Astronomers found six runaway stars when they applied a neural network to data from the European Space Agency’s Gaia mission, which is set to map a billion stars across the Milky Way and beyond.
If you’ve ever tried to look for a needle in the proverbial haystack, you might have wished for a computer that could do it for you. A team of European astronomers recently trained an artificial neural network to do just that — where the haystack is our galaxy, and the needle a runaway star.
Presenting at the European Week of Astronomy and Space Science in Prague, Czech Republic, Elena Maria Rossi and Tomasso Marchetti announced their finding of six hypervelocity stars flying away from the galactic center of the Milky Way at speeds greater than 350 kilometers per second (more than 780,000 mph). One of these stars has a high probability of escaping the Milky Way.
Hypervelocity stars are in high demand not only because of their origin in our galaxy’s hard-to-observe core but also because they can help us map the gravitational field of our galaxy, including its dark matter distribution. Previous efforts to find hypervelocity stars have focused on spectra — since most stars in the bulge and halo are typically old and red, any younger, bluer, interlopers would be prime candidates for recent ejectees from the Milky Way’s center. However, this method assumes the stellar properties of hypervelocity stars are all the same, which may not be the case. Other studies have tried to identify hypervelocity stars by their proper motion (their motion across the sky) and distance, but the measurements are too uncertain to say for sure where the stars originated.
ESA’s Gaia mission is changing all this. Gaia is a space-based telescope launched in 2013, designed to produce the largest and most accurate 3D map of stars in our galaxy to date. Its final catalog, to be released in 2022, will consist of precise positions, parallaxes, and proper motions for more than a billion stars — an order of magnitude more stars than covered by any current survey.
As a spiral galaxy, the Milky Way consists of a flat, pancake-like disk with a central spherical bulge. A spherical halo containing hundreds of thousands of stars surrounds the galaxy, but the vast majority of stars are located in the disk and bulge, orbiting at speeds between 210 and 240 km/s. Stars in the halo move more slowly, at roughly 150 km/s. It seems like it should be easy to spot stars with velocities more than 350 km/s. But with more than a hundred billion stars in the galaxy and less than a million of them with well-known velocities, identifying runaway stars with respect to slower moving background stars is no small task.
That’s where machine learning comes in. Rossi and Marchetti created an artificial neural network, a computer algorithm loosely based on the way neurons talk to each other in the brain (though a far cry from true artificial intelligence), to help differentiate hypervelocity stars from their “normal” brethren.
First, the researchers created a mock data set of hypervelocity stars and inserted them into a simulated Gaia data set. They then “trained” the algorithm to recognize the difference between a hypervelocity star and a normal star. Once it had “learned” what to look for, the algorithm then took in real data from Gaia’s first data release, which came out last year, and selected suspected hypervelocity stars. Follow-up observations narrowed the 80 initial candidates down to the six presented in Prague.
While the measurements of these stars’ distances are still too uncertain to precisely track down their origins, the team awaits more precise measurements from future Gaia observations. Rossi, Marchetti, and their colleagues will have the opportunity to put their neural network to the test again in April 2018, when Gaia’s second data release is scheduled.
The following video shows the motions of stars in our galaxy from more than 1 million years ago to present day, as calculated from data from the Gaia and Hipparcos missions. (The video is similar to the one released last April showing stars' predicted motions into the future.)