Astronomers have discovered more than 1,200 gravitational lens candidates, potentially doubling the number of similar objects known today. For this, the scientists used a residual neural network trained on a sample of more than 20 million images, according to an article accepted for publication in The Astrophysical Journal.
The gravity of massive celestial bodies, such as galaxies and their clusters, can change the direction in which electromagnetic radiation propagates. Being on the same line between the observer and the background source, they, like a lens, amplify the light of the latter, which allows scientists to study very distant and dim objects. Today, gravitational lensing is used to study the early universe and the distribution of dark matter in galaxy clusters. However, so far not so many such “lenses” are known – only a few hundred.
Xiaosheng Huang of the University of San Francisco and colleagues have published the search for gravitational lenses in the DESI (Dark Energy Spectroscopic Instrument) Legacy Imaging Surveys. They began their work back in 2018 and two years later submitted the first 335 candidates for gravitational lenses. Now scientists, using a similar method, have improved their results. Using a residual neural network developed as part of the Strong Gravitational Lens Finding Challenge, they found 1,210 new candidates.