Modeling martensitic transformation temperatures in Zirconia–Ceria solid solutions using machine learning interatomic potentials

Abstract Shape memory ceramics (SMCs), while exhibiting high strength, sizeable recoverable strain, and substantial energy damping, tend to shatter under load and have low reversibility. Recent developments in SMCs have shown significant promise in enhancing the reversibility of the shape memory phase transformation by tuning the lattice parameters and transformation temperatures through alloying. While first-principles methods, such as density functional theory (DFT), can predict the lattice parameters and enthalpy at zero Kelvin, calculating the transformation temperature from free energy at high temperatures is impractical. Empirical potentials can calculate transformation temperatures efficiently for large system sizes but lack compositional transferability. In this work, we develop a model to predict transformation temperatures and lattice parameters for the Zirconia–Ceria solid solutions. We construct a machine learning inter-atomic potential (MLIAP) using an initial dataset of DFT simulations, which is then iteratively expanded using active learning. We utilize reversible scaling to compute the free energy as a function of composition and temperature, from which the transformation temperatures are determined. These transformation temperatures match experimental trends and accurately predict the phase boundary. Finally, we compare other relevant design parameters (e.g. transformation volume change) to demonstrate the applicability of MLIAPs in designing SMCs.