image: From DFT calculation to ML prediction, the potential catalysts with highly active and selective performance are efficiently screened by four ML models, i.e. decision tree, random forest, support vector machine, and XGBoost classification, where ten-fold cross-validation is employed to reduce overfitting risks during model training.
Credit: Nano Research, Tsinghua University Press
Ammonia, as a crucial chemical raw material, holds irreplaceable strategic value in fields such as agriculture, pharmaceuticals, chemical industry, and national defense. The Haber-Bosch process for industrial production of ammonia requires harsh conditions (400-500 °C, 15-25 MPa) and will release amounts of carbon dioxide. The electrochemical nitrogen reduction reaction (eNRR) utilizes renewable energy to drive reactions under ambient conditions, which is regarded as the most promising green alternative for ammonia synthesis. Considering the metal type, coordination configuration and substrate combination, the catalysts varieties are enormous, hence to efficiently screen the potential candidates is challenging by traditional methods.
A groundbreaking study published in Nano Research by a collaborative team from Shaanxi Normal University and Xi’an Jiaotong University has made significant strides in addressing this challenge. The research team combined density functional theory (DFT) and machine learning (ML) to identify high-performance single-atom catalysts (SACs) for eNRR. They constructed a library of 196 TM@SxNy SACs and employed four classification models, where the XGBoost model is proved to possess the best performance with a remarkable 95% prediction accuracy in screening optimal catalysts. This ML-driven approach dramatically accelerated the discovery process, surpassing traditional trial-and-error methods.
The study screened 17 SACs with exceptional catalytic performance, where Mo@S₃N₁ and W@S₃N₁ exhibited remarkably low limiting potentials of -0.26 V and -0.25 V, respectively, under implicit solvation conditions. These catalysts demonstrated high activity and superior selectivity, effectively minimizing HER interference. Prof. Yuhong Huang emphasized the transformative potential of ML in catalyst screening: " The conventional methods of identifying high-performance catalysts are time-consuming and our approach can accelerate this process to a great extent, which is exciting and inspiring."
The researchers also identified critical descriptors for catalytic performance, such as the N≡N bond length and the number of outermost d electrons (Nd), which play pivotal roles in nitrogen activation and reduction. Through SHapley Additive exPlanations (SHAP) analysis, the team gained deeper insights into how these descriptors influence the catalytic activity. Additionally, the relationship between N≡N bond length, Nd, intermediate adsorption energy, and free energy changes in key reaction steps indicated that moderate nitrogen activation is more favorable for the NRR process. For the first time, a new descriptor φ was proposed to evaluate the catalytic performance of SACs, and the variation of φ with free energy in two hydrogenation steps displayed a "volcano plot," with qualified catalysts tending to appear on the slopes. Prof. Fei Ma highlighted the synergy between ML and DFT calculations: "This combination provides a robust framework for rational catalyst design, advancing both electrocatalysis and sustainable chemical production."
The implications of this study extend far beyond eNRR. The ML-driven methodology can be adapted to other catalytic processes, such as CO₂ reduction and water splitting, where efficient catalysts are equally crucial. By offering a blueprint for rational catalyst design, this research paves the way for innovations in green chemistry and renewable energy technologies, ultimately contributing to a more sustainable future.
Other contributors include XiuMei Wei and Haiping Lin from the School of Physics and Information Technology at Shaanxi Normal University.
This work was supported by National Natural Science Foundation of China (No. 52271136, No. 22373063), the Natural Science Foundation of Shaanxi Province in China (No. 2021JC-06, No. 2019TD-020), Fundamental Research Funds for the Central Universities of China (GK202203002).
About the Authors
Lintao Xu is currently a master's degree candidate at Shaanxi Normal University. His main research direction is the study of machine learning-assisted electrochemical ammonia synthesis. He has published a scientific research paper as the first author in Journal of Materials Chemistry A.
About Nano Research
Nano Research is a peer-reviewed, open access, international and interdisciplinary research journal, sponsored by Tsinghua University and the Chinese Chemical Society, published by Tsinghua University Press on the platform SciOpen. It publishes original high-quality research and significant review articles on all aspects of nanoscience and nanotechnology, ranging from basic aspects of the science of nanoscale materials to practical applications of such materials. After 17 years of development, it has become one of the most influential academic journals in the nano field. Nano Research has published more than 1,000 papers every year from 2022, with its cumulative count surpassing 7,000 articles. In 2023 InCites Journal Citation Reports, its 2023 IF is 9.6 (9.0, 5 years), and it continues to be the Q1 area among the four subject classifications. Nano Research Award, established by Nano Research together with TUP and Springer Nature in 2013, and Nano Research Young Innovators (NR45) Awards, established by Nano Research in 2018, have become international academic awards with global influence.
Journal
Nano Research
Article Title
Machine learning driven high-throughput screening of S and N-coordinated SACs for eNRR
Article Publication Date
20-Mar-2025