An AI found dozens of gravitational lens galaxies

  • The gravitational field surrounding Abell 68 - a cluster of galaxies - acts as a natural lens. Image credit: NASA/ESA
  • A closer look at how gravitational lensing works. Image credit: NASA
Date:25 October 2017 Tags:, , , , ,

The algorithm accomplished in a few weeks what it has taken astronomers decades to do. It found 56 gravitational lens galaxies in a universe of billions of galaxies.

By Avery Thompson

There are roughly 100 billion galaxies in our universe, and it would take many lifetimes for humans to look at them all. Even if our scientists only focused on a tiny sliver of the night sky, there’s still enough space to keep any person busy for decades, if not centuries.

Among the many things astronomers could spend decades searching for are gravitational lens galaxies. When one galaxy sits in front of another one, the light from the more distant galaxy in the background can curve around the nearer one due to its gravity. This causes a magnification effect. But in a universe of 100 billion galaxies, finding these arrangements can be tough. So astronomers are increasingly turning to artificial intelligence to find gravitational lenses for them.

A closer look at gravitational lens galaxies:

gravitational lens galaxies and how it works

One group of astronomers, from the universities of Groningen, Naples and Bonn, have used just such an AI to find 56 gravitational lens candidates. Their research was recently published in the most recent issue of the Monthly Notices of the Royal Astronomical Society.

Finding gravitational lens galaxies isn’t easy. Two galaxies have to be almost perfectly aligned for a gravitational lens to occur, which makes them very rare. Further, even if a possible candidate is found, it requires scientists to analyse it closely. They make sure that it is in fact a lens instead of just an irregularly shaped galaxy or other interference. The result is that scientists have spent decades combing through galaxy images. They still only find a few dozen gravitational lenses in that time.

Study author Carlo Enrico Petrillo was growing frustrated and overwhelmed by the sheer amount of data to comb through. “We don’t have enough astronomers to cope with this,” he says. To fix the problem, he and his colleagues developed a convolutional neural network. The network is similar to the image-recognition algorithms used by Google and other companies.

They trained the algorithm to spot gravitational lenses, then set it loose on data from the Kilo-Degree Survey. Analyzing 255 square degrees of the night sky, the algorithm found 761 gravitational lens candidates, which the team reviewed and narrowed down to 56. While these candidates still have to be confirmed, they should dramatically increase the number of known gravitational lenses.

Other applications

Once even more gravitational lenses are discovered, astronomers can use other artificial intelligence systems to analyse them. A few months ago, a different team of astronomers at the SLAC National Accelerator Laboratory and Stanford University developed another AI to study gravitational lenses. The system is designed to analyse images of gravitational lenses to work out what the more distant galaxy looks like and some of the properties of the lensing galaxy in the foreground. With the combination of these two machine learning systems, and the increasing use of artificial intelligence in astronomy, the wonders that will be discovered in the deep universe are anyone’s guess.

Source: Netherlands Research School for Astronomy
From: PM USA