López Pouso–Segura's Bessel-weight conjectures

What Lean checks

Lean proves all three signs, strict monotonicity and exact limits conditionally on the stated finite mixture, phase ODE, endpoint, zero/weight asymptotics and two resolvent expansions. It also proves the reference kernel integral and the dominated-convergence and comparison steps linking those inputs to the limits. The Bessel identification, special-function inputs, reference Euler–Maclaurin remainder and full repository gate remain open.

Research draft, 2026-11-21. Proposed complete proofs of Conjectures 2, 5, and 8 in Ó. López Pouso and J. Segura, Uniform relations between the Gauss–Legendre nodes and weights, Journal of Inequalities and Applications 2025, 40, DOI 10.1186/s13660-025-03283-w. The three assertions are treated together, using one common argument. No conclusion about the remaining uniform Gauss–Legendre conjectures is asserted here.

1. Statements and notation

Let jkj_k be the kth positive zero of J0J_0, and write

λk=jk2,ak=J1(jk)−2,Sm=∑k=1mak,Tm=∑k=1mλkak.\lambda_k=j_k^2,\qquad a_k=J_1(j_k)^{-2},\qquad S_m=\sum_{k=1}^m a_k,\qquad T_m=\sum_{k=1}^m\lambda_k a_k.

The source sequences, with exactly its indexing, are

Ci=2Si+ai+1−λi+12(i≥0),Ei=2Si−λi+λi+14(i≥1),C_i=2S_i+a_{i+1}-\frac{\lambda_{i+1}}2\quad(i\ge0), \qquad E_i=2S_i-\frac{\lambda_i+\lambda_{i+1}}4\quad(i\ge1),

and

Ki=Si−Ti2Si(i≥1),S0=0.K_i=S_i-\frac{T_i}{2S_i}\quad(i\ge1),\qquad S_0=0.

We prove that CiC_i is positive and strictly increasing to π2/12\pi^2/12; EiE_i is negative and strictly decreasing to −π2/6-\pi^2/6; and KiK_i is positive and strictly increasing to π2/12\pi^2/12.

2. A convexity property of the inverse Bessel phase

Put

M(x)2=J0(x)2+Y0(x)2,H(y)=yM(y)2(y>0).M(x)^2=J_0(x)^2+Y_0(x)^2,\qquad H(y)=yM(\sqrt y)^2\quad(y>0).

Nicholson's integral, specialized to order zero and followed by the substitution v=2ysinhuv=2\sqrt y\sinh u, gives

H(y)=4yπ2∫0∞K0(v)y+v2/4dv.H(y)=\frac{4y}{\pi^2}\int_0^\infty \frac{K_0(v)}{\sqrt{y+v^2/4}}\,dv. \tag{1}

Here K0(v)>0K_0(v)>0 for v>0v>0, and all its nonnegative integer moments are finite. The elementary beta identity

1AB=1π∫01duu(1−u)[uA+(1−u)B](A,B>0)\frac1{\sqrt{AB}}= \frac1\pi\int_0^1\frac{du}{\sqrt{u(1-u)}[uA+(1-u)B]} \qquad(A,B>0) \tag{2}

therefore expresses H(y)2H(y)^2 as a positive mixture of the functions y2/(y+q)y^2/(y+q), where q>0q>0:

H(y)2=∫(0,∞)y2y+qdσ(q).H(y)^2=\int_{(0,\infty)}\frac{y^2}{y+q}\,d\sigma(q). \tag{3}

Explicitly, σ\sigma is the pushforward under q=(uv2+(1−u)w2)/4q=(uv^2+(1-u)w^2)/4 of the measure

16π5K0(v)K0(w)u(1−u)dudvdw(0<u<1,v,w>0).\frac{16}{\pi^5} \frac{K_0(v)K_0(w)}{\sqrt{u(1-u)}}\,du\,dv\,dw \quad (0\lt u<1,\ v,w>0).

