underestimate of gap in G0W0 calculation

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
baiyunfei
Posts: 15
Joined: Thu Dec 01, 2022 12:53 pm

underestimate of gap in G0W0 calculation

Post by baiyunfei » Thu Dec 22, 2022 2:12 pm

I am trying to carry out G0W0 calculation in monolayer H-WSe2. I used SG15 pseudopotential to get nscf wavefunction on a 27*27*1 mesh of 40+200 bands ,Then I carried G0W0 calculation ,some convergence parameters are as follow :
EXXRLvcs= 60 Ry
VXCRLvcs= 10 Ry
NGsBlkXp= 12 Ry
GbndRnge 1 | 240 |
It gave a QP correction of about 0.6eV ,which results in a gap of around 1.9 eV. This result is considerably smaller than experiment and other calculation results(2.3-2.5eV). I also tried to use a denser mesh 33*33*1, more bands 300*300, bigger cutoff . The QP correction seems quite stable. In addition the QP correction seems vary a lot with k .I want to know where my problem is and how to solve it.
You do not have the required permissions to view the files attached to this post.
Yunfei Bai
Institute of Physics Chinese academy of sciences, Beijing

sdwang
Posts: 287
Joined: Fri Apr 09, 2010 12:30 pm

Re: underestimate of gap in G0W0 calculation

Post by sdwang » Fri Dec 23, 2022 6:43 am

Dear Yunfei,
I found two strange things in your output:
1). The cutoff box is set to 32 a.u. and the Z direction is 40 a.u. of your system (try to use 'Slab Z' and leave the cutbox value empty).
2). "ElecTemp= 0.025861" and "BoseTemp= 0.025861" are weired. The DFT gap is 1.26 eV of your system and I think these two options should be 0 instead of finite value.

I am not sure this is the reason for your problem. But you can improve these options to optimize your results.
Best

Shudong
S. D. Wang
IMU,HOHHOT,CHINA
E-mail: sdwang@imu.edu.cn

baiyunfei
Posts: 15
Joined: Thu Dec 01, 2022 12:53 pm

Re: underestimate of gap in G0W0 calculation

Post by baiyunfei » Sun Dec 25, 2022 9:13 am

I tried to use slab z , it only made results worse ,predicting a QP gap prediction of around 0.5 eV . Also , sticking to 'box z ' and changing the box size to 39 a.u seems have little influence on QP gap(resulting in a QP gap correction of 0.66eV) . As for ElecTemp, I think 0.026eV is pretty small for a gap of around 1-2 eV ,I don't konw what does BoseTemp mean, but it seems have nothing to do with G0W0 calculation. I think my problem may lie in somwhere else, but I can't figure it out.
Yunfei Bai
Institute of Physics Chinese academy of sciences, Beijing

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

Re: underestimate of gap in G0W0 calculation

Post by Daniele Varsano » Mon Dec 26, 2022 9:02 am

Dear Yunfei,

the main problem I can see here is the cutoff set for the <Vxc>, which is very low, if you do not have memory problem I would leave it to the maximum of the Gvectors (the one used for the density: 526167 RL, here you are using 969 RL, in Ry should be 4 times the cutoff you used for the wfs). This cause inconsistency with the <Vxc> calculated by QE that should be subtracted. Next, of course, check the convergence of other parameters, including k points.

* electronic temperature here should be not a problem
* cutoff, I suggest slab z as suggested by Shudong, but if you want to use the box, yes use a larger value (39/40 should work)

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/

baiyunfei
Posts: 15
Joined: Thu Dec 01, 2022 12:53 pm

Re: underestimate of gap in G0W0 calculation

Post by baiyunfei » Wed Dec 28, 2022 9:51 am

Dear Daniele
Thanks for your reply.
In fact , I have tried to increase the cutoff for exchange and Vxc, I also tried to use a densener k mesh. The QP gap correction convergences well with these parameters , but it convergences to a pretty small value of about 0.65eV. I used these parameter just because an article ''Exciton Radiative Lifetimes in Two-Dimensional Transition Metal Dichalcogenides'' adopt similiar parameter and got resonable results.
I doubt whether it has something to do with the pseudopotential I used (I heard that semi-core electrons may have great impact on the result , I am not sure whether sg15 pseudopotential is appropriate for G0W0 calculation) .Also I don't know whether a warning of ''Found non closed shells. Max cutoff will be reduced'' during setup is related to this problem.
The attachment is the result of denser kmesh and larger cutoff, the pseudopotentials I used, and the setup report

Sincerely
You do not have the required permissions to view the files attached to this post.
Yunfei Bai
Institute of Physics Chinese academy of sciences, Beijing

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

Re: underestimate of gap in G0W0 calculation

Post by Daniele Varsano » Wed Jan 04, 2023 12:43 pm

Dear Yunfei,

the report you posted in truncated, so not really useful.
From the output anyway I can see, you are using the cutoff box with a wrong z-side (30au) it should be something like 40 au.
As suggested before, it could be important to use all the G vectors of the density to evaluate correctly the <Vxc> term (here 526167).

Sure psp can have a great impact, you can try to use pseudodojo oncv psuedo potential, in this case you need to pay attention to the non-local core correction.
They are activated using the keyword UseNLCC in the gw input file, anyway if it is better to include or neglect them it is not totally clear for a GW caluclation as they will enter in the <Vxc> evaluation only and not in the other self energy terms.

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/

baiyunfei
Posts: 15
Joined: Thu Dec 01, 2022 12:53 pm

Re: underestimate of gap in G0W0 calculation

Post by baiyunfei » Mon Jan 09, 2023 2:47 am

Thanks a lot for your help.
All problems disapeared, when I use pseudodojo oncv psuedo potential for my calculation. Probably ,it is because
sg15 psp considers 28 valence electrons for W atom, which doesn't account for all electrons of n=4. On the other hand, pseudodojo with the number of valance electrons= 14 for W atom , include all electrons of n=5.
From somewhere, I heard that incomplete innclusion of valance electrons may bring problems to gw calculation, that possibly where my problem is.
Thanks again for your suggestions.
Yunfei Bai
Institute of Physics Chinese academy of sciences, Beijing

Post Reply