Skip to main content

Math 344-1,2: Kursobjekt

Section 1.10 Homeomorphisms

Consider the two spaces \(X=\{1,2\}\) and \(Y=\{3,4\}\text{,}\) both equipped with the discrete topology. As sets they are clearly not equal; and yet considered as topological spaces they are essentially the same. Indeed, we call both spaces the discrete topological space on a set of two elements. The notion of a homeomorphism gives us a rigorous way of determining when two spaces are “essentially the same thing, topologically speaking”.

Definition 1.10.1. Homeomorphism.

Let \(X\) and \(Y\) be topological spaces. A function \(f\colon X\rightarrow Y\) is a homeomorphism if the following conditions are satisfied:
  1. \(f\) is continuous;
  2. \(f\) is invertible (equivalently, \(f\) is bijective);
  3. the inverse function \(f^{-1}\colon Y\rightarrow X\) is continuous.
Two spaces are homeomorphic if there is a homeomorphism between them.

Remark 1.10.2. Homeomorphism: inverse must be continuous.

You may have met analogs of the homeomorphism in other areas of mathematics: e.g., in linear algebra or group theory. In category theory it is what is called an isomorphism. The basic idea is that once you decide what the “structure-preserving” maps are between objects (continuous functions for topological spaces, linear transformations for vector spaces, homomorphisms for groups), an isomorphism is essentially just an invertible structure-preserving map: a vector space isomorphism is an invertible linear transformation; a group isomorphism is an invertible group homomorphism. The only subtlety is that you want the inverse of this invertible function to also be structure preserving! For linear transformations and group homomorphisms this is automatic. Not so for continuous functions! The inverse of an invertible continuous function need not be continuous, as the example below illustrates.

Example 1.10.3. Homeomorphism: inverse must be continuous.

Let \(\R_{\operatorname{std}}\) and \(\R_{\operatorname{disc}}\) be the spaces obtained by equipping \(\R\) with the standard and discrete topologies, respectively. The identity function
\begin{align*} \id_\R\colon \R_{\operatorname{disc}}\amp\rightarrow \R_{\operatorname{std}}\\ x\amp\mapsto x \end{align*}
is clearly continuous and invertible. In fact, we have \(\id_\R^{-1}=\id_\R\text{.}\) However, \(\id_\R\colon \R_{\operatorname{std}}\rightarrow \R_{\operatorname{disc}}\) is not continuous. Indeed, \(U=[0,1)\) is open in \(\R_{\operatorname{disc}}\text{,}\) but \(\id_\R^{-1}(U)=[0,1)\) is not open in \(\R_{\operatorname{std}}\text{.}\) Thus \(\id_{\R_{\operatorname{disc}}}\colon \R_{\operatorname{std}}\rightarrow \R\) is an invertible continuous function that is not a homeomorphism.

Remark 1.10.4. Homeomorphic equivalence relation.

For topological spaces \(X, Y\text{,}\) define the relation \(X\sim Y\) to mean \(X\) is homeomorphic to \(Y\text{.}\) This relation is an equivalence relation. That is, for all topological spaces \(X, Y, Z\) we have
\begin{align*} X\amp\sim X \amp \text{(reflexive)}\\ X\sim Y\amp\iff Y\sim X \amp \text{(symmetric)}\\ X\sim Y \text{ and } Y\sim Z \amp \implies X\sim Z \amp \text{(transitive)}\text{.} \end{align*}
These three properties follow from the simple facts that (1) the identity map is a homeomorphism, (2) the inverse of a homeomorphism is a homeomomoprhism, (3) the composition of homeomorphisms is a homeomorphism.

Example 1.10.5. All open intervals of \(\R\) are homeomorphic.

