SIESTA Single-Element Density Dataset

SIESTA Single-Element Density Dataset (30k configurations) This dataset contains 30,000 fully converged density-functional theory (DFT) calculations for elemental Al, Fe, and Ni (10,000 configurations per element), generated using the SIESTA electronic structure code. Each configuration consists of: - Atomic structure (`.fdf`, `.XV`, `.STRUCT_OUT`, `.ion`) - SCF output (`.out`) - Electron density grid (`Rho.grid.nc`) - Electrostatic potential grid (`ElectrostaticPotential.grid.nc`) All calculations are converged and validated. Dataset Scope Element Configurations Al 10,000 Fe 10,000 Ni 10,000 Total 30,000 Each configuration corresponds to a two-atom FCC-derived primitive cell subjected to controlled perturbations (see below). Structure of the Archive public_core/ ├── data/ │ ├── Al/ │ │ ├── cfg_0000/ │ │ │ ├── Al.fdf │ │ │ ├── Al.ion │ │ │ ├── Al.ion.xml │ │ │ ├── Al.out │ │ │ ├── Rho.grid.nc │ │ │ ├── ElectrostaticPotential.grid.nc │ │ │ ├── Al.XV │ │ │ └── Al.STRUCT_OUT │ ├── Fe/ │ └── Ni/ │ ├── index_core.jsonl ├── index_core.parquet └── README.md Configuration Generation DFT Calculations Calculations were executed using: - Code: SIESTA - Fully self-consistent single-point calculations - All SCF cycles converged - Automatic pseudo-copying and convergence validation The execution pipeline: - Skips already converged runs - Validates convergence via parser (`parse_out`) - Ensures `.RHO` was produced (internal check) - Uses controlled threading for reproducibility Grid Data Format Electron density and electrostatic potential are stored in: Rho.grid.nc ElectrostaticPotential.grid.nc NetCDF variables include: - `gridfunc` → 3D scalar field - `cell` → lattice vectors Example loading (Python) ``` python from netCDF4 import Dataset import numpy as np ds = Dataset("Rho.grid.nc") rho = np.array(ds.variables["gridfunc"][:]) cell = np.array(ds.variables["cell"][:]) ds.close() ``` Index Files `index_core.jsonl` One JSON object per configuration. Contains: - `calc_id` - `element` - `calc_dir` - `scf_converged` - `scf_iterations` - `final_energy_ev` - `fermi_ev` - File paths relative to archive `index_core.parquet` Columnar equivalent for efficient ML loading. Validation All configurations satisfy: - SCF convergence: **100%** - Missing density files: **0** - Broken index references: **0** - NetCDF readability: verified - Size consistency: no outliers - Deterministic configuration ordering Intended Uses - Machine learning on charge density fields - Learning inverse mappings (structure → density) - DFT surrogate modeling - Electronic structure benchmarking - Representation learning on periodic systems License This dataset contains: - Structural perturbations - DFT outputs - Electron density grids It does **not** redistribute pseudopotentials. CC BY 4.0 Citation If you use this dataset, please cite: > Irina Arévalo, Pablo Olleros, SIESTA Single-Element Density Dataset, Zenodo, > 2026. DOI: 10.5281/zenodo.18925343 Contact For questions, please contact: Irina Arévalo Universidad Politecnica de Madrid, Spain [e-mail] Pablo Olleros CUNEF Universidad [e-mail]