The applicability of GW0 calculation on compounds

Concerns issues with computing quasiparticle corrections to the DFT eigenvalues - i.e., the self-energy within the GW approximation (-g n), or considering the Hartree-Fock exchange only (-x)

Moderators: Davide Sangalli, andrea.ferretti, myrta gruning, andrea marini, Daniele Varsano

Post Reply
hplan
Posts: 58
Joined: Thu Mar 26, 2009 2:26 pm

The applicability of GW0 calculation on compounds

Post by hplan » Fri Apr 24, 2009 11:23 am

Dear Developers:

I have some confusions about GW0's applicability on ternary system.
I have read several reviews and regular papers on GW0 calculations, and found that almost discussions were concentrated on elemental solids and binary compounds. There are rarely discussions on ternary or more compounds.
The limited application of GW0 calculation on these systems may be due to large computational loads required. But , i still wonder about its applicatbility from my initial testings on some systems and some literature works. Would you please give me some clarifications ?
I have read several papers done by Andrea and Lucia group. The impression i have is that a proper self-energy calculation requires exhausted examination on pseudopotentials. Therefore, when we turn to tenary compounds, it seems no unique choice of atomic configurations for pseudopotential generations. Recently, i did a series tests on a compound which has Zn, Rh and O elements. I found that no unique LDA-gap was predicted when i changed the configurations of pseudopotentials which though gave well transferrability and good lattice values. For example, if i just chose 3d states of Zn and 4d states of Rh as valence electrons, the LDA-gap of compound was 1.10eV, however the LDA-gap was 0.85 eV if i included 4s and 4p states of Rh as valence. But the story is not end. I did check the system with VASP package using USPP and PAW , and USPP gave me a 0.80 eV gap while PAW was a 1.0 eV gap although the valence configurations of Zn, Rh and O were the same. Therefore, i am confused to choose which combination of pseudopotentials if we need examine compound's GW gap, and how about its applicability on tenary compounds ?

Thanks!

Wishes,
Hai-Ping
Hai-Ping LAN,
Department Of Electronics,
Peking University, 100871,Beijing, CHINA

User avatar
myrta gruning
Posts: 239
Joined: Tue Mar 17, 2009 11:38 am
Contact:

Re: The applicability of GW0 calculation on compounds

Post by myrta gruning » Sat Apr 25, 2009 6:23 pm

Hallo Hai-Ping,

The problem you described has for the moment more to do with the LDA starting point, than with the GW calculation in itself. It is indeed disturbing that using the same valence configuration you obtained such differences on the electronic properties, and probably it is good to understand where this difference comes from before embarking on GW calculations. Does the difference concern just the band gap, or also the band structure? Do you use each time the equilibrium geometry?

At GW level what you have to take care of is that when you include d electrons in your PP, you should include all electrons of the same atomic shell (same principal quantum number). Not doing so generates large error in particular for the exchange-part of the self-energy. From a computational point of view this may indeed result in quite heavy calculations, because the large cutoff and the number of electrons.
Dr Myrta Grüning
School of Mathematics and Physics
Queen's University Belfast - Northern Ireland

http://www.researcherid.com/rid/B-1515-2009

hplan
Posts: 58
Joined: Thu Mar 26, 2009 2:26 pm

Re: The applicability of GW0 calculation on compounds

Post by hplan » Mon Apr 27, 2009 10:24 am

Dear Myrta,

Thanks for this comment!
myrta gruning wrote:Hallo Hai-Ping,

The problem you described has for the moment more to do with the LDA starting point, than with the GW calculation in itself. It is indeed disturbing that using the same valence configuration you obtained such differences on the electronic properties, and probably it is good to understand where this difference comes from before embarking on GW calculations. Does the difference concern just the band gap, or also the band structure? Do you use each time the equilibrium geometry?
Yes, all bands i compared were obtained from equilibrium lattices. The bands structures were very similar for two different combinations. Right now, the main difference is the LDA-gap: one is 0.80 eV, the other is 1.10 eV.
At GW level what you have to take care of is that when you include d electrons in your PP, you should include all electrons of the same atomic shell (same principal quantum number). Not doing so generates large error in particular for the exchange-part of the self-energy. From a computational point of view this may indeed result in quite heavy calculations, because the large cutoff and the number of electrons.
Is there any work disscussing this issue ? I read several works on 3d-metals and their oxides (i.e., Andrea's works on Cu, and Lucia's works on Cu2O and V2O et al.), and get impressed of the sublte role of 3s/p states for results. I donot know whether this subtle effect will be held for 4s/p staes of 4d metals.


