My question is related to this question: About two space with infinite circles.. But what I am curious about is not actually the same with the linked one, so I make here a new question.

For each positive integer $n$, let $C_n$ denote the circle in $\Bbb R^2$ of radius $n$ centered at $(n,0)$. Then the $C_n$'s meet each other at the origin $(0,0)$. Let $X$ be their union $\bigcup_{n=1}^\infty C_n$. Then evidently the closure of $X$ equals $X$ union the $y$-axis. What I want to know is whether $X$ and $\bar{X}$ are homotopy equivalent or not.

What I know is: $X$ is homotopy equivalent, but not homeomorphic, to the wedge sum of countably inifnitely many copies of $S^1$. Also $X$ is certainly not a deformation retract of $\bar{X}$.


Let us regard $X$ as subspace of $\mathbb C$.

That $X$ and $\overline X$ are not homeomorphic is easy to see: The space $\overline X \setminus \{i\}$ has two path components (proof left as an exercise!), but $X \setminus \{0\}$ has infinitely many path components and for $z \ne 0$ the space $X \setminus \{z\}$ has one path component.

We shall show that the inclusion $i : X \to \overline X$ is a homotopy equivalence.

The idea for constructing a homotopy inverse is this: Map the vertical line $L = \{z \in \mathbb C \mid \text{Re}(z) = 0 \}$ and the "left half circles" $C_n^l = \{ z \mid \lvert z-n \rvert = n, \text{Re}(z) \le n \}$ to $0$ and map the "right half circles" $C_n^r = \{ z \mid \lvert z-n \rvert = n, \text{Re}(z) \ge n \}$ onto the full circles $C_n$ such that the two boundary points $n(1 \pm i)$ of $C^r_n$ are sent to $0$. Let us make this precise.

Define $M = L \cup \bigcup_{n=1}^\infty C_n^l$ and $$H : M \times I \to \overline X, H(z,t) = \begin{cases} n \left(1 + \dfrac{tz-n}{\lvert tz-n \rvert} \right) & z \in C^l_n \\ tz & z \in L \end{cases} $$ The geometric idea for this definition is this: Given $z \in C^l_n$, consider the line $L(z,t)$ through $n$ and $tz$ (which lies on the secant segment connecting $0$ and $z$). It intersects $C^l_n$ in a single point $H(z,t)$. Note that by construction $H( C_n^l \times I) \subset C_n^l$ and $H(L \times I) \subset L$. Moreover $H(z,0) = 0$ and $H(\overline z,t) = \overline {H(z,t)}$. We shall show later that $H$ is continous.

The map $f_n : [0,\pi] \to C_n, f_n(s) = n(1+e^{is})$, is an embedding whose image is the closed upper half circle. Let $R = \bigcup_{n=1}^\infty C_n^r$. The $C^r_n$ are open subspaces of $R$. Therefore $$G : R \times I \to \overline X, G(z,t) = \begin{cases} G_+(z,t) = f_n\left(\frac{2f_n^{-1}(z)f_n^{-1}(H(n(1+i),t)}{\pi}\right)) & z \in C^r_n , \text{Im}(z) \ge 0 \\ \overline{G_+(\overline z,t)} & z \in C^r_n , \text{Im}(z) \le 0 \end{cases}$$ is continuous. The geometric idea for this definition is this: Pull the two the boundary points $n(1 \pm i)$ of $C^r_n$ to the left along the circle $C_n$ until they reach the points $H(n(1 \pm i),t)$. This induces a deformation of $C^r_n$ which takes place inside $C_n$.

Since $H, G$ are defined on closed subspaces of $\overline X \times I$ whose union is $\overline X \times I$ and $H, G$ agree on the intersection of these sets, we get a continuous homotopy $$K : \overline X \times I \to \overline X .$$ Note that $K(X \times I) \subset X$ and $K(z,1) \in X$ for all $z \in \overline X$. Let $$\rho : \overline X \to X, \rho(z) = K(z,1). $$ Clearly $i \circ \rho$ is homotopic via $K$ to $id$, similarly $\rho \circ i$ is homotopic via $K \mid_{X \times I}$ to $id$.

This proves that $i$ is a homotopy equivalence.

Let is finally verify that $H$ is continuous. This is technically somewhat nasty.

