Mather measures are probability measures supported on orbits of a Lagrangian system that minimize the action. They generalize KAM tori: when a KAM torus exists, the Mather measure sits on it, but Mather measures also exist when KAM tori have broken up. They are the dynamical system's way of remembering ordered motion even after the structures supporting it have been destroyed.
Note (2026-11-12): checked against the paper's full text (Sorrentino, Zhang and Zhu). The headline matches: for a perturbed Tonelli Lagrangian whose unperturbed Mather measure sits on a quasi-periodic torus with a Diophantine frequency, the perturbed measure is Hoelder continuous with an exponent that depends on the Diophantine index. The paper's statement is an upper bound on the rate, in specific classes of perturbation (the first is Mane's Lagrangian with a Lipschitz perturbation of a constant vector field, the frequency being finite Diophantine or algebraic), with a lower bound proved only in two dimensions and a gap remaining between them; for more than two dimensions it says a general lower bound is unattainable and discusses special arithmetic conditions. It also discusses the possibility of Lipschitz regularity through KAM theory. So 'better Diophantine properties give better regularity' and 'robustly or fragilely stable depending on how rational the frequency is' describe the guaranteed exponent, not a sharp dependence of the actual stability.
Perturb the Lagrangian and the Mather measure moves. The question is how smoothly: does a small perturbation produce a small change in the measure, or can the measure jump discontinuously?
For Mather measures on quasi-periodic tori with Diophantine frequency, the answer depends on the arithmetic of the frequency (arXiv:2603.11576). The perturbed measure is Hölder continuous in the perturbation parameter, and the Hölder exponent is determined by the Diophantine properties of the underlying frequency. Better Diophantine properties — frequencies that are harder to approximate by rationals — give better regularity. Worse Diophantine properties give weaker regularity.
The structural observation: the stability of a dynamical orbit under perturbation is controlled by a number-theoretic property of its frequency. The same orbit can be robustly or fragily stable depending on how rational its frequency is, and this arithmetic property is entirely extrinsic to the mechanics of the system. The dynamics doesn't know about number theory — but the number theory determines how the dynamics responds to perturbation. The arithmetic of the orbit's label controls the analysis of its persistence.