Gaia and Hipparcos map possible companions around 100,000 stars

A catalog of possibilities

The study reports results for 105,412 stars observed by both Hipparcos and Gaia, recommending 96,915 for downstream analysis. Its central product is a set of Bayesian posteriors: probability distributions describing possible companions and their orbits after combining observations with stated starting assumptions.

Unlike a conventional discovery list, the catalog includes stars without strong companion signals. Those results describe which companions could remain hidden, information useful for planning observations and eventually studying companion populations. A non-detection is therefore a constraint, not proof that a star has no planets.

Combining several kinds of evidence

The researchers use the G23H catalog to combine Hipparcos–Gaia motion measurements, differences between Gaia’s second and third data releases, and excess scatter in Gaia’s astrometric measurements. Together, these probe departures from the motion expected for an isolated star.

They add Gaia brightness measurements and, where available, excess radial-velocity variability: scatter in measurements of motion toward or away from Earth. This is a catalog-level variability constraint, not a newly collected precision radial-velocity time series.

Nearby Gaia sources provide another check. A proposed wide stellar companion should have a plausible brightness, position, distance and motion. If Gaia should resolve such an object but no suitable source appears, that possibility becomes less likely. This is an imaging-like constraint from the source catalog, not new direct photographs of candidate planets.

The model also accounts for blended starlight. When two stars are unresolved, the measured position follows their combined center of light rather than the primary star alone. Stellar-evolution models connect mass and brightness to help distinguish luminous companions from objects treated as dark.

Allowing the companion count to change

Each system can contain zero, one, two or three modeled companions. The statistical sampling explores both their number and orbital properties, producing 4,096 saved draws per star. This carries uncertainty about multiplicity into the inferred masses and orbits instead of subtracting one best-fitting companion before searching for another.

The starting assumptions matter. The prior assigns half its probability to no companions, with the remainder distributed over one to three. Where observations provide little information, the results retain substantial dependence on those assumptions.

What the analysis finds

Sensitivity varies strongly between targets and orbital periods. At some period, 72,721 stars—75% of the recommended sample—reach a 95th-percentile upper limit below a companion-to-host mass ratio of 0.01. That does not exclude such companions at every separation.

The paper also lists 582 companion identities with existence probabilities of at least 99% and median inferred masses below 13 Jupiter masses. These are not 582 confirmed new planets: the existence probability does not itself establish a planetary mass, the distributions can be broad, and known companions are included.

Comparisons with reference catalogs recover stellar binaries and exoplanets with published true masses reasonably well. Planets known only through radial velocities are recovered poorly as a group, because many are too small or too close to their stars for these data to constrain.

Where false signals can arise

The authors tested their statistical pipeline using 1,000 simulated systems with realistic observing patterns and uncertainties. The final implementation passed their calibration criteria after an earlier test exposed a software bug that they corrected. Such tests establish internal consistency, not that every real star obeys the model.

Important exceptions remain. White dwarfs and companions that are themselves unresolved binaries can be too faint for their mass under the adopted brightness model, potentially prompting an unnecessary extra companion. Crowding, bright backgrounds and stellar activity can also mimic companion evidence.

Nearly equal-brightness binaries are particularly challenging: they may be detected while receiving incorrect orbital parameters. Some Hipparcos solution types are excluded from the recommended sample because the model does not faithfully describe them. Since those exclusions preferentially remove binaries, raw detection fractions cannot be interpreted as population occurrence rates.

A resource awaiting follow-up

The study is arXiv:2611.12841v1, posted November 9, 2026, and presented as a preprint. Candidate companions require further vetting, and many planetary-mass solutions have broad, complicated uncertainties.

The authors report interactive access to reduced posterior data and plan full archival hosting at the Canadian Astronomy Data Centre upon acceptance. The immediate value is a framework for choosing follow-up targets and preserving uncertainty—not a definitive census of newly discovered worlds.