BSE oscillator strength from EVGW0 and QPGW0

[michal_otyepka1]

Dear admins,

I am calculating the excitation energies of the acetone molecule and testing BSE with various flavors of GW. No problem with calculation itself, it is well described in the vaspwiki. The issue I encounter is that I am getting quite different oscillator strengths in BSE depending on whether I use EVGW0 or QPGW0. While EVGW0 agrees very well to other calculations including CASPT2 benchmark, switching to QPGW0 gives non-zero values for excitations which are zero in EVGW0. Some of them are even unphysical, for instance QPGW0 yields non-zero osc. strength for the lowest singlet n-pi excitation, which is dipole forbidden (and EVGW0 and CASPT2 give exactly zero).

I strictly tested convergency, use converged cutoff in a large box and use NBANDS=maximum number of plane waves, but the problem persists at all settings I used. The setup is the same for EVGW0 and QPGW0, I literally change a single line in INCAR for GW step. Any idea what might be the problem?

On a note, the excitation energies themselves are much better in QPGW0, so I would like to stick with it.

Thanks for any help,
Petr Lazar

[alexey.tal]

Dear Petr,

It is hard to give a definitive answer without looking at your input files. Could you please attach the files as described in the forum guidelines.

The issue you are describing might be related to the WAVEDER file. When you update the wave functions in the QPGW0 calculation you should also recalculated the derivatives of the wave functions by setting LOPTICS=.TRUE. and LPEAD=.TRUE..

[michal_otyepka1]

Dear Alexey,

thanks for quick reply! I attach all relevant input and output files for BSE/QPGW0/PBE0 calculation. I got unphysical oscillator strengths also using PBE for the initial KS wavefunctions.
I did not use LPEAD=.TRUE. to update WAVEDER - I thought that it was not necessary, since I used just single k-point for the molecule. I can try it though.

Best, Petr

[michal_otyepka1]

job_files.zip

[alexey.tal]

LPEAD=.TRUE. and LOPTICS=.TRUE. should be used in QPGW calculations even for a Gamma-only calculation.
The unphysical oscillator strengths are often a result of incompatible WAVEDER and WAVECAR files, so you can also try to recalculate the derivatives after the QPGW steps as described in the wiki.

Do you find comparable results in your EVGW0 and QPGW0 calculations, i.e., dielectric constants and band gaps? Could you attach OUTCAR from your EVGW0 calculation and the BSE results for this calculation?

[michal_otyepka1]

As far as I can tell, the band gaps from EVGW0 and QPGW0 are comparable. I attached OUTCAR and vasprun from EVGW0 run, maybe you find something I missed. The osc. strength from EVGW0 are pretty spot on for the lowest excitations (CASPT2 gives 0.000 for n-pi and 0.0235 for the second n-3s excitation, see Pasteka et al. Molecular Physics, Vol. 110, 2012, 2219–2237) even though the excitation energies are too low.

evgw0_files.zip

Petr

[michal_otyepka1]

Hi,
the explicit recalculation of WAVEDER using ALGO=Nothing has cured the problem! Now, the oscillator strengths are similar to other calculations and zero for dipole forbidden excitations. Thanks for the help!
Petr