Active learning of interatomic potentials in the vicinity of dynamical instability for low-moduli bcc Ti-based alloys

The emergence of machine learning interatomic potentials (MLIP) made it possible to account simultaneously for chemical disorder and finite temperature effects in theoretical simulations of multicomponent alloys. Active learning of MLIPs allows one to predict elastic properties of technologically relevant alloys at temperatures and stresses relevant for their applications with the same accuracy as state-of-the-art ab initio molecular dynamics, but with efficiency which is several orders of magnitude higher [1]. Here we extend the scheme towards alloys with dynamical instability and demonstrate its applicability in studies of elastic properties of multicomponent low-moduli bcc Ti-based alloys, relevant e.g. for bio-medical applications. Varying Ti to Nb ratio in Ti_94-xNb_xZr₆ and Ti_94-xNb_xSn₆ alloys, we predict that in a vicinity of dynamical and mechanical instability both alloys have very low elastic moduli, comparable to those of human bones. Moreover, these alloys demonstrate anomalously weak dependence on temperature in a wide temperature interval, that is they can be classified as elinvar alloys. Simultaneously, we observe extremely strong anisotropy of directional Young’s modulus and argue that this could be useful for tailoring the alloys mechanical properties in applications.

[1] F. Bock, F. Tasnádi, and I. A. Abrikosov, “Active learning with moment tensor potentials to predict material properties: Ti_0.5Al_0.5N at elevated temperature”, J. Vac. Sci. Technol. A 42, 013412 (2024).