[Pw_forum] LDA+U in PWSCF

[Matteo Cococcioni]

Dear Jiang-long Wang

the full procedure to compute the Hubbard parameters (i.e. to
process the data coming from the linear response study with pwscf) is
explained in my thesis in full detail.
To obtain the necesary data using pwscf do the following:

1) lda (Hubbard_U = Hubbard_alpha = 0) calculation up to scf

2) put Hubbard_alpha =/= 0 on ONE of the d ions (it needs to be
inequivalent to others, so you have to describe it as of different kind
thus breaking some possible symmetries) and, starting from the
wavefunction and the scf potential obtained in 1) (startingwfc = 'file', startingpot =
'file') you make a run up to scf. Remember to lower the threshold for the
diagonalizaion of the Hamiltonian (ethr) to the value used in the last
iteration of the run in 1). Probably you will have to add ethr among the
input variables and make the code read it from the input namelist.
This is because you require high precision also at the first iteration of
the perturbed run.

3) repeat 2) for several values of Hubbard_alpha around 0 (typical
|values| are from 0.01 to 0.1 eV) checking a linear behaviour of the
d state occupations with respect to Hubbard_alpha expecially around 0.

4) repeat 2) and 3) for every non equivalent d-ion in your unit cell.

5) built the matrices \chi and \chi_0 and compute U from them.

6) repeat everything in larger and larger supercells up to convergence of
the U

This procedure guarantees plenty consideration of the screening of the
Hubbard parameter so that U should not be overstimated anymore.

Good luck,

		Matteo



On Mon, 18 Aug 2003 jlwang at theory.issp.ac.cn wrote:

>
> Dear Dr. Matteo,
>
>                  Thank you very much for your thesis !
>
>               Could  you give me a example or explanation to calculate
> Hubbardparameter with PWSCF code.   For example,  the vale of
> Hubbard_alpha, how to calculate the local dos.
>
>       I used to calculate the Hubbard parameter with TB-LMTO method by
> constrained LDA calculation. The U parameter is usually overestimated for
> metal e.g  3d transition metal series, it is hard to take the screening
> effect into account correctly.
>
>       I want do some calculation with the LDA+U utility in PWSCF code.
> --
> Yours sincerely,
>
> Jiang-long Wang
>
> .................................
>
> Institute of Solid State Physics,
> Academia Sinica, P. O. Box 1129,
> 230031 Hefei, Anhui, P. R. China
>
> Tel: 86-551-5591464
> Fax: 86-551-5591434
>
> E-mail: jlwang at theory.issp.ac.cn
>
>
>