Using the new Setonix supercomputer, researchers have created the largest simulation ever of our Milky Way galaxy, with 100 billion stars. This ambitious project leveraged an astonishing 7 million CPU cores. In the process, it created a stunning new paradigm for cosmological simulations, significantly advancing our knowledge of galactic dynamics and supernova feedback. Led by Keiya Hirashima, this simulation was performed at the RIKEN Center for Interdisciplinary Theoretical and Mathematical Sciences (iTHEMS) in Japan. As part of his work, he worked with teams from The University of Tokyo as well as Universitat de Barcelona in Spain.
This emulative simulation model was an advanced mix of deep learning surrogate modeling and classic physics-based simulations. This approach allowed researchers to predict how the surrounding gas expands in the aftermath of a supernova explosion, specifically over a period of 100,000 years. This breakthrough showcases the staggering possibilities AI holds for scientific discovery. It is a tremendous achievement that is now broadly in place.
The smallest simulation unit is a star cluster of stars with mass of 100 suns. This creation allows a more intimate portrayal of galactic reformation. The prior state-of-the-art models were ultimately extremely limited, simulating only up to one billion suns and portraying many fewer individual stars. By contrast, this new simulation is able to include 100 times more stars as its predecessors.
In fact, the simulation made this impressive speed-up possible, finishing the simulation of one billion years in a mere 115 days. Traditional physics-based simulations require a staggering 315 hours to simulate one million years. That’s right, a complete billion-year simulation would require more than 36 years of computation time! With the new model, the time required to simulate one million years shrank to just 2.78 hours.
Hirashima noted the implications of integrating AI with high-performance computing, stating, “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.”
This advancement has larger implications for understanding how galaxies form elements. Hirashima remarked on the potential of AI-accelerated simulations: “This achievement 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.”
This pioneering research makes an invaluable contribution in deepening our understanding of stellar dynamics. It’s pioneering a new standard for peer-reviewed research on the cutting edge of astrophysics and data-driven research more broadly. Using cutting edge computation techniques, researchers are finding new approaches to illuminating the universe’s deepest secrets. Their work will reveal unprecedented insights into the stellar structure, behavior, and formation history of galaxies, like our own Milky Way.
