dolfin team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Tue, Dec 05, 2006 at 11:25:50AM +0100, Garth N. Wells wrote:

> >First dg test
> >
> 
> Great!
> 
> Garth

We're far from finished, but we think we can do basic forms like

  a = v('+')*u('-')*dS

I'll commit some updates to FEM soon. Before, when we only had
boundary integrals over exterior facets (facets on the boundary), we
could treat the assembly in the same way as when we assemble over
cells (but with an extra "facet" argument). This gets a bit more
complicated when we do DG methods since we need to assemble over
interior facets that have two neighboring cells and we need to
integrate products of things from both sides (like in the example
above). The best way to treat this seems to be to separate the
assembly into assembly of three different types of tensors:

    A^K     : element tensor
    A^{S_e} : exterior facet tensor (on the boundary)
    A^{S_i} : interior facet tensor (facets inside the domain)

This means there will be some code duplication until we figure out how
to make this uniform over the tensor rank. Right now, we need separate
also over the type of global structure we are assembling into: matrix,
vector or scalar.

/Anders