Poisson process

In probability theory, a Poisson process is a stochastic process that counts the number of events^{[note 1]} in a given time interval. The time between each pair of consecutive events has an exponential distribution with parameter λ and each of these inter-arrival times is assumed to be independent of other inter-arrival times. The process is named after the Poisson distribution introduced by French mathematician Siméon Denis Poisson.^[1] It is useful for modeling radioactive decay,^[2] telephone calls at a call center,^[3] document requests on a web server,^[4] and many other punctual phenomena where events occur independently from each other.

The Poisson process is a continuous-time stochastic process; the sum of a Bernoulli process can be thought of as its discrete-time counterpart. A Poisson process is a pure-birth process, the simplest example of a birth-death process. It is also a point process on the real half-line.

Definition

The basic form of Poisson process, often referred to simply as "the Poisson process", is a continuous-time counting process {N(t), t ≥ 0} that possesses the following properties:^[5]

N(0) = 0
Independent increments (the numbers of occurrences counted in disjoint intervals are independent of each other)
Stationary increments (the probability distribution of the number of occurrences counted in any time interval only depends on the length of the interval)
Proportionality (the probability of an occurrence in a time interval is proportional to the length of the time interval, for "small" time intervals)
The probability of simultaneous occurrences equals zero.

Consequences of this definition include:

The probability distribution of the waiting time until the next occurrence is an exponential distribution.
For each t≥0, the probability distribution of N(t) is a Poisson distribution with parameter λt. Here λ>0 is called the rate of the Poisson process.
The occurrences are distributed uniformly on any interval of time. (Note that N(t), the total number of occurrences, has a Poisson distribution over the non-negative integers, whereas the location of an individual occurrence on $t \in (a, b]$ is uniform.)

Other types of Poisson process are described below.

Types

There are several generalizations of the basic Poisson process, also termed Poisson processes. The homogeneous Poisson process coincides with the basic Poisson process defined above, while in an inhomogeneous Poisson process, events occur at a variable rate. Generalizations to counts over a spatial region or on a spatial region over time are also useful.

Homogeneous

Sample path of a counting Poisson process N(t)

A homogeneous Poisson process counts events that occur at a constant rate; it is one of the most well-known Lévy processes. This process is characterized by a rate parameter λ, also known as intensity, such that the number of events in time interval (t, t + τ] follows a Poisson distribution with associated parameter λτ. This relation is given as

P[N(t+\tau )-N(t)=k]={\frac {e^{-\lambda \tau }(\lambda \tau )^{k}}{k!}}\qquad k=0,1,\ldots ,

where N(t + τ) − N(t) = k is the number of events in time interval (t, t + τ].

Just as a Poisson random variable is characterized by its scalar parameter λ, a homogeneous Poisson process is characterized by its rate parameter λ, which is the expected number of "events" or "arrivals" that occur per unit time.

N(t) is a sample homogeneous Poisson process, not to be confused with a density or distribution function.

Inhomogeneous

An inhomogeneous Poisson process counts events that occur at a variable rate. In general, the rate parameter may change over time; such a process is called a non-homogeneous Poisson process or inhomogeneous Poisson process. In this case, the generalized rate function is given as λ(t). Now the expected number of events between time a and time b is

N_{a,b}=\int _{a}^{b}\lambda (t)\,dt.

Thus, the number of arrivals in the time interval [a, b], given as N(b) − N(a), follows a Poisson distribution with associated parameter N_a,b

P[N(b)-N(a)=k]={\frac {e^{-N_{a,b}}(N_{a,b})^{k}}{k!}}\qquad k=0,1,\ldots .

A rate function λ(t) in a non-homogeneous Poisson process can be either a deterministic function of time or an independent stochastic process, giving rise to a Cox process. A homogeneous Poisson process may be viewed as a special case when λ(t) = λ, a constant rate.

Spatial

An important variation on the (notionally time-based) Poisson process is the spatial Poisson process. In the case of a one-dimension space (a line) the theory differs from that of a time-based Poisson process only in the interpretation of the index variable. For higher dimension spaces, where the index variable (now x) is in some vector space V (e.g. R² or R³), a spatial Poisson process can be defined by the requirement that the random variables defined as the counts of the number of "events" inside each of a number of non-overlapping finite sub-regions of V should each have a Poisson distribution and should be independent of each other.

