Harnack inequalities and Gaussian estimates for random walks on metric measure spaces

From MaRDI portal
(Redirected from Publication:6164929)



Abstract: We characterize Gaussian estimates for transition probability of a discrete time Markov chain in terms of geometric properties of the underlying state space. In particular, we show that the following are equivalent: (1) Two sided Gaussian bounds on heat kernel (2) A scale invariant Parabolic Harnack inequality (3) Volume doubling property and a scale invariant Poincar'{e} inequality. The underlying state space is a metric measure space, a setting that includes both manifolds and graphs as special cases. An important feature of our work is that our techniques are robust to small perturbations of the underlying space and the Markov kernel. In particular, we show the stability of the above properties under quasi-isometries. We discuss various applications and examples.


Consider a discrete time Markov chain on a metric measure space. Motivated by classical results for diffusions on Riemannian manifolds, the authors want to study the relationships (implication, equivalence) between bounds on the heat kernel (Gaussian bounds, parabolic Harnack inequality) and the geometric properties of the metric space (doubling property, Poincaré inequality). \par For the space, the authors want to include both Riemannian manifolds and discrete spaces such as graphs. The notion of Riemannian manifold can be generalised to the notion of length space, and in order to also include discrete spaces, they choose for their framework the notion of quasi-geodesic space. With this framework they are able to relate properties of the Markov chain and properties of the space. \par Moreover, an important benefit of their technique is the robustness of the results with respect to perturbations on the space and the Markov kernel.



Cites work









This page was built for publication: Harnack inequalities and Gaussian estimates for random walks on metric measure spaces

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6164929)