*) If update_usermod is used to set a parameter that is used in the definition
of the true solution, the graphics process does not get the new parameter, so
its display of the true solution and true error is wrong.

*) The quadrature in unowned elements is not exactly the same as the quadrature
in the processor that owns the element, which causes an inconsistency in
the parallel hierarchical basis multigrid method so that it does not reduce
the residual to 0.  It does, however, seem to reduce the error below the
discretization error, so the method does provide an adequate solution.  But
in some cases the test for termination of multigrid iterations based on the
size of the residual is never satisfied, so you should (for now) put a limit
on the number of iterations via the mg_cycles parameter to phaml_solve_pde,
e.g. mg_cycles=50, if running in parallel.

*) For systems of equations, the L1 sum and L2 sum of the components in the
graphics is wrong if using high order elements and subelement resolution.

*) The PETSc preconditioners ICC and ILU generate the PETSc error code 56.

*) the PETSc preconditioner Eisenstat crashes with 3D problems.

*) the test suite is not available for SPMD model

*) With Open MPI, the graphics test suite will report the tests failed, even
though they succeeded, because mpi_finalize is not called because, for some
unknown reason, the master and slaves get stuck in mpi_finalize if graphics
is used.

*) BLOPEX can have problems with agressively optimized versions of LAPACK,
because the optimization can break an invariance property that BLOPEX assumes.
If you find that "make test what=slepc" gives results that are significantly
different from the BLOPEX comparison files, try using "BLAS source LAPACK
source" when you run mkmkfile.sh.

*) the explicit and equilibrated residual error estimators do not get
the optimal convergence rate when the cross derivative term cxy is nonzero.
Use one of the other error estimators.

*) the color key does not get resized when the graphics window is resized.
You have to toggle the key off and back on again to resize it, or better yet,
resize the window with the key off and then toggle it on.

*) LAM and OpenMP do not work together, because LAM is not thread safe.  Try
using Open MPI instead of LAM.

*) isosurface and face transparency requires the drawing primitives be sorted
from back to front.  To reduce computation time, PHAML sorts the elements
instead, which may cause some artifacts.  More importantly, it does not
resort them while the image is rotated, so a sufficiently large rotation
will destroy the image.  In this case, redraw the image.

*) If both PETSc and ZOLTAN are used, and PETSc was built with download-parmetis
(which is required with download-mumps or download-superlu), then ParMETIS
will not work with ZOLTAN.

*) hypre's PILUT and EUCLID don't work.

*) In 3D it is possible for phaml_evaluate to report that a point is not in the
domain (i.e. return a 0 solution value) when it is contained in an element with
a circular boundary.
