betaFeedback

WorksMathematics

The sum of the reciprocal cubes of the Fibonacci numbers is irrational

Abstract

We prove that ζF(3)=∑n≥1Fn−3\zeta_F(3)=\sum_{n\ge1}F_n^{-3} is irrational, where FnF_n are the Fibonacci numbers. To our knowledge the only odd value of the Fibonacci zeta function previously known to be irrational was ζF(1)\zeta_F(1). The proof is a Hankel-determinant construction over Z[φ]\mathbb{Z}[\varphi], φ=(1+5)/2\varphi=(1+\sqrt5)/2, adapted from a recent unrefereed construction for the qq-zeta value ∑n2/(pn−1)\sum n^2/(p^n-1) at integers p≥2p\ge2. Four estimates are combined: a real decay of size exp⁡(−29K3log⁡φ)\exp(-\tfrac29K^3\log\varphi), a smallness exp⁡(−K3log⁡φ)\exp(-K^3\log\varphi) at the Galois conjugate, which replaces the base-prime content gain of the integer case, local cyclotomic denominator bounds at the twisted roots of unity that occur, and nonvanishing by rational interpolation. The whole argument is formalised in Lean 4 with Mathlib.

Introduction

Let FnF_n be the Fibonacci numbers and ζF(s)=∑n≥1Fn−s\zeta_F(s)=\sum_{n\ge1}F_n^{-s} the Fibonacci zeta function. André-Jeannin proved that ζF(1)\zeta_F(1) is irrational [1], by Padé approximation of a Lambert series. Duverney, Ke. Nishioka, Ku. Nishioka and Shiokawa proved that the even values ζF(2s)\zeta_F(2s) are transcendental, and so are ∑n≥1F2n−1−s\sum_{n\ge1}F_{2n-1}^{-s} for every s≥1s\ge1 [2]; Elsner, Shimomura and Shiokawa determined the algebraic relations among these numbers [3, 4]. All of these sums are values of quasi-modular forms, to which Nesterenko’s theorem [11] applies. The odd values ζF(2s+1)\zeta_F(2s+1), s≥1s\ge1, are not of this kind, and as far as we know nothing was proved about their arithmetic (see the survey [16] and the introduction of [10]).

Theorem 1.1. ζF(3)=∑n≥11Fn3\zeta_F(3)=\sum_{n\ge1}\frac{1}{F_n^3} is irrational.

With t=φ−1=(5−1)/2t=\varphi^{-1}=(\sqrt{5}-1)/2 and Binet’s formula, Fn=t−n(1−(−t2)n)/5F_n=t^{-n}(1-(-t^2)^n)/\sqrt{5}, so

Ξ:=ζF(3)55=∑N≥1t3N(1−(−t2)N)3.(1)\Xi:=\frac{\zeta_F(3)}{5\sqrt{5}}=\sum_{N\ge1}\frac{t^{3N}}{(1-(-t^2)^N)^3}. \tag*{(1)}

This is a qq-series at the quadratic unit tt, a relative of the qq-zeta value ζq(3)=∑nn2qn/(1−qn)\zeta_q(3)=\sum_n n^2q^n/(1-q^n) [9]. For q=1/pq=1/p with p≥2p\ge2 an integer, the irrationality of ζq(3)\zeta_q(3) remains open in the refereed literature: Krattenthaler, Rivoal and Zudilin proved that one of ζq(3),…,ζq(11)\zeta_q(3),\ldots,\zeta_q(11) is irrational [9], Jouhet and Mosaki reduced this to ζq(3),…,ζq(9)\zeta_q(3),\ldots,\zeta_q(9) [7], and Fischler and Zudilin proved linear independence results of the same kind [6]. In October 2026 a proof that ∑n≥1n2/(pn−1)\sum_{n\ge1}n^2/(p^n-1) is irrational for every integer p≥2p\ge2 was claimed in an unrefereed, AI-assisted research draft with a Lean formalisation, posted on GitHub under the pen name uplift-universal-inc [14, 13]. Its construction is a moving-denominator Hankel determinant in the line of Uvarov [15], Krattenthaler [8], Prévost [12], Zudilin [17, 18] and the unrefereed preprint of Fauzan [5].

Expanding (1.1) in Lambert series does not reduce Theorem 1.1 to a statement about ζq(3)\zeta_q(3): one obtains a Q(5)\mathbb{Q}(\sqrt{5})-linear combination of 11, odd parts of ζq(3)\zeta_q(3) and ζq(1)\zeta_q(1) at q=φ−2q=\varphi^{-2}, and values of holomorphic modular forms of weights 1 and 3 at q=φ−1,φ−2q=\varphi^{-1},\varphi^{-2}. We therefore treat the series (1.1) directly.

The method

We adapt the construction of [14] to the real quadratic field Q(5)\mathbb{Q}(\sqrt{5}). A linear functional LL, a second Euler derivative summed against the weights (−t)n(-t)^n at the nodes (−t2)n(-t^2)^n, has rational moments and pure simple-pole tails of Ξ\Xi (Lemma 2.2). The Hankel determinants ΔK(X)\Delta_K(X) of LL against the moving denominators DK(s)=∏a≤K(1−(−t2)as)D_K(s)=\prod_{a\le K}(1-(-t^2)^a s) are polynomials of degree ≤K\le K in XX with ΔK(Ξ)=det⁡[L(si+j/DK)]\Delta_K(\Xi)=\det[L(s^{i+j}/D_K)]. Four estimates are combined:

size (per K3log⁡φK^3\log\varphi)
real decay ∣ΔK(Ξ)∣|\Delta_K(\Xi)|−2/9-2/9Proposition 3.1
conjugate ∣σΔK(−Ξ)∣|\sigma\Delta_K(-\Xi)|−1-1Proposition 5.1
denominators log⁡∣NCK(t)∣\log|N C_K(t)|≤c∗=1.1954…<11/9\le c_* = 1.1954\ldots< 11/9Propositions 4.4, 4.6
nonvanishinginfinitely many KKProposition 6.1

Table 2.

If ζF(3)\zeta_F(3) were rational, CK(t)(25v)KΔK(Ξ)C_K(t)(25v)^K\Delta_K(\Xi) would be a nonzero element of Z[φ]\mathbb{Z}[\varphi] of norm <1<1 for large KK.

Three ingredients are new compared with [14]. First, the setting: the Fibonacci kernel is a single qq-series over Z[φ]\mathbb{Z}[\varphi], handled with a formal parameter τ\tau specialised at τ=t\tau=t and at its conjugate τ=−φ\tau=-\varphi. Second, since φ\varphi is a unit there is no base prime and no coefficient content to extract. Its role is taken by the smallness of the conjugate determinant at the conjugate point σΞ=−Ξ\sigma\Xi=-\Xi (Proposition 5.1). Its size, K3log⁡φ=12K3log⁡φ2K^3\log\varphi=\frac{1}{2}K^3\log\varphi^2, is the size of the content gain 12K3log⁡p\frac{1}{2}K^3\log p of [14] at p=φ2p=\varphi^2. Third, the poles of the coefficients sit at two kinds of twisted roots of unity, those ζ\zeta for which −ζ2-\zeta^2 has small order and those for which −ζ(−ζ2)k=1-\zeta(-\zeta^2)^k=1; we prove local denominator bounds of both types (Section 4). The present paper does not depend logically on [14]: every step is proved here, and the formal proof is self-contained.

Formal verification

The theorem and every step of its proof are formalised in Lean 4 with Mathlib (https://github.com/telaviv/fibonacci-zeta3-irrationality); Section 8 gives the statement, the map from the propositions below to Lean declarations, and what has to be trusted. The formal proof uses the elementary denominator constant

c∗=118β7+13=43933675c_*=\frac{11}{8}\beta_7+\frac{1}{3}=\frac{4393}{3675}

of Proposition 4.6, where β7=∏p≤7(1−p−2)\beta_7=\prod_{p\leq7}(1-p^{-2}); it needs no information on the distribution of Euler’s function beyond a sieve by 2, 3, 5, 7.

