Each result cited is universally quantified over the data in its own statement.
Throughout, N={1,2,3,…} as in Natural Numbers, and for a sequence (ai)i∈N in [0,∞] the sum ∑i∈Nai is the one of Measure, Measure Space, and Probability Measure: if every ai is real and the partial sums ∑i=1nai are bounded above, it is their least upper bound, and otherwise it is ∞. Two consequences of this definition are used repeatedly: if all ai are real and c is a real number with ∑i=1nai≤c for every n∈N, then ∑i∈Nai≤c; and if ∑i∈Nai is real, then every partial sum is at most ∑i∈Nai. In particular, if ai=0 for all i outside a finite set F⊆N, the partial sums are eventually constant equal to ∑i∈Fai, so ∑i∈Nai=∑i∈Fai. For a sequence (ai) of nonnegative reals, Series of Nonnegative Real Numbers, Comparison, and the Geometric Series §criterion shows that the series ∑i=1∞ai 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.
Step 0: elementary properties of A and m. Let A,B∈A. Then A∩B=Ω∖((Ω∖A)∪(Ω∖B))∈A and A∖B=A∩(Ω∖B)∈A, and by induction on the number of sets every finite union of members of A belongs to A, the empty union being ∅∈A. If A and B are disjoint, the sequence (Ci)i∈N with C1=A, C2=B and Ci=∅ for i≥3 consists of pairwise disjoint members of A with union A∪B∈A, so by the countable additivity of m the series ∑i=1∞m(Ci) converges to m(A∪B); since m(∅)=0 its partial sums equal m(A)+m(B) from n=2 on, so
m(A∪B)=m(A)+m(B)(A,B∈A disjoint).
If A⊆B, then B is the disjoint union of A and B∖A∈A, so m(B)=m(A)+m(B∖A)≥m(A). In particular m(A)≤m(Ω) for every A∈A.
Step 1: the outer measure m∗. For E⊆Ω, call a sequence (Ai)i∈N of members of A a cover of E if E⊆⋃i∈NAi, and call ∑i∈Nm(Ai)∈[0,∞] its weight. Let W(E) be the set of those weights of covers of E that are real numbers. The sequence Ω,∅,∅,… is a cover of E of weight m(Ω), so W(E) is a nonempty set of real numbers, and it is bounded below by 0. Define m∗(E) to be the greatest lower bound of W(E). Then m∗(E) is a real number with 0≤m∗(E)≤m(Ω), and:
(a) every cover of E has weight at least m∗(E) (for a real weight because m∗(E) is a lower bound of W(E), for the weight ∞ by the conventions of Measure, Measure Space, and Probability Measure);
(b) for every real ε>0 there is a cover of E with real weight less than m∗(E)+ε, because m∗(E)+ε is not a lower bound of W(E).
Step 2: m∗ is an outer measure on Ω in the sense of Outer Measure and Caratheodory Measurability. Its values lie in [0,∞)⊆[0,∞].
Property 1. The sequence all of whose terms are ∅ is a cover of ∅ of weight 0, so 0≤m∗(∅)≤0.
Property 2. If E⊆F⊆Ω, every cover of F is a cover of E, so W(F)⊆W(E) and therefore m∗(E)≤m∗(F).
Property 3. Let (Ek)k∈N be a sequence of subsets of Ω and put E=⋃k∈NEk. If ∑k∈Nm∗(Ek)=∞ there is nothing to prove, so assume that s=∑k∈Nm∗(Ek) is real; then ∑k=1Km∗(Ek)≤s for every K∈N. Let a real ε>0 be given. For each k∈N, by (b) choose a cover (Ak,i)i∈N of Ek whose weight wk is real and satisfies wk<m∗(Ek)+ε(1/2)k. The set N is countable by claim 1 of Basic Properties of Countable Sets, so N×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 in N×N whose set of terms is N×N. Call j∈N a first index if φ(j′)=φ(j) for every j′∈N with j′<j, and define Cj=Aφ(j) if j is a first index and Cj=∅ otherwise; here A(k,i) means Ak,i. Each Cj belongs to A.
The sequence (Cj)j∈N is a cover of E: if x∈E, choose k with x∈Ek and i with x∈Ak,i; the set of j∈N with φ(j)=(k,i) is nonempty, and its least element j0 is a first index, so x∈Ak,i=Cj0.
