output of GW calculation returns much higher excitation energies

[awang95]

Hi,

I've been trying to run a GW/BSE calculation from HSE06 ground state results, but I noticed that for some reason the GW calculation is outputting a band gap that is much larger than that predicted by the ground state HSE calculation, leading to excitation energies that are much higher than expected. This is observed when comparing the EIGENVAL files from the ground state calculation (whether it's the initial static calculation or the one with exact diagonalization/generating virtual orbitals) and the EIGENVAL file from the GW calculation - I'm looking at bands 248/249 for spin channel 1 and bands 229/230 for spin channel 2. I'm using only the gamma point and G0W0 since I run into memory issues when going beyond that.

Could I please know why there is this large difference in excitation energies, and how to best address it? Or are my INCAR files set up incorrectly (I have, however, tried my best to follow the tutorial on the wiki)?

My files are attached

Thanks

[merzuk.kaltak]

Dear awang,

you are studying a large system

Code: Select all

NiGa-CIF-20220522\(3\1\1)    
   1.00000000000000    
    10.1139001846000003    0.0000000000000000    0.0000000000000000
    -3.3713306695999998   11.1814057590000004    0.0000000000000000
     0.0000000000000000    0.0000000000000000   21.9368000031000001
   Ga   Ni   O    H   
    24    12    47     3   

Inspecting the OUTCAR_dftvo I find

Code: Select all

...
maximum number of plane-waves:     62965
...

which tells me that the maximum number of bands possible is around 62000.
In our tutorials, we usually recommend to set NBANDS to the maximum.

However, you are selecting only 800 bands in your GW calculation

Code: Select all

...
NBANDS = 800
...

Thus, I suspect that your calculation is not converged with respect to NBANDS.

I recommend following strategy:

* update vasp to 6.4.1
* switch to ALGO = EVG0W0R with NOMEGA = 12 or even 16
* increase NBANDS

A final remark to NBANDS is in order here. Typically the GW band gap that vasp prints into the OUTCAR starts making sense once the highest energy in the preceding DFT run is in the regime of the maximum frequency point.
In your case the latter around 200, while the energy of band 800 of your HSE calculation is around 21.03 eV.
This relation holds true also for ALGO=EVG0W0R.

N.B: 800 bands is enough for HSE, since you require only the occupied states for the ground state in this theory.

[awang95]

ok i'll try this; thanks so much!