|Author(s)||Kesheng Wu and Horst Simon|
|Date||March 30, 1999|
|Platform(s)||UNIX, need BLAS, LAPACK, parallel version needs MPI|
4.4 Lanczos Method, Hermitian eigenvalue problem
5.5 Lanczos Method, generalized Hermitian eigenvalue problem
The LANSO/PLANSO package implements the Lanczos method for hermitian eigenproblems. LANSO is the single processor version, PLANSO is the parallel version and requires MPI.
LANSO implements the Lanczos method for symmetric generalized eigenvalue problems. The current version uses the Omega-recurrence to simulate the loss of orthogonality among the Lanczos vectors and maintains semiorthogonality. This is sufficient to guarantee the eigenvalues are computed accurately, but under extreme conditions the eigenvectors may not be as accurate as the eigenvalues.
The parallel version uses Single-Program-Multiple-Data (SPMD) programming paradigm. All Lanczos vectors are conformally mapped onto all processors of a given MPI group. It should be fairly easy to interface our parallel LANSO (PLANSO) with other packages that provide parallel sparse matrix-vector multiplication and linear system solution.