PFT: Phonon Fine-tuning for Machine Learned Interatomic Potentials

TL;DR

This paper introduces phonon fine-tuning (PFT), a method to improve machine learned interatomic potentials by directly supervising second-order force constants, leading to better vibrational property predictions and generalization to higher derivatives.

Contribution

The paper presents PFT, a novel approach that enhances MLIPs by matching energy Hessians to DFT data, improving accuracy of vibrational and thermal properties.

Findings

PFT improves Nequix MP by 55% on phonon thermodynamic properties.

PFT achieves state-of-the-art accuracy on the MDR Phonon benchmark.

PFT enhances predictions of thermal conductivity involving third-order derivatives.

Abstract

Many materials properties depend on higher-order derivatives of the potential energy surface, yet machine learned interatomic potentials (MLIPs) trained with a standard loss on energy, force, and stress errors can exhibit error in curvature, degrading the prediction of vibrational properties. We introduce phonon fine-tuning (PFT), which directly supervises second-order force constants of materials by matching MLIP energy Hessians to DFT-computed force constants from finite displacement phonon calculations. To scale to large supercells, PFT stochastically samples Hessian columns and computes the loss with a single Hessian-vector product. We also use a simple co-training scheme to incorporate upstream data to mitigate catastrophic forgetting. On the MDR Phonon benchmark, PFT improves Nequix MP by 55% on average across phonon thermodynamic properties and achieves state-of-the-art accuracy among models trained on Materials Project trajectories. PFT also generalizes to improve properties beyond second-derivatives, improving thermal conductivity predictions that rely on third-order derivatives of the potential energy.