Franel polynomials as moment sequences: where they fail
This record contains the preprint "Franel polynomials as moment sequences: where they fail". Preprint, 18 November 2026. Not yet independently refereed. The check script, its log and its mutation log are in the companion code record doi:10.5281/zenodo.24410968. STATEMENT The Franel polynomials are F_n(x) = Σ_(k=0..n) C(n,k)³ x^k. At x = 1 they give the Franel numbers 1, 2, 10, 56, 346, …. The Franel numbers are a Hamburger moment sequence (Zhu and Sun) but not a Stieltjes one: the shifted 3 × 3 Hankel determinant is −504, worked here by hand. For each of the seven values x = 1/2, 11/10, 5/4, 3/2, 2, 3 and 5, the sequence (F_n(x)) is not even a Hamburger moment sequence: an exact Hankel determinant is negative. The note conjectures that (F_n(x)) is a Hamburger moment sequence only for x = 0 and x = 1, and a Stieltjes moment sequence only for x = 0 (x ≥ 0). MOTIVATION A sequence (m_n) is a Hamburger moment sequence when m_n = ∫ t^n dμ(t) for a positive measure μ on the real line, and a Stieltjes moment sequence when μ can be taken on [0, ∞). By Hamburger's and Stieltjes' theorems, the first holds exactly when every Hankel matrix [m_(i+j)] is positive semidefinite, and the second when, in addition, every shifted Hankel matrix [m_(i+j+1)] is. So one negative leading minor is a complete proof of failure, while a thousand positive minors prove nothing, because the next one could be negative. That asymmetry runs through the whole note, and it is why each result is stated value by value. The question arose while classifying the Apéry polynomials as moment sequences, when the shifted Hankel determinants of the Franel numbers were computed and the 3 × 3 one came out as −504. These polynomials are not the Franel x-analogues m_k(x) = Σ_j C(k,j)² C(2j,k) x^j of the companion Apéry paper (doi:10.5281/zenodo.24331816), which are Hamburger moment sequences for every x > 0. STATUS The Franel numbers (x = 1) are a Hamburger moment sequence: PROVED by Zhu and Sun (every Hankel determinant positive); cited. The Franel numbers are not a Stieltjes moment sequence: PROVED (one negative shifted determinant, −504, by hand); first stated by Liang, Wang and Zheng. F_n(x) is not a Hamburger moment sequence at x = 1/2, 11/10, 5/4, 3/2, 2, 3, 5: PROVED at each of these seven values (an exact negative Hankel determinant). The characteristic roots of the recurrence are (1 + ω^j x^(1/3))³, ω = e^(2πi/3): MEASURED at x = 1/2, 1, 2, 3, 5 (exact recurrences found by search); checked by hand at x = 1 and x = 3. F_n(x) is a Hamburger moment sequence only for x = 0 and x = 1, and a Stieltjes moment sequence only for x = 0: OPEN (conjecture). Not yet checked by an independent mathematician. RESULTS The size of the first non-positive Hankel determinant: 4 at x = 5 and x = 3, 5 at x = 2, 7 at x = 3/2, 9 at x = 5/4, 13 at x = 11/10, and 5 at x = 1/2. The failure comes later the closer x is to 1. At x = 1 there is none at any size, by Zhu and Sun's theorem, and the same exact test finds none up to size 26. Each row is a theorem about that one value of x; the pattern between them is the conjecture. A heuristic reading, not a proof. The recurrence for F_n(x) has characteristic roots (1 + ω^j x^(1/3))³ for j = 0, 1, 2. The real root dominates, and the other two form a complex pair, except at x = 1, where both equal −1. A measure on the real line can produce a negative real root; a complex pair forces oscillating corrections that no positive measure can match, and a large enough Hankel determinant sees them. The closer x is to 1, the closer the pair is to the real axis, and the larger the determinant that first sees it. A root at −1 is compatible with a measure on the line but not on [0, ∞): exactly the Hamburger-but-not-Stieltjes case at x = 1. What a proof for every x needs: an argument that turns "the secondary characteristic roots are a non-real