Al-Fe-Ni and Al-Fe-Ni-Co-Cr Alloy Electron-Density Datasets

This Zenodo deposit contains two data-only archives for volumetric electron-density prediction and representation-identifiability studies in alloy systems. The archives include curated atomic structures, DFT electron-density targets, metadata manifests, train/validation/test split definitions, and lightweight SIESTA provenance files. No neural-network model weights, generated audit bundles, manuscript figures, or manuscript source-data tables are included in this data-only upload. Files file size_bytes sha256 al_fe_ni_alloy_dataset.zip 1711953389 08b001fc5742b0442e9a0dcbaec455a0a6ed6d3c338c683f19955a8e2b281609 quinary_al_fe_ni_co_cr_dataset.zip 314276357 79d23516ce239aed410444d17fc0d88ba9586ddb1d28cb12f1cb3fbd5228750e checksums.sha256 194 checksums for the two ZIP archives Verify the downloaded archives with: sha256sum -c checksums.sha256 Archive Contents al_fe_ni_alloy_dataset.zip This archive expands to al_fe_ni/ and contains: README.md: dataset-specific documentation. manifest.csv and manifest.parquet: structure-level metadata. structures/structures_000.h5: HDF5 atomic structures. densities/densities_000.h5: HDF5 DFT density targets. splits/train.csv, splits/val.csv, splits/test.csv, splits/all.csv: grouped split definitions. dft_metadata/raw_light/: lightweight sanitized SIESTA .fdf, .out, STRUCT_OUT, and MESSAGES provenance files. Verified counts: Total structures: 4,002 Training: 3,204 Validation: 402 Test: 396 Alloy systems: Al-Fe, Al-Ni, Fe-Ni, and Al-Fe-Ni Stored structure format: HDF5 groups structures/{id}/cell, R_cart, and Z Stored density format: HDF5 groups rho/total/{id}/data, cell, cell_from_rho, and grid_shape quinary_al_fe_ni_co_cr_dataset.zip This archive expands to quinary_al_fe_ni_co_cr/ and contains: README.md: dataset-specific documentation. manifest.csv and manifest.parquet: structure-level metadata. structures/structures_000.h5: HDF5 atomic structures. densities/densities_000.h5: HDF5 DFT density targets. splits/train.csv, splits/val.csv, splits/test.csv, splits/all.csv: parent-grouped split definitions. dft_metadata/raw_light/: lightweight sanitized SIESTA .fdf, .out, STRUCT_OUT, and MESSAGES provenance files. Verified counts: Total structures: 500 Parent geometries: 100 Assignments per parent geometry: 5 Training: 350 structures from 70 parent geometries Validation: 75 structures from 15 parent geometries Test: 75 structures from 15 parent geometries Composition: Al2Fe2Ni2Co2Cr2 Structure description: a 10-atom supercell formed by a 5x1x1 replication of the two-atom parent cell This quinary dataset is a compact representation-identifiability stress test. It is not a comprehensive high-entropy-alloy benchmark, a state-of-the-art HEA benchmark, a transfer-learning benchmark, or a multi-seed benchmark. Data Conventions The HDF5 structure files store Cartesian coordinates and cells: R_cart: Cartesian atomic positions in Angstrom. cell: 3x3 simulation cell in Angstrom. Z: atomic numbers. The HDF5 density files store total SIESTA electron-density targets: rho/total/{id}/data: density array. rho/total/{id}/grid_shape: grid shape. rho/total/{id}/cell: cell associated with the density. Grid-axis ordering is (grid_nx, grid_ny, grid_nz). The exact text unit label for the stored density arrays was not recovered from project files. The arrays are the SIESTA total electron-density targets used by the training and evaluation pipelines. DFT Provenance The lightweight DFT provenance files record SIESTA calculation inputs and outputs needed to interpret the curated HDF5 data. Machine-specific absolute paths were sanitized in the uploaded copies. Verified DFT provenance: DFT code: SIESTA 5.4.2 Pseudopotential format used in the calculations: PSML 1.1 Pseudopotential generator provenance recorded in local files: ONCVPSP-3.3.0+psml-3.3.0-73, scalar-relativistic License This dataset is distributed under the Creative Commons Attribution 4.0 International license: https://creativecommons.org/licenses/by/4.0/ The PSML pseudopotential files used to perform the calculations are not redistributed as part of the dataset. Contact Irina ArévaloUniversidad Politécnica de Madrid, [e-mail]