Solution for unsatisfactory GW-corrected band-gap in GaAs before BSE

[peiogargor]

Dear Yambo code developers,

I want to get a satisfactory GW-corrected band-gap energy in bulk GaAs, whose experimental value is 1.4 eV. To that end, I have been using QE + Yambo codes. As I have read in literature, this exercise turns out to be problematic when using standard LDA or GGA pseudopotentials (PPs) and it is recommended to have a DFT-starting point based on hybrid PPs.

Nevertheless, I would like to avoid the use of such computationally heavy PPs. I have been doing some calculations with PPs within LDA (PZ) and GGA (PBE & PBEsol) from the Optimized Norm-Conserving Vanderbilt (ONCV) PP database (Ga: 3d10 4s2 4p1 Zval=13 & As: 4s2 4p3 4d0 Zval=5) for which I get DFT band-gaps in the range of 0.33-0.53 eV and GW-corrected values around 0.75-0.79 eV. On the other hand, I have constructed PPs using the ld1.x tool from QE in order to take explicitly into account 3d10 electrons of As as valence electrons in the PP for the same three approximations of the XC and I get DFT band-gaps in the range of 0.23-0.45 eV and GW-corrected values around 1.03-1.05 eV. It seems that more core electrons you take into account better you approach to the experimental value, but it is not yet enough.

So my question comes now: In order to get a subsequent BSE calculation, the scissors shift I use in the BSE solver must be the one that recovers a value of 1.4 eV. Nevertheless, for the static screening should I also use it or not? I would say no, as in a BSE calculation on top of GW QP bands the static screening corresponds to the RPA from DFT, and here the BSE calculation would be on top of scissors shifted QP bands, so similar staff in practice. Is not it? If not, which is the best way to proceed in these cases? Please, feel also free to recommend me any other strategy I have not thought about.

Thank you in advance for your answer and advices.

Peio.

[Daniele Varsano]

Dear Peio,

it seems that you have already done the best you can do with the current approximations.
What you can try to do further is to relax the plasmon pole approximation and try a more accurate treat of the dynamical effect of the screening. You can avoid to go through the full frequency approach and release with the multipole approximations. You can find the method here:
https://journals.aps.org/prb/abstract/1 ... 104.115157
and a tutorial here:
https://www.yambo-code.eu/wiki/index.ph ... tion_(MPA)

Having said that, it is not guaranteed that this will provide you a gap in agreement with the literature.

Another attempt you can do is to go for an eigenvalue self-consistent GW, this will increase the gap:
https://www.yambo-code.eu/wiki/index.ph ... alues_only

So my question comes now: In order to get a subsequent BSE calculation, the scissors shift I use in the BSE solver must be the one that recovers a value of 1.4 eV. Nevertheless, for the static screening should I also use it or not? I would say no, as in a BSE calculation on top of GW QP bands the static screening corresponds to the RPA from DFT, and here the BSE calculation would be on top of scissors shifted QP bands, so similar staff in practice. Is not it?

I would say not. In general, the screening is calculated on top of DFT to mimic also an optical gap. The screening on top of QP energies usually provide an underscreening effect.

Best,
Daniele

[peiogargor]

Dear Daniele,

Thank you for your advice, I am trying multipole approximation and self-consistent GW corrections, when they are finished I will update you with my results. In the meantime another question came to my mind: I noticed that the G0W0 band-gap + CBM & VBM convergences of bulk GaAs wrt k-mesh are slower than I thought. So, can RIM method be used also for bulk? If yes, which cutoff geometry should be used? I have tried to search within the forum and the wiki, but I did not find anything by the moment. I suppose RIM-W can not be used because in the result file of a monolayer I did before it is explicitly said: "[WARNING] RIM-W works only for 2D semiconductors with slab cutoff".

Thank you in advance,

Peio.

[Daniele Varsano]

Dear Peio,

yes you can use RIM for the bulk and also it is suggested! It will accelerate mostly the exchange part of the self-energy as you can see in this paper:
https://www.sciencedirect.com/science/a ... 5520300734

Moreover, it is advised because also the accuracy will improve. You need just to select the RIM options (Q points and G vector) without using any cutoff geometry. In this way coulomb integral will be performed stochastically using a 3D Coulomb potential.

Best,
Daniele

[peiogargor]

Dear yambo developers & users,

I woulk like just to say that self-consistent GW corrections enhanced the band gap energy to 1.2 eV (remember one-shot GW corrections gave 1.0 eV), but still not close to the experimental value of 1.4 eV. Likewise, the plamon multipole approximation did not give better result than the plasmon pole approximation for this case.

Therefore, I will proceed with my calculations using a scissors shift.

Best,
Peio.