An incomplete nested dissection algorithm for parallel direct solution of finite element discretizations of partial differential equations (Q1322460): Difference between revisions

The purpose of the paper is to present a highly parallel and memory efficient algorithm for direct \(LU\)-decomposition of large sparse systems of linear equations that appear when finite element and spectral element methods are applied to linear and nonlinear differential equations. The efficiency of the discussed algorithms is illustrated by the results of numerical experiments that show significant reduction of the number of messages and the total volume of messages passed during factorization among the processors in comparison to other algorithms.