[akidess]

By the way Andrea, another thing I'm not sure about is the relaxation described no page 11, 2.(e).(i.) - should we relax U and then determine phi, or determine phi and then relax it? What do you think?

- Anton

[Andrea_85]

i used adjustableRunTime.

andrea

[Andrea_85]

the authors are taking about "u" (u=0.7u^(n-1)+ 0.3u^n), actually i have seen you commented out the relaxation of phi in pEqn in last version. The problem is that eq 25 is not calculated explicitly at cell centres but it is just calculated once at face centres (eq 27). So the question is which velocity has to be relaxed? the only velocity we have at the cell centres is H/A, which is the predicted velocity so maybe they are referring to that velocity

andrea

[Andrea_85]

i made some other tests. I changed relaxation of phi in pEqn.H with relaxation of U=H/A, i added gradAlpha smooth in interfaceProperties.C and then i tryied to switch between mulesImplicit and mulesExplicit (actually i do not understand why you changed from explicit to implicit, the semi-implicit formulation to which they refer i think is related to the outerCorrector loop)
btw the results look better when mules explicit is used, see the attached picture.
the errors is capillary pressure are of the order of 0.02-0.04 which seems (more or less) in agreement with Fig 5 for deltax/R=0.2

best

andrea

[rcastilla]

I am making some testing with capillary rise. I am experiencing problems with the sign of the argument in sqrt in the factor w. I can overcome that with a higher value than 10^-6 in eq. (17), but I guess that is not a good solution. You have commented in a previous post that a better solution should be limiting the value of alpha1. How can that be done?

With respect to the sign of the curvature, I think that it should be enough to change the sign in line 125 of interfaceProperties.C :

K_ = -fvc::div(ns);

Is it correct?

Robert

[akidess]

Hey Andrea, I switched to MULES::implicit because the authors write they use Crank-Nicholson to advect the indicator function (page 10) - that implies that they use the implicit formulation as well.

What was the effect of changing the relaxation, and changing the smoothing? Can you check them separately? I won't have time until tonight to work on the code.

- Anton

Originally Posted by Andrea_85

i made some other tests. I changed relaxation of phi in pEqn.H with relaxation of U=H/A, i added gradAlpha smooth in interfaceProperties.C and then i tryied to switch between mulesImplicit and mulesExplicit (actually i do not understand why you changed from explicit to implicit, the semi-implicit formulation to which they refer i think is related to the outerCorrector loop)
btw the results look better when mules explicit is used, see the attached picture.
the errors is capillary pressure are of the order of 0.02-0.04 which seems (more or less) in agreement with Fig 5 for deltax/R=0.2

best

andrea

[akidess]

Robert, I pushed some more changes to address the problems you mentioned - have a look at the code.

Keep us posted on the capillary rise results!

Originally Posted by rcastilla

I am making some testing with capillary rise. I am experiencing problems with the sign of the argument in sqrt in the factor w. I can overcome that with a higher value than 10^-6 in eq. (17), but I guess that is not a good solution. You have commented in a previous post that a better solution should be limiting the value of alpha1. How can that be done?

With respect to the sign of the curvature, I think that it should be enough to change the sign in line 125 of interfaceProperties.C :

K_ = -fvc::div(ns);

Is it correct?

Robert

[rcastilla]

Anton,

have you tested it? I think that where you write

max(1.0, Foam::min(alpha1_, 0.0))

it should be

min(1.0, Foam::max(alpha1_, 0.0))

if functions max and min in OF are working as I expect.

On the other hand, w should be calculated with alpha1c instead of alpha1, isn't it?

Robert

[Andrea_85]

#Robert

I added these lines in interfaceProperties.C

volScalarField limitedAlpha1_
(
min(max(alpha1_, scalar(0)), scalar(1))
);


//factors
volScalarField w = Foam::sqrt(limitedAlpha1_*(1.0 - limitedAlpha1_) + 1e-6);
volScalarField factor = 2.0*Foam::sqrt(limitedAlpha1_*(1.0 - limitedAlpha1_)); // * pos(alpha1_-1e-6) * pos(0.9999999-alpha1_);


this should be enough to ensure a positive argument in the square root.

