General Topology

Topology is the branch of mathematics concerned with properties of spaces that are preserved under continuous deformations — stretching, bending, and twisting, but not tearing or gluing. Often called “rubber-sheet geometry,” it strips away the rigid metrical structure of real analysis and isolates the deeper concept of nearness, giving rise to a remarkably general and powerful framework. From the foundations laid by Georg Cantor and Henri Poincaré in the late nineteenth century through the modern synthesis of the mid-twentieth century, general topology has grown into one of the broadest and most internally coherent areas of pure mathematics, with connections to algebra, analysis, geometry, and theoretical physics.

Topological Spaces and Continuity

The central object of study is the topological space. To arrive at its definition, it helps to notice what the key properties of open sets in real analysis actually are — and then abstract those properties into axioms. Recall that in $\mathbb{R}$, an open set is a set where every point has an open ball contained in the set. Three structural facts hold: the empty set and all of $\mathbb{R}$ are open, any union of open sets is open, and any finite intersection of open sets is open. A topological space takes precisely these three facts as defining axioms.

Formally, a topological space is a pair $(X, \tau)$ where $X$ is a set and $\tau$ is a collection of subsets of $X$ satisfying: (i) $\emptyset \in \tau$ and $X \in \tau$; (ii) an arbitrary union of sets in $\tau$ lies in $\tau$; (iii) any finite intersection of sets in $\tau$ lies in $\tau$. The collection $\tau$ is called a topology on $X$, and its members are called open sets. A subset $C \subseteq X$ is closed if its complement $X \setminus C$ is open.

This definition encompasses an enormous variety of examples. Every metric space $(X, d)$ becomes a topological space when open sets are defined as unions of open balls — this is the metric topology, and it is the primary source of intuition for everything else. At the extremes sit the discrete topology $\tau = 2^X$ (every subset is open) and the indiscrete topology $\tau = \{\emptyset, X\}$ (only the trivial sets are open). The cofinite topology on an infinite set $X$ declares a set to be open if its complement is finite or all of $X$; this example is neither discrete nor metrizable but arises naturally in algebraic geometry as the Zariski topology on $\mathbb{A}^1$.

Continuous functions are defined without any mention of distance. A function $f : X \to Y$ between topological spaces is continuous if the preimage of every open set in $Y$ is open in $X$:

$f \text{ continuous} \;\Longleftrightarrow\; \forall V \in \tau_Y,\; f^{-1}(V) \in \tau_X.$

This condition recovers the familiar $\varepsilon$-$\delta$ definition exactly when $X$ and $Y$ are metric spaces, but it applies in full generality. A continuous bijection $f : X \to Y$ whose inverse is also continuous is called a homeomorphism, and two spaces that admit a homeomorphism between them are homeomorphic — they are, from a topological standpoint, the same space. The central problem of topology is to classify spaces up to homeomorphism.

A basis for a topology is a collection $\mathcal{B}$ of open sets such that every open set is a union of elements of $\mathcal{B}$. Working with a basis is far more convenient than specifying all open sets: the open balls $\{B(x, r) : x \in X, r > 0\}$ form a basis for the metric topology, and the open intervals $(a, b)$ form a basis for the standard topology on $\mathbb{R}$. The interior $\mathrm{int}(A)$ of a set $A$ is the largest open set contained in $A$; the closure $\overline{A}$ is the smallest closed set containing $A$; and the boundary $\partial A = \overline{A} \setminus \mathrm{int}(A)$ captures the edge of $A$. A point $x$ is an accumulation point (or limit point) of $A$ if every open neighbourhood of $x$ contains a point of $A$ other than $x$ itself.

Separation Axioms and Regularity

Different topological spaces exhibit vastly different separation properties — how well the topology can distinguish between distinct points or between points and closed sets. These properties are organized into a hierarchy called the separation axioms, labelled $T_0$ through $T_4$ (the subscript $T$ stands for the German Trennungsaxiom).

