Recombination of NH2 with alkyl radicals: VRC-TST rate constants from neural network potentials

Abstract

The recombination between CH₃ and NH₂ is an important reference reaction for describing the formation of chemical bonds between hydrocarbons and nitrogen compounds in combustion. This is for example the case when ammonia is burned together with hydrocarbon mixtures. Despite the important role played by this reaction in combustion processes, the theoretical studies on the accurate determination of its rate constant, or on the pressure dependence, are limited. At present, most existing kinetic mechanism use experimental measures performed at room temperatures, or detailed balance and the rate constants measured for the reverse process at high temperatures, thus in conditions in which the reaction rate is pressure dependent. This places some limits on the ability to accurately describe the reactivity of two key radical species: methyl and NH₂, in particular when this reaction pathway is in competition with others. The present work aims at filling this gap, providing ab-initio rate constant estimations for the recombination pathway of the reaction family C_nH_2n+1 + NH₂, with n = 1, 2, 3. Rate constants are estimated using Variable Reaction Coordinate – Transition State Theory (VRC-TST) and machine learning. VRC-TST is the golden standard for kinetic studies of barrierless reactions, which do not have a well-defined transition state. The rate constants estimated with VRC-TST approach the experimental accuracy, at the cost of a computationally demanding Monte Carlo sampling of the reactive Potential Energy Surface (PES). In this work we use Artificial Neural Network (ANN) to learn the portion of the multidimensional PES relevant to the reaction of interest as a function of the degrees of freedom describing the relative orientation of the two reacting fragments. The physics-informed ANN architecture significantly reduces the number of explicit electronic structure calculations needed by VRC-TST, thus gaining significant time savings without compromising accuracy. The calculated rate constants are in good agreement with the available experimental data and are thus expected to provide a useful reference for the kinetic modelling of the co-combustion of nitrogen compounds and hydrocarbons.