Researchers at the RIKEN Center for Interdisciplinary Theoretical and Mathematical Sciences in Japan, working with collaborators from the University of Tokyo and the Universitat de Barcelona, have developed an AI-driven simulation capable of tracking more than 100 billion stars in the Milky Way. As reported on SciTechDaily, the new model achieves far higher resolution and speed than previous galaxy-scale simulations.
The team combines conventional physics modeling with a deep learning component trained to reproduce how gas behaves after a supernova. This hybrid approach allows the simulation to capture both large-scale galactic motion and fine-scale stellar events while running more than 100 times faster than earlier methods.
By replacing the most computationally intensive supernova calculations with an AI surrogate model, the researchers reduced the time required for long-term galaxy simulations dramatically. A million years of simulated evolution now takes only 2.78 hours, compared with decades using traditional approaches.
The method was validated against results from major Japanese supercomputers and could benefit other fields that rely on multi-scale simulations, including climate and environmental modeling. Researchers say the approach offers a practical way to run large, long-term simulations that would otherwise require far more time and computing resources
Leave a Reply