GPT-5.2 Derives New Result in Theoretical Physics
GPT-5.2 identifies non-zero gluon tree amplitudes
OpenAI and a team of researchers from the Institute for Advanced Study, Vanderbilt University, University of Cambridge, and Harvard University have demonstrated that GPT-5.2 can derive new results in theoretical physics. Specifically, the team published a preprint titled "Single-minus gluon tree amplitudes are nonzero," which proves that certain particle interactions involving gluons—the particles that carry the strong nuclear force—can occur under specific conditions where they were previously assumed to be absent.
Technical discovery: The half-collinear regime
Theoretical physicists use scattering amplitudes to calculate the probability of particles interacting. In the case of gluons, many "tree-level" amplitudes (calculations that exclude quantum loops) are unexpectedly simple.
Historically, standard textbook arguments suggested that if one gluon has negative helicity (one of two possible spin orientations) and the remaining $n - 1$ gluons have positive helicity, the tree-level amplitude must be zero. This conclusion was based on the assumption of generic particle momenta, where directions and energies are not in special alignment.
The research identifies a specific, mathematically well-defined slice of momentum space called the half-collinear regime. In this regime, gluon momenta obey a special alignment condition, and the amplitude does not vanish. This discovery opens new avenues for investigation, including the computation of analogous amplitudes for gravitons, the particles that mediate the gravitational force.
AI-assisted methodology and derivation
GPT-5.2 Pro and a scaffolded version of GPT-5.2 were integral to the derivation of the final formula (Eq. 39 in the preprint) through a multi-step process:
- Complexity Reduction: Human authors first calculated amplitudes for integer $n$ up to $n = 6$ by hand. These expressions, derived from a Feynman diagram expansion, were superexponentially complex. GPT-5.2 Pro reduced the complexity of these expressions into much simpler forms.
- Pattern Recognition: Based on these simplified base cases, GPT-5.2 Pro spotted a pattern and conjectured a formula valid for all $n$.
- Formal Proof: An internal scaffolded version of GPT-5.2 spent approximately 12 hours reasoning through the problem to produce a formal proof of the formula's validity.
- Analytical Verification: The resulting equation was verified using the Berends-Giele recursion relation (a method for building multi-particle tree amplitudes) and checked against the soft theorem, which governs amplitude behavior when a particle becomes soft.
Expert perspectives on AI-assisted science
External physics professors have highlighted the importance of this result as a template for the future of scientific discovery.
"The physics of these highly degenerate scattering processes has been something I’ve been curious about since I first ran into them about fifteen years ago, so it is exciting to see the strikingly simple expressions in this paper."
"To me, ‘finding a simple formula’ has always been fiddly, and also something that I have long felt might be automatable by computers."
— Nima Arkani-Hamed, Professor of Physics, Institute for Advanced Study
"This is clearly journal-level research advancing the frontiers of theoretical physics, and its novelty will inspire future developments and subsequent publications. This preprint felt like a glimpse into the future of AI-assisted science, with physicists working hand-in-hand with AI to generate and validate new insights."
— Nathaniel Craig, Professor of Physics, University of California, Santa Barbara (UCSB)