An iterated shift-and-invert Arnoldi algorithm for quadratic matrix eigenvalue problems (Q2369227)

For solving the large scale quadratic eigenvalue problem \((A\lambda^2+ B\lambda+ C\lambda)x= 0\), a direct projection method based on the Krylov subspaces generated by a single matrix \(A^{-1}B\). Using the standard Arnoldi or the Lanczos algorithm is considered. A key observation in this paper is that when this direct projection is iteratively combined with the shift-and-invert technique, the influence of \(C\) matrix after the shift-and-invert transformation diminished and the Krylov subspace solely constructed by \(A^{-1}B\) is sufficient to extract good approximations. The important situations of inexact shift-and-invert are also discussed and numerical examples are presented to illustrate the new method.