ufl team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Mon, May 30, 2011 at 10:10:00PM +0200, Martin Sandve Alnæs wrote:
> On 30 May 2011 21:39, Anders Logg <logg@xxxxxxxxx> wrote:
> > How is this mapping created? Which degree is chosen for the elements?
> > If it's the same as currently used in FFC, we can drop that and rely
> > on UFL.
>
> It only adds missing cells. I think which degrees to choose is not
> something UFL should guess at, that is completely in the form
> compiler realm. In particular I think FFC should choose quadrature
> elements for quadrature representation and CG elements for tensor
> representation, which are concepts we do not want to introduce in UFL.
>
> But the original bug report here was that FFC does not pass any
> element mapping at all, so if you need to map degrees you should
> look into that.

So then what's the point of this function if it can't be used by a
form compiler? The form compiler will have to check for missing degree
and supply the mapping which will override any mapping that UFL
computes.

> > Here's what's used in FFC:
> >
> >  # Extract common degree
> >  common_degree = max([e.degree() for e in elements])
> >  if common_degree is None:
> >      common_degree = default_quadrature_degree
> >
> >  # Degree must be at least 1 (to work with Lagrange elements)
> >  common_degree = max(1, common_degree)
>
> So if f and v are CG2 functions, then f*v*dx which is an up to
> 4th degree polynomial, is integrated with 2nd degree accuracy?

No, it's integrated with with a quadrature rule exact for degree 4
polynomials. But for the form c*f*v*dx where c is an undefined
expression, c would be interpolated into P2, and then a quadrature
rule exact for polynomials of degree 6 is used.

--
Anders




> Martin
>
>
> >
> >
> > ---------- Forwarded message ----------
> > From: "Martin Sandve Alnæs" <785143@xxxxxxxxxxxxxxxxxx>
> > To: logg@xxxxxxxxx
> > Date: Mon, 30 May 2011 11:35:32 -0000
> > Subject: [Bug 785143] Re: Element mapping is not provided to ufl.Form.compute_form_data()
> > In an upcoming commit, the element mapping is now constructed
> > automatically by ufl in preprocess in the case of invalid cells. Thus
> > ffc (and sfc) do not need to pass a mapping unless you want to override
> > things like degree and family.
> >
> > Note that there may still be issues with the above code snippet, but
> > I'll close this bug and report eventual other bugs separately if
> > necessary.
> >
> > ** Changed in: ffc
> >       Status: Confirmed => Invalid
> >
> >
> > Title:
> >  Element mapping is not provided to ufl.Form.compute_form_data()
> >
> > Status in FEniCS Form Compiler:
> >  Invalid
> >
> > Bug description:
> >  FFC does not provide an element mapping to
> >  ufl.Form.compute_form_data(). This is needed when used from PyDOLFIN,
> >  to make forms well defined when coefficients are missing cells.
> >
> >  I have some failing code, will need to simplify it before I post it
> >  here.
> >
> > _______________________________________________
> > Mailing list: https://launchpad.net/~ufl
> > Post to     : ufl@xxxxxxxxxxxxxxxxxxx
> > Unsubscribe : https://launchpad.net/~ufl
> > More help   : https://help.launchpad.net/ListHelp
> >
> >