Parallel implementation of a nonsymmetric tridiagonal eigensolver
In this paper, we investigate parallel solution of the nonsymmetric tridiagonal eigenproblem by the LR method, inverse iteration, and Rayleigh quotient iteration on a distributed-memory computer. We examine serial computation of eigenvalues followed by parallel computation of block- or wrap-distributed eigenvectors and parallel computation of eigenvalues using a block algorithm. We also consider pipelining the calculation of eigenvalues with the calculation of the eigenvectors. Experiments on an Intel iPSC/860, where the communication to computation ratio is very large, show that serial eigenvalue computation with wrap-mapped eigenvector computation has the lowest execution time. 3 refs., 1 fig., 1 tab.
- Research Organization:
- Oak Ridge National Lab., TN (United States)
- Sponsoring Organization:
- USDOE; USDOE, Washington, DC (United States)
- DOE Contract Number:
- AC05-84OR21400
- OSTI ID:
- 5498542
- Report Number(s):
- CONF-910326-7; ON: DE91016773
- Resource Relation:
- Conference: 5. Society for Industrial and Applied Mathematics (SIAM) conference, Houston, TX (United States), 22-29 Mar 1991
- Country of Publication:
- United States
- Language:
- English
Similar Records
Parallel solution of the symmetric tridiagonal eigenproblem
Parallel tridiagonalization of a general matrix using distributed-memory multiprocessors