[andy.sun.wei]

I've been a COMSOL user for over 20 years (from when it was called FEMLAB). And there isn't a COMSOL user forum here at cfd-online. Is the software that unpopular?


Well, recently I ran into a COMSOL simulation problem and went to COMSOL's own forum to post and ask for user's input and help. You know what, COMSOL forum modulator deleted my post and banned my account very quickly. Wow, I'm surprised by the COMSOL treatment to their user. Maybe that's the reason nobody using it any more. By the way, that was the first time I posted anything there.



Any way, here is my COMSOL simulation problem and let me know if you can help. It's a long post, please bare with me.

All in all I got some weird results from COMSOL simulation and hope someone can explain where I got it wrong, or COMSOL is wrong?

I’m trying to study a very simple 2-gas mixing problem. I assume there is only diffusion between the gasses and ignore any convection flow or gravitational effect. The gases follow Fick’s diffusion law and diffusion coefficient is set at 1.2e-5[m^2/s].

This is a 2-D simple geometry problem, see picture 1 below. The 2 gases initially are separated in 2 adjacent boxes with gas 1 in box 1 with molar mass for gas 1Mw1=0.016[kg/mol] and gas 2 in box 2 withMw2=0.044[kg/mol]. Pressure is set at 1 [atm] and temperature at 293.15 [K] all over. The gases behavior like ideal gas.

Listed in table below are calculated volume, density, mass and molar amount (assume depth at 1 [m]) of the 2 gases. The 2 gases start to mix from complete separated state at beginning. Eventually when they are fully mixed the gas volume percentage will be 80% for gas 1 and 20% for gas 2, which is equal to the initial volume percentage of the 2 gasses. The gas mass faction will be w1= 0.5926 and w2=0.4074 everywhere when fully mixed.




Here I solve the problem as time dependent and got the solution up to 100 seconds. The 4 plots (picture 3) are for gas 1 mass faction w1 at time of 0, 1, 10 and 100 second mark. The results seem reasonable. At beginning (0 second), gas 1 is only at the left box (w1=1) and no gas 1 in right box (w1=0). As time evolve, it can be seen that the mass fraction of gas 1 gradually goes from left to right (see 1 and 10 second plots). At 100 second, the plot shows w1 is virtually uniform across all domains 1 and 2, that means 2 gases are fully mixed.



However, the results at 100 second show w1= 0.71149 all across. Or if plot the surface with gas volume concentration for gas 1 (use formula (w1/0.016)/(w1/0.16+w2/0.044) ), the result is 87.149% all across at 100 second. Those results are totally different from the theoretically calculated results. Where did I get it wrong? I’m carefully about mesh I used, please see the picture 4 below of mesh size. I think it is a good strategy to use extra fine mesh at the separate boundary between the 2 gases. Also, the separation boundary is assume to have no thickness.


The COMSOL Multiphysics module used for this problem is “Transport of Concentrated Species”.



If you want try it, you can compile this simulation in 5 minutes following the parameters shown in the pictures, COMSOL version 5 or above will do.


Thank you for reading this.

[arjun]

It is hard to say what is going on but we can try to guess based on few things.

My first guess would be lack of convergence during time step. So I would just reduce the simulation timestep by factor of 10 and run the whole thing. If the things become better I would then think what shall be causing this lack of convergence.

[sbaffini]

As per the COMSOL status on this Forum, you might want to look at this certainly biased poll that took place in 2018. 49 users answered and multiple answers were allowed. Only the 6% (3 among 49 users) use it.

[andy.sun.wei]

Originally Posted by arjun

It is hard to say what is going on but we can try to guess based on few things.

My first guess would be lack of convergence during time step. So I would just reduce the simulation timestep by factor of 10 and run the whole thing. If the things become better I would then think what shall be causing this lack of convergence.

Hi, Arjun,


I think you might onto something. One thing I didn't mention in the original post is that I tried to solve the problem I described using "stationary" solver to find the steady state solution. It should come to a final stead state when fully mixed with certain gas concentration. But I found that every time I solve it, the results change. The gas 1 mass fraction changes every time, jumping from 0.67, 0.76, 0.87, 0.93 etc. It's very weird.


As to the "time dependent" solver, unfortunately it's unfortunately very stable. As you suggested, I reduced the time step from 0.1 to 0.01 second, the results are exactly the same.


I came up with this simplified gas mixing model as a sanity check. I've been doing some gas dissipation study into complex 3D geometry. Later on, I was doubting if the mass was balanced, meaning the final concentration didn't seem to add up to match the total amount of mass I put into there. That's why I did some sanity checks and came to this.



Thank you for you time.

[andy.sun.wei]

Originally Posted by sbaffini

As per the COMSOL status on this Forum, you might want to look at this certainly biased poll that took place in 2018. 49 users answered and multiple answers were allowed. Only the 6% (3 among 49 users) use it.

Thank you sbaffini for the interesting poll. I know COMSOL is not a strong content as a CFD simulation tool, still it's sad to see it's so low in the ranking.

[sbaffini]

With pure diffusion, a correct steady solver should reach the exact same solution that would eventually be reached by an unsteady solver, provided the boundary conditions allow that.

If boundary conditions don't allow that (say Neumann over all boundaries), there might be a visible effect linked to the numerical discretizazion, but for a given initialization it should be always the same.

Actually, excluding you are in certain unrealistic enviromental conditions that would affect your machine, even straightly wrong codes should still behave fully deterministically, unless explicitly instructed to not do so. So, results changing so wildly with runs suggests there might some more general issue.

[andy.sun.wei]

Originally Posted by sbaffini

With pure diffusion, a correct steady solver should reach the exact same solution that would eventually be reached by an unsteady solver, provided the boundary conditions allow that.

If boundary conditions don't allow that (say Neumann over all boundaries), there might be a visible effect linked to the numerical discretizazion, but for a given initialization it should be always the same.

Actually, excluding you are in certain unrealistic enviromental conditions that would affect your machine, even straightly wrong codes should still behave fully deterministically, unless explicitly instructed to not do so. So, results changing so wildly with runs suggests there might some more general issue.

Yes, I'm afraid I have made some general mistakes somewhere but I can't see it.

[andy.sun.wei]

please see attached file for this problem, in version 5.5.



due to the file size limit, the mesh is cleared too, it needs to be re-built before solve.


thank you for your help!

[andy.sun.wei]

Can any one give a quick test run on the uploaded program? Really frustrated about this wrong result.

Originally Posted by andy.sun.wei

please see attached file for this problem, in version 5.5.



due to the file size limit, the mesh is cleared too, it needs to be re-built before solve.


thank you for your help!

[sbaffini]

Originally Posted by andy.sun.wei

Can any one give a quick test run on the uploaded program? Really frustrated about this wrong result.

I don't have nor I ever had COMSOL, so I can't help here