A team of researchers at the RIKEN Center for Interdisciplinary Theoretical and Mathematical Sciences (iTHEMS) in Japan has achieved a significant milestone in astrophysics by successfully simulating the Milky Way galaxy with remarkable detail. Collaborating with colleagues from the University of Tokyo and the Universitat de Barcelona, they have created the first model that accurately represents over 100 billion stars over a period of 10,000 years. This innovative simulation surpasses previous models by capturing 100 times more stars and executing the process 100 times faster.
The breakthrough was made possible by harnessing 7 million CPU cores, advanced machine learning algorithms, and sophisticated numerical simulations. The findings were published in a paper titled “The First Star-by-star N-body/Hydrodynamics Simulation of Our Galaxy Coupling with a Surrogate Model,” featured in the *Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis* (SC ’25).
Advancements in Galactic Simulation Techniques
Simulations that incorporate the dynamics of individual stars provide a vital platform for testing theories related to galactic formation, structure, and evolution. Until now, researchers faced considerable challenges in accurately modeling the myriad forces at play, including gravity, fluid dynamics, supernovae, and the influence of supermassive black holes. Traditional methods have been limited by the available computational power, with existing models typically capping at around 1 billion solar masses, which accounts for less than 1% of the Milky Way’s stellar population.
Compounding these challenges, state-of-the-art supercomputing systems would require approximately 315 hours—over 13 days—to simulate just 1 million years of galactic evolution. This represents only a fraction of the Milky Way’s age, estimated at 13.61 billion years. Consequently, only large-scale galactic events could be accurately simulated, and simply increasing supercomputer cores proved ineffective due to diminishing efficiency and escalating energy demands.
To overcome these barriers, the research team, led by Hirashima, integrated a machine learning surrogate model, which significantly reduced the resource consumption of the simulations. This AI-powered approach was trained on high-resolution simulations of supernova events, allowing the model to predict the impact of these explosions on surrounding gas and dust up to 100,000 years post-explosion.
Impressive Results and Future Implications
The performance of this innovative model was validated through extensive testing on the Fugaku and Miyabi Supercomputer Systems. The results confirmed that the new methodology could simulate the dynamics of galaxies housing over 100 billion stars in a mere 2.78 hours for every 1 million years of evolution. This efficiency means that simulating an entire 1 billion years of galactic history could be achieved in approximately 115 days.
These advancements not only provide astronomers with a powerful tool to explore theories about galactic evolution but also highlight the potential of integrating AI models into complex simulations across various fields. The implications extend beyond astrophysics, as this “AI shortcut” approach could facilitate intricate simulations in areas such as meteorology, ocean dynamics, and climate science.
By successfully merging advanced computational techniques with artificial intelligence, the research team at RIKEN has laid the groundwork for enhanced understanding of our universe and its intricate workings. This transformative work represents a notable leap forward in both astrophysics and the application of AI in scientific research.
