Ab Initio Thermodynamics and Atomistic Modeling of NiTi SMA with Machine Learning Interatomic Potentials

Equi-atomic Nickel-Titanium (NiTi) shape memory alloy (SMA) possesses interesting properties such as pseudo/super-elasticity and shape memory effect. They arise from the ability of the alloy to exist in two different phases and undergo a stress- or a temperature-induced reversible phase transformation. Austenite is the high-temperature parent phase which has a B2 cubic crystal structure, and it transforms into the low- temperature martensite, which has a B19' monoclinic structure. Modeling the behavior of NiTi is challenging, since there are other competing phases (orthorhombic B19 or the base-centered orthorhombic B33). The thermodynamic phase stability still remains ambiguous and one needs to resort to <i>ab initio</i>-based methods such as density functional theory (DFT). DFT predicts the B33 phase as the lowest energy structure at 0 K (Huang et. al., 2003), although it has never been experimentally observed. Ab initio molecular dynamics (AIMD) calculations have shown that the B19' and B2 phases are entropically stabilized at temperatures above ≈100 K (Haskins et. al., 2016), but the calculations were restricted to a single exchange correlational functional (XC). Additionally, studying kinetic effects during the phase transformation using only DFT is also severely expensive.<br/><br/>In this work, we try to address both the above issues. Using a recently developed thermodynamic integration (TI) technique (direct upsampling, Jung et. al., 2022) aided with machine-learning based moment tensor potentials (MTPs, Shapeev 2016), it is possible to efficiently calculate high-temperature thermodynamic properties and phase stability to DFT accuracy. We perform such calculations to analyze the prediction of three different XCs (GGA, LDA and SCAN) in the phase stability of NiTi in order to find the most accurate representation. We also perform large-scale molecular dynamics (MD) simulations using the MTPs fitted to high-temperature AIMD data from the different XCs, in order to study the kinetically-driven phase transformation behavior in each of these cases. Preliminary results show that the B2 and B19' phases do get en- tropically stabilized with temperature for all XCs. The MTPs are extremely accurate in predicting the energies and forces of various phases in comparison to previously existing conventional interatomic models such as the EAM and the MEAM. Only the MTP fitted to the SCAN XC data predicts a reversible B2-B19' transformation. In the case of GGA and LDA, the B2 phase transforms into a B19 phase on cooling, suggesting that there might be an energy barrier in the B2-B19' transformation path.<br/>References<br/>[1] Huang X., Ackland G.J., Rabe K.M.: Nature Materials 2, 307-311 (2003).<br/>[2] Haskins J.B., Thompson A.E., Lawson J.W.: Physical Review B 94, 214110 (2016)<br/>[3] Jung J.H., Srinivasan P., Forslund A., Grabowski, B.: npj Computational Materials, <i>under review</i> (2022)<br/>[4] Shapeev A.V.: Multiscale Modeling and Simulation 14, 1153-1173 (2016)