López Pouso–Segura's node and partial-moment conjectures

What Lean checks

Lean checks the two Legendre roots, standard weight formula, exact cubic quadrature, and the two failed inequalities; the degree-two polynomial and its derivative are connected to mathlib’s shifted Legendre polynomial. Strict source compilation, one-module kernel replay and endpoint axiom checks passed. Lean also verifies the explicit perturbation functions, derivatives, differential equations final coefficient algebra, and the deduction of the scaled limit and eventual violation from three separate factor approximations with vanishing second-order remainders. Controlled ODE and root asymptotics, the gamma-ratio expansion, those factor approximations and the full repository gate remain pending.

Complete proposed disproof, 21 November 2026. An independent mathematical review passed. The separate source audit records the prior searches and their explicit access limits. This work is unrefereed. The separate Bessel-limit assertions, Conjectures 2, 5, and 8, are not contradicted.

1. Exact source and scope

Óscar López Pouso and Javier Segura, Uniform relations between the Gauss–Legendre nodes and weights, Journal of Inequalities and Applications (2025), article 40, published full text, arXiv:2305.19128.

Let x1<⋯<xnx_1<\cdots<x_n and w1,…,wnw_1,\ldots,w_n be the ordinary Gauss–Legendre nodes and weights on [−1,1][-1,1], and let κ=n+1/2\kappa=n+1/2. The source defines

z‾0=−1,z‾i=−1+∑j=1iwj,zi=z‾i−1+z‾i2,αi=−2∑j=1ixjwj.\bar z_0=-1,\qquad \bar z_i=-1+\sum_{j=1}^i w_j,\qquad z_i=\frac{\bar z_{i-1}+\bar z_i}{2}, \qquad \alpha_i=-2\sum_{j=1}^i x_jw_j.

Published Conjecture 3 asserts, for the defined nonzero ratios,

0<xizi−1<π212κ2.0<\frac{x_i}{z_i}-1<\frac{\pi^2}{12\kappa^2}. \tag{1}

Published Conjecture 7 asserts, for the interior intermediate nodes,

0<αi1−z‾i2−1<π212κ2.0<\frac{\alpha_i}{1-\bar z_i^2}-1 <\frac{\pi^2}{12\kappa^2}. \tag{2}

The source also discusses asymptotic evidence and an optimal asymptotic constant. An asymptotic expansion with a remainder of unspecified sign does not imply either strict upper inequality. We disprove the displayed inequalities. In addition, we prove that (1) fails at the smallest positive node for every sufficiently large even degree. Thus the failure of (1) persists if its exact upper inequality is required only eventually in the degree.

2. Exact common counterexample

Take n=2n=2. Then

x1=−13,x2=13,w1=w2=1,κ=52,x_1=-\frac1{\sqrt3},\quad x_2=\frac1{\sqrt3}, \quad w_1=w_2=1,\quad \kappa=\frac52,

and consequently

z‾1=0,z2=12,α1=23.\bar z_1=0,\qquad z_2=\frac12,\qquad \alpha_1=\frac2{\sqrt3}.

Both left-hand expressions in the proposed upper inequalities are therefore

x2z2−1=α11−z‾12−1=23−1.\frac{x_2}{z_2}-1 =\frac{\alpha_1}{1-\bar z_1^2}-1 =\frac2{\sqrt3}-1.

The elementary inequalities 3<7/4\sqrt3<7/4 and π<22/7\pi<22/7 give the entirely exact certificate

23−1>17=5253675>4843675>π275=π212κ2.\frac2{\sqrt3}-1 >\frac17 =\frac{525}{3675} >\frac{484}{3675} >\frac{\pi^2}{75} =\frac{\pi^2}{12\kappa^2}. \tag{3}

For a self-contained check of the second elementary bound, polynomial division and integration give

0<∫01t4(1−t)41+t2dt=227−π.0<\int_0^1\frac{t^4(1-t)^4}{1+t^2}\,dt=\frac{22}{7}-\pi.

All quantities in (3) have nonzero denominators and the chosen index is within the source's intended interior range. This is not a counterexample arising from a central 0/00/0 ratio or an endpoint convention.

Equations (1) and (2), with their stated strict upper bounds, are false.

3. Failure at arbitrarily large degrees

Here we establish the stronger asymptotic statement

xm+1zm+1=1+π212κ2+π4120κ4+O(κ−6),n=2m,κ=2m+12.\frac{x_{m+1}}{z_{m+1}} =1+\frac{\pi^2}{12\kappa^2} +\frac{\pi^4}{120\kappa^4} +O(\kappa^{-6}), \qquad n=2m,\quad \kappa=2m+\frac12. \tag{4}

The positive fourth-order term proves failure of (1) for all sufficiently large even nn.

