TheoremBase

The outer measure defined by countable covers with members of the algebra agrees with m on the algebra, and every member of the algebra is Caratheodory measurable, so Caratheodory's theorem yields a finite measure on the generated sigma-algebra extending m. Uniqueness follows from the uniqueness of finite measures agreeing on a generating pi-system.

Proof

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

Throughout, N={1,2,3,… }\mathbb{N}=\{1,2,3,\dots\} as in Natural Numbers, and for a sequence (ai)i∈N(a_{i})_{i\in\mathbb{N}} in [0,∞][0,\infty] the sum ∑i∈Nai\sum_{i\in\mathbb{N}}a_{i} is the one of Measure, Measure Space, and Probability Measure: if every aia_{i} is real and the partial sums ∑i=1nai\sum_{i=1}^{n}a_{i} are bounded above, it is their least upper bound, and otherwise it is ∞\infty. Two consequences of this definition are used repeatedly: if all aia_{i} are real and cc is a real number with ∑i=1nai≤c\sum_{i=1}^{n}a_{i}\le c for every n∈Nn\in\mathbb{N}, then ∑i∈Nai≤c\sum_{i\in\mathbb{N}}a_{i}\le c; and if ∑i∈Nai\sum_{i\in\mathbb{N}}a_{i} is real, then every partial sum is at most ∑i∈Nai\sum_{i\in\mathbb{N}}a_{i}. In particular, if ai=0a_{i}=0 for all ii outside a finite set F⊆NF\subseteq\mathbb{N}, the partial sums are eventually constant equal to ∑i∈Fai\sum_{i\in F}a_{i}, so ∑i∈Nai=∑i∈Fai\sum_{i\in\mathbb{N}}a_{i}=\sum_{i\in F}a_{i}. For a sequence (ai)(a_{i}) of nonnegative reals, Series of Nonnegative Real Numbers, Comparison, and the Geometric Series §criterion shows that the series ∑i=1∞ai\sum_{i=1}^{\infty}a_{i} of Series of Real Numbers §convergent converges exactly when the partial sums are bounded above, and then its sum is their least upper bound, that is, it equals ∑i∈Nai\sum_{i\in\mathbb{N}}a_{i}.

Step 0: elementary properties of A\mathcal{A} and mm. Let A,B∈AA,B\in\mathcal{A}. Then A∩B=Ω∖((Ω∖A)∪(Ω∖B))∈AA\cap B=\Omega\setminus\bigl((\Omega\setminus A)\cup(\Omega\setminus B)\bigr)\in\mathcal{A} and A∖B=A∩(Ω∖B)∈AA\setminus B=A\cap(\Omega\setminus B)\in\mathcal{A}, and by induction on the number of sets every finite union of members of A\mathcal{A} belongs to A\mathcal{A}, the empty union being ∅∈A\varnothing\in\mathcal{A}. If AA and BB are disjoint, the sequence (Ci)i∈N(C_{i})_{i\in\mathbb{N}} with C1=AC_{1}=A, C2=BC_{2}=B and Ci=∅C_{i}=\varnothing for i≥3i\ge3 consists of pairwise disjoint members of A\mathcal{A} with union A∪B∈AA\cup B\in\mathcal{A}, so by the countable additivity of mm the series ∑i=1∞m(Ci)\sum_{i=1}^{\infty}m(C_{i}) converges to m(A∪B)m(A\cup B); since m(∅)=0m(\varnothing)=0 its partial sums equal m(A)+m(B)m(A)+m(B) from n=2n=2 on, so

m(A∪B)=m(A)+m(B)(A,B∈A disjoint).m(A\cup B)=m(A)+m(B)\qquad(A,B\in\mathcal{A}\text{ disjoint}).

If A⊆BA\subseteq B, then BB is the disjoint union of AA and B∖A∈AB\setminus A\in\mathcal{A}, so m(B)=m(A)+m(B∖A)≥m(A)m(B)=m(A)+m(B\setminus A)\ge m(A). In particular m(A)≤m(Ω)m(A)\le m(\Omega) for every A∈AA\in\mathcal{A}.