The functional and the determinants

Notation

Throughout, φ=(1+5)/2\varphi=(1+\sqrt{5})/2 is the golden ratio, t=φ−1t=\varphi^{-1}, O=Z[φ]\mathcal{O}=\mathbb{Z}[\varphi], and σ\sigma is the nontrivial automorphism of Q(5)\mathbb{Q}(\sqrt{5}), so σt=−φ\sigma t=-\varphi. Euler’s function is written φE(d)\varphi_E(d) and the Möbius function μ(d)\mu(d). We put

ρ=−t,w=−t2,αk=ρwk (k≥0),\rho=-t,\qquad w=-t^2,\qquad\alpha_k=\rho w^k\ (k\geq0),
ψN=t3N(1−wN)3,Sa=∑N=1aψN,\psi_N=\frac{t^{3N}}{(1-w^N)^3},\qquad S_a=\sum_{N=1}^{a}\psi_N,

AK=K(K+1)/2A_K=K(K+1)/2, and θ=sdds\theta=s\frac{d}{ds}. Then ρ2=−w\rho^2=-w, ∣w∣=t2|w|=t^2 and Ξ=∑NψN\Xi=\sum_N\psi_N. We write hmh_m for the complete homogeneous symmetric polynomial of degree mm (hm=0h_m=0 for m<0m<0). Constants in O(⋅)O(\cdot) are absolute and K≥2K\geq2.

Definition 2.1. For gg analytic on a neighbourhood of ∣s∣≤t2|s|\leq t^2 put L(g)=∑n≥1ρn[θ(θ+1)g](wn)L(g)=\sum_{n\geq1}\rho^n[\theta(\theta+1)g](w^n).

The series converges absolutely since θ(θ+1)g=O(s)\theta(\theta+1)g=O(s) and ∣ρnwn∣=t3n|\rho^n w^n|=t^{3n}.

Lemma 2.2. (i) L(sk)=νk:=k(k+1)αk/(1−αk)L(s^k)=\nu_k:=k(k+1)\alpha_k/(1-\alpha_k) for k≥0k\geq0. (ii) For a≥1a\geq1, L(1/(1−was))=λa(Ξ−Sa)L(1/(1-w^a s))=\lambda_a(\Xi-S_a) with λa=2ρ−a\lambda_a=2\rho^{-a}.

Proof. (i) θ(θ+1)sk=k(k+1)sk\theta(\theta+1)s^k=k(k+1)s^k; sum the geometric series. (ii) θ(θ+1)(1−cs)−1=2cs/(1−cs)3\theta(\theta+1)(1-cs)^{-1}=2cs/(1-cs)^3, so the left side is ∑nρn2wa+n/(1−wa+n)3=2ρ−a∑N>aρNwN/(1−wN)3\sum_n\rho^n 2w^{a+n}/(1-w^{a+n})^3=2\rho^{-a}\sum_{N>a}\rho^Nw^N/(1-w^N)^3, and ρNwN=t3N\rho^Nw^N=t^{3N}. □

Let DK(s)=∏a=1K(1−was)D_K(s)=\prod_{a=1}^{K}(1-w^a s), with poles xa=w−ax_a=w^{-a}, ∣xa∣=φ2a|x_a|=\varphi^{2a}. For a polynomial PP of degree <2K−1<2K-1,

PDK=Q+∑a=1Kres⁡a(P)1−was,res⁡a(P)=P(xa)∏b≠a(1−wb−a),\frac{P}{D_K}=Q+\sum_{a=1}^{K}\frac{\operatorname{res}_a(P)}{1-w^a s},\qquad\operatorname{res}_a(P)=\frac{P(x_a)}{\prod_{b\ne a}(1-w^{b-a})},

with QQ a polynomial of degree ≤deg⁡P−K\leq\deg P-K. Let LXL_X be the linear map with LX(sk)=νkL_X(s^k)=\nu_k and LX(1/(1−was))=λa(X−Sa)L_X(1/(1-w^a s))=\lambda_a(X-S_a), XX an indeterminate, and

ΔK(X)=det⁡[LX(si+jDK)]0≤i,j<K∈Q(5)[X],deg⁡ΔK≤K.\Delta_K(X)=\det\left[L_X\left(\frac{s^{i+j}}{D_K}\right)\right]_{0\leq i,j<K}\in\mathbb{Q}(\sqrt{5})[X],\qquad\deg\Delta_K\leq K.

By Lemma 2.2, ΔK(Ξ)=det⁡[L(si+j/DK)]\Delta_K(\Xi)=\det[L(s^{i+j}/D_K)].

The formal parameter

Every quantity above is the value at τ=t\tau=t of a rational function of an indeterminate τ\tau with rational coefficients: replace tt by τ\tau, so ρ=−τ\rho=-\tau, w=−τ2w=-\tau^2, αk=(−1)k+1τ2k+1\alpha_k=(-1)^{k+1}\tau^{2k+1} and ψN=τ3N/(1−(−τ2N)3)\psi_N=\tau^{3N}/(1-(-\tau^{2N})^3). Write ΔK,τ(X)∈Q(τ)[X]\Delta_{K,\tau}(X)\in\mathbb{Q}(\tau)[X] for the resulting polynomial, and w(ζ)=−ζ2w(\zeta)=-\zeta^2, αk(ζ)\alpha_k(\zeta) for the values at τ=ζ\tau=\zeta. Since σt=−φ\sigma t=-\varphi, the coefficientwise conjugate σΔK\sigma\Delta_K is the value of ΔK,τ\Delta_{K,\tau} at τ=−φ\tau=-\varphi. Both values are defined because no denominator vanishes at a real number that is neither 0 nor a root of unity.

Real decay

Proposition 3.1. log⁡∣ΔK(Ξ)∣≤−29(K3+3K2−4)log⁡φ+O(K2)\log|\Delta_K(\Xi)|\leq-\frac{2}{9}(K^3+3K^2-4)\log\varphi+O(K^2).

Proof. Let Vj(s)=∏u=1j(s−w⌈u/3⌉)V_j(s)=\prod_{u=1}^{j}(s-w^{\lceil u/3\rceil}), Rj=⌈j/3⌉R_j=\lceil j/3\rceil and aj=∑u≤j⌈u/3⌉a_j=\sum_{u\leq j}\lceil u/3\rceil. The change of basis from 1,s,…,sK−11,s,\ldots,s^{K-1} to V0,…,VK−1V_0,\ldots,V_{K-1} is unitriangular, so ΔK(Ξ)=det⁡[L(ViVj/DK)]\Delta_K(\Xi)=\det[L(V_iV_j/D_K)].

Since θ(θ+1)g(s0)\theta(\theta+1)g(s_0) depends only on the 2-jet of gg at s0s_0, the mm-th term of L(ViVjW)L(V_iV_jW), W=1/DKW=1/D_K, vanishes for m<R:=max⁡(Ri,Rj)m<R:=\max(R_i,R_j): then ViVjV_iV_j has a triple zero at wmw^m. For m≥Rm\geq R and s=wms=w^m every root β\beta of VjV_j satisfies ∣s∣≤∣β∣|s|\leq|\beta|, so ∣s−β∣≤2∣β∣|s-\beta|\leq2|\beta| and θ(s−β)=s\theta(s-\beta)=s. Hence ∣Vj(s)∣≤2j∣w∣aj|V_j(s)|\leq2^j|w|^{a_j} and ∣θuVj(s)∣≤ju2j∣w∣aj∣s∣/∣w∣Rj|\theta^uV_j(s)|\leq j^u2^j|w|^{a_j}|s|/|w|^{R_j} for u=1,2u=1,2. On ∣s∣≤t2|s|\leq t^2 we have ∣was∣≤t2a+2|w^as|\leq t^{2a+2}, so ∣W∣|W|, ∣θW∣/∣s∣|\theta W|/|s| and ∣θ2W∣/∣s∣|\theta^2W|/|s| are bounded by an absolute constant. Since θ(θ+1)=θ2+θ\theta(\theta+1)=\theta^2+\theta kills constants, by Leibniz every term of θ(θ+1)(ViVjW)\theta(\theta+1)(V_iV_jW) contains at least one θ\theta on some factor, hence a factor ≤∣s∣/∣w∣R\leq|s|/|w|^R. With ∣ρ∣m≤1|\rho|^m\leq1 and ∑m≥R∣w∣m−R≤(1−t2)−1\sum_{m\geq R}|w|^{m-R}\leq(1-t^2)^{-1},

∣L(ViVjW)∣≤c1(i+j+1)24i+j∣w∣ai+aj.|L(V_iV_jW)|\leq c_1(i+j+1)^2 4^{i+j}|w|^{a_i+a_j}.

