Problems running VASP: crashes, internal errors, "wrong" results.
Moderators: Moderator, Global Moderator
-
sc_han
- Newbie

- Posts: 6
- Joined: Mon Dec 12, 2022 1:22 pm
#1
Post
by sc_han » Wed Nov 18, 2026 12:07 pm
Dear VASP developers,
I observe that enabling LVHAR changes the LVEL band velocities and intraband plasma-frequency-squared tensor by approximately 6.5%, while the band energies remain essentially unchanged.
Independent finite differences of band energies agree with the LVHAR = .FALSE. velocities but not with LVHAR = .TRUE.
Since LVHAR is intended to control LOCPOT output, I would not expect it to affect optical matrix elements.
Could you clarify whether this is a known issue, whether LVHAR = .FALSE. is a reliable workaround, and whether it has been fixed in later releases?
Thank you for your clarification.
-
manuel_engel1
- Global Moderator

- Posts: 275
- Joined: Mon May 08, 2023 4:08 pm
#2
Post
by manuel_engel1 » Thu Nov 19, 2026 1:29 pm
Hello,
Thank you for the report and for the careful finite-difference check. You are right, this is a bug.
To write the Hartree-only LOCPOT, LVHAR temporarily switches off exchange-correlation, but VASP does not switch it back on afterwards. The optical routines (LOPTICS/LVEL) run after LOCPOT is written, so they recompute the potential and the PAW one-centre terms without exchange-correlation. This changes the velocity matrix elements, the plasma-frequency tensor and the dielectric function. The eigenvalues are unaffected because they come from the self-consistent cycle, which is why only the velocities disagree with your finite differences.
To your questions:
This was not known before your report. Optical results from runs with LVHAR = .TRUE. should be recomputed.
Yes, LVHAR = .FALSE. is a reliable workaround. If you need the electrostatic potential as well, write it in a separate run without LOPTICS/LVEL.
The bug is present in all releases up to and including 6.6.1. It will be fixed in the next release.
Manuel Engel
VASP developer