Acessibilidade / Reportar erro

Novas Versões para a Inversa Aproximada em Blocos: Uma Comparação Numérica

ABSTRACT

We propose two variations of the block approximate inverse preconditioner (BAINV), presented by Benzi, Kouhia and Tůma in 2001. The first variation, the stabilized block approximate inverse for non-symmetric matrices (SBAINV-NS), is used for non-symmetric and non-singular matrices. The second variation, the combined stabilized block approximate inverse (SBAINV-VAR), is based on the relations between the block approximate inverse factors with the block LDU factors of A, as we will demonstrate, and on the relations between the approximate inverse and Neumann series. We prove the mathematical consistency of these new versions and present the algorithms for each one. We also present the numerical experiments, where we compare the density of the preconditioners and the number of iterations when applying the biconjugate gradient stabilized method (Bi-CGSTAB). The main numerical results indicate that the use of the block structure can increase the performance of the Krylov’s iterative method compared to the scalar version. Furthermore, the experiments show that SBAINV-VAR preconditioners, in general, perform less iterations of Bi-CGSTAB and are less dense than SBAINV-NS preconditioners.

Keywords:
approximate inverse; block matrices; preconditioners; krylov methods

Sociedade Brasileira de Matemática Aplicada e Computacional - SBMAC Rua Maestro João Seppe, nº. 900, 16º. andar - Sala 163, Cep: 13561-120 - SP / São Carlos - Brasil, +55 (16) 3412-9752 - São Carlos - SP - Brazil
E-mail: sbmac@sbmac.org.br