Scientists have developed an artificial intelligence system that could sharpen our understanding of how the universe manufactures its heaviest elements. The tool, created by an international team based at GSI/FAIR, uses machine learning to simulate the extreme nuclear reactions that unfold during neutron star mergers and other violent cosmic events, doing so far more efficiently than previous methods. The findings were published in the journal Physical Review D.
Many of the elements scattered across the cosmos are born in cataclysmic events such as supernova explosions and neutron star mergers. These explosions release enough energy to trigger the r-process, or rapid neutron capture, in which atomic nuclei quickly absorb free neutrons. Some of those neutrons then convert into protons, allowing the nuclei to grow into heavier elements.
Modeling this process is notoriously difficult because it demands enormous computing power, forcing researchers to simplify their calculations.
"Researchers around the world strive to make these complex reactions understandable through theoretical simulations. However, modeling all parameters requires incredible computing power, which is why the models often have to be simplified," said Dr. Oliver Just, first author of the study and a researcher in the Nuclear Astrophysics & Structure department at GSI/FAIR. "Our new model RHINE, which uses artificial intelligence, offers an efficient alternative."
A Neural Network Trained on Nuclear Physics
The new system, called RHINE — short for r-process heating implementation in hydrodynamic simulations with neural networks — uses a deep learning neural network to estimate how much energy is released during r-process nuclear reactions while broader hydrodynamic simulations are running.
That energy release, known as heating, shapes how matter is ejected during stellar explosions, influencing both the speed of the expelled material and the light it produces afterward. In neutron star mergers, this light appears as a glowing kilonova.
Rather than performing every nuclear calculation from scratch during each simulation, the AI is first trained on an extensive library of reference calculations built from complete nuclear reaction networks. Once trained, it can estimate heating rates using only a fraction of the computational effort.
"First the ML models are trained using a large number of reference calculations produced with a full set of nuclear reactions. Subsequently, the models are adopted in running hydrodynamical simulations to approximate the heating rates during the r-process with minimal effort," said Dr. Zewei Xiong, a scientist in the same GSI/FAIR department and a key developer of the machine learning models.
Xiong said the team validated the approach against reference data and found a high degree of agreement, suggesting the machine learning models could save substantial computing time. The results also indicated that r-process heating is an important effect that should be better accounted for in future modeling.
The researchers say RHINE could enable far more detailed simulations while sharply cutting the computing resources needed. They hope the improved models will eventually help link experiments at the upcoming FAIR research facility with astronomical observations of stellar explosions and neutron star mergers.
The RHINE source code has been made publicly available so other researchers can build on the work. The project was co-funded, among other organizations, by the European Research Council.