On Wed, May 30, 2012 at 2:40 PM, Roy Stogner <roystgnr@ices.utexas.edu> wrote:

On Wed, 30 May 2012, John Peterson wrote:

It might be possible to have a more complete separation between the
mapping aspects of the FE class and the approximation space aspects,
i.e. have an "FEMap" base class and corresponding hierarchy?

I personally would be thrilled if we could generalize to non-Lagrange
maps.  I haven't drunk the NURBS-approximation-spaces Flavor-Aid yet,
but supporting NURBS geometric mappings would be fantastic.

I think it's a red herring from the vector valued elements questions,
however.  The (Lagrange-or-NURBs-or-high_p-etc) option on mappings
from reference to physical space ought to be completely independent of
the (covariant-or-contravariant-or-piola) option for the subsequently
induced mappings from reference to physical vectors (or tensors).

You're correct here I think. Although my comment before applies to H(curl), the actual transformation is different than the H(div) case, so seems to make sense to generalize the mapping concept.

But there's no reason to worry about that potential design change
while you are already immersed in creating FEAbstract... maybe it
could be looked at again later if it seems useful.


Cool. I think it'll come up again in a big hurry, but I think it will be better/easier to deal with once the details become more concrete.

FYI, I can reproduce the miscellaneous_ex6 problem in absence of my patch on my mac, so it seems orthogonal.

Other comments before I commit?