[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: Position of the Ef in LDOS
Dear Feffusers,
This is an update on Sergio Moreno's msg and our reply re
the position of the Fermi level calculated by FEFF8. Similar
comments apply in many other cases.
Our checks of Moreno's feff.inp show that part of the problem
is due to the possibilities b), d) and e) listed below, as the SCF loop
appears to be reasonably converged and cluster size is reasonable.
In particular the basis set for the O atoms in SnO_2
(lscf=1 i.e., s and p only) is too small! This is the original
list in feff.inp:
POTENTIALS
* ipot Z element l_scmt l_fms stoichiometry
0 8 O 1 1 0.001
1 50 Sn 3 3 2
2 8 O 1 1 4
The reason is that due to hybridization the O d bands are partly
occupied and take up about 0.05 electrons. This number is printed
out at the end of the scf loop in log1.dat. Without the d- valance
band charges, the Fermi level must rise to steal those electrons from
the conduction band!
iph il N_el
0 0 1.941
0 1 5.070
0 2 .050
0 3 .000
1 0 1.256
1 1 1.421
1 2 10.642
1 3 .000
2 0 1.894
2 1 4.387
2 2 .062
2 3 .000
The fix for this is to *usually* use the FEFF default basis
(set lscf = -1 in feff.inp)
POTENTIALS
0 8 O -1 -1 0.01
1 50 Sn -1 -1 1
2 8 O -1 -1 1.99
These defaults typically add one to the bare atomic configuration.
That is, lscf for Sn (with d10s2p2) is 3, that for O (s2p4) is 2.
Note also that I adjusted the stoichiometry so the total charge is
neutral - otherwise the Fermi level would again too high. This
is almost always a very minor tweak but possibly can be important for
big band gaps.
I also tried maxing out the lscf adding in f electrons, but it didn't change
the results much since the occupied f counts turn out to be negligible.
POTENTIALS
0 8 O 3 3 0.01
1 50 Sn 3 3 1
2 8 O 3 3 1.99
With these changes the Fermi level drops by a about 1.2 eV from -3.387 to
mu_new= -4.559
This is still a little high - about one eV above the conduction
band edge of -5.5 eV. This +1 eV error is typical of the
accuracy one can expect in FEFF. It's due to limitations of the basis
set, spherical potentials, and other numerics, even though FEFF
builds in overall charge neutrality. The DOS is very flat in the band
gap, so an eV error in E_F amount can come from very small charge
differences.
The large band gap in this material (look at the neutral O pDOS) goes
from about -5.5 to -9.5 so the midgap is about -7.5.
Thus we would recommend a -3 eV shift (Vr = 3.0 eV in the EXCHANGE card)
to set the Fermi level at mid-gap. Although this seems like a big
shift, it's source seems to be the big band gap in these insulators.
The error is typically smaller in metallic materials with non-zero
DOS at the Fermi level. In any case, we do hope that improvements
in the next releases of FEFF will reduce the errors further.
John Rehr
>
> We would have to see a feff.inp to say for sure. But here are some
> possibilities:
>
> a) the SCF loop hasn't completely converged
> b) you don't impose stoichiometry in the potentials, so
> the potentials are not completely neutral - it's important
> to do add stoichiometry
> c) the cluster is not large enough
> d) the basis is too small, so you miss some occupied states
> e) there is enough "leakage" to unoccupied states that
> the calculation is off. It's not unusual to be off by 1-2 eV.
>
> In any case, a simple fix is to move the E_0 with the
> EXCHANGE card to mid gap.
>
> Cheers,
> John
>
> > I would like to mention that I have found the same problem that Maurizio
> > has mentioned. In my case was for MO and MO2 oxides. My additional problem
> > is that my compounds has a band gap but the Fermi level is not falling
> > there (I have found states just on the fermi level as maurizio has
> > mentioned).
> >
> > Thank you for your help.
> >
> > Sergio.