A new parallel molecular dynamics algorithm for organic systems
- Sandia National Labs, Albuquerque, NM (United States)
A new parallel algorithm for simulating bonded molecular systems such as polymers and proteins by molecular dynamics (MD) is presented. In contrast to methods that extract parallelism by breaking the spatial domain into sub-pieces, the new method does not require regular geometries or uniform particle densities to achieve high parallel efficiency. For very large, regular systems spatial methods are often the best choice, but in practice the new method is faster for systems with tens-of-thousands of atoms simulated on large numbers of processors. It is also several times faster than the techniques commonly used for parallelizing bonded MD that assign a subset of atoms to each processor and require all-to-all communication. Implementation of the algorithm in a CHARMm-like MD model with many body forces and constraint dynamics is discussed and timings on the Intel Delta and Paragon machines are given. Example calculations using the algorithm in simulations of polymers and liquid-crystal molecules will also be briefly discussed.
- OSTI ID:
- 272824
- Report Number(s):
- CONF-930557-; ISSN 0003-0503; TRN: 96:002196-0034
- Journal Information:
- Bulletin of the American Physical Society, Vol. 38, Issue 5; Conference: Conference on physics computing, Albuquerque, NM (United States), 31 May - 4 Jun 1993; Other Information: PBD: May 1993
- Country of Publication:
- United States
- Language:
- English
Similar Records
Massively Parallel Simulations of Diffusion in Dense Polymeric Structures
Fast parallel algorithms for short-range molecular dynamics