My Community

Hi,

i`m calculating band structure but i have some problems.
My system is a graphene sheet on the top of some layers of a metal.
I impose the periodicity of the unit cell of graphene.
This is my POSCAR:
--------------------------------------
Graphene 2.465
1.00
2.13475262032864 -1.2325 0.0
0. 2.465 0.
0. 0. 25.0
2 6
Cartesian
0.00000 0.00000 13.84700
1.42317 0.00000 13.85903
0.71158 1.23250 0.00000
0.00000 0.00000 2.01266
1.42317 0.00000 4.02533
0.71158 1.23250 6.03799
0.00000 0.00000 8.05066
1.42317 0.00000 10.06332
---------------------------------------
To calculate the band structure, as first step, i run a job
with INCAR file:
---------------
NWRITE = 2
PREC = HIGH
ENCUT = 550
ISTART = 0
ICHARG = 2
NELM = 100
EDIFF = 1E-5
ISMEAR = -5
LDIPOL=.TRUE.
IDIPOL = 3
-------------------
and KPOINTS file:
-------------------
24x24x2
0
Gamma
24 24 2
0 0 0
------------------
to produce the charge density file.
From this run i get a fermi level.
Then to have the dipersion along some direction, i do another run with INCARfile:
-------------------
NWRITE = 2
PREC = HIGH
ENCUT = 550
ISTART = 1
ICHARG = 11
NELM = 100
EDIFF = 1E-5
ISMEAR = 0
SIGMA = 0.01
LDIPOL=.TRUE.
IDIPOL = 3
------------------
The fermi level i get from this run is different (the two number are not close...)
from the one of the first run.
This is something i don`t understand.
But if i calculate bands structure for a single sheet of graphene (using absolutely same files of input) i don`t get difference in the two fermi level.
Can some one tell me the reason?
Thank you.

Gianluca

please define 'not close'....
1) for a slab calculation, please choose just 1 k-point along the z-direction of the BZ.
2) the absolute values of Kohn-Sham energies (and hence the Fermi energy) cannot compared in different calculations, because the E=0 level is not defined in periodic supercells. (you could use the LVTOT=True option to obtain the constant energy level of the vauum of your slab as a point of reference) As a rough estimate for the common reference point, you can also check alpha+bet in both OUTCARs.
3) please check the highest occupied eigenvalues at k-points of the scf run which are close to the k-point which defines E_fermi in the bandstructure run.

[ Edited Wed Mar 14 2007, 04:07PM ]

Hi,
I have a problem with the bandstructure calculation. I think the difference may be due to the ISMEAR as I have kept all parameters the same. My system is a semiconductor bulk system.

For the first run, I used ISMEAR=-5 for single point energy.
For the second run bandstructure calculation, I used ISMEAR=0 since tetrahedron method is not possible for this case.

The end result is that the fermi energies are quite different, as large as 1eV for some systems.

If I used ISMEAR=0 for the first run (single point), the fermi energy is closer to the 2nd bandstructure run (difference of around 0.2eV).

Should we use ISMEAR=0 or -5 for optimization/single point, followed by bandstructure calculation? I have sufficient K points for either case. Is the difference due to some other reasons? The accuracy of the fermi energies is quite important to me.

Thanks for your help.



<span class='smallblacktext'>[ Edited Wed Apr 25 2007, 09:44AM ]</span>

the Fermi level is always calculated from the uppermost occupied state, therefore it may change significantly if you change the type of the k-mesh (depending on the band strucuture), especially if you have not made careful k-point set convergence tests before. Furthermore, the smearing width (SIGMA) is taken into account if you choose Gaussian integration,
whereas it is not for the tetrahedron method.