# Homeomorphism between derived sets implies homeomorphism

(Note: I asked this question at MSE days ago and received no answer, so I'm now reposting it here.)

I want to prove the following statement:

Let $K_1$ and $K_2$ be two countable, compact sets of $\mathbb R$ such that $K_1'$ and $K_2'$ are homeomorphic, then $K_1$ and $K_2$ are homeomorphic.

More generally, I want to prove that if $K_1$ and $K_2$ be two countable, compact sets of $\mathbb R$ have the same Cantor–Bendixson rank, $\alpha+1$, and $|K_1^{(\alpha)}|=|K_2^{(\alpha)}|$, then $K_1$ and $K_2$ are homeomorphic.

I know that a proof is given by Mazurkiewicz and Sierpinski in "Contribution à la topologie des ensembles dénombrables" (1920) (http://matwbn.icm.edu.pl/ksiazki/fm/fm1/fm114.pdf). And by Millient in "A remark in Cantor derivative" (http://arxiv.org/pdf/1104.0287v1.pdf). But I want a more direct proof in $\mathbb{R}$; can anybody help me? Thanks.

• As a consequence of Mazurkiewicz-Sierpinski theorem, any countable compact space is homeomorphic to some $\omega^{\alpha} \cdot n +1$, where $\alpha$ is a countable ordinal and $n \geq 1$; in particular, every such space is homeomorphic to a subspace of $\mathbb{R}$. Therefore, it cannot be simpler to prove Mazurkiewicz-Sierpinsky theorem restricted to subspaces of $\mathbb{R}$. Jul 1 '14 at 7:11
• @Seirios : I'm not sure this conclusion is inevitable (of course, this is an ill-defined philosophical question). The fact that MS for $\mathbb R$ implies MS, but by MS itself, does not exclude the possibility that MS for $\mathbb R$ might be "easier". Or here's a (very silly) analogy: The fact that every finite dimensional vector space $V$ has a basis implies that $V\cong F^n$. However, it's still "easier" to find a basis for $F^n$ than for $V$. Jul 1 '14 at 17:22
• @ChristianRemling: Right, so in fact it depends on the difficulty to prove that a compact countable space is homeomorphic to a subspace of $\mathbb{R}$; fortunately, Joseph Van Name gave a simple argument in his answer. Therefore, the two problems are roughly the same. Jul 2 '14 at 7:15
• By the way, I find the original proof of Mazurkiewicz and Sierpinsky already simple. Their article is written in French, but I wrote a part of their argument here: chiasme.wordpress.com/2013/06/19/… Jul 2 '14 at 7:17
• Thanks. I think that using the ordinal topology is a bit complicated. I found another proof that seems simpler (math.stackexchange.com/questions/853500/…), but there are some parts that I don't understand, can you help me please? Jul 2 '14 at 15:35

Let me give a proof that makes use of the ordering inherited from $\mathbb{R}$.

Suppose that $A\subseteq\mathbb{R}$ is a countable compact space. Then $A$ has no subset $B$ order isomorphic to the rational numbers (otherwise $A$ would be uncountable). I now claim that the order topology on $A$ is isomorphic to the subspace topology on $A$. To avoid confusion, let $\mathcal{O}$ be the order topology on $A$ and let $\mathcal{S}$ be the subspace topology on $A$. It is well known that for every ordered set, the subspace topology is finer than the order topology, i.e. $\mathcal{O}\subseteq\mathcal{S}$. However, since $(A,\mathcal{S})$ is compact, we have $\mathcal{O}=\mathcal{S}$, so the order topology and the subspace topology coincide on $A$. In particular, since $A$ is complete as an ordered set, $A$ is a complete lattice by this answer of mine.

Let $\mathcal{H}_{0}$ be the set of all finite and countable successor ordinals. Let $\mathcal{H}_{\alpha}$ be the collection of all sums $\sum_{i\in B}B_{i}$ where $B_{i}\in\bigcup_{\beta<\alpha}\mathcal{H}_{\beta}$ for each $\beta<\alpha$ and where $B=\omega+1$,$B=(\omega+1)^{*}$ or $B$ is finite.

I claim that $\bigcup_{\alpha<\omega_{1}}\mathcal{H}_{\alpha}$ is precisely the collection of all countable complete linear orders (this is a slight modification of a result by Hausdorff proven here). Suppose that $X$ is a countable compact linear ordering that is not in $\bigcup_{\alpha<\omega_{1}}\mathcal{H}_{\alpha}$. Then it is easy to show that there exists some $x\in X$ with $[0,x_{1/2}],[x_{1/2},1]\not\in\bigcup_{\alpha<\omega_{1}}\mathcal{H}_{\alpha}$. In other words, we can cut $X$ into two pieces not in $\bigcup_{\alpha<\omega_{1}}\mathcal{H}_{\alpha}$. Now cut each piece $[0,x_{1/2}],[x_{1/2},1]$ into pieces $[0,x_{1/4}],[x_{1/4},x_{1/2}],[x_{1/2},x_{3/4}],[x_{3/4},1]$. Continue this process until we have obtained an $x_{r/2^{n}}$ for all $n\in\omega$ and $0<r<\frac{1}{2^{n}}$ and we have obtained an isomorphic copy of the rational numbers. This is a contradiction since $X$ cannot contain a copy of the rational numbers.

