Hot subdwarfs come into focus as AI scans thousands of stars in Gaia data

There are myriad ways of “being a star,” and scientists spend years trying to uncover and understand them. One unusual type is the so-called hot subdwarf—small stars that burn mysteriously hot for their size. Scientists from the Faculty of Physics and the Faculty of Mathematics and Informatics at Vilnius University (VU) are working to explain this strange phenomenon and recently made important breakthroughs.

The paper is published in the journal Astronomy & Astrophysics.

Can a star be too strange?

The space observatory Gaia, launched in 2013, cataloged more than half a billion stars. Of those, only 60,000 belong to a peculiar class called hot subdwarfs. That’s less than one in a thousand. To understand what makes hot subdwarfs so unusual, one has to picture the standard map astronomers use to classify stars, a diagram where temperature runs along one axis and brightness along the other.

“Most stars, including our sun, sit along a broad band called the main sequence: stable, hydrogen-burning objects in the quiet middle of their lives. Hot subdwarfs don’t belong there. They’re hot and lie below this main branch. That’s where the name comes from,” says Dr. Markus Ambrosch, a VU researcher at the Astrospectroscopy and Exoplanets Group.

Ambrosch and his colleagues are trying to explain why subdwarfs are the way they are. According to Dr. Carlos Viscasillas Vázquez, who initiated the research project, hot subdwarfs are among the most fascinating objects in stellar astrophysics because they seem to borrow properties from many different kinds of stars while not fully belonging to any of them.

“They are hot and blue like massive OB-type stars, known for their extreme temperatures and brightness, but are much smaller, less luminous and extremely compact. They are exposed stellar cores, in most cases formed when a companion strips away most of the star’s outer envelope, changing its evolutionary path and leading it toward the white dwarf stage through a nonstandard route,” he explains.

This “companion star” is the leading explanation for why subdwarfs came to be. Theory holds that hot subdwarfs are products of binary systems in which the more massive star swells into a red giant while the other remains compact and gravitationally hungry. The compact companion then strips away the giant’s outer hydrogen envelope, leaving behind just the bare core.

How machine learning helps us detect subdwarf stars

Understanding why hot subdwarfs behave in a certain way is one thing, but they must first be identified. This is a complicated problem. As mentioned before, hot subdwarfs are incredibly rare. The latest Gaia data release contained 60,000 hot subdwarf candidates for examination. The research group examined one-third of them with XP spectra.

The standard approach of fitting models star by star, adjusting temperatures and checking the fit was simply not feasible at that scale. To tackle this challenge, the team built a convolutional neural network. To train the algorithm, the team repeatedly presented 2,500 stars with known classifications to the network. Once trained, the AI processed the remaining 17,500 unknown stars in seconds.

“The network had taught itself without any physical equations. It didn’t know any physics when it started. It was just randomly initialized. But through machine learning, it learned that this part of the spectrum is important. We’re glad to see this because it confirms what we know from physics. It makes sense. We would also do it in the same way,” Ambrosch says.

He also emphasizes that one of the more rewarding aspects of the project was the opportunity to work with colleagues from different disciplines. For example, training the AI would have been impossible without input from Dr. Aidas Medžiūnas from the Faculty of Mathematics and Informatics.

“We focused on optimizing the use of the Gaia XP spectra database by applying smoothing and dimensionality reduction methods. We also applied functional machine learning methods to improve model interpretability, which is crucial for physical interpretation. It was an intriguing challenge,” Medžiūnas says.

A spirit of collaboration

Ana Ulla, a key member of the research team, a professor at the University of Vigo and a member of the Gaia Data Processing and Analysis Consortium, is one of the field’s leading experts. She has conducted research on hot subdwarf stars for more than 30 years and emphasizes the importance of the multidisciplinary nature of the process.

“This collaboration is especially promising because the team is heterogeneous and truly complementary, combining expertise in many topics. For me, it is a beautiful example of how scientific ideas travel, evolve and take root in new places through international collaboration. This collaboration in particular has become a genuine success story,” she says.

What began as a new research direction has quickly become a success story. In a short time, the team researching hot subdwarfs wrote two papers, obtained two grants, built strong international collaborations and, importantly, observed hot subdwarfs from Lithuania for the first time at the VU Molėtai Astronomical Observatory.

“I find this research interesting due to the complex nature of hot subdwarfs. They exhibit a mix of properties that make them difficult to analyze. However, this challenge makes the work rewarding, as every new finding reveals further directions for our research,” says Vladas Šatas, another member of the team.

VU scientists emphasize that this project has provided significant opportunities for professional growth within the broader science community.

“With the new project that began in September, the release of Gaia DR4 in December and an expanding international team, we now have a unique opportunity to study these rare stars on a large scale and understand how binary companionship can change the fate of stars,” Vázquez says.

Who’s behind this story?

Swati Mestri

Swati Mestri holds a bachelor’s degree in Electronics Engineering and has worked as a content editor since 2019. She has experience editing research documents across technology, health care, and materials science, and has a particular interest in technology and space.

Full profile →

Robert Egan

Bachelor’s in mathematical biology, Master’s in creative writing. Well-traveled with unique perspectives on science and language.

Full profile →