An acceleration method for computing the generalized eigenvalue problem on a parallel computer (Q756366)

The authors discuss a computational technique for the generalized eigenvalue problem, which is the problem of finding \(\lambda\) such that \(\det (A-\lambda B)=0,\) where A and B are square \(n\times n\) matrices. Jacobi-like methods work by applying equivalence transformations which aim to reduce the subdiagonal elements of both A and B to zero. The authors propose an acceleration method which can be used when the subdiagonals of A and B have become sufficiently small. Their method applies transformations which force a norm of these elements to approach zero cubically. Convergence analysis and numerical experience are given, and the authors show how their technique can be implemented on a mesh- connected \(n\times n\) square array of processors.