Data-driven discovery of NOx suppression mechanisms in ammonia–methane combustion via ReaxFF and ensemble learning

• Integrates atomistic ReaxFF MD at elevated temperatures of 2000–3000 K. • Identifies radical pathways and bond restructuring responsible for NOx suppression. • Provides ML framework to extrapolate new blend ratios without costly simulations. • Accelerates discovery of cleaner combustion strategies and alternative fuels. Although ammonia–methane (NH 3 –CH 4 ) combustion is a viable low-carbon energy pathway, high NOx emissions remain a major barrier to practical deployment. In this study, reactive force field (ReaxFF) molecular dynamics (MD) simulations were carried out to investigate how ethanol and methanol additives influence the radical chemistry and NOx formation at 2000 K and 3000 K. Atomic-scale analysis was performed using mean squared displacement (MSD) and radial distribution functions (RDF). The results have revealed that alcohol additions alter species mobility and short-range ordering, weakening N–O correlations while enhancing H–O interactions. These effects promoted the radical redistribution and enabled NOx suppression, with the 10 % ethanol blend achieving the largest reduction (39.6 %) at 3000 K. To extend predictive capability beyond directly simulated cases, machine learning (ML) models were trained on MD-derived descriptors. Among the algorithms tested, Random Forest Regression (RFR) demonstrated the most reliable performance, accurately generalizing to untested alcohol ratios (2 %, 7 %, and 12 %) and reproducing consistent chemical trends. The integrated MD–ML framework shows how atomistic simulations and ensemble learning can together provide predictive insights into cleaner combustion regimes, offering a cost-effective tool for guiding additive selection and optimizing NOx control strategies.