TheoremBase

Build a countable algebra of balls and Ulam compacts generating the Borel sets, take Radon-Nikodym densities of B -> pi(. x B) on it, and show that off a null set they form a premeasure, countably additive by compact inner approximation; extend it and verify the kernel identity by a pi-lambda argument. Uniqueness follows from uniqueness of densities on the countable generating pi-system.

Proof

Each result cited is universally quantified over the data in its own statement.

Throughout, Td\mathcal{T}_{d} is the collection of the subsets of ZZ that are open in (Z,d)(Z,d), a topology on ZZ by Metric Open Sets Form a Topology, and "open", "closed" and "compact" for subsets of ZZ refer to (Z,Td)(Z,\mathcal{T}_{d}). We write Q>0\mathbb{Q}_{>0} for the set of positive rational numbers, Bd(q,s)B_{d}(q,s) for the open ball (open ball) and Bˉd(q,s)={z∈Z:d(q,z)≤s}\bar{B}_{d}(q,s)=\{z\in Z:d(q,z)\le s\} for the closed ball. For A∈YA\in\mathcal{Y} and B∈B(Z)B\in\mathcal{B}(Z) the rectangle A×BA\times B belongs to Y⊗B(Z)\mathcal{Y}\otimes\mathcal{B}(Z) by Product Sigma-Algebra. Since π(Y×Z)=1\pi(Y\times Z)=1, every value of π\pi lies in [0,1][0,1] by Basic Properties of a Measure §monotone; the same holds for every probability measure used below. "Measurable" for a real-valued function on YY means measurable with respect to Y\mathcal{Y} and B(R)\mathcal{B}(\mathbb{R}), and "a.e." and "null" refer to μ\mu.

Part 1: existence.

Step 0 (the second marginal). The projection (y,z)↦z(y,z)\mapsto z is measurable with respect to Y⊗B(Z)\mathcal{Y}\otimes\mathcal{B}(Z) and B(Z)\mathcal{B}(Z) by claim 5 of Sections of Product-Measurable Sets and Maps Are Measurable, and Insertion Maps into Products Are Measurable, so by claim 1 of Image Measures, Measures with Densities, and Change of Variables the formula ρ(B)=π(Y×B)\rho(B)=\pi(Y\times B), for B∈B(Z)B\in\mathcal{B}(Z), defines a measure ρ\rho on (Z,B(Z))(Z,\mathcal{B}(Z)) with ρ(Z)=1\rho(Z)=1; it is a Borel measure on (Z,d)(Z,d). Moreover Z≠∅Z\neq\varnothing, since otherwise Y×Z=∅Y\times Z=\varnothing and π(Y×Z)=0\pi(Y\times Z)=0.

Step 1 (an increasing sequence of compact sets). For each m∈Nm\in\mathbb{N}, Ulam's Theorem: a Finite Borel Measure on a Complete Separable Metric Space is Tight §tight, applied to the finite Borel measure ρ\rho on the complete separable metric space (Z,d)(Z,d) with ε=1/m\varepsilon=1/m, shows that the set Am\mathcal{A}_{m} of the compact sets L⊆ZL\subseteq Z with ρ(Z∖L)≤1/m\rho(Z\setminus L)\le 1/m is nonempty. By Axiom of Countable Choice, applied to the family (Am)m∈N(\mathcal{A}_{m})_{m\in\mathbb{N}} of nonempty subsets of the set of all subsets of ZZ, there is a sequence (Lm)m∈N(L_{m})_{m\in\mathbb{N}} with Lm∈AmL_{m}\in\mathcal{A}_{m} for every mm; that is, each LmL_{m} is compact and ρ(Z∖Lm)≤1/m\rho(Z\setminus L_{m})\le 1/m. Put Km=L1∪⋯∪LmK_{m}=L_{1}\cup\dots\cup L_{m}. Each KmK_{m} is compact: if (Oi)i∈I(O_{i})_{i\in I} is a family of open sets with Km⊆⋃i∈IOiK_{m}\subseteq\bigcup_{i\in I}O_{i}, then for each j≤mj\le m the criterion Compact Subset Criterion via Open Covers in the Ambient Space gives a finite Jj⊆IJ_{j}\subseteq I with Lj⊆⋃i∈JjOiL_{j}\subseteq\bigcup_{i\in J_{j}}O_{i}; the set J=J1∪⋯∪JmJ=J_{1}\cup\dots\cup J_{m} is finite by claim 3 of Peeling an Element off a Finite Set, and Unions of Finite Sets and induction, and Km⊆⋃i∈JOiK_{m}\subseteq\bigcup_{i\in J}O_{i}, so KmK_{m} is compact by the same criterion. By Compact Subsets of a Metric Space are Closed and Borel §closed and Compact Subsets of a Metric Space are Closed and Borel §borel each KmK_{m} is closed and belongs to B(Z)\mathcal{B}(Z). Clearly Km⊆Km+1K_{m}\subseteq K_{m+1}, and ρ(Z∖Km)≤ρ(Z∖Lm)≤1/m\rho(Z\setminus K_{m})\le\rho(Z\setminus L_{m})\le 1/m by Basic Properties of a Measure §monotone. Let K∞=⋃mKm∈B(Z)K_{\infty}=\bigcup_{m}K_{m}\in\mathcal{B}(Z). For every mm we have ρ(Z∖K∞)≤ρ(Z∖Km)≤1/m\rho(Z\setminus K_{\infty})\le\rho(Z\setminus K_{m})\le 1/m, so ρ(Z∖K∞)=0\rho(Z\setminus K_{\infty})=0, since a positive real number exceeds some 1/m1/m by claim 3 of The Archimedean Property of the Real Numbers. Consequently, by Basic Properties of a Measure §monotone,

π(A×(Z∖K∞))≤π(Y×(Z∖K∞))=ρ(Z∖K∞)=0(A∈Y).(1)\pi\bigl(A\times(Z\setminus K_{\infty})\bigr)\le\pi\bigl(Y\times(Z\setminus K_{\infty})\bigr)=\rho(Z\setminus K_{\infty})=0\qquad(A\in\mathcal{Y}).\tag{1}

