ProbXiv
Explorersign in
Problem archiveProblem record

Statement

We know that the following sub-modularity property of the entropy is true: if X,Y,X, Y, and ZZ are independent, real-valued random variables, then

h(X+Y+Z)+h(Z)≤h(X+Z)+h(Y+Z).h(X+Y+Z) + h(Z) \leq h(X+Z) + h(Y+Z).

We consider a conjecture relating to quantifying the gap (RHS - LHS) of the above inequality. The conjecture is the following: If XX, YY, and ZZ are i.i.d. with a given finite variance, is the gap minimized by Gaussians?

This would have an implication for the entropic CLT: it would imply that if SnS_n is the normalized sum that appears in the CLT, then

2D(S2)≤D(S1)+D(S3)2 D(S_2)\leq D(S_1) + D(S_3)

which suggests that the sequence D(Sn)D(S_n) in the entropic CLT is not just decreasing (as shown by ABBN) but also convex, where D(⋅)=D(⋅∣∣q)D(\cdot) = D( \cdot || q), i.e.\ it is the relative entropy to the mean- and variance-matched Gaussian.

Sergey Bobkov suggested the following follow-up question. If X∈RX \in \mathbb{R} and Y∈RY \in \mathbb{R} are random variables with Var(X)=Var(Y)=1Var(X) = Var(Y) = 1, then D(X+Y)≤12D(X)+12D(Y)D(X+Y) \leq \frac{1}{2} D(X) + \frac{1}{2} D(Y). Moreover, D(X+Y)=0D(X+Y) = 0 implies D(X)=D(Y)=0D(X) = D(Y) = 0. This is a special case of Cramer (1937). The question is, how do we prove Cramer (1937) using information-theoretic techniques?

Record

Comments

Nothing has been published against this problem yet, and nothing has been said about it.

Sign in with an institutional address to take part in the discussion. Reading every thread stays open to everyone.

Sign in

Solve with an agent

Open the statement in a chat, with the problem and the ground rules already written into the prompt.

This opens a third-party site. Nothing is posted back to ProbXiv and nothing you write there is recorded here — what a model gives you is an attempt, which a person still has to check.