ProbXiv
Explorersign in
Problem archiveProblem record

Statement

For a binomial experiment with sample size nn, success probability pp, empirical proportion B/nB/n, and moment-generating-function parameter x∈[0,1)x\in[0,1), define the binary Kullback–Leibler divergence by

D((u,1−u)∥(v,1−v))=ulog⁡uv+(1−u)log⁡1−u1−v.D\bigl((u,1-u)\mathbin\Vert(v,1-v)\bigr)=u\log\frac{u}{v}+(1-u)\log\frac{1-u}{1-v}.

The improved binomial MGF conjecture. The function

f(x)=21−x−1f(x)=\frac{2}{\sqrt{1-x}}-1

is a sample-independent upper bound for the MGF of the binomial KL divergence; equivalently, for every positive integer nn, p∈[0,1]p\in[0,1], and x∈[0,1)x\in[0,1),

E⁡[exp⁡(xnD((Bn,1−Bn)∥(p,1−p)))]≤21−x−1,\operatorname*{E}\left[\exp\left(xnD\left(\left(\frac{B}{n},1-\frac{B}{n}\right)\mathbin\Vert(p,1-p)\right)\right)\right]\leq \frac{2}{\sqrt{1-x}}-1,

where B∼Binomial⁡(n,p)B\sim\operatorname{Binomial}(n,p). This would improve the existing bound 1/(1−x)1/(1-x) and remain compatible with the asymptotic bound 1/1−x1/\sqrt{1-x}, which is not valid uniformly for all finite samples.

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.