Machine Learning-Assisted Screening of Stepped Alloy Surfaces for C1 Catalysis

Efficient reactivity assessment for stepped alloy surfaces presents a major challenge in designing effective catalysts for industrial catalytic conversion of single-carbon (C1) molecules. In this work, we propose a machine learning (ML)-assisted approach to screen active and stable binary alloys for various C1 catalytic processes. Leveraging only non-ab initio, simple bulk material properties as input features, the ML models exhibit impressive accuracy for predicting site-specific adsorption energies of atomic carbon and oxygen, which enable not only fast navigation through abundant material space but also extraction of explicable physical insights. The effectiveness of the ML models is further validated by applying their predictions to catalyst screening for common reactions in C1 catalysis, as well as a detailed kinetic study on one example candidate, Cu3Pd. This data-driven approach with fully interpretable physical features demonstrates the possibility of unearthing underlying catalyst design principles from apparent data and paves the road for the discovery of desirable alloy catalysts.

阶梯状合金表面的反应活性高效评估，是开发用于单碳（C1）分子工业催化转化的高效催化剂所面临的核心挑战。本研究提出一种机器学习（ML）辅助策略，用于针对各类C1催化过程筛选兼具催化活性与结构稳定性的二元合金。该策略仅以非从头算（non-ab initio）的简单块体材料属性作为输入特征，所构建的机器学习模型在预测原子碳与氧的位点特异性吸附能方面展现出优异精度，这不仅能够快速遍历海量候选材料空间，还可提取出可阐释的物理机理认知。通过将模型预测结果应用于C1催化常见反应的催化剂筛选，以及针对一例候选材料Cu3Pd开展的详细动力学研究，进一步验证了该机器学习模型的有效性。这种采用完全可解释物理特征的数据驱动研究方法，证明了从表观数据中挖掘催化剂设计内在原理的可行性，并为高性能合金催化剂的发现铺平了道路。