I'm experiencing something strange when calculating a band structure with the HSE functional.
The steps I follow are :
i) run self-consistent calculation at PBE level (EDIFF=1E-7), the resulting WAVECAR is used in the next step
ii) run self-consistent calculation at HSE level after adding the following lines to the INCAR:
The K-point file is the same as i), and looks like:LHFCALC = .TRUE. ; HFSCREEN = 0.2
ALGO = Damped ; TIME = 0.4
PRECFOCK = Fast
The resulting WAVECAR is used in the next step4x4x4
0
Monkhorst Pack
4 4 4
0 0 0
iii) run self-consistent calculation at HSE level after the bands structure k-point path as been APPENDED
to the previous KPOINT file, assigning to these k-points zero weight. Of course, self-consistency is immediately reached
because the starting WAVECAR already contains the converged wave functions, as calculated in ii)
The point is that this procedure FAILS, because the band structure is meaningless.
The way in which I manage to obtain an acceptable band structure is to modify step ii) adding NKRED=2 to INCAR, whereas this flag
is remove in iii).
This second procedure should give just the SAME result as the first one, because in this case the wave functions in ii) are more approximate,
but iii) does not immediately converge, and several scf steps are required to converge to the ground state. What is really surprising is that
the two procedures give the SAME ground state energy and the eigenvalues, calculated at the k-points of the 4x4x4 MP grid are the same,
whereas the additional eigenvalues in step iii) relative to the k-points in the band structure path are completely different.
So, I would conclude that the two procedures lead to exactly the same ground state, but for some reason the first one fails at
diagonalizing at the additional k-points (not involved in the ground-state charge density).
I would really appreciated any hint to understand what is happening.
Thanks,
Giovanni