First-Principles Investigation Of Mg-Based Semiconductors And Application Of Artificial Intelligence In Cluster Physics

Many members of the ABC2 ternary semiconductors show good performance in the fields of nonlinear optics and photo-electric solar energy converters.In 2018,MgSiAs2,a new member in the ABC2 family,has been synthesized for the first time by K.E.Woo et al.,and was found to have good nonlinear optical performance in the infrared region.Meanwhile,K.E.Woo et al.explored a new structure Mg3Si6As8 in experiment.The band structure of Mg3Si6As8 has pseudo-direct band gap with width 2.02 eV.The valence band top of Mg3Si6As8 has multiple local maxima with small energy differences between each other.The small energy-differences between these local maxima provides multiple light absorption channels,therefore the light absorption coefficient of the material tend to be large.Accordingly,we can assume that the photovoltaic efficiency of Mg3Si6As8 is probably high.Defects can highly reform the properties of a material,including its optical,electrical and magnetic performances.For instance,an infrared nonlinear optical material should be transparent in the infrared region.However,the transparency window of the material can be limited by the photo-ionization of native defects.In the case of photovoltaic materials,substitutional doping in a multi-component semiconductor is a widely employed strategy to tune the band-gap width and the band-gap width are directly related to the photovoltaic efficiency.What's more,in the field of magnetic engineering,inducing magnetic property in nonmagnetic semiconductors througb magnetic doplng has been paid much attention.This dissertation is divided in two parts.In the first part,we systematically investigated the effects of doping on the optical,electrical and magnetic properties of MgSiAs2 and Mg3Si6As8.The first part contains five chapters.In the first chapter,we analyze the structure and electronic structures of MgSiAs2 and Mg3Si6As8,and make bold assumptions about their nonlinear optical and photovoltaic properties accordingly.In the second chapter,the first-principles calculations that used in our calculations are briefly introduced.In the third chapter,we systematically investigated the native point defects of MgSiAs2 including vacancies,interstitials and antisites using first-principles calculations with the hybrid functional.Of the thirteen different point defects studied,nine kinds of defects show deep transition levels,which might contribute to the limiting of the transparency spectrum of MgSiAs2 compounds.The defects with the highest concentration at equilibrium are found to be cation antisites,Mgsi and SiMg,serving as acceptors and donors respectively.Because of their significantly low formation energies,they lead to Fenni level pinning in the band gap and constrain the carrier concentration doping.Our calculations show that a change in growth conditions affects modestly the formation energy of defects or the concentration of charge carriers.In the fourth chapter,through substituting the ? and the ? elements in Mg3Si6As8,we theoretically predicted a new class of crystals denoted as Mg3?6?8.Nine stable crystals in this new class are theoretically predicted,and the band-gaps of these compounds range from 2.5 eV to 0.88 eV.It is found that each of these compounds shows good performance in photovoltaic efficiencies.Furthermore,multilayer tandem solar cells consisting of these compounds are proposed,and the photovoltaic efficiencies of these solar cells are predicted to reach 30%exceeding that of silicon and most silicon containing tandem cells.In the fifth chapter,we present a study of Co doping in the wide-band-gap semiconductor Mg3Si6Ass.It is found that among all considered doping sites in the compound,the substitutional doping at the tetrahedral sites are by far the most stable,and the exchange interactions between substitutional Co ions at the tetrahedral sites yield antiferromagnetic order.The crystal field effect in the splitting of the Co-d orbitals as well as the nature in the magnetic couplings between CotetMg is revealed,and the optimal doping for the highest Neel temperature is found.The second part of this dissertation contains two chapters and is an investigation of the capability of machine learning based potential(MLP)models in the field of cluster physics.In the sixth chapter,we briefly introduced neural networks and machine learning,and some most pronounced MLP models.We also made comparison on the basic ideas,pros and cons of these MLP models.In the seventh chapter,a ML based deep potential model for Al clusters is developed through training with an extended database including ab initio data of both bulk and several clusters.Based on our developed DP model,the low-lying structures of 101 different sized Al clusters are extensively searched,among which the lowest energy candidates of 69 sized clusters are updated.Our calculations demonstrate that machine-learning is indeed powerful in generating potentials to describe the interaction of atoms in complex materials.