Exact rational certificates in Lean 4 and a Meixner–Pollaczek moment representation

This record contains the preprint "The Apéry polynomials as Hamburger and Stieltjes moment sequences". Preprint, November 2026. Not yet independently refereed. The code, registrations, logs and the exact recurrence certificate are in the companion record doi:10.5281/zenodo.24331682. STATEMENT For real x ≥ 0 let A_n(x) = Σ_(k=0..n) C(n,k)² C(n+k,k)² x^k be the Apéry polynomials. A sequence (a_n) is a Hamburger moment sequence if a_n = ∫ s^n dμ(s) for a positive measure μ on the real line, and a Stieltjes moment sequence if μ can be taken on [0, ∞). Then (A_n(x))_(n≥0) is a Hamburger moment sequence exactly when x ∈ {0, 1/4} ∪ [1, ∞), and a Stieltjes moment sequence exactly when x = 0 or x ≥ 1. At x = 1/4 the representing measure lives on [−1, 7 + 4√3] and puts mass on (−1, 0), so there is none on [0, ∞). MOTIVATION At x = 1 these are the Apéry numbers 1, 5, 73, 1445, …, which appear in Apéry's proof that ζ(3) is irrational. A. D. Sokal conjectured that they form a Stieltjes moment sequence, and G. A. Edgar proved it. In a 2018 lecture Sokal proposed the polynomial version: Stieltjes for every x ≥ 1, but not for 0 < x < 1. He restated it in 2025, and B. Deb's database of positivity problems lists the case x ≥ 1 as problem P1.1. Both directions have since been proved: the case x ≥ 1 by Knowledge Sense (preprint dated 13 October 2026), and the full classification by B. Li and E. Yeroushalmi (November 2026). This paper gives another proof of both directions and decides the weaker Hamburger property for every x ≥ 0, which neither work classifies. The exceptional point x = 1/4 is the new feature. The whole answer can be read off one random variable. Walk three steps in the plane, of lengths 1, √x and √x, each in an independent uniformly random direction. Let R be the distance from the start and put W = 2R² − 1. Then A_n(x) = E[P_n(W)], where P_n is the n-th Legendre polynomial, and the moment property is decided by one number, the folded length |1 − 2√x| (the distance when both short steps point backwards): A_n(x) is a Hamburger moment sequence exactly when that length is 0 or at least 1, and a Stieltjes one exactly when it is at least 1. At x = 1/4 the length is 0, which is why that point behaves differently from its neighbours. STATUS Theorem A (no representing measure on the real line for 0 < x < 1, x ≠ 1/4): PROVED, not yet checked by an independent mathematician. Its Stieltjes half was first proved by Li and Yeroushalmi, whose negative Hankel minors also exclude Hamburger representations on parts of (0, 1); the band lemma gives the exclusion for every x in (0, 1) other than 1/4. Theorem B (a representing measure on [0, ∞) for x ≥ 1): PROVED, not yet checked by an independent mathematician. First proved by Knowledge Sense; this is another proof. Theorem D (x = 1/4: a measure on the real line, none on [0, ∞)): PROVED, not yet checked by an independent mathematician. Theorems C and C′ (the Franel numbers and their x-analogues): PROVED, not yet checked by an independent mathematician. Remark 3.4 (the recurrence's characteristic polynomial factors through the critical values): DERIVED by exact algebra and checked in exact arithmetic; a consistency check, not part of any proof. The computational checks of Section 9 (thirteen rows, each test registered before it was run): REGISTERED PASS, with two exceptions the paper states in full: one registered rule was met at six of seven points (a fault in the script's choice of root, supported by an unregistered rerun), and one test failed on its first run for numerical reasons and passed on rerun. As far as a non-exhaustive search shows, these are new: the Hamburger classification, Theorem D, the walk form of the representation (Proposition 3.2), the elliptic-surface reading of the classification, and the Franel x-analogues of Theorem C′. RESULTS Theorem A. If 0 < x < 1 and x ≠ 1/4, no positive measure μ on the