Recently, astronomers have found hundreds of “dirty” white dwarf stars in our home galaxy, the Milky Way. These are white dwarfs that are actively absorbing planets in their orbits. They are a valuable resource for studying the internal workings of these distant, destroyed planets. They are also difficult to find. Historically, astronomers had to manually sift through mountains of survey data to find signs of these stars. Further observations would either confirm or disprove their suspicions.

A team led by University of Texas at Austin graduate student Malia Kao sped up the process using a new form of artificial intelligence called multi-faceted learning, achieving a 99% identification success rate. The results were published July 31 Astrophysics Journal.

White dwarfs are stars in the final stages of life. They have exhausted their fuel, shed their outer layers into space, and are slowly cooling. Our Sun will one day become a white dwarf, but not for another 6 billion years.

Sometimes planets orbiting a white dwarf are pulled in by the gravity of their star, torn apart, and destroyed. When this happens, the star becomes “contaminated” with heavy metals from the planet’s interior. Since the atmospheres of white dwarfs are almost entirely hydrogen and helium, the presence of other elements can be reliably attributed to external sources.

“For dirty white dwarfs, the interior of the planet has literally burned onto the surface of the star so we can look at it,” Kao said. “Contaminated white dwarfs are the best way we can characterize inner planets right now.”

“In other words, it’s the only sure way to find out what kind of material exoplanets are made of, which means finding dirty white dwarfs is critical,” added Keith Hawkins, an astronomer at UCLA and co-author of the paper.

Unfortunately, the signatures of these stars, defined by the amount of polluting metals in their atmospheres, can be subtle and difficult to detect, and astronomers need to find them in a relatively short time.

Although astronomers can identify these stars by manually examining astronomical survey data, this can be time-consuming. To test the faster process, the team applied artificial intelligence to data from the Gaia space telescope. “Gaia provides one of the largest spectroscopic surveys of white dwarfs to date, but the data is at such low resolution that we thought it would be impossible to find contaminated white dwarfs,” Hawkins said. “This study shows what you can do.”

To find these elusive stars, the team used an artificial intelligence technique called multivariate learning. With its help, the algorithm looks for similar features in the dataset and combines similar elements into a simplified visual diagram. The researchers can then examine the table and decide which clots warrant further investigation.

Astronomers created an algorithm to rank more than 100,000 possible white dwarfs. Of these, a group of 375 stars looked promising: they showed the key feature of having heavy metals in their atmospheres. Subsequent observations using the Hobby-Eberly Telescope at UT’s McDonald Observatory confirmed the astronomers’ suspicions.

“Our method could increase the number of known contaminated white dwarfs by a factor of ten, allowing us to better study the diversity and geology of planets outside our solar system,” Kao said. “Ultimately, we want to determine whether life could exist outside our solar system. If our system is unique among planetary systems, it may also be unique in its ability to support life.”

This innovative approach is just one example of how researchers at the University of Texas at Austin are using AI to solve scientific mysteries. To promote and showcase these innovations, UT Austin has declared 2024 the Year of Artificial Intelligence.

This study used data from the European Space Agency (ESA) Gaia mission. The data was processed by the Gaia Data Processing and Analysis Consortium.

Follow-up observations were obtained with the Hobby-Eberly Telescope (HET), a joint project of the University of Texas at Austin, Pennsylvania State University, Ludwig Maximilian University of Munich, and Georg-August University of Göttingen, as well as the Very Large Telescope (VLT) at the European Southern Observatory (ESO).

The Texas Advanced Computing Center at UT Austin provided high-performance computing, visualization, and storage resources for this work.