Finite volume interface kernels for porosity jumps, contact, ..

[GiudGiud]

Reason

For many different physics, we do not want to resolve sharp transitions at interfaces with the mesh, and would rather introduce a discontinuity that captures this transition.

Design

A kernel that only contributes to the residual on faces that belong to an interface.
There will probably be many similarities to FVFluxKernels, which only act on (all element's) faces.

Impact

Huge boost to finite volume modeling of nuclear reactors

[lindsayad]

@GiudGiud the way I handled this for CNSFV was to apply the divergence theorem to the epsilon * grad_pressure term. The resulting epsilon*pressure should be much smoother so I just use the same epsilon-pressure FVFluxKernel across the whole domain with no special treatment at porosity jump interfaces.

[lindsayad]

Although the math is not quite right on that, is it...

[lindsayad]

Actually the math is good as long as epsilon doesn't have a gradient within a cell. If it does, then you would need an additional term:

-\int Ip * \nabla \epsilon dV

The problem is in implementation...we know that Green-Gauss computations of either grad_p or grad_eps will be problematic near interfaces where both p and eps are discontinuous

[GiudGiud]

What do you do at boundaries then? Is that kernel executed on boundaries?

I think Paolo wanted to have the option to define a form loss coefficient too.

We dont need grad_eps for this term?

[lindsayad]

But at least in the case of epsilon it will almost always be constant within cells or at worst have a user-prescribed function within a cell whose gradient is clearly defined...e.g. we wouldn't have to do Green-Gauss with information from discontinuous faces

[lindsayad]

What do you do at boundaries then? Is that kernel executed on boundaries?

Yes in general this FVFluxKernel or rather a FVFluxBC relative must be executed on boundaries.

I think Paolo wanted to have the option to define a form loss coefficient too.

In this case we would want a FVInterfaceKernel for sure

We dont need grad_eps for this term?

Which term?

[GiudGiud]

The momentum pressure term.
EDIT: ok it is in the extra term you would need to add