Let $D^2$ be the closed disk on the plane.

First we pick an arbitrary point $x\in bd(D^2)$ on $D^2$, and define $X = D^2-\{x\}$.

Then define another space $Y$ by removing a homeomorphic image of the closed interval $I$ from the boundary, that is, $Y = D^2-h(I)$.(The original question defines $Y$ by removing the upper closed semi-circle of $D^2$ while I think this can be generalized.)

I tried to construct the homeomorphism between two spaces but failed. The two spaces are both connected, non-compact and convex so I also failed to prove they are not homeomorphic.

  • $\begingroup$ The boundary-punctured disc is homeomorphic to the closed upper half plane, and if you delete an arc, you instead get the upper half plane with $(-\infty, t_0) \cup (t_1, \infty)$ where $t_0 < t_1$. By scaling and translation, you see that the result of deleting any arc from the boundary of $D^2$ is homeomphic to any other. This still leaves the question of point vs arc open, but that is true (I just need to come up with the argument...) $\endgroup$
    – user98602
    Jul 18, 2018 at 2:13

2 Answers 2


Let $e : \mathbb{R} \to S^1, e(t) = e^{it}$. Then a homeomorphic image of a closed interval in $S^1$ is any set having the form $J = e([a,b])$ with $a < b$ and $b - a < 2\pi$.

We have $X = D^2 \backslash \{e(c) \}$ for some $c \in \mathbb{R}$ and $Y = D^2 \backslash e([a,b])$ for suitable $a,b$ as above. Define $r : [a,a + 2 \pi] \to [c,c+2\pi]$, $r(x) = c$ for $t \in [a,b]$, $r(x) = c + \frac{2\pi (x - b)}{a+2\pi-b}$ for $t \in [b,a+2\pi]$. This is a continuous map. Define

$$H : [a,a + 2 \pi] \times [0,1] \to [c,c+2\pi], H(x,t) = (1-t)(x -a + c) + t r(x) .$$

This is a continuous map such that (with $H_t(x) = H(x,t)$)

(1) $H_1 = r$

(2) $H_t(a) = c, H_t(a+2\pi) = c + 2\pi$ for all $t$

(3) $H_t$ is strictly increasing, in particular bijective, for all $t < 1$.


$$G : [a,a + 2 \pi] \times [0,1] \to D^2, G(x,t) = te(H(x,t))$$

which is again continuous. Consider the continuous map

$$p : [a,a + 2 \pi] \times [0,1] \to D^2, p(x,t) = te(x) .$$

Since domain and range are compact, it is an identification map. It is easy to check that if $p(\xi) = p(\xi')$, then $G(\xi) = G(\xi')$. Therefore we obtain a unique continuous $g : D^2 \to D^2$ such that $g \circ p = G$. By construction $g$ maps $Y$ bijectively onto $X$. The restriction $g' : Y \to X$ is an open map which shows that it is a homeomorphism: Let $U \subset Y$ be open. Then $D^2 \backslash U$ is compact, thus $g(D^2 \backslash U) ) = D^2 \backslash g(U)$ is compact so that $g'(U) = g(U)$ is open in $X$.


The answer of @PaulFrost is awesome, and should be the accepted answer. I thought I'd just give an explicit description of his approach.

We'll write the unit disk as $$ \mathbb{D} = \{re^{\pi it}\mid 0\le r\le1,\ -1\le t\le1\}$$

And define the lower hemisphere as $H=\{e^{\pi it}\mid -1\le t\le0\}$, and the point $q=-1$.

We have a continuous map $h:\mathbb{D}\rightarrow\mathbb{D}$, given by $$ h(re^{\pi it}) = re^{t-r+r|t|}$$ It should be checked this is well-defined (plugging in $t=1$ and $t=-1$). It's also straightforward that $h^{-1}(q)=H$. Thus we get a map $h:\mathbb{D}\setminus H\rightarrow \mathbb{D}\setminus\{q\}$.

An explicit inverse can be given too: $$ h^{-1}(\rho e^{\pi i\theta}) = \rho e^{(\rho+\theta)^2/(\rho+\theta+\rho|\rho+\theta|)}$$

One has to worry about that denominator being zero. The quadratic formula shows that happens when $\rho+\theta=0$. If $\rho<1$, then $-1<\theta<0$, and it is easy to check the fraction in the exponent goes to zero. So $h^{-1}(\rho e^{-\pi i\rho})=\rho$.

When $\rho=1$, that's a prpblem, because then we have $\theta=\pm1$. Luckily we are ignoring that point ($q$), so we get a well-defined map $$ h^{-1}:\mathbb{D}\setminus \{q\}\rightarrow\mathbb{D}\setminus H$$ and so $h$ is the desired homeomorphism.


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.