Re: [Getfem-users] Problem assembling convection contribution

[Cédric Venet]

Le 09/02/2010 22:44, Cédric Venet a écrit :
> Le 09/02/2010 17:37, Yves Renard a écrit :
> > On lundi 8 février 2010, Cédric Venet wrote:
> > > Hi,
> > >
> > > I am trying to assemble the term "v.Grad u" where v is a field of vector
> > > data on my mesh. u is a scalar field, and the unknow of my equation. I
> > > am trying to do this:

> > It is possible to have vector data, but you declaration seems to be
> > wrong. It
> > should be something like that :
> >
> > assem.set("F=data(1); G=data$2(qdim(#1), #1); M$1
> > (#1,#1)+=comp(Base(#1).Grad(#1).Base(#1))(i,:,k,:).G(k,i).F(p)");
> > assem.assembly();
> >
> > Yves.
>
> Hi,
>
> While this solve the parse problem, the answer is more complex and still
> elude me.
> My fem_mesh #1 has a qdim of 1, so with this solution, I hit the assert
> invalid size for data argument. (my data vector is 3 times the number of
> dof in #1). So my understanding is that I need a new fem_mesh #2, with a
> qdim of 3. However, I didn't find a way to do this without duplicating a
> lot of work (ie, without respecifying all the finite element on each
> convex). Is there a way to avoid this?
>
> regards,
> Cédric

ok, it seems to work (at least it does not assert) if I create a fake 
fem mesh mf2 using the same linked mesh as #1 but without seting fem 
element on it. But I am not sure this is valid, or if it does what I want.

  getfem::generic_assembly assem;
  assem.push_mi(fm.GetIntegrationMesh());
  assem.push_mf(fm.GetShapeMesh());
  getfem::mesh_fem mf2(fm.GetShapeMesh().linked_mesh(),3);
  assem.push_mf(mf2);
  std::vector<T> tmpve(1,m_coef);
  assem.push_data(tmpve);
  assem.push_data(m_speed);
  assem.push_mat(fm.GetAMatrix());
  // warning: non symetric!!
  // row/column major... check the order of data$2
  assem.set("F=data(1);G=data$2(qdim(#2),#1); 
M$1(#1,#1)+=comp(Base(#1).Grad(#1).Base(#1))(i,:,k,:).G(k,i).F(p)");
  assem.assembly();