Its weight is at most s+ε. Let n∈N. If j<j′≤n are both first indices, then φ(j)=φ(j′) by the definition of a first index applied to j′; so the pairs φ(j), for the first indices j≤n, are pairwise distinct. Let K and I be the largest first, respectively second, coordinates of the finitely many pairs φ(1),…,φ(n). Since all terms are nonnegative and distinct first indices contribute distinct pairs in {1,…,K}×{1,…,I},
j=1∑nm(Cj)≤k=1∑Ki=1∑Im(Ak,i)≤k=1∑Kwk≤k=1∑Km∗(Ek)+εk=1∑K(1/2)k≤s+ε,
where the second inequality holds because each partial sum ∑i=1Im(Ak,i) is at most the real weight wk, and the last because ∑k=1K(1/2)k≤1: by Series of Nonnegative Real Numbers, Comparison, and the Geometric Series §geometric the series ∑k=1∞(1/2)k converges with sum 1, 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 is at most s+ε, so the weight of (Cj) is a real number at most s+ε, and by (a) m∗(E)≤s+ε. As ε>0 was arbitrary, m∗(E)≤s, which is property 3.
Step 3: m∗(A)=m(A) for every A∈A. The sequence A,∅,∅,… is a cover of A of weight m(A), so m∗(A)≤m(A). Conversely, let (Ai)i∈N be a cover of A with real weight w. For i∈N let Ui be the union of the sets Aj with j∈N and j<i (so U1=∅), and let Bi=(A∩Ai)∖Ui; by Step 0, Ui and Bi belong to A. The sets Bi are pairwise disjoint: if i<i′, then Bi⊆Ai⊆Ui′ while Bi′∩Ui′=∅. Their union is A: each Bi⊆A, and if x∈A, the set of i with x∈Ai is nonempty, and for its least element i we have x∈/Ui, so x∈Bi. By the countable additivity of m, the series ∑i=1∞m(Bi) converges with sum m(A). Since Bi⊆Ai, Step 0 gives m(Bi)≤m(Ai), so ∑i=1nm(Bi)≤∑i=1nm(Ai)≤w for every n; by Series of Nonnegative Real Numbers, Comparison, and the Geometric Series §criterion, m(A) is the least upper bound of these partial sums, so m(A)≤w. Thus m(A) is a lower bound of W(A), and m(A)≤m∗(A).
Step 4: every A∈A is Carath'eodory measurable with respect to m∗. Let A∈A and E⊆Ω. Applying property 3 of Step 2 to the sequence E∩A, E∖A, ∅,∅,…, whose union is E, and using m∗(∅)=0, we get m∗(E)≤m∗(E∩A)+m∗(E∖A). For the reverse inequality let a real ε>0 be given and, by (b), choose a cover (Ai)i∈N of E with real weight w<m∗(E)+ε. Then (Ai∩A)i∈N is a cover of E∩A and (Ai∖A)i∈N is a cover of E∖A, all terms belonging to A by Step 0, and Step 0 also gives m(Ai)=m(Ai∩A)+m(Ai∖A). Write Pn=∑i=1nm(Ai∩A) and Qn=∑i=1nm(Ai∖A); these are nondecreasing in n and Pn+Qn=∑i=1nm(Ai)≤w. For n,n′∈N and N=max{n,n′} we get Pn+Qn′≤PN+QN≤w; hence the weights w′=supnPn and w′′=supnQn of the two covers are real and w′+w′′≤w. By (a),
m∗(E∩A)+m∗(E∖A)≤w′+w′′≤w<m∗(E)+ε.
As ε>0 was arbitrary, m∗(E∩A)+m∗(E∖A)≤m∗(E), and so m∗(E)=m∗(E∩A)+m∗(E∖A).
Step 5: existence (claim 1). Let M be the family of subsets of Ω that are Carath'eodory measurable with respect to m∗. By claim 1 of Caratheodory Extension Theorem, M is a σ-algebra on Ω, and by Step 4 it contains A; hence σ(A)⊆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∗ to M is a measure on (Ω,M). Let mˉ be the restriction of m∗ to σ(A). It takes values in [0,∞], satisfies mˉ(∅)=m∗(∅)=0, and every sequence of pairwise disjoint members of σ(A) is a sequence of pairwise disjoint members of M whose union lies in the σ-algebra σ(A); so the countable additivity of m∗ on M gives that of mˉ on σ(A), and mˉ is a measure on (Ω,σ(A)) in the sense of Measure, Measure Space, and Probability Measure. By Step 3, mˉ(A)=m∗(A)=m(A) for every A∈A; in particular mˉ(Ω)=m(Ω) is a real number, so mˉ(Ω)<∞ and mˉ is finite.
Step 6: uniqueness (claim 2). The family A is a π-system in the sense of Dynkin's Pi-Lambda Theorem: it is nonempty since Ω∈A, and it is closed under finite intersections by Step 0. The σ-algebra σ(A) of Generated Sigma-Algebra, the family of subsets of Ω belonging to every σ-algebra on Ω containing A, is the intersection of all σ-algebras on Ω containing A, which is the generated σ-algebra of Generated Sigma-Algebra. Let mˉ1 and mˉ2 be as in claim 2. They agree on A, and mˉ1(Ω)=mˉ2(Ω)=m(Ω)<∞. 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)), the π-system A and the measures mˉ1 and mˉ2, gives mˉ1=mˉ2.