[gpaw-users] Summary of conference call experiment

Jens Jørgen Mortensen jensj at fysik.dtu.dk
Wed Mar 20 08:32:33 CET 2013


Den 13-03-2013 00:34, Christopher O'Grady skrev:
> The phone conference has (country-specific) toll-free numbers for
> everybody, I believe:
>
> It starts at 430pm in Denmark (830am in USA).  The phone numbers and
> access code:
> 	
> U.S. & Canada:    866.740.1260
> Denmark: 70142286 (Toll)
> Denmark: 80886327 (Toll Free)
> Spain: 900800233 (Toll Free)
> Spain: 917911831 (Toll)
> 	
> Access Code: 9262693
>
> I have reserved 2 hours of time.

Here is a quick summary:

Marcin showed some results using Jussi's new RMM-DIIS solver that looked 
very promising.  Maybe this will be our new default and robust eigensolver.

Ask mentioned that when parallelizing over bands and real-space it is 
important that band parallelization happens within a node so that 
communication is fast.  This is not currently how we lay out ranks. Ask 
will send us his 55 atom cluster example so we can do some measurements 
on Niflheim.  We should also do some timings for other typical systems 
before we change the rank-layout.

We discussed merging of Chris's new libxc wrapper to trunk.  Can we rely 
on people installing GPAW having a libxc on their system or should we 
have a copy of libxc's source code in GPAW's code repository and compile 
it for people?  We decided that it is best not to have a copy of libxc 
in GPAW since compiling libxc is easy to do.  This does however create a 
problem for Ask that wants to package GPAW for Debian which doesn't have 
a libxc package ...

We also decided that "smearing=('Fermi-Dirac', 0.1)" will be the new 
ASE-way to set smearing of occupation numbers.

In the end we evaluated the conference call vs. IRC meeting and we all 
thought that the conference call was a very effective way to communicate 
and that we should do more of those.

Jens Jørgen

>
> chris
> _______________________________________________
> gpaw-developers mailing list
> gpaw-developers at listserv.fysik.dtu.dk
> https://listserv.fysik.dtu.dk/mailman/listinfo/gpaw-developers



More information about the gpaw-users mailing list