It is a finite nonzero positive measure supported on (0,∞)(0,\infty). On compact subintervals of y>0y>0, differentiation under the integral is justified by the finite mass and elementary bounds on the rational integrand. Consequently

(H2)″(y)=∫(0,∞)2q2(y+q)3dσ(q)>0.(H^2)''(y)=\int_{(0,\infty)}\frac{2q^2}{(y+q)^3}\,d\sigma(q)>0. \tag{4}

Choose the continuous phase θ(x)=arg(J0(x)+iY0(x))\theta(x)=\arg(J_0(x)+iY_0(x)) with θ(0+)=−π/2\theta(0+)=-\pi/2. The Bessel Wronskian gives

θ′(x)=2πxM(x)2>0.\theta'(x)=\frac{2}{\pi xM(x)^2}>0.

Thus t=(θ(x)+π/2)/πt=(\theta(x)+\pi/2)/\pi increases from zero to infinity. Let x=x(t)x=x(t) be its inverse and set λ(t)=x(t)2\lambda(t)=x(t)^2. Consecutive zeros of J0J_0 give

λ(k)=λk,λ′(t)=π2H(λ(t)).\lambda(k)=\lambda_k,\qquad \lambda'(t)=\pi^2 H(\lambda(t)). \tag{5}

At x=jkx=j_k, the same Wronskian and J0′=−J1J_0'=-J_1 yield Y0(jk)=2/[πjkJ1(jk)]Y_0(j_k)=2/[\pi j_kJ_1(j_k)], hence

λ′(k)=4ak.\lambda'(k)=4a_k. \tag{6}

Differentiating (5) twice and applying (4) proves

λ‴(t)=π6H((H′)2+HH″)(λ(t))=π62H(λ(t))(H2)″(λ(t))>0.\lambda'''(t) =\pi^6H\bigl((H')^2+HH''\bigr)(\lambda(t)) =\frac{\pi^6}{2}H(\lambda(t))(H^2)''(\lambda(t))>0. \tag{7}

Finally, the small-argument expansions J0(x)=1+O(x2)J_0(x)=1+O(x^2) and Y0(x)=(2/π)logx+O(1)Y_0(x)=(2/\pi)\log x+O(1) imply H(y)=O(ylog2y)H(y)=O(y\log^2 y). It follows from (5) that λ(0+)=λ′(0+)=0\lambda(0+)=\lambda'(0+)=0. Therefore g(t)=λ′(t)g(t)=\lambda'(t) extends continuously to g(0)=0g(0)=0 and is strictly convex on [0,∞)[0,\infty). Endpoint convexity follows by taking limits from positive arguments; no differentiated small-argument remainder is needed.

3. Positivity and monotonicity

By (6), CiC_i is half a composite trapezoidal error for gg:

2Ci=∑k=1ig(k)+g(i+1)2−∫0i+1g(t)dt.2C_i= \sum_{k=1}^{i}g(k)+\frac{g(i+1)}2- \int_0^{i+1}g(t)\,dt. \tag{8}

Strict convexity makes every unit-interval trapezoidal error positive. Equation (8) gives Ci>0C_i>0, and adjoining one interval gives

2(Ci+1−Ci)=g(i+1)+g(i+2)2−∫i+1i+2g(t)dt>0.2(C_{i+1}-C_i)= \frac{g(i+1)+g(i+2)}2-\int_{i+1}^{i+2}g(t)\,dt>0. \tag{9}

For convenience put E0=−λ1/4<0E_0=-\lambda_1/4<0. For every integer i≥1i\ge1,

Ei−Ei−1=14(2g(i)−∫i−1i+1g(t)dt)<0,E_i-E_{i-1} =\frac14\left(2g(i)-\int_{i-1}^{i+1}g(t)\,dt\right)<0, \tag{10}

by the strict midpoint inequality for a convex function. Thus all EiE_i are negative and the sequence is strictly decreasing.

4. A convergent Bessel resolvent sum

We next establish the exact identity

∑k=1∞ak(λk+s2)2=1−I0(s)−24s2(s>0).\sum_{k=1}^\infty\frac{a_k}{(\lambda_k+s^2)^2} =\frac{1-I_0(s)^{-2}}{4s^2}\qquad(s>0). \tag{11}

This is a classical Fourier–Bessel resolvent identity; a derivation is included to fix normalization and avoid importing an unverified formula. Use the complete orthogonal Dirichlet Fourier–Bessel system ϕk(r)=J0(jkr)\phi_k(r)=J_0(j_kr) in L2((0,1),rdr)L^2((0,1),r\,dr), with

∫01rϕk(r)2dr=J1(jk)22.\int_0^1r\phi_k(r)^2\,dr=\frac{J_1(j_k)^2}{2}.