Step 1: the outer measure m∗m^{*}. For E⊆ΩE\subseteq\Omega, call a sequence (Ai)i∈N(A_{i})_{i\in\mathbb{N}} of members of A\mathcal{A} a cover of EE if E⊆⋃i∈NAiE\subseteq\bigcup_{i\in\mathbb{N}}A_{i}, and call ∑i∈Nm(Ai)∈[0,∞]\sum_{i\in\mathbb{N}}m(A_{i})\in[0,\infty] its weight. Let W(E)W(E) be the set of those weights of covers of EE that are real numbers. The sequence Ω,∅,∅,…\Omega,\varnothing,\varnothing,\dots is a cover of EE of weight m(Ω)m(\Omega), so W(E)W(E) is a nonempty set of real numbers, and it is bounded below by 00. Define m∗(E)m^{*}(E) to be the greatest lower bound of W(E)W(E). Then m∗(E)m^{*}(E) is a real number with 0≤m∗(E)≤m(Ω)0\le m^{*}(E)\le m(\Omega), and:

(a) every cover of EE has weight at least m∗(E)m^{*}(E) (for a real weight because m∗(E)m^{*}(E) is a lower bound of W(E)W(E), for the weight ∞\infty by the conventions of Measure, Measure Space, and Probability Measure);

(b) for every real ε>0\varepsilon>0 there is a cover of EE with real weight less than m∗(E)+εm^{*}(E)+\varepsilon, because m∗(E)+εm^{*}(E)+\varepsilon is not a lower bound of W(E)W(E).

Step 2: m∗m^{*} is an outer measure on Ω\Omega in the sense of Outer Measure and Caratheodory Measurability. Its values lie in [0,∞)⊆[0,∞][0,\infty)\subseteq[0,\infty].

Property 1. The sequence all of whose terms are ∅\varnothing is a cover of ∅\varnothing of weight 00, so 0≤m∗(∅)≤00\le m^{*}(\varnothing)\le0.

Property 2. If E⊆F⊆ΩE\subseteq F\subseteq\Omega, every cover of FF is a cover of EE, so W(F)⊆W(E)W(F)\subseteq W(E) and therefore m∗(E)≤m∗(F)m^{*}(E)\le m^{*}(F).

Property 3. Let (Ek)k∈N(E_{k})_{k\in\mathbb{N}} be a sequence of subsets of Ω\Omega and put E=⋃k∈NEkE=\bigcup_{k\in\mathbb{N}}E_{k}. If ∑k∈Nm∗(Ek)=∞\sum_{k\in\mathbb{N}}m^{*}(E_{k})=\infty there is nothing to prove, so assume that s=∑k∈Nm∗(Ek)s=\sum_{k\in\mathbb{N}}m^{*}(E_{k}) is real; then ∑k=1Km∗(Ek)≤s\sum_{k=1}^{K}m^{*}(E_{k})\le s for every K∈NK\in\mathbb{N}. Let a real ε>0\varepsilon>0 be given. For each k∈Nk\in\mathbb{N}, by (b) choose a cover (Ak,i)i∈N(A_{k,i})_{i\in\mathbb{N}} of EkE_{k} whose weight wkw_{k} is real and satisfies wk<m∗(Ek)+ε(1/2)kw_{k}<m^{*}(E_{k})+\varepsilon(1/2)^{k}. The set N\mathbb{N} is countable by claim 1 of Basic Properties of Countable Sets, so N×N\mathbb{N}\times\mathbb{N} is countable by claim 1 of Products and Powers of Countable Sets; as it is nonempty, Countable Set provides a sequence (φ(j))j∈N(\varphi(j))_{j\in\mathbb{N}} in N×N\mathbb{N}\times\mathbb{N} whose set of terms is N×N\mathbb{N}\times\mathbb{N}. Call j∈Nj\in\mathbb{N} a first index if φ(j′)≠φ(j)\varphi(j')\ne\varphi(j) for every j′∈Nj'\in\mathbb{N} with j′<jj'<j, and define Cj=Aφ(j)C_{j}=A_{\varphi(j)} if jj is a first index and Cj=∅C_{j}=\varnothing otherwise; here A(k,i)A_{(k,i)} means Ak,iA_{k,i}. Each CjC_{j} belongs to A\mathcal{A}.

