Large deviations for random walks on Galton-Watson trees: Averaging and uncertainty

In the study of large deviations for random walks in random environment, a key distinction has emerged between quenched asymptotics, conditional on the environment, and annealed asymptotics, obtained from averaging over environments. In this paper we consider a simple random walk {X_n} on a Galton-Watson tree T, i.e., on the family tree arising from a supercritical branching process. Denote by |X_n| the distance between the node X_n and the root of T. Our main result is the almost sure equality of the large deviation rate function for |X_n|/n under the "quenched measure" (conditional upon T), and the rate function for the same ratio under the "annealed measure" (averaging on T according to the Galton-Watson distribution). This equality hinges on a concentration of measure phenomenon for the momentum of the walk. (The momentum at level n, for a specific tree T, is the average, over random walk paths, of the forward drift at the hitting point of that level). This concentration, or certainty, is a consequence of the uncertainty in the location of the hitting point. We also obtain similar results when {X_n} is a λ-biased walk on a Galton-Watson tree, even though in that case there is no known formula for the asymptotic speed. Our arguments rely at several points on a "ubiquity" lemma for Galton-Watson trees, due to Grimmett and Kesten (1984).