|About this Abstract
||2018 TMS Annual Meeting & Exhibition
||Algorithm Development in Materials Science and Engineering
||Parallel Algorithms for Hyperdynamics in LAMMPS
||Steve Plimpton, Danny Perez, Art Voter
|On-Site Speaker (Planned)
Hyperdynamics (HD) is a method used in molecular dynamics (MD) for performing time-accurate accelerated MD, typically for solids where relatively rare events trigger transitions between potential energy basins. In bond-boost HD, transitions are accelerated by adding a bias potential to pairs of atoms, which can increase simulated time by many orders-of-magnitude. In global HD only one pair of atoms is biased at a time; in local HD multiple pairs can be biased at the same time, if separated by a cutoff distance. We have implemented both HD flavors in the LAMMPS MD code, which enables parallel HD simulations using any of the interatomic potentials LAMMPS supports. I'll describe the algorithmic ideas we use for efficient parallelization of HD and give performance and scalability results on large machines. I'll also present a large-scale surface diffusion model we've simulated via HD.
||Planned: Supplemental Proceedings volume