Define

Fs(r)=K0(sr)−K0(s)I0(s)I0(sr),F0(r)=−logr.F_s(r)=K_0(sr)-\frac{K_0(s)}{I_0(s)}I_0(sr),\qquad F_0(r)=-\log r.

Both belong to this Hilbert space and vanish at r=1r=1. For L=−(1/r)(rd/dr)′L=-(1/r)(r\,d/dr)', the equations Lϕk=λkϕkL\phi_k=\lambda_k\phi_k, LFs=−s2FsLF_s=-s^2F_s, and limr↓0rFs′(r)=−1\lim_{r\downarrow0}rF_s'(r)=-1 give, by integration by parts,

∫01rFs(r)ϕk(r)dr=1λk+s2.\int_0^1rF_s(r)\phi_k(r)\,dr=\frac1{\lambda_k+s^2}.

The coefficient of ϕk\phi_k in F0−FsF_0-F_s is therefore

2ak(1λk−1λk+s2).2a_k\left(\frac1{\lambda_k}-\frac1{\lambda_k+s^2}\right).

The standard large-zero estimates λk≍k2\lambda_k\asymp k^2 and ak≍ka_k\asymp k show that these coefficients are O(k−3)O(k^{-3}). Because |J0(x)|≤1|J_0(x)|\le1 for real xx, the series converges uniformly on [0,1][0,1]. Its Hilbert-space equality with the continuous function F0−FsF_0-F_s is consequently a pointwise equality, including at zero. Equivalently, DLMF 10.23.18–10.23.20 directly gives the interior pointwise equality: F0−FsF_0-F_s is bounded at zero and smooth on (0,1)(0,1), so it satisfies the required weighted integrability and local bounded variation. Uniform convergence then extends this equality to zero. Taking that limit gives

log(s/2)+γ+K0(s)I0(s)=2∑k=1∞ak(1λk−1λk+s2).\log(s/2)+\gamma+\frac{K_0(s)}{I_0(s)} =2\sum_{k=1}^\infty a_k \left(\frac1{\lambda_k}-\frac1{\lambda_k+s^2}\right). \tag{12}

Differentiate on compact subintervals of s>0s>0. The modified Bessel Wronskian says (K0/I0)′=−1/(sI02)(K_0/I_0)'=-1/(sI_0^2), while the series derivative is 4s∑ak/(λk+s2)24s\sum a_k/(\lambda_k+s^2)^2. This proves (11).

The elementary series lower bound I0(s)≥s2/4I_0(s)\ge s^2/4 implies, without needing an exponential asymptotic,

∑k=1∞ak(λk+s2)2=14s2+O(s−6)(s→∞).\sum_{k=1}^\infty\frac{a_k}{(\lambda_k+s^2)^2} =\frac1{4s^2}+O(s^{-6})\quad(s\to\infty). \tag{13}

5. The exact renormalized sum

Let βk=π(k−1/4)\beta_k=\pi(k-1/4). Standard Bessel and zero asymptotics give

λk=βk2+14+O(k−2),ak=π2βk+O(k−3).\lambda_k=\beta_k^2+\frac14+O(k^{-2}),\qquad a_k=\frac\pi2\beta_k+O(k^{-3}). \tag{14}

For clarity, the cancellation responsible for the second remainder is visible from

jk=βk+18βk+O(βk−3),M(x)2=2πx(1−18x2+O(x−4)),j_k=\beta_k+\frac1{8\beta_k}+O(\beta_k^{-3}),\qquad M(x)^2=\frac2{\pi x}\left(1-\frac1{8x^2}+O(x^{-4})\right),

and ak=π2jk2M(jk)2/4a_k=\pi^2j_k^2M(j_k)^2/4. Both leading reciprocal terms cancel. The expansion of M2M^2 also follows directly from (1) and the finite moments ∫K0(v)dv=∫v2K0(v)dv=π/2\int K_0(v)\,dv=\int v^2K_0(v)\,dv=\pi/2.

