Negative frequencies in SCPH calculations

[RicardoCecconello]

Dear Dr. Tadano,

I'm finding negative frequencies for converged SCPH calculations, even though they are not present in the harmonic calculations.
This is the result of the harmonic calculation (fitting error = 0.6 %), with displacement patterns:

This is the result of the SCPH calculation at 0 K, with cutoff None 15 8, NBODY 232, fixing the 2nd order IFCs with the result above, and the DFSET was made with 320 structures randomly displaced:

This result is similar to temperatures up to 50 K, and the negative frequencies also appear for cutoff None 15 7 6 6 and NBODY 23322. The fitting error (with ENET) is about 4 %. Changing the SCPH meshes or MIXALPHA doesn't make it better, and with SELF_OFFDIAG = 1 it doesn't converge. Any ideas or suggestions on the reason for that?

[ttadano]

Please tell us

• the supercell sizes for IFC2 and IFC4
• KMESH_INTERPOLATE and KMESH_SCPH.

[RicardoCecconello]

The supercell is 4x4x2 for all IFCs, and the meshes are 4 4 2.

[ttadano]

Thanks. The KMESH_* values look reasonable.

In addition to the negative frequency along the Z-l1 line, small imaginary pockets appear near the \Gamma point in the SCPH band structure. This indicates the accuracy of the anharmonic IFC is not enough, and you may need to increase the cutoff radius for the forth-order term to resolve it.

In some cases, similar problem could be resolved by using a coarse KMESH_* values, say KMESH_INTERPOLATE = 2 2 2 or 2 2 1.

Also, did you use the same NONANALYTIC option for the harmonic and SCPH calculations? If not, please do so.

[RicardoCecconello]

I used the default value of NONANALYTIC for both of them.

Using 12 12 6 for KMESH_INTERPOLATE and KMESH_SCPH gives the following:

Those negative frequencies "move" to the Z-G direction, and those imaginary pockets near G get bigger. For KMESH_INTERPOLATE = 2 2 1 and KMESH_SCPH = 4 4 2 the spectrum is

and similarly for KMESH_INTERPOLATE = 2 2 2 and KMESH_SCPH = 2 2 2.

We'll try increasing the 4th order cutoff.