for an absolute constant c1c_1. Pull ∣w∣ai|w|^{a_i} out of row ii and column ii and apply Hadamard’s inequality: ∣ΔK(Ξ)∣≤∣w∣2∑j<Kaj(c1(2K)242KK)K|\Delta_K(\Xi)|\leq|w|^{2\sum_{j<K}a_j}(c_1(2K)^2 4^{2K}\sqrt K)^K. Finally 2∑j<Kaj=(K3+3K2−ιK)/92\sum_{j<K}a_j=(K^3+3K^2-\iota_K)/9 with ιK∈{0,4,2}\iota_K\in\{0,4,2\} according to K mod 3K\bmod3. □

Denominators

The integrality ring and the root types

Lemma 4.1. The coefficients of ΔK,τ(X)\Delta_{K,\tau}(X) lie in Z[τ,τ−1,Φd(τ)−1:d∈DK]\mathbb{Z}[\tau,\tau^{-1},\Phi_d(\tau)^{-1}:d\in D_K], where DKD_K is the set of orders of roots of unity ζ\zeta such that (W) w(ζ)w(\zeta) has order e≤Ke\leq K, or (α\alpha) αk(ζ)=1\alpha_k(\zeta)=1 for some 1≤k≤K−21\leq k\leq K-2.

Proof. The inputs are the Laurent monomials waw^a and λa=2(−τ)−a\lambda_a=2(-\tau)^{-a}; the factors 1/(1−wb−a)1/(1-w^{b-a}) and (1−wN)−3(1-w^N)^{-3} (N≤KN\leq K), where 1−wn=1−(−1)nτ2n1-w^n=1-(-1)^n\tau^{2n} is ±\pm a product of cyclotomic polynomials Φd(τ)\Phi_d(\tau) with w(ζ)n=1w(\zeta)^n=1; and νk=k(k+1)αk/(1−αk)\nu_k=k(k+1)\alpha_k/(1-\alpha_k), of which only k≤K−2k\leq K-2 occur (the quotient has degree ≤2K−2−K\leq2K-2-K), with 1−αk=1∓τ2k+1=±∏Φd(τ)1-\alpha_k=1\mp\tau^{2k+1}=\pm\prod\Phi_d(\tau). Division by DKD_K is integral because its leading coefficient ∏a(−wa)=±τK(K+1)\prod_a(-w^a)=\pm\tau^{K(K+1)} is a unit. No rational constant is inverted. □

For d∈DKd \in D_K the type depends only on dd. If ζ\zeta has order dd, then w(ζ)w(\zeta) has order e(d)=2de(d)=2d (dd odd), dd (d≡2(mod4)d \equiv2 \pmod{4}), d/4d/4 (d≡4(mod8)d \equiv4 \pmod{8}), d/2d/2 (d≡0(mod8)d \equiv0 \pmod{8}), and for each ee the orders dd with e(d)=ee(d)=e satisfy ∑e(d)=eφE(d)=2φE(e)\sum_{e(d)=e}\varphi_E(d)=2\varphi_E(e). The least k≥1k \ge1 with αk(ζ)=1\alpha_k(\zeta)=1 is kmin⁡(d)=(d−1)/2k_{\min}(d)=(d-1)/2 (d≡3(mod4)d \equiv3 \pmod{4}), (3d−1)/2(3d-1)/2 (d≡1(mod4)d \equiv1 \pmod{4}), (d−2)/4(d-2)/4 (d≡2(mod8)d \equiv2 \pmod{8}, d≥10d \ge10), (3d−2)/4(3d-2)/4 (d≡6(mod8)d \equiv6 \pmod{8}), and kmin⁡(2)=2k_{\min}(2)=2 (the formula (d−2)/4(d-2)/4 would give 00); there is none if 4∣d4 \mid d. (Every primitive dd-th root is a power ζj\zeta^j with jj coprime to 2d2d; then −ζ2j=(−ζ2)j-\zeta^{2j}=(-\zeta^2)^j and αk(ζj)=αk(ζ)j\alpha_k(\zeta^j)=\alpha_k(\zeta)^j, and powers coprime to the order preserve the order.) All d∈DKd \in D_K satisfy d≤4Kd \le4K: w(ζ)e=1w(\zeta)^e=1 gives ζ4e=1\zeta^{4e}=1, and αk(ζ)=1\alpha_k(\zeta)=1 gives ζ4k+2=1\zeta^{4k+2}=1.

Local setting. Fix ζ\zeta of order d∈DKd \in D_K, let ϖ=τ−ζ\varpi=\tau-\zeta and let v=ord⁡ϖv=\operatorname{ord}_{\varpi} on Q(ζ)((ϖ))\mathbb{Q}(\zeta)((\varpi)), extended to polynomials in XX as the Gauss valuation (the minimum over the coefficients; it is multiplicative). Then:

(a) if e(d)=e≤Ke(d)=e \le K, then xa(ζ)=xb(ζ)x_a(\zeta)=x_b(\zeta) iff e∣a−be \mid a-b, and then v(xa−xb)=1v(x_a-x_b)=1: indeed xa−xb=xa(1−wb−a)x_a-x_b=x_a(1-w^{b-a}), and 1−wn1-w^n has only simple zeros, since ddτwn≠0\frac{d}{d\tau}w^n \ne0 at τ≠0\tau\ne0;

(b) ψN\psi_N has a pole of order exactly 33 if e∣Ne \mid N and is regular otherwise, by (a); so SaS_a has a pole of order ≤3\le3, and is regular if a<ea<e;

(c) 1−αk=1∓τ2k+11-\alpha_k=1\mp\tau^{2k+1} is separable, so it has at most a simple zero at ζ\zeta;

(d) λa\lambda_a and xax_a are Laurent monomials, hence units.

Local bounds

Put c(m)=2∑j=0m−2max⁡(0,m+22−j)c(m)=2\sum_{j=0}^{m-2}\max(0,\frac{m+2}{2}-j) and c′(m)=2∑j=0m−1max⁡(0,m+22−j)c'(m)=2\sum_{j=0}^{m-1}\max(0,\frac{m+2}{2}-j); thus c(1)=0c(1)=0, c(m)=4(m−1)c(m)=4(m-1) for m≤6m\le6, c(m)≤(m+3)2/4c(m)\le(m+3)^2/4, c′(1)=3c'(1)=3 and 0≤c′−c≤30\le c'-c\le3.

Proposition 4.2. Let e=e(d)≤Ke=e(d)\le K and partition {1,…,K}\{1,\ldots,K\} into the clusters Ir={a≡r(mode)}I_r=\{a\equiv r\pmod e\} of sizes mrm_r, 1≤r≤e1\le r\le e. Then

−v(ΔK,τ)≤∑r=1e−1c(mr)+c′(me)+ϵmax⁡(0,K−1−kmin⁡(d)),-v(\Delta_{K,\tau})\le\sum_{r=1}^{e-1}c(m_r)+c'(m_e)+\epsilon\max(0,K-1-k_{\min}(d)),

with ϵ=1\epsilon=1 if ζ\zeta is also an α\alpha-root and ϵ=0\epsilon=0 otherwise.

Proof. Let F(s)=∏a(s−xa)F(s)=\prod_a(s-x_a) and Fr=∏a∈Ir(s−xa)F_r=\prod_{a\in I_r}(s-x_a). Fix a cluster rr, write m=mrm=m_r and Ir={a1<⋯<am}I_r=\{a_1<\cdots<a_m\}.

Newton cluster basis. For 0≤j<m0\le j<m put Nr,j=(F/Fr)∏ℓ=m−j+1m(s−xaℓ)N_{r,j}=(F/F_r)\prod_{\ell=m-j+1}^{m}(s-x_{a_\ell}); it vanishes at the top jj nodes of the cluster. At ϖ=0\varpi=0 these reduce to (F/Fr)(s−yr)j⋅(unit)(F/F_r)(s-y_r)^j\cdot(\mathrm{unit}), yr=xr(ζ)y_r=x_r(\zeta), which form a basis of the polynomials of degree <K<K by the Chinese remainder theorem (distinct residues rr give distinct points yry_r). So the change of basis from 1,…,sK−11,\ldots,s^{K-1} has unit determinant, and v(ΔK,τ)v(\Delta_{K,\tau}) is the valuation of G=[LX(NN′/DK)]G=[L_X(NN'/D_K)] in this basis.

