A parallel multisplitting solution of the least squares problem (Q4940812)
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 1406959
| Language | Label | Description | Also known as |
|---|---|---|---|
| default for all languages | No label defined |
||
| English | A parallel multisplitting solution of the least squares problem |
scientific article; zbMATH DE number 1406959 |
Statements
27 February 2000
0 references
linear least-squares problem
0 references
QR factorization
0 references
iterative refinement
0 references
parallel algorithms
0 references
multisplitting
0 references
numerical examples
0 references
convergence
0 references
A parallel multisplitting solution of the least squares problem (English)
0 references
0.8644874691963196
0 references
0.8100131750106812
0 references
0.7860445976257324
0 references
Modified author's abstract: The linear least squares (LS) problem, \(\min _x\|Ax-b\|_2\), is solved by applying a multisplitting (MS) strategy in which the matrix \(A\) is decomposed into \(p\) blocks of columns. The vectors \(b\) and \(x\) are partitioned consistently with the matrix decomposition. The global least squares problem is then replaced by a sequence of local least squares problems solved in parallel. Basically, two algorithms are suggested to solve the global LS problem. In the LSMS method, the solutions to the local problems are recombined using fixed weighting matrices. If, moreover, an optimal recombination (OR) is introduced as an auxiliary LS problem of dimension \(p\), one gets a two-stage algorithm (ORLSMS). The above-mentioned algorithms can be modified by means of the global residual vector (iterative refinement, IR) but numerical examples show that the convergence is not improved. Overlaps of local subproblems can also be considered in the algorithms. Convergence analysis for (OR)LSMS(IR) is performed as well as numerical tests. The latter demonstrate the iteration behaviour with respect to the subproblem size, convergence criteria, recombination techniques and overlap values. Results are discussed and conclusions drawn. In general, the ORLSMS method seems to converge better than the LSMS one though the contrary is also shown in a few cases. The method is effective for near-separable problems.
0 references