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 (S,𝒮), (T,𝒯) be two measurable spaces. A function

κ:S×𝒯→[0,+∞]

is called a (transition) kernel from S to T if the following two conditions hold:

s↦κ(s,B)
is 𝒮/ℬ([0,+∞])-measurable;
B↦κ(s,B)
is a measure on (T,𝒯).

Classification of transition kernels

Transition kernels are usually classified by the measures they define. Those measures are defined as

κs:𝒯→[0,+∞]

with

κs(B)=κ(s,B)

for all B∈𝒯 and all s∈S. With this notation, the kernel κ is called

Operations

In this section, let (S,𝒮), (T,𝒯) and (U,𝒰) be measurable spaces and denote the product σ-algebra of 𝒮 and 𝒯 with 𝒮⊗𝒯

Product of kernels

Definition

Let κ1 be a s-finite kernel from S to T and κ2 be a s-finite kernel from S×T to U. Then the product κ1⊗κ2 of the two kernels is defined as

κ1⊗κ2:S×(𝒯⊗𝒰)→[0,∞]
κ1⊗κ2(s,A)=∫Tκ1(s,dt)∫Uκ2((s,t),du)𝟏A(t,u)

for all A∈𝒯⊗𝒰.

Properties and comments

The product of two kernels is a kernel from S to T×U. It is again a s-finite kernel and is a σ-finite kernel if κ1 and κ2 are σ-finite kernels. The product of kernels is also associative, meaning it satisfies

(κ1⊗κ2)⊗κ3=κ1⊗(κ2⊗κ3)

for any three suitable s-finite kernels κ1,κ2,κ3.

The product is also well-defined if κ2 is a kernel from T to U. In this case, it is treated like a kernel from S×T to U that is independent of S. This is equivalent to setting

κ((s,t),A):=κ(t,A)

for all A∈𝒰 and all s∈S.

Composition of kernels

Definition

Let κ1 be a s-finite kernel from S to T and κ2 a s-finite kernel from S×T to U. Then the composition κ1⋅κ2 of the two kernels is defined as

κ1⋅κ2:S×𝒰→[0,∞]
(s,B)↦∫Tκ1(s,dt)∫Uκ2((s,t),du)𝟏B(u)

for all s∈S and all B∈𝒰.

Properties and comments

The composition is a kernel from S to U that is again s-finite. The composition of kernels is associative, meaning it satisfies

(κ1⋅κ2)⋅κ3=κ1⋅(κ2⋅κ3)

for any three suitable s-finite kernels κ1,κ2,κ3. Just like the product of kernels, the composition is also well-defined if κ2 is a kernel from T to U.

An alternative notation is for the composition is κ1κ2

Kernels as operators

Let 𝒯+,𝒮+ be the set of positive measurable functions on (S,𝒮),(T,𝒯).

Every kernel κ from S to T can be associated with a linear operator

Aκ:𝒯+→𝒮+

given by

(Aκf)(s)=∫Tκ(s,dt)f(t).

The composition of these operators is compatible with the composition of kernels, meaning

Aκ1Aκ2=Aκ1⋅κ2