3.1 The central node and its exact weight

By symmetry, the total weight of the negative nodes is 1. Thus

z‾m=0,zm+1=wm+12.\bar z_m=0,\qquad z_{m+1}=\frac{w_{m+1}}2. \tag{5}

Write x=xm+1>0x=x_{m+1}>0 and w=wm+1w=w_{m+1}. For the usual normalization Pn(1)=1P_n(1)=1, the exact Gauss–Legendre weight formula is

w=2(1−x2)Pn′(x)2.w=\frac{2}{(1-x^2)P_n'(x)^2}. \tag{6}

Set t=arcsinxt=\arcsin x and

Y(t)=costPn(sint).Y(t)=\sqrt{\cos t}\,P_n(\sin t).

The Legendre differential equation gives

Y″(t)+(κ2+14cos2t)Y(t)=0.Y''(t)+\left(\kappa^2+\frac{1}{4\cos^2t}\right)Y(t)=0. \tag{7}

At a zero tt, one has

Y′(t)=cos3/2tPn′(x),2xw=tantY′(t)2.Y'(t)=\cos^{3/2}t\,P_n'(x), \qquad \frac{2x}{w}=\tan t\,Y'(t)^2. \tag{8}

3.2 A regular perturbation at the centre

Put ε=κ−2\varepsilon=\kappa^{-2} and u=κtu=\kappa t. For even nn define

fε(u)=cos(u/κ)Pn(sin(u/κ))Pn(0).f_\varepsilon(u) =\frac{\sqrt{\cos(u/\kappa)}\,P_n(\sin(u/\kappa))} {P_n(0)}.

Then

fε″(u)+(1+ε4+ε2u24+O(ε3))fε(u)=0,fε(0)=1,fε′(0)=0.f_\varepsilon''(u)+ \left(1+\frac{\varepsilon}{4} +\frac{\varepsilon^2u^2}{4} +O(\varepsilon^3)\right)f_\varepsilon(u)=0, \qquad f_\varepsilon(0)=1,\quad f_\varepsilon'(0)=0. \tag{9}

The error is uniform with all required derivatives on any fixed bounded uu interval. More explicitly, the coefficient before expansion is 1+(ε/4)sec2(εu)1+(\varepsilon/4)\sec^2(\sqrt\varepsilon u); it is analytic in ε\varepsilon near zero on such an interval. Standard analytic dependence for the initial-value problem, or the corresponding Volterra integral equation, therefore justifies the expansion and its uniform remainder below.

Write

fε(u)=cosu+εf1(u)+ε2f2(u)+O(ε3).f_\varepsilon(u)=\cos u+\varepsilon f_1(u) +\varepsilon^2f_2(u)+O(\varepsilon^3).

Coefficient comparison in (9) gives

f1(u)=−usinu8,f2(u)=∫0usin(u−v)(vsinv32−v2cosv4)dv.f_1(u)=-\frac{u\sin u}{8}, \qquad f_2(u)=\int_0^u\sin(u-v) \left(\frac{v\sin v}{32}-\frac{v^2\cos v}{4}\right)\,dv. \tag{10}

Equivalently, the second coefficient has the explicit form

f2(u)=9128(usinu−u2cosu)−u3sinu24.f_2(u)=\frac9{128}(u\sin u-u^2\cos u)-\frac{u^3\sin u}{24}.

Let a=π/2a=\pi/2. Elementary integration yields

f1(a)=−a8,f1′(a)=−18,f1″(a)=a8,f_1(a)=-\frac a8,\qquad f_1'(a)=-\frac18, \qquad f_1''(a)=\frac a8,

f2(a)=9a128−a324,f2′(a)=9128−7a2128.f_2(a)=\frac{9a}{128}-\frac{a^3}{24}, \qquad f_2'(a)=\frac9{128}-\frac{7a^2}{128}. \tag{11}

Since cosu\cos u has a simple first positive zero at aa, the implicit-function theorem gives a simple zero uεu_\varepsilon near aa. Uniform convergence before that zero shows that it is the first positive zero. Substituting (10)–(11) gives

uε=a−a8ε+(11a128−a324)ε2+O(ε3),u_\varepsilon =a-\frac a8\varepsilon+ \left(\frac{11a}{128}-\frac{a^3}{24}\right)\varepsilon^2 +O(\varepsilon^3), \tag{12}

and

fε′(uε)2=1+ε4+a2−18ε2+O(ε3).f_\varepsilon'(u_\varepsilon)^2 =1+\frac{\varepsilon}{4} +\frac{a^2-1}{8}\varepsilon^2+O(\varepsilon^3). \tag{13}

3.3 The central normalization

The exact central value is

P2m(0)=(−1)m(2mm)4m=(−1)mΓ(m+1/2)πΓ(m+1).P_{2m}(0)=(-1)^m\frac{\binom{2m}{m}}{4^m} =(-1)^m\frac{\Gamma(m+1/2)} {\sqrt\pi\,\Gamma(m+1)}.

