G0W0 @ PBE+U with Yambo

[Daniele Varsano]

Dear Haseeb,
Indeed at the moment it is not possible to perform GW on top of DFT+U. Hopefully it will be coded in the near future. Anyway, if you are interested in BSE spectra you can do it provided that you use the dft+u gap or you model the qp correction by a scissor operator.
Best,
Daniele

[haseebphysics1]

Dear Haseeb,
Indeed at the moment it is not possible to perform GW on top of DFT+U. Hopefully it will be coded in the near future. Anyway, if you are interested in BSE spectra you can do it provided that you use the dft+u gap or you model the qp correction by a scissor operator.
Best,
Daniele

Thanks dear Daniele.

[haseebphysics1]

Dear Yambo community,
I'm trying calculating optical properties using BSE.

I want to apply scissor shift to conduction band KS energies to correct the bandgap from DFT. But I'm having following minor issues.
1: I don't know the exact syntax of "KfnQP_E"
% KfnQP_E
scissor in eV | shift in CB | shift in VB |
%
I have seen in some examples that shifts in CB and VB are kept at exactly 1.0000! What does that mean? Does not that implies that "scissor" will apply to the conduction band as well as valence band and if it raises the energy of CBs then will it also raise the energy of VBs (then no change in the gap!) ?

2: I have a semiconductor with small bandgap (1.6 eV) but DFT underestimates it so much that it gives the metallic state! Now the yambo output files are also sending the warning messages in report files that system is metallic and Drude frequency is missing etc. (Even I have applied scissor shift of 1.6 eV : 1.60000 | 1.00000 | 1.0000|). What we can do in this sort of situation? I have applied gaussian smearing and degauss value 0.02 in pwscf calculations?

3: Why we need nscf calculation for yambo? Because we already have generated KS orbitals from SCF.

Regards,

Haseeb Ahmad
MS-Physics,
LUMS, Pakistan.

[Daniele Varsano]

Dear Haseeb,

I don't know the exact syntax of "KfnQP_E"

The first entry is the scissor in energy (eV), ie the opening of the gap.
A common effect of the GW correction besides the opening of the gap is a stretching of the QP bands wrt the DFT ones. This can be obtained by a linear
fit od the QP energy wrt KS energies. The slope for valence and conduction band can be different. So, the second and third fields are 2 adimensional numbers indicating the slope for the conduction and valence band. If they are both equal to 1 it means that only the scissor is applied without changing the slope of the bands.

2: I have a semiconductor with small bandgap (1.6 eV) but DFT underestimates it so much that it gives the metallic state! Now the yambo output files are also sending the warning messages in report files that system is metallic and Drude frequency is missing etc. (Even I have applied scissor shift of 1.6 eV : 1.60000 | 1.00000 | 1.0000|). What we can do in this sort of situation? I have applied gaussian smearing and degauss value 0.02 in pwscf calculations?

Note that KfnQP_E apply a shift in the energies entering in the kernel of the BSE and not in other parts of the code (e.g. screening). If you want to apply a scissor also in the screening you need to use the XfnQP_E variable and GfnQP_E for applying the scissor in the Green function (same syntax of KfnQP_E).

3: Why we need nscf calculation for yambo? Because we already have generated KS orbitals from SCF.

Usually, you need many empty bands for the calculation of the screening and response function in general. Nscf allows calculation of empty states once you have a converged ground state density (i.e. the Hamiltonian). You can calculate them also using scf but if you need many of them this is not advantageous.

Best,
Daniele

[haseebphysics1]

Respected Daniele,

The first entry is the scissor in energy (eV), ie the opening of the gap.
A common effect of the GW correction besides the opening of the gap is a stretching of the QP bands wrt the DFT ones. This can be obtained by a linear
fit od the QP energy wrt KS energies. The slope for valence and conduction band can be different. So, the second and third fields are 2 adimensional numbers indicating the slope for the conduction and valence band. If they are both equal to 1 it means that only the scissor is applied without changing the slope of the bands.

