There are phrases that end up in manuals and seem destined to remain there forever. One of those said, in essence, that a certain interaction between gluons – the particles that hold the nucleus of atoms together – simply . But no. A new study, to which he contributed GPT-5.2shows that that interaction is not at all impossible. It occurs under very particular conditions, but it exists. And this is enough to reopen an issue that was thought closed.
The work, entitled “Single-minus gluon tree amplitudes are nonzero”was published on arXiv and is under scientific review. The authors include physicists from institutions such as the Institute for Advanced Study, Vanderbilt University, the University of Cambridge, Harvard and OpenAI.
What does all this mean?
When two particles collide, physicists don’t just observe what happens: they calculate the chance that a certain type of interaction occurs. This calculation is based on a called number scattering amplitude. Without these amplitudes, we could not predict what happened in particle accelerators or in the energetic phenomena of the early universe.
In the case of gluons, responsible for the strong nuclear force, the one that holds protons and neutrons together, many of these amplitudes are surprisingly simple when only the most “direct” interactions are considered, without additional quantum complications. It is what physicists call the “tree” level. But there was one exception that was considered definitive. If a gluon has a certain spin configuration (called negative helicity) and all the others have the opposite one (positive helicity), the amplitude was given as zero. Translated: that interaction doesn’t happen.
The study shows that this conclusion only holds if the particles move in a “generic” way. However, there is a special configuration, called half-collinear regimein which the particles are aligned in a particular way. In that case, the interaction is nothing. It’s not a minor detail. It’s like discovering that a door we thought was walled up actually opens… if you find the right angle.
How GPT-5.2 helped find a formula that physicists struggled to see
This is where artificial intelligence comes into the picture. The researchers had already done the calculations by hand for some specific cases. The problem? The formulas quickly became very long, almost unmanageable. The more the number of particles involved increased, the more the complexity grew explosively.
GPT-5.2 Pro took those complicated expressions and simplified them. But above all he identified a recurring pattern, managing to propose a general formula valid for any number of particles. It wasn’t a “fluke”. An internal version of the model worked for about twelve hours, reconstructing the mathematical reasoning step by step until arriving at the same formula and providing a formal proof.
The validity of the result was then verified with standard methods of theoretical physics, including the recursive Berends-Giele relation and the so-called soft theoremwhich imposes very precise rules on the behavior of low-energy interactions. In other words: it is not an AI suggestion. It’s math checked and rechecked.
AI and science: collaboration, not replacement
Theoretical physicist Nima Arkani-Hamed enthusiastically commented on the appearance of such simple expressions in a field known for its complexity. Often, he recalled, formulas that seem unmanageable with traditional methods turn out to be very elegant once the right key has been found. And the search for these simple structures could be one of the most promising fields for intelligent automation.
Nathaniel Craig also underlined how this work represents advanced academic research and offers a concrete model of collaboration between physicists and large-scale linguistic models. The question, at this point, is no longer whether artificial intelligence can enter theoretical laboratories. It’s how the way of doing science will change when the dialogue between humans and AI becomes structural.
In a historical moment in which technological innovation is often described only in an economic or commercial perspective, this discovery reminds us of something deeper: AI can also help us understand the universe better. And maybe, every now and then, to question what we thought we knew.
