Developing an analytical bond-order potential for Hf/Nb/Ta/Zr/C system using machine learning global optimization

Despite being effective in unveiling the complicated nanostructure-property relation of (HfNbTaZr)C high-entropy carbide (HEC), applications of atomistic simulations are restricted by the lack of reliable many-body interatomic potentials. In this study, we propose a machine learning (ML) global optimization framework to parameterize an analytical bond-order potential (ABOP) with massive internal parameters for the Hf/Nb/Ta/Zr/C system. The combined distributed breeder genetic algorithm (DBGA), extensive ab initio training data and a normalized objective function markedly accelerate the convergence towards global optimum. By performing a cross-validation for the established potential, we demonstrate that the ABOP can accurately predict fundamental energetic, structural and mechanical properties of metals, intermetallic compounds, alloys and carbides in both the training and test datasets. Moreover, the potential is employed in dynamic simulations such as thermal annealing, simple loadings, nanoindentation and irradiation collision, from which some typical experimental observations are successfully reproduced. This not only further verifies the reliability of ABOP, but also implies its broad future applicability in nanoscale simulations.