The neural network found 335 candidates in gravitational lens

Astronomers have discovered 335 new candidates for gravitational lenses among millions of images from ground-based overview of the DECaLS. This has been done by neural network which was trained on a sample of known objects of this type, and the objects, just non lenses. Article published in the journal The Astrophysical Journal.

The phenomenon of gravitational lensing was predicted in the framework of the General theory of relativity over a hundred years ago. Its essence is that the gravitational field of massive objects can change the direction of propagation of the photons, and most strongly this effect will be noticeable for very massive lenses, such as galaxies and clusters of galaxies. The lensing allows us to study the distribution of dark matter in the Universe, to obtain constraints on the value of the Hubble constant and to study very distant galaxies behind the lens, by exploring their multiple distorted images, however, we know of such objects is extremely small — only a few hundred.

A team of astronomers led by Saosin Juan (Xiaosheng Huang) published the results of a search for candidates for gravitational lenses in ground-based overview of the DECaLS (Dark Energy Camera Legacy Survey) with the residual neural network, developed in 2018 in the framework of the contest “Strong Gravitational Lens Finding Challenge”. As training samples we used the 423 known today gravitational lenses and 9451 of the object, just non lenses. Then the researchers tested 50 thousand images that were selected by the neural network of 5.7 million photographs from the review, and divided them into three classes. Class a was considered the most likely candidates for gravitational lenses and demonstrated one or more sections of the arcs (arches or arkleton), mainly blue, which are considered distorted images of galaxies behind the lens. Class B like class a, but had a smaller and dim sections of arcs. Finally, the class were objects that showed even more dull and shallow arches, if they are indeed lenses, the angle of deflection of photons from the original path in their case, is small enough.

In the end, scientists were able to find 335 candidates in a gravitational lens: 60 pieces of class A, 105 units of class b and 176 units of class S. Some of these objects have since been confirmed by observations of the space telescope Hubble, launched in late 2019. This is a good result, given that the training sample was small. Now astronomers are going to increase it and use a neural network for processing data related DESI, part of which was the DECaLS.

Previously we described how gravitational lensing has allowed scientists to find the sleeves from the most ancient spiral galaxy, a record of distant supernovae type Ia and the dim galaxy in the early Universe.

Leave a Reply

Your email address will not be published.