Let $X$ be a locally compact, metrizable, path-connected space which is not compact. Is the one-point compactification of $X$ also path-connected?
-
2No. Specifically: "The extended long ray $L^$ is compact. It is the one-point compactification of the closed long ray $L$ (...). $L^$ is also connected, but not path-connected." – Aloizio Macedo May 29 '18 at 06:13
-
@AloizioMacedo Thanks for the pointer. I think I'm more interested in the metrizable case, so I'll edit the question. – Cihan May 29 '18 at 06:57
-
It definitely is true if $X^*=X\sqcup{\infty}$ is locally path connected around the $\infty$ point. Does path connectedness of $X$ imply this? I'm not sure, probably not. Hmm... – freakish May 29 '18 at 09:08
-
2A very similar, but not equivalent question was asked here: https://math.stackexchange.com/q/1869794/3457 – Niels J. Diepeveen May 29 '18 at 11:13
-
You might want $X$ to be separable (or equivalently, second countable) as well as otherwise the one-point compactification is not metrisable, and I suppose you want to stay within metrisable spaces? – Henno Brandsma May 29 '18 at 17:54
2 Answers
The answer is "no"'. We shall construct a path-connected closed $X \subset \mathbb{R}^3$ (which is a fortiori locally compact and $\sigma$-compact) such that the one-point compactification is not path-connected.
Let $$\Sigma = \lbrace (x,\sin(1/x)) \mid x \in (0,1] \rbrace \subset \mathbb{R}^2 ,$$ $$S = \lbrace 0 \rbrace \times [-1,1] \cup \Sigma \subset \mathbb{R}^2,$$ $$S_n = [0,2^{-n}] \times [-1,1] \cup \Sigma \subset \mathbb{R}^2, n \in \mathbb{N}.$$ $S$ is the closed topologist's sine curve which has the two path-components $\lbrace 0 \rbrace \times [-1,1]$ and $\Sigma$. The $S_n$ are path-connected. Define $$T_n = S \times \lbrace 0 \rbrace \cup \bigcup_{i=n}^\infty S_n \times \lbrace 2^{-n} \rbrace\subset \mathbb{R}^3 .$$ All $T_n$ are compact. Next define translations
$$t_n : \mathbb{R}^3 \to \mathbb{R}^3, t_n(x_1,x_2,x_3) = (n + x_1,x_2,x_3)$$ and set $$T = \bigcup_{n=1}^\infty t_{n-1}(T_n) .$$
This is a closed subspace of $\mathbb{R}^3$ which has infinitely many path components. Its intersection with $\mathbb{R}^2 \times \lbrace 2^{-n} \rbrace$ is $$R_n = \bigcup_{i=1}^n t_{i-1}(S_n \times \lbrace 2^{-n} \rbrace)$$ which is path-connected. Its intersection with $\mathbb{R}^2 \times \lbrace 0 \rbrace$ is $$R = \bigcup_{i=1}^\infty t_{i-1}(S \times \lbrace 0 \rbrace)$$ which splits into infinitely many path components (these are $C_0 = \lbrace 0 \rbrace \times [-1,1] \times \lbrace 0 \rbrace$ and $C_n = t_{n-1}((\Sigma \cup (\lbrace 1 \rbrace \times [-1,1])) \times \lbrace 0 \rbrace)$).
Define $$K_0 = \lbrace (0,0) \rbrace \times [0,1/2],$$ $$K_n = \lbrace n \rbrace \times [1,2] \times \lbrace 0, 2^{-n} \rbrace \cup \lbrace (n,2) \rbrace \times [0,2^{-n}], n \in \mathbb{N}.$$ Then $$K = K_0 \cup \bigcup_{n=1}^\infty K_n$$ is closed in $\mathbb{R}^3$ and $$X = T \cup K$$ is a closed path-connected subset of $\mathbb{R}^3$ ($K_0$ connects $C_0$ and the $R_n$, $K_n$ connects $C_n$ and $R_n$). Let $$A_n = X \cap (\lbrace n \rbrace \times \mathbb{R}^2).$$ Let $X^+ = X \cup \lbrace \infty \rbrace$ be the one-point compactification of $X$. Any path in $X^+$ from $A_0$ to $\infty$ must go through all $A_n$ since $X^+ \backslash A_n$ is not connected.
Assume $X^+$ is pathwise connected. Let $u : I \to X^+$ be a path with $u(0) \in A_0$ and $u(1) = \infty$. Let $a = sup \lbrace t \in I \mid u(t) \in A_0 \rbrace$ and $b = inf \lbrace t \in I \mid u(t) = \infty \rbrace$. Then $u(a) \in A_0, u(b) = \infty$ and $u((a,b)) \subset X \backslash A_0$. Moreover, $u((a,b))$ must intersect all $A_n$ with $n \ge 1$.
