9

Suppose we have a premeasure $\mu$ on a space $X$ such that $\mu(X) < \infty$. Prove that $E \subset X$ is Caratheodory measurable iff $ \mu^*(E)+ \mu^*(E^C) = \mu(X)$.

Going in the forward direction is fairly easy. Assuming that E is Caratheodory measurable, we can just substitute X into $\mu^*(A) = \mu^*(A \cap E) + \mu^*(A \cap E^C) $, and then we note that the outer measure and the premeasure of X themselves would have the same value.

The other direction is more difficult however. My primary idea of how to solve this part is to show that $\mu^*(A)$ and $\mu^*(A \cap E) + \mu^*(A \cap E^C) $ are both greater than and less than each other to show equality. However, I am not completely sure how to proceed with this. Can anybody provide any pointers as to how to prove the equality between these two expressions?

D.R.
  • 10,556
sjgandhi2312
  • 155
  • 2
  • 6
  • You get that $\mu^(A)\leq \mu^(A\cap E)+\mu^(A\cap E^c)$ from subadditivity. For the other inequality I'm not entirely sure. Things that are worth looking into is making some argument about a cover of $A$ that's $\epsilon/2$ greater than $\mu^(A)$, or possibly considering the case where $\mu^(X)$ is infinite and finite seperately. – Mark Schultz-Wu Nov 10 '16 at 21:02
  • We are given that $\mu(X)$ is finite. The subadditivity should have been obvious, so I suppose the other way will be more difficult. – sjgandhi2312 Nov 10 '16 at 21:08
  • Not sure if it's useful at all (I've only had basic measure theory that only introduces the outer measure to build the Lebesgue measure) but $\mu^(X)$ being finite makes things like the approximation theorem for measurable sets much more powerful if* you can use it with an outer measure (no clue if that's valid).

    If you can you should be done with just that and $\mu^*(X)$ being finite.

    – Mark Schultz-Wu Nov 10 '16 at 21:16

1 Answers1

15

Step 0: The inequality $\mu^{\ast}(A) \leqslant \mu^{\ast}(A\cap E) + \mu^{\ast}(A \cap E^C)$ for all $A \subset X$ follows directly from the subadditivity of outer measures.

Step 1: For a Carathéodory-measurable $B \subset X$ we have

$$\mu^{\ast}(E) = \mu^{\ast}(E \cap B) + \mu^{\ast}(E \cap B^C)\quad\text{and}\quad \mu^{\ast}(E^C) = \mu^{\ast}(E^C \cap B) + \mu^{\ast}(E^C \cap B^C),$$

whence

\begin{align} \mu(X) &= \mu^{\ast}(E) + \mu^{\ast}(E^C)\\ &= \mu^{\ast}(E\cap B) + \mu^{\ast}(E\cap B^C) + \mu^{\ast}(E^C\cap B) + \mu^{\ast}(E^C \cap B^C) \\ &= \bigl( \mu^{\ast}(E \cap B) + \mu^{\ast}(E^C \cap B)\bigr) + \bigl(\mu^{\ast}(E \cap B^C) + \mu^{\ast}(E^C \cap B^C)\bigr) \\ &\geqslant \mu^{\ast}(B) + \mu^{\ast}(B^C) \\ &= \mu(X), \end{align}

and so we must have

$$\mu^{\ast}(B) = \mu^{\ast}(B\cap E) + \mu^{\ast}(B \cap E^C)$$

for every Carathéodory-measurable $B$.

Step 2: Let $A\subset X$ arbitrary. For each $n \in \mathbb{N}$, there is a Carathéodory-measurable $B_n \supset A$ with $\mu^{\ast}(B_n) < \mu^{\ast}(A) + 2^{-n}$. Let

$$B = \bigcap_{n = 0}^\infty B_n.$$

Then $B \supset A$, and $B$ is Carathéodory-measurable with $\mu^{\ast}(B) = \mu^{\ast}(A)$, therefore

\begin{align} \mu^{\ast}(A) &= \mu^{\ast}(B) \\ &= \mu^{\ast}(B \cap E) + \mu^{\ast}(B \cap E^C) \tag{Step 1}\\ &\geqslant \mu^{\ast}(A\cap E) + \mu^{\ast}(A \cap E^C). \tag{$A \subset B$} \end{align}

Thus $E$ is Carathéodory-measurable.

Daniel Fischer
  • 211,575
  • 2
    Great answer, as always. Here are a few duplicates of this question with answers less valuable than yours: http://math.stackexchange.com/questions/506090/prove-that-a-set-a-is-mu-star-measurable-is-and-only-if-mu-star-a-l?rq=1 , http://math.stackexchange.com/questions/1031303/prove-that-a-subset-is-measurable-is-and-only-if-the-measurable-of-the-set-equal , http://math.stackexchange.com/questions/1301860/how-can-i-show-the-equivalent-condition-for-mu-measurability?noredirect=1&lq=1 – Gabriel Romon Nov 12 '16 at 20:57
  • Does this use finiteness of the pre-measure $\mu_0$ anywhere? – Muno Feb 22 '18 at 19:38
  • 2
    @Muno Yes. We can't conclude $a = c$ and $b = d$ from $0 \leqslant a \leqslant c$, $0 \leqslant b \leqslant d$, and $a + b = c + d$ when the sum is infinite. So if $\mu(X) = \infty$, we can't take step 1 if one of $B$ and $B^C$ has nonzero finite measure, and of course all that comes after that then also only works only in the trivial cases $\mu^{\ast}(A) = 0$ or $\mu^{\ast}(A) = \infty$. – Daniel Fischer Mar 03 '18 at 10:39