ACCELERATION OF THE SUBSPACE ITERATION METHOD BY SELECTIVE REPEATED INVERSE ITERATION

Acceleration of the subspace iteration method is achieved by performing an extra inverse iteration step on trial vectors associated with selected higher required eigenvalues. A criterion, based on an estimate of each eigenvalue's ultimate rate of convergence, is used to choose the trial vectors which will undergo another inverse iteration. The technique is found to be efficient and can be used in conjunction with other acceleration techniques. The results of a number of demonstrative numerical analyses are presented and discussed.