I am reading Real Analysis by Yeh and have a question about the following result (Thm 3.25 in the book)
Theorem If $E \in \mathfrak{M}_L$ and $\mu_L(E)< \infty$, then $\forall \epsilon>0$ $\exists$ finitely many open intervals $I_1, \ldots, I_N$ s.t. $\mu_L(E \triangle \cup_{n=1}^NI_n)< \epsilon$.

Notations: $\mathfrak{M}_L$ is the $\sigma$-algebra of Lebesgue measurable subsets of $\mathbb{R}$, $\mu_L$ denotes Lebesgue measure. If we're not sure $E \in \mathfrak{M}_L$, then $\mu_L^*(E)$ denotes the Lebesgue outer measure of E. For intervals $I$, Yeh sometimes uses $\ell(I)$ to denote their length. The symmetric difference of 2 sets is $A \triangle B:=(A \setminus B) \cup (B \setminus A)$.

I didn't know how to prove this, so I read Yeh's proof. After a while I tried to prove it again but came up with something simpler that seems to work, so wanted to ask if I might have missed something. My attempt is below, followed by Yeh's proof (it's long, so I of course want something simpler in my notes if possible). Thanks in advance for any help.

My attempt: Fix $\epsilon>0$. Use the definition of outer measure as an infimum to pick a sequence $(I_n)$ of open intervals s.t. $E \subseteq \cup_{n=1}^ \infty I_n$ and $\sum_{n=1}^\infty \mu_L(I_n) \leq \mu_L(E) + \epsilon$. Since $\mu_L(E)< \infty$, $\sum_{n=1}^\infty \mu_L(I_n)$ converges, so pick $N \in \mathbb{N}$ s.t. $\sum_{n=N+1}^\infty \mu_L(I_n)< \epsilon$. Then \begin{split} \mu_L(E \triangle \cup_{n=1}^NI_n) &= \mu_L \big( (E \setminus \cup_{n=1}^N I_n) \cup ( \cup_{n=1}^N I_n \setminus E) \big) \\ & \leq \mu_L(E \setminus \cup_{n=1}^N I_n)+\mu_L( \cup_{n=1}^N I_n \setminus E) \\ & \leq \mu_L(\cup_{n=1}^\infty I_n \setminus \cup_{n=1}^N I_n)+\mu_L( \cup_{n=1}^\infty I_n \setminus E) \\ & \leq \mu_L(\cup_{n=N+1}^\infty I_n)+\mu_L( \cup_{n=1}^\infty I_n)- \mu_L(E) \\ & \leq \sum_{n=N+1}^\infty \mu_L(I_n)+ \sum_{n=1}^\infty \mu_L(I_n)- \mu_L(E) <2 \epsilon \end{split} where we have repeatedly used that a measure is monotone and subadditive. In the 4th line, we have also used an earlier fact that $A,B \in \mathfrak{M}_L$, $A \subseteq B$, $\mu_L(A)< \infty$ implies $\mu_L(B \setminus A)= \mu_L(B)- \mu_L(A)$. Since $\epsilon$ is arbitrary, this is the desired result.

enter image description here
(I'm cutting the proof here, but Yeh pretty much bounds each of the 3 pieces on the RHS separately after this).

  • $\begingroup$ Do you know that a Lebesgue measurable subset of finite measure is approximable, in measure, by compact subsets? $\endgroup$
    – zhw.
    Mar 14 '17 at 22:51
  • $\begingroup$ Sorry, I have not heard of it $\endgroup$
    – Jason
    Mar 14 '17 at 23:34
  • $\begingroup$ It seems to be the modern teaching style to avoid mentioning inner measure. If $m$ is a measure for which closed sets are measurable, the inner measure of $S$ is $m^i(S)=\sup \{m(C): C=\overline C\subset S\}.$ $\endgroup$ Apr 13 '18 at 18:26
  • 1
    $\begingroup$ When $m$ is Lebesgue measure and $m^o$ is outer Lebesgue measure and $m^i$ is inner Lebesgue measure we have $S\in dom (m)$ iff $m^i(A\cap S)=m^o(A\cap S)$ for every bounded interval $A.$ It follows that $S\in dom (m)$ iff there exists an $F_{\sigma}$-set $C$ and a $G_{\delta}$-set $D$ such that $C\subset S\subset D$ and such that $m(C)=m^i(S)=m^o(S)=m(D).$ $\endgroup$ Apr 13 '18 at 19:41

Let $m$ be Lebesgue measure on $\Bbb R.$ Let $E\in dom (m)$ with $m(E)<\infty.$ We have $m(E)=\sum_{n\in \Bbb Z}m(E\cap (n,n+1]).$ Let $\epsilon>0.$

There exists $a\in \Bbb Z^+$ such $m(E\cap (-a,a))=\sum_{n=-a}^{a-1}m(E\cap (n,n+1])>m(E)-\epsilon /3.$

If U is an open subset of $\Bbb R$ then $U=\cup F$ for some countable family $F$ of pair-wise disjoint convex open sets. If $U$ is a bounded open set then each $f\in F$ is a bounded open interval.

Let $U$ be open with $U\supset E$ and $m(U\setminus E)<\epsilon /3.$ Let $V=U\cap (-a,a+\epsilon /3).$ Let $F$ be a countable family of pair-wise disjoint bounded open intervals such that $\cup F=V.$

Now $\sum_{f\in F}m(f)=m(\cup F)=m(V)<\infty,$ so there exists a finite $G\subset F$ such that $m(\cup G)=\sum_{g\in G}m(g)>-\epsilon /3+\sum_{f\in F}m(f)=m(V)-\epsilon /3.$

You may now confirm that $m(E\Delta (\cup G))<\epsilon.$

  • $\begingroup$ $E\Delta (\cup G)\subset P\cup Q\cup R$ where $Q=E\setminus (-a,a)$ and $Q=(\cup (F)\setminus (\cup G)$ and $R=E\setminus (\cup F)$. $\endgroup$ Apr 13 '18 at 19:22

Your proof is correct.

Yeh's proof is longer (more complicated?) because he wants to use (ii) of Theorem 3.22 ($E$ measurable iff $\forall\ \epsilon > 0\ \exists$ an open set $O \supseteq E$ such that $\mu_L(O \setminus E) < \epsilon$). Directly using the definition of measurability on $E$ as you have done also works.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.