dnagpt/riemann-clock-spectra

Riemann Clock spectra

This dataset accompanies the exploratory spectroscopy study in maris205/riemann_clock, with analysis base commit 321ae1cd48d69fe4d1091326167a8aaa65f2e597. It mirrors public, previously reduced, extracted, or coadded quasar spectra and the study's processed arrays. It contains no newly acquired observations and no raw CCD frames. The project directory name data/raw/ means downloaded inputs. This exploratory study reports no detection of Riemann-clock physics, varying fundamental constants, or a cosmological time signal.

Contents	Files	Bytes
ESPRESSO extracted, wavelength-calibrated exposure FITS	17	1,108,454,400
ESPRESSO final combined FITS	1	39,666,240
UVES SQUAD individual final-spectrum archives	34	218,173,262
Study processed arrays, metadata, and coverage table	424	260,049,012

The 52 unchanged source products total 1,366,293,902 bytes; the source and processed products together total 1,626,342,914 bytes. Provenance, licence copies, catalogues, and this card add a small amount of storage. Counts refer to files, not independent observations or independent measurements: the ESPRESSO combined spectrum and its 17 exposures reuse the same photons, and many processed windows reuse pixels from the same coadd. The UVES subset contains 34 sightlines, including HE 0515−4414 also observed with ESPRESSO.

Download and restore

From a current project checkout, restore all 52 source products to their original project locations with:

python code/fetch_huggingface_spectra.py

That recovery script uses the dataset revision and manifest checksum pinned in data/huggingface_release.json, verifies file hashes, and preserves the original source manifests. Add --include-processed to restore the processed files too, or use --verify-only to check an existing local restoration.

Install huggingface_hub, then download the complete dataset:

from huggingface_hub import snapshot_download

snapshot_download(
    repo_id="dnagpt/riemann-clock-spectra",
    repo_type="dataset",
    local_dir="riemann-clock-spectra",
)

For repeatable downloads, supply revision="<dataset commit SHA>" after selecting a published dataset revision. The Hugging Face viewer previews the 52-row catalogue.csv; the processed_files configuration previews a separate 424-row index. FITS, NPZ, and tar archives remain downloadable binary files. The CSV rows are file metadata, not one row per wavelength pixel.

manifest.json lists each distributed file except itself, with dataset_path, project_relative_path, byte count, SHA-256, category, and source family. Source products also carry their original URL, citation DOI, and existing provenance manifest references. The two CSV indexes carry the same source and processed file hashes. source_dataset_paths in processed records connects each product to its source download; several files have more than one source.

The dataset uses short source/ and processed/ paths. To restore the study's layout, copy each file with a non-null project_relative_path to that path relative to the cloned riemann_clock repository root, checking its size and SHA-256 first. For example, source/espresso_null/hes0515m4414.fits maps to experiments/highz_feasibility_2026-09-20/data/raw/espresso_null/hes0515m4414.fits. Keep existing files when their hashes agree and reject mismatches rather than silently replacing local changes. Dataset-only cards and indexes have null restoration paths. Additional study code, fit inputs, results, and documentation are available in the linked GitHub repository; this is a curated spectrum package, not a replacement for the entire project checkout.

Sources and attribution

ESPRESSO HE 0515−4414: the 17 extracted exposures and final combined spectrum come from the authors' repository, pinned at 09ee2cb32fb57ace9cda2943e997930bc7813366; data DOI 10.5281/zenodo.5512490. Cite Murphy et al. (2022), “Fundamental physics with ESPRESSO: Precise limit on variations in the fine-structure constant towards the bright quasar HE 0515−4414,” Astronomy & Astrophysics 658, A123, 10.1051/0004-6361/202142257, arXiv:2112.05819. The source release uses CC BY 4.0. Its original licence and README are retained.

UVES SQUAD DR1: the 34 final-spectrum .tar.gz downloads come from the Data Central spectrum portal, whose archived metadata retains its earlier Swinburne hosting URLs. The exact download URL for each file is in catalogue.csv. The supporting metadata/code repository is pinned at a0cdc8e7b99f2b01a45d919988af9d60e6447d19; data DOI 10.5281/zenodo.1345974. Cite Murphy, Kacprzak, Savorgnan & Carswell (2019), “The UVES Spectral Quasar Absorption Database (SQUAD) Data Release 1: The first 10 million seconds,” Monthly Notices of the Royal Astronomical Society 482(3), 3458–3479, 10.1093/mnras/sty2834, arXiv:1810.06136. The spectrum portal specifies Creative Commons Attribution Share-Alike without a version number. The metadata/code repository separately provides CC BY 4.0. These are distinct notices; this mirror does not relicense the spectra under the metadata repository's licence. See LICENSE.md.

Processed products and limitations

The 424 processed files are preserved exactly as present in the study:

• processed/archive_expansion/: 32 UVES coadd NPZ files, 32 metadata JSON files, and 324 absorption-window NPZ files for 36 absorber entries.
• The root processed/J034943-381030_* and processed/J051707-441055_* files: two UVES spectra, their metadata, and 12 Fe II windows. The latter target is HE 0515−4414, but these files come from UVES, not ESPRESSO.
• processed/feii_coverage_audit.csv: the initial two-target UVES coverage audit.
• processed/exposures/: 17 ESPRESSO native-pixel segment NPZ files and exposure metadata, with the study's masking and count scaling recorded there.
• processed/espresso_noise_controls.npz: ESPRESSO coadd control windows for the study's noise diagnostics.

The producing scripts are retained in the project's experiment code directory: fetch_public_spectra.py, archive_fetch.py, archive_expand_all.py, exposure_analysis.py, and empirical_noise_controls.py.

NPZ files use NumPy named arrays; inspect their array names and the accompanying metadata before analysis. Original FITS headers and upstream FITS format notes are retained. Source and processed wavelength conventions differ by product (for example, native ESPRESSO vacuum barycentric wavelengths and published UVES vacuum heliocentric coadds); follow each product's metadata. Processed masks reflect the specific study choices and do not guarantee an uncontaminated or precision-ready line sample. Coaddition, interpolation, blends, continuum choices, and shared photons can induce correlations. Complete pixel covariance and raw detector calibration material are not supplied by this package.

The selected archives are a study subset of SQUAD DR1, not the entire release or a statistically complete population sample. The archived source manifests and catalogues can mention upstream files beyond this package; only files in manifest.json are distributed here. No publisher PDFs, general software manuals, or unrelated analysis scripts are included.

Rebuild and verify

With the original project data restored, run the repository's python code/prepare_huggingface_dataset.py. It stages this release under the ignored build/huggingface_dataset/, verifies all 52 source hashes against existing download manifests, and checks every staged file. Re-running the builder with unchanged inputs produces the same metadata and manifest. Use --copy to stage independent copies; the default uses hardlinks where possible, so treat staged binary files as immutable. --verify-only checks an existing package. The script does not access the network or upload data.