Deflated block Krylov subspace methods for large scale eigenvalue problems (Q966072)

The aim of the paper is to study a class of deflated block Krylov subspace methods, for solving large scale matrix eigenvalue problems, techniques which are different from conventional block Krylov subspace methods in the way of generating the approximate subspaces. The first section is of introductory nature. The second section briefly recalls the concept of block Krylov subspace and related procedures for solving large scale eigenvalue problems, giving then a brief introduction of the deflated block Krylov subspace. The authors also present an Arnoldi-type algorithm which can accomplished the orthogonalization procedure presented above. The third section deals with some deflated Krylov subspace methods for solving large scale eigenvalue problems; particularly, an Arnoldi-type method and its refined variant will be outlined and compared. The authors prove in this section that due to deflation, approximate subspaces constructed by the new method have similar properties to ones constructed by the implicitly restarted Arnoldi method, or the restarted Arnoldi methods augmented with approximate eigenvectors. The new methods preserve the superiorities of regular block Krylov subspace methods and regular Krylov subspace methods augmented with approximate eigenvectors. In the fourth section, the authors present seven numerical examples computed by the Arnoldi-type method and its refined variant. These examples demonstrate the efficiency of the presented methods for computing partial or nearly clustered eigenvalues of large matrices. Also, the test results show that the refined variant can improve the performance of the Arnoldi-type method in computing partial eigenvalues of large matrices.