Research
Understanding Catalysts through Structure, Descriptors, and Data
Our research combines computational chemistry and artificial intelligence to understand how catalysts function and to develop principles for their rational design. We connect atomic-scale structures with catalytic activity, selectivity, and stability, while developing methods to extract useful knowledge from computational results and scientific literature. Our current research focuses on three closely connected directions.
1. Dynamic Catalyst Structures under Reaction Conditions
催化剂动态结构
Catalysts can undergo substantial structural changes during reactions. Identifying the structures and active sites relevant to working conditions is therefore central to understanding catalytic performance.
We combine density functional theory, machine-learning interatomic potentials, and global structure searches to investigate surface reconstruction, defect formation, changes in coordination environments, and metal–support interactions. We seek to understand how these structural changes influence reaction pathways and catalytic behavior.
A particular focus is the relationship between structural diversity and catalytic function. Our recent work on γ-Al₂O₃ shows how an ensemble of defect-rich configurations can provide a more realistic description of catalyst supports than a single idealized structure. Related studies examine how evolving oxygen vacancies regulate oxide catalysts during chemical-looping reactions.
γ-Al₂O₃ structural ensembles, 2026; oxygen-vacancy regulation, 2025
2. Physically Interpretable Descriptors for Catalyst Design
物理可解释描述符与催化剂设计
We develop descriptors that connect atomic geometry, electronic structure, and chemical bonding with catalytic activity and selectivity. By combining mechanistic understanding with physically informed feature engineering and interpretable machine learning, we aim to identify simple relationships that explain catalytic trends and guide material selection.
Representative work includes the degree-of-isolation descriptor for single-site alloy catalysts and interpretable descriptor models connecting the atomic properties and local environments of dual-atom sites with multiple electrocatalytic reactions. Building on these studies, we seek to extend descriptor-based design to more complex and structurally evolving catalytic systems, with particular attention to physical meaning and transferability.
Degree-of-isolation descriptor, 2023; interpretable dual-atom descriptors, 2024
3. AI-Assisted Knowledge Discovery and Data Mining
AI 辅助知识发现与数据挖掘
Scientific literature contains extensive information about catalysts, reaction conditions, and catalytic performance, but much of this knowledge remains difficult to compare and reuse systematically.
We develop AI-assisted approaches to literature retrieval, information extraction, evidence organization, and data mining. Our published framework integrates literature search, topic formulation, knowledge extraction, and review composition, together with a data-mining module. Traceability and quality assessment are central to this approach.
Automated literature research and review generation, 2025
Beyond review generation, we aim to connect literature-derived knowledge with computational datasets to identify relationships among catalyst composition, structure, reaction conditions, and performance. The broader goal is to support research-question formulation, catalyst screening, and the discovery of testable design hypotheses.