Unable to generate input file

[xjxiao]

Dear all.
After generating the X.save folder using Quantum ESPRESSO, I entered the folder and used the p2y command to create the SAVE folder. Then, I initialized it by entering the yambo command

yambo_setup.png

, and successfully generated the r_setup file (). However, when I entered yambo -gw0 p -g n -V par -F gw.in, the generated input file was very strange (), with very few parameters. What could be the issue here?

Thanks!!!

yours,
Xiao

[Daniele Varsano]

Dear Xiao,
this is because the k grid is not recognized as uniform, so GW calculation are not possible.
This is indicated in the r_setuo file:

Code: Select all

 X grid is uniform                                : no

You can try to delete the./SAVE/ndb.kindx
and run again the setup adding this string in the input:
NoDiagSC
and see if it does the problem.

If not, check your k point sampling in the QE file.

Best,

Daniele

[xjxiao]

Dear Xiao,
this is because the k grid is not recognized as uniform, so GW calculation are not possible.
This is indicated in the r_setuo file:

Code: Select all

 X grid is uniform                                : no

You can try to delete the./SAVE/ndb.kindx
and run again the setup adding this string in the input:
NoDiagSC
and see if it does the problem.

If not, check your k point sampling in the QE file.

Best,

Daniele

Dear Daniele,
Thank you very much for your response! I have successfully generated the input file, but it seems a bit strange that in the resulting `r` file, it still shows `Grid dimensions: 1 9 1` (As shown in blue)(I set it to 9 9 1). However, later in the file, I can see 41 irreducible k-points(As shown in green). Will this discrepancy affect the calculations?
The 2D material I am calculating is a semimetal under PBE, with the Fermi energy crossing both the 60th and 61st bands. Why does the GW calculation result show that only bands 1-58 are fully occupied(As shown in red), and why is the 59th band not fully occupied? Is there an issue with the calculation?

r-gw_HF_and_locXC_gw0_dyson_em1d_ppa_01.png

This system has an indirect bandgap. For the convergence criteria of GW calculation parameters, should I focus on the minimum direct bandgap or the indirect bandgap (the current calculation is time-consuming, and I hope to reduce the k-point range)?
Under HSE06, the indirect bandgap of this system is positive. Would calculating GW based on HSE06 results be less time-consuming compared to calculating GW based on PBE results?
Thanks!!!
Yours,
Xiao

[Daniele Varsano]

Dear Xiao,

`Grid dimensions: 1 9 1` (As shown in blue)(I set it to 9 9 1). However, later in the file, I can see 41 irreducible k-points(As shown in green). Will this discrepancy affect the calculations?

It seems Yambo is having difficulties in recognizing your cell, are you using ibrav=0, or a shifted k point grid?
In any case, it seems that the Q points are correctly generated in 2D. The calculation should be fine. You can check in the r_setup the list of k points and see if they are equivalent to the one of QE.

Why does the GW calculation result show that only bands 1-58 are fully occupied(As shown in red), and why is the 59th band not fully occupied? Is there an issue with the calculation?

This should be a matter of the Occupation Threshold, you can change it, anyway also this should not be an issue.

This system has an indirect bandgap. For the convergence criteria of GW calculation parameters, should I focus on the minimum direct bandgap or the indirect bandgap (the current calculation is time-consuming, and I hope to reduce the k-point range)?

That's up to you, it should be more or less equivalent. For convergence purpose you can focus of few k points and bands.

Under HSE06, the indirect bandgap of this system is positive. Would calculating GW based on HSE06 results be less time-consuming compared to calculating GW based on PBE results?

No, Yambo calculations will be equivalent in terms of resources. In QE instead calculating many unoccupied states at HSE06 level is painful.
Please note that at moment, HSE06 is not supported, while PBE0 and GAUPBE are supported.

Best,
Daniele

[xjxiao]

Dear Daniele,

It seems Yambo is having difficulties in recognizing your cell, are you using ibrav=0, or a shifted k point grid?
In any case, it seems that the Q points are correctly generated in 2D. The calculation should be fine. You can check in the r_setup the list of k points and see if they are equivalent to the one of QE.

Yes, as you mentioned, I used ibrav=0 in the scf.in file and set:
K_POINTS { automatic }
9 9 1 1 1 1
I just tried changing K_POINTS { automatic } to 9 9 1 0 0 0, but I still cannot correctly generate the input file when using the command yambo -g n -p p -F gw_ppa.in. Is it not allowed to use ibrav=0? Or, if ibrav=0 is used, I need to add NoDiagSC in r_setup.

No, Yambo calculations will be equivalent in terms of resources. In QE instead calculating many unoccupied states at HSE06 level is painful.
Please note that at moment, HSE06 is not supported, while PBE0 and GAUPBE are supported.

I am a bit confused by your explanation. I have seen GW calculations based on HSE in the literature. What do you mean by "HSE06 is not supported"?


