trureturingGitHub MATHEMATICAL ATLAS

RESEARCH / Focused target

Bottom heaviness of the four q-Stirling coefficient families

Literature status: not recheckedRepository record: proved

RELEASED FOUNDATIONS / PROPOSED CONNECTION

Research connections

0 released anchors
All source anchors (1)

Problem

The question

Bruce E. Sagan and Joshua P. Swanson, q-Stirling numbers in type B, arXiv:2205.14078v1, Section 7.5, Conjecture 7.6, printed page 40:

For each n, k, the coefficients of S[n, k], c[n, k], S_B[n, k], and c_B[n, k] are all bottom heavy.

The preceding definition, printed page 39, is:

A sequence is bottom heavy if aₖ ≥ aₙ₋ₖ for k < n/2.

The published version, European Journal of Combinatorics 118 (2024), 103899, DOI 10.1016/j.ejc.2023.103899, repeats the conjecture on page 31.

Statement echo

For every $n\in\mathbb N$, $k\in\mathbb Z$, and $P\in{S[n,k],c[n,k],S_B[n,k],c_B[n,k]}$, the conclusion is

$$ 2r<\operatorname{natDegree}(P)\quad\Longrightarrow\quad [q^{\operatorname{natDegree}(P)-r}]P\leq[q^r]P \qquad(r\in\mathbb N). $$

The Lean definitions S, c, SB, and cB use the source's common delta initial row and its recurrences (arXiv equations (2), (5), (8), and (9)). The names SB and cB denote $S_B$ and $c_B$. The existing D5.S0.Certificates.Polynomials.QProductNecessityRefutation.qInteger supplies $[a]q=\sum{j<a}q^j$ for natural $a$, including $[0]_q=0$. The second-kind definitions explicitly give zero for negative $k$; this is the zero branch of the source recurrence, so no negative q-integer is represented by an incorrect polynomial. All integer $k$ remain quantified. The first-kind multipliers have natural lengths $n$ and $2n+1$ in successor rows. Coefficients start at exponent zero. The zero polynomial has $\operatorname{natDegree}(0)=0$; it and the constant polynomials have no strict-half index. There is no additional positivity or valid-index hypothesis. The separately shifted $\overline S[n,k]=q^{\binom{k}{2}}S[n,k]$ is excluded.

Motivation

Our foothold

The coefficient comparison concerns each fixed polynomial, rather than log-concavity of a row indexed by $k$. It permits nonunimodal polynomials. The settling theorem is D5/S3/Combinatorics/QStirling/SaganSwansonBottomHeavy.result. The complete source contract and literature screen are in preregistration #21443.

Gap

Missing bridges

The bounded literature screen recorded in #21443 found no earlier settlement at preregistration. Its completeness is ASSUMED-UNVERIFIED: a complete forward-citation inventory and all publisher-only full texts were unavailable. The screen distinguishes Conjecture 7.6 from the separate Conjectures 7.4, 7.5, and the resolved 7.9. The journal statement and arXiv conventions agree.

Independent later settlement

PrincipiaMathOrg/math-solutions, immutable source, path site-backup/mathdb-354788/mathdb-354788-solution.tex, commit 47371706c400c79d06e3fb62190fcf6f7a725c85 (47371706), contains an independent settlement. Its first public commit was 2026-10-27T02:59Z, after preregistration at 2026-10-25T04:27:04Z, as recorded in #21443. Eligibility under CLAUDE.md §3.2 is assessed at preregistration; this later public appearance therefore preserves the preregistered eligibility. This delivery claims a formal settlement without exclusive priority. The third-party implementation is not imported or copied.

Route

Proposed approach

Write $U_a=1+q+\cdots+q^{a-1}$ for $a>0$. The recursive identity

$$ U_{a+1}U_{b+1}=U_{a+b+1}+qU_aU_b $$

constructs concentric interval layers. The private predicate Centered d p is generated by the interval $U_{d+1}$, addition at a common center, and the shift $(d,p)\mapsto(d+2,qp)$. Structural induction proves support at most $d$, reflection symmetry about $d/2$, and coefficient increase toward the center. The private theorem centered_mul_interval proves closure under multiplication by $U_{a+1}$.

For a centered layer with positive constant coefficient, symmetry makes its coefficient at $d$ positive. Any ambient support bound $D$ therefore satisfies $d\leq D$. If $D-r>d$, the reflected coefficient is zero. Otherwise put $t=d-(D-r)$. For $2r<D$ one has $t\leq r$ and $t+r\leq d$; symmetry followed by increase toward the center proves $[q^{D-r}]p\leq[q^r]p$.

