The ABINIT Users Mailing List ( CLOSED )

[Jorge Iniguez <jiniguez@nist.gov>]

Hi Artem,

Wow!, you are fast answering emails :)

> Try any system (or even look at the tutorial) - you will see small
> departures from exact degeneracies. I guess (didn't check) that these
> departures would increase as your calculation gets cruder (or Ecut gets
> lower).

Hmm, I guess small departures from degeneracy pertaining *very small*
numbers can occur, and they are caused by numerical noise in the
diagonalization of the dynamical matrix.

If you look at the tutorial, the result for AlAs phonons at Gamma using a
3 a.u. cutoff is:

  Phonon wavevector (reduced coordinates) :  0.00000  0.00000  0.00000
 Phonon energies in Hartree :
   2.586632E-06  2.590723E-06  2.614440E-06  1.568560E-03  1.568560E-03
   1.568560E-03

I have just repeated the calculation with a 10 a.u. cutoff; the result is:

  Phonon wavevector (reduced coordinates) :  0.00000  0.00000  0.00000
 Phonon energies in Hartree :
  -5.857224E-07 -5.034097E-07 -4.878934E-07  1.667388E-03  1.667388E-03
   1.667388E-03

Finally, I have also used a 1 a.u. cutoff, and got:

  Phonon wavevector (reduced coordinates) :  0.00000  0.00000  0.00000
 Phonon energies in Hartree :
   4.639947E-05  4.639949E-05  4.639950E-05  2.492164E-03  2.492164E-03
   2.492164E-03

As you see, in the three cases the optical modes are always exactly
degenerate. The acoustic modes are funny: They are most degenerate for the
smallest cutoff! This makes sense to me because they are larger in that
case (i.e. they are calculated less accurately) and thus less affected by
numerical noise in the diagonalization. Convincing?


> Now, the acoustic frequencies you get at Gamma are SO large that there
> is obviously something very wrong in the calculation. Which
> pseudopotentials do you use? For some types (e.g., HGH) the needed Ecut
> is very high (easily above 100-200 Ha). Though from the small changes in
> your results I suppose your calculation is likely converged...

I am using tm pseudos from the Khein-Allan distribution. The 1500 eV
cutoff should be more than fine...


> Some people have studied hexagonal materials with ABINIT, and did not
> seem to get in any troubles - e.g., Xavier studied quartz and got
> perfectly reasonable results. All I can suggest is a problem with your
> input file.

That's an important piece of information!  Oh well, I guess a mistake in
the input is always possible (!) but I still vote for the bug possibility
(even if it did not affect Xavier's quartz calculations...).

Cheers,
Jorge