Appendix F

Disintegration of measures

[9/0%]
Show
LaTeX
Problem F.1

(repeated from Chapter 3) Let μ\mu be a sigma-finite measure on (T, B). Suppose h1h_{1} and h2h_{2} are functions in M+(T,B)\mathcal{M}^{+}(\mathcal{T}, \mathcal{B}) for which μ(g(t)h1(t))≤μ(g(t)h2(t))\mu \left(g(t) h_{1}(t)\right) \leq \mu \left(g(t) h_{2}(t)\right) for all gg in M+(T,B)\mathcal{M}^{+}(\mathcal{T}, \mathcal{B}). Show that h1≤h2h_{1} \leq h_{2} a.e. [ μ\mu ].

?
Problem F.2

Establish Lemma ⟨3⟩\langle 3 \rangle by the following steps. Write γ\gamma for the XX-marginal of Γ\Gamma.

Lemma 3: Let Γ\Gamma be a finite measure on A⊗B\mathcal{A} \otimes \mathcal{B} whose T\mathcal{T}-marginal is absolutely continuous with respect to the sigma-finite measure μ\mu on B\mathcal{B}. For f∈M+(X,A)f \in \mathcal{M}^{+}(X, \mathcal{A}), define the measure Γf\Gamma_{f} on B\mathcal{B} by Γfg:=Γx,t(f(x)g(t))\Gamma_{f} g: = \Gamma^{x, t}(f(x) g(t)) for g∈M+(T,B)g \in \mathcal{M}^{+}(\mathcal{T}, \mathcal{B}). Then the density λ(t,f):=dΓf/dμ\lambda (t, f): = d \Gamma_{f} / d \mu is well defined and determined uniquely up to a μ\mu-equivalence, and t↦λ(t,f)t \mapsto \lambda (t, f) has the following measure-like properties, for all α1,α2\alpha_{1}, \alpha_{2} in M+(T,B)\mathcal{M}^{+}(\mathcal{T}, \mathcal{B}) and all f,f1,f2,…f, f_{1}, f_{2}, \ldots in M+(X,A)\mathcal{M}^{+}(X, \mathcal{A}): (i) λ(t,α1f1+α2f2)=α1(t)λ(t,f1)+α2(t)λ(t,f2)\lambda \left(t, \alpha_{1} f_{1}+\alpha_{2} f_{2}\right) = \alpha_{1}(t) \lambda \left(t, f_{1}\right)+\alpha_{2}(t) \lambda \left(t, f_{2}\right) a.e. [μ][\mu ]; (ii) if f1≤f2f_{1} \leq f_{2} then λ(t,f1)≤λ(t,f2)\lambda \left(t, f_{1}\right) \leq \lambda \left(t, f_{2}\right) a.e. [μ][\mu ]; (iii) if 0≤f1≤f2≤…↑f0 \leq f_{1} \leq f_{2} \leq \ldots \uparrow f then λ(t,fn)↑λ(t,f)\lambda \left(t, f_{n}\right) \uparrow \lambda (t, f) a.e. [μ][\mu ].

?
(i)

For fixed ff in M+(X,A)\mathcal{M}^{+}(\mathcal{X}, \mathcal{A}) and each nn in N\mathbb {N}, write λn(t,f)\lambda_{n}(t, f) for any choice of the Radon-Nikodym derivative dΓn∧f/dμd \Gamma_{n \wedge f} / d \mu. That is, λn(t,f)\lambda_{n}(t, f) is a B\mathcal{B}-measurable function for which Γx,tg(t)(n∧f(x))=μtg(t)λn(t,f)\Gamma^{x, t} g(t)(n \wedge f(x)) = \mu^{t} g(t) \lambda_{n}(t, f) for all gg in M+(T,B)\mathcal{M}^{+}(\mathcal{T}, \mathcal{B}). Use Problem [1] to show that 0≤λn(t,f)↑0 \leq \lambda_{n}(t, f) \uparrow a.e. [ μ\mu ].

(ii)

Show that λ(t,f):=lim sup⁡λn(t,f)\lambda (t, f): = \limsup \lambda_{n}(t, f) is a density for Γf\Gamma_{f} with respect to μ\mu. Use Problem [1] to show that the density is unique up to a μ\mu-equivalence.

(iii)

For fixed α1,α2,g\alpha_{1}, \alpha_{2}, g in M+(T,B)\mathcal{M}^{+}(\mathcal{T}, \mathcal{B}) and f,f1,f2f, f_{1}, f_{2} in M+(X,A)\mathcal{M}^{+}(\mathcal{X}, \mathcal{A}), show that μt(g(t)α1(t)λ(t,f1))+μt(g(t)α2(t)λ(t,f2))=Γg(t)(α1(t)f1(x)+α2(t)f2(x))\mu^{t}\left(g(t) \alpha_{1}(t) \lambda \left(t, f_{1}\right)\right)+\mu^{t}\left(g(t) \alpha_{2}(t) \lambda \left(t, f_{2}\right)\right) = \Gamma g(t)\left(\alpha_{1}(t) f_{1}(x)+\alpha_{2}(t) f_{2}(x)\right). Deduce via (ii) that λ(t,α1f1+α2f2)=α1(t)λ(t,f1)+α2(t)λ(t,f2)\lambda \left(t, \alpha_{1} f_{1}+\alpha_{2} f_{2}\right) = \alpha_{1}(t) \lambda \left(t, f_{1}\right)+\alpha_{2}(t) \lambda \left(t, f_{2}\right) a.e. [ μ\mu ].

