Recently, Associate Professor Liang Zhang from the Center for Combustion Energy / School of Vehicle and Mobility, Tsinghua University and Professor Bin Cai from the Shandong University have made new progress in the study of machine-learning-driven design of high-entropy alloy electrocatalysts. The related work, entitled “Hierarchical Site-to-Composition Machine Learning for High-Entropy Electrocatalysts Design”, has been published online in Advanced Functional Materials.
High-entropy alloys feature highly disordered surface structures and a wide variety of local active sites, showing considerable promise for applications such as hydrogen production via water electrolysis. However, the numbers of possible elemental combinations and surface configurations increase combinatorially with the number of constituent elements, making it difficult for conventional approaches based on first-principles calculations or experimental trial and error to efficiently screen such an enormous materials space. To address this challenge, the research team developed a hierarchical site-to-composition machine-learning framework, termed HS2C-ML, which integrates atomic-scale activity prediction with composition-level materials screening.
At the site level, the team fine-tuned a pretrained interatomic potential through transfer learning, enabling rapid prediction of the adsorption free energies of the key hydrogen evolution intermediates *H and *OH on high-entropy alloy surfaces. By further combining two-dimensional kernel density analysis with an activity volcano model, the researchers statistically evaluated more than 1,000 surface sites for each material. The maximum-density centers predicted by the model showed excellent agreement with first-principles calculations, with correlation coefficients of 0.99 and 0.98 for the *H- and *OH-adsorption descriptors, respectively.
At the composition level, the team employed the SISSO method to construct interpretable composition–activity descriptors based on intrinsic elemental properties, including d-electron count, atomic radius, and electronegativity. These descriptors were then used to conduct high-throughput screening of approximately 16,000 five-component high-entropy alloy compositions, leading to the identification of a series of candidate compositions with promising activity for the alkaline hydrogen evolution reaction.
To validate the predictions, the research team synthesized a previously unreported CuZnPtRuRh high-entropy alloy nanocatalyst. In 1 mol/L KOH electrolyte, the catalyst required an overpotential of only 26 mV to reach a current density of 10 mA cm-2, substantially lower than the 96 mV required for commercial Pt/C. It also exhibited a Tafel slope of 60 mV dec-1 and maintained stable operation for 100 h. The close agreement between the experimental results and theoretical predictions demonstrates the reliability of the framework for the design of complex multicomponent catalytic materials. This study provides a new strategy for the cross-scale, interpretable, and high-throughput design of complex materials such as high-entropy alloys, and may be extended to other electrocatalytic reactions and multicomponent functional material systems.
Figure: Overview of the HS2C-ML framework for HEA catalyst discovery.
2022 Ph.D. student Shiyu Zhen from the Center for Combustion Energy / School of Vehicle and Mobility, Tsinghua University and Ph.D. student Lingwei Wang from the Shandong University contributed equally as co-first authors of this paper. Associate Professor Liang Zhang from the Center for Combustion Energy / School of Vehicle and Mobility, Tsinghua University and Professor Bin Cai from the Shandong University served as co-corresponding authors. This work was supported by the National Natural Science Foundation of China and other funding programs.
Provided by: Liang Zhang's Group
Approved by: Yu Cheng Liu, Xiaoqing You