Space-time

A further variation on the Poisson process, the space-time Poisson process, allows for separately distinguished space and time variables. Even though this can theoretically be treated as a pure spatial process by treating "time" as just another component of a vector space, it is convenient in most applications to treat space and time separately, both for modeling purposes in practical applications and because of the types of properties of such processes that it is interesting to study.

In comparison to a time-based inhomogeneous Poisson process, the extension to a space-time Poisson process can introduce a spatial dependence into the rate function, such that it is defined as $\lambda (x,t)$ , where $x\in V$ for some vector space V (e.g. R² or R³). However a space-time Poisson process may have a rate function that is constant with respect to either or both of x and t. For any set $S\subset V$ (e.g. a spatial region) with finite measure $\mu (S)$ , the number of events occurring inside this region can be modeled as a Poisson process with associated rate function λ_S(t) such that

\lambda _{S}(t)=\int _{S}\lambda (x,t)\,d\mu (x).

Separable space-time processes

In the special case that this generalized rate function is a separable function of time and space, we have:

\lambda (x,t)=f(x)\lambda (t)\,

for some function $f(x)$ . Without loss of generality, let

\int _{V}f(x)\,d\mu (x)=1.

(If this is not the case, λ(t) can be scaled appropriately.) Now, $f(x)$ represents the spatial probability density function of these random events in the following sense. The act of sampling this spatial Poisson process is equivalent to sampling a Poisson process with rate function λ(t), and associating with each event a random vector $X$ sampled from the probability density function $f(x)$ . A similar result can be shown for the general (non-separable) case.

Characterisation

In its most general form, the only two conditions for a counting process to be a Poisson process are:^{[citation needed]}

Orderliness: which roughly means

\lim _{\Delta t\to 0}P(N(t+\Delta t)-N(t)>1\mid N(t+\Delta t)-N(t)\geq 1)=0

which implies that arrivals don't occur simultaneously (but this is actually a mathematically stronger statement).

Memorylessness (also called evolution without after-effects): the number of arrivals occurring in any bounded interval of time after time t is independent of the number of arrivals occurring before time t.

These seemingly unrestrictive conditions actually impose a great deal of structure in the Poisson process. In particular, they imply that the time between consecutive events (called interarrival times) are independent random variables. For the homogeneous Poisson process, these inter-arrival times are exponentially distributed with parameter λ (mean 1/λ).

Also, the memorylessness property entails that the number of events in any time interval is independent of the number of events in any other interval that is disjoint from it. This latter property is known as the independent increments property of the Poisson process.

Properties

As defined above, the stochastic process {N(t)} is a Markov process, or more specifically, a continuous-time Markov process.^{[citation needed]}

To illustrate the exponentially distributed inter-arrival times property, consider a homogeneous Poisson process N(t) with rate parameter λ, and let T_k be the time of the kth arrival, for k = 1, 2, 3, ... . Clearly the number of arrivals before some fixed time t is less than k if and only if the waiting time until the kth arrival is more than t. In symbols, the event [N(t) < k] occurs if and only if the event [T_k > t] occurs. Consequently the probabilities of these events are the same:

P(T_{k}>t)=P(N(t)<k).\,

In particular, consider the waiting time until the first arrival. Clearly that time is more than t if and only if the number of arrivals before time t is 0. Combining this latter property with the above probability distribution for the number of homogeneous Poisson process events in a fixed interval gives:

P(T_{1}>t)=P(N(t)=0)=P[N(t)-N(0)=0]={\frac {e^{-\lambda t}(\lambda t)^{0}}{0!}}=e^{-\lambda t}.

And therefore:

P(T_{1}\leq t)=P(N(t)>0)=1-e^{-\lambda t}

(which is the CDF of the exponential distribution).

Consequently, the waiting time until the first arrival T₁ has an exponential distribution, and is thus memoryless. One can similarly show that the other interarrival times T_k − T_k−1 share the same distribution. Hence, they are independent, identically distributed (i.i.d.) random variables with parameter λ > 0; and expected value 1/λ. For example, if the average rate of arrivals is 5 per minute, then the average waiting time between arrivals is 1/5 minute.

