Parallel reordering algorithms and parallel direct solvers

Abstract

Sparse direct solvers represent a robust approach for the solution of large-scale linear systems arising from complex engineering problems. The primary challenge in their efficient execution is the management of fill-in, which directly affects memory consumption and computational complexity. This thesis provides a clear description of contemporary reordering algorithms employed in reducing these memory requirements, as well as techniques used to parallelize the factorization. Their performance is demonstrated on a selected set of matrices.