dolfin team mailing list archive

[Marie Rognes <meg@xxxxxxxxxxx>]

Anders Logg wrote:
> On Mon, Apr 28, 2008 at 02:24:38PM +0200, Anders Logg wrote:
>   
> > On Mon, Apr 28, 2008 at 03:00:26PM +0200, Marie Rognes wrote:
> >     
> > > Error:
> > >
> > >     Traceback (most recent call last):
> > >       File "test.py", line 33, in ?
> > >         u0.assign(w)
> > >       File
> > >     "/home/meg/src/dolfin/dolfin/local/lib/python2.4/site-packages/dolfin/dolfin.py",
> > >     line 3172, in assign
> > >         return _dolfin.cpp_Function_assign(*args)
> > >     RuntimeError: *** Error: Unable to find finite element in library:
> > >     "Mixed finite element: [Crouzeix-Raviart finite element of degree 1
> > >     on a triangle, Crouzeix-Raviart finite element of degree 1 on a
> > >     triangle]".
> > >     Segmentation fault
> > >
> > >
> > > Code:
> > >
> > >     from dolfin import *
> > >
> > >     # Geometry of the domain and mesh
> > >     shape = "triangle"
> > >     mesh = UnitSquare(16, 16)
> > >
> > >     # Setting up the elements
> > >     V = VectorElement("CR", shape, 1)
> > >
> > >     u0 = Function(V, mesh, 0.0)
> > >
> > >     v = TestFunction(V)
> > >     u = TrialFunction(V)
> > >     a = dot(grad(u), grad(v))*dx
> > >     L = dot(u0, v)*dx
> > >
> > >     pde = LinearPDE(a, L, mesh)
> > >     w = pde.solve()
> > >     u0.assign(w)
> > >
> > >    
> > > Easy fix?
> > >       
> > Probably not. The message comes from the copy constructor (in C++) of
> > DiscreteFunction, which looks in the list of precompiled elements and
> > that list does not include vector-valued CR elements.
> >
> > I'll see if I can find the problem. You are doing this from Python so
> > it should be possible to get around this problem and not need to
> > create the element from precompiled code.
> >     
>
> ok, I found a solution (temporary).
>
> If you change the line
>
>   u0 = Function(V, mesh, 0.0)
>
> to
>
>   u0 = Function(V, mesh, Vector())
>
> it should work.
>
>   

Perfect. Thanks!

> Then both w and u0 will be discrete functions (i.e. storing their
> values in a vector). They will also be defined on the same space so
> then just the values in the vector need to get copied and that works
> nicely (in operator= in DiscreteFunction).
>
> The problem with your code is that u0 was not a discrete function, but
> instead a "constant function" which does not store its values in a
> vector. So then what happens is that a new discrete function must be
> created (including a new finite element) and then DOLFIN must (in C++)
> look in its library of precompiled elements (which is limited to
> reduce compile times).
>
> We can fix this and related issues when we rework/cleanup the function
> classes (now that the linear algebra seems to be ok).
>
>   

Good.

--
Marie E. Rognes
Ph.D Fellow, 
Centre of Mathematics for Applications, 
University of Oslo
http://folk.uio.no/meg