For the first time, an international research team has succeeded in completely simulating the Milky Way down to its approximately 100 billion stars, over a period of 10,000 years. This was announced by the Japanese research institute Riken, which led the work. Accordingly, the simulation not only includes 100 times as many stars as its most powerful predecessors, it was also completed 100 times faster. This was made possible by the inclusion of AI algorithms. The research group adds that the approach is now not only of great value for astrophysics, it could also help in researching climate change and weather patterns.
Read more after the ad
Significantly faster than previous simulations
As the researchers explain, the best simulations to date could only model a maximum of one billion stars, so calculations of the Milky Way always only had clusters of 100 stars as the smallest unit. What happens to individual stars has been calculated. But this was particularly problematic because an exploding single star can have an influence far beyond its immediate surroundings as a supernova. At the same time, it would have taken 36 years if one had wanted to simulate a million years in the evolution of the Milky Way with the level of detail that has now been achieved. With the method now developed, this would only take 115 days.
For the improved simulation, the team trained, among other things, an AI model with high-resolution models of stellar explosions, from which the spread of the emitted gas over 100,000 years could be derived. As a result, resources no longer need to be diverted to these processes when simulating the entire galaxy, the team writes. This would allow dynamics in the entire galaxy to be simulated at the same time and more limited phenomena such as supernovae to be included. To verify this, the team compared the output data with those of established supercomputer simulations.
Those responsible are convinced that their approach can also be of great help in other computationally intensive branches of science. They point to simulations of the weather, oceans and the climate as a whole. In all of them, narrowly defined processes would have to be linked to system-wide processes. In addition, their work shows once again that simulations accelerated by AI can be a real scientific tool. In this specific case, for example, they could show us how the substances in the Milky Way from which life on Earth ultimately developed were created. The research work on this was presented at the supercomputing conference SC 25 and is publicly available.
(my)