Given an n x n symmetric positive definite (SPD) matrix A and an SPD preconditioner P we propose a new class of generalized block tuned (GBT) preconditioners. These are defined as a p-rank correction of P with the property that arbitrary (positive) parameters are eigenvalues of the preconditioned matrix. We propose to employ these GBT preconditioners to accelerate the iterative solution of linear systems like (A - c I) s = r in the framework of iterative eigensolvers. We give theoretical evidence that a suitable, and effective, choice of the scalars is able to shift p eigenvalues of P (A - c I) very close to one. Numerical experiments on various matrices of very large size show that the proposed preconditioner is able to yield an almost constant number of iterations, for different eigenpairs, irrespective of the relative separation between consecutive eigenvalues. We also give numerical evidence that the GBT preconditioner is always far superior to the spectral preconditioner on matrices with highly clustered eigenvalues.

Generalized block tuned preconditioners for SPD eigensolvers

Luca Bergamaschi
;
Angeles Martinez Calomardo
2019

Abstract

Given an n x n symmetric positive definite (SPD) matrix A and an SPD preconditioner P we propose a new class of generalized block tuned (GBT) preconditioners. These are defined as a p-rank correction of P with the property that arbitrary (positive) parameters are eigenvalues of the preconditioned matrix. We propose to employ these GBT preconditioners to accelerate the iterative solution of linear systems like (A - c I) s = r in the framework of iterative eigensolvers. We give theoretical evidence that a suitable, and effective, choice of the scalars is able to shift p eigenvalues of P (A - c I) very close to one. Numerical experiments on various matrices of very large size show that the proposed preconditioner is able to yield an almost constant number of iterations, for different eigenpairs, irrespective of the relative separation between consecutive eigenvalues. We also give numerical evidence that the GBT preconditioner is always far superior to the spectral preconditioner on matrices with highly clustered eigenvalues.
2019
Proceedings of the 2018 Cortona meeting:Structured Matrices in Numerical Linear Algebra: Analysis, Algorithms and Applications
978-3-030-04087-1
File in questo prodotto:
Non ci sono file associati a questo prodotto.
Pubblicazioni consigliate

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11577/3278969
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 1
  • ???jsp.display-item.citation.isi??? ND
social impact