Suppose that we have an uncountable collection $C_\alpha$ of disjoint Cantor Sets contained in the closed unit interval $I$. Suppose we have ordered the indices $\alpha \in [0,1]$ as well. Then is it possible for the collection to satisfy the following property: If $x_n \in C_{\alpha_n}$ and $x_n \rightarrow y$ with $y \in C_\beta$ from the left then every accumulation point of $(\alpha_n)$ is contained in $[0, \beta]$ and similarly if they converge to $y$ from the right then every accumulation point of $(\alpha_n)$ is contained in $[\beta, 1]$.

In other words, if you converge to a point in one of the Cantor Sets $C_\beta$ from the left, then you can only do so via points coming out of Cantor Sets that are less than $\beta$ in the order and vice-versa. Can you order the Cantor Sets in this way that agrees with the index order 'up to the existence of limit points'?

This problem is coming from a general problem concerning admissible ensembles of continua in the plane. But this order theory aspect is very far removed from my expertise, what little I have haha So I need some help! My intuition is that it is impossible.



Such an enumeration does not exist. To derive a contradiction, take any family of Cantor sets $(C_\alpha)_{\alpha\in A}$ in $[0,1]$, indexed by an uncountable subset $A\subset[0,1]$.

Let $\mathcal K[0,1]$ be the space of non-empty compact subsets of $[0,1]$, endowed with the Hausdorff metric $d_H$. It is well-known that the metric space $\mathcal K[0,1]$ is compact and hence second-countable. Since second-countable spaces do not contain uncountable discrete subspaces, the uncountable set $\{(\alpha,C_\alpha):\alpha\in A\}\subset [0,1]\times\mathcal K[0,1]$ contains a non-isolated point $(\beta,C_\beta)$, which allows us to find a sequence $(\alpha_n)_{n\in\omega}$ in $A$ such that $\alpha_n\to\beta$ and $C_{\alpha_n}\to C_\beta$.

Using the Ramsey Theorem, we can find an increasing sequence $(n_k)_{k\in\omega}$ such that the sequence $(\alpha_{n_k})_{k\in\omega}$ is either strictly increasing or strictly decreasing. Replacing $(\alpha_n)$ by this subsequence, we can assume that the sequence $(\alpha_n)_{n\in\omega}$ is either increasing or decreasing.

First assume that the sequence $(\alpha_n)_{n\in\omega}$ is increasing. Let $x_\beta$ be the smallest element of the Cantor set $C_\beta\subset[0,1]$. Since $C_\beta$ has no isolated points, the Cantor set $C_\beta$ contains a strictly decreasing sequence $(y_n)_{n\in\omega}$ of real numbers that converges to the point $x_\beta$. For every $n\in\omega$ select a neighborhood $U_n\subset [0,1]$ of the point $y_n$ such that for any $n<m$ and $u\in U_n$, $v\in U_m$ we have $v<u$ and $|u-y_n|<1/2^n$. Since $C_{\alpha_n}\to C_\beta$, for every $k\in\omega$ there exists a number $n_k$ such that $U_k\cap C_{\alpha_n}\ne\emptyset$ for every $n\ge n_k$. Replacing each $n_k$ by a larger number, we can assume that the seqeunce $(n_k)_{k\in\omega}$ is strictly increasing. Then $U_k\cap C_{\alpha_{n_k}}\ne\emptyset$ for all $k\in\omega$.

Replacing the sequence $(\alpha_n)_{n\in\omega}$ by $(\alpha_{n_k})_{k\in\omega}$, we can assume that $U_k\cap C_{\alpha_k}\ne\emptyset$ for all $k\in\omega$. Then for every $k\in\omega$ we can choose a point $x_{\alpha_n}\in C_{\alpha_n}\cap U_n$. The choice of the sequence $(U_n)_{n\in\omega}$ ensures that the sequence $(x_{\alpha_n})_{n\in\omega}$ is strictly decreasing and converges to $x_\beta$ (whereas the sequence $(\alpha_n)_{n\in\omega}$ is increasing and converges to $\beta$. But this contradicts the choice of the enumeration $(C_\alpha)_{\alpha\in A}$.

By analogy we can consider the case of decreasing sequence $(\alpha_n)_{n\in\omega}$. In this case we should take $x_\beta$ to be the largest element of $C_\beta$.

| cite | improve this answer | |
  • $\begingroup$ Is $(\beta, \omega)$ the same as $(\beta, 1]$? When you say that "the following lemma implies," the implication would definitely prove the theorem. But if you take just one point $x_\alpha$ from each $ C_\alpha$, then what if we just let $\alpha = x_\alpha$? Then the lemma will be true, but left/right convergence will agree with the order. Maybe I am just not seeing how to go from the lemma to the result. I am really stumped! $\endgroup$ – John Samples Aug 7 '18 at 7:57
  • $\begingroup$ @JohnSamples $(\beta,\omega_1)$ is not the same as $(\beta,1]$: $\beta$ is just an index (a countable ordinal) - it is an element of $\omega_1$, which is not order embeddable into $[0,1]$. So, the notation $(\beta,1]$ is incorrect: $\beta$ and $1$ live in different sets. $\endgroup$ – Taras Banakh Aug 7 '18 at 21:45
  • 1
    $\begingroup$ @JohnSamples Aha! I have just realized that the set of your indices $\beta$ is a subset of $[0,1]$! I thought it is $\omega_1$. This changes the situation. Let me think on the problem a bit more. $\endgroup$ – Taras Banakh Aug 7 '18 at 21:49
  • $\begingroup$ Ok, thanks! I am trying to prove the other direction and see if I can construct such a family. Here are some possibly relevant results that MSE dug up: math.stackexchange.com/questions/2874786/… $\endgroup$ – John Samples Aug 8 '18 at 3:49
  • $\begingroup$ @JohnSamples I have rewritten my answer to fit the right interpretation of your question. $\endgroup$ – Taras Banakh Aug 8 '18 at 11:11

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.