Cox point process

In previous posts I have stressed the importance of the Poisson point process, but it can be unsuitable for certain mathematical models.  We can generalize this point process by first considering a non-negative random measure, called a driving or directing measure. Then a Poisson point process, which is independent of the random driving measure, is generated by using the random measure as its intensity or mean measure. This doubly stochastic construction gives what is called a Cox point process.

Typically, we don’t observe the driving measure, so it’s impossible to distinguish a Cox point process from a Poisson point if there’s only one realization available. The properties of a Cox point process are derived by conditioning on the random driving measure, and then using the properties of the Poisson point process.

Cox point processes are also known as doubly stochastic Poisson point processes. Guttorp and Thorarinsdottir argue that it should be called the Quenouille point process, as Maurice Quenouille introduced an example of it before Sir David Cox, but I opt for the more common name.

In this post I’ll cover a couple examples of Cox point processes, but first I’ll give a more precise mathematical definition.

Definition

We consider a point process defined on some underlying mathematical space \(\mathbb{S}\), which is sometimes called the carrier space or state space.  For applications, the underlying space is often the real line \(\mathbb{R}\), the plane \(\mathbb{R}^2\), or some other familiar mathematical space like a square lattice.

For the first definition, we use the concept of a random measure.

Let \(M\) be a non-negative random measure on \(\mathbb{S} \). Then a point process \(\Phi\) defined on some underlying space \(\mathbb{S}\) is a Cox point process driven by the intensity measure \(M\) if the conditional distribution of \(\Phi\) is a Poisson point process with intensity function \(M\).

Alternatively, we can give a slightly less general definition of a Cox  point process by using a random intensity function.

Let \(Z=\{Z(x):x\in\mathbb{S} \}\) be a non-negative random field such that with probability one, \(x\rightarrow Z(x)\) is a locally integrable function. Then a point process \(\Phi\) defined on some underlying space \(\mathbb{S}\) is a Cox point process driven by \(Z\) if the conditional distribution of \(\Phi\) is a Poisson point process with intensity function \(Z\).

The random driving measure \(M\) is then simply the integral
$$
M(B)=\int_B Z(x)\, dx , \quad B\subseteq S.
$$

Over-dispersion

We will soon see that the random driving measures take different forms, giving different Cox point processes. But there is a general observation that can be made for all Cox point processes. For any region \(B \subseteq S\), it can be shown that the number of points \(\Phi (B)\) adheres to the inequality
$$
\mathbb{Var} [\Phi (B)] \geq \mathbb{E} [\Phi (B)],
$$

where \(\mathbb{Var} [\Phi (B)] \) is the variance of the random variable \(\Phi (B)\).  As a comparison, for a Poisson point process \(\Phi’\), the variance of \(\Phi’ (B)\) is simply \(\mathbb{Var} [\Phi’ (B)] =\mathbb{E} [\Phi’ (B)]\).  Due to its greater variance, the Cox point process is said to be over-dispersed compared to the Poisson point process.

Special cases

There is an virtually unlimited number of ways to define a random driving measure, where each one yields a different a Cox point process. But in general mathematicians are and practitioners are restricted by examining only tractable and interesting Cox point processes. I will give some common examples of such Cox point processes, but I stress that the Cox point process family is very large.

Mixed Poisson point process

For the random driving measure \(M\), an obvious example is the product form \(M= Y \mu \),  where \(Y\) is some independent non-negative random variable and \(\mu\) is the Lebesgue measure on \(\mathbb{S}\), meaning \(Y\) is the only source of randomness. This driving measure gives the mixed Poisson point process.

Log-Gaussian Cox point process

Instead of a random variable, we can use a non-negative random field to define a random driving measure.  We then have the product form \(M= Y \mu \), where \(Y\) is now some independent non-negative random field. (A random field is a collection of random variables indexed by some set, which in this case is the underlying space \(\mathbb{S}\).)

Arguably the most tractable and used random field is the Gaussian random field. This random field, like Gaussian or normal random variables, takes negative and positive values, but if we define the random field such that its logarithm is a Gaussian field \(Z\), then we obtain the non-negative random driving measure \(M=\mu e^Z \), which gives the log-Gaussian Cox point process.

