Ferreira–Simon's diagonal Mittag–Leffler conjecture

What Lean checks

Complete proposed proof of strict increase throughout the full unit parameter interval. The independent audit checks the Pollard integral, classical sine quantile, derivative signs, reciprocal pairing and strictness. No numerical or finite-case argument is used.

Date: 21 November 2026. Status: complete proposed proof, independently checked within the research campaign. A separate source audit records the prior-resolution search and its limits. The note is unrefereed.

Statement

For positive parameters, define

Eα,β(z)=∑n=0∞znΓ(αn+β),Eα=Eα,1.E_{\alpha,\beta}(z)=\sum_{n=0}^{\infty}\frac{z^n}{\Gamma(\alpha n+\beta)}, \qquad E_\alpha=E_{\alpha,1}.

Ferreira and Simon, Mittag-Leffler functions and convex ordering, arXiv:2512.04940v1, Remark 5(b), printed page 14, conjecture that the diagonal value at minus one increases with the parameter throughout the unit interval.

Theorem. For every pair of real parameters satisfying 0<α<β<10<\alpha<\beta<1, one has

Eα,α(−1)<Eβ,β(−1).E_{\alpha,\alpha}(-1)<E_{\beta,\beta}(-1).

This is the entire parameter-monotonicity conjecture in that remark. The fixed argument minus one is part of the original statement; the proof covers every allowed real parameter and does not infer a theorem from a numerical case.

1. A bounded integral for the diagonal value

We use the classical Pollard representation, stated as equation (8) in the source:

Eα(−xα)=∫0∞e−xrgα(r)dr,gα(r)=sin(πα)πrα−1r2α+2cos(πα)rα+1,x≥0.E_\alpha(-x^\alpha)=\int_0^\infty e^{-xr}g_\alpha(r)\,dr, \qquad g_\alpha(r)= \frac{\sin(\pi\alpha)}{\pi} \frac{r^{\alpha-1}}{r^{2\alpha}+2\cos(\pi\alpha)r^\alpha+1}, \quad x\ge0.

Here 0<α<10<\alpha<1 and gαg_\alpha is a probability density. Its normalization also follows from the change of variables in the next section.

Differentiating the defining entire series gives

αEα′(z)=Eα,α(z).\alpha E_\alpha'(z)=E_{\alpha,\alpha}(z).

Differentiating the integral at x=1x=1 is justified by domination: for 1/2≤x≤3/21/2\le x\le3/2, the absolute derivative is at most re−r/2gα(r)r e^{-r/2}g_\alpha(r), and the first factor is bounded. Consequently,

Eα,α(−1)=∫0∞re−rgα(r)dr.E_{\alpha,\alpha}(-1) =\int_0^\infty r e^{-r}g_\alpha(r)\,dr. \tag{1}

The factors of alpha cancel in the chain rule. In particular, there is no extra factor of alpha in (1).

2. A common quantile parametrization

For 0<u<10<u<1, put

Qα(u)=[sin(παu)sin(πα(1−u))]1/α.Q_\alpha(u)= \left[ \frac{\sin(\pi\alpha u)} {\sin(\pi\alpha(1-u))} \right]^{1/\alpha}. \tag{2}

This classical quantile is given by Lancelot F. James, Lamperti-type laws, Proposition 2.3(ii), equation (2.13). It is a strictly increasing map of the unit interval onto the positive half-line. We verify its change-of-variables formula directly.

Write a=παa=\pi\alpha, θ=au\theta=au, and y=rαy=r^\alpha. Then

y=sinθsin(a−θ),dydθ=sinasin2(a−θ)>0,y=\frac{\sin\theta}{\sin(a-\theta)}, \qquad \frac{dy}{d\theta}=\frac{\sin a}{\sin^2(a-\theta)}>0,

and the elementary sine identity

y2+2(cosa)y+1=sin2asin2(a−θ)y^2+2(\cos a)y+1 =\frac{\sin^2a}{\sin^2(a-\theta)}

yields

gα(r)dr=sinaπαdyy2+2(cosa)y+1=dθa=du.g_\alpha(r)\,dr =\frac{\sin a}{\pi\alpha}\, \frac{dy}{y^2+2(\cos a)y+1} =\frac{d\theta}{a}=du. \tag{3}

The endpoint values of yy are zero and infinity. Thus (1) becomes

Eα,α(−1)=∫01Qα(u)e−Qα(u)du.E_{\alpha,\alpha}(-1) =\int_0^1 Q_\alpha(u)e^{-Q_\alpha(u)}\,du. \tag{4}

The identity Qα(1−u)=Qα(u)−1Q_\alpha(1-u)=Q_\alpha(u)^{-1} will make the integrand monotone after pairing complementary points.