I don't want to calculation the QP corrections using GW calculation since my system is highly correlated containing transition metals oxide. I only want to do scissor corrections to KS eigenvalues and then calculate BSE absorption spectra. So, my questions are:

Q1: How can I know the | shift in CB | shift in VB in the % KfnQP_E, for I am not doing any GW. Moreover, since the slope of the bands is related to effective carrier mass and is also critical in the positions of the peaks of the absorption spectra, so, will it affect the BSE spectra as well?

Q2: Do we need to apply the scissor shift in the static dielectric screening as well and again what will be the source of the slope of the CB and VB values?

Q3: Does the new Yambo version supports Ultrasoft or PAW PPs? NCPP takes large K.E. cutoff which I think to increase a lot of CPU time and memory. And can Yambo supports advanced functionals like those in meta-GGAs? I know a few hybrid functionals are supported.

Regards,
Haseeb Ahmad,
MS - Physics
LUMS - Pakistan.

[haseebphysics1]

kindly provide an answer (whenever you get some time), I am waiting for your kind responce.

[Daniele Varsano]

Dear Haseeb,

I've been travelling these days, please be patient.

A1: When scissor+stretching approximation is feasible, usually the idea is to calculate GW corrections for a few states and from them, it is possible to obtain the scissor and the stretching parameters by a fit. In your case, if you do not want/or can't calculate GW spectra, also for the scissor I think you need to resort to some others consideration (e.g. experimental gap? I do not know how you have estimated your scissor). About the stretching parameters, usually, they change from 1 from a few percentages, so probably, if you do not have a reference band structure it is better to leave to 1.0.
Yes, the stretching of the bands can affect the absorption spectra, but again, should be not dramatic.

A2: Usually the scissor shift is applied in the screening when doing eigenvalue GW self-consistency, in a standard BSE calculation on top of G0W0 the static screening is calculated by using the KS structure, so no need to apply scissor and stretching.

A3: Yambo support Ultrasoft pseudopotential, but this is an experimental feature and it is discouraged. Please note that new Hamman pseudopotentials (Optimized Norm-Conservinng Vanderbilt PSeudopotentials) are not that hard and are perfectly suitable for Yambo.
Meta-GGA functionals at the moment are not supported.

Best,
Daniele

[haseebphysics1]

Thank you very much and sorry for the inconveniene due to my 2nd message.


Regards,
Haseeb Ahmad
MS - Physics.

[haseebphysics1]

Dear Daniele,

Q1: If I am just willing to apply the scissor operator to open the gap to the experimental gap then can I avoid this parameter altogether and manually shift the x-axis energies to x > x + shift (in eV) at the end? Will this do the job?

% KfnQP_E
scissor in eV | shift in CB | shift in VB |
%

Q2: Is it okay and a good idea to apply sophisticated optical absorption theories just on top of DFT KS energies without applying the GW QP calculations (using scissor corrections only). Actually, I have transition metal oxide as a system and I have had heard Dr. Andrea Marini saying Yambo is not for d-block elements! But recently I have found your paper on TiO2 using QE and Yambo! My system is a similar one, so can I also go with the GW corrections before BSE?

https://arxiv.org/abs/1811.04280

Q3: And finally, In BSE calculations static screening, yambo -b -F screening.in

why the q-point parameter like QpntsRXs not created? It is not there to select the absorption or eels spectra?
1 | 1 | # [Xd] Transferred momenta

Is it not essential in the BSE?

Thanks,

[Daniele Varsano]

Dear Haseeb,

A1: Yes, that variable it is meant for applying a scissor
A2: Applying a scissor or calculate GW energies are two methods to consider quasiparticle correction, note in both case the wavefunctions are the KS wavefunction. If KS wavefunctions are not accurate, GW does not solve the problem. I do not which claim of Andrea you are referring, as he calculated GW band structure for copper and silver (see e.g. Phys. Rev. Lett. 88, 016403 (2001). You can do the calculation and evaluate the quality: for doped TiO2 we realized that the effect of GW correction was to locate the defect states, found in conduction at PBE level, inside the gap. There are indeed other cases, where GW@PBE fails.
A3: It is not created, because it is calculated for all the q by default. Have you tried?

Best,
Daniele