Gaussian Elimination Method with Partial Pivoting for Sparse Matrices

• Main Authors: Wang Liang-wei, 王良偉
• Other Authors: Liu Jen-Shiuh
• Format: Others
• Language: zh-TW
• Published: 2000
• Online Access: http://ndltd.ncl.edu.tw/handle/53994875136897626166

碩士 === 逢甲大學 === 資訊工程學系 === 88 === Matrix computation is at the core of many engineering and scientific. Sparse matrices are the most part of those matrices. If sparse matrices aren’t been compressed, it will cause much overhead in computation and storage. So efficient compute and store sparse matrices been a researching topic. In this thesis we use partial pivoting method and dynamic memory allocate method factorize sparse matrix in parallel, and suppose a precise and efficient computing sparse matrix system. In this thesis we compressed sparse matrix with CCS compressed scheme, distributed matrix data with column block-cyclic(k) distribution, maintained computation stability with partial pivoting, and did Gaussian Elimination with matrix in parallel. Our method can avoid diagonal-zero matrices such as b1_ss, and bcsstm19, and can factorized matrices efficiently and precisely in parallel. Keyword: sparse matrix, Gaussian Elimination, partial pivoting, block-cyclic(k)