[gpaw-users] New finite-difference stencil for Poisson equation

[Jens Jørgen Mortensen]

On Wed, 2010-03-03 at 12:34 +0100, Jens Jørgen Mortensen wrote:
> Hi!
> 
> Now that GPAW can handle non-orthogonal unit cells, we need to change
> the default finite-difference stencil for the Poisson equation.
> Currently we use a Mehrstellen stencil for orthogonal cells and a
> central FD stencil with n=2 neighbors in all directions for the
> non-orthogonal case.
> 
> We need to use n=3 in the future so that we don't get a jump in the
> energy when going from an orthogonal cell to a slightly non-orthogonal
> cell.  The switch from Mehrstellen to n=3 will change total energies
> slightly.  In my experience it is in the meV/atom range, but it will
> depend on the system and the grid spacing used.
> 
> I think we should make the change as soon as possible so that it will be
> in the upcoming version 0.7.
> 
> You could already now try to use poissonsolver=PoissonSolver(nn=3) in
> your calculations and check how big the change is.

WARNING:

PoissonSolver(nn=3) is now the default!  Use PoissonSolver(nn='M') to
get the old behavior.

JJ