[jameswilson620]

All,

I have modified interFoam to include a different form for the CSF, surface tension force. Instead of a force that is proportional to the interface curvature, I am simply adding a constant field where the sign may be positive or negative, reversing the direction of the force.

In an attempt to remedy the problem, I removed features from my solver only to find that the problem lies solely in the surface tension force field.

In pEqn.H from interFoam 2.3.0, the CSF is represented by:

    
    surfaceScalarField phig
    (
        (
            fvc::interpolate(interface.sigmaK())*fvc::snGrad(alpha1)
          - ghf*fvc::snGrad(rho)
        )*rAUf*mesh.magSf()
    );

where the surface force is bolded.

Here is what i have added in interFoam:

    surfaceScalarField phig
    (
        (
            fvc::interpolate(surfaceTensionForce)*fvc::snGrad(alpha1)
          - ghf*fvc::snGrad(rho)
        )*rAUf*mesh.magSf()
    );

I have defined surfaceTensionForce in createFields and it is updated in the main .C file as series of constants that results in a uniform field:

			
        surfaceTensionForce = interface.sigmaK();
        forAll(surfaceTensionForce,celli)
	{
	    if((porosity[celli]!=1)&(capillaryPressurePorous))
	    {
        	surfaceTensionForce[celli] = -porosity[celli]*4*interface.sigma().value()*Foam::cos(contactAnglePorous*2*3.14159/360)/dPores.value();
            }
		    		
        }

As you will see in images 4 and 5, this operation is carried out successfully.

So heres the situation:

In serial mode, this implementation is flawless. In parallel, the capillary force at the liquid/vapor interface, at the processor boundary, seems to be too high (i believe), resulting in the regression of the liquid/vapor interface back to the processor boundary. I have tried scotch and simple with 3-6 nodes. there is no problem in regions using the traditional interface.sigmaK() field, only in the updated, uniform cells along processor boundaries.

I have attached a few images:

1. Initial starting point for the simulation. By design, the liquid/vapor interface is set ON TOP/ACROSS the processor boundary so that the very next write interval will reveal whether or not any solver remedies/modifications have worked.

2. next write interval in SERIAL. Gravitational and capillary forces pulling the liquid/vapor interface down as expected.

3. same as 2. but PARALLEL. Note that the processor boundary is horizontal and aligns with the flat spots on the liquid/vapor interface. This behaviour is incorrect and is definitely inconsistent with the volScalarField, surfaceTensionForce that I have defined.

4. This is the surfaceTensionForce field in SERIAL. Note that interface.sigmaK() can be seen as active in the upper portion of the domain where the modified region (uniform value of -398) is active in the lower region.

5. same as 4. but in PARALLEL. Note the uniform, properly defined, surfaceTensionForce field in the lower region where the interface gets, for a lack of a better word, wonky.

The new surfaceTensionForce field is simply a modified version of the original, i.e. the field is a place holder for whatever i want to put in it, so long as the units are consistent. I can't figure out what is happening leading up to interpolation in pEqn.H (which I will be investigating next), i.e. the volScalarField looks great so I cant explain why modifying it causes the solver to fail at the processor boundaries in the uniform region.

Any help or suggestions are welcome.

Thank you, James W.