Step 2 (a countable algebra generating B(Z)\mathcal{B}(Z)). Since (Z,d)(Z,d) is separable, fix a countable set D⊆ZD\subseteq Z that is dense in (Z,Td)(Z,\mathcal{T}_{d}). It is nonempty: pick z∈Zz\in Z (Step 0); then zz lies in the closure of DD, which is ZZ, and ZZ is an open set containing zz, so Z∩D≠∅Z\cap D\ne\varnothing by Closure of a Subset of a Topological Space. Hence ((Z,d),D)\bigl((Z,d),D\bigr) is a separable metric datum in the sense of Borel Sets and Measurable Maps in a Separable Metric Space; let E\mathcal{E} be the family of the open balls Bd(q,s)B_{d}(q,s) with q∈Dq\in D and s∈Q>0s\in\mathbb{Q}_{>0}, which is countable by claim 1 of that lemma. Let Gcl\mathcal{G}_{\mathrm{cl}} be the family of the closed balls Bˉd(q,s)\bar{B}_{d}(q,s) with q∈Dq\in D and s∈Q>0s\in\mathbb{Q}_{>0}. The set Q>0\mathbb{Q}_{>0} is countable by claim 2 of The Integers and the Rational Numbers are Countable and claim 3 of Basic Properties of Countable Sets, so D×Q>0D\times\mathbb{Q}_{>0} is countable by claim 1 of Products and Powers of Countable Sets, and Gcl\mathcal{G}_{\mathrm{cl}}, its image under (q,s)↦Bˉd(q,s)(q,s)\mapsto\bar{B}_{d}(q,s), is countable by claim 4 of Basic Properties of Countable Sets. The family Gcp={Km:m∈N}\mathcal{G}_{\mathrm{cp}}=\{K_{m}:m\in\mathbb{N}\}, the image of N\mathbb{N} under m↦Kmm\mapsto K_{m}, is countable by claims 1 and 4 of Basic Properties of Countable Sets. Let G=E∪Gcl∪Gcp\mathcal{G}=\mathcal{E}\cup\mathcal{G}_{\mathrm{cl}}\cup\mathcal{G}_{\mathrm{cp}}. Putting X1=E\mathcal{X}_{1}=\mathcal{E}, X2=Gcl\mathcal{X}_{2}=\mathcal{G}_{\mathrm{cl}} and Xi=Gcp\mathcal{X}_{i}=\mathcal{G}_{\mathrm{cp}} for i≥3i\ge3, we obtain a family (Xi)i∈N(\mathcal{X}_{i})_{i\in\mathbb{N}} of countable subsets of the set of all subsets of ZZ whose union is G\mathcal{G}, so G\mathcal{G} is countable by A Countable Union of Countable Sets is Countable. It is nonempty, as K1∈GK_{1}\in\mathcal{G}, so by Countable Set there is a sequence (Gi)i∈N(G_{i})_{i\in\mathbb{N}} whose set of terms is G\mathcal{G}.

For a set G⊆ZG\subseteq Z write G1=GG^{1}=G and G0=Z∖GG^{0}=Z\setminus G. For n∈Nn\in\mathbb{N} and ϵ=(ϵ1,…,ϵn)∈{0,1}n\epsilon=(\epsilon_{1},\dots,\epsilon_{n})\in\{0,1\}^{n} let Pϵ=G1ϵ1∩⋯∩GnϵnP_{\epsilon}=G_{1}^{\epsilon_{1}}\cap\dots\cap G_{n}^{\epsilon_{n}}. For z∈Zz\in Z let ϵn(z)∈{0,1}n\epsilon^{n}(z)\in\{0,1\}^{n} have ii-th entry 11 exactly when z∈Giz\in G_{i}; then z∈Pϵz\in P_{\epsilon} if and only if ϵ=ϵn(z)\epsilon=\epsilon^{n}(z), so the sets PϵP_{\epsilon}, ϵ∈{0,1}n\epsilon\in\{0,1\}^{n}, are pairwise disjoint and cover ZZ. Let Un\mathcal{U}_{n} be the family of the sets ⋃ϵ∈SPϵ\bigcup_{\epsilon\in S}P_{\epsilon} with S⊆{0,1}nS\subseteq\{0,1\}^{n} (the empty union being ∅\varnothing). It is finite, being the image of the finite set of subsets of {0,1}n\{0,1\}^{n}. It contains ZZ (take S={0,1}nS=\{0,1\}^{n}); the complement of ⋃ϵ∈SPϵ\bigcup_{\epsilon\in S}P_{\epsilon} is ⋃ϵ∉SPϵ\bigcup_{\epsilon\notin S}P_{\epsilon}, and the union of the members given by SS and S′S' is the member given by S∪S′S\cup S'. Since Pϵ=P(ϵ,1)∪P(ϵ,0)P_{\epsilon}=P_{(\epsilon,1)}\cup P_{(\epsilon,0)} with (ϵ,1),(ϵ,0)∈{0,1}n+1(\epsilon,1),(\epsilon,0)\in\{0,1\}^{n+1}, we have Un⊆Un+1\mathcal{U}_{n}\subseteq\mathcal{U}_{n+1}. Let U=⋃nUn\mathcal{U}=\bigcup_{n}\mathcal{U}_{n}; it is countable by claim 2 of Basic Properties of Countable Sets and A Countable Union of Countable Sets is Countable. It contains ZZ, so it is nonempty, and by Countable Set we fix a sequence (Uj)j∈N(U_{j})_{j\in\mathbb{N}} whose set of terms is U\mathcal{U}; for B∈UB\in\mathcal{U} let jBj_{B} be the least j∈Nj\in\mathbb{N} with Uj=BU_{j}=B. The family U\mathcal{U} is closed under complements, and if B∈UnB\in\mathcal{U}_{n} and B′∈Un′B'\in\mathcal{U}_{n'} then both lie in Umax⁡(n,n′)\mathcal{U}_{\max(n,n')}, so B∪B′∈UB\cup B'\in\mathcal{U}. Hence U\mathcal{U} is also closed under B∩B′=Z∖((Z∖B)∪(Z∖B′))B\cap B'=Z\setminus\bigl((Z\setminus B)\cup(Z\setminus B')\bigr) and under B′∖B=B′∩(Z∖B)B'\setminus B=B'\cap(Z\setminus B); in particular U\mathcal{U} is a π\pi-system in the sense of Dynkin's Pi-Lambda Theorem, and ∅∈U\varnothing\in\mathcal{U}. Each GiG_{i} is the union of the PϵP_{\epsilon} with ϵ∈{0,1}i\epsilon\in\{0,1\}^{i} and ϵi=1\epsilon_{i}=1, so G⊆U\mathcal{G}\subseteq\mathcal{U}.

We show that the σ\sigma-algebra generated by U\mathcal{U}, denoted σ(U)\sigma(\mathcal{U}), is B(Z)\mathcal{B}(Z). Open balls are open by Open Ball in a Metric Space is Open, closed balls are closed by claim 3 of Elementary Properties of the Closed Ball in a Metric Space, and open and closed sets belong to B(Z)\mathcal{B}(Z) by claim 1 of Borel Measurability and Bounded Integration on a Metric Space; with Step 1, G⊆B(Z)\mathcal{G}\subseteq\mathcal{B}(Z). As B(Z)\mathcal{B}(Z) is a σ\sigma-algebra, closed under complements and finite unions and intersections (Sigma-Algebra and Measurable Space), every PϵP_{\epsilon} and every member of U\mathcal{U} belongs to B(Z)\mathcal{B}(Z), so σ(U)⊆B(Z)\sigma(\mathcal{U})\subseteq\mathcal{B}(Z) by claim 2 of Intersections of Sigma-Algebras and Minimality of the Generated Sigma-Algebra. Conversely E⊆U⊆σ(U)\mathcal{E}\subseteq\mathcal{U}\subseteq\sigma(\mathcal{U}), so by the same claim and claim 2 of Borel Sets and Measurable Maps in a Separable Metric Space, B(Z)=σ(E)⊆σ(U)\mathcal{B}(Z)=\sigma(\mathcal{E})\subseteq\sigma(\mathcal{U}).

