Machine Learning Enabled Potential for (BA)2(MA)(n−1)PbnI3n+1 2D Ruddlesden–Popper Perovskite Materials

Lead-halide organic–inorganic perovskite material has recently been the focus of investigation by numerous research groups due to its favorable properties when employed as an active layer in a wide range of photovoltaic and optoelectronic devices. 2D perovskite layered type was introduced as a solution to the inherent moisture instability of the 3D counterpart, while at the same time enabling the tunability of the aforementioned properties through a spacer to perovskite layer ratio. However, theoretical studies of the layered 2D perovskites have been limited to the density functional level of theory (DFT) due to the lack of reliable force-fields that are necessary to explore the properties of this material observable only on a large scale. In this work, we employed the machine learning enabled Spectral Neighbor Analysis Potential (SNAP) to obtain the quantum accurate description of energies and forces in 2D layered Ruddlesden–Popper perovskite material, with butylammonium (BA) molecule included as a spacer. The trained SNAP potential reproduces both energies and forces of the reference atomic configurations with high fidelity and comparable with DFT calculations. Furthermore, the potential is stable at both 300 and 400 K which is verified for the first five 2D perovskite members under the canonical ensemble in bulk phase for 0.5 ns.