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

TL;DR

This paper constructs particle models approximating a class of Dean-Kawasaki equations with logarithmic interactions, analyzing their convergence, regularity, and entropy decay depending on inverse temperature thresholds.

Contribution

It introduces a spectral approximation to the noise in Dean-Kawasaki equations and establishes convergence rates and phase transitions based on temperature.

Findings

Particle models converge to solutions of the Dean-Kawasaki equations.

Higher inverse temperature leads to more regular solutions.

Entropy of solutions decays exponentially at high temperatures.

Abstract

We consider 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 $\mathbb{P}(\mathbb{T})$. 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 this class of equations. Specifically, we give a quantitative convergence rate of the particle approximation, which allows us to identify a unique limit distribution depending on $\beta$. As the inverse temperature rises, the regularizing effect of repulsive interactions becomes stronger. We prove that there exists three thresholds $0<\lambda_0\leq\lambda_1<\lambda_2$ 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 R\'enyi entropy of $p_t^\beta$ satisfies an exponential decay estimate.