Step 3 (inner approximation by closed and by compact members of U\mathcal{U}). Call a set B⊆ZB\subseteq Z closed-approximable if there is a sequence (Ck)k∈N(C_{k})_{k\in\mathbb{N}} of closed members of U\mathcal{U} with Ck⊆Ck+1C_{k}\subseteq C_{k+1} for every kk and ⋃kCk=B\bigcup_{k}C_{k}=B. We show that every member of U\mathcal{U} is closed-approximable.

(a) Let G±\mathcal{G}^{\pm} consist of the members of G\mathcal{G} and their complements; all lie in U\mathcal{U}. Each is closed or open: the sets Bˉd(q,s)\bar{B}_{d}(q,s) and KmK_{m} are closed, Bd(q,s)B_{d}(q,s) is open, Z∖Bd(q,s)Z\setminus B_{d}(q,s) is closed and Z∖Bˉd(q,s)Z\setminus\bar{B}_{d}(q,s), Z∖KmZ\setminus K_{m} are open, by Closed Subset of a Topological Space. A closed member BB is closed-approximable by the constant sequence Ck=BC_{k}=B. Let V∈UV\in\mathcal{U} be open and let KV\mathcal{K}_{V} be the family of the members of Gcl\mathcal{G}_{\mathrm{cl}} contained in VV. We claim ⋃KV=V\bigcup\mathcal{K}_{V}=V. One inclusion is clear. Let z∈Vz\in V. By claim 1 of Borel Sets and Measurable Maps in a Separable Metric Space there are p∈Dp\in D and t∈Q>0t\in\mathbb{Q}_{>0} with z∈Bd(p,t)⊆Vz\in B_{d}(p,t)\subseteq V. As t−d(p,z)>0t-d(p,z)>0, claim 2 of The Archimedean Property of the Real Numbers gives n∈Nn\in\mathbb{N} with t<n(t−d(p,z))t<n\bigl(t-d(p,z)\bigr); then n≥2n\ge2 (for n=1n=1 this would say d(p,z)<0d(p,z)<0), and d(p,z)<t(n−1)/n=:t′d(p,z)<t(n-1)/n=:t'. Now t′∈Q>0t'\in\mathbb{Q}_{>0} and t′<tt'<t, so Bˉd(p,t′)⊆Bd(p,t)⊆V\bar{B}_{d}(p,t')\subseteq B_{d}(p,t)\subseteq V; thus Bˉd(p,t′)∈KV\bar{B}_{d}(p,t')\in\mathcal{K}_{V} and z∈Bˉd(p,t′)z\in\bar{B}_{d}(p,t'). If V=∅V=\varnothing, take Ck=∅C_{k}=\varnothing. Otherwise KV\mathcal{K}_{V} is nonempty and countable (claim 3 of Basic Properties of Countable Sets), so there is a sequence (Fj)j∈N(F_{j})_{j\in\mathbb{N}} whose set of terms is KV\mathcal{K}_{V}, and Ck=F1∪⋯∪FkC_{k}=F_{1}\cup\dots\cup F_{k} is closed by claim 2 of Complements, Unions and Intersections of Closed Sets in a Topological Space, belongs to U\mathcal{U}, increases with kk and has union VV. So every member of G±\mathcal{G}^{\pm} is closed-approximable.

(b) If B,B′B,B' are closed-approximable through (Ck)(C_{k}) and (Ck′)(C'_{k}), then B∪B′B\cup B' and B∩B′B\cap B' are closed-approximable through (Ck∪Ck′)(C_{k}\cup C'_{k}) and (Ck∩Ck′)(C_{k}\cap C'_{k}): these are closed members of U\mathcal{U} by claims 2 and 3 of Complements, Unions and Intersections of Closed Sets in a Topological Space, they increase with kk, every point of B∪B′B\cup B' lies in some CkC_{k} or some Ck′C'_{k}, and a point of B∩B′B\cap B' lying in CkC_{k} and in Ck′′C'_{k'} lies in Ck′′∩Ck′′′C_{k''}\cap C'_{k''} for k′′=max⁡(k,k′)k''=\max(k,k'); the reverse inclusions are clear.

(c) The empty set is closed-approximable by the constant sequence ∅\varnothing, which is closed by claim 1 of Complements, Unions and Intersections of Closed Sets in a Topological Space. By (a), (b) and induction, every PϵP_{\epsilon}, being a finite intersection of members of G±\mathcal{G}^{\pm}, and every finite union of such sets is closed-approximable; that is, every member of U\mathcal{U} is.

For j∈Nj\in\mathbb{N} let Cj\mathcal{C}_{j} be the set of the sequences (Ck)k∈N(C_{k})_{k\in\mathbb{N}} of closed members of U\mathcal{U} with Ck⊆Ck+1C_{k}\subseteq C_{k+1} for every kk and ⋃kCk=Uj\bigcup_{k}C_{k}=U_{j}; it is nonempty, as UjU_{j} is closed-approximable. By Axiom of Countable Choice, applied to the family (Cj)j∈N(\mathcal{C}_{j})_{j\in\mathbb{N}} of nonempty subsets of the set of all sequences in U\mathcal{U}, there is a sequence (cj)j∈N(c_{j})_{j\in\mathbb{N}} with cj∈Cjc_{j}\in\mathcal{C}_{j} for every jj. For B∈UB\in\mathcal{U} let (CB,k)k∈N(C_{B,k})_{k\in\mathbb{N}} be the sequence cjBc_{j_{B}}, which belongs to CjB\mathcal{C}_{j_{B}} with UjB=BU_{j_{B}}=B, and put EB,k=CB,k∩KkE_{B,k}=C_{B,k}\cap K_{k}. Then EB,k∈UE_{B,k}\in\mathcal{U}, EB,kE_{B,k} is compact by claim 1 of Compactness of Intersections with Closed Sets and of Level Sets of Semicontinuous Functions, EB,k⊆EB,k+1E_{B,k}\subseteq E_{B,k+1}, EB,k⊆BE_{B,k}\subseteq B, and ⋃kEB,k=B∩K∞\bigcup_{k}E_{B,k}=B\cap K_{\infty}: a point of B∩K∞B\cap K_{\infty} lies in some CB,kC_{B,k} and some Kk′K_{k'}, hence in EB,k′′E_{B,k''} for k′′=max⁡(k,k′)k''=\max(k,k').

