SkyScan

Stellar streams are elongated structures of stars that form from the tidal disruption of dwarf galaxies or globular clusters orbiting the Milky Way. Although simulations predict that on the order of a thousand streams should be detectable today, only about a hundred have been identified so far, leaving many still hidden. These undiscovered streams could provide crucial insights into the formation history of the Milky Way and into the structure of its dark matter potential—one of the greatest unknowns of our Galaxy.

Detecting stellar streams is particularly challenging: each stream typically consists of only a few hundred stars, stretched thinly across the sky. These stars form narrow, line-like structures and exhibit strongly localized proper motions due to their coherent orbital motion.

This localization in proper-motion space makes stellar streams ideal targets for classification without labels (CWoLa) in combination with machine-learning–based template interpolation methods. We employ TRANSIT and RAD-OT as fast template-building techniques to interpolate stellar features across the Gaia sky, using neighboring regions as references. The interpolated templates are then compared to the original Gaia data to identify anomalous stars, which are subsequently clustered to assemble line-like stream candidates.

These modern machine-learning approaches, building on existing stream-identification techniques, show strong promise for constraining the remaining unknowns of our Galaxy and shedding light on the nature of dark matter.

Researchers involved: 
Jona Ackerschott, Debajyoti Sengupta, Stephen Mulligan, Ivan Oleksiyuk, Frank Rothen, Tobias Golling.