Rise of machine learning potentials in heterogeneous catalysis: Developments, applications, and prospects

The urgency of tackling climate change is driving a global shift towards renewable sources of energy, with a growing contribution from alternative energy sources such as solar, wind and hydroelectric power. With the global push for the sustainable energy, the demand for effective catalysts for sustainable chemical production and energy storage has been rapidly increasing. Computational simulations have contributed to the rational design of catalysts by allowing profound analysis of catalyst properties. Machine learning potential (MLP) has emerged as a potential tool to bridge the gap between quantum mechanical accuracy and computational efficiency, overcoming the computational cost limitations of quantum chemistry-based simulations. This review discusses the development and application of MLP in multiscale simulations of heterogeneous catalysis. It covers the basic concepts of computational catalysis, the construction of MLP focusing on efficient datasets, atomic structure representations, and the process of training and evaluating of ML models. Furthermore, the potential applications of MLP are discussed in addressing computational challenges within the field, as MLP has potential to overcome limitations in simulation time and length scale. Lastly, the prospects for MLP are presented, taking advantage of the rapid advancements in artificial intelligence architectures. It is expected that the integration of MLP will accelerate progress within the catalyst research community and will bridge the gap between theoretical and experimental approaches in catalytic research.