Clearly $H$ is continuous in all points of $(M \setminus L) \times I$ because $M \setminus L = \bigcup_{n=1}^\infty \left(C_n^l \setminus \{0\}\right)$ and the sets $C_n^l \setminus \{0\}$ are open in $M$. Let us show that $H$ is continuous in all points $(\zeta,\tau) \in L \times I$. Since $H \mid_{L \times I}$ is continuous, it suffices to consider a point $(\zeta,\tau) \in L \times I$ and to prove that for each $\epsilon > 0$ one has $\lvert H(z,t) - \tau \zeta \rvert < \epsilon$ for all $(z,t) \in M \setminus L$ which are sufficiently close to $(\zeta,\tau)$. To do so, it clearly suffices to show that $$\lvert H(z,t) - t z \rvert < \epsilon \text{ if } (z,t) \text{ is sufficiently close to } (\zeta,\tau) .$$ Let $n$ denote the unique index such that $z \in C^l_n$ and write $z = x +iy$. The point $H(z,t)$ was obtained as the intersection of the line $L(z,t)$ with $C^l_n$. For $z \ne n(1 \pm i)$ the line $L(z,t)$ also intersects $L$ and it is easy to see that the intersection point is $g(z,t) = i\dfrac{nty}{n-tx}$. Note that for $z \in C^l_n \setminus \{n(1+i), n(1-i)\}$ we have $tx \le x < n$. Clearly $\lvert H(z,t) - t z \rvert \le \lvert g(z,t) - t z \rvert$. Straightforward calculations show that $$\lvert g(z,t) - t z \rvert^2 = t^2x^2 + \dfrac{t^4x^2y^2}{(n - tx)^2} \le x^2(1 + \dfrac{y^2}{(n - tx)^2}).$$ Write $\zeta = i\eta$ with $\eta \in \mathbb R$. Then for $\lvert z - \zeta \rvert < \frac{1}{2}$ we have $0 \le tx \le x \le \sqrt{x^2 +(y - \eta)^2} = \lvert z - \zeta \rvert < \frac{1}{2} < n$, thus we are in the situation $z \ne n(1 \pm i)$. Moreover $n - tx \ge n - x \ge 1 - \frac{1}{2} = \frac{1}{2}$ and $\lvert y - \eta \rvert \le \lvert z - \zeta \rvert < \frac{1}{2}$. In particular $\lvert y \rvert < \lvert \eta \rvert + \frac{1}{2}$ and therefore $$\lvert g(z,t) - t z \rvert^2 \le x^2(1 + 4(\lvert \eta \rvert + \frac{1}{2})^2)$$ which completes the proof because $x \to 0$ as $z \to \zeta$.

  • $\begingroup$ This is a very nice answer. +1 $\endgroup$ – Tyrone Oct 27 '20 at 18:58

Never mind, this does not work.

$X$ and $\overline{X}$ are homotopy equivalent by the following:

Let $C_0$ be the circle of radius $1$ and in general let $C_n$ denote the circle of radius $n+1$. Consider the homotopy that, on the interval $[0,\tfrac{1}{2}]$, shrinks $C_0$ so as to have radius $$(1-2t)+(2t)tan^{-1}(1)$$ and leaves the remaining $C_n$ and the line $\mathbb{R}\times \{0\}$ fixed. More generally, on the interval $[\tfrac{2^{n}-1}{2^n},\tfrac{2^{n+1}-1}{2^{n+1}}]$ the map deforms $C_n$ into the circle of radius $$\Big(1-2^{n+1}\Big(t-\frac{2^n-1}{2^n}\Big)\Big)(n+1)+\Big(2^{n+1}\Big(t-\frac{2^n-1}{2^n}\Big)\Big)tan^{-1}(n+1)$$ leaving everything else fixed. At the end of the homotopy, all of our circles will be bunched up inside of a missing circle having radius $\frac{\pi}{2}$. Since $\mathbb{R}\times \{0\}$ is fixed at each stage, the restriction of the homotopy to this subset is the identity. With the circles unable to interfere, we can crush $\mathbb{R}\times \{0\}$ and then launch all the circles back to their original positions with a homotopy constructed from the tangent function.

  • 1
    $\begingroup$ This does not work. Shrinking the $C_n$ is possible in $\mathbb R^2$, but you need a homotopy taking place in $\overline X$. It is impossible to find a homotopy shifting all points of $C_n$ (except $0$) into $\overline X \setminus (C_n \setminus \{0\})$. $\endgroup$ – Paul Frost Oct 25 '20 at 22:51
  • $\begingroup$ Oh my, this really was wrongheaded. Thank you for pointing out my error. $\endgroup$ – Gary D Oct 25 '20 at 23:25

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.