3. The logarithmic quantile contracts

Fix 1/2<u<11/2<u<1 and define

sα(u)=logQα(u)=logsin(παu)−logsin(πα(1−u))α.s_\alpha(u)=\log Q_\alpha(u) =\frac{\log\sin(\pi\alpha u) -\log\sin(\pi\alpha(1-u))}{\alpha}.

It is strictly positive, because

sin(παu)−sin(πα(1−u))=2cos(πα/2)sin(πα(u−1/2))>0.\sin(\pi\alpha u)-\sin(\pi\alpha(1-u)) =2\cos(\pi\alpha/2)\sin(\pi\alpha(u-1/2))>0.

Consider

h(z)=zcotz−logsinz,0<z<π.h(z)=z\cot z-\log\sin z,\qquad 0<z<\pi.

Its derivative is

h′(z)=−zcsc2z<0.h'(z)=-z\csc^2z<0.

Therefore

∂sα(u)∂α=h(παu)−h(πα(1−u))α2<0.\frac{\partial s_\alpha(u)}{\partial\alpha} = \frac{h(\pi\alpha u)-h(\pi\alpha(1-u))}{\alpha^2}<0. \tag{5}

Both arguments of hh lie in its stated interval, and the first is larger. Thus the positive logarithmic quantile strictly decreases with alpha. By reciprocal symmetry, its absolute value decreases on the other half of the unit interval as well.

4. Pairing the integral

For s≥0s\ge0, set

f(s)=12[es−es+e−s−e−s].f(s)=\frac12\left[e^{s-e^s}+e^{-s-e^{-s}}\right].

For every s>0s>0,

2f′(s)=es−es(1−es)−e−s−e−s(1−e−s)<0.2f'(s) =e^{s-e^s}(1-e^s) -e^{-s-e^{-s}}(1-e^{-s})<0. \tag{6}

Both terms on the last line are strictly negative. Pairing uu with 1−u1-u in (4) gives

Eα,α(−1)=2∫1/21f(sα(u))du.E_{\alpha,\alpha}(-1) =2\int_{1/2}^1 f(s_\alpha(u))\,du. \tag{7}

If 0<α<β<10<\alpha<\beta<1, equation (5) gives sβ(u)<sα(u)s_\beta(u)<s_\alpha(u) for every u∈(1/2,1)u\in(1/2,1). Equation (6) reverses this inequality for ff. Integrating the resulting strict pointwise inequality in (7) proves the theorem. All integrands are bounded by 1/e1/e, so no convergence issue is involved.

5. Endpoints and the source's zero bound

For each fixed u∈(1/2,1)u\in(1/2,1), the logarithmic quantile tends to infinity as alpha decreases to zero, since the sine ratio tends to u/(1−u)>1u/(1-u)>1. It tends to zero as alpha increases to one, by the identity sin(πu)=sin(π(1−u))\sin(\pi u)=\sin(\pi(1-u)). Dominated convergence in (7) therefore gives

limα↓0Eα,α(−1)=0,limα↑1Eα,α(−1)=e−1.\lim_{\alpha\downarrow0}E_{\alpha,\alpha}(-1)=0, \qquad \lim_{\alpha\uparrow1}E_{\alpha,\alpha}(-1)=e^{-1}.

The source's Proposition 1 says that

xβ−1Eβ,β(−xβ)−xα−1Eα,α(−xα)x^{\beta-1}E_{\beta,\beta}(-x^\beta) -x^{\alpha-1}E_{\alpha,\alpha}(-x^\alpha)

is positive precisely between its two positive zeros, denoted there by yα,βy_{\alpha,\beta} and zα,βz_{\alpha,\beta}. The theorem makes this expression positive at one, so zα,β>1z_{\alpha,\beta}>1. Together with the already proved Theorem B, this gives the complete bound proposed in Remark 5(b):

xα,β<1<zα,β.x_{\alpha,\beta}<1<z_{\alpha,\beta}.

This consequence is not counted as a separate conjecture.

Attribution and scope

The Pollard density and the sine quantile are classical. The proof includes the exact substitution and both derivative-sign arguments, so the new application can be checked without a numerical package. It proves the full real-parameter statement, not a finite sample of alpha values.

This assertion is distinct from the mean-normalized convex ordering proved in the source's Theorem C and from its Remark 7(b) stochastic-order conjecture. The latter has a separate Poisson-series proof in this research campaign. The present proof does not depend on it.

The source audit records searches for an earlier resolution of this exact assertion and the limits of those searches. No Lean work is performed.