The rational numbers $\Bbb Q$ are dense in $\Bbb R$, but they are still a set of measure zero, i.e.

$$\begin{align} \mu(\Bbb Q \cap [a,b]) &= 0 \\ \mu((\Bbb R\!\setminus\! \Bbb Q) \cap [a,b]) &= b-a \\ \tag 1 \end{align}$$ for any finite interval $[a,b]$.

Is it possible to have more equally distributed sets, so that neither of them is a set of measure 0, and this holds on any interval similar to (1)?

More specifically, are there decompositions $A, B\subset \Bbb R$ and a measure $\mu$ such that all of the following conditions hold?

$$\begin{align} A\cap B = \emptyset\quad&\text{ and }\quad A\cup B = \Bbb R \\ \mu ([a,b]) &= b-a \\ \mu (A\cap [a,b]) &= (b-a) / 2 \\ \mu (B\cap [a,b]) &= (b-a) / 2 \\\tag 2 \end{align}$$ for any finite interval $[a,b]\subset\Bbb R$? The first line just states that $A,B$ is a decomposition of $\Bbb R$, the second line is a common normalizing condition for $\mu$.

Or, at your option, that $$\begin{align} \mu(A\cap[a,b]) &= (b-a)\kappa \qquad\text{ for some } 0<\kappa<1 \\ \mu(B\cap[a,b]) &= (b-a)(1-\kappa) \end{align}$$ again for any finite interval $[a,b]$. And it might even be in order if $\kappa=\kappa(a,b)$ depends on $a$ and $b$ provided $0<\kappa(a,b)<1$ for finite intervals.

My intuition says that there is no such decomposition, but maybe I am wrong.

  • $\begingroup$ You can't get them to both take up half of each interval - this is a reasonably elementary reason. You can get them both to take up some nonzero amount of each interval (depending on the interval) - eyeballfrog links a question about this below, but my personal favourite construction is this one, with Markov chains. $\endgroup$ May 16 at 21:50
  • $\begingroup$ @Izaak van Dongen: Isn't that answer with the Markov chains a constructions of a measurable set $A$ that has non-zero, non-full measure on any subinterval of $[0,1]$? Then by repeating $[0,1]$ to cover all of $\Bbb R$ would give a similar measurable set over all of $\Bbb R$? $\endgroup$ May 17 at 8:29
  • $\begingroup$ That's right (see the last paragraph of the Markov chains answer) - and if you cover all of $\Bbb R$ in this way, then $A$ and its complement satisfy the "$\kappa(a, b)$" part of your question (if I have understood correctly). $\endgroup$ May 17 at 9:13
  • $\begingroup$ @Izaak van Dongen: Ok bit doesn't that contradict David C. Ullrich's answer, because on the left side of the limit equation there would be $\kappa(a,b)$ or $1-\kappa(a,b)$, and on the right side there is 1? $\endgroup$ May 17 at 10:29
  • $\begingroup$ David C. Ullrich has only addressed the part where you're asking for exactly half-measure in each interval. In his answer, you can't directly evaluate the limit for the $\kappa(a, b)$ version, because the limit simplifies to $\lim_{h \to 0} \kappa(a - h, a + h)$ (which is only absurd if you can show this limit is less than $1$ on some non-null set). He is exploiting the fact that in the half-measure version, $\kappa$ is constant so this limit is that constant value. $\endgroup$ May 17 at 13:25

1 Answer 1



First, it's clear that $\mu([a,b])=b-a$, and hence that $\mu$ is just Lebesgue measure. Now the Lebesgue Differentiation Theorem (applied to $\chi_A$) shows that $$\frac12=\lim_{h\to0}\frac{\mu(A\cap[a-h,a+h])}{2h}=1$$for almost every $a\in A$, contradiction.

  • $\begingroup$ How does that square with this? $\endgroup$ May 16 at 13:35
  • 1
    $\begingroup$ @eyeballfrog That other post is perfectly consistent with what I say above - why do you evidently think they contradict each other? $\endgroup$ May 16 at 13:48
  • $\begingroup$ @eyeballfrog Because $\mu(A\cap[a,b])>0$ does not say $\mu(A\cap[a,b])=\frac12(b-a)$. $\endgroup$ May 16 at 13:49
  • 1
    $\begingroup$ I see now. For points not in $A$, $\mu(A\cap [a-h, a+h])$ will decrease faster than $h$ while still always being positive. And similarly for points in $A$ having that limit approach $1$. $\endgroup$ May 16 at 13:51
  • $\begingroup$ @eyeballfrog Those examples from your link don't satisfy the strong homogeneity condition $\mu[a,b]=b-a$ for all $a<b$ in $\mathbb R$. $\endgroup$
    – Lee Mosher
    May 16 at 15:08

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.