Wasserstein-p Bounds via Cumulant-Based Edgeworth Expansion for $α$-Mixing Random Fields

Tianle Liu,Morgane Austern

2024-09-22

Abstract:There has been considerable recent progress in establishing normal approximation bounds in terms of the Wasserstein-$p$ distance for \emph{i.i.d.} or locally dependent random variables. However, for $p>1$ no such result has previously been shown for the case of dependent variables with $\alpha$-mixing conditions. In this paper, we generalize the Wasserstein-$p$ bounds to work with $\alpha$-mixing random fields. We prove that under proper conditions the rescaled average over random fields can converge to the standard normal in the Wasserstein-$p$ distance at the rate of $\mathcal O(|T|^{-\beta})$, where $\lvert T\rvert$ is the size of the index set, and $\beta\in (0, 1/2]$ depends on $p$, the dimension $d$ of random fields, and the decay rate of the $\alpha$-mixing coefficients. In particular, $\beta=1/2$ can be achieved if the coefficients decay at a fast polynomial rate. Our results are obtained via a carefully constructed cumulant-based Edgeworth expansion and an adaptation of recent progress on Stein's method. Notably we propose a novel constructive graph approach that utilizes techniques in combinatorics to establish the desired expansion for general dependent variables.

Probability