[behzad-cfd]

Hi dear foamers,

I want to solve magnetic potential equation.

I defined this equation like this:

fvScalarMatrix ksiEqn
(
    fvm::laplacian(interfacePsi.muM() ,ksi)
);
ksiEqn.solve();

Boundary conditions for achieving the uniform vertical magnetic field is:

inlet
    {
        type            zeroGradient;
    }

    outlet
    {
        type            zeroGradient;
    }

    up
    {
        type            fixedGradient;
        gradient       uniform -1234;
    }

    down
    {
        type            fixedGradient;
        gradient       uniform 1234;
    }

    frontAndBack
    {
        type            empty;
    }

fvSolution dictionary for ksi equation is:

ksi
    {
        solver           smoothSolver;
        smoother      DICGaussSeidel; //DIC-FDIC-GAMG-diagonal-none
        tolerance      1e-6;
        relTol            0;
    }

When i choose any solver except selected one (smoothSolver) after some time steps courant number rises dramatically and solver diverges. also with mentioned fvSolution setup, solving process is very slow and ksi equtaion iterates 1000 times in every time step!
The log file is attached.
Any suggestion is appreciated.

Thanks in advance,
Behzad

[behzad-cfd]

Hi,

I solved this problem by adding equation in a simple loop. If there is a better solution I'm very glad to know.

Regards,

Behzad

[Zeppo]

What kind of loop did you add? Could you please post the peace of code which solved your problem.

[behzad-cfd]

Originally Posted by Zeppo

What kind of loop did you add? Could you please post the peace of code which solved your problem.

Hi Sergei,

1- In your solver *.C file include "simpleControl.H".
2- Then create an object from simpleControl class like this:

simpleControl simple(mesh);

3- Put your equation in simple loop:

while (simple.correctNonOrthogonal())
{
        yourEqn.solve();
}

Don't forget to add simple control parameters in fvSolution.

SIMPLE
{
    nNonOrthogonalCorrectors 1; //or whatever number of correctors you set
}

I hope this could be helpful.

[Zeppo]

You are right. There should be 2 loops. The outer one to advance over time and the inner one to deal with mesh nonorthogonality which strongly affects laplacian discretisation. For the nonorthogolal correction loop to work don't forget to specify in fvSchemes dictionary:

gradSchemes 
{ 
    default         Gauss linear; 
} 
laplacianSchemes 
{ 
    default         Gauss linear corrected; 
}

[behzad-cfd]

Originally Posted by Zeppo

You are right. There should be 2 loops. The outer one to advance over time and the inner one to deal with mesh nonorthogonality which strongly affects laplacian discretisation. For the nonorthogolal correction loop to work don't forget to specify in fvSchemes dictionary:

gradSchemes 
{ 
    default         Gauss linear; 
} 
laplacianSchemes 
{ 
    default         Gauss linear corrected; 
}

Thanks for confirming. And yes I've set the laplacianSchemes as Gauss linear corrected.
Rgards,
Behzad