Computational Study And Machine Learning On Photochemical Process Of Molecular Systems

The photochemical process is one of the most common processes in life,responsible for human vision,plant photosynthesis,and the optical modification of materials.Photochemistry is the study of the interaction between matter and electromagnetic radiation,and is an important research area within physical chemistry,encompassing numerous fields such as spectral characterization,organic photoluminescent materials,organic light-emitting diodes,and photocatalysis.Spectroscopy has always been an important tool for exploring the microstructure and properties of materials,with the characterization and measurement of materials involving the digitization and reduction of complex,high-dimensional information to a few dimensions.Traditional approaches to exploring the microscale properties of materials typically involve using spectroscopic tools to identify the structure of the material,and then inferring the desired properties through expert knowledge or theoretical simulations.However,in recent years,with the development of spectroscopic techniques and high-throughput experiments,massive amount of spectral data has been generated,and the traditional approach of relying on experts to analyze spectra one by one,deduce structures,and estimate properties has become inefficient.Artificial intelligence(AI)is adept at exploring correlations in high-dimensional and complex data,and has introduced a new paradigm of data-driven research.Therefore,developing AI strategies that can replace human experts has become an important research direction for the future.This paper starts from the basic theory of molecular photochemistry and employs first-principles density functional theory calculations and modern data science methods of machine learning.It focuses on the artificial intelligence interpretation of spectra to obtain the microscopic physical and chemical properties of materials molecules,as well as the rational design and mechanistic exploration of photo-functional materials.The paper is divided into four parts:The first part,namely Chapter 1,discusses the basic principles of molecular photochemistry and the research progress of machine learning in photochemistry.We discuss the movement of molecules under photoexcitation from the perspective of quantized electronic energy levels,vibrational energy levels,and rotational energy levels,including electronic transitions,vibrational transitions,and rotational transitions.The absorption and emission processes of electronic transitions correspond to ultraviolet-visible spectroscopy and fluorescence and phosphorescence,while infrared spectroscopy and raman spectroscopy complement each other in describing vibrational transitions.These spectroscopies play an important role in experimental characterization and theoretical analysis.In recent years,data-driven research paradigms have gradually entered the field of photochemistry,and we discuss the research progress of machine learning in photochemistry in two stages:machine learning prediction of spectra to accelerate quantitative calculations,and machine learning spectroscopic inversion to explore hidden correlations.The second section,namely Chapter 2,introduces the theoretical methods involved in the work,including electronic structure theory and machine learning methods.Electronic structure theory is based on the fundamental laws of quantum mechanics,and calculates the electronic wave function to obtain the properties of molecules and atoms.Density functional theory uses electronic spatial density instead of the wave function coordinates,ushering in a new era of theoretical calculations.Machine learning is a data science method that learns the underlying rules of things from big data.We discuss machine learning classification,machine learning basics,and common algorithm implementations.The third section,namely Chapter 3,introduces our "white-box" intelligent model for constructing spectra-property relationship for imperfect chemical small data.Due to experimental and computational costs,high-quality chemical data is usually sparse and lacking,making it difficult to meet the data requirements of intelligent models.To address this problem,we used copper-based metal-organic frameworks as the research object,used infrared/raman spectroscopic features as descriptors,and established an interpretable intelligent model for predicting the performance indicators of metal catalysts based on compressed sensing algorithms.This "spectra-property relationship"appears in a clear mathematical analytical expression in a "white-box" form,exhibiting excellent generalization ability.Even in the case of small data or partial errors,the model can accurately predict and transfer generalization,and even identify erroneous data,achieving data cleaning.This research approach for interpretable intelligent models will greatly enhance the applicability of machine learning methods in the field of photochemistry.The fourth section,consisting of Chapters 4 and 5,explores the mechanisms and design of organic photoluminescent materials using first-principles calculations based on the fundamental principles of photochemistry.Typically,the design of organic long persistent luminescence materials involves enhancing the intersystem crossing process or increasing the lifetime and stability of the excited state.In Chapter 4,we formed a D-S-A molecular configuration by connecting photo-isomerized azobenzene molecule with D-A type fluorescent molecule,thereby increasing the rate of intersystem crossing and inducing the formation of a photo-induced charge separation state,which is favorable for the realization of organic long persistent luminescence materials.The study of luminescent radicals as low-cost alternatives to organic light-emitting diodes is still in its infancy and faces challenges such as poor stability and lack of blue-green emission.In Chapter 5,we theoretically explored the ground state stability of diradical molecules and the energy levels of molecular orbitals,electron population,and orbital transitions information of the molecules upon photoexcitation.Our findings indicate that the emission mechanism of the radical molecules with doublet states is in accordance with the anti-Kasha rule,thus verifying the possibility of blue-green photoluminescence from radicals from a theoretical perspective.