Tutorials › Topology (Undergraduate) › Uncountably Many Loop Classes In The Hawaiian Earring

Counterexamples in Topology · Tutorial 781 of 799

Uncountably Many Loop Classes In The Hawaiian Earring

Learn how shrinking circles make selected-circle concatenations continuous, and how winding numbers distinguish uncountably many based loop classes.

Intermediate 12 min read

What You'll Learn

  • Define the Hawaiian earring as a subspace of the plane and parametrize each circle.
  • Construct a continuous based loop from any subset of the positive integers.
  • Use a coordinate retraction to detect how often a loop traverses one circle.
  • Prove that winding number is invariant under based homotopy.
  • Deduce that the Hawaiian earring has uncountably many based loop classes.
  • Recognize why shrinking tails do not imply that a loop is homotopic to a finite truncation.

Loops That Visit Selected Circles

The Hawaiian earring consists of infinitely many circles that meet at one point, with the circles shrinking toward that common point. A loop can traverse some of the circles in sequence, using shorter and shorter time intervals. Shrinking diameters make the resulting path continuous even though infinitely many traversals accumulate before its final time. The key question is whether different choices of circles can give different homotopy classes. We will detect each choice by retracting the whole space onto one circle and measuring a winding number.

Use the Euclidean topology on \(\mathbb{R}^2\) and the subspace topology on the Hawaiian earring \(H\). Write \(p=(0,0)\) for its common point, and, for \(n\in\mathbb{N}\), let \(C_n\) be the circle with center \((1/n,0)\) and radius \(1/n\). Thus \(H=\bigcup_{n\in\mathbb{N}}C_n\). The circles meet pairwise only at \(p\): subtracting the equations for two distinct circles shows that any intersection has first coordinate \(0\), and substitution then gives second coordinate \(0\).

$$ C_n=\left\{(x,y)\in\mathbb{R}^2: \left(x-\frac{1}{n}\right)^2+y^2=\frac{1}{n^2}\right\}. $$

A based loop in \(H\) is a continuous map \(\gamma:[0,1]\to H\) with \(\gamma(0)=\gamma(1)=p\). Two such loops are based-homotopic if there is a continuous map \(F:[0,1]\times[0,1]\to H\) such that \(F(0,t)\) is the first loop, \(F(1,t)\) is the second, and \(F(s,0)=F(s,1)=p\) for all \(s,t\in[0,1]\). We construct a loop for every subset \(S\subseteq\mathbb{N}\).

Definition: For \(n\in\mathbb{N}\), set \(c_n(u)=\left(\frac{1}{n}(1-\cos(2\pi u)),\frac{1}{n}\sin(2\pi u)\right)\), for \(0\le u\le1\). This parametrizes \(C_n\) once, beginning and ending at \(p\). Given \(S\subseteq\mathbb{N}\), define \(t_0=0\) and \(t_n=1-2^{-n}\). On \([t_{n-1},t_n]\), let \(\gamma_S\) traverse \(c_n\) once if \(n\in S\), and stay at \(p\) if \(n\notin S\); set \(\gamma_S(1)=p\).

More explicitly, on an interval where \(n\in S\), use \(c_n((t-t_{n-1})/(t_n-t_{n-1}))\). The interval has length \(t_n-t_{n-1}=2^{-n}\). Each piece begins and ends at \(p\), so adjacent pieces agree at their common endpoint. The only continuity issue not settled by ordinary pasting on finitely many intervals is at \(t=1\).

Why the Infinite Concatenation Is Continuous

Proposition: For every subset \(S\subseteq\mathbb{N}\), the map \(\gamma_S:[0,1]\to H\) defined above is a continuous based loop.

Proof. On the interior of each interval \([t_{n-1},t_n]\), the map is either a rescaled continuous parametrization \(c_n\) or the constant map with value \(p\). The formulas meet at \(p\) at each shared endpoint, so \(\gamma_S\) is continuous at every \(t<1\). To check continuity at \(1\), let \(\varepsilon>0\). Choose \(N\) so large that \(2/(N+1)<\varepsilon\). Every point of \(C_n\) is at distance at most \(2/n\) from \(p\), because \(C_n\) has center at distance \(1/n\) from \(p\) and radius \(1/n\). If \(t_N<t<1\), then \(t\) lies in an interval with index \(n>N\), so either \(\gamma_S(t)=p\), or \(\|\gamma_S(t)-p\|\le 2/n\le 2/(N+1)<\varepsilon\). Also \(\gamma_S(1)=p\). This proves continuity at \(1\). The endpoint values are \(p\), so \(\gamma_S\) is a based loop. \(\square\)

