Centered gamma moment-generating-function conjecture for empirical relative entropy

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. Let tt be a real parameter satisfying 0t<n0\leq t<n.

Centered gamma moment-generating-function conjecture. For every nn, kk, and PP,

logE[exp(t(Vn,k,PE[Vn,k,P]))](k1)log(exp(t/n)1t/n).\log\mathbb{E}\left[\exp\left(t\left(V_{n,k,P}-\mathbb{E}[V_{n,k,P}]\right)\right)\right] \leq (k-1)\log\left(\frac{\exp(-t/n)}{1-t/n}\right).

The right-hand side is the centered log moment-generating function of a gamma distribution with shape k1k-1 and rate nn. The conjecture would imply a one-sided form of the preceding concentration conjecture; it was attributed in the source to Agrawal and remains unresolved here.

Sources & referencesView supporting material

Primary source

Rohit Agrawal, “Finite-sample concentration of the empirical relative entropy around its mean”, arXiv:2203.00800 (2022).

Progress summary

Never refreshed

Nothing recorded yet. Refresh searches the literature and the public web for attempts on this problem, and writes the first summary here.

Solutions 0

No solutions have been posted yet.