Proceedings of Internet Conference on Theory and Computation of Halide Perovskites (ComPer)
Publication date: 4th September 2020
Finite-temperature simulations of complex dynamic solids are a formidable challenge for first-principles methods. Long simulation times and large length scales under isothermal-isobaric (NPT) conditions are required, demanding years of compute time. We applied the recently developed on-the-fly Machine-Learning Force Field (MLFF) scheme[1] to generate force fields for several hybrid and inorganic perovskites (APbX3, A={MA,Cs},X={I,Br,Cl}). The MLFF is trained on the potential energy surface of the state-of-the-art SCAN density functional. By benchmarking to high quality reference calculations applying the Random Phase Approximation, the SCAN functional is shown to result in the 'best' potential energy for hybrid perovskites [2]. The MLFFs open up the required time and length scales, while retaining the distinctive chemical precision of first principles methods. We study the entropy driven phase transitions of hybrid perovskites, which have never been accurately described in simulations. Simulations using machine learned potentials give direct insight into the underlying microscopic mechanisms. The ordering of the Methylammonium (MA) molecules as function of temperature beyond the accuracy of toy Hamiltonians is obtained [3]. Furthermore, we relate the phase transition temperatures of different perovskites to the radii of the involved species, and we determine the order of the transitions in Landau theory.
[1] R. Jinnouchi et al., Phys. Rev. Lett. 122, 225701 (2019)
[2] M. Bokdam et al., Phys. Rev. Lett. 119, 145501 (2017)
[3] J. Lahnsteiner et al., Phys Rev. B 100, 094106 (2019)