Escape Rates in a Stochastic Environment with Multiple Scales

TL;DR

This paper develops and compares two stochastic reduction methods for multi-scale systems to accurately predict escape rates from basins of attraction, validated through numerical simulations on a nonlinear oscillator.

Contribution

It introduces a general theory comparing deterministic and stochastic center manifold reduction methods for escape rate prediction in multi-scale stochastic systems.

Findings

Both reduction methods provide similar escape rate scaling.

Center manifolds predict high probability escape regions.

Numerical simulations confirm theoretical predictions.

Abstract

We consider a stochastic environment with two time scales and outline a general theory that compares two methods to reduce the dimension of the original system. The first method involves the computation of the underlying deterministic center manifold followed by a naive replacement of the stochastic term. The second method allows one to more accurately describe the stochastic effects and involves the derivation of a normal form coordinate transform that is used to find the stochastic center manifold. The results of both methods are used along with the path integral formalism of large fluctuation theory to predict the escape rate from one basin of attraction to another. The general theory is applied to the example of a surface flow described by a generic, singularly perturbed, damped, nonlinear oscillator with additive, Gaussian noise. We show how both nonlinear reduction methods compare in escape rate scaling. Additionally, the center manifolds are shown to predict high pre-history probability regions of escape. The theoretical results are confirmed using numerical computation of the mean escape time and escape prehistory, and we briefly discuss the extension of the theory to stochastic control.