Recent advances in the application of machine-learning algorithms to predict adsorption energies

Abstract

Adsorption energies on the surface sites of heterogeneous catalysts, together with the Sabatier volcano plot correlating them with the reaction activation barrier (Δ<i>G_a</i>) along the catalytic reaction pathways through Brønsted−Evans−Polanyi (BEP) relations, determine the catalytic activity. This review categorizes machine-learning (ML) models into on-lattice and off-lattice models, discusses the different approaches to build models predicting adsorption energies, and summarizes several recent advances in the use of ML algorithms. These developed ML models enable researchers to rationally design high-performance heterogeneous catalysts by identifying active surface features. Furthermore, this review concludes with the limitations of current models and the challenges needing to be addressed to build sophisticated models that are more consistent with the real-life operating conditions of catalytic reactions without sacrificing speed.