New parallel algorithms for eigenvalue problems /
This work presents a parallelization of a variant form film Inc. of the Davidson algorithm. This variant allows the calculation of several eigenvalues including the case where the multiplicity is higher than one. In addition, new preconditioning strategies deliver better convergence. The disadvantag...
| Main Author: | |
|---|---|
| Format: | Thesis Book |
| Language: | English |
| Published: |
[Place of publication not identified] :
[publisher not identified] ;
1998.
|
| Subjects: | |
| Online Access: | http://proxy.library.tamu.edu/login?url=http://proquest.umi.com/pqdweb?did=733050111&sid=1&Fmt=2&clientId=2945&RQT=309&VName=PQD |
| Summary: | This work presents a parallelization of a variant form film Inc. of the Davidson algorithm. This variant allows the calculation of several eigenvalues including the case where the multiplicity is higher than one. In addition, new preconditioning strategies deliver better convergence. The disadvantage related to dense projected eigenproblems is diminished by representing the projected problems in an arrowhead matrix format. Theoretical timing estimates are developed. The parallel code is compared against these estimates. Also a Newton-Multigrid method is presented. The new iterative method does not rely on Rayleigh quotient iterations, which makes the algorithm distinct from the standard inverse iteration. A multigrid solver is adopted to handle the nearly singular systems which arise from this class of algorithms. Numerical experiments are reported. |
|---|---|
| Item Description: | Vita. "Major Subject: Computer Science". |
| Physical Description: | x, 136 leaves : illustrations ; 28 cm. |
| Bibliography: | Includes bibliographical references (leaves 121-131). |