Training sets for A unified microscopic picture of cation and anion migration in MAPbI3

The training sets of the MAPbI3 machine learning potentials used in arXiv.2605.02685 are provided. These training sets, in combination with the training parameters provided in the publication, can be used for generating the force fields. The training sets are provided in the .extxyz format, with the stress tensors and total energy written in the header, and forces written in the last three columns for each structure. In the publication, these sets are used for training Allegro models, but in principle, this data can be used to train any interatomic potential that reads .extxyz as a valid input training format. The training sets are: training_set_MAPbI3_charged_defects.extxyz: Includes negatively charged iodide interstitial (I_I^-), positively charged iodide vacancy (V_I^+), positively charged methylammonium interstitial (I_MA^+), and negatively charged methylammonium vacancy (V_MA^+). training_set_MAPbI3_PT.extxyz: Phase transition model training_set_MAPbI3_neutral_MA_interstitial.extxyz: Neutral methylammonium interstitial (I_MA^0) training_set_MAPbI3_neutral_MA_vacancy.extxyz: Neutral methylammonium vacancy (V_MA^0)