Research On Nucleation Mechanism Of Acetic Acid Dimethylamine Based On Neural Network Potential Energy Surface

Atmospheric aerosols play an important role in global radiation balance,global climate change and human health.New particle generation(NPF)is the main source of atmospheric aerosols.Due to the diversity of global atmospheric environment and the limitations of current detection methods,people’s understanding of the mechanism of new particle generation is still unclear,especially the process of the formation of critical clusters of meteorological substances in the early stage of new particle generation.At present,some field observations and experiments show that organic acids and amines can participate in the formation of new particles.In this paper,acetic acid(HOAc)and dimethylamine(DMA)with high content in the air are selected for research.In the process of studying the nucleation mechanism of acetic acid-dimethylamine system,molecular dynamics simulation is one of the important research contents,and obtaining accurate potential energy surface is the basis of molecular dynamics simulation.Ac-cording to the calculation of potential energy surface,information such as energy and force can be obtained.Among the commonly used methods of building potential en-ergy surface,ab initio calculation and density functional theory are of high accuracy,but they are slow in speed and high in cost,while semi empirical potential method is a little fast in speed but its accuracy is not up to the requirements.Because the neural network has the characteristics of flexible function form,high accuracy and fast speed,it can make up for the limitations of the above three methods.Therefore,this paper also applies neural network to the construction of potential energy surface in the study of the nucleation mechanism of acetic acid-dimethylamine,in order to obtain the molecular potential energy surface with high fitting accuracy,so as to obtain reliable molecular dynamics results,while greatly improving the calculation speed,which provides tech-nologic support for the study of the physical and chemical characteristics of the system and the impact on the formation of aerosols.The main research contents are as follows:1.The potential energy surface of molecules is obtained by neural network method.In this paper,a feedforward neural network with three hidden layers is constructed to predict the potential energy surface information of water clusters,including energy and force.Data set is a water cluster with 1-21 water molecules,with more than 300000 con-figurations,symmetry function and environment vector of single atom atom are selected as the input of molecular expression network.As for the energy of clusters,compared with the results of traditional quantum chemistry,the RMSE error of the neural network is 0.053 kcal mol-1,and the RMSE error of force is 0.48 kcal mol-1A-1.The change of energy with the change of bond angle between molecules is analyzed experimen-tally,the results show that the neural network can fit the data of traditional quantum mechanics well.2.Extending neural network model to the construction of potential energy surface of organic macromolecules,134 macromolecules with different heavy atom numbers are selected for prediction.Then,the correlation and error analysis between the prediction results of neural network and traditional quantum mechanics are analyzed.It is shown that the prediction results are linearly correlated with a slope of 0.969,approaching to 1,and the standard deviation of error is 0.76.In addition,compared with quantum chemistry,the neural network method can increase the computational speed by 105.This is of great significance to the prediction of potential energy surface of other molecular systems and the study of molecular dynamics.3.The low energy structure and thermodynamic properties of acetic acid-dimethylamine hydration system were obtained by combining the neural network model with PW91PW91/6-311++G(3df,3pd)method group,And further study the physic-ochemical properties of the system and its impact on aerosol formation.The nucleation mechanism of the system is introduced in detail.We found that acetic acid and dimethy-lamine form a relatively stable cluster,and proton transfer enhances the hydrogen bond in the(HOAc)(DMA)(H2O)n(n=2-4)cluster.strength.Temperature has an impor-tant effect on the distribution of isomers,especially for(HOAc)(DMA)(H2O)2 clusters.In addition,all isomers are part of the cluster.The effect of various relative humidity on the distribution of hydrates is negligible.However,from the comparison of hydrate distribution and cluster formation rate,unhydrate clusters always dominate,and they tend to form stable clusters.The above analysis shows that(HOAc)(DMA)is rela-tively stable,and some larger clusters based on(HOAc)(DMA)may be involved in the formation of new particles.