Step 4 (densities with values in [0,1][0,1]). For B∈B(Z)B\in\mathcal{B}(Z) define λB(A)=π(A×B)\lambda_{B}(A)=\pi(A\times B) for A∈YA\in\mathcal{Y}. Then λB(∅)=0\lambda_{B}(\varnothing)=0, and for pairwise disjoint Am∈YA_{m}\in\mathcal{Y} the rectangles Am×BA_{m}\times B are pairwise disjoint with union (⋃mAm)×B\bigl(\bigcup_{m}A_{m}\bigr)\times B, so λB\lambda_{B} is a measure on (Y,Y)(Y,\mathcal{Y}). As A×B⊆A×ZA\times B\subseteq A\times Z, Basic Properties of a Measure §monotone gives λB(A)≤μ(A)\lambda_{B}(A)\le\mu(A); so λB\lambda_{B} is finite and λB(A)=0\lambda_{B}(A)=0 whenever μ(A)=0\mu(A)=0. The measure μ\mu is σ\sigma-finite, as μ(Y)=1\mu(Y)=1 (Measure, Measure Space, and Probability Measure).

Let B∈UB\in\mathcal{U}. By The Radon-Nikodym Theorem for a Finite Measure and a Sigma-Finite Measure, and Uniqueness of Densities §existence there is a measurable h~B:Y→R\tilde{h}_{B}:Y\to\mathbb{R} with h~B≥0\tilde{h}_{B}\ge0 and λB(A)=∫Y1Ah~B dμ\lambda_{B}(A)=\int_{Y}\mathbf{1}_{A}\tilde{h}_{B}\,d\mu for every A∈YA\in\mathcal{Y}. For n∈Nn\in\mathbb{N} let tn=1+1/nt_{n}=1+1/n and An={y∈Y:tn≤h~B(y)}A_{n}=\{y\in Y:t_{n}\le\tilde{h}_{B}(y)\}, which belongs to Y\mathcal{Y} by The Lebesgue Integral and Null Sets: Almost-Everywhere Comparison, Markov's Inequality, and Dominated Convergence Almost Everywhere §markov. The function 1Anh~B\mathbf{1}_{A_{n}}\tilde{h}_{B} is measurable by claims 1 and 3 of Arithmetic, Absolute Values, and Pointwise Limits of Measurable Real-Valued Functions, it is nonnegative, and the set where it is at least tnt_{n} is AnA_{n}, since tn>0t_{n}>0. By The Lebesgue Integral and Null Sets: Almost-Everywhere Comparison, Markov's Inequality, and Dominated Convergence Almost Everywhere §markov and Basic Properties of a Measure §monotone,

tn μ(An)≤∫Y1Anh~B dμ=λB(An)≤μ(An),t_{n}\,\mu(A_{n})\le\int_{Y}\mathbf{1}_{A_{n}}\tilde{h}_{B}\,d\mu=\lambda_{B}(A_{n})\le\mu(A_{n}),

and as μ(An)≤1\mu(A_{n})\le1 is real this gives μ(An)/n≤0\mu(A_{n})/n\le0, so μ(An)=0\mu(A_{n})=0. If h~B(y)>1\tilde{h}_{B}(y)>1, claim 3 of The Archimedean Property of the Real Numbers gives nn with 1/n<h~B(y)−11/n<\tilde{h}_{B}(y)-1, so y∈Any\in A_{n}; hence {h~B>1}\{\tilde{h}_{B}>1\} is contained in the null set ⋃nAn\bigcup_{n}A_{n} (The Lebesgue Integral and Null Sets: Almost-Everywhere Comparison, Markov's Inequality, and Dominated Convergence Almost Everywhere §null-union). Let hB=min⁡(h~B,1)h_{B}=\min(\tilde{h}_{B},1), measurable by claims 1 and 4 of Arithmetic, Absolute Values, and Pointwise Limits of Measurable Real-Valued Functions, with values in [0,1][0,1] and equal to h~B\tilde{h}_{B} off ⋃nAn\bigcup_{n}A_{n}. For A∈YA\in\mathcal{Y} the nonnegative measurable functions 1AhB\mathbf{1}_{A}h_{B} and 1Ah~B\mathbf{1}_{A}\tilde{h}_{B} agree a.e., so by The Lebesgue Integral and Null Sets: Almost-Everywhere Comparison, Markov's Inequality, and Dominated Convergence Almost Everywhere §comparison

π(A×B)=λB(A)=∫Y1AhB dμ(A∈Y, B∈U),(2)\pi(A\times B)=\lambda_{B}(A)=\int_{Y}\mathbf{1}_{A}h_{B}\,d\mu\qquad(A\in\mathcal{Y},\ B\in\mathcal{U}),\tag{2}

that is, hBh_{B} is a density of λB\lambda_{B} with respect to μ\mu. Consequently, for every j∈Nj\in\mathbb{N} the set Hj\mathcal{H}_{j} of the measurable functions h:Y→[0,1]h:Y\to[0,1] with π(A×Uj)=∫Y1Ah dμ\pi(A\times U_{j})=\int_{Y}\mathbf{1}_{A}h\,d\mu for every A∈YA\in\mathcal{Y} is nonempty. By Axiom of Countable Choice, applied to the family (Hj)j∈N(\mathcal{H}_{j})_{j\in\mathbb{N}} of nonempty subsets of the set of all functions from YY to R\mathbb{R}, there is a sequence (fj)j∈N(f_{j})_{j\in\mathbb{N}} with fj∈Hjf_{j}\in\mathcal{H}_{j} for every jj. From now on, for B∈UB\in\mathcal{U}, hBh_{B} denotes fjBf_{j_{B}}; since UjB=BU_{j_{B}}=B, it is a measurable function Y→[0,1]Y\to[0,1] satisfying (2).

Step 5 (countably many identities almost everywhere).

(5.1) The constant function 11 is measurable by claim 1 of Arithmetic, Absolute Values, and Pointwise Limits of Measurable Real-Valued Functions, and ∫Y1A⋅1 dμ=μ(A)=π(A×Z)=λZ(A)\int_{Y}\mathbf{1}_{A}\cdot1\,d\mu=\mu(A)=\pi(A\times Z)=\lambda_{Z}(A) by The Integral of an Indicator Function is the Measure of the Set. So 11 and hZh_{Z} are densities of λZ\lambda_{Z}, and by The Radon-Nikodym Theorem for a Finite Measure and a Sigma-Finite Measure, and Uniqueness of Densities §uniqueness the set NZ={y:hZ(y)≠1}N_{Z}=\{y:h_{Z}(y)\ne1\} belongs to Y\mathcal{Y} and μ(NZ)=0\mu(N_{Z})=0.

(5.2) Let P\mathcal{P} be the set of pairs (B,B′)∈U×U(B,B')\in\mathcal{U}\times\mathcal{U} with B∩B′=∅B\cap B'=\varnothing; it is countable by claim 1 of Products and Powers of Countable Sets and claim 3 of Basic Properties of Countable Sets. For (B,B′)∈P(B,B')\in\mathcal{P} we have B∪B′∈UB\cup B'\in\mathcal{U}, the function hB+hB′h_{B}+h_{B'} is measurable (claim 2 of Arithmetic, Absolute Values, and Pointwise Limits of Measurable Real-Valued Functions) and nonnegative, and for A∈YA\in\mathcal{Y}, by Linearity and Monotonicity of the Lebesgue Integral §nonnegative, (2) and Basic Properties of a Measure §additivity applied to the disjoint sets A×BA\times B and A×B′A\times B',

∫Y1A(hB+hB′) dμ=∫Y1AhB dμ+∫Y1AhB′ dμ=π(A×B)+π(A×B′)=π(A×(B∪B′)).\int_{Y}\mathbf{1}_{A}(h_{B}+h_{B'})\,d\mu=\int_{Y}\mathbf{1}_{A}h_{B}\,d\mu+\int_{Y}\mathbf{1}_{A}h_{B'}\,d\mu=\pi(A\times B)+\pi(A\times B')=\pi\bigl(A\times(B\cup B')\bigr).

