The ABINIT Users Mailing List ( CLOSED )

[Fabien Bruneval <fabien.bruneval@cea.fr>]

Dear Thanusit,

Let's say that if you perform your convergence tests and also your 
production runs with the complete k-point grid, you are on the safe side.
Maybe you can do some convergence tests with a reduced k-point grid, but 
when turning to production, you should not forget to have a complete grid.

My main worry was that most of the beginners, which train with the 
tutorial using only 1 k-point, think that only one k-point is needed in 
general.
I have the feeling that it is not clear from the tutorial that the 1 
k-point run is just meant for testing purposes, and that for all 
real-life GW, one absolutely needs the complete Monkhorst-Pack grid...
That's why I was proposing to change the tutorial.

Best regards.

Fabien


ธนูสิทธิ์ บุรินทร์ประโคน wrote:
> Dear Fabien,
>
> Acoording to your recent sugestion on the k-point grid needed in GW 
> calculations, does it mean that the KSS file must be generated by 
> using a complete k-point grid in the first place before we perform the 
> convergence tests and all sucessive calculations?
>
> Kind Regards,
>
> Thanusit
>
>
> */On April 22, 2008 4:01:54 PM ICT, "Fabien Bruneval" 
> <fabien.bruneval@cea.fr> wrote:/*
>
>     Dear Sophie,
>
>     This tutorial is rather misleading and should be changed in the
>     future.
>
>     The GW calculations needs a k-point grid similar to the one uses
>     in the
>     usual DFT calculations.
>     The only requirement is that the k-point, where you want to know
>     the GW
>     band structure, needs to be part of the k-point grid you use.
>
>     A shortcoming of the current state of the GW part is that it is not
>     really clever in using the symmetries. Therefore, it is generally
>     a good
>     policy to
>     choose a k-point grid that is centered in Gamma or with the usual
>     shift
>     0.5 0.5 0.5. Other choices usually perturb the code, which
>     produces tons
>     of k-points
>     in the Brillouin Zone.
>
>     If you want to calculate the GW correction in a weird k-point,
>     such as
>     [0.3 0.3 0.], I advise to you to calculate the GW corrections in a
>     "nice" neighboring point like [0.25 0.25 0], which is part of you
>     k-point grid. And then use this GW correction on top of DFT
>     calculation
>     made in the precise k-point you wanted.
>
>     e_GW(k=0.3) = e_KS(k=0.3) + [ e_GW(k=0.25) - e_KS(k=0.25) ]
>
>     This is justified because usually the term within squared brackets is
>     rather k-point insensitive.
>
>     I'm not sure to be clear enough... If not, just tell me.
>
>     Fabien
>
>
>     us_sophie.green@yahoo.com wrote:
>     > Dear all users,
>     >
>     > I looked the tests in abinit about the GW calculations. In the
>     GW corrections
>     > for the band gap, the convergence tests (including ecutwfn,
>     ecutsigx, nband,
>     > etc) have done within only one point (the gamma point for Si). I
>     am not sure of
>     > its accuracy. Besides, the band gap correction for Si is in
>     gamma point (direct
>     > band gap semiconductor), whether the convergence test on gamma
>     point is
>     > revelant to the following calculation of the band gap correction
>     in Gamma
>     > point?
>     > If I want to calculate the GW correction for the band gap with a
>     nonspecial
>     > point, for example (0.3 0.3 0), should I test the parameters
>     with the (0.3 0.3
>     > 0) point, other than gamma point? The point (0.3 0.3 0) should
>     be included in
>     > the KSS calculation, then I wrote my input file in the Dataset2
>     part as
>     > follows. I give three points including (0.3 0.3 0) ( the other
>     two points are
>     > interesting points, with which I will also do GW corrections). I
>     am not sure
>     > whether it is feasible.
>     > Need your help. Thanks in advance.
>     >
>     >
>     > # Dataset2: calculation of kss file
>     > # Definition of k-points
>     > kptopt2 0
>     > nkpt2 3 # A set of 19 k-points containing Gamma
>     > kpt 0.3 0.3 0
>     > 0.5 0.5 0
>     > -0.5 0.5 0
>     > #istwfk2 19*1 # Option needed for Gamma
>     > istwfk2 3*1
>     > iscf2 -2 # Non self-consistent calculation
>     > getden2 -1 # Read previous density file
>     > nband2 13
>     > nbandkss2 -1 # Number of bands to store in KSS file
>     >