Transition kernel
In the mathematics of probability, a transition kernel or kernel is a function in mathematics that has different applications. Kernels can for example be used to define random measures or stochastic processes. The most important example of kernels are the Markov kernels.
Definition
Let [math]\displaystyle{ (S, \mathcal S) }[/math], [math]\displaystyle{ (T, \mathcal T) }[/math] be two measurable spaces. A function
- [math]\displaystyle{ \kappa \colon S \times \mathcal T \to [0, +\infty] }[/math]
is called a (transition) kernel from [math]\displaystyle{ S }[/math] to [math]\displaystyle{ T }[/math] if the following two conditions hold:[1]
- For any fixed [math]\displaystyle{ B \in \mathcal T }[/math], the mapping
- [math]\displaystyle{ s \mapsto \kappa(s,B) }[/math]
- is [math]\displaystyle{ \mathcal S/ \mathcal B([0, +\infty]) }[/math]-measurable;
- For every fixed [math]\displaystyle{ s \in S }[/math], the mapping
- [math]\displaystyle{ B \mapsto \kappa(s, B) }[/math]
- is a measure on [math]\displaystyle{ (T, \mathcal T) }[/math].
Classification of transition kernels
Transition kernels are usually classified by the measures they define. Those measures are defined as
- [math]\displaystyle{ \kappa_s \colon \mathcal T \to [0, + \infty] }[/math]
with
- [math]\displaystyle{ \kappa_s(B)=\kappa(s,B) }[/math]
for all [math]\displaystyle{ B \in \mathcal T }[/math] and all [math]\displaystyle{ s \in S }[/math]. With this notation, the kernel [math]\displaystyle{ \kappa }[/math] is called[1][2]
- a substochastic kernel, sub-probability kernel or a sub-Markov kernel if all [math]\displaystyle{ \kappa_s }[/math] are sub-probability measures
- a Markov kernel, stochastic kernel or probability kernel if all [math]\displaystyle{ \kappa_s }[/math] are probability measures
- a finite kernel if all [math]\displaystyle{ \kappa_s }[/math] are finite measures
- a [math]\displaystyle{ \sigma }[/math]-finite kernel if all [math]\displaystyle{ \kappa_s }[/math] are [math]\displaystyle{ \sigma }[/math]-finite measures
- a s-finite kernel if all [math]\displaystyle{ \kappa_s }[/math] are [math]\displaystyle{ s }[/math]-finite measures, meaning it is a kernel that can be written as a countable sum of finite kernels
- a uniformly [math]\displaystyle{ \sigma }[/math]-finite kernel if there are at most countably many measurable sets [math]\displaystyle{ B_1, B_2, \dots }[/math] in [math]\displaystyle{ T }[/math] with [math]\displaystyle{ \kappa_s(B_i) \lt \infty }[/math] for all [math]\displaystyle{ s \in S }[/math] and all [math]\displaystyle{ i \in \N }[/math].
Operations
In this section, let [math]\displaystyle{ (S, \mathcal S) }[/math], [math]\displaystyle{ (T, \mathcal T) }[/math] and [math]\displaystyle{ (U, \mathcal U) }[/math] be measurable spaces and denote the product σ-algebra of [math]\displaystyle{ \mathcal S }[/math] and [math]\displaystyle{ \mathcal T }[/math] with [math]\displaystyle{ \mathcal S \otimes \mathcal T }[/math]
Product of kernels
Definition
Let [math]\displaystyle{ \kappa^1 }[/math] be a s-finite kernel from [math]\displaystyle{ S }[/math] to [math]\displaystyle{ T }[/math] and [math]\displaystyle{ \kappa^2 }[/math] be a s-finite kernel from [math]\displaystyle{ S \times T }[/math] to [math]\displaystyle{ U }[/math]. Then the product [math]\displaystyle{ \kappa^1 \otimes \kappa^2 }[/math] of the two kernels is defined as[3][4]
- [math]\displaystyle{ \kappa^1 \otimes \kappa^2 \colon S \times (\mathcal T \otimes \mathcal U) \to [0, \infty] }[/math]
- [math]\displaystyle{ \kappa^1 \otimes \kappa^2(s,A)= \int_T \kappa^1(s, \mathrm d t) \int_U \kappa^2((s,t), \mathrm du) \mathbf 1_A(t,u) }[/math]
for all [math]\displaystyle{ A \in \mathcal T \otimes \mathcal U }[/math].
Properties and comments
The product of two kernels is a kernel from [math]\displaystyle{ S }[/math] to [math]\displaystyle{ T \times U }[/math]. It is again a s-finite kernel and is a [math]\displaystyle{ \sigma }[/math]-finite kernel if [math]\displaystyle{ \kappa^1 }[/math] and [math]\displaystyle{ \kappa^2 }[/math] are [math]\displaystyle{ \sigma }[/math]-finite kernels. The product of kernels is also associative, meaning it satisfies
- [math]\displaystyle{ (\kappa^1 \otimes \kappa^2) \otimes \kappa^3= \kappa^1 \otimes (\kappa^2\otimes \kappa^3) }[/math]
for any three suitable s-finite kernels [math]\displaystyle{ \kappa^1,\kappa^2,\kappa^3 }[/math].
