Environment Life

Milky Way simulation created by AI

Admin

November 26, 2025
3 min read

Researchers have created the first Milky Way simulation that can track over 100 billion individual stars across ten thousand years of evolution. The team was led by Keiya Hirashima at the RIKEN Center for Interdisciplinary Theoretical and Mathematical Sciences (iTHEMS) in Japan, working with partners from The University of Tokyo and Universitat de Barcelona in Spain. The research was presented at the international supercomputing conference SC ’25

This was achieved by using artificial intelligence (AI), paired with advanced numerical simulation techniques. The model includes one hundred times more stars than the most sophisticated earlier simulations and was generated over a hundred times faster. A similar technique could also be used on large-scale systems on Earth, such as climate and weather research.

Solar Flare Triggers Widespread Radio Blackouts

A powerful solar flare disrupted radio communication across parts of Africa and Europe, raising fresh concerns about solar storm impacts.

New Research Suggests the Universe May Be Slowing Down

A groundbreaking study challenges long-standing dark energy theories, proposing the universe may be decelerating rather than accelerating.

JWST Creates First Detailed Temperature Map of Distant Exoplanet

The James Webb Space Telescope has produced an unprecedented heat map of an exoplanet, offering new insight into atmospheric circulation beyond our solar system.

For years, astronomers had hoped to create simulations of the Milky Way that are detailed enough to follow each individual star. Doing so would allow them to compare different ideas about galactic evolution, structure, and star formation against data from observations. But simulating the entire Milky Way galaxy accurately would need calculating gravity, fluid behaviour, chemical element formation, and supernova activity over massive spans of time and space.

To overcome the challenges, the team designed a method that combined a deep learning surrogate model with standard physical simulations. The surrogate was trained with high-resolution supernova simulations, learning to predict how gas spreads over the 100,000 years after a supernova explosion, without further resources from the main simulation.

The AI aspect allowed the team to capture the overall behaviour of a galaxy, while still being able to monitor smaller-scale events, including the fine details of an individual supernova. This was validated by comparing its results against large-scale runs on supercomputers including The University of Tokyo’s Miyabi Supercomputer System.

Stay tuned to EyeOnLondon for the latest news and expert opinions.