Accurate Crystal Structure Prediction of New 2D Hybrid Organic Inorganic Perovskites

We have included the machined-learned interatomic potential (MLIP) using MACE trained to a dataset of hybrid organic-inogranic perovskites. The model*_swa.model are the trained potentials that can be used in either molecular dynamics or geomtery optimizations. The three models only differ in the random seed that was used in initializing the weight of the graph tensor network. They can be used together to determine the uncerainty of the model in predicing the energy or forces of a given system.The entire training samples are included as an xyz file (all_configurations.xyz), with lables for energy, forces, stress calcualted using density functional theory (DFT). Details of the compositions used in the training, including the DOI references are included in composition_dataframe.pkl. Structural files saved as train.xyz and test.xyz correspoding to the compostions that are in the training and test set, respectively.

The experimentally synthesized structure, together with the top 6 minima predictions are included in the synthesized_structure.zip.

Further details and results are available in the arxiv: https://doi.org/10.48550/arXiv.2403.06955