Data-Efficient Construction of High-Fidelity Graph Deep Learning Interatomic Potentials

TL;DR

This paper presents a multi-fidelity approach to efficiently develop high-accuracy graph neural network interatomic potentials by combining low- and high-fidelity data, reducing computational costs significantly.

Contribution

It introduces a data-efficient multi-fidelity training method for graph neural network potentials that achieves high accuracy with less high-fidelity data.

Findings

Multi-fidelity models match high-fidelity-only models in accuracy.

Using 10% high-fidelity data yields comparable results to larger high-fidelity datasets.

Method reduces computational costs for developing interatomic potentials.

Abstract

Machine learning potentials (MLPs) have become an indispensable tool in large-scale atomistic simulations because of their ability to reproduce ab initio potential energy surfaces (PESs) very accurately at a fraction of computational cost. For computational efficiency, the training data for most MLPs today are computed using relatively cheap density functional theory (DFT) methods such as the Perdew-Burke-Ernzerhof (PBE) generalized gradient approximation (GGA) functional. Meta-GGAs such as the recently developed strongly constrained and appropriately normed (SCAN) functional have been shown to yield significantly improved descriptions of atomic interactions for diversely bonded systems, but their higher computational cost remains an impediment to their use in MLP development. In this work, we outline a data-efficient multi-fidelity approach to constructing Materials 3-body Graph Network (M3GNet) interatomic potentials that integrate different levels of theory within a single model. Using silicon and water as examples, we show that a multi-fidelity M3GNet model trained on a combined dataset of low-fidelity GGA calculations with 10% of high-fidelity SCAN calculations can achieve accuracies comparable to a single-fidelity M3GNet model trained on a dataset comprising 8x the number of SCAN calculations. This work paves the way for the development of high-fidelity MLPs in a cost-effective manner by leveraging existing low-fidelity datasets.