Difference in Fermi level convention in scf calculations between VASP 6.x and 5.x?

[sophie_weber]

Hello,

I have only just started using VASP 6.3 instead of VASP 5.4, and I wanted to sanity check something to make sure nothing problematic is going on. If I do identical calculations using FeF2 (a wide-gap antiferromagnetic insulator) for a single-point scf calculation, and then a non-self consistent band structure, the band structures for the two VASP versions are identical and the Fermi levels are almost equivalent; from the plots it seems clear that the convention for insulators is to put E-fermi at the top of the valence band.

However, a big discrepancy in the self-consistent fermi level occurs between the two versions. For VASP 5.4, the scf and non-scf Fermi energies are almost identical to order meV (a slight difference would be expected due to not using a complete k-mesh for the band structure plot). However, in VASP 6.3, the fermi level in the self-consistent calculation is more than 2 eV larger than in the non-self-consistent calculation which reads in the scf CHGCAR. The value in the scf calculation looks to be halfway between the valence band and conduction band; so the only explanation I can think of is that in going from version 5 to 6, the convention for choosing the Fermi level in insulators was changed for scf calculations, but NOT for non-scf band structure calculations, which seems extremely odd and confusing.

Could one of the administrators comment on this? If this is not the case the the convention for Ef was changed for scf and not non-scf calculations, there seems to be a big inconsistency in VASP6. Thank you!

[martin.schlipf]

This question comes up occasionally see here, here, or here. You can read those answers for a more complete story, but a short summary is that in 6.1 a bug in the tetrahedron method was fixed. However that fix changed the way the Fermi energy is computed so that it is now more likely to find values other that the VB top. In 6.4, we will revert the tetrahedron method to the old behavior and introduce a new INCAR tag (EFERMI = MIDGAP) that produces a deterministic Fermi energy in the middle of the gap.

[sophie_weber]

I see; thank you so much for clarifying! I'm glad to hear I'm not making some severe mistake