1
$\begingroup$

I saw an answer here on the page stating the fact that for a dense sequence $(a_n)$ in $[0, 1]$ and a set of positive measure $A$, the set $\bigcup_{n \geq 1} A + a_n$ has measure $1$ (with respect to the Lebesgue-measure). So I tried to prove it.

Let $B := \left( \bigcup_{n \geq 1} A + a_n \right) \cap \mathbf{R}/ \mathbf{Z}$.

My attempt was to pick an open set $U \supset B$ with $\lambda(U) \leq \lambda (B) + \varepsilon$ for an arbitrary $\varepsilon >0$. This can be done since the Lebesgue measure is outer regular.

But then $U$ contains a dense sequence, hence any non-empty open set $V$ intersects it (the set $U$ being non-empty, as $A$ has positive measure). But then again, as $[0, 1]$ is Hausdorff, we can only have $U = [0, 1]$. As $\varepsilon$ was arbitrary, the set $B$ has measure $1$.

However I'm a bit confused about my "proof". If $B$ were $\mathbf{Q} \cap [0, 1]$, the same argument would show that the measure of $\mathbf{Q} \cap [0, 1]$ is one.

Where is my mistake?

And also: How can one prove the statement?

Thanks!

$\endgroup$
  • $\begingroup$ If anyone is interested in the question where I found the fact, here's the link: math.stackexchange.com/questions/1033251/… $\endgroup$ – Steven Jun 2 '17 at 16:36
  • $\begingroup$ Do you mean $(\cup (A+a_n)) \cap [0,1]?$ $\endgroup$ – zhw. Jun 2 '17 at 16:59
  • $\begingroup$ yes, of course, I'll edit it $\endgroup$ – Steven Jun 2 '17 at 17:03
  • $\begingroup$ Perhaps you mean that there exists a translation of $\bigcup_n A+a_n$ which has full measure in $[0,1]$. $\endgroup$ – Rigel Jun 2 '17 at 17:10
  • $\begingroup$ I mean the elements seen to be in $[0, 1) $ modulo $1$. so yes, you're right. $\endgroup$ – Steven Jun 2 '17 at 17:17
1
$\begingroup$

Let us prove that there exists a translation of the set $\bigcup_n (A + a_n)$ which has full measure in $[0,1]$.

Assume by contradiction that $B :=\bigcup_n (A + a_n)$ has not full measure in $[0,1]$, so that there exists a point $x\in B$ with density $0$ in $B$. Let us fix $\epsilon \in (0,1/3)$. Hence there exists $\rho_0 > 0$ such that $$ \frac{\lambda((x-\rho, x+\rho)\cap B)}{2\rho} < \epsilon \qquad \forall \rho \in (0, \rho_0). $$ To simplify the argument, we also assume that $(x-\rho_0, x+\rho_0) \subset [0,1]$.

W.l.o.g. we can assume that $0$ is a Lebesgue point of $A$, i.e., there exists $r \in (0, \rho_0)$ such that $$ \frac{\lambda(A_r)}{2r} > 1 - \epsilon, \quad\text{where}\quad A_r := ( - r, r) \cap A. $$ It is not restrictive to assume $r < 2\epsilon\rho_0$.

On the other hand, $B \supset C_r := \bigcup_n (A_r + a_n)$.

Since $(a_n)$ is dense in $[0,1]$, given $\rho :=r/{2\epsilon} \in (r, \rho_0)$ there exists an index $j\in\mathbb{N}$ such that $A_r + a_j \subset (x-\rho, x+\rho)$, hence $$ A_r + a_j \subset (A_r + a_j) \cap B \subset (x-\rho, x+\rho)\cap B $$ so that $$ 2 r (1-\epsilon) < 2\rho \epsilon = r, $$ a contradiction.

$\endgroup$
  • $\begingroup$ I don't quite see why the last inequality holds. Could you maybe elaborate on that? $\endgroup$ – Steven Jun 2 '17 at 17:30
  • $\begingroup$ I have modified the proof. (Not sure of the choice of the constants, but the idea should be correct.) $\endgroup$ – Rigel Jun 2 '17 at 18:04

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.