NaN values appear in the assembled system when facet integrals are included. about firedrake

[kynan]

I can reproduce that issue and looking at the wrapper code we generate for the interior facet integral I strongly suspect it's a firedrake bug. In fact looking through our existing test cases it seems (interior) facet integrals on vector function spaces are untested and therefore broken.

[wence-]

I think this is caused by #165.

[dorugeber]

Facet integrals won't support VFS for quite a while. Use a proper space, like BDM ;-)

[dorugeber]

@ctjacobs do you still get this error?

I cannot reproduce this; it was probably 'fixed' by a combination of
OP2/PyOP2#319
https://bitbucket.org/mapdes/ffc/pull-request/17/fix-problems-with-existing-facet-integral
https://bitbucket.org/mapdes/ffc/pull-request/21/fix-ir-for-interior-facets

Note that you want n = FacetNormal(mesh) instead of n = FacetNormal(mesh.ufl_cell()).

[dorugeber]

(I expect that the calculation is still wrong, though)

[wence-]

There will now, hopefully, not be a NaN error. But the calculation will still be wrong because we haven't fixed interior facet integrals on vector function spaces yet. That's still #165.

[ctjacobs]

The NaN issue has indeed been fixed.

[dorugeber]

What remains is a duplicate of #165, so I will close.