Iterative Pretraining Framework for Interatomic Potentials

TL;DR

The paper introduces IPIP, an iterative pretraining framework for machine learning interatomic potentials that enhances accuracy and efficiency in molecular simulations with less data and lightweight models.

Contribution

IPIP is a novel iterative pretraining method that improves MLIP performance by preventing local minima and reducing data and model complexity requirements.

Findings

Achieves over 80% reduction in prediction error.

Up to 4x speedup in simulations.

Outperforms general-purpose force fields.

Abstract

Machine learning interatomic potentials (MLIPs) enable efficient molecular dynamics (MD) simulations with ab initio accuracy and have been applied across various domains in physical science. However, their performance often relies on large-scale labeled training data. While existing pretraining strategies can improve model performance, they often suffer from a mismatch between the objectives of pretraining and downstream tasks or rely on extensive labeled datasets and increasingly complex architectures to achieve broad generalization. To address these challenges, we propose Iterative Pretraining for Interatomic Potentials (IPIP), a framework designed to iteratively improve the predictive performance of MLIP models. IPIP incorporates a forgetting mechanism to prevent iterative training from converging to suboptimal local minima. Unlike general-purpose foundation models, which frequently underperform on specialized tasks due to a trade-off between generality and system-specific accuracy, IPIP achieves higher accuracy and efficiency using lightweight architectures. Compared to general-purpose force fields, this approach achieves over 80% reduction in prediction error and up to 4x speedup in the challenging Mo-S-O system, enabling fast and accurate simulations.