#1063 UGGrid parallel with cgstokes example
Metadata
| Property | Value |
|---|---|
| Reported by | Manuel Hofmann (manuel.hofmann@iwr.uni-heidelberg.de) |
| Reported at | Apr 11, 2012 11:00 |
| Type | Bug Report |
| Version | Git (pre2.4) [autotools] |
| Operating System | Unspecified / All |
| Last edited by | Christoph Grüninger (gruenich@iws.uni-stuttgart.de) |
| Last edited at | Apr 11, 2012 12:01 |
| Closed by | Christoph Grüninger (gruenich@iws.uni-stuttgart.de) |
| Closed at | Apr 11, 2012 12:01 |
| Closed in version | Unknown |
| Resolution | Not a bug |
| Comment |
Description
Hi,
i tried to run the cgstokes example in pdelab-howto parallel. This example uses Taylorhood elements q2/q1 with uggrid. Im using the latest version of dune-grid/dune-geometry/... however I attached an older version of cg_stokes.hh of dune-pdelab because I modified it alot and didnt update my program to the new stokesparameter.hh
If i run this example with ISTLBackend_NOVLP_BCGS_NOPREC as linear solver it seems to work, however it converges really slow and even breaks down (ISTLError) at some point (but only after a few nonlinear iterations). If i use a preconditoned method like ISTLBackend_NOVLP_BCGS_SSORk i get a Dune::GridError at the first nonlinear iteration. I tracked it down and i think it happens in novlpistlsolverbackend.hh:line 1018, where the first linear system is being solved.
Oliver suggested I should file a bug if this still happens with an unrefined grid. I also attached an error logfile of the case with ISTLBackend_NOVLP_BCGS_SSORk as linear solver.
Am I doing anything wrong or is this still a bug of UGGrid?
Regards, Manuel