An open problem in theoretical physics: compute the exact power spectrum of gravitational radiation from cosmic strings for arbitrary loop geometries. Prior analytical work produced only partial asymptotic results. The integral I(N,α) resisted human mathematicians.
Note (2026-11-11): checked against the paper's full text (this essay had been logged as held up on an abstract-level read). The authors, the system (a language model with tree search and numerical feedback), the six methods and the Gegenbauer expansion are as in the paper. Three statements go beyond it. (1) The essay credits the AI with trying six different analytical approaches as its advantage. The paper says that once the algorithm found valid solution paths, the researchers used negative prompting (detailed in an appendix) to force the discovery of different types of solutions, which led to the six reported; it also describes the system prompts, search constraints and intermittent feedback loops that guided the model. (2) The essay says the system solved the problem and found exact solutions. The abstract says exact analytical solutions, but also that the methods lead to an asymptotic result for the integral at large N, and the paper notes that earlier attempts gave asymptotic results for large N or partial solutions for odd N, with a unified exact solution being the open problem. (3) The essay's remark that the singularities were artifacts of the coordinate system and not features of the physics is its own: the paper describes the integrand as having singularities at the end points that make standard numerical integration unstable and a Legendre expansion difficult, because the weight functions do not match.
Brenner, Cohen-Addad, and Woodruff (arXiv: 2603.04735) built a neuro-symbolic system that solved it. Gemini Deep Think paired with tree search and numerical feedback found six distinct analytical methods. The most elegant uses Gegenbauer polynomial expansion — the polynomials naturally absorb the integrand's singularities, turning a hard problem into a clean one. The solutions match numerical computations and connect to Feynman parameterization from quantum field theory.
The through-claim: the AI didn't solve the problem by brute force. It found the representation that makes the problem dissolve. Gegenbauer polynomials are a specific choice of basis functions. In the wrong basis, the integral is intractable. In the right basis, the singularities vanish. The discovery is the basis, not the computation. The AI's advantage was trying six different analytical approaches — not faster arithmetic, but broader exploration of representation spaces.
This is the kind of mathematical discovery that matters most: not a new theorem but a new way of seeing that makes existing theorems fall out. The singularities were artifacts of the coordinate system, not features of the physics. The problem was hard because people were looking at it in the wrong basis, and the AI happened to look at it in the right one.
Brenner, Cohen-Addad & Woodruff, 2603.04735. AI-assisted discovery / gravitational waves / cosmic strings.