Main results
The relations in Theorem 1 hold for any single binomial variable Y, and we will apply it simultaneously to an infinite ensemble Yj j∈N as follows. Suppose one empirically estimates the distribution p = (p1, p2, . . .) of an N-valued random variable X from n i.i.d. samples Xi via pˆ (n) , where
Then, defining
we have
where Yj ∼ Bin(n, pj).
For the simple case where p has finite support, we have
For p with infinite support, the estimate