AI-driven hunt through Gaia data uncovers record haul of hot subdwarf stars

Scientists from the Faculty of Physics and the Faculty of Mathematics have leveraged artificial‑intelligence techniques to comb through the European Space Agency's Gaia catalog, identifying thousands of previously unrecognized hot subdwarf stars. The effort dramatically expands the known population of these compact, intensely hot objects and demonstrates the power of machine learning in large‑scale stellar surveys.

Hot subdwarfs are small, luminous stars that burn at temperatures far exceeding those of ordinary main‑sequence stars of comparable size. They are thought to be helium‑core burners that have lost most of their outer hydrogen envelope, often through interaction with a companion star. Their unusual properties make them valuable probes of late‑stage stellar evolution and binary dynamics.

The research team trained a supervised learning model on a set of confirmed hot subdwarfs, teaching the algorithm to recognize the subtle combination of Gaia‑measured brightness, color, and parallax that typifies the class. The model was then applied to Gaia's third data release, which contains precise measurements for more than a billion objects. By filtering candidates that matched the learned pattern, the AI flagged several thousand new hot subdwarf prospects.

Collaboration between the two faculties blended astrophysical insight with advanced data‑science expertise. Physicists supplied the theoretical framework for what defines a hot subdwarf, while mathematicians refined the algorithm's classification criteria and evaluated its performance across the massive dataset. This interdisciplinary approach was essential for handling the sheer volume of Gaia entries without sacrificing reliability.

The enlarged sample offers fresh opportunities to test competing models of how hot subdwarfs form. In particular, it may clarify the relative importance of common‑envelope ejection, stable mass transfer, and mergers in stripping a star’s envelope. A richer inventory also aids in interpreting the ultraviolet excess observed in old stellar populations, where hot subdwarfs are thought to be a primary contributor.

Future work will focus on spectroscopic follow‑up to confirm the candidates' temperatures and surface compositions. Ground‑based telescopes equipped with high‑resolution spectrographs are slated to verify the AI selections, while refinements to the machine‑learning pipeline aim to reduce false positives further. As more data from Gaia and upcoming missions become available, the synergy between AI and astronomy is expected to reveal additional rare stellar classes, deepening our grasp of the life cycles of stars.