5

Prove that the open interval $(0, 1)$ contains uncountably infinite numbers.

Apparently, there is a way to prove this proposition using Cantor's diagonalization argument.

How does that work? How does looking at the diagonal of a big grid of numbers help? Is there a more straightforward proof?

Asaf Karagila
  • 393,674
  • 4
    Maybe you might have a look at this question: How does Cantor's diagonal argument work? – Martin Sleziak May 04 '12 at 19:30
  • 1
    Given a set $A$, it can't be a bijection between $A$ and $\mathcal{P}(A)$. In particular $\mathcal{P}(\mathbb{N})$ is not numerable. Construct a bijection between $(0,1)$ and $\mathcal{P}(\mathbb{N})$. – leo May 04 '12 at 19:33
  • 1
    A numerable set has measure $0$. $(0,1)$ has measure $1$, therefor $(0,1)$ can't be numerable. – leo May 04 '12 at 19:35
  • @leo, but you define the Lebesgue measure to be translation invariant and sigma additive; if you do not know that $(0,1)$ is uncountable you cannot assume the definition is well. You use the fact that you aim to prove in that second comment. – Asaf Karagila May 04 '12 at 19:40
  • 1
    @AsafKaragila, We can construct the Lebesgue measure from the outer measure ussually defined as the inf taken over the sums of lengths of intervals that cover the set what we are measuring, that properties are deduced from the construction. However in this particular case, given a set $A$ of real numbers you can say that $m(A)=0$ if given $\epsilon\gt 0$, there exist a collection ${(a_k,b_k)}{k\in\Delta}$, $\Delta\subseteq\mathbb{N}$ such that $$\sum{k\in\Delta}b-k-a_k\lt\epsilon.$$ This is enough to sustain the above argument. Even we can avoid to say that $(0,1)$ has measure $1$. – leo May 04 '12 at 19:55
  • @leo: But you are using the fact that it is possible to define a sigma additive measure for which $m(A)=m(A+x\pmod 1)$ for all $x$. This is in fact a uniform measure when you consider singletons. It is easy to prove that such measure cannot be defined on a countable set unless it is the zero measure. – Asaf Karagila May 04 '12 at 19:58
  • @leo, please show me an outer measure on $P(\mathbb N)$ which gives all the singletons zero; is sigma-additive and gives $\mathbb N$ a positive measure. – Asaf Karagila May 04 '12 at 20:09
  • Let $(a,b)$ a finite no trivial interval. We can prove (though not very enjoyable) that if $\Delta\subseteq\mathbb{N}$ and $(a,b)\subseteq\bigcup_{k\in\Delta} (a_k,b_k)$ then $b-a\leq \sum_{k\in\Delta}b_k-a_k$. As consequence $(0,1)$ can't have measure $0$ with the above definition. In the other hand consider ${x_k}{k\in\mathbb N}$, by taking the covering ${(x_k-\epsilon/2^{k+1},x-k+\epsilon/2^{k+1})}$ it is easy to see that ${x_k}{k\in\mathbb N}$ has measure $0$ with the above definition. – leo May 04 '12 at 20:11
  • @AsafKaragila, my last comment was not in response to your comment. It is an attempt to sustain a variant of the above comment without measure theoretic considerations. – leo May 04 '12 at 20:13

5 Answers5

19

We can take a small twist of Matt N.'s hint to make sure we don't run into trouble with dual representations. To see what the problem is, note that, for example, there are two ways of writing $\frac{1}{2}$ in binary: $$0.100000\ldots\quad\text{and}\quad 0.011111\ldots$$ In principle, it would be that your list begins with $0.1000\ldots$, and then the numbers in position $n$ all have $n$th digit equal to $0$ for all $n\gt 1$. Then the straight "change the digit" process leads to $0.01111\ldots$, which is on the list (it's a different representation of the first number on the list).

So the first thing we need to do is specify that we will use one particular representation in our "list"; usually the one that terminates if there is a choice.

That done, instead of looking at the single diagonal digit, work with diagonals in blocks of $2$; that is, look at the digits in position $1$ and $2$ first, then the digits in positions $3$ and $4$ for the second number, and so on, looking at the digits in position $2k+1$ and $2k+2$ for the $k$th number. Then make the "switch" ensuring that the number you construct does not have two different representations. For instance, if the $k$th number in the list has $00$, $01$, or $11$ in the $2k+1$ and $2k+2$ positions, then put $10$ in your number on those positions; if the $k$th number in the list has $10$, then put $01$ on your list. Then the number you construct does not have two representations, so you can test that it is not on the list by simply comparing the $2k+1$ and $2k+2$ digits with the $k$th number on the list.

So if your list looks like: $$\begin{align*} &0.{\color{red}{a_{11}a_{12}}}a_{13}a_{14}a_{15}a_{16}\cdots a_{1,2k+1}a_{1,2k+2}\cdots\\ &0.a_{21}a_{22}{\color{red}{a_{23}a_{24}}}a_{25}a_{26}\cdots a_{2,2k+1}a_{2,2k+2}\cdots\\ &0.a_{31}a_{32}a_{33}a_{34}{\color{red}{a_{35}a_{36}}}\cdots a_{3,2k+1}a_{3,2k+2}\cdots\\ &\vdots\\ &0.a_{k1}a_{k2}a_{k3}a_{k4}a_{k5}a_{k6}\cdots {\color{red}{a_{k,2k+1}a_{k,2k+2}}}\cdots\\ &\vdots \end{align*}$$ You then take each red block and construct a new number $$0.\color{blue}{b_{1}b_{2}}\color{green}{b_3b_4}\color{brown}{b_5b_6}\cdots\color{magenta}{b_{2k+1}b_{2k+2}}$$ where $\color{blue}{b_1b_2}$ is chosen so that it is different from $\color{red}{a_{11}a_{12}}$; $\color{green}{b_3b_4}$ is chosen so that it is different from $\color{red}{a_{23}a_{24}}$; $\color{brown}{b_5b_6}$ is chosen so that it is different from $\color{red}{a_{35}a_{36}};\ldots,\color{magenta}{b_{2k+1}b_{2k+2}}$ is chosen so that it is different from $\color{red}{a_{k,2k+1}a_{2k+2}}$, etc. So you are "going down the diagonal" and changing things so that the resulting number is not on the list: it's not the first number on the list, because it differs from it in the first two entries and your number has only one representation. It's not the second number on the list because it differs from that one in the third and fourth digits; given any $k$, the number constructed is not the $k$th number on the list because it differs from it in the $2k+1$ and $2k+2$ positions. Etc.

Arturo Magidin
  • 398,050
  • This deserves a multiple vote, because it is so often missed or dismissed. – Mark Bennet May 04 '12 at 19:39
  • To deal with this, you can restrict to only infinite representations, which are unique. – leo May 04 '12 at 19:41
  • 1
    @MarkBennet: I agree. You can give him a second vote on the answer linked by Martin on the question. – Asaf Karagila May 04 '12 at 19:41
  • @leo: But you run into exactly the same problem: if we restrict to infinite representations, what is to stop the first number in the list form being $0.01111\ldots$, and for $n\gt 1$ the $n$th number having $0$ on the $n$th position; then the number you get is $0.10000\ldots$ which is a finite representation of a number which is on the list. Doesn't matter which representation you pick, if you are not careful when constructing the diagonal number you can end up with a different presentation for a number already on the list. – Arturo Magidin May 04 '12 at 19:48
  • @ArturoMagidin, in the post linked above, we consider $0.10000\ldots$ as a finite representation. – leo May 04 '12 at 19:58
  • @leo: I know. Do you understand my point, or not? The number you construct by simply switching the diagonal digits may be a finite presentation; that means that it could be on the list with its infinite presentation. You have control over the representation of the numbers on the list, but if you specify your procedure as "just change the $k$th digit of the $k$th number", then you don't have a choice of presentation in the number you construct. In the example I write in the comment above, all numbers which I choose have their infinite presentation. – Arturo Magidin May 04 '12 at 20:02
  • I see. Thanks for show me that. – leo May 04 '12 at 20:24
  • Arturo, as a color blinded person I really cannot read the yellow(? the $b_5b_6$) font. Is it possible to use a different color please? – Asaf Karagila May 04 '12 at 20:27
  • @Asaf: Is cyan any better? – Arturo Magidin May 04 '12 at 20:30
  • @Arturo: Better, but far from great. It doesn't hurt my eyes, though. So I can live with that... thanks! – Asaf Karagila May 04 '12 at 20:33
  • @Asaf: How about brown? – Arturo Magidin May 04 '12 at 20:40
  • Bless you.${}{}$ – Asaf Karagila May 04 '12 at 20:44
  • 1
    @Asaf: That wasn't me sneezing... There's someone else in the room! Run! – Arturo Magidin May 04 '12 at 20:45
  • I don't see what is the advantage of working with pairs of binary digits. The diagonalization argument will work using single digits in any base. (well, of course, a pair of binary digits is effectively a single digit in base 4). – Kaz May 04 '12 at 22:19
  • @Kaz: If you try to use a single digit in base 2, you run the risk of producing a number that is on the list. It's the same risk that you run in the decimal case if you carelessly say you will make your number have a 9 in the $n$th position if the $n$th number does not have a $9$ in the $n$th position, an a $0$ otherwise. You could end up with a number that has two representation, and in that case it is not clear that your number is not on the list in its other representation. In base $b$, one usually tries to avoid $0$ and $b-1$, but this is not possible in base $2$. – Arturo Magidin May 05 '12 at 01:02
  • @Kaz: (Cont) So you may end up constructing a number that can be expressed in two different ways. If so, then it is not enough to know that it differs from the $n$th number in the $n$th digit to be able to conclude that it is not the $n$th number. Like I say in my examples: say you end up with the "diagonal number" 0.10000000000.... and the first number on your list is 0.01111111111.... (in binary). Then the diagonal number's expansion differs from the first on the list in its first digit, but the two expressions represent the same number, so your "diagonal number" is on the list. – Arturo Magidin May 05 '12 at 01:06
15

