In this paper we discuss sparse matrix computational methods, and their parallel implementations, for evaluating matrix-vector products in iterative solution of coupled, nonlinear equations encountered in finite element flow simulations. Based on sparse computation schemes, we introduce globally-defined preconditioners by mixing clustered element-by-element preconditioning concept with incomplete factorization methods. These preconditioners are implemented on a CRAY T3D parallel supercomputer. In addition to being tested in a number of benchmarking studies, the sparse schemes discussed here are applied to 3D simulation of incompressible flow past a circular cylinder.
|ジャーナル||Computer Assisted Mechanics and Engineering Sciences|
|出版ステータス||Published - 1998 12 1|
ASJC Scopus subject areas
- コンピュータ サイエンスの応用