So hB+hB′h_{B}+h_{B'} and hB∪B′h_{B\cup B'} are densities of λB∪B′\lambda_{B\cup B'}, and by The Radon-Nikodym Theorem for a Finite Measure and a Sigma-Finite Measure, and Uniqueness of Densities §uniqueness the set NB,B′={y:hB∪B′(y)≠hB(y)+hB′(y)}N_{B,B'}=\{y:h_{B\cup B'}(y)\ne h_{B}(y)+h_{B'}(y)\} belongs to Y\mathcal{Y} and has μ(NB,B′)=0\mu(N_{B,B'})=0. The set P\mathcal{P} is nonempty, as (∅,∅)∈P(\varnothing,\varnothing)\in\mathcal{P}, so by Countable Set there is a sequence (pj)j∈N(p_{j})_{j\in\mathbb{N}} whose set of terms is P\mathcal{P}. Write Np=NB,B′N_{p}=N_{B,B'} for p=(B,B′)∈Pp=(B,B')\in\mathcal{P}, and let Nadd=⋃j∈NNpjN_{\mathrm{add}}=\bigcup_{j\in\mathbb{N}}N_{p_{j}}, which is the union of the NB,B′N_{B,B'} over (B,B′)∈P(B,B')\in\mathcal{P}. Each NpjN_{p_{j}} belongs to Y\mathcal{Y}, so Nadd∈YN_{\mathrm{add}}\in\mathcal{Y} because Y\mathcal{Y} is a σ\sigma-algebra (Sigma-Algebra and Measurable Space), and μ(Nadd)≤∑j∈Nμ(Npj)=0\mu(N_{\mathrm{add}})\le\sum_{j\in\mathbb{N}}\mu(N_{p_{j}})=0 by Basic Properties of a Measure §subadditivity; so μ(Nadd)=0\mu(N_{\mathrm{add}})=0, and in particular NaddN_{\mathrm{add}} is null. If y∉Naddy\notin N_{\mathrm{add}} and B⊆B′B\subseteq B' are members of U\mathcal{U}, then (B,B′∖B)∈P(B,B'\setminus B)\in\mathcal{P} with union B′B', so

hB′(y)=hB(y)+hB′∖B(y)≥hB(y).(3)h_{B'}(y)=h_{B}(y)+h_{B'\setminus B}(y)\ge h_{B}(y).\tag{3}

(5.3) Let B∈UB\in\mathcal{U}. Let gB(y)=sup⁡khEB,k(y)g_{B}(y)=\sup_{k}h_{E_{B,k}}(y); it takes values in [0,1][0,1] and is measurable by claim 1 of Measurability of Countable Suprema, Bounded Pointwise Limits, Monotone Functions, and Continuous Functions, applied with the bound 11. For k∈Nk\in\mathbb{N} let sB,k=max⁡(hEB,1,…,hEB,k)s_{B,k}=\max(h_{E_{B,1}},\dots,h_{E_{B,k}}), measurable by claim 4 of Arithmetic, Absolute Values, and Pointwise Limits of Measurable Real-Valued Functions and induction on kk. At every yy the sequence (sB,k(y))k(s_{B,k}(y))_{k} is nondecreasing with supremum gB(y)g_{B}(y), and for y∉Naddy\notin N_{\mathrm{add}}, (3) applied to EB,i⊆EB,kE_{B,i}\subseteq E_{B,k} for i≤ki\le k gives sB,k(y)=hEB,k(y)s_{B,k}(y)=h_{E_{B,k}}(y). Let A∈YA\in\mathcal{Y}. The measurable functions 1AsB,k\mathbf{1}_{A}s_{B,k} (claims 1 and 3 of Arithmetic, Absolute Values, and Pointwise Limits of Measurable Real-Valued Functions) are nondecreasing in kk at every point, with supremum 1AgB\mathbf{1}_{A}g_{B}, so by Monotone Convergence Theorem

∫Y1AgB dμ=sup⁡k∫Y1AsB,k dμ=sup⁡k∫Y1AhEB,k dμ=sup⁡kπ(A×EB,k),\int_{Y}\mathbf{1}_{A}g_{B}\,d\mu=\sup_{k}\int_{Y}\mathbf{1}_{A}s_{B,k}\,d\mu=\sup_{k}\int_{Y}\mathbf{1}_{A}h_{E_{B,k}}\,d\mu=\sup_{k}\pi(A\times E_{B,k}),

where the second equality is The Lebesgue Integral and Null Sets: Almost-Everywhere Comparison, Markov's Inequality, and Dominated Convergence Almost Everywhere §comparison, the two integrands agreeing off the null set NaddN_{\mathrm{add}}, and the third is (2). The sets A×EB,kA\times E_{B,k} increase with kk and have union A×(B∩K∞)A\times(B\cap K_{\infty}) by Step 3, so by Basic Properties of a Measure §continuity-below the last supremum is π(A×(B∩K∞))\pi\bigl(A\times(B\cap K_{\infty})\bigr). By Basic Properties of a Measure §additivity, Basic Properties of a Measure §monotone and (1), π(A×B)=π(A×(B∩K∞))+π(A×(B∖K∞))=π(A×(B∩K∞))\pi(A\times B)=\pi\bigl(A\times(B\cap K_{\infty})\bigr)+\pi\bigl(A\times(B\setminus K_{\infty})\bigr)=\pi\bigl(A\times(B\cap K_{\infty})\bigr). Hence ∫Y1AgB dμ=λB(A)\int_{Y}\mathbf{1}_{A}g_{B}\,d\mu=\lambda_{B}(A) for every A∈YA\in\mathcal{Y}, so gBg_{B} and hBh_{B} are densities of λB\lambda_{B}, and by The Radon-Nikodym Theorem for a Finite Measure and a Sigma-Finite Measure, and Uniqueness of Densities §uniqueness the set NB={y:gB(y)≠hB(y)}N^{B}=\{y:g_{B}(y)\ne h_{B}(y)\} belongs to Y\mathcal{Y} and has μ(NB)=0\mu(N^{B})=0.

