Diabatization And Nonadiabatic Dynamics Study Of Conjugated Molecular Chains

Conjugated molecular chains are widely present in chemistry,biology,and material sciences,and the nonadiabatic dynamics(such as charge transport)processes determine their intrinsic properties.Since neither the fully-quantum nor the fullyclassical dynamics can give effective simulations of the electron dynamics in these large extended systems,mixed quantum-classical dynamics has gained great attention.The trajectory surface hopping method has been widely used to study nonadiabatic dynamics because of its ease of combination with first principles electronic structure calculations,good detailed balance,and clear physical and chemical pictures.However,because of the intrinsic difficulties to describe complex surface crossings and the high demand of efficient electronic structure calculations,existed surface hopping methods cannot be used to simulate the nonadiabatic dynamics in long molecular chains.This thesis makes theoretical studies to solve these two problems and the new methods are applied to study the charge transport in DNA chains.In the first part of the thesis,additional decoherence correction is shown to intrinsically enhance the difficulty to treat complex surface crossings in the traditional surface hopping framework and the underlying mechanism is given.Furthermore,we propose a restricted decoherence(RD)strategy and incorporate it into the selfconsistent fewest switches surface hopping(SC-FSSH)method.The resulting SCFSSH-RD approach reduces the simulation time of nonadiabatic dynamics by about 50 times.Note that the RD strategy is highly universal and can be combined with the global flux,crossing classified,and subspace surface hopping methods to further improve the numerical stability and simulation efficiency when dealing with complex nonadiabatic dynamics problems.In the second part of the thesis,we propose an efficient machine learning approach for diabatization,which could construct the diabatic Hamiltonian of flexible molecular systems with high efficiency and reliability.Using the wavefunctions and energies of molecular orbitals(MOs),this approach can realize the diabatization of any MO independently by training the specific artificial neural network(ANN),without the need for nonadiabatic coupling calculations.The trained ANN can be used to predict the diabatic Hamiltonian of any geometry directly without additional electronic structure calculations,greatly reducing the computational and storage costs of electronic structure calculations in on-the-fly nonadiabatic dynamics simulations.In the third part of the thesis,we study the charge transport in real DNA chains using the new diabatization and surface hopping methods proposed in the first two parts,which have been integrated in the SPADE(simulation package for nonadiabatic dynamics in extended systems)software developed by our group.DNA can be regarded as a heavily doped system due to its four kinds of diverse nucleobases.Our nonadiabatic dynamics simulations reproduce the complex charge transport mechanisms and find that the charge transport ability of different sequence differs by several orders of magnitude.We focus on the microsatellite sequences,which are abundant in the genome.The effects of local and nonlocal electron-phonon couplings on the nonadiabatic dynamics are systematically examined,and the femtosecond dynamics of charge transfer observed in experiment is reproduced.Especially,to a certain degree,we observe linear relationship between the charge transport abilities of specific DNA sequences and their distributions in the genome.And this linear relationship seems to vary regularly with species evolution.