Dear Foamers, we found very
we found very strange behavior in solving purely conductive heat transfer in parallel using tetrahedral mesh in version 1.4.1.
We were able to reproduce the error in a simple case named flate plate.
The case is composed by two separate region one of which is tetrahedral and the other hexahedral. The simulation is only conductive in a steady manner and represents the 1-D conduction between two surfaces at fixed temperature. So isolines in the two region should be almost aligned. While in the serial simulation the results are ok the solution of the parallel simulation is influenced by the number of partition in the tetrahedral domain. No matter what gradient scheme was used.
Some picture can explain better the phenomenon
Serial Temperature contour
Parallel Temperature contour
Mesh detail around processor patch and Temperature contour
As we can see from figure the tetrahedral region for the parallel run has not the same isolines temperature like in the serial simulation.
The conductions is pure steady simulation (only laplacian) and the resudual are of the order of 1.e-10, so we believe it is not a convergence problem.
we have uplodaded also the case at flate plate case
Lu & Co
Sorry guys for warning you abo
Sorry guys for warning you about something we just discovered was not a bug. It was simply the value of the diffusivity on the processor patch that was a wrong one.
Anyhow sometimes the forum is helpful also for this: you spend time searching for a difficult answer but at the end the bug is right in front of your nose but you see it only when you post the question.
Sorry again for this half-an-hour lasting alarm,
Luca & Cosimo
Hallo Luca M., I also have
Hallo Luca M.,
I also have the problem, that parallel running cases are instable and the results are kind of wrong. Please explain what you mean by "diffusivity of the processor patch" and how to change it right.
|All times are GMT -4. The time now is 15:42.|