Skip to main content
U.S. Department of Energy
Office of Scientific and Technical Information

Parallel inverse iteration with reorthogonalization

Conference ·
OSTI ID:6519502
A parallel method for finding orthogonal eigenvectors of real symmetric tridiagonal is described. The method uses inverse iteration with repeated Modified Gram-Schmidt (MGS) reorthogonalization of the unconverged iterates for clustered eigenvalues. This approach is more parallelizable than reorthogonalizing against fully converged eigenvectors, as is done by LAPACK's current DSTEIN routine. The new method is found to provide accuracy and speed comparable to DSTEIN's and to have good parallel scalability even for matrices with large clusters of eigenvalues. We present al results for residual and orthogonality tests, plus timings on IBM RS/6000 (sequential) and Intel Touchstone DELTA (parallel) computers.
Research Organization:
Pacific Northwest Lab., Richland, WA (United States)
Sponsoring Organization:
DOE; USDOE, Washington, DC (United States)
DOE Contract Number:
AC06-76RL01830
OSTI ID:
6519502
Report Number(s):
PNL-SA-21790; CONF-930331--10; ON: DE93012867
Country of Publication:
United States
Language:
English