The private predicate Cone p includes zero, positive-constant centered atoms, and their sums. In natural coefficients, a support bound on a sum bounds each summand separately. Consequently cone_ambient proves the comparison at the ambient degree of the entire sum. cone_mul and families_cone show that all four recurrences preserve this class. Applying the support bound at $D=\operatorname{natDegree}(P)$ gives result.

Falsifier

What would falsify this route

A counterexample would specify one of the four unshifted source polynomials, an integer $k$, a natural $n$, and a natural strict-half index $r$ whose reflected coefficient is larger. The theorem rules out such an instance for the displayed definitions. A differently shifted polynomial or a comparison below the polynomial's support bound is outside the conclusion.

Evidence

Evidence to collect

Kernel-checked: D5.S3.Combinatorics.QStirling.SaganSwansonBottomHeavy.result proves the full four-family claim. The axiom closure of every public declaration is contained in ${\mathrm{propext},\mathrm{Classical.choice},\mathrm{Quot.sound}}$. The corresponding Blueprint displays all five public definitions and the expanded proposition of result.

Computed: immutable experiment directory. Command: python3 -I -S /Users/auric/trureturing-experiments/docs/reports/ss2205-bottomheavy/check.py; exit 0. The checked file is byte-identical to the version at that commit, SHA-256 62e24222171d76397e28b0485fe8300358aa15159282611dccf9abfa0ca70d45. It reproduces the complete nonunimodal but bottom-heavy coefficient lists of $S_B[6,4]$ and $c_B[7,5]$, and $S[4,3]=3+2q+q^2$. Its exact ranges are: all four families with $0\leq n\leq50$ and $-3\leq k\leq53$ (11,628 polynomials and 1,187,487 comparisons); 216 independent elementary/complete-homogeneous identities with $0\leq n\leq8$, $0\leq k\leq n+1$; 1,555 products of zero through four q-integers of lengths 1 through 6; 161,736 ambient comparisons with $d\leq D\leq d+12$; and 7,056 weighted sums of two short products with weights zero through three. These finite checks are not the universal proof.

The four-slot escape audit of the sole public theorem result is unfinished: issue #22255. No Reg registration or declared_validated status is asserted. The missing source reconstruction, enrolled readout, and variation/sensitivity evidence must preserve the polynomial-selection disjunction and degree guard.

Triage

Scope assessment

What the settlement shows

  • Mechanism — proved, kernel-checked: interval_mul and centered_mul_interval establish the interval-layer multiplication invariant; cone_mul, families_cone, and cone_ambient preserve it through the four recurrences and compare at the actual degree of the whole sum. These are private declarations of the settling module, used by result.
  • Formal extension — proved, kernel-checked: the private cone_ambient proves the comparison for every Cone p and every natural ambient $D$ above all nonzero coefficients, under $2r<D$. It is not restricted to the four source recurrences. The coefficients remain natural.
  • Finite validation — computed: the experiment in Evidence reproduces $S_B[6,4]$ and $c_B[7,5]$ as nonunimodal and bottom heavy; every finite and structural check in the stated ranges exits 0. No larger numerical range is claimed.
  • Ambient-degree condition — computed: the checker's convolution of $(1,1)$ with itself gives $(1,2,1)$, of actual degree $2$. At $D=1$, $r=0$, the reflected coefficient is $2>1$. Thus removing $D\geq d$ invalidates the generic comparison, although the original claim always uses the actual degree. This product occurs in the experiment's length-two product range; the failure at $D=1$ is direct arithmetic on its displayed coefficients.
  • Exponent-zero anchor — computed: the checker's table gives $S[2,2]=1$. The source's shifted family gives $\overline S[2,2]=q$, coefficient list $(0,1)$. At actual degree $1$ and $r=0$, its reflected coefficient is $1>0$. Exponent shifts therefore need not preserve bottom heaviness. The table entry lies in the experiment's stated range; the shifted comparison is direct arithmetic using the source's definition.
  • Surviving neighbours — open: coefficient log-concavity, parity log-concavity, general unimodality, asymptotic normality, Conjectures 7.4 and 7.5, and a formal version for arbitrary real nonnegative weights are not settled by this module. The source's other established results remain statements of that source; the bounded screen establishes no consequence for another result depending on Conjecture 7.6.

ASSUMED-UNVERIFIED

Unverified assumptions

Literature completeness and exclusive novelty are not established. Source and literature checks attributed to the orchestrator in #21443 are supplied reports. The incomplete citation inventory and unscreened publisher/web material remain possible locations for other settlements. The separate escape audit remains unfinished under #22255; this does not add an axiom to the mathematical proof.

Dossier & anchor history

Loading release observations...