[Pw_forum] PBE Hybrid funcitonals and LDA pseudopotentials
hannu.komsa at epfl.ch
hannu.komsa at epfl.ch
Fri Jul 22 11:51:46 CEST 2011
Dear Gabriel,
> I have been attempting to use the hybrid functional PBE0 to tune the
> bandgap of InGaAs, while using Virtual Crystal Approximation to
> obtain In(0.57)Ga(0.43)As. I initially tried mixing the PBE pseudos
> of Ga and In, but ran into a number of problems (unequal nqf mesh
> when mixing, and using ultrasoft pseudos with hybrid functionals).
When I tested VCA for InGaAs, it didn't look so good. At least the
lattice constant dependence on Ga/In composition came out pretty bad
as Ga and In are of quite different sizes. I used pseudos taken from
the abinit page and converted to QE. VCA seemed to work fine for
those, also at the PBE level.
> I then tried the Norm-conserving LDA pseudos (which also have no nqf
> information to worry about) for mixing In and Ga (also using LDA for
> As pseudo, obviously). I then found I could tune the bandgap by
> specifying "PBE0" in the &SYSTEM card of the input file by changing
> the proportion of Fock exchange (alpha, or "exx_fraction" in
> funct.f90), while specifying LDA pseudos for each atomic species.
You can mix and mash pseudos and functionals, but if you are not
absolutely sure of what you are doing, it will likely fall in the
category of garbage-in-garbage-out. PW also warns you about this.
> I am surprised this would work because my understanding is that LDA
> contains no exchange term, yet modifying the value of alpha has an
> effect on the bandgap just like if I had specified PBE
> pseudopotentials in the ATOMIC_SPECIES card. I initially set up this
> calculation as a test to just prove to myself that the results
> obtained using LDA pseudopotentials are independant of the amount of
> exchange, yet this is not so.
You might want to read some more literature about DFT and functionals,
as the above description is mostly wrong. All functionals have
exchange term, but with different (level of) approximation.
Regards,
Hannu-Pekka Komsa
EPFL, Switzerland
More information about the users
mailing list