Kernel Transformation of Measure is Measure
Theorem
Let $\struct {X, \Sigma, \mu}$ be a measure space.
Let $N: X \times \Sigma \to \overline \R_{\ge0}$ be a kernel.
Then $\mu N: X \to \overline \R$, the kernel transformation of $\mu$, is a measure.
Proof
From the definition of the kernel transformation of $\mu$, we have:
- $\ds \map {\paren {\mu N} } E = \int \map N {x, E} \rd \map \mu x$
for each $E \in \Sigma$.
We verify each of the conditions for a measure in turn.
Proof of $(1)$
Let $E \in \Sigma$.
From the definition of a kernel, we have that:
- $x \mapsto \map N {x, E}$ is a positive $\Sigma$-measurable function.
From the definition of the $\mu$-integral of a positive $\Sigma$-measurable function, we have:
- $\ds \int \map N {x, E} \rd \map \mu x \ge 0$
So:
- $\map {\paren {\mu N} } E \ge 0$
for each $E \in \Sigma$, verifying $(1)$.
$\Box$
Proof of $(2)$
Let $\sequence {E_n}_{n \mathop \in \N}$ be a sequence of pairwise disjoint $\Sigma$-measurable sets.
Then:
\(\ds \map {\paren {\mu N} } {\bigcup_{n \mathop = 1}^\infty E_n}\) | \(=\) | \(\ds \int \map N {x, \bigcup_{n \mathop = 1}^\infty E_n} \rd \map \mu x\) | ||||||||||||
\(\ds \) | \(=\) | \(\ds \int \paren {\sum_{n \mathop = 1}^\infty \map N {x, E_n} } \rd \map \mu x\) | since for each $x \in X$, $E \mapsto \map N {x, E}$ is a measure, we use countable additivity in the second argument | |||||||||||
\(\ds \) | \(=\) | \(\ds \sum_{n \mathop = 1}^\infty \paren {\int \map N {x, E_n} \rd \map \mu x}\) | Integral of Series of Positive Measurable Functions | |||||||||||
\(\ds \) | \(=\) | \(\ds \sum_{n \mathop = 1}^\infty \map {\paren {\mu N} } {E_n}\) |
So $\mu N$ is countably additive, and we have $(2)$.
$\Box$
Proof of $(3)$
We have:
\(\ds \map {\paren {\mu N} } \O\) | \(=\) | \(\ds \int \map N {x, \O} \rd \map \mu x\) | ||||||||||||
\(\ds \) | \(=\) | \(\ds \int 0 \rd \map \mu x\) | Empty Set is Null Set | |||||||||||
\(\ds \) | \(=\) | \(\ds 0\) | Measurable Function Zero A.E. iff Absolute Value has Zero Integral | |||||||||||
\(\ds \) | \(\in\) | \(\ds \R\) |
So there exists $E \in \Sigma$ such that:
- $\map {\paren {\mu N} } E \in \R$
verifying $(3)$.
$\blacksquare$
Sources
- 2005: René L. Schilling: Measures, Integrals and Martingales ... (previous) ... (next): $\S 9$: Problem $11 \ \text{(i)}$