Simulation of Noncircular Rigid Bodies: Machine Learning Based Overlap Calculation Technique with System Size Independent Computational Cost

TL;DR

This paper introduces a machine learning-based method for simulating noncircular rigid bodies in two dimensions, significantly reducing computational costs while maintaining accurate structural features, thus enabling more efficient non-spherical particle simulations.

Contribution

The paper presents a novel ML model for overlap prediction that is independent of the number of constituent disks, speeding up non-spherical body simulations in Monte Carlo methods.

Findings

Significant speed-up in MC simulations using the ML model.

Structural features of the system are preserved with the ML approach.

The ML model's computational cost remains constant regardless of constituent disk number.

Abstract

Standard molecular dynamics (MD) and Monte Carlo (MC) simulation deals with spherical particles. Extending these standard simulation methodologies to the non-spherical cases is non-trivial. To circumvent this problem, non-spherical bodies are considered as a collection of constituent spherical objects. As the number of these constituent objects becomes large, the computational burden to simulate the system also increases. In this article, we propose an alternative way to simulate non-circular rigid bodies in two dimensions having pairwise repulsive interactions. Our approach is based on a machine learning (ML) based model which predicts the overlap between two non-circular bodies. The machine learning model is easy to train and the computation cost of its implementation remains independent of the number of constituents disks used to represent a non-circular rigid body. When used in MC simulation, our approach provides significant speed up in comparison to the standard implementation where overlap determination between two rigid bodies is done by calculating the distance of their constituent disks. Our proposed ML based MC method provided very similar structural features (in comparison to standard implementation) of the systems. We believe this work is a very first step towards a time-efficient simulation of non-spherical rigid bodies.