TheoremBase

Uniqueness: the integral of f−g over the set where f>g vanishes, on all of X for integrable functions and on pieces of finite measure where g is bounded otherwise. Existence follows Billingsley: for a finite measure, a maximal function among those whose integrals stay below ν is a density, by a Hahn-set argument; the σ-finite case pastes densities over a disjoint sequence, and signed measures follow from the Jordan decomposition. Dependent choice enters through two uses of countable choice and the Hahn decomposition.

Proof

Throughout, arithmetic and order in R\mathbb{R} use the field and ordered-ring laws (The Structure of ℝ as a Field and an Ordered Ring (i), The Structure of ℝ as a Field and an Ordered Ring (ii)) and the rules Rules of Arithmetic and Order in an Ordered Field (iv), Rules of Arithmetic and Order in an Ordered Field (vi) and Rules of Arithmetic and Order in an Ordered Field (viii). For a measure ρ\rho on (X,F)(X,\mathcal{F}), a measurable h:X→Rh:X\to\mathbb{R} with h≥0h\ge0 lies in M+\mathcal{M}^{+}, and for A∈FA\in\mathcal{F} it is integrable on AA with respect to ρ\rho exactly when its integral over AA in M+\mathcal{M}^{+} is finite, the two integrals then agreeing (Integrals over Measurable Subsets (iii)); and integrals over AA are integrals over XX of the product with 1A\mathbf{1}_{A} (Integrals over Measurable Subsets (i), Integrals over Measurable Subsets (ii)).

Uniqueness: a local statement. Let B∈FB\in\mathcal{F} be a set on which ff and gg are integrable, with ∫Af dμ=∫Ag dμ\int_{A}f\,d\mu=\int_{A}g\,d\mu for the integrals as integrable functions for every A∈FA\in\mathcal{F} with A⊆BA\subseteq B, and let

C=B∩{x∈X:f(x)>g(x)}.C=B\cap\{x\in X:f(x)>g(x)\}.

We show μ(C)=0\mu(C)=0. The set CC lies in F\mathcal{F}, as f−gf-g is measurable (Arithmetic of Measurable Real- and [0,∞]-Valued Maps (i), Measurable Maps: Generators, Composition, Continuous Maps, Restrictions, and Criteria for Real- and [0,∞]-Valued Maps (v)). Since f1B∈L1(μ)f\mathbf{1}_{B}\in\mathcal{L}^{1}(\mu) and ∣f1C∣≤∣f1B∣|f\mathbf{1}_{C}|\le|f\mathbf{1}_{B}|, also f1C∈L1(μ)f\mathbf{1}_{C}\in\mathcal{L}^{1}(\mu) (Linearity and Monotonicity of the Lebesgue Integral, and the Integrability Criterion (v), Linearity and Monotonicity of the Lebesgue Integral, and the Integrability Criterion (x)), and likewise g1C∈L1(μ)g\mathbf{1}_{C}\in\mathcal{L}^{1}(\mu). By linearity (Linearity and Monotonicity of the Lebesgue Integral, and the Integrability Criterion (vii)),

∫X(f−g)1C dμ=∫Cf dμ−∫Cg dμ=0.\int_{X}(f-g)\mathbf{1}_{C}\,d\mu=\int_{C}f\,d\mu-\int_{C}g\,d\mu=0 .

The function (f−g)1C(f-g)\mathbf{1}_{C} is nonnegative, so its integral in M+\mathcal{M}^{+} is 00 as well (Linearity and Monotonicity of the Lebesgue Integral, and the Integrability Criterion (vi)), and the set where it is positive, which is CC, has measure 00 (The Lebesgue Integral and Null Sets (iv)).

Uniqueness for integrable functions. If f,g∈L1(μ)f,g\in\mathcal{L}^{1}(\mu), both are integrable on B=XB=X (Integrals over Measurable Subsets (vi)), so μ({f>g})=0\mu(\{f>g\})=0 by the local statement, and μ({g>f})=0\mu(\{g>f\})=0 by the same argument with ff and gg exchanged. The set {f≠g}\{f\neq g\} is the union of these two sets of F\mathcal{F}, so it lies in F\mathcal{F} and has measure 00 (Basic Properties of a Measure (vii)). This proves (i).

