GetGaia 2.0 separates stellar companions from the Milky Way foreground

A distant dwarf galaxy can hide in plain sight. Its stars appear among thousands of foreground stars in the Milky Way, some with similar colors and apparent motions. Before astronomers can measure how the galaxy moves, they must decide which points of light belong together.

In a preprint posted on November 4, Andrés del Pino of the Instituto de Astrofísica de Andalucía describes GetGaia 2.0, a Python package designed to automate that task. Applied to Gaia’s third data release, the software reports collective motions for 55 dwarf-galaxy fields and 93 globular clusters. The results broadly match published measurements, but the paper also documents situations where automation cannot reliably distinguish the target from its surroundings.

Combining imperfect clues

Gaia measures proper motion—the slow angular drift of a source across the sky—and parallax, the apparent positional shift associated with Earth’s orbit. Neither measurement alone guarantees membership. A faint satellite star may have large uncertainties, making it compatible with both the target and unrelated foreground populations.

GetGaia 2.0 replaces the earlier software’s sequential selection cuts with a joint probabilistic model. It combines motion and parallax with each star’s location on the sky and its color and brightness. Where available, velocities toward or away from Earth supply another clue, using Gaia measurements or an external spectroscopic catalog.

The model repeatedly updates its description of the stellar system and surrounding field, assigning membership probabilities along the way. Crucially, it uses each source’s measurement uncertainties and the correlations between its astrometric errors. A poorly measured star therefore does not receive the same evidential weight as a precisely measured one.

The color-and-brightness pattern is learned from the data rather than imposed through a predetermined stellar-population model. The output includes likely members, collective motion, correlated uncertainties, structural estimates and a contamination estimate derived from a surrounding control region.

How much prior knowledge is needed?

Del Pino tested the software in stages, beginning with only a target’s sky coordinates. Subsequent stages added a published distance, then published motion and structural information, and finally tailored settings where necessary.

Of 103 dwarf galaxies examined within 850 kiloparsecs of the Sun, the paper reports recovering 55, including 29 from coordinates alone. Among 95 globular clusters selected outside the most difficult crowded and reddened regions, it recovered 93, with 86 requiring only coordinates.

The Hercules dwarf illustrates the challenge: the software selected 31 members among roughly 63,500 sources. Their individual motions were too uncertain to establish high membership probabilities alone; their concentration on the sky and shared color-and-brightness sequence supplied essential information.

For 47 dwarfs, the reported proper motions differ from published values by less than 0.1 milliarcseconds per year. The median difference is 0.032 milliarcseconds per year for dwarfs and 0.009 for globular clusters. These comparisons demonstrate agreement, not access to an independently known true motion.

There is also an important qualification to the distant-galaxy tally. In the fields of M32 and NGC 205, the paper says the population identified is predominantly the disk of Andromeda behind them. Those returned motions should not be interpreted as clean measurements of the two companions themselves.

Where the method reaches its limits

Synthetic catalogs provide a test with known membership. In the main simulation grid, selecting stars with membership probabilities of at least 0.9 produced samples with 98–100% purity, while retaining 78–99% of genuine members. Raising the selection threshold generally removes contaminants at the cost of losing real members.

Performance deteriorates when the target shares the foreground’s proper motion. Additional simulations found lower completeness, residual motion biases and, in some distant, dense cases, failure to recover the system. Intermediate membership probabilities were also somewhat optimistic rather than exact frequencies.

The spatial model assumes a uniform background, an approximation that becomes problematic near strong density gradients or overlapping systems. Many unsuccessful dwarf-galaxy measurements instead reflect Gaia’s brightness limit: too few usable stars remain to establish a convincing population.

Systematic errors matter even when thousands of members make statistical uncertainties tiny. GetGaia estimates local reference-frame errors using surrounding quasars, whose detectable motions should be negligible, and propagates their uncertainty into the reported collective motion.

Preparing for another release

Simulations using expected Gaia DR4 uncertainties suggest roughly halved statistical proper-motion errors. That is a forecast, not a result from DR4 observations; release-specific calibrations still need updating.

Although the paper describes GetGaia as public software, its data-availability statement says version 2.0 and the associated catalogs will be released upon acceptance. The preprint’s contribution is a tested route toward more consistent automated membership analysis—not a claim that difficult stellar backgrounds have been solved.