Application Of Symbolic Regression In The Design Of A₂BX₆ Perovskite Materials

Since halogen perovskites were first used as solar cell materials,the power conversion efficiency of halogen perovskites solar cells has jumped from 3.8% to 26.7% in just over a decade.Halogen perovskites have become ideal candidates for solar cell light absorption.Perovskite materials because its composition diversity,low manufacturing cost and many excellent properties have become the focus in the field of science.However,the chemical stability prediction of the popular lead-based perovskite materials is still a challenge,and the search for stable lead-free perovskite materials has become a hot research topic.This paper mainly includes the following contents:In the first part,this paper uses the open-source Material Project database to obtain the structural data of compounds,and systematically studies the physical properties of 752 materials whose general formula is A₂BX₆（X=F,Cl,Br,I）by using first-principles calculation.For the 752 compounds,the formation energy,band gap and effective mass of the typical perovskite structure crystal phase（Fm3?m）and non-perovskite structure crystal phase（P3?m1）were calculated.The calculation accuracy of the decomposition energy was more than 93%,which ensured the reliability of the calculation.The changes of elements at different sites were summarized when the properties of compounds changed,which provided a new idea for the design of A₂BX₆ materials with specific functions.In the second part,the first principles calculation is combined with machine learning.Using the decomposition energy data obtained in the first part,the 88 atomic structural characteristics that may be related to the stability of A₂BX₆ materials are extensively searched.The important factors affecting the stability of A₂BX₆ compound,such as ionic radius,ionization energy and molar heat capacity,were found out and used as the characteristic quantity of machine learning.Finally,a symbolic regression algorithm based on genetic programming was used to find the interpretable A₂BX₆ material stability descriptor with an accuracy of nearly 90%.