A space is $T_0$ (Kolmogorov) if for any two distinct points, at least one has an open neighbourhood not containing the other. This is the weakest separation condition and is often too weak to be useful. A space is $T_1$ (Fréchet) if for any two distinct points, each has an open neighbourhood not containing the other; equivalently, every singleton $\{x\}$ is a closed set. The most important condition in practice is $T_2$, or Hausdorff: any two distinct points can be separated by disjoint open sets. Formally, for every $x \neq y$ there exist open sets $U \ni x$ and $V \ni y$ with $U \cap V = \emptyset$. Felix Hausdorff introduced this axiom in his foundational 1914 textbook Grundzüge der Mengenlehre, arguing that it was necessary to ensure limits of sequences are unique. In a Hausdorff space, if a net (or sequence in a first-countable space) converges, its limit is uniquely determined — a basic sanity condition for analysis.

Moving further up the hierarchy, a regular space (or $T_3$ space) can separate a point from any closed set not containing it by disjoint open sets. A normal space (or $T_4$ space) can separate any two disjoint closed sets by disjoint open sets. The culminating result in this hierarchy is Urysohn’s Lemma, proved by Pavel Urysohn in 1925: a space is normal if and only if for any two disjoint closed sets $A$ and $B$ there exists a continuous function $f : X \to [0,1]$ with $f|_A = 0$ and $f|_B = 1$. Urysohn’s Lemma is a cornerstone result — it provides a way to construct continuous functions from purely topological hypotheses, and it implies the Tietze Extension Theorem: any continuous real-valued function on a closed subspace of a normal space extends to the whole space.

Metric spaces are automatically normal (hence Hausdorff and regular), so for most analysis the separation axioms are a non-issue. But in algebraic geometry, functional analysis, and general topology itself, non-Hausdorff and non-normal spaces arise naturally, and tracking the axioms carefully matters. The metrization problem asks which topological spaces are homeomorphic to metric spaces; the answer is given by the Urysohn metrization theorem (second-countable regular Hausdorff spaces) and the deeper Nagata-Smirnov theorem (regular Hausdorff spaces with a $\sigma$-locally finite basis). Partitions of unity, which exist on paracompact Hausdorff spaces, allow locally defined continuous functions to be patched together globally — a technique central to differential geometry and the theory of manifolds.

Compactness and Connectedness

Compactness is the topological generalization of the finite. A topological space $X$ is compact if every open cover of $X$ has a finite subcover: whenever $X = \bigcup_{\alpha} U_\alpha$ with each $U_\alpha$ open, there exist finitely many indices $\alpha_1, \ldots, \alpha_n$ with $X = U_{\alpha_1} \cup \cdots \cup U_{\alpha_n}$. The idea crystallized through the work of many analysts in the late nineteenth century, with Eduard Heine and Émile Borel identifying the key covering property in $\mathbb{R}^n$. The Heine-Borel theorem characterizes compact subsets of $\mathbb{R}^n$: a subset is compact if and only if it is closed and bounded. In general metric spaces, compactness is equivalent to sequential compactness (every sequence has a convergent subsequence), but these notions diverge in general topological spaces.

Compact spaces behave like finite sets in many respects. A closed subset of a compact space is compact. A compact subset of a Hausdorff space is closed. The continuous image of a compact space is compact — a fact that immediately implies the extreme value theorem: a continuous real-valued function on a compact space attains its maximum and minimum. If $f : X \to Y$ is a continuous bijection with $X$ compact and $Y$ Hausdorff, then $f$ is automatically a homeomorphism, because $f^{-1}$ is automatically continuous.

When a space is not compact, one can sometimes embed it into a compact space. The one-point compactification of a locally compact Hausdorff space $X$ is the space $X^* = X \cup \{\infty\}$ with the topology in which a neighbourhood of $\infty$ is the complement of a compact subset of $X$. The one-point compactification turns the real line into a circle and the plane into a sphere. The far richer Stone-Čech compactification $\beta X$ is the largest compactification in a precise universal sense: every bounded continuous function $f : X \to \mathbb{R}$ extends uniquely to a continuous function $\beta X \to \mathbb{R}$. It plays a fundamental role in functional analysis and model theory.