(iv)

If f1,f2∈M+(X,A)f_{1}, f_{2} \in \mathcal{M}^{+}(X, \mathcal{A}) and f1(x)≤f2(x)f_{1}(x) \leq f_{2}(x) a.e. [ γ\gamma ], show that μt(g(t)λ(t,f1))≤μt(g(t)λ(t,f1))\mu^{t}\left(g(t) \lambda \left(t, f_{1}\right)\right) \leq \mu^{t}\left(g(t) \lambda \left(t, f_{1}\right)\right) for all gg in M+(T,B)\mathcal{M}^{+}(\mathcal{T}, \mathcal{B}). Deduce that λ(t,f1)≤λ(t,f2)\lambda \left(t, f_{1}\right) \leq \lambda \left(t, f_{2}\right) a.e. [μ][\mu ].

(v)

If f1,f2,…∈M+(X,A)f_{1}, f_{2}, \ldots \in \mathcal{M}^{+}(\mathcal{X}, \mathcal{A}) and fn(x)↑f(x)f_{n}(x) \uparrow f(x) a.e. [ γ\gamma ], show that

μt(g(t)λ(t,fn))=Γ(g(t)fn(x))↑Γ(g(t)f(x))=μt(g(t)λ(t,f)) \mu ^{t}\left(g(t) \lambda \left(t, f_{n}\right)\right) = \Gamma \left(g(t) f_{n}(x)\right) \uparrow \Gamma (g(t) f(x)) = \mu ^{t}(g(t) \lambda (t, f))

for all gg in M+(T,B)\mathcal{M}^{+}(\mathcal{T}, \mathcal{B}). Deduce that λ(t,fn)↑λ(t,f)\lambda \left(t, f_{n}\right) \uparrow \lambda (t, f) a.e. [μ][\mu ].

Problem F.3

Let λ\lambda be a sigma-finite Radon measure on (X,A)(X, \mathcal{A}), with A\mathcal{A} the Borel sigma-field. Show that there exists a sequence of disjoint compact sets {Ki:i∈N}\left\{ K_{i}: i \in \mathbb {N}\right\} for which λ(∪iKi)c=0\lambda \left(\cup_{i} K_{i}\right)^{c} = 0.

?
Problem F.4

Let λ\lambda be a finite measure on the Borel sigma-field of a complete, separable metric space XX. Write K\mathcal{K} for the class of all compact subsets of XX.

?
(i)

Show that λ\lambda is tight: for each ϵ>0\epsilon > 0 there exists a KϵK_{\epsilon } in X\mathcal{X} such that λKϵc<ϵ\lambda K_{\epsilon }^{c} < \epsilon. Hint: For each positive integer nn, show that the space XX is a countable union of closed balls with radius 1/n1 / n. Find a finite family of such balls whose union BnB_{n} has λ\lambda measure greater than λX−ϵ/2n\lambda X-\epsilon / 2^{n}. Show that ∩nBn\cap_{n} B_{n} is compact, using the total-boundedness characterization of compact subsets of complete metric spaces (Simmons 1963, Section 25).

(ii)

Deduce that λB=sup⁡{λK:B⊇K∈X}\lambda B = \sup \left\{ \lambda K: B \supseteq K \in \mathcal{X}\right\} for each Borel set BB. Hint: Compare with the Problems to Chapter 2.

(iii)

Extend the result to sigma-finite measures λ\lambda on a complete, separable metric space for which λK<∞\lambda K < \infty for each compact KK.

Problem F.5

Let TT be an A\B\mathcal{A} \backslash \mathcal{B}-measurable map from (X,A)(X, \mathcal{A}) into (y,By, \mathcal{B}). Suppose B\mathcal{B} contains all singleton sets {y}\left\{ y\right\}, for y∈yy \in y, and is countably generated: B:=σ(B0)\mathcal{B}: = \sigma \left(\mathcal{B}_{0}\right) for some countable subclass B0\mathcal{B}_{0}. With no loss of generality you may assume that B0\mathcal{B}_{0} is stable under complements. Show that graph⁡(T)∈A⊗B\operatorname {graph}(T) \in \mathcal{A} \otimes \mathcal{B}, by following these steps.

