A team led by Keiya Hirashima has created the first simulation of the Milky Way tracking over 100 billion individual stars across 10,000 years. By combining artificial intelligence with advanced numerical techniques, the model runs hundreds of times faster than previous methods. The breakthrough was presented at the SC '25 supercomputing conference.
Researchers at the RIKEN Center for Interdisciplinary Theoretical and Mathematical Sciences (iTHEMS) in Japan, led by Keiya Hirashima, collaborated with partners from The University of Tokyo and Universitat de Barcelona in Spain to achieve a milestone in astrophysical modeling. Their simulation individually tracks more than 100 billion stars over 10,000 years of galactic evolution, representing 100 times more stars than the most advanced prior simulations.
The challenge of modeling the Milky Way stems from the need to compute gravity, fluid dynamics, chemical processes, and supernova events across vast scales while maintaining detail at the individual star level. Traditional simulations group stars into particles representing about 100 each, limiting accuracy for small-scale phenomena like supernovae. These require tiny time steps, making full simulations computationally intensive: modeling 1 million years of evolution takes about 315 hours, so 1 billion years would require over 36 years of real time.
To address this, the team integrated a deep learning surrogate model trained on high-resolution supernova data. This AI predicts gas behavior post-supernova over 100,000 years without burdening the main computation. Validated on RIKEN's Fugaku and The University of Tokyo's Miyabi supercomputers, the approach simulates 1 million years in just 2.78 hours, enabling a 1-billion-year run in about 115 days.
The work, detailed in a paper titled 'The First Star-by-star N-body/Hydrodynamics Simulation of Our Galaxy Coupling with a Surrogate Model' (DOI: 10.1145/3712285.3759866), was presented at SC '25. Hirashima stated, "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." He added, "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."
The technique holds promise for other fields like climate and weather modeling, where multi-scale simulations are essential.