About this site

• Catalogue release: wodvik/gaia-slow-vgrf-catalogue · concept DOI 10.5281/zenodo.20116134
• Catalogue tables served here are converted 1:1 from the release FITS/CSV products (catalogue_expanded_master.fits, catalogue_expanded_orbits_tierABC.csv, orbit Monte Carlo percentiles, McMillan-2017 comparison table). Gaia source_ids are handled as strings throughout — they exceed 2⁵³.
• Potential models are the release's own AGAMA files (Hunter+2024 axisymmetrised and full barred, Portail+2017; McMillan 2017 from AGAMA's distribution), exported as dense force/potential grids by this repository's pipeline.
• Context-star astrometry (Barnard's Star, Kapteyn's Star, Arcturus) is fetched live from the Gaia archive / SIMBAD at build time, with per-object provenance stored in the data files.
• Underlying survey data: ESA Gaia DR3; distances from Bailer-Jones et al. (2021); parallax zero-points from Lindegren et al. (2021); chemistry cross-matches from APOGEE DR17 and GALAH DR3.

The 3D viewer does not play back pre-rendered orbits — it integrates them in your browser, on interpolation grids sampled from the exact release potentials. Three validation gates run before any data ships:

• interpolated forces vs direct AGAMA evaluation at random points (typical relative error ~10⁻⁶, worst ~3×10⁻⁴);
• grid-integrated 4-Gyr test orbits vs AGAMA DOP853 in the pipeline (pericentre agreement ≲0.2 pc for the axisymmetric models);
• the browser engine itself, re-integrating the full Tier A+B+C sample and comparing against the published orbit columns:

The ICRS→Galactocentric transform used on the Methods page reproduces the release's astropy implementation to <1 milliparsec and <4 cm s⁻¹ over all 1,952 orbit-sample stars.

• Orbit quantities are model outputs, not observables. R_peri, R_apo, eccentricity, z_max and actions all depend on the assumed potential, the solar frame, and the 4-Gyr static-model idealisation.
• Point estimate ≠ posterior.The paper's headline Tier A+B+C medians are Monte Carlo posterior medians (e = 0.949, R_peri = 154 pc, R_apo = 7.35 kpc); the point-estimate medians (e = 0.964, R_peri ≈ 116 pc) are different numbers. Star pages show both.
• R_peri is cylindrical. It is the closest approach to the rotation axis; closest approach to Sgr A* itself is the separate spherical min_r_sph column. The paper finds zerostars with P(r<10 pc) > 0.5.
• Vertical actions are unreliable here. The Stäckel fudge degrades for near-radial orbits: J_z has a median fractional error of ~62% in this sample (1,164 of 1,952 stars flagged sampled_poor). Treat J_z as qualitative.
• Selection is severe. The Gaia RVS selection function is sparse exactly where these stars concentrate (83.2% of Tier A+B in prior-dominated cells); end-to-end injection recovery is 60.4%. Counts are lower limits on the underlying population, not measurements of it.
• Tiers are probabilities. Tier A+B is expected to contain ~31 false positives (score-implied purity 94%). A Tier C star is a coin-flip-to-5:1 candidate, not a confirmed slow star.
• The barred results are frame-dependent. Published barred columns use Ω_p = 37.5 km s⁻¹ kpc⁻¹ and a bar angle of −0.44 rad; the paper's inner-reach counts vary from 3 (static) to 7–31 across the Ω_p grid. The viewer's other Ω_p choices are sensitivity explorations, not published values.
• This site's live integrations reproduce the published columns to the tolerances in the table above — for anything requiring exact release values, use the release files themselves.

If you use the catalogue, cite Humble (2026) and the Zenodo DOI above. Catalogue data: CC BY 4.0. Release code: MIT. This website is a visualisation layer; when numbers here and the paper disagree, the paper and release files win.

Built with Next.js, react-three-fiber and AGAMA-derived data products. Site source: wodvik/GAIAweb.