A variant of the subspace iteration algorithm for generalized eigenproblems

A variant of the subspace iteration algorithm for the generalized eigenproblem, Kp= λMp, is proposed. The algorithm does not require the computation of the projected stiffness matrix, and hence, the projected eigenproblem turns out to be a standard eigenproblem as against the generalized eigenproblem of the classical algorithm. This results in appreciable saving in the number of operations of the Rayleigh–Ritz step and hence the solution time, particularly when a large number of eigenvalues are to be computed. The numerical experiments suggest that the proposed algorithm can yield a saving in solution time of up to 28% for certain eigenproblems.