In a groundbreaking achievement that pushes the boundaries of astrophysics and computational science, researchers led by Keiya Hirashima at the RIKEN Center for Interdisciplinary Theoretical and Mathematical Sciences (iTHEMS) in Japan, in collaboration with esteemed partners from The University of Tokyo and Universitat de Barcelona in Spain, have successfully developed the first Milky Way simulation capable of tracking an astonishing number of individual stars – exceeding 100 billion – across 10,000 years of cosmic evolution. This monumental feat was realized through the innovative integration of artificial intelligence (AI) with sophisticated numerical simulation techniques. The resulting model boasts a star count 100 times greater than previous state-of-the-art simulations and was generated with unprecedented speed, completing the simulation process over 100 times faster than conventional methods. This pioneering work, unveiled at the prestigious international supercomputing conference SC ’25, represents a significant leap forward not only for astrophysics and high-performance computing but also for the burgeoning field of AI-assisted scientific modeling. Crucially, the underlying strategy employed by Hirashima’s team holds immense promise for application to vast and complex Earth system studies, including critical areas like climate and weather research.

The pursuit of accurately modeling the Milky Way at the individual star level has long been a central aspiration for astrophysicists. Such detailed simulations would provide an invaluable platform for directly comparing theoretical frameworks of galactic evolution, structure, and star formation against real-world observational data. However, the sheer complexity of simulating a galaxy with fidelity presents formidable challenges. The process necessitates the intricate calculation of gravitational interactions, fluid dynamics, the nucleosynthesis of chemical elements, and the explosive phenomena of supernovae across immense spatial and temporal scales. This multifaceted computational burden has historically rendered simulations of the Milky Way, with its estimated 100 to 400 billion stars, beyond the reach of individual star resolution.

Prior to this breakthrough, even the most advanced simulations were limited to representing systems with a mass equivalent to roughly one billion suns. This meant that the smallest computational "particle" in these models typically represented a collective of approximately 100 stars. Such averaging inevitably smoothed out the behavior of individual stars, thereby compromising the accuracy of small-scale processes crucial for understanding galactic dynamics. A primary bottleneck lies in the time step required for simulations. To faithfully capture rapid and energetic events, such as the lifecycle of a supernova, simulations must advance in infinitesimally small temporal increments.

Reducing the time step, while necessary for accuracy, dramatically escalates computational demands. Even with today’s most sophisticated physics-based models, simulating the Milky Way with individual star resolution would demand approximately 315 hours of computation for every 1 million years of simulated galactic evolution. Extrapolating this to simulate a mere 1 billion years of activity would translate to over 36 years of continuous real-time computation. Simply scaling up by adding more supercomputer cores is not a sustainable or efficient solution. As the number of cores increases, energy consumption becomes prohibitively high, and computational efficiency tends to diminish due to communication overhead.

To surmount these formidable obstacles, Hirashima and his team devised a novel hybrid approach that ingeniously combines a deep learning surrogate model with traditional physical simulations. The deep learning surrogate was meticulously trained on high-resolution simulations of supernovae. Through this training, it acquired the ability to predict the behavior of gas dispersal in the 100,000 years following a supernova explosion, all without imposing additional computational load on the main simulation. This AI component proved instrumental in enabling the researchers to capture the overarching dynamics of the galaxy while simultaneously modeling fine-grained, small-scale events, including the intricate details of individual supernovae. The efficacy and accuracy of this pioneering approach were rigorously validated through extensive comparisons against large-scale simulations executed on RIKEN’s powerful Fugaku supercomputer and The University of Tokyo’s Miyabi Supercomputer System.

The remarkable outcome of this AI-enhanced methodology is its capacity to achieve true individual-star resolution for galaxies containing over 100 billion stars, and critically, to do so with astonishing speed. The simulation of 1 million years of galactic evolution, which previously would have taken an unfeasible amount of time, was completed in a mere 2.78 hours. This translates to a staggering acceleration, allowing for the simulation of 1 billion years of galactic history in approximately 115 days, a stark contrast to the previous 36-year estimate.

The implications of this hybrid AI approach extend far beyond the realm of astrophysics, offering transformative potential for numerous fields within computational science that grapple with the challenge of linking microscopic physical processes with macroscopic behavior. Disciplines such as meteorology, oceanography, and climate modeling, which are inherently multi-scale and multi-physics in nature, stand to benefit immensely from tools that can dramatically accelerate complex simulations.

"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," stated Hirashima. "This achievement also demonstrates that AI-accelerated simulations can transcend mere pattern recognition to become a genuine engine for scientific discovery – empowering us to trace the very origins of the elements that underpin life itself within our galaxy." This sentiment underscores the profound impact of this research, not just as a technological triumph, but as a pivotal moment in the evolution of scientific inquiry, opening new avenues for understanding the universe and our place within it.