17

It is known that the closed unit ball $\overline{B_1(0)}$ in a normed space $X$ is compact if and only if $\dim X < \infty$. In particular, the $\overline{B_1(0)}$ is not compact if $\dim X = \infty$. The proof of this involves finding a sequence $\{ x_n \}_{n\in\mathbb{N}} $ with $||x_n|| = 1 $ for all $n \in \mathbb{N}$ such that $||x_n - x_m|| > \frac{1}{2}$ for all $n \neq m$. Then this sequence is a bounded sequence that is not a Cauchy-sequence, so it does not have a converging subsequence, so $X$ is not (sequentially) compact. The construction then goes with Riesz's lemma, by subsequently finding points with norm 1 that have distance greater than $\frac{1}{2}$ to the subspace generated by the preceeding points.

Now my question is, is any closed ball $\overline{B_r(0)}$ with $r>0$ non-compact in an infinite-dimensional space? It seems very intuitive, since topologically the balls are all diffeomorphic, and it would seem unlikely that for $r>1$ the larger ball is compact while $\overline{B_1(0)}$ is not. However, when I look at the proof, I notice that Riesz' lemma really only works for norm 1 and not for some arbitary norm. Is there any way to adapt this lemma or make use of a different construction in order to say something about all the closed balls?

PianoEntropy
  • 2,776
  • 9
    Yes. Take your sequence ${x_n}$ and consider the sequence ${r x_n}$. – David Mitra Dec 07 '12 at 14:24
  • 1
    The norm in Riesz's lemma is one due to a normalization step in the proof. Multiply this normalization by arbitrary constant will adapt Riesz's lemma to any given norm. – Bihu Duo Dec 29 '20 at 20:49

2 Answers2

23

If $x$ is a vector in your infinite dimensional space and $\alpha\neq 0$, then the functions $x\mapsto v+x$ and $x\mapsto \alpha x$ are homeomorphisms of the space onto itself. It follows that all closed balls are homeomorphic and therefore fail to be compact.

  • 4
    Ok, thanks, I did not see that compactness is a topological property, so it is preserved under homeomorphisms. – PianoEntropy Dec 07 '12 at 14:47
  • 4
    @Yiteng That's why topology is so interesting. A ton of properties we study in metric spaces actually depend only on the induced topology: convergence, compactness, (path-)connectedness, dense sets, separability, hausdorffness (You'll lose that when allowing prenorms)… Because of that, it makes sense to differentiate between the category of normed spaces with contractions (isos are isometries) and with bounded operators (isos are homeomorphisms) as morphisms. – Lukas Juhrich Nov 28 '19 at 10:11
9

Is there any way to adapt this lemma or make use of a different construction in order to say something about all the closed balls?

Yes. Consider the proof of Theorem 1.22 in Rudin's book, according to which Every locally compact topological vector space $X$ has finite dimension. If $X$ is a normed space and we consider the particular neighborhood $V=B(0,1)$ then such a proof takes this simpler form. By rescaling and translating, we get the argument below.

Claim. Any closed ball is non-compact in an infinite dimensional space.

Proof: Let $X$ be a normed space and $B$ an arbitrary closed ball in $X$, say with center $a$ and radius $r$.

Assume that $B$ is compact.

As $\{B(x;\frac{r}{2})\}_{x\in B}$ is an open cover of $B$, there are $x_1,...,x_n\in B$ such that $$B\subset \bigcup_{i=1}^n B(x_i;\tfrac{r}{2}).$$ As $$B(x_i;\tfrac{r}{2})= x_i-a+B(a;\tfrac{r}{2})=x_i-\tfrac{a}{2}+\tfrac{1}{2}B(a;r)$$ it follows that $$B(a;r)\subset B\subset \bigcup_{i=1}^n \big(x_i-\tfrac{a}{2}+\tfrac{1}{2}B(a;r)\big)\subset Y+\tfrac{1}{2}B(a;r),$$ where $Y=\text{span}\{a,x_1,...,x_n\}$. Thus \begin{align*} B(a;r)\subset Y+\tfrac{1}{2}[Y+\tfrac{1}{2}B(a;r)]&=Y+\tfrac{1}{2^2} B(a;r)\\ &\subset Y+\tfrac{1}{2^2} [Y+\tfrac{1}{2}B(a;r)]=Y+\tfrac{1}{2^3}B(a;r). \end{align*} In general, $$B(a;r)\subset Y+\tfrac{1}{2^n} B(a;r)=Y+B\big(a,\tfrac{r}{2^n}\big),\quad \forall\ n\in\mathbb N.$$ So, each $x\in B(a;r)$ has the form $x=y_n+x_n$ with $y_n\in Y$ and $x_n\in B\big (a;\tfrac{r}{2^n}\big)$. As $x_n\to a$, $y_n\to x-a$ and thus $x-a\in \overline{Y}=Y$. As $a\in Y$, it follows that $x=(x-a)+a\in Y$. This shows that $B(a;r)\subset Y$ which implies that $X\subset Y$ and thus $$\dim X\leq\dim Y=n+1<\infty.$$ Therefore if $\dim X=\infty$ then $B$ is not compact. $\square$

Pedro
  • 19,965
  • 9
  • 70
  • 138
  • The problem that I corrected was that $frac{1}{2}B(a;r) \neq B(a;{r \over 2})$ – davik May 12 '17 at 07:21
  • @davik But $B(a;\tfrac{r}{2})=\tfrac{1}{2}B(a;r)+\tfrac{a}{2}$ and this corrects the old proof, right? – Pedro May 12 '17 at 13:21
  • I tried to change it that way but I don't think the extra complexity is justified given we can shift in the beginning – davik May 12 '17 at 13:25
  • As it is, the in general line is incorrect – davik May 12 '17 at 13:27
  • @davik WIth respect to the complexity, I agree with you. But this is a requirement of the question. By "a different construction in order to say something about all the closed balls" I understand "an argument that does not use the fact that the ball has center zero and radius one". Also, the simplified case is linked in the post. With respect to the said line, I can't see the mistake. As $a\in Y$ we have $Y+\tfrac{1}{2^n}B(a;r)=Y+(\tfrac{1}{2^n}-1)a+B(a;\tfrac{r}{2^n})=Y+B(a;\tfrac{r}{2^n})$. – Pedro May 12 '17 at 13:50
  • Yes, sorry you are correct – davik May 12 '17 at 14:40