ffc team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Wed, Apr 16, 2008 at 09:24:02PM +0200, Johan Hoffman wrote:
> > Hi all, I'm trying to do a survey of all the different ways of enforcing
> > incompressibility in the Navier-Stokes equations, and FFC/FIAT of course
> > has
> > the most advanced element capabilities for doing this...
> >
> > Unfortunately, I have hit a snag (either in my ignorance or through a bug)
> > in my grand scheme, starting with the Oseen equations (fix the convective
> > velocity for fixed points, don't do the full Newton linearization).
> >
> > It seems that if you put a vector element into a mixed element and then
> > take
> > it back out, you can no longer index it.
> >
> > Putting the following in a .form (everything that doesn't involve pressure
> > spaces) works beatifully
> > #start .form file here
> > vel_element = VectorElement("Lagrange","triangle",2)
> > u = TrialFunction( vel_element )
> > v = TestFunction( vel_element )
> > u0 = Function( vel_element )
> >
> > a = D(v[i],j)*D(u[i],j)*dx + v[i]*u0[j]*D(u[i],j)*dx
> > #end .form file
> >
> > On the other hand, if I do:
> > # start new .form file here
> > vel_element = VectorElement("Lagrange","triangle",2)
> > pre_element = FiniteElement("Lagrange","triangle",1)
> > element = vel_element + pre_element
> > (u,p) = TrialFunctions( element )
> > (v,q) = TestFunctions( element )
> > u0 = Function( vel_element )
> >
> > a = D(v[i],j)*D(u[i],j)*dx + v[i]*u0[j]*D(u[i],j)*dx
> > # end new .form file
> >
> > I get an error:
> >
> > Preprocessing form file: Oseen.form --> Oseen.py
> >
> > *** object cannot be interpreted as an index
> > *** To get more information about this error, rerun ffc with the option
> > -d1.
> >
> > I've attached the output using -d1.
> >
> > Any corrections either of my stupidity or of FFC would be appreciated :)
> 
> Hi Rob!
> 
> I don't think you have to question your sanity, we have experienced
> similar problems when dealing with the compressible Euler equations: with
> scalar density and energy together with a momentum vector. We have been
> forced to use built in operators instead, which is non optimal since the
> definitions of the action of some operators is a bit unclear.
> 
> I do not know what the underlying problem is, but we have brought this up
> before and it seems to be non-trivial? Or maybe (hopefully) I am wrong?
> 
> This is also something I want to stress for the coming UFL; that to me in
> terms of operators, the most important is to have a fully reliable system
> for indexing. Then to add a nice collection of well defined operators on
> top is great, but secondary (to me).

Indexing is one of many priorities for UFL.

> I wasn't sure I was supposed to edit the UFL page directly with this
> wishes? What do you say Anders?

Yes, just edit the page with suggestions. That's what it's for.

-- 
Anders