Study of the novel boron nitride polymorphs: First- principles calculations and machine learning

Abstract

This study explores two new boron nitride polymorphs, namely C2221 BN and I-4 BN, characterized by sp2 hybridization. The investigation is conducted through first-principles calculations, including assessments of their structural properties, stability, elastic properties, anisotropy, and electronic properties. The newly discovered boron nitride polymorphs exhibit mechanical stability, dynamic stability, and thermodynamic stability, as ascertained by analyzing their elastic constants, phonon spectra, and associated enthalpies. The B/G values for both C2221 BN and I-4 BN far exceed the threshold of 1.75, thus confirming that they are ductile materials. The BN polymorphs investigated in this work exhibit different degrees of anisotropy in Young's modulus, shear modulus, and Poisson's ratio. Meanwhile, two different datasets were trained using four machine learning algorithms, and the random forest model with the m2ax dataset showed a high fitting degree, small error, and good stability. Then, the machine learning model, which has undergone training, is used to predict the elastic modulus of boron nitride polymorphs. Upon comparing the computed GGA values with the experimental values of c-BN, it is evident that the random forest model outperforms in accurately predicting the elastic modulus of boron nitride polymorphs.