Análise da influência de algoritmos de reordenação de matrizes esparsas no desempenho do método CCCG(n)

Abstract

In this work we investigate the influence of reordering algorithms on the performance of controlled Cholesky conjugate gradient method - CCCG(ç). This method has been proved to be ecient in solution of high linear systems with positive definite coecient matrix. It has been considered simple algorithms like column count and reverse thill-McKeeand more sophisticated algorithms like approximate minimum degree. Some numerical results on the eect of orderings on the fill-in and the iteration number have been presented. It is shown experimentally that certain reorderings like aproximated minimum degree can be very beneficial. The benefits consist of a faster convergence of the method and a lower storage requirements. Finally, the situation where a reordering can improve the CCCG(ç) was estabilished.