Hamiltonian perturbation theory for ultra-differentiable functions

TL;DR

This paper develops a framework of ultra-differentiable function spaces to extend Hamiltonian perturbation theory, proving invariant torus and Nekhoroshev stability theorems under broader conditions, and analyzing stability thresholds and destruction of tori.

Contribution

Introduces new ultra-differentiable function spaces with stability properties, extending Hamiltonian perturbation results beyond analytic functions, and provides criteria for stability and instability in this setting.

Findings

Proves invariant torus theorem for ultra-differentiable functions under BR M condition.

Establishes Nekhoroshev stability estimates depending on the growth sequence M.

Constructs examples of tori destruction and diffusing orbits in ultra-differentiable classes.

Abstract

Some scales of spaces of ultra-differentiable functions are introduced, having good stability properties with respect to infinitely many derivatives and compositions. They are well-suited for solving non-linear functional equations by means of hard implicit function theorems. They comprise Gevrey functions and thus, as a limiting case, analytic functions. Using majorizing series, we manage to characterize them in terms of a real sequence M bounding the growth of derivatives. In this functional setting, we prove two fundamental results of Hamiltonian perturbation theory: the invariant torus theorem, where the invariant torus remains ultra-differentiable under the assumption that its frequency satisfies some arithmetic condition which we call BR M , and which generalizes the Bruno-R{\"u}ssmann condition ; and Nekhoroshev's theorem, where the stability time depends on the ultra-differentiable class of the pertubation, through the same sequence M. Our proof uses periodic averaging, while a substitute of the analyticity width allows us to bypass analytic smoothing. We also prove converse statements on the destruction of invariant tori and on the existence of diffusing orbits with ultra-differentiable perturbations, by respectively mimicking a construction of Bessi (in the analytic category) and Marco-Sauzin (in the Gevrey non-analytic category). When the perturbation space satisfies some additional condition (we then call it matching), we manage to narrow the gap between stability hypotheses (e.g. the BR M condition) and instability hypotheses, thus circumbscribing the stability threshold. The formulas relating the growth M of derivatives of the perturbation on the one hand, and the arithmetics of robust frequencies or the stability time on the other hand, bring light to the competition between stability properties of nearly integrable systems and the distance to integrability. Due to our method of proof using width of regularity as a regularizing parameter, these formulas are closer to optimal as the the regularity tends to analyticity.