Hankel failure sizes for the zeta(2) Apéry polynomials

This record contains the preprint "Hankel failure sizes for the zeta(2) Apéry polynomials". Preprint, 21 November 2026, version 2. Not yet independently refereed. The scripts, registrations, logs and records are in the companion code record doi:10.5281/zenodo.[CODE v8 DOI] (its version 8; version 7 is doi:10.5281/zenodo.24421335). VERSION 2 Adds Appendix W12 (not yet read by the author), three rows of the status table and a pointer in Section 2a. W12 compares the wall with the ζ(3) family of the companion paper [P1]: there the critical value w* = 1 − 8√x + 8x gives w* − 1 = −8s(1 − s) and w* + 1 = 2(1 − 2s)² (s = √x), so the wall x = 1 is a crossing (a first-power factor) and the isolated Hamburger point x = 1/4 a touch (a squared factor); here both happen at x = 27/32, which is the cube in x²(32x − 27)³. Carried over, [P1]'s rule suggests a conjecture: B_n(x) is Hamburger for every x ≥ 27/32 and Stieltjes for no x > 0. Nothing in version 1 changes. STATEMENT Let B_n(x) = Σ_k C(n,k)² C(n+k,k) x^k; at x = 1 these are Apéry's numbers for ζ(2), 1, 3, 19, 147, 1251, …. For x = 1, 2, …, 8 the sequence (B_n(x)) is not a Stieltjes moment sequence: the shifted Hankel determinant first becomes negative at sizes 11, 23, 33, 42, 52, 61, 71 and 80, found by exact integer elimination. At x = 1/4, 1/2, 3/4, 13/16, 53/64, 107/128 and 215/256, all below x = 27/32, it is not even a Hamburger moment sequence: the first non-positive ordinary Hankel minor is strictly negative, at sizes 5, 8, 22, 48, 80, 131 and 218. Conditional on two standard results of singularity theory, it fails to be Hamburger for every x in (0, 27/32). The Apéry numbers for ζ(2) cannot be written as E[P_n(T)] for a trigonometric kernel T with integer, Gaussian-integer or algebraic-integer coefficients, because such a kernel would need E[T³] = 303/5. MOTIVATION The companion paper classifies the Apéry polynomials for ζ(3) as moment sequences, by writing them as a Legendre layer over a walk with integer moments. This note asks the same questions of the ζ(2) family and finds a different answer. A sequence is a Hamburger moment sequence exactly when every Hankel matrix [a_(i+j)] is positive semidefinite, and a Stieltjes one when [a_(i+j+1)] is too. So one negative minor is a complete proof of failure, and the size at which it first appears measures how far the sequence gets. Below 27/32 the explanation is geometric: B_n(x) is the constant term of the n-th power of a two-variable symbol W_x, two of its three critical values form a non-real pair, and a moment sequence cannot have a generating function with a non-real singularity. STATUS First shifted failures at sizes 11, 23, 33, 42, 52, 61, 71, 80 for x = 1 to 8: REPLICATED (two independent exact runs). Hence not Stieltjes at x = 1 to 8: PROVED (exact negative minors). x = 16, first failure at size 152: REPLICATED (two independent exact runs). The endpoint budget law at x = 16 (prediction 152.0, window 144.4 to 159.6): REGISTERED PASS. Not Hamburger at x = 1/4, 1/2, 3/4, 13/16, 53/64, 107/128: PROVED (exact strictly negative minors, two independent programs). Not Hamburger at x = 215/256 (size 218): PROVED by one registered exact run. Not Hamburger for every x in (0, 27/32): PROVED, conditional on two standard results not consulted directly (Picard–Lefschetz for the complement; Milnor's curve selection lemma). |κ| = 1 at x = 1/2: REGISTERED PASS (certified, integer ball arithmetic). |κ| = 1 at x = 1/10, 3/10, 3/5, 83/100: REGISTERED PASS (double precision). The wall at 27/32 (failures below, none above to size 256): REGISTERED PASS. The endpoint budget law N* ≈ 1.02·√(r_1/|r_2|) at x = 3/2, 12, 16, 24: REGISTERED PASS (a heuristic with a fitted constant). No failure at x = 55/64 through size 100: MEASURED. No integer-type trigonometric kernel: PROVED. P_x is the characteristic polynomial of the recurrence for B_n(x): MEASURED (an order-3 recurrence fitted to exact values reproduces