Let N0=NZ∪Nadd∪⋃j∈NNUjN_{0}=N_{Z}\cup N_{\mathrm{add}}\cup\bigcup_{j\in\mathbb{N}}N^{U_{j}}; since the terms of (Uj)(U_{j}) are exactly the members of U\mathcal{U}, NB⊆N0N^{B}\subseteq N_{0} for every B∈UB\in\mathcal{U}. The set N0N_{0} is the union of the sequence NZ,Nadd,NU1,NU2,…N_{Z},N_{\mathrm{add}},N^{U_{1}},N^{U_{2}},\dots of members of Y\mathcal{Y} of μ\mu-measure 00, so N0∈YN_{0}\in\mathcal{Y} because Y\mathcal{Y} is a σ\sigma-algebra (Sigma-Algebra and Measurable Space), and μ(N0)=0\mu(N_{0})=0 by Basic Properties of a Measure §subadditivity.

Step 6 (a premeasure on U\mathcal{U} for each y∉N0y\notin N_{0}). Fix y∈Y∖N0y\in Y\setminus N_{0} and let τy(B)=hB(y)∈[0,1]\tau_{y}(B)=h_{B}(y)\in[0,1] for B∈UB\in\mathcal{U}.

(6.1) τy(Z)=1\tau_{y}(Z)=1, as y∉NZy\notin N_{Z}.

(6.2) τy(B∪B′)=τy(B)+τy(B′)\tau_{y}(B\cup B')=\tau_{y}(B)+\tau_{y}(B') for disjoint B,B′∈UB,B'\in\mathcal{U}, as y∉Naddy\notin N_{\mathrm{add}}. Taking B=B′=∅B=B'=\varnothing gives τy(∅)=2τy(∅)\tau_{y}(\varnothing)=2\tau_{y}(\varnothing), so τy(∅)=0\tau_{y}(\varnothing)=0.

(6.3) If B⊆B′B\subseteq B' in U\mathcal{U}, then τy(B′∖B)=τy(B′)−τy(B)\tau_{y}(B'\setminus B)=\tau_{y}(B')-\tau_{y}(B) and τy(B)≤τy(B′)\tau_{y}(B)\le \tau_{y}(B'), by (6.2).

(6.4) For B1,…,Br∈UB_{1},\dots,B_{r}\in\mathcal{U}, τy(B1∪⋯∪Br)≤τy(B1)+⋯+τy(Br)\tau_{y}(B_{1}\cup\dots\cup B_{r})\le \tau_{y}(B_{1})+\dots+\tau_{y}(B_{r}): for r=2r=2, B1∪B2B_{1}\cup B_{2} is the disjoint union of B1B_{1} and B2∖B1B_{2}\setminus B_{1}, so (6.2) and (6.3) give τy(B1∪B2)=τy(B1)+τy(B2∖B1)≤τy(B1)+τy(B2)\tau_{y}(B_{1}\cup B_{2})=\tau_{y}(B_{1})+\tau_{y}(B_{2}\setminus B_{1})\le \tau_{y}(B_{1})+\tau_{y}(B_{2}); the general case follows by induction on rr.

(6.5) For B∈UB\in\mathcal{U}, τy(B)=hB(y)=gB(y)=sup⁡kτy(EB,k)\tau_{y}(B)=h_{B}(y)=g_{B}(y)=\sup_{k}\tau_{y}(E_{B,k}), as y∉NBy\notin N^{B}. Hence for every real η>0\eta>0 there is kk with τy(B)−τy(EB,k)≤η\tau_{y}(B)-\tau_{y}(E_{B,k})\le\eta, since otherwise τy(B)−η\tau_{y}(B)-\eta would be an upper bound of the τy(EB,k)\tau_{y}(E_{B,k}) smaller than their least upper bound.

(6.6) Continuity at ∅\varnothing. Let (Rn)n∈N(R_{n})_{n\in\mathbb{N}} be a sequence in U\mathcal{U} with Rn+1⊆RnR_{n+1}\subseteq R_{n} for every nn and ⋂nRn=∅\bigcap_{n}R_{n}=\varnothing. We show that τy(Rn)\tau_{y}(R_{n}) converges to 00. By (6.3) the sequence (τy(Rn))(\tau_{y}(R_{n})) is nonincreasing and bounded below by 00; let δ≥0\delta\ge0 be its greatest lower bound, and suppose, for a contradiction, that δ>0\delta>0. Then τy(Rn)≥δ\tau_{y}(R_{n})\ge\delta for every nn. With δ\delta fixed, for each nn let knk_{n} be the least k∈Nk\in\mathbb{N} with τy(Rn)−τy(ERn,k)≤δ2−n−1\tau_{y}(R_{n})-\tau_{y}(E_{R_{n},k})\le\delta2^{-n-1}, which exists by (6.5). Put Wn=ER1,k1∩⋯∩ERn,kn∈UW_{n}=E_{R_{1},k_{1}}\cap\dots\cap E_{R_{n},k_{n}}\in\mathcal{U}. Then Wn+1⊆WnW_{n+1}\subseteq W_{n} and Wn⊆ERn,kn⊆RnW_{n}\subseteq E_{R_{n},k_{n}}\subseteq R_{n}. Since Rn⊆RiR_{n}\subseteq R_{i} for i≤ni\le n,

Rn∖Wn=⋃i=1n(Rn∖ERi,ki)⊆⋃i=1n(Ri∖ERi,ki),R_{n}\setminus W_{n}=\bigcup_{i=1}^{n}\bigl(R_{n}\setminus E_{R_{i},k_{i}}\bigr)\subseteq\bigcup_{i=1}^{n}\bigl(R_{i}\setminus E_{R_{i},k_{i}}\bigr),

so by (6.3) and (6.4)

τy(Rn∖Wn)≤∑i=1n(τy(Ri)−τy(ERi,ki))≤∑i=1nδ2−i−1<δ2,\tau_{y}(R_{n}\setminus W_{n})\le\sum_{i=1}^{n}\bigl(\tau_{y}(R_{i})-\tau_{y}(E_{R_{i},k_{i}})\bigr)\le\sum_{i=1}^{n}\delta2^{-i-1}<\frac{\delta}{2},

and therefore, by (6.3), τy(Wn)=τy(Rn)−τy(Rn∖Wn)>δ−δ/2>0\tau_{y}(W_{n})=\tau_{y}(R_{n})-\tau_{y}(R_{n}\setminus W_{n})>\delta-\delta/2>0. By (6.2), Wn≠∅W_{n}\ne\varnothing for every nn. Each ERi,kiE_{R_{i},k_{i}} is compact, hence closed by Compact Subsets of a Metric Space are Closed and Borel §closed, so every WnW_{n} is closed by claim 3 of Complements, Unions and Intersections of Closed Sets in a Topological Space, and W1=ER1,k1W_{1}=E_{R_{1},k_{1}} is compact. The sets Z∖WnZ\setminus W_{n} are open, and since ⋂nWn⊆⋂nRn=∅\bigcap_{n}W_{n}\subseteq\bigcap_{n}R_{n}=\varnothing they cover W1W_{1}. By Compact Subset Criterion via Open Covers in the Ambient Space there is a finite J⊆NJ\subseteq\mathbb{N} with W1⊆⋃n∈J(Z∖Wn)W_{1}\subseteq\bigcup_{n\in J}(Z\setminus W_{n}); JJ is nonempty because W1≠∅W_{1}\ne\varnothing. Let n∗n^{*} be the largest element of JJ. For n∈Jn\in J we have Wn∗⊆WnW_{n^{*}}\subseteq W_{n}, so Z∖Wn⊆Z∖Wn∗Z\setminus W_{n}\subseteq Z\setminus W_{n^{*}}, and hence W1⊆Z∖Wn∗W_{1}\subseteq Z\setminus W_{n^{*}}. But ∅≠Wn∗⊆W1\varnothing\ne W_{n^{*}}\subseteq W_{1}, a contradiction. Thus δ=0\delta=0. Given a real η>0\eta>0 there is then n0n_{0} with τy(Rn0)<η\tau_{y}(R_{n_{0}})<\eta, and 0≤τy(Rn)≤τy(Rn0)<η0\le \tau_{y}(R_{n})\le \tau_{y}(R_{n_{0}})<\eta for n≥n0n\ge n_{0}; so τy(Rn)\tau_{y}(R_{n}) converges to 00.

