Let $\lambda( \cdot )$ denote Lebesgue measure on $[0,1]$. Let $(A_n)_{n=1}^\infty$ be a decreasing sequence of Borel subsets of $[0,1]$ such that $\bigcap_{n=1}^\infty A_n = \emptyset$. Given $\epsilon > 0$ does there exist an open set $U \subseteq [0,1]$ such that

(i) $\lambda(A_n \setminus U) = 0$ for some $n$,

(ii) $\lambda(U) \leq \epsilon$, and

(iii) $\lambda(\partial U) = 0$,

where $\partial U$ is the boundary of $U$?

I don't have a good intuition whether this should be true or not. One idea might be to cover some $A_n$ by an open set with measure much smaller than $\epsilon$, then hope that we can expand that open set somewhat to make its boundary null.

  • $\begingroup$ How did this arise? It looks and quacks a lot like homework. $\endgroup$ Mar 31, 2015 at 16:48
  • 3
    $\begingroup$ If this were true, it would have been an easy way to solve a much harder problem. Ergodic averages of circle rotations converge uniformly for open sets with null boundary. I want to know that every set of small measure is contained in a set of small measure for which ergodic averages converge uniformly. So no, this is not homework. $\endgroup$ Mar 31, 2015 at 16:55

1 Answer 1


No, such a $U$ does not exist in general. You can have (i)+(ii), but (iii) is too much to ask. The problem is that the $A_n$ might all be dense in $[0,1]$ so that $\partial U = \overline{U} \setminus U =[0,1]\setminus U$ has measure $\geq 1-\epsilon$ by (ii).

  • $\begingroup$ This isn't quite a counterexample, because I only want $\lambda(A_n \setminus U) = 0$, not $A_n \subseteq U$. Perhaps there's some way to tweak it so that it is a counterexample? $\endgroup$ Mar 31, 2015 at 16:37
  • 2
    $\begingroup$ Enumerate the subintervals of $[0,1]$ with rational endpoints as $I_n$. Let $K_n$ for $n\ge 2$ be a sequence of disjoint "fat Cantor sets" (nowhere dense compact sets of positive measure) with $K_n \subset I_n$. Let $A_n = \bigcup_{m \ge n} K_m$. $\endgroup$ Mar 31, 2015 at 16:54
  • $\begingroup$ @burtonpeterj: Yes, I was sloppy there. Nevertheless the argument works, because the $A_n$ could have the property that not only $A_n$ itself is dense but also every $A_n' \subseteq A_n$ with $\lambda(A_n\setminus A_n')=0$ is dense as well (including the set $A_n \cap U$) $\endgroup$ Mar 31, 2015 at 18:13

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.