Question: Show that if $0 \leq \text{m}(A) \leq \infty$, then for each positive $q < \text{m}(A)$ there is a perfect set $B \subset A$ of measure $q$. (All subsets are in $\mathbb{R}$ and the measure is the Lebesgue measure).

I can find a $B$ without the restriction of being perfect. Now for any given $\epsilon > 0 $ thre is a closed $F$ such that $\text{m}(B) - \text{m}(F) < \epsilon$. and so by Cantor Bendixson we get a perfect set with measure $q' = \text{m}(F)$ where $q - q' < \epsilon$ and this can be done for any $\epsilon$ but I am unable to find a perfect set with exact $q$ as the measure.

How to proceed?

This is problem $2.1.31$ from Kaczor and Nowak, Problems in Analysis III, Integration. I am unable to follow the solution given in the book.

  • 1
    $\begingroup$ Appropriately modify the construction process for any specific construction of a perfect nowhere dense set having positive measure. $\endgroup$ Oct 2 at 14:20
  • 1
    $\begingroup$ I am unable to follow up from your hint. I can make a nowhere dense perfect for any given measure $\alpha$ by translating the Cantor type sets. But how to ensure that the construction gives a subset inside $A$? I only know how to do it inside intervals. $\endgroup$ Oct 2 at 14:56
  • 1
    $\begingroup$ Sorry, I missed the part about doing it inside an arbitrary set of positive measure. What I said was intended for when $A$ is an interval of positive length. In case you're interested, see this MSE answer for similar results -- more general in some ways, less in other ways. $\endgroup$ Oct 2 at 14:59
  • $\begingroup$ @DaveL.Renfro Thank you for the link. The conjecture is interesting. for any $a < b$ we can make a set $A$ with $m_*(A)= a$ , $m^{*}(A) = b$(say using Vitali Sets), but with the added restriction of finding it in a given subset seems open. Of course in this $A$ cannot be perfect. $\endgroup$ Oct 4 at 13:04

1 Answer 1


You can apply this countable procedure:

Since $m(A)>q$, for some $n_1$ sufficiently large $m(A)>q+\frac{1}{n_1}$. Now, choose $B\subseteq A$ arbitrary with $\mu(B)=q+\frac{1}{n_1}$ and set $\varepsilon_1:=\frac{1}{2n_1}$. We can find a closed set $F_{n_1}\subseteq B$ with measure at least $q+\frac{1}{2n_1}$ (and at most $q+\frac{1}{n_1}$), taking a countable set out we can assume that $F_{n_1}$ is perfect. Now take $F_{n_1}$ in place of $A$, the same procedure gives $F_{n_2}\subseteq F_{n_1}$ with measure between $q+\frac{1}{2n_2}$ to $q+\frac{1}{n_2}$, where $n_2>n_1$ (and we have complete freedom choosing $n_2$).

Continue this way (using induction) choosing $n_k\rightarrow\infty$, we get a sequence of perfect sets $...\subseteq F_{n_k}\subseteq F_{n_{k-1}}\subseteq...\subseteq F_{n_1}$.

The intersection of a decreasing sequence of perfect sets is a perfect set $F=\bigcap_{i=1}^\infty F_{n_i}$ and by construction and regularity of the Lebesgue measure, we have $m(F)=q.$

  • $\begingroup$ Thank you, I also wanted to approximate from above to preserve the perfectness but this idea didn't strike me. $\endgroup$ Oct 3 at 1:47

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.