Preconditioning eigenvalues and some comparison of solvers (Q1591177)

This is nice summary of preconditioning techniques for eigenvalue problems \(Ax=\lambda x\), including Davidson, Jacobs-Davidson, Rayleigh quotient iteration and preconditioned Lanczos methods. It is shown that these preconditioning methods are related by the same preconditioned operator \(M^{-1}(A-\theta I)\). Here \(\theta\) is an approximate eigenvalue and \(M\) is an approximation of \(A-\theta I\). Hence, the convergence rates of preconditioning methods are often similar. However, proper implementation can make a difference. This is, in addition to the comparisons of these methods, illustrated by numerical examples.