Exchangeable random measures in the plane (Q909335)

A random measure \(\xi\) on \([0,1]^ 2\), \([0,1]\times R_+\) or \(R^ 2_+\) is said to be separately exchangeable if its distribution is invariant under measurable mappings of the form \(h(x,y)=f(x)g(y)\), where f and g preserve the respective one-dimensional Lebesgue measure. If \(\xi\) is defined on \([0,1]^ 2\) or \(R^ 2_+\) and the above invariance property holds whenever \(f=g\), then \(\xi\) is said to be jointly exchangeable. The paper derives de Finetti-type representations for separately or jointly exchangeable random measures in terms of independent Poisson processes and i.i.d. random variables. There are five representation theorems, corresponding to the five cases mentioned above, and to give the flavour of these results we state here the simplest: A random measure \(\xi\) on \([0,1]^ 2\) is separately exchangeable if and only if it has an almost sure representation of the form \[ \xi =\sum_{i}\sum_{j}\alpha_{ij}\delta_{\tau_ i,\tau_ j'}+\sum_{j}\{\beta_ j(\delta_{\tau_ j}\times \lambda)+\beta_ j'(\lambda \times \delta_{\tau_ j'})\}+\gamma \lambda^ 2 \] for some \(R_+\)-valued random variables \(\alpha_{ij},\beta_ i,\beta_ j',\gamma\) (i,j\(\in N)\) and some independent set of independent random variables \(\tau_ i,\tau_ j'\) (i,j\(\in N)\) uniformly distributed on (0,1). (\(\lambda\) denotes the Lebesgue measure on (0,1)).