Simon's stable-infimum conjecture: independent audit

What Lean checks

Complete proposed proof of the full parameterwise convex order, including the Brownian endpoint. The independent review checks the prior Mellin transform, increasing logarithmic Lévy measures, exponential integrability and conditional Jensen argument. No numerical or finite-case inference is used.

Audit date: 21 November 2026. Coordination: issue #175. Manuscript reviewed: PROOF.md, initially committed as f01ba454b97ea702b4eb3a764bd133ef4d2a953a.

Verdict and limits

The proof establishes the convex-order statement in Simon's Remark 6 for every pair 1/2≤β<α<11/2\leq\beta<\alpha<1. I found no mathematical gap. The boundary case β=1/2\beta=1/2, the moment normalization, the construction of the independent multiplier, and the convergence requirements of the Lévy integrals all check.

This is a mathematical audit, not a certification of priority. The factorization and Mellin transforms are established results and must be credited as such. The separate source search determines whether this proof qualifies as a new resolution.

1. Exact statement and normalization

I checked Remark 6 on printed page 13 of the final EJP article, Thomas Simon, Comparing Fréchet and positive stable laws, Electronic Journal of Probability 19 (2014), no. 16, DOI 10.1214/EJP.v19-3058, and the normalization preceding Theorem B. The same conjecture appears on printed pages 13–14 of arXiv:1310.1888v2.

For the spectrally negative strictly 1/α1/\alpha-stable process normalized by 𝔼eλXt=etλ1/α\mathbb E e^{\lambda X_t}=e^{t\lambda^{1/\alpha}}, put Iα=−inf0≤t≤1XtI_\alpha=-\inf_{0\le t\le1}X_t and Vα=Γ(1+α)IαV_\alpha=\Gamma(1+\alpha)I_\alpha. The requested ordering is exactly

Vβ≼cxVα,12≤β<α<1.V_\beta\preceq_{\mathrm{cx}}V_\alpha, \qquad \frac12\le\beta<\alpha<1.

The proof covers this entire range and has the correct direction. The corresponding ordering for the suprema runs in the other direction; it is not substituted for the infimum conjecture.

2. Established transform and independent algebra check

The factorization Iα=dMαYαI_\alpha\stackrel d=M_\alpha Y_\alpha with independent factors is recorded in the original remark. The density of YαY_\alpha there agrees with the density used in the manuscript.

The fractional moments are also explicitly present in Pierre Patie and Thomas Simon, Intertwining certain fractional derivatives, arXiv:1008.4713v1: Lemma 3.3 and Proposition 3.6. Their stability parameter lies in (1,2)(1,2) and must be replaced by 1/α1/\alpha. Their reflected process at time one has the required infimum law by the identity explained in their proof. This confirms the manuscript's transform from an independent primary source.

I also checked the integral calculation directly. After u=t1/αu=t^{1/\alpha}, the two exponents in the elementary quadratic-denominator integral are 1−α+αs1-\alpha+\alpha s and 1+αs1+\alpha s. Their simultaneous integrability interval is

1−1α<ℜs<1α.1-\frac1\alpha<\Re s<\frac1\alpha.

The sine addition formula gives

𝔼Yαs=αsin(πα)sin(πs)sin(παs)sin(πα(1−s)).\mathbb E Y_\alpha^s =\frac{\alpha\sin(\pi\alpha)\sin(\pi s)} {\sin(\pi\alpha s)\sin(\pi\alpha(1-s))}.

Both apparent singularities at s=0,1s=0,1 are removable. Multiplying by 𝔼Mαs=Γ(1+s)/Γ(1+αs)\mathbb E M_\alpha^s=\Gamma(1+s)/\Gamma(1+\alpha s) and applying the reflection and recurrence formulas gives

𝔼Iαs=Γ(1−αs)Γ(1+α−αs)Γ(1−α+αs)Γ(1+α)Γ(1−α)Γ(2−s).\mathbb E I_\alpha^s =\frac{\Gamma(1-\alpha s)\Gamma(1+\alpha-\alpha s) \Gamma(1-\alpha+\alpha s)} {\Gamma(1+\alpha)\Gamma(1-\alpha)\Gamma(2-s)}.

One way to check the cancellation is first to replace α/[Γ(1+αs)sin(παs)]\alpha/[\Gamma(1+\alpha s)\sin(\pi\alpha s)] by Γ(1−αs)/(πs)\Gamma(1-\alpha s)/(\pi s), and then replace Γ(1+s)sin(πs)/(πs)\Gamma(1+s)\sin(\pi s)/(\pi s) by 1/Γ(1−s)1/\Gamma(1-s). The remaining reflection identity and one recurrence yield the displayed formula.

