Skip to main content

Section Path Connectedness and Connectedness in Infinite Topological Spaces

Given that connectedness and path connectedness are equivalent in finite topological spaces, a reasonable question now is whether the converse of Theorem 19.10 is true in arbitrary topological spaces. As we will see, the answer is no. To find a counterexample, we need to look in an infinite topological space. There are many examples, but a standard example to consider is the topologist’s sine curve. This curve \(S\) is defined as the union of the sets
\begin{equation*} S_1 = \{(0,y) \mid -1 \leq y \leq 1\} \ \text{ and } \ S_2 = \left\{ \left(x,\sin\left(\frac{1}{x}\right)\right) \Big| \ 0 \lt x \leq 1\right\}\text{.} \end{equation*}
A picture of \(S\) is shown in Figure 19.16.
Figure 19.16. The topologist’s sine curve.
To understand if \(S\) is connected, let us consider the relationship between \(S\) and \(S_2\text{.}\) Figure 19.16 seems to indicate that \(S = \overline{S_2}\text{.}\) To see if this is true, let \(q=(0,y) \in S_1\text{,}\) and let \(N\) be a neighborhood of \(q\text{.}\) Then there is an \(\epsilon \gt 0\) such that \(B = B(q, \epsilon) \subseteq N\text{.}\) Choose \(K \in \Z^+\) such that \(\frac{1}{\arcsin(y)+2 \pi K} \lt \epsilon\text{,}\) and let \(z = \frac{1}{\arcsin(y)+2 \pi K}\text{.}\) Then
\begin{align*} d_E\left(q,\left(z, \sin\left(\frac{1}{z}\right)\right)\right) \amp = d_E\left((0,y),(z, \sin(\arcsin(y)+2 \pi K)\right)\\ \amp = d_E((0,y), (z,\sin(\arcsin(y))))\\ \amp = d_E((0,y), (z,y))\\ \amp = \la z \ra\\ \amp \lt \epsilon\text{,} \end{align*}
and so \(\left(z, \arcsin(z)\right) \in B(q, \epsilon)\) and every neighborhood of \(q\) contains a point in \(S_2\text{.}\) Therefore, \(S_1 \subseteq S_2' \subseteq \overline{S_2}\) and \(\overline{S_2} = S\) in \(S\text{.}\) The fact that \(S\) is connected follows from Theorem 18.8.
Now that we know that \(S\) is connected, the following theorem demonstrates that \(S\) is a connected space that is not path connected.

Proof.

