Screening And Performance Research Of Lead-Free Antiperovskite Based On Machine Learning

In recent years,perovskite materials have received widespread attention because of its excellent photovoltaic properties.There are a wide range of applications in the fields of solar cells,photodetectors and photoelectric sensors.Their excellent light absorption properties and ability to modulate the energy band structure make them ideal for a new generation of optical communication devices,such as optical modulators,optical amplifiers and optical switches.In addition,perovskite materials can also be applied to flexible electronics,quantum computing and optical storage,offering new possibilities for the development of electronics and communications technology.However,the toxicity of lead-based perovskite poses environmental and human hazards,and their inherent instability has hindered their commercialisation to a certain extent.For anti-perovskite,they are becoming a hot research topic in the field of chalcogenides due to their adjustable band gap,superconductivity,high carrier mobility and low cost.Therefore,one of the effective strategies to solve this problem is to find lead-free anti-chalcogenides with a band gap in the ideal light absorption range.Machine learning,on the other hand,rapidly identifies materials with potential applications from a large number of candidates by constructing appropriate models and algorithms,thus saving time and resources compared to traditional approaches of material exploration.This work combines density functional theory with machine learning to construct structure-property relationships for X₃BA-type anticalcite and to rapidly predict the band gap of anticalcite.The predicted compounds are further screened for their stability,light absorption ability and other related properties through DFT calculations,and lead-free anticalcite materials with suitable band gaps and room temperature stability are obtained.The main content of this article has the following two aspects:（1）We establish a machine learning model of the X₃BA antiperovskite material about the band gap,and introduce the machine learning algorithm used in this work.The data of band gap about 128 antiperovskite compounds were generated by first-principle calculations,while 20 original features were selected as the initial feature sets.The data set is divided into training set and test set based on the ratio of 4:1,and the best festure subset is selected with the method of feature engineering.In the next step,the performance of the support vector machine,random forest,gradient boosting regression,decision tree,and Gaussian regression algorithm was compared.It was found that the gradient boosting regression has the best prediction performance for the band gap of X₃BA perovskite materials.After hyper-parameter optimization,the coefficient of determination and root mean square errors of GBR model are 0.982 and 0.0268.At the same time,the impact of important features on the target variable is analyzed.The optimized model was used to predict the band gap,and the predicted data were further screened based on structural stability and appropriate band gap.From 832 X₃BA prediction samples,71 antiperovskite candidate materials were screened out.（2）DFT calculations and analysis were performed on screened 71 materials predicted by machine learning.Twenty-nine of these structures were found to have cubic or pseudo-cubic cells,and 18 of these 29 structures have band gaps calculated by HSE that matched the values of machine learning band gap prediction.Ultimately,nine bandgaps were selected as suitable candidates for solar photovoltaic applications.The stability,electronic and optical properties of these nine structures were investigated by density functional theory calculation,and their potential photovoltaic applications were explored.Three materials that are Rb₃ON₃,Rb₃ONCO,Rb₃OPF₆ are found to have suitable band gaps,good stability and high optical absorption coefficients.In this work,by combining first-principles calculation and machine learning,a prediction model was successfully established to screen 3 kinds of promising lead-free antiperovskite materials,and they were verified by calculations to have good performance.They can be candidate material in the field of lead-free perovskite solar cells.