On Stopping Sets and Continuous-Time Decision Trees

Abstract

We study stopping sets and continuous-time decision trees (CTDTs) for Poisson point processes. We prove that every $\mu$-continuous CTDT has an explicit exploration representation: between successive discoveries, the explored set grows according to a deterministic rule depending on the points already observed. If $\mu$ is diffuse, we prove that the points are revealed one by one almost surely; this need not be imposed as a separate assumption. We also allow atoms of $\mu$, in which case all points at one location are revealed together. For diffuse $\mu$, we also prove that a locally bounded stopping set that reveals at most one Poisson point is attainable by a $\mu$-continuous exploration. More generally, an increasing sequence of locally bounded stopping sets, each revealing at most one additional point, can be attained by a single exploration viewed at successive stopping times. Finally, we prove the spatial Markov property of the Poisson point process without assuming that the stopping set contains finitely many points or admits a finite approximation. The proof is based on a general criterion for characterizing conditionally Poisson point processes using moment measures, and uses the multivariate Mecke formula on the complement of the stopping set.

arXiv comment20 pages