Chapter 19: Machine Learning for Heterogeneous Catalysis: Global Neural Network Potential from Construction to Applications
Published:15 Jul 2020
S. Ma, P. Kang, C. Shang, and Z. Liu, in Machine Learning in Chemistry
Download citation file:
While the potential energy surface (PES) determines the physicochemical properties of matter, chemical system surfaces are often too complex to solve even with modern computing facilities. Heterogeneous catalysis, being widely utilized in industry, calls for new techniques and methods to resolve the active site structure and reaction intermediates at the atomic scale. In this chapter, we provide an overview of recent theoretical progress on large-scale atomistic simulation via the machine learning global neural network (G-NN) potential developed by our research group in recent years, focusing on methodology and representative applications in heterogeneous catalysis. The combination of global optimization and machine learning provides a convenient and automated way to generate the transferable and robust G-NN potential, which can be utilized to reveal new chemistry from unknown regions of the PES at an affordable computational cost. The predictive power of the G-NN potential is demonstrated in several examples, where the method is applied to explore the material crystal phases and the structure of supported catalysts, to follow surface structure evolution under high-pressure hydrogen and to determine the ternary oxide phase diagram. Limitations and future directions of the G-NN potential method are also discussed.