Asymptotics for the second-largest Lyapunov exponent for some Perron-Frobenius operator cocycles

TL;DR

This paper establishes bounds on the second-largest Lyapunov exponent for certain cocycles of Perron-Frobenius operators, providing new insights into mixing rates and decay of correlations in random dynamical systems.

Contribution

It proves a generalized Perron-Frobenius theorem for operator cocycles and applies it to bound the second Lyapunov exponent in parametrized families of maps.

Findings

Upper bounds on $$ for Perron-Frobenius cocycles.

Asymptotic linearity of $$ in perturbation scale.

Sharp estimates showing $$ approaches -2 times the scale parameter.

Abstract

Given a discrete-time random dynamical system represented by a cocycle of non-singular measurable maps, we may obtain information on dynamical quantities by studying the cocycle of Perron-Frobenius operators associated to the maps. Of particular interest is the second-largest Lyapunov exponent, $\lambda_2$, which can tell us about mixing rates and decay of correlations in the system. We prove a generalized Perron-Frobenius theorem for cocycles of bounded linear operators on Banach spaces that preserve and occasionally contract a cone; this theorem shows that the top Oseledets space for the cocycle is one-dimensional, and there is an readily computed lower bound for the gap between the largest Lyapunov exponent $\lambda_1$ and $\lambda_2$ (that is, an upper bound for $\lambda_2$ which is strictly less than $\lambda_1$). We then apply this theorem to the case of cocycles of Perron-Frobenius operators arising from a parametrized family of maps to obtain an upper bound on $\lambda_2$; to the best of our knowledge, this is the first time $\lambda_2$ has been upper-bounded for a family of maps. To do this, we utilize a new balanced Lasota-Yorke inequality. We also examine random perturbations of a fixed map with two invariant densities and show that as the perturbation is scaled back down to the unperturbed map, $\lambda_2$ is asymptotically linear in the scale parameter. Our estimates are sharp, in the sense that there is a sequence of scaled perturbations of the fixed map that are all Markov, such that $\lambda_2$ is asymptotic to $-2$ times the scale parameter.