Entries. G(r,j),(r′,k)=P+∑ares⁡a(NN′)λa(X−Sa)G_{(r,j),(r',k)}=P+\sum_a\operatorname{res}_a(NN')\lambda_a(X-S_a), where PP is LXL_X of the polynomial part. (i) If r≠r′r\ne r' then F∣Nr,jNr′,kF\mid N_{r,j}N_{r',k} and all res⁡a=0\operatorname{res}_a=0. (ii) If r=r′r=r', then res⁡ai=0\operatorname{res}_{a_i}=0 unless i≤m−max⁡(j,k)i\le m-\max(j,k), and for such ii

res⁡ai=(F/Fr)(xai)2∏l>m−j(xai−xal)∏l>m−k(xai−xal)∏b≠ai(1−wb−ai).\operatorname{res}_{a_i}=\frac{(F/F_r)(x_{a_i})^2\prod_{l>m-j}(x_{a_i}-x_{a_l})\prod_{l>m-k}(x_{a_i}-x_{a_l})}{\prod_{b\ne a_i}(1-w^{b-a_i})}.

Here (F/Fr)(xai)(F/F_r)(x_{a_i}) is a unit, each displayed difference has valuation exactly 11 by (a), and the denominator has valuation exactly m−1m-1. So v(res⁡ai)=j+k−(m−1)v(\operatorname{res}_{a_i})=j+k-(m-1). (iii) v(λa)=0v(\lambda_a)=0, and by (b) the Gauss valuation of X−SaX-S_a is ≥−3\ge-3 if a≥ea\ge e and ≥0\ge0 if a<ea<e. (iv) For r<er<e the node a1=ra_1=r is regular and aia_i is singular for i≥2i\ge2; for r=er=e every node is singular. So the proper part of the (j,k)(j,k) entry has v≥j+k−(m−1)−3v\ge j+k-(m-1)-3 if some singular node has i≤m−max⁡(j,k)i\le m-\max(j,k), and v≥j+k−(m−1)v\ge j+k-(m-1) otherwise. The second case occurs exactly when r<er<e and max⁡(j,k)=m−1\max(j,k)=m-1.

Weights. Let κj=max⁡(0,m+22−j)\kappa_j=\max(0,\frac{m+2}{2}-j) for j≤m−2j\le m-2 and κm−1=0\kappa_{m-1}=0 if r<er<e, and κj=max⁡(0,m+22−j)\kappa_j=\max(0,\frac{m+2}{2}-j) for all j<mj<m if r=er=e. We claim, following the two cases of (iv):

(A) if the (j,k)(j,k) entry has a singular node (all pairs when r=er=e; the pairs with max⁡(j,k)≤m−2\max(j,k)\le m-2 when r<er<e), then κj+κk≥m+2−j−k\kappa_j+\kappa_k\ge m+2-j-k;

(B) if r<er<e and max⁡(j,k)=m−1\max(j,k)=m-1, then κj+κk≥0≥m−1−j−k\kappa_j+\kappa_k\ge0\ge m-1-j-k.

In case (A) both indices carry the weight max⁡(0,m+22−j)\max(0,\frac{m+2}{2}-j) (for r=er=e this includes κm−1=max⁡(0,4−m2)\kappa_{m-1}=\max(0,\frac{4-m}{2}). If both weights are positive the inequality is an equality. If κk=0\kappa_k=0, then k≥m+22k\ge\frac{m+2}{2}, so κj≥m+22−j≥m+2−j−k\kappa_j\ge\frac{m+2}{2}-j\ge m+2-j-k. If both vanish, then j,k≥m+22j,k\ge\frac{m+2}{2} and the right side is ≤0\le0. Case (B) holds because j+k≥m−1j+k\ge m-1.

Scale row (r,j)(r,j) by ϖ⌈κj⌉\varpi^{\lceil\kappa_j\rceil} and column (r,j)(r,j) by ϖ⌊κj⌋\varpi^{\lfloor\kappa_j\rfloor}. The required exponents are integers, and a+b≥n∈Za+b\ge n\in\mathbb{Z} implies ⌈a⌉+⌊b⌋≥n\lceil a\rceil+\lfloor b\rfloor\ge n, so by (iv), (A) and (B) every scaled proper entry is integral. The total scaling is ∑(⌈κj⌉+⌊κj⌋)=2∑κj\sum(\lceil\kappa_j\rceil+\lfloor\kappa_j\rfloor)=2\sum\kappa_j, which is c(m)c(m) for r<er<e and c′(m)c'(m) for r=er=e. All exponents are ≥0\ge0.

Polynomial part. The quotient of NN′NN' by DKD_K has degree ≤K−2\le K-2 and coefficients in Z[τ±1]\mathbb{Z}[\tau^{\pm1}], so P=∑k≤K−2(coeff⁡)νkP=\sum_{k\le K-2}(\operatorname{coeff})\nu_k. At a pure W-root every νk\nu_k, k≤K−2k\le K-2, is regular, so PP is integral, and so is the scaled matrix; this gives the bound with ϵ=0\epsilon=0.

At an α\alpha-root, put Pα={k≤K−2:αk(ζ)=1}P_\alpha=\{k\le K-2:\alpha_k(\zeta)=1\} and write νk=ϖ−1uk\nu_k=\varpi^{-1}u_k with uku_k regular for k∈Pαk\in P_\alpha (by (c)). Split P=Apol+ϖ−1BP=A^{\mathrm{pol}}+\varpi^{-1}B, where B(f,g)=∑k∈Pαuk(0)[sk](fg/DK)+B(f,g)=\sum_{k\in P_\alpha}u_k(0)[s^k](fg/D_K)_+ (here (⋅)+(\cdot)_+ is the polynomial part) and ApolA^{\mathrm{pol}} collects ϖ−1(uk−uk(0))[sk](⋅)+\varpi^{-1}(u_k-u_k(0))[s^k](\cdot)_+ for k∈Pαk\in P_\alpha and νk[sk](⋅)+\nu_k[s^k](\cdot)_+ for k∉Pαk\notin P_\alpha; both ApolA^{\mathrm{pol}} and BB have regular entries. If deg⁡f≤kmin⁡\deg f\le k_{\min}, then deg⁡(fg)≤kmin⁡+K−1\deg(fg)\le k_{\min}+K-1, so the quotient has degree <kmin⁡<k_{\min} and B(f,⋅)=0B(f,\cdot)=0. Since the polynomials of degree ≤kmin⁡\le k_{\min} form a space of dimension kmin⁡+1k_{\min}+1, rank⁡B≤max⁡(0,K−1−kmin⁡)\operatorname{rank}B\le\max(0,K-1-k_{\min}). After scaling, the matrix is A+ϖ−1B′A+\varpi^{-1}B'. Here AA is the sum of the scaled proper parts, integral by (A) and (B), and the scaled ApolA^{\mathrm{pol}}, integral because all exponents are ≥0\ge0. The matrix B′B' is BB in the Newton basis, scaled; its entries are regular and its rank is that of BB, since the scalings are invertible over the fraction field. Expand the determinant multilinearly in the columns: a term taking more than max⁡(0,K−1−kmin⁡)\max(0,K-1-k_{\min}) columns from ϖ−1B′\varpi^{-1}B' vanishes by linear dependence, and every other term has valuation ≥−max⁡(0,K−1−kmin⁡)\ge-\max(0,K-1-k_{\min}). □\square

Lemma 4.3. If e(d)>Ke(d)>K and ζ\zeta is an α\alpha-root, then −v(ΔK,τ)≤max⁡(0,K−1−kmin⁡(d))-v(\Delta_{K,\tau})\le\max(0,K-1-k_{\min}(d)).

