Font Size: a A A

A Preliminary Study Of Application Of Atomic Neural Network And Ring Polymer Molecular Dynamics In Surface Chemical Reactions

Posted on:2022-10-07Degree:DoctorType:Dissertation
Country:ChinaCandidate:Q H LiuFull Text:PDF
GTID:1481306323481924Subject:Physical chemistry
Abstract/Summary:PDF Full Text Request
The microscopic core step in the fields of heterogeneous catalysis,corrosion,and material preparation is the chemical reaction process between molecules and surfaces.In heterogeneous catalysis,an in-depth understanding of the adsorption,diffusion,dissociation,scattering and desorption processes of molecules on the surface of transition metals,and understanding of the elementary reactions on the surface during the reaction process are of great help to the design of catalysts and industrial control.Theoretical researchers focus on complex chemical reactions at the atomic and molecular scale based on first-principles calculations to obtain kinetic and thermodynamic information during the reaction process,and they provide reaction mechanism support with experimental scientists to broaden the understanding of elementary reaction on the metal surface.With the improvement of modern computer technology and computing power,scientific research,which theoretical researchers focus on,have gradually shifted from simple chemical reactions between small molecules in the gas phase to molecular dynamics simulations of large systems at the interface.In the molecular dynamics simulation of large systems,although the speed of using the empirical force field is fast,the accuracy is not high.The high-precision first-principles-based molecular dynamics calculations are expensive.Recently theorists have simulated the reaction process by constructing a potential energy surface based on first-principles calculations and performing quantum wave packet evolution dynamics or quasi-classical trajectory dynamics on the potential energy surface,where the kinetic information of reaction can be obtained.However,as the number of atoms in the system increases,it becomes much more difficult to construct the potential energy surface of the system and to accurately calculate the quantum dynamics.With the development of neural network technology,the fitting method of potential energy surface has also developed from the parameter fitting of semi-empirical potential energy surface to the fitting method based on neural network.Different from the computer industry,in the chemical reaction system,because the atoms and molecules in the system must meet the invariant symmetry of translation,rotation,and replacement,they cannot simply be fitted with atomic coordinates.Behler and Parrinello proposed the concept of atomic neural network in 2007.They believe that the energy of a single atom is affected by the environment near it.The interaction between the environment near a single atom and the central atom is the input layer of the neural network,which is fitted by the neural network.Finally,the energy of the entire system is equal to the sum of the energy of all atoms in the system.This method uses the symmetry function of the atomic center as the descriptor of the neural network,so that the dimensionality of the system has good scalability,and ensures the invariant symmetry of the potential energy surface with respect to the translation,rotation,and displacement of the molecule.This method has been widely used in very complex systems,condensed phase systems.For the construction of the potential energy surface of the high-dimensional system,based on the data points obtained by first-principles calculations,we use the atomic neural network(Atomic Neural Network,AtNN)method to construct the 60-dimensional potential energy surface of HCl on the Au(111)surface for research The scattering and dissociation process.We exhibited the morphology of the rigid surface potential energy surface at different sites,calculated the static properties of the reaction energy barrier and the sticking probability of DCl on the Au(111)surface,sticking probability of HCl on the Au(111)surface,and energy transfer of HCl inelastic scattering from metal surface.We reproduce the results of DFT on the energy barrier when the surface atoms move,and the dissociation probability of DC1 calculated by the AIMD method.It proves that our potential energy surface can accurately predict the energy of the system under the accuracy of first-principles calculations.We have studied the energy loss during the inelastic scattering process of the HCl state in consideration of the channel to the surface,compared with the previous rigid PES.In consideration of the energy loss to the electron-hole pair excitation channel in the surface metal,it is found that only a small amount of energy is lost to the electron-hole pair excitation channel during the scattering process,and a large amount of energy is lost to the surface lattice vibration.Nevertheless,compared with the experiment,we still underestimated the energy loss to the surface lattice vibration.In addition,focusd on the dissociation process of HCl,QCT results was similar to the previous results based on rigid PES and did not narrow the difference between experimental results and theoretical results.Comparing QCT with QD results,ZPE leakage is serious in this system,which also promotes us into performing accurate molecular dynamics simulations in large systems.A new accurate molecular dynamics simulation method is required,for the zero-point energy leakage in quasi-classical trajectory method.ZPE leakage will affect the observation of dynamic properties,such as mode specificity,sticking probability and energy transfer of scattering products etc.Ring polymer molecular dynamics(RPMD)can strictly approximate the average value of quantum statistics in equilibrium statistics,and has also been successfully applied to the calculation of rate constants including quantum zero-dot energy effects and tunneling effects.Therefore,we took the lead in applying RPMD to the reaction process of molecules colliding with the metal surface.The H2+Cu(111)and D2O+Ni(111)reaction system have been used for tests.We found the zero-point vibrational energy of H2(0.27 eV)and D2O(0.41 eV)can be well included in the canonical ensemble(NVT).The sticking probability of molecules on the metal surface is studied in the micro-canonical ensemble(NVE).It is found that the QCT in the region underestimate the sticking probability for ignoring the tunneling effect where normal incidence energy is low(Ei<0.55 eV)in the H2+Cu(111)system.The RPMD results are in good agreement with the quantum dynamics results at 300K,while QCT seriously overestimates the sticking probability due to a large amount of zero-point energy leakage in the D2O+Ni(111)system.The great performance of RPMD in the two systems provides a possibility for accurate high-dimensional quantum dynamics simulation with quantum effects included,ZPE and tunneling effect.
Keywords/Search Tags:Potential Energy Surface, Atomic Neural network, Ring Polymer Molecular Dynamics, Zero-point Energy leakage, Tunneling effects, Energy Transfer
PDF Full Text Request
Related items