Worked Example: A Loop That Selects Two Circles

Take \(S=\{1,3\}\). The first three interval endpoints are \(t_1=1-2^{-1}=1/2\), \(t_2=1-2^{-2}=3/4\), and \(t_3=1-2^{-3}=7/8\). Thus \(\gamma_S\) traverses \(C_1\) on \([0,1/2]\), stays at \(p\) on \([1/2,3/4]\), traverses \(C_3\) on \([3/4,7/8]\), and stays at \(p\) on every later interval. For instance, on \([3/4,7/8]\), the rescaled parameter is \((t-3/4)/(7/8-3/4)=8t-6\), which runs from \(0\) to \(1\). The loop therefore makes exactly one positive traversal of \(C_3\). The tail is constant, so continuity at \(1\) also follows directly in this example.

Reading One Circle at a Time

For each \(k\), we need a map that retains \(C_k\) and sends every other circle to \(p\). Define \(r_k:H\to C_k\) by \(r_k(x)=x\) for \(x\in C_k\) and \(r_k(x)=p\) for \(x\in C_n\) with \(n\ne k\). This is well-defined because the circles intersect only at \(p\).

Proposition: Each coordinate map \(r_k\) is continuous.

Proof. Let \(V\) be open in \(C_k\). If \(p\in V\), then \(C_k\setminus V\) is closed in the compact circle \(C_k\), hence compact and closed in the Hausdorff space \(\mathbb{R}^2\). Since \(r_k^{-1}(V)=H\setminus(C_k\setminus V)\), this preimage is open in \(H\). If \(p\notin V\), then \(V\) is open in \(H\): each point of \(C_k\setminus\{p\}\) has a neighborhood in \(H\) meeting no other circle. Indeed, circles with sufficiently large index lie arbitrarily close to \(p\), away from that point, and the finitely many remaining circles are compact and disjoint from it. Consequently \(r_k^{-1}(V)=V\) is open in \(H\). Both cases show continuity. \(\square\)

Identify \(C_k\) with the unit circle \(S^1=\{z\in\mathbb{C}:|z|=1\}\) by the homeomorphism sending \(c_k(u)\) to \(e^{2\pi i u}\). A loop \(u:[0,1]\to S^1\) based at \(1\) has a continuous angle lift \(\theta:[0,1]\to\mathbb{R}\) such that \(u(t)=e^{2\pi i\theta(t)}\) and \(\theta(0)=0\). Its winding number is the integer \(\theta(1)\). The lift can be obtained by subdividing the interval into pieces on which \(u(t)/u(t_i)\) avoids \(-1\), and using the continuous argument with values in \((-1/2,1/2)\) on each piece. The endpoint is an integer because \(u(1)=1\); the lift is unique, since two lifts differ by a continuous integer-valued function and agree at \(0\).

Theorem: Winding number is invariant under based homotopy of loops in \(S^1\).

Proof. First suppose two based loops \(u,v:[0,1]\to S^1\) satisfy \(|u(t)-v(t)|<2\) for every \(t\). Then \(v(t)/u(t)\ne-1\), because the distance between \(1\) and \(-1\) is \(2\). There is a continuous function \(\delta:[0,1]\to(-1/2,1/2)\) with \(v(t)/u(t)=e^{2\pi i\delta(t)}\). Since both loops start and end at \(1\), \(\delta(0)=\delta(1)=0\). If \(\theta\) is the lift of \(u\) starting at \(0\), then \(\theta+\delta\) is the lift of \(v\) starting at \(0\), and its endpoint equals \(\theta(1)\). The winding numbers are equal.

Now let \(F:[0,1]\times[0,1]\to S^1\) be a based homotopy. Uniform continuity on the compact square gives a partition \(0=s_0<s_1<\cdots<s_m=1\) fine enough that \(|F(s_j,t)-F(s_{j+1},t)|<2\) for every \(t\) and every \(j\). The preceding argument shows that the loops \(F(s_j,\cdot)\) and \(F(s_{j+1},\cdot)\) have equal winding numbers. Applying this equality across the finite partition proves that \(F(0,\cdot)\) and \(F(1,\cdot)\) have equal winding numbers. \(\square\)

Composing a based homotopy in \(H\) with \(r_k\), and then with the identification of \(C_k\) with \(S^1\), gives a based homotopy in \(S^1\). The theorem therefore supplies a homotopy invariant for each circle-coordinate of a loop in \(H\).

