#928 breakdown of BiCGSTAB
Metadata
| Property | Value |
|---|---|
| Reported by | Bernd Flemisch (bernd@iws.uni-stuttgart.de) |
| Reported at | Jun 8, 2011 14:49 |
| Type | Bug Report |
| Version | Git (pre2.4) [autotools] |
| Operating System | Unspecified / All |
| Last edited by | Markus Blatt (markus@dr-blatt.de) |
| Last edited at | Jun 9, 2011 09:40 |
| Closed by | Markus Blatt (markus@dr-blatt.de) |
| Closed at | Jun 9, 2011 09:40 |
| Closed in version | Unknown |
| Resolution | Not a bug |
| Comment | see my comment |
Description
Dear Dune,
in some cases, the BiCGSTABSolver breaks down with the message Dune reported error: ISTLError [apply:../../../dune/istl/solvers.hh:791]: h=0 in BiCGSTAB
I attach an example with a particular case, to be run with ./test_bicgstab matrix.txt
I checked the matrix with Matlab, it is a perfectly fine positive definite matrix. As can be seen from the example, the LoopSolver works (so does the CGSolver). The preconditioner does not seem to matter.
What seems to matter, is the right hand side. If I change line 82 to b = 1; everything is fine.
It would be great if you could check. Thank you!
BTW: The example contains a routine void readBCRSMatrixFromMatlabFile(const char *fileName, Matrix& A) which I think could be interesting for inclusion in dune/istl/io.hh. It is the reverse to writeMatrixToMatlab.