Topologist's Sine Curve

This scene draws the topologist's sine curve. Watch it oscillate endlessly near zero, and the limit segment it sticks to.

The closer to zero, the faster the curve wiggles. So a whole segment on the axis sticks to the curve. The combined figure is in one piece. Yet no path walks from the segment into the curve. Being in one piece and having a path are different things.

For example, shake one end of a 🧡 string faster and faster. The shaking end blurs into a band. The band is visible, but you cannot walk from one point of it onto a chosen part of the string.

S={(x,sin⁑(1/x)):x>0},Sβ€Ύ=Sβˆͺ({0}Γ—[βˆ’1,1])S = \{(x, \sin(1/x)) : x > 0\}, \quad \overline{S} = S \cup \bigl(\{0\} \times [-1,1]\bigr)
Sβ€ΎΒ connected,Sβ€ΎΒ notΒ path-connected\overline{S} \text{ connected},\quad \overline{S} \text{ not path-connected}

Examples & Exercises

Prove completely that S={(x,sin⁑(1/x)):x>0}βŠ†R2S = \{(x, \sin(1/x)) : x > 0\} \subseteq \mathbb{R}^2 is connected.

S=f((0,∞)),f(x)=(x, sin⁑(1/x))S = f((0,\infty)), \qquad f(x) = (x,\, \sin(1/x))
xmin⁑=0.05x_{\min} = 0.05: the curve SS alone oscillates infinitely near 0, yet it is one unbroken continuous curve.
Show proof
  1. (0,∞)(0,\infty) is connected (it is an interval β€” the same sup⁑\sup argument as for R\mathbb{R} applies verbatim). The map f(x)=(x,sin⁑(1/x))f(x) = (x, \sin(1/x)) is continuous since both coordinate functions are, and its image is exactly SS.

  2. The continuous image of a connected set is connected: if f(X)=AβˆͺBf(X) = A \cup B were a separation into disjoint nonempty relatively open sets, then X=fβˆ’1(A)βˆͺfβˆ’1(B)X = f^{-1}(A) \cup f^{-1}(B) would be a separation of XX into disjoint nonempty relatively open sets (A,BA, B nonempty implies their preimages are nonempty), contradicting the connectedness of XX.

  3. Hence S=f((0,∞))S = f((0,\infty)) is connected. The closure of a connected set is connected, so T=Sβ€ΎT = \overline{S} is connected as well β€” problem 2 below computes this closure.

Prove completely that Sβ€Ύ=Sβˆͺ({0}Γ—[βˆ’1,1])\overline{S} = S \cup \big(\{0\} \times [-1,1]\big): the closure of SS is SS together with the vertical segment {0}Γ—[βˆ’1,1]\{0\} \times [-1,1].

Sβ€Ύ=Sβˆͺ({0}Γ—[βˆ’1,1])\overline{S} = S \cup \big(\{0\} \times [-1,1]\big)
xmin⁑=0.01x_{\min} = 0.01: zooming closer to 0 shows the sequences (xn,1)(x_n, 1), (xn,βˆ’1)(x_n, -1) converging onto the endpoints of the red segment.
Show proof
  1. βŠ‡\supseteq: SβŠ†Sβ€ΎS \subseteq \overline{S} is trivial. For (0,y)(0,y) with ∣yβˆ£β‰€1|y| \le 1, put xn=1arcsin⁑y+2Ο€nx_n = \frac{1}{\arcsin y + 2\pi n} (nβ‰₯1n \ge 1); then xnβ†’0+x_n \to 0^+ and sin⁑(1/xn)=sin⁑(arcsin⁑y+2Ο€n)=y\sin(1/x_n) = \sin(\arcsin y + 2\pi n) = y.

    Hence (xn,y)∈S(x_n, y) \in S converges to (0,y)(0,y), so every point of {0}Γ—[βˆ’1,1]\{0\} \times [-1,1] is a limit point of SS.

  2. βŠ†\subseteq: let p∈Sβ€Ύp \in \overline{S}. If p=(x,y)p = (x,y) with x<0x < 0, the ball B(p,∣x∣/2)B(p, |x|/2) contains no point of SS (whose first coordinates are all positive) β€” impossible. If x>0x > 0 and pβˆ‰Sp \notin S, then K=f([x/2, 3x/2])K = f([x/2,\, 3x/2]) (ff from the problem above) is compact as a continuous image of a compact set, hence closed.

    Now let r>0r > 0 be arbitrary: points of SS meeting B(p,min⁑(r,x/2))B(p, \min(r, x/2)) have first coordinate in (x/2, 3x/2)(x/2,\, 3x/2), so they lie in KK. Since p∈Sβ€Ύp \in \overline{S}, this ball meets SS, hence meets KK, so dist⁑(p,K)<r\operatorname{dist}(p, K) < r. As r>0r > 0 was arbitrary, p∈Kβ€Ύ=KβŠ†Sp \in \overline{K} = K \subseteq S β€” a contradiction. Thus x>0x > 0 implies p∈Sp \in S.

