dolfin team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Thu, May 24, 2007 at 09:02:08PM +0200, Garth N. Wells wrote:
> 
> 
> Kristian Oelgaard wrote:
> > Quoting "Garth N. Wells" <g.n.wells@xxxxxxxxxx>:
> > 
> >> I'm getting a lot of output about mesh ordering. Is the mesh ordering 
> >> new, or is it just that the output messages are new?
> > 
> > The messages are new I think, and also the structure of when mesh.order()
> > is called has changed.
> > 
> > Before results (for DG) where different for:
> > 
> > Mesh mesh(5,5,5)
> > 
> > and 
> > 
> > Mesh mesh(5,5,5)
> > mesh.order()
> > 
> 
> Should it be called for a given mesh only once? For some problems it's 
> being called many times.

It is called in the constructor of DofMap. Right now it is called
after each call to mesh.init(), but we could call it after all those
calls.

The important thing is that we use an ordered mesh in the following
cases:

  1. For CG with q >= 3
  2. For all DG elements
  3. For all BDM elements
  4. Other

We could figure out some way to do a minimal amount of ordering.

For example, for standard first order Lagrange, no ordering is needed.

As long as the convergence test (higher order Lagrange) in
src/bench/fem/convergence, the DG demo and the BDM demo work, we can
do the ordering anywhere we like. (But it's important that it is done
before the call to the assembleFoo functions so that all contributions
are assembled on the same mesh.)

/Anders