Simulated Milky Way with AI and supercomputing sets star modeling milestone

by Riko Seibo
Tokyo, Japan (SPX) Nov 18, 2025
Researchers at the RIKEN Center for Interdisciplinary Theoretical and Mathematical Sciences in Japan, collaborating with The University of Tokyo and Universitat de Barcelona, have delivered the first simulation of the Milky Way that models over 100 billion individual stars spanning ten thousand years. The project utilized artificial intelligence alongside numerical simulations to achieve this scale and speed. According to the team, the result exceeds previous star-level models by a factor of 100, both in the number of stars represented and the speed of computation.

Current galaxy simulations have struggled to display high-resolution star modeling. Traditional approaches treat large clusters as single simulation particles and only estimate the evolution of individual stars. Simulating rapid changes, such as supernova events, has been limited by the time between computational steps. These limitations, combined with the vast resources needed for smaller time intervals, prevented accurate modeling of large galaxies.

To resolve this, lead researcher Keiya Hirashima and his group integrated a deep learning surrogate model into physical galaxy simulations. This AI model, trained on detailed supernova simulations, predicts gas expansion following a supernova event, without relying on the primary simulation for those calculations. This integration allowed the simulation to simultaneously portray both large-scale galactic dynamics and fine-scale phenomena. Verification runs were completed using RIKEN's Fugaku supercomputer and the Miyabi system at The University of Tokyo.

The new method achieved individual star resolution in the simulated Milky Way and reduced computational time dramatically. One million years of galaxy evolution could now be modeled in less than three hours, making billion-year scenarios possible in just over three months.

The team notes that this methodology can be adapted to other fields, including climate and weather modeling, for systems requiring linkage between fine-scale and broad-scale processes in simulation.

"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," said 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."

SC '25 Milky Way Simulation:The First Star-by-star $N$-body/Hydrodynamics Simulation of Our Galaxy Coupling with a Surrogate Model