OpenAI Reveals Unexpected Graviton Interactions

A Breakthrough in Graviton Study by OpenAI and Its Partners

A new preprint, recently published by a team of researchers from OpenAI and several prestigious universities, explores scattering amplitudes in the realm of quantum gravity. This innovative work extends previous results obtained for gluons and applies them to the gravitational context. The researchers discovered that a class of graviton interactions, long considered nonexistent, can actually manifest under certain kinematic conditions. The preprint is available online, and the authors encourage feedback from the scientific community.

The article, titled "One Minus Tree Amplitudes for Gravitons are Non-Zero," was written by Alfredo Guevara from the Institute for Advanced Study, Alexandru Lupsasca from Vanderbilt University and OpenAI, David Skinner from the University of Cambridge, Andrew Strominger from Harvard University, and Kevin Weil from OpenAI.

Decoding One Minus Amplitudes in Gravity

Scattering amplitudes are crucial mathematical tools for physicists, allowing them to calculate the probability of specific interactions between particles. Rather than detailing each step of a collision through numerous diagrams, these amplitudes condense the final observable results into a compact form. Over the decades, researchers have found that these amplitudes often reveal unexpected simplicity, uncovering an underlying mathematical structure that is not immediately apparent in traditional calculations.

In this preprint, the study focuses on gravitons, the quantum particles associated with gravity within the framework of quantum field theory. The authors particularly examine a configuration called one minus amplitude, where one particle has negative helicity while the others have positive helicity. Helicity, which describes the orientation of a particle's spin relative to its direction of motion, plays a key role in determining interactions.

Standard theories suggest that these amplitudes should be zero at the simplest approximation level, known as tree level, where only the most direct interaction diagrams are considered, ignoring quantum loop effects.

Reevaluating Traditional Assumptions

The preprint demonstrates that this conclusion relies on the assumption of generic particle motion. However, when the momenta of the particles adhere to a particular alignment, called the half-collinear regime, the standard argument no longer applies. In this context, the amplitudes do not vanish but exist as well-defined mathematical distributions, confined to a specific region of momentum space. The authors derived explicit formulas describing these interactions, showing that they arise from symmetry principles and recursive relations that construct complex interactions from simpler ones.

This result represents a step forward in resolving the central problem of reconciling quantum mechanics with Einstein's theory of general relativity. The one minus amplitudes embody an infinite-dimensional symmetry "w-(1+∞)," discovered by Penrose half a century ago in the context of classical gravity. This symmetry is expected by many to play a central role in the quantization of the gravitational field. The new preprint illustrates how, in the simplest possible framework, this symmetry acts on gravitons, the fundamental quantum elements of the gravitational field.

Methodology and Verification of Results

Although gravity and gauge theory share profound conceptual relationships, their calculations differ significantly in practice. The previous result on gluons had shown that a previously overlooked helicity configuration could produce non-zero amplitudes under particular conditions. After finalizing this work, the document on gluons was provided to GPT‑5.2 Pro as context. Relying on this model as a reference point, the researchers invited the model to construct the corresponding amplitudes in quantum gravity, a task that would have taken human researchers considerable time to derive. GPT‑5.2 Pro not only solved this problem using an elegant and surprising technique, the directed tree theorem, but also...