real line has A_n(x) = ∫ s^n dμ(s) for all n. The tool is a band lemma (Lemma 4.1): if a positive measure on the real line has a jump or a logarithmic singularity strictly inside (−1, 1), its Legendre moments ∫ P_n dρ are not a Hamburger moment sequence. This excludes every positive measure on the real line, not only measures on [0, ∞). Theorem B. If x ≥ 1, there is such a measure on [0, ∞). On each piece of its support the density is a non-negative integer combination of positive period integrals of a rational elliptic surface, and the integers are orders of vanishing of its discriminant. Theorem D. At x = 1/4 there is a measure on the real line, supported on [−1, 7 + 4√3], but none on [0, ∞): a singular fibre of additive type III places positive mass on (−1, 0). Geometrically, the level curves of W form a rational elliptic surface over the w-line. The sequence is Hamburger exactly when no singular fibre lies over the open band (−1, 1), and Stieltjes exactly when, in addition, the fibre over w = −1 is semistable. Theorem C. The Franel numbers f_k = Σ_j C(k,j)³ are the moments of V = (W − 1)/2 on the x = 1 torus, a positive measure on [−1, 8]. Hence (f_k) is a Hamburger moment sequence. It is not a Stieltjes moment sequence: the shifted Hankel determinants det[f_(i+j+1)] for N = 1, 2, 3 are 2, 12, −504. Theorem C′. For every x > 0 the polynomials m_k(x) = Σ_j C(k,j)² C(2j,k) x^j are the moments of V = (W − 1)/2, so (m_k(x)) is a Hamburger moment sequence and never a Stieltjes one. Their binomial transform is Sun's g_m(x) = Σ_j C(m,j)² C(2j,j) x^j = E[((W + 1)/2)^m], a Stieltjes moment sequence for every x > 0. Remark 3.4. With s = √x, w_max = 1 + 8s + 8s² and w* = 1 − 8s + 8s², the characteristic polynomial of the recurrence equals 8(4x − 1) Q(t), where Q(t) = (t² − 2 w_max t + 1)(t² − 2 w* t + 1). So each characteristic root t satisfies (t + 1/t)/2 = w_max or w*: the recurrence sees exactly the critical values of W. At x = 1/4 the factor 4x − 1 vanishes, and the paper treats that point separately. APPENDIX: THE MATHEMATICS Each step below is in Appendix A of the paper, worked with pencil and followed by a nearby case where the method correctly says no. Why one negative determinant settles the question. If a_n = ∫ s^n dμ(s) with μ ≥ 0 on [0, ∞), then for any real c_0, …, c_(N−1), Σ c_i c_j a_(i+j+1) = ∫ s (Σ c_i s^i)² dμ(s) ≥ 0, so every determinant det[a_(i+j+1)] is ≥ 0. Without the extra factor s, the same argument gives det[a_(i+j)] ≥ 0, the Hamburger condition. A thousand positive determinants prove nothing; one negative one is a complete proof of failure. The Franel numbers by hand. f_0, …, f_5 = 1, 2, 10, 56, 346, 2252. Expand the shifted 3 × 3 determinant det[[2, 10, 56], [10, 56, 346], [56, 346, 2252]] along its first row: 2 × (56·2252 − 346²) = 2 × 6396 = 12792; −10 × (10·2252 − 346·56) = −10 × 3144 = −31440; 56 × (10·346 − 56²) = 56 × 324 = 18144. The total is 12792 − 31440 + 18144 = −504 < 0, so the Franel numbers are not a Stieltjes moment sequence. The case that must pass: the unshifted determinant det[[1, 2, 10], [2, 10, 56], [10, 56, 346]] = 324 − 264 + 120 = 180 > 0, as the Hamburger property requires. The critical values decide each x. With s = √x, compute w* = 1 − 8s + 8s² and w_max = 1 + 8s + 8s². x = 1/16 (s = 1/4): w* = 1 − 2 + 1/2 = −1/2, strictly inside the band (−1, 1), so (A_n) is not Hamburger (Theorem A). x = 1/4 (s = 1/2): w* = 1 − 4 + 2 = −1, the edge of the band; w_max = 7, and the growth rate w_max + √(w_max² − 1) = 7 + √48 = 7 + 4√3 is the right end of the support in Theorem D. x = 1 (s = 1): w* = 1 and w_max = 17, with growth rate 17 + √288 = 17 + 12√2 = (1 + √2)⁴ ≈ 33.97, the known growth rate of the Apéry numbers. x = 4 (s = 2): w* = 1 − 16 + 32 = 17 ≥ 1, so (A_n) is Stieltjes (Theorem B). Why the growth rate is w + √(w² − 1): for w = cosh θ that expression is e^θ, and |P_n(w)| ≤ e^(nθ). An honest limit. At x = 1 the first Hankel determinants are 73 − 25 = 48 and 5·1445 − 73² = 7225 − 5329 = 