Negative Jacobians encountered in dynamic cuboid problem

[ikalash]

When running the single-domain AHeaD dynamic cuboid problem, negative Jacobians are encountered when adaptive time-stepping is disabled: https://github.com/sandialabs/Norma.jl/blob/main/examples/ahead/single/cuboid/dynamic/cuboid.yaml . This happens with dt = 0.01 and larger dt, even though the larger dts should work. When running the problem with adaptive time-stepping, the velocity and acceleration fields look wrong, as shown in the images below.

It is not 100% clear if the problem had issues from the beginning or if this started recently. It definitely ran initially with dt = 0.01 and no adaptive time-stepping.

[lxmota]

I think this issue and #50 are related, and that the problem is older, because even reverting to 060269f shows incorrect velocities and accelerations for me. I'm not sure how the tests pass. Working on this.

[ikalash]

@lxmota they could be related. I have been focusing on the clamped problem issue. I've narrowed it down a bit more where this commit is bad 25f3b46 whereas this commit is good c72ab22 .

[ikalash]

Wait, 060269f is a merge of Brian's work on contact + inclined support. The clamped problem was still OK after that merge.

[ikalash]

Also, I think we should add a test for the clamped problem once this is fixed where the disp, velo and acce are all compared to a reference. This will catch issues like this coming up in the future, as the clamped problem is a pretty stringent test case.