In the following, all intervals are treated as metric spaces with the Euclidean metric on \(\R\text{.}\) Make use of any function whose continuity properties are well known.
  1. Prove that any finite open interval \((a,b)\) is homeomorphic to \((0,1)\text{.}\)
  2. Prove that \((-\pi/2, \pi/2)\) and \(\R\) are homeomorphic.
  3. Prove that \((a,\infty)\) and \((-\infty, b)\) are both homeomorphic to \(\R\text{.}\)
  4. Explain why we may now conclude that all open intervals of \(\R\) are homeomorphic.
Solution.
  1. The function \(f\colon (0,1)\rightarrow (a,b)\) defined as \(f(x)=a+(b-a)x\) is linear (hence continuous) and has continuous inverse \(f^{-1}(x)=\frac{x-a}{b-a}\text{.}\)
  2. The function \(\tan\colon (-\pi/2,\pi/2)\rightarrow \R\) is continuous and has continuous inverse \(f^{-1}(x)=\arctan x\text{.}\)
  3. The functions \(f\colon (a,\infty)\rightarrow \R\text{,}\) \(f(x)=\ln(x-a)\text{,}\) and \(g\colon (-\infty, b)\rightarrow \R\text{,}\) \(g(x)=\ln(b-x)\) are continuous and have continuous inverses \(f^{-1}(x)=e^x+a\) and \(g^{-1}(x)=b-e^{x}\text{.}\)
  4. This is now the result of the fact that being homeomorphic is an equivalence relation. In more detail, from (1) we know that all finite open intervals are homeomorphic; this in conjunction with (2) shows that all finite open intervals are homemorphic to \(\R\text{.}\) From (3) we conclude that all half-finite intervals are homeomorphic to \(\R\text{.}\) Since any open interval is either finite, half-finite, or equal to \(\R\text{,}\) we conclude that all open intervals are homeomorphic.

Definition 1.10.6. Open and closed maps.

Let \(f\colon X\rightarrow Y\) be a function between topological spaces.
  1. \(f\) is open if for all open \(U\subseteq X\) we have \(f(U)\) open in \(Y\text{.}\)
  2. \(f\) is closed if for all closed \(C\subseteq X\) we have \(f(C)\) closed in \(Y\text{.}\)

Proof.

Assume \(f\) is continuous and invertible throughout. To be a homeomorphism we need \(g=f^{-1}\) to be continuous. Since \(g(y)=x\) if and only if \(f(x)=y\text{,}\) we see that \(g^{-1}(A)=f(A)\text{.}\) Thus \(g\) continuous if and only if \(g^{-1}(U)\) is open for all open \(U\subseteq X\text{,}\) if and only if \(f(U)\) is open for all open \(U\subseteq X\text{,}\) if and only if \(f\) is open. The same argument with “closed” replacing “open” shows \(g\) is continuous if and only if \(f\) is closed. The equivalence of (1)-(3) now follows.

Remark 1.10.8. Open and closed maps.

Don’t be mislead by Theorem 1.10.7 into thinking that being an open continuous map is equivalent to being a closed continuous map; we need the additional assumption of invertibility to guarantee equivalence. As an example, consider any constant function from \(\R\) to itself: this is clearly continuous and closed, but not open.

Definition 1.10.9. Topological properties.

Let \(\mathcal{P}\) be a property that is either satisfied or not satisfied by any given topological space. We say \(\mathcal{P}\) is a topological property if it is invariant under homeomorphisms: i.e., if \(X\) and \(Y\) are homeomorphic, then \(X\) satisfies \(\mathcal{P}\) if and only if \(Y\) satisfies \(\mathcal{P}\text{.}\)

Example 1.10.10. Topological properties.

It is easy to see that the properties of being discrete, \(T_1\text{,}\) Hausdorff, or a metric space are topological.

Proof.