it at x = 16 and 27/32). B_n(x) is the constant term of W_x^n: PROVED. Its critical values are the roots of P_x for every x: PROVED (Appendix W9). Why the wall is at 27/32 and why the discriminant carries a cube: PROVED (Appendix W11). The same pattern in the ζ(3) family: its wall x = 1 is a crossing (first power) and its isolated Hamburger point x = 1/4 a touch (a square); here both fall at 27/32: PROVED (Appendix W12). For every x ≥ 27/32, P_x has a negative real root: PROVED (Appendix W12). B_n(x) is Hamburger for every x ≥ 27/32 and Stieltjes for no x > 0: OPEN (a conjecture suggested by W12). Is B_n(x) Hamburger for every x ≥ 27/32? A law for N*(x)? OPEN. Not yet checked by an independent mathematician. RESULTS Section 2. The exact failure sizes for x = 1 to 8 and x = 16, and why failure must move outward as x grows: x^(−n) B_n(x) tends to C(2n, n), a Stieltjes sequence. Section 2a. The characteristic polynomial of the recurrence, P_x(z) = z³ − (3 + 8x)z² + (16x² − 20x + 3)z + (x − 1), has discriminant x²(32x − 27)³, so below 27/32 two roots are a non-real pair. B_n(x) is the constant term of W_x^n with W_x(u, v) = (1 + xu)(1 + v)(1 + u + v)/(uv), proved by Vandermonde's identity, and the critical values of W_x are the roots of P_x. The registered wall test and endpoint budget law, the route to every x below the wall through an integer invariant κ, and a worked case at x = 1/2. Section 3. No integer kernel: the Legendre layer forces E[T] = 3, E[T²] = 13 and E[T³] = 303/5, and 303/5 is not an algebraic integer. Section 4. Why the ζ(3) route does not transfer: inverting the Legendre layer gives h_3 = 1 + (57/10)x + 6x² + x³ or −(3/10)x + 3x² + x³, a non-integer or a negative coefficient. A lead: b_n = [t^n] (1/(1 − t)) P_n((1 + t)/(1 − t)), with a one-line proof. APPENDIX: THE MATHEMATICS Appendix W works every step with pencil, and each block has a case where the method must fail. The first polynomials. B_1 = 1 + 2x, B_2 = 1 + 12x + 6x², B_3 = 1 + 36x + 90x² + 20x³, and b_4 = 1251, b_5 = 11253. The first shifted minors at x = 1, by fraction-free elimination: 3, then 3·147 − 19² = 80, then (80·12150 − 960²)/3 = 16800, all positive below N*(1) = 11. Dividing by 19 instead of the pivot 3 leaves a remainder, which is how a wrong pivot shows. The idea in one variable. W(u) = u + 1/u has constant terms 1, 0, 2, 0, 6, …, generating function 1/√(1 − 4z²), critical values ±2 and singularities at z = ±1/2, their reciprocals; the Hankel minors 1, 2, 4 are positive. With equal weights on e^(iπ/3) and its conjugate, a_n = 2, 1, −1, … and the 2 × 2 minor is −3: a non-real singularity rules out a moment sequence. The critical values for every x. Setting the logarithmic derivatives of W_x to zero gives v² = 1 + u and v = xu² − 1, hence x²u³ − 2xu − 1 = 0. With s = xu this is s³ − 2xs − x = 0, the critical value is s²(1 + s)²/(s² − x), and the identity (1 + s)(1 + 2s)(s² − x) − s²(1 + s)² = (1 + s)(s³ − 2xs − x) turns it into (1 + s)(1 + 2s). Newton's identities then give the sum 8x + 3, the pairwise sum 16x² − 20x + 3 and the product 1 − x: exactly P_x. At x = 1, P_1 = z(z² − 11z − 1), whose discriminant 125 equals x²(32x − 27)³ = 5³. Why the wall, and why the cube. The cubic s³ − 2xs − x has discriminant x²(32x − 27), so two critical points meet exactly at x = 27/32, at s = −3/4, with the third at 3/2; the map gives −1/8 (twice) and 10, the roots of P_x there. Since w_i − w_j = (s_i − s_j)(3 − 2s_k) and the product of the 3 − 2s_k is 8Q(3/2) = 27 − 32x, the discriminant of P_x is x²(32x − 27)·(27 − 32x)² = x²(32x − 27)³. The extra square appears because the meeting point s = −3/4 is the turning point of w = 2s² + 3s + 1; the map (1 + s)(1 + 3s), which turns at −2/3, gives only the first power. The same pattern in the ζ(3) family. For [P1]'s critical value w* = 1 − 8s + 8s², s = √x: w* − 1 = −8s(1 − s) and w* + 1 = 2(1 − 2s)². The first factor changes sign at x = 1, where the pair crosses onto the real axis; the second is a square, so at x = 1/4 