Connectedness captures the intuitive idea that a space is in one piece. A space $X$ is connected if it cannot be written as a disjoint union of two nonempty open sets; equivalently, the only sets that are simultaneously open and closed (clopen) are $\emptyset$ and $X$ itself. The continuous image of a connected space is connected — this is the topological core of the intermediate value theorem: a continuous function on a connected space cannot jump between values without passing through all intermediate values. A stronger notion is path connectedness: $X$ is path connected if for any two points $x, y \in X$ there exists a continuous path $\gamma : [0,1] \to X$ with $\gamma(0) = x$ and $\gamma(1) = y$. Path connectedness implies connectedness but not vice versa — the topologist’s sine curve $\{(x, \sin(1/x)) : x > 0\} \cup \{(0,0)\}$ is connected but not path connected.

Every topological space decomposes uniquely into maximal connected subsets called connected components. A space is locally connected if every neighbourhood of every point contains a connected neighbourhood; local connectedness ensures that connected components are open sets, which simplifies many arguments.

Homotopy and Fundamental Group

Two continuous maps $f, g : X \to Y$ are homotopic if one can be continuously deformed into the other: there exists a continuous map $H : X \times [0,1] \to Y$ with $H(\cdot, 0) = f$ and $H(\cdot, 1) = g$. The map $H$ is called a homotopy from $f$ to $g$. Homotopy is an equivalence relation on continuous maps, and two spaces $X$ and $Y$ are homotopy equivalent if there exist continuous maps $f : X \to Y$ and $g : Y \to X$ such that $g \circ f$ is homotopic to $\mathrm{id}_X$ and $f \circ g$ is homotopic to $\mathrm{id}_Y$. Homotopy equivalence is coarser than homeomorphism: a disk is homotopy equivalent to a point, and a Möbius strip is homotopy equivalent to a circle, even though these spaces are not homeomorphic.

The fundamental group $\pi_1(X, x_0)$ of a space $X$ at a basepoint $x_0$ is the group whose elements are homotopy classes of loops based at $x_0$ — continuous maps $\gamma : [0,1] \to X$ with $\gamma(0) = \gamma(1) = x_0$ — and whose group operation is concatenation of loops. This construction was introduced by Henri Poincaré in his 1895 Analysis Situs, arguably the founding document of algebraic topology. The fundamental group is a topological invariant: homeomorphic spaces have isomorphic fundamental groups (at corresponding basepoints).

Computing $\pi_1$ for familiar spaces reveals the geometry. The fundamental group of the circle $S^1$ is infinite cyclic:

$\pi_1(S^1) \cong \mathbb{Z},$

where the generator is the class of a loop that winds once around the circle. The fundamental group of the sphere $S^2$ is trivial — every loop on the sphere can be contracted to a point. A space with trivial fundamental group is called simply connected. The torus $T^2 = S^1 \times S^1$ has fundamental group $\mathbb{Z} \times \mathbb{Z}$, reflecting the two independent directions in which loops can wind.

The most powerful computational tool for fundamental groups is the Seifert-van Kampen theorem, formulated in the early 1930s by Herbert Seifert and Egbert van Kampen. If a path-connected space $X = U \cup V$ where $U$, $V$, and $U \cap V$ are all path connected and open, then $\pi_1(X)$ is the free product of $\pi_1(U)$ and $\pi_1(V)$ amalgamated over $\pi_1(U \cap V)$:

$\pi_1(X) \cong \pi_1(U) *_{\pi_1(U \cap V)} \pi_1(V).$

This theorem allows fundamental groups to be computed by cutting a space into simpler pieces, much as inclusion-exclusion works for counting. It shows, for instance, that the wedge sum $S^1 \vee S^1$ has fundamental group the free group on two generators $F_2$.

Covering Spaces

A covering space of a topological space $X$ is a space $\tilde{X}$ together with a continuous surjection $p : \tilde{X} \to X$ such that every point $x \in X$ has an open neighbourhood $U$ with $p^{-1}(U)$ a disjoint union of open sets each mapped homeomorphically onto $U$ by $p$. The archetypal example is the exponential map $p : \mathbb{R} \to S^1$ defined by $p(t) = e^{2\pi i t}$: every small arc of the circle has a preimage consisting of infinitely many disjoint intervals in $\mathbb{R}$, each mapping homeomorphically to that arc.

