Direct discovery · Physics · Mathematics
Claude autonomously computes a nine-loop amplitude in planar N=4 super-Yang-Mills
Two independently derived representations of the nine-loop six-gluon MHV amplitude in planar N=4 super-Yang-Mills, released as machine-readable coefficient files.
Summary
Anthropic physicists prompted Claude to compute a six-particle amplitude at nine loops, one loop beyond the previous published frontier. Claude produced the result through two separate bootstrap routes; the representations agree on every compared coefficient, and computer-readable files and validation records were released. Amplitude expert Lance Dixon independently validated the result, while a human team using some AI assistance concurrently obtained the symbol-level result.
AI role
Claude selected the challenge, organized week-long symbolic computations and produced the amplitude by both a form-factor route and a direct bootstrap with limited human prompting.
Narrative role
Adds a physics result where an agent executed and coordinated a frontier-scale symbolic computation on an academic-scale budget, with an external expert check and a concurrent human-AI result providing unusually concrete comparison points.
Caveat
N=4 super-Yang-Mills is a tractable toy theory rather than a direct prediction for real particles. The computation applies known bootstrap machinery rather than introducing a new physical principle, the programs were not released, and the human team was already close to the same result.