Active rejection enables reliable generalization of universal machine-learning interatomic potentials

2026-07-10

AI Summary

This work addresses the challenge of ensuring reliable predictions with universal machine-learned interatomic potentials (uMLIPs) under conditions of scarce training data and vast materials space. The authors propose an Adaptive Teacher Routing (ATR) framework that introduces, for the first time, an active rejection mechanism to transform high-fidelity training data construction into a structure-level uncertainty decision problem. By integrating structural descriptors, teacher identities, and model disagreement, ATR evaluates the reliability of multiple pretrained uMLIPs and generates pseudolabels only for high-confidence structures. Using merely 0.2% of true r²SCAN labels, the method constructs a large-scale dataset of 2.89 million high-quality pseudolabels. The resulting lightweight CHGNet model significantly outperforms baseline methods on the MP-r²SCAN benchmark and in high-temperature molecular dynamics simulations, effectively preventing structural collapse and enhancing dynamic stability.

Abstract

Universal machine learning interatomic potentials (uMLIPs) bridge quantum-mechanical accuracy and large-scale molecular dynamics, but the cost of high-accuracy calculations such as r$^2$SCAN limits training to datasets that remain small relative to the open materials space. Strong average benchmark performance also does not guarantee reliable energy--force predictions for every structure. We propose Adaptive Multi-Teacher Routing (ATR), which reformulates high-fidelity data construction as a structure-wise decision problem under uncertainty. Using a small set of real r$^2$SCAN labels, ATR calibrates multiple pretrained uMLIP teachers and combines structural descriptors, teacher identity, and inter-teacher disagreement to estimate the reliability of each structure--teacher pair. It selects high-confidence predictions for pseudo-label generation and rejects structures for which no teacher is sufficiently reliable. With real r$^2$SCAN labels for only 0.2\% of candidate structures, ATR distils 2.89 million traceable r$^2$SCAN-level pseudo-labels for pretraining. On held-out r$^2$SCAN structures and the MP-r$^2$SCAN benchmark, a lightweight CHGNet trained on the ATR-generated dataset consistently outperforms the baseline and non-routed controls. Finite-temperature molecular dynamics further shows that ATR improves dynamical robustness across multiple material systems, maintaining stable trajectories where baseline simulations undergo catastrophic structural collapse. These results establish active rejection as an effective mechanism for converting multiple pretrained uMLIPs into a scalable and reliable data-construction system for high-fidelity uMLIPs.

Problem

• universal machine learning interatomic potentials
• high-fidelity data construction
• reliable generalization
• active rejection
• pseudo-labeling

Innovation

• active rejection
• universal machine learning interatomic potentials
• adaptive multi-teacher routing
• pseudo-labeling
• high-fidelity data construction