I am new to point processes. I know there are a number of theorems along the lines that if a point process $\eta$ satisfies:

  1. Complete independence (the random variables $\eta(B_1), \ldots, \eta(B_n)$ are independent for pairwise disjoint bounded measurable $B_1, \ldots, B_n$) and

  2. Some regularity conditions like being simple and uniform $\sigma$-finite on a Borel subset of a complete separable metric space,

then $\eta$ is a Poisson process.

It seems that something like the following must be known. I think (I haven't worked through all the details) that if $\eta$ satisfies:

  1. (a) A condition one might call "complete exchangeability": for any disjoint bounded measurable $B_1, \ldots, B_n$, there are $A_i \subseteq B_i$, with equality for at least one $i$, such that $\eta(A_1), \ldots, \eta(A_n)$ are exchangeable random variables;

or equivalently

  1. (b) For any disjoint bounded measurable $B_1, \ldots, B_n$ with $\mathbb{E}\eta(B_1) = \ldots = \mathbb{E}\eta(B_n)$, $\eta(B_1),\ldots, \eta(B_n)$ are exchangeable random variables;

as well as

  1. Similar regularity conditions including $\mathbb{E}\eta(B)<\infty$ for all bounded measurable $B$; and
  2. $\mathbb{E}\eta(\text{whole space}) = \infty$

then there exists a nonnegative-valued scalar random variable $G$ with $\mathbb{E}G = 1$ such that conditioned on $G$, $\eta$ is a Poisson process with intensity measure $G\mathbb{E}\eta$.

Without (3) there are simple counterexamples, e.g. $\eta = \delta_x$ where $x$ is distributed according to a nonatomic probability distribution.

Could anyone provide a reference for such a point process de Finetti theorem?


1 Answer 1


This is Theorem 3.34 in

Kallenberg, Olav, Random measures, Berlin: Akademie-Verlag. London - New York - San Francisco: Academic Press. 104 p. M 28.00 (1976). ZBL0345.60032.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.