Two Algorithms Designed for Realizing Efficient Combination of Fast Multipole Method and Dedicated Hardware for Molecular Dynamics Simulations

Abstract

This paper concerns two algorithms designed for acceleration of molecular dynamics simulations using the fast multipole method (FMM) and special-purpose machine (MD-Engine II) for calculating pair interactions. One algorithm is based on a three-dimensional cyclic buffer, which enables efficient cooperation between FMM and MD-Engine II. The algorithm is implemented as part of the particle-processor of MD-Engine II. The processor specifically extracts a set of particles which are contained in nearby regions defined in FMM. The other algorithm is based on a recursive bisection for static load-balancing technique. The efficacy of these algorithms was confirmed by numerical tests.