I hope to achieve rapid convergence of GW calculations for a 2D system with fewer k-points. However, the GW calculation still yields a negative bandgap, which does not match the positive bandgap obtained from HSE calculations. Can rim-cut and rim-w help me achieve this?

rim.png

I would like to ask how to set the parameters RandQpts, RandGvec, and RandGvecW. I have seen different settings for CUTGeo in various tutorials. What is the difference between setting it to box Z and slab Z?

I am a beginner with Yambo and find many aspects confusing. Thank you very much for your patient guidance!

yours,
Xiao

[Daniele Varsano]

Dear Xiangjun,

Is it not allowed to use ibrav=0? Or, if ibrav=0 is used, I need to add NoDiagSC in r_setup.

Yes it is allowed but there might be difficulties in recognizing the k sampling as uniform. The alternative NoDiagSC algorithm helps in that.

What do you mean by "HSE06 is not supported"?

It means that it is not supported. Due to the different way of treating its long wavelength limit in Yambo and QE, a systematic error arises in the GW QP energies. We have now developed a different algorithm to support any functional that not require the evaluation of Vxc, and hopefully it will be released soon. If you have seen GW@HSE in literature, either calculation were not done with Yambo, either warning were ignored and the calculations suffers of a systematic error.

Can rim-cut and rim-w help me achieve this?

You need to check the convergences (bands, cutoff in the dielectric matrix etc..), and yes rim and rim-w helps for converngences wrt k points.
Anyway, please note that it is not guaranteed that GW will open the gap as, being GoWo, at difference of HSE06 calculations it lacks of self consistency in the wavefunctions.

I would like to ask how to set the parameters RandQpts, RandGvec, and RandGvecW. I have seen different settings for CUTGeo in various tutorials. What is the difference between setting it to box Z and slab Z?

These are the Q points for the stochastic integration and the number of G vec outside the BZ to be considered for the integration of the exchange and W terms respectiuvely. You can set RandQpts few millions (2 or 3 millions usually is accurate) and one or few tenths for the G vectors, for large vectors is not needed.

box Z and slab Z are two different geometries, slab Z is analytic, and it is suggested. RimW works only using "slab z".

Best,

Daniele

[xjxiao]

Dear Daniele,
Thank you for your answer, but I still have some details unclear, please help me again.

These are the Q points for the stochastic integration and the number of G vec outside the BZ to be considered for the integration of the exchange and W terms respectiuvely. You can set RandQpts few millions (2 or 3 millions usually is accurate) and one or few tenths for the G vectors, for large vectors is not needed.

1. Do the parameters RandQpts, RandGvec, and RandGvecW need convergence testing?
2. For the following parameters, how should the z-component be set? Should it refer to the value in the r_setup file or the vacuum layer thickness in the QE calculation file?

Code: Select all

% CUTBox
 0.000000 | 0.000000 | 0.000000 |        # [CUT] [au] Box sides
%

3. My system is a negative-gap semimetal before GW calculation and becomes a small-gap semiconductor after GW calculation. How should rimw_type be set?

Code: Select all

rimw_type= ""             # [RIM-W] Screened interaction limit metal/semiconductor/graphene

Best,
Xiao

[Daniele Varsano]

Dear Xiangjun,

1. Do the parameters RandQpts, RandGvec, and RandGvecW need convergence testing?

Usually not, RandQpts should be set to a large value (few millions) to ensure a small stochastic error. G vectors can be checked but the order of ten is usually enough.

2. For the following parameters, how should the z-component be set?

The RIMW technique needs the "slab z" geometry, box is not allowed so the CUTBox field is ignored. You can delete it from the input or leave as it is.
In slab z geometrical parameter are automatically set.

3. My system is a negative-gap semimetal before GW calculation and becomes a small-gap semiconductor after GW calculation. How should rimw_type be set?

In this case you can set it as semiconductor.

Best,
Daniele

[xjxiao]

Dear Daniele,
Thanks for your reply!
May I ask why the input file cannot be generated using the following command? I am using Yambo version 5.1.0, and I noticed that the reference files provided on the Yambo website were also generated using version 5.1.0.

problem.png

I just downloaded Yambo version 5.3.0, and it can successfully generate the input file. However, my previous parameter convergence tests were all done using version 5.1.0, so I suspect it would be better to continue using version 5.1.0.
When using the command `yambo -g n -p p`, the input file generated by Yambo 5.3.0 includes the following additional items compared to Yambo 5.1.0.

Code: Select all

el_el_corr                       # [R] Electron-Electron Correlation

Will there be significant differences in the calculation results between the two versions?
Best wishes!
Xiao

[Daniele Varsano]

Dear Xiangjun,

I suggest you to upgrade to yambo.5.3, the old version 5.1 is not maintained anymore.
The problem here arise because the stochastic integration of W (-rw command) is a new feature and was not present in the old version, and it is misinterpreted.

The el_el_corr is ok, and you do not need to worry about it.

Best,
Daniele