Proof. Work in the monomial basis. Since e>Ke>K, no difference ∣a−b∣<K|a-b|<K is divisible by ee, so by (a)–(d) the coefficients res⁡a\operatorname{res}_a, the λa\lambda_a and the SaS_a (a≤K<ea\le K<e) are regular at ζ\zeta. Only the νk\nu_k, k∈Pαk\in P_\alpha, carry poles, and the rank argument of Proposition 4.2 applies with no scaling. □\square

Proposition 4.4. Let bK,db_{K,d} be the bound of Proposition 4.2 or Lemma 4.3 (00 if d∉DKd\notin D_K) and CK(τ)=∏dΦd(τ)bK,dC_K(\tau)=\prod_d\Phi_d(\tau)^{b_{K,d}}. Then CK(τ)ΔK,τ(X)∈Z[τ±1][X]C_K(\tau)\Delta_{K,\tau}(X)\in\mathbb{Z}[\tau^{\pm1}][X]; hence CK(t)ΔK(X)∈O[X]C_K(t)\Delta_K(X)\in\mathcal{O}[X], and CK(−φ)σΔK(X)∈O[X]C_K(-\varphi)\sigma\Delta_K(X)\in\mathcal{O}[X] is its conjugate.

Proof. By Lemma 4.1 a coefficient is P/(τn∏dΦdnd)P/(\tau^n\prod_d\Phi_d^{n_d}) with P∈Z[τ]P\in\mathbb{Z}[\tau]. The bound vζ≥−bK,dv_\zeta\ge-b_{K,d} at every primitive dd-th root forces Φdnd−bK,d∣P\Phi_d^{n_d-b_{K,d}}\mid P in Q[τ]\mathbb{Q}[\tau], hence in Z[τ]\mathbb{Z}[\tau] since Φd\Phi_d is monic. Specialise τ↦t\tau\mapsto t, a unit of O\mathcal{O}, and τ↦−φ\tau\mapsto-\varphi. □\square

The size of CK(t)C_K(t)

Lemma 4.5. For d≥1d\ge1, ∣log⁡∣Φd(t)Φd(−φ)∣−ϕE(d)log⁡φ∣≤2E1|\log|\Phi_d(t)\Phi_d(-\varphi)|-\phi_E(d)\log\varphi|\le2\mathcal{E}_1, where

E1=∑a≥1∣log⁡(1−φ−a)∣=2.1136…\mathcal{E}_1=\sum_{a\ge1}|\log(1-\varphi^{-a})|=2.1136\ldots

Proof. Φd(x)=∏a∣d(xa−1)μ(d/a)\Phi_d(x)=\prod_{a\mid d}(x^a-1)^{\mu(d/a)} and ∑a∣dμ(d/a)a=ϕE(d)\sum_{a\mid d}\mu(d/a)a=\phi_E(d). For ∣x∣=φ|x|=\varphi write xa−1=xa(1−x−a)x^a-1=x^a(1-x^{-a}), and ∣log⁡∣1−x−a∣∣≤∣log⁡(1−φ−a)∣|\log|1-x^{-a}||\le|\log(1-\varphi^{-a})|; for x=tx=t use ∣1−ta∣|1-t^a| in the same way. □\square

Let β7=∏p≤7(1−p−2)=7681225\beta_7=\prod_{p\le7}(1-p^{-2})=\frac{768}{1225} and c∗=118β7+13=43933675=1.19537…c_*=\frac{11}{8}\beta_7+\frac{1}{3}=\frac{4393}{3675}=1.19537\ldots.

Proposition 4.6. log⁡∣N CK(t)∣≤c∗K3log⁡φ+O(K2log⁡K)\log|\mathrm{N}\,C_K(t)|\le c_*K^3\log\varphi+O(K^2\log K).

Proof. By Lemma 4.5, log⁡∣N CK(t)∣≤log⁡φ∑dbK,dϕE(d)+2E1∑dbK,d\log|\mathrm{N}\,C_K(t)|\le\log\varphi\sum_d b_{K,d}\phi_E(d)+2\mathcal{E}_1\sum_d b_{K,d}. Here ∑dbK,d=O(K2log⁡K)\sum_d b_{K,d}=O(K^2\log K): for each e≤Ke\le K at most four dd have e(d)=ee(d)=e, each with a W part ≤4K+K2/(4e)+3\le4K+K^2/(4e)+3 (see below), and the at most 4K4K orders d≤4Kd\le4K have α\alpha parts ≤K\le K. Split bK,db_{K,d} into its W part and its α\alpha part.

W part. Since mr≥1m_r\ge1, ∑rmr=K\sum_r m_r=K and mr<K/e+1m_r<K/e+1, and since c(m)=4(m−1)+δ(m)c(m)=4(m-1)+\delta(m) with 0≤δ(m)≤max⁡(0,m−5)2/40\le\delta(m)\le\max(0,m-5)^2/4 (for m≥6m\ge6, c(m)=(m+2)(m+4)/4c(m)=(m+2)(m+4)/4 or (m+3)2/4(m+3)^2/4 according to the parity of mm),

∑r<ec(mr)+c′(me)≤4(K−e)+max⁡(0,K−4e)24e+3.\sum_{r<e}c(m_r)+c'(m_e)\le4(K-e)+\frac{\max(0,K-4e)^2}{4e}+3.

Summing over the orders dd with e(d)=ee(d)=e gives the weight ∑e(d)=eϕE(d)=2ϕE(e)\sum_{e(d)=e}\phi_E(d)=2\phi_E(e). Bound ϕE(e)≤eη(e)\phi_E(e)\le e\eta(e) with η(e)=∏p≤7p∣e(1−1/p)\eta(e)=\prod_{\substack{p\le7\\p\mid e}}(1-1/p), a function of period P=210P=210 with 0≤η≤10\le\eta\le1 and mean β7\beta_7. So the W part is at most ∑e≤Kη(e)fK(e)\sum_{e\le K}\eta(e)f_K(e), where

fK(y)=8y(K−y)+12max⁡(0,K−4y)2+6y.f_K(y)=8y(K-y)+\frac{1}{2}\max(0,K-4y)^2+6y.

We use the elementary estimate: if η\eta has period PP, 0≤η≤10 \le\eta\le1 and mean β\beta, and ff is a function on [0,K][0,K] with ∣f∣≤M|f| \le M and total variation VV, then

∣∑1≤e≤Kη(e)f(e)−β∫0Kf(y) dy∣≤2P(M+V).\left|\sum_{1\le e\le K}\eta(e)f(e)-\beta\int_0^K f(y)\,dy\right|\le2P(M+V).

(Cut [0,K][0,K] into blocks of length PP. On a full block (e0,e0+P](e_0,e_0+P] both ∑η(e)f(e)\sum\eta(e)f(e) and β∫f\beta\int f differ from Pβf(e0)P\beta f(e_0) by at most PP times the oscillation of ff on the block; the oscillations add up to at most VV, and the last, incomplete block costs at most 2PM2PM.) For fKf_K we have M,V=O(K2)M,V=O(K^2), so the W part is at most

β7∫0KfK(y) dy+O(K2)=(43+124)β7K3+O(K2)=118β7K3+O(K2).\beta_7\int_0^K f_K(y)\,dy+O(K^2)=\left(\frac{4}{3}+\frac{1}{24}\right)\beta_7K^3+O(K^2)=\frac{11}{8}\beta_7K^3+O(K^2).

α\alpha part. max⁡(0,K−1−kmin⁡)\max(0,K-1-k_{\min}) counts the k∈[kmin⁡,K−2]k\in[k_{\min},K-2], so

∑dmax⁡(0,K−1−kmin⁡(d))ϕE(d)≤∑k=1K−2∑j=1k#{ζ:αj(ζ)=1}=∑k=1K−2(k2+2k)≤13K3+O(K2),\begin{aligned} \sum_d\max(0,K-1-k_{\min}(d))\phi_E(d)\le\sum_{k=1}^{K-2}\sum_{j=1}^k\#\{\zeta:\alpha_j(\zeta)=1\} \\ &=\sum_{k=1}^{K-2}(k^2+2k)\le\frac{1}{3}K^3+O(K^2), \end{aligned}

since αj(ζ)=1\alpha_j(\zeta)=1 means ζ2j+1=(−1)j+1\zeta^{2j+1}=(-1)^{j+1}, which has 2j+12j+1 solutions. □

Remark 4.7. The following is a sketch and is not used in the proof. Using ∑n≤yϕE(n)=3y2/π2+O(ylog⁡y)\sum_{n\le y}\phi_E(n)=3y^2/\pi^2+O(y\log y) in residue classes modulo 8 and the exact kmin⁡k_{\min}, the same bound holds with c∗c_* replaced by (12I+80/27)/π2≤1.130790(12I+80/27)/\pi^2\le1.130790, where I=∫01ℓ(x)x dx∈[0.6822834,0.6831236]I=\int_0^1\ell(x)x\,dx\in[0.6822834,0.6831236] and ℓ\ell is the piecewise linear function with ℓ(1/m)=c(m)/m\ell(1/m)=c(m)/m. Numerically, ∑dbK,dϕE(d)/K3\sum_{d}b_{K,d}\phi_E(d)/K^3 is 1.19 at K=20K=20 and 1.16 at K=40K=40, decreasing towards 1.1308.