Uniqueness for nonnegative functions. Let f,g≥0f,g\ge0 be measurable, with equal integrals in M+\mathcal{M}^{+} over every A∈FA\in\mathcal{F}. As μ\mu is σ\sigma-finite, there is a sequence (Xk)(X_{k}) in F\mathcal{F} with ⋃kXk=X\bigcup_{k}X_{k}=X and μ(Xk)<∞\mu(X_{k})<\infty for every kk; we fix one. Let E={f>g}∈FE=\{f>g\}\in\mathcal{F}, and for k,j∈Nk,j\in\mathbb{N} let

Bk,j=Xk∩{x∈X:g(x)≤j}∈F.B_{k,j}=X_{k}\cap\{x\in X:g(x)\le j\}\in\mathcal{F}.

Since g1Bk,j≤j1Xkg\mathbf{1}_{B_{k,j}}\le j\mathbf{1}_{X_{k}}, we get ∫Bk,jg dμ≤j μ(Xk)<∞\int_{B_{k,j}}g\,d\mu\le j\,\mu(X_{k})<\infty (Basic Properties of the Integral of Nonnegative Measurable Maps (iii), Basic Properties of the Integral of Nonnegative Measurable Maps (iv), Basic Properties of the Integral of Nonnegative Measurable Maps (ii)), and ∫Af dμ=∫Ag dμ\int_{A}f\,d\mu=\int_{A}g\,d\mu is finite for every A∈FA\in\mathcal{F} with A⊆Bk,jA\subseteq B_{k,j} (Integrals over Measurable Subsets (viii)). So ff and gg are integrable on Bk,jB_{k,j} with equal integrals over its measurable subsets, and the local statement gives μ(Bk,j∩E)=0\mu(B_{k,j}\cap E)=0. For fixed kk the sets Bk,j∩EB_{k,j}\cap E increase with jj, and their union is Xk∩EX_{k}\cap E because gg is real-valued; hence μ(Xk∩E)=sup⁡jμ(Bk,j∩E)=0\mu(X_{k}\cap E)=\sup_{j}\mu(B_{k,j}\cap E)=0 (Basic Properties of a Measure (viii)). As E=⋃k(Xk∩E)E=\bigcup_{k}(X_{k}\cap E), countable subadditivity gives μ(E)=0\mu(E)=0 (Basic Properties of a Measure (vi)). Exchanging ff and gg and combining as before proves (ii). All exceptional sets here are explicit members of F\mathcal{F}; no choice is used.

Existence when the measure is finite. We first prove: if ρ\rho is a finite measure on (X,F)(X,\mathcal{F}) and ν\nu a finite measure with ν≪ρ\nu\ll\rho, there is a measurable f:X→Rf:X\to\mathbb{R} with f≥0f\ge0 and ν(A)=∫Af dρ\nu(A)=\int_{A}f\,d\rho in M+\mathcal{M}^{+} for every A∈FA\in\mathcal{F}. Here ν(A)≤ν(X)<∞\nu(A)\le\nu(X)<\infty for every AA (Basic Properties of a Measure (iii)). Let K\mathcal{K} be the set of measurable g:X→Rg:X\to\mathbb{R} with g≥0g\ge0 and

∫Ag dρ≤ν(A)(A∈F).\int_{A}g\,d\rho\le\nu(A)\qquad(A\in\mathcal{F}).

The zero function lies in K\mathcal{K} (Basic Properties of the Integral of Nonnegative Measurable Maps (v)), and ∫Xg dρ≤ν(X)\int_{X}g\,d\rho\le\nu(X) for g∈Kg\in\mathcal{K}; let α=sup⁡{∫Xg dρ:g∈K}\alpha=\sup\{\int_{X}g\,d\rho:g\in\mathcal{K}\}, a real number with 0≤α≤ν(X)0\le\alpha\le\nu(X).

