Claude Opus 4.6 Solves Don Knuth's Open Math Problem

AI Collaborates with a Computer Science Icon

Donald Knuth, Stanford professor emeritus and author of The Art of Computer Programming, published a formal paper on March 2, 2026, documenting how Anthropic's Claude Opus 4.6 solved an open mathematical problem he had been working on for weeks.

The Problem

While writing a future volume of TAOCP, Knuth encountered a directed Hamiltonian cycle decomposition problem in a digraph with m³ vertices. His friend Filip Stappers had empirically found solutions for 4 ≤ m ≤ 16, strongly suggesting general solutions exist, but no mathematical proof had been established.

Claude's Solution

Knuth presented the problem to Claude Opus 4.6. The model deduced where to look for a solution and found one. As Knuth notes in the paper, Claude's approach was just one of hundreds of valid solutions — yet the model arrived at it through logical reasoning rather than brute force.

Why It Matters

The paper, drafted February 28 and revised March 2, 2026, is available on Knuth's Stanford faculty page. It represents one of the clearest endorsements yet of an LLM's mathematical reasoning capability by a top figure in computer science. The Hacker News post garnered a score of 188, sparking debate about whether results like this constitute true mathematical understanding or sophisticated pattern matching.

Either way, the fact that Knuth chose to formally document Claude's contribution signals a meaningful shift in how leading researchers are beginning to view AI as a legitimate research collaborator.