2

In preparation for an exam I am struggling with the following problem. We let $A:=\{x=(x_{n})_{n}\in \ell^{2}| \phantom{x} \|x\|\leq 1\}$ and consider $d:A\times A\rightarrow \mathbb{R}_{+}$ defined by

$$d(x,y)=\sum_{n=1}^{\infty}(1/2)^{n}|x_{n}-y_{n}|.$$ Prove that $(A,d)$ is sequentially compact.

My question is can I do this by showing it is closed and bounded or is it not allowed because of dimensionality issues? Or is there another way I can show it is sequentially compact without resorting to a diagonal argument?

Lech121
  • 343
  • 1
  • 9
  • You can only use the closed-boundedness of a space to prove compacity in the metric spaces $\mathbb R^n$ with the Euclidean metric. Otherwise it's not a general result. – Patrick Da Silva Sep 29 '13 at 14:57
  • @Patrick so is a diagonal argument the only way? Or is there another possibility? – Lech121 Sep 29 '13 at 14:58
  • I'm trying to work out what you mean by diagonal argument, but I believe that's your only option. – Patrick Da Silva Sep 29 '13 at 15:06
  • Alternately you could prove this space is compact, or that it is (complete and totally bounded). Both of these conditions are equivalent to sequential compactness in a general metric space. I would try for (complete and totally bounded). – Daron Sep 29 '13 at 15:12
  • I've answered what I believe is the only possible (and reasonable) answer to this problem. The option of showing that the closed ball is complete and totally bounded is also legitimate, but uses a bit of heavy machinery : this link http://math.stackexchange.com/questions/109550/totally-bounded-complete-implies-compact – Patrick Da Silva Sep 29 '13 at 15:57
  • explains how complete and totally bounded implies sequentially compact. To show that your closed ball is complete, all you have to show is that it is closed (which follows by continuity of the norm) because $\ell^2$ is complete. Totally bounded does not seem that hard either I guess, but I didn't give it a try. – Patrick Da Silva Sep 29 '13 at 15:59
  • I wasn't suggesting you prove it is totally bounded & complete, and then prove that this in general implies sequential compactness. I believe those three conditions being equivalent is a pretty standard result in metric topology which was probably covered on your course. – Daron Sep 29 '13 at 16:06

1 Answers1

1

Let $x = (x_n)_{n \in \mathbb N} \in \ell^2$ denote an element of $\ell^2$ and $x^k = (x_n^k)_{n \in \mathbb N}$ denote a sequence of elements in $\ell^2$. Since for every $x^k$, $\sum_{n \ge 1} |x_n^k|^2 \le 1$, we have $x_n^k \in [-1,1]$ for every element of the sequence. Note that $[-1,1]$ is closed and bounded, hence sequentially compact.

From the sequence $x^k$ satisfying $\| x^k \| \le 1$, the sequence $\{x_1^k\}_{k \in \mathbb N}$ is a sequence in $[-1,1]$, hence admits a convergent subsequence to $x_1$. Choose the sequence such that $|x_1^k - x_1| \le 2^{-k}$.

Inductively, if you have a subsequence of $x^k$ such that the first $N$ component sequences $\{x_1^k\}_{k \in \mathbb N}$, $\dots$, $\{x_N^k\}_{k \in \mathbb N}$ converge respectively to $x_1, \dots, x_N$ and $|x_i^k - x_i| \le 2^{-k}$ for $k \ge i$, $1 \le i \le N$, keep the first $N$ terms of the subsequence $x^k$ previously chosen as they are, and choose a subsequence that begins at the $N+1^{\text{th}}$ such that $x_{N+1}^k$ converges to $x_{N+1}$ and $|x_{N+1}^k - x_{N+1}| \le 2^{-k}$ for $k \ge N+1$.

It's important that you require $k \ge i$ and not $k \ge 1$ ; otherwise you cannot "let the algorithm run to infinity". Letting $k \ge i$ allows you to keep the first $i$ terms of your subsequence satisfying the constraints on their components, and at the next iteration of the process, the first $i$ terms can be preserved, and all you need to choose the rest of the sequence, starting at the $(i+1)^{\text{th}}$ term. Also note that if $|x_n^k - x_n| \le 2^{-k}$, then a subsequence of $x^k$, say $x^{k_{\ell}}$, will satisfy $|x_n^{k_{\ell}} - x_n| \le 2^{-k_{\ell}} \le 2^{-k}$ because $k_{\ell} \ge k$.

In other words, each iteration only builds the $n^{\text{th}}$ term of the subsequence, but in such a way that some bounds on the convergence are obtained along the way.

This gives you a subsequence of $x^k$ and an element of $\ell^2$, namely $x = (x_1,\dots,x_n,\dots)$ (although we still have to prove $\|x\| \le 1$) such that $$ |x_n^k - x_n| \le 2^{-k}, \quad k,n \in \mathbb N, \quad k \ge n. $$ Now $$ d(x^k,x) = \sum_{n \ge 1} \frac{|x_n^k - x_n|}{2^n} = \sum_{n=1}^k \frac{|x_n^k - x_n|}{2^n} + \sum_{n > k} \frac{|x_n^k - x_n|}{2^n} \le \sum_{n=1}^k \frac{1/2^k}{2^n} + \sum_{n > k} \frac 1{2^n} $$ and $$ \sum_{n=1}^k \frac{1/2^k}{2^n} + \sum_{n > k} \frac 1{2^n} = \frac 1{2^k} \left( \sum_{n=1}^k \frac 1{2^n} \right) + \sum_{n > k} \frac 1{2^n} \le \frac 1{2^k}(1) + \frac 1{2^k} = \frac 1{2^{k-1}} \underset{ k \to \infty}{\longrightarrow} 0. $$ Since $\ell^2$ is complete, $x \in \ell^2$ and since $\|x^k\| \le 1$ and $x^k \to x$ in $\ell^2$, $\|x\| \le 1$.

Hope that helps,