Very interesting. At this very moment I happen to have 30,000 cores
worth of jobs queued on our supercomputer that match this description!
We have noticed some problems when trying to do mesh refinement
studies for hermites when starting with analytic initial conditions
(like your smooth cross).
A quick warning for those using 2D and/or 3D HERMITE elements:
there was a bug, now fixed in the svn head, in
System::project_vector(fptr,gptr,etc), which caused the
mixed-derivative-based degree of freedom coefficients to be set to
zero rather than calculated from finite differencing of the gptr
