5
$\begingroup$

Vitali sets are nonmeasurable and in particular are not null sets. But all Vitali sets are in some sense small, as described below. Let $V$ be any Vitali set and let $k \in \mathbb{N}$. For each $i \in \mathbb{Z}$, consider the set $A_i = V \cap [\frac{i}{k}, \frac{i+1}{k} )$ and let $B_i = A_i - \frac{i}{k} = \{x-\frac{i}{k} : x \in A_i\}$. Since the difference between two different elements of $V$ is irrational, the sets $B_i$ for $i \in \mathbb{Z}$, are pairwise disjoint. The union of the $B_i$ is a subset of the open interval $[0,\frac{1}{k})$. This shows that given any interval, no matter how small, it is possible to partition any Vitali set $V$, by intersecting $V$ with half-open intervals, into countably many pieces and then translate those pieces so that they all fit, without overlap, inside the interval.

Here is the formalization I have in mind. We say $A \subseteq \mathbb{R}$ is "small" if for all $\epsilon>0$ there exists a sequence of half-open intervals $I_n = [a_n, b_n)$ which together partition $\mathbb{R}$ and a sequence of real numbers $t_n$ such that the sequence $(A \cap I_n) - t_n$ is a pairwise disjoint sequence of sets whose union is contained in the interval $[0,\epsilon)$.

Is there a natural finitely additive measure on $\mathbb{R}$ whose null sets are exactly the sets that are small in the above sense? Can Lebesgue measure be extended to a finitely (but not countably!) additive measure so that the sets that are small in the sense above are precisely the null sets?

$\endgroup$
5
  • 3
    $\begingroup$ If such a measure exists, then in particular a union of two small sets has to be small. Do you know if this is indeed the case? (If so, then the measure giving 0 to small sets and infinite mass to the others fits the bill, but it is not particularly enlightening.) $\endgroup$
    – Pierre PC
    Commented Nov 3, 2022 at 18:33
  • $\begingroup$ I don't know whether the union of two small sets must be small, but suspect that it must. (That there is no Banach-Tarski decomposition in one dimension, using only finitely many pieces, seems like it might be relevant here.) Since countably many translates of a Vitali set covers the line, this notion of small is not closed under countable unions. $\endgroup$
    – Aaron Hill
    Commented Nov 3, 2022 at 19:05
  • $\begingroup$ For the union, it should just work to fit $A$ into $[0,\varepsilon/2)$ and $B$ into $[\varepsilon/2,\varepsilon)$, right? More generally, the natural candidate would seem to be to set $\mu(A)$ to be the $\inf$ of all $\varepsilon$ where it works. Problem just is that this should be finitely sub-additive, but I don't see why it would finite additive. $\endgroup$
    – Arno
    Commented Nov 3, 2022 at 19:29
  • 1
    $\begingroup$ I don't think finite unions is that easy. When you are partitioning the set $A \cup B$ into pieces you must be using half-open intervals. That means that elements of your partition may have a combination of points from $A$ and points from $B$. You can't get all of the points in $A$ to go in $[0, \frac{\epsilon}{2})$ and all of the points of $B$ to go in $[\frac{\epsilon}{2}, \epsilon)$. Maybe the way you can get the points of $A$ to fit in a small interval is incompatible (somehow) with the way you can get the points of $B$ to fit in a small interval. $\endgroup$
    – Aaron Hill
    Commented Nov 3, 2022 at 21:46
  • $\begingroup$ @AaronHill Ah, of course. Thanks for clearing up my confusion. $\endgroup$
    – Arno
    Commented Nov 4, 2022 at 14:55

0

You must log in to answer this question.

Browse other questions tagged .