A new variant of Arnoldi method for approximation of eigenpairs

Arnoldi method approximates exterior eigenvalues of a large sparse matrix, but may fail to approximate corresponding eigenvectors. The refined Arnoldi method approximates an eigenpair by solving a related singular value problem. In this paper, we propose a new procedure to extract an approximate eigenpair from a Krylov subspace in Arnoldi method, using a minimization problem. Unlike the refined Arnoldi method, the suggested procedure requires solving a linear system.