This point process has found applications in spatial statistics.

Cox-Poisson line-point process

To construct this Cox point process, we first consider a Poisson line process, which I discussed previously.  Given a Poisson line process, we then place an independent one-dimensional Poisson point process on each line, then we obtain an example of a Cox point process, which we could call a Cox line-point process orCox-Poisson line-point process, but I am not sure of the name.

Researchers have recently used this point process to study wireless communication networks in cities, where the streets correspond to Poisson lines. For example, see these two preprints:

  1. Continuum percolation for Cox point processes
  2. Poisson Cox Point Processes for Vehicular Networks

Shot-noise Cox point process

We construct the next Cox point process by first considering a Poisson point process on the space \(\mathbb{S}\) to create a shot noise term, which we then use as the driving measure of the Cox point process.

More specifically, we first introduce a kernel function \(k(\cdot,\cdot)\) on \(\mathbb{S}\), where \(k(x,\cdot)\) is a probability density function for all points \(x\in \mathbb{S}\). We then consider a Poisson point process \(\Phi’\) on \(\mathbb{S}\times (0,\infty)\), which we assume has a locally integrable intensity function \(\mu \). (We can interpret  the point process \(\Phi’\) as a spatially-dependent marked Poisson point process,  where the unmarked Poisson point process is defined on \(\mathbb{S}\), and each point \(X\) of this unmarked point process has a mark  \(T \in (0,\infty)\) with probability density \(\mu(X,t)\).) The resulting shot noise

$$
Z(x)= \sum_{(Y,T)\in \Phi’} T \, k(Y,x)\,.
$$

gives the random field, which we then use as the random intensity function to drive the shot-noise Cox point process.

In previous posts, I have detailed how to simulate non-Poisson point processes such as the Matérn and Thomas cluster point processes. These are, more specifically, examples of a Neyman-Scott point process, which is a special case of a shot noise Cox point process. All these point processes find applications in spatial statistics.

Simulation

Unfortunately, there is no universal way to simulate all Cox point processes — and even if there were one, it would not be the most optimal way for every Cox point process. The simulation method depends on how the Cox point process is constructed, which usually means how its directing or driving measure is defined.

In previous posts I have presented ways (with code) to simulate the following Cox point processes:

In addition to the Matérn and Thomas point processes, there are ways to simulate to more general shot noise Cox point processes, which I will cover in another post.

Further reading

For general Cox point processes, I suggest the following: Chapter 6 in the monograph Poisson processes by Kingman; Chapter 5 in Statistical Inference and Simulation for Spatial Point Processes by Møller and Waagepetersen; and Section 5.2 in Stochastic Geometry and its Applications by Chiu, Stoyan, Kendall and Mecke. For a much more mathematical treatment, see Chapter 13 in Lectures on the Poisson Process by Last and Penrose. Grandell wrote two detailed monographs titled Mixed Poisson Process and Doubly Stochastic Poisson Processes.

Motivated by applications in spatial statistics, Jesper Møller has (co)-written papers on specific Cox point processes. For example:

  • 2001, Møller, Syversveen, and Waagepetersen, Log Gaussian Cox Processes;
  • 2003, Møller, Shot noise Cox Processes;
  • 2005, Møller and Torrisi,Generalised shot noise Cox processes.

I also suggest the survey article:

  • 2003, Møller and Waagepetersen, Modern statistics for spatial point processes.

Shot noise

Given a mathematical model based on a point process, a quantity of possible interest is the sum of some function applied to each point of the point process. This random sum is called shot noise, where the name comes from developing mathematical models of the noise measured in old electronic devices, which was likened to shot (used in guns) hitting a surface.

Researchers have long studied shot noise induced by a point process. One particularly application is wireless network models, in which the interference term is an example of shot noise. It is also possible to construct new point processes, called shot noise Cox point processes, by using based on the shot noise of some initial point process.

For such applications, we need a more formal definition of shot noise.

Definition

Shot noise of a point process

We consider a point processes \(\Phi=\{X_i\}_i\) defined on some space \(\mathbb{S}\), which is often \(\mathbb{R}^n\), and a non-negative function \(f\) with the domain \(\mathbb{S}\), so \(f:\mathbb{S} \rightarrow [0,\infty)\). This function \(f\) is called the response function.