The classical gamma-ratio expansion (DLMF 5.11.8 and 5.11.13), with m+1/2=κ/2+1/4m+1/2=\kappa/2+1/4 and m+1=κ/2+3/4m+1=\kappa/2+3/4, gives

P2m(0)2=2πκ(1−ε8+11ε2128+O(ε3)).P_{2m}(0)^2 =\frac{2}{\pi\kappa} \left(1-\frac{\varepsilon}{8} +\frac{11\varepsilon^2}{128}+O(\varepsilon^3)\right). \tag{14}

For clarity, the logarithm of the parenthesized factor is −ε/8+5ε2/64+O(ε3)-\varepsilon/8+5\varepsilon^2/64+O(\varepsilon^3), so its exponential gives exactly the coefficient 11/12811/128 in (14). These expansions have ordinary remainder estimates along the positive real axis.

3.4 The resulting ratio

Equations (5), (8), and the definition of fεf_\varepsilon imply

xm+1zm+1=2xw=tan(uε/κ)κ2P2m(0)2fε′(uε)2.\frac{x_{m+1}}{z_{m+1}} =\frac{2x}{w} =\tan(u_\varepsilon/\kappa)\, \kappa^2P_{2m}(0)^2 f_\varepsilon'(u_\varepsilon)^2. \tag{15}

Multiplying (13) and (14), the relevant factor is

1+ε8+(a28−9128)ε2+O(ε3).1+\frac{\varepsilon}{8} +\left(\frac{a^2}{8}-\frac9{128}\right)\varepsilon^2 +O(\varepsilon^3).

Also,

κtan(uε/κ)=a+(−a8+a33)ε+(11a128−a324−a38+2a515)ε2+O(ε3).\kappa\tan(u_\varepsilon/\kappa) =a+\left(-\frac a8+\frac{a^3}{3}\right)\varepsilon +\left(\frac{11a}{128}-\frac{a^3}{24} -\frac{a^3}{8}+\frac{2a^5}{15}\right)\varepsilon^2 +O(\varepsilon^3).

Using 2a/π=12a/\pi=1, the constant and quadratic-in-aa fourth-order correction terms cancel. Equation (15) becomes

xm+1zm+1=1+a23ε+2a415ε2+O(ε3),\frac{x_{m+1}}{z_{m+1}} =1+\frac{a^2}{3}\varepsilon +\frac{2a^4}{15}\varepsilon^2 +O(\varepsilon^3),

which is exactly (4).

In particular,

κ4(xm+1zm+1−1−π212κ2)→π4120>0.\kappa^4\left( \frac{x_{m+1}}{z_{m+1}}-1-\frac{\pi^2}{12\kappa^2} \right)\longrightarrow\frac{\pi^4}{120}>0.

The exact upper bound in Conjecture 3 therefore fails for every sufficiently large even degree, at a specifically identified node.

4. What this does and does not settle

The exact inequalities in the published Conjectures 3 and 7 are disproved by (3). For Conjecture 3, (4) also rules out an eventual version of its exact upper inequality. We do not use a finite computation as evidence for an unrestricted theorem: the negative answer is the exact certificate (3), and the infinite family is justified by the differential equation and its controlled asymptotic expansion.

The leading asymptotic constant π2/12\pi^2/12 remains compatible with (4). Neither an O(κ−2)O(\kappa^{-2}) estimate nor an expansion with an O(κ−4)O(\kappa^{-4}) remainder is disproved. The source's Bessel-limit Conjectures 2, 5, and 8 concern different limits and can all be true while the strict finite-degree upper bounds fail.

The counterexample is one joint obstruction to the two related assertions, not two unrelated named results. The source audit records the checked prior literature and the limits of the negative search.

5. Existing asymptotic ingredients

The node and weight asymptotics themselves are classical inputs. In particular, Gil, Segura and Temme, Fast and accurate computation of classical Gaussian quadratures, arXiv:2509.16716v2, equation (2.9), printedp5, display the fourth- and sixth-order expansions, crediting the corresponding Riemann–Hilbert calculations of Opsomer and Huybrechs. At the smallest positive even-degree node, their auxiliary cotangent is O(κ−1)O(\kappa^{-1}), so the two displayed normalization brackets agree through order κ−4\kappa^{-4}. Their quotient and the elementary tangent expansion therefore also imply (4).

The source article's first terms and the later general node/weight expansions are credited as prior mathematics. The proposed contribution here is the explicit contradiction to the two published strict upper inequalities, with an elementary exact witness and an all-large-even-degree obstruction. No earlier explicit refutation was found in the sources inspected; the separate literature audit states the search coverage and access limits. The differential-equation derivation above makes the particular sign and remainder needed for that contradiction independently checkable.