ABINIT Discussion Forums

Dear All

I've tried to run the self-consistent gw calculations, by using the "tests/wannier90/Input/t03.in" for Si as a guide but running it only the gw part.
In order to speed up the convergence study of gw corrections with respected to the number of self-consistent iterations, I tried reducing the "nkptgw" from 8 to 1
and used only "kptgw" at gamma point. The remaining part of the input were kept the same as the original run, excepting that the parts involving Wannier interface
were commented out. Please see the attached files: t03_gw_8kptgw.in and t03_gw_at_gamma.in, for details. The psp was 14si.pspnc.

However, I've found that the "E^GW_gap" for the gamma point in the 2nd iteration were different between the runs using nkptgw=8 and nkptgw=1, as given below.

*For the Run using nkptgw = 8, kptgw = 0 0 0 and others 7 kptgw's
GW iteration 1st 2nd
E^0_gap 2.521 3.312
E^GW_gap 3.328 3.428
DeltaE^GW_gap 0.807 0.116

*For the Run using nkptgw = 1, kptgw = 0 0 0 only
GW iteration 1st 2nd
E^0_gap 2.521 3.312
E^GW_gap 3.328 3.120
DeltaE^GW_gap 0.807 -0.192

Is the above result a common thing for self-consistent gw calcuations? Or is there something wrong?
I cannot figure out what might cause the difference in the E^GW_gap.

Looking forward to your replies,

Kind regards,
Thanusit

It's normal as the self-energy and the QP density are obtained by integrating the
wavefunctions in k-space.
In your gamma-only calculation only the wave-functions at Gamma have been updated
while the remaining k-points in the mesh are not updated and are treated at the Kohn-Sham level.
This is erroneous and the final SCGW results are not reliable.

Long time ago I've added a check in order to stop the code if SCGW is done on a subset
of the k-mesh used in the KSS file.
Unfortunately this prevented users from performing one-shot calculations on top of SCGW results.
I will reintroduce a warning to warn the user that not all the k-points are included in the SCGW run.