Gradient Gibbs measures and fuzzy transformations on trees

Abstract

We study Gibbsian models of unbounded integer-valued spins on trees which possess a symmetry under height-shift. We develop a theory relating boundary laws to gradient Gibbs measures, which applies also in cases where the corresponding Gibbs measures do not exist. Our results extend the classical theory of Zachary beyond the case of normalizable boundary laws, which implies existence of Gibbs measures, to periodic boundary laws. We provide a construction for classes of tree-automorphism invariant gradient Gibbs measures in terms of mixtures of pinned measures, whose marginals to infinite paths on the tree are random walks in a q-periodic environment. Here the mixture measure is the invariant measure of a finite state Markov chain which arises as a mod-q fuzzy transform, and which governs the correlation decay. The construction applies for example to SOS-models and discrete Gaussian models and delivers a large number of gradient Gibbs measures. We also discuss relations of certain gradient Gibbs measures to Potts and Ising models.