Hello,
As explained in wiki/index.php/BSE_calculations, for BSE calculations with ANTIRES=2, it is recommended to set LORBITALREAL=TRUE, I'm wondering what's the physical origin for that? I compared finite momentum BSE calculations with and without LORBITALREAL, and found that for LORBITALREAL=FALSE the exciton dispersion is more continuous or smooth from Gamma to the second Q point. However, for LORBITALREAL=TRUE, there is a slight discontinuity. So which should I trust?
Best,
Xiaoming
BSE beyond TDA
Moderators: Global Moderator, Moderator
-
kresse_georg
- Global Moderator
- Posts: 10
- Joined: Tue Nov 05, 2019 2:49 pm
Re: BSE beyond TDA
Dear Xiaoming,
The orbitals phi(r) can be chosen to be real valued at the Gamma point, as well as at the Brillouin zone boundary. The VASP code can enforce this by
) performing an FFT of the orbitals to real space, taking the real part, and transforming the result back to reciprocal space;
) re-orthogonalizing the orbitals afterwards.
Usually the procedure is robust. However, if many degenerate orbitals exist, taking the real part can lead
to linear dependent orbitals. To be entirely sure that this problem does not occur, set LREALORBITAL = .TRUE.,
in the mean field calculation and rerun VASP reading the WAVECAR file and setting ALGO = SUBROT. This should give exactly the same energy as in the preceding DFT/HF calculation (make sure to select the correct exchange correlation functional).
For ANTIRES=2, one or two routines assume that the orbitals are chosen to be real (in real space). So LREALORBITAL = .TRUE. should yield the correct answer. However, we have never observed any noticeable difference, in practice, when selecting or omitting LREALORBITAL = .TRUE. I would tentatively rather trust the LREALORBITAL = .TRUE. calculations.
You might be aware that very few calculations have been published using the BSE at finite wave vectors. Although, we have tested the code in quite some detail (by comparing to the dielectric function obtained at the RPA level in the GW routines), we can not guarantee that the BSE code is entirely bug free. We advise to perform tests at the RPA level (LADDER = .FALSE.). Also you can attempt to calculate the dispersion along different symmetry equivalent lines, and see whether the results are robust.
Sincerely,
Georg Kresse
The orbitals phi(r) can be chosen to be real valued at the Gamma point, as well as at the Brillouin zone boundary. The VASP code can enforce this by
) performing an FFT of the orbitals to real space, taking the real part, and transforming the result back to reciprocal space;
) re-orthogonalizing the orbitals afterwards.
Usually the procedure is robust. However, if many degenerate orbitals exist, taking the real part can lead
to linear dependent orbitals. To be entirely sure that this problem does not occur, set LREALORBITAL = .TRUE.,
in the mean field calculation and rerun VASP reading the WAVECAR file and setting ALGO = SUBROT. This should give exactly the same energy as in the preceding DFT/HF calculation (make sure to select the correct exchange correlation functional).
For ANTIRES=2, one or two routines assume that the orbitals are chosen to be real (in real space). So LREALORBITAL = .TRUE. should yield the correct answer. However, we have never observed any noticeable difference, in practice, when selecting or omitting LREALORBITAL = .TRUE. I would tentatively rather trust the LREALORBITAL = .TRUE. calculations.
You might be aware that very few calculations have been published using the BSE at finite wave vectors. Although, we have tested the code in quite some detail (by comparing to the dielectric function obtained at the RPA level in the GW routines), we can not guarantee that the BSE code is entirely bug free. We advise to perform tests at the RPA level (LADDER = .FALSE.). Also you can attempt to calculate the dispersion along different symmetry equivalent lines, and see whether the results are robust.
Sincerely,
Georg Kresse