Artificial intelligence has helped astronomers from the University of Warwick find hundreds of previously hidden exoplanets in data from NASA's TESS mission. The new RAVEN system processed observations of more than 2,2 million stars from the telescope's first four years of operation.
RAVEN combines automatic signal detection, machine learning and statistical validation in a single pipeline. As a result, the team confirmed 118 new planets and identified more than 2000 high-quality candidates, almost a thousand of which were previously unknown.
Among the discoveries are ultra-short-period worlds that orbit their star in less than 24 hours. Such objects are especially interesting because they exist under extreme conditions.
The analysis showed that 9–10% of Sun-like stars have close-in planets. This is consistent with Kepler data, but with ten times greater precision. In addition, for the first time it was possible to accurately measure the rarity of the “Neptune desert” — a region where planets the size of Neptune are almost never found. Such worlds have been detected around only 0,08% of solar-type stars.
“Using our new RAVEN pipeline, we were able to confirm 118 new planets and more than 2000 candidates, almost 1000 of which are completely new,” noted Dr Marina Lafarga Magro, lead author of the study.
The team also discovered previously unknown multi-planet systems with tight orbits. The resulting catalogues and tools are already available to other researchers for planning observations.
How will these data influence the selection of targets for future missions, such as the European Space Agency's PLATO?
The work has been published in Monthly Notices of the Royal Astronomical Society. The scientists emphasise that a clean and well-characterised sample will make it possible to better understand how planetary systems form and evolve near stars.