the pair only touches −1, at the turning point of s → w*. The product (z² − 2w_max z + 1)(z² − 2w* z + 1) = (z² − 2(1 + 8x)z + 1)² − 256x z² has discriminant 2³²x³(x − 1)(4x − 1)². For ζ(2) both meetings fall at 27/32. For every x ≥ 27/32, P_x has a negative root: for x > 1, P_x(0) = x − 1 > 0; at x = 1, (11 − √125)/2 < 0; for 27/32 ≤ x < 1 the roots are real and their pairwise sum 16x² − 20x + 3 is negative (−159/64 at the wall). The case that must fail: 2^n has the recurrence roots 2 and −1/2 yet is a Stieltjes moment sequence, so a negative root alone proves nothing. The certified κ. |V(z0)| in [0.3671072825, 0.3679146911] times |√det H| = 2.7212827 gives |κ| in [0.99900, 1.00120], so |κ| = 1; a wider enclosure such as [0.816, 2.177] would decide nothing. Every number in the appendix is checked in exact arithmetic (hand_values_check_2.py; mustfail_check.py, 29 checks; zeta2_hand_blocks_check_3.py, 27 checks, and zeta2_hand_blocks_check_4.py, 12 checks, and zeta2_hand_blocks_check_5.py, 24 checks, with 9 of 9, 8 of 8 and 8 of 8 deliberate errors caught). A further appendix records how the results were found, including a prediction that missed (about 161 at x = 16 against the exact 152), reported as a miss. FILES The preprint (PDF, 24 pages). The code record, doi:10.5281/zenodo.[CODE v8 DOI], holds zeta2_code_record_v8.zip (SHA-256 beginning 90b5d758c6068f50; SHA256SUMS.txt lists 86 files) under the MIT licence: the exact programs and their logs, the registrations written before their tests ran, the 7 November wall and budget-law records, the certified κ run with every attempt, and the checks of every number in the appendix. 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,

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-21
DOI
https://doi.org/10.5281/zenodo.24421504
Citations
1
Primary Topic
Advanced Mathematical Identities
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Hankel failure sizes for the zeta(2) Apéry polynomials

Joshua W. Davis
1 citations
Zenodo (CERN European Organization for Nuclear Research)
Advanced Mathematical Identities
preprint

Hankel failure sizes for the zeta(2) Apéry polynomials

Joshua W. Davis
preprint en
1 citations

Abstract

This record contains the preprint "Hankel failure sizes for the zeta(2) Apéry polynomials". Preprint, 21 November 2026, version 2. Not yet independently refereed. The scripts, registrations, logs and records are in the companion code record doi:10.5281/zenodo.[CODE v8 DOI] (its version 8; version 7 is doi:10.5281/zenodo.24421335). VERSION 2 Adds Appendix W12 (not yet read by the author), three rows of the status table and a pointer in Section 2a. W12 compares the wall with the ζ(3) family of the companion paper [P1]: there the critical value w* = 1 − 8√x + 8x gives w* − 1 = −8s(1 − s) and w* + 1 = 2(1 − 2s)² (s = √x), so the wall x = 1 is a crossing (a first-power factor) and the isolated Hamburger point x = 1/4 a touch (a squared factor); here both happen at x = 27/32, which is the cube in x²(32x − 27)³. Carried over, [P1]'s rule suggests a conjecture: B_n(x) is Hamburger for every x ≥ 27/32 and Stieltjes for no x > 0. Nothing in version 1 changes. STATEMENT Let B_n(x) = Σ_k C(n,k)² C(n+k,k) x^k; at x = 1 these are Apéry's numbers for ζ(2), 1, 3, 19, 147, 1251, …. For x = 1, 2, …, 8 the sequence (B_n(x)) is not a Stieltjes moment sequence: the shifted Hankel determinant first becomes negative at sizes 11, 23, 33, 42, 52, 61, 71 and 80, found by exact integer elimination. At x = 1/4, 1/2, 3/4, 13/16, 53/64, 107/128 and 215/256, all below x = 27/32, it is not even a Hamburger moment sequence: the first non-positive ordinary Hankel minor is strictly negative, at sizes 5, 8, 22, 48, 80, 131 and 218. Conditional on two standard results of singularity theory, it fails to be Hamburger for every x in (0, 27/32). The Apéry numbers for ζ(2) cannot be written as E[P_n(T)] for a trigonometric kernel T with integer, Gaussian-integer or algebraic-integer coefficients, because