dolfin team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Wed, Feb 16, 2011 at 09:21:23AM -0800, Johan Hake wrote:
> On Wednesday February 16 2011 09:07:57 Garth N. Wells wrote:
> > On 16/02/11 16:54, Johan Hake wrote:
> > > Question #145534 on DOLFIN changed:
> > > https://answers.launchpad.net/dolfin/+question/145534
> > >
> > > Johan Hake proposed the following answer:
> > >
> > > On Wednesday February 16 2011 08:48:38 Anders Logg wrote:
> > >> Question #145534 on DOLFIN changed:
> > >> https://answers.launchpad.net/dolfin/+question/145534
> > >>
> > >>     Status: Open => Answered
> > >>
> > >> Anders Logg proposed the following answer:
> > >>
> > >> On Tue, Feb 15, 2011 at 10:30:06PM -0000, Yi Jiang wrote:
> > >>> New question #145534 on DOLFIN:
> > >>> https://answers.launchpad.net/dolfin/+question/145534
> > >>>
> > >>> Hi all,
> > >>>
> > >>> I am using the dolfin to solve a PDE problem by finite element method.
> > >>> However, I met a problem when I try to define the linear form L(v) of
> > >>> right hand side. Suppose the PDE is:
> > >>>
> > >>> F(x) = f.
> > >>>
> > >>> What should I do if the f is a Dirac Delta function. How to define it
> > >>> in ufl? Anybody met and solved the similar problems before?
> > >>> Thank you for attention and kind help.
> > >>
> > >> This is not supported.
> > >
> > > I am also interested in this. What would it require to implement? Changes
> > > to UFL, FFC, DOLFIN?
> >
> > A lot of changes. It would require being able to integrate over
> > surfaces/lines within a cell.
> >
> > UFL can be stretched to do this, but extentions to FFC and DOLFIN are
> > needed. Take a look at the modelling discontinuities chapter in the
> > FEniCS book ;).
>
> Looking forward to do that!
>
> A naive thought would be to introduce a DeltaDiracFunction which evaluates to
> zero for all but a given coordinates which need to coincide with a vertex. But
> I guess we need to prevent DOLFIN to interpolate the function during assemble,
> and just evaluate it at the quadrature points, where it evaluates to zero?
>
> Johan

Anyone interested in this, please add a blueprint so we can remember
it for later.

--
Anders