The fundamental theorem of covering space theory establishes a deep correspondence: for a path-connected, locally path-connected, and semi-locally simply connected space $X$ with basepoint $x_0$, there is a bijection between connected covering spaces of $X$ (up to isomorphism over $X$) and conjugacy classes of subgroups of $\pi_1(X, x_0)$. The correspondence is given by $p : \tilde{X} \to X$ corresponding to the subgroup $p_*(\pi_1(\tilde{X}, \tilde{x}_0)) \leq \pi_1(X, x_0)$. This bijection is a striking analogy with Galois theory, where subgroups of the Galois group correspond to intermediate field extensions; indeed, it is one of the motivating examples for the modern notion of a Galois category.

The universal covering space $\tilde{X}$ corresponds to the trivial subgroup — it is simply connected and covers all other covering spaces. Its deck transformation group (the group of homeomorphisms $\tilde{X} \to \tilde{X}$ lying over the identity on $X$) is isomorphic to $\pi_1(X, x_0)$ itself. This makes the universal cover a powerful tool: results about the simply connected cover often descend to the original space via the deck group action. For instance, the circle’s universal cover is $\mathbb{R}$, whose fundamental group is trivial, and the deck transformations are the integer translations $t \mapsto t + n$, isomorphic to $\mathbb{Z} \cong \pi_1(S^1)$.

The path-lifting property is central: given any path $\gamma : [0,1] \to X$ and a point $\tilde{x}_0 \in p^{-1}(\gamma(0))$, there is a unique lift $\tilde{\gamma} : [0,1] \to \tilde{X}$ with $\tilde{\gamma}(0) = \tilde{x}_0$ and $p \circ \tilde{\gamma} = \gamma$. Homotopy lifting similarly holds. These lifting properties are used to prove the fundamental theorem and to compute $\pi_1$ for many spaces by analyzing their universal covers.

Manifolds and Knot Theory

A topological manifold of dimension $n$ is a Hausdorff, second-countable topological space in which every point has an open neighbourhood homeomorphic to $\mathbb{R}^n$ (or to the closed half-space $\mathbb{R}^n_+ = \{x \in \mathbb{R}^n : x_n \geq 0\}$ for manifolds with boundary). Manifolds are the geometric objects at the heart of modern mathematics and physics. Familiar examples include curves (1-manifolds), surfaces (2-manifolds), and the 3-dimensional space we inhabit (locally a 3-manifold). The $n$-sphere $S^n$, the $n$-torus $T^n$, real projective space $\mathbb{RP}^n$, and complex projective space $\mathbb{CP}^n$ are all compact manifolds of the indicated dimensions.

The classification of compact surfaces (compact 2-manifolds without boundary) is one of the most beautiful theorems in all of topology. Every such surface is homeomorphic to exactly one of: the sphere $S^2$, a connected sum of $g$ tori $T^2 \# \cdots \# T^2$ (the orientable surface of genus $g$), or a connected sum of $k$ real projective planes $\mathbb{RP}^2 \# \cdots \# \mathbb{RP}^2$ (the non-orientable case). The Euler characteristic $\chi = 2 - 2g$ for orientable surfaces provides a complete numerical invariant: the sphere has $\chi = 2$, the torus $\chi = 0$, and higher genus surfaces have $\chi < 0$. This classification was completed in the early twentieth century through the combined work of many mathematicians.

In dimension 3, the landscape is far more complex. The geometrization conjecture, formulated by William Thurston in 1982 and proved by Grigori Perelman in 2003 (a feat that also resolved the century-old Poincaré conjecture), asserts that every compact orientable 3-manifold decomposes along spheres and tori into pieces each admitting one of eight model geometries. The most ubiquitous is hyperbolic geometry; most 3-manifolds, in a precise statistical sense, are hyperbolic.

Knot theory studies embeddings of circles in 3-dimensional space. A knot is a smooth (or piecewise-linear) embedding $K : S^1 \hookrightarrow S^3$; two knots are equivalent if one can be continuously deformed into the other within $S^3$ without crossing itself (ambient isotopy). The simplest knot is the unknot (a trivially embedded circle); the simplest non-trivial knot is the trefoil, which appears in two mirror-image versions. Reidemeister moves — three elementary moves on knot diagrams — generate all ambient isotopies, so knot invariants can be verified by checking they are unchanged under these moves.

