Reinforced random walks are random walks on graphs whose transition probabilities along edges from a vertex are proportional to the weights of those edges, but where the weight of an edge evolves in a way that depends on the past traversals across it. In an edge-reinforced random walk (ERRW), the weight of an edge increases by $1$ whenever that edge is traversed, in either direction. On a finite graph, an ERRW admits a remarkable representation as a random walk in a random environment. The law of the environment is given by the so-called {\em magic formula}, with this law depending on the initial edge weights. This representation provides a natural route for studying statistical properties of ERRWs. This work focuses on various information-theoretic quantities associated with ERRWs on finite graphs, motivated in part by the problem of statistically distinguishing between different ERRW models from observed trajectories. In particular, we study the entropy rate of an ERRW. We also study the Kullback--Leibler divergence (KL divergence) between two ERRW environment laws, and the KL divergence between the corresponding finite-trajectory distributions. Leveraging structural properties of the underlying random environment, we derive an annealed representation of the entropy rate, a closed-form formula for the environment-level KL divergence, and quantitative bounds on the convergence of trajectory-level KL divergence toward environment-level KL divergence. These information-theoretic quantities are motivated by the two-point hypothesis testing problem for ERRW trajectories, and in particular by the associated Stein exponent. We also expect them to play a fundamental role in the study of other testing problems for ERRWs, including identity testing and closeness testing.

翻译：暂无翻译