Dataset: Accelerated Dopant Screening in Oxide Semiconductors via Multi-Fidelity Contextual Bandits and a Three-Tier DFT Validation Funnel

Dataset: Accelerated Dopant Screening in Oxide Semiconductors via Multi-Fidelity Contextual Bandits and a Three-Tier DFT Validation Funnel Contents dft/ — DFT Calculations (516 total) production/ — PBE-SCF calculations (245 systems across 5 oxide hosts) prospective_zno/ — 529-candidate ZnO co-doping campaign (229 unique calcs) relaxation/ — Ionic relaxation calculations (32 systems) hse06/ — HSE06 hybrid functional spot checks (5 systems) convergence/ — Basis set and k-point convergence tests (12 settings) Each subdirectory contains `pw.in` (Quantum ESPRESSO input) and `pw.out` (output). results/ — Parsed Results and Benchmarks results.csv — Prospective ZnO campaign: all 529 candidates with bandgaps step_log.csv — Step-by-step MF-OFUL decisions (DFT vs surrogate) qe_retroactive/ — 5-host retroactive oracle benchmark (223 candidates) prospective_mgo/ — MgO retroactive benchmark (49 candidates, 50 seeds) gp_ucb_benchmark/ — GP-UCB vs BO-GP-EI comparison thompson_benchmark/ — Thompson Sampling benchmark mace_comparison/ — MACE-MP-0 neural network potential comparison regret_bound/ — OFUL regret bound verification (60 traces) code/ — Screening Pipeline Source Code src/rr/ — Python package implementing all 10 screening methods suite.yaml — Main configuration file pseudos.yaml — Pseudopotential mapping lean4_proofs/ — Formal Verification (Lean 4) Basic.lean — Secretary problem optimal stopping OFULRegret.lean — OFUL regret bound MFLyapunov.lean — Multi-fidelity Lyapunov stability RiemannSum.lean — Supporting analysis lemmas DFT Settings Code: Quantum ESPRESSO 7.2, PBE-PAW Production: ecutwfc=40 Ry, 2×2×2 k-grid, Gaussian smearing 0.01 Ry Relaxation: ecutwfc=50 Ry, 3×3×3 k-grid, force threshold 0.001 Ry/Bohr HSE06: ecutwfc=40 Ry, 2×2×2 k-grid + 2×2×2 q-grid, mixing=0.25 Supercell: 2×2×2 (32-atom for wurtzite/rocksalt, 40-atom for perovskite/rutile) Reproducibility Install dependencies: `pip install -e .` in the `code/` directory.Run screening: `rr suite run --config code/suite.yaml` Citation If you use this dataset, please cite the accompanying paper. License CC-BY 4.0