A reactive neural network framework for water-loaded acidic zeolites

Area of Science:

• Computational Chemistry
• Materials Science
• Catalysis

Background:

• Water and ion dynamics in zeolite micropores dictate hydrothermal stability, acidity, and catalytic activity.
• Operando studies of acidic zeolites are computationally expensive, limiting research to simplified models and specific cases.

Purpose of the Study:

• To develop a versatile reactive neural network potential (NNP) for acidic zeolites.
• To enable accurate and efficient simulations of zeolite catalysts under realistic operating conditions.

Main Methods:

• Development of a reactive neural network potential (NNP) applicable to the entire class of acidic zeolites.
• The NNP aims to cover a wide range of water concentrations and Si/Al ratios.
• Incorporation of techniques for data-efficient learning and accelerated rare event sampling.

Main Results:

• The NNP achieves (meta)GGA DFT level accuracy, significantly improving computational sampling.
• The potential facilitates the discovery of novel chemical phenomena, such as collective defect formation.
• Demonstration of NNP's extensibility for incorporating higher-level references and enhancing sampling efficiency.

Conclusions:

• The developed NNP represents a significant advancement for simulating realistic catalysts under operando conditions.
• This tool opens new avenues for understanding zeolite behavior and discovering new catalytic chemistry.