高等学校化学学报 ›› 2026, Vol. 47 ›› Issue (2): 20250266.doi: 10.7503/cjcu20250266

• 物理化学 • 上一篇    下一篇

催化电子捐赠的机器学习描述符: 预测氮固定中的吸附能和极限电位

赵迎1(), 杨海迪1,2, 柴玉春1,2, 高帅帅1, 原鹏飞1, 陈雪波1,3()   

  1. 1.烟台先进材料与绿色制造山东省实验室, 烟台 264000
    2.哈尔滨工程大学烟台研究院, 烟台 264000
    3.北京师范大学化学学院, 北京 100091
  • 收稿日期:2025-09-17 出版日期:2026-02-10 发布日期:2025-10-31
  • 通讯作者: 赵迎,陈雪波 E-mail:yingz@amgm.ac.cn;xuebochen@bnu.edu.cn
  • 基金资助:
    国家自然科学基金(22503078);山东省重点研发计划项目-竞争性创新平台(2024CXPT036);山东省泰山学者工程项目(tstp20240844)

Machine Learning Descriptors for Catalytic Electron Donation: Predicting Adsorption Energies and Limiting Potentials in Nitrogen Fixation

ZHAO Ying1(), YANG Haidi1,2, CHAI Yuchun1,2, GAO Shuaishuai1, YUAN Pengfei1, CHEN Xuebo1,3()   

  1. 1.Shandong Laboratory of Advanced Materials and Green Manufacturing at Yantai,Yantai 264000,China
    2.Yantai Research Institute,Harbin Engineering University,Yantai 264000,China
    3.College of Chemistry,Beijing Normal University,Beijing 100091,China
  • Received:2025-09-17 Online:2026-02-10 Published:2025-10-31
  • Contact: ZHAO Ying, CHEN Xuebo E-mail:yingz@amgm.ac.cn;xuebochen@bnu.edu.cn
  • Supported by:
    the National Natural Science Foundation of China(22503078);the Key Research and Development Program of Shandong Province-Competitive Innovation Platforms, China(2024CXPT036);the Program of Taishan Scholars of Shandong Province, China(tstp20240844)

摘要:

通过高通量密度泛函计算筛选出一系列具有氮还原反应(NRR)活性的B和双金属原子组成的CN-B@M2催化剂. CN-B@Fe2, CN-B@Tc2, CN-B@Os2和CN-B@Re2被认为是具有良好选择性和NRR活性的催化剂, 其极限电位(UL)分别为-0.24, -0.34, -0.31和-0.38 V. 计算结果表明, N2在B@M2上的吸附呈周期性演变, 吸附构型和能量受d带中心调节. UL随转移电荷呈火山型分布. 具有中等电子给体能力(中等电荷转移)的B@M2催化剂表现出优异的NRR活性. 通过量化催化剂的原子电子特性和拓扑结构, 构建了用于描述给电子能力的描述 符Φ. 结果表明, 给电子能力与氮还原反应的极限电位呈火山关系. 使用描述符Φ和催化剂的内在特性作为特征预测了吸附能和极限电位, 由于R2值为0.99, 梯度提升回归(GBR)被认为是构建机器学习预测模型的最恰当方法.

关键词: 氮还原, 高通量计算, 机器学习, 供电子能力描述符

Abstract:

In this work, a series of CN-B@M2 catalysts composed of B and bimetallic atoms with nitrogen reduction reaction(NRR) activity were screened by high-throughput density functional calculations. CN-B@Fe2, CN-B@Tc2, CN-B@Os2, and CN-B@Re2 were considered as catalysts with good selectivity and NRR activity, with the limiting potentials(UL) of -0.24, -0.34, -0.31 and -0.38 V, respectively. Calculation results show that the adsorption configuration of N2 at B@M2 shows a periodic evolution, and adsorption configuration and energy are regulated by d-band center. UL shows a volcanic distribution with adsorbed N2 charge. B@M2 catalyst with moderate electron donor capacity shows excellent NRR activity. Descriptor Φ used to describe electron donating ability is constructed by quantifying atom electronic properties and topology structure of catalysts. Φ shows a strong linear correlation with adsorption energy, and describes limiting potential of NRR by volcano diagram. Φ and intrinsic properties of catalyst are used as features to predict the adsorption energy and UL. Gradient boosting regression(GBR) is considered the most appropriate method for building a machine learning prediction model due to an R2 of 0.99. This work provides novel insights into the design of rational and efficient NRR catalysts and construction of their descriptors.

Key words: Nitrogen reduction, High-throughput computing, Machine learning, Descriptor of supply electrons ability

中图分类号: 

TrendMD: