GW Band structure changes caused by QP correction

[CJS]

Dear all,

After performing standard SCF and NSCF calculations with Quantum ESPRESSO, I carried out a G₀W₀ quasiparticle band calculation with Yambo and interpolated the band structure using ypp_bands.in. The results show a severe distortion of the band structure after applying the QP correction. Specifically, the conduction band corrections near the Gamma point are much smaller than those in other regions, leading to a strongly deformed band structure. What could be the reason for this?

I have attached the following for reference:

Band structure with and without QP correction

Input files: gw_ppa_all_Bz.in and ypp_bands.in

Yours
chenjiashu
Master student, Xiangtan University

[Daniele Varsano]

Dear Chenjiashu,

this is due to interpolation problem of ypp that in some cases is not accurate enough.
My suggestion is to perform the interpolation using the yambopy utility which is much more accurate:
Here you can find a tutorial:
https://wiki.yambo-code.eu/wiki/index.p ... structures


Best,
Daniele

[Davide Sangalli]

If you try to interpolate the DFT bands, which result do you get ?
(Same input file for the GW bands, without GfnQPdb= "E < ./all_Bz/ndb.QP")

Also, uncomment BANDS_built_in. You will get a file with the points which are on the initial grids, and belongs to the path.
You can plot them together with the interpolated bands, to double check if the distortion comes from the GW correction or from the interpolation.

Last, you can try to increase the value of

Code: Select all

INTERP_Shell_Fac= 20.00000       # The bigger it is a higher number of shells is used

Best,
D.

[CJS]

Dear Daniele and Davide,

First of all, thank you both very much for your prompt and highly valuable suggestions!

I have carefully followed both of your advice and performed the corresponding tests:

1. Regarding the yambopy interpolation (Daniele's suggestion):
I successfully used yambopy to interpolate the band structure, and it indeed solved the severe deformation problem . Thank you for pointing me to the tutorial!

2. Regarding the ypp tests (Davide's suggestion):
I also followed your steps to investigate the ypp behavior. I uncommented BANDS_built_in, increased INTERP_Shell_Fac= 40.00000, and ran ypp in two cases: with and without commenting out GfnQPdb= "E < ./all_Bz/ndb.QP". Unfortunately, the severe distortion in the ypp GW interpolated curves remains present in both cases when compared to the built-in points. I have attached these new comparison plots for your reference.

My New Concern:
Although yambopy successfully bypassed the interpolation issue, I noticed a new problem. After the calculation, I plotted the DFT vs. GW energies to check their linear relationship. As shown in the attached file, the linear relationship looks very bad and highly scattered.

Given this poor DFT-GW correlation, I strongly suspect there might still be some fundamental issues in my G0W0 calculation step (perhaps related to convergence or parameter settings), rather than it just being a post-processing interpolation artifact.

Could this bad linear relationship indicate a problem with my GW setup? Do you have any suggestions on which parameters I should double-check to improve this?


The k-grid used in the QE calculation is 36×36×1. Please let me know if you need me to provide any other parameter settings or input files.


Thank you again for your time and continuous help!

Best regards,

chenjiashu

[Davide Sangalli]

Two comments:

1) non-linear behaviour of GW correction is not necessarly bad, and, indeed, I'd say it is physical. There is not strict rule stating that the energy dependence should be strictly linear.

2) My impression is that the quality of the interpolation is the same in bewteen ypp and yambopy.
However, ypp keeps using the bands ordering of DFT when drawing lines, while yambopy uses the new bands ordering imposed by the GW correction.
(There are clearly some bands inverted here). The ypp ordering gives a bad looking band structure

D.