A Scalable, Linear-Time Dynamic Cutoff Algorithm for Molecular Simulations of Interfacial Systems

TL;DR

This thesis presents a scalable, linear-time dynamic cutoff algorithm for molecular dynamics simulations that efficiently handles interfacial systems, improving performance and scalability on large parallel architectures.

Contribution

It introduces a real-time, parallelizable interface detection method and integrates it with a dynamic cutoff algorithm, enhancing large-scale molecular simulations.

Findings

Linear-time complexity and nearly ideal scaling demonstrated.

Comparable accuracy to traditional PPPM algorithm.

Superior performance for large systems due to FFT limitations in PPPM.

Abstract

This master thesis introduces the idea of dynamic cutoffs in molecular dynamics simulations, based on the distance between particles and the interface, and presents a solution for detecting interfaces in real-time. Our dynamic cutoff method (DCM) exhibits a linear-time complexity as well as nearly ideal weak and strong scaling. The DCM is tailored for massively parallel architectures and for large interfacial systems with millions of particles. We implemented the DCM as part of the LAMMPS open-source molecular dynamics package and demonstrate the nearly ideal weak- and strong-scaling behavior of this method on an IBM BlueGene/Q supercomputer. Our results for a liquid/vapor system consisting of Lennard-Jones particles show that the accuracy of DCM is comparable to that of the traditional particle-particle particle- mesh (PPPM) algorithm. The performance comparison indicates that DCM is preferable for large systems due to the limited scaling of FFTs within the PPPM algorithm. Moreover, the DCM requires the interface to be identified every other MD timestep. As a consequence, this thesis also presents an interface detection method which is (1) applicable in real time; (2) parallelizable; and (3) scales linearly with respect to the number of particles.