Problem with GW0

[asaaravindh2019]

Dear All,
We are now facing some issues with G0W0 calculations implemented in VASP. We are calculating the nonself consistent GW band gap
with HSE06 wavefunctions as the starting wavefunctions. In order to use HSE06 we set

LHFCALC = .TRUE, HFSCREEN = 0.2, with a mixing fraction AEXX = 0.2.

However, we have observed that in GW calculation VASP set the value of
HFSCREEN to 0.000 and AEXX to 1.

This essentially means that we are running full Hartree - Fock calculation in the GW step in spite of setting
screened-hybrid parameters in the INCAR. For all the steps prior to the actual GW step (HSE ground state
calculation, making Virtual orbitals) the calculation is behaving according to the INCAR.

Why does this happen in the actual GW step? Are we missing something?

To test the set up we have run GW calculation on Silicon following the official VASP tutorial. Unfortunately the problem remains in that
case too in the actual GW calculation step. In our case calculated silicon GW bandgap is overestimated Eg = 1.4 eV. (INCAR file attached)
# INCAR for Silicon GW #
System = fcc Si; ISMEAR = 0; SIGMA = 0.05; EDIFF = 1E-8
#HSE Tags
LHFCALC = .TRUE. ; HFSCREEN = 0.2 ; AEXX = 0.20

NBANDS = 64
#GW tags
ALGO = GW0; LSPECTRAL = .TRUE.; NOMEGA = 50

Any suggestion would be appreciated!
thanks
Assa

[merzuk.kaltak]

Dear Assa,

The GW step does not support mixing of exchange and screened exchange. The entire idea of GW is to approximate the exchange-correlation effects with the self-energy calculated with orbitals and energies of a preceding DFT step. It does not really make sense to define a screened exchange, since W (screened Coulomb interaction) takes care of that in the GW approximation. This is the reason why VASP sets HFSCREEN = 0 and AEXX = 1 in GW calculations.

Please note that the starting point for GW, that is the preceding calculation, has an impact on the quasi-particle energies. That is, the qp-energies differ whether you start from PBE, HSE03, ... WAVECAR (and WAVEDER) files.

[asaaravindh2019]

Thank you! Much appreciated