Extending single-minus amplitudes to gravitons
OpenAI
A new physics preprint pushes 'single-minus' amplitude tricks into gravity, using GPT-5.2 Pro to grind through the math. It found nonzero graviton tree amplitudes that shouldn't cancel out — a real physics result, not a chatbot party trick.
Based on reporting by OpenAI — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Single-minus amplitudes are one of those quirky corners of particle physics where the usual rules about helicity and symmetry seem to bend. For years, theorists have studied them in gauge theories like Yang-Mills, where a single negative-helicity gluon amid a sea of positive-helicity ones can still produce a nonzero scattering amplitude, defying naive expectations. The new preprint takes that same logic and pushes it into gravity, asking whether gravitons — the hypothetical particles that would carry the gravitational force in a quantum theory — behave the same way at tree level.
The answer, according to the paper, is yes. The researchers derived nonzero graviton tree amplitudes using the single-minus setup, a result that adds a small but meaningful brick to the wall of quantum gravity scattering theory. Tree-level amplitudes are the simplest, most fundamental calculations in this kind of physics — no loops, no renormalization headaches, just the raw combinatorics of how particles interact. Getting a clean nonzero result here matters because it tells you something concrete about the structure of gravity's interactions, independent of whatever full theory eventually reconciles gravity with quantum mechanics.
What's notable is how the work got done. GPT-5.2 Pro reportedly assisted directly in deriving and verifying these amplitudes, handling the kind of dense symbolic manipulation that graviton calculations are infamous for. Gravity amplitudes are brutal to compute by hand because the vertices multiply in complexity fast — gravitons interact with themselves in ways that gluons simply don't. Having an AI system check derivations step by step, and apparently catch or confirm cancellations correctly, is the kind of grunt work that used to eat weeks of a grad student's time.
This isn't a flashy announcement about AI discovering new physics on its own. It's a narrower, more useful story: a model built for reasoning got pointed at a genuinely hard calculation in theoretical physics and held up. That's a different bar than writing an essay or debugging code, and clearing it says something about how far symbolic and mathematical reasoning in these systems has come.
My take — AI-written commentary, not fact-checked reporting
This is the kind of AI-in-science story I actually trust — not a press release about an LLM 'discovering' a theorem, but a physicist using a model as a very fast, very patient calculator for a problem that's tedious rather than mysterious. The real headline isn't GPT-5.2 Pro, it's that single-minus tricks might genuinely extend to gravity, and the model just helped get there faster. I'll take ten of these over one more chatbot-writes-a-sonnet demo.
Read more about this at: OpenAI