Some strategies for enhancing the performance of the block Lanczos method
Abstract
The block Lanczos method is used to calculate the eigenfunctions for a generalized eigenvalue problem constructed as a finite element solution to a 2dimensional Schrödinger equation on the surface of a hypersphere. This equation results from a treatment of the 3dimensional reactive scattering problem using Adiabatically adjusting, Principal axes Hyperspherical (APH) coordinates. Three strategies are considered with respect to increasing the CPU performance of the block Lanczos (with selective orthogonalization) method: (1) the effect of varying the Lanczos block size; (2) the effect of block tridiagonal ordinary eigenvalue problem upon every other Lanczos iteration; and (3) the effect of dividing a single problem of finding ϱ eigenvalue into a set of ϱ _{i} problems, where each subproblem consists of finding ϱ/ϱ _{i} eigenvalues.
 Publication:

Computer Physics Communications
 Pub Date:
 May 1989
 DOI:
 10.1016/00104655(89)901513
 Bibcode:
 1989CoPhC..53..109K