such a kernel would need E[T³] = 303/5. MOTIVATION The companion paper classifies the Apéry polynomials for ζ(3) as moment sequences, by writing them as a Legendre layer over a walk with integer moments. This note asks the same questions of the ζ(2) family and finds a different answer. A sequence is a Hamburger moment sequence exactly when every Hankel matrix [a_(i+j)] is positive semidefinite, and a Stieltjes one when [a_(i+j+1)] is too. So one negative minor is a complete proof of failure, and the size at which it first appears measures how far the sequence gets. Below 27/32 the explanation is geometric: B_n(x) is the constant term of the n-th power of a two-variable symbol W_x, two of its three critical values form a non-real pair, and a moment sequence cannot have a generating function with a non-real singularity. STATUS First shifted failures at sizes 11, 23, 33, 42, 52, 61, 71, 80 for x = 1 to 8: REPLICATED (two independent exact runs). Hence not Stieltjes at x = 1 to 8: PROVED (exact negative minors). x = 16, first failure at size 152: REPLICATED (two independent exact runs). The endpoint budget law at x = 16 (prediction 152.0, window 144.4 to 159.6): REGISTERED PASS. Not Hamburger at x = 1/4, 1/2, 3/4, 13/16, 53/64, 107/128: PROVED (exact strictly negative minors, two independent programs). Not Hamburger at x = 215/256 (size 218): PROVED by one registered exact run. Not Hamburger for every x in (0, 27/32): PROVED, conditional on two standard results not consulted directly (Picard–Lefschetz for the complement; Milnor's curve selection lemma). |κ| = 1 at x = 1/2: REGISTERED PASS (certified, integer ball arithmetic). |κ| = 1 at x = 1/10, 3/10, 3/5, 83/100: REGISTERED PASS (double precision). The wall at 27/32 (failures below, none above to size 256): REGISTERED PASS. The endpoint budget law N* ≈ 1.02·√(r_1/|r_2|) at x = 3/2, 12, 16, 24: REGISTERED PASS (a heuristic with a fitted constant). No failure at x = 55/64 through size 100: MEASURED. No integer-type trigonometric kernel: PROVED. P_x is the characteristic polynomial of the recurrence for B_n(x): MEASURED (an order-3 recurrence fitted to exact values reproduces it at x = 16 and 27/32). B_n(x) is the constant term of W_x^n: PROVED. Its critical values are the roots of P_x for every x: PROVED (Appendix W9). Why the wall is at 27/32 and why the discriminant carries a cube: PROVED (Appendix W11). The same pattern in the ζ(3) family: its wall x = 1 is a crossing (first power) and its isolated Hamburger point x = 1/4 a touch (a square); here both fall at 27/32: PROVED (Appendix W12). For every x ≥ 27/32, P_x has a negative real root: PROVED (Appendix W12). B_n(x) is Hamburger for every x ≥ 27/32 and Stieltjes for no x > 0: OPEN (a conjecture suggested by W12). Is B_n(x) Hamburger for every x ≥ 27/32? A law for N*(x)? OPEN. Not yet checked by an independent mathematician. RESULTS Section 2. The exact failure sizes for x = 1 to 8 and x = 16, and why failure must move outward as x grows: x^(−n) B_n(x) tends to C(2n, n), a Stieltjes sequence. Section 2a. The characteristic polynomial of the recurrence, P_x(z) = z³ − (3 + 8x)z² + (16x² − 20x + 3)z + (x − 1), has discriminant x²(32x − 27)³, so below 27/32 two roots are a non-real pair. B_n(x) is the constant term of W_x^n with W_x(u, v) = (1 + xu)(1 + v)(1 + u + v)/(uv), proved by Vandermonde's identity, and the critical values of W_x are the roots of P_x. The registered wall test and endpoint budget law, the route to every x below the wall through an integer invariant κ, and a worked case at x = 1/2. Section 3. No integer kernel: the Legendre layer forces E[T] = 3, E[T²] = 13 and E[T³] = 303/5, and 303/5 is not an algebraic integer. Section 4. Why the ζ(3) route does not transfer: inverting the Legendre layer gives h_3 = 1 + (57/10)x + 6x² + x³ or −(3/10)x + 3x² + x³, a non-integer or a negative coefficient. A lead: b_n = [t^n] (1/(1 − t)) P_n((1 + t)/(1 − t)), with a one-line proof. APPENDIX: THE MATHEMATICS Appendix W works every step