AI analysis of Gaia data yields unprecedented haul of hot subdwarf stars

Researchers from the Faculty of Physics together with colleagues from the Faculty of Mathematics have applied artificial‑intelligence methods to sift through the European Space Agency's Gaia catalogue, uncovering thousands of hot subdwarf stars that had escaped detection until now. This discovery vastly enlarges the roster of these compact, extremely hot objects and showcases how machine‑learning can transform massive stellar surveys.

Hot subdwarfs are diminutive yet bright stars whose surface temperatures far surpass those of ordinary main‑sequence stars of similar size. They are believed to be helium‑core burners that have been stripped of most of their hydrogen envelope, often as a result of interaction with a binary companion. Because of these atypical traits, they serve as important probes of late‑stage stellar evolution and binary dynamics.

The team built a supervised learning classifier using a training set of confirmed hot subdwarfs, teaching the algorithm to detect the subtle mix of Gaia‑measured brightness, colour, and parallax that characterises the class. The trained model was then run on Gaia's third data release, which provides precise measurements for over a billion objects. By selecting entries that matched the learned signature, the AI identified several thousand new hot subdwarf candidates.

The partnership merged astrophysical knowledge with cutting‑edge data‑science skills. Physicists supplied the theoretical criteria that define a hot subdwarf, while mathematicians honed the classification algorithm and assessed its performance across the enormous dataset. This interdisciplinary strategy was crucial for processing the vast Gaia inventory without compromising accuracy.

The expanded catalogue opens fresh avenues for evaluating competing formation scenarios for hot subdwarfs. In particular, it may help disentangle the roles of common‑envelope ejection, stable mass transfer and stellar mergers in removing a star’s outer layers. A more complete inventory also aids the interpretation of the ultraviolet excess observed in aged stellar populations, where hot subdwarfs are thought to be the dominant contributors.

Upcoming efforts will centre on spectroscopic follow‑up to verify the temperatures and surface compositions of the candidates. Ground‑based observatories equipped with high‑resolution spectrographs are scheduled to test the AI‑selected objects, while further tweaks to the machine‑learning pipeline aim to cut down false positives. As additional Gaia data and observations from future missions arrive, the combination of AI and astronomy is poised to uncover more rare stellar types, deepening our understanding of stellar life cycles.