1896, both positive, as Theorem B requires. For 0 < x < 1, size 2 shows nothing: A_2 − A_1² = 28x + 20x² > 0 for every x > 0. A machine search finds the first negative determinant at size 4 for x = 1/16 and at size 5 for x = 1/10. Those sizes are computer evidence, not part of the proof; the proof is the analytic argument of Sections 4 and 5. That is the paper's main lesson: a negative finite determinant disproves the moment property at one point, but only an analytic argument settles a whole interval. Every number in Appendix A is checked in exact arithmetic by the two scripts in hand_blocks/ of the companion record (37 and 54 checks, all passing). Appendix B records how the results were found, including the two tests that did not pass cleanly the first time and a credit that was corrected. FILES The preprint (PDF, 17 pages). The code record, doi:10.5281/zenodo.24331682, holds apery_deposit_v2.zip (SHA-256 beginning 8076b9c5101ff722): every script, registration and log, including those that failed, were voided or were superseded, and the exact recurrence certificate. 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 [2] A. K. Common, Properties of Legendre expansions related to series of Stieltjes and applications to π–π scattering, Nuovo Cimento A 63 (1969) 863–891, doi:10.1007/BF02760744. [3] B. Deb, Positivity problems in enumerative and algebraic combinatorics, bishaldeb.com/positivity, problem P1.1. [5] G. A. Edgar, The Apéry numbers as a Stieltjes moment sequence, arXiv:2005.10733. [6] Knowledge Sense, Apéry polynomials as Stieltjes moment sequences, preprint dated 13 October 2026, github.com/KnowledgeSense/math. [7] B. Li and E. Yeroushalmi, Stieltjes moment cla

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-21
DOI
https://doi.org/10.5281/zenodo.24409973
Primary Topic
Mathematical functions and polynomials
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Exact rational certificates in Lean 4 and a Meixner–Pollaczek moment representation

Joshua W. Davis
Zenodo (CERN European Organization for Nuclear Research)
Mathematical functions and polynomials
preprint

Exact rational certificates in Lean 4 and a Meixner–Pollaczek moment representation

Joshua W. Davis
preprint en

Abstract

This record contains the preprint "The Apéry polynomials as Hamburger and Stieltjes moment sequences". Preprint, November 2026. Not yet independently refereed. The code, registrations, logs and the exact recurrence certificate are in the companion record doi:10.5281/zenodo.24331682. STATEMENT For real x ≥ 0 let A_n(x) = Σ_(k=0..n) C(n,k)² C(n+k,k)² x^k be the Apéry polynomials. A sequence (a_n) is a Hamburger moment sequence if a_n = ∫ s^n dμ(s) for a positive measure μ on the real line, and a Stieltjes moment sequence if μ can be taken on [0, ∞). Then (A_n(x))_(n≥0) is a Hamburger moment sequence exactly when x ∈ {0, 1/4} ∪ [1, ∞), and a Stieltjes moment sequence exactly when x = 0 or x ≥ 1. At x = 1/4 the representing measure lives on [−1, 7 + 4√3] and puts mass on (−1, 0), so there is none on [0, ∞). MOTIVATION At x = 1 these are the Apéry numbers 1, 5, 73, 1445, …, which appear in Apéry's proof that ζ(3) is irrational. A. D. Sokal conjectured that they form a Stieltjes moment sequence, and G. A. Edgar proved it. In a 2018 lecture Sokal proposed the polynomial version: Stieltjes for every x ≥ 1, but not for 0 < x < 1. He restated it in 2025, and B. Deb's database of positivity problems lists the case x ≥ 1 as problem P1.1. Both directions have since been proved: the case x ≥ 1 by Knowledge Sense (preprint dated 13 October 2026), and the full classification by B. Li and E. Yeroushalmi (November 2026). This paper gives another proof of both directions and decides the weaker Hamburger property for every x ≥ 0, which neither work classifies. The exceptional point x = 1/4 is the new feature. The whole answer can be read off one random variable. Walk three steps in the plane, of lengths 1, √x and √x, each in an independent uniformly random