Influence of matrix reordering on the performance of iterative methods for solving linear systems arising from interior point methods for linear programming

CNPQ - CONSELHO NACIONAL DE DESENVOLVIMENTO CIENTÍFICO E TECNOLÓGICOFAPESP - FUNDAÇÃO DE AMPARO À PESQUISA DO ESTADO DE SÃO PAULOThis study analyzes the influence of sparse matrix reordering on the solution of linear systems arising from interior point methods for linear programming. In particular, such linear systems are solved by the conjugate gradient method with a two-phase hybrid preconditioner that uses the controlled Cholesky factorization during the initial iterations and later adopts the splitting preconditioner. This approach yields satisfactory computational results for the solution of linear systems with symmetric positive-definite matrices. Three reordering heuristics are analyzed in this study: the reverse Cuthill-McKee heuristic, the Sloan algorithm, and the minimum degree heuristic. Through numerical experiments, it was observed that these heuristics can be advantageous in terms of accelerating the convergence of the conjugate gradient method and reducing the processing time.This study analyzes the influence of sparse matrix reordering on the solution of linear systems arising from interior point methods for linear programming. In particular, such linear systems are solved by the conjugate gradient method with a two-phase hybrid preconditioner that uses the controlled Cholesky factorization during the initial iterations and later adopts the splitting preconditioner. This approach yields satisfactory computational results for the solution of linear systems with symmetric positive-definite matrices. Three reordering heuristics are analyzed in this study: the reverse Cuthill-McKee heuristic, the Sloan algorithm, and the minimum degree heuristic. Through numerical experiments, it was observed that these heuristics can be advantageous in terms of accelerating the convergence of the conjugate gradient method and reducing the processing time85197112CNPQ - CONSELHO NACIONAL DE DESENVOLVIMENTO CIENTÍFICO E TECNOLÓGICOFAPESP - FUNDAÇÃO DE AMPARO À PESQUISA DO ESTADO DE SÃO PAULOCNPQ - CONSELHO NACIONAL DE DESENVOLVIMENTO CIENTÍFICO E TECNOLÓGICOFAPESP - FUNDAÇÃO DE AMPARO À PESQUISA DO ESTADO DE SÃO PAULOsem informaçãosem informação13th EUROPT Workshop on Advances in Continuous OptimizationJUL 08-10, 2015Edinburgh, SCOTLAN