Three mathematicians have cracked a problem that stumped one of the world's greatest mathematicians for three decades—a conjecture that Michel Talagrand, an Abel Prize winner, declared unsolvable even as he posed it. Dongming Hua and Antoine Song from the California Institute of Technology, alongside Stefan Tudose from Princeton University, have finally proven Talagrand's convexity conjecture, a geometric mystery that bridges the hidden order within high-dimensional randomness.
The problem itself, born in 1995, asks a deceptively elegant question: can convexity—the mathematical property that ensures a shape bends outward with no inward dents—be created in a fixed, uniform number of steps using operations called Minkowski sums, no matter how many dimensions you're working in? A circle or a sphere are convex; anything with a gap or indent is not. But in higher dimensions, the problem becomes ferociously complex. Each additional dimension multiplies the computational difficulty exponentially, a phenomenon mathematicians call the "curse of dimensionality." Talagrand was so skeptical of his own conjecture that he offered $2,000 to anyone who could prove it, telling Scientific American: "I made this bold conjecture really without any ground for it, you know—it's just a shot in the dark. When you say something like that, you feel it cannot possibly be true."
The breakthrough came through an unexpected pivot. Rather than attacking the geometric problem head-on, the three mathematicians reformulated Talagrand's conjecture as a problem in probability theory. They proved an equivalent statement: that any 1-subgaussian random vector in n dimensions can be expressed as the sum of three standard Gaussian random vectors. This elegant translation from geometry to probability unlocked the solution. Their work, published on the arXiv preprint server, establishes that for any large enough set in Gaussian space, a convex set of significant measure can be found nested inside a triple sum of the original set. The proof also confirms a combinatorial analog of the problem, extending the solution's reach into discrete mathematics.
The path to the answer was collaborative and surprisingly modern. Song and Hua initially consulted ChatGPT as they worked through the problem, using the AI to help clarify questions and explore promising directions. Yet it was Tudose, who joined the project after learning of their work, who ultimately provided the rigorous proof—one that proved "more general and conceptual" than the AI-assisted approaches, according to the team's published paper.
The implications ripple far beyond abstract mathematics. While problems involving high-dimensional randomness might sound obscure, they underpin technologies woven into daily life. Machine learning algorithms navigate high-dimensional data spaces. Data scientists deploy tools built on these mathematical foundations. Logistics optimization relies on similar models of complex randomness to route goods and services efficiently. By solving Talagrand's conjecture, the three mathematicians have illuminated structures that countless algorithms depend upon, potentially opening new avenues for more efficient and powerful computational tools. The solution stands as a reminder that sometimes the deepest mysteries yield only to those bold enough to ask the right question—and creative enough to approach it from an entirely unexpected angle.