  1. Fix any \(i\in I\text{.}\) We showed \(\pi_i\) is continuous in Example 1.9.5. To see that \(\pi_i\) is surjective, given any \(x_i\in X_i\text{,}\) we have \(\pi_i^{-1}(\{x_i\}=\{x_i\}\times \prod_{j\ne i} X_j\text{,}\) which is nonempty. Lastly, suppose \(U\subseteq X\) is open. We wish to show that \(\pi_i(U)\) is open in \(X_i\text{.}\) To this end, given any \(x_i=\pi_i(\boldx)\in \pi_i(U)\text{,}\) we have \(x\in B\subseteq U\) for some open base set
    \begin{equation*} B=\prod_{i\in I}U_i\text{,} \end{equation*}
    where \(U_j\) is open in \(X_j\) and \(U_j=X_j\) for all but finitely many \(j\text{.}\) But then \(\pi_(\boldx)\in \pi_i(B)=U_i\subseteq \pi_i(U)\text{.}\) Since \(U_i\) is open, we have shown that given any \(x_i\in \pi_i(U)\text{,}\) there is an open set \(U_i\) such that \(x\in U_i\subseteq \pi_i(U)\text{.}\) It follows that \(\pi_i(U)\) is open, as desired.
  2. First observe that equality for elements \(\boldx=(x_i)_{i\in I}, \boldy=(y_i)_{i\in I}\) in \(X\) can be expressed in terms of the projection maps as follows:
    \begin{equation*} \boldx=\boldy \iff x_i=y_i \text{ for all } i\in I \iff \pi_i(x)=\pi_i(y) \text{ for all } i\in I\text{.} \end{equation*}
    Using this fact, we see that given a function \(f\colon Y\rightarrow X\) we have
    \begin{align*} f_i=\pi_i\circ f \text{ for all } i\in I \amp \iff f_i(y)=\pi_i(f(y)) \text{ for all } y\in Y, i\in I \\ \amp \iff f(y)=(f_i(y))_{i\in I} \text{.} \end{align*}
    Thus the function \(f(y)=(f_i(y))_{i\in I}\) is the unique function from \(Y\) to \(X\) satisfying \(f_i=\pi_i\circ f\) for all \(i\in I\text{.}\) To see that this \(f\) is continuous, we show that \(f^{-1}(B)\) is open for any base open set \(B\subseteq X\text{.}\) For such a \(B\) we have \(B=\prod_{j=1}^n U_{i_j}\times \prod_{i\ne j}X_i\) for some open sets \(U_{i_j}\subseteq X_{i_j}\text{,}\) \(1\leq j\leq n\text{.}\) Note that for all \(1\leq j \leq n\) we have
    \begin{equation*} \pi_{i_j}^{-1}(U_{i_j})=U_{i_j}\times \prod_{i\ne i_j}X_i \end{equation*}
    from which it follows that
    \begin{align*} B=\prod_{j=1}^n U_{i_j}\times \prod_{i\ne j}X_i\amp = \bigcap_{j=1}^nU_{i_j}\times \prod_{i\ne i_j}X_i \\ \amp = \bigcap_{j=1}^n\pi_{i_j}^{-1}(U_{i_j})\text{.} \end{align*}
    Thus
    \begin{align*} f^{-1}(B) \amp =f^{-1}\left( \bigcap_{j=1}^n\pi_{i_j}^{-1}(U_{i_j})\right)\\ \amp = \bigcap_{j=1}^nf^{-1}(\pi_{i_j}^{-1}(U_{i_j}))\\ \amp = \bigcap_{j=1}^n(\pi_{i_j}\circ f)^{-1}(U_{i_j}) \amp \left((\pi_{i_j}\circ f)^{-1}(A)=f^{-1}(\pi_{i_j}^{-1}(A)) \right)\\ \amp = \bigcap_{j=1}^nf_{i_j}^{-1}(U_{i_j}) \amp (f_i=\pi_i\circ f \text{ for all } i\in I)\text{.} \end{align*}
    Since each \(f_{i_j}\) is continuous, this last set is a finite intersection of open sets. Thus \(f^{-1}(B)\) is open, as desired.
  3. Let \(X'\) together with continuous maps \(\pi_i'\colon X\rightarrow X_i\) be any other space that satisfies Product universal mapping property. Letting \(Y\) be the space \(X'\) together with its maps \(\pi_i'\text{,}\) there is a continuous function \(f\colon X'\rightarrow X\) satisfying \(\pi_i'=\pi_i\circ f\) for all \(i\text{.}\) Similarly, since \(X'\) also satisfies the product mapping property, taking \(Y=X\) along with its projection maps, we see there is a continuous map \(g\colon X\rightarrow X'\) satisfying \(\pi_i=\pi_i'\circ g\text{.}\) We claim \(f\circ g=\id_X\) and \(g\circ f=\id_{X'}\text{:}\) i.e., \(g=f^{-1}\) is a continuous inverse of \(f\text{,}\) and thus \(f\) is a homeomorphism between \(X'\) and \(X\text{.}\) To see why this is true, we use the uniqueness claim in the product mapping property (twice). Indeed, the function \(f\circ g\colon X\rightarrow X\) is continuous and satisfies
    \begin{equation*} \pi_i\circ (g\circ f)=(\pi_i\circ g)\circ f=\pi_i'\circ f=\pi_i \end{equation*}
    for all \(i\in I\text{.}\) But the function \(\id_X\colon X\rightarrow X\) also satisfies \(\pi_i=\pi_i\circ\id_X\) for all \(i\in I\text{.}\) By the uniqueness claim of the product mapping property, we conclude \(\id_X=f\circ g\text{,}\) as desired. The proof that \(g\circ f=\id_{X'}\) is nearly identical, now applying the product mapping property to maps into \(X'\text{.}\)
In order to give a plain English description of Theorem 1.10.11 it helps to define the purely set-theoretic notion of a product of maps.

Definition 1.10.12. Product of functions.