(6.7) Countable additivity. Let (Bi)i∈N(B_{i})_{i\in\mathbb{N}} be pairwise disjoint members of U\mathcal{U} whose union BB belongs to U\mathcal{U}. Put Rn=B∖(B1∪⋯∪Bn)∈UR_{n}=B\setminus(B_{1}\cup\dots\cup B_{n})\in\mathcal{U}. Then Rn+1⊆RnR_{n+1}\subseteq R_{n}, and ⋂nRn=∅\bigcap_{n}R_{n}=\varnothing because each point of BB lies in some BiB_{i} and hence not in RnR_{n} for n≥in\ge i. As BB is the disjoint union of B1,…,Bn,RnB_{1},\dots,B_{n},R_{n}, (6.2) and induction give ∑i=1nτy(Bi)=τy(B)−τy(Rn)\sum_{i=1}^{n}\tau_{y}(B_{i})=\tau_{y}(B)-\tau_{y}(R_{n}), which converges to τy(B)\tau_{y}(B) by (6.6). So the series ∑iτy(Bi)\sum_{i}\tau_{y}(B_{i}) converges and its sum is τy(B)\tau_{y}(B), in the sense of Series of Real Numbers §convergent.

By Step 2, (6.2) and (6.7), the family U\mathcal{U} of subsets of ZZ and the function τy\tau_{y} satisfy the hypotheses of Extension of a Finite Premeasure from an Algebra of Sets to the Generated Sigma-Algebra. By Extension of a Finite Premeasure from an Algebra of Sets to the Generated Sigma-Algebra §extension there is a measure on σ(U)=B(Z)\sigma(\mathcal{U})=\mathcal{B}(Z) that agrees with τy\tau_{y} on U\mathcal{U}, and by Extension of a Finite Premeasure from an Algebra of Sets to the Generated Sigma-Algebra §uniqueness there is only one; we call it κy\kappa_{y}. By (6.1), κy(Z)=1\kappa_{y}(Z)=1, so κy\kappa_{y} is a probability measure on (Z,B(Z))(Z,\mathcal{B}(Z)), and

κy(B)=hB(y)(y∈Y∖N0, B∈U).(4)\kappa_{y}(B)=h_{B}(y)\qquad(y\in Y\setminus N_{0},\ B\in\mathcal{U}).\tag{4}

For y∈N0y\in N_{0} put κy=ρ\kappa_{y}=\rho, a probability measure on (Z,B(Z))(Z,\mathcal{B}(Z)) by Step 0. Define κ(y,B)=κy(B)\kappa(y,B)=\kappa_{y}(B) for y∈Yy\in Y and B∈B(Z)B\in\mathcal{B}(Z); its values lie in [0,1][0,1].

Step 7 (measurability and the rectangle identity). Let L\mathcal{L} be the family of those B∈B(Z)B\in\mathcal{B}(Z) such that y↦κ(y,B)y\mapsto\kappa(y,B) is measurable and

π(A×B)=∫Y1A(y) κ(y,B) μ(dy)for every A∈Y.(5)\pi(A\times B)=\int_{Y}\mathbf{1}_{A}(y)\,\kappa(y,B)\,\mu(dy)\qquad\text{for every }A\in\mathcal{Y}.\tag{5}

U⊆L\mathcal{U}\subseteq\mathcal{L}: for B∈UB\in\mathcal{U}, (4) gives κ(⋅,B)=1Y∖N0hB+ρ(B)1N0\kappa(\cdot,B)=\mathbf{1}_{Y\setminus N_{0}}h_{B}+\rho(B)\mathbf{1}_{N_{0}}, which is measurable by claims 1, 2 and 3 of Arithmetic, Absolute Values, and Pointwise Limits of Measurable Real-Valued Functions, as N0∈YN_{0}\in\mathcal{Y}. For A∈YA\in\mathcal{Y} the nonnegative measurable functions 1Aκ(⋅,B)\mathbf{1}_{A}\kappa(\cdot,B) and 1AhB\mathbf{1}_{A}h_{B} agree off the null set N0N_{0}, so (5) follows from The Lebesgue Integral and Null Sets: Almost-Everywhere Comparison, Markov's Inequality, and Dominated Convergence Almost Everywhere §comparison and (2).

L\mathcal{L} is a λ\lambda-system in the sense of Dynkin's Pi-Lambda Theorem. (1) Z∈U⊆LZ\in\mathcal{U}\subseteq\mathcal{L}. (2) Let B⊆B′B\subseteq B' be members of L\mathcal{L}. For each yy, κy(B′∖B)=κy(B′)−κy(B)\kappa_{y}(B'\setminus B)=\kappa_{y}(B')-\kappa_{y}(B) by Basic Properties of a Measure §differences, so κ(⋅,B′∖B)\kappa(\cdot,B'\setminus B) is measurable by claim 2 of Arithmetic, Absolute Values, and Pointwise Limits of Measurable Real-Valued Functions. For A∈YA\in\mathcal{Y} we have 1Aκ(⋅,B′)=1Aκ(⋅,B)+1Aκ(⋅,B′∖B)\mathbf{1}_{A}\kappa(\cdot,B')=\mathbf{1}_{A}\kappa(\cdot,B)+\mathbf{1}_{A}\kappa(\cdot,B'\setminus B) pointwise, by Basic Properties of a Measure §additivity for each κy\kappa_{y}, so by Linearity and Monotonicity of the Lebesgue Integral §nonnegative and (5) for BB and B′B',

