Nobel Laureate and AI Collaborate to Prove 10-Year-Old Physics Conjecture

Nobel laureate Giorgio Parisi and physicist Francesco Zamponi have published a proof of a mathematical conjecture about jamming — a phenomenon where fluid-like particle systems suddenly become rigid — that had resisted solution for over a decade. The key collaborator? Anthropic’s Claude (Sonnet 4.6 and Opus 4.7).

The conjecture, dating to 2014, stated that two critical exponents in the full replica symmetry breaking solution of jammed matter satisfy the identity a + b = 1. Numerical evidence supported it to extraordinary precision, but an analytic proof remained elusive. Parisi, recognizing the problem’s well-defined structure as an ideal test case for AI, turned to Claude. After 40 prompts and several rounds of verification — during which Claude independently corrected some of its own errors — the model produced an “essentially correct” proof, which the physicists verified, refined, and published in the Journal of Statistical Mechanics: Theory and Experiment.

The proof confirms that two independently developed theoretical descriptions of jamming — one by Parisi’s group and another by French physicist Matthieu Wyart — are mathematically equivalent. The full conversation logs have been deposited in a Zenodo repository for transparency.

Why it matters: This is among the first demonstrations of an AI model contributing directly to a publishable proof in theoretical physics, suggesting that LLMs can serve as genuine research collaborators — not just assistants — in mathematical and scientific discovery.