A uranium nucleus is like a tight wallet snapped shut on two halves. To make it burst and release energy (that is, to make the nucleus split) you have to climb over an energy barrier. The height of this threshold determines whether a substance is stable or ready to fall apart at any moment. Physicists have been calculating it with formulas for decades, but for the most finicky nuclei, highly elongated ones with unusual particle counts, the calculations lied.
In the new study, a machine learning algorithm acted as a meticulous proofreader. It was fed hundreds of experimental measurements and asked to correct the predictions of the ETFSI model, one of the standard theoretical approaches. The XGBoost algorithm learned from its mistakes, like a student checking answers against a solution manual, and ended up making errors of only 0.3–1.2 MeV. For comparison, the typical binding energy of a single particle in a nucleus is about 8 MeV.
In this way, the machine's eye became a magnifying glass through which cracks in old formulas are visible. Now physicists know where to look to improve their calculations, and perhaps to describe more accurately how heavy elements are born in stellar explosions.
The fission energy of a single uranium nucleus is about 200 MeV. If you could split all the nuclei in one gram of uranium-235, it would release as much heat as burning 2.5 tons of coal.