fiat team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Tue, Jan 22, 2008 at 12:23:29PM -0600, Robert Kirby wrote:
> 1.) It will get rid of all of the if numbering == "UFC" type code that has been
> inserted into FIAT.   These ifs are difficult to maintain and have crept into
> several parts of the code.
> 2.) Within FFC, you will instantiate a FIAT "ReferenceElement" such as 
> FIAT.reference_element.UFCTriangle() and pass that to the constructor for the
> particular finite element.  Inside, it will automatically handle the
> transformations and ordering.

Sounds excellent!

Maybe it's time for a release soon since all the new spaces seem to
work now (BDM, BDFM, RT, Nedelec) so perhaps this could wait until
after the release?

We still haven't implemented the code generation for evaluate_dof
which depends on getting dof information from FIAT (which will enable
us to set essential BCs for these elements) but depending on how that
goes, it could be included before or after the release. (?)

-- 
Anders


> On Jan 22, 2008 12:17 PM, Marie E. Rognes < meg@xxxxxxxxxxx> wrote:
> 
>     Robert Kirby wrote:
>     > This should replace the transformed spaces.  FIAT's client would say
>     > that they want a 5th order Lagrange over such and such a reference
>     > element.  I would provide the "old" reference element and the UFC
>     > reference element as particular cases.
>     >
> 
> 
>     How does this differ from the setup before the transformed spaces?
> 
> 
>     > We were facing problems in both geometry and topology in the FFC/FIAT
>     > hookup.  The numbering.py module attempted to solve the topology
>     > problem and the transformedspace.py module the geometry problem.
>     >
>     > I think we should be able to have a single point of entry for all the
>     > geometry & topology information about reference elements in FIAT.
>     >
>     > On Jan 22, 2008 11:58 AM, Anders Logg <logg@xxxxxxxxx
>     > <mailto:logg@xxxxxxxxx >> wrote:
>     >
>     >     On Tue, Jan 22, 2008 at 11:48:23AM -0600, Robert Kirby wrote:
>     >     > Hi, with all of the difficulties regarding (0,1) vs (-1,1) and
>     >     handling
>     >     > different kinds of reference orderings,
>     >     > we seem to have scattered reference element info in about 3
>     >     different modules.
>     >     >  It might be a good time
>     >     > to refactor some of this.  I propose having a "ReferenceElement"
>     >     class that
>     >     > will encapsulate the
>     >     > geometry and topology of the particular reference elements,
>     >     parametrizing the
>     >     > rest of FIAT on top of
>     >     > this.  This will make FIAT much cleaner and more extensible for
>     >     people who want
>     >     > other orderings.
>     >     >
>     >     > Rob
>     >
>     >     Is this instead of or in addition to the transformed spaces?
>     >
>     >     If instead of, this is what we started out with (trying to
>     parametrize
>     >     FIAT over the reference element).
>     >
>     >     At some point, we may also want to think of having some common module
>     >     or interface for finite elements. Now, each of FIAT, SyFi, FFC, UFC
>     >     and UFL have a FiniteElement class of their own, some of which are
>     >     wrappers for the others.
>     >
>     >
>     >
>     > ------------------------------------------------------------------------
>     >
>     > _______________________________________________
>     > FIAT-dev mailing list
>     > FIAT-dev@xxxxxxxxxx
>     > http://www.fenics.org/mailman/listinfo/fiat-dev
>     >
> 
> 
> 

> _______________________________________________
> FIAT-dev mailing list
> FIAT-dev@xxxxxxxxxx
> http://www.fenics.org/mailman/listinfo/fiat-dev