Modified Krylov acceleration for parallel environments
The authors present a few variants of Krylov subspace techniques for solving systems of linear equations mainly on parallel computers. A strategy to avoid the main difficulty when implementing this class of methods in parallel environment, computation of inner products, is given. It is based on replacing the standard Eulerian inner product by a discrete inner product over a space of polynomials. In this approach, the Krylov subspace is actually viewed as a space of polynomials. The set of nodes for the discrete inner product is obtained by estimating the eigenvalues of the coefficient matrix. The authors consider first the case when the spectrum is real and then extend the technique to the case when the spectrum is complex. Results obtained from a lot of numerical test examples are given and discussed. All tests have been performed on CRAY T3E computer with up to eight processors and using matrices which have been taken from the well known Harwell-Boeing and SPARSKIT collections. The numerical results show that the method proposed in this paper is attractive in both cases, without and with preconditioners, but the new algorithm is much more sensitive to poor preconditioners than classical ones.
- Parallel computation of a Krylov matrix for a sparse and structured input
- Hardware-oriented Krylov methods for high-performance computing
- High Performance Computing for Computational Science - VECPAR 2004
- Krylov subspace acceleration for nonlinear multigrid schemes
- Alternatives for parallel Krylov subspace basis computation
- Parallel implementation of multifrontal schemes
- Parallel variable-triangular iterative methods in Krylov subspaces
- scientific article; zbMATH DE number 2080875
- Experiments with Krylov subspace methods on a massively parallel computer
- Interpolating discrete advection--diffusion propagators at Leja sequences
- Krylov Subspace Methods on Supercomputers
- Alternatives for parallel Krylov subspace basis computation
- General framework for deriving reproducible Krylov subspace algorithms: BiCGStab case
- A performance model for Krylov subspace methods on mesh-based parallel computers
- Krylov subspace methods for structural finite element analysis
This page was built for publication: Modified Krylov acceleration for parallel environments
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q1294592)
