2

I came across the following theorem:

Let $M$ be a metric space. $M$ is compact if and only $M$ is the continuous image of the cantor set. (see reference)

An immediate consequence of this theorem is that any compact metric space has cardinality no larger than $\mathbb{R}$. This seems wrong to me. Consider the space of equibounded functions in $L_1[0,1]$. That is, take $M > 0$, and let $$ X = \big{\{} f \in L_1[0,1] : |f(x)| \leq M \ \text{for all } x \in [0,1] \big{\}} $$

I have the following observations:

  1. $X$ is obviously a metric space with the $L_1$ norm.
  2. $X$ is obviously totally bounded.
  3. $X$ obviously inherits completeness from $L_1[0,1]$.
  4. $X$ obviously inherits its cardinality $\# X = 2^\mathbb{R}$ from $L_1[0,1]$.

One of the above four observations must be wrong, or else we have violated the above theorem. An alternative explanation would be that theorem is wrong and the proof presented in Willard's General Topology makes an implicit assumption on the cardinality of the metric space.

Overall I am least confident in my second observation. While my intuition tells me this set is totally bounded, this could be an example of a space that is bounded but not totally bounded.

Any thoughts on which of these observations is/are wrong would be appreciated. Thanks!

  • ... has cardinality not larger than that of $\mathbb{R}$. For example, a singleton is compact and metric with the only possible metric, it is the continuous image of the Cantor set by a constant map. – blueInk Mar 21 '18 at 01:51
  • Why is $X$ not of cardinality $2^\mathbb{R}$ though? – OJdidntdoitlol Mar 21 '18 at 01:55
  • I meant the conclusion that you drew immediately after quoting the theorem. – blueInk Mar 21 '18 at 01:56
  • Oh oops I mean no larger than $\mathbb{R}$, not the same cardinality. I edited my post. – OJdidntdoitlol Mar 21 '18 at 01:57
  • The other thing is that you can't apply the theorem without the condition of compactness. 1-4 hold. the one that fails is 5. compactness. – blueInk Mar 21 '18 at 02:06
  • @blueInk Doesn't complete and totally bounded imply compact in a metric space? – OJdidntdoitlol Mar 21 '18 at 02:09
  • Apply Riesz's lemma to the closed unit ball of $C[0,1]$. – blueInk Mar 21 '18 at 02:15
  • @OJdidntdoitlol Indeed, in metric spaces compactness is equivalent to completeness and total boundedness. – Math1000 Mar 21 '18 at 02:17
  • Now I notice that it is totally bounded the property you mentioned. Every time you were mentioning it in my mind I was reading uniformly bounded, from the way $X$ is defined and the obviously from where you concluded the **** boundedness. No, it is not totally bounded (Use Riesz lemma). – blueInk Mar 21 '18 at 02:24
  • I am rather curious what made you think total boundedness is "obvious"... – Eric Wofsey Mar 21 '18 at 04:23

2 Answers2

2

The space $L^1[0,1]$ does not have cardinality $2^{|\mathbb{R}|}$; it has cardinality $|\mathbb{R}|$. For instance, this is immediate from the fact that $L^1[0,1]$ is a separable metric space. Choosing a countable dense subset $D$, every element of $L^1[0,1]$ is the limit of some sequence in $D$, showing that $|L^1[0,1]|\leq|D^{\mathbb{N}}|=|\mathbb{R}|$.

(It is true that there are $2^{|\mathbb{R}|}$ Lebesgue-measurable functions $[0,1]\to\mathbb{R}$, but remember that elements of $L^1[0,1]$ are equivalence classes of such functions modulo equality on sets of measure $0$.)


More subtly, $X$ is not totally bounded either. Here's a nice way to prove this: let $f_n(x)$ be the $n$th digit of the binary expansion of $x\in[0,1]$. Then $f_n\in X$ (assuming $M\geq 1$). However, if $m\neq n$, then $\|f_n-f_m\|_1=1/2$: if you pick a number randomly from $[0,1]$, there is a $1/2$ chance its $m$th and $n$th binary digits are the same and a $1/2$ chance they are different, so $|f_n-f_m|=0$ on a set of measure $1/2$ and $|f_n-f_m|=1$ on its complement. So we have infinitely many elements of $X$ such that the distance between any two is $1/2$, so $X$ is not totally bounded.

Eric Wofsey
  • 330,363
0

Take $M=1.$ For distinct $m,n\in \mathbb N,$

$$\int_0^1|\sin (2\pi mx) - \sin (2\pi nx)|\,dx =2\int_0^1\frac{|\sin (2\pi mx) - \sin (2\pi nx)|}{2}\,dx$$ $$ \ge 2\int_0^1\left (\frac{\sin (2\pi mx) - \sin (2\pi nx))}{2}\right)^2\,dx = \frac{1}{2}\int_0^1(\sin^2 2\pi mx + \sin^2 2\pi n)\,dx = \frac{1}{2}.$$

The inequality follows because if $0\le f\le 1,$ then $\int_0^1 f\ge \int_0^1f^2.$ Squaring the integrand allows us to use the orthogonality of the functions $\sin (2\pi nx).$ Thus the sequence $\sin (2\pi nx)$ in $X$ cannot have a convergent subsequence, showing $X$ is not compact.

zhw.
  • 105,693