Continued-fraction certificates for Sokal's Conjecture 2: what is excluded and what is open

This record contains the preprint "Continued-fraction certificates for Sokal's Conjecture 2: what is excluded and what is open". Preprint, 21 November 2026, version 4 (version 1 is doi:10.5281/zenodo.24342237). Not yet independently refereed. The scripts, registrations, certificates and logs are in the companion code record doi:10.5281/zenodo.24429534 (its version 5). VERSION 4 New Appendix D: contiguous minors alone do not decide coefficientwise Hankel positivity. An explicit sequence a_0, ..., a_4 = 2 + y², 1 + y, 1 + y, 2 + y², 5 + 4y + 9y² + y³ has every contiguous minor with nonnegative coefficients (1 + y³, 1 + y³, 1 + 9y + 9y² + 10y³, and the 3 × 3 determinant 9y + 9y³ + 9y⁵), while the minor with one skipped column, a_0a_3 − a_1a_2 = 3 − 2y + 3y² + y⁴, has the negative coefficient −2; the identity a_1(a_0a_3 − a_1a_2) = a_0Δ(1) + a_2Δ(0) shows how the contiguous minors hide it. Appendix D also reports a registered check of all 224 one-gap minors of A_n(1+y) of sizes 2 to 8 built from a_23 to a_30, beyond Sokal's 12 × 12 region: none has a negative coefficient, confirmed by a separately written program. Section 4 now points to Appendix D. Appendix D has been read by the author. The code record is version 5, which adds these files. No earlier result or status label changes. STATEMENT Sokal conjectured that the Apéry polynomials A_n(x) = Σ_k C(n,k)² C(n+k,k)² x^k, written in y = x − 1, form a coefficientwise Hankel-totally positive sequence: every minor of the Hankel matrix [a_(i+j)(y)] is a polynomial in y with nonnegative coefficients. The usual proof of such a statement exhibits a continued fraction, or more generally a production matrix, with nonnegative coefficients. This note shows that most of the usual certificate types cannot exist for this sequence. Every m-branched S-fraction, for every m, is ruled out by the first two moments alone; J-fractions are ruled out at level 3 and classical T-fractions at level 4; nineteen branches of lower-Hessenberg production matrices are ruled out at level 5 by exact Farkas certificates. Three interleaved bidiagonal templates survive as far as they were searched. On the other side, every contiguous Hankel minor up to 25 × 25 and 300 random non-contiguous minors have nonnegative coefficients. The conjecture stays open; the note establishes which certificates cannot settle it. MOTIVATION By the theorems of Pétréolle, Sokal and Zhu, one continued fraction or totally positive production matrix with nonnegative polynomial coefficients would prove the conjecture. So the first question is which kinds of certificate are still possible. Each type is matched to the moments a_1, a_2, … level by level, and a type is excluded at the first level where no nonnegative choice survives. Exclusion is a statement about the certificate, not about the conjecture: a sequence can be Hankel-totally positive with no certificate of a given type. STATUS Two-level lemma (in any m-branched S-fraction the first two moments force α_m = a_1 and α_(m+1) + … + α_(2m) = (a_2 − a_1²)/a_1): PROVED (two lines); novelty not checked. For A_n(1+y), (a_2 − a_1²)/a_1 has the negative coefficient −92/125 at y², so branched S-fractions are excluded at level 2 for every m: PROVED. J-fractions excluded at level 3: PROVED. Classical T-fractions excluded at level 4: PROVED. One lower-Hessenberg production-matrix branch, and 18 further branches sampled under a registered rule, excluded at level 5: PROVED for those branches (Farkas certificates, re-verified by code that shares nothing with the search). Interleaved bidiagonal templates ULL, LUL, LLUL: OPEN (they survive to levels 3, 2 and 3). The (1 + y) normal form: OPEN (survives to level 2; not canonical). Templates ULLL, LULL: NO VERDICT (the search engine failed its planted control). Contiguous Hankel minors up