dolfin team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Fri, Aug 06, 2010 at 06:52:58PM +0100, Garth N. Wells wrote:
> On Fri, 2010-08-06 at 10:06 -0700, Johan Hake wrote:
> > On Friday August 6 2010 09:52:20 Garth N. Wells wrote:
> > > We have an issue with MixedFunctionSpace which makes incorrect usage
> > > very easy. For example, if I have
> > >
> > >     V = VectorFunctionSpace(mesh, "CG", 2)
> > >     Q = FunctionSpace(mesh, "CG", 1)
> > >     W = V * Q
> > >
> > > The dof maps for Q and W.sub(1) are not the same. I've seen a number of
> > > users using V or Q in setting boundary conditions rather than W.sub(0)
> > > and W.sub(1).
> >
> > I have done this mistake myself :P
> >
> > I guess this problem only excist in POyDOLFIN?
> >
>
> Yes.
>
> > > When using
> > >
> > >     W = MixedFunctionSpace([ V, Q])
> > >
> > > could we create W and then inside the MixedFunctionSpace constructor set
> > >
> > >    V = W.sub(0)
> > >    Q = W.sub(1)
> >
> > No as you are then just changing the local references of V and Q inside the
> > __init__ method.
>
> I realise this, but then at least Q and W.sub(1) will point to the same
> space. It may be confusing the creating a mixed space changes V and Q.

I don't understand how that will work. There is no way that the call
to MixedFunctionSpace can change the meaning of V and Q. It would also
be confusing and dangerous. What we could do is change some property
of V and Q (like the offset) but that would lead to other problems
(see below).

> > What you need is a method of FunctionSpace that transform it
> > to a SubSpace. Not sure that is possible?
> >
> > Is there a user case where V and Q are used for what they are (not subspaces),
> > after a declaration of a mixed space? We will screw such a use if we turn V
> > and Q into a subspaces.
> >
> > If this is not a consern, we might be able to tag V and Q as beeing used to
> > create a mixed space inside the __init__ of mixedfunctionspace and issue some
> > kind of warning (or exception!) whenever these are used to create a Function
> > or someother FunctionSpace dependent object. This can be done purely in the
> > Python layer.
> >
>
> I'm really not sure what the best approach is.

Tagging V and Q as special because they have been used to create a
mixed function space would be bad since V and Q might very well be
used to create other functions which have no relation to the mixed
space. A simple example is a right-hand side f = Function(V) in
Stokes.

A solution would be to let the BoundaryCondition class check that the
global size of the dofmap (of which the sub dofmaps are parts) and
compare with the vector or matrix it is manipulating. That check would
fail for V and Q but work for W.sub(0) and W.sub(1).

--
Anders

Attachment: signature.asc
Description: Digital signature