The proton looks like a solid ball, but inside it's a hive of activity: quarks exchange gluons — the particles that bind them together. Gluons pop in and out of existence, and their distribution within the proton is not just a curious detail but a key to predictions at the Large Hadron Collider. Previously, these distributions (known as parton distribution functions) were extracted from data like fish from water: the catch was sorted into bins, and each bin was counted. But this sorting loses fine details.
Now physicists have tried a different approach: feed the entire catch to a neural network without any bins. The network learns from millions of collision events where pairs of top quarks, the heaviest known particles, are produced. It looks at raw, unbinned data from the detector and teases out patterns on its own. This is the first time such an approach has been applied to the gluon distribution, and the precision has improved noticeably.
This method doesn't just improve the numbers. It removes the reliance on crude assumptions about how errors are correlated. When data isn't chopped into pieces, the neural network sees the full picture and makes fewer mistakes. For the future High-Luminosity LHC, where data will be many times more abundant, such precision will become a necessity.
Gluons were discovered in 1979 in an experiment at the PETRA accelerator in Hamburg: three jets of particles from an electron-positron collision revealed the presence of a third, invisible particle.
In the series 'Star Trek', quarks and gluons are mentioned as the energy source for the warp drive, although in reality their internal dynamics still baffle physicists.