to 25 × 25 and 300 sampled non-contiguous minors have nonnegative coefficients: MEASURED (registered, controls passed); evidence, not proof. The same searches for Sokal's little Domb and Domb conjectures: MEASURED (registered). A sequence with every contiguous minor nonnegative and a negative one-gap minor (Appendix D1): PROVED (by hand, and by two programs). All 224 one-gap minors of sizes 2 to 8 from a_23 to a_30 have nonnegative coefficients (Appendix D2): REGISTERED PASS (replicated by a separately written program); evidence, not proof. Sokal's Conjecture 2 itself: OPEN. Not yet checked by an independent mathematician. RESULTS Section 2. The two-level lemma, with its path-counting proof, and the level-2 obstruction for the Apéry polynomials. The same test obstructs Sokal's other three conjectures of this kind in Deb's database: little Domb (−2/9 at y²), Domb (−1/4 at y²) and Boros–Moll (−3/16 at y³). Section 3. The ladder of exclusions: J-fractions at level 3 (b_1 = 35/2 + (295/24)y − (419/288)y² + …); classical T-fractions at level 4, by reducing each level to a factorisation into nonnegative polynomials; production-matrix branches at level 5, by a Farkas certificate with 55 nonnegative multipliers and constant K = −281468537621471/1600148025, and 18 sampled branches the same way. Section 4. The registered minor searches: all contiguous minors up to 25 × 25 (degree up to 600) and 300 random non-contiguous minors of the 20 × 20 matrix, with controls that catch a non-totally-positive sequence every time. The power of (y + 1) dividing the contiguous minor Δ_k is 1, 2, 3, 5 for k = 2 to 5, and the cofactors have positive coefficients. Appendix D (new in version 4). D1 works the counterexample by hand: the three contiguous 2 × 2 minors, the 3 × 3 determinant by Desnanot–Jacobi, the one-gap minor with its negative coefficient, the identity behind it, and a case that must fail. D2 reports the registered one-gap check, with a Kronecker-substitution example done by hand (10283 read in base 10 as the digits 3, −2, 3, 0, 1 after carrying). Every number in Appendix D is checked in exact arithmetic (paper3_appD_hand_check.py, 19 checks; a mutation test catches 10 of 10 deliberate errors). Section 5. Three open routes: deeper levels for the surviving templates, a lattice-path model through the Lindström–Gessel–Viennot lemma, and induction through the Desnanot–Jacobi identity. APPENDIX: THE MATHEMATICS Appendix A works every step with pencil, and each block ends with a case where the method must fail, or a control that must not. The certificate types, in words. A production matrix P produces a_n = (P^n)_00. For P = UL with β_0, β_1 = 2, 5 and α_1, α_2 = 3, 7: P_00 = 5, P_10 = 15, P_11 = 12 and a_2 = 25 + 15 = 40. Positive entries are not enough: [[1, 1], [2, 1]] has determinant −1, which is why the templates are built from bidiagonal factors. The data. A_1 = 1 + 4x and A_2 = 1 + 36x + 36x², so at x = 1 + y: a_1 = 5 + 4y and a_2 = 73 + 108y + 36y². The level-2 obstruction. a_2 − a_1² = 48 + 68y + 20y² = 4(y + 1)(5y + 12). Divide by a_1 = 5 + 4y as a power series: 5c_0 = 48 gives 48/5; 5c_1 + 4c_0 = 68 gives 148/25; 5c_2 + 4c_1 = 20 gives −92/125. The control (n!)² gives (4 − 1)/1 = 3 > 0. The case that must fail: 1, 2, 3 gives (3 − 4)/2 = −1/2. Why a rational series with nonnegative coefficients is a polynomial. By Pringsheim's theorem such a series with finite radius has a singularity on the positive axis. By long division, (a_2 − a_1²)/a_1 = 5y + 43/4 − (23/4)/(4y + 5), whose only pole, y = −5/4, is off that axis; so it cannot have nonnegative coefficients. The case the hypothesis leaves out: 1/(1 − 2y) = 1 + 2y + 4y² + … has its pole at y = 1/2. Classical T-fractions, worked. With G = 1/(1 − δ_0t − α_1t·F_1), level 2 forces α_1 = c, c(y + 1) or c(5y + 12). At level 3 the first and third fail, and α_1 = c(y + 1) survives only for c in [96/35, 4], because the

