I'm really stuck on the following problem.

Let $E\subset[0,1]\times[0,1]$ be a Lebesgue measurable subset of positive measure. Show that there exists $x,y\in E$, with $x\neq y$, such that $x-y\in\mathbb{Q}\times\mathbb{Q}$.

I was thinking of doing a proof by contradiction, but I'm not sure what I would try to establish as my contradiction (perhaps using density/countability of $\mathbb{Q}$?). Any help is greatly appreciated!


You need to use the Steinhaus theorem from which you get that $E-E$ contains an open set.

  • $\begingroup$ We've never gone over the Steinhaus Theorem in any of my lectures. Is there another way to tackle this? $\endgroup$ – Sir_Math_Cat Oct 4 '18 at 17:48
  • $\begingroup$ @Sir_Math_Cat yes, I post another answer. $\endgroup$ – Yanko Oct 4 '18 at 20:09

As a request from the OP I post another answer to this question:

Suppose by contradiction that for all $x,y\in E$ we have $x-y\in \mathbb{Q} \Rightarrow x=y$.

Lemma 1: Let $r=(p,q)\in\mathbb{Q}^2$ be non-zero then $E,E+r$ are disjoint.

Indeed if by contradiction $e,e'\in E$ are such that $e=e'+r$ then $e-e' = (p,q)$ and so $e=e'$ which is absurd. $\blacksquare$

Lemma 2: Let $r$ as before, and let $E_r:=(E+r)\cap [0,1]\times [0,1]$ then $\mu(E_r)\geq \mu(E)-pq$

The square $[1-p,1]\times [1-q,1]$ is pushed out after translating by $r$. The size of the square is $pq$ and so the measure of $E$ inside this square must be less than that. $\blacksquare$

Let $c>0$ denote the measure of $E$. Choose infinitely many rational pairs $r_1,r_2,...$ (in fact we only need finitely many depending on $c$) such that $E_{r_i}$ is of measure at least $c/2>0$. This is possible by Lemma 2, also by Lemma 1 all $E_{r_i}$ are disjoint. Hence

$\mu(\bigsqcup_{i=1}^{\infty} E_i) = \sum_{i=1}^\infty \mu(E_i)(= c/2\cdot \infty)=\infty$

Which is absurd as all $E_i$ lies in $[0,1]^2$.


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.