ELPH_IGNORE_IMAG_PHONONS: What is skipped, and what is the threshold?

[saranya_velliyarat]

Hi,

I am computing electronic conductivity (ELPH_MODE = TRANSPORT, SERTA) of a system with VASP and phelel. I used the phelel workflow (displaced supercells → phelel_params.hdf5 → VASP transport run).

For the relaxed structure, the phonons computed from the force constants in phelel_params.hdf5 are all real, except the acoustic modes at Gamma, which come out slightly negative (about -0.006 THz). Without ELPH_IGNORE_IMAG_PHONONS, VASP stops transport calculation with an imaginary-frequency error. With ELPH_IGNORE_IMAG_PHONONS = .TRUE., the run completes.

I have two questions about what the tag does:

1) When a q-point has an imaginary frequency, does VASP skip the entire q-point (all phonon branches and all k -> k+q scattering processes with that q), or only the imaginary mode(s) at that q-point?

2) What criterion does VASP use to classify a frequency as imaginary? Is it strictly omega^2 < 0, or is there a numerical tolerance below which small negative values are still accepted?

Thank you!
Saranya

[martin.schlipf]

1. Only the imaginary mode(s) are excluded, not the whole q-point. VASP stores an imaginary frequency as a negative number, ω = −√|ω²|. For every mode with ω ≤ 0, the phonon displacement (and therefore the electron-phonon matrix element g) is set to zero, so that mode does not contribute to the scattering rates. All other branches at the same q, and all k → k+q processes for that q, are kept. In addition, any mode with |ħω| < 10⁻⁴ eV is skipped, whatever its sign; this removes the acoustic modes at and very near Γ. ELPH_IGNORE_IMAG_PHONONS only disables the stop. It does not change how these modes are treated.
2. The stop is triggered only if some mode at some q-point of the irreducible q-mesh has ħω < −10⁻⁴ eV (about −0.024 THz, or −0.8 cm⁻¹). Smaller negative values are accepted.

Your Γ-point value of −0.006 THz is inside this tolerance, so the stop must have come from another q-point of the transport mesh. The error message prints that q-point and its three lowest frequencies. If these are small acoustic dips near Γ, ignoring them is harmless. If they come from a genuine soft mode, the corresponding scattering channels are missing from the result, and the structure or force constants should be checked.

PS: I'm assuming that you saw an error message like this. Note that the frequencies are printed in THz here.

Code: Select all

 -----------------------------------------------------------------------------
|                                                                             |
|     EEEEEEE  RRRRRR   RRRRRR   OOOOOOO  RRRRRR      ###     ###     ###     |
|     E        R     R  R     R  O     O  R     R     ###     ###     ###     |
|     E        R     R  R     R  O     O  R     R     ###     ###     ###     |
|     EEEEE    RRRRRR   RRRRRR   O     O  RRRRRR       #       #       #      |
|     E        R   R    R   R    O     O  R   R                               |
|     E        R    R   R    R   O     O  R    R      ###     ###     ###     |
|     EEEEEEE  R     R  R     R  OOOOOOO  R     R     ###     ###     ###     |
|                                                                             |
|     Negative frequencies found in the phonon dispersion. At q-point =       |
|     0.062 0 0 the 3 lowest freqs are -0.041 0.499 0.812. You can            |
|     silence this error by setting ELPH_IGNORE_IMAG_PHONONS in the INCAR     |
|     file.                                                                   |
|                                                                             |
|       ---->  I REFUSE TO CONTINUE WITH THIS SICK JOB ... BYE!!! <----       |
|                                                                             |
 -----------------------------------------------------------------------------