Machine-learning potentials for structurally and chemically complex MAB phases: strain hardening and ripplocation-mediated plasticity (2503.16018v1)

Abstract: Though offering unprecedented pathways to molecular dynamics (MD) simulations of technologically-relevant materials and conditions, machine-learning interatomic potentials (MLIPs) are typically trained for ``simple'' materials and properties with minor size effects. Our study of MAB phases (MABs) - alternating transition metal boride (MB) and group A element layers - exemplifies that MLIPs for complex materials can be fitted and used in a high-throughput fashion: for predicting structural and mechanical properties across a large chemical/phase/temperature space. Considering group 4-6 transition metal based MABs, with A=Al and the 222, 212, and 314 type phases, three MLIPs are trained and tested, including lattice and elastic constants calculations at temperatures $T\in{0,300,1200}$ K, extrapolation grade and energy (force, stress) error analysis for $\approx{3\cdot10^6}$ ab initio MD snapshots. Subsequently, nanoscale tensile tests serve to quantify upper limits of strength and toughness attainable in single-crystal MABs at 300~K as well as their temperature evolution. In-plane tensile deformation is characterised by relatively high strength, {110}$\langle001\rangle$ type slipping, and failure by shear banding. The response to [001] loading is softer, triggers work hardening, and failure by kinking and layer delamination. Furthermore, W$_2$AlB$_2$ able to retard fracture via ripplocations and twinning from 300 up to 1200~K.