[Pw_forum] Negative frequencies in phonon calculation

[Alberto Milani]

Dear all,
I am calculating phonon frequencies for an infinite linear carbon chain and up 
to now I have obtained puzzling results.
The two degenerate acoustic branches relative to transverse modes show large 
negative frequencies. At Gamma I found omegas of -100 cm-1. Then the curve 
begins to rise up to a value of 50 cm-1 but then a sharp discontinuity occurs 
and the frequencies fall down to -700 cm-1.
In my calculation, I studied a well relaxed structure (Peierls distorted) and 
I tried also to increase the quality of the results by using all the tricks 
explained in the manual. My cell has parameters a=10, b=10 and c=2.5671, I am 
using  Vanderbilt ultrasoft pseudopotential with PBE GGA and a K-POINTS grid 
of 1 1 70. By increasing ecutwfc and/or ecutrho (at the beginning I chose 
35-350 Ry) I have obtained no better results and also by reducing convergence 
threshold both for SCF and phonons calculation nothing goes better. The same 
bad results are found by improving the k points grid (e.g. 10 10 70). In my 
calculation, I have directly computed the dynamical matrices without 
interpolating with q2r.x and matdyn.x codes and I have chosen 15 q-points in 
the irreducible Brillouin zone (the q vector has been chosen to be parallel 
to the linear chain axis).
All the other branches (one longitudinal acoustic and three optical branches) 
are good and I don't understand what is going bad in the computation.
Does anybody know what is the source of the problem?
Thanks in advance.
A. M.