A block inverse-free preconditioned Krylov subspace method for symmetric generalized eigenvalue problems (Q1044859)

An analysis is presented, explaining some observed behavior of the preconditioned strategies suggested by \textit{G. H. Golub} and \textit{Q. Ye} [SIAM J. Sci. Comp. 24, No.~1, 312--334 (2002; Zbl 1016.65017)] for the symmetric eigenvalue problem. First, an analysis of the preconditioned strategy based on incomplete factorizations is given. Then the method is extended by developing a block generalization for computing multiple or severely clustered eigenvalues and a robust black-box algorithm implementation is developed. Numerical examples show that this algorithm is competitive and and can effectively detect multiple and clustered eigenvalues. As might be expected, this gain in reliability is achieved with some but not significant loss in performance in terms of extra matrix-vector multiplications.