G_δ < 0 on the whole cube: standalone argument for every k ≥ 4, and for k = 3 when α ≥ 1/10

This record contains the preprint "G_δ < 0 on the whole cube: standalone argument for every k ≥ 4, and for k = 3 when α ≥ 1/10". Version 2, 21 November 2026; version 1 is doi:10.5281/zenodo.24424876. Version 2 adds Part 7, "Every step by hand". Preprint, not yet independently refereed. This record re-posts version 2 (first posted as doi:10.5281/zenodo.24429476) so that it is again the latest version; the record doi:10.5281/zenodo.24429600 was a mistaken upload of a different paper and should not be cited for this one. The scripts, logs and proof texts are in the companion code record doi:10.5281/zenodo.24429363. STATEMENT The six-variable Dumont–Foata polynomials have moments m_0, m_1, m_2, ... that are total weights of Motzkin paths. On the cube [α, 1]^6 with coordinates x, y, z, x̄, ȳ, z̄, the moments depend on the six coordinates only through X = x + z̄, Y = x̄ + y, Z = ȳ + z and δ = xȳ + yz̄ + zx̄. The ratio R_k = m_k / m_(k−1)² decreases as δ grows exactly where G_δ = k·m_(k−1)² − 2(k−1)·m_k·m_(k−2) is negative. This paper proves G_δ < 0 at every point of the cube [α, 1]^6 for every k ≥ 4 and every α > 0, and for k = 3 when α ≥ 1/10. For k = 3 the condition on α cannot be dropped: G_δ > 0 at the vertex (α, α, α, 1, 1, 1) once α is below about 0.0675. Even k follow from the Cauchy–Schwarz inequality. For odd k ≥ 5, G_δ decreases as δ grows, so the worst case is δ = 0, where the claim reduces to one inequality on path weights, Lemma L: μ_j² < (8/5)·μ_(j−1)·μ_(j+1) for every even j ≥ 2 and all X, Y, Z in (0, 2], where μ_j is the total weight of the j-step paths from height 1 back to height 1. Lemma L is proved with a 2 × 2 Hankel minor after the path matrix is shifted by an explicit constant c ≤ (3/5)(1 + X + Y + Z). The case k = 3 is settled by an exact Bernstein certificate. The final proof uses no cited theorem. MOTIVATION The published article [D] proves that R_k decreases in each of the six coordinates on [1/2, 1]^6 for every k ≥ 3 (for k ≥ 21 with the article's own note that independent expert review is pending), and the thresholds paper [T] asks how far below 1/2 this survives. By the chain rule, the derivative of R_k along a coordinate is a combination of its derivatives along X, Y, Z and δ. This paper settles the δ part on the whole cube. The standard sufficient conditions for log-convexity of moments ([LRZ], [WZ]) fail at the bottom of the path chain here, because at δ = 0 the first two flat weights give F₀F₁ − λ₁ = −XYZ < 0. Lemma L asks for less than log-convexity, a constant 8/5 in place of 1, and gets it from a constant shift instead. STATUS Even k ≥ 4 (Cauchy–Schwarz): PROVED. G_δ decreases in δ, and the reduction to δ = 0: PROVED. Lemma L, and G_δ < 0 for every odd k ≥ 5: PROVED (machine-checked in exact arithmetic; read by the author). k = 3 for α ≥ 1/10: PROVED (exact Bernstein certificate, 15625 coefficients, three separately written programs). k = 3 fails at the vertex (0.03, 0.03, 0.03, 1, 1, 1): PROVED (G₃ = +1.9201 > 0). The crossing for k = 3 near α = 0.0675: MEASURED. Odd k = 5 to 41 by a second route that does not use Lemma L: PROVED (exact machine certificates). Rebuild from the moment definition at sample points, k = 3 to 21: MEASURED. Part 7, every step of the proof by hand (H1 to H11), each with a case that must fail: PROVED (67 checks in exact fractions; a mutation test catches 14 of 14 deliberate errors; read by the author). The X, Y and Z parts of the derivative of R_k: OPEN. Not yet checked by an independent mathematician. RESULTS Part 1. Even k by Cauchy–Schwarz: G_δ ≤ (2 − k)·m_k·m_(k−2) < 0, with a pencil case at X = Y = Z = 1, k = 4. Part 2. dm_n/dδ = n·m_(n−1), so G_δ is strictly decreasing in δ for odd k ≥ 5; at δ = 0, m_n = XYZ·μ_(n−2), and G_δ < 0 becomes Lemma L. Part 3. The proof of Lemma L: r_n ≥ 1 + X + Y + Z, Cauchy–Schwarz on the ratios, a shifted 2 × 2 Hankel minor, and a two-variable polynomial Φ that is positive on the needed range. Part 4. The shift lemma: the path matrix plus c = max(0, (1 + Z(W − 1))/2, (Z + 1)W/4), with W = X + Y, is non-negative on finitely supported vectors, and c ≤ (3/5)(1 + X + Y + Z) by a four-corner check. Part 5. k = 3: the closed form of G₃ by hand, and the exact Bernstein certificate on [α₀, 1]^6 for α₀ = 1/10, 3/20, 21/100 and 531/2500, with a control at α₀ = 1/20 that fails as it must. Part 6. Pencil cases at X = Y = Z = 1 (μ_0 to μ_5 = 1, 4, 33, 392, 6497, 143244), Lemma L at j = 2 step by step, and the cases that must fail: X = Y = Z = 32 for Lemma L, X = Y = Z = 16 for the bound X, Y, Z ≤ 2, and α = 0.03 for k = 3. How we got here. The path to the result in the order it was found, wrong turns included: the dual Hahn route, the error in the first version of the proof and how it was caught, and a teaching note on each idea. What was checked. Every claim with its status, its machine check and the file and fingerprint behind it, and the steps a reader should check first. Part 7 (new in version 2). Every step by hand: eleven blocks H1 to H11 that redo each step of the proof with pencil-sized numbers, each in plain words, by hand, and with a case where the step must fail. FILES The preprint, lemmaL_paper_v2_2bada83e44a78ed8.pdf (22 pages). The code record, doi:10.5281/zenodo.24429363 (version 2 of the record), holds lemmaL_code_record_v2_8dce9f11e4448157.zip (SHA-256 beginning 8dce9f11e4448157; SHA256SUMS.txt lists 54 files) under the MIT licence: every script the paper cites, with its log, including the check and mutation test of Part 7, and the proof of Lemma L and its worked-by-hand block as plain text. 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 [D] J. W. Davis, A monotonicity property of the six-variable Dumont–Foata polynomials, Zenodo, doi:10.5281/zenodo.24158125. [T] J. W. Davis, Monotonicity thresholds for the six-variable Dumont–Foata ratios, Zenodo, doi:10.5281/zenodo.24423914. [GZ] I. M. Gessel and J. Zeng, Moments of orthogonal polynomials and exponential generating functions, Ramanujan J., doi:10.1007/s11139-022-00548-6; arXiv:2107.00255. [KLS] R. Koekoek, P. A. Lesky and R. F. Swarttouw, Hypergeometric Orthogonal Polynomials and Their q-Analogues, Springer, 2010, Section 9.3. [LRZ] H. Liang, J. Remmel and S. Zheng, Stieltjes moment sequences of polynomials, arXiv:1710.05795. [WZ] Y. Wang and Z.-H. Zhang, Combinatorics of generalized Motzkin numbers, J. Integer Seq. 18 (2015), Article 15.2.4.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-21
DOI
https://doi.org/10.5281/zenodo.24424875
Primary Topic
Advanced Combinatorial Mathematics
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

