Abstract
A comparison is made between Davidson's method for the real, symmetric matrix eigenproblem and a version of the Lanczos method obtained by removing the perturbation theory 'corrections' from Davidson's algorithm. It is found that the convergence of Davidson's method is superior to that of Lanczos only if the matrix is quite strongly diagonally dominant. Applications to typical matrices from nuclear structure calculations, which are not very diagonally dominant, show no essential difference between the convergence rates. The Davidson-Lanczos method as used here is capable, unlike the usual versions of the Lanczos method, of direct application to the generalised eigenproblem Ax= lambda Bx. The author shows how this can be implemented and gives some examples that illustrate the convergence properties.