Given a collection of functions \(\{f_i\colon A\rightarrow B_i\}_{i\in I}\text{,}\) their product, denoted \(f=(f_i)_{i\in I}\) is the the function
\begin{align*} f\colon A \amp \rightarrow \prod_{i\in I} B_i\\ x \amp\mapsto (f_i(x))_{i\in I} \text{.} \end{align*}
For each \(i\in I\) we call \(f_i\) the \(i\)-th component function of \(f=(f_i)_{i\in I}\text{.}\)

Proof.

  1. This is just a rewording of Item 2 using the notion of a product of functions.
  2. The fact that \(f=(\pi_i\circ f)_{i\in I}\) is purely set theoretic. Check for yourself.
    If \(f\) is continuous, then since \(\pi_i\) is continuous we have \(f_i=\pi_\circ f\) continuous.
    Conversely, if \(f_i\) is continuous, then the product \(f=(f_i)_{i\in I}\) by (1).

Remark 1.10.14. Universal mapping property of product space.

There is quite a bit to unpack in Theorem 1.10.11. The result is useful on a number of levels, some theoretical and some practical.
Its practical utility is nicely articulated by Corollary 1.10.13, which tells us that a continuous function to a product space is the same thing as a product of continuous functions. You may have seen this property at work in an analysis or multivariable calculus course. For example, oftentimes in these settings a function
\begin{align*} f\colon \R^n \amp \rightarrow \R^m\\ \boldx \amp\mapsto (f_1(\boldx),f_2(\boldx),\dots, f_m(\boldx)) \end{align*}
is simply defined to be continuous if each component functions \(f_i\colon \R^n\rightarrow \R\) is continuous. Thanks to Theorem 1.10.11 we understand that this is equivalent to \(f\) being continuous with respect to the standard topology on \(\R^n\) and the product topology on \(\R^m\) (which is equal to the standard Euclidean topology on \(\R^m\)).
On the theoretical side, the theorem identifies precisely what is so special about the product topology. Namely, it is the unique (up to homeomorphism) space satisfying the Product universal mapping property. The property establishes \(X\) with the product topology as the category theoretic product of the spaces \(X_i\text{.}\) You may have seen analogues of this type of object in other categories (e.g., category of groups or category of rings).

Example 1.10.15. Polar transformation.

Call the map
\begin{align*} f\colon \R^2\amp\rightarrow \R^2 \\ (r,\theta) \amp\mapsto (r\cos\theta, r\sin\theta) \end{align*}
the polar transformation.
  1. Let \(U\subseteq \R^2\) be the infinite horizontal strip \(U=\{(r,\theta)\in \R^2\colon r> 0, -\pi < \theta < \pi\}\text{,}\) and let \(V=\R^2- \{(x,0)\colon x\leq 0\}\text{,}\) the plane with the nonpositive real axis deleted.
    Prove that \(f\vert_U\colon U\rightarrow V\) is a homeomorphism.
  2. Let \(U'\subseteq \R^2\) be the infinite horizontal strip \(U'=\{(r,\theta)\in \R^2\colon r> 0, 0 < \theta < 2\pi\}\text{,}\) and let \(V'=\R^2- \{(x,0)\colon x\geq 0\}\text{,}\) the plane with the nonnegative real axis deleted.
    Prove that \(f\vert_U'\colon U'\rightarrow V'\) is a homeomorphism.
  3. Conclude that “open polar rectangles” are open in \(\R^2-\{(0,0)\}\text{.}\) In other words, given any \(r_1,r_2,\theta_1,\theta_2\in \R\) satisfying \(0< r_1< r_2\) and \(\theta_1< \theta_2\text{,}\) the region \(R\subseteq \R^2\) consisting of all points with polar coordinates \((r,\theta)\) satisfying
    \begin{align*} r_1 \amp< r < r_2 \\ \theta_1 \amp < \theta < \theta_2 \end{align*}
    is open. In fact, you can show that such polar rectangles form a basis for \(\R^2-\{(0,0)\}\text{.}\)
Solution.
We leave most of this as an exercise. Note that to show \(f\) is continuous, by Theorem 1.10.11 we need only show that the two “component functions”
\begin{align*} f_1=\pi_1\circ f\colon \R^2\amp \rightarrow \R \amp f_2=\pi_2\circ f\colon \R^2\amp \rightarrow \R \\ (r,\theta)\amp \mapsto r\cos\theta \amp (r,\theta)\amp \mapsto r\sin\theta \end{align*}
are continuous. To see that \(f_1\) is continuous, note that we have
\begin{equation*} f_1=m\circ g \end{equation*}
where \(g\colon \R^2\rightarrow \R^2 \) is the function \(g(r,\theta)=(r,\cos \theta)\) and \(m\colon \R^2\rightarrow \R\) is the multiplication map \(m(x,y)=xy\text{.}\) Both \(g\) and \(m\) are continuous: the two component functions of \(g\) are \(\pi_1\circ g=\pi_1\) and \(\pi_2\circ g=\cos\circ \pi_2\text{,}\) both of which are continuous; we showed in homework that more generally scalar multiplication \(m\colon \R\times \R^n\rightarrow \R^n\) is continuous for any \(n\text{.}\)

Example 1.10.16. Continuous bijection onto circle.