  3. Finally, if p=(0,y)p = (0,y): for (u,v)∈S(u,v) \in S, ∣pβˆ’(u,v)∣β‰₯∣yβˆ’v∣β‰₯∣yβˆ£βˆ’1|p - (u,v)| \ge |y - v| \ge |y| - 1 (since ∣v∣=∣sinβ‘βˆ£β‰€1|v| = |\sin| \le 1). If ∣y∣>1|y| > 1, the ball B(p,(∣yβˆ£βˆ’1)/2)B(p, (|y|-1)/2) misses SS, contradicting p∈Sβ€Ύp \in \overline{S}.

    Hence ∣yβˆ£β‰€1|y| \le 1, i.e. p∈{0}Γ—[βˆ’1,1]p \in \{0\} \times [-1,1]. The three cases together give Sβ€Ύ=Sβˆͺ({0}Γ—[βˆ’1,1])\overline{S} = S \cup (\{0\} \times [-1,1]).

Prove completely that T=Sβ€ΎT = \overline{S} is not path-connected: no path in TT leads from a point of the segment {0}Γ—[βˆ’1,1]\{0\} \times [-1,1] to a point of SS.

Ξ³(0)∈{0}Γ—[βˆ’1,1],Β Ξ³(1)∈Sβ€…β€ŠβŸΉβ€…β€ŠnoΒ continuousΒ Ξ³:[0,1]β†’T\gamma(0) \in \{0\} \times [-1,1],\ \gamma(1) \in S \implies \text{no continuous } \gamma: [0,1] \to T
Try "walking" from the red segment into the curve β€” the infinite oscillation near 0 blocks every path. The proof captures this obstruction with sequences.
Show proof
  1. Assume a continuous Ξ³:[0,1]β†’T\gamma: [0,1] \to T with Ξ³(0)∈{0}Γ—[βˆ’1,1]\gamma(0) \in \{0\} \times [-1,1] and Ξ³(1)∈S\gamma(1) \in S, aiming for a contradiction. Write Ξ³(t)=(x(t),y(t))\gamma(t) = (x(t), y(t)) and put c=sup⁑{t∈[0,1]:x(t)=0}c = \sup\{t \in [0,1] : x(t) = 0\} (nonempty, containing 00). By continuity x(c)=0x(c) = 0, and c<1c < 1 since x(1)>0x(1) > 0.

    Moreover TβŠ†{xβ‰₯0}T \subseteq \{x \ge 0\} by the definition of T=Sβ€ΎT = \overline{S} (points of SS have positive first coordinate and the limit segment has first coordinate 00), so x(t)β‰₯0x(t) \ge 0 for all tt. For t∈(c,1]t \in (c,1] we also have x(t)β‰ 0x(t) \ne 0 (otherwise the supremum would be larger), hence x(t)>0x(t) > 0 on (c,1](c,1].

  2. For t∈(c,1]t \in (c,1], Ξ³(t)∈T\gamma(t) \in T has positive first coordinate, so Ξ³(t)∈S\gamma(t) \in S and y(t)=sin⁑(1/x(t))y(t) = \sin(1/x(t)). Since vn=12Ο€n+Ο€/2β†’0v_n = \frac{1}{2\pi n + \pi/2} \to 0, we have vn<x(1)v_n < x(1) for large nn. On [c,1][c,1], x(c)=0<vn<x(1)x(c) = 0 < v_n < x(1), so the intermediate value theorem gives tn∈(c,1]t_n \in (c,1] with x(tn)=vnx(t_n) = v_n.

    Then y(tn)=sin⁑(2Ο€n+Ο€/2)=1y(t_n) = \sin(2\pi n + \pi/2) = 1. Every subsequential limit tβˆ—βˆˆ[c,1]t^* \in [c,1] of (tn)(t_n) satisfies x(tβˆ—)=lim⁑vn=0x(t^*) = \lim v_n = 0, forcing tβˆ—=ct^* = c (as x>0x > 0 on (c,1](c,1]). Hence tnβ†’ct_n \to c, and by continuity y(c)=lim⁑y(tn)=1y(c) = \lim y(t_n) = 1.

  3. Likewise, with wn=12Ο€n+3Ο€/2w_n = \frac{1}{2\pi n + 3\pi/2}, pick sn∈(c,1]s_n \in (c,1] with x(sn)=wnx(s_n) = w_n; then y(sn)=sin⁑(2Ο€n+3Ο€/2)=βˆ’1y(s_n) = \sin(2\pi n + 3\pi/2) = -1, and the same argument gives snβ†’cs_n \to c, so y(c)=βˆ’1y(c) = -1. This contradicts y(c)=1y(c) = 1, so the assumed path cannot exist.

    TT is connected but not path-connected.

5. Topology