Smallness at the conjugate

Proposition 5.1. log⁡∣σΔK(−Ξ)∣≤−K2(K+1)log⁡φ+Klog⁡(KMK)\log|\sigma\Delta_K(-\Xi)|\le-K^2(K+1)\log\varphi+K\log(\sqrt{K}M_K), where MK=C0(K2/((1−t3)(1−t2))+φ2)M_K=C_0(K^2/((1-t^3)(1-t^2))+\varphi^2) and C0=∏i≥1(1−t2i)−1=2.0770906…C_0=\prod_{i\ge1}(1-t^{2i})^{-1}=2.0770906\ldots.

Proof. At τ=−φ\tau=-\varphi the data become: w↦σw=w−1w\mapsto\sigma w=w^{-1}, λa↦σλa=2ta\lambda_a\mapsto\sigma\lambda_a=2t^a, Sa↦σSa=−SaS_a\mapsto\sigma S_a=-S_a (as σψN=−ψN\sigma\psi_N=-\psi_N) and νk↦σνk=−k(k+1)/(1+αk)\nu_k\mapsto\sigma\nu_k=-k(k+1)/(1+\alpha_k) (as σαk=−1/αk\sigma\alpha_k=-1/\alpha_k). Let LXσL_X^\sigma be the corresponding formal functional: LXσ(sk)=σνkL_X^\sigma(s^k)=\sigma\nu_k and LXσ(1/(1−w−as))=2ta(X+Sa)L_X^\sigma(1/(1-w^{-a}s))=2t^a(X+S_a). The conjugate poles are ra=war_a=w^a, ∣ra∣=t2a|r_a|=t^{2a}, and σDK(s)=∏a(1−w−as)=(−1)Kw−AKFσ(s)\sigma D_K(s)=\prod_a(1-w^{-a}s)=(-1)^Kw^{-A_K}F^\sigma(s) with Fσ(s)=∏a(s−ra)F^\sigma(s)=\prod_a(s-r_a). Hence

σΔK(X)=(−1)K2wAKKdet⁡[Yi+j(X)],Yk(X)=LXσ(sk/Fσ).\sigma\Delta_K(X)=(-1)^{K^2}w^{A_KK}\det[Y_{i+j}(X)],\qquad Y_k(X)=L_X^\sigma(s^k/F^\sigma).

Partial fractions give sk/Fσ=∑j=0k−Khk−K−j(r)sj+∑arak/((Fσ)′(ra)(s−ra))s^k/F^\sigma=\sum_{j=0}^{k-K}h_{k-K-j}(r)s^j+\sum_a r_a^k/((F^\sigma)'(r_a)(s-r_a)). The pole moments at X=−ΞX=-\Xi. Since 1/(1−w−as)=−ra/(s−ra)1/(1-w^{-a}s)=-r_a/(s-r_a), we have LXσ(1/(s−ra))=−ra−12ta(X+Sa)L_X^\sigma(1/(s-r_a))=-r_a^{-1}2t^{a}(X+S_a), and at X=−ΞX=-\Xi this is 2tara−1(Ξ−Sa)=2tara−1∑N>aψN2t^ar_a^{-1}(\Xi-S_a)=2t^ar_a^{-1}\sum_{N>a}\psi_N. Writing N=a+MN=a+M, t3N=t3at3Mt^{3N}=t^{3a}t^{3M}, t4a=ra2t^{4a}=r_a^2 and wa+M=rawMw^{a+M}=r_aw^M, this equals Γ(ra)\Gamma(r_a), where

Γ(r)=2r∑M≥1t3M(1−rwM)3=∑n≥1γnrn,γn=n(n+1)∑M≥1t3MwM(n−1),\Gamma(r)=2r\sum_{M\geq1}\frac{t^{3M}}{(1-rw^M)^3}=\sum_{n\geq1}\gamma_nr^n,\qquad\gamma_n=n(n+1)\sum_{M\geq1}t^{3M}w^{M(n-1)},

analytic on ∣r∣<φ2|r|<\varphi^2, with ∣γn∣≤n(n+1)t2n+1/(1−t2n+1)≤n(n+1)t2n+1/(1−t3)|\gamma_n|\leq n(n+1)t^{2n+1}/(1-t^{2n+1})\leq n(n+1)t^{2n+1}/(1-t^3). Since ∑aram/(Fσ)′(ra)=hm−K+1(r)\sum_a r_a^m/(F^\sigma)'(r_a)=h_{m-K+1}(r),

Yk(−Ξ)=∑j=0k−Khk−K−j(r) σνj+∑n≥1γnhk+n−K+1(r).Y_k(-\Xi)=\sum_{j=0}^{k-K}h_{k-K-j}(r)\,\sigma\nu_j+\sum_{n\geq1}\gamma_nh_{k+n-K+1}(r).

Bounds. As hm(r1,…,rK)=wmhm(1,w,…,wK−1)h_m(r_1,\ldots,r_K)=w^mh_m(1,w,\ldots,w^{K-1}) and the coefficients of ∏i≥0(1−∣w∣iu)−1=∑mum∏i=1m(1−t2i)−1\prod_{i\geq0}(1-|w|^iu)^{-1}=\sum_m u^m\prod_{i=1}^m(1-t^{2i})^{-1} are at most C0C_0, we get ∣hm(r)∣≤C0t2m|h_m(r)|\leq C_0t^{2m}. For j≥1j\geq1, ∣σνj∣≤j(j+1)/(1−t3)|\sigma\nu_j|\leq j(j+1)/(1-t^3), so the first sum is at most C0K2/((1−t3)(1−t2))C_0K^2/((1-t^3)(1-t^2)) for k≤2K−2k\leq2K-2. For the second, ∑nn(n+1)t2n+1=2t3/(1−t2)3=2\sum_n n(n+1)t^{2n+1}=2t^3/(1-t^2)^3=2 (as 1−t2=t1-t^2=t) and 1/(1−t3)=φ2/21/(1-t^3)=\varphi^2/2, so ∑n∣γn∣≤φ2\sum_n|\gamma_n|\leq\varphi^2. Hence ∣Yk(−Ξ)∣≤MK|Y_k(-\Xi)|\leq M_K for 0≤k≤2K−20\leq k\leq2K-2. Hadamard’s inequality and ∣w∣AKK=φ−K2(K+1)|w|^{A_KK}=\varphi^{-K^2(K+1)} finish the proof. □

Remark 5.2. Coefficientwise, the factor 2ta=2∣ra∣1/22t^a=2|r_a|^{1/2} is not analytic in rar_a; only at the point X=σΞ=−ΞX=\sigma\Xi=-\Xi does it combine with the tail ∑N>aψN≈t3a\sum_{N>a}\psi_N\approx t^{3a} into ra2r_a^2 times an analytic function. This is why the conjugate point, rather than the coefficients of σΔK\sigma\Delta_K, carries the gain.

Nonvanishing

Proposition 6.1. ΔK(Ξ)≠0\Delta_K(\Xi)\ne0 for infinitely many KK.

