A group scientists may have stumbled upon a radical new way to study cosmology.
Cosmologists usually determine the composition of the universe by observing it as much as possible. But these researchers have discovered that a machine learning algorithm can scrutinize a single simulated galaxy and predict the overall structure of the digital universe in which it exists — a feat is similar to analyzing a random grain of sand under a microscope and working out the mass of Eurasia. The machines seem to have found a pattern that could one day allow astronomers to draw far-reaching conclusions about the real universe simply by studying its elemental building blocks.
“This is a completely different idea,” says Francisco Villaescusa-Navarro, a theoretical astrophysicist at the Flatiron Institute in New York and lead author of the work. “Instead of measuring these millions of galaxies, you can take just one. It’s really great when it works. ”
It’s not that. The unpredictable discovery arose from an exercise Villaescusa-Navarro gave Jupiter Ding, a Princeton University student: Build a neural network that, knowing the properties of the galaxy, can estimate a some cosmological properties. The exercise is only intended to familiarize Ding with machine learning. Then they noticed that the computer was nailing down the overall density of matter.
Villaescusa-Navarro said: “I think the students made a mistake. “Honestly, it’s a bit hard to believe for me.”
The results of the ensuing investigation appeared in print before January 6 has been submitted for publication. Researchers analyzed 2,000 digital universes created by Cosmology and Astrophysics with Machine Learning Simulation (CAMELS) project. These universes have a wide range of compositions, containing between 10 percent and 50 percent matter with the rest made up of dark energy, which drives the universe to expand faster and faster. (Our actual universe consists of about one-third dark and visible matter and two-thirds dark energy.) As the simulation runs, dark matter and visible matter swirl into each other into particles. Galaxy. The simulations also include rough treatments for complex events such as supernovas and jet ejections from supermassive black holes.
Ding’s neural network studied nearly 1 million simulated galaxies in these diverse digital universes. From a divine perspective, it knows the size, composition, mass of each galaxy, and more than a dozen other characteristics. It tries to relate this list of numbers to the density of matter in the parent universe.
It was successful. When tested on thousands of new galaxies from dozens of universes it had never tested before, the neural network was able to predict the cosmic density of matter to within 10 percent. “It doesn’t matter which galaxy you’re looking at,” says Villaescusa-Navarro. “No one imagined this could happen.”
“A possible galaxy [the density to] 10% or so, that’s very surprising to me,” Volker Springelan expert in simulating galaxy formation at the Max Planck Institute for Astrophysics, who was not involved in the study.
The performance of the algorithm astounded researchers because galaxies are inherently chaotic objects. Some form all in one go, and others grow by eating their neighbors. Giant galaxies tend to hold onto their matter, while supernovas and black holes in dwarf galaxies can eject most of their visible matter. However, every galaxy has somehow managed to keep a close eye on the overall density of matter in its universe.
One interpretation is that “the universe and/or galaxies are in some ways much simpler than we imagine,” Pauline Barby, an astronomer at Western University in Ontario. Another thing is that the simulations have unrecognized flaws.
