View Issue Details
ID | Project | Category | View Status | Date Submitted | Last Update |
---|---|---|---|---|---|
0000616 | OpenFOAM | Bug | public | 2012-08-06 12:53 | 2012-08-06 16:28 |
Reporter | akidess | Assigned To | henry | ||
Priority | normal | Severity | minor | Reproducibility | always |
Status | closed | Resolution | no change required | ||
Platform | Linux | OS | Other | OS Version | (please specify) |
Summary | 0000616: fvMatrix relaxation does not give expected behaviour | ||||
Description | I'd expect the results of "fvScalarMatrix.solve();" and "fvScalarMatrix.relax(1); fvScalarMatrix.solve();" to be exactly the same, as in section 4.5.2 of the user guide it is stated "0<a<=1 specifies the amount of under-relaxation, ranging from none at all for a=1". However, if the relaxation is included the solution seems to diverge (in any case, the results differ from the solution without relax() ). | ||||
Steps To Reproduce | Compile the attached solver, then blockMesh and run on the test case. | ||||
Tags | No tags attached. | ||||
|
Accidental duplicate of #615 |
|
fvScalarMatrix.relax(1) is not equivalent to no under-relaxation; it forces the matrix to be at least diagonally equal which is a necessary initial condition to apply an under-relaxation factor. |
|
Alright, well that explains the divergence for relax(1), since the resulting matrix will not be strictly diagonally dominant. However, I even get a divergent solution with relaxation factors a<<1, which should ensure diagonal dominance. The solution without relaxation is fine. |
|
Diagonal dominance should ensure that the particular matrix may be solved with the current solvers but does not guarantee anything with respect to overall convergence of the multi-equation non-linear system. The relaxation procedure currently requires the diagonal to be positive. It is not clear that this report relates to a bug in OpenFOAM but is more likely to be a bug in the solver or the case you are working with. |
|
Excuse my persistence, but I think that's making things too easy. All I'm solving is a simple Laplace equation (see attached tar file), there is no inter-equation dependency, and the only thing hindering convergence to the proper solution is the relaxation itself. |
|
|
|
NOTE: The relaxation procedure currently requires the diagonal to be positive. |
Date Modified | Username | Field | Change |
---|---|---|---|
2012-08-06 12:53 | akidess | New Issue | |
2012-08-06 12:55 | akidess | Note Added: 0001557 | |
2012-08-06 14:43 | henry | Note Added: 0001558 | |
2012-08-06 14:43 | henry | Status | new => resolved |
2012-08-06 14:43 | henry | Resolution | open => no change required |
2012-08-06 14:43 | henry | Assigned To | => henry |
2012-08-06 15:33 | akidess | Note Added: 0001559 | |
2012-08-06 15:33 | akidess | Status | resolved => feedback |
2012-08-06 15:33 | akidess | Resolution | no change required => reopened |
2012-08-06 15:47 | henry | Note Added: 0001560 | |
2012-08-06 15:47 | henry | Status | feedback => closed |
2012-08-06 15:47 | henry | Resolution | reopened => fixed |
2012-08-06 16:21 | akidess | Note Added: 0001561 | |
2012-08-06 16:21 | akidess | Status | closed => feedback |
2012-08-06 16:21 | akidess | Resolution | fixed => reopened |
2012-08-06 16:21 | akidess | File Added: relaxBugRep.tar | |
2012-08-06 16:28 | henry | Note Added: 0001562 | |
2012-08-06 16:28 | henry | Status | feedback => closed |
2012-08-06 16:28 | henry | Resolution | reopened => no change required |