phase transition between recurrent and transient behavior in dimension \(d\geq3\). \textit{C. Sabot} and \textit{P. Tarrès} [J. Eur. Math. Soc. (JEMS) 17, No. 9, 2353--2378 (2015; Zbl 1331.60185)] have proved an almost sure positive recurrence of ERRW on any graph of bounded degree for any dimension \(d\geq3\) and large reinforcement, i.e., if \(a_e<\tilde{a}_c\) for all \(e\in V\) and some \(\tilde{a}_c>0\) depending on that degree. This paper completes the solution of an open problem stated by P.\,Diaconis in 1986 by proving an almost sure transience of the ERRW in dimension \(d\geq3\) under small reinforcement, i.e., if \(a_e>a_c(d)\) for all \(e\in V\) and some \(a_c(d)>0\).