Imginary phonons obtained by auxiliary matrices

[hellolori]

Recently, I performed SSCHA calculations on a two-dimensional material that exhibits imaginary phonon modes in the harmonic approximation. The harmonic phonons were calculated using Quantum ESPRESSO on a 6×6×1 q-grid, as illustrated below.

Based on the harmonic dynamical matrices, the SSCHA calculations were then carried out using a 3×3×1 supercell. The SSCHA procedure converged within three populations. The evolution of the free energy, force constant (FC) gradients, and other related quantities are shown below.

From these figures, it appears that the SSCHA calculations converged well. However, I found that the phonon frequencies obtained from both the auxiliary and free energy Hessian matrices still display imaginary modes, as illustrated below.

This result is puzzling, as I have understood that the auxiliary phonon frequencies should always be positive. I have also recalculated the phonons directly using the auxiliary matrices on a 3×3×1 q-grid without referencing the harmonic matrices, and obtained similar results, as shown below.

Could you provide any suggestions regarding the appearance of imaginary phonon modes when using auxiliary matrices? Is there potentially something wrong with my calculations, or is it possible for the auxiliary phonon modes to be imaginary under certain circumstances?

[DjordjeDangic]

Hello,

I would guess the imaginary phonon frequencies appear in the q points that are not in the supercell (for the q points that are not in the 3x3x1 grid in your case). This might be an interpolation issue, especially for the phonon modes close to Gamma. Try with a larger supercell or increase the number of configurations and see if that improves your results.

[adenchfi]

Hello,

I've also encountered this issue. Perhaps there can be an alternative linear interpolation procedure for when Fourier interp is producing artificial imaginary frequencies?

[mesonepigreco]

Yes, this is on our TODO list; however, we need a good guy to implement and test it.