tolerances too small for shape function checks

[jacobmerson]

The tolerances on these checks are quite strict and I found that they fail in many practical cases where Omega_h::barycentric_from_global is used.

meshFields/src/MeshField_Shape.hpp

Lines 22 to 28 in 26f923c

	KOKKOS_INLINE_FUNCTION bool greaterThanOrEqualZero(Array &xi) {
	auto gt = true;
	for (int i = 0; i < xi.size(); i++) {
	gt = gt && (xi[i] >= 0);
	}
	return gt;
	}

meshFields/src/MeshField_Shape.hpp

Lines 13 to 19 in 26f923c

	template <typename Array> KOKKOS_INLINE_FUNCTION bool sumsToOne(Array &xi) {
	auto sum = 0.0;
	for (int i = 0; i < xi.size(); i++) {
	sum += xi[i];
	}
	return (Kokkos::fabs(sum - 1) <= MeshField::MachinePrecision);
	}

For the cases I was working with I found setting the tolerances to 1E-12 was adequate. I don't have a good reason that this is a good tolerance to pick in general. I'm open to alternate suggestions.

[cwsmith]

OK. Thank you. We'll put together a small test case (assuming you don't have one on hand) and come up with a fix.

[jacobmerson]

Thanks. Based on a discussion with Mark, we are going to update the localization search to have a hierarchy of tolerances based on the topological entity type so we provide a consistent answer when close to the element edges. In this case we could clamp the parametric coordinate so that the coordinate is exactly on an edge or vertex.

[jacobmerson]

@cwsmith this is now blocking some things from moving forward in PCMS. Can you take a look at #79 ?

[cwsmith]

resolved with #79