← All papers
First page of Finite-order exponential mixing does not imply the central limit theorem for unique equilibrium states

Finite-order exponential mixing does not imply the central limit theorem for unique equilibrium states

Matheus M Castro

math.DS Nov 16, 2026 · v1
Main theorems (A, 7.3, 8.5, Corollary 1.2) on unique equilibrium states failing the CLT are formalised in Lean 4 using Mathlib, with AI assistance.
Let $M$ be a compact smooth manifold of dimension at least three, $\varphi:M\to\mathbb R$ be continuous, and $R\ge2$ be an integer. We construct a $\mathcal C^\infty$ diffeomorphism $T:M\to M$ and an observable $f\in\mathcal C^\infty(M,\mathbb R)$ such that $(T,\varphi)$ has a unique equilibrium state $μ$ and $μ(f)=0$. The system $(T,μ)$ is exponentially mixing of order $R$ but not of order $R+1$ for Hölder observables, and $n^{-1/2}\sum_{i=0}^{n-1}f\circ T^i$ does not satisfy the central limit theorem. The main results have been formalised in Lean 4.

Whether exponential mixing of finite order implies the central limit theorem for Hölder observables. The question is posed here for unique equilibrium states of smooth diffeomorphisms. Björklund–Gorodnik showed that mixing of all orders suffices.

A symbolic subshift Y is built by adding auxiliary coordinates to a (2R+1)-tuplewise independent process of Bassols Cornudella and Castro, using recursively constructed block measures based on the Hoeffding–Efron–Stein decomposition. Exact independence of distant blocks gives order-R mixing, and an entropy gap gives uniqueness of the equilibrium state. Y is then realised inside a smooth diffeomorphism of any manifold of dimension at least 3 via a horseshoe and the Lin–Tian isolation construction. The main results were formalised in Lean 4 with Mathlib, with AI assistance.

For every R≥2 and continuous potential, there is a C^∞ diffeomorphism with a unique equilibrium state that is exponentially mixing of order R but not R+1, and a smooth observable for which the CLT fails. The main theorems are checked by Lean's kernel.