Hankel positivity of row polynomials of column-weight triangles: a necessary condition from the top degree
Year: 2026
Type: preprint
Abstract: This record contains the preprint "Hankel positivity of row polynomials of column-weight triangles: a necessary condition from the top degree". Preprint, 21 November 2026. Not yet independently refereed. The check scripts, the two registered tests and their logs are in the companion code record doi:10.5281/zenodo.24332213. STATEMENT For a triangle defined by T(0, 0) = 1 and T(n, k) = T(n−1, k−1) + w_k T(n−1, k), let a_n(q) = Σ_k T(n, k) q^k be its row polynomials and H_k(q) = det[a_(i+j)(q)] (0 ≤ i, j ≤ k−1) their leading Hankel determinants. Theorem 1: for every k ≥ 2, H_k(q) has degree at most (k−1)² in q, and its coefficient of q^((k−1)²) is the Hankel determinant of the forward differences of the weights at 0, det[(Δ^(i+j+1) w)(0)] over 0 ≤ i, j ≤ k−2. If that number is negative for some k, the row polynomials are not coefficientwise Hankel-totally positive, and for all large real q they are not even a Hamburger moment sequence. MOTIVATION Coefficientwise Hankel-total positivity of row polynomials is a strong positivity property studied by Sokal, Pétréolle, Zhu and others; it is usually proved through continued fractions or production matrices, and it is hard to refute, because a failure has to be found in some minor of an infinite matrix. Zhu conjectured it for the generalised Jacobi–Stirling triangle, with weights w_k = xk² + yk. Theorem 1 gives a test that needs only the weights: the top coefficient of each Hankel determinant can be read off from the differences of the weights, without computing a single row polynomial. At size 3 it is d f − e², where d, e, f are the first, second and third differences of the weights at 0. STATUS Theorem 1 (every k): PROVED, by a six-step argument; not yet checked by an independent mathematician. It was also confirmed by computer: symbolically for general weights at k = 3 and 4, and at k = 5 and 6 in a test registered before it ran (fingerprint 759001ec19a4c518: 12 of 12 fresh random weight sequences at each size, with controls): REGISTERED PASS. Corollaries 1 to 5 (consequences for quadratic, polynomial and power weights): PROVED, not yet checked by an independent mathematician. The five examples from Gelineau and Zeng: an exact negative value of H_3 for each failing triangle, in a registered test (fingerprint 765a8b04aa22ade4; 9 of 9 explicit negatives found; Stirling clean to 5 × 5): REGISTERED PASS. Theorem 2 and Proposition 3 (Appendix B, more than one weight per column): PROVED, and checked on 28 random cases. Zhu's conjecture for the generalised Jacobi–Stirling triangle: answered in the negative (Theorem 1 and an explicit negative coefficient of H_3). Novelty is not established, and the paper claims no priority. RESULTS Corollary 2: quadratic weights w_k = Ak² + Bk + C (A ≠ 0) give d = A + B, e = 2A, f = 0, so the q⁴ coefficient of H_3 is −4A² < 0. Quadratic weights never give coefficientwise Hankel-total positivity. Corollary 3: for a polynomial weight of degree r with leading coefficient L, the top coefficient at size r + 1 is (−1)^(r(r−1)/2)(r!·L)^r; for L > 0 it is negative exactly when r ≡ 2 or 3 (mod 4). Corollary 4: for w_k = k^r (r ≥ 2) the size-3 coefficient is 3^r + 2^r − 4^r − 1 < 0. Corollary 5: reading Theorem 1's matrix as a Hankel matrix of moments, polynomial weights of every degree r ≥ 2 fail by size ⌈r/2⌉ + 2, unless an earlier determinant is exactly 0. Examples: the row polynomials of the central factorial, odd central factorial, Legendre–Stirling and Jacobi–Stirling triangles are not coefficientwise Hankel-totally positive; the Stirling triangle passes the test, as it must. For the generalised Jacobi–Stirling triangle, H_3 = 2q³(6x³ + 11x²y + 6xy² + y³ − 2x²q), which has the negative coefficient −4x²; for example H_3 = −7/250 at x = 1/10, y = 0, q = 1. Appendix B: Theorem 1 extends to the Catalan–Stieltjes