Point-centered domain decomposition for parallel molecular dynamics simulation (Q1971523)

Molecular dynamics (MD) simulation can throw light upon subnanosecond phenomena and processes in biological macromolecules. They require large amounts of computation. The main result of the paper is a new point-centered domain decomposition method for molecular dynamics simulation on parallel computers. The method is based on Voronoi-like domain decomposition technique: cells (or diagrams) of \textit{G. Voronoï} [J. Reine Angew. Math. 134, 198-287 (1908; JFM 39.0274.01)]. On the base of this idea the notion of a cluster of atoms is defined. It is one of the main ingredients of the method. Each cluster is described by a point in space, called its ''center''. Atoms belong to the cluster whose center is closest to the atom. The initial distribution of atoms into clusters is obtained with an iterative algorithm. The authors carefully describe the parallelization strategy. The description includes goals of the strategy, atom distribution, pair list construction, molecular dynamics algorithm, dynamic load balancing and energy minimization. Both vectoring and parallelization on computers with many standard CPUs are used. The method and algorithm are implemented in the new program Opalp using a standard message passing interface [Message Passing Interface Forum, Int. J. Supercomputing Appl. 8, No. 1 (1994)]. The point-centered domain decomposition algorithm was tested in a free MD simulation of the protein cyclophilin A, starting from the structure that has been determined by NMR in the work of \textit{M. Ottiger, O. Zerbe, P. Güntert} and \textit{K. Wüthrich} [J. Mol. Biol. 272, No. 64 (1997)]. The entire system consisted of 16708 atoms. 300 steps of energy minimization were performed. Authors tested the performance of the algorithm on the vector computer Cray J90 and on the massively parallel machine Cray T3D and gave two tables with benchmark results for the computers.