The sequence (Cj)j∈N(C_{j})_{j\in\mathbb{N}} is a cover of EE: if x∈Ex\in E, choose kk with x∈Ekx\in E_{k} and ii with x∈Ak,ix\in A_{k,i}; the set of j∈Nj\in\mathbb{N} with φ(j)=(k,i)\varphi(j)=(k,i) is nonempty, and its least element j0j_{0} is a first index, so x∈Ak,i=Cj0x\in A_{k,i}=C_{j_{0}}.

Its weight is at most s+εs+\varepsilon. Let n∈Nn\in\mathbb{N}. If j<j′≤nj<j'\le n are both first indices, then φ(j)≠φ(j′)\varphi(j)\ne\varphi(j') by the definition of a first index applied to j′j'; so the pairs φ(j)\varphi(j), for the first indices j≤nj\le n, are pairwise distinct. Let KK and II be the largest first, respectively second, coordinates of the finitely many pairs φ(1),…,φ(n)\varphi(1),\dots,\varphi(n). Since all terms are nonnegative and distinct first indices contribute distinct pairs in {1,…,K}×{1,…,I}\{1,\dots,K\}\times\{1,\dots,I\},

∑j=1nm(Cj)≤∑k=1K∑i=1Im(Ak,i)≤∑k=1Kwk≤∑k=1Km∗(Ek)+ε∑k=1K(1/2)k≤s+ε,\sum_{j=1}^{n}m(C_{j})\le\sum_{k=1}^{K}\sum_{i=1}^{I}m(A_{k,i})\le\sum_{k=1}^{K}w_{k}\le\sum_{k=1}^{K}m^{*}(E_{k})+\varepsilon\sum_{k=1}^{K}(1/2)^{k}\le s+\varepsilon ,

where the second inequality holds because each partial sum ∑i=1Im(Ak,i)\sum_{i=1}^{I}m(A_{k,i}) is at most the real weight wkw_{k}, and the last because ∑k=1K(1/2)k≤1\sum_{k=1}^{K}(1/2)^{k}\le1: by Series of Nonnegative Real Numbers, Comparison, and the Geometric Series §geometric the series ∑k=1∞(1/2)k\sum_{k=1}^{\infty}(1/2)^{k} converges with sum 11, and its partial sums are at most its sum by Series of Nonnegative Real Numbers, Comparison, and the Geometric Series §dominates. Hence every partial sum of (m(Cj))j∈N(m(C_{j}))_{j\in\mathbb{N}} is at most s+εs+\varepsilon, so the weight of (Cj)(C_{j}) is a real number at most s+εs+\varepsilon, and by (a) m∗(E)≤s+εm^{*}(E)\le s+\varepsilon. As ε>0\varepsilon>0 was arbitrary, m∗(E)≤sm^{*}(E)\le s, which is property 3.

