Machine Learning Accelerates The Design Of New Functional Carbon Crystal Structures

With the rapid development of science and technology,functional materials have been widely used in various fields of modern industry.The structural design and synthesis of new functional materials is one of the current research hotspots in materials science.This paper adopts the crystal structure prediction technology based on particle swarm optimization algorithm.First,by using the CALYPSO structure search software,combined with the direct search method that only limits the number of atoms,a new large-cell superhard carbon crystal structure Fmmm-C80 with a direct band gap is obtained.Then,using the functional material reverse search method,in the process of using the CALYPSO code to search the crystal structure,set the band gap as the secondary screening target parameter in each generation of structure,and finally obtain a new monoclinic superhard carbon crystal structure C₁₂.Finally,a machine learning method is used to train a machine learning model that predicts the elastic modulus of carbon crystal structure,and the machine learning model is used in the reverse search process of functional materials instead of VASP to calculate material properties,thereby speeding up the reverse design process of superhard carbon.The main research results of this paper are as follows:1.By using CALYPSO software and density functional theory calculations,a new orthogonal superhard carbon phase Fmmm-C80（space group:Fmmm）with full-sp³hybridization is proposed.The unit cell of the Fmmm-C80 phase contains 80 carbon atoms,indicating that it is a large cell structure.The thermal and mechanical stability of the new structure was verified by calculating the enthalpy of Fmmm-C80 relative to diamond,molecular dynamics simulation,phonon spectrum and elastic constant.By using PBE functional and HSE06 hybridization functional to calculate the electronic band structure of Fmmm-C80,it is found that the band gap of the HSE06 level of the new phase is 4.11 e V,and it appears as a direct band gap,which confirms that Fmmm-C80 has semiconductor characteristics,and has the ability to become the future industrial semiconductor material.The Vickers hardness of Fmmm-C80 is 51 GPa,which means that Fmmm-C80 has the potential to become a superhard semiconductor material.2.By using an unbiased crystal structure search method combined with density functional theory calculations,a new superhard carbon crystal structure C₁₂（space group:C2/m）is proposed based on CALYPSO software.The C₁₂ phase has a unit cell of 12 atoms in the sp²-sp³ hybrid bonding network.By calculating the enthalpy relative to diamond,elastic constant,and phonon spectrum of the new phase,the mechanical and thermal stability of C₁₂ under ambient pressure can be determined.The electronic band structure of C₁₂ calculated using PBE functional and HSE06 hybrid functional respectively shows that C₁₂ is a narrow indirect band gap semiconductor.Although the band structure of the C₁₂ phase is metallic at the PBE level.But at the HSE06 level,its energy band structure shows a narrow band gap.The hardness of C₁₂ is 67 GPa,which means that C₁₂ is a potential superhard narrow band gap semiconductor material.3.Based on the unbiased crystal structure search method,machine learning（ML）methods are used instead of density functional theory（DFT）methods to predict the properties of carbon materials,thereby accelerating the search process of superhard carbon materials.After obtaining the available carbon crystal structure from the SACADA,a carbon material data set was established based on the high-throughput calculation method.Based on the data set,a machine learning model that specifically predicts the elastic modulus（Bulk modulus,Shear modulus,and Young’s modulus）of carbon structures is trained.Through comparison with AFLOW-ML,it is found that the new model has better predictability in predicting the elastic modulus of carbon allotropes.Then,in the process of reverse design of superhard carbon structure,ML model is used to predict young’s modulus of carbon structure instead of VASP,to speed up the search for carbon structure with high Young’s modulus.Through this method,a new carbon allotrope Cmcm-C24 was found,and its hardness was higher than80 GPa.The electronic band structure of the new phase shows that Cmcm-C24 is a direct band gap semiconductor.The calculated phonon spectra and elastic constants show that Cmcm-C24 has mechanical and thermal stability.