[Pw_forum] LDA+U calculations to get U and J

[Dmitry Korotin]

Dear Gianluca,

> i need U and J for a DMFT (homemade) code.
> this is the reason i would prefer to calculate U and J via Wannier
> formalism.
> Wannier functions i have calculated including d and p states/bands in order
> to have more localized states and i can have eigenvalues for self-consistent
> unperturbed (alpha=0) and perturbed (alpha finite) runs.
> i don`t understand if U and J calculated via derivates of eigenvalues
> respect occupancies of this two runs are correct.
> what is your opinion?

To calculate U via Wannier formalism one should perturb Wannier states
(please take a look on Eq. 26 of http://xxx.lanl.gov/pdf/0801.3500).
Usage of Hubbard_alpha perturbes d-states of correlated atom instead.
Moreover one needs to subtract the amount of "contrained potential"
from Wannier-function energy to compensate the influence of
perturbation.

The calculation described in http://xxx.lanl.gov/pdf/0801.3500 (The
European Physical Journal B 65 1 (2008) 91-98) was performed with
modified QE code and I am ready to provide it to you (please, email me
directly in that case - there is no any manual, so you will have to
ask me questions)

Actually, if WF were constructed for p- and d-states there is only
slightly difference between d-Wannier and d-atomic states - so you can
use traditional QE procedure for U and J calculation.

> i have the feeling that i should take care to extrapolate for different
> sizes of my system in what i`m doing.

It depends on compound under consideration, but in general - yes, you should.

Best regards,
Dmitry Korotin

Ph. D. Student,
Institute of Metal Physics
S. Kovalevskaya, 18
620041 Ekaterinburg GSP-170
Russia

email: dmitry at korotin.name