VibeMathedMath problems solved with AI

Ziegler's Cross-Polytope Conjecture (simplicial 0/1-polytopes)

Ziegler proved every simplicial dd-dimensional 0/1-polytope has at most 2d2d vertices, and asked whether attaining 2d2d vertices forces central symmetry (i.e. a 0/1 cross-polytope). Known true for d6d \le 6; open since ~2000.

Result
Disproved
Status
Resolved
AI contribution
AI-discovered
Method
Construction
Field
Combinatorics, Discrete Geometry
Posed by
Günter M. Ziegler
Year posed
2000
Years open
26y
Solved
2026-06-30
Model
DeepSeek V4 Flash, GLM 5.2
Vendor
DeepSeek / Zhipu AI
Collaborators
Volker Kaibel, Sebastian Pokutta
Verification
Unreviewed
Publication
Preprint
Significance
15 / 100
Disclosed cost
Wikipedia
No dedicated article

What the AI did

An agentic research framework (locally-deployed open-weights DeepSeek V4 Flash + GLM 5.2, augmented with reflection prompts) first produced a flawed proof that the conjecture holds, then attempted a Lean 4 formalization as verification. The formalization failed, and from that failure the agent extracted the combinatorial condition that yielded an explicit counterexample: 14 vertices in {0,1}7\{0,1\}^7 whose convex hull is simplicial but not centrally symmetric.

Verification

arXiv preprint 2606.31640 (30 Jun 2026) by Volker Kaibel and Sebastian Pokutta. The counterexample is explicit and computer-checkable (exhaustive enumeration finds exactly five such non-centrally-symmetric polytopes in dimension 7, of two combinatorial types); a domain-expert preprint, not yet peer-reviewed. Notably found with locally-run open-weights models, not closed frontier LLMs.

Source

Changelog1 change
  • Curatoradded this entry

Discussion