(a) K\mathcal{K} is closed under maxima. Let g,g′∈Kg,g'\in\mathcal{K} and A∈FA\in\mathcal{F}. The function max⁡(g,g′)\max(g,g') is measurable (Arithmetic of Measurable Real- and [0,∞]-Valued Maps (i)), and A1=A∩{g≥g′}A_{1}=A\cap\{g\ge g'\} and A2=A∩{g<g′}A_{2}=A\cap\{g<g'\} are disjoint sets of F\mathcal{F} with union AA, on which max⁡(g,g′)\max(g,g') equals gg and g′g' respectively. By Integrals over Measurable Subsets (vii) and the countable additivity of ν\nu (Basic Properties of a Measure (ii)),

∫Amax⁡(g,g′) dρ=∫A1g dρ+∫A2g′ dρ≤ν(A1)+ν(A2)=ν(A),\int_{A}\max(g,g')\,d\rho=\int_{A_{1}}g\,d\rho+\int_{A_{2}}g'\,d\rho\le\nu(A_{1})+\nu(A_{2})=\nu(A),

so max⁡(g,g′)∈K\max(g,g')\in\mathcal{K}.

(b) A maximizing sequence (countable choice). For n∈Nn\in\mathbb{N} the set Kn={g∈K:∫Xg dρ>α−1/n}\mathcal{K}_{n}=\{g\in\mathcal{K}:\int_{X}g\,d\rho>\alpha-1/n\} is nonempty, by the definition of α\alpha. Under the axiom of dependent choice, countable choice (Countable Choice Follows from Dependent Choice (i), with bb the set of measurable functions X→RX\to\mathbb{R}) gives a sequence (gn)(g_{n}) with gn∈Kng_{n}\in\mathcal{K}_{n} for every nn. Put f1=g1f_{1}=g_{1} and fn+1=max⁡(fn,gn+1)f_{n+1}=\max(f_{n},g_{n+1}). By (a) and induction, fn∈Kf_{n}\in\mathcal{K}; the sequence (fn)(f_{n}) is increasing; and α−1/n<∫Xgn dρ≤∫Xfn dρ≤α\alpha-1/n<\int_{X}g_{n}\,d\rho\le\int_{X}f_{n}\,d\rho\le\alpha. Let fˉ:X→[0,∞]\bar f:X\to[0,\infty] be the function fˉ(x)=sup⁡nfn(x)\bar f(x)=\sup_{n}f_{n}(x). By the monotone convergence theorem, applied to the increasing sequences (fn1A)(f_{n}\mathbf{1}_{A}) with supremum fˉ1A\bar f\mathbf{1}_{A}, fˉ∈M+\bar f\in\mathcal{M}^{+} and

∫Afˉ dρ=sup⁡n∫Afn dρ≤ν(A)(A∈F),∫Xfˉ dρ=α\int_{A}\bar f\,d\rho=\sup_{n}\int_{A}f_{n}\,d\rho\le\nu(A)\qquad(A\in\mathcal{F}),\qquad\int_{X}\bar f\,d\rho=\alpha

(Monotone Convergence Theorem (i), Monotone Convergence Theorem (ii)).

(c) A real-valued maximal element. As ∫Xfˉ dρ=α<∞\int_{X}\bar f\,d\rho=\alpha<\infty, the set N={fˉ=∞}N=\{\bar f=\infty\} lies in F\mathcal{F} and ρ(N)=0\rho(N)=0 (The Lebesgue Integral and Null Sets (v)). Let f=fˉ1X∖Nf=\bar f\mathbf{1}_{X\setminus N}; it is measurable (Arithmetic of Measurable Real- and [0,∞]-Valued Maps (iii)), real-valued and nonnegative, hence measurable as a function into R\mathbb{R} (Measurable Maps: Generators, Composition, Continuous Maps, Restrictions, and Criteria for Real- and [0,∞]-Valued Maps (vii)). For A∈FA\in\mathcal{F} the functions f1Af\mathbf{1}_{A} and fˉ1A\bar f\mathbf{1}_{A} agree outside the null set NN, so ∫Af dρ=∫Afˉ dρ\int_{A}f\,d\rho=\int_{A}\bar f\,d\rho (The Lebesgue Integral and Null Sets (ii)). Hence f∈Kf\in\mathcal{K} and ∫Xf dρ=α\int_{X}f\,d\rho=\alpha.