Knot invariants are functions on knots that take the same value on equivalent knots. The fundamental group of the knot complement $\pi_1(S^3 \setminus K)$, known as the knot group, is a powerful invariant — it distinguishes the unknot (which has knot group $\mathbb{Z}$) from most non-trivial knots. Polynomial invariants provide more computable alternatives. The Alexander polynomial $\Delta_K(t)$, introduced by James Alexander in 1928, was the first knot polynomial. The Jones polynomial $V_K(t)$, discovered by Vaughan Jones in 1984 using operator algebras, was revolutionary: it distinguishes many knots that the Alexander polynomial cannot, and it connects knot theory to quantum groups and statistical mechanics. The HOMFLY polynomial (1985) generalizes both the Jones and Alexander polynomials in a two-variable family.

The Seifert surface of a knot $K$ is a compact orientable surface embedded in $S^3$ whose boundary is $K$. Its existence, proved by Herbert Seifert in 1934, allows one to define the knot genus — the minimum genus of any Seifert surface — a subtle invariant detecting the knot’s essential complexity. A knot of genus zero is the unknot. The genus is related to the degree of the Alexander polynomial and can be computed via the Thurston norm on the homology of the knot complement.

Dimension Theory

The concept of dimension for topological spaces requires care: naive geometric intuition suggests that $\mathbb{R}^n$ and $\mathbb{R}^m$ have different dimensions, but this is not obvious from the topological definition. Indeed, it was initially surprising when Georg Cantor (1878) showed that the unit interval and the unit square have the same cardinality. Giuseppe Peano’s 1890 construction of a continuous surjection from $[0,1]$ onto $[0,1]^2$ (the Peano curve) further muddied the waters. The resolution came when L.E.J. Brouwer proved in 1911 that $\mathbb{R}^n$ and $\mathbb{R}^m$ are not homeomorphic for $n \neq m$ — the invariance of domain theorem, one of the first deep applications of algebraic topology.

Three independent notions of topological dimension were developed in the early twentieth century and shown to coincide on the class of separable metrizable spaces. The small inductive dimension $\mathrm{ind}(X)$ is defined recursively: $\mathrm{ind}(\emptyset) = -1$; a space has $\mathrm{ind}(X) \leq n$ if every point has a neighbourhood basis of open sets whose boundaries have $\mathrm{ind} \leq n-1$. The large inductive dimension $\mathrm{Ind}(X)$ replaces points by closed sets in the same recursive scheme. The Lebesgue covering dimension $\dim(X)$ measures the minimum order of finite open covers: $\dim(X) \leq n$ if every finite open cover has a refinement in which every point is covered by at most $n+1$ sets. For compact metrizable spaces, all three dimensions agree and coincide with the intuitive dimension: $\dim(\mathbb{R}^n) = n$.

The Hausdorff dimension, introduced by Felix Hausdorff in 1918, is a finer notion that takes non-integer values for fractals. A subset $A \subseteq \mathbb{R}^n$ has Hausdorff dimension $d$ if the $d$-dimensional Hausdorff measure of $A$ is positive and finite. The Cantor middle-thirds set, for example, has Hausdorff dimension $\log 2 / \log 3 \approx 0.631$, capturing the sense in which it is more than a point but less than a curve. Self-similar fractals like the Sierpiński triangle (dimension $\log 3 / \log 2 \approx 1.585$) and the von Koch snowflake (dimension $\log 4 / \log 3 \approx 1.262$) provide further examples. The Hausdorff dimension is not a topological invariant — homeomorphic spaces can have different Hausdorff dimensions when embedded in Euclidean space — but it measures geometric complexity in a way that the topological dimensions cannot.

The interplay between dimension and other topological properties is rich. The product formula $\dim(X \times Y) \leq \dim(X) + \dim(Y)$ holds in general, with equality for compact metrizable spaces. Embedding theorems connect dimension to ambient Euclidean space: every compact $n$-dimensional metrizable space embeds in $\mathbb{R}^{2n+1}$, and this bound is sharp. The Menger-Nöbeling theorem sharpens this to show that every $n$-dimensional compact metrizable space embeds in the $(2n+1)$-dimensional Menger cube. These embedding results reflect the deep interplay between topology, dimension theory, and the geometry of Euclidean space, and they continue to influence contemporary research in topological data analysis, where the intrinsic dimension of a data cloud is estimated using persistent homology and related invariants drawn from the algebraic topology of metric spaces.