🧮 Brain Teaser

The Hawaiian Earring Is Not Semi-Locally Simply Connected

Consider the Hawaiian earring HH: the subspace of R2\mathbb{R}^2 defined as the union of circles

H=n=1Cn,Cn={(x,y):(x1n)2+y2=1n2}.H = \bigcup_{n=1}^{\infty} C_n, \quad C_n = \left\{(x,y) : \left(x - \frac{1}{n}\right)^2 + y^2 = \frac{1}{n^2}\right\}.

All circles pass through the origin p=(0,0)p = (0,0), and the radii shrink to zero.

Show that HH is not semi-locally simply connected at pp. That is, show that for every open neighborhood UU of pp in HH, there exists a loop based at pp in UU whose homotopy class is nontrivial in π1(H,p)\pi_1(H, p).

Why does this matter? What classical theorem does it obstruct?

fundamental groupcovering spacessemi-local simple connectivityHawaiian earringcounterexample

Answer: Hawaiian Earring Is Not Semi-Locally Simply Connected

Key Idea / Intuition

The Hawaiian earring has circles piling up at pp with radii 0\to 0. Any open neighborhood UU of pp, no matter how small, must contain entire small circles CnC_n for all sufficiently large nn (because eventually CnUC_n \subset U). Each such circle represents a loop that is not contractible in HH — it wraps around a hole that genuinely exists in the global space. Since UU cannot "kill" the loop (the loop goes around a hole in the ambient HH, not just in UU), the semi-local simple connectivity condition fails.

The punchline: this failure is exactly what blocks the existence of a universal cover for HH. The standard theorem says a path-connected, locally path-connected space has a universal cover if and only if it is semi-locally simply connected. The Hawaiian earring is a clean counterexample showing why that hypothesis is necessary.


Formal Proof / Solution

Setup. Let UU be any open neighborhood of p=(0,0)p = (0,0) in HH. We need to find a loop γ\gamma based at pp, lying entirely in UU, such that the induced homomorphism

i:π1(U,p)π1(H,p)i_* : \pi_1(U, p) \to \pi_1(H, p)

does not send [γ][\gamma] to 00.

Step 1: Every small neighborhood contains an entire circle CnC_n.

Since UU is open in HH and pUp \in U, there exists ε>0\varepsilon > 0 such that B(p,ε)HUB(p, \varepsilon) \cap H \subset U, where B(p,ε)B(p,\varepsilon) is the Euclidean open ball. The circle CnC_n has center (1/n,0)(1/n, 0) and radius 1/n1/n, so every point of CnC_n satisfies

(x,y)p(x,y)(1/n,0)+(1/n,0)p=1n+1n=2n.\|(x,y) - p\| \leq \|(x,y) - (1/n,0)\| + \|(1/n,0) - p\| = \frac{1}{n} + \frac{1}{n} = \frac{2}{n}.

For all n>2/εn > 2/\varepsilon, every point of CnC_n lies within distance ε\varepsilon of pp, so CnUC_n \subset U.

Step 2: The loop around CnC_n is nontrivial in π1(H,p)\pi_1(H,p).

Fix such an nn. Let γn\gamma_n be the loop that traverses CnC_n once (based at pp, the unique point of Cn{p}C_n \cap \{p\}). We claim [γn]0[\gamma_n] \neq 0 in π1(H,p)\pi_1(H, p).

Consider the retraction r:HCnr: H \to C_n defined by: r(x)={xif xCn,pif xCk, kn.r(x) = \begin{cases} x & \text{if } x \in C_n, \\ p & \text{if } x \in C_k,\ k \neq n. \end{cases} This map is continuous: on each CkC_k it is continuous, and at pp it sends every neighborhood to a neighborhood of r(p)=pr(p) = p. (Continuity at pp follows because any open set in CnC_n containing pp pulls back to an open set in HH containing pp, as the circles only meet at pp.)

Since rr is a continuous retraction onto CnS1C_n \cong S^1, it induces a surjection

r:π1(H,p)π1(Cn,p)Z.r_* : \pi_1(H, p) \to \pi_1(C_n, p) \cong \mathbb{Z}.

But r([γn])=[γn]π1(Cn,p)r_*([\gamma_n]) = [\gamma_n] \in \pi_1(C_n, p), which is a generator of Z\mathbb{Z}, hence nonzero. Therefore [γn]0[\gamma_n] \neq 0 in π1(H,p)\pi_1(H,p).

Step 3: The inclusion i:UHi: U \hookrightarrow H does not kill [γn][\gamma_n].

The loop γn\gamma_n lies in UU (by Step 1) and i([γn])=[γn]0i_*([\gamma_n]) = [\gamma_n] \neq 0 in π1(H,p)\pi_1(H,p) (by Step 2). Hence the homomorphism i:π1(U,p)π1(H,p)i_*: \pi_1(U,p) \to \pi_1(H,p) does not send [γn][\gamma_n] to the identity. \blacksquare

Consequence. Since every neighborhood UU of pp fails to "kill" some loop, HH is not semi-locally simply connected at pp.

By the Fundamental Theorem of Covering Space Theory (Munkres §82, Lee Chapter 12):

A path-connected, locally path-connected space XX admits a universal covering space if and only if XX is semi-locally simply connected.

The Hawaiian earring is path-connected and locally path-connected, but fails the third condition — so it has no universal cover. This is the canonical example showing the hypothesis is not vacuous.

Remark on π1(H,p)\pi_1(H,p). The fundamental group of the Hawaiian earring is genuinely exotic: it is not the free group on countably many generators (which would be the naïve guess). It contains elements corresponding to infinite products of loops γ1γ2γ3\gamma_1 \gamma_2 \gamma_3 \cdots, since such infinite concatenations converge uniformly to a continuous loop (the speed can be arranged so the nn-th circle is traversed in time 1/2n1/2^n). This makes π1(H,p)\pi_1(H,p) an uncountable, non-free group — a beautiful pathology.

Source: Munkres, Topology, §82; Lee, Introduction to Topological Manifolds, Ch. 12

Type: topologySource: Munkres, Topology, §82; Lee, Introduction to Topological Manifolds, Ch. 12Edit on GitHub ↗