Markov kernel: Difference between revisions

Revision as of 15:06, 30 November 2022

Latest revision as of 14:25, 11 September 2024

(11 intermediate revisions by 8 users not shown)

Line 1:

{{Short description|Concept in probability theory}}

In [[probability theory]], a '''Markov kernel''' (also known as a '''stochastic kernel''' or '''probability kernel''') is a map that in the general theory of [[Markov process]]es plays the role that the [[Stochastic matrix|transition matrix]] does in the theory of Markov processes with a [[finite set|finite]] [[state space]].<ref>{{Cite book | last1 = Reiss | first1 = R. D. | title = A Course on Point Processes | doi = 10.1007/978-1-4613-9308-5 | series = Springer Series in Statistics | year = 1993 | isbn = 978-1-4613-9310-8 }}</ref>

== Formal definition ==

Let <math>(X,\mathcal A)</math> and <math>(Y,\mathcal B)</math> be [[measurable space]]s. A ''Markov kernel'' with source <math>(X,\mathcal A)</math> and target <math>(Y,\mathcal B)</math>, sometimes written as <math>\kappa:(X,\mathcal{A})\to(Y,\mathcal{B})</math>, is a mapfunction <math>\kappa : \mathcal B \times X \to [0,1]</math> with the following properties:

# For every (fixed) <math>BB_0 \in \mathcal B</math>, the map <math> x \mapsto \kappa(BB_0, x)</math> is <math>\mathcal A</math>-[[measurable function|measurable]]

# For every (fixed) <math> xx_0 \in X</math>, the map <math> B \mapsto \kappa(B, xx_0)</math> is a [[probability measure]] on <math>(Y, \mathcal B)</math>

In other words it associates to each point <math>x \in X</math> a [[probability measure]] <math>\kappa(dy|x): B \mapsto \kappa(B, x)</math> on <math>(Y,\mathcal B)</math> such that, for every measurable set <math>B\in\mathcal B</math>, the map <math>x\mapsto \kappa(B, x)</math> is measurable with respect to the [[Σ-algebra|<math>\sigma</math>-algebra <math>\mathcal A</math>]].<ref>{{cite book |last1=Klenke |first1=Achim |title=Probability Theory: A Comprehensive Course|series=Universitext |year=2014 |publisher=Springer|page=180|edition=2|doi=10.1007/978-1-4471-5361-0|isbn=978-1-4471-5360-3 }}</ref>

== Examples ==

Line 47 ⟶ 48:

===[[Galton–Watson process]]===

As a less obvious example, take <math>X = \N, \mathcal A = \mathcal P(\N)</math>, and <math>(Y, \mathcal B)</math> the real numbers <math>\R</math> with the standard sigma algebra of [[Borel set]]s. Then

:<math>\kappa(B|n)=\begin{cases} \mathbf{1}_B(0) & n=0\\ \Pr(\xi_1 + \cdots + \xi_x \in B) & n \neq 0 \\ \end{cases} </math>{{Clarify|reason=what is $x$?|date=May 2022}}

withwhere <math> x </math> is the number of element at the state <math> n </math>, <math>\xi_i</math> are [[Independent and identically distributed random variables|i.i.d.]] [[random variable]]s <math>\xi_i</math> (usually with mean 0) and where <math>\mathbf{1}_B</math> is the indicator function. For the simple case of [[Bernoulli distribution|coin flips]] this models the different levels of a [[Galton board]].

== Composition of Markov Kernels and the Markov Category==

Given measurable spaces <math>(X, \mathcal A)</math>, <math>(Y, \mathcal B) </math> we consider a Markov kernel <math>\kappa: \mathcal B \times X \to [0,1]</math> as a morphism <math>\kappa: X \to Y</math>. Intuitively, rather than assigning to each <math>x \in X</math> a sharply defined point <math> y \in Y</math> the kernel assigns a "fuzzy" point in <math>Y</math> which is only known with some level of uncertainty, much like actual physical measurements. If we have a third measurable space <math>(Z, \mathcal C)</math>, and probability kernels <math>\kappa: X \to Y</math> and <math>\lambda: Y \to Z</math>, we can define a composition <math>\lambda \circ \kappa : X \to Z</math> by the [[Chapman-Kolmogorov equation]]

:<math>(\lambda \circ \kappa) (dz|x) = \int_Y \lambda(dz | y)\kappa(dy|x)</math>.

The composition is associative by the Monotone Convergence Theorem and the identity function considered as a Markov kernel (i.e. the delta measure <math> \kappa_{1}(dx'|x) = \delta_x(dx')</math>) is the unit for this composition.

This composition defines the structure of a [[category (mathematics)|category]] on the measurable spaces with Markov kernels as morphisms, first defined by Lawvere.,<ref>{{cite web|author = F. W. Lawvere|title = The Category of Probabilistic Mappings|date = 1962|url = https://ncatlab.org/nlab/files/lawvereprobability1962.pdf}}</ref> Thethe [[category has the empty set as initial object and the one point set <math>*</math> as the terminal object. From this point of view a probability space <math>(\Omega, \mathcal A, \mathbb P)</math> is the same thing as a pointed space <math>* \to \Omega</math> in the [[Markov categorykernels]].

== Probability Space defined by Probability Distribution and a Markov Kernel==

ByA composition, of a probability space <math>(X, \mathcal A, P_X)</math> and a probability kernel <math>\kappa: (X, \mathcal A) \to (Y, \mathcal B) </math> defines a probability space <math>(Y, \mathcal B, P_Y = \kappa \circ P_X)</math>., Itwhere isthe concretelyprobability measure is definedgiven by

:<math> P_Y(B) = \int_X \int_B \kappa(dy|x) P_X(dx) = \int_X \kappa(B|x)P_X(dx) = \mathbb{E}_{P_X}\kappa(B|\cdot) .</math>

== Properties ==

Line 83 ⟶ 84:

can be any type of (non negative) measure, not necessarily a probability measure.

== External links ==

* [https://ncatlab.org/nlab/show/Markov+kernel Markov kernel] in [https://ncatlab.org/ nLab].

== References ==

Line 88 ⟶ 93:

* {{citation|first1=Heinz|last1=Bauer|title=Probability Theory|publisher=de Gruyter|year=1996|isbn=3-11-013935-9}}

:: §36. Kernels and semigroups of kernels

== See also ==

* [[Category of Markov kernels]]

[[Category:Markov processes]]