?
(i)

Say that a set BB separates points y1y_{1} and y2y_{2} if either y1∈By_{1} \in B and y2∈Bcy_{2} \in B^{c} or y1∈Bcy_{1} \in B^{c} and y2∈By_{2} \in B. For fixed y1y_{1} and y2y_{2}, show that

B1:={B∈B:B does not separate y1 and y2} \mathcal{B}_{1}: = \left\{ B \in \mathcal{B}: B \text{ does not separate } y_{1} \text{ and } y_{2}\right\}

is a sigma-field. Deduce that B1\mathcal{B}_{1} cannot contain every B0\mathcal{B}_{0} set.

(ii)

If y≠Txy \neq T x, for some x∈Xx \in X and y∈yy \in y, show that (x,y)∈(T−1Bc)⊗B(x, y) \in \left(T^{-1} B^{c}\right) \otimes B for some B∈B0B \in \mathcal{B}_{0}.

(iii)

Deduce that graph⁡(T)c\operatorname {graph}(T)^{c} is a countable union of measurable rectangles.

Problem F.6

Suppose H:y→RH: y \rightarrow \mathbb {R} is one-to-one. Show that σ(H)\sigma (H) is countably generated and contains all the singleton sets.

?
Problem F.7

Suppose C\mathcal{C} is a countably generated sigma-field on a set yy, that contains all the singleton subsets. That is, C:=σ(C0)\mathcal{C}: = \sigma \left(\mathcal{C}_{0}\right), for some countable C0:={C1,C2,…}\mathcal{C}_{0}: = \left\{ C_{1}, C_{2}, \ldots \right\}. Define a real-valued, C -measurable function H(y):=∑i3−i{y∈Ci}H(y): = \sum_{i} 3^{-i}\left\{ y \in C_{i}\right\} on yy.

?
(i)

Show that HH is one-to-one. Hint: If H(y1)=H(y2)H\left(y_{1}\right) = H\left(y_{2}\right), with y1≠y2y_{1} \neq y_{2}, show that C\mathcal{C} cannot separate the points y1,y2y_{1}, y_{2}, which implies that {y1}∉C\left\{ y_{1}\right\} \notin \mathcal{C}.

(ii)

Equip the range R:=Hy\mathcal{R}: = H^{y} (which might not be a Borel set) with the trace sigma-field D:={BR:B∈B(R)}\mathcal{D}: = \left\{ B \mathcal{R}: B \in \mathcal{B}(\mathbb {R})\right\}. Show that the map G=H−1G = H^{-1} from R\mathcal{R} onto yy is D\C\mathcal{D} \backslash \mathcal{C}-measurable. Hint: Show that G−1CiG^{-1} C_{i} is the set of all points in R\mathcal{R} whose base- 3 expansions have a 1 in the ii th place.

(iii)

Show that C=σ(H)\mathrm{C} = \sigma (H).

Problem F.8

Suppose TT is a function from a set XX into a set T\mathcal{T} equipped with a sigma-field B\mathcal{B}. Suppose S:X→yS: X \rightarrow y is σ(T)\C\sigma (T) \backslash \mathcal{C}-measurable, where C\mathcal{C} is a countably generated sigmafield on yy containing all the singleton sets. Show that there exists a B\C\mathcal{B} \backslash \mathcal{C}-measurable map γ:T→y\gamma : \mathcal{T} \rightarrow y such that S=γ∘TS = \gamma \circ T.

?
Problem F.9

Let X0X_{0} be a countable dense subset of a compact metric space XX. Write Bˉ(x,r)\bar{B}(x, r) for the closed ball with radius rr and center xx. Define X1\mathcal{X}_{1} as the collection of all finite unions of balls Bˉ(x,r)\bar{B}(x, r) with rr rational and x∈X0x \in X_{0}. Write X0\mathcal{X}_{0} for the collection of all finite intersections of sets from X1\mathcal{X}_{1}.

?
(i)

Show that X0\mathcal{X}_{0} is a countable (∅,∪f,∩f\emptyset , \cup f, \cap f)-paving of compact subsets of X\mathcal{X}.

(ii)

For each closed subset FF of XX show that there exists a decreasing sequence of sets {Kn}\left\{ K_{n}\right\} from X0\mathcal{X}_{0} for which F=∩nKnF = \cap_{n} K_{n}. Hint: The union of the open balls B(x,1/n)B(x, 1 / n) with x∈X0∩Fx \in X_{0} \cap F covers the compact FF. Find a finite subcover. Consider the union of the corresponding closed balls.

(iii)

For each pair of closed subsets K1,K2K_{1}, K_{2} in X0\mathcal{X}_{0} for which K1⊇K2K_{1} \supseteq K_{2} show that there exists an increasing sequence of sets {Hn}\left\{ H_{n}\right\} from K0\mathcal{K}_{0} for which UnHn=K1\K2U_{n} H_{n} = K_{1} \backslash K_{2}. Hint: Cover the compact set {x∈K1:d(x,K2)≥2/n}\left\{ x \in K_{1}: d\left(x, K_{2}\right) \geq 2 / n\right\} by finitely many closed balls with radius 1/n1 / n, then intersect their union with K1K_{1}.