1: \begin{abstract}

2: R\'enyi's {\em thinning} operation on a discrete random variable 

3: is a natural discrete analog of the scaling operation for 

4: continuous random variables. The properties of thinning 

5: are investigated in an information-theoretic context,

6: especially in connection with information-theoretic inequalities

7: related to Poisson approximation results. The classical 

8: Binomial-to-Poisson convergence (sometimes referred to 

9: as the \textquotedblleft law of small numbers\textquotedblright) 

10: is seen to be a special case of a thinning limit theorem 

11: for convolutions of discrete distributions. A rate of convergence 

12: is provided for this limit, and nonasymptotic bounds are also

13: established. This development parallels, in part, the 

14: development of Gaussian inequalities leading to the 

15: information-theoretic version of the central limit theorem.

16: In particular, a ``thinning Markov chain'' is introduced,

17: and it is shown to play a role analogous to that 

18: of the Ornstein-Uhlenbeck process in connection to the

19: entropy power inequality.

20: \end{abstract}

21: