My Community

Dear VASP team,
I am going to compute the absorption spectrum of a graphene-like system. To do so, I started with the GW/BSE approach for an ideal graphene. The PBE starting point provides well-known features, and the PBE-G0W0 level gives qualitatively similar results with wider gaps at the Gamma point, e.g. For your reference, I ran this job using the following INCAR and with/without WAVEDER from PBE_unocc calculations.

System = Gr_GW_PBE
ISMEAR = 0 ; SIGMA = 0.05
ISPIN = 2
IVDW = 11
NBANDS = 672
NELM = 1
ALGO = GW0
NOMEGA = 100
LASPH = .TRUE.
LSPECTRAL = .TRUE.
OMEGAMIN= 0.0100

So far everything looks satisfying. However, solving BSE for these GW results leads to zero strength in the optical transitions given in vasprun.xml for both jobs with/without the WAVEDER file. My INCAR file is as follows and the job report says all gone successfully including the read of WFULL*.tmp files, BSE diagonalizing, and reading the WAVEDER when it is available.

System = Gr_BSE_GW_PBE
ISTART = 1
ISMEAR = 0 ; SIGMA = 0.05
ISPIN = 2
IVDW = 11
NBANDS = 672
NBANDSO = 2; NBANDSV = 12
ALGO = BSE
OMEGAMAX = 60
###CSHIFT = 0.1

What have I misunderstood/missed in this last step?
I use vasp 5.4.4, and since the actual system that I want to study is spin-polarized, I run this test calculation with ISPIN=2 and a small smearing.

Regards,
Masoud

Hi Masoud,

Could you provide us with your INCAR, POSCAR, POTCAR, and KPOINTS input files, and OUTCAR and stdout output files, please?

Hello,
I attached those files to this reply. Thanks in advance and regards,
Masoud

Thank you, we're just trying to repeat your results and will get back to you then.

Hi Masoud,

Sorry for the slow response, it took a while to get the calculations running. I could obtain non-zero strength for BSE optical transitions, so as far as I can see, there doesn't appear to be anything wrong with your input files. However, I did significantly reduce the complexity of the calculation. Although you have the computational resources to do large calculations, I would reduce the settings to something cheaper and try to obtain the absorption spectrum, to work out if any setting could be improved. Unfortunately, GW calculations are not straightforward and have many variables that can be changed. I'd recommend reducing the settings and testing each parameter to work out what the most important ones are for your system.

E.g. reducing k-points from 12x12x1 to 6x6x1. This would have the advantage of reducing NBANDS to a much smaller number, which should help with testing suitable settings. Additionally, I would reduce the NBANDS used in the GW calculation while you are struggling to produce an absorption plot. You could try reducing NOMEGA but this doesn't seem to make so large a difference on computational cost. It would also simplify the calculation to do non-spin-polarised calculations, to begin with.

Incidentally, you mentioned that you are using PBE, but in your script it is PBEsol. Is this intentional? It shouldn't make a difference but just for certainty's sake.

I would also consider reducing the vacuum of the POSCAR from ~20 A to ~10 A. By using a large volume, you increase the number of plane waves and this becomes significant for expensive methods like GW.

Have you followed the tutorial for optical absorption? This might be a good place to start.

Best,

Chris

Dear Masoud,

while Christopher is right, I write below the reason for changing the k-point grid in more detail.
I was able to reproduce this with version 6.4.3 and found the following. I guess the same applies to version 5.4.4 as well, but I haven't checked that.

The zeros in the output are due to the fact that the optical transition matrix elements written in WAVEDER are zero (see BSE equations page).
These zeros are produced by the fact that your system contains partial occupancies and that you use LPEAD. As explained on the wiki page, this tag should not be used for metals.
After removing LPEAD from your INCARs, the optical transition matrix elements (written to WAVEDER) are non-vanishing.

Christopher's answer is correct for following reason.
LPEAD allows also to update the WAVEDER after an GW calculation.
By reducing (or changing) the k-point sampling, such that your hamiltonian does not include partial occupancies the system becomes effectively an insulator and LPEAD can be used for the calculation of WAVEDER (for the DFT step, as well as the GW one).
However, if the changes in the optical transition matrix elements from DFT to GW do not change that much, I suggest you remove the LPEAD tag from your calculations and work with the WAVEDER of your DFT step.