I am observing discrepancies in the phonon dispersion results obtained using the Density Functional Perturbation Theory (DFPT) method implemented in Phonopy. Specifically, when I use the same INCAR settings, I obtain correct phonon dispersion results for a 4x4x1 supercell. However, when I use 1x1x1 or 3x3x1 supercells, I encounter negative frequencies. What could be causing this inconsistency in the results, and how can I address it?:

enter image description here

  • 1
    $\begingroup$ Please change the image by the code it self. $\endgroup$
    – Camps
    Mar 18 at 13:20
  • 1
    $\begingroup$ Please use a code block instead of a screenshot image. $\endgroup$ Mar 18 at 14:41
  • $\begingroup$ please also post k-points for each supercell size. $\endgroup$ Mar 18 at 16:43
  • $\begingroup$ @TylerSterlingFor a 4x4x1 supercell, I utilized a Gamma-centered k-point grid with 1x1x1 spacing (0 0 0). However, when employing a 16x16x1 k-point grid for a 3x3x1 supercell, I obtained accurate results. I set the calculation algorithm to 'veryfast' for the 3x3x1 supercell simulations, which yielded the correct outcomes.does ALGO plays important role? $\endgroup$ Mar 18 at 18:07
  • $\begingroup$ Thanks for the info. For "I obtained accurate results", do you mean no-imaginary modes and the phonon dispersions looked as expected? Also please note that DFPT with 4x4x1 vs e.g. 1x1x1 sample different q-points. VASP DFPT only works for q=(0,0,0), so a supercell is needed to sample non-zero q phonons. Other q-points are calculated by interpolation in phonopy, so you need to converge wrt. supercell size. $\endgroup$ Mar 18 at 18:54

1 Answer 1


Based on your comments above, it looks like you increased k-point sampling for the 3x3x1 supercell and found the correct dispersions. It seems like the imaginary modes in the smaller cells is due to insufficient sampling of the BZ. Indeed, converging phonon dispersions requires fairly dense BZ sampling, since forces have to be very accurate. The fact that you had unstable modes for the small supercells with coarse BZ sampling, while you had stable modes in the large supercell (implicitly more dense BZ sampling!) is expected. I am glad you got it sorted out!

  • $\begingroup$ yes for some different ALGO methods i got pathological instabilities.Is there any way to decide to ALGO according to crystal structure? $\endgroup$ Mar 19 at 18:09
  • $\begingroup$ I think, according to the rules, you’re supposed to open a new question instead of asking a new one here. $\endgroup$ Mar 20 at 20:30

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .