2

In this answer and this question, we saw how to construct open intervals covering $\Bbb Q \cap [0,1]$ with total length being $\epsilon$, an arbitrarily small positive real number.

If we do it with disjoint open intervals, can we still make the total length arbitrary?

In symbols: $$\forall \epsilon > 0: \exists \langle I_n \rangle: \left(\left(\sum_{i \in I_n}|i| < \epsilon \right) \land \left(\forall a,b \in \Bbb N: I_a \cap I_b = \varnothing \right) \land \left(\Bbb Q \cap [0,1] \subset \bigcup I_n\right)\right)$$

DHMO
  • 3,014
  • 12
  • 20
  • 1
    Any open set in the real line can be written as a countable union of disjoint open intervals http://math.stackexchange.com/questions/2077116/open-set-of-reals-as-union-of-open-intervals – Del Mar 28 '17 at 10:00

1 Answers1

2

You can make it arbitrarily small even with disjoint open intervals.

Take an arbitrary enumeration of $\mathbb Q\cap [0,1]$, as $\{q_i\}_{i=0}^\infty$. We will make sure that each interval we take has irrational valued endpoints. Also take $\gamma=\epsilon/2$

At step $i$ we have a finite number of open intervals that have irrational endpoints call them $Q_i=\{(a_j,b_j)\;|\; (j<i)\;\wedge\; (\forall j<i \exists l<i \;q_j\in(a_l,b_l))\}$ which are all disjoint and the sum of whose sizes add up to some $\epsilon_i<\gamma$. We also have $Q_i\subseteq Q_k$ when $i<k$.

We now need to create $Q_{i+1}$ which still satisfies all the criteria.

If $q_{i+1}$ is in one of the intervals that is in $Q_i$ we set $Q_{i+1}=Q_i$ and are done.

If $q_{i+1}$ is in none of the intervals in $Q_i$ we look at the distances $d_j$ of $q_{i+1}$ from $(a_j,b_j)$ for all $j<i+1$. There are only finitely many such distances and since $a_j,b_j$ are all irrational none of those distances is 0. Take the smallest of those distances and call it $D_i$. Let $s=\min\{D_i,(\gamma-\epsilon_i)/2\}$. Than let us take some $s^\prime<s/2$ irrational and define $a_i=q_i-s^\prime$ and $b_i=q_i+s^\prime$. Define $Q_{i+1}=Q_{i}\cup \{(a_i,b_i)\}$.

Note that the fact that $s^\prime<s/2$ makes sure that $|(a_i,b_i)|<s$ and so $\epsilon_{i+1}=\epsilon_i+|(a_i,b_i)|<\gamma$. At the same time the same condition ensures that $(a_i,b_i)\cup (a_j,b_j)=\emptyset$ for all $j<i$. Obviously $q_i\in(a_i,b_i)$.

Now we just define $Q=\bigcup_{i=0}^\infty Q_i$.

Note

This only shows you can make the total length arbitrarily small. Not arbitrary. I'm not sure how to hit any specific $\epsilon$.

DRF
  • 5,167
  • But we can't do it with a finite amount of intervals, right? – DHMO Mar 28 '17 at 10:02
  • @DHMO Of course not. Any finite amount of intervals covering all rationals in $[0,1]$ will have to have total length 1. Assuming they are disjoint. – DRF Mar 28 '17 at 10:05
  • And why is that? – DHMO Mar 28 '17 at 10:05
  • 1
    @DHMO There must be one containing $0$. This one will have a right endpoint call it $a_1$. Since there are only finitely many left there must be one with a left most endpoint. This endpoint must be $a_1$ since otherwise we're leaving out a rational. That interval is then $(a_1,a_2)$. Rinse and repeat. – DRF Mar 28 '17 at 10:27
  • Basically the same line of "induction only works for $n \in \Bbb N$, but not $\Bbb N$ itself (AKA $\omega$)" – DHMO Mar 28 '17 at 10:38
  • @DHMO hmm not sure really. The point is more that finite and infinite sets behave very differently. I'm not exactly sure what saying that induction only works for $n\in\mathbb N$ and not $\omega$ really means. Standard induction just says: If $A\subseteq \mathbb N; \wedge 0\in A ; \wedge \forall i ; (i\in A\implies S(i)\in A)$ then $A=\mathbb N$. – DRF Mar 28 '17 at 10:59