Then the shot noise is defined as
$$
I= \sum_{X_i\in \Phi} f(X_i)\,.
$$

Shot noise of a marked point process

The previous definition of shot noise can be generalized by considering a marked point process \(\Phi’=\{(X_i, M_i)\}_i\), where each point \(X_i\) now has a random mark \(M_i\), which can be a random variable some other random object taking values in some space \(\mathbb{M}\). Then for a response function \(g:\mathbb{S}\times \mathbb{M} \rightarrow [0,\infty)\) , the shot noise is defined as
$$
I’= \sum_{(X_i, M_i)\in \Phi’} g(X_i,M_i)\,.
$$

Properties

Given a point process on a space, like the plane, at any point the shot noise is simply a random variable. If we consider a subset of the space, then shot noise forms a random field, where we recall that a random field is simply a collection of random variables indexed by some set. (By convention, the set tends to be Euclidean space or a manifold). The shot noise can also be considered as a random measure, for example
$$
I(B)= \sum_{X_i\in \Phi\cap B} f(X_i)\,,
$$
where \(B\subseteq \mathbb{S}\). This makes sense as the point process \(\Phi\) is an example of a random (counting) measure.

For Poisson point processes, researchers have studied resulting shot noise random variable or field. For example, given a homogeneous Poisson point process on \(\mathbb{R}^d\), if the response function is a simple power-law \(f(x)=|x|^{-\beta}\), where \(\beta> d\) and \(|x|\) denotes the Euclidean distance from the origin, then the resulting shot noise is alpha stable random variable with parameter \(\alpha=d/\beta\).

For a general point process \(\Phi\) with intensity measure \(\Lambda\), the first moment of the shot noise is simply
$$
\mathbb{E}(I)= \int_{\mathbb{S}} f(x) \Lambda (dx) \,.
$$

This is a result of Campbell’s theorem or formula. A similar expression exists for the shot noise of a marked point process.

Some history

Shot noise has been studied for over a century in science. In physics, Walter Schottky did research on shot noise in Germany at the beginning of the 20th century. In the same era, Norman R. Campbell studied shot noise in Britain and wrote two key papers, where one of them contains a result now called Campbell’s theorem or Campbell’s formula, among other names, which is a fundamental result in point process theory. Campbell was a physicist, but his work contains this mathematical result for which he credited the famed pure mathematician G. H. Hardy.

(It’s interesting to note that Hardy claimed years later that, given he did pure mathematics, none of his work would lead to applications, but that claim is simply not true for this and other reasons.)

The work on the physical process of shot noise motivated more probability-oriented papers on shot noise, including:

  • 1944, S. O. Rice, Mathematical Analysis of Random Noise;
  • 1960, Gilbert and Pollak. Amplitude distribution of shot noise;
  • 1971, Daley, The definition of a multi-dimensional generalization of shot noise;
  • 1977, J. Rice, On generalized shot noise;
  • 1990 Lowen and Teich, Power-law shot noise.

Further reading

As a model for interference in wireless networks, shot noise is covered in books such as the two-volume textbooks Stochastic Geometry and Wireless Networks by François Baccelli and Bartek Błaszczyszyn, where the first volume is on theory and the second volume is on applications. Martin Haenggi wrote a very readable introductory book called Stochastic Geometry for Wireless networks.

Point process definition and notation

A non-random collection of points located on some space is called a point pattern in spatial statistics. Informally, you can interpret a point process as a collection of random points scattered over some underlying mathematical space, meaning each outcome or realization of a point process forms a point pattern. Using such intuition gets you pretty far. But if we want to be more mathematically formal, we need to use precise mathematical objects.

Historically, there’s a couple main interpretations of a point process, which is also called a random point field. The different interpretations partly explain the various terminology and notation used in the theory of point processes, but now a standard mathematical approach is used.

In this post I will cover the main definitions, terminology and notation used in the theory and application of point processes. I won’t delve too much into the precise details, giving just an outline with a references at the end.