There have already been posted a few really good answers explaining how to prove this by diagonalization. Just for your awareness, here is an alternative way to show it.

Look at $[0,1]$ as a metric space (a subspace of $\mathbb{R}$ with the usual Euclidean distance). It is complete since $\mathbb{R}$ is complete and $[0,1]$ is closed. But we may write $$ [0,1] = \bigcup_{x \in [0,1]}\{x\}. $$

Singletons are closed and have empty interior, so it follows that $[0,1]$ must be uncountable, otherwise this would contradict the Baire Category Theorem. From here, we have that $(0,1)$ is obtained by removing two elements from an uncountable set, and so $(0,1)$ is uncountable.

nullUser
  • 27,877
  • Plus one for using Baire. : ) This is my favourite answer! – Rudy the Reindeer May 04 '12 at 21:26
  • 2
    I find metric space / topology arguments like this more enlightening than decimal representations, although they might need more machinery. Another way is to show that every compact Hausdorff space without isolated points is uncountable. This is actually how Munkres proves that $[0,1]$ is uncountable in his Topology. – Cihan May 04 '12 at 21:42
6

Hint: Represent a number in $(0,1)$ in base $2$. Write the numbers below each other line by line. Then take $+1$ mod $2$ of the diagonal...

4

$\newcommand{\N}{\mathbb N}$ Let $f:\N\to (0,1)$ a function. Given $n\in\N$ consider the infinite representation of $f(n)$, $$f(n)=0.a^n_1a^n_2\ldots a_n^n\ldots$$ in base $10$. This infinite representation is unique.

For each $n\in\N$ define $$b_n=\begin{cases} 5 & \text{if} &a_n^n\neq 5\\ 7 & \text{if} & a_n^n=5\end{cases}$$ we will see that $y=0.b_1b_2b_3\ldots$ can not be in the image of $f$.

Suppose that there is an $l\in\N$ such that $$y=f(l)=0.a_1^la_2^l\ldots a_l^l\ldots$$ Then $a_l^l=b_l$, i.e. $$a_l^l=\begin{cases} 5&\text{if}& a_l^l\neq 5\\ 7&\text{if}&a_l^l=5\end{cases}$$ you can see that such an $l$ is not possible.

Therefore $f$ is not surjective and $(0,1)$ is not numerable.

leo
  • 10,433
3

Cantor diagonal argument says: Suppose there is any countable collection of numbers, then there is another number which is not in the collection.

To use it, assume that $\{a_n\mid n\in\mathbb N\}$ is a countable collection of numbers from $(0,1)$, let $a_n^i$ be the $i$-th digit in the decimal representation of $a_n$, let $a$ be a number such that $a^i\neq a_i^i$ for all $i$ (if you also ensure that $a^i\neq0,9$ then you have ensured that you do not run into problems of numbers having two representations), this number is between $0$ and $1$ and is definitely not on the list. We have shown that there cannot be a surjective function from $\mathbb N$ onto $(0,1)$.

Another way is to use Cantor's theorem that $\aleph_0<2^{\aleph_0}$. Define the following injection from $P(\mathbb N)$ into $(0,1)$:

$$A\mapsto\sum_{n\in A}\frac2{3^{n+1}}$$

Show that this is an injective function, and therefore $|(0,1)|\geq2^{\aleph_0}>\aleph_0$, as wanted.

Asaf Karagila
  • 393,674