Averaging of Hamiltonian flows with an ergodic component

TL;DR

This paper studies the limiting behavior of a stochastic process on a torus with an ergodic component, showing convergence to a Markov process that reflects the flow's ergodic structure under certain conditions.

Contribution

It extends previous results by analyzing non-periodic stream functions with ergodic components, proving convergence to a Markov process with positive time in the ergodic region.

Findings

Process converges to a Markov process on the graph

Markov process spends positive time in ergodic component

Convergence holds for Diophantine rotation numbers

Abstract

We consider a process on $\mathbb{T}^2$, which consists of fast motion along the stream lines of an incompressible periodic vector field perturbed by white noise. It gives rise to a process on the graph naturally associated to the structure of the stream lines of the unperturbed flow. It has been shown by Freidlin and Wentzell [Random Perturbations of Dynamical Systems, 2nd ed. Springer, New York (1998)] and [Mem. Amer. Math. Soc. 109 (1994)] that if the stream function of the flow is periodic, then the corresponding process on the graph weakly converges to a Markov process. We consider the situation where the stream function is not periodic, and the flow (when considered on the torus) has an ergodic component of positive measure. We show that if the rotation number is Diophantine, then the process on the graph still converges to a Markov process, which spends a positive proportion of time in the vertex corresponding to the ergodic component of the flow.