Worked Example: Distinguishing Odd and Even Selections

Let \(O=\{1,3,5,\ldots\}\) and \(E=\{2,4,6,\ldots\}\). At coordinate \(k=1\), \(r_1\circ\gamma_O\) traverses \(C_1\) once, while \(r_1\circ\gamma_E\) is constant. Their winding numbers are respectively \(1\) and \(0\). At coordinate \(k=2\), the roles reverse: \(r_2\circ\gamma_O\) is constant and \(r_2\circ\gamma_E\) traverses \(C_2\) once, giving winding numbers \(0\) and \(1\). Since even one coordinate has different winding numbers, the two loops cannot be based-homotopic.

Uncountably Many Different Classes

Theorem: The Hawaiian earring has uncountably many based loop homotopy classes.

Proof. For \(S\subseteq\mathbb{N}\), consider the loop \(\gamma_S\). For each \(k\), applying \(r_k\) gives one traversal of \(C_k\) if \(k\in S\), and the constant loop if \(k\notin S\). Under the identification with \(S^1\), the winding number is therefore \(1\) when \(k\in S\), and \(0\) when \(k\notin S\). If \(S\ne T\), there is some \(k\) belonging to exactly one of them. The corresponding winding numbers differ, so homotopy invariance proves that \(\gamma_S\) and \(\gamma_T\) are not based-homotopic. Thus distinct subsets give distinct loop classes.

For completeness, the set of subsets of \(\mathbb{N}\) is uncountable: if they could be listed as \(S_1,S_2,\ldots\), the subset \(D=\{j\in\mathbb{N}:j\notin S_j\}\) would differ from \(S_j\) at the element \(j\) for every \(j\), so it would not occur on the list. The injection from subsets to loop classes consequently proves that the collection of based loop classes is uncountable. \(\square\)

Worked Example: Comparing All Circles with the Even Circles

Let \(A=\mathbb{N}\) and \(B=E\), the set of even positive integers. At \(k=2\), both loops traverse \(C_2\), so both winding numbers are \(1\). That coordinate does not distinguish them. At \(k=1\), \(\gamma_A\) traverses \(C_1\), giving winding number \(1\), whereas \(\gamma_B\) is collapsed to the constant loop by \(r_1\), giving winding number \(0\). Hence \(\gamma_A\) and \(\gamma_B\) are not based-homotopic. Checking a coordinate where the selections agree is not enough; a single coordinate where they differ certifies distinctness.

A Common Pitfall: Shrinking Does Not Mean Homotopically Trivial

It is tempting to reason that the circles become so small that the infinite tail of a loop can be discarded by a homotopy. The diameters tending to zero prove continuity of the concatenation at its limiting time; they do not prove that the tail is homotopically trivial in \(H\). For example, compare \(\gamma_{\mathbb{N}}\) with the loop that traverses only \(C_1,\ldots,C_N\) and then stays at \(p\). At coordinate \(k=N+1\), the first loop has winding number \(1\), while the finite loop has winding number \(0\). The retraction and homotopy-invariance argument proves that these loops are not based-homotopic. The distinction is that smallness controls distances in the plane, whereas winding number records a topological obstruction that remains visible after retracting onto a particular circle.

The coordinate winding numbers are used here to distinguish the constructed family; no claim is being made that they classify every loop in the Hawaiian earring. Their strength is sufficient for the conclusion: each subset of circle indices gives a loop, and any change in the subset changes at least one homotopy invariant.

Check Your Understanding

Use the coordinate retractions and the winding-number argument to justify each answer.

  1. For \(S=\{2,5\}\), what are the winding numbers of the coordinate loops at \(k=2\), \(k=3\), and \(k=5\)?
  2. Why is the estimate \(\|\gamma_S(t)-p\|\le 2/n\) enough to prove continuity at \(1\), even when infinitely many circles are selected?
  3. If \(S\) and \(T\) differ at exactly one index \(k\), why can \(\gamma_S\) and \(\gamma_T\) not be based-homotopic?
  4. Explain why the shrinking diameters establish continuity of the concatenation but do not establish that its infinite tail is homotopic to a constant loop.
  5. Give the diagonal argument showing that the set of subsets of \(\mathbb{N}\) is uncountable.