#Anton
i'll give a look at the two changes separately this afternoon if i have time. You can use mulesExplicit keeping crackNick as time scheme. why do you think a implicit formulation must be used when crackNick is selected? From my experiences i've never got good results using mulesImplicit.

andrea

[akidess]

Robert, you are right about the min/max, I messed it up! Sorry about that. If you have problems on the square-root of w use alpha1c there as well, but you should think about why you get unboundedness > 10e-6 and how much unboundedness you are willing to accept.

Andrea, if you limit alpha1_ and use it to determine w, you don't need to add 1e-6 into the square root any more ([edit] I take this back, you still have to avoid div-0[/edit]). About the time stepping - if you set MULES::explicit, you can remove the ddt(alpha1) entry from fvSchemes without the code complaining. Ergo, it doesn't matter any more what scheme you enter in the fvSchemes any more I think.

[Andrea_85]

you are right but apparently the results look better. the magnitude of spurious current is 10^-9 in liquid-liquid case when mules::explicit is used whereas is 10^-4 if implicit formulation is used. I am a bit confused.

andrea

[rcastilla]

Anton,

I get negative values of alpha1, up to 1e-5. With the same case in interFoam (not SSF) I didn't have this problem. The negative values of alpha1 where under 1e-40. Can that be due to the explicit MULES solver? I will try to switch to implicit solver for testing.

Robert

[akidess]

I do have the feeling that there is less unboundedness using MULES::implicit. See attachments for my results using MULES::implicit - it's very good for the liquid-liquid system, and less good for liquid-gas.

[Andrea_85]

I can't get the same results with implicit mules and i dont know why. Did you change something in fvSchemes or fvSolution?. Can you post again the static test case. my plots look really different from yours in case of mules::implicit.

andrea

[rcastilla]

Actually, I realized that it was yet with MULES::implicit. interFoam uses MULES::explicit. I have switched, and, though the simulation is very slow, it seems to behave better (at least the negative values of alpha1 are of the order of 1e-40.

Robert

[akidess]

Andrea, I pushed the testcase to the repository.

Robert, I didn't quite understand which one you think works better now - implicit or explicit?

[Andrea_85]

i lowered the tolerance in p, pFinal, pcFinal, Ufinal and alpha1 to match your fvSolution and i got really nice results, even in the gas-liquid case (why we get different results in this case is a mystery, maybe you are using different tolerance in two case). I used 1e-6 s for the time-step, 2 piso and 3 outCorr and mules::implicit. I also compared the case with 3 Piso and 4 outCorr for liquid-liquid system and results are the same but the sim takes much more time.

Then i compared the case in which it is performed a smoothig on gradAlpha (green curve in the first plot) with the case in which is not (yellow and blue curve in the first plot). they look both nice but different, I do not know exactly how to interpret them

best
andrea

[akidess]

Andrea, I'd be thankful if you could use an unmodified build of interFoamSSF and the test case (it takes about 30 min on my system to run) and check your results, and then try to isolate which of your modifications leads to a change of the convergence...

I'm using exactly the same settings in the liquid-gas and liquid-liquid cases, only the transport properties change.

- Anton

[edit] It would also be a good idea if you push your version of the code to a clone repository on google code so we both can compare [/edit]

[edit no 2] I think I narrowed it down: Can you cross-check constant/transportProperties? A liquid droplet immersed in a gas phase gives a different convergence than a gas bubble in a liquid phase!

[Andrea_85]

The two version we are using are the same, the only changes are PIMPLE controls. The test case is the same, the only difference at this point is the time step, it takes 5min wiht 1e-6 (I just checked the difference using 2 PISO and 3 outCorr as it is described in the paper, pag 11, but the results are the same so we can exclude it). I think i undrestood why we get different results in the liquid-gas case: you are simulating a liquid droplet immersed in a gas while, as in the paper, i'm simulating a gas droplet immersed in a liquid. check your transport properties and your setFieldDict. the properties are inverted.
btw attached you can find my results for liquid droplet immersed in a gas (unmodified test case) and they are equal to what you got in #73

best

andrea

[akidess]

You are right about the ordering of the phases. Do you think it's a concern that the liquid droplet performs worse?