On the other hand, $u((a,b))$ must be contained in a path component of $X \backslash A_0$. But none of these intersects all $A_n$ which is a contradiction.

- 76,394
- 12
- 43
- 125
-
Nice example. Next you may want to consider the problem here: https://math.stackexchange.com/questions/2020740/equivalence-relation-on-a-continuum. It is a major unsolved problem. Unfortunately your space here does not have the properties it asks for; the point at $\infty$ is contained in the nowhere dense continuum $X\cap (\mathbb R ^2\times{0})$. It is tempting to mimic your construction with a Solenoid in place of your $S$, so that the covergence continuum at the $0$-level only goes through certain composants of $S$. But then there is the problem of path-connecting the other composants. – Forever Mozart May 31 '18 at 02:29
-
What I mean in my last comment is that you could fatten the lines to include plane regions. Then $X$ would be contained in an equivalence class of $X\cup {\infty}$, in the sense of my linked post. But the goal is to make $\infty$ be in an eq. class of its own. This problem has haunted me for years. – Forever Mozart May 31 '18 at 02:57
-
@ForeverMozart I looked at the problem - seems in fact to be very difficult! I can generalize my construction, perhaps it may be useful. Since there is not enough space in comments, I give another answer. – Paul Frost May 31 '18 at 11:02
My example can be generalized as follows. Let $Q$ denote the Hilbert cube, $Q' = [0,\infty) \times Q$, $Q_n = [n,n+1] \times Q$, $n \ge 0$, and $a_n = (n,q_0)$ with $q_0 = (0,0,0,...) \in Q$.
Take any sequence of connected but not path-connected compact metric spaces $X_n$ together with points $x_n \in X_n$ and closed $Y_n \subset X_n$ such that $x_n \notin Y_n$ and $\lbrace x_n \rbrace \cup Y_n$ contains at least one point from each path component of $X_n$. It may happen that already $Y_n$ has this property, but we do not exclude the case that $x_n$ is not contained in any path component meeting $Y_n$.
Via a suitable embedding we may assume that the $X_n$ are subsets of $Q_n$ with the following properties:
(1) $x_n = a_n$
(2) $X_n \cap (\lbrace n \rbrace \times Q) = \lbrace a_n \rbrace$
(3) $a_{n+1} \in X_n$
(4) $a_n$ and $a_{n+1}$ belong to distinct path components of $X_n$.
We can write $X_n = \bigcap_{i=0}^\infty X_n^i$ with a sequence of path-connected closed $X_n^i \subset Q_n$ such that $X_n^{i+1} \subset int(X_n^i)$. They all contain $a_n, a_{n+1}$. We define subspaces of $Q'' = I \times Q' \times I$ as follows (brackets $\lbrace$ and $\rbrace$ are omitted for sets containing only a single point):
$$T_n = 0 \times X_n \times 0 \cup \bigcup_{i=n}^\infty 2^{-i} \times X_n^i \times 0 \subset I \times Q' \times 0 .$$
These are compact and we set
$$T = \bigcup_{n=0}^\infty T_n $$
which is closed in $Q''$ and has infinitely many path components. Its intersection with $2^{-m} \times Q' \times 0$ is $$R_m = \bigcup_{n=0}^m 2^{-m} \times X_n^m \times 0$$ which is path-connected. Its intersection with $0 \times Q' \times 0$ is $$R = \bigcup_{n=0}^\infty 0 \times X_n \times 0$$ which splits into infinitely many path components.
The set $$K_{-1} = I \times (a_0,0)$$ connects all $R_m$. To connect the path components of $R$ with $T \backslash R$ we introduce
$$K_n = 2^{-n} \times Y_n \times I \cup [0,2^{-n}] \times Y_n \times 1 \cup 0 \times Y_n \times I $$ They are pairwise disjoint since $Y_n \subset (n,+1] \times Q$. The set $$K = \bigcup_{n=0}^\infty K_n$$ is closed in $Q''$ and $$X = K_{-1} \cup K \cup T$$ is a closed path-connected subset of $Q''$. This is due to the fact that each path component $P$ of $X_n$ either meets $Y_n$ or contains $a_n$. Then $0 \times P \times 0$ is connected to $R_n$ via $K_n$ in the first case and to $R_{n-1}$ via $K_{n-1}$ in the second case since $a_n$ is contained in a path component $P'$ of $X_{n-1}$ meeting $Y_{n-1}$ (so that $P \cup P'$ is path-connected).
Define $A_n = X \cap (n \times Q \times I)$. Then the rest of the proof goes as in my first answer.
Of course we can replace $Q$ by any other compact metric space in which case only very special $X_n$ will be allowed.

- 76,394
- 12
- 43
- 125