Classical and Quantum Molecular Dynamics Simulation on Distributed-Memory Massively Parallel Computers

Abstract

The implementations of classical and quantum molecular dynamics simulations on distributed-memory massively parallel computers are presented. First, we discuss the implementation of large-scale classical molecular dynamics (MD) simulations on SIMD architecture parallel computers, and in particular, on the family of MasPar distributed-memory data parallel computers. We describe methods of mapping the problem onto the Processing Elements (PE's) of the SIMD architecture, and assess the performance of each strategy. The detailed implementations of this data parallel construct are illustrated for two case studies: classical MD simulation of a two-dimensional lattice and the photodissociation mechanisms of a diatomic iodine impurity in a three-dimensional argon lattice. We also present a study of quantum dynamics using the Time-Dependent Self-Consistent Field (TDSCF) method. These calculations demonstrate the potential of using massively parallel computers in MD simulations of thermodynamic properties and chemical reaction dynamics in condensed phases.