with pencil, and each block has a case where the method must fail. The first polynomials. B_1 = 1 + 2x, B_2 = 1 + 12x + 6x², B_3 = 1 + 36x + 90x² + 20x³, and b_4 = 1251, b_5 = 11253. The first shifted minors at x = 1, by fraction-free elimination: 3, then 3·147 − 19² = 80, then (80·12150 − 960²)/3 = 16800, all positive below N*(1) = 11. Dividing by 19 instead of the pivot 3 leaves a remainder, which is how a wrong pivot shows. The idea in one variable. W(u) = u + 1/u has constant terms 1, 0, 2, 0, 6, …, generating function 1/√(1 − 4z²), critical values ±2 and singularities at z = ±1/2, their reciprocals; the Hankel minors 1, 2, 4 are positive. With equal weights on e^(iπ/3) and its conjugate, a_n = 2, 1, −1, … and the 2 × 2 minor is −3: a non-real singularity rules out a moment sequence. The critical values for every x. Setting the logarithmic derivatives of W_x to zero gives v² = 1 + u and v = xu² − 1, hence x²u³ − 2xu − 1 = 0. With s = xu this is s³ − 2xs − x = 0, the critical value is s²(1 + s)²/(s² − x), and the identity (1 + s)(1 + 2s)(s² − x) − s²(1 + s)² = (1 + s)(s³ − 2xs − x) turns it into (1 + s)(1 + 2s). Newton's identities then give the sum 8x + 3, the pairwise sum 16x² − 20x + 3 and the product 1 − x: exactly P_x. At x = 1, P_1 = z(z² − 11z − 1), whose discriminant 125 equals x²(32x − 27)³ = 5³. Why the wall, and why the cube. The cubic s³ − 2xs − x has discriminant x²(32x − 27), so two critical points meet exactly at x = 27/32, at s = −3/4, with the third at 3/2; the map gives −1/8 (twice) and 10, the roots of P_x there. Since w_i − w_j = (s_i − s_j)(3 − 2s_k) and the product of the 3 − 2s_k is 8Q(3/2) = 27 − 32x, the discriminant of P_x is x²(32x − 27)·(27 − 32x)² = x²(32x − 27)³. The extra square appears because the meeting point s = −3/4 is the turning point of w = 2s² + 3s + 1; the map (1 + s)(1 + 3s), which turns at −2/3, gives only the first power. The same pattern in the ζ(3) family. For [P1]'s critical value w* = 1 − 8s + 8s², s = √x: w* − 1 = −8s(1 − s) and w* + 1 = 2(1 − 2s)². The first factor changes sign at x = 1, where the pair crosses onto the real axis; the second is a square, so at x = 1/4 the pair only touches −1, at the turning point of s → w*. The product (z² − 2w_max z + 1)(z² − 2w* z + 1) = (z² − 2(1 + 8x)z + 1)² − 256x z² has discriminant 2³²x³(x − 1)(4x − 1)². For ζ(2) both meetings fall at 27/32. For every x ≥ 27/32, P_x has a negative root: for x > 1, P_x(0) = x − 1 > 0; at x = 1, (11 − √125)/2 < 0; for 27/32 ≤ x < 1 the roots are real and their pairwise sum 16x² − 20x + 3 is negative (−159/64 at the wall). The case that must fail: 2^n has the recurrence roots 2 and −1/2 yet is a Stieltjes moment sequence, so a negative root alone proves nothing. The certified κ. |V(z0)| in [0.3671072825, 0.3679146911] times |√det H| = 2.7212827 gives |κ| in [0.99900, 1.00120], so |κ| = 1; a wider enclosure such as [0.816, 2.177] would decide nothing. Every number in the appendix is checked in exact arithmetic (hand_values_check_2.py; mustfail_check.py, 29 checks; zeta2_hand_blocks_check_3.py, 27 checks, and zeta2_hand_blocks_check_4.py, 12 checks, and zeta2_hand_blocks_check_5.py, 24 checks, with 9 of 9, 8 of 8 and 8 of 8 deliberate errors caught). A further appendix records how the results were found, including a prediction that missed (about 161 at x = 16 against the exact 152), reported as a miss. FILES The preprint (PDF, 24 pages). The code record, doi:10.5281/zenodo.[CODE v8 DOI], holds zeta2_code_record_v8.zip (SHA-256 beginning 90b5d758c6068f50; SHA256SUMS.txt lists 86 files) under the MIT licence: the exact programs and their logs, the registrations written before their tests ran, the 7 November wall and budget-law records, the certified κ run with every attempt, and the checks of every number in the appendix. 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,

Zenodo (CERN European Organization for Nuclear Research)
Advanced Mathematical Identities
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