Underlying mathematical space

We consider a point process defined on some underlying mathematical space \(\mathbb{S}\), which is sometimes called the carrier space or state space.  We further assume that the space is measurable by having a Borel \(\sigma\)-algebra \(\mathcal{S}\).

In practice, the underlying space is usually the real line \(\mathbb{R}\), the plane \(\mathbb{R}^2\), or some other familiar mathematical space like a square lattice. More generally, a point process can be defined on any metric space, allowing for the notion of distance. Mathematicians study point processes in even more general settings by defining them on, for example, a locally compact second countable Hausdorff space, but such generality is not needed for most people and their applications.

Modern probability approach

In modern probability theory, if we want to define a random mathematical object, we start with a random experiment in the context of a probability space or triple \((\Omega,\mathcal{A},\mathbb{P})\), where:

  1. \(\Omega\) is a sample space, which is the set of all possible outcomes;
  2. \(\mathcal{A}\) is a \(\sigma\)-algebra or \(\sigma\)-field, which is a family of events (subsets of \(\Omega\));
  3. \(\mathbb{P}\) is a probability measure, which assigns probability to each event in \(\mathcal{A}\).

To gain some intuition, David Williams says to imagine that Tyche, Goddess of Chance, chooses a point \(\omega\in\Omega\) at random according to the law \(\mathbb{P}\) such that an event \(A\in \mathcal{A}\) has a probability given by \(\mathbb{P}(A)\), where we understand probability with our own intuition. (We can also choose \(\omega\in\Omega\) by using some physical experiment, as long as it is truly random.)

Now we can define random objects by using a certain measurable function or mapping that maps to a suitable space of objects. For example, a real-valued random variable is a measurable function from \(\Omega\) to the real line; a random matrix is a measurable function from \(\Omega\) to some space of matrices; and, as John Kingman quips, a random elephant is just a measurable function from \(\Omega\) to some suitable space of elephants.

But what space should we use for a point process? To answer that, we need to interpret a point process as a suitable mathematical object.

Interpretation

There are different ways to interpret a point process, which is often denoted by a single letter, for example, \(N\) or \(\Phi\). (The convention of using the Greek letter \(\Phi\) comes from German mathematicians, but some prefer not to use \(\Phi\), as it’s often used for the normal cumulative distribution function.) If the point process is defined on a space like the real like, where the points can be ordered, then additional interpretations exist, but mathematicians assume the order of the points does not matter, limiting the possible interpretations.

Random closed set

In mathematics a collection of distinct things is formalized by a mathematical object called a set. We say that a set contains elements or members, and a set never contains more than one of the same element. Sets are fundamental objects with set concepts and notation being found everywhere in mathematics.

We now define a common type of point process, which we can formalize with the concept of a set.

A point process is simple if the probability of all points of the point process being distinct is one.

In other words, for a simple point process, there is zero probability of two or more of its points being found in the same location of the underlying state space \(\mathbb{S}\), which brings us to our first interpretation.

A simple point process can be interpreted as a random closed set.

More specifically, we can interpret a simple point process as a (measurable) mapping from a sample space \(\Omega\) to the space of closed sets \(\mathbb{F}\), meaning that each realization of a simple point process is a closed set \(\phi\in \mathbb{F}\).

Point process theory has adopted the notation from set theory. For example, if we want to say some point, which we denote by \(x\), of the underlying space \(\mathbb{S}\) belongs to or is a member of a simple point process \(\Phi\), then we can simply write \(x\in \Phi\). We can also write a point process as \(\{x\}_i\) to highlight its interpretation as a random closed set of points.

The theory of random sets, which is a field of study in its own right, can be applied to point processes owing to this interpretation. But for non-simple point processes, we need another point process interpretation.

Random measures

Modern integration theory is based on measure theory, which revolves around the concept of a set function known as a measure. In addition to a couple of other properties, when you apply this function to a set, it gives a number, such as a integer or real number. For example, a counting measure gives you the number elements in a set, which could be a subspace \(B\), such as a region of the plane \(B \subset \mathbb{R}^2\). (The letter \(B\) is often used for sets in measure and probability theory as it’s typically assumed that the sets are Borel sets, which form a very large family of well-behaved sets in terms of measurability.) The concept of a counting measure gives us the second and now standard interpretation of a point process.

