Fermi level with tetrahedral integration [v6.1.0]

[chengcheng_xiao1]

Dear all,

I found a strange behavior of the tetrahedral method (ISMEAR=-5) with VASP 6.1.0 where the Fermi level is located directly on top of the conduction band minimum. Using the same input, VASP 5.4.4 gives a "normal" answer where the Fermi level is directly on top of the valence band maximum:
[DOS With 6.1.0, Fermi energy set to 0]

dos611.png

[DOS With 5.4.4, Fermi energy set to 0]

dos544.png

Please see attached for my input.

I'm not sure if this is a bug at the moment.

Best wishes,
Chengcheng

[martin.schlipf]

Thank you for reporting this. There was a bugfix for the tetrahedral integration in 6.1, perhaps that had unintended side effects. I'll investigate to see how this occurs.

[martin.schlipf]

Okay, it doesn't seem to be a bug, but perhaps the algorithm can still be improved. The Fermi energy is determined with a bisection scheme always replacing the lower or the upper barrier until the Fermi energy is converged to 1meV and the number of electrons is converged to 1e-10. In you particular case, the sum of all weights up to the valence band top ended up being just a tiny bit less than the actual number of electrons, so the Fermi energy increased to the bottom of the conduction band. Due to a slightly different algorithm the previous version converged to the valence band top.
Until we come up with a better solution here is a workaround, set the number of electrons to a tiny bit less than the actual number, i.e. in your case NELECT = 55.99999999. That will converge to the valence band maximum.

[chengcheng_xiao1]

That works!
Thank you, Martin!