What about the make_pressure_rhs_compatible function with prescribed velocities?

[gassmoeller]

What about the make_pressure_rhs_compatible function with prescribed velocities?

[tjhei]

The current logic is:
//TODO: The "correct" condition would be to not do the correction for compressible
// models if there are either open boundaries, or if there are boundaries
// where the prescribed velocity is so that in- and outflux do not exactly
// match
do_pressure_rhs_compatibility_modification = (material_model->is_compressible()
&&
(parameters.prescribed_velocity_boundary_indicators.size() == 0)
&&
(open_velocity_boundary_indicators.size() == 0));

And I don't understand why prescribed should be handled differently than zero boundaries. @bangerth ?

[bangerth]

Hm, I need to remind me again about that as well. The issue was that we rewrite the condition
div(rho u) = 0
as
div u = b
with some fixed b. Integrating over the domain and integrating by parts yields the condition
int_Gamma u.n = int_Omega b

So, if we have (i) zero boundary conditions, (ii) tangential boundary conditions, (iii) prescribed boundary conditions that happen to be tangential as is often the case, then the left hand side is zero and we need to make sure the same is true for the right hand side -- i.e., do the modification.

On the other hand, if there is are boundaries where there is some normal flow (i.e., there are open boundaries, or prescribed boundary conditions with unknown normal conditions), then the left hand side is nonzero and we cannot do the modification.

There is one case that we probably currently handle wrongly: if the user prescribes the velocity at some part but the prescribed values happen to be tangential (a common case) then we should do the modification, but don't right now. Similarly for cases where the flow happens to be not just purely tangential, but happens to cancel when integrated over the entire boundary.

I think one can construe another case where there is normal flow but left and right hand side are not compatible. In that case one would need to adjust the right hand side so that
int_Omega b = int_Gamma normal_component_of_prescribed_velocity
I think this is what @gassmoeller is referring to, right?

We could decide to just always do the rhs modification by always computing the boundary integral. Is this what you suggest, Rene?

Does anyone

[gassmoeller]

It seems like the end of your comment got truncated somehow. Anyway, if we would do the compatibility modification in any case with the integral of the normal velocity that would solve my problems very well (my models are in the described category that we currently handle wrong).
Maybe I missed a mathematical detail, but I just wonder what then happens in cases, where somebody prescribes a completely senseless boundary condition (like inflow at any boundary)? Would the compatibility correction you described then result in a solution that has no meaning, or would the solver still get a convergence error?

[bangerth]

As long as you have a compressible model, inflow on all boundaries is completely valid from a physical point of view -- you just increase the pressure and compress the material some more :-)

I do wonder what happens to our formulation, though. We make the assumption that the pressure is dominated by the hydrostatic pressure. I imagine that in the situation you describe you just get garbage.

In any case, I think doing the modification in all cases where there is no open/traction boundary makes sense but we need to compute the boundary integral, at least over all boundaries with prescribed velocity.

[gassmoeller]

Ok I guess, if you prescribe such boundary velocities you will notice soon enough that something is odd.

Yes then the boundary integral seems like a good way. Do you have some time to implement it before the release?

[gassmoeller]

Otherwise I would take a look at it, when I have some more time, but can not promise on when that would be.

[bangerth]

I'm afraid I won't. Let's postpone the issue until after the release.

[gassmoeller]

Is merged and done. I will close the issue.