Let \(X=[0,1)\subseteq \R\) and \(S^1=\{(x,y)\in \R^2\colon x^2+y^2=1\}\subseteq \R^2\text{,}\) and equip each set with the subspace topology inherited from \(\R\) (resp. \(\R^2\)) with its Euclidean topology.
Prove: the function \(f\colon X\rightarrow S^1\) defined as \(f(t)=(\cos 2\pi t, \sin 2\pi t)\) is continuous and invertible, but not a homeomorphism.
Solution.
Recall first that on \(\R^2=\R\times \R\) the product topology, Euclidean metric topology, and box topology are all equal. Since \(f_1(x)=\cos 2\pi t\) and \(f_2(x)=\sin 2\pi t\) are continuous, it follows from Theorem 1.10.11 that the map \(g\colon [0,1)\rightarrow \R\times \R=\R^2\) is continuous. Next, since \(g=i\circ f\text{,}\) where \(i\colon S^1\hookrightarrow \R^2\text{,}\) it follows that \(f\) is continuous: indeed, for any open set \(U\subseteq S^1\text{,}\) we have \(U=U'\cap S^1\) for some open \(U'\subseteq \R^2\text{,}\) and \(f^{-1}(U)=g^{-1}(U')\text{,}\) which is open since \(g\) is continuous.
The usual trigonometric arguments show that \(f\) is injective and surjective. However, \(f\) is not open, and hence not a homeomorphism. Indeed, let \(U=[0,1/2)\text{,}\) which is open in \([0,1)\text{.}\) The image \(f(U)\) is easily seen to be the upper half of the circle minus the point \(Q=(-1,0)\text{:}\) in particular, all points \(f(U)\) have nonnegative \(y\)-coordinates. On the other hand, we have \((1,0)\in f(U)\) and with respect to the subspace topology on \(S^1\) any open set containing \((1,0)\) contains elements of \(S^1\) whose \(y\)-coordinate is negative. It follows that \(f(U)\) is not open.
In light of Theorem 1.10.11, it is natural to ask whether given topological spaces \((X_i, \mathcal{T}_i)\) there is a space \(X\) with continuous maps \(X_i\rightarrow X\) that satisfies a dual mapping property to that of the product. Indeed, there is, and we call it the coproduct of the \(X_i\text{.}\)

Topological specimen 13. Coproduct space.

Let \(\{X_i, \mathcal{T}_i\}_{i\in I}\) be a collection of topological spaces. Let \(X\) be the disjoint union \(\coprod_{i\in I}X_i\) of the \(X_i\text{;}\) in more detail, for each \(i\in I\text{,}\) define \(\widetilde{X}_i=X_i\times \{i\}\text{,}\) and define
\begin{equation*} \coprod_{i\in I}X_i=\bigcup_{i\in I}\widetilde{X}_i\text{.} \end{equation*}
For each \(i\in I\) the function
\begin{align*} \iota_i\colon X_i \amp \rightarrow \widetilde{X}_i\\ x \amp \mapsto (x,i) \end{align*}
is a bijection between \(X_i\) and \(\widetilde{X}_i\text{,}\) and it is easy to see that the collection
\begin{equation*} \widetilde{\mathcal{T}}_i=\iota_i(\mathcal{T}_i)=\{U\times\{i\}\colon U\in \mathcal{T}_i\} \end{equation*}
is a topology on \(\widetilde{X}_i\) making the map \(\iota_i\colon X_i\rightarrow \widetilde{X}_i\) a homeomorphism. 1  The coproduct topology on \(X=\coprod_{i\in I}X_i\) is the topology \(\mathcal{T}\) generated by the basis \(\mathcal{B}\) defined as follows:
\begin{equation*} \mathcal{B}=\{\iota_i(U)=U\times\{i\} \colon i\in I, U\in \mathcal{T}_i\} \text{.} \end{equation*}
We call \(X\) together with the topology \(\mathcal{T}\) the coproduct of the \(X_i\text{.}\)

Proof.