Researchers Develop Groundbreaking Milky Way Simulation with AI

Researchers from the RIKEN Center for Interdisciplinary Theoretical and Mathematical Sciences (iTHEMS) in Japan, in collaboration with the University of Tokyo and the Universitat de Barcelona, have achieved a remarkable milestone by creating the world’s first hyper-realistic simulation of the Milky Way galaxy. This groundbreaking model accurately simulates over 100 billion stars over a span of 10,000 years, setting a new standard in the field of astrophysics.

The simulation surpasses previous models by representing 100 times more individual stars and being produced 100 times faster. This was made possible by harnessing the power of 7 million CPU cores, advanced machine learning algorithms, and state-of-the-art numerical simulations. The findings were detailed in a paper titled “The First Star-by-star N-body/Hydrodynamics Simulation of Our Galaxy Coupling with a Surrogate Model,” published in the *Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis* (SC ’25).

Advancements in Galactic Simulation

Simulations that capture the dynamics of the Milky Way down to individual stars are essential for testing theories regarding galactic formation, structure, and evolution. Researchers have faced challenges in creating increasingly complex simulations due to the difficulty of accurately capturing the various forces at play, including gravity, fluid dynamics, supernovae, and the influence of supermassive black holes (SMBHs).

Historically, scientists have struggled to simulate galaxies at such a high level of detail and complexity. Current models are limited to around one billion solar masses, which represents less than 1% of the total stellar mass of the Milky Way. Traditional supercomputing systems require approximately 315 hours (over 13 days) to simulate just one million years of galactic evolution, a minuscule fraction of the Milky Way’s age of 13.61 billion years.

To overcome these limitations, Hirashima and his team introduced an innovative AI approach involving a machine learning surrogate model. This model was designed to predict the impact of supernova explosions on surrounding gas and dust, 100,000 years after the event, without utilizing the same computational resources as the primary simulation.

Transforming Astrophysical Research

The research team validated their model’s effectiveness through extensive testing on the Fugaku and Miyabi Supercomputer systems at RIKEN and the University of Tokyo. The results demonstrated that their method could simulate star resolution in galaxies containing more than 100 billion stars, completing a simulation of one million years in just 2.78 hours. At this pace, a simulation covering 1 billion years of galactic history could be achieved in approximately 115 days.

These advancements offer astronomers a powerful tool for examining theories surrounding galactic evolution and the formation of the universe. Furthermore, the integration of surrogate AI models highlights significant potential for enhancing advanced simulations across various scientific disciplines, including meteorology, ocean dynamics, and climate science.

As this research continues to unfold, it stands to reshape our understanding of the Milky Way and beyond, marking a significant leap forward in the capabilities of astrophysical simulations.