The ABINIT Users Mailing List ( CLOSED )

["Matteo Giantomassi" <Matteo.Giantomassi@uclouvain.be>]

> Dear Matteo,
>
> we (me and Pierre-Yves Prodhomme during his thesis) have done some GW
> calculations using
> some HGH pseudos one year ago, especially for Hf-based oxides.
>
> May be, it's time to clarify the limitations of such an approach when
> using the GW functionalities of abinit ?

 Dear Philippe,

 The main limitation is in the generation of the KSS file. For
 HGH or GTH pseudopotentials, one cannot use
 the direct diagonalization of the Hamiltonian (kssform ==1 or 2)
 since the nonlocal part of the Hamiltonian is not correctly assembled
 in the outkss.F90 routine.

 As a consequence, both HGH and GTH pseudos require the use of the
 conjugate gradient method (kssform==3) in order generate a KSS file.

 For what concerns the GW part (screening and sigma part),
 there's no serious limitation in the implementation
 since the only ingredients needed for GW are the eigenvalues
 and the wave functions stored in the KSS file. The only big constraint is
that
 non-collinear calculations ( i.e nspinor==2 )
 are not yet fully supported. In particular SO coupling is not implemented
at all.
 There's also a small issue related to the inclvkb variable that is
described below.

>
> The first work around is to nullify the spin-orbit coupling terms in the
> hgh file.
> In fact for our calculations, semi-core states have been proven useful,
> even without the SO term,
> for DFT and GW calculations, and as far as I understand, by default the
> SO terms are not really used
> (no spin orbit coupling).
>
> Then, I remember also that we have had to modify some irrelevant tests
> in abinit, and to play
> a little bit with the kssform keyword (probably we have set kssform = 0,
> but now it is obsolete).

 Yes, indeed these are the small changes needed to have HGH pseudos work
in GW I was referring to in my previous mail.
 One should get rid of the checks on the number of projectors done
 in the code and make sure that inclvkb==0 is always used (see below)

>
> I'm not totally confident with that, but anyway, at the end the results
> seemed to be quite correct,
> by providing at least realistic band gaps, compared to other TM pseudos,
> i.e. without semicore states.

 Good news,

>
> Also, I'm quite sure that we didn't use the
>
> inclvkb
>
> variable - in fact it seems to be undocumented, what is its exact meaning
> ?

 It's documented on the web page in the section describing the GW input
variables

 http://www.abinit.org/Infos_v5.6/input_variables/vargw.html#inclvkb

 Anyway I agree with you that it's meaning should be explained in more
detail.

 The correct treatment of the q-->0 limit in the irreducible polarizability
 would require the evaluation of the term: iq \cdot <b1,k| -2\nabla + 
[V_nl, r] |b2, k>

 You might also have a look at appendix B of Phys. Rev. B 35, 5585 - 5601
(1987)
  http://prola.aps.org/abstract/PRB/v35/i11/p5585_1

 The term involving the gradient operator is straightforward in a plane
wave basis set and
 is always calculated.

 The second contribution requires the evaluation of the commutator of the
nonlocal
 part of the Hamiltonian with the position operator.
 This term is quite CPU demanding, moreover its exact expression depends
of the kind of
 pseudopotential you are using.
 At present the evaluation of this term is not coded in case of HGH, GTH
pseudos or
 "standard" pseudos with more than one projector per angular channel.

 The inclusion of the commutator is important if one is interested in
optical properties but
 it plays a marginal role in the GW results if the weight associated to
 the Gamma point is negligible when  compared to the other q-points..
 For systems of low dimensionality and|or anisotropic compounds it's worth
to check whether
 the correct inclusion of the commutator affects the final results.

 Best Regards
 Matteo Giantomassi


>
> Best regards,
>
>  Philippe
>
> Matteo Giantomassi a écrit :
>>> Dear Seungmi,
>>>
>>> I don't exactly if HGH pseudos are working in the GW code.
>>>
>>
>>  HGH pseudos contain additional projectors for the spin-orbit coupling
>> term.
>>  Since GW is not yet able to deal with non-collinear systems
>>  HGH pseudos, at present, can only be used to generate a KSS file in the
>>  collinear case employing kssform ==3
>>
>>  Please note however that, to the best of my knowledge, nobody has ever
>> used HGH to perform
>>  GW calculations. I ran some tests and indeed the code stops complaining
>> about the number of
>>  projectors found in the pseudopotential file.
>>  Making the GW part work with HGH will require some change in the code
>> as
>> well as
>>  additional testing. For the time being I suggest use "standard"
>> norm-conserving pseudos
>>  or the files available in the Goedecker-Teter-Hutter (GTH) table.
>>
>>  I agree with Fabien on the need of adding semicore states during the
>> generation of the
>>  pseudo. Please keep in mind, however, that an accurate
>>  pseudo with semicore states would require more than one projector per
>>  angular state. If you don't find the elements you need in the GTH
>> table,
>>  you might use the opium generator to have more than one projector per a
>> particular l
>>  (FHI doesn't support multiprojector pseudopotentials)
>>
>>  By the way:
>>  As already explained in the forum, in case of pseudos with more than
>> one
>> projector per angular state,
>>  the treatment of the non analytic behavior of the inverse dielectric
>> matrix is not
>>  coded. In this case one has to use inclvkb==0 during the screening run.
>>
>>  Hope it helps,
>>  Best Regards,
>>  Matteo Giantomassi
>>
>>
>>
>>> But something I had repeated many times in this forum: when performing
>>> a
>>> GW calculation, one should always choose a pseudopotential, which
>>> considers the semicore states as valence. Here for indium, I think the
>>> 4s4p4d should be treated as valence, else the exact-exchange operator
>>> could be off by several eV.
>>> Eventually, I advise you to create your own pseudo...
>>>
>>> Cheers,
>>>
>>> Fabien
>>>
>>
>>
>>
>
>
>