AI generates first full-scale Milky Way simulation with 100 billion stars
Researchers led by Keiya Hirashima at the RIKEN Center for Interdisciplinary Theoretical and Mathematical Sciences (iTHEMS) have produced what they describe as the first Milky Way simulation that models roughly 100 billion stars using artificial intelligence. The work, carried out with partners from The University of Tokyo and other collaborating institutions, marks a step forward in how scientists can recreate complex galactic structure at star-by-star scales.
Why this simulation matters
Large-scale galaxy simulations traditionally trade resolution for scope: they either model millions of particles across cosmological volumes or simulate small regions of galaxies in exquisite detail. This new approach uses AI-driven techniques to bring star-level detail to the scale of the entire Milky Way, enabling researchers to explore the galaxy’s structure, stellar populations, and kinematics with unprecedented completeness.
By modeling the Milky Way at the scale of approximately 100 billion stars, the simulation offers a synthetic laboratory for testing theories of galactic formation and evolution, refining models of stellar dynamics, and improving interpretation of observational surveys.
How AI was applied
The research team employed artificial intelligence methods to accelerate and enhance the simulation workflow. AI can interpolate and infer detailed stellar properties and motions across regions where direct high-resolution simulation would be computationally prohibitive. This allows the combined model to represent star-by-star features while remaining computationally feasible for large volumes.
While technical details of the algorithms and computing resources are described in the team’s report, the primary innovation is the integration of machine learning with established astrophysical simulation techniques to produce a coherent, galaxy-scale, high-resolution result.
Collaboration and computational effort
The project is led by Keiya Hirashima at RIKEN iTHEMS, with contributions from researchers at The University of Tokyo and additional collaborators. The work leveraged modern computational infrastructure and cross-disciplinary expertise in theoretical astrophysics, numerical simulation, and machine learning.
Applications and future uses
High-fidelity Milky Way simulations at star-by-star resolution have multiple scientific applications. They can be used to validate and interpret data from observational missions and ground-based surveys, test models of stellar populations and chemical evolution, and investigate dynamical features such as spiral arms, the central bar, or stellar streams.
The new AI-enabled simulation also provides a controlled dataset for developing and benchmarking analysis tools used by astronomers, including techniques for detecting substructure or inferring the history of past merger events in the Milky Way.
Limitations and next steps
Despite the advance, the authors acknowledge that simulations remain models built on assumptions about input physics and data. AI components can introduce uncertainties and biases if training data are incomplete or the models extrapolate outside their validated regimes. Continued comparison with observational data and refinement of the physical prescriptions remain essential.
Future work will likely expand the types of physics included, improve the realism of stellar and interstellar processes, and make the datasets more accessible to the broader research community for cross-comparison with observations.
Conclusion
The development of an AI-created Milky Way simulation containing about 100 billion stars represents a notable milestone in computational astrophysics. By marrying machine learning with large-scale simulation, the team led by Keiya Hirashima at RIKEN iTHEMS has opened new possibilities for studying our galaxy in detail and preparing for the next generation of observational surveys.
Source: ScienceDaily report on the RIKEN Milky Way simulation
