TheoremBase

Conditional Kernels of a Borel Probability Measure on a Polish Space Given a Borel Map: Existence, Uniqueness and Concentration on the Fibres

Given a Borel map from a complete separable metric space into a separable metric space, every Borel probability measure has a conditional kernel given the map, unique up to a null set of the image measure; almost every conditional measure is carried by its fibre, and integrals against the measure are iterated integrals against the kernel and the image measure.

Statement

In the setting of Measure Spaces and the Lebesgue Integral: Standing Notation, let (Z,d)(Z,d) be a complete and separable metric space and (Y,dY)(Y,d_{Y}) a separable metric space, with Borel σ\sigma-algebras B(Z)\mathcal{B}(Z) and B(Y)\mathcal{B}(Y); let q:Z→Yq:Z\to Y be measurable with respect to B(Z)\mathcal{B}(Z) and B(Y)\mathcal{B}(Y); let π\pi be a Borel measure on (Z,d)(Z,d) with π(Z)=1\pi(Z)=1, with image measure ν=q#π\nu=q_{\#}\pi; and let conditional kernels of π\pi given qq be those of The Conditional Kernel of a Probability Measure Given a Measurable Map §conditional-kernel, with κy=κ(y,⋅)\kappa_{y}=\kappa(y,\cdot). For y∈Yy\in Y the singleton {y}\{y\} is closed in (Y,dY)(Y,d_{Y}), since for w≠yw\ne y the open ball about ww of radius dY(w,y)d_{Y}(w,y) misses yy, hence belongs to B(Y)\mathcal{B}(Y) by claim 1 of Borel Measurability and Bounded Integration on a Metric Space, so the fibre q−1({y})q^{-1}(\{y\}) belongs to B(Z)\mathcal{B}(Z). For a measure λ\lambda and a function hh integrable with respect to it we also write ∫h(t) λ(dt)\int h(t)\,\lambda(dt) for ∫h dλ\int h\,d\lambda.

1. (Existence) There is a conditional kernel of π\pi given qq.

2. (Uniqueness) If κ\kappa and κ′\kappa' are conditional kernels of π\pi given qq, the set of the y∈Yy\in Y with κy≠κy′\kappa_{y}\ne\kappa'_{y} belongs to B(Y)\mathcal{B}(Y) and has ν\nu-measure 00.

3. (Fibres) If κ\kappa is a conditional kernel of π\pi given qq, the set of the y∈Yy\in Y with κy(q−1({y}))=1\kappa_{y}(q^{-1}(\{y\}))=1 belongs to B(Y)\mathcal{B}(Y) and has ν\nu-measure 11.

4. (Bounded functions) If κ\kappa is a conditional kernel of π\pi given qq and f:Z→Rf:Z\to\mathbb{R} is measurable with respect to B(Z)\mathcal{B}(Z) and bounded, then ff is integrable with respect to π\pi and to every κy\kappa_{y}, the function y↦∫Zf dκyy\mapsto\int_{Z}f\,d\kappa_{y} is measurable with respect to B(Y)\mathcal{B}(Y), bounded and integrable with respect to ν\nu, and

∫Zf dπ=∫Y(∫Zf dκy)ν(dy).\int_{Z}f\,d\pi=\int_{Y}\Bigl(\int_{Z}f\,d\kappa_{y}\Bigr)\nu(dy).

5. (Nonnegative functions) Let κ\kappa be a conditional kernel of π\pi given qq and let f:Z→Rf:Z\to\mathbb{R} be measurable with respect to B(Z)\mathcal{B}(Z), with f≥0f\ge0. Then the set NN of the y∈Yy\in Y for which ff is not integrable with respect to κy\kappa_{y} belongs to B(Y)\mathcal{B}(Y), the function g:Y→Rg:Y\to\mathbb{R} equal to ∫Zf dκy\int_{Z}f\,d\kappa_{y} for y∉Ny\notin N and to 00 for y∈Ny\in N is measurable with respect to B(Y)\mathcal{B}(Y), and ff is integrable with respect to π\pi if and only if ν(N)=0\nu(N)=0 and gg is integrable with respect to ν\nu; in that case ∫Zf dπ=∫Yg dν\int_{Z}f\,d\pi=\int_{Y}g\,d\nu.

Proofs

Log in to submit a proof.

Loading...

Citations

Loading…

Dependencies

Loading…

Related

0 relations

Curated associations between results. These are editable and subjective — they do not replace the dependency graph, which is derived from the references in the text.

No relations recorded yet.

Comments

Log in to comment.

Loading…