Applications

The classic example of phenomena well modelled by a Poisson process is deaths due to horse kick in the Prussian army, as shown in 1898 by Ladislaus Bortkiewicz, a Polish economist and statistician who also examined data of child suicides.^[6]^[7] The following examples are also well-modeled by the Poisson process:

Number of road crashes (or injuries/fatalities) at a site or in an area
Goals scored in a association football match.^[8]
Requests for individual documents on a web server.^[4]
Particle emissions due to radioactive decay by an unstable substance. In this case the Poisson process is non-homogeneous in a predictable manner—the emission rate declines as particles are emitted.
Action potentials emitted by a neuron.^[9]
L. F. Richardson showed that the outbreak of war followed a Poisson process from 1820 to 1950.^[10]
Photons landing on a photodiode, in particular in low light environments. This phenomenon is related to shot noise.
Opportunities for firms to adjust nominal prices.^[11]
Arrival of innovations from research and development.^[12]
Requests for telephone calls at a switchboard.^{[citation needed]}
In queueing theory, the times of customer/job arrivals at queues are often assumed to be a Poisson process.
The evolution (changes on pages) of Internet, in general (although not in the particular case of Wikipedia)^[13]

Occurrence

The Palm–Khintchine theorem provides a result that shows that the superposition of many low intensity non-Poisson point processes will be close to a Poisson process.

Notes

^ The word event used here is not an instance of the concept of event as frequently used in probability theory.

References

^ S.D. Poisson, Probabilité des jugements en matière criminelle et en matière civile, précédées des règles générales du calcul des probabilitiés (Paris, France: Bachelier, 1837), page 206.
^ Cannizzaro, F.; Greco, G.; Rizzo, S.; Sinagra, E. (1978). "Results of the measurements carried out in order to verify the validity of the poisson-exponential distribution in radioactive decay events". The International Journal of Applied Radiation and Isotopes. 29 (11): 649. doi:10.1016/0020-708X(78)90101-1.
^ Willkomm, D.; Machiraju, S.; Bolot, J.; Wolisz, A. (2009). "Primary user behavior in cellular networks and implications for dynamic spectrum access". IEEE Communications Magazine. 47 (3): 88. doi:10.1109/MCOM.2009.4804392.
^ ^a ^b Arlitt, M. F.; Williamson, C. L. (1997). "Internet Web servers: Workload characterization and performance implications". IEEE/ACM Transactions on Networking. 5 (5): 631. doi:10.1109/90.649565.
^ Birolini, A., Reliability Engineering, 7th edition, Springer, Heidelberg, 2014
^ Ladislaus von Bortkiewicz, Das Gesetz der kleinen Zahlen [The law of small numbers] (Leipzig, Germany: B.G. Teubner, 1898). On page 1, Bortkiewicz presents the Poisson distribution. On pages 23-25, Bortkiewicz presents his famous analysis of "4. Beispiel: Die durch Schlag eines Pferdes im preussischen Heere Getöteten." (4. Example: Those killed in the Prussian army by a horse's kick.).
^ Gibbons, Robert D.; Bhaumik, Dulal; Aryal, Subhash (2009). Statistical Methods for Groundwater Monitoring. John Wiley and Sons. p. 72. ISBN 0-470-16496-4.
^ Heuer, A.; Müller, C.; Rubner, O. (2010). "Soccer: Is scoring goals a predictable Poissonian process?". EPL (Europhysics Letters). 89 (3): 38007. doi:10.1209/0295-5075/89/38007. To a very good approximation scoring goals during a match can be characterized as independent Poissonian processes with pre-determined expectation values.
^ Brunel, N. (2000). "Phase diagrams of sparsely connected networks of excitatory and inhibitory spiking neurons". Neurocomputing. 32–33: 307–312. doi:10.1016/S0925-2312(00)00179-X.
^ Hayes, B. (2002). "Statistics of Deadly Quarrels". American Scientist. 90: 10–14. doi:10.1511/2002.1.10.
^ Calvo, G. A. (1983). "Staggered prices in a utility-maximizing framework". Journal of Monetary Economics. 12 (3): 383–398. doi:10.1016/0304-3932(83)90060-0.
^ Aghion, Philippe; Howitt, Peter (1992). "A Model of Growth Through Creative Destruction". Econometrica. 60 (2): 323–351. JSTOR 2951599.
^ Almeida, R. B.; Mozafari, B., y Cho, J. (2007). On the evolution of Wikipedia. ICWSM (Boulder, Colorado) (Retrieved May 31, 2014)