pair" into "some Hankel determinant is negative", with the size of that determinant controlled. An explicit determinant for each x cannot cover infinitely many x. The note claims nothing for x < 0, and nothing at other x beyond the seven values computed. APPENDIX: THE MATHEMATICS The Franel numbers. f_n = F_n(1) = Σ C(n,k)³; for example f_2 = 1 + 8 + 1 = 10 and f_3 = 1 + 27 + 27 + 1 = 56. The first eight are 1, 2, 10, 56, 346, 2252, 15184, 104960. The first Hankel determinants are 1, 6, 180, 28296; divided by 1, 6, 36, 216 they are 1, 1, 5, 131, all odd, as Zhu and Sun's theorem says. Not Stieltjes. Expand det[[2, 10, 56], [10, 56, 346], [56, 346, 2252]] along the first row: 2 × (56·2252 − 346²) = 2 × (126112 − 119716) = 12792; −10 × (10·2252 − 346·56) = −10 × (22520 − 19376) = −31440; 56 × (10·346 − 56²) = 56 × (3460 − 3136) = 18144. The sum is −504 < 0. A Stieltjes sequence needs every shifted minor to be ≥ 0, so the Franel numbers are not Stieltjes. Not Hamburger at x = 5, by a 4 × 4 elimination. The values are F_0, …, F_6 = 1, 6, 66, 936, 14346, 231876, 3885456; for example F_3(5) = 1 + 27·5 + 27·25 + 125 = 936. Eliminate below the diagonal of [F_(i+j)], i, j = 0, …, 3. Row 2 − 6·row 1, row 3 − 66·row 1 and row 4 − 936·row 1 give pivot 30; then row 3 − 18·row 2 and row 4 − 291·row 2 give pivot 270; then row 4 − 48·row 3 leaves 468930 − 48·12960 = 468930 − 622080 = −153150. The pivots are 1, 30, 270, −153150, so the leading minors are 1, 30, 8100, −1240515000. The fourth is negative, so (F_n(5)) is not a Hamburger moment sequence. The checks that must not fail. At x = 1 the same exact test, run to size 13 (where it catches x = 11/10) and on to size 26, finds no non-positive determinant. At x = 0 the determinants are 1, 0, 0, …, never negative. So the test does not call a known moment sequence a non-moment sequence. The characteristic roots by hand. Franel's recurrence (n+1)² f_(n+1) = (7n² + 7n + 2) f_n + 8n² f_(n−1), divided by n² as n grows, gives t² = 7t + 8, that is (t − 8)(t + 1) = 0: the roots are 8 = (1 + 1)³ and −1 = (1 + ω)³, since 1 + ω = −ω². At x = 3, with a = 3^(1/3), the three roots sum to 3 + 3a³ = 12 and multiply to (1 + a³)³ = 64, matching the polynomial t³ − 12t² − 33t − 64 that the search found. Every number in the note is checked in exact rational arithmetic with the Python standard library (franel_note_check.py, 17 checks, all pass; 9 of 9 deliberate errors caught, including an error in the row-swap step that is caught only by a control that forces a swap). The note's Section 7 records how the results were found, including two references first recorded wrongly, one by a survey and one by a quick search, both caught only by reading the original. FILES The preprint (PDF, 7 pages). The code record, doi:10.5281/zenodo.24410968, holds franel_code_record_v1.zip (SHA-256 beginning aa26284d539e0f7a; SHA256SUMS.txt lists 5 files) under the MIT licence: franel_note_check.py (cb727fcbe06108ea), its log franel_note_check.log (0d72212ebc64eb39), and mutation_franel.log (86ff9af06c0be54b). 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 [ZS] B.-X. Zhu, Z.-W. Sun, Hankel-type determinants for some combinatorial sequences, Int. J. Number Theory 14 (2018), no. 5, 1265–1277; arXiv:1609.06810. [LWZ] H. Liang, Y. Wang, S. Zheng, Hamburger moment sequences in combinatorics, Acta Math. Sin. (Engl. Ser.) 34 (2018), no. 7, 1101–1109, doi:10.1007/s10114-017-7230-z (Section 4, p. 1108). [S] A. D. Sokal, The Euler and Springer numbers as moment sequences, Expo. Math. 38 (2020) 1–26; arXiv:1804.04498 (Theorems 2.1 and 2.2, the Hamburger and Stieltjes criteria). [BCHKSW] A. Bostan et al., arXiv:2001.00393 (section 1.6; cited only for a misreading of [LWZ]).
Authors
- Joshua W. Davis
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-21
- DOI
- https://doi.org/10.5281/zenodo.24410975
- Primary Topic
- Mathematical functions and polynomials
- Type
- preprint