TheoremBase

After the elementary properties of a finite signed measure, a maximizing sequence of sets is chosen (the only use of dependent choice) and refined, through the finite partitions it generates, into a set of maximal measure, which is a Hahn set; uniqueness, and then the Jordan decomposition with its independence of the Hahn set and its absolute continuity, follow directly.

Proof

Step 1: elementary properties of ν\nu. Throughout, ν\nu is the finite signed measure of the statement, and the field and order laws of R\mathbb{R} are used freely (The Structure of ℝ as a Field and an Ordered Ring (i), The Structure of ℝ as a Field and an Ordered Ring (ii), Rules of Arithmetic and Order in an Ordered Field (iv), Rules of Arithmetic and Order in an Ordered Field (v), Rules of Arithmetic and Order in an Ordered Field (viii)).

(a) Finite additivity. Let JJ be a finite set and (Bj)j∈J(B_{j})_{j\in J} a family in F\mathcal{F} with Bi∩Bj=∅B_{i}\cap B_{j}=\emptyset for i≠ji\neq j. Then, with the finite sum over JJ,

ν(⋃j∈JBj)=∑j∈Jν(Bj).\nu\Bigl(\bigcup_{j\in J}B_{j}\Bigr)=\sum_{j\in J}\nu(B_{j}).

For J=∅J=\emptyset both sides are 00. Otherwise list JJ as j1,…,jrj_{1},\dots,j_{r} and apply countable additivity to the pairwise disjoint sequence Bj1,…,Bjr,∅,∅,…B_{j_{1}},\dots,B_{j_{r}},\emptyset,\emptyset,\dots: since ν(∅)=0\nu(\emptyset)=0, its partial sums equal ∑j∈Jν(Bj)\sum_{j\in J}\nu(B_{j}) from the rr-th on, so this is the sum of the series (Limits of Sequences of Real Numbers: Uniqueness, Boundedness, Constants, Tails, Arithmetic, Quotients, Absolute Values, Finite Sums, Order, Squeezing, Domination and Subsequences (i), Limits of Sequences of Real Numbers: Uniqueness, Boundedness, Constants, Tails, Arithmetic, Quotients, Absolute Values, Finite Sums, Order, Squeezing, Domination and Subsequences (v)). In particular, for A,B∈FA,B\in\mathcal{F} with B⊆AB\subseteq A, AA is the disjoint union of BB and A∖BA\setminus B, so

ν(A∖B)=ν(A)−ν(B).\nu(A\setminus B)=\nu(A)-\nu(B).

(b) Continuity from below. Let (Bm)(B_{m}) be a sequence in F\mathcal{F} with Bm⊆Bm+1B_{m}\subseteq B_{m+1} for every mm, and B=⋃m∈NBmB=\bigcup_{m\in\mathbb{N}}B_{m}. Its disjointing sequence (Dm)(D_{m}) is a pairwise disjoint sequence in F\mathcal{F} with union BB (The Disjointing Sequence of a Sequence of Sets (v), The Disjointing Sequence of a Sequence of Sets (iii), The Disjointing Sequence of a Sequence of Sets (iv)), and since (Bm)(B_{m}) increases, D1∪⋯∪Dm=BmD_{1}\cup\dots\cup D_{m}=B_{m} (The Disjointing Sequence of a Sequence of Sets (i), Monotone Sequences of Sets, Unions of Traces, and Disjointness of Smaller Sets (i)). By (a) the mm-th partial sum of ∑kν(Dk)\sum_{k}\nu(D_{k}) is ν(Bm)\nu(B_{m}), so countable additivity gives

ν(Bm)→ν(B).\nu(B_{m})\to\nu(B).

(c) Continuity from above. Let (Em)(E_{m}) be a sequence in F\mathcal{F} with Em+1⊆EmE_{m+1}\subseteq E_{m} for every mm, and E=⋂m∈NEmE=\bigcap_{m\in\mathbb{N}}E_{m}. The sets E1∖EmE_{1}\setminus E_{m} increase and have union E1∖EE_{1}\setminus E, so by (b) and (a), ν(E1)−ν(Em)→ν(E1)−ν(E)\nu(E_{1})-\nu(E_{m})\to\nu(E_{1})-\nu(E), that is (Limits of Sequences of Real Numbers: Uniqueness, Boundedness, Constants, Tails, Arithmetic, Quotients, Absolute Values, Finite Sums, Order, Squeezing, Domination and Subsequences (vi))

ν(Em)→ν(E).\nu(E_{m})\to\nu(E).

