Integrals over named internal faces behaves unexpectedly for DG
blakseth opened this issue · comments
For the purpose of solving a heat transfer problem with thermal contact resistance, I wish to include an integral over a named internal interface in my bilinear form. For standard discontinuous Galerkin, this produces unexpected results (cf. gists [1] and [2]).
In [1], I attempt to compute the lifting of an arbitrary CoefficientFunction that lives on an internal face. The resulting lifting function is non-zero only on neighboring elements on one side of the face (cf. [3]). I expected it to be non-zero on both sides. Moreover, the
In [2], I have copied the code from tutorial 2.8 [4] and applied it to a domain with a named internal interface. The definition of the domain is the only modification I have made to the code (apart from renaming some variables to distinguish DG and hybrid DG, and removing the wildcard import). I have not made any specifications with regards to the named interface, so the facets it contains should be treated as all other facets. For hybrid DG, this is what I observe, and I get the same output as when using the geom2d.unit_square as my domain. However, for DG, the results are not the same as for the geom2d.unit_square.
PS: I saw that you recommended HDG over DG on the forum recently. Since HDG appears to be working, I will try to make progress on my problem using that. In any case, I thought it would be nice if someone could look into the identified DG issue.
[1] https://gist.github.com/blakseth/4f6313b9d9e200801d29f4bf3bd3600e
[2] https://gist.github.com/blakseth/460bac09d60b064c6fa1938f5f63c0f5
[3] https://github.com/blakseth/images/blob/main/MWE%20Lifting%20output.PNG
[4] https://docu.ngsolve.org/latest/i-tutorials/unit-2.8-DG/DG.html
Best,
Sindre
Such specific questions are better suited for the forum: https://forum.ngsolve.org
It was not a question though. Just an observation about something that appeared to be a bug. Do you want bug reports on the forum instead of here?