direction. Let R be the distance from the start and put W = 2R² − 1. Then A_n(x) = E[P_n(W)], where P_n is the n-th Legendre polynomial, and the moment property is decided by one number, the folded length |1 − 2√x| (the distance when both short steps point backwards): A_n(x) is a Hamburger moment sequence exactly when that length is 0 or at least 1, and a Stieltjes one exactly when it is at least 1. At x = 1/4 the length is 0, which is why that point behaves differently from its neighbours. STATUS Theorem A (no representing measure on the real line for 0 < x < 1, x ≠ 1/4): PROVED, not yet checked by an independent mathematician. Its Stieltjes half was first proved by Li and Yeroushalmi, whose negative Hankel minors also exclude Hamburger representations on parts of (0, 1); the band lemma gives the exclusion for every x in (0, 1) other than 1/4. Theorem B (a representing measure on [0, ∞) for x ≥ 1): PROVED, not yet checked by an independent mathematician. First proved by Knowledge Sense; this is another proof. Theorem D (x = 1/4: a measure on the real line, none on [0, ∞)): PROVED, not yet checked by an independent mathematician. Theorems C and C′ (the Franel numbers and their x-analogues): PROVED, not yet checked by an independent mathematician. Remark 3.4 (the recurrence's characteristic polynomial factors through the critical values): DERIVED by exact algebra and checked in exact arithmetic; a consistency check, not part of any proof. The computational checks of Section 9 (thirteen rows, each test registered before it was run): REGISTERED PASS, with two exceptions the paper states in full: one registered rule was met at six of seven points (a fault in the script's choice of root, supported by an unregistered rerun), and one test failed on its first run for numerical reasons and passed on rerun. As far as a non-exhaustive search shows, these are new: the Hamburger classification, Theorem D, the walk form of the representation (Proposition 3.2), the elliptic-surface reading of the classification, and the Franel x-analogues of Theorem C′. RESULTS Theorem A. If 0 < x < 1 and x ≠ 1/4, no positive measure μ on the real line has A_n(x) = ∫ s^n dμ(s) for all n. The tool is a band lemma (Lemma 4.1): if a positive measure on the real line has a jump or a logarithmic singularity strictly inside (−1, 1), its Legendre moments ∫ P_n dρ are not a Hamburger moment sequence. This excludes every positive measure on the real line, not only measures on [0, ∞). Theorem B. If x ≥ 1, there is such a measure on [0, ∞). On each piece of its support the density is a non-negative integer combination of positive period integrals of a rational elliptic surface, and the integers are orders of vanishing of its discriminant. Theorem D. At x = 1/4 there is a measure on the real line, supported on [−1, 7 + 4√3], but none on [0, ∞): a singular fibre of additive type III places positive mass on (−1, 0). Geometrically, the level curves of W form a rational elliptic surface over the w-line. The sequence is Hamburger exactly when no singular fibre lies over the open band (−1, 1), and Stieltjes exactly when, in addition, the fibre over w = −1 is semistable. Theorem C. The Franel numbers f_k = Σ_j C(k,j)³ are the moments of V = (W − 1)/2 on the x = 1 torus, a positive measure on [−1, 8]. Hence (f_k) is a Hamburger moment sequence. It is not a Stieltjes moment sequence: the shifted Hankel determinants det[f_(i+j+1)] for N = 1, 2, 3 are 2, 12, −504. Theorem C′. For every x > 0 the polynomials m_k(x) = Σ_j C(k,j)² C(2j,k) x^j are the moments of V = (W − 1)/2, so (m_k(x)) is a Hamburger moment sequence and never a Stieltjes one. Their binomial transform is Sun's g_m(x) = Σ_j C(m,j)² C(2j,j) x^j = E[((W + 1)/2)^m], a Stieltjes moment sequence for every x > 0. Remark 3.4. With s = √x, w_max = 1 + 8s + 8s² and w* = 1 − 8s + 8s², the characteristic polynomial of the recurrence equals 8(4x − 1) Q(t), where Q(t) = (t² − 2 w_max t + 1)(t² − 2 w* t + 1). So each characteristic