Step 2: a maximizing sequence. Assume the axiom of dependent choice. Let

α=sup⁡{ν(A):A∈F}∈[−∞,∞];\alpha=\sup\{\nu(A):A\in\mathcal{F}\}\in[-\infty,\infty];

α≥ν(∅)=0\alpha\ge\nu(\emptyset)=0. For n∈Nn\in\mathbb{N} put tn=α−1/nt_{n}=\alpha-1/n if α∈R\alpha\in\mathbb{R}, and tn=nt_{n}=n if α=∞\alpha=\infty. Since tn<αt_{n}<\alpha, the set {A∈F:ν(A)>tn}\{A\in\mathcal{F}:\nu(A)>t_{n}\} is nonempty (Order and Suprema in the Extended Real Line (x)). By countable choice, which follows from dependent choice by Countable Choice Follows from Dependent Choice (i), there is a sequence (An)(A_{n}) in F\mathcal{F} with

ν(An)>tn(n∈N).\nu(A_{n})>t_{n}\qquad(n\in\mathbb{N}).

This is the only use of the axiom of dependent choice.

Step 3: the atoms generated by (An)(A_{n}). For n∈Nn\in\mathbb{N} let En\mathcal{E}_{n} be the finite set of maps ε:[n]→{0,1}\varepsilon:[n]\to\{0,1\}, and for ε∈En\varepsilon\in\mathcal{E}_{n} let

Bn,ε=⋂k=1nAkε(k),Ak1=Ak,Ak0=X∖Ak.B_{n,\varepsilon}=\bigcap_{k=1}^{n}A_{k}^{\varepsilon(k)},\qquad A_{k}^{1}=A_{k},\quad A_{k}^{0}=X\setminus A_{k}.

For fixed nn, these atoms of generation nn lie in F\mathcal{F}, are pairwise disjoint and cover XX: a point x∈Xx\in X lies in Bn,εB_{n,\varepsilon} exactly for the ε\varepsilon with ε(k)=1⇔x∈Ak\varepsilon(k)=1\Leftrightarrow x\in A_{k} (k∈[n]k\in[n]). Consequently AnA_{n} is the disjoint union of the atoms Bn,εB_{n,\varepsilon} with ε(n)=1\varepsilon(n)=1. Let CnC_{n} be the union of the atoms of generation nn of positive measure:

Cn=⋃{Bn,ε:ε∈En, ν(Bn,ε)>0}.C_{n}=\bigcup\{B_{n,\varepsilon}:\varepsilon\in\mathcal{E}_{n},\ \nu(B_{n,\varepsilon})>0\}.

By Step 1(a), dropping the terms that are not positive and then adding positive ones,

ν(An)=∑ε∈En, ε(n)=1ν(Bn,ε)≤∑ε∈En, ε(n)=1, ν(Bn,ε)>0ν(Bn,ε)≤∑ε∈En, ν(Bn,ε)>0ν(Bn,ε)=ν(Cn).\nu(A_{n})=\sum_{\varepsilon\in\mathcal{E}_{n},\,\varepsilon(n)=1}\nu(B_{n,\varepsilon})\le\sum_{\varepsilon\in\mathcal{E}_{n},\,\varepsilon(n)=1,\,\nu(B_{n,\varepsilon})>0}\nu(B_{n,\varepsilon})\le\sum_{\varepsilon\in\mathcal{E}_{n},\,\nu(B_{n,\varepsilon})>0}\nu(B_{n,\varepsilon})=\nu(C_{n}).

Step 4: the unions Cn∪⋯∪CmC_{n}\cup\dots\cup C_{m}. Let j≤mj\le m. An atom Bm,δB_{m,\delta} of generation mm lies in the single atom Bj,δ∣[j]B_{j,\delta|_{[j]}} of generation jj and is disjoint from all other atoms of generation jj; as CjC_{j} is a union of atoms of generation jj, Bm,δB_{m,\delta} is therefore contained in CjC_{j} or disjoint from it. Now fix nn and put Fn,m=Cn∪Cn+1∪⋯∪CmF_{n,m}=C_{n}\cup C_{n+1}\cup\dots\cup C_{m} for m≥nm\ge n. Let m>nm>n. By what was just shown, every atom of generation mm lies in Fn,m−1F_{n,m-1} or is disjoint from it, so Cm∖Fn,m−1C_{m}\setminus F_{n,m-1} is the union of the atoms Bm,εB_{m,\varepsilon} with ν(Bm,ε)>0\nu(B_{m,\varepsilon})>0 that are disjoint from Fn,m−1F_{n,m-1}. By Step 1(a), ν(Cm∖Fn,m−1)≥0\nu(C_{m}\setminus F_{n,m-1})\ge0, and since Fn,mF_{n,m} is the disjoint union of Fn,m−1F_{n,m-1} and Cm∖Fn,m−1C_{m}\setminus F_{n,m-1},

