ABCD-Solver: A hybrid method for solving large sparse linear systems - Algorithmes Parallèles et Optimisation Accéder directement au contenu
Communication Dans Un Congrès Année : 2015

ABCD-Solver: A hybrid method for solving large sparse linear systems

Résumé

Although current work on direct methods has enabled the solution of very large systems efficiently, there are still applications that can cause problems particularly because of memory requirements. On the other hand, iterative methods can often be very slow to converge even with a good preconditioner. We are thus drawn to examine techniques that combine the best features of both approaches, and we term these hybrid methods. We look in particular at the Block Cimmino method which is a rather elementary iterativetechnique but can be accelerated in a number of ways. We examine a derivative of thismethod which is effectively a direct method and compare the performance of this against the MUMPS multifrontal solver and Block Cimmino accelerated by block conjugate gra-dients. A major strength of the block Cimmino method is that it is easily parallelizable. We also comment on this aspect of the approach.
Fichier non déposé

Dates et versions

hal-03213967 , version 1 (30-04-2021)

Identifiants

  • HAL Id : hal-03213967 , version 1

Citer

Iain Duff, Ronan Guivarch, Daniel Ruiz, Mohamed Zenadi. ABCD-Solver: A hybrid method for solving large sparse linear systems. 18th International Conference on Finite Elements in Flow Problems (FEF 2015), International Association of Computational Mechanics (IACM); National Tsing Hua University; Ministry of Science of Technology of Taiwan, Mar 2015, Taipei, Taiwan. ⟨hal-03213967⟩
32 Consultations
0 Téléchargements

Partager

Gmail Facebook X LinkedIn More