Quasi-classical Trajectory Analysis Of Multichannel Reaction N⁺+H₂ Based On Dimensionality Reduction Approaches

As an important method for molecular dynamics research,the quasi-classical trajectory method has been widely used in the theoretical research of ion-molecular reaction dynamics due to its low computational cost and good accuracy.In order to reduce statistical errors,a large number of trajectories are usually calculated in the study of quasi-classical trajectories.For the analysis of massive trajectories,efficient and reasonable data processing methods are required.In recent years,machine learning has been applied in many fields such as medicine,image recognition and autonomous driving.In studies related to chemistry,machine learning（ML）has also made rapid developments in quantum chemical calculations,potential energy surface fitting,and the design of new materials.In this work,we apply the unsupervised machine learning method to the theoretical study of chemical reaction dynamics.Taking the multi-channel reaction NH⁺+H₂ as an example,the reaction trajectory simulated by the classical trajectory method were quantitative analyzed,and the reaction mechanism and reactivity characteristics were discussed.This dissertation is composed of three chapters.The first chapter introduces the background and significance of this study,the theories and models involved in the work of ion-molecule reaction dynamics,and related machine learning methods.In Chapter2,based on the newly constructed NH⁺+H₂ reaction potential energy surface,two manifold learning methods isometric mapping（ISOMAP）and local linear embedding（LLE）were used to analyze the quasi-classical trajectory of multi-channel reaction NH⁺+H₂.The results show that the isometric feature mapping can clearly identify the different reaction path trajectories,and the local linear embedding can distinguish the non-reactive trajectory.Both of the two methods are conducive to quantitative analysis.Through trajectory analysis,the competition between different reaction channels can be attributed to different capture modes.For the NH⁺+H₂→N+N′H₂⁺reaction,at a lower collision energy,both the complex-forming mechanism and the direct extraction mechanism are significant,while the direct extraction mechanism is dominant at higher collision energy.However,NH′⁺+H₂→H+N′NH⁺responds to the rebound mechanism in a certain collision energy range.In total,through the analysis of the two manifold learning methods,the competitive relationship between the two reactions and the details of the reaction mechanism are well explained.In chapter 3,the initial state selected quasi-classical trajectory method was used to study the effects of the vibration and rotation excitation of the reactants in the NH⁺+H₂ reaction on different product channels,and the isometric feature mapping（ISOMAP）and k-means method were performed to analyze the trajectory.In the process of accessing to the products,two reactions,in which both NH and HH bonds are broken（NH′⁺+H₂→NH′⁺+HH′and NH′⁺+H₂→H′₊HNH′⁺）are promoted by all the excitation modes,but inhibit the formation of N+H₃⁺.However,for the reaction R3where the product is NH₂⁺+H,the vibration and rotational excitation of H₂ can enhance the reactivity.On the other hand,the vibration of the reactant NH inhibits the reaction,but the rotation excitation has a significant promotion effect on the reaction.R3 has two reaction paths,and the trajectory analysis shows that the effect of H₂ motions on the two mechanisms is the same,but the different effects of NH ro-vibration on the mechanism make the reactivity more complicated.For such ion-molecular reaction with multiple reaction potential wells,the mode specificity is not clear,and more detailed analysis is needed to obtain reasonable explanations.Machine learning method is a potentially powerful tool to analyze the reaction trajectory.