I claim that for each ordinal $\alpha$, each $X\in\mathcal{H}_{\alpha}$ is homeomorphic to some countable ordinal, and this claim shall be proven by transfinite induction. Suppose that $\alpha$ is an ordinal and assume that whenever $\beta<\alpha$ each $X\in\mathcal{H}_{\beta}$ is homeomorphic to some ordinal. Now assume that $X\in\mathcal{H}_{\alpha}$. If $X=\sum_{i\in B}B_{i}$ where $B$ is finite and for each $i\in B$ we have $B_{i}\in\mathcal{H}_{\beta_{i}}$ for some $\beta_{i}<\alpha$, then clearly $X$ is homeomorphic to some ordinal. Therefore without loss of generality, we can assume that $X=\sum_{\alpha\in\omega+1}B_{\alpha}$. Then let $W_{n}$ be a well ordered set and let $h_{n}:B_{n}\rightarrow W_{n}$ be a homeomorphism for all $n\in\omega$.

Let $0_{\omega}$ be the least element in $B_{\alpha}$.

If $0_{\omega}$ is not a limit point of $B_{\alpha}$, then since $B_{\alpha}$ is homeomorphic to a well ordered set, one can easily show that there exists an isomorphism $h_{\omega}:0_{\omega}\rightarrow W_{\omega}$ such that $0_{\omega}$ is the least element of $W_{\omega}$. Then define a mapping $h:\sum_{\alpha\in\omega+1}B_{\alpha}\rightarrow \sum_{\alpha\in\omega+1}W_{\alpha}$ by letting $h(x,\alpha)=(h_{\alpha}(x),\alpha)$. Then $H$ can easily be seen to be a homeomorphism.

Now assume that $0_{\omega}$ is a limit point of $B_{\alpha}$. Then let $(x_{n})_{n\in\omega}$ be a descending sequence in $B_{\omega}$ that converges to $0_{\omega}$ and such that $x_{0}$ is the greatest element of $B_{\omega}$ and where if $n\neq 0$, then $x_{n}$ has an immediate successor $y_{n-1}$.

For all $n\in\omega$, let $Y_{n}$ be a well ordered set and let $j_{n}:[y_{n},x_{n}]\rightarrow Y_{n}$ be an homeomorphism.

Let $Z_{2n}=W_{n}$ and let $Z_{2n+1}=Y_{n}$ for each $n\in\mathbb{N}$, and let $Z_{\omega}=\{0^{\omega}\}$ for some element $0^{\omega}$. Let

$k:\sum_{\alpha\in\omega+1}B_{\alpha}\rightarrow\sum_{\alpha\in\omega+1}Z_{\alpha}$ be the mapping where $k(x,n)=(h_{n}(x),2n)$ for $n\in\omega$, $k(x,\omega)=(j_{n}(x),2n+1)$ for $x\in[y_{n},x_{n}]$ and $k(0_{\omega},\omega)=(0^{\omega},\omega)$. Then $k$ is a homeomorphism. Therefore each countable compact subspace $A\subseteq\mathbb{R}$ is homeomorphic to some successor ordinal.

Suppose that $\omega^{\alpha_{n}}a_{n}+...+\omega^{\alpha_{0}}a_{0}$ is an ordinal in Cantor's normal form. Then by putting the initial segment $\omega^{\alpha_{n}}a_{n}+1$ last instead of first, we conclude that $\omega^{\alpha_{n}}a_{n}+...+\omega^{\alpha_{0}}a_{0}$ is homeomorphic to $\omega^{\alpha_{n}}a_{n}+1$. In particular, if two countable compact sets $X,Y$ have the same Cantor-Bendixson rank $\alpha+1$ and $|X^{(\alpha)}|=|Y^{(\alpha)}|$, then $X\simeq Y$.

Also take note that it is easy to prove that every compact countable metric space is isomorphic to a subspace of $\mathbb{R}$. If $X$ is a countable compact space, then $C(X)$ is a Banach space. However, whenever $x,y\in X$ are distinct elements, then $A_{x,y}\subseteq\{f\in C(X)|f(x)=f(y)\}$ is a closed nowhere dense subspace of $C(X)$. Therefore by the Baire category theorem, $\bigcup_{x\neq y}A_{x,y}$ contains no open subset of $C(X)$ and if $f\in C(X)\setminus\bigcup_{x\neq y}A_{x,y}$, then $f$ is injective. Since $X$ is compact, the set $X$ and its image $f[X]$ are homeomorphic.

• I find your proof that a compact countable metric space is homeomorphic to a subspace of $\mathbb{R}$ very nice. A possibility to prove that a compact countable metric space is metrizable is to introduce $$e : \left\{ \begin{array}{ccc} X & \to & \mathbb{R}^A \\ x & \mapsto & (f(x)) \end{array} \right.$$ where $A$ is the set of continuous functions $X \to \mathbb{R}$. It is an embedding, and because $X$ is countable and compact, we may deduce an embedding into $[0,1]^{\omega}$, which is metrizable. Jul 2 '14 at 7:24
• By the way, I find the original argument of Mazurkiewicz and Sierpinski more elementary. Jul 2 '14 at 7:25
• @Seirios In your post, why do we need the sequence $(p_k)$ in $X^{(1)}$ converging to $p$? Jul 3 '14 at 13:01
• The sequence is just useful to exhibit the "onion structure"; then, the induction hypothesis may be applied to any connected component. Perhaps a more appropriate place to deal with such questions would be my blog itself; you can write your questions as comments. Jul 3 '14 at 21:02