∫Y1Aκ(⋅,B′∖B) dμ=π(A×B′)−π(A×B)=π(A×(B′∖B)),\int_{Y}\mathbf{1}_{A}\kappa(\cdot,B'\setminus B)\,d\mu=\pi(A\times B')-\pi(A\times B)=\pi\bigl(A\times(B'\setminus B)\bigr),

the subtraction being legitimate since π(A×B)≤1\pi(A\times B)\le1 is real, and the last equality being Basic Properties of a Measure §differences for A×B⊆A×B′A\times B\subseteq A\times B', whose difference is A×(B′∖B)A\times(B'\setminus B). So B′∖B∈LB'\setminus B\in\mathcal{L}. (3) Let (Bn)n∈N(B_{n})_{n\in\mathbb{N}} be a nondecreasing sequence in L\mathcal{L} with union BB. For each yy, κ(y,Bn)\kappa(y,B_{n}) is nondecreasing in nn by Basic Properties of a Measure §monotone, and, its values lying in [0,1][0,1], it converges to κ(y,B)\kappa(y,B), which is also its least upper bound, by Basic Properties of a Measure §continuity-below; hence κ(⋅,B)\kappa(\cdot,B) is measurable by claim 5 of Arithmetic, Absolute Values, and Pointwise Limits of Measurable Real-Valued Functions. For A∈YA\in\mathcal{Y} the measurable functions 1Aκ(⋅,Bn)\mathbf{1}_{A}\kappa(\cdot,B_{n}) are nondecreasing in nn with supremum 1Aκ(⋅,B)\mathbf{1}_{A}\kappa(\cdot,B), so by Monotone Convergence Theorem, (5) for each BnB_{n}, and Basic Properties of a Measure §continuity-below for the increasing sets A×BnA\times B_{n} with union A×BA\times B,

∫Y1Aκ(⋅,B) dμ=sup⁡n∫Y1Aκ(⋅,Bn) dμ=sup⁡nπ(A×Bn)=π(A×B).\int_{Y}\mathbf{1}_{A}\kappa(\cdot,B)\,d\mu=\sup_{n}\int_{Y}\mathbf{1}_{A}\kappa(\cdot,B_{n})\,d\mu=\sup_{n}\pi(A\times B_{n})=\pi(A\times B).

So B∈LB\in\mathcal{L}.

By Dynkin's Pi-Lambda Theorem, applied to the π\pi-system U\mathcal{U} and the λ\lambda-system L\mathcal{L}, we get σ(U)⊆L\sigma(\mathcal{U})\subseteq\mathcal{L}; with Step 2 and L⊆B(Z)\mathcal{L}\subseteq\mathcal{B}(Z) this gives L=B(Z)\mathcal{L}=\mathcal{B}(Z).

Step 8 (conclusion of existence). By Step 6 each κy\kappa_{y} is a probability measure on (Z,B(Z))(Z,\mathcal{B}(Z)), and by Step 7 each κ(⋅,B)\kappa(\cdot,B), B∈B(Z)B\in\mathcal{B}(Z), is measurable with respect to Y\mathcal{Y}; so κ\kappa is a probability kernel from (Y,Y)(Y,\mathcal{Y}) to (Z,B(Z))(Z,\mathcal{B}(Z)) by Probability Kernels Between Measurable Spaces §kernel. By Step 7, (5) holds for all A∈YA\in\mathcal{Y} and B∈B(Z)B\in\mathcal{B}(Z), which is the displayed identity of claim 1, and Integration Against a Probability Kernel: Measurable Sections, the Composite Measure on the Product and the Iterated Integral §rectangles, applied to the probability measure μ\mu, the kernel κ\kappa and the measure π\pi, gives π=μ⊗κ\pi=\mu\otimes\kappa.

Part 2: uniqueness. Let κ\kappa and κ′\kappa' be probability kernels from (Y,Y)(Y,\mathcal{Y}) to (Z,B(Z))(Z,\mathcal{B}(Z)) with π=μ⊗κ=μ⊗κ′\pi=\mu\otimes\kappa=\mu\otimes\kappa', and let U\mathcal{U} be the family fixed in Step 2, with its enumeration (Uj)j∈N(U_{j})_{j\in\mathbb{N}}; it is a countable π\pi-system that contains ZZ and satisfies σ(U)=B(Z)\sigma(\mathcal{U})=\mathcal{B}(Z). Let B∈UB\in\mathcal{U}. By Integration Against a Probability Kernel: Measurable Sections, the Composite Measure on the Product and the Iterated Integral §composite, for every A∈YA\in\mathcal{Y},

λB(A)=π(A×B)=(μ⊗κ)(A×B)=∫Y1A(y) κ(y,B) μ(dy),\lambda_{B}(A)=\pi(A\times B)=(\mu\otimes\kappa)(A\times B)=\int_{Y}\mathbf{1}_{A}(y)\,\kappa(y,B)\,\mu(dy),

and likewise with κ′\kappa'. By Probability Kernels Between Measurable Spaces §kernel, κ(⋅,B)\kappa(\cdot,B) and κ′(⋅,B)\kappa'(\cdot,B) are measurable with values in [0,1][0,1], so both are densities of the finite measure λB\lambda_{B} (Step 4) with respect to μ\mu, and by The Radon-Nikodym Theorem for a Finite Measure and a Sigma-Finite Measure, and Uniqueness of Densities §uniqueness the set MB={y:κ(y,B)≠κ′(y,B)}M_{B}=\{y:\kappa(y,B)\ne\kappa'(y,B)\} belongs to Y\mathcal{Y} and has μ(MB)=0\mu(M_{B})=0. Let M=⋃j∈NMUjM=\bigcup_{j\in\mathbb{N}}M_{U_{j}}, which is the union of the MBM_{B} over B∈UB\in\mathcal{U}, since the terms of (Uj)(U_{j}) are exactly the members of U\mathcal{U}. Each MUjM_{U_{j}} belongs to Y\mathcal{Y}, so M∈YM\in\mathcal{Y} because Y\mathcal{Y} is a σ\sigma-algebra (Sigma-Algebra and Measurable Space), and μ(M)≤∑j∈Nμ(MUj)=0\mu(M)\le\sum_{j\in\mathbb{N}}\mu(M_{U_{j}})=0 by Basic Properties of a Measure §subadditivity, so μ(M)=0\mu(M)=0.

We claim that MM is exactly the set of yy with κy≠κy′\kappa_{y}\ne\kappa'_{y}. If y∈My\in M, then κy(B)≠κy′(B)\kappa_{y}(B)\ne\kappa'_{y}(B) for some BB, so κy≠κy′\kappa_{y}\ne\kappa'_{y}. If y∉My\notin M, then κy\kappa_{y} and κy′\kappa'_{y} are measures on (Z,B(Z))(Z,\mathcal{B}(Z)) with κy(Z)=κy′(Z)=1<∞\kappa_{y}(Z)=\kappa'_{y}(Z)=1<\infty that agree on the π\pi-system U\mathcal{U} generating B(Z)\mathcal{B}(Z), so κy=κy′\kappa_{y}=\kappa'_{y} by claim 1 of Uniqueness of Finite Measures on a Generating Pi-System and the Density of the Exponential Law. Hence the set of yy with κy≠κy′\kappa_{y}\ne\kappa'_{y} is MM, which belongs to Y\mathcal{Y} and has μ\mu-measure 00. ■\blacksquare

Citations

Loading…

Dependencies

Uses0

Loading…

Comments

Log in to comment.

Loading…