Phonon Calculation Skill

This skill provides tools for calculating vibrational properties of materials using Machine Learning Interatomic Potentials (MLIPs).

1. Prerequisites

• The appropriate MLIP wrapper must be available (MACEWrapper, MatGLWrapper, or FAIRCHEMWrapper).
• matcalc, phonopy, and phono3py must be installed in the relevant conda environment.

2. Choosing a Foundation Potential

Phonon calculations are highly sensitive to the quality of the potential energy surface (PES).

[!IMPORTANT]

• Use OMAT or MatPES trained models: These models (e.g., MACE-OMAT-0-small, TensorNet-MatPES-r2SCAN) are specifically optimized for forces and vibrational stability.
• Avoid MPtrj-trained models: Models trained primarily on the MPtrj dataset (e.g., CHGNet-MPtrj) suffer from the "softening" problem, where the calculated phonon frequencies are significantly lower than DFT values.

Refer to the foundation-potentials skill for more details.