[Attesz]

Hi all,

if I set the final residual tolerances, finally the solver doesn't reach the prescribed level. However, if I adjust the tolerance/reltol in the simple U/p etc. it can reach whatever I want. Is it a bug maybe?

/*--------------------------------*- C++ -*----------------------------------*\
| =========                 |                                                 |
| \\      /  F ield         | OpenFOAM: The Open Source CFD Toolbox           |
|  \\    /   O peration     | Version:  2.1.x                                 |
|   \\  /    A nd           | Web:      www.OpenFOAM.org                      |
|    \\/     M anipulation  |                                                 |
\*---------------------------------------------------------------------------*/
FoamFile
{
    version     2.0;
    format      ascii;
    class       dictionary;
    object      fvSolution;
}
// * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * //

solvers
{
    pcorr
    {
        solver          GAMG;
        tolerance       1e-2;
        relTol          0;
        smoother        DICGaussSeidel;
        cacheAgglomeration true;
        nCellsInCoarsestLevel 10000;
        agglomerator    faceAreaPair;
        mergeLevels     1;
        maxIter         10;
    }

    p
    {
        $pcorr;
        tolerance       1e-5;
        relTol          0.01;
    }

    pFinal
    {
        $p;
        tolerance       1e-8;
        relTol          0;
    }

    
//parameters for k-omega model
  
    U
    {
        solver          GAMG;
        smoother        DILUGaussSeidel;
    nPreSweeps       0;
        nPostSweeps      2;
        tolerance        1e-4;
        relTol           0;
        cacheAgglomeration true;
        nCellsInCoarsestLevel 10000;
        agglomerator    faceAreaPair;
        mergeLevels     1;
        maxIter         5;
    }

    k
    {
        solver          GAMG;
        smoother        DILUGaussSeidel;
    nPreSweeps       0;
        nPostSweeps      2;
        tolerance        1e-4;
        relTol           0;
        cacheAgglomeration true;
        nCellsInCoarsestLevel 10000;
        agglomerator    faceAreaPair;
        mergeLevels     1;
        maxIter         5;
    }

    omega
    {
        solver          GAMG;
        smoother        DILUGaussSeidel;
    nPreSweeps       0;
        nPostSweeps      2;
        tolerance        1e-4;
        relTol           0;
        cacheAgglomeration true;
        nCellsInCoarsestLevel 10000;
        agglomerator    faceAreaPair;
        mergeLevels     1;
        maxIter         5;
    }
      
    "(U|k|omega)Final"
    {
        solver          PBiCG;
        preconditioner  DILU;
        tolerance       1e-8;
        relTol          0;
    }

}


SIMPLE
{
    nNonOrthogonalCorrectors 5;        //because the mesh is highly not orthogonal. maximum 20, 0 is cartesian hexahedral!
    pRefCell        0;
    pRefValue       0;
    residualControl
    {
        p               1e-4;
        U               1e-4;
        "(k|epsilon|omega)" 1e-4;
    }

}

potentialFlow
{
nNonOrthogonalCorrectors 10;
}

relaxationFactors
{
    fields
    {
        p               0.3;
    }
    equations
    {
        U   0.7;
    k   0.5;
    omega   0.5;
    }
}



// ************************************************************************* //

[marupio]

Try prescribing something nonesensical in pFinal to see if it's actually even reading it. e.g. tolerance banana;

[alberto]

I see where your confusion can come from. The SIMPLE algorithm does not have any "final" iteration, so the pFinal entry won't be read. The "Final" entries are only used in solvers based on the PIMPLE algorithm, where indeed each integration (time or pseudo-time) step has more than one iteration. In short, the linear solver will give you the final residual you require in "p".

Also, please note that if tolerance is too small, the linear solver will stop at 1000 iterations by default (you can change this value). If you set a high tolerance (say 1.0e-2), the solver will stop at the first iteration below that value, so it would not surprise me if that would be smaller (say something like 0.004).

P.S. I am not so sure it's convenient to use GAMG on anything but p. For steady-state solutions, smoothSolver seems to work better for me.

Best,

[Attesz]

Alberto, thanks for your comments. Indeed, I didn't see any pFinal nor GAMG for U in tutorial MRFSimpleFoam cases and on the Wiki.

Have a nice day,
Attila