Infinite divisibility (probability)

In probability theory, a probability distribution is infinitely divisible if it can be expressed as the probability distribution of the sum of an arbitrary number of independent and identically distributed random variables. The characteristic function of any infinitely divisible distribution is then called an infinitely divisible characteristic function.^[1]

More rigorously, the probability distribution F is infinitely divisible if, for every positive integer n, there exist n independent identically distributed random variables X_n1, ..., X_nn whose sum S_n = X_n1 + … + X_nn has the distribution F.

The concept of infinite divisibility of probability distributions was introduced in 1929 by Bruno de Finetti. This type of decomposition of a distribution is used in probability and statistics to find families of probability distributions that might be natural choices for certain models or applications. Infinitely divisible distributions play an important role in probability theory in the context of limit theorems.^[1]

Examples

The Poisson distribution, the negative binomial distribution, the Gamma distribution and the degenerate distribution are examples of infinitely divisible distributions; as are the normal distribution, Cauchy distribution and all other members of the stable distribution family. The uniform distribution and the binomial distribution are not infinitely divisible.^[2] The Student's t-distribution is infinitely divisible, while the distribution of the reciprocal of a random variable having a Student's t-distribution, is not.^[3]

All the Compound Poisson distributions are infinitely divisible, but inverse proposition is not true.

Limit theorem

Infinitely divisible distributions appear in a broad generalization of the central limit theorem: the limit as n → +∞ of the sum S_n = X_n1 + … X_nn of independent uniformly asymptotically negligible (u.a.n.) random variables within a triangular array

\begin{array}{cccc} X_{11} \\ X_{21} & X_{22} \\ X_{31} & X_{32} & X_{33} \\ \vdots & \vdots & \vdots & \ddots \end{array}

approaches — in the weak sense — an infinitely divisible distribution. The uniformly asymptotically negligible (u.a.n.) condition is given by

\lim_{n\to\infty} \, \max_{1 \le k \le n} \; P( \left| X_{nk} \right| > \varepsilon ) = 0 \text{ for every }\varepsilon > 0.

Thus, for example, if the uniform asymptotic negligibility (u.a.n.) condition is satisfied via an appropriate scaling of identically distributed random variables with finite variance, the weak convergence is to the normal distribution in the classical version of the central limit theorem. More generally, if the u.a.n. condition is satisfied via a scaling of identically distributed random variables (with not necessarily finite second moment), then the weak convergence is to a stable distribution. On the other hand, for a triangular array of independent (unscaled) Bernoulli random variables where the u.a.n. condition is satisfied through

\lim_{n\rightarrow\infty} np_n = \lambda,

the weak convergence of the sum is to the Poisson distribution with mean λ as shown by the familiar proof of the law of small numbers.

Lévy process

Main article: Lévy process

Every infinitely divisible probability distribution corresponds in a natural way to a Lévy process. A Lévy process is a stochastic process { L_t : t ≥ 0 } with stationary independent increments, where stationary means that for s < t, the probability distribution of L_t − L_s depends only on t − s and where independent increments means that that difference L_t − L_s is independent of the corresponding difference on any interval not overlapping with [s, t], and similarly for any finite number of mutually non-overlapping intervals.

If { L_t : t ≥ 0 } is a Lévy process then, for any t ≥ 0, the random variable L_t will be infinitely divisible: for any n, we can choose (X_n0, X_n1, …, X_nn) = (L_t/n − L₀, L_2t/n − L_t/n, …, L_t − L_(n-1)t/n). Similarly, L_t − L_s is infinitely divisible for any s < t.

On the other hand, if F is an infinitely divisible distribution, we can construct a Lévy process { L_t : t ≥ 0 } from it. For any interval [s, t] where t − s > 0 equals a rational number p/q, we can define L_t − L_s to have the same distribution as X_q1 + X_q2 + … + X_qp. Irrational values of t − s > 0 are handled via a continuity argument.

Footnotes

1 2 Lukacs, E. (1970) Characteristic Functions, Griffin , London. p. 107
↑ Sato, Ken-iti (1999). Lévy Processes and Infinitely Divisible Distributions. Cambridge University Press. p. 31. ISBN 978-0-521-55302-5.
↑ Johnson, N.L., Kotz, S., Balakrishnan, N. (1995) Continuous Univariate Distributions, Volume 2, 2nd Edition. Wiley, ISBN 0-471-58494-0 (Chapter 28, page 368)

References

Domínguez-Molina, J.A.; Rocha-Arteaga, A. (2007) "On the Infinite Divisibility of some Skewed Symmetric Distributions". Statistics and Probability Letters, 77 (6), 644–648 doi:10.1016/j.spl.2006.09.014

Steutel, F. W. (1979), "Infinite Divisibility in Theory and Practice" (with discussion), Scandinavian Journal of Statistics. 6, 57–64.

Steutel, F. W. and Van Harn, K. (2003), Infinite Divisibility of Probability Distributions on the Real Line (Marcel Dekker).

Probability distributions

Discrete univariate with finite support

Benford Bernoulli Beta-binomial binomial categorical hypergeometric Poisson binomial Rademacher discrete uniform Zipf Zipf–Mandelbrot

Discrete univariate with infinite support

beta negative binomial Borel Conway–Maxwell–Poisson discrete phase-type Delaporte extended negative binomial Gauss–Kuzmin geometric logarithmic negative binomial parabolic fractal Poisson Skellam Yule–Simon zeta

Continuous univariate supported on a bounded interval, e.g. [0,1]

Arcsine ARGUS Balding–Nichols Bates Beta Beta rectangular Irwin–Hall Kumaraswamy logit-normal Noncentral beta raised cosine Reciprocal Triangular U-quadratic uniform Wigner semicircle

[[List of probability distributions#Supported_on_semi-infinite_intervals.2C_usually_.5B0.2C.E2.88.9E.29|Continuous univariate supported on a semi-infinite interval, usually [0,∞)]]

Continuous univariate supported on the whole real line (−∞, ∞)

Cauchy exponential power Fisher's z generalized normal generalized hyperbolic geometric stable Gumbel Holtsmark hyperbolic secant Johnson S_U Landau Laplace Asymmetric Laplace Linnik logistic noncentral t normal (Gaussian) normal-inverse Gaussian skew normal slash stable Student's t type-1 Gumbel Tracy–Widom variance-gamma Voigt

Continuous univariate with support whose type varies

generalized extreme value generalized Pareto Tukey lambda q-Gaussian q-exponential q-Weibull shifted log-logistic

Mixed continuous-discrete univariate distributions

rectified Gaussian

Multivariate (joint)

Discrete Ewens multinomial Dirichlet-multinomial negative multinomial Continuous Dirichlet Generalized Dirichlet multivariate normal Multivariate stable multivariate Student normal-scaled inverse gamma normal-gamma Matrix-valued inverse matrix gamma inverse-Wishart matrix normal matrix t matrix gamma normal-inverse-Wishart normal-Wishart Wishart

Directional

Univariate (circular) directional Circular uniform univariate von Mises wrapped normal wrapped Cauchy wrapped exponential wrapped asymmetric Laplace wrapped Lévy Bivariate (spherical) Kent Bivariate (toroidal) bivariate von Mises Multivariate von Mises–Fisher Bingham

Degenerate and singular

Degenerate discrete degenerate Dirac delta function Singular Cantor

Families

Circular compound Poisson elliptical exponential natural exponential location-scale maximum entropy mixture Pearson Tweedie wrapped

This article is issued from Wikipedia - version of the Tuesday, January 06, 2015. The text is available under the Creative Commons Attribution/Share Alike but additional terms may apply for the media files.