Rational Design of Catalysts

Developing new and improved catalysts is crucial for enhancing energy efficiency, promoting environmental sustainability, optimizing resource utilization, enabling novel chemical transformations, boosting economic competitiveness, and advancing fundamental scientific understanding. Catalysts facilitate chemical reactions at lower temperatures, reducing energy input and greenhouse‑gas emissions across various industries. They enable cleaner processes, such as converting biomass to biofuels or selectively removing pollutants, while improving reaction selectivity and yield and minimizing waste generation. Novel catalysts unlock previously inaccessible chemical transformations, paving the way for new materials, pharmaceuticals, and high‑value‑added compounds. Well‑designed catalytic systems also bring competitive advantages for industries by cutting operating costs, elevating product yields, and supporting the development of high‑value products.

Exploring catalyst reaction mechanisms lays a fundamental foundation for optimizing existing catalysts and designing entirely new catalytic systems. In recent years, density functional theory (DFT) and complementary computational methods have been widely applied to unravel complex catalytic processes. Coupled with advances in computational chemistry algorithms, parallel computing, and high‑performance computing clusters, modern computational chemistry not only reveals the intrinsic nature of known catalytic pathways but also serves as a fast, low‑cost pre‑screening tool to guide rational catalyst discovery. Notably, the integration of artificial intelligence (AI) and machine‑learning (ML) techniques has greatly accelerated catalyst development. Machine‑learning models can learn from massive computational datasets and generate predictive descriptors to rapidly shortlist promising candidate catalysts, streamlining the resource‑intensive trial‑and‑error cycle of catalyst development. The synergy between computational chemistry and AI‑driven workflows reshapes catalyst exploration, allowing efficient navigation of high‑dimensional chemical space and faster identification of high‑performance catalysts for energy‑related applications.
Our current research efforts are centered on the following topics:

i. Mechanistic Studies of Light‑Alkane Dehydrogenation (Propane & Ethane):
To deepen the mechanistic understanding of saturated‑hydrocarbon dehydrogenation, we focus on propane and ethane dehydrogenation as model catalytic systems. Beyond conventional Pt‑based dehydrogenation catalysts, our scope extends to chemical‑looping oxidative dehydrogenation (CL‑ODH) and tandem catalytic strategies that break thermodynamic equilibrium limits via in‑situ hydrogen removal. We investigate catalytic behaviors on flat/stepped Pt and Pt‑alloy surfaces, as well as metal‑oxide interfaces (MgO, Al₂O₃, TiO₂, perovskites, carbonates). Model supports are selected for well‑defined lattice matching with Pt facets, which facilitates establishing systematic activity trends, Brønsted‑Evans‑Polanyi (BEP) correlations at metal‑support boundaries. By systematically tuning alloy compositions and support properties, we build reactivity descriptors, BEP and volcano‑type relationships linking catalytic performance to key material parameters. Insights are further extended to broader supported‑transition‑metal systems. Recent work highlights hydrogen‑spillover‑mediated tandem catalysis, e.g., CaCO₃‑mediated ethane dehydrogenation, and lattice‑oxygen regulation for chemical‑looping propane dehydrogenation toward low‑carbon olefin production under carbon‑neutral scenarios.

ii. Development of Global‑Optimization Algorithms for Catalyst‑Surface Modeling:
One major bottleneck for building realistic catalyst models is locating thermodynamically stable global‑minimum or low‑energy surface structures under actual reaction conditions. Though local‑minimization algorithms are mature, they cannot guarantee access to the true global minimum under complex operating environments, such as high adsorbate coverage or dynamic surface reconstruction. Complete sampling of the full potential‑energy‑surface requires numerous local geometry optimizations. Our group aims to improve the efficiency of global‑optimization workflows by combining genetic‑algorithm schemes and machine‑learning surrogate models, so that target low‑energy structures can be obtained with fewer local‑optimization cycles.

iii. Interpretable Machine‑Learning Descriptors for Electrocatalysis:
We develop physically interpretable descriptors for multi‑electron electrocatalytic reactions including CO₂ reduction, nitrogen reduction, and ethylene electro‑epoxidation. Instead of black‑box data fitting, our ML workflows extract intrinsic material properties to unify activity‑selectivity trends across multiple electrocatalytic transformations. These descriptor‑based models offer physical insights and guide the design of alloy, single‑atom, dual‑atom, and high‑entropy‑hydroxide electrocatalysts for carbon‑cycle‑related energy conversion.

iv. Automated Review Generation Driven by Large‑Language Models:
With explosive growth of scientific publications, researchers face heavy burdens in tracking state‑of‑the‑art progress and extracting core conclusions efficiently. To address this challenge, we build an automated‑review‑generation framework powered by large‑language models (LLMs). Leveraging advanced natural‑language‑processing capacities of LLMs, our tool streamlines literature‑review workflows and outputs concise, comprehensive, up‑to‑date summaries for different sub‑fields of catalysis. This tool is intended to boost the efficiency and quality of literature digestion for catalysis researchers and accelerate the discovery and optimization of new catalytic materials, with potential benefits for interdisciplinary collaboration and innovation across catalysis‑related disciplines.