A point process can be interpreted a random counting measure.

More specifically, we define a point process as a mapping from a sample space \(\Omega\) to the space of counting measures \(\mathbb{M}\), meaning that each realization of a point process is a counting measure \(\phi\in \mathbb{M}\). Some mathematicians even say a point process is just another name for a random counting measure. The techniques of random measure theory provide alternative (and arguably main) approach to study point processes.

This standard interpretation of a point process means that point process theory borrows heavily notation from measure theory and calculus. For example, in measure theory we can write a (non-random) counting measure as \(\#\), so \(\#(B)=n\) is how we write that the set \(B\) contains \(n\) points. We can then write the the number of points of a point process \(\Phi\) located in some (Borel) set \(B\) as \(N(B) =\#( B \cap \Phi)\), where \(N(B)\) is a random variable. In this expression, the point process is denoted by \(\Phi\), while\(N(B)\) is the number of points of \(\Phi\) in \(B\), meaning \(N\) is a random counting measure .

Notation

The main interpretations of point processes as random sets and counting measures is captured with the notation:

  • \(\Phi\) is a set of random points.
  • \(\Phi(B)\) is a random variable that gives the number of points of \(\Phi\) located in the (Borel) set \(B\).

This is the notation often used in point process theory. It implies
$$
\Phi(B) =\#(B \cap \Phi).
$$

We now look at how this notation is used in point process theory.

Sums

If \(f\) is some (measurable) function on the underlying space \(\mathbb{S}\), such as Euclidean space \(\mathbb{R}^d\), then we can write the sum of \(f(x)\) over all the points of a simple point process \(\Phi\) as
$$
\sum_{x\in \Phi}f(x)\,,
$$
where we are using the random set interpretation.

For any point process \(\Phi\), we can also write the sum as
$$
\int_{\mathbb{S}} f(x) \,\Phi(dx) \,,
$$
which highlights the interpretation of the point process \(\Phi\) as a random counting measure. Of course, we can use different integral notation, giving, for example, the expression
$$
\int_{\mathbb{S}} f \,d\Phi \,,
$$
which denotes the same sum.

We can illustrate the dual interpretation of a point process by writing the number of point of a simple point process \(\Phi\) existing in a set \(B\) as
$$
\Phi(B)= \sum_{x\in \Phi}1_B(x)\,,
$$
where the indicator function \(1_B(x) =1\) if the point \(x\) is exists in the set \(B\), and \(1_B(x) =0\) otherwise. In this setting, \(1_B(x)\) is also known as a Dirac measure, as it gives a measure of the set \(B\). We can see in this expression that the random measure interpretation is on the left-hand side, while the random set notation is on the right-hand side.

Expectations

We can write the average or expected value of a sum of functions over a simple point process \(\Phi\) as
$$
\mathbb{E}\left[\sum_{x\in \Phi}f(x)\right] \,,
$$
or for any point process \(\Phi\) as
$$
\int_{\textbf{N}}\sum_{x\in \Phi}f(x) \mathbb{P}(d\Phi)\,,
$$
where \(\mathbb{P}\) is an appropriate probability measure defined on the space of counting functions \(\textbf{N}\), thus illustrating the random measure interpretation.

We can write the expected value of \(\Phi(B)\), which is the definition of the intensity measure of a point process \(\Phi\), as
$$
\mathbb{E}[\Phi(B)]=\mathbb{E}\left( \sum_{x\in \Phi}1_B(x)\right) \qquad \text{or} \qquad \mathbb{E}[\Phi(B)]=\int_{\textbf{N}}\sum_{x\in \Phi}1_B(x) P(d\Phi) \,,
$$
which is also known as the mean measure or first moment measure of \(\Phi\).

Events

In probability we want to describe the behaviour of certain events, such as flipping at last three heads across ten coin flips. For point processes, events are simply configurations with a certain (geometric) property, such as no points existing in a certain region or all the points being a fixed minium distance from each other.

