出版社：SISSA, Scuola Internazionale Superiore di Studi Avanzati
摘要：We present a new variant of a Krylov space algorithm for finding the eigenpairs of a large hermitean matrix where the eigenvalues lie in a specified low density part of the spectrum. The method uses selective re-orthogonalization and re-starting to find each eigenpair once. We present theoretical bounds on the convergence rate, and show that these work well in practice for the hermitean Wilson Dirac operator. We have implemented the method in Chroma, and we show that it is significantly faster than the Ritz method currently available.
