Reactive Machine Learning Potential Models for the NO Formation Reaction

Machine learning potentials provide a method for accurate energy assessment at affordable computational cost. In this work, we produce machine learning potentials for CO₂, O₂, N₂, and NO, with the goal of modeling reactive equilibria between the latter three species.

Potential energy surfaces are sampled via molecular dynamics trajectories and extracted geometries are evaluated using DFT. Neural networks are then trained on datasets of energies, producing machine learning potential models capable of capturing disperse interactions. These models are then retrained on CCSD(T) and CASSCF derived energies to produce wavefunction level models transferable to ensembles of molecules. These models are used in Monte Carlo simulations to predict vapor-liquid equilibria and reactive equilibria and to predict second virial coefficients.

Current results include models for CO₂ and N₂ that reproduce DFT assessment of disperse intermolecular interactions to within 0.1kcal/mol. Future work includes producing potentials for open shell species, retraining potentials with wavefunction data, and including electrostatics.