The international team of scientists has developed a tool to help you study the "Space Network", the huge network of galaxies and empties, which is the structure of the universe. The new EFFORT.JL program can make calculations that previously demanded supercomputers in just a few minutes, on the usual laptop. The work was published in the "Magazine of Cosmology and Star Magazine" (JCAP).
The construction of the models of large-scale structure of the universe is a very resource problem. One of the most advanced theories, the theory of the effective field of large structure, statistically describes the "skeleton" of the universe, connecting the observation data with basic physics. However, the processing of constantly increasing data collections, such as Desi project or European Euclid Mission, requires a huge computing power.
"Our emulator is actually backing up what the model does but much faster and at a lower price." "It seems like a neural network has studied theoretical reactions and learn to predict new parameters, without recalculating everything," said Marco Bonici, researcher at the University of Vategoo.
According to the scientist, Effort.JL is unique in the fact that it uses the knowledge of how the predictions change along with the change of parameters. This allows the system to train much smaller copies, saving resources. Moreover, in some cases, the emulator has shown even greater accuracy than the model itself, as it allowed to include the analysis of the parts that were usually missed.
"We have shown that the accuracy of Effort.JL is close to the model accuracy of both modeled and real data," Bonichi said. "This means that we now have a powerful tool that can effectively develop the forthcoming data collections from Desi and Euclid."
According to the authors, the new emulator can become a basic tool for astrophysicos in the coming years, when humanity will receive the "Space Network" in the history of observation history.
Translation of: Euromedia24.com