Typically, when being mathematically abstract, we denote an event with a single letter, such as \(\Gamma\). Then to denote that a point process satisfies this condition we write \(\Phi\in \Gamma\). In other words, the point process \(\Phi\) has the property \(\Gamma\). We can then write the probability of the event (or configuration) \(\Gamma\) of occurring as
$$
\mathbb{P}(\Gamma)= \mathbb{P}(\Phi\in \Gamma ) \,.
$$

Uppercase and subscript notation

The convention in probability is usually to denote random objects, such as random variables and point processes, with uppercase (or capital) letters. Conversely, a non-random object, such as the realization of a random variable or point process is denoted by a lowercase letter. For example, \(\Phi\) is a point processes, while \(\phi\) is a point pattern, which may be a realization of the point process \(\Phi\).

With this convention, we can denote an arbitrary point process of a point process \(\Phi\) by \(X\), meaning \(X\in \Phi\). (But such a point is also a point on the underlying non-random space \(\mathbb{S}\) on which the point process \(\Phi\) is defined.) We also see lowercase used for the point, giving \(x\in \Phi\).

Sometimes subscripts are used to emphasize some type of numbering of points, giving, for example, two points \(X_1\in \Phi\) and \(X_2\in \Phi\). Sometimes authors will write something like

$$
\sum_{X_i\in \Phi}f(X_i)\,,
$$

but this redundant notation as \(X_i\) is a dummy variable, so you can omit the subscript in such an expression.

Some authors use a notation where the letter with and without a subscript denotes, respectively, the point process and a point belonging to the point process. Using this convention, we write, for example, \(X=\{ X_i\}_i\) and \(X_i\in X\).

Important concepts

In the theory point processes, like any other field of mathematics, there are various important concepts for understanding and proving various results. Without going in the details, these include shot noiseCampbell’s theorem, Laplace functional, Palm calculus, void probability, and factorial moment measures. In future posts, I’ll detail some of these concepts.

Further reading

There are many, many books covering the fundamentals of modern probability theory, including those (in roughly order of difficult) by Grimmett and Stirzaker, Karr,  Rosenthal, Durrett, Shiryaev,  and Billingsley.  A very quick introduction is given in this web article.

For point process theory, Wikipedia is a good start place to start, particularly the articles on point processes and point process notation, though the former is too mathematical for a Wikipedia article. The standard reference on point processes was the An Introduction to the Theory of Point Processes by Daley and Vere-Jones, which now spread across volume one and two, but I would not learn the subject with these books. The classic text Stochastic Geometry and its Applications by Chiu, Stoyan, Kendall and Mecke covers point processes and the varying notation in Chapters 2 and 4. The similar material is covered in the previous edition by Stoyan, Kendall and Mecke. A more mathematical book that covers point processes and random sets is Stochastic and integral geometry by Schneider and Weil. Point processes are also covered in a recent readable tutorial on Palm calculus and Gibbs point processes, which will be the subject of another post.

Spatial statistics builds of point process theory, giving good texts for learning the basics of point processes. I suggest the lectures notes by Baddeley, which form Chapter 1 of these published lectures, edited by Baddeley, Bárány, Schneider, and Weil. I always recommend the book Spatial Point Patterns: Methodology and Applications with R written by spatial statistics experts Baddeley, Rubak and Turner, which covers the spatial statistics (and point process simulation) R-package spatstat. Another good book is Statistical Inference and Simulation for Spatial Point Processes by Møller and Waagepetersen, but there are many more.

In recent years, point process theory (under the guise of stochastic geometry) has been used to model wireless networks. An early treatment of the subject is the two-volume textbook Stochastic Geometry and Wireless Networks by Baccelli and Błaszczyszyn, where the first volume is on theory and the second volume is on applications. Haenggi wrote a very readable introductory book called Stochastic Geometry for Wireless networks.

For the more mathematically brave, there’s the recent book Random Measures, Theory and Applications by Kallenberg who is the authority of the subject, having written earlier books, but these have now become obsolete with this recent publication. Another mathematically challenging book is The Theory of Random Sets by Molchanov, but it has less emphasis on point processes.

A very recent book (manuscript) is Random Measures, Point Processes, and Stochastic Geometry by Baccelli, Błaszczyszyn, and Karray, which contains much material, including new results and proofs. Finally, Last and Penrose wrote a mathematical monograph Lectures on the Poisson process, which is freely available online here.