Machine Learning on a Synergistic Transition Metal Dual-Atom Surface for Efficient Decomposition of Ammonia.

Abstract

The technology of H₂ production through NH₃ decomposition is of great importance for finding clean energy alternatives to fossil fuels. Here, a framework for screening dual-atom catalysis by integrating machine learning (ML) with high-throughput (HT) calculations to predict the catalytic performance of dual-atom systems for NH₃ decomposition is designed. First-principles-based HT calculations are conducted on 62 randomly selected systems for intermediate steps. Then, feature engineering is employed to obtain features with high importance and low correlation. The results of the HT calculations are subsequently used as a training set to train the ML model. The well-trained model is subsequently used to predict the catalytic performance of 2187 structures. Several potentially good dual-atom catalysts (RuMo─O─C, ScOs─N─C, and OsV─N─C) for NH₃ decomposition are obtained. Finally, the density of states and differential charge analysis show the presence of the synergistic catalytic process in these dual-atom catalysts.