When two neutron stars collide, the explosion is one of the most violent events in the universe. In milliseconds, matter is compressed and heated to billions of degrees, creating conditions extreme enough to forge elements heavier than iron — including the gold in your jewelry and the uranium in nuclear reactors. But simulating exactly how this happens has been a computational nightmare.
An international team of researchers at GSI/FAIR in Germany has built a machine-learning tool called RHINE that cuts through this bottleneck. RHINE — which stands for r-process heating implementation in hydrodynamic simulations with neural networks — uses a deep neural network trained on thousands of detailed reference calculations. Once trained, it can estimate the energy released by nuclear reactions during a neutron star merger in a fraction of the computing time a full calculation would require.
The challenge lies in tracking the r-process (rapid neutron capture), a chain reaction in which atomic nuclei absorb free neutrons faster than they can decay. Each step produces a different element, and the full network involves thousands of isotopes interacting simultaneously. Simulating that alongside the hydrodynamics of the explosion itself has traditionally required weeks on supercomputers — far too expensive for routine research.
"Researchers around the world strive to make these complex reactions understandable through theoretical simulations. However, modeling all parameters requires incredible computing power," said Dr. Oliver Just, first author of the study published in Physical Review D. RHINE solves this by approximating the nuclear heating rate — the energy released as nuclei form — with minimal loss of accuracy. Dr. Zewei Xiong, a key developer, explained the two-stage approach: first the ML models are trained using a large number of reference calculations, then they are adopted in running hydrodynamic simulations to approximate the heating rates during the r-process with minimal effort.
This heating shapes how the ejected material behaves. It determines how fast matter is thrown outward from the merger and how bright the resulting kilonova appears — the electromagnetic afterglow that astronomers observe from Earth. By making these simulations faster, RHINE allows researchers to run many more scenarios and better interpret observations from gravitational-wave detectors like LIGO and Virgo.
The 2017 detection of a neutron star merger (GW170817) confirmed that these cosmic collisions are indeed factories for heavy elements. RHINE now gives scientists a practical tool to understand exactly which elements are produced and in what quantities — connecting the observations of kilonovae to the periodic table.
Knowledge takeaway: The RHINE AI system at GSI/FAIR uses deep learning to cut the computing cost of modeling neutron star merger nucleosynthesis, enabling scientists to simulate how gold, platinum, uranium, and other heavy elements are forged in these cosmic collisions, directly linking gravitational-wave observations to element formation.