Phase Transitions in Inorganic Halide Perovskites from Machine-Learned Potentials

The atomic scale dynamics of halide perovskites have a direct impact not only on their thermal stability but also optoelectronic properties. Progress in machine-learned potentials has recently enabled modeling the finite temperature behavior these materials using fully atomistic methods with near first-principles accuracy. Here, we systematically analyze heating and cooling rate, simulation size, model uncertainty, role underlying exchange-correlation functional phase CsPbX3 X = Cl, Br, I, including both perovskite δ-phases. We show that rates below approximately 60 K/ns system sizes at least few tens thousands atoms should be used to achieve convergence regard parameters. By controlling factors constructing models are specific for different functionals, then assess seven widely semilocal functionals (LDA, vdW-DF-cx, SCAN, SCAN+rVV10, PBEsol, PBE, PBE+D3). based LDA, SCAN+rVV10 agree well experimental data tetragonal-to-cubic-perovskite transition CsPbI3 reasonable agreement perovskite-to-delta temperature. They underestimate, however, orthorhombic-to-tetragonal All other models, those CsPbBr3 CsPbCl3, predict temperatures experimentally observed values all transitions considered here. Among vdW-DF-cx yield closest experiment, followed by PBE+D3. Our work provides guidelines systematic analysis inorganic similar systems. It serves as benchmark further development functionals.