(d) ff is a density (Hahn decomposition). We show ν(X)=α\nu(X)=\alpha. Suppose instead that δ=ν(X)−α>0\delta=\nu(X)-\alpha>0, and put

ε=δ2(ρ(X)+1)>0,so thatε ρ(X)<δ.\varepsilon=\frac{\delta}{2(\rho(X)+1)}>0,\qquad\text{so that}\qquad\varepsilon\,\rho(X)<\delta .

The function f+εf+\varepsilon lies in M+\mathcal{M}^{+} with ∫X(f+ε) dρ=α+ερ(X)<∞\int_{X}(f+\varepsilon)\,d\rho=\alpha+\varepsilon\rho(X)<\infty (Linearity and Monotonicity of the Lebesgue Integral, and the Integrability Criterion (i), Basic Properties of the Integral of Nonnegative Measurable Maps (iv), Basic Properties of the Integral of Nonnegative Measurable Maps (ii)), so the measure (f+ε)ρ(f+\varepsilon)\rho with density f+εf+\varepsilon is a finite measure (Image Measures and Measures with a Density: Measures, Integrals and Change of Variables (v)), and

λ(A)=ν(A)−∫A(f+ε) dρ(A∈F)\lambda(A)=\nu(A)-\int_{A}(f+\varepsilon)\,d\rho\qquad(A\in\mathcal{F})

defines a finite signed measure λ\lambda (Finite Signed Measures from Finite Measures and from Integrals (i)). Under the axiom of dependent choice, λ\lambda has a Hahn set PP (Hahn and Jordan Decompositions of a Finite Signed Measure (i)). For A∈FA\in\mathcal{F} we have A∩P⊆PA\cap P\subseteq P, so λ(A∩P)≥0\lambda(A\cap P)\ge0, that is, ∫A∩P(f+ε) dρ≤ν(A∩P)\int_{A\cap P}(f+\varepsilon)\,d\rho\le\nu(A\cap P). Let g=f+ε1Pg=f+\varepsilon\mathbf{1}_{P}, a nonnegative measurable function equal to f+εf+\varepsilon on PP and to ff off PP. Since f∈Kf\in\mathcal{K},

∫Ag dρ=∫A∩P(f+ε) dρ+∫A∖Pf dρ≤ν(A∩P)+ν(A∖P)=ν(A)\int_{A}g\,d\rho=\int_{A\cap P}(f+\varepsilon)\,d\rho+\int_{A\setminus P}f\,d\rho\le\nu(A\cap P)+\nu(A\setminus P)=\nu(A)

(Integrals over Measurable Subsets (vii), Basic Properties of a Measure (ii)). Thus g∈Kg\in\mathcal{K}, and α≥∫Xg dρ=α+ερ(P)\alpha\ge\int_{X}g\,d\rho=\alpha+\varepsilon\rho(P) (Linearity and Monotonicity of the Lebesgue Integral, and the Integrability Criterion (i), Basic Properties of the Integral of Nonnegative Measurable Maps (iv), Basic Properties of the Integral of Nonnegative Measurable Maps (ii)), whence ρ(P)=0\rho(P)=0 and therefore ν(P)=0\nu(P)=0, as ν≪ρ\nu\ll\rho. On the other hand, the Hahn property applied to X∖PX\setminus P itself gives λ(X∖P)≤0\lambda(X\setminus P)\le0, that is, ν(X∖P)≤∫X∖Pf dρ+ερ(X∖P)≤α+ερ(X)\nu(X\setminus P)\le\int_{X\setminus P}f\,d\rho+\varepsilon\rho(X\setminus P)\le\alpha+\varepsilon\rho(X) (Integrals over Measurable Subsets (viii), Integrals over Measurable Subsets (v), Basic Properties of a Measure (iii)). Then

