The parallelisation of an iterative method for the resolution of an eigenvalue problem is efficient on a shared memory computer and on a distribued memory computer
The parallelisation of an iterative method for the resolution of an eigenvalue problem is efficient on a shared memory computer and on a distribued memory computer

Thomas Grosges

Unité de Physico-Chimie et de Physique des Matériaux
Université Catholique de Louvain
Place Croix du Sud, 1 (Bat. BOLTZMANN)
B-1348 Louvain-La-Neuve
Belgium




Abstract

The study of more and more complex systems, like electronic structures, necessitates to solve numerically large eigenvalue problems. Due to the large size of their matrix representation, the utilization of traditionnal resolution methods are inadequate. With the development of parallel computers and iterative methods, some of these problems can be solved.

This study must be included into a large project, consisting in parallelizing the DFT2000 code of total energy computation and must serve to analyze the efficiency of the parallelization of iterative methods on shared and distribued memory computers.

We use a modified block Davidson's iterative method and parallelize it using the MPI (Message Passing Interface) Standard and Fortran 90. The efficiency of parallelization on the two computer architectures are presented.








oo BACK -goto- HOME oo

© Copyright 2007-2036 Thomas Grosges.