AI Lets Scientists Simulate 100 Billion Stars in the Milky Way for the First Time

The Challenge of Modeling a Galaxy Star by Star

Simulating a massive spiral galaxy like the Milky Way is notoriously difficult. Galaxy evolution requires tracking gravity, fluid dynamics, explosions, and chemical enrichment across scales ranging from individual stars to the galactic disk, each governed by processes unfolding over very different time steps. Until now, computational limits forced simulations to treat clusters of about 100 stars as a single “particle,” smoothing out the physics of individual stellar evolution.

Moreover, conventional simulations struggle because high-resolution time steps are too computationally expensive. For example, the best physical simulation available would require 315 hours of real time for every one million years of simulated galactic history. Simulating a single billion years of Milky Way evolution would take more than 36 years - even on top-tier supercomputers - while consuming massive amounts of energy.

AI as a Surrogate for Fine-Scale Physics

To break this barrier, the research team developed a deep learning surrogate model trained on high-resolution supernova simulations. The AI learned to predict how gas evolves over the first 100,000 years after a supernova explosion, enabling the main simulation to bypass computationally intensive physics calculations without sacrificing accuracy.

This hybrid technique allows the galaxy-scale simulation to maintain fine-scale fidelity while running dramatically faster. The surrogate model handles local supernova dynamics, while the global simulation continues evolving the galaxy as a whole. The team validated the method by comparing results with large-scale runs on RIKEN’s Fugaku supercomputer and the University of Tokyo’s Miyabi system.

Speed and Scale: A Breakthrough in Computational Astrophysics

• More than 100 billion stars simulated individually
• 100× higher resolution than previous state-of-the-art models
• 100× faster computation than traditional methods
• Simulating 1 million years now requires only 2.78 hours
• A full one billion-year simulation would take ~115 days, down from decades

This advance demonstrates that star-by-star modeling of large galaxies is now feasible, enabling unprecedented tests of theories of galactic formation, stellar evolution, and chemical enrichment.

Implications Beyond Astrophysics

The methodology extends far beyond galactic science. Many fields - from climate modeling to ocean dynamics and extreme weather prediction - face similar challenges coupling processes across vastly different scales. AI-accelerated surrogate modeling could provide a template for improving accuracy while dramatically reducing computational cost.

“A Fundamental Shift”

“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.”