[gpaw-users] odd SCF convergence behavior

[naromero at alcf.anl.gov]

Ask,

I added 50 empty bands and the problem essentially went away.

32 MPI tasks:
http://en.pastebin.ca/1973609

64 MPI tasks
http://en.pastebin.ca/1973610

Up to iteration 20, the numerics are identical. Perhaps this
is worth documenting on our GPAW web page? I can do it
if there are no objections.

----- "Ask Hjorth Larsen" <askhl at fysik.dtu.dk> wrote:

> Hi
> 
> On Mon, 25 Oct 2010, Nichols A. Romero wrote:
> 
> > Hi,
> >
> > Here is the input file in question.
> > http://en.pastebin.ca/1973390
> >
> > The SCF cycle is well-behaved at 64 MPI tasks,
> > http://en.pastebin.ca/1973391
> >
> > but is poorly behaved at 32 MPI tasks.
> > http://en.pastebin.ca/1973392
> >
> > Is there a simple way to explain this difference?
> >
> > This behavior is present irregardless of using
> > GS or Jacobi for the poisson solver?
> 
> Seems to me that they are both performing a bit 'randomly', one of
> them 
> just happens to converge by chance.  What happens if you use 100 more
> 
> empty bands?
> 
> As for things behaving a bit differently depending on parallelization:
> It 
> has been like always (or a very long time), and I'm not sure why. 
> Does 
> anyone know?  It seems to be more than just numerical noise.
> 
> Regards
> Ask

-- 
Nichols A. Romero, Ph.D.
Argonne Leadership Computing Facility
Argonne National Laboratory
Building 240 Room 2-127
9700 South Cass Avenue
Argonne, IL 60490
(630) 252-3441