[Pw_forum] Negative frequencies from ph.x

[Jan Gryko]

 Dear QE users,

 The problem of negative frequencies from ph.x was discussed several times
 on the PWSCF Forum.  From what I read in the archives, the negative frequencies
 were due to too small number of k-points, or too small energy cutoff, or
 non-zero forces acting on atoms (i.e. non-equilibrium geometry).

 Unfortunately, I am still getting negative frequencies for my system
 (FCC cell, 34 Si atoms, semiconductor with ~0.7 eV gap).  
 The results were practically the same for:

 number of k-points:  1 (gamma point), 10, and 8x8x8 points (=29 k-points)
 ecutwcf = 25, 30, and 55 Ry, and ecutrho = 4*ecutwfc

 In all cases, the geometry was optimized so the total force was 0.000002 Ry/au, 
 and the max. force acting on an atom was 0.00000044 Ry/au. 
 Therefore, all atoms were in equilibrium positions, but yet about 10 first phonon
 frequencies were negative, and phonon dos (calculated with nq1=nq2=nq3=4) was wrong.
 
 I noticed that ionic forces acting on atoms were large, ~0.7 Ry/atom, but these 
 forces were compensated by opposite sign local forces.  

 Is it possible that during perturbation calculations, these two opposite forces are not
 evaluated accurately, giving unstable system and negative frequencies? 

 Any suggestions?  Thank you very much.

 JG