1
$\begingroup$

$\forall \beta \in I$, $A_{\beta }$ is connected, and $\left ( \bigcup_{\alpha < \beta }A_{\alpha } \right )\cap A_{\beta }\neq \varnothing$ . Is $\bigcup_{\alpha \in I}A_{\alpha } $connected? For the index set $I$ , when it is countable ,the answer is obvious. I want to know the general conclusion. Thanks!

$\endgroup$
6
  • $\begingroup$ Some details seem to be missing. Are the sets $A_\alpha$ subsets of a Euclidean (or another fixed topological) space? You seem to have an order on the index set, is it well-ordered or of some other special type? $\endgroup$ May 5, 2015 at 10:12
  • $\begingroup$ It is an open problem. You can add some conditions $\endgroup$
    – 4869
    May 5, 2015 at 11:14
  • $\begingroup$ And the order on the index set should not be well-ordered,or it is obvious as the countable one. $\endgroup$
    – 4869
    May 5, 2015 at 11:16
  • $\begingroup$ The hypothesis as stated is impossible for a well-ordered set, since if $\beta$ is the least element, then $\bigcup_{\alpha<\beta}A_\alpha$ is itself empty. $\endgroup$ May 5, 2015 at 12:00
  • $\begingroup$ @JoelDavidHamkins : And if you drop the assumption when $\beta$ is the least element ? Like when you require $\left ( \bigcup_{k < n }A_k \right )\cap A_n\neq \emptyset$ for $n \geq 1$ instead of $n \geq 0$ ? (The whole family is $(A_n)_{n\in \mathbb{N}}$.) $\endgroup$
    – Hachino
    May 5, 2015 at 12:18

1 Answer 1

6
$\begingroup$

I assume that you intend $I$ is linearly ordered by $<$. The hypothesis is impossible when $I$ has a least element $\beta$ — in particular, it is impossible when $I$ is nonempty and well-ordered — since if $\beta$ is the least element, then $\bigcup_{\alpha<\beta}A_\alpha$ is empty itself. But if we insist on that requirement only when $\beta$ is not the least element of $I$, then your conclusion holds if and only if $I$ is well-ordered (or nonempty and well-ordered, depending on whether you consider $\emptyset$ to be connected).

Theorem. The following are equivalent for any linearly ordered set $\langle I,<\rangle$.

  1. $\langle I,<\rangle$ is a nonempty well-order.

  2. Whenever $X$ is a topological space and $A_\alpha\subset X$ is a connected subspace for every $\alpha\in I$ and whenever $\beta\in I$ is not least, then $\bigcup_{\alpha<\beta}A_\alpha\cap A_\beta\neq\emptyset$, then $\bigcup_{\alpha\in I}A_\alpha$ is connected.

Proof. ($1\to 2$) Assume $I$ is nonempty and well-ordered by $<$. We prove that $\bigcup_{\alpha\in I}A_\alpha$ is connected by transfinite induction on the order type of $I$. Suppose that $U\sqcup V$ is a nontrivial open separation of $\bigcup_{\alpha\in I}A_\alpha$. Let $\beta$ be least such that $\bigcup_{\alpha\leq\beta}A_\alpha$ has points from both $U$ and $V$. But this union is equal to $\bigcup_{\alpha<\beta}A_\alpha\cup A_\beta$, which by induction is the union of two overlapping connected subspaces, and hence is connected. So it cannot have points from both sides of the separation, a contradiction. So there is no nontrivial open separation of $\bigcup_{\alpha\in I}A_\alpha$, and so it is connected.

($2\to 1$) We prove the contrapositive. For the main case, suppose that $I$ is not a well-order. Thus, it has a strictly descending sequence $\alpha_0>\alpha_1>\alpha_2>\cdots$ and so on. If $\alpha$ is below all the $\alpha_n$, then let $A_\alpha=\{0\}$ have just the point $0$. Otherwise, $\alpha_n\leq \alpha$ for some smallest $n$, and in this case, we let $A_\alpha$ be $\{0\}$, if $n$ is even, and otherwise $\{1\}$. So each $A_\alpha$ is connected, and furthermore, for every $\beta\in I$ (except the least element, if there is one), we have $\bigcup_{\alpha<\beta}A_\alpha\cap A_\beta\neq\emptyset$, since below the descending sequence, both of these sets are just $\{0\}$, and above any point in the descending sequence, the union set is $\{0,1\}$, which meets $A_\beta$. Meanwhile, the union $\bigcup_{\alpha\in I}A_\alpha$ is $\{0,1\}$, which is not connected, by design. Lastly, if $I$ is empty, then the hypothesis of 2 is vacuous, but the union set is empty, which is technically disconnected. QED

$\endgroup$
0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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