Showing countable additivitiy of Lebesgue measure

530 Views Asked by At

The following is taken from the classic Probability and Measure by Patrick Billingsley, Theorem 2.2 (page 26 in the 3rd edition). I have a question on his proof, but I give the necessary defintions to make my question self-contained.

Denote by $\mathcal B_0$ the field of all finite unions of half-open sub-intervals $(a,b]$ of $(0,1]$, i.e. it is a system of sets containing all half-open sub-intervals of $(0,1]$ and closed under finite intersection and complementation (and therefore also finite unions). It is easy to see that $\mathcal B_0$ is exactly the set $$ \mathcal B_0 = \left\{ \bigcup_{i=1}^n (a_i, b_i] : (a_i, b_i] \subseteq (0,1] ~ \mbox{disjoint} \right\} $$ of all subsets of $(0,1]$ that could be written as finite unions of disjoint subintervals of $(0,1]$. The following property is needed in the following: If $I = \bigcup_k I_k$ and the $I_k$ are disjoint, then $|I| = \sum_k |I_k|$. (*)

Then define a mapping $\lambda : \mathcal B_0 \to [0,1]$ by $$ \lambda(A) = \sum_{i=1}^n (b_i - a_i). $$ Then $\lambda : \mathcal B_0 \to [0,1]$ is countable additive, i.e. we have for $A \in \mathcal B_0$ and $A_1, A_2, \ldots \in \mathcal B_0$ disjoint and $A = \bigcup_{k=1}^{\infty} A_k$ $$ \lambda(A) = \lambda(\bigcup_{k=1}^{\infty} A_k) = \sum_{k=1}^{\infty}\lambda(A_k). $$

Proof: Suppose that $A = \bigcup_{k=1}^{\infty} A_k$, where $A$ and the $A_k$ are $\mathcal B_0$-sets and the $A_k$ are disjoint. Then $A = \bigcup_{i=1}^n I_i$ and $A_k = \bigcup_{j=1}^{m_k} J_{kj}$ are disjoint unions of subintervals of $(0,1]$, and the above, and (*) give

\begin{align*} \lambda(A) & = \sum_{i=1}^{n} |I_i| = \sum_{i=1}^n \sum_{k=1}^{\infty} \sum_{j=1}^{m_k} |I_i \cap J_{kj}| \\ & = \sum_{k=1}^{\infty} \sum_{j=1}^{m_k} |J_{kj}| = \sum_{k=1}^{\infty} \lambda(A_k). \qquad \square \end{align*}

To quote the book:

[...] proving countable additivity on $\mathcal J$ [i.e. the set of all subintervals of $(0,1]$] requires the deeper property of compactnness.

Where exactly is compactness used in the above proof? I do not see where it is applied.

1

There are 1 best solutions below

1
On BEST ANSWER

Billingsley uses compactness to prove $*$. If you cover $I=(a,b]$ with countably many intervals $I_k=(a_k,b_k]$ (disjoint or not), $(a,b]\subset\bigcup_{k=1}^{\infty} (a_k,b_k]$, how can you be sure that $b-a=|I|\leq \sum_k|I_k|$? (Elementary to show for the finite case.) At the moment, we only know how to handle the length $|\!\cdot\!|$ of a finite union of intervals.

Here is a sketch of the argument from Billingsley (proof of $*$, Th. 1.3). The idea is that for $0<\varepsilon<b-a$, the open intervals $(a_k,b_k+\varepsilon2^{-k})$ cover the interval $[a+\varepsilon,b]$. But the latter interval is closed and bounded, so it is compact. By compactness, there exists a finite $n$ such that $[a+\varepsilon,b]\subset \bigcup_{k=1}^n (a_k,b_k+\varepsilon2^{-k})$. Now $(a+\varepsilon,b]\subset \bigcup_{k=1}^n (a_k,b_k+\varepsilon2^{-k}]$. By the finite case, $b-(a+\varepsilon)\leq \sum_{k=1}^n b_k+\varepsilon2^{-k}-a_k\leq \varepsilon+\sum_{k=1}^{\infty} b_k-a_k$, and the result follows because $\varepsilon$ was arbitrary.