triangles of Zhu's Question 5.1 (down-step weights do not change the top coefficient), while weights on the up-steps change the leading degree. APPENDIX: THE MATHEMATICS Appendix A of the paper works every step with pencil (A1 to A9), each with a nearby case where the method must pass. The idea of the proof. Write the triangle as paths: T(n, k) = (P^n)(0, k) for the matrix P with w_m on the diagonal and 1 above it. The Hankel matrix then factors through a unitriangular matrix, and after factoring out q^m from row m every entry has the same top term q^j: a rank-one top layer. Subtracting neighbouring rows removes it, and what is left at the top degree is det[w_(m+j) − w_m], which column operations turn into the Hankel determinant of the differences. At size 3, by hand (A3). Steps 5 and 6 give det[[w_1 − w_0, w_2 − w_0], [w_2 − w_1, w_3 − w_1]]; subtracting column 1 from column 2 gives Δw(0)Δw(2) − Δw(1)². With Δw(1) = d + e and Δw(2) = d + 2e + f this is d(d + 2e + f) − (d + e)² = df − e². For the Jacobi–Stirling weights w_k = xk² + yk: d = x + y, e = 2x, f = 0, so df − e² = −4x². The central factorial triangle, w_k = k² (A4). The rows are a_0 = 1, a_1 = q, a_2 = q + q², a_3 = q + 5q² + q³, a_4 = q + 21q² + 14q³ + q⁴. The differences of 0, 1, 4, 9 are d = 1, e = 2, f = 0, so Theorem 1 predicts a q⁴ coefficient of −4, and indeed H_3 = −4q⁴ + 12q³ = −4q³(q − 3). At q = 10 the rows give 1, 10, 110, 1510, 26110 and H_3 = 592,000 − 950,000 + 330,000 = −28,000 = −4 × 1000 × 7. The case that must pass. The Stirling weights w_k = k give p, r, s = 1, 2, 3 in the formula of A5, so H_3 = 2q³, with no negative coefficient, consistent with the known positivity of the Stirling (Bell/Touchard) row polynomials. In Corollary 4, r = 1 gives 3 + 2 − 4 − 1 = 0: the test is silent there, as it must be. Where positivity stops (A6). For the Jacobi–Stirling triangle at x = y = 1 the 2 × 2 minors are (1 + y)q and (3 + y)q³, both nonnegative, while the 3 × 3 determinant is 48q³ − 4q⁴. Coefficientwise positivity holds at size 2 and fails at size 3. Appendix A9 works the table of Section 5, Corollaries 3 and 4, and the last step of the proof at size 2. Every number in Appendix A is checked in exact arithmetic by the scripts in the code record; A9 is checked by two separately written scripts. Appendix C records how the result was found, including the wrong turns. FILES The preprint (PDF, 19 pages). The code record, doi:10.5281/zenodo.24332213, holds paper2_code_record_v6.zip (SHA-256 beginning 0ab917bb9cd957be; SHA256SUMS.txt lists 34 files) under the MIT licence: the hand-check scripts for every block of Appendix A, the two registered tests with their frozen registrations and results, an independent re-verification script, and all logs. AI ASSISTANCE The work was directed by the author. Most of the computational work was done with Claude (Anthropic) and Gemini (Google), with support from Grok (xAI). The text was drafted with AI assistance from the project's records and edited by the author, who takes responsibility for its contents. REFERENCES [DS] B. Deb, A. D. Sokal, Higher-order Stirling cycle and subset triangles: total positivity, continued fractions and real-rootedness, arXiv:2507.18959 (2025). [GZ] Y. Gelineau, J. Zeng, Combinatorial interpretations of the Jacobi–Stirling numbers, arXiv:0905.2899 (2009). [LZ] Z. Lin, J. Zeng, Positivity properties of Jacobi–Stirling numbers and generalized Ramanujan polynomials, arXiv:1309.4237 (2013). [PSZ] M. Pétréolle, A. D. Sokal, B.-X. Zhu, Lattice paths and branched continued fractions, arXiv:1807.03271 (2018). [Zhu] B.-X. Zhu, Coefficientwise Hankel-total positivity of the row-generating polynomials for the output matrices of certain production matrices, Adv. Math. 447 (2024), 109695; arXiv:2202.03793.
Author: Davis, Joshua W
Language: English
Subfield: Discrete Mathematics and Combinatorics
Field: Mathematics
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7237340820
Cites:
Cited by:
Related to:
