Known mathematical problems and OpenAI
Lecture hosted by the Icelandic Mathematical Society on November 17th.
Tuesday November 17th, Anders Karl Claesson, Gianmarco Brocchi and Matthew McGeagh Roberts will give a lecture hosted by the Icelandic Mathematical Society, in which they will discuss and explain well-known mathematical problems that OpenAI recently claimed to have solved. These are:
* Local smoothing and the Kakeya conjecture
* The exponential time hypothesis
* The Yang–Mills problem
The lecture will take place in room N-132 in Askja, from 17:30 to 18:30.
See announcement here: https://stae.is/ísf/tilkynning/51690
Abstracts
- Anders: The exponential time hypothesis. The exponential time hypothesis (ETH) asserts that 3-SAT cannot be solved in time $2^{o(n)}$, where $n$ is the number of variables. It underpins many conditional lower bounds in algorithms and combinatorics. In October 2026, OpenAI and Anthropic released AI-generated papers, accompanied by Lean formalizations, claiming to refute ETH by giving deterministic subexponential-time algorithms for counting satisfying assignments of $k$-CNF formulas, for every fixed $k$. I will give some background on ETH and outline these claims, with a brief look at the role of Lean. Time permitting, I will sketch a connection, following Jay Pantone, to counting 1324-avoiding permutations in subexponential time.
- Gianmarco: Local smoothing and the Kakeya conjecture. The local smoothing conjecture (Sogge, 1991) says that waves cannot stay focused for long. It is the strongest of a family of open problems in harmonic analysis that includes the Fourier restriction and Kakeya conjectures. I will review its history, up to recent claims, and explain how these problems are connected.
- Matthew: The Yang–Mills problem. Yang-Mills theory, a mathematical model for the strong nuclear force, is a both beautiful and notoriously difficult physical system to study. An alternative description, which comes from discretizing space and time, works well as a physical model, but there are many open questions about whether the continuum limit of the lattice theory is equivalent to the continuum Yang-Mills theory. The recent OpenAI result attempts to rigorously study the continuum scaling limit, and makes some progress, but fails to answer the Millennium Prize Problem.