Proof. Let C(z)=L(1/(z−s))=∑nρn2wnz/(z−wn)3C(z)=L(1/(z-s))=\sum_n\rho^n2w^n z/(z-w^n)^3 for ∣z∣>t2|z|>t^2. It continues to C∖{0}\mathbb{C}\setminus\{0\} with a pole of order exactly 33 at each wnw^n (leading coefficient 2ρnw2n≠02\rho^nw^{2n}\ne0), so it is not rational; it is analytic at ∞\infty and O(1/z)O(1/z). For deg⁡Q≤n−1\deg Q\leq n-1 put Hn,Q(z)=L(Q(s)/((z−s)Dn(s)))H_{n,Q}(z)=L(Q(s)/((z-s)D_n(s))), analytic for ∣z∣>t2|z|>t^2 and in particular at every xax_a, and

Tn,Q(z)=L(Q(z)Dn(s)−Dn(z)Q(s)(z−s)Dn(s)),T_{n,Q}(z)=L\left(\frac{Q(z)D_n(s)-D_n(z)Q(s)}{(z-s)D_n(s)}\right),

a polynomial in zz of degree ≤n−1\leq n-1. Then CQ−Tn,Q=DnHn,QCQ-T_{n,Q}=D_nH_{n,Q}. If Δn(Ξ)=det⁡[L(si+j/Dn)]=0\Delta_n(\Xi)=\det[L(s^{i+j}/D_n)]=0, a kernel vector gives Qn≠0Q_n\ne0 of degree ≤n−1\leq n-1 with L(sjQn/Dn)=0L(s^jQ_n/D_n)=0 for j<nj<n. Put Tn=Tn,QnT_n=T_{n,Q_n} and En=CQn−TnE_n=CQ_n-T_n. For ∣z∣>t2+ϵ|z|>t^2+\epsilon the series (z−s)−1=∑jsjz−j−1(z-s)^{-1}=\sum_{j}s^jz^{-j-1} converges, together with its first two ss-derivatives, uniformly on ∣s∣≤t2|s|\leq t^2, so LL may be applied termwise; hence Hn,Qn(z)=∑j≥nL(sjQn/Dn)z−j−1=O(z−n−1)H_{n,Q_n}(z)=\sum_{j\geq n}L(s^jQ_n/D_n)z^{-j-1}=O(z^{-n-1}). So En=O(z−1)E_n=O(z^{-1}) and En(xa)=0E_n(x_a)=0 for a≤na\leq n. If Δn(Ξ)=Δn+1(Ξ)=0\Delta_n(\Xi)=\Delta_{n+1}(\Xi)=0, then V=Tn+1Qn−TnQn+1=Qn+1En−QnEn+1V=T_{n+1}Q_n-T_nQ_{n+1}=Q_{n+1}E_n-Q_nE_{n+1} is a polynomial, O(z−n−1)O(z^{-n-1}), vanishing at x1,…,xnx_1,\ldots,x_n; so V=0V=0 and Tn/Qn=Tn+1/Qn+1T_n/Q_n=T_{n+1}/Q_{n+1}. If Δn(Ξ)=0\Delta_n(\Xi)=0 for all n≥Nn\geq N, all Tn/QnT_n/Q_n equal one reduced fraction T/QT/Q, so Qn=UnQQ_n=U_nQ and Tn=UnTT_n=U_nT, and E:=CQ−T≠0E:=CQ-T\ne0 since CC is not rational. Then UnE=En=O(1/z)U_nE=E_n=O(1/z) forces E=c0z−q(1+o(1))E=c_0z^{-q}(1+o(1)) with c0≠0c_0\ne0, q≥1q\geq1, and deg⁡Un≤q−1\deg U_n\leq q-1. But E(xa)≠0E(x_a)\ne0 for all large aa, while En(xa)=0E_n(x_a)=0; so Un(xa)=0U_n(x_a)=0 for all large a≤na\leq n, too many zeros for a polynomial of degree <q<q once nn is large. □\square

Proof of the theorem

Suppose ζF(3)=u/v\zeta_F(3)=u/v with u∈Zu\in\mathbb{Z}, v∈Z>0v\in\mathbb{Z}_{>0}. Since 5=1+2t\sqrt{5}=1+2t, (1.1) gives Ξ=u(1+2t)/(25v)\Xi=u(1+2t)/(25v) and σΞ=−Ξ\sigma\Xi=-\Xi. For KK in the infinite set of Proposition 6.1 put yK=CK(t)(25v)KΔK(Ξ)y_K=C_K(t)(25v)^K\Delta_K(\Xi). By Proposition 4.4, yK∈Oy_K\in\mathcal{O}, and yK≠0y_K\ne0 because ΔK(Ξ)≠0\Delta_K(\Xi)\ne0 and CK(t)≠0C_K(t)\ne0 (tt is not a root of unity, so Φd(t)≠0\Phi_d(t)\ne0). Moreover σyK=CK(−φ)(25v)KσΔK(−Ξ)\sigma y_K=C_K(-\varphi)(25v)^K\sigma\Delta_K(-\Xi), so ∣N(yK)∣=∣yKσyK∣≥1|\mathrm{N}(y_K)|=|y_K\sigma y_K|\geq1, and by Propositions 3.1, 5.1 and 4.6

0≤log⁡∣N CK(t)∣+2Klog⁡(25v)+log⁡∣ΔK(Ξ)∣+log⁡∣σΔK(−Ξ)∣≤(c∗−29−1)K3log⁡φ+O(K2log⁡K).\begin{aligned} 0&\leq\log|\mathrm{N}\,C_K(t)|+2K\log(25v)+\log|\Delta_K(\Xi)|+\log|\sigma\Delta_K(-\Xi)|\\ &\leq\left(c_*-\frac{2}{9}-1\right)K^3\log\varphi+O(K^2\log K). \end{aligned}

Since c∗−119=118β7−89=−0.02684…<0c_*-\frac{11}{9}=\frac{11}{8}\beta_7-\frac{8}{9}=-0.02684\ldots<0, this fails for large KK. □\square

Remark 7.1. Not used in the proof. With the sharp constant of Remark 4.7 the margin is 119−1.130790≥0.0914\frac{11}{9}-1.130790\geq0.0914 per K3log⁡φK^3\log\varphi. At finite KK, computed with exact determinants and interval arithmetic, the sum log⁡∣N CK(t)∣+log⁡∣ΔK(Ξ)∣+log⁡∣σΔK(−Ξ)∣\log|\mathrm{N}\,C_K(t)|+\log|\Delta_K(\Xi)|+\log|\sigma\Delta_K(-\Xi)| (without the term 2Klog⁡(25v)2K\log(25v)) is already negative, about −0.12K3log⁡φ-0.12K^3\log\varphi for 8≤K≤268\leq K\leq26; and the local bounds of Section 4 agree with the exact multiplicities of the cyclotomic factors in Q(τ)[X]\mathbb{Q}(\tau)[X] for small KK. The scripts are available from the author.

Formal verification

The Lean development is the repository https://github.com/telaviv/fibonacci-zeta3-irrationality; all file names below refer to it. It (library FibZeta, Lean 4.35.0, Mathlib v4.35.0, about 8500 lines) proves

theorem FibZeta.irrational zetaF3 : Irrational FibZeta.zetaF3,

where zetaF3 is the series ∑n≥01/Fn+13\sum_{n\geq0}1/F_{n+1}^3, written with Mathlib’s Nat.fib and a tsum (FibZeta/Final.lean; hasSum_zetaF3 shows that the series converges to it). The proof uses only the axioms propext, Classical.choice and Quot.sound (Axioms.lean); there is no sorry, no native_decide and no added axiom.

What has to be trusted. For the theorem itself, only the definition of zetaF3 (and Mathlib’s Nat.fib, tsum and Irrational) has to be read: the kernel checks everything else. Since a divergent tsum has the value 0, which is rational, the statement cannot hold for a degenerate reason. To see that the formal proof follows the argument of this paper one also reads the definitions in FibZeta/Defs.lean and the intermediate statements in FibZeta/Statements.lean; an internal comparison of these with an earlier draft of this section is in the repository (FIDELITY_REFEREE.md).

Definitions

FibZeta/Defs.lean defines the construction of Section 2 over an arbitrary field T\mathbb{T} with a parameter τ∈T\tau\in\mathbb{T}: alpha, nu, psi, S, the poles node =w−a= w^{-a}, the monic polynomial F=∏a(s−xa)F = \prod_a(s-x_a) (so DK=(−1)KwμKAKFD_K = (-1)^K w^{\mu_K} A_K F), the residues res, the two coefficients mConst, mLin of LX(sk/F)L_X(s^k/F), the normalisation nrm, and the determinant Delta. Specialised at τ=t\tau= t (tR) and at τ=−φ\tau= -\varphi it gives ΔK\Delta_K and σΔK\sigma\Delta_K; at τ=RatFunc.X\tau= \mathrm{RatFunc.X} it gives ΔK,τ\Delta_{K,\tau}. The real functional is Lfun, with θ(θ+1)g=s2g′′+2sg′\theta(\theta+1)g = s^2g'' + 2sg' written with deriv, and XiR is Ξ\Xi. The clearing polynomial is CKpoly =∏d≤4KΦdbK,d= \prod_{d\leq4K}\Phi_d^{b_{K,d}}, with exponents nExp == nW ++ nAlpha: the W part (cluster costs cW, cW’ and cluster sizes clusterSize) and the α\alpha part. The root types are defined through the chosen root exp⁡(2πi/d)\exp(2\pi i/d) (zetaC, eW, IsAlphaRoot), and FibZeta/Roots.lean proves that they are the same for every primitive dd-th root.

