[borose]

Hi there, i'm pretty new to CFD moddeling. (COMSOL) Is there any criteria, if the solver reaches convergence, that the solution is valid? For example, you can post-process the local cell reynolds number. Is this a reasonable value for verification? Should the mesh density be increased where the cell Re-Nnumber is too high? Or peclet number for heat transfering ... What is your experience? Thanks in advance!

[hectorgabriel85]

One way you can do it is solve the problem with a given mesh. Then refine it (i.e. new elements half the size of original ones). You can refine everywhere and/or in regions where the unknowns (pressure, velocity, temperature) change fast in space (AKA the gradient of the unknown is high)
Finally you compare both solutions. If they are very similar you are done. If not something can be wrong.

[borose]

thanks for reply

this was exactly my "problem". The solver converged, but then i refined the mesh and i had +15K in another domain cause of an bigger heat transfer through convection. Further refinement had only little change.

Often, if the solver doesn't converges, then i look up the cell Re-number and i refine areas with the highest Re. This usually works for me. So I thought, the cell Re number is maybe kind of important - even for verification of the solution. But it seems to me, that the cell Re should not get too much attention or should not used for evaluation for the question "were the equations solved right?". I read some papers where the cell Re was used for adaptive mesh refinement - so this was my starting point.

What do you usally do to verificate your results? coarsening the mesh? Refinement of the mesh? Use higher order elements (fem)? Changing the tolerances in the solver settings? I really don't know, but just changing the mesh should not be the only way for verification of the solution.

regards