A.I. outperforms astronomers, predicts whether exoplanets will survive

It’s been just thirty years since the first exoplanet was scientifically detected. At the time of this publication, astronomers have added 3,767 to the list.

Most of these far off planets are cruel and inhospitable places, but a few of them may have just the right conditions to harbor life. That is, they’re not too hot and not too cold for liquid water to exist. Like Goldilocks picking porridge, scientists think conditions have to fall between two extremes for life to take hold.

At its core, the search for exoplanets is the search for habitable exoplanets and a new system developed by astronomers at Columbia University may help made that hunt easier. Using machine learning algorithms, the researchers were able to make better predict about whether certain exoplanets could survive in stable orbits.

The work focused on “Tatooines,” or exoplanets that orbit two stars, much like Luke Skywalker’s desert home planet in Star Wars. These circumbinary planets, as they’re more formally known, can undergo huge orbital changes as they’re tugged between stars, sometimes causing them to get ejected from the system all together or crash into one of their host stars.

There’s an equation out there that astronomers use to determine longterm stability of a circumbinary planet, but lead researcher Chris Lam explained that it doesn’t give an accurate answer in all circumstances.

“The trouble is that motion becomes what physicists and mathematicians call ‘chaotic’ when you have three or more bodies in a system,” Lam, a recent graduate of Columbia University, told Digital Trends. “So there are some boundary cases where the equation predicts an unstable system where it’s stable and vice versa, and we honed in on that as something a neural network could potentially address.”

Predicting whether or not a planet gets flung out of its solar system may seem like little more than a galactic drinking game, but it actually determines life’s ability to exist. It takes billions of years for life as we know it to establish. There’s no hope for life on a planet floating aimlessly through space.

So to determine whether or not a Tatooine has survivability potential, Lam and his colleagues built a machine learning algorithm, which they trained on ten million simulated Tatooines. After a few hours and a bit of tuning, the system was able to outperform the conventional equation on “all metrics,” Lam said.

A paper detailing the study was recently published in the journal Monthly Notices of the Royal Astronomical Society.

Editors' Recommendations