[gpaw-users] bug on GW-BSE at q!=0 ?

[Sponza Lorenzo]

Hello Gpaw users,

I’m currently using GPAW to compute the exciton spectra at q!=0. This means that I’m performing GW-BSE calculations at finite q. The system I'm working with has hexagonal symmetry. What happened is that I encounter a striking differences in the spectra and in the eigenvalues between my GW-BSE calculations and the SO-BSE calculations (SO=scissor operator), although the GW corrections are reasonably reproduced by SO. These discrepancies do not arise at q=0, but only at q!=0. 

In order to pin down the problem I made the following test. I would like some of you to tell me if it’s meaningful.
1) Starting from a DFT calculation recorded in a .gpw file, I computed GW corrections and saved the result on a .pckl file.
2) I run BSE from the same .gpw file with gw_skn=None at q=0 and q!=0. 
3) I extracted the DFT energies from the pckl file and saved them in an array (call it dften).
4) I run BSE from the same .gpw file with gw_skn=dften at q=0 and q!=0.
If I understood well how GPAW handles these files, the two sets of calculations should be equivalent. Indeed they give the same eigenvalues (up to the forth digit) in the q=0 case, but I fund differences of the order of tenths of eV in the q!=0 case.

If the test is correct, it seems to point to a bug in the selection of the elements of the gw_skn array when valence and conduction states do not belong to the same k-point. 
If the test is wrong, then there should be a mistake in the way I pass the energy arrays.
Please, if you can have a look at this issue and keep me updated on this. 

Many thanks 

====================
Dr. Lorenzo Sponza	
LEM
ONERA-CNRS
BP72
92322 Châtillon Cedex
Tel: +33146734546