What type of study is this?

September 10, 2025Open Access

Reactive Active Learning: An Efficient Approach for Training Machine Learning Interatomic Potentials for Reacting Systems

Key Points

Reactive active learning effectively trains machine learning interatomic potentials to mimic quantum mechanical accuracy.
This approach combines automated exploration for reactions with uncertainty-driven active learning to sample transition states.
Real-world demonstrations included ammonia synthesis and methane activation, revealing significant predictive accuracy in complex systems.
The method's ability to simulate large systems over nanosecond timescales opens new avenues for investigating reaction mechanisms.

Abstract

Discovering chemical reaction pathways using quantum mechanics is impractical for many systems of practical interest because of unfavorable scaling and computational cost. While machine learning interatomic potentials (MLIPs) trained on quantum mechanical data offer a promising alternative, they face challenges for reactive systems due to the need for extensive sampling of the potential energy surface in regions that are far from equilibrium geometries. Unfortunately, traditional MLIP training protocols are not designed for comprehensive reaction exploration. We present a reactive active learning (RAL) framework that is designed to efficiently train MLIPs to achieve near-quantum mechanical accuracy for reactive systems for situations where one may not have prior knowledge of the possible transition states, reaction pathways, or even the potential products. Our method combines automated reaction exploration, uncertainty-driven active learning, and transition state sampling to build accurate potentials. We demonstrate RAL's effectiveness across three different systems: uncatalyzed ammonia synthesis (gas-phase), methanimine hydrolysis (solution phase), and methane activation on titanium carbide surfaces (heterogeneous). The resulting MLIPs accurately predict reaction barriers and transition states. For catalysis, we show that RAL-trained MLIPs identify Ti2C as the most active methane activation surface (90% decomposition at 1000 K) through C-vacancy mediated mechanisms. The framework's ability to simulate large systems (∼900 atoms) over nanosecond time scales provides previously inaccessible insights into surface poisoning and reaction networks. We show that reactive exploration is essential for adequately capturing the potential energy surface, with chemical and configurational sampling working synergistically to improve model accuracy. Our results establish general guidelines for training robust reactive potentials and open new possibilities for computational discovery of catalysts and reaction mechanisms.

Read Full Paperexternally

AIに質問

Bookmark

View Full Paper