← All papers
First page of Conditioned stochastic stability of equilibrium states on uniformly hyperbolic sets

Conditioned stochastic stability of equilibrium states on uniformly hyperbolic sets

Bernat Bassols Cornudella, Matheus M. Castro

math.DS Jun 18, 2025 · v3 math.PR
Main theorems, the anisotropic Banach space construction, and the cited spectral results are formalised in Lean 4 with Mathlib, sorry-free, with the formalisation produced by AI models.
We establish the conditioned stochastic stability of equilibrium states for Hölder potentials on uniformly hyperbolic sets. While standard stochastic stability characterises measures on attractors, we analyse the statistics of transient dynamics on non-attracting sets by conditioning small random perturbations of the dynamics to not escape from our regions of interest. We prove that as the noise intensity vanishes, the quasi-ergodic measure of the $e^φ$-weighted process generated by $\e$-small random perturbations of the deterministic dynamics converges to the unique equilibrium state associated with the potential $φ- \log \left|\det \left. D T\right|_{E^u}\right|$. The results are obtained via perturbative spectral analysis of transfer operators acting on anisotropic Banach spaces and topological hyperbolic dynamics arguments. Furthermore, we extend this framework globally to Axiom A diffeomorphisms with multiple basic sets using dynamical filtrations. This work provides a rigorous characterisation of natural measures on uniformly hyperbolic repellers, which are fundamental in the context of transient chaos. Our results have been formalised in Lean 4.

Standard stochastic stability selects invariant measures on attractors, so it cannot characterise statistics of transient dynamics on non-attracting (repelling) uniformly hyperbolic sets. The goal is to identify which equilibrium states are selected when small random perturbations are conditioned not to escape a region.

Random perturbations are conditioned on survival in an isolating neighbourhood, and the resulting quasi-ergodic measures are studied. The main tools are perturbative spectral analysis of transfer operators on Baladi's anisotropic Banach spaces, Keller–Liverani perturbation theory, and topological hyperbolic dynamics arguments. The framework is extended to Axiom A diffeomorphisms with multiple basic sets via dynamical filtrations. All main theorems and the cited supporting results are formalised in Lean 4 with Mathlib, and the formalisation is attributed to AI models.

As the noise vanishes, the quasi-ergodic measure of the e^φ-weighted perturbed process converges to the unique equilibrium state of φ − log|det DT|_{E^u}|. The Lean development contains no sorry placeholders and depends only on the standard axioms propext, Classical.choice and Quot.sound.