Anthropic AI model helps physicists calculate record nine-loop particle physics amplitude.

Anthropic reports that its Claude model helped physicists Liam Fitzpatrick and Siddharth Mishra-Sharma calculate a six-particle scattering amplitude in planar N=4 super Yang-Mills theory at nine loops, exceeding the previous eight-loop record. The work answered a challenge from physicist and science writer Matt von Hippel to use AI and academic-scale computing resources to tackle a major problem in theoretical physics. Claude worked from a single prompt for several days, using established methods; Anthropic says each of the two calculation approaches cost about $1,000 to $2,000. SLAC and Stanford physicist Lance Dixon, whose team had reached eight loops by an indirect route, checked the result. The achievement demonstrates AI’s potential to help carry out demanding calculations, though the reported work relied on human physicists, existing techniques and substantial computation.
Von Hippel’s challenge offered two possible targets: calculate the N=4 super Yang-Mills six-particle amplitude at nine loops, or determine whether N=8 supergravity diverges at seven loops. He argued the bottleneck was computational power rather than a lack of known methods, with the calculations’ complexity growing exponentially or even factorially with loop count.
The reported run used 96 processors for about a week. The calculation concerned the “hexagon” amplitude, so named because it involves six particles; in the model, “planar” refers to a limit that removes some of the more complex terms.
The earlier eight-loop result by Dixon and collaborators was obtained indirectly through a related form factor, using a symmetry called antipodal duality, rather than by calculating the amplitude directly.
Von Hippel said he set the challenge partly because he wanted to assess AI on a familiar problem he knew to be difficult, and partly to test a computational bottleneck rather than a task whose difficulty might hinge on an unpredictable new idea.
Anthropic disclosed that it invited von Hippel to write the guest post and compensated him for his time; the company also gave feedback on drafts, while von Hippel said the content and opinions were his own.
Publishers
9
Articles
6
Reach
15