A hybrid machine learning and DFT-informed microkinetic model for ammonia decomposition on Ru/Al2O3

Ammonia (NH 3 ) decomposition is a key reaction for hydrogen storage and transport applications, and its mechanistic understanding is critical for catalyst and reactor design. Microkinetic modeling enables molecular-level insight into reaction networks; however, models built purely on first-principles methods such as density functional theory (DFT) often suffer from quantitative inaccuracies due to the exponential sensitivity of rate expressions to small errors in activation energies. Consequently, ab initio models are typically limited to qualitative predictions. In this work, we present a hybrid modeling framework that combines DFT-informed kinetic parameters with a neural network-enhanced global optimization routine to refine a microkinetic model for ammonia decomposition over a commercial catalyst. Literature DFT values serve as initial estimates, while high-fidelity experimental data from a microreactor setup guide the fitting. Neural networks are trained as surrogates of the microkinetic model to accelerate convergence and efficiently explore the high-dimensional parameter space. The resulting model identifies bond scission as the rate-determining step, with scission and N 2 association as additional slow steps. Surface coverage analysis reveals NH 2 ∗ and H ∗ as the most abundant reaction intermediates, while N ∗ remains low. Thermodynamic consistency is enforced via partition function analysis, yielding excellent agreement with experimental enthalpies and entropies. We further extend the model to a non-interacting dual-site description, showing that terrace sites contribute negligibly under the studied conditions. This work highlights the power of integrating mechanistic microkinetic models with data-driven optimization to overcome the limitations of purely first-principles approaches. The proposed framework offers a scalable and accurate strategy for guiding catalyst design and mechanistic analysis in ammonia decomposition and related catalytic systems. • Developed a DFT-informed microkinetic model for NH 3 decomposition on Ru/Al 2 O 3 . • Integrated neural networks for accelerated global optimization of kinetic parameters. • Identified NH–H scission as the rate-determining step via sensitivity analysis. • Ensured thermodynamic consistency using partition functions and experimental data. • Demonstrated low terrace-site activity using a dual-site kinetic model extension.