Due to unplanned maintenance of the back-end systems supporting article purchase on Cambridge Core, we have taken the decision to temporarily suspend article purchase for the foreseeable future. We apologise for any inconvenience caused whilst we work with the relevant teams to restore this service.
In this chapter, we introduce four additional point process models: cluster processes, hard-core processes, Cox processes, and Gibbs processes. Poisson point processes exhibit complete spatial randomness due to their independence property. Cox processes model less regular spatial distributions – they are overdispersed relative to PPPs, which means that the ratio of the variance of the number of nodes in a set to its mean is larger than 1. Gibbs processes, on the other hand, may be overdispersed or underdispersed.
Figure 3.1 illustrates the two directions along which a point process model may depart from the point of complete spatial randomness, the PPP. In terms of the J function introduced in Definition 2.40, regular processes have J values larger than 1, since the probability of having a nearby neighbor is small; conversely, clustered processes have J values smaller than 1.
General finite point processes
A general finite point process is a generalization of the binomial point process to a process where the total number of points is itself a random variable. Compared with a random vector, there are three differences: The randomness of the number of points and the facts that the point process is unordered and simple and thus is better represented as a set.