Quantcast
ESA/Hubble & NASA/Creative Commons

New York cosmologist leads study that uses AI analysis of one galaxy to infer properties of others

Galaxies have many different features such as stellar mass, gas metallicity and star-formation rate. To study these properties, cosmologists have traditionally studied a wide diversity of galaxies.


Marjorie Hecht
Apr 1, 2022

Galaxies have many different features such as stellar mass, gas metallicity and star-formation rate. To study these properties, cosmologists have traditionally studied a wide diversity of galaxies.

However, in a fortuitous sequence of discoveries cosmologists learned that in simulated galaxies the internal properties of one galaxy allowed the cosmic density of other galaxies to be inferred within 10%.

The team of researchers hopes to perfect the method in order to help astronomers more easily determine the features of real galaxies. It will also shed light on the relationship of cosmology, the study of the universe as a whole, to astrophysics, which looks at the physics of individual objects.

The international group of cosmologists, led by astrophysicist Francisco Villaescusa-Navarro of the Flatiron Institute in New York, reported their work in arxiv, an open-access preprint archive, Jan. 6. 

The Flatiron Institute's mission is to advance scientific research using computational methods and artificial intelligence.

The paper addressed the question: "... can we infer the value of the cosmological parameters from a single, generic galaxy?" The researchers looked specifically at stellar mass, gas metallicity, the rate of star formation, the total mass in the galaxy's subhalo and the stellar radius. 

The database the researchers used comprises hundreds of thousands of galaxies from 2,000 digital universes generated by the Cosmology and Astrophysics with Machine Learning Simulations project, known as CAMELS. These state-of-the-art hydrodynamic galaxy simulations include different cosmologies and astrophysical models.

The researchers publicly released all of the data used, plus the codes, allowing other researchers can reproduce the results. 

Surprising results

The researchers describe the model results as "surprising" and at first didn't believe their findings. They made use of several neural network algorithms to look for patterns among the simulated galaxies. Unexpectedly, they found that once they knew the internal properties of a single galaxy, their models could infer the mass of the universe involved, Ωm

The Ωm is important because, the article notes, understanding the dark energy provides the key to the rate at which the universe is expanding.

The research group then ran different machine learning algorithms to determine that the results were not dependent on a particular algorithm and to "provide some interpretability."

They concluded that "there is evidence showing that the value of Ωm can be inferred from the properties of individual galaxies for the vast majority of cases." The researchers further conclude that this holds true for "galaxies with very different cosmologies, astrophysics and almost independently on whether the galaxy is massive or dwarf central or satellite...."

In future work the researchers will investigate "the improvement on the parameter constraints when considering several galaxies instead of just one." They also intend to investigate some astrophysical parameters.

____

Francisco Villaescusa-Navarro et al. "Cosmology with one galaxy?" arxiv, Jan, 6, 2022. 

https://doi.org/10.48550/arXiv.2201.02202


RECOMMENDED