infinitely divisible distribution

infinitely divisible distribution
Name	Infinitely divisible distribution
Field	Probability theory
Notable examples	Normal distribution; Poisson distribution; Cauchy distribution; Gamma distribution; Stable distribution; Compound Poisson distribution; Negative binomial distribution; Log-normal distribution

infinitely divisible distribution An infinitely divisible distribution is a probability distribution on the real line that for every positive integer n can be expressed as the distribution of the sum of n independent identically distributed random variables. This concept underpins central results in Andrei Kolmogorov's and Paul Lévy's work on limit distributions and underlies the structure of Lévy processes and many stochastic models used in Alan Turing-era and contemporary applied probability. Infinitely divisible laws connect to classical limit theorems associated with Pafnuty Chebyshev and Simeon Denis Poisson and to modern developments in Émile Borel-style measure theory.

Definition and basic properties

A probability distribution μ on ℝ is infinitely divisible if for every n≥1 there exists a distribution μ_n such that μ is the n-fold convolution μ_n^{*n}. This property links to the concept of convolution semigroups studied by Andrey Kolmogorov, Paul Lévy, and Boris Gnedenko. Basic properties include closure under weak convergence, scaling relations found in the theory of Gennady Margulis-style ergodic limits, and preservation under convolution powers appearing in the work of William Feller and Kolmogorov. If X has an infinitely divisible law then for any t>0 there exists a distribution for X_t such that X has the law of the sum of t independent copies of X_1 when t is an integer, extending to continuous convolution semigroups examined by Kiyosi Itô and Kai Lai Chung.

Characteristic functions and Lévy–Khintchine representation

A distribution μ is infinitely divisible exactly when its characteristic function φ(θ)=E[e^{iθX}] admits the Lévy–Khintchine representation. The canonical form was established by Aleksandr Khinchin and Paul Lévy and is typically written as φ(θ)=exp{i a θ - 1/2 σ^2 θ^2 + ∫_{ℝ\{0\}} (e^{iθx}-1-iθx 1_{|x|<1}) ν(dx)}, where a∈ℝ, σ≥0, and ν is the Lévy measure. This representation is central to the classification program pursued by Kolmogorov and later refined by Kiyosi Itô and Ken-iti Sato. The parameters (a,σ,ν) provide a bijection between infinitely divisible laws and Lévy triplets, an idea used extensively by Paul Lévy in his studies of stable laws and by Igor Girsanov in stochastic calculus.

Examples and classes of infinitely divisible distributions

Prominent examples include the Normal distribution (Gaussian), the Poisson distribution, the Compound Poisson distribution, the Cauchy distribution, the family of Stable distributions, the Gamma distribution, the Inverse Gaussian distribution, and the Negative binomial distribution. The class of self-decomposable distributions studied by Olof Thorin and Lennart Bondesson sits inside the infinitely divisible laws and includes generalized gamma convolutions related to Gottfried E. Noether-style invariants. Compound structures investigated by William Feller and S. Bernstein yield mixtures and subordinated laws that remain infinitely divisible, a theme revisited by Takaaki Kajita and Ken-iti Sato in modern treatments.

Lévy processes and connections to infinite divisibility

A stochastic process with stationary independent increments, a Lévy process, has marginal distributions that are infinitely divisible. The canonical examples—Brownian motion (connected to Norbert Wiener), the Poisson process, and stable Lévy motions—were developed in work by Paul Lévy, Kiyosi Itô, and Salomon Bochner. Subordination, a technique due to J. L. Doob and S. Bochner, constructs new Lévy processes by random time change, preserving infinite divisibility. Lévy–Khintchine parameters determine sample path properties studied by Itô and Henry McKean, and hitting-time distributions link to problems explored by Andrei Kolmogorov and William Feller.

Factorization and decomposition results

Infinitely divisible laws admit factorization and decomposition theorems: the Lévy–Khintchine representation yields canonical decomposition into Gaussian, pure jump (compound Poisson), and small-jump components—ideas traced to Paul Lévy and formalized by Kiyosi Itô and A. N. Kolmogorov. Self-decomposability and urbanik classes were studied by J. Wojciechowski and I. E. Volkov with contributions from Olof Thorin; factorization results connect to the Krein–Milman framework explored by Mark Krein and David Milman in functional analysis contexts. Unique factorization up to convolution factors appears in results by Ken-iti Sato and Boris Oksendal.

Applications in probability and statistics

Infinitely divisible distributions appear in limit theorems (extensions of the Central Limit Theorem due to Paul Lévy and Andrey Kolmogorov), in queueing models analyzed by John Kingman, in actuarial science (collective risk models used by Harold Jeffreys-era practitioners), and in finance through models of returns and jumps developed by Robert C. Merton and Peter Carr. In signal processing and telecommunications, compound Poisson and stable laws are used in impulse modeling studied by Claude Shannon and Harry Nyquist. Statistical inference for Lévy-driven models engages methods from the work of Murray Rosenblatt, David Cox, and Bradley Efron.

Historical development and key contributors

The concept matured through contributions by Paul Lévy and Aleksandr Khinchin in the 1930s, with formal structure added by Andrey Kolmogorov and later extensions by Kiyosi Itô in stochastic calculus. Subsequent advances came from William Feller in probability foundations, Ken-iti Sato in modern Lévy process theory, and Olof Thorin in self-decomposable classes. Contemporary development involves researchers at institutions associated with Princeton University, University of Cambridge, University of Tokyo, and University of Oxford, with applications driven by interdisciplinary teams influenced by work at Bell Labs, Institute for Advanced Study, and Centre National de la Recherche Scientifique.

Category:Probability distributions