dolfin team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Wed, Nov 09, 2005 at 03:01:52PM +0100, Johan Jansson wrote:
> On Wed, Nov 09, 2005 at 10:48:37AM +0100, hetzel.devel@xxxxxx wrote:
> > 
> > Hi,
> > I'm having two functions say u1 und u2 and two test functions v1 and v2. At the moment I'm using the "Vector Lagrange" Functions to declare u and v as a vector.
> > 
> > I would like to solve my PDE with u1 and v1 defined on mesh1 (approx. 2000 nodes) and u2 and v2 defined on a different mesh2 (approx. 20 nodes). I would like to get a 2020x2020 matrix. Both meshes will be contain exectly the same area.
> > Is there a change to declare the bilinearform in such a manner?
> > 
> > /Haiko
> > 
> 
> Hi Haiko,
> 
> A form is not connected to a specific mesh, so it's perfectly ok to
> assemble the same bilinear form on two different meshes. But then you
> would get two different matrices. Why do you want the result in the
> same matrix? Isn't it ok to get matrix A1 (2000x2000) corresponding to
> the first mesh and matrix A2 (20x20) corresponding to the second mesh?
> 
>   Johan

I suspect he wants to have two functions defined on different meshes
but appear in the same form so that for example u1 would couple to v2
in some way.

This is currently not possible. Each form must be assembled on only
one mesh.

Or are the two systems uncoupled? In that case, you can assemble A1
and A2 and then copy them to one big matrix with some PETSc calls.

(We might want to add a new class BlockMatrix as a subclass of
VirtualMatrix that automates the creation of block matrices. I think
Andy Terrel did some initial work in this direction.)

/Anders