ffc team mailing list archive

["Garth N. Wells" <gnw20@xxxxxxxxx>]

Kristian Oelgaard wrote:
> Quoting "Garth N. Wells" <gnw20@xxxxxxxxx>:
>
> > Anders Logg wrote:
> > > On Fri, Apr 17, 2009 at 03:57:09PM +0800, Garth N. Wells wrote:
> > > > Anders Logg wrote:
> > > > > On Fri, Apr 17, 2009 at 03:07:17PM +0800, Garth N. Wells wrote:
> > > > > > Anders Logg wrote:
> > > > > > > On Fri, Apr 17, 2009 at 08:59:43AM +0200, Johan Hake wrote:
> > > > > > > > On Friday 17 April 2009 03:32:26 Garth N. Wells wrote:
> > > > > > > > > Should we start moving to UFL-based forms for the DOLFIN demos? Is
> > the
> > > > > > > > > only serious outstanding issue on the FFC side the correct
> > determination
> > > > > > > > > of the quadrature order?
> > > > > > > > Not sure how this could be best done in PyDOLFIN.
> > > > > > > >
> > > > > > > > We could add
> > > > > > > >
> > > > > > > >   from dolfin.ufl import *
> > > > > > > >
> > > > > > > > after each
> > > > > > > >
> > > > > > > >   from dolfin import *
> > > > > > > >
> > > > > > > > in the demos. When this is done and it all just works(TM) we can move
> > the
> > > > > > > > files in the site-packages/dolfin/ufl directory down to
> > site-packages/dolfin,
> > > > > > > > and remove
> > > > > > > >
> > > > > > > >   from dolfin.ufl import *
> > > > > > > >
> > > > > > > > from the demos?
> > > > > > > >
> > > > > > > > Johan
> > > > > > > I suggest we just move everything to UFL at once and then solve any
> > > > > > > problems that we encounter.
> > > > > > >
> > > > > > > I'm intending to remove the .form support from FFC quite soon (maybe
> > > > > > > today).
> > > > > > >
> > > > > > > It's easier to fix things so that they work with UFL than to have two
> > > > > > > conflicting form languages work at the same time.
>
> I agree on this.
>
> > > > > > The quadrature issue is serious, so I think that it should be tackled
> > > > > > before the transition, or at least a short-term plan for how to solve it
> > > > > > should be in place.
> > > > > >
> > > > > > Garth
> > > > > Is it so serious? It's a small thing compared to what will happen when
> > > > > the .form support is removed from FFC (as in everything will be broken
> > > > > until fixed).
> > > > Results computed with the new and old formats will not, in general, be
> > > > the same, so I considered it to be serious.
> > > Yes, they won't be the same but I thought they would still be correct
> > > (as in possibly using a better quadrature rule than needed).
> > No, it uses a lower degree of quadrature. For example, for the Poisson
> > demo using the new UFL format only one quadrature point is used for the
> > source term, whereas enough points for a quadratic polynomial are required.
> >
> > Not quite sure what 'not the same but still correct' means ;).
>
> As far as I can tell the current implementation in UFL will result in a
> sufficient degree of quadrature being used.

I've tested it, and not in all cases are enough quadrature points used 
for exact integration.

Martin made some changes to UFL yesterday to help in getting the 
quadrature order right.

Garth

> Kristian
>
> > Garth
> >
> >
> > > Kristian?
> >
> > _______________________________________________
> > FFC-dev mailing list
> > FFC-dev@xxxxxxxxxx
> > http://www.fenics.org/mailman/listinfo/ffc-dev