ν(Fn,m)=ν(Fn,m−1)+ν(Cm∖Fn,m−1)≥ν(Fn,m−1).\nu(F_{n,m})=\nu(F_{n,m-1})+\nu(C_{m}\setminus F_{n,m-1})\ge\nu(F_{n,m-1}).

By induction on mm and Step 3,

ν(Fn,m)≥ν(Fn,n)=ν(Cn)≥ν(An)(m≥n).\nu(F_{n,m})\ge\nu(F_{n,n})=\nu(C_{n})\ge\nu(A_{n})\qquad(m\ge n).

Step 5: a set of maximal measure. Let En=⋃m≥nCmE_{n}=\bigcup_{m\ge n}C_{m}, the union of the increasing sequence (Fn,n+k−1)k∈N(F_{n,n+k-1})_{k\in\mathbb{N}}. By Step 1(b) and Step 4 (Limits of Sequences of Real Numbers: Uniqueness, Boundedness, Constants, Tails, Arithmetic, Quotients, Absolute Values, Finite Sums, Order, Squeezing, Domination and Subsequences (x)),

ν(En)≥ν(An)>tn(n∈N).\nu(E_{n})\ge\nu(A_{n})>t_{n}\qquad(n\in\mathbb{N}).

The sets EnE_{n} decrease; let D=⋂n∈NEn∈FD=\bigcap_{n\in\mathbb{N}}E_{n}\in\mathcal{F}. By Step 1(c), ν(En)→ν(D)\nu(E_{n})\to\nu(D). If α=∞\alpha=\infty, then ν(En)>n\nu(E_{n})>n for every nn, which is impossible for a convergent, hence bounded, sequence (Limits of Sequences of Real Numbers: Uniqueness, Boundedness, Constants, Tails, Arithmetic, Quotients, Absolute Values, Finite Sums, Order, Squeezing, Domination and Subsequences (iii)). So α∈R\alpha\in\mathbb{R}, and for every N∈NN\in\mathbb{N} we have ν(En)>α−1/N\nu(E_{n})>\alpha-1/N for all n≥Nn\ge N, hence ν(D)≥α−1/N\nu(D)\ge\alpha-1/N (Limits of Sequences of Real Numbers: Uniqueness, Boundedness, Constants, Tails, Arithmetic, Quotients, Absolute Values, Finite Sums, Order, Squeezing, Domination and Subsequences (x)). Thus ν(D)≥α\nu(D)\ge\alpha, and ν(D)≤α\nu(D)\le\alpha by the definition of α\alpha:

ν(D)=α=sup⁡{ν(A):A∈F}.\nu(D)=\alpha=\sup\{\nu(A):A\in\mathcal{F}\}.

Step 6: DD is a Hahn set. Let A∈FA\in\mathcal{F}. If A⊆DA\subseteq D and ν(A)<0\nu(A)<0, then ν(D∖A)=ν(D)−ν(A)>α\nu(D\setminus A)=\nu(D)-\nu(A)>\alpha by Step 1(a), contradicting the definition of α\alpha. If A⊆X∖DA\subseteq X\setminus D and ν(A)>0\nu(A)>0, then ν(D∪A)=ν(D)+ν(A)>α\nu(D\cup A)=\nu(D)+\nu(A)>\alpha, again impossible. So DD is a Hahn set for ν\nu. This proves (i).

Step 7: uniqueness. Let P,P′P,P' be Hahn sets for ν\nu and A∈FA\in\mathcal{F} with A⊆(P∖P′)∪(P′∖P)A\subseteq(P\setminus P')\cup(P'\setminus P). The set A∩(P∖P′)A\cap(P\setminus P') lies in PP and in X∖P′X\setminus P', so its measure is both ≥0\ge0 and ≤0\le0, hence 00; likewise ν(A∩(P′∖P))=0\nu(A\cap(P'\setminus P))=0. These two sets are disjoint with union AA, so ν(A)=0\nu(A)=0 by Step 1(a).

