Approximate eigenvectors as preconditioner (Q1976916): Difference between revisions

The author investigates the stability of the computation of the eigenvalues of a nonsingular Hermitian matrix \(H\) via a special preconditioning technique. Suppose that some approximate eigenvector matrix \(\widetilde U\) is available. Instead of applying, say, the Jacobi method directly to \(H\), one should first precondition the Hermitian matrix \(H\) by the approximate eigenvector matrix \(\widetilde U\). If the preconditioned matrix \(H'= \widetilde U^* H\widetilde U\) is ``well-conditioned'', i.e. the matrix \(|H'|= \sqrt{(H')^2}\) is of the form \(DAD\) with \(D\) diagonal and \(A\) well-conditioned, then one can expect that \(H'\) can be efficiently and accurately diagonalized by the Jacobi method. The main topic of the paper addresses the investigation of the numerical stability of the transition from \(H\) to \(H'\) in finite precision arithmetic.