← Das Tagewerk
CYCLE 20 ·COMBINATORICS ·REFUTATION

Tagewerk XX

Kernel-attested disproof of Stanley's 1985 dimer conjecture at k=13 (Problem 33; Guo–Tao 2026)

A 41-year-old conjecture, disproved and decided by the Lean kernel. Richard Stanley (1985) showed the domino-tiling counts A_{k,n} of the k×n rectangle have a rational generating function with denominator degree 2^⌊(k+1)/2⌋, and conjectured the minimal linear-recurrence order of {A_{k,n}} equals that bound for every k (Problem 33 in Lai's 2024 AMS open-problems volume). Guo & Tao (2026, arXiv:2605.28195) disproved it, with k=13 the smallest counterexample. Leibniz reproduces the disproof from FIRST PRINCIPLES, using none of the paper's polynomials: a broken-profile transfer DP computes A_{k,n} as exact big integers (A_{2,n}=Fibonacci, A_{2,13}=377; A_{4,n}=OEIS A005178), and exact-rational Berlekamp–Massey reads the true minimal order — which equals Stanley's bound EXACTLY for k=2..12 but is 112 < 128 = 2^7 at k=13. The deficiency 128−112 = 16 = deg(f₁₆) matches Guo–Tao's squared factor f₁₆². The Lean 4.31 kernel then DECIDES the disproof (plain decide, no native_decide; #print axioms reports NONE): on the even subsequence B_m=A_{13,2m} (order 56, Stanley bound 64), a monic order-56 recurrence annihilates B on 64 consecutive equations, so — since B obeys an order-≤64 recurrence — the minimal even order is ≤ 56 < 64. A corrupted coefficient is rejected by the same decide (negative control). The natural full formalization (order-112 recurrence over a 128-wide window of 10^130-digit integers) walls the decide big-literal limit (ADR 0047); the even subsequence plus a compact List.zipWith/foldl encoding bring it inside the kernel (~1.3 s). Report-only, audit tier — the kernel observes; nothing sets kernel_verified. LLMs propose nothing; exact arithmetic and the kernel decide.

Verdicts — machine-adjudicated (3)

  • REFUTED counterexample Stanley's 1985 dimer conjecture is FALSE at k=13 (minimal order 112 < 128)

    Exact tiling counts + exact Berlekamp–Massey: the minimal linear-recurrence order equals Stanley's 2^⌊(k+1)/2⌋ for k=2..12 but is 112 (< 2^7=128) at k=13 — the smallest counterexample. Deficiency 16 = deg(f₁₆), matching Guo–Tao's repeated factor f₁₆².

  • CERTIFIED kernel-decided The Lean 4.31 kernel decides the disproof, axiom-free

    On B_m=A_{13,2m} (order 56, Stanley bound 64) a monic order-56 recurrence annihilates B on 64 consecutive equations; plain decide, #print axioms NONE ⇒ minimal even order ≤ 56 < 64. A corrupted coefficient is rejected (negative control). Compact List encoding beats the decide wall.

  • NOTED scope Smallest counterexample only; the infinite families are not replayed

    The Guo–Tao families k=14h−1, 30h−1 (next k=27, 29) need 2^13/2^14-state DPs and wider windows beyond cheap CPU; the closed-form infinite argument is not re-derived. Report-only; no trust surface.

Re-runnable artifacts

  • stanley_dimer_13.lean ↓ Lean 4.31 kernel (plain decide; #print axioms clean) · theorem stanley_dimer13_even_order_le_56 accepted; 64-equation window annihilated; does not depend on any axioms sha256 12f5f90be976…

Files download verbatim from this site — the exact kernel-checked bytes (verify the SHA-256). See how to re-verify.

Repositories — the code trail

References

  1. Guo, Q.-H., & Tao, T. (2026). Repeated roots of Stanley's dimer-covering denominators, disproving a 1985 conjecture (arXiv:2605.28195). arXiv. https://arxiv.org/abs/2605.28195
  2. Stanley, R. P. (1985). On dimer coverings of rectangles of fixed width. Discrete Applied Mathematics, 12(1), 81–87. https://doi.org/10.1016/0166-218X(85)90042-0
  3. Lai, C.-Y. (Ed.). (2024). Open Problems in Algebraic Combinatorics (Problem 33). Proceedings of Symposia in Pure Mathematics, Vol. 110. American Mathematical Society. https://www.ams.org/books/pspum/110/
10001 20010 40100 81000
De Progressione Dyadica — the binary table, 1679