Step 3: m∗(A)=m(A)m^{*}(A)=m(A) for every A∈AA\in\mathcal{A}. The sequence A,∅,∅,…A,\varnothing,\varnothing,\dots is a cover of AA of weight m(A)m(A), so m∗(A)≤m(A)m^{*}(A)\le m(A). Conversely, let (Ai)i∈N(A_{i})_{i\in\mathbb{N}} be a cover of AA with real weight ww. For i∈Ni\in\mathbb{N} let UiU_{i} be the union of the sets AjA_{j} with j∈Nj\in\mathbb{N} and j<ij<i (so U1=∅U_{1}=\varnothing), and let Bi=(A∩Ai)∖UiB_{i}=(A\cap A_{i})\setminus U_{i}; by Step 0, UiU_{i} and BiB_{i} belong to A\mathcal{A}. The sets BiB_{i} are pairwise disjoint: if i<i′i<i', then Bi⊆Ai⊆Ui′B_{i}\subseteq A_{i}\subseteq U_{i'} while Bi′∩Ui′=∅B_{i'}\cap U_{i'}=\varnothing. Their union is AA: each Bi⊆AB_{i}\subseteq A, and if x∈Ax\in A, the set of ii with x∈Aix\in A_{i} is nonempty, and for its least element ii we have x∉Uix\notin U_{i}, so x∈Bix\in B_{i}. By the countable additivity of mm, the series ∑i=1∞m(Bi)\sum_{i=1}^{\infty}m(B_{i}) converges with sum m(A)m(A). Since Bi⊆AiB_{i}\subseteq A_{i}, Step 0 gives m(Bi)≤m(Ai)m(B_{i})\le m(A_{i}), so ∑i=1nm(Bi)≤∑i=1nm(Ai)≤w\sum_{i=1}^{n}m(B_{i})\le\sum_{i=1}^{n}m(A_{i})\le w for every nn; by Series of Nonnegative Real Numbers, Comparison, and the Geometric Series §criterion, m(A)m(A) is the least upper bound of these partial sums, so m(A)≤wm(A)\le w. Thus m(A)m(A) is a lower bound of W(A)W(A), and m(A)≤m∗(A)m(A)\le m^{*}(A).

Step 4: every A∈AA\in\mathcal{A} is Carath'eodory measurable with respect to m∗m^{*}. Let A∈AA\in\mathcal{A} and E⊆ΩE\subseteq\Omega. Applying property 3 of Step 2 to the sequence E∩A, E∖A, ∅,∅,…E\cap A,\ E\setminus A,\ \varnothing,\varnothing,\dots, whose union is EE, and using m∗(∅)=0m^{*}(\varnothing)=0, we get m∗(E)≤m∗(E∩A)+m∗(E∖A)m^{*}(E)\le m^{*}(E\cap A)+m^{*}(E\setminus A). For the reverse inequality let a real ε>0\varepsilon>0 be given and, by (b), choose a cover (Ai)i∈N(A_{i})_{i\in\mathbb{N}} of EE with real weight w<m∗(E)+εw<m^{*}(E)+\varepsilon. Then (Ai∩A)i∈N(A_{i}\cap A)_{i\in\mathbb{N}} is a cover of E∩AE\cap A and (Ai∖A)i∈N(A_{i}\setminus A)_{i\in\mathbb{N}} is a cover of E∖AE\setminus A, all terms belonging to A\mathcal{A} by Step 0, and Step 0 also gives m(Ai)=m(Ai∩A)+m(Ai∖A)m(A_{i})=m(A_{i}\cap A)+m(A_{i}\setminus A). Write Pn=∑i=1nm(Ai∩A)P_{n}=\sum_{i=1}^{n}m(A_{i}\cap A) and Qn=∑i=1nm(Ai∖A)Q_{n}=\sum_{i=1}^{n}m(A_{i}\setminus A); these are nondecreasing in nn and Pn+Qn=∑i=1nm(Ai)≤wP_{n}+Q_{n}=\sum_{i=1}^{n}m(A_{i})\le w. For n,n′∈Nn,n'\in\mathbb{N} and N=max⁡{n,n′}N=\max\{n,n'\} we get Pn+Qn′≤PN+QN≤wP_{n}+Q_{n'}\le P_{N}+Q_{N}\le w; hence the weights w′=sup⁡nPnw'=\sup_{n}P_{n} and w′′=sup⁡nQnw''=\sup_{n}Q_{n} of the two covers are real and w′+w′′≤ww'+w''\le w. By (a),

m∗(E∩A)+m∗(E∖A)≤w′+w′′≤w<m∗(E)+ε.m^{*}(E\cap A)+m^{*}(E\setminus A)\le w'+w''\le w<m^{*}(E)+\varepsilon .