We know that \(S\) is connected, so it remains to show that \(S\) is not path connected. The sets \(S_1\) and \(S_2\) are connected (as continuous images of the interval \([0,1]\) and \((0,1]\text{,}\) respectively). We will prove that there is no path \(p\) in \(S\) from \(p(0) = (0,0)\) to \(p(1) = b\) for any point \(b \in S_2\) by contradiction. Assume the existence of such a path \(p\text{.}\) Let \(U = p^{-1}(S_1)\) and \(V = p^{-1}(S_2)\text{.}\) Then
\begin{equation} [0,1] = p^{-1}(S) = p^{-1}(S_1 \cup S_2) = p^{-1}(S_1) \cup p^{-1}(S_2) = U \cup V\text{.}\tag{19.2} \end{equation}
Note that \(S_2\) is an open subset of \(S\text{,}\) since \(S_2 = \left( \bigcup_{z = (x,y) \in S_2} B\left(z, \frac{x}{2}\right)\right) \cap S\text{.}\) So the continuity of \(p\) implies that \(V\) is an open subset of \([0,1]\text{.}\) Also, the fact that \(p(0) \in S_1\) means that \(U \neq \emptyset\text{,}\) and the fact that \(p(1) \in S_2\) means that \(V \neq \emptyset\text{.}\) If we demonstrate that \(U\) is an open subset of \([0,1]\text{,}\) then Equation (19.2) will imply that \([0,1]\) is not connected, a contradiction. So we proceed to prove that \(U\) is open in \([0,1]\text{.}\)
Let \(x \in U\text{,}\) and so \(p(x)\) in \(S_1\text{.}\) The set \(O = B_S\left(p(x), \frac{1}{2}\right) \cap S\) is open in \(S\text{.}\) The continuity of \(p\) then tells us that \(p^{-1}(O)\) is open in \([0,1]\text{.}\) So there is a \(\delta \gt 0\) such that the open ball \(B=B_{[0,1]}(x, \delta)\) is a subset of \(p^{-1}(O)\text{.}\) We will prove that \(p(B) \subseteq S_1\text{.}\) This will imply that \(B \subseteq U\) and so \(U\) is a neighborhood of each of its points, and \(U\) is therefore an open set.
Figure 19.18. The set \(O\text{.}\)
Every element in \(B\) is mapped into \(O\) by the path \(p\text{.}\) The set \(O\) is complicated, consisting of infinitely many sub-curves of the curve \(S_2\text{,}\) along with points in \(S_1\text{,}\) as illustrated in Figure 19.18. To simplify our analysis, let us consider the projection onto the \(x\)-axis. The function \(P_x : \R^2 \to \R\) defined by \(P_x(x,y) = x\) is a continuous function. Let \(I = P_x(p(B))\text{.}\) Since \(p(B) \subseteq O\text{,}\) we know that \(I \subseteq P_x(O)\text{.}\) Let \(Z = P_x(O)\text{.}\) So \(I \subseteq Z\text{.}\) Since \(B\) is a connected set (\(B\) is an interval), we know that \(p(B)\) is a connected set. The fact that \(P_x\) is continuous means that \(I = P_x(p(B))\) is connected as well. Now \(I\) is a bounded subset of \(\R\text{,}\) so \(I\) must be a bounded interval. Recall that \(x \in B\) and so \(p(x) \in p(B)\text{.}\) The fact that \(p(x) \in S_1\) tells us that \(0 = P_x(p(x)) \in P_x(p(B)) = I\text{.}\) So \(I \neq \emptyset\text{.}\) There are two possibilities for \(I\text{:}\) either \(I = \{0\}\text{,}\) or \(I\) is an interval of positive length. We consider the cases.
Suppose \(I = \{0\}\text{.}\) Then the projection of \(p(B)\) onto the \(x\)-axis is the single point \(0\) and \(p(B) \subseteq S_1\) as desired. Suppose that \(I\) is an interval of the form \([0,d]\) or \([0,d)\) for some positive number \(d\text{.}\) The structure of \(O\) would indicate that there must be some gaps in the set \(Z\text{,}\) the projection of \(O\) onto the \(x\)-axis. This implies that \(I\) cannot be a connected interval. We proceed to show this. In other words, we will prove that \(I \setminus Z \neq \emptyset\) (which is impossible since \(I \subseteq Z\)). Remember that \(p(x) \in S_1\text{,}\) so let \(p(x) = (0, q)\text{.}\) We consider what happens if \(q \lt \frac{1}{2}\) and when \(q \geq \frac{1}{2}\text{.}\)
Suppose \(q \lt \frac{1}{2}\text{.}\) Then the ball \(B_S\left(p(x), \frac{1}{2}\right)\) contains only points with \(y\) value less than 1. Let \(N \in \Z^+\) so that \(t=\frac{1}{\pi/2+2N\pi} \lt d\text{.}\) Then \(t \in I\text{.}\) But \(\sin\left(\frac{1}{t}\right) = \sin(\pi/2 + 2N\pi) = \sin(\pi/2) = 1\text{,}\) and so \(\left(t,\sin\left(\frac{1}{t}\right)\right)\) is not in \(O\text{.}\) Thus, \(t \notin Z\text{.}\) Thus we have found a point in \(I \setminus Z\text{.}\)
Finally, suppose \(q \geq \frac{1}{2}\text{.}\) Then the ball \(B_S\left(p(x), \frac{1}{2}\right)\) contains only points with \(y\) value greater than \(-1\text{.}\) Let \(N \in \Z^+\) so that \(t=\frac{1}{3\pi/2+2N\pi} \lt d\text{.}\) Then \(t \in I\text{.}\) But \(\sin\left(\frac{1}{t}\right) = \sin(3\pi/2 + 2N\pi) = \sin(3\pi/2) = -1\text{,}\) and so \(t \notin Z\text{.}\) Thus we have found a point in \(I \setminus Z\text{.}\)
We conclude that there can be no path in \(S\) from \((0,0)\) to any point in \(S_2\text{,}\) completing our proof that \(S\) is not path connected. (In fact, the argument given shows that there is no path in \(S\) from any point in \(S_1\) to any point in \(S_2\text{.}\)