G_δ < 0 on the whole cube: standalone argument for every k ≥ 4, and for k = 3 when α ≥ 1/10

Joshua W. Davis
Zenodo (CERN European Organization for Nuclear Research)
Advanced Combinatorial Mathematics
preprint

G_δ < 0 on the whole cube: standalone argument for every k ≥ 4, and for k = 3 when α ≥ 1/10

Joshua W. Davis
preprint en

Abstract

This record contains the preprint "G_δ < 0 on the whole cube: standalone argument for every k ≥ 4, and for k = 3 when α ≥ 1/10". Version 2, 21 November 2026; version 1 is doi:10.5281/zenodo.24424876. Version 2 adds Part 7, "Every step by hand". Preprint, not yet independently refereed. This record re-posts version 2 (first posted as doi:10.5281/zenodo.24429476) so that it is again the latest version; the record doi:10.5281/zenodo.24429600 was a mistaken upload of a different paper and should not be cited for this one. The scripts, logs and proof texts are in the companion code record doi:10.5281/zenodo.24429363. STATEMENT The six-variable Dumont–Foata polynomials have moments m_0, m_1, m_2, ... that are total weights of Motzkin paths. On the cube [α, 1]^6 with coordinates x, y, z, x̄, ȳ, z̄, the moments depend on the six coordinates only through X = x + z̄, Y = x̄ + y, Z = ȳ + z and δ = xȳ + yz̄ + zx̄. The ratio R_k = m_k / m_(k−1)² decreases as δ grows exactly where G_δ = k·m_(k−1)² − 2(k−1)·m_k·m_(k−2) is negative. This paper proves G_δ < 0 at every point of the cube [α, 1]^6 for every k ≥ 4 and every α > 0, and for k = 3 when α ≥ 1/10. For k = 3 the condition on α cannot be dropped: G_δ > 0 at the vertex (α, α, α, 1, 1, 1) once α is below about 0.0675. Even k follow from the Cauchy–Schwarz inequality. For odd k ≥ 5, G_δ decreases as δ grows, so the worst case is δ = 0, where the claim reduces to one inequality on path weights, Lemma L: μ_j² < (8/5)·μ_(j−1)·μ_(j+1) for every even j ≥ 2 and all X, Y, Z in (0, 2], where μ_j is the total weight of the j-step paths from height 1 back to height 1. Lemma L is proved with a 2 × 2 Hankel minor after the path matrix is shifted by an explicit constant c ≤ (3/5)(1 + X + Y + Z). The case k = 3 is settled by an exact Bernstein certificate. The final proof uses no cited theorem. MOTIVATION The published article [D] proves that R_k decreases in each of the six coordinates on [1/2, 1]^6 for every k ≥ 3 (for k ≥ 21 with the article's own note that independent expert review is pending), and the thresholds paper [T] asks how far below 1/2 this survives. By the chain rule, the derivative of R_k along a coordinate is a combination of its derivatives along X, Y, Z and δ. This paper settles the δ part on the whole cube. The standard sufficient conditions for log-convexity of moments ([LRZ], [WZ]) fail at the bottom of the path chain here, because at δ = 0 the first two flat weights give F₀F₁ − λ₁ = −XYZ < 0. Lemma L asks for less than log-convexity, a constant 8/5 in place of 1, and gets it from a constant shift instead. STATUS Even k ≥ 4 (Cauchy–Schwarz): PROVED. G_δ decreases in δ, and the reduction to δ = 0: PROVED. Lemma L, and G_δ < 0 for every odd k ≥ 5: PROVED (machine-checked in exact arithmetic; read by the author). k = 3 for α ≥ 1/10: PROVED (exact Bernstein certificate, 15625 coefficients, three separately written programs). k = 3 fails at the vertex (0.03, 0.03, 0.03, 1, 1, 1): PROVED (G₃ = +1.9201 > 0). The crossing for k = 3 near α = 0.0675: MEASURED. Odd k = 5 to 41 by a second route that does not use Lemma L: PROVED (exact machine certificates). Rebuild from the moment definition at sample points, k = 3 to 21: MEASURED. Part 7, every step of the proof by hand (H1 to H11), each with a case that must fail: PROVED (67 checks in exact fractions; a mutation test catches 14 of 14 deliberate errors; read by the author). The X, Y and Z parts of the derivative of R_k: OPEN. Not yet checked by an independent mathematician. RESULTS Part 1. Even k by Cauchy–Schwarz: G_δ ≤ (2 − k)·m_k·m_(k−2) < 0, with a pencil case at X = Y = Z = 1, k = 4. Part 2. dm_n/dδ = n·m_(n−1), so G_δ is strictly decreasing in δ for odd k ≥ 5; at δ = 0, m_n = XYZ·μ_(n−2), and G_δ < 0 becomes Lemma L. Part 3. The proof of Lemma L: r_n ≥ 1 + X + Y + Z, Cauchy–Schwarz on the ratios, a shifted 2 × 2 Hankel minor, and a two-variable polynomial Φ that is positive on the needed range. Part 4. The shift lemma: the path matrix plus c = max(0, (1 + Z(W − 1))/2, (Z + 1)W/4), with W = X + Y, is non-negative on finitely supported vectors, and c ≤ (3/5)(1 + X + Y + Z) by a four-corner check. Part 5. k = 3: the closed form of G₃ by hand, and the exact Bernstein certificate on [α₀, 1]^6 for α₀ = 1/10, 3/20, 21/100 and 531/2500, with a control at α₀ = 1/20 that fails as it must. Part 6. Pencil cases at X = Y = Z = 1 (μ_0 to μ_5 = 1, 4, 33, 392, 6497, 143244), Lemma L at j = 2 step by step, and the cases that must fail: X = Y = Z = 32 for Lemma L, X = Y = Z = 16 for the bound X, Y, Z ≤ 2, and α = 0.03 for k = 3. How we got here. The path to the result in the order it was found, wrong turns included: the dual Hahn route, the error in the first version of the proof and how it was caught, and a teaching note on each idea. What was checked. Every claim with its status, its machine check and the file and fingerprint behind it, and the steps a reader should check first. Part 7 (new in version 2). Every step by hand: eleven blocks H1 to H11 that redo each step of the proof with pencil-sized numbers, each in plain words, by hand, and with a case where the step must fail. FILES The preprint, lemmaL_paper_v2_2bada83e44a78ed8.pdf (22 pages). The code record, doi:10.5281/zenodo.24429363 (version 2 of the record), holds lemmaL_code_record_v2_8dce9f11e4448157.zip (SHA-256 beginning 8dce9f11e4448157; SHA256SUMS.txt lists 54 files) under the MIT licence: every script the paper cites, with its log, including the check and mutation test of Part 7, and the proof of Lemma L and its worked-by-hand block as plain text. 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 [D] J. W. Davis, A monotonicity property of the six-variable Dumont–Foata polynomials, Zenodo, doi:10.5281/zenodo.24158125. [T] J. W. Davis, Monotonicity thresholds for the six-variable Dumont–Foata ratios, Zenodo, doi:10.5281/zenodo.24423914. [GZ] I. M. Gessel and J. Zeng, Moments of orthogonal polynomials and exponential generating functions, Ramanujan J., doi:10.1007/s11139-022-00548-6; arXiv:2107.00255. [KLS] R. Koekoek, P. A. Lesky and R. F. Swarttouw, Hypergeometric Orthogonal Polynomials and Their q-Analogues, Springer, 2010, Section 9.3. [LRZ] H. Liang, J. Remmel and S. Zheng, Stieltjes moment sequences of polynomials, arXiv:1710.05795. [WZ] Y. Wang and Z.-H. Zhang, Combinatorics of generalized Motzkin numbers, J. Integer Seq. 18 (2015), Article 15.2.4.

Zenodo (CERN European Organization for Nuclear Research)
Advanced Combinatorial Mathematics
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon