Is the iterative refinement of eigenelements an expensive technique (Q1099587)

The eigenvalue problem \(T\phi =\lambda \phi\) with a linear compact integral operator T with continuous kernel on a Banach space X is considered. In classical methods for the numerical computation of eigenelements based on projections or on approximate quadrature one has to solve the corresponding matrix eigenproblem with a dense matrix of order n. In practice X is replaced by a finite dimensional subspace \(X_ N\) of dimension \(N\gg n\). The iterative refinement formula which uses an approximate inverse allows to improve the approximation of an eigenvalue \(\lambda_ n\) and of the left eigenvector \(\phi^*_ n\) in \(X_ n\) only using the product \(T_ Nx\); one needs not to solve the eigenproblem in \(X_ N\). The authors compare the applicability conditions of the different methods and their number of operations as well as the Padé technique. Two numerical examples and four tables are presented.