I understand the construction and proof that no measure on the Vitali set can meet the conditions that a measure should: correspond to length for intervals; be countably additive; be monotone; be translation invariant.

The Carathéodory construction starts with a pre-measure $\mu_0$ on a semi-ring on a set $X$ and extends it to an outer measure $\mu^*$ on the power set $\mathscr P (X)$ then provides a restriction that measurable sets must satisfy, that if $M$ is measurable then $\mu^*(A) = \mu^* (A \cup M) + \mu^*(A \cup M^c)$ for all $A \in \mathscr P (X)$.

So, my question is if one starts with the Lebesgue pre-measure defined as the length of half open intervals, and $V$ is the vitali set, is there a demonstrable set $ A \subset \mathbb R$ such that $\mu^*(A) \ne \mu^* (A \cup V) + \mu^*(A \cup V^c)$ ?



Note that for every $B\subset \mathbb{R}^n$ its outer measure also equals $$\mu^{\star}(B) = \inf \{ \mu(M) \ | \ M \textrm{ measurable}, M \supset B\} $$

Let's consider for every subset $A$ of $\mathbb{R}^n$ its inner measure defined as

$$\mu_{\star}(B) = \sup \{ \mu(M) \ | \ M \textrm{measurable} , M \subset B\}$$

It's fairly easy to show that for every $A$ measurable and any $B\subset A$ we have $$\mu(A) = \mu^{\star} (A\backslash B) + \mu_{\star} (B)$$

Consider now a Vitali set $V$, i.e. such a set so that $$\mathbb{R}^n = \sqcup_{q \in \mathbb{Q}^n} (V+q)$$ Then we have

  1. $\mu_{\star}(V)= 0$.

  2. $\mu^{\star}(V) > 0$.

For the first statement: Take a measurable $M \subset V$, and let's show that $\mu(M) = 0$. We may assume $M$ bounded. We have $$\sum_{q \in \mathbb{Q}^n \cap [0,1]^n} \mu( M+q) = \mu( \sqcup_{q \in \mathbb{Q}^n\cap [0,1]^n}\mu( M+q)) < \infty$$ so all terms ( being equal) are $0$, that is $\mu(M)=0$.

For the second: $$\sum_{q \in \mathbb{Q}^n}\mu^{\star}(V+q) \ge \mu^{\star}( \cup_{q \in \mathbb{Q}^n} M+q) = \mu^{\star}(\mathbb{R}^n) = \infty$$ so $\mu^{\star}(V) > 0$.

Now it's easy to find $A$ as required. Let's assume that $V$ is contained in a subset $A$ of finite measure. Since

$$\mu(A) = \mu^{\star}(A\backslash V) + \mu_{\star}(V)$$ we get $\mu^{\star}(A\backslash V) = \mu(A)$. Since $\mu^{\star}(V) >0$ we get the strict inequality.

Observation: if $B$ is a subset of a measurable set $A$ of finite measure and $$\mu(A) = \mu^{\star}(A\backslash B) + \mu^{\star}(B)$$ then $B$ is measurable. Indeed, we get $\mu_{\star}(B) = \mu^{\star}(B)$ and so $B$ is measurable (easy to prove, use completeness of the measure).

  • $\begingroup$ Thanks for the detailed response. From my starting point, it will take me a while to work through it. $\endgroup$ – Tom Collinge Aug 24 '17 at 7:55

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.