still weird band behavior

[giacomo giorgi]

Dear Vaspers,
I ask your opinion about this result. It seems that the more I try to get confident with band calculations the more I fail... excuse me for the harassment

my INCAR (PAW, 10x10x10 k-points according MP, 4 Ga and 4 As atoms, Zincblende structure, a total of 72 electrons, spinpolarized)

SYSTEM=GaAs 8 atoms
ISTART=0
ISMEAR=0
PREC=High
ISIF=3
NSW=300
IBRION=2
EDIFF=1E-5
EDIFFG=-0.01
GGA=91
ISPIN=2
ENCUT=287.594
ENAUG=531.356

once I optimize I get a presumably correct ordered band structure, i.e.,

E-fermi : 3.0947 XC(G=0): -9.4225 alpha+bet : -8.5373
add alpha+bet to get absolut eigen values

spin component 1

k-point 1 : 0.0500 0.0500 0.0500
band No. band energies occupation
1 -12.1568 1.00000
2 -12.1568 1.00000
3 -12.1568 1.00000
4 -12.0653 1.00000
5 -12.0630 1.00000
....
....
....
33 0.2025 1.00000
34 1.9885 1.00000
35 2.6641 0.99884
36 2.6641 0.99884
37 3.4877 0.00273
38 4.2868 0.00000
39 4.2868 0.00000

i.e. a band gap of about 0.81 eV at gamma (direct band gap, as it should be) that at PAW/GGA is good too.

When I perform non-selfconsistent (ICHARG=11) calculation on the optimized geometry, i.e.

INCAR (kpoint increased up to 100)

SYSTEM=GaAs 8 atoms ZB bands
ISTART=0
ISMEAR=0
ICHARG=11
PREC=High
EDIFF=1E-5
EDIFFG=-0.01
GGA=91
ISPIN=2
ENCUT=287.594
ENAUG=531.356

Now the corrected and so plottable band structure becomes:

E-fermi : 2.8892 XC(G=0): -9.4141 alpha+bet : -8.5212
add alpha+bet to get absolut eigen values

spin component 1

k-point 1 : 0.0000 0.0000 0.0000
band No. band energies occupation
1 -12.1704 1.00000
2 -12.1704 1.00000
3 -12.1704 1.00000
4 -12.0769 1.00000
5 -12.0769 1.00000
6 -12.0769 1.00000
7 -12.0351 1.00000
....
....
....
32 0.1666 1.00000
33 0.1666 1.00000
34 2.6873 0.92327
35 2.6873 0.92327
36 2.6873 0.92327
37 2.8589 0.58464
38 4.2323 0.00000
....
....

k-point 2 : 0.0149 0.0000 0.0149
band No. band energies occupation
1 -12.1703 1.00000
2 -12.1703 1.00000
3 -12.1703 1.00000
4 -12.0770 1.00000
5 -12.0768 1.00000
....
....
.....
33 0.1680 1.00000
34 2.5678 0.98847
35 2.6817 0.92880
36 2.6865 0.92410
37 2.9734 0.27564
38 4.2333 0.00000
39 4.2333 0.00000


and the situation seems to get normal on the 3rd kpoint for which is
....
....
34 2.3794 0.99984
35 2.6642 0.94420
36 2.6839 0.92667
37 3.1459 0.03471
38 4.2382 0.00000
39 4.2382 0.00000
40 4.2606 0.00000
....
....

My question is why the 37th band is so "strangely" occupied? Is it smearing? I have 72 electrons then 37th should be the first unoccupied band, shouldn't it? Is it a ghost state? But all the parameters seem so accurate that I would not expect a ghost state.

Many thanks and still excuse me,
Giac

[admin]

please note that the default smearing width (SIGMA) is 0.2eV. Therefore the occuancy of the (uppermost) bands depend on their dispersion at the respective k-point. You will immediately see if you compare the eigenvalue spectra of the uppermost states: δE (37-36) are 0.83 at the R point of the scf run, but only 0.17, 0.28 and 0.46 at the first 3 k-points of the band structure run. correspondingly, the partial occupancies of band decrease with increasing δE

[giacomo giorgi]

Dear admin,
you are illuminating me.thanks!
I reduced SIGMA and now the occupancy is OK! Nevertheless, I found in literature that the same gap at the same level of calculation (PAW/GGA) is ~0.5 eV. Which are the parameters on which should I mainly focus in order to "open" the band? my 0.17 one seems quite poor.....

Still many thanks,
Giac