On August 17, 2017, astronomers watching the galaxy NGC 4993 witnessed something extraordinary: two neutron stars colliding, their gravitational waves rippling across the universe and their explosion blazing across telescope screens as a kilonova. That cosmic event, and thousands of simulations like it, have long posed a puzzle for scientists trying to understand how the heaviest elements in existence are born in these violent stellar mergers.
Now, a breakthrough from researchers at GSI/FAIR, an international research facility, is bringing fresh precision to that ancient cosmic mystery. Using artificial intelligence, the team has created a model called RHINE—r-process heating implementation in hydrodynamic simulations with neural networks—that for the first time uses deep learning to accurately simulate how heavy elements form during neutron star collisions. The results, published in Physical Review D, reveal that machine learning can dramatically simplify the enormous computational burden that has long constrained scientists' ability to model these events.
The challenge is fundamentally about energy and chemistry at the extreme. When neutron stars merge, they release unimaginable amounts of energy that allows nuclei to capture free neutrons and transform them into protons, creating heavier and heavier atomic nuclei. This process, called the rapid neutron-capture process or r-process, is responsible for forging roughly half of all the heavy elements in the universe—including gold, platinum, and uranium. Understanding it means understanding where many of the atoms in our bodies and our world originate.
"Modeling all parameters requires incredible computing power, which is why the models often have to be simplified," explained Dr. Oliver Just, first author of the publication and a researcher in the Nuclear Astrophysics & Structure Department at GSI/FAIR. "Our new model, RHINE, which uses artificial intelligence, offers an efficient alternative."
The RHINE system works by training neural networks on large numbers of reference calculations that include a complete set of nuclear reactions. Once trained, these machine learning models can approximate the energy release from r-process heating in hydrodynamic simulations with minimal computational effort. The innovation matters because the heating produced during these nuclear reactions significantly impacts the velocity and distribution of material ejected from the explosion—factors that determine the electromagnetic radiation observed as a kilonova, like the one glimpsed from NGC 4993 in 2017.
Dr. Zewei Xiong, also at GSI/FAIR's Nuclear Astrophysics & Structure Department and key to designing the ML models, emphasized the validation behind the work. "With detailed comparisons, we validated our ML scheme against reference data. The high degree of agreement suggests that the use of ML models can save a tremendous amount of computing time."
What makes this advance hopeful is not only the computing efficiency it unlocks but what comes next. The FAIR facility is currently under construction, and experiments conducted there could soon be directly linked with observations of real neutron star mergers and stellar explosions—allowing researchers to test their simulations against genuine cosmic events. With RHINE making detailed modeling far more feasible, the gap between laboratory data and the cosmos narrows. The ancient question of how the universe creates its heaviest elements now has a faster, smarter path toward an answer.