[Pw_forum] Na phonon issues

Nicola Marzari marzari at MIT.EDU
Thu May 17 18:20:04 CEST 2007

Dear Miguel,

I do not think we have ever managed, even for very high-thresholds
everywhere , to converge the lowest frequencies so that at Gamma
or close to Gamma they would come out "right" in a direct phononq
calculation. In other words, it does seem that the converged values that
phonon gives, for the lowest frequency at Gamma or nearby, it's not the
value that you would expect, i.e. sitting on a linear dispersion from
Gamma, with slopes determined by the elastic constants (if the q is on
a high symmetry direction).

Diamond carbon was the worst case, with a converged value at Gamma that
(if I recall correctly - Nicola B ?) could be anywhere between -10
and 50 cm^-1 , depending on the pseudo (note that these numbers are
small, compared to the optical frequencies of diamond, around 1500

Why is that ? Not sure - Stefano de Gironcoli advice was to try
and remove the xc part, to see if that was the culprit, but we never

The only solution is to calculate the dynamical matrix explicitly on a
homogeneous q grid that is not too fine (so that the q closest to gamma
but not gamma has all frequencies well converged), then transform in
real space, impose acoustic sum rules, and then obtain the phonon
dispersions everywhere from the fourier transform and diagonalization of 
the interatomic force constants.

But it's true that it's only partially satisfactory, and I wouldn't mind
figuring out a solution, especially because accurate results at long
wavelengths are very important for thermal transport.

One additional note about worst case scenarios: when two dispersions are 
degenerate, but maybe numerically slightly different, we found an 
instability in the back-transform (I think because diagonalization and 
ft do not commute) with the two degenerate branches being degenerate at 
the original q's, and mostly everywhere, but with one of the two
displaying a infinite derivative at a critical q close to gamma, in 
between gamma and one of the original q's. This was the case for CNTs.


> Dear Nicola,
> thanks for your advice. I've been reading this, as well as the 
> troubleshooting part of the manual. Please note that I am *not* talking 
> about \Gamma. I have, then, increased the K_POINTS to 30 30 30, which 
> seems pretty dense to me, unless I were to do electron-phonon 
> calculations. This still yields the following omegas:
>      omega( 1) =      -0.102074 [THz] =      -3.404853 [cm-1]
>      omega( 2) =       0.766422 [THz] =      25.565257 [cm-1]
>      omega( 3) =       1.200480 [THz] =      40.043969 [cm-1]
> These aren't really that different from the ones I got with a 20 20 20 
> grid (although the longitudinal mode is 2.5% higher). Just in case, I'm 
> running the script with an increased degauss value, because it's a bit 
> too small.
> The thing is, the pseudopotential is norm-conserving (or so it says on 
> the header), and thus it's not an issue of not setting ecutrho. And the 
> positive values are close to the numbers I get with the other pseudos. I 
> am also reading the correct file, as I delete the temporary dir at the 
> beginning and the end of my script. Am I being to harsh or lenient with 
> the phonon threshold?
> Thanks again for your comments,
> Miguel

Prof Nicola Marzari   Department of Materials Science and Engineering
13-5066   MIT   77 Massachusetts Avenue   Cambridge MA 02139-4307 USA
tel 617.4522758 fax 2586534 marzari at mit.edu http://quasiamore.mit.edu

More information about the users mailing list