Identifying split vacancy defects with machine-learned foundation models and electrostatics

Dataset accompanying the publication "Identifying split vacancy defects with machine-learned foundation models and electrostatics" in JPhys Energy, 2025 (https://doi.org/10.1088/2515-7655/ade916), including: Summary dataframes (in `csv.gz` format) of all materials investigated (i.e. all compounds on the Materials Project), giving the chemical formula, MP ID, space group number, and summary info on predicted split vacancy configurations from electrostatic ("ES") and MLIP ('MACE') analyses. Can be loaded with `pandas.read_csv()`, or decompressed and opened with Excel etc. Jupyter notebooks and parsing scripts used in this work Raw DFT calculation outputs Relevant parsed data files In addition, at 10.5281/zenodo.15830542 you can find: - Individual `json.gz` files for each material investigated (from the Materials Project), containing information on the split vacancy configurations (with site positions, point symmetries, and multiplicities), initial and MLIP-relaxed structures, calculated electrostatic formation energies and more. These can be loaded with `loadfn` from `monty.serialization` (as shown in notebooks), or with `json` etc. These also include the DFT calculated relative energies in applicable cases. The lean version of these individual `json.gz` files is hosted at https://github.com/kavanase/Split_Vacancies, for use with `doped` for auto-generation of pre-calculated split vacancy states. Feel free to contact me (`sk2045[at]cam.ac.uk`) if you have any issues or questions.