Beyond Force Metrics: Pre-Training MLFFs for Stable MD Simulations

TL;DR

Pre-training graph neural network-based machine-learning force fields on large datasets enhances the stability and reliability of molecular dynamics simulations, even when force prediction errors are similar.

Contribution

This work demonstrates that pre-training on large datasets improves MD simulation stability and the quality of learned force fields beyond just minimizing force MAE.

Findings

Pre-trained models sustain longer MD trajectories.

Pre-training leads to more structured and smoother force responses.

Force MAE alone does not predict MD stability.

Abstract

Machine-learning force fields (MLFFs) have emerged as a promising solution for speeding up ab initio molecular dynamics (MD) simulations, where accurate force predictions are critical but often computationally expensive. In this work, we employ GemNet-T, a graph neural network model, as an MLFF and investigate two training strategies: (1) direct training on MD17 (10K samples) without pre-training, and (2) pre-training on the large-scale OC20 dataset followed by fine-tuning on MD17 (10K). While both approaches achieve low force mean absolute errors (MAEs), reaching 5 meV/A per atom, we find that lower force errors do not necessarily guarantee stable MD simulations. Notably, the pre-trained GemNet-T model yields significantly improved simulation stability, sustaining trajectories up to three times longer than the model trained from scratch. By analyzing local properties of the learned force fields, we find that pre-training produces more structured latent representations, smoother force responses to local geometric changes, and more consistent force differences between nearby configurations, all of which contribute to more stable and reliable MD simulations. These findings underscore the value of pre-training on large, diverse datasets to capture complex molecular interactions and highlight that force MAE alone is not always a sufficient metric of MD simulation stability.