Hi Sander,
I think it is not working by now, but we are not far away ...
(A) I would use a hybrid version, with an additional variable (approximating the state) on the interface. Either H1(definedonbound=...), or better FacetFESpace(definedonbound=..), what is currently missing. This does not need additional matrix entries (no dgjumps flag required).
(
I would use skeleton-DG, on BND, for integration. We have to tweak it to evaluate on both sides of the interface. Have to think how we can add this, without breaking backward compatibility.
Christoph, Michael, what do you think ?
Joachim