BOUNDS ON BLOCK DIAGONAL PRECONDITIONING
Authors:
Mark Yan-Ming Chang a;
Martin H. Schultz a
| Affiliation: | a Department of Computer Science, Yale University, New Haven, CT |
DOI:
10.1080/10637199308915437
Publication Frequency:
6 issues per year
Published in:
International Journal of Parallel, Emergent and Distributed Systems,
Volume
1,
Issue
2
1993
, pages 141
- 164
Subjects:
Algorithms & Complexity;
Computer Engineering;
Computer Science (General);
Distributed Network Systems;
Distributed Systems;
Internet & Multimedia;
Neural Networks;
Parallel Algorithms;
Parallel Systems;
Programming & Programming Languages;
Quantum Information;
Systems & Computer Architecture;
Formats available:
PDF
(English)
Previously published as:
Parallel Algorithms and Applications
(1063-7192)
until 2005
View Article:
View Article (PDF)
Abstract
We consider the preconditioned conjugate gradient (PCG) method for solving the linear equations derived from difference approximations to the Laplace operator on 1D line, 2D square or 3D cube domains. If the preconditioner is a block diagonal matrix obtained from cutting the original domain by lines or planes, we can End an upper bound for the condition number of the preconditioned system. For several different parallel machine models, we derive the optimal number of processors to achieve the best asymptotic run time.
|
| Keywords: Block diagonal PCG; condition number bound; parallel asymptotic time; optimal number of processors |
| view references (10) |

Download Citation

CiteULike
Del.icio.us
BibSonomy
Connotea