Slow convergences of numerical integration of MFIE double layer on pair of triangles about beast.jl HOT 5 CLOSED

You are right, it does not choose the correct code path. It assumes trial and testing meshes are conforming and so it sometimes uses the wrong quadrature rules. For example, when a fine triangle is included in a large triangle, it erroneously relies on the SauterSchwabCommonVertex rule.

There is some support for nested meshes but it is very fragile. In particular it only works for barycentric refinements and fails quietly in other cases.

I will try to come up with a more robust strategy, but this will probably involve expensive on-the-fly generation of triangulations of intersections of trial and test elements.

from beast.jl.

For now I will create a new code path for the case you tested. Keep in mind this only works when Y is defined on the barycentric refinement of the mesh X is defined on.

from beast.jl.

I think this should be fixed by PR #81

from beast.jl.

I think there were still some issues correctly passing on the information on the quadrature order in the case that a Sauter-Schwab scenario occurs. PR #82 should fix this.

from beast.jl.

Excellent. Thank you. PR #82 is merged.

from beast.jl.