A new machine learning-powered simulation is helping scientists understand how some of the universeโs heaviest elements are created during the most violent events in space.
Elements such as gold, uranium, and many other heavy materials do not form easily. Scientists believe they are produced in extreme cosmic environments, especially during events such as neutron star mergers and supernova explosions. These events release enormous energy and create the conditions needed to build elements heavier than iron.
However, simulating these processes in full detail is extremely difficult. The reactions happen under extreme temperatures, densities, and pressures, and they involve thousands of atomic nuclei and nuclear reactions. Capturing all of these details in a single simulation requires enormous computing power.
Now, researchers at GSI Helmholtzzentrum fรผr Schwerionenforschung, the Facility for Antiproton and Ion Research, and their international collaborators have developed a new artificial intelligence-based model that makes this task more efficient.
The model gives scientists a clearer way to study how heavy elements are formed during extreme astrophysical events, especially neutron star mergers. For the first time, the researchers incorporated a deep learning neural network directly into hydrodynamic simulations to model the energy released during the rapid neutron-capture process.
Their findings were published in Physical Review D.
Many chemical elements are created in powerful cosmic explosions. In environments such as neutron star mergers, matter is extremely dense and rich in free neutrons. This allows atomic nuclei to rapidly capture neutrons in a process known as the rapid neutron-capture process.
During this process, nuclei absorb neutrons very quickly. Some of those neutrons later transform into protons, allowing the nuclei to become heavier and heavier. This is how many elements heavier than iron are believed to form.
One of the most important examples came on 17 August 2017, when scientists detected gravitational waves from the collision of two neutron stars in the galaxy NGC 4993. The event was also followed by a bright stellar flare called a kilonova, which was observed by the Hubble Space Telescope as it gradually faded over several days.
That discovery strongly supported the idea that neutron star mergers are important factories for heavy elements in the universe.
Even with such observations, scientists still need detailed theoretical simulations to understand exactly how the elements are produced. But these simulations are difficult because they must combine the motion of exploding matter with the complicated nuclear reactions happening inside it.
According to Dr. Oliver Just, first author of the study and a researcher in the Nuclear Astrophysics and Structure department at GSI and the Facility for Antiproton and Ion Research, scientists around the world are working to make these complex reactions understandable through simulations. But because modelling all the required parameters takes enormous computing power, many existing models have to be simplified.
To address this challenge, the team developed a new model called RHINE, which stands for r-process heating implementation in hydrodynamic simulations with neural networks.
RHINE uses artificial intelligence to represent the energy released by nuclear reactions during the rapid neutron-capture process. This energy release, known as heating, plays an important role in shaping the movement of matter thrown out during explosions.
The heating can affect how fast the ejected material moves, how its speed is distributed, and what kind of electromagnetic signals are produced. These signals include kilonova light, which can be observed after neutron star mergers.
To build the model, the researchers first trained machine learning systems using a large number of detailed reference calculations. These reference calculations included a full set of nuclear reactions and thousands of isotopes.
Once trained, the machine learning models were placed inside hydrodynamic simulations. Instead of calculating every nuclear reaction directly during the simulation, the artificial intelligence model could quickly estimate the heating rates caused by the rapid neutron-capture process.
This made the simulations far more efficient while still preserving important physical information.
Dr. Zewei Xiong, a scientist in the Nuclear Astrophysics and Structure department at GSI and the Facility for Antiproton and Ion Research, played a central role in designing the machine learning models. He explained that the models were first trained on detailed nuclear reaction calculations and then used during hydrodynamic simulations to approximate heating rates with much less computational effort.
The researchers tested the machine learning approach by comparing it with detailed reference data. They found a high level of agreement, showing that the artificial intelligence model could reproduce the important effects of rapid neutron-capture heating while saving a large amount of computing time.
The results also showed that heating from the rapid neutron-capture process is not a small detail. It can significantly influence the evolution of material ejected during explosive cosmic events and should be included more carefully in future models.
This is important because better simulations can help scientists connect different areas of research. They can link nuclear physics experiments on Earth with astronomical observations from telescopes and gravitational-wave detectors.
The researchers say RHINE could make future simulations more detailed and realistic. It may help scientists better understand the signals produced by neutron star mergers and other stellar explosions, while also improving knowledge of how the universe creates its heaviest elements.
The work is especially relevant for the upcoming Facility for Antiproton and Ion Research, where scientists will study exotic nuclei and nuclear reactions that play a role in cosmic element formation.
In simple terms, artificial intelligence is helping scientists simulate one of the universeโs most powerful creation processes. By making complex nuclear reactions easier to include in large-scale simulations, RHINE offers a new window into how gold, uranium, and many other heavy elements are forged in the aftermath of cosmic collisions.
Reference: Oliver Just, Zewei Xiong, and Gabriel Martรญnez-Pinedo, โr-process heating implementation in hydrodynamic simulations with neural networks,โ Physical Review D, 16 April 2026. ย DOI: 10.1103/gl2l-7f3g
ย