ProbXiv
Explorersign in
Problem archiveProblem record

Statement

Let Vn,k,PV_{n,k,P} denote the empirical relative entropy statistic for a multinomial distribution with sample size nn, alphabet size kk, and distribution PP. For θ≥0\theta\geq 0, write

θ=ε.\theta=\varepsilon.

Concentration conjecture. There are positive constants c1c_1 and c2c_2 such that for every nn, kk, PP, and ε≥0\varepsilon\geq 0,

Pr⁡{∣Vn,k,P−E[Vn,k,P]∣≥ε}≤c1exp⁡{−c2min⁡{n2ε2k−1,nε}}.\Pr\left\{\left|V_{n,k,P}-\mathbb{E}[V_{n,k,P}]\right|\geq\varepsilon\right\}\leq c_1\exp\left\{-c_2\min\left\{\frac{n^2\varepsilon^2}{k-1},n\varepsilon\right\}\right\}.

This conjecture would give finite-sample two-sided concentration of empirical relative entropy around its mean, improving bounds that focus only on deviations above zero. It was posed in the cited work and remains unresolved here.

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.