The ABINIT Users Mailing List ( CLOSED )

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

> Dear all,
> I'm trying to perform a G0W0-PAW calculation for CdTe using Abinit
> v5.7.3. Since there are no PAW-datasets available for Cd nor Te, I have
> constructed two sets using Atompaw and converted them to abinit-format
> via AtomPAW2Abinit v3.2.0. They are both quite expensive, but at least
> they produce the same lattice constants as the PAW-sets available in
> vasp for Cd, Te and CdTe. Details (projectors, reference energies, etc.)
> can be found in the attached file. I should also mention that the
> ground-state calculations I've tried with these PAW-sets finish without
> any warnings. However, during the self-energy calculation I encounter
> two warnings concerning the densities:
>
> <snip>
> mkdenpos : WARNING -
>    Density went < 0 at      1281 points
>    and was set to  1.00E-14.  Lowest was  -0.31E-01.
> </snip>
>
> and
>
> <snip>
>  mkdenpos : WARNING -
>   Density went < 0 at       240 points
>   and was set to  1.00E-14.  Lowest was  -0.60E-01.
> </snip>
>
> The problem firstly occurs when calculating the exchange potential for
> trho1+nhat (in pawcxm.F, tVxc1_val term) and later for rhor+nhat in
> rhohxc when called from calc_vHxc_braket. The input file is a somewhat
> modified copy of the the test abinit-5.7.3/tests/v5/Input/t66.in. The
> above warnings occur regardless of different cutoff-energies (ecut,
> ecutwfn, ecutsigx), but I may of course have missed something important.
> Any comments or suggestions are welcome.

 Dear Daniel,

 For GW calculations, one has to calculate the contribution to the XC
 potential given by the valence electrons.
 In the standard approach used to decouple the core from the valence
electrons,
 the valence-only XC term has to be subtracted out from the self-energy.
 You might have a look at Arnaud and Alouani's paper
http://prola.aps.org/abstract/PRB/v62/i7/p4464_1

 Therefore, using Kresse and Joubert's notation
(http://link.aps.org/abstract/PRB/v59/p1758),
 one has to calculate

 vxc[\tilde n + \hat n]

 instead of

 vxc[\tilde n + \hat n + \tilde n_{core}]


 The "smooth" charge, $\tilde n$, being the sum of the squared modulus of the
 pseudized wavefunctions, is always positive. The same property doesn't
hold anymore for \hat n.

 Usually when the pseudized core charge is added (i.e. $\tilde n + \hat n
+ \tilde n_{core}$ )
 the overall density turns out to be positive anywhere.
 This is not true anymore when only the partial  $\tilde n + \hat n$ is
considered.
 This explains your last WARNING related to the treatment of the plane
wave contribution
 to the valence charge density.
 A similar problem arises in the treatment of the on-site representation
of the pseudized
 valence charge (the first WARNING you got).

 The only solution I see consists of evaluating the valence-only
contribution to vxc using the expression

 vxc[\tilde + \hat n + \tilde n_{core}] - vxc[\tilde n_{core])

 In this case the densities in both XC functionals are always positive and
 we should get rid of the WARNINGs.
 It's not difficult but it requires some coding.
 Don't know whether this change will be included in 5.7.4 or in the
forthcoming 6.0


 Best Regards
 Matteo Giantomassi