Set

ak0=π22(k−1/4),λk0=π2(k−1/4)2+14,δk=ak−ak0.a_k^0=\frac{\pi^2}{2}(k-1/4),\qquad \lambda_k^0=\pi^2(k-1/4)^2+\frac14,\qquad \delta_k=a_k-a_k^0.

Then D=∑k≥1δkD=\sum_{k\ge1}\delta_k is absolutely convergent. We claim

D=116−π2192.D=\frac1{16}-\frac{\pi^2}{192}. \tag{15}

Write R(s)R(s) for the sum in (11) and R0(s)=∑ak0/(λk0+s2)2R_0(s)=\sum a_k^0/(\lambda_k^0+s^2)^2. The numerator difference satisfies

s4∑k≥1δk(λk+s2)2→Ds^4\sum_{k\ge1}\frac{\delta_k}{(\lambda_k+s^2)^2}\longrightarrow D

by dominated convergence. The remaining denominator difference tends to zero after multiplication by s4s^4. Indeed, (14) and the mean-value theorem bound its absolute value by

Cs4∑k≥1k−1(s2+ck2)3=O((1+logs)/s2)=o(1),Cs^4\sum_{k\ge1}\frac{k^{-1}}{(s^2+ck^2)^3} =O((1+\log s)/s^2)=o(1), \tag{16}

where the estimate follows on splitting at k=⌊s⌋k=\lfloor s\rfloor. Thus

s4(R(s)−R0(s))→D.s^4(R(s)-R_0(s))\longrightarrow D. \tag{17}

To calculate R0R_0, put A=s2+1/4/πA=\sqrt{s^2+1/4}/\pi and h(u)=u/(1+u2)2h(u)=u/(1+u^2)^2. Then

R0(s)=12π2A3∑m=0∞h((m+3/4)/A).R_0(s)=\frac1{2\pi^2A^3} \sum_{m=0}^\infty h((m+3/4)/A).

The shifted Euler–Maclaurin formula applies since every derivative of hh needed here is integrable at infinity. With ∫0∞h=1/2\int_0^\infty h=1/2, h(0)=h″(0)=0h(0)=h''(0)=0, h′(0)=1h'(0)=1, it gives

∑m=0∞h((m+3/4)/A)=A2−B2(3/4)2A+O(A−3),B2(3/4)=−148.\sum_{m=0}^\infty h((m+3/4)/A) =\frac A2-\frac{B_2(3/4)}{2A}+O(A^{-3}), \qquad B_2(3/4)=-\frac1{48}.

The remainder is bounded by the usual bounded-periodic-Bernoulli remainder after four integrations by parts; the fourth derivative of hh is absolutely integrable. Hence

R0(s)=14(s2+1/4)+π2192(s2+1/4)2+O(s−6)=14s2+−1/16+π2/192s4+O(s−6).R_0(s)=\frac1{4(s^2+1/4)} +\frac{\pi^2}{192(s^2+1/4)^2}+O(s^{-6}) =\frac1{4s^2}+\frac{-1/16+\pi^2/192}{s^4}+O(s^{-6}). \tag{18}

Combining (13), (17), and (18) proves (15).

6. Limits of the two quadrature-error sequences

The identity

Cm−1=2Sm−am−λm/2C_{m-1}=2S_m-a_m-\lambda_m/2

and the explicit arithmetic sum

∑k=1mak0=π2m24+π2m8\sum_{k=1}^m a_k^0=\frac{\pi^2m^2}{4}+\frac{\pi^2m}{8}

give, using (14)–(15),

limm→∞Cm−1=2D+3π232−18=π212.\lim_{m\to\infty}C_{m-1} =2D+\frac{3\pi^2}{32}-\frac18=\frac{\pi^2}{12}. \tag{19}

Also

Ei−Ci−1=ai−λi+1−λi4→−π24,E_i-C_{i-1}=a_i-\frac{\lambda_{i+1}-\lambda_i}{4} \longrightarrow-\frac{\pi^2}{4}, \tag{20}

again by (14). Equations (19)–(20), together with Section 3, prove Conjectures 2 and 5 in their entirety.

