Machine-learned interatomic potentials for accurate analysis of the mechanical properties of boron nitride sheets

Abstract

We introduced a novel machine-learned interatomic potential (MLIP) by thoroughly discussing the step–by–step MLIP creation process using precise but limited data. This study explored the mechanical properties of hexagonal boron nitride (hBN) nanosheets and addressed the challenges of accurately predicting their structural properties. We explored the use of ab initio molecular dynamics (AIMD) and classical molecular dynamics (CMD) simulation techniques, emphasizing the necessity for a more effective and efficient solution. We also discussed the machine learning procedure to construct an effective interatomic potential. Furthermore, we address techniques for evaluating the performance and robustness of MLIPs on unseen datasets. Using the newly formed MLIP in a CMD simulation, we investigated the mechanical attributes of hBN nanosheets, exploring the fluctuations in sheet strength across a range of dimensions, temperatures, and varying numbers of layers. We obtained an average Young’s modulus in the range of 980–1000 GPa at 1 K, whereas the average failure stress and strain were approximately 106 GPa and 0.16, respectively. Our results demonstrate significant improvements in the accuracy of hBN nanosheets compared to prior studies, highlighting the effectiveness of MLIP in achieving higher precision with minimal computational cost. This study offers comprehensive analysis and theoretical exploration, delivering valuable insights into MLIP and the mechanical properties of hBN nanosheets, and paves the way for future applications in materials science and engineering.