Data
Files behind the figures, tables and numbers of the paper. The labels and descriptors are derived from the Alexandria phonon and electron-phonon release of 11 August 2025, which is published under CC BY 4.0 by Cavignac and coauthors.
- h_alexandria.json (35 KB). Census of the scattering strength per proton over the Alexandria electron-phonon release: counts, distributions, largest values.
- labels.csv (4,390 KB). One row per compound of the release with a spectral function: cell volume, hydrogen density, Hopfield sum, its hydrogen part, coupling constant, frequencies and the release's own transition temperatures.
- descriptors.csv (1,836 KB). Descriptors of each cell used in the census and by the model.
- summary.json (1 KB). Record counts of the release as read.
- h_alexandria_points.json (119 KB). Hydrogen density and Hopfield parameter of the 4,619 labelled hydrides, as plotted.
- h_census.json (177 KB). Scattering strength per proton from published tables and the megabar superhydrides.
- hydrogen_density.json (38 KB). Hydrogen number density of known and proposed hydrides.
- jellium.json (94 KB). A proton, and other nuclei, in an electron gas: phase shifts and scattering strength against density.
- results.json (44 KB). Solutions of the Eliashberg equations for single-mode and two-mode spectra.
- requirements.json (29 KB). Frequency and Hopfield parameter needed for a given transition temperature.
- spectra.json (4 KB). Hopfield sums and transition temperatures recomputed from published spectra.
- tails.json (15 KB). Tail of the calculated transition temperature near the convex hull.
- tails_checks.json (3 KB). Checks of that tail: other thresholds, shapes and cuts.
- frontier.json (3 KB). Ambient-pressure candidates: published range of calculated transition temperature, hull distance and outcome.
- hexahydride_neighbours.json (5 KB). Known hydrides of the alkali platinum and alkali nickel systems in the Alexandria PBEsol database, with hull distance and gap.
- records.json (2 KB). Record transition temperatures by year.
- record_claims.json (3 KB). Record claims since 1986 that survived, failed or are open.
- plane.json (3 KB). Published coupling constants and logarithmic frequencies.
- paper-v4.md (177 KB). Draft 4 of the first paper, as text, in which the tests of the model were fixed before any training.
- gao2025_fig2_wlog_lambda_tc.csv (226 KB). Marker coordinates read from a figure of Gao et al., Nat. Commun. 16, 8253 (2025), CC BY 4.0: frequency, coupling and transition temperature.
- gao2025_fig4_tc_ehull.csv (142 KB). Marker coordinates read from a figure of the same paper: transition temperature against hull distance.
Scripts, as text: numerics/eliashberg.py, numerics/h_alexandria.py, numerics/h_census.py, numerics/jellium.py, numerics/packing.py, numerics/requirements.py, numerics/selection.py, numerics/selection_checks.py, numerics/spectra.py, numerics/supercon_counts.py, numerics/tails.py, numerics/tails_checks.py, numerics/derived/closure.out.txt, numerics/derived/closure.py, numerics/derived/kinetics.out.txt, numerics/derived/kinetics.py, numerics/derived/pairing_megabar.out.txt, numerics/derived/pairing_megabar.py, numerics/derived/pairing_reductions.out.txt, numerics/derived/pairing_reductions.py, numerics/derived/pairing_single_mode.out.txt, numerics/derived/pairing_single_mode.py, numerics/derived/pairing_spectra_response.out.txt, numerics/derived/pairing_spectra_response.py, numerics/derived/pairing_transfer.out.txt, numerics/derived/pairing_transfer.py, numerics/derived/run_all.sh, numerics/derived/screening_selection.out.txt, numerics/derived/screening_selection.py, numerics/derived/test_statistics.out.txt, numerics/derived/test_statistics.py, model/dataset.py, model/deposit.py, model/deposit2.py, model/extract.py, model/fetch_alexandria.sh, model/leak_check.py, model/predict.py, model/screen_unlabelled.py, model/serve.py, model/structure_types.py, model/train.py, model/train2.py, model/train_final.py, model/deploy/Dockerfile, model/deploy/compose.yaml, figures.mjs.
Deposited before the model was trained: manifest, splits, baseline predictions.