7. The first-moment sequence is a weighted average

Rearrange the definition of Cj−1C_{j-1} as

λj=4Sj−1+2aj−2Cj−1.\lambda_j=4S_{j-1}+2a_j-2C_{j-1}.

Multiplication by aja_j and summation use ∑j(4ajSj−1+2aj2)=2Si2\sum_j(4a_jS_{j-1}+2a_j^2)=2S_i^2 to give

Ti=2Si2−2∑j=1iajCj−1,Ki=∑j=1iajCj−1Si.T_i=2S_i^2-2\sum_{j=1}^i a_j C_{j-1},\qquad K_i=\frac{\sum_{j=1}^i a_j C_{j-1}}{S_i}. \tag{21}

Thus KiK_i is a positive weighted average of the first ii terms of the strictly increasing sequence CiC_i. Adding the larger next term makes that average strictly larger. Since Si→∞S_i\to\infty by (14), the standard finite-prefix/tail argument for weighted averages and (19) give Ki→π2/12K_i\to\pi^2/12. This proves Conjecture 8 completely.

8. Additional consequence for the circle-theorem constants

The unnumbered sequence in source equation (10) is

ki=λi+1−4ai+12π2.k_i=\lambda_{i+1}-\frac{4a_{i+1}^2}{\pi^2}.

Its asserted monotonicity also follows from (3). The integral moments give σ((0,∞))=4/π2\sigma((0,\infty))=4/\pi^2 and ∫qdσ(q)=1/π2\int q\,d\sigma(q)=1/\pi^2. Therefore

y−π24H(y)2=π24∫yqy+qdσ(q),y-\frac{\pi^2}{4}H(y)^2 =\frac{\pi^2}{4}\int\frac{yq}{y+q}\,d\sigma(q), \tag{22}

which is positive, strictly increasing, and tends to 1/41/4. At y=λi+1y=\lambda_{i+1} this is exactly kik_i. This consequence supplies a Bessel assertion used in the source's discussion of its uniform circle conjecture. It does not by itself establish that separate finite-degree Gauss–Legendre conjecture.

Classical ingredients

All formula links below refer to the NIST Digital Library of Mathematical Functions. The proof specializes them at order zero.

  • 10.9.30: Nicholson's integral.
  • 10.5.2 and 10.6.3: the ordinary Bessel Wronskian and J0′=−J1J_0'=-J_1.
  • 10.18.1–10.18.3 and 10.18.8: phase normalization and derivative. Equation 10.18.18 also gives that the phase tends to infinity.
  • 10.32.9: a positive integral for K0K_0. Equation 10.43.19 gives its Mellin moments; the instances used here are ∫0∞K0(v)dv=∫0∞v2K0(v)dv=π/2\int_0^\infty K_0(v)\,dv=\int_0^\infty v^2K_0(v)\,dv=\pi/2.
  • 10.7.1–10.7.2 give the ordinary small-argument behavior. Equation 10.31.2 gives K0(z)=−log(z/2)−γ+O(z2|logz|)K_0(z)=-\log(z/2)-\gamma+O(z^2|\log z|), including the constant needed in (12).
  • 10.23.18–10.23.20: the Fourier–Bessel expansion and its convergence hypotheses. Equation 10.9.1 yields |J0(x)|≤1|J_0(x)|\le1 for real xx.
  • 10.28.2: the modified Bessel Wronskian. Equation 10.25.2 gives the power series of I0I_0 and its lower bound in (13).
  • 10.21.19: McMahon's zero expansion. Equation 10.18.17 is the modulus expansion used in (14), also derivable from (1).
  • 2.10.1: Euler–Maclaurin with a periodic Bernoulli remainder. Applied to f(x)=h((x+3/4)/A)f(x)=h((x+3/4)/A) and expanded at zero, it gives (18). The proof uses this only for h(u)=u/(1+u2)2h(u)=u/(1+u^2)^2.

All conclusions above are analytic statements for every index. Numerical evaluations are optional diagnostics and are not premises of the proof.