In a monumental leap for space science, researchers from Japan and Spain have successfully created the first-ever comprehensive simulation of our Milky Way galaxy. This groundbreaking model possesses the unprecedented ability to track the movements and evolution of more than 100 billion individual stars across a staggering timeline of 10,000 years.
The international team, spearheaded by researcher Keiya Hirashima from the RIKEN Centre for Interdisciplinary Theoretical and Mathematical Sciences (iTHEMS) in Japan, collaborated with experts from the University of Tokyo and Spain's Universitat de Barcelona. They achieved this feat by ingeniously merging deep learning algorithms with high-resolution physics, creating a hybrid model that overcomes previous computational barriers.
Overcoming the Galactic Modelling Challenge
For decades, scientists have grappled with the immense challenge of modelling a galaxy as vast as the Milky Way with sufficient detail to monitor single stars. Existing simulations were powerful but limited, capable of handling systems equivalent to roughly one billion suns. This fell drastically short of accurately representing our galaxy, which is home to over 100 billion stars.
The core problem lay in the computational intensity. To capture fast-paced cosmic events like supernovae, models require extremely small time steps. This demands colossal computing power, making a full star-by-star simulation of the Milky Way an endeavour that would take decades, even with the most advanced supercomputers. Furthermore, simply adding more computer cores proved to be an inefficient and impractical solution.
How AI and Deep Learning Solved the Puzzle
Hirashima and his team ingeniously tackled these limitations. Their innovative approach involved training a surrogate model using deep learning. This AI first learned how gas behaves from highly detailed, small-scale supernova simulations.
Once trained, this AI surrogate could accurately predict how gas spreads for nearly 100,000 years after each supernova blast without slowing down the primary simulation. This hybrid method maintained the accurate large-scale structure of the galaxy while simultaneously capturing the fine details of individual stellar events. The result is a model that is hundreds of times faster than older methods and uses 100 times more stars than any previous work.
A New Era for Astrophysics and Earth Sciences
This breakthrough, which was showcased at the prestigious SC '25 supercomputing conference, marks a transformative moment for astrophysics, high-performance computing, and AI-supported scientific modelling.
The implications extend far beyond our galaxy. The researchers confirmed that their testing showed close agreement with large runs on powerful systems like Fugaku and Miyabi. The same revolutionary method is poised to transform large-scale Earth system studies, including climate modeling and weather prediction.
"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," ANI quoted lead researcher Keiya Hirashima as saying.
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."
