Physics-Informed Neural Networks for Modeling Galactic Gravitational Potentials (NeurIPS Submission): Datasets and Trained Models

This archive contains the training and validation datasets, along with the corresponding trained network weights, for a physics-informed neural framework developed to model both static and time-dependent galactic gravitational potentials. Two datasets are provided. The first consists of a single snapshot (t=0) of a toy Milky Way (MW) + Large Magellanic Cloud (LMC) system and is used to evaluate the performance of the static model. The second is a time-dependent dataset describing an evolving MW + LMC system, comprising six training snapshots and eight evaluation snapshots spanning a total modeling period of 120 Myr. See load_instructions.md for a detailed description of the dataset structure and step-by-step instructions for loading the data and trained models.