Overlapping domain decomposition preconditioners for the generalized Davidson method for the eigenvalue problem
- Vanderbilt Univ., Nashville, TN (United States)
The solution of the large, sparse, symmetric eigenvalue problem, Ax = {lambda}x, is central to many scientific applications. Among many iterative methods that attempt to solve this problem, the Lanczos and the Generalized Davidson (GD) are the most widely used methods. The Lanczos method builds an orthogonal basis for the Krylov subspace, from which the required eigenvectors are approximated through a Rayleigh-Ritz procedure. Each Lanczos iteration is economical to compute but the number of iterations may grow significantly for difficult problems. The GD method can be considered a preconditioned version of Lanczos. In each step the Rayleigh-Ritz procedure is solved and explicit orthogonalization of the preconditioned residual ((M {minus} {lambda}I){sup {minus}1}(A {minus} {lambda}I)x) is performed. Therefore, the GD method attempts to improve convergence and robustness at the expense of a more complicated step.
- Research Organization:
- Front Range Scientific Computations, Inc., Boulder, CO (United States); USDOE, Washington, DC (United States); National Science Foundation, Washington, DC (United States)
- OSTI ID:
- 223845
- Report Number(s):
- CONF-9404305--Vol.1; ON: DE96005735; CNN: Grant ASC-9005687; Grant DMR-9217287; Grant DAAL03-89-C-0038
- Country of Publication:
- United States
- Language:
- English
Similar Records
Reducing synchronization on the parallel Davidson method for the large, sparse, eigenvalue problem
The two-phase method for finding a great number of eigenpairs of the symmetric or weakly non-symmetric large eigenvalue problems