Now let A∈FA\in\mathcal{F} be arbitrary. The set A∩PA\cap P is the disjoint union of A∩P∩P′A\cap P\cap P' and A∩(P∖P′)A\cap(P\setminus P'), and A∩P′A\cap P' is the disjoint union of A∩P∩P′A\cap P\cap P' and A∩(P′∖P)A\cap(P'\setminus P); the second sets have measure 00 by what was just shown. By Step 1(a),

ν(A∩P)=ν(A∩P∩P′)=ν(A∩P′),ν(A∖P)=ν(A)−ν(A∩P)=ν(A)−ν(A∩P′)=ν(A∖P′).\nu(A\cap P)=\nu(A\cap P\cap P')=\nu(A\cap P'),\qquad\nu(A\setminus P)=\nu(A)-\nu(A\cap P)=\nu(A)-\nu(A\cap P')=\nu(A\setminus P').

This proves (iii).

Step 8: the Jordan decomposition. Let PP be a Hahn set for ν\nu. For A∈FA\in\mathcal{F} we have A∩P⊆PA\cap P\subseteq P and A∖P=A∩(X∖P)⊆X∖PA\setminus P=A\cap(X\setminus P)\subseteq X\setminus P, so ν+(A)=ν(A∩P)\nu^{+}(A)=\nu(A\cap P) and ν−(A)=−ν(A∖P)\nu^{-}(A)=-\nu(A\setminus P) are real and nonnegative, hence elements of [0,∞][0,\infty]; and ν±(∅)=0\nu^{\pm}(\emptyset)=0. Let (Am)(A_{m}) be a pairwise disjoint sequence in F\mathcal{F} with union AA. The sequences (Am∩P)(A_{m}\cap P) and (Am∩(X∖P))(A_{m}\cap(X\setminus P)) are pairwise disjoint, with unions A∩PA\cap P and A∖PA\setminus P (Monotone Sequences of Sets, Unions of Traces, and Disjointness of Smaller Sets (v), Monotone Sequences of Sets, Unions of Traces, and Disjointness of Smaller Sets (iii)). Countable additivity of ν\nu, together with Elementary Properties of Series of Real Numbers: Linearity, Null Terms, the Cauchy Criterion, Index Shifts, Tails, Order and Telescoping (i) for ν−\nu^{-}, gives convergent real series

∑m=1∞ν+(Am)=ν+(A),∑m=1∞ν−(Am)=ν−(A),\sum_{m=1}^{\infty}\nu^{+}(A_{m})=\nu^{+}(A),\qquad\sum_{m=1}^{\infty}\nu^{-}(A_{m})=\nu^{-}(A),

and these are also the sums in [0,∞][0,\infty] (Sums of Sequences in [0,∞]: Real Series, Linearity, Comparison, Rearrangement, Double Series and Monotone Convergence (vi)). So ν+\nu^{+} and ν−\nu^{-} are measures; they are finite because ν+(X)=ν(P)\nu^{+}(X)=\nu(P) and ν−(X)=−ν(X∖P)\nu^{-}(X)=-\nu(X\setminus P) are real (Small Facts about [0,∞]: Zero, Finiteness and Sums of Zeros (ii)). By Step 1(a),

ν(A)=ν(A∩P)+ν(A∖P)=ν+(A)−ν−(A)(A∈F).\nu(A)=\nu(A\cap P)+\nu(A\setminus P)=\nu^{+}(A)-\nu^{-}(A)\qquad(A\in\mathcal{F}).

By (iii), ν+\nu^{+} and ν−\nu^{-} do not change when PP is replaced by another Hahn set P′P'. Finally, let μ\mu be a measure on (X,F)(X,\mathcal{F}) with ν≪μ\nu\ll\mu in the sense of Absolute Continuity of Measures (ii), and let A∈FA\in\mathcal{F} with μ(A)=0\mu(A)=0. By monotonicity of μ\mu (Basic Properties of a Measure (iii)), μ(A∩P)=0\mu(A\cap P)=0 and μ(A∖P)=0\mu(A\setminus P)=0, so ν+(A)=ν(A∩P)=0\nu^{+}(A)=\nu(A\cap P)=0 and ν−(A)=−ν(A∖P)=0\nu^{-}(A)=-\nu(A\setminus P)=0. Thus ν+≪μ\nu^{+}\ll\mu and ν−≪μ\nu^{-}\ll\mu in the sense of Absolute Continuity of Measures (i). This proves (ii).

Citations

Loading…

Dependencies

Uses0

Loading…

Comments

Log in to comment.

Loading…