The product is also well-defined if [math]\displaystyle{ \kappa^2 }[/math] is a kernel from [math]\displaystyle{ T }[/math] to [math]\displaystyle{ U }[/math]. In this case, it is treated like a kernel from [math]\displaystyle{ S \times T }[/math] to [math]\displaystyle{ U }[/math] that is independent of [math]\displaystyle{ S }[/math]. This is equivalent to setting
- [math]\displaystyle{ \kappa((s,t),A):= \kappa(t,A) }[/math]
for all [math]\displaystyle{ A \in \mathcal U }[/math] and all [math]\displaystyle{ s \in S }[/math].[4][3]
Composition of kernels
Definition
Let [math]\displaystyle{ \kappa^1 }[/math] be a s-finite kernel from [math]\displaystyle{ S }[/math] to [math]\displaystyle{ T }[/math] and [math]\displaystyle{ \kappa^2 }[/math] a s-finite kernel from [math]\displaystyle{ S \times T }[/math] to [math]\displaystyle{ U }[/math]. Then the composition [math]\displaystyle{ \kappa^1 \cdot \kappa^2 }[/math] of the two kernels is defined as[5][3]
- [math]\displaystyle{ \kappa^1 \cdot \kappa^2 \colon S \times \mathcal U \to [0, \infty] }[/math]
- [math]\displaystyle{ (s, B) \mapsto \int_T \kappa^1(s, \mathrm dt) \int_U \kappa^2((s,t), \mathrm du) \mathbf 1_B(u) }[/math]
for all [math]\displaystyle{ s \in S }[/math] and all [math]\displaystyle{ B \in \mathcal U }[/math].
Properties and comments
The composition is a kernel from [math]\displaystyle{ S }[/math] to [math]\displaystyle{ U }[/math] that is again s-finite. The composition of kernels is associative, meaning it satisfies
- [math]\displaystyle{ (\kappa^1 \cdot \kappa^2) \cdot \kappa^3= \kappa^1 \cdot (\kappa^2 \cdot \kappa^3) }[/math]
for any three suitable s-finite kernels [math]\displaystyle{ \kappa^1,\kappa^2,\kappa^3 }[/math]. Just like the product of kernels, the composition is also well-defined if [math]\displaystyle{ \kappa^2 }[/math] is a kernel from [math]\displaystyle{ T }[/math] to [math]\displaystyle{ U }[/math].
An alternative notation is for the composition is [math]\displaystyle{ \kappa^1 \kappa^2 }[/math][3]
Kernels as operators
Let [math]\displaystyle{ \mathcal T^+, \mathcal S^+ }[/math] be the set of positive measurable functions on [math]\displaystyle{ (S, \mathcal S), (T, \mathcal T) }[/math].
Every kernel [math]\displaystyle{ \kappa }[/math] from [math]\displaystyle{ S }[/math] to [math]\displaystyle{ T }[/math] can be associated with a linear operator
- [math]\displaystyle{ A_\kappa \colon \mathcal T^+ \to \mathcal S^+ }[/math]
given by[6]
- [math]\displaystyle{ (A_\kappa f)(s)= \int_T \kappa (s, \mathrm dt)\; f(t). }[/math]
The composition of these operators is compatible with the composition of kernels, meaning[3]
- [math]\displaystyle{ A_{\kappa^1} A_{\kappa^2}= A_{\kappa^1 \cdot \kappa^2} }[/math]
References
- ↑ 1.0 1.1 Klenke, Achim (2008). Probability Theory. Berlin: Springer. p. 180. doi:10.1007/978-1-84800-048-3. ISBN 978-1-84800-047-6. https://archive.org/details/probabilitytheor00klen_341.
- ↑ Kallenberg, Olav (2017). Random Measures, Theory and Applications. Switzerland: Springer. p. 30. doi:10.1007/978-3-319-41598-7. ISBN 978-3-319-41596-3.
- ↑ 3.0 3.1 3.2 3.3 3.4 Kallenberg, Olav (2017). Random Measures, Theory and Applications. Switzerland: Springer. p. 33. doi:10.1007/978-3-319-41598-7. ISBN 978-3-319-41596-3.
- ↑ 4.0 4.1 Klenke, Achim (2008). Probability Theory. Berlin: Springer. p. 279. doi:10.1007/978-1-84800-048-3. ISBN 978-1-84800-047-6. https://archive.org/details/probabilitytheor00klen_341.
- ↑ Klenke, Achim (2008). Probability Theory. Berlin: Springer. p. 281. doi:10.1007/978-1-84800-048-3. ISBN 978-1-84800-047-6. https://archive.org/details/probabilitytheor00klen_341.
- ↑ Kallenberg, Olav (2017). Random Measures, Theory and Applications. Switzerland: Springer. pp. 29-30. doi:10.1007/978-3-319-41598-7. ISBN 978-3-319-41596-3.
 |  |
