Discussion Forum

Hi

I agree that 20-25 is too much, for me (my rule of thumb) FEM are precise within 1-5% in the good cases, should always be better than 10%, if not the model is too weak. As you are testing fundamental variables I would have expected to be in the "higher" precision range..

When I meet such discrepancies, I return to the testing, of simple cases, checking that I have not missed a variable name, or that it's one of he variables based on boundary normals (especially for internal boundaries) that is plying me a game, as for interior boundaries you have a sign/direction issue (no error in COMSOl it's a convention issue), so one must take some care when adding or checking the flows. But this is issue limited to "interior boundaries".

So questions to check:
- are you sure about the equations/variable integrated ?
- are you sure about the domains/boundaries used (none missing or some too many) ?
- integrating over interior boundaries ? are the direction/normals univoque ?
- are we in a steady state case, and has the solver fully converged ?
- can you make a simple cube/rectangle or sphere/circle model and do you see the same behavior ?
- ...

Once I have passed such check list, and I'm still not happy with the results, I discuss the case with a colleague (not many playing with COMSOL yet, but I'm working on that too) or I send the model and my explanations off to COMSOL support (or here on the Forum) often having someone else to take a look helps

By the way which version are you using ? be sure it's up to date, V4 is good now but there are still some ironing going on, so until 2-3 releases after a major change I'm never 100% confident with any software (or any man made produced item ;)

--
Good luck
Ivar

Hi,
thank you for the quick replay, I'm using version 4.0a, I checked all of the above except simple case, I will try that.

Thank you mean time

Yes. I am having this same problem in 4.0a with dilute transport (diffusion only) with some thermodynamic partition functions. My model is similar to the "Separation Through Dialysis" example without convection. This is a model that I remade in 4.0a and previously (in v3.5) there was never a problem with the overall mass balance.

Please let me know if there is any resolution on this problem.

Thanks.

Hi
Though I did not solve this problem completely yet, it seems that if you refine the mesh most extremely in the outlets it getting better (my computer currently does not have enough memory), though the general solution is seems to be unchanged. I basically had about 30,000 elements when I increased it to 70,000 the solution did not change at least significantly but the mass and energy balance became better

Hi

My advice, update to most recent version and check again. V4.0 was an early version after the "big changes" of the v4 revolution and has several weaknesses

--
Good luck
Ivar

Beware of the Velocity magnitude and the Total heat flux magnitude, they square the values, so if you have backflow on a boundary, it is reversed in the calculation and you get wrong mass/energy balance. If I'm not clear, look at the equation of the magnitudes in the Equation View\Variables.

Aloha

Hi

That is true, just as for frequency sweeps analysis, there are many RMS values that can be selected, but these should also be considered as such ;)

--
Good luck
Ivar