Multi-field in mixed-dimensional FE spaces

[oriolcg]

Gridap cannot handle a MultiFieldFESpace with single field FE spaces defined in different dimensions, e.g. a field defined in a domain Ω and another field defined on (part of) its boundary Γ⊆∂Ω. As @fverdugo pointed out:

Wait ⚠️, perhaps it is not as easy as that with the MultiFieldFESpace. I see a conceptual problem. when evaluating the shapefuns at some gps you now get a block matrix representing the different fields (which is what one would expect). This intrinsically assumes that all the fields can be evaluated at the same points... which is not true in your case

The error that I get when executing this case is:

ERROR: AssertionError: Incorrect number of point components
Stacktrace:
 [1] field_cache(::Gridap.Polynomials.MonomialBasis{1,Float64}, ::Array{VectorValue{2,Float64},1}) at C:\Users\ocolomesgene\.julia\packages\Gridap\0VcDN\src\Polynomials\MonomialBases.jl:117

[oriolcg]

In the latest version the error appears in another place, when defining the MultiFEFunction. In particular, there is a check in MultiFieldCellFields that fails when the fields are defined in different triangulations. This goes in line with the comments in previous post.

If I'm not wrong, having a MultiField defined in different triangulations (even if one is a subset of the other) would require a rethinking / deep refactoring of the MultiField data structure, right @fverdugo? Is there a "simple" alternative to that? Maybe use the union triangulation and "fix" the DOFs that don't belong to each triangulation? (not sure if this would work in general and for sure it isn't the most efficient/clean way)

[fverdugo]

I think this has already been fixed in #567