The chain

FibZeta/Statements.lean states the ingredients as propositions, and FibZeta.irrational_zetaF3_of (in FibZeta/Main.lean) derives the theorem from them by the argument of Section 7. Instead of norms in O\mathcal{O} it uses that tt and −φ-\varphi are the two roots of τ2+τ−1\tau^2+\tau-1: for Q∈Z[X]Q \in\mathbb{Z}[X] the product Q(t)Q(−φ)Q(t)Q(-\varphi) is an integer, nonzero if Q(t)≠0Q(t) \ne0 (norm_ne_zero). The docstrings of the Lean files refer to the numbering of the author’s notes; the header of FibZeta/Statements.lean translates it to the numbering of this paper.

PaperLean statementLean proof
Lemma 2.2, entriesEntryIdentityFibZeta.Analytic.entryIdentity
Prop. 3.1DecayBoundFibZeta.Analytic.decayBound
Lemma 4.1–Prop. 4.4IntegralityFibZeta.Denom.integrality
Lemma 4.5, Prop. 4.6CostBoundFibZeta.Cost.costBound
Prop. 5.1ConjBoundFibZeta.Conj.conjBound
Prop. 6.1NonvanishingFibZeta.NonVan.nonvanishing
Section 7FibZeta.irrational_zetaF3_of

Table 1.

EntryIdentity is used only in the proof of DecayBound, not by the assembly. The formal statements are equivalent to, or slightly weaker than, the propositions, in the form the assembly needs. DecayBound is ∣ΔK(Ξ)∣≤CK2t2bK|\Delta_K(\Xi)| \leq C^{K^2}t^{2b_K} with bK=2∑j<Kajb_K = 2\sum_{j<K}a_j (proved with C=4096C = 4096). ConjBound is ∣σΔK(−Ξ)∣≤CK2tK2(K+1)|\sigma\Delta_K(-\Xi)| \leq C^{K^2}t^{K^2(K+1)}; its proof bounds the complete homogeneous sums hm(r)h_m(r) by φK\varphi^K instead of C0t2mC_0t^{2m}, which still gives entries of size ≤DK\leq D^K. CostBound is log⁡∣CK(t)CK(−φ)∣≤121100K3log⁡φ+CK2(1+log⁡K)\log|C_K(t)C_K(-\varphi)| \leq\frac{121}{100}K^3\log\varphi+ CK^2(1+\log K); its proof follows Proposition 4.6, with cubic constant c∗=43933675c_* = \frac{4393}{3675} and the constant 12 in place of 2E12\mathcal{E}_1 in the upper half of Lemma 4.5, and c∗≤121100<119c_* \leq\frac{121}{100} < \frac{11}{9}. Integrality is Proposition 4.4 for every real τ≠0\tau\ne0 that is not a root of unity, with one integer polynomial for all such τ\tau; this covers tt and −φ-\varphi at once. Its proof works in the local rings of C(τ)\mathbb{C}(\tau) at the roots of unity: the Newton cluster basis at WW-roots, the monomial basis elsewhere, and the rank correction at α\alpha-roots (FibZeta.Denom.local_bound_cluster, FibZeta.Denom.local_bound_monomial, FibZeta.Denom.integrality_of_local_bounds). The proof of Nonvanishing uses the Cauchy transform CC directly: its values at the interpolation nodes, C(xa)=2ta(Ξ−Sa)C(x_a) = 2t^a(\Xi-S_a), and its expansion C(1/u)=∑kνkuk+1C(1/u) = \sum_k \nu_k u^{k+1} at ∞\infty replace the functional LL in the interpolation identities.

Provenance

Parts of the formal development are adapted from the Lean formalisation QZeta3 [13] of [14] (Apache License 2.0): the localisation, specialisation and descent machinery, the weighted and rank determinant lemmas, the Cauchy-transform and normality arguments for nonvanishing, the decay estimates, the totient sieve and the cyclotomic size bounds. Every adapted file says so in its header, and the repository’s NOTICE file records it.

Use of AI tools

This work was carried out with substantial assistance from AI systems (Anthropic’s Claude models), which were used to search the literature, to develop and check the proof, to write the Lean formalisation and to draft this text. The author directed the project and is responsible for its content. The formal proof is checked by the Lean kernel.

References

  1. [1]R. André-Jeannin, Irrationalité de la somme des inverses de certaines suites récurrentes, C. R. Acad. Sci. Paris Sér. I Math. 308 (1989), 539–541.
  2. [2]D. Duverney, Ke. Nishioka, Ku. Nishioka, and I. Shiokawa, Transcendence of Rogers–Ramanujan continued fraction and reciprocal sums of Fibonacci numbers, Proc. Japan Acad. Ser. A 73 (1997), 140–142.DOI
  3. [3]C. Elsner, S. Shimomura, and I. Shiokawa, Algebraic relations for reciprocal sums of Fibonacci numbers, Acta Arith. 130 (2007), 37–60.DOI
  4. [4]————, Algebraic relations for reciprocal sums of odd terms in Fibonacci numbers, Ramanujan J. 17 (2008), 429–446.DOI
  5. [5]A. Faunzan, ζ(5) is irrational, unrefereed preprint, 17 September 2026, Zenodo record 22826419.
  6. [6]S. Fischler and W. Zudilin, A refinement of Nesterenko’s linear independence criterion with applications to zeta values, Math. Ann. 347 (2010).
  7. [7]F. Jouhet and E. Mosaki, Irrationalité aux entiers impairs positifs d’un q-analogue de la fonction zêta de Riemann, Int. J. Number Theory 6 (2010), 959–988.arxiv.org/abs/0712.1762
  8. [8]C. Krattenthaler, A determinant identity for moments of orthogonal polynomials that implies Uvarov’s formula for the orthogonal polynomials of rationally related densities, 2021, preprint, arXiv:2103.03969.arxiv.org/abs/2103.03969
  9. [9]C. Krattenthaler, T. Rivoal, and W. Zudilin, Séries hypergéométriques basiques, q-analogues des valeurs de la fonction zêta et séries d’Eisenstein, J. Inst. Math. Jussieu 5 (2006), 53–79.DOI
  10. [10]N. K. Meher and D. K. Sahoo, Analytic continuation of the ℓ-generalized Fibonacci zeta function, Integers 24 (2024), #A85.DOI
  11. [11]Yu. V. Nesterenko, Modular functions and transcendence questions, Sb. Math. 187 (1996), 1319–1348.DOI
  12. [12]M. Prévost, A new proof of the irrationality of ζ(2) and ζ(3) using Padé approximants, J. Comput. Appl. Math. 67 (1996), 219–235.
  13. [13]F. Solomon Admasu, QZeta3: Lean 4 formalisation of the irrationality of ζ₁/ₚ(3) (moving-denominator Hankel argument), version 0.1.0, 2026-10-11, Apache License 2.0; directory lean/ of [14].
  14. [14]uplift-universal-inc, Irrationality of the Lambert-series value ζ₁/ₚ(3), unrefereed AI-assisted research draft with Lean formalisation, version 0.3.0-two-proofs, 2026-10-12, https://github.com/uplift-universal-inc/qzeta_3-irrationality (commit d12d56f).
  15. [15]V. B. Uvarov, The connection between systems of polynomials that are orthogonal with respect to different distribution functions, U.S.S.R. Comput. Math. Math. Phys. 9 (1969), no. 6, 25–36.DOI
  16. [16]M. Waldschmidt, Recent Diophantine results on zeta values: a survey, RIMS Kôkyûroku 1710 (2010), 166–179.
  17. [17]W. Zudilin, On the irrationality of generalized q-logarithm, Res. Number Theory 2 (2016), article 15.DOI
  18. [18]———, A determinantal approach to irrationality, Constr. Approx. 45 (2017), 301–310.DOI

Paper details

vlnc:W_j9deaphg
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Anthropic Claude Opus 5.5

View statement
Looking for related papers…
Contents