Catalyze Materials Science with Machine Learning

Abstract

Discovering and understanding new materials with desired properties are at the heart of materials science research, and machine learning (ML) has recently offered special shortcuts to the ultimate goal. Thanks to the nourishment of computer hardware and computational chemistry, the development of calculated scientific data repositories could fuel the ML models to investigate the vast materials space. At this moment, understanding this revolutionary paradigm is urgent, and this Review aims to deliver comprehensive information about the collaboration of ML with materials science. This Review summarizes recent achievements in catalysts design, which can be benefitted from ML because of the complex nature of catalytic reactions and vast candidate materials space. ML models for catalyst design could be transferred to applications in other domains and vice versa. The basic concepts of ML algorithms and practical guides to materials scientists are also demonstrated. Moreover, challenges and strategies of applying ML are discussed, which should be addressed collaboratively between materials scientists and ML communities. Ultimate integrations of ML in materials science are expected to accelerate the design, discovery, optimization, and interpretation of materials in both industry and academia, and this Review hopes to be the informative base camp for that journey.

How robust are modern graph neural network potentials in long and hot molecular dynamics simulations?

“…To bridge this gap between computational cost and accuracy, machine learned interatomic potentials have recently gained popularity in computational chemistry [8][9][10][11] and materials science [12][13][14]. In particular, a range of neural network [15][16][17] and kernel based potentials [18,19] have been developed and applied to a wide variety of chemical systems.…”

Machine learning for design principles for single atom catalysts towards electrochemical reactions

“…Once enough high quality databases are provided, a reliable ML model can be trained and constructed to address the electroreduction challenges. 148,149 ML in combination with DFT calculations commences a new direction for rapid and low cost rational design of SACs predicted to have optimal electroreduction catalytic activity. 150,151 For example, several studies have used ML to design single atom alloy catalysts (SAACs) with excellent stability and activity by predicting the E ads , DG, or aggregation energies.…”

Toward Multicomponent Single-Atom Catalysis for Efficient Electrochemical Energy Conversion

“…Furthermore, because of the availability of computer hardware and computational chemistry, machine learning (ML) has recently offered special shortcuts in the rational design and development of catalysts, including advanced SACs . For example, Kim et al recently predicted efficient electrocatalysts for the electrochemical NRR using a deep neural network .…”