Regards,
Hai-Ping
Hai-Ping LAN,
Department Of Electronics,
Peking University, 100871,Beijing, CHINA

User avatar
myrta gruning
Posts: 239
Joined: Tue Mar 17, 2009 11:38 am
Contact:

Re: The applicability of GW0 calculation on compounds

Post by myrta gruning » Mon Apr 27, 2009 11:48 am

hplan wrote: Is there any work disscussing this issue ? I read several works on 3d-metals and their oxides (i.e., Andrea's works on Cu, and Lucia's works on Cu2O and V2O et al.), and get impressed of the sublte role of 3s/p states for results. I donot know whether this subtle effect will be held for 4s/p staes of 4d metals.
You can see e.g.
Phys. Rev. Lett. 75, 3489 - 3492 (1995)
Quasiparticle Band Structure of CdS
Michael Rohlfing, Peter Krüger, and Johannes Pollmann

It happens to treat a system with 4d. In fact there is nothing special about the 3d (or the 4d). What does matter is the spatial localization of the d electron with respect to the s and p electron of the same atomic shell. This is illustrated e.g. in Fig.2 of the above-mentioned paper. You can also plot the radial wave function of your pseudopotential to realize that even if the s and p electron have usually higher binding energy than the d of the same shell, they are localized in the same region. For that reason since the exchange part of the self energy does not depend on the energy but on the localization and extent of the wavefunctions one can expect a large error when considering the d as valence, and s.p as core.
So it is a general feature, one has to worry about each time "the valence states of the relevant shell have a very similar spatial localization".

Cheers
m
Dr Myrta Grüning
School of Mathematics and Physics
Queen's University Belfast - Northern Ireland

http://www.researcherid.com/rid/B-1515-2009

hplan
Posts: 58
Joined: Thu Mar 26, 2009 2:26 pm

Re: The applicability of GW0 calculation on compounds

Post by hplan » Tue Apr 28, 2009 2:11 pm

Dear Myrta,
Thank you for the literature and experience shared.

Best Wishes,

Hai-Ping
Hai-Ping LAN,
Department Of Electronics,
Peking University, 100871,Beijing, CHINA

haseebphysics1
Posts: 169
Joined: Sat Aug 17, 2019 2:48 pm

Re: The applicability of GW0 calculation on compounds

Post by haseebphysics1 » Tue Jul 27, 2021 1:30 pm

myrta gruning wrote: Sat Apr 25, 2009 6:23 pm Hallo Hai-Ping,
...

At GW level what you have to take care of is that when you include d electrons in your PP, you should include all electrons of the same atomic shell (same principal quantum number). Not doing so generates large error in particular for the exchange-part of the self-energy. From a computational point of view this may indeed result in quite heavy calculations, because the large cutoff and the number of electrons.
Dear Developers,

I was wondering that the same rule as referred above holds for those orbitals which don't have much density (using PDOS from DFT) near the Fermi level?

For instance, the Bi has the following configuration, [Xe] 4f^[14]5d^[10]6s^[2]6p^[3], here in SG15 PP, the 5d^[10]6s^[2]6p^[3] are taken in the valance which means, 5s^[2] and 5p^[6] are part of the core. If, 5d^[10] has negligible density near the Fermi level, so in that case, can we avoid including 5s^[2] and 5p^[6] in the core explicitly?

Thanks,
Haseeb Ahmad
MS - Physics,
LUMS - Pakistan

User avatar
Daniele Varsano
Posts: 3773
Joined: Tue Mar 17, 2009 2:23 pm
Contact:

Re: The applicability of GW0 calculation on compounds

Post by Daniele Varsano » Tue Jul 27, 2021 1:51 pm

Dear Haseeb,
I suggest you to test your pseudo, e.g. calculating and HF gap in order to see the effect of the semi-core electrons.

Best,
Daniele
Dr. Daniele Varsano
S3-CNR Institute of Nanoscience and MaX Center, Italy
MaX - Materials design at the Exascale
http://www.nano.cnr.it
http://www.max-centre.eu/

Post Reply