1: \begin{definition}[Residual sampling estimators] 

2: \label{def:residual} %{{{

3: Let $n\in\N$, define $n_i \defeq \lfloor n p_i \rfloor$

4: and let $r \defeq n - \sum_{i\ge 1} n_i>0$. Define

5: the `residual' probability distribution $(p_i^*)_{i\ge 1}$ as 

6: $p_i^* \defeq (n p_i - n_i)/r$,

7: and let $Q^{(j)}$ be independent random variables such that 

8: $\P(Q^{(j)}=i)=p_i^*$ for $j=1,\ldots,r$. 

9: \begin{enumerate}[(i)]

10: \item 

11: Define $N_i \defeq n_i + N_i^*$ where 

12:     $N_i^* \defeq \textstyle\sum_{j=1}^r \charfun{Q^{(j)}=i}$,

13: then the estimator

14: $Z_\res^{(n)}$ defined as in Theorem 

15: \ref{thm:general} is the \emph{residual sampling single term estimator}.

16: \item Define 

17:    $\tilde{N}_i \defeq \tilde{n}_i + \tilde{N}_i^*$, where

18:    $\tilde{n}_i \defeq \textstyle\sum_{k=1}^i n_i$ and

19:    $\tilde{N}_i^* \defeq \textstyle\sum_{j=1}^r \charfun{Q^{(j)}\ge

20:      i}$,

21:   then the estimator

22:   $Z_{\sumsymb,\res}^{(n)}$ defined as in 

23:   Theorem \ref{thm:general} is the \emph{residual sampling sum

24:     estimator}.

25: \end{enumerate}

26: \end{definition}