As ε>0\varepsilon>0 was arbitrary, m∗(E∩A)+m∗(E∖A)≤m∗(E)m^{*}(E\cap A)+m^{*}(E\setminus A)\le m^{*}(E), and so m∗(E)=m∗(E∩A)+m∗(E∖A)m^{*}(E)=m^{*}(E\cap A)+m^{*}(E\setminus A).

Step 5: existence (claim 1). Let M\mathcal{M} be the family of subsets of Ω\Omega that are Carath'eodory measurable with respect to m∗m^{*}. By claim 1 of Caratheodory Extension Theorem, M\mathcal{M} is a σ\sigma-algebra on Ω\Omega, and by Step 4 it contains A\mathcal{A}; hence σ(A)⊆M\sigma(\mathcal{A})\subseteq\mathcal{M} by claim 2 of Intersections of Sigma-Algebras and Minimality of the Generated Sigma-Algebra. By claim 2 of Caratheodory Extension Theorem, the restriction of m∗m^{*} to M\mathcal{M} is a measure on (Ω,M)(\Omega,\mathcal{M}). Let mˉ\bar{m} be the restriction of m∗m^{*} to σ(A)\sigma(\mathcal{A}). It takes values in [0,∞][0,\infty], satisfies mˉ(∅)=m∗(∅)=0\bar{m}(\varnothing)=m^{*}(\varnothing)=0, and every sequence of pairwise disjoint members of σ(A)\sigma(\mathcal{A}) is a sequence of pairwise disjoint members of M\mathcal{M} whose union lies in the σ\sigma-algebra σ(A)\sigma(\mathcal{A}); so the countable additivity of m∗m^{*} on M\mathcal{M} gives that of mˉ\bar{m} on σ(A)\sigma(\mathcal{A}), and mˉ\bar{m} is a measure on (Ω,σ(A))(\Omega,\sigma(\mathcal{A})) in the sense of Measure, Measure Space, and Probability Measure. By Step 3, mˉ(A)=m∗(A)=m(A)\bar{m}(A)=m^{*}(A)=m(A) for every A∈AA\in\mathcal{A}; in particular mˉ(Ω)=m(Ω)\bar{m}(\Omega)=m(\Omega) is a real number, so mˉ(Ω)<∞\bar{m}(\Omega)<\infty and mˉ\bar{m} is finite.

Step 6: uniqueness (claim 2). The family A\mathcal{A} is a π\pi-system in the sense of Dynkin's Pi-Lambda Theorem: it is nonempty since Ω∈A\Omega\in\mathcal{A}, and it is closed under finite intersections by Step 0. The σ\sigma-algebra σ(A)\sigma(\mathcal{A}) of Generated Sigma-Algebra, the family of subsets of Ω\Omega belonging to every σ\sigma-algebra on Ω\Omega containing A\mathcal{A}, is the intersection of all σ\sigma-algebras on Ω\Omega containing A\mathcal{A}, which is the generated σ\sigma-algebra of Generated Sigma-Algebra. Let mˉ1\bar{m}_{1} and mˉ2\bar{m}_{2} be as in claim 2. They agree on A\mathcal{A}, and mˉ1(Ω)=mˉ2(Ω)=m(Ω)<∞\bar{m}_{1}(\Omega)=\bar{m}_{2}(\Omega)=m(\Omega)<\infty. Claim 1 of Uniqueness of Finite Measures on a Generating Pi-System and the Density of the Exponential Law, applied to the measurable space (Ω,σ(A))(\Omega,\sigma(\mathcal{A})), the π\pi-system A\mathcal{A} and the measures mˉ1\bar{m}_{1} and mˉ2\bar{m}_{2}, gives mˉ1=mˉ2\bar{m}_{1}=\bar{m}_{2}.

Citations

Loading…

Dependencies

Uses0

Loading…

Comments

Log in to comment.

Loading…