GSI/FAIR Research Team Uses Artificial Intelligence to Improve Simulations of Neutron Star Mergers and Heavy Element Formation

Artist’s impression of a neutron star merger
© Dana Berry, SkyWorks Digital, Inc.

(IN BRIEF) An international research team at GSI/FAIR has developed a new artificial intelligence-based simulation model called RHINE to improve understanding of how heavy elements form during stellar events such as neutron star mergers. The model uses machine learning and deep-learning neural networks to estimate the energy released during r-process nucleosynthesis, a key process in which nuclei capture neutrons and form heavier atomic nuclei. Traditional simulations of these events require enormous computing power because they involve highly complex nuclear reactions and physical parameters, often forcing researchers to simplify their models. RHINE offers a more efficient alternative by training machine learning models on reference calculations from full nuclear reaction networks and then applying them within hydrodynamic simulations to approximate r-process heating rates with much less computational effort. The results, published in Physical Review D, showed strong agreement with reference data and suggest that r-process heating plays an important role in the dynamics of material ejected during neutron star mergers and in the electromagnetic signals observed as kilonovae. The publicly available RHINE model could support future simulations linking experiments at FAIR with astronomical observations of stellar explosions and neutron star mergers.

(PRESS RELEASE) DARMSTADT, 8-Jun-2026 — /EuropaWire/ — An international research team at GSI/FAIR has developed a new artificial intelligence-based simulation model that provides deeper insight into how heavy elements are formed during powerful stellar events such as neutron star mergers.

The model, called RHINE, uses machine learning to describe the energy released during r-process nucleosynthesis in hydrodynamic simulations. For the first time, the researchers applied deep learning with a neural network to model this energy release in simulations of such astrophysical events. The findings have been published in the journal Physical Review D.

Many chemical elements are created during extreme cosmic events, including stellar explosions and neutron star mergers. These events release vast amounts of energy, creating the conditions required for the formation of heavy nuclides. One of the most important mechanisms behind this process is rapid neutron capture, known as the r-process, in which atomic nuclei capture free neutrons that later convert into protons, producing larger and heavier nuclei.

Scientists worldwide use theoretical simulations to better understand these complex reactions. However, modeling all of the physical and nuclear parameters involved requires enormous computing power, which often forces researchers to simplify their models.

“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 publication and researcher in the Nuclear Astrophysics & Structure department at GSI/FAIR. “Our new model RHINE, which uses artificial intelligence, offers an efficient alternative.”

RHINE stands for r-process heating implementation in hydrodynamic simulations with neural networks. The model uses deep-learning-based neural networks to estimate the heating rates generated by nuclear reactions during the r-process. This heating can influence the dynamics and velocity distribution of material ejected during stellar explosions or neutron star mergers, and it may also affect the electromagnetic radiation observed from neutron star mergers as a kilonova.

To build the system, the researchers first trained the machine learning models using a large set of reference calculations generated from a full network of nuclear reactions. Once trained, the models were integrated into hydrodynamic simulations, where they could approximate r-process heating rates with significantly lower 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 GSI/FAIR’s Nuclear Astrophysics & Structure department who played an important role in designing the machine learning models.

The researchers validated the machine learning approach through detailed comparisons with reference data. The strong agreement between the machine learning results and the reference calculations indicates that the method could substantially reduce computing time while preserving a high level of accuracy.

Dr. Xiong added that the findings show r-process heating is an important effect that should be more thoroughly included in future modeling of neutron star mergers and related stellar events.

The use of RHINE could make it possible to conduct more detailed astrophysical simulations in the future. These simulations may help connect experimental results from the future FAIR facility with astronomical observations of stellar explosions and neutron star mergers, supporting a better understanding of how the universe produces heavy elements.

The RHINE source code has been made publicly available. The project was co-funded, among others, by the European Research Council.

Media Contact:

Telephone: 06159-71-0
E-Mail: info(at)gsi.de

SOURCE: GSI/FAIR