At α=1/2\alpha=1/2, duplication reduces this expression to

2sΓ((1+s)/2)π=Γ(1+s)Γ(1+s/2),\frac{2^s\Gamma((1+s)/2)}{\sqrt\pi} =\frac{\Gamma(1+s)}{\Gamma(1+s/2)},

which is the half-Gaussian moment formula in this normalization. Thus the proof includes the endpoint rather than obtaining it from an unproved continuity assertion. Evaluation at s=1s=1 gives 𝔼Iα=1/Γ(1+α)\mathbb E I_\alpha=1/\Gamma(1+\alpha), hence 𝔼Vα=1\mathbb E V_\alpha=1.

3. Lévy representation

The logarithm of the gamma quotient is analytic on a strip containing the imaginary axis and [0,1][0,1]. Its factors have neither zeros nor poles in that strip. The logarithm with value zero at the origin is therefore well defined.

The integral identity used for each gamma term follows by differentiating twice and using the integral for the trigamma function; fixing the value and derivative at zero determines the linear term. It does not assume infinite divisibility in advance.

I checked the contributions after changing variables t=αut=\alpha u in the three gamma terms whose slopes have magnitude α\alpha:

Term in the cumulant Jump side Contribution to the kernel before division by tt
logΓ(1−αs)\log\Gamma(1-\alpha s) positive 1/(et/α−1)1/(e^{t/\alpha}-1)
logΓ(1+α−αs)\log\Gamma(1+\alpha-\alpha s) positive e−t/(et/α−1)e^{-t}/(e^{t/\alpha}-1)
−logΓ(2−s)-\log\Gamma(2-s) positive −e−2t/(1−e−t)-e^{-2t}/(1-e^{-t})
logΓ(1−α+αs)\log\Gamma(1-\alpha+\alpha s) negative et/(et/α−1)e^t/(e^{t/\alpha}-1)

The resulting kernels are exactly

kα(t)=1+e−tet/α−1−e−2t1−e−t,hα(t)=etet/α−1.k_\alpha(t)=\frac{1+e^{-t}}{e^{t/\alpha}-1} -\frac{e^{-2t}}{1-e^{-t}}, \qquad h_\alpha(t)=\frac{e^t}{e^{t/\alpha}-1}.

For every fixed t>0t>0, their parameter derivatives are strictly positive. Also k1/2(t)=0k_{1/2}(t)=0 by direct cancellation, so neither kernel has a hidden negative part in the required parameter interval.

Near zero, the kernels satisfy kα(t)=(2α−1)/t+O(1)k_\alpha(t)=(2\alpha-1)/t+O(1) and hα(t)=α/t+O(1)h_\alpha(t)=\alpha/t+O(1). Dividing by tt produces a measure integrable against t2t^2 near zero, as Lévy–Khintchine requires. The first absolute moment of the small jumps need not be finite, but the compensated integrands have second-order zeros and converge.

At infinity, the positive-jump kernel is O(e−t/α)O(e^{-t/\alpha}) and the negative-jump kernel is O(e−(1/α−1)t)O(e^{-(1/\alpha-1)t}). Both large-jump first moments are finite. The compensation by the linear term over the whole half-line is therefore valid. Since 1/α>11/\alpha>1, the positive jumps also have a finite exponential moment at parameter one. These bounds apply to each nonnegative difference of kernels as well.

4. Difference of cumulants and convex order

Let Kα(s)=log𝔼VαsK_\alpha(s)=\log\mathbb E V_\alpha^s. For β<α\beta<\alpha, both kernel differences are positive. Consequently Kα−KβK_\alpha-K_\beta is a valid Lévy–Khintchine exponent, with the difference of the drifts in the manuscript's common compensation convention.

The resulting real random variable Lβ,αL_{\beta,\alpha} has a finite exponential moment at one. The integral representation and the normalization already checked give

𝔼eLβ,α=exp(Kα(1)−Kβ(1))=1.\mathbb E e^{L_{\beta,\alpha}} =\exp\bigl(K_\alpha(1)-K_\beta(1)\bigr)=1.

Take Lβ,αL_{\beta,\alpha} independent of VβV_\beta and set Rβ,α=eLβ,αR_{\beta,\alpha}=e^{L_{\beta,\alpha}}. Comparing characteristic functions of logarithms proves

Vα=dVβRβ,α.V_\alpha\stackrel d=V_\beta R_{\beta,\alpha}.

This step uses the full transform on the imaginary axis, not an assertion that finitely many moments determine the law. The product is integrable and has conditional mean VβV_\beta. Conditional Jensen proves the stated convex order for every convex test function with defined expectations. Subtracting a supporting affine function gives the usual extended interpretation when the positive expectation is infinite.

5. Conclusion

The mathematical conclusion is valid for all parameter pairs in the source conjecture. The proof uses known distributional formulas followed by an explicit increasing-kernel argument. No numerical evidence, finite parameter reduction, or Lean verification is needed for this audit. Priority remains the responsibility of the separately documented literature search.