The stabilised diagonals of n(g,t) and ideals of numerical semigroups
Abstract
Let n(g,t) be the number of numerical semigroups of genus g and type t. Chappelon, Ramírez Alfonsín and Stamate proved that each diagonal t=g-ℓ of this array is eventually constant, n(g,g-ℓ)=|V(ℓ)| for g ≥ 3ℓ-1, where V(ℓ) is a set of balanced binary words of length 2ℓ prescribed by a condition on their difference set; they tabulated |V(ℓ)| for ℓ ≤ 11, asked for |V(ℓ)| in general, and proved only |V(ℓ)| ≥ ℓ²-3ℓ+10 — which at ℓ=11 reads 98 against the true value 3436. We show that the defining condition is a statement about numerical semigroups after all: the cotype of a word is the genus of the multiplier semigroup of an associated ideal, so that V(ℓ) is the set of subsets I ⊆ ℕ with ℓ gaps whose multiplier semigroup has genus ℓ. From the largest fibre of this description we deduce |V(ℓ)| ≥ 2^(k(⌊ ℓ/(k+1)⌋-1)) for every k ≥ 1, hence |V(ℓ)| ≥ 2^(ℓ-2√(ℓ)+O(1)), and a sharper union bound 2^(ℓ-log₂ℓ-O(1)); the same fibre turns out to be a sequence counted by Marzuola and Miller in 2010, in a paper on a different indexing of numerical sets, which upgrades the answer to |V(ℓ)| ≥ γ_∞ · 2^(ℓ-1) for their density constant γ_∞=0.4844 ± 0.0051. We also compute |V(12)|=7234, the first value beyond the published table, so that n(g,g-12)=7234 for every g ≥ 35, and prove n(35,23) ≥ 7234 from explicit numerical semigroups; since the published table gives n(33,21)=7214, the value 7234 is not recoverable from it. All statements are machine-checked in Lean 4.
Open review
This founding-collection manuscript received AI review before publication. Independent human review is open. Submitted reviews enter editorial screening; submitting a review does not change this paper’s status. Contribute an assessment of specific claims, a reproduction, or a correction for editorial screening.
Archived files
- Version 1 · current (opens in a new tab)
Source snapshot 2026-08-30 15:34 UTC
File fingerprint
741fe8f52723523cac38300f41695b2e5c583ec9a15f4c3485d29f4840761e40
Claim ledger
Stated results
NS1routine2026-08-22
n(g,t) is a finite computation: it equals a count over the genus-g level of the numerical semigroup tree
NS2known data2026-08-22
All 210 values n(g,t) for 1 <= t <= g <= 20, and the 20 row totals (A007323)
NS3known data2026-08-22
|V(l)| = 1, 3, 7, 15, 35, 78, 161, 367, 757, 1632, 3436 for l <= 11 – the source's Table 1
NS4candidate2026-08-22
|V(12)| = 7234 – one term past the source's Table 1, hence n(g, g-12) = 7234 for every g >= 35
NS5candidate2026-08-22
n(35,23) >= 7234 from 7234 explicit, distinct, individually verified gap sets
NS6known data2026-08-22
Witness lower bounds inside the published range: n(25,1) >= 1182, n(30,21) >= 757, n(33,21) >= 7214
NS7routine2026-08-22
The source's stabilisation threshold g >= 3l - 1 is sharp, and t <= g
NS8known data2026-08-22
The source's printed worked examples reproduce
NS-stable2known2026-08-22
Stable gap vectors classified: cotype = genus of the multiplier semigroup; |V(l)| counts ideals of genus-l semigroups with full multiplier; |V(l)| >= 2^(l-2sqrt(l)+O(1))
NS-stable3aknown2026-08-22
Top-half reduction: IsGood iff large (Marzuola-Miller Lemma LargeSmallAtom)
NS-stable3bknown2026-08-22
Doubling: goodCount(2m) = 2*goodCount(2m-1) (MM Theorem 11), first formalisation
NS-stable3croutine2026-08-22
Union lower bound on |V(l)| — 2ˡ⁻ᵗ - (l-t-floor(l/2))*2ˡ⁻²ᵗ⁻¹
NS-stable3dknown data2026-08-22
goodCount 19 = 130787, hence |V(20)| >= 261574 — two past Stable2's wall
NS-stable3eroutine2026-08-22
Negative controls: odd l does not double; no non-middle toggle; BadAt inhabited
NS-upper1routine2026-08-30
For a numerical set S and any gap f of it, genus S + #x: 0 < x < f, x ∈ S, f-x ∈ S ≤ genus (A S): the gaps of S and the f-complementary pairs of S are disjoint subsets of the gaps of the atom monoid
NS-upper2candidate2026-08-30
|V(ℓ)| ≤ ubound ℓ where ubound (ℓ+2) = 2·ubound (ℓ+1) + 2·ubound ℓ + 3, hence |V(ℓ)| = O((1+√3)^ℓ) and |V(ℓ)| ≤ 3^ℓ; kernel-computed |V(12)| ≤ 136383, |V(28)| ≤ 1313964867583
NS-upper3routine2026-08-30
Negative controls for the upper bound: ℓ ≥ 1 is not removable (|V(0)| = 1 > 0 = ubound 0), the bound is attained at ℓ = 1 and strict at ℓ = 4, the mirror term is what beats 4^ℓ, the counting step is exact, and |V(5)| ≤ 34 / |V(6)| = 79 are refuted
Provenance
- Generated by
- Machina Mathematica
- Released by
- Korea Superintelligence Labs
- Source context
- Source. J. Chappelon, J. L. Ramírez Alfonsín, D. I. Stamate, *Numerical semigroup tree: type-representation*, arXiv:2507.15006; *Experimental Mathematics*, DOI 10.1080/10586458.2025.2604786 (6 Jan 2026).
- Snapshot
- 2026-09-07 03:53 UTC
- Ledger commit
801848d7