v t e Stochastic processes
Discrete time	Bernoulli process Branching process Chinese restaurant process Galton–Watson process Independent and identically distributed random variables Markov chain Moran process Random walk Loop-erased Self-avoiding Biased Maximal entropy
Continuous time	Additive process Bessel process Birth–death process pure birth Brownian motion Bridge Excursion Fractional Geometric Meander Cauchy process Contact process Continuous-time random walk Cox process Diffusion process Dyson Brownian motion Empirical process Feller process Fleming–Viot process Gamma process Geometric process Hawkes process Hunt process Interacting particle systems Itô diffusion Itô process Jump diffusion Jump process Lévy process Local time Markov additive process McKean–Vlasov process Ornstein–Uhlenbeck process Poisson process Compound Non-homogeneous Quasimartingale Schramm–Loewner evolution Semimartingale Sigma-martingale Stable process Superprocess Telegraph process Variance gamma process Wiener process Wiener sausage
Both	Branching process Gaussian process Hidden Markov model (HMM) Markov process Martingale Differences Local Sub- Super- Random dynamical system Regenerative process Renewal process Stochastic chains with memory of variable length White noise
Fields and other	Dirichlet process Gaussian random field Gibbs measure Hopfield model Ising model Potts model Boolean network Markov random field Percolation Pitman–Yor process Point process Cox Poisson Random field Random graph
Time series models	Autoregressive conditional heteroskedasticity (ARCH) model Autoregressive integrated moving average (ARIMA) model Autoregressive (AR) model Autoregressive–moving-average (ARMA) model Generalized autoregressive conditional heteroskedasticity (GARCH) model Moving-average (MA) model
Financial models	Binomial options pricing model Black–Derman–Toy Black–Karasinski Black–Scholes Chan–Karolyi–Longstaff–Sanders (CKLS) Chen Constant elasticity of variance (CEV) Cox–Ingersoll–Ross (CIR) Garman–Kohlhagen Heath–Jarrow–Morton (HJM) Heston Ho–Lee Hull–White Korn-Kreer-Lenssen LIBOR market Rendleman–Bartter SABR volatility Vašíček Wilkie
Actuarial models	Bühlmann Cramér–Lundberg Risk process Sparre–Anderson
Queueing models	Bulk Fluid Generalized queueing network M/G/1 M/M/1 M/M/c
Properties	Càdlàg paths Continuous Continuous paths Ergodic Exchangeable Feller-continuous Gauss–Markov Markov Mixing Piecewise-deterministic Predictable Progressively measurable Self-similar Stationary Time-reversible
Limit theorems	Central limit theorem Donsker's theorem Doob's martingale convergence theorems Ergodic theorem Fisher–Tippett–Gnedenko theorem Large deviation principle Law of large numbers (weak/strong) Law of the iterated logarithm Maximal ergodic theorem Sanov's theorem Zero–one laws (Blumenthal, Borel–Cantelli, Engelbert–Schmidt, Hewitt–Savage, Kolmogorov, Lévy)
Inequalities	Burkholder–Davis–Gundy Doob's martingale Doob's upcrossing Kunita–Watanabe Marcinkiewicz–Zygmund
Tools	Cameron–Martin formula Convergence of random variables Doléans-Dade exponential Doob decomposition theorem Doob–Meyer decomposition theorem Doob's optional stopping theorem Dynkin's formula Feynman–Kac formula Filtration Girsanov theorem Infinitesimal generator Itô integral Itô's lemma Karhunen–Loève theorem Kolmogorov continuity theorem Kolmogorov extension theorem Lévy–Prokhorov metric Malliavin calculus Martingale representation theorem Optional stopping theorem Prokhorov's theorem Quadratic variation Reflection principle Skorokhod integral Skorokhod's representation theorem Skorokhod space Snell envelope Stochastic differential equation Tanaka Stopping time Stratonovich integral Uniform integrability Usual hypotheses Wiener space Classical Abstract
Disciplines	Actuarial mathematics Control theory Econometrics Ergodic theory Extreme value theory (EVT) Large deviations theory Mathematical finance Mathematical statistics Probability theory Queueing theory Renewal theory Ruin theory Signal processing Statistics Stochastic analysis Time series analysis Machine learning
List of topics Category