Massively Parallel Iterative Methods: Multiscale Preconditioners and Implicit Methods.
Abstract
Nonlinear and linear systems of equations often arise in scientific computation, for example in implicit methods in Computational Fluid Dynamics (CFD). It is important to find cost-effective and accurate methods to solve such systems. Iterative methods are among those widely used, especially for 3D problems. In this project, we consider iterative methods which are especially suited to massively parallel architectures. To accelerate convergence of these iterative methods, preconditioners are often used. Good preconditioners reduce the number of iterations and involves few arithmetic operations per iteration. Effective parallel preconditioners must account for the global coupling inherent in elliptic problems. On the other hand, efficient parallel implementation often favors local computations. Multiscale iterative methods represent a good compromise between these two conflicting goals. We focused our attention on two classes of multiscale preconditioners: multilevel basis preconditioners and domain decomposition preconditioners.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 22, 1995
- Accession Number
- ADA295790
Entities
People
- Tony F. Chan
Organizations
- University of California, Los Angeles