[Pw_forum] very large value 'U' obtained for 'GaAs' by linear response method

Giuseppe Mattioli giuseppe.mattioli at ism.cnr.it
Mon Oct 8 11:11:56 CEST 2012


Dear all

Burak has indeed described well the problem, but something more can be  
said: the Ga 3d shell (as well as the Zn 3d shell in ZnO) is almost  
atomic-like and give rise to a very narrow band in III-V  
semiconductors, horribly affected by delocalisation (aka  
self-interaction) problems. This band is raised in energy far more  
than the As (3p) valence band; so the two bands are mixed (and they  
should not) and the valence band maximum is pushed up towards the  
conduction band minimum. Moreover, you cannot extract meaningful  
self-consistent U values by linear response procedures in the limit of  
a simplified "on-site, one-shell" approach. However, it is not  
forbidden to use "semi-empirical" U values, as long as you have a good  
knowledge of the physics of your system... The above problems in DFT  
calculations of GaAs and ZnO are addressed in a Zunger paper, very  
long and quite... difficult to flick through.

Lany, S. and Zunger, A.; Phys. Rev. B 78, 235104 ͑(2008͒).

In a short, you may try to decouple the Ga (3d) band from the As (3p)  
band by hammering with some empirical U value the head of the Ga 3d  
band. And you can recover a very good band gap value by pushing down  
(more softly...) the As 3p band. You may test your results against a  
hybrid functional. I've done it in the case of ZnO, and I've obtained  
quite good results.

Matteo suggest that the Ga 3d has to be included in the calculation.  
In my experience (but I'm the "empirical" kind of computational  
guy...) The Ga 3d shell is mandatory in the case of GaN if you want a  
good lattice parameter. In the case of GaAs you can obtain good  
results (band gap included) by embedding the Ga 3d electrons within  
the Ga pseudopotential, and by using a NLCC approach.
This leads to some kind of decoupling of the Ga 3d and As 3p bands. I  
remember some old papers (2000-2005) by Van de Walle and Neugebauer on  
GaN in which a Ga PP without 3d shell was used after a lattice  
parameter optimisation performed by using a Ga PP with 3d shell.

HTH

Giuseppe

Quoting Burak Himmetoglu <himm0013 at umn.edu>:

