25
$\begingroup$

Given a subset X of unit interval, can we find a subset Y of X of same outer measure as X such that Y does not contain three points of the form x, y and (x+y)/2?

I can do this assuming CH but can we do this without this assumption? Thanks!!

$\endgroup$
7
  • 6
    $\begingroup$ This is a beautiful question. $\endgroup$ Apr 9, 2015 at 15:32
  • 4
    $\begingroup$ Tangential remark: Note that you can always find a non null $Y \subseteq X$ avoiding mid points. This follows from: (R. Rado) One can partition $\mathbb{R}^n$ into countably many sets such that every piece is mid point free. See P. Komjath: Set theoretic constructions in Euclidean spaces, New Trends in Discrete and Computational Geometry (J. Pach, ed.), Springer, 1993, 303-325 $\endgroup$
    – Ashutosh
    Apr 9, 2015 at 16:00
  • $\begingroup$ First thoughts. For $\varepsilon>0$ call $X$ $\varepsilon$-free if there are no $x,y,z\in X$ with $x+\varepsilon=y$ and $y+\varepsilon=z$. Given $\varepsilon$, is there $Z$ of inner measure $0$ such that $X\setminus Z$ is $\varepsilon$-free? (Note that it could conceivably happen that $m^*(X)=m^*(Z)=m^*(X\setminus Z$), see this answer.) Could we also get that $X\setminus Z$ is $\varepsilon'$-free for every $\varepsilon'\ge\varepsilon$ (this would solve the problem)? $\endgroup$
    – Mirko
    Apr 10, 2015 at 0:25
  • $\begingroup$ Yes, under the additional assumption that $X$ is measurable. I posted an answer that has a gap, and my proof does not seem to work when $X$ is non-measurable with positive outer measure, but it does work when $X$ is measurable with positive measure. (The case when $X$ has measure zero is trivial, since then we could take $Y$ to be the empty set.) $\endgroup$
    – Mirko
    Apr 10, 2015 at 3:41
  • 2
    $\begingroup$ @Mirko If X is a Borel and has positive measure, then it has a similar copy of every finite set of reals. You can prove this using Lebesgue density theorem. $\endgroup$
    – Ashutosh
    Apr 23, 2015 at 19:51

2 Answers 2

6
$\begingroup$

I think I can do it if every subset of $\mathbb R$ of cardinality $< \mathcal c$ has measure $0$ (which is a consequence of Martin's axiom).

Let $m$ be the outer measure of $X$. WLOG we can assume $m > 0$. Let $G$ be a $G_\delta$ containing $X$ with outer measure $m$. Index the $G_\delta$ subsets of $G$ with measure $< m$ by ordinals: $G_\alpha: \alpha < A$ where $A$ is the least ordinal of cardinality $\mathcal c$. Proceed by transfinite induction to choose, for each $\alpha < A$, $x_\alpha \in X \backslash G_\alpha$ such that $\{x_\beta: \beta \le \alpha\}$ is midpoint free. This is possible because $X \backslash G_\alpha$ has cardinality $c$, while the set of "forbidden" points ($(x_\beta + x_\beta')/2$ and $2 x_\beta - x_\beta'$ for $\beta, \beta' < \alpha$) has cardinality $< \mathcal c$.

$\endgroup$
0
0
$\begingroup$

Warning: The following answer has a gap, and I do not know if it could be fixed. I leave it for consideration, now that I already wrote it. (It does work, in ZFC, but under the additional assumption that $X$ is measurable.)

Repeat the usual construction of a Bernstein set: Let $\{F_\alpha: \alpha < \mathfrak c = 2^{\aleph_0}\}$ be a list of all closed uncountable subsets of the real line. Call a set $S$ free if it contains no three different points $x,y,z$ with $y-x=z-y$ (that is, no points with $y=\frac{x+z}2$). Let $E(S)=\{z\in\mathbb R\setminus S: \exists x,y\in S, z=\frac{x+y}2,\mathrm{\ or\ } z=y+(y-x)\}$, that is $E(S)$ are the excluded (or "forbidden") points for $S$. (Note that $z=y+(y-x)$ could be to the left or to the right of $y$, depending on whether $y-x$ is negative or positive.) For each $\alpha$ pick different $x_\alpha,y_\alpha\in F_\alpha$ such that if $P_\alpha=\{x_\beta:\beta<\alpha\}$ and $Q_\alpha=\{y_\beta:\beta<\alpha\}$ then we have :
$x_\alpha\not\in E(P_\alpha)\cup P_\alpha\cup Q_\alpha$, and $y_\alpha\not\in P_\alpha\cup Q_\alpha$.
(We could do this since $|F_\alpha|=\mathfrak c$, but $|E(P_\alpha)|\le\aleph_0\cdot |P_\alpha|\le\aleph_0\cdot|\alpha|<\mathfrak c$.)

Let $P=\{x_\alpha:\alpha<\mathfrak c\}$ and $Q=\mathbb R\setminus P$. Then each of $P$ and $Q$ is a Bernstein set of inner measure zero and full outer measure. Moreover, by construction $P$ is free. We have that $X\cap P=X\setminus Q$ and since $Q$ is of inner measure zero, the outer measure $m^*(X)=m^*(X\setminus Q)$ (this is the gap in my proof, this is not necessarily the case, unless we assume that $X$ is measurable). That is $m^*(X\cap P)=m^*(X)$, and $X\cap P$ is free since $P$ is.

Well, the gap is, I only know that $X$ has positive outer measure, but this does not rule out the possibility that $X$ is of inner measure zero, so it might happen that $X\subset Q$ and in that case $X\cap P$ would be empty.

I got mislead by the following answer (even if the answer is fine, but I didn't read it right, it only applies when we start with a measurable set, and the set $X$ above is not necessarily measurable):
A set $E$ with positive Lebesgue measure can be decomposed as a union $E=A\cup B$ where each of $A$ and $B$ have zero inner measure, and therefore each of $A$ and $B$ are nonmeasurable with $m^*(A)=m^*(B)=m(E)$.
Non Lebesgue measurable subsets with "large" outer measure

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.