Johan Hake wrote:
> On Wednesday 19 August 2009 14:57:20 Garth N. Wells wrote:
>> Johan Hake wrote:
>>> On Wednesday 19 August 2009 14:21:12 Garth N. Wells wrote:
>>>> Garth N. Wells wrote:
>>>>> Kent Andre wrote:
>>>>>>> Not sure what we have to do actually. I suppose I want get any clue
>>>>>>> from the
>>>>>>> demo/function/restriction
>>>>>>>
>>>>>>> demo?
>>>>>>> I see that Kent has modified the dolfin.FunctionSpace to reflect the
>>>>>>> present (broken) implementation of restriction, anything in that
>>>>>>> line? Should such an implementation has been done in
>>>>>>> FunctionSpaceBase instead of FunctionSpace? Kent?
>>>>>>>
>>>>>>> Johan
>>>>>> This restriction is different in the sense that the restriction is
>>>>>> made by the standard element on a sub mesh.
>>>>>> The restriction in focus now is on the complete mesh but for only
>>>>>> a part of the element.
>>>>>> Combined, these two features will be very powerful. I guess the
>>>>>> sub-mesh restriction stuff is broken due to the parallel mesh.
>>>>> I've attached an example solver. To make it work for P^k with k > 2 we
>>>>> need to restrict some functions to cell facets. Using FFC, we do
>>>>>
>>>>> scalar = FiniteElement("Lagrange", "triangle", 3)
>>>>> dscalar = FiniteElement("Discontinuous Lagrange", "triangle", 3)
>>>>>
>>>>> # This syntax restricts the space 'scalar' to cell facets.
>>>>> mixed = scalar["facet"] + dscalar
>>>>>
>>>>> In PyDOLFIN we could do
>>>>>
>>>>> V_dg = FunctionSpace(mesh, "DG", 3)
>>>>> V_cg = FunctionSpace(mesh, "CG", 3, "facet")
>>>>>
>>>>> mixed = V_cg + V_dg
>>>> I see now that we need to have the above option since the everything
>>>> (e.g. the dof map) is created and compiled inside functionspace.py.
>>> We could implement it in FunctionSpaceBase, as an operator, returning a
>>> new FunctionSpaceBase, with the restricted element as arguemnt.
>>>
>>> However my concerns go to a sub space of a restricted mixed space (if
>>> such exists) see the other email.
>>>
>>> If we go for 1, (as you have already done ;) ), we cannot create a
>>> restricted mixed space using the addition syntax.
>>>
>>> mixed = V1 + V2
>> Do you mean to then restrict mixed? I'm already doing
>
> Precisely.
>
> After a second thought I think that doing it in the constructor is more
> clean.
> Makes life simpler. Is this also possible i UFL, I mean:
>
> e = FiniteElement("CG",cell,0,"facet")
>
> Should it?
>
Is that a PyDOLFIN finite element? If it is, then yes, that's how it
should work.
Somewhat related, it's a bit annoying in PyDOLFIN that the code is
recompiled when the mesh is changed. Can we avoid this? Should one then
create a finite element and a dof map with which to initialise the
Pythin FunctionSpace?
Garth
> Johan
>
>> V0 = FunctionSpace(mesh, "CG", 3, "facet")
>> V1 = FunctionSpace(mesh, "DG", 3)
>>
>> mixed = V0 + V1
>>
>> and it works. This type of restriction (we'll need to sort out our
>> terminology) impacts the dofmap, which I why I think it needs to be
>> involved in the creation of the function space.
>>
>> Garth
>>
>>> If we want a restricted mixed space we need to use the constructor.
>>>
>>> mixed = MixedFunctionSpace((V1,V2),restriction="facet")
>>>
>>> Johan
>>>
>>>> Garth
>>>>
>>>>> or
>>>>>
>>>>> V_dg = FunctionSpace(mesh, "DG", 3)
>>>>> V_cg = FunctionSpace(mesh, "CG", 3)
>>>>>
>>>>> mixed = V_cg["scalar"] + V_dg
>>>>>
>>>>> Garth
>>>>>
>>>>>> Kent
>>>>>>
>>>>>> _______________________________________________
>>>>>> DOLFIN-dev mailing list
>>>>>> [email protected]
>>>>>> http://www.fenics.org/mailman/listinfo/dolfin-dev
>>>>> -----------------------------------------------------------------------
>>>>> -
>>>>>
>>>>> _______________________________________________
>>>>> DOLFIN-dev mailing list
>>>>> [email protected]
>>>>> http://www.fenics.org/mailman/listinfo/dolfin-dev
>>>> _______________________________________________
>>>> DOLFIN-dev mailing list
>>>> [email protected]
>>>> http://www.fenics.org/mailman/listinfo/dolfin-dev
_______________________________________________
DOLFIN-dev mailing list
[email protected]
http://www.fenics.org/mailman/listinfo/dolfin-dev
