Numerical error in Fermi level - whether system should be classified as insulator or metal

[alpinnovianus]

Hi all,

I have a question about the methodology for deciding whether my system would a metal or insulator. The material doesn't have experimental data, so there is no real-life comparison unfortunately.
My system has a gap between the valence and conduction bands (or I should say between the two bands closest to the estimated Fermi level value on either side).

I read from the numerous posts previously in this forum that the Fermi level for gapped systems can be placed by VASP anywhere from the bottom of conduction band to the top of valence bands.
What I am interested in, is the case where the Fermi level at the end of OUTCAR is determined to be close to the top of VB/bottom of CB but differ very closely to the Kohn-Sham energies list in the OUTCAR file.

For example, at the end of OUTCAR file the fermi energy is printed as:

BZINTS: Fermi energy: 3.103416; 200.000000 electrons
Band energy:-0.123909E+04; BLOECHL correction: -0.000000

while the K-S energies around this value has occupancy very close to 1 but not exactly 1.

k-point 209 : 0.5000 0.0000 0.3750
band No. band energies occupation
....
98 1.9169 1.00000
99 3.1034 0.99988
100 3.1034 0.99978
101 4.4895 0.00000
102 4.4895 0.00000
....

Similarly, there are also other cases where the Fermi energy is placed near the bottom of conduction band but has occupancy very close but not equal to 0: e. g. 0.00015.
In both cases, there is a clear gap between the VB and CB.

I know that ideally insulators' occupancy should be a step function going abruptly from 1 to 0 across fermi level, but because even for insulating systems with confirmed experimental data sometimes VASP also gives this minute deviation from zero/unity occupation, I think my material is an insulator and this small deviation could be numerical error-related.

Since I do not know exactly whether my material is metal or insulator in reality (no experiments yet), are there things that I can do more to further convince myself that my system is an insulator or a metal?
I tried using denser k-meshes, small sigma (<= 0.01) but this doesn't always eliminate these occurrences.

May I know if you can share a more systematic approach to deal with determining whether an unknown material is insulator/metal?

thank you.

[merzuk.kaltak]

Hello,

from your post is is not possible to judge if you are using ISMEAR=-4 or ISMEAR=-5.
Please upload an error report in form of POSCAR, INCAR, KPOINTS, POTCAR and OUTCAR.

Also note, VASP ignores SIGMA for ISMEAR=-4 and ISMEAR=-5, (see here).

[alpinnovianus]

I use ISMEAR = -5 for this example.

I am more interested in how experienced users interpret such minute deviations from unity/zero occupancies.
How small of a deviation could be safely considered as numerical error? is there any rule of thumb ?

roughly speaking, i know that if my partial occupancy is 0.900 instead of 0.999 then I should consider my system as a metal with partially filled band, but in the more refined cases of 0.998 or 0.999 or in this case even 0.99978, what should be the reference line?

[merzuk.kaltak]

The reference line for the occupancies depends on convergence parameter.
For instance if EDIFF is reasonable small, say smaller 1E-5 you can expect the convergence of the occupancies is of roughly the same order.