AI-Accelerated Simulation Tracks 100 Billion Milky Way Stars Over 10,000 Years
Researchers 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 the Universitat de Barcelona in Spain, have created the first Milky Way simulation capable of tracking more than 100 billion individual stars across 10,000 years of evolution.
The team achieved this breakthrough by combining artificial intelligence (AI) with advanced numerical simulation techniques. Their model includes 100 times more stars than the most sophisticated earlier simulations and was generated more than 100 times faster.
Presented at the international supercomputing conference SC ’25, the work represents a major advance for astrophysics, high-performance computing, and AI-assisted modeling. The same methodology could also be applied to large-scale Earth system studies, including climate and weather research.
Why Modeling Every Star Is So Difficult
For years, astrophysicists have sought to construct Milky Way simulations detailed enough to track each individual star. Such models would allow scientists to test theories of galactic evolution, structure, and star formation directly against observational data.
However, achieving this detail requires calculating gravity, fluid dynamics, chemical evolution, and supernova activity across vast time and spatial scales. This makes the task computationally enormous.
Until now, scientists have been unable to model a galaxy as massive as the Milky Way with single-star resolution. State-of-the-art simulations typically represent systems with the equivalent mass of around one billion suns—far below the galaxy’s more than 100 billion stars. In these models, the smallest “particle” often represents about 100 stars, averaging away critical small-scale behaviors.
The key challenge lies in timestep length: simulating fast events like supernovae requires extremely small increments. Shrinking the timestep dramatically increases the computational load.
Even with current top-tier physics-based models, simulating the Milky Way star-by-star would require about 315 hours for every 1 million years of evolution. At that rate, modeling 1 billion years would take more than 36 years of real time. Simply adding more supercomputer cores is not practical, as energy demands rise steeply and efficiency declines.
A New Deep Learning Approach
To break through this barrier, Hirashima and colleagues developed a method that combines a deep learning surrogate model with a conventional physical simulation. The surrogate was trained on high-resolution supernova simulations and learned to predict how gas behaves over the 100,000 years following a supernova explosion.
This approach allowed the main simulation to capture detailed, small-scale processes without directly computing them at every timestep. The team validated the method by comparing its results with large-scale runs on RIKEN’s Fugaku supercomputer and The University of Tokyo’s Miyabi Supercomputer System.
The hybrid technique achieves true individual-star resolution for galaxies containing more than 100 billion stars—and does so at unprecedented speed. Simulating 1 million years took just 2.78 hours, meaning that modeling 1 billion years could be completed in about 115 days instead of 36 years.
Broader Potential for Climate, Weather, and Ocean Modeling
This hybrid AI strategy may transform multiple scientific fields that require linking small-scale physical processes with large-scale behavior. Meteorology, oceanography, and climate science all face similar multi-scale challenges and could benefit from faster, more efficient simulation tools.
“I believe that integrating AI with high-performance computing marks a fundamental shift in how we tackle multi-scale, multi-physics problems across the computational sciences,” says Hirashima. “This achievement also shows that AI-accelerated simulations can move beyond pattern recognition to become a genuine tool for scientific discovery — helping us trace how the elements that formed life itself emerged within our galaxy.”