| Abstract Scope |
Boron carbide is light, super hard, and intensely studied as one of the boron-based ceramics due to its importance and applications in refractory material, neutron absorbent in nuclear reactors, high-temperature semiconductors, and abrasive resistant material. The structure of boron carbide B4C can be viewed as an arrangement of 3-atom linear chains diagonally connecting 12-atom icosahedra across a rhombohedral unit cell. The potential energy surface of B4C is complex due to the presence of an icosahedral unit in its structure. Here, we present the generation of two forms of machine learning interatomic potentials: a bispectrum-based spectral neighbor analysis interatomic potential (SNAP) and deep neural network-based deepMD potential for B4C. I will compare the result of these machine learning interatomic potentials with the analytic angular dependent interatomic potential (ADP) against the cohesive energy, lattice constants, and elastic constants calculated with the ab initio method and experimental values reported in the literature. |