Premium
A heterogeneous space–time full approximation storage multilevel method for molecular dynamics simulations
Author(s) -
Waisman Haim,
Fish Jacob
Publication year - 2007
Publication title -
international journal for numerical methods in engineering
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.421
H-Index - 168
eISSN - 1097-0207
pISSN - 0029-5981
DOI - 10.1002/nme.2078
Subject(s) - iterated function , smoothing , stability (learning theory) , space time , coupling (piping) , mathematics , computer science , waveform , algorithm , space (punctuation) , mathematical optimization , mathematical analysis , mechanical engineering , telecommunications , radar , machine learning , chemical engineering , engineering , computer vision , operating system
A heterogeneous space–time full approximation storage (HFAS) multilevel formulation for molecular dynamics simulations is developed. The method consists of a waveform Newton smoothing that produces initial space–time iterates and a coarse model correction. The formulation is coined as heterogeneous since it permits different interatomic potentials to be applied at different physical scales. This results in a flexible framework for physics coupling. Time integration is performed in windows using the implicit Newmark predictor–corrector method that permits larger time integration steps than the explicit method. The size of the time steps is governed by accuracy rather than by stability considerations of the algorithm. We study three different variants of the method: the Picard iteration, constrained dynamics and force splitting. Numerical examples show that FAS based on force splitting provides significant time savings compared to standard explicit methods and alternative implicit space–time schemes. Parallel studies of the Picard iteration on harmonic problems illustrate the time parallelization effect that leads to a superior parallel performance compared to explicit methods. Copyright © 2007 John Wiley & Sons, Ltd.