Fig. 9: Generality of our MLIP training strategy for ceramic materials. | npj Computational Materials

Fig. 9: Generality of our MLIP training strategy for ceramic materials.

From: Machine-learning potentials for nanoscale simulations of tensile deformation and fracture in ceramics

Fig. 9

a α-TaB2, b ω-WB2, c γ-ReB2, d NaCl-type TiN, and e the orthorhombic Ti2AlB2 MAB phase, as illustrated by stress/strain curves for room-temperature uniaxial tensile deformation. Specifically, the [0001] and the [001] loading directions are chosen as representative examples for hexagonal systems (TaB2, WB2, ReB2) and for the cubic and the orthorhombic systems (TiN, Ti2AlB2), respectively. The supercell sizes and computational setup are equivalent to the atomic scale tensile tests for TiB2, defined by the first bullet point in Section MLIPs’ up-fitting for nanoscale tensile tests. Note that the stress values should not be over-interpreted, as they were obtained for atomic scale supercells (to make a fair comparison with ab initio data) and---in case of negligible size effects---are the ideal upper bounds attainable by a perfect single crystal.

Back to article page