Quasi-classical Trajectoryanalysis Of Typical Reactions With Or Without Barriers Based On Unsupervised Learning Method

The calculation of quasi-classical trajectory（QCT）is an important tool for studying the dynamics of chemical reactions.Trajectory contains the information of the whole reaction process from reactants to products.Analysis of trajectory characteristics is of great significance to identify and understand the reaction mechanism.QCT studies usually require a large number of trajectories to reduce statistical error.For a large number of trajectories,it is difficult to identify different types of reaction paths,find"typical"reaction processes and classify them.The quantitative analysis is even more difficult.The trajectories of polyatomic molecular reactions are represented by data points in high-dimensional coordinate space.If these high-dimensional data are converted to low-dimensional space and their distribution rules are maintained,it will provide very useful help for trajectory analysis.The combination of several unsupervised learning methods in the field of machine learning provides a feasible solution for data processing of reaction trajectory.In this thesis,two typical small molecule reactions are chosen to explore the advantages of different machine learning methods in the analysis of quasi-classical trajectories,providing useful references for the further development of efficient orbital analysis schemes.This thesis is divided into three chapters.Chapter 1 is the introduction,which mainly introduces the characteristics of small molecule collision reaction,the background and model of reaction dynamics theory studies,the quasi-classical trajectory method and related machine learning methods including isometric feature mapping,k-means,dynamic time warping,etc.In chapter 2,taking the barrierless reaction between Be⁺and H₂O as an example,we analyze the quasi-classical trajectories by using isometric mapping and dynamic time warping method.The microscopic mechanism of the reaction is well revealed,where the existence of deep potential well and submerged energy barrier play a decisive role in the reaction dynamics.There is a potential well as deep as 2.78e V in the reaction,and all the reaction paths go through this potential well to form the Be-OH₂⁺complex before continuing to move towards the products.No direct reaction process is found.From the potential well to the product,it is necessary to go over a potential barrier.Although this is a submerged barrier whose energy is lower than the reactant limit,the barrier height is up to 2e V relative to the potential well,which has an important influence on the reaction probability.A preliminary study of the isotopic effect of H₂O molecule in the reaction shows that the relative height of the barrier is the determining factor of the reactivity.In addition,the relative velocity of the two reactants has the greatest influence on the ratio of the unreacted/reacted trajectories which trapped in the potential well.Therefore,although there is no reaction barrier between Be⁺and H₂O,the domination of the reaction is not the long-range interaction,and the traditional capture model is not suitable for this reaction.Chapter 3 takes the reaction of HCl with OH as an example and carefully analyzes the reaction mechanism when the reaction has typical reaction energy barrier.Although the transition state corresponding to the barrier position dominates the reaction,the analysis of the trajectories shows that the reaction is generally divided into two types.One is the direct process of overturning a potential barrier and the other is the indirect process of forming a complex in a shallow potential well before overturning the barrier.Using data in different coordinate systems as input for the dimensionality reduction calculations,the trajectories are classified consistently,but the features differ slightly in detail,causing different effects for further similarity analysis.The selection of suitable trajectories as a criterion,aided by the fitting of already trend lines,can be better used for quantitative analysis and discussion.