Machine Learning Accelerated Density Functional Theory Approach for Predicting Activation Barriers and Reaction Pathways in Methane-to-C ₂ Conversion

Computational advancements have accelerated density functional theory (DFT)-based catalyst design for CH4 conversion to value-added products. While the C1 reaction network study is computationally less demanding, exploring the C2 network comprising nearly 250 elementary reactions is challenging. To address this, we developed two machine learning (ML) models to predict activation energies correlating six and 37 descriptors, respectively. Among the models developed, gradient boost regression (GBR) with 37 descriptors exhibited the lowest mean absolute error (0.177 eV), predicting accurate C2 barriers. Leveraging this, we adopted a combined strategy: rate-determining step identification by GBR, followed by DFT validation. Applying this approach to oxidative coupling of methane on Ni(111) and NiB surfaces revealed that boron doping significantly lowers the CH2* dimerization barrier from 0.84 eV on Ni(111) to 0.49 eV, forming a C2 intermediate. The resulting ethylene subsequently hydrogenates to ethane while the integrated ML-DFT approach confirmed that competing desorption, hydroxylation, and oxidation exhibited higher barriers.