Waveform Relaxation with Fast Direct Methods as Preconditioner (Q4509799)
From MaRDI portal
!
This is the item page for this Wikibase entity, intended for internal use and editing purposes. Please use the normal view instead:
scientific article; zbMATH DE number 1519794
| Language | Label | Description | Also known as |
|---|---|---|---|
| default for all languages | No label defined |
||
| English | Waveform Relaxation with Fast Direct Methods as Preconditioner |
scientific article; zbMATH DE number 1519794 |
Statements
Waveform Relaxation with Fast Direct Methods as Preconditioner (English)
0 references
19 October 2000
0 references
semidiscretization
0 references
waveform relaxation
0 references
dynamic iteration
0 references
fast Fourier transform
0 references
parallel computation
0 references
preconditioning
0 references
numerical examples
0 references
linear parabolic equation
0 references
complexity
0 references
implicit linear multistep method
0 references
banded Toeplitz systems
0 references
cyclic reduction
0 references
convergence
0 references
0.813376784324646
0 references
0.813376784324646
0 references
The authors consider the numerical solution of linear parabolic equations in a rectangle with the aim to achieve optimal parallel complexity. They propose to use the fast Fourier transformation in one or both space directions and identify the differential operators for which this approach leads to a decoupling of the semidiscretized equations. In time direction, an implicit linear multistep method is assumed which, considered over all time steps, leads to banded Toeplitz systems which are solved by parallel cyclic reduction. In case the semidiscretized equations can not be decoupled, a preconditioned waveform relaxation is proposed in which the preconditioner leads to decoupled equations (at least in one space direction). Several versions of this efficient but suboptimal method are discussed and numerical results are exhibited which show good convergence, but it becomes clear that for problems with coefficients varying over several orders of magnitude, convergence can be arbitrarily slow.
0 references