Researchers spearheaded by Keiya Hirashima at the RIKEN Center for Interdisciplinary Theoretical and Mathematical Sciences (iTHEMS) in Japan, in collaboration with The University of Tokyo and Universitat de Barcelona in Spain, have achieved a groundbreaking feat: the creation of the first Milky Way simulation capable of individually tracking over 100 billion stars through 10,000 years of cosmic evolution. This monumental achievement, presented at the prestigious SC ’25 international supercomputing conference, was realized through the innovative synergy of artificial intelligence (AI) and sophisticated numerical simulation techniques. Their revolutionary model boasts an astonishing 100-fold increase in star representation compared to previous state-of-the-art simulations, all while generating results more than 100 times faster. This pioneering work signifies a paradigm shift in astrophysics, high-performance computing, and the burgeoning field of AI-assisted scientific modeling, with profound implications for our understanding of galactic dynamics and extending to critical Earth system studies, including climate and weather research.

For decades, astrophysicists have harbored the ambition of constructing Milky Way simulations with the fidelity to follow the trajectory of every single star. Such detailed models are crucial for directly comparing theoretical frameworks of galactic evolution, structure, and star formation against observational data. However, the sheer complexity of accurately simulating a galaxy presents formidable computational hurdles. It necessitates the precise calculation of gravitational interactions, intricate fluid dynamics, the nucleosynthesis of chemical elements, and the energetic outbursts of supernovae across vast expanses of time and space. This multifaceted challenge has historically rendered simulations of the Milky Way’s full grandeur, with its estimated 100 billion stars, an insurmountable task while maintaining the fine-grained detail of individual stellar behavior. Existing cutting-edge simulations, while advanced, typically represent systems with a total mass equivalent to about one billion suns. Within these models, the smallest computational "particle" often represents a cluster of approximately 100 stars, a compromise that inevitably averages out the unique behaviors of individual stars and consequently limits the accuracy of simulating small-scale phenomena. A primary constraint stems from the required interval between computational timesteps; to faithfully capture rapid and dynamic events like supernova evolution, simulations must advance in infinitesimally small time increments.

The computational burden escalates dramatically with each reduction in timestep. Even with the most powerful physics-based models available today, simulating the Milky Way at the individual star level for just one million years of galactic evolution would demand approximately 315 hours of processing time. Extrapolating this to one billion years of cosmic activity would translate to over 36 years of real-time computation. The intuitive solution of simply augmenting the number of supercomputer cores is not a practical or efficient remedy, as energy consumption becomes prohibitively high, and computational efficiency tends to diminish with an increasing number of cores. This inherent scalability issue necessitates a more radical approach to computational astrophysics.

In response to these formidable obstacles, Hirashima and his team devised a novel methodology that harmoniously integrates a deep learning surrogate model with traditional physical simulations. The surrogate model, meticulously trained on high-resolution supernova simulations, acquired the remarkable ability to predict the intricate dispersal of gas in the aftermath of a supernova explosion over a period of 100,000 years, without imposing any additional computational load on the primary simulation. This AI component proved instrumental in enabling the researchers to simultaneously capture the overarching behavior of the galaxy while also resolving the nuanced details of individual stellar events, including the complex processes within supernovae. The team rigorously validated their groundbreaking approach by comparing its outputs against large-scale simulations executed on RIKEN’s Fugaku supercomputer and The University of Tokyo’s Miyabi Supercomputer System, confirming its accuracy and reliability.

This innovative hybrid approach delivers true individual-star resolution for galaxies containing over 100 billion stars, achieving this with astonishing computational speed. The simulation of one million years of galactic evolution, which previously would have taken an astronomical amount of time, was completed in a mere 2.78 hours. This remarkable acceleration means that simulating one billion years of galactic history could now be accomplished in approximately 115 days, a staggering improvement over the previously estimated 36 years. This leap in efficiency fundamentally alters the landscape of astrophysical research, opening up possibilities for exploring longer timescales and more detailed galactic phenomena than ever before.

The broader implications of this hybrid AI approach extend far beyond astrophysics, holding transformative potential for numerous fields within computational science that grapple with the challenge of linking small-scale physical processes to large-scale emergent behavior. Disciplines such as meteorology, oceanography, and climate modeling face analogous computational hurdles. These fields could significantly benefit from tools that dramatically accelerate complex, multi-scale simulations, enabling more accurate predictions and deeper insights into Earth’s intricate systems.

"I believe that the integration of AI with high-performance computing represents a fundamental paradigm shift in how we approach and solve multi-scale, multi-physics problems across the computational sciences," states Hirashima. "This achievement demonstrably proves that AI-accelerated simulations can transcend mere pattern recognition to become a genuine instrument for scientific discovery, empowering us to unravel the complex cosmic origins of the very elements that gave rise to life itself within our own galaxy." The potential for this technology to unlock new avenues of scientific inquiry and accelerate breakthroughs across diverse scientific domains is immense.