Property Prediction And Analysis Of Two-dimensional Transition Metal Materials Based On Distributed Feature Representation

The calculation of electron band structure is one of the important application di-rections of density functional theory.The first principle calculation method based on density functional theory can approximately solve all kinds of material properties of a given molecular structure,but it also has some shortcomings,such as slow calculation speed and so on.In recent years,machine learning model has been widely used in material property prediction because of its fast reasoning speed and strong mapping and fitting ability.Based on the material structure data in oqmd database and the two-dimensional transition metal sulfide and halide data in c2db database,this paper constructs an effective machine learning model to predict its G₀W₀bandgap value.Aiming at the characteristics of non-uniform dimension and complex composition of traditional atomic features,atomtransmachine（ATM）model is introduced to extract the distributed features of atoms.The cluster analysis shows that the ATM feature effectively extracts the features of the main group elements and some transition metal elements,especially shows high similarity in the same group elements and adjacent group elements of the same period,and preliminarily proves the effectiveness of the ATM feature.Magpie feature and Atom2Vec feature are compared in this paper.Numerical experiments show that the computational accuracy of the ATM feature based method is generally higher than that of the traditional method in the current task,which further proves that the ATM feature is suitable for the current bandgap prediction task.It is found in the experiment that the distributed atomic features like ATM feature and Atom2Vec feature have stronger linear correlation with the band gap of two-dimensional transition metal materials,and are more stable in the face of different machine learning regression models.This paper also discusses the performance of the combination feature based on ATM feature and the combination model based on support vector regression model,ridge regression model and gradient ascending tree model in the current task.Through ten fold cross validation analysis,the prediction results of the model meet the actual demand.The model established in this paper can be applied to assist the screening of feasible two-dimensional tran-sition metal structure schemes,which greatly improves the computational efficiency compared with the traditional first-principles calculation method.