Application Of Conformational Adaptive Charge Model In The Development Of Azobenzene Force Field

The reversible conversion of azobenzene between cis and trans structures is the basis of molecular switching in the study of photodynamics.During cis-trans isomerization,the electron polarization caused by conformational change is obvious.The traditional force fields with fixed point charge model cannot accurately describe the polarization effect,so it is necessary to develop a charge model that can describe complex spatial conformation to capture the effect of electron polarization caused by conformational change.In recent years,machine learning has been gradually applied to the field of quantum chemistry.The efficiency of machine learning could be combined with the accuracy of ab initio method to improve molecular simulation.In this paper,atom type symmetry function was developed to describe atoms and their surrounding chemical environment based on atom centered symmetry function.By classifying atoms of the same element with different properties,the molecular structure provided a more reasonable machine learning descriptor.This method had been proved to have good mobility and was suitable for other molecular systems generally.The conformational adaptive charge prediction models of ground state S₀ and excited state S₁ were obtained by using the random forest regression algorithm,which used to replace fixed point charge model in traditional force fields.For the excited state charge,the accurate prediction results obtained only by using the geometric properties of ground state,which provided references for the development of other excited state force fields.Using the atomic descriptor generated by the central representative atom perceived the whole structure of azobenzene molecule.The prediction model provided accurate calculation results of molecular energies,excited energies and dipole moments.In conclusion,the atomic charges of ground state S₀ and excited state S₁ were predicted respectively based on atom type symmetry function and random forest regression algorithm,which provided a conformational adaptive charge model for azobenzene and effective parameters for the development of azobenzene force field.The atom type symmetry function descriptor based on the central representative atom could be used for the prediction of molecular integrity.This work provided an effective method for improving molecular dynamics simulation of azobenzene and provided references for the development of other excited state force fields.