Parallel Fast Multipole Method For Molecular Dynamics

摘要

We report on a parallel version of the Fast Multipole Method (FMM) implemented in the classical molecular dynamics code, NAMD (Not Another Molecular Dynamics program). This novel implementation of FMM aims to minimize interprocessor communication through the modification of the FMM grid to match the hybrid force and spatial decomposition scheme already present in NAMD. This new implementation has the benefit of replacing all-to-all communications broadcasts with direct communications between nearest neighbors. This results in a significant reduction in the amount of communication compared to earlier attempts to integrate FMM into common molecular dynamics programs. The early performance of FMM is similar to the existing electrostatics methods already in NAMD. In addition, tests of the stability and accuracy of the FMM algorithm in molecular dynamics as applied to several common solvated protein structures are discussed.