Creating a 3x More Precise Dark Matter Distribution Map Around the Milky Way Using Artificial Intelligence

International Collaborative Research Team Led by Korea Astronomy and Space Science Institute

Galaxy distribution within a radius of approximately 2 billion light-years centered on the Milky Way observed through the Sloan Digital Sky Survey (SDSS) (Source: SDSS Legacy Survey)

[Asia Economy Reporter Kim Bong-su] A dark matter distribution map around the Milky Way, three times more precise than previous ones, has been created using artificial intelligence.

On the 27th, the Korea Astronomy and Space Science Institute unveiled a dark matter distribution map around the Milky Way that is more than three times as precise as previous studies by applying artificial intelligence to external galaxy data surrounding the Milky Way. Dark matter is a substance that is indirectly inferred to exist in the universe through gravity due to its mass. It neither emits nor reflects light, making it invisible to the naked eye. This dark matter is estimated to account for about 26% of the energy composing the universe.

An international joint research team led by Dr. Hong Sung-wook of the Korea Astronomy and Space Science Institute applied deep learning technology to information from about 1,900 external galaxies to predict the density distribution of dark matter within 100 million light-years from the Milky Way. Through this result, the detailed structure of the large-scale structure of the universe around the Milky Way with a resolution of about 3 million light-years was confirmed.

3D distribution and motion direction of dark matter within 100 million light-years around the Milky Way predicted by artificial intelligence.

To train the artificial intelligence model, the research team utilized a large-scale cosmic structure simulation called ‘Illustris-TNG.’ The trained dark matter prediction model reconstructed the filament structures between galaxies in great detail. In particular, when both the positions and spatial velocities of galaxies were input simultaneously, it was confirmed that the model could predict a very high level of dark matter distribution similar to existing simulations. To verify the performance of the trained dark matter prediction model, actual galaxy data within 100 million light-years around the Milky Way was applied, confirming that filament structures connecting known galaxy groups and clusters, such as the Local Group including the Milky Way and the Virgo Cluster, were well reproduced.

Artificial intelligence model used to predict the distribution of dark matter

The reason why revealing the distribution of invisible dark matter is important is that the cosmic web connecting galaxies is mostly composed of dark matter. The distribution of dark matter forms the fundamental framework of cosmic expansion models that explain how each galaxy in the universe was formed in the past and how it will evolve in the future. Previous attempts to reconstruct cosmic web maps required establishing hypotheses about early universe models and simulating billions of years of cosmic evolution, which demanded enormous computational power and resources, making it impossible to view detailed dark matter distribution around the Milky Way. The significance of this study lies in efficiently reproducing dark matter distribution predictions by building probabilistic statistical models of various galaxy data through a completely new approach using deep learning technology.

Dr. Hong Sung-wook said, “With the operation of next-generation advanced astronomical observation equipment, new galaxies that have not been discovered so far will continuously be added to galaxy catalogs, which will further improve the reliability of dark matter prediction models. If detailed maps of the large-scale cosmic structure beyond the Milky Way can be obtained using the deep learning technology applied in this study, it will ultimately provide decisive clues to uncover the identity of dark matter, one of the major challenges in modern astronomy.”

The research results were published in the June 26 issue of The Astrophysical Journal.

This content was produced with the assistance of AI translation services.