The ABINIT Users Mailing List ( CLOSED )

[Martin Fuchs <fuchs@fhi-berlin.mpg.de>]

Hello,

the following is just a comment on the pseudopotentials for 
alkali/earthalkali,
in the case where it is them that cause trouble:

The 4s valence density (all there is in Mg and Ca) is a nonmonotonic
function of the distance from the ion center, i.e. it has an "off center"
maximum, unlike the physical (and I believe all other pseudo) atoms.
What I wonder is whether this might render the xc energy "unusually"
sensitive to moves of the atoms with respect to the fft grid.
If so, adding a (small) core density (nlcc pseudopotential) might help,
makeing the density monotonically decaying. How much is needed could be
seen by inspecting the xv...density file of the fhipp pseudopotential 
generator.
There is no extra cost for plane wave cutoff, it should be high enough
already due to the F or B.

Cheers,
Martin


verstraete@pcpm.ucl.ac.be wrote:

> Hello Jorge,
>
> I had a similar problem with CaF2 in a recent paper (90 cm-1 acoustic
> modes and nothing to do). CaF2 is insulating. What I ended up doing was
> changing the pseudopotential: the ASR sum rule error comes from the XC
> energy. The XC energy is calculated on the FFT grid, which breaks
> translational symmetry.
>
> If you do total energy calculations while moving the atoms by a fraction
> of the FFT grid spacing, you will see oscillations of the energy with the
> FFT period, and you can extract the phonon modes. Often (this was my case)
> the ASR breaking is only due to one element: if you do the same
> calculation with only 1 atom in the unit cell at a time (Mg then B) you
> will see which one contributes the most, and you can replace its
> pseudopotential. Are your psp hgh? For TM (fhi) I always have good ASR
> values, but the HGH F pseudo was responsible for all of my high acoustic
> modes. All of this method comes from Philippe Ghosez's thesis on BaTiO3
> http://dept.phys.ulg.ac.be/mate/phythema/PhD-Ph.Ghosez.ps.gz
>
> The disappearance of the acoustic modes for insulating occupations could
> be due to the states that actually contribute to the XC energy variation
> with translation. If they are the first unoccupied ones, there you go.
> Otherwise, it may be that upon a small translation, the states below Ef
> simply transform amongst themselves, whereas the partially filled ones
> above Ef mix with others that were unoccupied before. If the occupations
> are finite-T then you will see this effect in the energy.
>
> Your optical modes move a lot, though, even with the very dense kpoint
> sampling. Your cutoff is also not that high. I wouldn't be surprised if
> some of the error needed a 60 Ha cutoff or more to go away (your
> frequencises look converged though, I agree).
>
> Hope it helps.
>
> Matthieu
>
> On Thu, 4 Sep 2003, Jorge Iniguez wrote:
>
> > Dear abinitioners,
> >
> > This email is a (long-due) follow up to an early-August discussion on
> > phonon calculations for MgB2. (Sorry for the delay. Summer is summer!)
> > The problem then was that I was not getting the right degeneracies for the
> > phonons (even though abinit was recognizing the right symmetry). As Lu
> > Fu-Fa pointed out, the solution was to specify kptopt=2 in the Gamma
> > phonon calculation.  I do not yet understand why kptopt=2 solves the
> > problem (which is probably related with the construction of the dynamical
> > matrix from an irreducible set of distortions), but OK, it works now.
> >
> > HOWEVER, once kptopt is set to 2 and the degeneracies recovered, there is
> > the remaining problem of the acoustic-mode frequencies.  One obtains the
> > following Gamma phonons for MgB2:
> >
> >   Phonon wavevector (reduced coordinates) :  0.00000  0.00000  0.00000
> >  Phonon energies in Hartree :
> >    8.717196E-04  8.717208E-04  1.195609E-03  2.028753E-03  2.028757E-03
> >    3.068304E-03  3.068306E-03  3.335226E-03  4.394181E-03
> >
> > where the system is treated as a metal, cutoff=1000eV and k-point grid is
> > 16x16x12.  The acoustic frequencies are terribly large! (around 200 cm-1)
> >
> > If I increase the cutoff to 1500 eV, I get:
> >
> >   Phonon wavevector (reduced coordinates) :  0.00000  0.00000  0.00000
> >  Phonon energies in Hartree :
> >    8.718879E-04  8.719018E-04  1.195568E-03  2.029070E-03  2.029101E-03
> >    3.070568E-03  3.070573E-03  3.335321E-03  4.394402E-03
> >
> > which is the same result (cutoff=2000eV does not change anything).  If I
> > use an even denser grid of 20x20x16 k points, I get:
> >
> >   Phonon wavevector (reduced coordinates) :  0.00000  0.00000  0.00000
> >  Phonon energies in Hartree :
> >    8.723311E-04  8.723325E-04  1.195953E-03  2.031243E-03  2.031243E-03
> >    3.045633E-03  3.045633E-03  3.340805E-03  4.400362E-03
> >
> > So, the calculation was very well converged with the initial conditions.
> >
> > Now, if I treat the system as an *insulator*, for 1000 eV and 16x16x12, I
> > get:
> >
> >   Phonon wavevector (reduced coordinates) :  0.00000 0.00000 0.00000
> >  Phonon energies in Hartree :
> >   -1.347742E-05 -1.098398E-05  4.037062E-06  1.539327E-03  1.539351E-03
> >    1.760802E-03  3.431914E-03  4.124680E-03  4.124681E-03
> >
> > with reasonably small acoustic frequencies!!!  So, it seems there is
> > something funny about the acoustic phonons in the metallic case. I guess
> > this is not an input-file problem...  (a typical input file goes
> > attached).  Any suggestions?
> >
> > Cheers,
> > Jorge
>
> --
> ===================================================================
> Matthieu Verstraete               mailto:verstraete@pcpm.ucl.ac.be
> PCPM, Boltzmann, pl. Croix du Sud, 1            tel: 010/ 47 86 81
> B-1348 Louvain-la-Neuve  Belgium                fax: 010/ 47 34 52

--
Martin Fuchs
Fritz-Haber-Institut der MPG
Faradayweg 4-6
14195 Berlin
Germany
Phone: +30 8413 4802
Fax:     +30 8413 4701
E-Mail: fuchs@fhi-berlin.mpg.de