[rdbisme]

Hello,

I'm trying to simulate a two-phase problem in a pipe using twoPhaseEulerFoam. I noticed that when I run the case in parallel and I have GAMG as solver for p_rgh the computation crashes at first iteration on p_rgh.

Seems related to this (http://www.openfoam.org/mantisbt/view.php?id=1247), but I'm on OpenFOAM 3.0.1.

Any of you had experience about that? There are additional things to setup in decomposeParDict?

I tried to setup preservePatches for the cyclic patches but no luck since now.

The serial simulation instead proceeds and gives reasonable results.

(Courant Number mean: 0.278509 max: 0.299811
Max Ur Courant Number = 0.170775
deltaT = 0.000320513
Time = 0.0131983

PIMPLE: iteration 1
MULES: Solving for alpha.SLN2
MULES: Solving for alpha.SLN2
alpha.SLN2 volume fraction = 0.13  Min(alpha.SLN2) = 0.0625242  Max(alpha.SLN2) = 0.236231
Constructing momentum equations
Pressure gradient source: uncorrected Ubar = 1.3, pressure gradient = 217.899
Pressure gradient source: uncorrected Ubar = 1.3, pressure gradient = 1280.75
min T.SLN2 63.1
min T.LN2 63.2
GAMG:  Solving for p_rgh, Initial residual = 0.282455, Final residual = 5.98708e+40, No Iterations 500
Pressure gradient source: uncorrected Ubar = -5.01445e+35, pressure gradient = 2.45967e+41
Pressure gradient source: uncorrected Ubar = -2.08706e+35, pressure gradient = 7.42086e+41
PIMPLE: iteration 2
MULES: Solving for alpha.SLN2
MULES: Solving for alpha.SLN2
smoothSolver:  Solving for alpha.SLN2, Initial residual = 2.96995e-17, Final residual = 6.90969e-18, No Iterations 1
alpha.SLN2 volume fraction = 1.02942e+45  Min(alpha.SLN2) = -7.5363e+59  Max(alpha.SLN2) = 4.85275e+59
Constructing momentum equations
Pressure gradient source: uncorrected Ubar = 2.34146e+22, pressure gradient = -6.61316e+84
Pressure gradient source: uncorrected Ubar = 4.48319e+21, pressure gradient = -6.54619e+84)

[anishtain4]

I'm having similar issue too. I'm running channel359 case with dynamicLagranian model, it runs well when I'm using 4 nodes, and crashes on 8 nodes. However it runs for a considerable number of time steps before it crashes. It just happened recently so I didn't know where it's coming from but now that I checked I have similar settings for pressure solving method.

I suspect it could be a result of domain decomposition as it is sensitive to number of nodes. I have used the scotch method. Is this the same method you are using too?

PS: running on OF3.0.1 too.

[rdbisme]

Originally Posted by anishtain4

I'm having similar issue too. I'm running channel359 case with dynamicLagranian model, it runs well when I'm using 4 nodes, and crashes on 8 nodes. However it runs for a considerable number of time steps before it crashes. It just happened recently so I didn't know where it's coming from but now that I checked I have similar settings for pressure solving method.

I suspect it could be a result of domain decomposition as it is sensitive to number of nodes. I have used the scotch method. Is this the same method you are using too?

PS: running on OF3.0.1 too.

Since your case crashes after a while it seems a problem with your settings. Try to lower the Courant number or the underrelaxation factors. Are the results with 4 nodes reasonable?

For the moment I solved the problem changing the solver for pressure:

    p_rgh
    {
        solver          PCG;
        preconditioner  DIC;
        tolerance       1e-6;
        relTol          0;
        minIter         1;
    }

[anishtain4]

results of solution in serial and with 2 and 4 nodes converge to a reasonable solution. But using 6 and 8 nodes one of the variables diverges in one step, so I assume it should be the domain decomposition that triggers the problem.

Thanks for sharing the solver method. I'll use it and will see how it goes.