[Libmesh-devel] Adaptivity on non-Lagrangian elements

[Roy S. <roy...@ic...>]

I've committed adaptivity support for Clough-Tocher elements this
morning, and support for Hierarchic elements tonight.

Since I had to make a couple library-wide changes to make that work,
I'd appreciate help regression testing.  Just doing a "cvs update" and
rechecking your own application code would be a good start; John
already caught one embarassing bug I'd missed.

In particular, I've mostly tested with Clough-Tocher elements, so if
anyone's got Hierarchic code to try, I'd appreciate hearing about the
results.  Regression testing with uniform refinement is most
important, but if you've got adaptive code that doesn't assume
Lagrangian elements (in boundary conditions, for example) and you want
to try switching to hierarchics (and possibly bumping up the
polynomial degree), that would be interesting too.

There's one change I'd like to make but haven't yet (since it would
hurt memory usage).  Currently the DofConstraintRow type uses floats
to store constraint entries.  This is adequate for the Lagrange
element constraints that fit in floats without rounding error, but the
constraints for Clough-Tocher elements depend on element angles and
should be stored in double precision.  If I were just writing my own
code I'd replace the floats with doubles and be done with it; however
before changing libMesh I thought it would be more polite and see if
I need to make this a compile-time or run-time selectable option
instead.
---
Roy Stogner