dolfin team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Thu, Apr 08, 2010 at 08:44:21AM -0700, Johan Hake wrote:
> On Thursday April 8 2010 08:36:24 Marie Rognes wrote:
> > Johan Hake wrote:
> > > On Thursday April 8 2010 04:51:24 Marie Rognes wrote:
> > >> There is something suboptimal with regard to assembly over sub_domains
> > >> specified
> > >> by domains (in contrast to markers)
> > >>
> > >> Say I have some functional
> > >>
> > >> (*)    M = f*ds
> > >>
> > >> where f is some function. Let 'domain' be some sub-domain.
> > >>
> > >> Now, it is not clear to me what
> > >>
> > >>     v = assemble(M, mesh=mesh, exterior_facet_domains=domain)
> > >>
> > >> gives. It would be really convenient (for my purposes) if it gave
> > >> the same as
> > >>
> > >>     v = assemble(M, mesh=mesh)
> > >
> > > Considering that ds defaults to ds(0) I think it is a logical behaviour.
> >
> > So
> >
> >     ds == ds(0)
> >
> > ?
>
> In ufl/objects.py
>
>   ds = Measure(Measure.EXTERIOR_FACET, 0)
>
> So I guess yes.

Yes, this is the case.

I think the current behavior is correct, if one knows that ds = ds(0).

But I understand it can be confusing. Marie and I discussed this last
week and I suggested that we let

  ds = ds(-1)

and that should denote integration over all sub domains, and ds(0)
would mean integration specifically over sub domain 0.

I was going to suggest this but I realize now we would need to change
the UFC interface (and FFC). It is currently based on the function

  num_foo_integrals

returning the number of integrals and

  create_foo_integral(i)

returning the integral for the terms on sub domain i.

This would need to be extended, possibly by overloading with

  create_foo_integral()

which would return the terms that are present on all sub domains.

Is it worth the effort of updating ufc.h, the UFC manual, the
assembler and FFC?

--
Anders

Attachment: signature.asc
Description: Digital signature