Unstable phonons after fine tunning

[Harpriyaminhas]

Hlo CHGNet team,

I am exploring CHGNet universal potentials for phonon calculations, but for a few systems, I am encountering unstable modes with negative phonon frequencies. Could you suggest ways to improve these unstable modes?

I have followed the fine-tuning tutorial on GitHub and fine-tuned the pretrained CHGNet model using a dataset of 10,000 VASP AIMD results. Despite this effort, I am still unable to achieve fully positive phonon frequencies. I would appreciate your guidance on resolving this issue.