A team of researchers from the RIKEN Center for Interdisciplinary Theoretical and Mathematical Sciences (iTHEMS) in Japan, along with colleagues from the University of Tokyo and the Universitat de Barcelona, has achieved a significant milestone in astrophysics. They successfully conducted the first simulation of the Milky Way that models over 100 billion stars across a span of 10,000 years. This groundbreaking simulation represents a leap forward, as it includes 100 times more stars than previous models and was executed 100 times faster.
The innovative project combined the power of 7 million CPU cores, machine learning algorithms, and advanced numerical simulations. The results were detailed in a paper titled “The First Star-by-star N-body/Hydrodynamics Simulation of Our Galaxy Coupling with a Surrogate Model,” which was published in the Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis (SC ’25).
Advancements in Galactic Simulations
Simulating the Milky Way with such granularity allows astronomers to test theories regarding galactic formation and structure. By capturing the dynamics down to individual stars, researchers can align their findings with astronomical observations. Historically, creating such simulations has been challenging due to the immense computational power required to model various forces at play, including gravity, fluid dynamics, supernova events, and the influence of supermassive black holes.
Previously, scientists faced a significant limitation in simulating galaxies, with the ability to model only about one billion solar masses, representing less than 1% of the Milky Way’s total stellar content. Traditional supercomputers would require approximately 315 hours (over 13 days) to simulate just 1 million years of galactic evolution, which equates to merely 0.00007% of the Milky Way’s age of 13.61 billion years. This limitation restricted researchers to studying only large-scale galactic events.
To overcome these challenges, the research team, led by Hirashima, introduced a machine learning surrogate model. This AI-driven approach predicts the effects of supernova explosions on surrounding gas and dust, providing insights into their dynamics up to 100,000 years post-explosion. By integrating this AI model with physical simulations, the researchers were able to simultaneously analyze both the large-scale structure of the Milky Way and the intricate behaviors of individual stars.
Efficiency and Future Implications
The team validated their model through extensive tests on the Fugaku and Miyabi Supercomputer Systems. The results demonstrated that their new method could simulate the evolution of galaxies containing more than 100 billion stars, achieving this in just 2.78 hours for 1 million years of evolution. At this rate, simulating 1 billion years of galactic history would take approximately 115 days, a dramatic reduction in time compared to previous methods.
These findings not only provide astronomers with a powerful tool for studying galactic evolution but also illustrate the potential benefits of incorporating AI models in advanced simulations. The implications of this breakthrough extend beyond astrophysics. The “AI shortcut” approach could also enhance simulations in fields such as meteorology, ocean dynamics, and climate science, where both large and small-scale factors are crucial.
By pioneering this complex simulation of the Milky Way, the researchers have opened up new avenues for understanding the universe, offering a detailed platform for exploring how galaxies evolve over time. As the field of astrophysics continues to advance, the integration of such innovative technologies will likely play a vital role in unraveling the mysteries of our cosmos.