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-21
DOI
https://doi.org/10.5281/zenodo.24342236
Primary Topic
Advanced Combinatorial Mathematics
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Continued-fraction certificates for Sokal's Conjecture 2: what is excluded and what is open

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

Continued-fraction certificates for Sokal's Conjecture 2: what is excluded and what is open

Joshua W. Davis
preprint en

Abstract

This record contains the preprint "Continued-fraction certificates for Sokal's Conjecture 2: what is excluded and what is open". Preprint, 21 November 2026, version 4 (version 1 is doi:10.5281/zenodo.24342237). Not yet independently refereed. The scripts, registrations, certificates and logs are in the companion code record doi:10.5281/zenodo.24429534 (its version 5). VERSION 4 New Appendix D: contiguous minors alone do not decide coefficientwise Hankel positivity. An explicit sequence a_0, ..., a_4 = 2 + y², 1 + y, 1 + y, 2 + y², 5 + 4y + 9y² + y³ has every contiguous minor with nonnegative coefficients (1 + y³, 1 + y³, 1 + 9y + 9y² + 10y³, and the 3 × 3 determinant 9y + 9y³ + 9y⁵), while the minor with one skipped column, a_0a_3 − a_1a_2 = 3 − 2y + 3y² + y⁴, has the negative coefficient −2; the identity a_1(a_0a_3 − a_1a_2) = a_0Δ(1) + a_2Δ(0) shows how the contiguous minors hide it. Appendix D also reports a registered check of all 224 one-gap minors of A_n(1+y) of sizes 2 to 8 built from a_23 to a_30, beyond Sokal's 12 × 12 region: none has a negative coefficient, confirmed by a separately written program. Section 4 now points to Appendix D. Appendix D has been read by the author. The code record is version 5, which adds these files. No earlier result or status label changes. STATEMENT Sokal conjectured that the Apéry polynomials A_n(x) = Σ_k C(n,k)² C(n+k,k)² x^k, written in y = x − 1, form a coefficientwise Hankel-totally positive sequence: every minor of the Hankel matrix [a_(i+j)(y)] is a polynomial in y with nonnegative coefficients. The usual proof of such a statement exhibits a continued fraction, or more generally a production matrix, with nonnegative coefficients. This note shows that most of the usual certificate types cannot exist for this sequence. Every m-branched S-fraction, for every m, is ruled out by the first two moments alone; J-fractions are ruled out at level 3 and classical T-fractions at level 4; nineteen branches of lower-Hessenberg production matrices are ruled out at level 5 by exact Farkas certificates. Three interleaved bidiagonal templates survive as far as they were searched. On the other side, every contiguous Hankel minor up to 25 × 25 and 300 random non-contiguous minors have nonnegative coefficients. The conjecture stays open; the note establishes which certificates cannot settle it. MOTIVATION By the theorems of Pétréolle, Sokal and Zhu, one continued fraction or totally positive production matrix with nonnegative polynomial coefficients would prove the conjecture. So the first question is which kinds of certificate are still possible. Each type is matched to the moments a_1, a_2, … level by level, and a type is excluded at the first level where no nonnegative choice survives. Exclusion is a statement about the certificate, not about the conjecture: a sequence can be Hankel-totally positive with no certificate of a given type. STATUS Two-level lemma (in any m-branched S-fraction the first two moments force α_m = a_1 and α_(m+1) + … + α_(2m) = (a_2 − a_1²)/a_1): PROVED (two lines); novelty not checked. For A_n(1+y), (a_2 − a_1²)/a_1 has the negative coefficient −92/125 at y², so branched S-fractions are excluded at level 2 for every m: PROVED. J-fractions excluded at level 3: PROVED. Classical T-fractions excluded at level 4: PROVED. One lower-Hessenberg production-matrix branch, and 18 further branches sampled under a registered rule, excluded at level 5: PROVED for those branches (Farkas certificates, re-verified by code that shares nothing with the search). Interleaved bidiagonal templates ULL, LUL, LLUL: OPEN (they survive to levels 3, 2 and 3). The (1 + y) normal form: OPEN (survives to level 2; not canonical). Templates ULLL, LULL: NO VERDICT (the search engine failed its planted control). Contiguous Hankel minors up to 25 × 25 and 300 sampled non-contiguous minors have nonnegative coefficients: MEASURED (registered, controls passed); evidence, not proof. The same searches for Sokal's little Domb and Domb conjectures: MEASURED (registered). A sequence with every contiguous minor nonnegative and a negative one-gap minor (Appendix D1): PROVED (by hand, and by two programs). All 224 one-gap minors of sizes 2 to 8 from a_23 to a_30 have nonnegative coefficients (Appendix D2): REGISTERED PASS (replicated by a separately written program); evidence, not proof. Sokal's Conjecture 2 itself: OPEN. Not yet checked by an independent mathematician. RESULTS Section 2. The two-level lemma, with its path-counting proof, and the level-2 obstruction for the Apéry polynomials. The same test obstructs Sokal's other three conjectures of this kind in Deb's database: little Domb (−2/9 at y²), Domb (−1/4 at y²) and Boros–Moll (−3/16 at y³). Section 3. The ladder of exclusions: J-fractions at level 3 (b_1 = 35/2 + (295/24)y − (419/288)y² + …); classical T-fractions at level 4, by reducing each level to a factorisation into nonnegative polynomials; production-matrix branches at level 5, by a Farkas certificate with 55 nonnegative multipliers and constant K = −281468537621471/1600148025, and 18 sampled branches the same way. Section 4. The registered minor searches: all contiguous minors up to 25 × 25 (degree up to 600) and 300 random non-contiguous minors of the 20 × 20 matrix, with controls that catch a non-totally-positive sequence every time. The power of (y + 1) dividing the contiguous minor Δ_k is 1, 2, 3, 5 for k = 2 to 5, and the cofactors have positive coefficients. Appendix D (new in version 4). D1 works the counterexample by hand: the three contiguous 2 × 2 minors, the 3 × 3 determinant by Desnanot–Jacobi, the one-gap minor with its negative coefficient, the identity behind it, and a case that must fail. D2 reports the registered one-gap check, with a Kronecker-substitution example done by hand (10283 read in base 10 as the digits 3, −2, 3, 0, 1 after carrying). Every number in Appendix D is checked in exact arithmetic (paper3_appD_hand_check.py, 19 checks; a mutation test catches 10 of 10 deliberate errors). Section 5. Three open routes: deeper levels for the surviving templates, a lattice-path model through the Lindström–Gessel–Viennot lemma, and induction through the Desnanot–Jacobi identity. APPENDIX: THE MATHEMATICS Appendix A works every step with pencil, and each block ends with a case where the method must fail, or a control that must not. The certificate types, in words. A production matrix P produces a_n = (P^n)_00. For P = UL with β_0, β_1 = 2, 5 and α_1, α_2 = 3, 7: P_00 = 5, P_10 = 15, P_11 = 12 and a_2 = 25 + 15 = 40. Positive entries are not enough: [[1, 1], [2, 1]] has determinant −1, which is why the templates are built from bidiagonal factors. The data. A_1 = 1 + 4x and A_2 = 1 + 36x + 36x², so at x = 1 + y: a_1 = 5 + 4y and a_2 = 73 + 108y + 36y². The level-2 obstruction. a_2 − a_1² = 48 + 68y + 20y² = 4(y + 1)(5y + 12). Divide by a_1 = 5 + 4y as a power series: 5c_0 = 48 gives 48/5; 5c_1 + 4c_0 = 68 gives 148/25; 5c_2 + 4c_1 = 20 gives −92/125. The control (n!)² gives (4 − 1)/1 = 3 > 0. The case that must fail: 1, 2, 3 gives (3 − 4)/2 = −1/2. Why a rational series with nonnegative coefficients is a polynomial. By Pringsheim's theorem such a series with finite radius has a singularity on the positive axis. By long division, (a_2 − a_1²)/a_1 = 5y + 43/4 − (23/4)/(4y + 5), whose only pole, y = −5/4, is off that axis; so it cannot have nonnegative coefficients. The case the hypothesis leaves out: 1/(1 − 2y) = 1 + 2y + 4y² + … has its pole at y = 1/2. Classical T-fractions, worked. With G = 1/(1 − δ_0t − α_1t·F_1), level 2 forces α_1 = c, c(y + 1) or c(5y + 12). At level 3 the first and third fail, and α_1 = c(y + 1) survives only for c in [96/35, 4], because the

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