QUEST-wannier_v40: physics-driven synthetic ARPES training corpus from JARVIS Wannier tight-binding models (Part 1/2)

QUEST-wannier_v40 is a large-scale synthetic angle-resolved photoemission spectroscopy (ARPES) corpus for training and benchmarking machine-learning models that extract quasiparticle self-energy effects (mass renormalization, scattering rate, band narrowing) from momentum–energy images. This is Part 1 of 2 (samples 000000–012499, shards 00–04); Part 2 (samples 012500–024999, shards 05–09) carries identical documentation and the full manifest. Each of the 25,000 samples is a simulated constant-ky E–k cut (kx ∈ [−0.4, 0.4] Å⁻¹ at a random ky, E ∈ [−3, 0] eV vs the Fermi level) through the Brillouin zone of one of 1,772 inorganic materials from the JARVIS-DFT Database of Wannier tight-binding Hamiltonians. The QUEST pipeline solves the Wannier tight-binding Hamiltonian, applies randomized FL/MFL/EPC self-energy models with a Kramers–Kronig-consistent real part, and synthesizes photoemission intensity I(k,ω) = Σn |Mn(k)|²·An(k,ω)·f(ω) with full coherent (chinook-aligned) dipole matrix elements. Images are clean (/max-normalized); noise and background are applied online by the consumer's code. All samples are non-spin-polarized (no SOC) and unsymmetrized (symmetrized=false). Each HDF5 file contains the clean image, bare and quasiparticle dispersions (ground truth), the complex self-energy Σ(ω) on the energy axis (ground truth), orbital-character weights, full matrix elements, and per-band forward visibility. See the attached README.md and data_dictionary.md for the complete schema, and manifest.csv for per-sample parameters. License & attribution: CC-BY-4.0. Users must cite this dataset and the upstream database: Garrity & Choudhary, Database of Wannier tight-binding Hamiltonians using high-throughput density functional theory, Scientific Data 8, 106 (2021), DOI 10.1038/s41597-021-00885-z. Limitations: the corpus is a frozen, validated artifact (25,000/25,000 files pass structural validation); it is not bit-reproducible with the current public code because the per-sample matrix-element configuration was not recorded in file metadata and the self-energy normalization constants were revised afterwards. Access by request; fully open upon journal publication.