ν(X)=ν(P)+ν(X∖P)≤α+ερ(X)<α+δ=ν(X)\nu(X)=\nu(P)+\nu(X\setminus P)\le\alpha+\varepsilon\rho(X)<\alpha+\delta=\nu(X)

(Basic Properties of a Measure (ii)), a contradiction. Hence ν(X)=α\nu(X)=\alpha. Finally, for A∈FA\in\mathcal{F}, ∫Af dρ≤ν(A)\int_{A}f\,d\rho\le\nu(A) and ∫X∖Af dρ≤ν(X∖A)\int_{X\setminus A}f\,d\rho\le\nu(X\setminus A), while

∫Af dρ+∫X∖Af dρ=α=ν(X)=ν(A)+ν(X∖A)\int_{A}f\,d\rho+\int_{X\setminus A}f\,d\rho=\alpha=\nu(X)=\nu(A)+\nu(X\setminus A)

(Integrals over Measurable Subsets (vii), Integrals over Measurable Subsets (v), Basic Properties of a Measure (ii)). Both inequalities between real numbers are therefore equalities, and ν(A)=∫Af dρ\nu(A)=\int_{A}f\,d\rho.

Existence when μ\mu is σ\sigma-finite. Fix a sequence (Xm)(X_{m}) in F\mathcal{F} with ⋃mXm=X\bigcup_{m}X_{m}=X and μ(Xm)<∞\mu(X_{m})<\infty, and let (Bm)(B_{m}) be its disjointing sequence: the BmB_{m} lie in F\mathcal{F}, are pairwise disjoint, have union XX, and Bm⊆XmB_{m}\subseteq X_{m}, so μ(Bm)<∞\mu(B_{m})<\infty (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), The Disjointing Sequence of a Sequence of Sets (ii)). For m∈Nm\in\mathbb{N} let μm=1Bmμ\mu_{m}=\mathbf{1}_{B_{m}}\mu be the measure with density 1Bm\mathbf{1}_{B_{m}}, a measure (Image Measures and Measures with a Density: Measures, Integrals and Change of Variables (v)) with

μm(A)=∫X1Bm1A dμ=μ(A∩Bm)(A∈F)\mu_{m}(A)=\int_{X}\mathbf{1}_{B_{m}}\mathbf{1}_{A}\,d\mu=\mu(A\cap B_{m})\qquad(A\in\mathcal{F})

(Basic Properties of the Integral of Nonnegative Measurable Maps (ii)), so finite; and let νm(A)=ν(A∩Bm)\nu_{m}(A)=\nu(A\cap B_{m}), a finite measure (Restricting a Measure to a Measurable Set (iii), Restricting a Measure to a Measurable Set (iv)). If μm(A)=0\mu_{m}(A)=0 then μ(A∩Bm)=0\mu(A\cap B_{m})=0, so νm(A)=0\nu_{m}(A)=0: thus νm≪μm\nu_{m}\ll\mu_{m}. By the finite case, the set Dm\mathcal{D}_{m} of measurable h:X→Rh:X\to\mathbb{R} with h≥0h\ge0 and νm(A)=∫Ah dμm\nu_{m}(A)=\int_{A}h\,d\mu_{m} for every A∈FA\in\mathcal{F} is nonempty. Under the axiom of dependent choice, countable choice (Countable Choice Follows from Dependent Choice (i)) gives a sequence (hm)(h_{m}) with hm∈Dmh_{m}\in\mathcal{D}_{m} for every mm. For A∈FA\in\mathcal{F} and m∈Nm\in\mathbb{N}, integrating against the density (Image Measures and Measures with a Density: Measures, Integrals and Change of Variables (vi)),

ν(A∩Bm)=∫Xhm1A dμm=∫Xhm1Bm1A dμ.\nu(A\cap B_{m})=\int_{X}h_{m}\mathbf{1}_{A}\,d\mu_{m}=\int_{X}h_{m}\mathbf{1}_{B_{m}}\mathbf{1}_{A}\,d\mu .

