Two-dimensional Material Structure Design And Properties Research Based On Machine Learning

Two-dimensional materials have unique electrical,optical and mechanical properties,the most representative of which is graphene.Graphene is an electronic material with excellent physical and chemical properties.In order to apply it in a wider range of fields,it needs to be doped to precisely control its electronic properties.Among them,the work function is a good indicator,which can reflect the difference between the vacuum energy level and the Fermi energy level.On the other hand,boron doping has been widely studied because it is energetically more stable and can maintain the graphene honeycomb structure.However,due to the numerous structures of boron-doped graphene,research based on theoretical calculations is costly and time-consuming.Machine learning can quickly screen out materials with potentially excellent properties through the analysis and modeling of a small part of data,thereby accelerating the material research and development process.Therefore,this thesis uses machine learning methods to successfully predict boron-doped graphene with high work function and explore its performance as an anode material for alkali metal ion batteries.First,the work function prediction of boron-doped graphene is performed based on machine learning.Combined with the concentration of boron doping in the current experiment,2004 structures were selected from graphene supercells with boron doping numbers ranging from 4 to 8,and the work function value was obtained through first-principle calculations,and a data set was established.In order to obtain accurate results,four graph neural network models were used to train and predict the work function of boron-doped graphene,and the test error of the best model was only 0.054e V.By predicting all the structures that do not appear in the data set,the 50 structures with the highest work function are obtained,and the accuracy of the model prediction is verified using the first-principle calculations.Considering the possibility of experimental preparation,material stability was taken as one of the criteria,and the structure(B₅C₂₇)with the highest work function and lower formation energy（0.18 e V/atom）was found.A linear relationship between the work function and adsorption energy of boron-doped graphene was further found:the higher the work function,the stronger the adsorption capacity.Finally,the performance of B₅C₂₇as anode material for alkali metal ion batteries is studied.The calculation of the mass transfer process shows that it has a low diffusion barrier（0.41 e V,0.18 e V,and 0.1 e V for Li,Na,and K,respectively）,and the theoretical capacity and open circuit voltage of its Li/Na/K stable adsorption system have reached 2262/1546/1131 m Ah/g and 0.41/0.18/0.10 V,far superior to some traditional two-dimensional materials.In summary,this thesis uses machine learning to predict the highest work function of boron-doped graphene,combined with first-principle calculations to verify its stability,electronic properties and alkali metal ion adsorption performance,which lays a theoretical foundation for the rapid discovery of two-dimensional alkali metal ion battery anode materials with high work function,stability and high capacity.