Anthropic published account of Claude solving nine-loop scattering amplitude, surpassing eight-loop record; Dixon verification reported.
Claude Solves Nine-Loop Scattering Amplitude, Breaking Physics Record
The result advances the frontier of a notoriously hard class of physics calculations that directly bear on particle interaction probabilities. It also demonstrates that an AI model can run a multi-day autonomous symbolic computation, catch or avoid its own errors, and produce output compatible with existing physics literature.
The full picture
Anthropic's Claude computed a nine-loop scattering amplitude in the planar N=4 super-Yang-Mills model, surpassing the previous eight-loop record held by SLAC physicist Lance Dixon and collaborators. Given a single prompt, Claude ran largely unsupervised for days inside the Claude Science environment, at a total cost of a few thousand dollars. Dixon independently verified the result using his group's existing verification tools. The problem involves calculating six-particle scattering amplitudes, where each additional loop exponentially multiplies computational complexity; most such calculations stop at two or three loops. The N=4 super-Yang-Mills model is a simplified setting physicists use as a testing ground. The challenge was originally posed by a physicist and science writer known as @4gravitons, asking whether AI could advance past eight loops within an academic compute budget. Dixon said Claude "understands our 2019 and 2023 papers better than any human, aside from my co-authors." The computation ran with only periodic continuation prompts and no mid-run scientific corrections. Model inference costs dominated the overall expense, while the numerical calculation itself used modest conventional computing resources.
How it developed
Sources
Want this in your inbox?
I send one email each morning with the stories that moved. If you would rather just read here, that works too.
Subscribe free