GW band interpolation

Anything regarding the post-processing utility (e.g. excitonic wavefunction analysis) is dealt with in this forum.

Moderators: Davide Sangalli, andrea marini, Daniele Varsano

Post Reply
Li Tenghui
Posts: 6
Joined: Mon Nov 15, 2021 8:58 am

GW band interpolation

Post by Li Tenghui » Wed Jan 05, 2022 2:04 pm

Dear developers:
I'm trying to plot a GW band structure for CdSe using Yambo 5.0.4, following the tutorial link below:
http://www.yambo-code.org/wiki/index.ph ... rial:_h-BN
I've calculated the DFT band structure and the GW band structure. However, It seems that the GW correction only changed the gamma point energy and output a weird structure.
band_structure.pdf
Can you help me deal with it? I uploaded some of the calculation input and output files, please let me know if you need any other file.
Thanks,
Tenghui
You do not have the required permissions to view the files attached to this post.
Li Tenghui,
Department of Chemistry,
Zhejiang University, Hangzhou, China

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

Re: GW band interpolation

Post by Daniele Varsano » Thu Jan 06, 2022 6:08 pm

Dear Li,

a possible reason is that you have the QP energies calculated in a subset of k points and these are not enough to have a good interpolation.
I'm not sure that this is the problem but I would try to calculate the QP for all the k points avoiding calculating them for deep states as you do not need them.
So something like:
1|35|53|62|

and see if the quality of the interpolation improves.


Next, please note that your GW calculation is extremely non-converged:
1)

Code: Select all

NGsBlkXp= 1  

here you are considering a scalar approximation which is not reliable. This is a parameter that has to be converged (you should make a series of calculations varying this cutoff e,g. 2 Ry, 4 Ry , 6 Ry etc.)

2) 100 bands in the polarization and GW summation could be not enough considering you have 52 occupied bands.

Here you can find a tutorial on how to perform convergences for a GW calculation:
http://www.yambo-code.org/wiki/index.ph ... 21_version

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/

Li Tenghui
Posts: 6
Joined: Mon Nov 15, 2021 8:58 am

Re: GW band interpolation

Post by Li Tenghui » Thu Feb 17, 2022 4:49 am

Dear Daniele:
Thanks for your advice, after checking the convergence, I can get a normal band structure now. However, the calculated band gap (about 0.83 eV) is small than the experimental band gap, which is about 1.66 eV. The following file shows the convergence test I've done before.
convergence_test.pdf
Then I tried to perform self-consistent GW on the eigenvalues, the band gap would finally converged at approximately 1.10eV after two or three iterations. The calculation costs a lot of time so I have to reduce the number of bands, can you give me some suggestions about improving the calculation efficiency and accuracy? The following input file shows the parameters currently used, please tell me if there are any mistakes.
yambo.in.txt
It will be nice of you if you can help me handle this.
Thanks,
Tenghui
You do not have the required permissions to view the files attached to this post.
Li Tenghui,
Department of Chemistry,
Zhejiang University, Hangzhou, China

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

Re: GW band interpolation

Post by Daniele Varsano » Thu Feb 17, 2022 10:00 am

Dear Li,

I can see a couple of issues here:

1) are you using pseudo-dojo pseudopotential?
These pseudos contain non-linear core correction and this produces a mismatch in the <Vxc> value. Check if this is the case: this is reported in the psp file. If it is the case you need to take nlcc into account in the Vxc calculation by using the flag UseNLCC in your input file (you should remove the ndb.HF* database first).
Alternatively and may be suggested, is to use pseudopotential without nlcc that can be found here:
http://www.quantum-simulation.org/potentials/sg15_oncv/


2) You are using the automatic GW ev self-consistency. Some problems have been observed in this procedure and it is suggested to follow this flow for the self-consistency:
http://www.yambo-code.org/wiki/index.ph ... alues_only

Anyway, before going to evGW self-consistency I would inspect point one first. Next, maybe you can resort to GWo so you do not need to recompute the screening and hence you do not need to calculate QP corrections for many bands as you are doing now.

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/

Li Tenghui
Posts: 6
Joined: Mon Nov 15, 2021 8:58 am

Re: GW band interpolation

Post by Li Tenghui » Sun Feb 20, 2022 5:44 am

Dear Daniele:
Thanks for your advice. I tried to use pseudopotentials without nlcc. However, the G0W0 calculation still gives a band gap about 0.87eV. The following files are the input and output file in the calculation.
yambo.in.txt
o.qp_G0W0.txt
Here are pseudopotential files I used in QE calculation.
pseudopotentials.zip
Please let me know if there are any mistakes.

Thanks,
Tenghui
You do not have the required permissions to view the files attached to this post.
Li Tenghui,
Department of Chemistry,
Zhejiang University, Hangzhou, China

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

Re: GW band interpolation

Post by Daniele Varsano » Mon Feb 21, 2022 10:28 am

Dear Li,

it seems you are not including the 3d electrons in your Se pseudo. This can have a very large impact! In particular in the exchange self energy.

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/

Li Tenghui
Posts: 6
Joined: Mon Nov 15, 2021 8:58 am

Re: GW band interpolation

Post by Li Tenghui » Mon Feb 28, 2022 3:24 am

Dear Daniele:
Thanks a lot for point out the mistake! However, I cannot find a proper norm-conserving pseudopotential which considers the 3d electron of Se and without nlcc at the same time. It will be kind of you to tell me where to find one. Otherwise maybe I need to generate it by myself.
Then I tried to calculate the band structure for CdS bulk to avoid d electrons. I just used the same parameters as in CdSe calculation without checking convergence, and only performed a G0W0 calculation. It works out band gaps with 1.66 eV for zinc-blende CdS, and 1.72 eV for wurtzite CdS, as the experimental band gaps are 2.3eV and 2.4eV, respectively. The calculated CdSe band gaps are 0.87 eV and 0.95 eV for zinc-blende and wurtzite structures, while the experimental results are 1.66 eV and 1.85 eV.
It seems that my GW calculation has a discrepancy of 0.8~0.9 eV with experimental value. Can you please help me deal with it? I've reproduced the band gap and GW band gap of hBN in tutorial, so the compilation seems successful.
Thanks,
Tenghui
Li Tenghui,
Department of Chemistry,
Zhejiang University, Hangzhou, China

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

Re: GW band interpolation

Post by Daniele Varsano » Mon Feb 28, 2022 6:24 pm

Dear Li Tenghui,
there could be so many reasons why GW calculations and experiments do not agree.
I suggest you to look in the literature and compare with other GW calculations. Be sure you are using the right structure and you have a correct starting DFT point.
Next as already stated you are using a possible wrong procedure to include eV self consitency.
Finally you can try to repeate your calculation using nlcc pseudo with d electrons activating the flag UseNLCC in the input file (remove first the ndb.HF* database).
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