Let f=∑mhm1Bmf=\sum_{m}h_{m}\mathbf{1}_{B_{m}}, pointwise in [0,∞][0,\infty]. Each x∈Xx\in X lies in exactly one BkB_{k}, where only the kk-th term is nonzero, so f(x)=hk(x)f(x)=h_{k}(x): ff is real-valued and nonnegative. By Linearity and Monotonicity of the Lebesgue Integral, and the Integrability Criterion (ii), applied to the sequences (hm1Bm1A)m(h_{m}\mathbf{1}_{B_{m}}\mathbf{1}_{A})_{m} with sum f1Af\mathbf{1}_{A}, f∈M+f\in\mathcal{M}^{+} and

∫Af dμ=∑m∈N∫Xhm1Bm1A dμ=∑m∈Nν(A∩Bm)=ν(A),\int_{A}f\,d\mu=\sum_{m\in\mathbb{N}}\int_{X}h_{m}\mathbf{1}_{B_{m}}\mathbf{1}_{A}\,d\mu=\sum_{m\in\mathbb{N}}\nu(A\cap B_{m})=\nu(A),

the last equality by the countable additivity of ν\nu, the sets A∩BmA\cap B_{m} being pairwise disjoint with union AA (Measure, Measure Space, and Probability Measure (i)). So ff is measurable as a function into R\mathbb{R} (Measurable Maps: Generators, Composition, Continuous Maps, Restrictions, and Criteria for Real- and [0,∞]-Valued Maps (vii)) with ∫Xf dμ=ν(X)<∞\int_{X}f\,d\mu=\nu(X)<\infty, hence f∈L1(μ)f\in\mathcal{L}^{1}(\mu) (Linearity and Monotonicity of the Lebesgue Integral, and the Integrability Criterion (vi)), and ν(A)=∫Af dμ\nu(A)=\int_{A}f\,d\mu for the integral over AA as an integrable function. This proves (iii).

Finite signed measures. Under the axiom of dependent choice, ν\nu has a Hahn set PP (Hahn and Jordan Decompositions of a Finite Signed Measure (i)), and the measures ν+\nu^{+} and ν−\nu^{-} of the Jordan decomposition are finite, absolutely continuous with respect to μ\mu, and satisfy ν=ν+−ν−\nu=\nu^{+}-\nu^{-} on F\mathcal{F} (Hahn and Jordan Decompositions of a Finite Signed Measure (ii)). By (iii) there are f1,f2∈L1(μ)f_{1},f_{2}\in\mathcal{L}^{1}(\mu) with ν+(A)=∫Af1 dμ\nu^{+}(A)=\int_{A}f_{1}\,d\mu and ν−(A)=∫Af2 dμ\nu^{-}(A)=\int_{A}f_{2}\,d\mu for every A∈FA\in\mathcal{F}. Then f=f1−f2∈L1(μ)f=f_{1}-f_{2}\in\mathcal{L}^{1}(\mu) and, by linearity (Linearity and Monotonicity of the Lebesgue Integral, and the Integrability Criterion (vii)),

∫Af dμ=∫Xf11A dμ−∫Xf21A dμ=ν+(A)−ν−(A)=ν(A)(A∈F),\int_{A}f\,d\mu=\int_{X}f_{1}\mathbf{1}_{A}\,d\mu-\int_{X}f_{2}\mathbf{1}_{A}\,d\mu=\nu^{+}(A)-\nu^{-}(A)=\nu(A)\qquad(A\in\mathcal{F}),

which proves (iv).

Uses of dependent choice. Dependent choice enters only through countable choice in (b) and the σ\sigma-finite step and through the Hahn decomposition in (d) and the signed case, so (i) and (ii) use no choice.

Citations

Loading…

Dependencies

Uses0

Loading…

Comments

Log in to comment.

Loading…