💥 An AI to understand where gold, uranium, and all heavy elements come from

The gold in your jewelry and the uranium in nuclear power plants share a common and cataclysmic origin: they were born during collisions of neutron stars in what are called kilonovae. Until now, scientists were unable to simulate these events with enough precision to understand the formation of these heavy elements. A new approach has just unlocked this barrier.

To carry out this simulation, it is necessary to track thousands of nuclear reactions under extreme conditions. Classical methods require colossal computing power, which forces simplifications in the models. Essential details about the energy released during the rapid neutron capture process, called the r-process, were often omitted.


To overcome these limitations, researchers at GSI/FAIR and their collaborators created RHINE, a machine learning tool. RHINE stands for "r-process heating implementation in hydrodynamic simulations with neural networks". This model uses deep learning to predict the energy released by nuclear reactions, without having to perform complete nuclear calculations. It thus saves valuable computing time.

The central idea is elegant: first, a neural network is trained on a large number of reference calculations. Once trained, the network can instantly estimate heating rates during a hydrodynamic simulation. This allows researchers to include the effect of nuclear energy on the explosion dynamics, which was previously too computationally expensive.

The team validated RHINE by comparing its predictions with detailed reference data. The agreement was remarkably good, confirming that machine learning can capture the physics at play. This validation indicates that future simulations will be able to realistically integrate r-process heating, improving models of kilonovae and other electromagnetic signals.

Thanks to RHINE, researchers can now perform more detailed simulations of neutron star mergers in kilonovae but also in supernovae. This paves the way for links between theoretical models and observations. Understanding r-process heating is essential for interpreting kilonova light curves and the abundance of heavy elements in the Universe.

Neutron star mergers

Neutron stars are the ultra-dense remnants of supernova explosions. When two of these celestial bodies orbit each other, they eventually collide in a cataclysm. This merger releases a gigantic amount of energy in the form of gravitational waves, detectable on Earth, and ejects neutron-rich matter.

The ejected matter gives rise to a kilonova, a transient luminous flash. It is in this matter that the r-process operates, producing a large part of the heavy elements. The neutron star merger observed in 2017 (GW170817) confirmed this scenario.

Simulations of these mergers are demanding because they involve nuclear physics, hydrodynamics, and gravitation. Models like RHINE allow nuclear heating to be integrated into simulations, improving our understanding of kilonovae and element synthesis. This work helps interpret current and future observations.