Artificial Intelligence Independently Solves Sixty-Year-Old Math Problem

2026-10-16 data

Eindhoven, Friday, 16 October 2026.
An AI model has independently solved a 60-year-old mathematical problem, producing a concise proof in hours and signaling a major leap from simple research assistant to autonomous problem solver.

A Decades-Old Conjecture Solved in Eindhoven

The breakthrough centers on Lehmer’s permutation conjecture, a problem first formulated by American mathematician D.H. Lehmer in 1965 [1][4]. The conjecture asks whether it is possible to iterate through every possible ordering of a set of colored items or dancers in a row by swapping only adjacent elements of differing colors, while allowing minor detours known as ‘spurs’ to avoid duplicate or skipped states [1][4][5]. Tom Verhoeff, a retired mathematician and computer scientist at Eindhoven University of Technology (TU/e) in the Netherlands, had previously solved the binary two-color variant of the problem in 2017 [1][2][4]. However, proving the general case involving multiple colors and varying quantities remained elusive for six decades [1][4].

From Human-Assisted Computing to Autonomous AI Proofs

In early September 2026, Verhoeff initially completed a dense, brute-force proof of the general conjecture after nearly a year of intensive research [5]. That initial derivation spanned approximately 150 to 175 pages of mathematical reasoning and required over 100,000 lines of formal verification code [2][5][6]. The workflow relied on an iterative human-in-the-loop cycle: Verhoeff utilized Anthropic’s Claude Opus 4.7 and DeepSeek V4.1 Flash on TU/e’s ‘Spike-1’ supercomputer to test reasoning steps, while Harmonic’s Aristotle system handled the formal proof checking in Lean 4 [1][4][5].

From Human-Assisted Computing to Autonomous AI Proofs

Later in September 2026, Verhoeff asked Anthropic’s newly released Claude Opus 5.5 model to find an elegant proof from scratch without relying on the original 150-page draft [1][4][5]. Working autonomously across multiple AI agents, Opus 5.5 generated a completely original proof in just a few hours [3][5][6]. The resulting document came under 30 pages [1][2][6]. When formally verified in Lean, the verification code plummeted from 100,000 lines to approximately 35,000 lines [1][2][4]—a reduction of 65% in code complexity.

The Rapid Acceleration of AI Capabilities

The speed of progress in artificial intelligence’s mathematical reasoning capabilities between late 2025 and late 2026 has been notable [1][5]. As recently as December 2025, OpenAI’s GPT-5.2 exhibited elementary mathematical errors when tasked with logic problems [1][5]. An attempt by Verhoeff to apply AI models to Lehmer’s conjecture in March 2026 similarly failed to generate useful results [1][2]. However, by May 2026, Anthropic’s Opus 4.7 demonstrated advanced comprehension as a research assistant, paving the way for Opus 5.5’s autonomous problem-solving breakthrough in September 2026 [1][5].

Implications for Research and Higher Education

Verhoeff submitted the concise proof as a preprint to arXiv (arXiv:2610.01240) in October 2026, following a visual presentation published on his website on 12 October 2026 [1][4]. The achievement coincides with broader industry developments, including OpenAI releasing hundreds of AI-generated mathematical proofs during the week of 4 October 2026 [2][5]. Reflecting on the impact of autonomous reasoning systems, Verhoeff emphasized that the evolution from basic computational tools to fully autonomous AI provers represents a paradigm shift larger than the introduction of electronic calculators in the 1970s or computer algebra systems in the 1980s, forcing academic institutions like TU/e to fundamentally rethink how mathematics is taught and researched [1][4][6].

Bronnen

Artificial Intelligence Mathematical Reasoning