TheoremBase

Proof

Throughout, (X,F,μ)(X,\mathcal{F},\mu) and the [0,∞][0,\infty] conventions are as in the statement. We use repeatedly that F\mathcal{F} is closed under differences: for A,B∈FA,B\in\mathcal{F} we have B∖A=X∖((X∖B)∪A)B\setminus A=X\setminus\bigl((X\setminus B)\cup A\bigr), which lies in F\mathcal{F} by the closure properties recorded in Sigma-Algebra and Measurable Space.

Claim 1. Define a sequence (Dm)m∈N(D_m)_{m\in\mathbb{N}} in F\mathcal{F} by Dm=AmD_m=A_m for m≤rm\le r and Dm=∅D_m=\varnothing for m>rm>r. Its members are pairwise disjoint and ⋃m∈NDm=⋃i=1rAi\bigcup_{m\in\mathbb{N}}D_m=\bigcup_{i=1}^{r}A_i, so countable additivity in Measure, Measure Space, and Probability Measure gives

μ(⋃i=1rAi)=∑m∈Nμ(Dm).\mu\Bigl(\bigcup_{i=1}^{r}A_i\Bigr)=\sum_{m\in\mathbb{N}}\mu(D_m).

Since μ(∅)=0\mu(\varnothing)=0, the kk-th partial sum of (μ(Dm))m∈N(\mu(D_m))_{m\in\mathbb{N}} equals ∑i=1min⁡{k,r}μ(Ai)\sum_{i=1}^{\min\{k,r\}}\mu(A_i), and in particular equals ∑i=1rμ(Ai)\sum_{i=1}^{r}\mu(A_i) for every k≥rk\ge r.

If every μ(Ai)\mu(A_i) with i≤ri\le r is real, then every μ(Dm)\mu(D_m) is real and the partial sums form a nondecreasing sequence that is constant with value ∑i=1rμ(Ai)\sum_{i=1}^{r}\mu(A_i) from index rr on; that value is therefore an upper bound for the partial sums and is itself a partial sum, hence it is their least upper bound. By the definition of the sum of a sequence in [0,∞][0,\infty], the displayed sum equals ∑i=1rμ(Ai)\sum_{i=1}^{r}\mu(A_i).

If μ(Ai)=∞\mu(A_i)=\infty for some i≤ri\le r, then not every term of (μ(Dm))(\mu(D_m)) is real, so by that same definition ∑mμ(Dm)=∞\sum_{m}\mu(D_m)=\infty; and the finite sum ∑i=1rμ(Ai)\sum_{i=1}^{r}\mu(A_i), formed in [0,∞][0,\infty] with the convention a+∞=∞+a=∞a+\infty=\infty+a=\infty, is also ∞\infty. In either case the two sides agree, and the finite sum is ∞\infty precisely when some term is ∞\infty, since a finite sum of nonnegative real numbers is real.

Claim 2. The sets AA and B∖AB\setminus A are disjoint members of F\mathcal{F} whose union is BB, so claim 1 with r=2r=2 gives

μ(B)=μ(A)+μ(B∖A).\mu(B)=\mu(A)+\mu(B\setminus A).

If μ(A)=∞\mu(A)=\infty or μ(B∖A)=∞\mu(B\setminus A)=\infty, then μ(B)=∞\mu(B)=\infty and μ(A)≤μ(B)\mu(A)\le\mu(B) holds, because every element of [0,∞][0,\infty] is at most ∞\infty by the convention a<∞a<\infty for real a≥0a\ge0. Otherwise both are nonnegative real numbers and μ(A)≤μ(A)+μ(B∖A)=μ(B)\mu(A)\le\mu(A)+\mu(B\setminus A)=\mu(B) by the order of the ordered field of real numbers.

Claim 3. By claim 2, μ(A)≤μ(B)<∞\mu(A)\le\mu(B)<\infty and μ(B∖A)≤μ(B)<∞\mu(B\setminus A)\le\mu(B)<\infty, so both μ(A)\mu(A) and μ(B∖A)\mu(B\setminus A) are real. The identity μ(B)=μ(A)+μ(B∖A)\mu(B)=\mu(A)+\mu(B\setminus A) of claim 2 is then an identity between real numbers, and subtracting μ(A)\mu(A) gives μ(B∖A)=μ(B)−μ(A)\mu(B\setminus A)=\mu(B)-\mu(A). If μ\mu is finite, apply this with B=XB=X, which is permitted since μ(X)<∞\mu(X)<\infty and A⊆XA\subseteq X.

Claim 4. Set B1=A1B_1=A_1 and, for m≥2m\ge2,

Bm=Am∖⋃i=1m−1Ai.B_m=A_m\setminus\bigcup_{i=1}^{m-1}A_i .

Each BmB_m lies in F\mathcal{F}, and Bm⊆AmB_m\subseteq A_m. They are pairwise disjoint: if m<m′m<m' then Bm⊆Am⊆⋃i=1m′−1AiB_m\subseteq A_m\subseteq\bigcup_{i=1}^{m'-1}A_i, while Bm′B_{m'} is disjoint from that union. An induction on kk shows ⋃m=1kBm=⋃m=1kAm\bigcup_{m=1}^{k}B_m=\bigcup_{m=1}^{k}A_m: this is clear for k=1k=1, and if it holds for kk then

⋃m=1k+1Bm=(⋃m=1kAm)∪(Ak+1∖⋃m=1kAm)=⋃m=1k+1Am.\bigcup_{m=1}^{k+1}B_m=\Bigl(\bigcup_{m=1}^{k}A_m\Bigr)\cup\Bigl(A_{k+1}\setminus\bigcup_{m=1}^{k}A_m\Bigr)=\bigcup_{m=1}^{k+1}A_m .

