Localization for linearly edge reinforced random walks (Q2450255)

The linearly edge reinforced random walk (LRRW) is a model of a self-interacting random walk which is defined as follows. Each edge \(e\) of a graph \(G\) has an initial weight \(a_e>0\). A starting vertex \(v_0\) is given. The walker starts at \(v_0\), examines the weights on the edges around it, normalizes them to be probabilities, and then chooses an edge with these probabilities. The weight of the edge traversed is then increased by 1 (the edge is ``reinforced''). The process then repeats with the new weights. In the present paper, it is proved that the LRRW on any graph with bounded degrees is recurrent for sufficiently small initial weights. For non-amenable graphs, the LRRW is transient for sufficiently large initial weights. Also, the existence of a transient regime is shown in the case of non-amenable graphs. Similar results are derived for the vertex reinforced jump process.