DSMC: A Statistical Mechanics Perspective

TL;DR

This paper reinterprets DSMC as a statistical mechanics tool, using analytical methods to reveal its stochastic nature as beneficial for modeling physical phenomena and guiding machine learning applications.

Contribution

It introduces a novel perspective on DSMC, applying statistical mechanics techniques to analyze its properties and highlighting its usefulness in simulating physical fluctuations.

Findings

DSMC's stochastic noise can model Brownian motion.

Analytical techniques reveal DSMC's connection to thermodynamic fluctuations.

Statistical mechanics principles can guide machine learning applications in DSMC.

Abstract

This paper presents a perspective in which Direct Simulation Monte Carlo (DSMC) is viewed not in its traditional role as an algorithm for solving the Boltzmann equation but as a numerical method for statistical mechanics. First, analytical techniques such as the collision virial and Green-Kubo relations, commonly used in molecular dynamics, are used to study the numerical properties of the DSMC algorithm. The stochastic aspect of DSMC, which is often viewed as unwanted numerical noise, is shown to be a useful feature for problems in statistical physics, such as Brownian motion and thermodynamic fluctuations. Finally, it is argued that fundamental results from statistical mechanics can provide guardrails when applying machine learning to DSMC.