A Vitali set is a subset $V$ of $[0,1]$ such that for every $r\in \mathbb R$ there exists one and only one $v\in V$ for which $v-r \in \mathbb Q$. Equivalently, $V$ contains a single representative of every element of $\mathbb R / \mathbb Q$.

The proof I read is in this short article on Wikipedia: https://en.wikipedia.org/wiki/Vitali_set

Under "proof", the second to last inequality $1 \leq \sum \lambda (V_k) \leq 3$ is claimed to result from the previous inequality $[0,1] \subset \bigcup V_k \subset [-1,2]$ simply using sigma-additivity. There must be some missing argument to claim that the sum of the measures, although greater than the measure of the union, is still less than the measure of $[-1,2]$.

What is the missing argument ?

  • $\begingroup$ Note that the $V_k$ are pairwise disjoint. $\endgroup$ – Suzet Aug 1 '18 at 0:07

The sets $V_k$ are disjoint and countable, hence the measure of the union is exactly equal to the sum of measures.


We assume for the sake of contradiction $V$ is measurable, so $\lambda(V)$ is a real number. By construction, each $V_k$ is disjoint, and is merely a shifted copy of $V$, so each of these sets has the same measure. The sequence $(V_k)_{k=1}^\infty$ is countable.

Sigma additivity (the assumption in the definition of Lebesgue measure) guarantees that the measure of a countable union of disjoint sets will be the sum of the measures of these sets. Applying this principle along with the equal measures of all $V_k$, we find:

$$ \lambda\left(\bigcup_k V_k\right)=\sum_k\lambda\left(V_k\right)=\sum_k\lambda(V). $$

What can the infinite sum of a constant term be? If the value of the constant term $\lambda(V)$ is $0$, then the sum is zero. On the other hand, if $\lambda(V)>0$, then the sum is unbounded. It follows that $\lambda\left(\bigcup_k V_k\right)$ is either $0$ or $\infty$.

However, by construction, this union is a superset of the interval $[0,1]$, which has measure $1$, and a subset of the interval $[-1, 2]$, which has a measure $3$. The measure must therefore fall between $1$ and $3$.

We cannot have a real number be simultaneously in $\{0, \infty\}$ and $[1,3]$, so we have our contradiction.


Notation: $x+S=S+x=\{x+s:s\in S\}$ when $x\in \Bbb R$ and $S\subset \Bbb R.$

Let $q_0$ be the unique member of $V\cap \Bbb Q.$ Let $V_0=V+(-q_0).$ Let $W_0=\cup_{n\in \Bbb Z}(n+V_0).$ For $q\in (0,1)\cap \Bbb Q$ let $W_q=q+W_0.$

Observe that $(-q)+W_q=W_0$ when $q\in \Bbb Q\cap [0,1)$ and that if $q,q'$ are distinct members of $[0,1)\cap \Bbb Q$ then $W_q$ and $W_{q'}$ are disjoint.

Observe that $\{W_q\cap [0,1):q\in \Bbb Q \cap [0,1)\}$ is a partition of $[0,1).$

Observe that if $V$ is Lebesgue-measurable then so is $W_q\cap [0,1)$ for each $q\in \Bbb Q \cap [0,1).$

Let $m'(S)$ denote the Lebesgue outer measure of any $S\subset \Bbb R.$

For any $q\in \Bbb Q \cap [0,1)$ we have $$m'(W_q\cap [0,1))=m'((\;(-q)+W_q)\;)\cap [-q,1-q))=m'(W_0\cap [-q,1-q))=$$ $$=m'(W_0\cap [-q,0))+m'(W_0\cap [0,1-q))=$$ $$=m'(1+(W_0\cap [-q,0)\;)+m'(W_0\cap [0,1-q))=$$ $$=m'(W_0\cap [1-q,1))+m'(W_0\cap [0,1-q))=$$ $$=m'(W_0\cap [0,1)).$$

Suppose $V$ is Lebesgue-measurable. Then $\{W_q\cap [0,1): q\in \Bbb Q\cap [0,1)\}$ is a partition of $[0,1)$ into a countably infinite family of pair-wise disjoint measurable subsets, each with the same measure $r.$ But if $r=0$ this implies that $[0,1)$ is a measure-$0$ set, and if $r>0$ this implies the measure of $[0,1)$ is $\infty.$

The sets $W_q$ can be described as follows: Consider $\Bbb R$ as a vector space over the field $\Bbb Q.$ Let $B$ be a Hamel (vector-space ) basis for $\Bbb R$ over $\Bbb Q,$ with $1=b_1\in B.$ Let $p:\Bbb R\to \Bbb Q$ be the projection of $\Bbb R$ onto the "$b_1$" co-ordinate. Let $W_q=\{x\in \Bbb R: p(x)-q\in \Bbb Z\}.$ The Axiom of Choice is needed to prove that $B$ exists.


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.