Hermitian Subspace Iteration

Abstract

The subspace iteration for symmetric real matrices is extended to the partial eigensolutions for Hermitian matrices. Only the lower triangle of the Hermitian matrix in variable band form is required and processed. Efficient and compact fortran subroutines are developed for computing the lowest modes in ascending order of absolute eigenvalues. The method is suitable for incore microcomputer solution of the sparse Hermitian eigenvalue problem.