Integral representation of Lyapunov exponents

TL;DR

This paper introduces an operator-theoretic variational principle for Lyapunov exponents, extending classical formulas and providing new asymptotic representations for random linear systems.

Contribution

It develops a novel variational framework for Lyapunov exponents applicable to Markov-driven systems, including singular cocycles and conditional growth analysis.

Findings

The growth rate equals the supremum over invariant lifts for each measure.

The principle extends classical formulas to include singular cocycles.

Pointwise Lyapunov exponents depend only on current noise state and initial position.

Abstract

We develop an abstract operator-theoretic variational principle for asymptotic growth rates arising from subadditive processes driven by Markov operators: for each invariant measure on the base, the growth rate equals the supremum of fiber integrals over invariant lifts to the bundle, and this supremum is attained on an ergodic lift. Applied to (random) linear bundle morphisms, the principle extends the classical projective formulas for sums of Lyapunov exponents, including singular cocycles, and yields new asymptotic representations in terms of conditional annealed growth along individual directions. As an application, we prove that for random linear bundle morphisms driven by Markovian place-dependent noise, the pointwise Lyapunov exponents depend only on the current noise state and initial position, not on the full noise realization.