covering a compact set $X$ with two closed balls, and covering each of these sets with two closed balls, etc so the diameters go to zero

105 Views Asked by At

I'm working on proving that every compact metric space is the continuous image of the Cantor set (we use the definition of the Cantor set being a set homeomorphic to $\{0,1\}^{\mathbb{N}}$). The question of proving this has been asked several times here on stackexchange, so hopefully this doesn't get flagged; my question is more on justifying a technical point.

Attempt

I understand the general form (roughly) of the proof is as follows: Take the compact set $X$ and cover it with two closed balls $U_{0},U_{1}$. Then take $U_1$ and $U_2$ and cover each with two closed balls $U_{0,0},U_{0,1}$, and $U_{1,0}$, $U_{1,1}$, respectively. By replacing $U_{0,0}$ and $U_{0,1}$ with $U_{0}\cap U_{0,0}$ and $U_{0}\cap U_{0,1}$, respectively, we can assume that $U_{0,0}$ and $U_{0,1}$ are contained in $U_{0}$; for the same reasons we can also assume that $U_{1,0},U_{1,1}$ are contained in $U_{1}$. Continuing this process inductively yields sets $$\mathscr{U}:=\big\{U_{t}\,:\,t\in\{0,1\}^{\mathbb{N}}\big\}.$$ Now each set $U_t$ is essentially constructed as the infinite intersection of nested compact sets, so if I could further rig up the construction to have these nested sets decrease to zero in diameter, each $U_{t}\in\mathscr{U}$ is a singleton, so the function $f:\{0,1\}^{\mathbb{N}}\to X$ given by $f(t)=U_t$ is well-defined. From here it's not hard to show that this map is continuous and surjective, which completes the proof.

Questions

I'm trying to come up with a way (if it's even possible) to cover up a compact set by two closed balls, then cover up each of those closed with two more closed balls, etc in such a way that the diameters of these coverings decrease to zero. I suppose my conjecture is better stated more generally and formally is as follows:

Conjecture: Let $X$ be a nonempty compact metric space. There exists a family of compact sets $\big\{U_t\,:\,t\in\{0,1\}^{n},n\in\mathbb{N}\big\}$ such that $$ X=U_{0}\cup U_{1}, $$ and in general $$ U_t=U_{\{t,0\}}\cup U_{\{t,1\}} $$ for each $t\in\{0,1\}^n$. Moreover $$ \lim_{n\to\infty}\max_{t\in\{0,1\}^n}\text{diam}(U_t)=0. $$

Since $X$ is compact for each $n$ one could cover $X$ with finitely many closed balls $\mathscr{U}_n=\{U_{n,1},\ldots U_{n,k_n}\}$ of diameter say $2^{-n}$ so that $\mathscr{U}_{n+1}$ is a refinement of $\mathscr{U}_n$ for each $n$, but is there a way to alter this construction to where each refinement adds to more balls for each set?

This is part of a homework exercise so please only hints. Thanks in advance. Any help is greatly appreciated.

1

There are 1 best solutions below

4
On BEST ANSWER

Hints: In the general case, I don't think it is wise to aim for precisely 2 balls in each refinement. Instead going from level n-1 to n you look at the max number, say $M$, of closed balls of radius $2^{-n}$ you need to cover a ball from level n-1. Then find $m$ so that $M\leq 2^m$. Now instead of using 1 binary symbol to describe the refinement you use m binary symbols. So your symbol sequence will consists of 'trunks' of binary symbols describing the nesting.

An example: Suppose you cover $X$ first by 3 balls, $U_0,U_1,U_2$. This may be encoded by 2 bits (as $2^2=4\geq 3$). So we map the 'cylinders' $[00],[01],[10]$ to $U_0$,$U_1$,$U_2$, respectively. The remaining $[11]$ you map to whatever you want, e.g. some arbitrary but fixed point $\xi\in X$. In the next level suppose you may cover with maximally 5 balls, so e.g. $U_1\subset U_{1,0},...,U_{1,4}$. You may encode all possibilities by 3 bits, so as before map e.g. $[01 \ 010]$ to $U_1\cap U_{1,2}$ (if non-empty). You create this way a map from $\{0,1\}^5=\{0,1\}^2\times \{0,1\}^3$ to the collection of intersection of balls at this level. Again for the leftover symbols (or if an intersection is empty) you map the cylinder to your chosen point $\xi$. Continuing this way, using compactness, you obtain the wanted continuous surjective map from $\{0,1\}^{\Bbb N}$ to $X$.

The above is not the most economical way of coding. A better, but for which more book-keeping is needed, is to let the number m depend on each ball at level n-1. But your binary tree at a given level $n$ will then not have uniform length in each branch.