Predicting Catalytic Activity for CH4 Combustion on Pd-Exchanged Zeolite Catalysts Using Automated Reaction Route Mapping

While computational predictions of catalytic activity are highly desired, conventional methods have difficulty capturing the complexity of reactions on solid catalyst surfaces. To address this issue, we employed a novel approach combining neural network potentials (NNPs) with automated reaction route mapping to explore reaction mechanisms of CH₄ combustion on Pd-exchanged zeolites (Pd-CHA, Pd-beta, and Pd-MOR). The predicted reaction map of CH₄ combustion over Pd²⁺ site revealed partially oxidized species such as CH₂O, HCOOH, and bicarbonate as the potential intermediates toward CO₂ + 2H₂O. Activation energies (E_a) of the rate-determining step (RDS) were evaluated, revealing the order of E_a as Pd-MOR < Pd-beta < Pd-CHA. A kinetic analysis using rate constant matrix contraction (RCMC) method estimated the catalytic activities of these catalysts. No reaction intermediates with significant lifetimes were observed on Pd-beta and Pd-MOR, whereas stable bicarbonate intermediates were present on Pd-CHA, decreasing the formation rate of CO₂ + 2H₂O. Kinetic analysis further predicted the pseudo CO₂ formation rate with activity order Pd-MOR > Pd-beta > Pd-CHA, aligning with experimental results. These findings demonstrate the potential of the automated reaction route mapping with NNP for predicting the catalytic activity of solid catalysts, enabling their efficient prescreening and rational design.