exciton fine splitting from BSE

[pooja_basera]

Dear all,
In the BSE calculation, How to determine the 1s 2s excitonic peaks in the very low energy range (let say in the range < 2 eV) from the vasprun.xml file?
I am interested mostly in the very low energy range i.e exciton fine splitting? Is it possible to capture the exciton fine splitting from the vasp, where we can determine the 1s 2s peaks from the vasp?

Thankyou in advance

[henrique_miranda]

Two possibilities:
1. You perform the BSE calculation with a small broadening in CSHIFT in the INCAR file:
https://www.vasp.at/wiki/index.php/CSHIFT
You can change the energy range in which you plot the dielectric function with:
https://www.vasp.at/wiki/index.php/OMEGAMAX

2. Look at the 'opticaltransitions' entry in the vasprun.xml file as indicated here:
https://www.vasp.at/wiki/index.php/Plot ... ture_of_Si
The first entry is the energy of the exciton and the second is the intensity.

[pooja_basera]

Thankyou for your reply.
Could we get zoomed section of excitonic peak below electronic bandgap. May be it requires high K-grid sampling. I am interested in very low energy range (between GW bandgap and the BSE optical gap), let say in meV for bulk perovskites, to observe 1s, 2s excitonic peaks. Is it possible from vasp code to capture exciton fine splitting?

[henrique_miranda]

Yes, you can change the energy range in which you plot the exciton spectrum, using https://www.vasp.at/wiki/index.php/OMEGAMAX.
What exactly do you mean by exciton fine splitting?
If you are referring to small energy differences between the excitons at the BSE level of theory then yes.
You can get an explicit list of excitons and their energy in the 'opticaltransitions' entry in the vasprum.xml file

[pooja_basera]

From exciton fine splitting I mean 1s 2s peaks basically the Rydberg series. Here I understood that 1st column corresponds to exciton energy and 2nd column corresponds to intensity.

Code: Select all

<varray name="opticaltransitions" >
  <v>       1.52089262      52.31788087 </v>
  <v>       1.53102469    5606.77917615 </v>
  <v>       1.53254843    8301.89702608 </v>
  <v>       1.53257179    8306.79999559 </v>
  <v>       2.42469311     208.34444100 </v>
  <v>       2.43993235    5363.08934234 </v>
  <v>       2.45237494   12157.88950776 </v>
  <v>       2.45279074   14474.74709241 </v>
  <v>       2.50953937     574.11894131 </v>

In the above example, the electronic bandgap is 1.94 eV, so I want optical transitions below 1.94 eV. Here we can see that in 5 th row, I observe sudden transition from 1.53 to 2.42 in energy eigenvalues. It does not contain an electronic bandgap point i.e. 1.94 eV value. However, I want that after 1.53 eV, It should also have points like 1.54, 1.55, 1.56.....1.94 eV, and so on. So that I will have sufficient point (optical peaks) below the electronic bandgap.

[henrique_miranda]

Ok, I understand what you mean.
Probably you need to increase your k-point sampling but it is difficult to say without seeing the input files.
Try increasing your k-point sampling and check if you get more exciton energies below the gap.

[pooja_basera]

Is it possible to do BSE calculations with high K-grid sampling, if previous steps i.e. GW is performed with less k-grid (let say 4x4x4). In my case, I have performed GW with 4x4x4 K-points, now I want BSE@GW with 12x12x12 K-points. For this purpose again I have performed Algo=NONE with 12x12x12 K-grid, so that there is no issue of WAVEDER compatibility. In BSE calculations, it reads WAVECAR as well as WAVEDER both successfully, but the calculations stops at reading WFULL0001.tmp and not even calculating the oscillator strength. How to resolve this issues, so that I can include high k-grid sampling for the BSE calculations?

[henrique_miranda]

The computation of the BSE matrix elements requires the knowledge of the microscopic dielectric function at q-points commensurate with the k-mesh.
Furthermore, in case you are doing BSE@GW you also need to obtain the QP energies for the same k-mesh that you will use in the BSE calculation.
While in principle these quantities could be interpolated from coarse to denser grids this feature is not implemented currently.

Furthermore, an accurate description of both the QP energies and microscopic dielectric function often requires dense k-point grids.
You should perform convergence tests at the different stages of your calculation: GW and BSE.
You might find that you need more than 4x4x4 k-points to get converged results.

[pooja_basera]

Is it possible in vasp to have 1000 of K-point near the band extrema (valence band maximum and conduction band minimum) (let say at Gamma)? I mean to sample the Brillouin zone only along the band extrema symmetry point with 1000 of K-points? How does the KPOINTS input file look in that case? Or I need to use any specific parameters in the INCAR file.

[henrique_miranda]

You can reduce the size of your BSE matrix by using NBANDSO/NBANDSV and/or OMEGAMAX.
If you accept some degree of approximation in your calculation you can use the model dielectric function to avoid computing the macroscopic dielectric function and a scissor shift to avoid the GW calculation.
You should nonetheless cross-validate your results with actual calculations.

I recommend you to have a look (and maybe do) the BSE tutorials:
https://www.vasp.at/wiki/index.php/BSE_-_Tutorial

[pooja_basera]

Is it possible to get these type of excitonic feature 1s 2s peaks as shown in this paper in Figure2 https://doi.org/10.1103/PhysRevLett.81.2312

To observe this type of peaks it definitely requires very high K-grid sampling along the Gamma centred, as you have mentioned there is no interpolation scheme to go from coarser to denser Kgrid in BSE calculations, so I think it wouldn't be possible with vasp. Please correct me If I am wrong.

However in this paper PHYSICAL REVIEW B 99, 035443 (2019), they have performed the vasp calculations for MoS2 and they are able to observe 1s 2s excitonic features in Figure2. Therefore, I am totally confused. For this calculations, shall I move to other codes or try to do same with vasp?

[henrique_miranda]

Yes, you can perform such calculations and observe the 1s and 2s splitting using VASP.
The interpolation scheme is not crucial. While it might help you to reduce the computational cost of the calculation it also adds complexity to the calculation so it would be an advanced feature.
Please follow the advice I suggested in my previous post of doing the BSE tutorials.
There you can learn how to use a model dielectric function and a scissor operator to avoid the most expensive parts of the computation.