gmsh-generated unstructured hexahedral mesh fails Gauss's Law test

[MTCam]

The attached simple mesh was created in gmsh by splitting tets into hexs. It has multiple volumes. It fails the Gauss's Law test in grudge, and fails to hold a quiescent fluid state in mirgecom, even with inviscid operators.

so_weird.msh.txt

>       assert abs(int_1 - int_2) < 1e-13
E       assert cl.TaggableCLArray(1.38647505e-09) < 1e-13
E        +  where cl.TaggableCLArray(1.38647505e-09) = abs((cl.TaggableCLArray(0.00285925) - cl.TaggableCLArray(0.00285925)))

The only weird thing I can find about this mesh is that its faces aren't exactly coplanar. The worst face I found was basically a tet with a volume of about -1e-11.

[inducer]

As discussed during meeting: I think this may be explained by not accurately integrating the mapping Jacobians (and other geometric factors). If increasing the quadrature order gets this to be accurate, then IMO there's nothing to see here. (But please do this experiment and follow up.)

[MTCam]

As discussed during meeting: I think this may be explained by not accurately integrating the mapping Jacobians (and other geometric factors). If increasing the quadrature order gets this to be accurate, then IMO there's nothing to see here. (But please do this experiment and follow up.)

FYI, the ACTII mesh that MikeA showed in the meeting does pass the Gauss's Law test. 😮

[MTCam]

Here's another, simpler mesh that reproduces the Gauss's Law test failure. Overintegration did not appear to have an affect on the results for any of the hex meshes. Setting ($p=3$) actually gets a passing test.

Here's the hex version that fails the test for ($p<3$):
simple_hexs.msh.txt

Here's the tet version that the hex mesh was generated from:
simple_tets.msh.txt