We present a parallel implementation of the Car-Parrinello method (1985) for first principles molecular dynamics simulations, in the Parallel Virtual Machine (PVM, version 3.3.4) environment on the CRAY-T3D computer. Parallelism is achieved at the Single Program Multiple Data (SPMD) level, dividing among processors the calculations on different columns of a large matrix, containing the electronic wavefunctions of a condensed matter system object of the simulation, written in a plane-wave basis. Fast Fourier Transforms (FFTs) of these electronic states are the heaviest computations in the program. Communications are avoided an most of the code, except where the columns must be orthogonalized among each other. Even if it is not possible to make parallel the whole code, we have reached a parallel implementation of the most time-consuming part. We get an excellent behaviour of this part, and a satisfactory global behaviour.
Ab-initio molecular dynamics calculations: a parallel implementation of the Car-Parrinello method / Di Felice, R.; Mantovani, M. L.; Bertoni, C. M.. - (1996), pp. 109-114. ( 4th Euromicro Workshop on Parallel and Distributed Processing, PDP 1996 prt 1996) [10.1109/empdp.1996.500576].
Ab-initio molecular dynamics calculations: a parallel implementation of the Car-Parrinello method
Di Felice, R.
;Mantovani, M. L.
;Bertoni, C. M.
1996
Abstract
We present a parallel implementation of the Car-Parrinello method (1985) for first principles molecular dynamics simulations, in the Parallel Virtual Machine (PVM, version 3.3.4) environment on the CRAY-T3D computer. Parallelism is achieved at the Single Program Multiple Data (SPMD) level, dividing among processors the calculations on different columns of a large matrix, containing the electronic wavefunctions of a condensed matter system object of the simulation, written in a plane-wave basis. Fast Fourier Transforms (FFTs) of these electronic states are the heaviest computations in the program. Communications are avoided an most of the code, except where the columns must be orthogonalized among each other. Even if it is not possible to make parallel the whole code, we have reached a parallel implementation of the most time-consuming part. We get an excellent behaviour of this part, and a satisfactory global behaviour.
Pubblicazioni consigliate
I metadati presenti in IRIS UNIMORE sono rilasciati con licenza Creative Commons CC0 1.0 Universal, mentre i file delle pubblicazioni sono rilasciati con licenza Attribuzione 4.0 Internazionale (CC BY 4.0), salvo diversa indicazione.
In caso di violazione di copyright, contattare Supporto Iris
