A High-Fidelity Reactive Force Field ReaxFFCHON-S25 for C/H/O/N Systems via Successive Partition Optimization: Unraveling Mechanistic Insights into Hydrocarbon and Ammonia Combustion
Qi He, Qing-qing Wang, Bin Xiao, Xudong He, William A. Goddard, Yi LiuAbstract
Accurate reactive molecular dynamics (RMD) simulations of hydrocarbon and ammonia combustion are often hindered by obsolete quantum mechanical (QM) training data and the large computational cost of conventional ReaxFF parametrization. To address these challenges, we develop ReaxFFCHON-S25, a high-fidelity reactive force field for C/H/O/N systems, parametrized via Successive Partition Optimization (SuPO), an efficient and scalable strategy for force field refinement. Parameters were trained against a comprehensive QM data set computed at the M06–2X/6–311++G** level within the meta-GGA framework. The reference library comprises 30,267 configurations (>29,256 data points) spanning 356 representative systems (C0–C8, including nitrogenous species). The training subset (2570 configurations, >2497 data points) covers 171 systems and incorporates exhaustive scans of bond lengths, angles, torsions, transition states, and elementary reactions. The benchmarking reveals that ReaxFFCHON-S25 significantly outperforms the legacy ReaxFF-lg models, reducing RMSEs by 21.0% (C/H/O) and 13.1% (C/H/O/N) for the C0–C4 training sets, and by 34.4%–46.2% for the out-of-sample C5–C8 test sets. The RMD simulations further validate the force field’s robustness: the simulated apparent activation energy for ammonia oxidation aligns closely with the reported experimental measurements, and the resolved reaction network accurately captures critical intermediates (e.g., NH2, HNO) and pathways consistent with both the experimental observations and high-level theoretical benchmarks. Crucially, ReaxFFCHON-S25 elucidates the mechanistic intricacies of NOx formation and elimination, establishing a powerful computational microscope for unraveling nitrogen combustion chemistry. This work provides a validated computational tool for gas-phase hydrocarbon and ammonia combustion, offering a foundation for future extension to more complex carbon-neutral fuel systems.