Possibilities and Limitations of Kinematically Identifying Stars from Accreted Ultra-Faint Dwarf Galaxies

While Gaia kinematics data has unlocked a better understanding of the Milky Way’s major mergers, the history of low-mass mergers involving ultra-faint dwarf galaxies (UFDs; L<10⁵ solar luminosities) remains poorly understood. Uncovering this history requires searching for the remnants of these accreted UFDs in the Galaxy’s stellar halo. A common method of doing this search uses stellar kinematics. Because the kinematics of halo stars retain information about their accretion origins, a cluster of halo stars in kinematic phase space may be the remnant of a progenitor UFD. However, the strength (or weakness) of this kinematic cluster-UFD remnant correspondence has not been thoroughly quantified.

To investigate how well kinematic clustering recovers UFD remnants, we test seven clustering algorithms on a large suite of 32 Milky Way-like halos from the Caterpillar simulations. First, we find that kinematic clustering recovers only 0-2% of all UFD remnants. We also find extremely high false positive rates of ~90% across all algorithms. Second, we find that among the algorithms—all of which perform poorly—HDBSCAN performs best. It recovers ~2% of UFD remnants within 5 kpc of the Sun and has a false positive rate of ~80%. Lastly, we conclude that although kinematic clustering is severely limited and error-prone, it becomes more reliable if applied to a sample of remnants with higher energy and more recent accretion times.