GW band structure with SOC

[8813204602]

Dear Yambo community forum

I calculated the band structure for a 2D magnetic material with and without SOC effect. Without SOC, both the DFT and GW band structures are correct, but with SOC, as shown in the figure, the GW calculation is incorrect. I would be very grateful if you help in resolving this issue.

It should be noted that the effect of the spin-orbit in the material under study is weak, so it is expected that it does not change the DFT band structure. While the DFT band structure obtained from Yambo with and without SOC,in the valence band, is different.

Best regards,
Mitra Helmi

[Daniele Varsano]

Dear Mitra,

please note that Yambo does not calculate the DFT band structure. The DFT ground state is calculated by QE (or Abinit) and yambo takes it as input.
So probably you should inspect your DFT calculation to spot the source of the discrepancy (e.g. semicore electrons in PSP, different parameters...etc.).
Best,
Daniele

[8813204602]

Dear Daniele,

Thank you very much for your reply.

please note that Yambo does not calculate the DFT band structure. The DFT ground state is calculated by QE (or Abinit) and yambo takes it as input.
So probably you should inspect your DFT calculation to spot the source of the discrepancy (e.g. semicore electrons in PSP, different parameters...etc.).

The problem of the DFT band structure with the spin-orbit effect was solved, but as shown in the figure, the GW band structure with the spin-orbit still has a problem in valence band and its shape is completely different from the DFT band structure. The plot of Eqp vs Eks shows that the correction of GW on the energies of DFT in this region is nonlinear while without spin-orbit this correction is linear. How can I solve this problem?

I did the GW calculation with both versions of yambo-4.5.1 and yambo-5.0.3 but the result was the same.
The scf,nscf and gw input files with r-setup are attached.

Best regards,
Mitra Helmi

[Daniele Varsano]

Dear Mitra,

as a first check I wouldcheck the convergence of your calculation, in particolar the number of bands in Xp (and probably in GbndRnge) seems a bit low. Note that in presence of soc (spinors) the bumber of bands is doubled wrt the case without SOC.

Best,
Daniele

[8813204602]

Dear Daniele,

Thanks a lot for your support.

I did the convergence tests before the GW calculations without SOC and with SOC. In without SOC mode, because the result of the GW calculations in the output is reported separately for both the up and down spins, I performed the convergence tests separately for both of these spin channels. For up spin, as can be seen in the attached figures, convergence occurs in more bands than down spin.
However, in the case of with SOC, because the result of GW calculations in the output for spin channels is not reported separately, I examined the general gap convergence. Interestingly, in this case, convergence occurred in a smaller number of bands, which was almost the same as the number of down spin bands without spin-orbit. What is the reason for this? Did I do this convergence correctly according to the attached figures?

According to the diagrams, I considered the converged BndsRnXP and GbndRnge parameters without spin-orbit equal to 80 and 120 and with spin-orbit equal to 60 and 80.

The diagrams obtained from these convergence tests are attached.

Best regards,
Mitra

[Daniele Varsano]

Dear Mitra,

the convergence with respect to the dielectric matrix seems ok, while the ones wrt the number of bands in presence of SOC are rather strange.
I do not understand how you could calculate the qp energies considering 20 bands as you have 32 occupied bands.

In any case, if you think that the converge are correct you can try to perform a better interpolation of the GW bands using wannier90 which is surely more precise than the one done by ypp.
Here an example on how to use wannier90 using the QP energies calculated by yambo:
viewtopic.php?f=9&t=1661&p=11005&hilit=wannier#p11005

Best,
Daniele

[8813204602]

Dear Daniele,

the convergence with respect to the dielectric matrix seems ok, while the ones wrt the number of bands in presence of SOC are rather strange.
I do not understand how you could calculate the qp energies considering 20 bands as you have 32 occupied bands.

You were right and I was done the convergence wrong. By repeating the calculation, convergence for GbndRnge =160 was obtained but the qp energies obtained from the GW calculation with this new convergent value are still wrong and there is the same problem as before.
As shown in the figure, quasi-particle energies in valence band do not behave in a linear and regular manner in the presence of SOC, as in the case without SOC (this problem does not exist in the conduction band). Because the effect of the SOC is weak in my material, the DFT band structure with and without the SOC is very similar, so it is expected that there is not much difference between the quasi-particle energies with and without the SOC.
I would be very grateful if you could help me in solve this problem. Do I have to consider a specific parameter, for example, for symmetries or anything else when calculating with SOC?

[Daniele Varsano]

Dear Mitra,
can you post the input/output/report files for the two cases?

Best,
Daniele

[8813204602]

Dear Daniele,

Files are attached.
It should be noted that the attached files related to GW calculations include the SOC related to the calculation with GbndRnge= 80. The result of the calculation with convergent GbndRnge= 160 was similar and there was still the mentioned problem.

Thanks and Regards,
Mitra

[8813204602]

This is also r-setup without SOC: