Thanks to AI: First Complete Simulation of the Milky Way Including All Stars
Our Milky Way consists of 100 billion stars. Until now, simulations could not include them individually; thanks to AI, this was achieved for the first time.
The Milky Way above a telescope of the European Southern Observatory in Chile
(Image: Y. Beletsky (LCO)/ESO)
An international research team has succeeded for the first time in completely simulating the Milky Way down to its approximately 100 billion stars for a period of 10,000 years. This was announced by the Japanese research institute Riken, which led the work. According to the announcement, the simulation not only includes 100 times more stars than previous high-performance simulations but was also completed 100 times faster. This was reportedly made possible by the integration of AI algorithms. The approach is now not only of great value to astrophysics but could also help in researching climate change and weather patterns, the research group adds.
Significantly Faster Than Previous Simulations
As the researchers explain, previous best simulations could only model a maximum of one billion stars, meaning calculations for the Milky Way always used star clusters of 100 stars as the smallest unit. What happens to individual stars was calculated out. However, this was particularly problematic because an exploding single star as a supernova can have an influence far beyond its immediate surroundings. At the same time, it would have taken 36 years to simulate one million years of the Milky Way's development with the detail achieved now. With the method now developed, this would only take 115 days.
For the improved simulation, the team trained an AI model with high-resolution models of stellar explosions, from which the spread of ejected gas over 100,000 years could be derived. As a result, no resources need to be diverted for these processes in the simulation of the entire galaxy, the team writes. This would allow dynamics in the entire galaxy to be simulated simultaneously and more narrowly defined phenomena such as supernovae to be included. For verification, the team compared the output data with that of established simulations from supercomputers.
Videos by heise
Those responsible are convinced that their approach can also be of great help in other computationally intensive scientific fields. They refer to simulations of weather, oceans, and climate as a whole. In all of these, narrowly defined processes must be linked to system-wide ones. Furthermore, their work shows once again that AI-accelerated simulations can be a real scientific tool. In the specific case, they could, for example, show us how those substances in the Milky Way originated from which life on Earth eventually developed. The research paper was presented at the Supercomputing Conference SC 25 and is publicly available.
(mho)
