3 The Poisson Process
3.1 Motivation: Complete Spatial Randomness
In many applications, it is useful to have a baseline model representing no spatial structure.
Intuitively, this corresponds to a situation where:
- events occur independently of one another
- there is no clustering or inhibition
- the spatial distribution is entirely random
This notation is formalized by the Poisson process, often referred to as a model of complete spatial randomness (CSR).
3.2 Definition (Informal)
A Poisson process is a point process with two key properties:
Counts in disjoint regions are independent
The number of points in a region depends only on its size (or intensity).
3.3 Homogeneous Poisson Process
We first consider the simplest case.
A point process \(N(\cdot)\) on \(W \subset \mathbb{R}^d\) is called a homogeneous Poisson process (HPPP) with intensity \(\lambda > 0\) if it satisfies:
(i) Poisson counts
For any region \(B \subset W\):
\[ N(B) \sim \text{Poisson}(\lambda |B|) \]
where \(|B|\) is the volume (area in 2D) of \(B\).
(ii) Independent increments
If \(B_1, \dots, B_k\) are disjoint regions, then:
\[ N(B_1), \dots, N(B_k) \ \text{are independent} \]
3.4 Interpretation
The parameter \(\lambda\) has a clear meaning:
- it represents the average number of points per area
- and for any region \(B\), we have:
\[ \mathbb{E}[N(B)] = \lambda |B| \]
3.5 Equivalent construction
The Poisson process can be constructed by the following two steps:
- Draw the number of points:
\[ N \sim \text{Poisson}(\lambda |W|) \]
- Given \(N = n\), sample locations:
\[ x_1, \dots, x_n \stackrel{\text{iid}}{\sim} \text{Uniform}(W) \] ## Inhomogeneous Poisson process
In many applications, the density of events varies across space.
A point process is called an inhomogeneous Poisson point process (IPPP) with intensity function \(\lambda(u)\) if:
(i) Poisson counts
For any region \(B \subset W\):
\[ N(B) \sim \text{Poisson}(\int_B \lambda(u)\ du) \]
where \(|B|\) is the volume (area in 2D) of \(B\).
(ii) Independent increments
If \(B_1, \dots, B_k\) are disjoint regions, then:
\[ N(B_1), \dots, N(B_k) \ \text{are independent} \]
3.6 Interpretation of \(\lambda(u)\)
The function \(\lambda(u)\) represents:
The expected number of points per unit area at location \(u\)
More precisely,
\[ \mathbb{E}[N(B)] = \int_B \lambda(u) \ du \]
3.7 Why the Poisson process matters
The Poisson process plays a central role because:
- it represents complete spatial randomness
- it serves as a baseline for detecting structure
- many summary statistics are defined relative to it