×

Une mesure d’information caractérisant la loi de Poisson. (An information measure characterising the Poisson distribution). (French) Zbl 0621.60028

Sémin. probabilités XXI, Lect. Notes Math. 1247, 563-573 (1987).
[For the entire collection see Zbl 0606.00022.]
After a brief introduction in which ”information theoretic” arguments in probability are reviewed, a discrete analog of Fisher information measure is defined by \[ I(P)=\sum^{\infty}_{k=0}[P(k)-P(k-1)]^ 2/P(k) \] where supp P\(=\{0,1,...\}\) and \(P(-1):=0\). It is shown that I(P) satisfies: \[ I(P)Var(P)\geq 1;\quad I(P*Q)\leq \min \{I(P),I(Q)\};\quad 4I(P*Q)\leq I(P)+I(Q). \] In Section 4, the quantity I(P) is used to establish a criterium for convergence of some discrete measures to the Poisson distribution. This is applied e.g. to show that if \(X_{n,k}\) is a triangle array of row-wise independent \(\{\) 0,1\(\}\)-valued r.v. \(P(X_{nk}=1)=p_{nk}\) and \(\max \{p_{kn}:k\}\to 0\) as \(n\to \infty,\quad \sum_{k}p_{kn}\to \lambda\) then \(\sum_{k}X_{kn}\) has Poison distribution in the limit \(n\to \infty.\)
Additional references to information-theoretic methods besides the very complete references of the paper can be found in I. J. Good’s review-article J. Am. Stat. Assoc. 78, 987-989 (1983), which gives 36 references related to ”maximum entropy formalism”; S. Watanabe, Knowing and guessing, A quantitative study of inference and information (1969; Zbl 0206.209), provides additional insight into meaning of ”measures of information”, and H. J. Landau, Maximum entropy and the moment problem, Bull. Am. Math. Soc., New Ser. 16, 47-77 (1987), is a recent expository paper.
Reviewer: W.Bryc

MSC:

60F05 Central limit and other weak theorems
60E05 Probability distributions: general theory
60E99 Distribution theory
94A15 Information theory (general)
Full Text: Numdam EuDML