DATE2025.07.01 #Press Releases
High-resolution galaxy simulation made possible by AI
Disclaimer: machine translated by DeepL which may contain errors.
-Computation of Galaxy Evolution is now completed in about 2 months, 6 months shorter than before...
Announcement Summary
Keiya Hirashima, Special Postdoctoral Researcher in Basic Sciences, Mathematical Basis Division, RIKEN Center for Mathematical Innovation; Associate Professor Michiko Fujii and Assistant Professor Kana Moriwaki, Graduate School of Science, The University of Tokyo; Lecturer Yutaka Hirai, Department of Public Service, Tohoku University of Mathematical Sciences; Associate Professor Takayuki Saitoh and Senior Professor Junichiro Makino, Graduate School of Science, Kobe University, The international research group, led by Dr. Ulrich Philipp Steinwandel, a postdoctoral fellow at the Max Planck Institute for Astrophysics, and Dr. Shirley Ho, group leader of the Center for Computational Astrophysics at the Flatiron Institute, has used deep learning in artificial intelligence (AI) to predict the complex physical processes of supernova explosions. The research group has developed a surrogate model for predicting the complex physical processes of supernova explosions using artificial intelligence (AI) deep learning, and integrated this model into a galaxy formation simulation code for the first time.
This achievement is the first example of accelerated "star-by-star" high-resolution galaxy simulations that directly treat individual stars, which has been difficult to achieve in the past, by performing deep learning inference in real time during the simulation. The newly developed method is expected to contribute to the detailed analysis of supernova feedback in the formation and evolution of our Milky Way Galaxy.
In the past, the reproduction of gas expansion caused by the shock wave from a supernova explosion (supernova feedback) has been a bottleneck in the simulation process, but the AI model developed by the international research group has solved this problem by predicting this process with high speed and accuracy. As a result, the calculation of galaxy evolution, which previously took about eight months, can now be completed in about two months, six months less.
This research was published in the online edition of the scientific journal The Astrophysical Journal on July 1 (JST).
Figure: Gas and star distribution in a dwarf galaxy simulation (© Keiya Hirashima)
Related Links
RIKEN, Tohoku University of Science and Letters, Kobe University, Max Planck Institute for Astrophysics, Flatiron Institute Center for Computational Astrophysics
Published Journals
-
Journal Name The Astrophysical JournalTitle