Researchers at the RIKEN Center for Interdisciplinary Theoretical and Mathematical Sciences (iTHEMS) in Japan have achieved a groundbreaking milestone in astrophysics. In collaboration with colleagues from the University of Tokyo and the Universitat de Barcelona, the team created the world’s first detailed simulations of the Milky Way galaxy, accurately modeling over 100 billion stars over a span of 10,000 years. This simulation not only surpasses previous models by a factor of 100 in terms of star count but also accelerates processing speed by a similar margin.
The project utilized an impressive combination of 7 million CPU cores, machine learning algorithms, and advanced numerical simulations. The results, detailed in a paper titled “The First Star-by-star N-body/Hydrodynamics Simulation of Our Galaxy Coupling with a Surrogate Model,” were presented at the Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis (SC ’25). This innovative approach has significant implications for studying stellar and galactic evolution on a grand scale.
Revolutionizing Galactic Simulations
Traditionally, astronomers have faced challenges in developing simulations that account for complex forces like gravity, fluid dynamics, and supernovae, which all operate over different scales. Past simulations were limited to modeling galaxies with a mass of approximately one billion solar masses, representing less than 1% of the Milky Way’s total star count. The computational demands were immense, with existing supercomputers requiring about 315 hours (over 13 days) to simulate just 1 million years of evolutionary time, a fraction of the Milky Way’s age of approximately 13.61 billion years.
The research team, led by Hirashima, addressed these limitations by implementing an AI-driven surrogate model. This model, trained on high-resolution simulations of supernovae, enables predictions on how these cosmic explosions influence surrounding gas and dust up to 100,000 years post-explosion. By integrating this AI component with physical simulations, the team successfully captured both large-scale galaxy dynamics and small-scale stellar phenomena.
A New Era for Astrophysical Research
The performance of this new simulation method was verified through extensive testing on the Fugaku and Miyabi Supercomputer Systems, confirming its ability to simulate star resolution in galaxies with over 100 billion stars. Remarkably, the model can simulate 1 million years of galactic evolution in just 2.78 hours. At this accelerated pace, simulating 1 billion years of galactic history could be accomplished in roughly 115 days.
This advancement not only furnishes astronomers with a powerful tool for examining theories of galactic evolution but also demonstrates the potential of integrating AI into complex simulations across various scientific fields. Beyond astrophysics, this “AI shortcut” approach could revolutionize simulations in meteorology, ocean dynamics, and climate science, allowing for more efficient and detailed analyses.
The implications of this research extend far beyond theoretical astrophysics. By enhancing the capabilities of supercomputing and AI, scientists are now better equipped to unravel the mysteries of our universe, shedding light on how galaxies like our own came into existence and evolved over billions of years.
