The ABINIT Users Mailing List ( CLOSED )

[Philippe Blaise <philippe.blaise@cea.fr>]

asadidomakani@yahoo.com a écrit :
> Dear Abinit users,
> I want to know,how to determine nband1,nband2 and nband3 in GW ?
> thank you so much for any suggestions,
> Best regardes,
> Asadi
>
>   
As Fabien has just said, only a convergence study can give you a useful 
range of values for nband.
There is no convergence criteria like the one that is used in a typical 
SCF cycle in DFT.
Of course the numerical convergence is a trade-off between accuracy and 
computing time.
Some GW calculations are so long that  the convergence study can consume 
a lot.
So just to give you an idea, for a small system, like bulk silicon, 
probably less than 100 bands
are sufficient to obtain a converged band gap. For a slightly complex 
insulator made of some tens of electrons a nband between 500 and
1000 is required  to converge the different eigenvalues close to the 
conduction or the valence band with an accuracy of +/- 0.1 eV ;
of course your convergence study should also take into account the other 
parameters (all the ecutxx) and you must be aware
of the various dielectric models, and the influence of the 
pseudopotentials you use.
Last, the massive parallelization of GW calculations is effective with 
Abinit which is just fabulous.
So try to choose this approach : this requires at least the nbands to be 
a multiple of the number of cores or cpus.

Best Regards,

Philippe