[gpaw-users] Speeding up calculation for DFT/MD?

[Ask Hjorth Larsen]

Dear Susi

(Please use reply to all so it stays on the list)

2013/4/22 Susi Toma <toma.susi at aalto.fi>:
> Dear Ask,
>
> Thanks a lot for the tips, I'll be sure to test them. I'm a bit hesitant
> about using LCAO since I'm at the first stage trying to validate a method
> that was developed for VASP, and I want to follow that as closely as
> possible. But of course, doing tests to see if the property of interest
> comes out correctly would probably be enough to assuage these doubts – I'm
> just not sure I need that big of speed boost.
>
> One question about this comment:
>
>
> *Do* set the entirely different "nbands" variable (e.g.
> GPAW(nbands=27)) to something which is sufficient to accommodate all
> electrons plus a few.
>
>
> Currently I am using the default settings to converge occupied states only:
>
> Number of Atoms: 96
> Number of Atomic Orbitals: 384
> Number of Bands in Calculation:         384
> Bands to Converge:                      Occupied States Only
> Number of Valence Electrons:            384
>
>
> I don't quite understand what the would be the purpose of setting the nbands
> variable to something else?
>
>
> -Toma

Bands are basically Kohn-Sham states.  With 384 electrons you probably
need just above 200 bands (to have a few empty ones; perhaps more if
there is trouble converging things).  Right now the calculation will
be much more expensive than it could be, because of the very large
number of bands.

Regards
Ask