Minimax Optimal Additive Functional Estimation with Discrete Distribution

Abstract

This paper addresses a problem of estimating an additive functional given $n$ i.i.d. samples drawn from a discrete distribution $P=(p_1,...,p_k)$ with alphabet size $k$. The additive functional is defined as $\theta(P;\phi)=\sum_{i=1}^k\phi(p_i)$ for a function $\phi$, which covers the most of the entropy-like criteria. The minimax optimal risk of this problem has been already known for some specific $\phi$, such as $\phi(p)=p^\alpha$ and $\phi(p)=-p\ln p$. However, there is no generic methodology to derive the minimax optimal risk for the additive function estimation problem. In this paper, we reveal the property of $\phi$ that characterizes the minimax optimal risk of the additive functional estimation problem; this analysis is applicable to general $\phi$. More precisely, we reveal that the minimax optimal risk of this problem is characterized by the divergence speed of the function $\phi$.