Consequently ⋃m∈NBm=⋃m∈NAm\bigcup_{m\in\mathbb{N}}B_m=\bigcup_{m\in\mathbb{N}}A_m, and countable additivity gives

μ(⋃m∈NAm)=∑m∈Nμ(Bm),\mu\Bigl(\bigcup_{m\in\mathbb{N}}A_m\Bigr)=\sum_{m\in\mathbb{N}}\mu(B_m),

while μ(Bm)≤μ(Am)\mu(B_m)\le\mu(A_m) for every mm by claim 2.

If some μ(Am)=∞\mu(A_m)=\infty, or if every μ(Am)\mu(A_m) is real but the partial sums of (μ(Am))(\mu(A_m)) are not bounded above, then ∑mμ(Am)=∞\sum_{m}\mu(A_m)=\infty and the asserted inequality holds. Otherwise every μ(Am)\mu(A_m) is real with partial sums bounded above; then every μ(Bm)\mu(B_m) is real, and comparing the sums term by term, the kk-th partial sum of (μ(Bm))(\mu(B_m)) is at most the kk-th partial sum of (μ(Am))(\mu(A_m)), hence at most ∑mμ(Am)\sum_{m}\mu(A_m). Thus ∑mμ(Am)\sum_{m}\mu(A_m) is an upper bound for the partial sums of (μ(Bm))(\mu(B_m)), and their least upper bound ∑mμ(Bm)\sum_{m}\mu(B_m) is at most ∑mμ(Am)\sum_{m}\mu(A_m).

Claim 5. Set C1=A1C_1=A_1 and Cm=Am∖Am−1C_m=A_m\setminus A_{m-1} for m≥2m\ge2; these lie in F\mathcal{F}. They are pairwise disjoint: if m<m′m<m' then Cm⊆Am⊆Am′−1C_m\subseteq A_m\subseteq A_{m'-1} by the nesting hypothesis, while Cm′C_{m'} is disjoint from Am′−1A_{m'-1}. An induction on kk shows ⋃m=1kCm=Ak\bigcup_{m=1}^{k}C_m=A_k: this is clear for k=1k=1, and if it holds for kk then ⋃m=1k+1Cm=Ak∪(Ak+1∖Ak)=Ak+1\bigcup_{m=1}^{k+1}C_m=A_k\cup(A_{k+1}\setminus A_k)=A_{k+1}, using Ak⊆Ak+1A_k\subseteq A_{k+1}. Hence ⋃m∈NCm=A\bigcup_{m\in\mathbb{N}}C_m=A, and countable additivity gives μ(A)=∑m∈Nμ(Cm)\mu(A)=\sum_{m\in\mathbb{N}}\mu(C_m), while by claim 1 the kk-th partial sum of (μ(Cm))(\mu(C_m)) is

∑m=1kμ(Cm)=μ(⋃m=1kCm)=μ(Ak).\sum_{m=1}^{k}\mu(C_m)=\mu\Bigl(\bigcup_{m=1}^{k}C_m\Bigr)=\mu(A_k).

Suppose first that every μ(Am)\mu(A_m) is real and that S={μ(Am):m∈N}S=\{\mu(A_m):m\in\mathbb{N}\} is bounded above. Each μ(Cm)\mu(C_m) is then real, since μ(Cm)≤μ(Am)<∞\mu(C_m)\le\mu(A_m)<\infty by claim 2, and the partial sums of (μ(Cm))(\mu(C_m)) are exactly the members of SS, which are bounded above. By the definition of the sum of a sequence in [0,∞][0,\infty], μ(A)\mu(A) is the least upper bound σ\sigma of SS. The sequence (μ(Ak))k∈N(\mu(A_k))_{k\in\mathbb{N}} is nondecreasing by claim 2 and the nesting hypothesis. Given a real ε>0\varepsilon>0, the number σ−ε\sigma-\varepsilon is not an upper bound of SS, so there is K∈NK\in\mathbb{N} with σ−ε<μ(AK)\sigma-\varepsilon<\mu(A_K); for every k≥Kk\ge K we then have σ−ε<μ(AK)≤μ(Ak)≤σ\sigma-\varepsilon<\mu(A_K)\le\mu(A_k)\le\sigma, whence ∣μ(Ak)−σ∣<ε|\mu(A_k)-\sigma|<\varepsilon. Thus (μ(Ak))(\mu(A_k)) converges to σ=μ(A)\sigma=\mu(A) in the sense of Limit of a Sequence of Real Numbers.

Suppose instead that some μ(Am0)=∞\mu(A_{m_0})=\infty. Then Am0⊆AA_{m_0}\subseteq A and claim 2 give μ(A)=∞\mu(A)=\infty. Finally, suppose every μ(Am)\mu(A_m) is real but SS is not bounded above. The partial sums of (μ(Cm))(\mu(C_m)) are the members of SS and are therefore not bounded above, so by the definition of the sum of a sequence in [0,∞][0,\infty] we again get μ(A)=∑mμ(Cm)=∞\mu(A)=\sum_{m}\mu(C_m)=\infty.

Citations

Loading…

Dependencies

Uses0

Loading…

Comments

Log in to comment.

Loading…