[Pw_forum] no. of nbnd

[Nicola Marzari]

vicky singh wrote:
> Hi
> I am studying a supercell for Ni with 64 atoms. while specifying nbnd 
> how much i select i am not sure. if i go by document PW.html it says 
> that for metal it should be nelec/2 + 20 %. then should i take nbnd = 
> 400. also it says minimum 4 more i.e. should i take nbnd = 330. Thanks 
> in advance.
Dear Vicky,

short answer: it depends on "degauss".

You are studying a metallic system, and introducing some fictitious 
temperature
for the electronic degrees of freedom (this is the "broadening" or 
"smearing") in
order to improve the quality of your k-point sampling (otherwise at zero 
temperature
there are sharp discontinuities in the integrals over the brillouin zone).

you need to have enough bands so that the highest bands are close to 
empty - having
more would be a waste of cpu, having less is somewhat incorrect because you
force the higher bands to be more occupied than they would like to be 
according
to the broadening you have chosen.

In a gaussian/m-p/m-v scheme, it is safe to say that the occupations 
becomes negligible
around 3 times degauss above the fermi energy; check for yourself (by 
asking the code
to print occupations) that the highest bands have occupation less than 
0.01 .

If you were to use Fermi-Dirac, that has longer tails, you would need 
more bands (hence
better not to use it, unless you are keen on real fermi-dirac statistics).


             nicola


-- 
---------------------------------------------------------------------
Prof Nicola Marzari   Department of Materials Science and Engineering
13-5066   MIT   77 Massachusetts Avenue   Cambridge MA 02139-4307 USA
tel 617.4522758 fax 2586534 marzari at mit.edu http://quasiamore.mit.edu