ウェーブレットを前処理に用いた共役勾配法の並列処理

概要

論文の詳細を見る
In computational fluid dynamics, most of the computing time, especially for large-scale problems, is occupied by linear system solvers. The reason for this is that computing time of the solvers exponentially increases with increase of grid points. However, the increase of computing time can be suppressed by the incomplete discrete wavelet transform (iDWT). In this study, we first briefly explain the method of iDWT and how to apply it to the preconditioning for a conjugate gradient (CG) solver. The iDWTCG solver is advantageous for large-scale problems, which are the main object to be handled in recent parallel processing studies of CFD. Therefore, we consider parallel implementation of the iDWTCG on the Cray T3D parallel processing system. The results show that modifying data arrangement in iDWT makes it possible to achieve good performance with 128 processors up to 102 times faster than with 1 processor.
一般社団法人日本機械学会の論文
1997-05-25