Countability of R in diagonalization argument of Arzela-Ascoli theorem proof

264 Views Asked by At

I have an intricate issue with the diagonalization argument used in the proof of Arzela-Ascoli theorem. It goes as follows:

So assume that $\scr F$ has these three properties [closed, bounded, equicontinuous] and let $(f_n)$ be a sequence in $\scr F$. We will construct a convergent subsequence. By Corollary 8.6.8, there is a sequence $(x_i)$ such that for each $r > 0$, there is an integer $N$ such that $\{x_1, \dots, x_N\}$ forms and $r$-net for $K$.

We claim that there is a subsequence of $(f_n)$, call it $(f_{n_k})$, such that $$\lim_{k \to \infty} f_{n_k}(x_i) = L_i\quad\text{exists for all }i \ge 1.$$ To prove this, let $\Lambda_0$ denote the set of positive integers. Since $\big(f_n(x_1)\big)$ is a bounded sequence, the Bolzano-Weierstrass Theorem (2.7.2) provides a convergence subsequence. That is, there is an infinite subset $\Lambda_1\subset \Lambda_0$ such that $$\lim_{n \in \Lambda_1} f_n(x_1) = L_1\quad\text{exists.}$$ Next, $\big(f_n(x_2)\big)_{n\in\Lambda_1}$ is a bounded sequence, so there is an infinite subset $\Lambda_2 \subset \Lambda_1$ such that $$\lim_{n \in \Lambda_2} f_n(x_2) = L_2\quad\text{exists.}$$ Continuing in this way, we obtain a decreasing sequence $\Lambda_0 \supset\Lambda_1 \supset\Lambda_2 \supset\cdots$ of infinite sets such that $$\lim_{n \in \Lambda_i} f_n(x_i) = L_i\quad\text{converges for each }i \ge 1.$$

We now use a diagonalization method, similar to the proof that $\Bbb R$ is uncountable, Theorem 2.9.8. Let $n_k$ be the $k$th entery of $\Lambda_k$; and let $\Lambda = \{n_k : k \in \Bbb N\}$. Then for each $i$, there are at most $i-1$ entries of $\Lambda$ that are not in $\Lambda_i$. Thus $$\lim_{k\to\infty} f_{n_k}(x_i) = \lim_{n \in \Lambda_i} f_n(x_i) = L_i\quad\text{for all}\quad i \ge 1,$$ proving the claim.

For simplicity of notation, we use $g_k$ for $f_{n_k}$ in the remainder of the proof. Now fix $\varepsilon > 0$. By uniform equicontinuity, there is $r > 0$ such that $$\|f(x) - f(y)\| < \dfrac{\varepsilon}3\quad\text{for all}\quad f\in \mathscr F\text{ and }\|x-y\|< r.$$ Choose $N$ such that $\{x_1, \dots,x_N\}$ is an $r$-net for $K$. Since the $g_k$ converge at each of the $N$ points, there is some $M$ such that $$\|g_k(x_i) - g_l(x_i)\| < \dfrac{\varepsilon}3\quad\text{for all}\quad k,l \ge M\text{ and }l \le i \le N.$$

Let $k,l \ge M$ and pick $x \in K$. Since $\{x_1, \dots, x_N\}$ is an $r$-net for $K$, there is some $i \le N$ such that $\|x - x_i\| < r$. We need an $\varepsilon/3$-argument to finish the proof: $$\begin{align}\|g_k(x) - g_l(x)\|&\le \|g_k(x) - g_k(x_i)\| + \|g_k(x_i) - g_l(x_i)\| + \|g_l(x_i) - g_l(x)\|\\ &\le\dfrac{\varepsilon}3+\dfrac{\varepsilon}3+\dfrac{\varepsilon}3 = \varepsilon\end{align}$$ Thus $g_k$ is uniformly Cauchy, and so converges uniformly by Theorem 8.2.2. The limit $g$ belongs to $\scr F$ because $\scr F$ is closed. Finally, since every sequence in $\scr F$ has a convergent subsequence, it follows that $\scr F$ is compact.

  1. Corollary 8.6.8

Initially my question was:

As far as I understand, the diagonalization argument has $r\rightarrow0$ and $N\rightarrow\infty$ which will map every point in the set $K$ to some sequence. But real numbers on any set like $[a,b]$, $a<b$ are uncountable and by taking $N\rightarrow\infty$ we assume they are countable? The proof maps every point in set K to some sequence, correct? The proof presumes we can count these sequences since we will take $k$-th element from $k$-th sequence but the way these sequences are constructed and mapped to in the diagonalization argument signals we can't. Am I missing something?

After a discussion with Paul Sinclair below, I figured out I missed the meaning of Corollary 8.6.8 somehow and $r\rightarrow0$ and $N\rightarrow\infty$ is not even considered in the proof. The sequence will be the same but for every $r$-net we might take a head of the sequence of different length $N$ every time.

1

There are 1 best solutions below

1
On BEST ANSWER

Am I missing something?

Not so much "missing something" as "seeing things that are not there".

You have read far too much into what is just an off-hand comment in the proof. Note that the author says "a diagonalization method", not "a diagonalization argument". The difference is that author is just building a sequence. They are not making an argument about countability or uncountability at all. The comment was just to give another example where a sequence was built by following a diagonal. The purposes of these two sequences are entirely different. And the conclusion obtained from this diagonalization is the proof of the opening claim: the existence of a subsequence of functions $f_{n_k}$ such that for all $i, \lim_{k \to \infty} f_{n_k}(x_i)$ converges. the line "proving the claim" at the top of the second page signals the end of anything related to diagonalization.

As far as I understand, the diagonalization argument has $r\rightarrow0$ and $N\rightarrow\infty$

As already noted, the diagonalization bit is already finished at this point, but the proof also doesn't have $r \to 0$ and $N \to \infty$ in it. Instead, it fixes an arbitrary $\varepsilon > 0$, and uses that to fix a value of $r$ per equicontinuity. Then it uses that fixed $r$ to fix a value of $N$. It is true that a smaller $\varepsilon$ will require a smaller $r$, and that in turn will require a larger $N$. But that is immaterial to the argument being made. That argument only requires that for any given $\varepsilon$, you can find a single corresponding value of $r$ and a single corresponding value of $N$.

which will map every point in the set $K$ to some sequence.

No. Just, no. There is no such claim. There is not even a suggestion of any such thing in the proof. I can only presume you came up with the idea based on the misunderstanding of what they meant by the reference to Cantor's diagonalization argument.

The rationals form an $r$-net for $\Bbb R$ for every value of $r > 0$. But no one suggests that means $\Bbb R$ is mapped to $\Bbb Q$. It obviously is not.

The rest of your post just builds on these false ideas.

But real numbers on any compact set like $[a,b]$, $a<b$ are uncountable

As an aside - not a problem with your overall argument, just an easily fixed incorrect detail - compactness does not guarantee that a set of real numbers is uncountable. All finite sets are compact. Many countable sets, such as $\{0, \frac 1n \mid n \in \Bbb N^+\}$, are also compact. What assures $[a,b]$ is uncountable is that it contains $(a,b)$, and all non-empty open sets of real numbers are uncountable.