Towards DFT-accurate prediction of twin interface structure and motion in NiTi shape memory alloys

Recent atomistic simulations have suggested that twin boundary motion, rather than interface energy, governs twin formation in NiTi shape memory alloys (SMAs). Yet, these findings rely on empirical interatomic potentials (IAPs), whose intrinsic inaccuracies pose uncertainties regarding the quantitative prediction of interface energetics, driving force and transformation mechanisms. In this study, we address these limitations by developing a machine learning IAP using the Performant Atomic Cluster Expansion (PACE) framework, trained on a comprehensive database of density functional theory (DFT) calculations. The resulting PACE-IAP outperforms state-of-the-art empirical and neural network-based potentials, by reproducing accurate lattice parameters, improved elastic constants, and correct features of the B2–B19’ phase transformation. Leveraging this increased accuracy, we model the structure, energetics, and motion of twin interfaces in NiTi. By computing the extrapolation grade, we verify that the local atomic environments at the predicted interfaces are well contained within the DFT configurational space. Our simulations confirm that the driving force for twin boundary motion, rather than the interface energy, controls the hierarchy of twin formation in NiTi. These atomistic insights can be used into mesoscale models of microstructural formation, ultimately enhancing predictions of variant selection and enabling the design of high-performance SMAs. • A machine learning interatomic potential for NiTi is developed using the PACE (Atomic Cluster Expansion) framework. • The interatomic potential accurately reproduces lattice parameters, elastic constants, and phase transformation features of B2 and B19’ phases. • Twin interface structures and motion are captured with DFT-level accuracy, validated via extrapolation grade analysis. • The driving force for twin boundary motion, rather than interface energy, is confirmed to control twin formation in NiTi shape memory alloys.