AIMNet2-NSE: A Transferable Reactive Neural Network Potential for Open-Shell Chemistry.

Area of Science:

• Computational Chemistry
• Materials Science
• Chemical Engineering

Background:

• Open-shell systems, including radical intermediates, are crucial in diverse chemical processes like polymerization, combustion, and catalysis.
• Accurate computational modeling of these systems is challenging due to their complex spin states.
• Existing machine learning interatomic potentials often fail to account for spin multiplicities, limiting their application in reactive chemistry.

Purpose of the Study:

• To develop a novel machine learning interatomic potential, AIMNet2-NSE (neural spin-charge equilibration), capable of accurately treating molecules and reactions with arbitrary charge and spin multiplicities.
• To enable efficient and accurate modeling of open-shell radical chemistry, which is computationally prohibitive with traditional quantum mechanical methods.

Main Methods:

• AIMNet2-NSE is built upon the AIMNet2 framework, incorporating a spin-charge equilibration mechanism.
• The model is trained on an extensive dataset of 20 million closed-shell molecules, 13 million open-shell radical configurations, and 200,000 radical reaction profiles.
• Explicit handling of spin charges allows for the prediction of spin-resolved properties.

Main Results:

• AIMNet2-NSE achieves near-density functional theory (DFT) accuracy for spin-resolved properties.
• The model exhibits favorable linear scaling, significantly outperforming the polynomial scaling of traditional electronic structure methods.
• Evaluations on radical test sets, the BASChem19 benchmark, and radical polymerization reactions demonstrate strong predictive capabilities and generalizability.

Conclusions:

• AIMNet2-NSE represents a significant advancement in machine learning interatomic potentials for open-shell systems.
• The model facilitates efficient exploration of complex radical reaction pathways and reactive intermediates.
• This work overcomes computational limitations, enabling broader application in chemical and industrial processes involving radical chemistry.