XBandgap: Experimental Bandgaps of 3,571 Inorganic Crystals with X-ray Diffraction Patterns and Structural Features

Abstract

XBandgap is a curated, machine-learning-ready dataset of 3,571 inorganic crystalline compounds, each pairing an experimentally measured electronic bandgap with its crystal structure and a suite of compositional and diffraction-based feature representations. It is designed for bandgap regression, metal/semiconductor classification, and X-ray-diffraction-driven property prediction. What makes it distinct: unlike the many bandgap collections built from DFT-computed values, every bandgap here is experimentally measured (optical, UV-Vis, or electrical) and aggregated from peer-reviewed literature. This makes XBandgap a realistic benchmark for models intended to predict real, observable material properties. Highlights 3,571 compounds spanning 81 chemical elements. Experimental bandgaps in eV (range 0.00–11.70, 58.9% metallic (Eg = 0), semiconductor mean 2.36 eV). 3,570 CIF crystal structures (Materials Project, COD). 132 Magpie composition descriptors per compound. Full simulated powder XRD profiles on a 1,701-point 2θ grid (5–90°). Three reproducible, leakage-free stratified splits (seed = 42). Contents metadata.csv — compound, averaged experimental bandgap (eV), measurement count. cif_files/ — 3,571 CIF structures (unit cell, space group, fractional coordinates, site occupancies). magpie_features.csv — 132 Magpie composition descriptors. structural_features.csv — lattice parameters, density, coordination statistics, crystal-system encodings. xrd_patterns.csv — simulated powder XRD intensities, 1,701-point 2θ grid (5–90°). dspacing_features.csv — top-20 XRD peak descriptors (d-spacing, intensity, 2θ, prominence). composition_labels.csv, lattice_labels.csv, spacegroup_labels.csv — auxiliary label tables. split_70_30.csv, split_semiconductors_70_30.csv, split_train_semi_test_30.csv — predefined splits. DATA_CARD.md, SPLITS.md, DISTRIBUTION.md — full documentation. Supported tasks Bandgap regression — predict Eg (eV) from composition, structure, or XRD. Metal vs. semiconductor classification — Eg = 0 against Eg > 0. XRD-based property prediction — learn directly from diffraction profiles. Provenance Experimental bandgaps are aggregated from Zhuo et al. (2018), The Journal of Physical Chemistry Letters 9(7), 1668–1673 (DOI: 10.1021/acs.jpclett.8b00124); compounds with repeated measurements are averaged. Crystal structures are sourced primarily from the Materials Project, with additional structures from the Crystallography Open Database (COD). Citation Tawfik, M., & Badawy, W. (2026). XBandgap: Experimental Bandgaps of Inorganic Crystals with X-ray Diffraction Patterns and Structural Features (Version 1.0) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.20777921 License: Creative Commons Attribution 4.0 International (CC BY 4.0).