Did not fix the Fedora 33 build, just keeping it commented.

Allow pointing to existing dXvdXs, or recompute it.

Note that it does not give the correct dRdX for h-adaptive tests.

Please refer to issue:
#48 (comment)

for more details.

Allows the use of ld.gold with Ubuntu's MPI bug.

First, use cell volume to determine cell stiffness.
Second, use better linear solver parameters.

Do not be so quick to accept a higher residual. Try out other direction first.

And some snippets of code to test out the h-adaptivity.

At compile time, amplitude isn't necessarily seen.

Inverted cells will results in normals pointing in the same direction instead of opposite direction.

In that scenario, just return a huge residual to signify that the constraints are not satisfied.

…ers.

Also allows to pass a precomputed dXvdXp matrix at construction.

CFL should be negative in this scenario since we use dRdW = -R

In the ODE solver, we current use M/dt - dRdW = R

Therefore, the full-space basically switched sides of dRdW and R, and the mass matrix sign must be changed consequentially.

Allows working dRdX test in case of hanging nodes.