Homeomorphisms

Homeomorphism is the backbone concept of topology as a mathematical discipline, and it radiates outward into many fields. In geometric topology and low-dimensional topology, classification problems (which surfaces, knots, or manifolds are "the same") are literally questions of homeomorphism (or the finer notion of diffeomorphism). In complex analysis, the Riemann mapping theorem produces homeomorphisms (indeed biholomorphisms) between simply connected planar domains and the unit disk, underlying conformal mapping techniques used in fluid dynamics and electrostatics. In data analysis, topological data analysis and persistent homology use homeomorphism-invariant features to detect the "shape" of high-dimensional data robust to noisy deformation. In physics, configuration spaces and phase spaces are often studied up to homeomorphism to classify qualitatively distinct dynamical regimes, and in condensed matter physics, topological phases of matter are classified by invariants unchanged under continuous (homeomorphic) deformations of the underlying Hamiltonian's band structure. In computer graphics and mesh processing, algorithms that need to compare or morph 3D shapes rely on detecting homeomorphisms (or approximate ones) between meshes. And pedagogically, homeomorphism sets up the entire subsequent machinery of Topological Invariants and The Fundamental Group, since these tools exist precisely to distinguish spaces that are not homeomorphic when open-set arguments alone cannot decide the question.

Ask the tutor: Why does the compact–Hausdorff shortcut theorem fail if we only assume $Y$ is metrizable but not necessarily Hausdorff in the relevant sub-argument, and can you construct an explicit example showing the failure?

Think of continuity as a one-way promise: if points of $X$ are close, their images in $Y$ are close (in the precise open-set sense). A homeomorphism upgrades this to a two-way promise, valid in both directions simultaneously. This is why homeomorphic spaces are considered topologically identical: any statement about $X$ that can be phrased using only open sets translates, word for word, into a true statement about $Y$, and vice versa. The classical slogan is that a coffee mug and a donut (solid torus with a handle-shaped indentation, more precisely a torus surface) are homeomorphic, because one can be continuously deformed into the other without cutting or gluing. Deformability by bending is a sufficient — but not necessary — condition for homeomorphism; the actual definition never mentions bending at all. It is purely about open sets and inverses. This distinction matters: some homeomorphic pairs (like a square and a circle) look nothing alike geometrically, and some non-homeomorphic pairs might seem "deformable" at a glance but are secretly obstructed (a circle is never homeomorphic to a line, because removing one point disconnects the line but not the circle). The first figure below depicts the open-interval-to-real-line homeomorphism $\tan$, illustrating that boundedness is not a topological invariant — a bounded space can be homeomorphic to an unbounded one. The second figure depicts the classic failure of bicontinuity: a continuous bijection from a half-open interval to a circle whose inverse is discontinuous at the gluing point.

This confirms that "homeomorphic to" partitions the class of topological spaces into equivalence classes — exactly the sense in which topology only cares about a space "up to homeomorphism." The next theorem addresses the most common practical difficulty: checking condition (3) directly is often painful, so we want conditions under which it is automatic.

Example 1 (an explicit, asymmetric homeomorphism of subspaces of $\mathbb{R}^2$). Let $X = \{(x,y) \in \mathbb{R}^2 : x^2 + y^2 = 1\}$ be the unit circle, and let

be a "diamond" (a rhombus with horizontal diagonal length $2$ and vertical diagonal length $1$) — deliberately asymmetric in scale so the map is not a trivial rotation. We construct an explicit homeomorphism $f: X \to Y$. Define $f$ radially: for $p = (\cos\theta, \sin\theta) \in X$, let

This is well-defined since the denominator $|\cos\theta| + 2|\sin\theta|$ is never zero (it vanishes only if both $\cos\theta=0$ and $\sin\theta=0$, impossible). Write $r(\theta) = |\cos\theta|+2|\sin\theta| > 0$; note $r$ is continuous in $\theta$ as a sum of continuous functions (absolute values of continuous functions are continuous). $f$ is continuous. Each coordinate of $f(p)$ is a quotient of continuous functions of $(x,y)=(\cos\theta,\sin\theta)$ with nonvanishing denominator, hence continuous, by the algebra of continuous real-valued functions (sums, products, and quotients — with nonvanishing denominator — of continuous functions are continuous). $f$ maps into $Y$. Writing $f(p) = (x/r, y/r)$ with $(x,y)=(\cos\theta,\sin\theta)$, we compute

so $f(p) \in Y$. $f$ is a bijection onto $Y$. Given $q \in Y$, write $q = (a,b)$ with $|a|+2|b|=1$; then $q \neq (0,0)$ (else $0=1$), so $q$ determines a unique angle $\theta$ with $(a,b) = t(\cos\theta,\sin\theta)$ for some $t>0$ (polar decomposition of a nonzero vector, unique with $\theta \in [0,2\pi)$). Then $p=(\cos\theta,\sin\theta) \in X$ satisfies $f(p) = q$, since $f(p)$ is the unique point on ray $\theta$ satisfying the diamond's equation, and $q$ is that point. This shows surjectivity; injectivity follows because $f(p)=f(p')$ forces $p,p'$ to lie on the same ray from the origin (as $f$ preserves the direction/angle $\theta$ up to positive scaling) and both lie on the unit circle, forcing $p=p'$. Conclusion via the Corollary. $X$ is the continuous image of $[0,2\pi]$ under $\theta \mapsto (\cos\theta,\sin\theta)$ hence compact (continuous image of a compact interval), and $Y \subset \mathbb{R}^2$ is Hausdorff (subspace of a metric space). By the Corollary above, since $f$ is a continuous bijection from the compact space $X$ to the Hausdorff space $Y$, $f$ is automatically a homeomorphism — we need not separately verify continuity of $f^{-1}$. This example shows homeomorphism is a genuinely flexible notion: a perfectly round circle and a lopsided rhombus, which look geometrically quite different (different curvature, corners versus smoothness, different symmetry group), are topologically identical.

Example 2 (a case where the compact–Hausdorff shortcut does not apply, requiring direct verification). Let $X = (0,1)$ with the usual topology (not compact!) and let $Y = (1,\infty)$ with the usual topology. Define

$f$ is a bijection. If $x \in (0,1)$ then $1/x > 1$, so $f(x) \in Y$. Given $y \in (1,\infty)$, $x = 1/y \in (0,1)$ and $f(x)=y$, giving surjectivity. If $f(x_1)=f(x_2)$ then $1/x_1 = 1/x_2$ forces $x_1=x_2$, giving injectivity. $f$ is continuous. $f(x) = 1/x$ is continuous on $(0,1)$ since it is a quotient of continuous functions ($1$ and $x$) with nonvanishing denominator on that domain. $f^{-1}$ is continuous. We compute $f^{-1}(y) = 1/y$ for $y \in (1,\infty)$ directly: if $f(x) = y$ then $x = 1/y$. This is again a quotient of continuous functions with nonvanishing denominator on $(1,\infty)$, hence continuous. Since here $X = (0,1)$ is not compact, the compact–Hausdorff shortcut theorem does not apply, and we were obligated to check continuity of $f^{-1}$ by hand — which we did directly, since in this case $f^{-1}$ has the same simple algebraic form as $f$ itself. This illustrates why the shortcut theorem's hypotheses matter operationally, not just theoretically: dropping compactness removes the free pass, though the map may still happen to be a homeomorphism (as here) — one just cannot conclude it for free.