> Dear Brahma,
>
> Ga has a full d-shell. Therefore, if you try to perturb the d-shell of Ga,
> you can reduce the number of electrons localized on d-states by a positive
> Hubbard_alpha but there is no way to increase their occupations with a
> negative Hubbard_alpha. Therefore, the response matrices will be non-linear
> and you will get very small numbers in them. When you invert these
> matrices, very large numbers will appear.
>
> Calculating U for full d-shells is not easy, and there is no simple
> solution (Another known example is ZnO). My personal opinion is that the
> use of LDA+U is questionable for filled d-shells. In fact if you check the
> paper mentioned by Matteo, the gap for GaAs is corrected by applying U and
> V (inter-site interactions) on Ga-p states and states which overlap with
> Ga-p. No correction on Ga-d is applied.
>
> Best regards,
>
> Burak
>
> On Sun, Oct 7, 2012 at 12:02 PM, Bramha Pandey   
> <pandey.bramha at gmail.com>wrote:
>
>> sorry for wrong matrix copied in previous mail.
>>
>> CHI0^-1 - CHI^-1 Matrix
>>
>>     0.0000    0.0000    0.0000
>>     0.0000    0.0000    0.0000
>>
>>     0.0000    0.0000    0.0000
>>   type:            1  U0 =   0.000000000000000E+000
>>
>> On Sun, Oct 7, 2012 at 10:30 PM, Bramha Pandey   
>> <pandey.bramha at gmail.com>wrote:
>>
>>> Dear Prof. Matteo,
>>> i am using Ga, pseudo with 3d as a valence, Hubbard_l=2 is set for 'Ga'.
>>> If it should change how much value i have to set it?
>>>
>>> Dear sir i am totally struck by zero matrix element in ch0^-1 - chi^-1,
>>> so always it gifted me ZERO value where as ch0 and chi matrix are NON ZERO
>>> with same element value.
>>> CHI_0 Matrix
>>>    -0.0002  -0.0002   0.0005
>>>   -0.0002  -0.0002   0.0005
>>>    0.0005   0.0005  -0.0010
>>>    CHI Matrix
>>>    -0.0002  -0.0002   0.0005
>>>   -0.0002  -0.0002   0.0005
>>>    0.0005   0.0005  -0.0010
>>>
>>>   CHI0^-1 - CHI^-1 Matrix
>>>
>>>     0.0000    0.0000
>>>     0.0000    0.0000
>>>   type:            1  U0 =   0.000000000000000E+000
>>> This above is for all the  2 atoms supercell combination when we take the
>>> ZB structure with 2 atoms: (1) when both atoms are 'Ga'
>>> (2) when both atoms are 'Ga1' and 'Ga' (3) when both atoms are different
>>> 'Ga' and 'As'.
>>>
>>> All above three conditions are tested under one atom(Ga1 or Ga) is
>>> perturbed.
>>> Please Sir give me some ray of hope so that i can get rid of this
>>> situation.
>>>
>>>
>>> On Sun, Oct 7, 2012 at 8:10 PM, Matteo Cococcioni <matteo at umn.edu> wrote:
>>>
>>>> Dear Bramha,
>>>>
>>>> in J. Phys.: Condens. Matter 22 (2010) 055602 we did a similar
>>>> calculation but we got pretty different numbers (see Table 2 of the
>>>> reference). Maybe the states of Ga you applied U or alpha to are
>>>> completely full. what is the value of l set for Ga in set_hubbard_l ?
>>>> did you change that? also, for Ga, you better use a pseudo with 3d
>>>> states in the valence rather than frozen in the core.
>>>>
>>>> Matteo
>>>>
>>>>
>>>>
>>>> On Sun, Oct 7, 2012 at 7:31 AM, Bramha Pandey <pandey.bramha at gmail.com>
>>>> wrote:
>>>> > First of all thank you very much Dear Prof.Stefano.
>>>> > I have change the input as ZB 'GaAs' structure and calculate the 'U'
>>>> value
>>>> > for both 'Ga' and 'As' which is given bellow.
>>>> > But i wondered to see the 'U' value of 'Ga' 1010.79 which is quite
>>>> large so
>>>> > how can i figure out this is  correct or wrong?
>>>> > No of Atom        U_Ga                       U_As
>>>> > 2                1043.35792360332    2.87817590827985
>>>> > 16              1010.81003743078    2.82970777976411
>>>> > 54              1010.79529573193    2.82968584264055
>>>> > 128             1010.79528904780    2.82968583269392
>>>> > 250             1010.79528904476    2.82968583268940
>>>> > Any comment is welcome.
>>>> > --
>>>> > Thanks and Regards
>>>> > Bramha Prasad Pandey
>>>> > Ph.D Student Indian School of Mines(ISM)
>>>> > Dhanbad, INDIA.
>>>> >
>>>> >
>>>> > _______________________________________________
>>>> > Pw_forum mailing list
>>>> > Pw_forum at pwscf.org
>>>> > http://pwscf.org/mailman/listinfo/pw_forum
>>>>
>>>>
>>>>
>>>> --
>>>> Matteo Cococcioni
>>>> Department of Chemical Engineering and Materials Science,
>>>> University of Minnesota
>>>> 421 Washington Av. SE
>>>> Minneapolis, MN 55455
>>>> Tel. +1 612 624 9056    Fax +1 612 626 7246
>>>> _______________________________________________
>>>> Pw_forum mailing list
>>>> Pw_forum at pwscf.org
>>>> http://pwscf.org/mailman/listinfo/pw_forum
>>>>
>>>
>>>
>>>
>>> --
>>> Thanks and Regards
>>> Bramha Prasad Pandey
>>> Ph.D Student Indian School of Mines(ISM)
>>> Dhanbad, INDIA.
>>>
>>>
>>
>>
>> --
>> Thanks and Regards
>> Bramha Prasad Pandey
>> Ph.D Student Indian School of Mines(ISM)
>> Dhanbad, INDIA.
>>
>>
>> _______________________________________________
>> Pw_forum mailing list
>> Pw_forum at pwscf.org
>> http://pwscf.org/mailman/listinfo/pw_forum
>>
>

Giuseppe Mattioli
ISM-CNR
Italy







More information about the users mailing list