Prof. Jinlan Wang's Team At The School Of Physics, Southeast University, Make New Progress In Mechanism-Guided Intelligent Discovery Of Catalysts For Methane C–H Activation

Recently, the research team led by Prof. Jinlan Wang, Assoc. Prof. Qiang Li, and Assoc. Prof. Qionghua Zhou from the School of Physics, Southeast University, has made new progress in data-driven catalyst design. The related work, entitled “Mechanism-Guided Catalyst Discovery for Methane C−H Activation via Structure-Aware Multisource Transfer Learning,” was published in the internationally leading chemistry journal Journal of the American Chemical Society.

Artificial intelligence(AI) is accelerating the transition of catalyst discovery from experience-based trial and error to high-throughput intelligent screening. However, existing AI-assisted catalyst design methods mostly rely on readily available thermodynamic descriptors, such as adsorption energies and reaction energies. In contrast, the kinetic activation barriers that truly determine catalytic rates usually require costly transition-state searches, resulting in a severe scarcity of large-scale kinetic datasets. Therefore, achieving fast and accurate prediction of reaction barriers in small kinetic datasets remains a key challenge for realistic catalytic performance prediction and intelligent catalyst design.

To address this challenge, the team took methane C–H bond activation as a representative reaction and proposed a mechanism-guided, structure-aware multisource transfer learning framework, as shown in Figure 1. This framework encodes the structural and elemental information of the local coordination environment around active sites and integrates multisource thermodynamic information related to the reaction mechanism, enabling accurate prediction of C–H activation barriers with limited high-precision kinetic data. The optimal multisource model achieved an R² of 0.89 and an MAE of 0.15 eV on the test set, approaching the accuracy of DFT calculations. It can also be generalized to different alloy compositions, crystal facets, and single-atom alloy systems, demonstrating strong transferability. Further analysis revealed that the activation barrier of methane C–H activation is not determined by a single adsorption energy, but is synergistically regulated by the reactivity of O species, the ability to form O–H bonds, and the stability of methyl intermediates, providing multidimensional optimization criteria for the rational design of complex alloy catalysts.

In addition, the team proposed a composite kinetic descriptor combining temperature and activation barrier, establishing a quantitative link between microscopicallypredicted C–H activation barriers and experimentally measured macroscopic methane oxidation rates. Based on this descriptor, several potentially highly active candidates were identified, including AgOs₃, AuPd₃, AgPd₃, Ag₃Os, and FePd₃. Among them, AuPd₃ has been experimentally reported to exhibit excellent performance in selective methane oxidation, validating the reliability of the proposed approach. This work establishes a new kinetics-oriented paradigm for intelligent catalyst discovery under realistic reaction conditions.

The first authors of the paper are Ph.D. students Wangqiang Lin and Huiyang Zhang from Southeast University. Prof. Jinlan Wang, Assoc. Prof. Qiang Li, and Assoc. Prof. Qionghua Zhou from the School of Physics, Southeast University, are the corresponding authors. This work was supported by the National Key Research and Development Program of China, the National Natural Science Foundation of China, and the Basic ResearchProgram of Jiangsu Province.