← All problems

A subunit Rényi threshold for exponential-maximal iid entropy increments

For a probability density ff on R\mathbb R and 0<p<10<p<1, define

hp(f)=log⁡∫Rf(x)p dx1−p,(f∗f)(s)=∫Rf(t)f(s−t) dt,Δp(f)=hp(f∗f)−hp(f),h_p(f)=\frac{\log\int_{\mathbb R}f(x)^p\,dx}{1-p},\qquad (f*f)(s)=\int_{\mathbb R}f(t)f(s-t)\,dt,\qquad \Delta_p(f)=h_p(f*f)-h_p(f),

whenever the two power integrals are finite and strictly positive. Thus Δp(f)\Delta_p(f) is the Rényi entropy increment of the sum of two independent random variables with density ff. Let e(x)=e−x1[0,∞)(x)e(x)=e^{-x}\mathbf1_{[0,\infty)}(x).

Subunit threshold question. Does there exist p0∈(0,1)p_0\in(0,1) such that for every p0<p<1p_0<p<1 and every log-concave probability density ff with the required finite positive moments,

Δp(f)≤Δp(e)?\Delta_p(f)\le\Delta_p(e)?

This is the one-dimensional necessary consequence of the proposed exponential-maximizer branch in the unnumbered paragraph after (1.2), p. 2, of Fu–Li. It is distinct from the numbered Shannon conjecture (p=1p=1), and does not ask about the proposed cube-uniform branch below the threshold.

References

Fu–Li, A reverse entropy power inequality for iid log-concave random variables, arXiv:2510.09206v3, p. 2, unnumbered paragraph after (1.2). https://arxiv.org/abs/2510.09206v3

Progress summary

Refreshed
Open · progress reported

A reader-submitted argument claims to disprove the proposed threshold for every subunit exponent, but no independent verification was found.

Fu–Li proposed, but did not prove, that exponential variables maximize the Rényi entropy increment above some threshold p0∈(0,1)p_0\in(0,1); the question is distinct from their Shannon conjecture.

Known results

Fu–Li prove only a one-dimensional min-entropy comparison, while the Shannon inequality and the subunit threshold remain conjectural.

Community submission (unverified), November 7, 2026

A submitted argument claims that, for every 0<p<10<p<1, a finite truncated exponential has strictly larger Δp\Delta_p than the exponential, which would rule out every threshold p0∈(0,1)p_0\in(0,1). The claimed proof has no independent mathematical assessment in the retrieved sources.

Current status (as of November 2026): the threshold question remains open; a community submission claims a counterexample for every 0<p<10<p<1, but that claim is unverified.

Sources

Solutions 1

For every real 0<p<1 there exists a finite normalized truncated exponential whose iid positive-sum Rényi entropy increment strictly exceeds the exponential increment. Consequently no threshold p0 in (0,1) has the proposed exponential-maximizer property above p0.See full solutionHide full solution

Result. For every real 0<p<1 there exists a finite normalized truncated exponential whose iid positive-sum Rényi entropy increment strictly exceeds the exponential increment. Consequently no threshold p0 in (0,1) has the proposed exponential-maximizer property above p0.

Paper and complete proof. Failure of the exponential Rényi-increment threshold below one. The paper page provides the proof PDF and its accompanying source and verification package.

Scope and source. Fu–Li, A reverse entropy power inequality for iid log-concave random variables, arXiv:2510.09206v3, printed p2 unnumbered paragraph after (1.2). The result addresses exactly the question stated above; adjacent conjectures and broader formulations are not claimed.

Provenance. AI-assisted research. Formalization and verification claims are only those explicitly established in the paper and its attached artifacts; this entry does not assert MathDB endorsement or independent acceptance.