HOMOTOPY CONTINUATION METHOD FOR THE NUMERICAL-SOLUTIONS OF GENERALIZED SYMMETRICAL EIGENVALUE PROBLEMS

We consider a generalised symmetric eigenvalue problem Ax = lambda-Mx, where A and M are real n by n symmetric matrices such that M is positive semidefinite. The purpose of this paper is to develop an algorithm based on the homotopy methods in [9, 11] to compute all eigenpairs, or a specified number of eigenvalues, in any part of the spectrum of the eigenvalue problem Ax = lambda-Mx. We obtain a special Kronecker structure of the pencil A-lambda-M, and give an algorithm to compute the number of eigenvalues in a prescribed interval. With this information, we can locate the lost eigenpair by using the homotopy algorithm when multiple arrivals occur. The homotopy maintains the structures of the matrices A and M (if any), and the homotopy curves are n disjoint smooth curves. This method can be used to find all/some isolated eigenpairs for large sparse A and M on SIMD machines.