Three central limit theorems for the unbounded excursion component of a Gaussian field
Michael McAuley
TL;DR
The work analyzes the geometry of the unbounded excursion component $\{f\ge\ell\}_{\infty}$ for smooth, stationary Gaussian fields $f$ with fast-decaying covariance, proving law of large numbers and central limit theorems for three global statistics—volume, surface area, and Euler characteristic—restricted to growing boxes. The authors develop a white-noise-based CLT for nonlocal functionals, combine it with a Morse-theoretic stability framework to control perturbations, and derive explicit (semi-explicit) formulas for limiting variances, establishing positivity under suitable conditions. In the planar case, the results hold for all supercritical levels $\ell<\ell_c$, with higher dimensions requiring stronger decay and a truncated-connection decay assumption; a finitary CLT is also obtained for practical observations in finite domains. The methods highlight the efficacy of martingale approaches for nonlocal geometric functionals of Gaussian fields and lay groundwork for extensions to broader classes of fields and topological statistics.
Abstract
For a smooth, stationary Gaussian field $f$ on Euclidean space with fast correlation decay, there is a critical level $\ell_c$ such that the excursion set $\{f\geq\ell\}$ contains a (unique) unbounded component if and only if $\ell<\ell_c$. We prove central limit theorems for the volume, surface area and Euler characteristic of this unbounded component restricted to a growing box. For planar fields, the results hold at all supercritical levels (i.e. all $\ell<\ell_c$). In higher dimensions the results hold at all sufficiently low levels (all $\ell<-\ell_c<\ell_c$) but could be extended to all supercritical levels by proving the decay of truncated connection probabilities. Our proof is based on the martingale central limit theorem.
