Diffusions and random walks with prescribed sub-Gaussian heat kernel estimates

TL;DR

This paper characterizes when metric measure spaces and diffusions have sub-Gaussian heat kernel estimates with given volume and escape time profiles, constructing new examples of diffusions and graphs with these properties.

Contribution

It provides necessary and sufficient conditions for sub-Gaussian heat kernel estimates and introduces a new family of diffusions and graphs with prescribed profiles.

Findings

Constructed diffusions with martingale dimension one and high spectral dimension.

Developed a framework for existence of spaces with specific heat kernel estimates.

Generalized earlier results on random walks on graphs with sub-Gaussian behavior.

Abstract

Given suitable functions $V, \Psi:[0,\infty) \to [0,\infty)$, we obtain necessary and sufficient conditions on $V,\Psi$ for the existence of a metric measure space and a symmetric diffusion process that satisfies sub-Gaussian heat kernel estimates with volume growth profile $V$ and escape time profile $\Psi$. We prove sufficiency by constructing a new family of diffusions. Special cases of this construction also leads to a new family of infinite graphs whose simple random walks satisfy sub-Gaussian heat kernel estimates with prescribed volume growth and escape time profiles. In particular, these random walks on graphs generalizes earlier results of Barlow who considered the case $V(r)=r^\alpha$ and $\Psi(r)=r^\beta$ (Rev Mat Iberoam 2004). The family of diffusions we construct have martingale dimension one but can have arbitrarily high spectral dimension. Therefore our construction shows the impossibility of obtaining non-trivial lower bounds on martingale dimension in terms of spectral dimension which is in contrast with upper bounds on martingale dimension using spectral dimension obtained by Hino (Probab Theory Relat Fields 2013).