Potential Energy Surface And Dynamics Of Polyatomic Molecular Reaction

Molecular reaction dynamics can provide a direct understanding of chemical reactions at the atomic and molecular level,which are helpful in determining the microscopic mechanism of the reaction.The potential energy surface is the premise of molecular dynamics computation.Based on the potential energy surface,dynamic results such as the integral cross section,differential cross section,product energy partitioning,vibrational state distributions of product,and thermal rate coefficients of the reaction can be calculated accurately.In the past few decades,many high-dimensional potential energy surfaces(PES)have been fitted based on the neural network(NN)method,making accurate dynamic calculations possible.But this method has certain random errors and the risk of overfitting.The risk has a impact on the dynamics results,especially for systems with strong long-range interactions.This project proposes a NN-CPDA(Neural Network-Combined with Partial Derivative Approximation)fitting method based on neural NN method that combines energy and partial energy derivatives to simultaneously approximate the energy analysis function.This method can accurately reproduce energy and energy derivatives,which is of great significance for dynamic simulation.Then the method was verified in the H2O system,H+H2 hydrogen exchange reaction and H2+CO+?H+HCO+/HOC+reaction.In addition to the traditional Euclidean distance,the relative value and absolute value of the force were used to selecting the points,which greatly reduces the ab initio energy points of the product channel and reactant channel,and well avoids the possibility of energy oscillation in the long-range region.In addition,this project also develops an accurate full-dimensional ab initio PES for F+NH3?HF+NH2 reaction.The fundamental-invariant neural network(FI-NN)method is utilized to fit the PES based on 41282 ab initio energy points at the level of UCCSD(T)-F12/aug-cc-pVTZ,resulting in a total root mean square error(RMSE)of 0.13 kcal mol-1.The quasi-classical trajectory(QCT)dynamics study was performed on this potential energy surface.Combining the simulation results and experimental results,we found that the reaction is dominated by the direct abstraction and stripping mechanisms while a considerable amount of reaction takes place by the indirect "yo-yo"mechanism.So that there is no temperature dependence of the thermal rate constant within the statistical error can be reasonably explained.On the other hand,the calculated product energy partition shows that the energy could not effectively flow into the rotational mode of the two products.The vibrational state distribution of HF follows PvHF=2?PvHF=1>PvHF=0>PvHF=3,which explains the previous dispute about the vibration dynamic distribution of the product HF among different experiments.