Inconsistency in boundary conditions for nonconservative terms

[andrewwinters5000]

When computing the boundary conditions for systems with nonconservative terms there is a potential to accidentally add the boundary contribution incorrectly on P4estMesh (https://github.com/trixi-framework/Trixi.jl/blob/main/src/solvers/dgsem_p4est/dg_2d.jl#L365-L371 ) or UnstructuredMesh2D (https://github.com/trixi-framework/Trixi.jl/blob/main/src/solvers/dgsem_unstructured/dg_2d.jl#L443-L449).

If one does not use the surface_flux_function or nonconservative_flux to compute the boundary condition, e.g. as we do for something like the Euler slip walls (https://github.com/trixi-framework/Trixi.jl/blob/main/src/equations/compressible_euler_2d.jl#L299-L337), then the boundary condition will incorrectly be added as something like

surface_flux_values = flux + 0.5 *  noncons

because flux and noncons will be identical and we actually get the boundary flux one and a half times.

It would be better from my perspective if we did not assume / require that the boundary condition is set via a numerical flux for the sake of flexibility.

[andrewwinters5000]

@ranocha I think this is related to issues that Marco discussed at the last Trixi meeting.

[andrewwinters5000]

As a starting idea, it would be handy if the boundary condition routine(s) could use a different surface_flux than that used for the interior interfaces taken from surface_integral. This would already grant extra flexibility.

[ranocha]

CC @MarcoArtiano

[ranocha]

As discussed in the last meeting, we should consider passing (flux_conservative, flux_nonconservative) as flux and call the BC only once. This would need to be tested in practice and documented appropriately.

[MarcoArtiano]

Unfortunately I couldn't join the Trixi Meeting yesterday. My suggestion and my understanding would have been that I could theoretically use an anti-symmetric volume flux for the non-conservative part, that is consistent with 0, so that no boundary conditions needs to be defined for the non-conservative flux. As far as I know, every non-conservative formulation in Flux Differencing form can be written as a product of a symmetric flux times a jump (i.e. anti-symmetric flux) (at the moment I don't know of any non-conservative flux that doesn't boil down to this form. This is basically, an equivalent form to the one presented here, https://arxiv.org/pdf/2211.14009, where a non-conservative flux can be written as a product of a symmetric times a local term). The anti-symmetric "formulation" would be also consistent with the point-wise source term, where no boundary conditions are imposed. However, that would require to modify all the existing non-conservative fluxes and might not be the best idea at the moment.

Anyway, I can definitely work on the suggestion presented and pass the tuple directly to the boundary condition call and return the correct value flux + 0.5*cons.

[andrewwinters5000]

Anyway, I can definitely work on the suggestion presented and pass the tuple directly to the boundary condition call and return the correct value flux + 0.5*cons.

That would be great! Feel free to ping me for a review whenever you have a PR ready.

[patrickersing]

My suggestion and my understanding would have been that I could theoretically use an anti-symmetric volume flux for the non-conservative part, that is consistent with 0, so that no boundary conditions needs to be defined for the non-conservative flux.

Even with anti-symmetric non-conservative terms, these don't need to be consistent with zero at the boundary, since u_ll and u_rr may differ and the jump term is non-zero. So there are actually some cases where boundary conditions need to be imposed for the nonconservative terms.

[MarcoArtiano]

My suggestion and my understanding would have been that I could theoretically use an anti-symmetric volume flux for the non-conservative part, that is consistent with 0, so that no boundary conditions needs to be defined for the non-conservative flux.

Even with anti-symmetric non-conservative terms, these don't need to be consistent with zero at the boundary, since u_ll and u_rr may differ and the jump term is non-zero. So there are actually some cases where boundary conditions need to be imposed for the nonconservative terms.

You are definitely right! I didn't consider that option, it would not allow for such flexibility.