Fig. 8: Illustration of simulations to which the here-developed MLIP (MLIP-4) is transferable (a–d) or for which it requires up-fitting (e–f). | npj Computational Materials

Fig. 8: Illustration of simulations to which the here-developed MLIP (MLIP-4) is transferable (a–d) or for which it requires up-fitting (e–f).

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

Fig. 8

a Comparison of AIMD (dash-dotted line) and ML-MD (solid line) stress/strain curves for TiB2 subject to (0001)\([\overline{1}2\overline{1}0],(10\overline{1}0)[\overline{1}2\overline{1}0]\), and \((10\overline{1}0)[0001]\) room-temperature atomic-scale shear deformation. b–d Representative snapshots at strain steps marked by shaded rectangles in a. The dashed lines in b–d guide the eye for slip directions described in the text. e Differences in ML-MD and AIMD stresses (σ(ML-MD) − σ(AIMD)), resolved in the basal plane and [0001] direction (σx,y and σz) of TiB2 subject to room-temperature volumetric compression, plotted as a function of the compression percentage. f Blue and red data points indicate maximum extrapolation grades returned by MLIP-[4] and its up-fitted version, MLIP-[4]Plus, during TiB2 compression.

Back to article page