root t satisfies (t + 1/t)/2 = w_max or w*: the recurrence sees exactly the critical values of W. At x = 1/4 the factor 4x − 1 vanishes, and the paper treats that point separately. APPENDIX: THE MATHEMATICS Each step below is in Appendix A of the paper, worked with pencil and followed by a nearby case where the method correctly says no. Why one negative determinant settles the question. If a_n = ∫ s^n dμ(s) with μ ≥ 0 on [0, ∞), then for any real c_0, …, c_(N−1), Σ c_i c_j a_(i+j+1) = ∫ s (Σ c_i s^i)² dμ(s) ≥ 0, so every determinant det[a_(i+j+1)] is ≥ 0. Without the extra factor s, the same argument gives det[a_(i+j)] ≥ 0, the Hamburger condition. A thousand positive determinants prove nothing; one negative one is a complete proof of failure. The Franel numbers by hand. f_0, …, f_5 = 1, 2, 10, 56, 346, 2252. Expand the shifted 3 × 3 determinant det[[2, 10, 56], [10, 56, 346], [56, 346, 2252]] along its first row: 2 × (56·2252 − 346²) = 2 × 6396 = 12792; −10 × (10·2252 − 346·56) = −10 × 3144 = −31440; 56 × (10·346 − 56²) = 56 × 324 = 18144. The total is 12792 − 31440 + 18144 = −504 < 0, so the Franel numbers are not a Stieltjes moment sequence. The case that must pass: the unshifted determinant det[[1, 2, 10], [2, 10, 56], [10, 56, 346]] = 324 − 264 + 120 = 180 > 0, as the Hamburger property requires. The critical values decide each x. With s = √x, compute w* = 1 − 8s + 8s² and w_max = 1 + 8s + 8s². x = 1/16 (s = 1/4): w* = 1 − 2 + 1/2 = −1/2, strictly inside the band (−1, 1), so (A_n) is not Hamburger (Theorem A). x = 1/4 (s = 1/2): w* = 1 − 4 + 2 = −1, the edge of the band; w_max = 7, and the growth rate w_max + √(w_max² − 1) = 7 + √48 = 7 + 4√3 is the right end of the support in Theorem D. x = 1 (s = 1): w* = 1 and w_max = 17, with growth rate 17 + √288 = 17 + 12√2 = (1 + √2)⁴ ≈ 33.97, the known growth rate of the Apéry numbers. x = 4 (s = 2): w* = 1 − 16 + 32 = 17 ≥ 1, so (A_n) is Stieltjes (Theorem B). Why the growth rate is w + √(w² − 1): for w = cosh θ that expression is e^θ, and |P_n(w)| ≤ e^(nθ). An honest limit. At x = 1 the first Hankel determinants are 73 − 25 = 48 and 5·1445 − 73² = 7225 − 5329 = 1896, both positive, as Theorem B requires. For 0 < x < 1, size 2 shows nothing: A_2 − A_1² = 28x + 20x² > 0 for every x > 0. A machine search finds the first negative determinant at size 4 for x = 1/16 and at size 5 for x = 1/10. Those sizes are computer evidence, not part of the proof; the proof is the analytic argument of Sections 4 and 5. That is the paper's main lesson: a negative finite determinant disproves the moment property at one point, but only an analytic argument settles a whole interval. Every number in Appendix A is checked in exact arithmetic by the two scripts in hand_blocks/ of the companion record (37 and 54 checks, all passing). Appendix B records how the results were found, including the two tests that did not pass cleanly the first time and a credit that was corrected. FILES The preprint (PDF, 17 pages). The code record, doi:10.5281/zenodo.24331682, holds apery_deposit_v2.zip (SHA-256 beginning 8076b9c5101ff722): every script, registration and log, including those that failed, were voided or were superseded, and the exact recurrence certificate. 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 [2] A. K. Common, Properties of Legendre expansions related to series of Stieltjes and applications to π–π scattering, Nuovo Cimento A 63 (1969) 863–891, doi:10.1007/BF02760744. [3] B. Deb, Positivity problems in enumerative and algebraic combinatorics, bishaldeb.com/positivity, problem P1.1. [5] G. A. Edgar, The Apéry numbers as a Stieltjes moment sequence, arXiv:2005.10733. [6] Knowledge Sense, Apéry polynomials as Stieltjes moment sequences, preprint dated 13 October 2026, github.com/KnowledgeSense/math. [7] B. Li and E. Yeroushalmi, Stieltjes moment cla

Zenodo (CERN European Organization for Nuclear Research)
Mathematical functions and polynomials
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