Point process operation

In probability and statistics, a point process operation is a type of mathematical operation performed on a random object known as a point process, which are often used as mathematical models of phenomena that can be represented as points randomly located in space. These operations can be purely random, deterministic or both, and are used to construct new point processes, which can be then also used as mathematical models. The operations may include removing or thinning points from a point process, combining or superimposing multiple point processes into one point process or transforming the underlying space of the point process to another space.

One point process that gives particular convenient results under point process operations is the Poisson point process, hence it is often used as a mathematical model.

Point process operations are used in the theory of point processes and related fields such as stochastic geometry and spatial statistics.

Examples of operations

To develop suitable models with point processes in stochastic geometry, spatial statistics and related fields, there are number of useful transformations that can be performed on points processes including: thinning, superposition, mapping (or transformation of space), clustering, and random displacement.

Thinning

The thinning operation entails using some predefined rule to remove points from a point process $\textstyle \Phi$ to form a new point process $\textstyle \Phi _{p}$. These rules may be completely random, which is the case for one of the simplest rules known as $\textstyle p$-thinning: each point of $\textstyle \Phi$ is independently removed (or kept) with some probability $\textstyle p$ (or $\textstyle 1-p$). This rule may be generalized by introducing a non-negative function $\textstyle p(x)\leq 1$ in order to define the located-dependent $\textstyle p(x)$-thinning where now the probability of a point being removed is $\textstyle p(x)$ and is dependent on where the point of $\textstyle \Phi$ is located on the underlying space. A further generalization is to have the thinning probability $\textstyle p$ random itself.

These three operations are all types of independent thinning, which means the interaction between points has no effect on the where a point is removed (or kept). Another generalization involves dependent thinning where points of the point process are removed (or kept) depending on their location to other points of the point process. Thinning can be used to create new point processes such as hard-core processes where points do not exist (due to thinning) within a certain radius of each point in the thinned point process.

Superposition

If there is a countable collection of point processes $\textstyle \Phi _{1},\Phi _{2}\dots$ with mean measures $\textstyle \Lambda _{1},\Lambda _{2},\dots$, then their superposition

$\Phi =\bigcup _{i=1}^{\infty }\Phi _{i},$

also forms a point process. In this expression the superposition operation is denoted by a set union), which implies the random set interpretation of point processes; see Point process notation for more information.

Poisson point process case

In the case where each $\textstyle \Phi _{i}$ is a Poisson point process, then the resulting process $\textstyle \Phi$ is also a Poisson point process with mean intensity

$\Lambda =\sum \limits _{i=1}^{\infty }\Lambda _{i}.$

Clustering

The point operation known as clustering entails replacing every point $\textstyle x$ in a given point process $\textstyle \Phi$ with a cluster of points $\textstyle N^{x}$. Each cluster is also a point process, but with a finite number of points. The union of all the clusters forms a cluster point process

$\Phi _{c}=\bigcup _{x\in \Phi }N^{x}.$

Often is it assumed that the clusters $\textstyle N^{x}$ are all sets of finite points with each set being independent and identically distributed. Furthermore, if the original point process $\textstyle \Phi$ has a constant intensity $\textstyle \lambda$, then the intensity of the cluster point process $\textstyle \Phi _{c}$ will be

$\lambda _{c}=c\lambda ,$

where the constant $\textstyle c$ is the mean of number of points in each $\textstyle N^{x}$.

Random displacement

A mathematical model may require randomly moving or displacing points of a point process to other locations on the underlying mathematical space.

Poisson point process case

The result known as the displacement theorem effectively says that the random independent displacement of points of a Poisson point process (on the same underlying space) forms another Poisson point process.

Transformation of space

Another property that is considered useful is the ability to map a point process from one underlying space to another space. For example, a point process defined on the plane 'R² can be transformed from Cartesian coordinates to polar coordinates.

Poisson point process case

Provided that the mapping (or transformation) adheres to some conditions, then a result sometimes known as the mapping theorem says that if the original process is a Poisson point process with some intensity measure, then the resulting mapped (or transformed) collection of points also forms a Poisson point process with another intensity measure.

Convergence of point process operations

A point operation performed once on some point process can be, in general, performed again and again. In the theory of point processes, results have been derived to study the behaviour of the resulting point process, via convergence results, in the limit as the number of performed operations approaches infinity. For example, under certain mathematical conditions, if each point in a general point process is repeatedly displaced in a random and independent manner, then the new point process, loosely speaking, will more and more resemble a Poisson point process. Similar convergence results have been developed for the operations of thinning and superposition (with suitable rescaling of the underlying space).