Press release · 9 October 2026 · version 1.1.0-candidate
Positive coefficients, failed majorisation
A cyclotomic generating function has strictly positive coefficients, yet its largest parameters break a conjectured ordering—even while every lower-prefix inequality holds.
Summary
What the question is about. Many counting problems produce polynomials that can be written as a fraction built from simple blocks of the form $1+q+\dots+q^{n-1}$, with equally many blocks on top and bottom. When such a fraction turns out to have nonnegative whole-number coefficients, it is a cyclotomic generating function—a family that appears in combinatorial counting, algebra and probability. Deciding whether a proposed fraction really belongs to this family can be hard, because the polynomial may be far too large to expand.
What was conjectured. Billey and Swanson proposed a quick necessary test (their Conjecture 33). Sort the block sizes on top and on bottom into two lists. The test predicts that the top list wins whether we add up the smallest entries or the largest. A test like this would let researchers rule out impossible formulas just by comparing two lists of integers.
What was found. Positive coefficients do not guarantee that ordering. This candidate gives an exact counterexample to one half of the conjecture. Every coefficient of the constructed polynomial is strictly positive, but the largest 5,591 numerator entries lose to the corresponding denominator entries. Every smallest-entry comparison still holds for this example, so the other half of the conjecture is left open. The construction links a known arithmetic fact—a smoothed Möbius sum turns negative at 18350—to the classical device of making coefficients positive by multiplying by a large power of $1+q$, which does not disturb the failed comparison.
Status. An anonymous, unrefereed written-proof candidate. The witness is enormous but exactly specified, and the decisive arithmetic is a small integer certificate that can be rerun. The positivity proof is a written argument; there has been no external peer review, unaffiliated replication or formal verification.
Summary for specialists
Set $X=18350$, let $N$ be the product of the primes at most $X$, and put $D=\varphi(N)$ and $K=D^2N^D$. Split the divisors of $N$ by the sign of $\mu(N/d)$ into $A_0$ and $B_0$, then append $K$ twos to $A_0$ and $K$ ones to $B_0$.
The resulting equal-length lists represent
The sorted lists have common length $m=2^{2101}+K$ (Theorem 1 of the paper). Every coefficient is a strictly positive integer, yet $U_A(5591)<U_B(5591)$, where $U_V(r)$ sums the largest $r$ entries. This refutes the suffix assertion—and hence the conjunction—in Billey–Swanson Conjecture 33. It does not settle the universal lower-prefix assertion.
Technical account
For a list $V$, its hinge at threshold $t$ is $h_V(t)=\sum_{v\in V}(v-t)_+$. At $t=N/X$, Möbius inversion gives
The smoothed Möbius sum $m_1(18350)$ is negative. There are 5,591 denominator entries above the threshold but only 5,570 numerator entries. Equal counts are not needed: the denominator count gives equality in the hinge-to-suffix bound, while an inequality suffices on the numerator side.
The positivity multiplier adds only ones and twos, all below $t$. It therefore leaves the negative hinge unchanged. A binomial/hypergeometric coupling estimate and a lower bound on conjugate root pairs establish positivity for all coefficients; the polynomial is never expanded.
The general proposition exposes the reusable mechanism: a polynomial strictly positive on the nonnegative real axis can acquire positive coefficients under sufficiently large powers of $1+q$, while a negative parameter hinge above 2 survives.
Two remarks locate the failure more precisely. Cancelling one common 1 and one common 2 produces disjoint parameter supports without changing the selected suffix. In the original padded presentation, direct estimates prove every lower-prefix inequality.
Evidence, assurance and limitations
The finite certificate uses integer and rational arithmetic. Its normalised suffix defect is approximately $-2.2401856719\times10^{-6}$; exact directed bounds, not that decimal, determine the sign. Two differently implemented programs regenerate and check the arithmetic, and eight deliberately corrupted certificates must fail in both normal and optimized Python.
The analytic positivity proof and the interpretation of the certificate remain written mathematical arguments. The supplied model-assisted review is not authenticated external peer review. No end-to-end formal verification or unaffiliated reproduction is established.
The witness is exceptionally large but finitely specified. No smallest example, optimal exponent, absolute priority or broader classification is claimed.
Relationship to earlier work
Billey and Swanson supplied the conjecture. Positivity multiplication is established mathematics, including its cyclotomic application by Gatzweiler, Levicán-Santibáñez and Yoshida. Daval recorded the negative arithmetic endpoint. The candidate contribution is the connection between that signed divisor tail and a parameter-order counterexample—not rediscovery of either ingredient.
Who should care, and why
| Audience | Potential use | Required caution |
|---|---|---|
| Algebraic combinatorialists | Correct the proposed necessary conditions for cyclotomic generating functions | The universal prefix assertion remains open here |
| Researchers studying positive polynomials | Reuse the distinction between coefficient smoothing and parameter tails | The positivity principle itself is classical |
| Certificate and verification developers | Replay a small exact computation attached to a huge symbolic object | Arithmetic replay does not prove the analytic bridge |
Why the problem matters
Necessary conditions help rule out impossible generating functions without expanding them. This example shows that strict coefficient positivity alone cannot justify the proposed upper-tail test. It also explains why simply making coefficients more positive cannot rescue that test: the relevant tail does not change.
How to inspect or reproduce the recorded checks
Download the versioned evidence ZIP from the linked GitHub release or Zenodo record. Start with AI_INDEX.md and claims.md, then run:
python3 code/check_package.py
python3 -O code/check_package.py
The programs need Python 3.11 or later and its standard library only. paper.tex builds with two pdfLaTeX passes. The verification document explains which conclusions require reading the proof rather than executing code.
The most valuable next projects
Find a substantially smaller witness; determine which useful parameter conditions survive positivity multiplication; or address the universal lower-prefix assertion separately. These are research directions, not conclusions of this release.
What is in the evidence package
The seven-page paper and source, unchanged exact arithmetic certificate, strengthened trial verifier, corruption controls, claim map, AI index, source audit, review response, licensing map and hash manifest. The public package is a research handoff, not a claim of journal acceptance.
Media
The audio briefing is provided in the header above. Download the MP3 briefing · read the transcript.
Verification status
Unrefereed written-proof candidate with exact internal arithmetic replay. No external peer review, unaffiliated reproduction, formal verification, minimality or absolute priority is established. The supplied model-assisted review is not authenticated external peer review.
Cite
BibTeX
@misc{cgfmajorisationcounterexample2026,
title = {Positive coefficients, failed majorisation},
author = {Anonymous},
year = {2026},
doi = {10.5281/zenodo.23270364},
url = {https://doi.org/10.5281/zenodo.23270364},
version = {1.1.0-candidate},
howpublished = {Zenodo},
note = {Unrefereed; internally replayed evidence package. Press page: https://evidencepress.org/releases/cgf-majorisation-counterexample/}
}Also: cite.bib · paper.json · this page as Markdown