[Pw_forum] further questions about 'nscf' and system control in 'nscf'

degironc degironc at sissa.it
Thu Dec 14 11:41:17 CET 2006

W. YU wrote:

>Thanks Stefano,
>Now some other questions have come to my mind:
>>'nscf' was in the past the only option for a non-scf
>>calculation and 
>>this  has been in the past a source of
>>problems because it was used for 1) band structure
>>calculation, 2)  BZ 
>>refining in order to obtain nicer
>>dos etc, and 3) preparation-run for phonon
>>calculations and maybe other 
>>uses as well...
>>The needs of these different types of calculation
>>are different... and 3) should be made, to be
>>with exactly the same sampling
>>as the  original scf calculation... 
>Does this mean the k-point grid for 'phonon'
>calculation should ALWAYS (at least in principle) be
>the same as the 'scf' part? If so, what will happen if
>it is not?
'phonon' flag should be used with exactly the same grid that generated
the scf potential... Its aim is to generate the SAME unperturbed information
on the k and k+q points that are needed for the linear response 
In the past  there was a check that the newly calculated charge density
agreed with the scf one... now I think this check has got lost but we
should reintroduce it.

>Now this also reminds me of questions about general
>setup of SYSTEM control and K_POINTS for nscf
>calculations including 'phonon','bands', and 'nscf'.
>This is how I understant it: 
>1. The ecut and ecutrho could be different from the
>scf part concerning all the three nscf
>calculations,even the smearing method and parameters
>supplied when dealing with a metallic system.
I think that ecut and ecutrho should be kept the same as in the
nscf/lband/phonon case the scf density and potential are read
from files into arrays with dimensions determined on the basis
of those numbers.
smearing parameteres should be irrelevant for bands as the potential
is fixed and no occupation is calculated... for the phonon case one
should be consistent... for the  nscf case I think one can experiment
with these features knowing that the resulting charge density or dos
will be slightly inconsistent.

>2. The k-points can and really should be different for
>band and dos type calculations. For phonon, I am not
>sure,I need your confirmation.
As I said above the phonon calculation should really be the same as
the scf calculation plus the indication of the desired q vector.

>3. If it is a spin-polarized calculation, if I already
>set the starting_magnetization say equals 0.5, why
>should I set it again in the band and other nscf
>calculation since the selfconsistency is already
>achieved and the magnetization is determined by the
>scf cycle.
The code check this quanties even if they are not needed in a nscf
calculation or in a restart from previously interrupted run.
The code is not using this information in tese cases.
The problem in this "illogical" behavior is the number of special cases
that one should keep in mind when trying to make it more logical.

best wishes,

>Best wishes,
>Wen YU
>Cheap talk?
>Check out Yahoo! Messenger's low PC-to-Phone call rates.
>Pw_forum mailing list
>Pw_forum at pwscf.org

More information about the users mailing list