On a particle approximation to the Dean-Kawasaki type equation with logarithmic interactions

Hao Ding

2024-12-04

Abstract:Following the idea of intrinsic constructions of Brownian motions on the Wasserstein space, we construct a class of particle models whose fluctuating hydrodynamic limits, denoted as $p_t^\beta$, are solutions to the martingale problems of a class of Dean-Kawasaki type equations on $\mathbb{T}$ with logarithmic repulsive interactions depending on the inverse temperature $\beta$ and a new spectral approximation to the noise part, which approximately features Otto's metric in $\mathbb{P}(\mathbb{T})$. Specifically, we give a quantitative convergence rate of the particle approximation using the convexity of the logarithmic interaction. This allows us to identify a unique limit distribution only depending on $\beta$. As the inverse temperature rises, the regularizing effect of repulsive interactions becomes stronger. We prove that there exists three thresholds $0<\lambda_0<\lambda_1<\lambda_2$ depending on the noise such that, when $\beta>\lambda_0$, $p_t^\beta$ is a non-atomic measure process in $\mathbb{P}(\mathbb{T})$; when $\beta>\lambda_1$, $p_t^\beta$ is absolutely continuous with respect to Lebesgue measure almost surely; when $\beta>\lambda_2$, the expectation of the Rényi entropy of $p_t^\beta$ satisfies an exponential decay estimate.

Probability