Quantitative Marcinkiewicz's theorem and central limit theorems: applications to spin systems and point processes

Abstract: The classical Marcinkiewicz theorem states that if an entire characteristic function $\Psi$ of a non-degenerate real-valued random variable $X$ is of the form $\exp(P(u))$ for some polynomial $P$, then $X$ has to be a Gaussian. In this work, we obtain a broad, quantitative extension of this framework in several directions, establish central limit theorems (CLTs) with explicit rates of convergence, and demonstrate Gaussian fluctuations in continuous spin systems and general classes of point processes. In particular, we obtain quantitative decay estimates on the Kolmogorov-Smirnov distance between $X$ and a Gaussian under the condition that $\Psi$ does not vanish only on a bounded disk. This leads to quantitative CLTs applicable to very general and possibly strongly dependent random systems. In spite of the general applicability, our rates for the CLT match the classic Berry-Esseen bounds for independent sums up to a log factor. We implement this programme for two important classes of models in probability and statistical physics. First, we extend to the setting of continuous spins a popular paradigm for obtaining CLTs for discrete spin systems that is based on the theory of Lee-Yang zeros, focussing in particular on the XY model, Heisenberg ferromagnets and generalised Ising models. Secondly, we establish Gaussian fluctuations for linear statistics of so-called $\alpha$-determinantal processes for $\alpha \in \mathbb{R}$ (including the usual determinantal, Poisson and permanental processes) under very general conditions, including in particular higher dimensional settings where structural alternatives such as random matrix techniques are not available. Our applications demonstrate the significance of having to control the characteristic function only on a (small) disk, and lead to CLTs which, to the best of our knowledge, are not known in generality.