Skip to content
🔬 Scientist Pure mathematics — no exam syllabus

Topology — The Mathematics of Continuity

Topology studies properties that remain unchanged under continuous deformations — stretching, bending, twisting, but not tearing or gluing. It asks what it means for two shapes to be "the same" without measuring distances. To a topologist, a coffee cup and a donut are the same shape (both have one hole). Topology provides the language for continuity across all of mathematics — from analysis to geometry to data science — and its deepest question, the Poincaré conjecture, was only solved in 2003.

"Topology is the study of those properties of geometric figures that persist even when the figures are subjected to deformations so drastic that all their metric and projective properties are lost."
— James W. Alexander

🧭 Open Sets, Topologies, and the Shape of Space

When Distance Doesn't Matter
Imagine a world made of rubber — you can stretch it, bend it, twist it, but you can't cut it or glue parts together. In this world, a circle and a square are the same shape. The triangle and the pentagon are the same. But a figure-8 is different from a circle — because to turn a circle into a figure-8, you'd have to glue two points together. Topology formalizes this intuition. Instead of distances, it studies open sets: collections of points that define "nearness" without measuring. A topological space is just a set X together with a collection of subsets (called open sets) that satisfy three simple rules: the empty set and X are open; any union of open sets is open; any finite intersection of open sets is open. That's it — three rules, and suddenly you have a language for continuity, convergence, and the shape of space.
Topology strips away measurement and keeps only the essential structure of "what's near what."

Metric Spaces — The First Topology

The most intuitive way to get a topology is from a metric — a distance function d(x,y) satisfying positivity, symmetry, and the triangle inequality. The open balls B(x,r) = { y : d(x,y) < r } generate the metric topology. Euclidean space ℝⁿ with the usual distance is the canonical example, but metrics also appear on function spaces, graphs, and even the space of DNA sequences (edit distance). Every metric space is a topological space, but not every topological space comes from a metric — those that do are called metrizable.

Topological Space
(X, τ) with τ ⊆ 𝒫(X)
Metric Space
d(x,y) ≥ 0, symmetric, Δ
Open Ball
B(x,r) = { y : d(x,y) < r }
Discrete Topology
Every subset is open
✨
The Standard Topology on ℝ ▶
On the real line, the standard topology has open sets that are unions of open intervals (a,b). This comes from the metric d(x,y) = |x − y|.

Key observation: In this topology, (0,1) is open, but [0,1] is not (because the endpoints don't have any open ball fully inside the set). The set (0,1) ∪ (2,3) is open — it's a union of open intervals. The set (0,1] is neither open nor closed — it's a half-open interval.

A set can be open, closed, both (clopen), or neither. In the discrete topology, every set is clopen. In a connected space like ℝ, the only clopen sets are the empty set and the whole space.

Bases and Subbases

A basis for a topology is a collection of open sets such that every open set is a union of basis elements. The open intervals (a,b) form a basis for the standard topology on ℝ. The open balls form a basis for any metric topology. A subbasis is an even smaller generating set — finite intersections of subbasis elements give a basis. This machinery lets us define topologies efficiently without listing every open set.

Product Topology
Given two topological spaces X and Y, the product topology on X×Y has a basis of sets U×V where U is open in X and V is open in Y. For ℝ² = ℝ×ℝ, the basis is open rectangles (a,b)×(c,d). These generate the same topology as open balls, so the product topology matches the usual Euclidean topology.

🌉 Connectedness and Compactness — The Two Pillars

What Makes a Space "One Piece" and "Finite"?
Two properties dominate topology more than any others. Connectedness captures the intuitive idea that a space is "all one piece" — it cannot be split into two disjoint nonempty open subsets. The interval [0,1] is connected; the set [0,1]∪[2,3] is not. A space that is connected but not "path-connected" is a weird intermediate — the topologist's sine curve { (x, sin(1/x)) : x > 0 } ∪ { (0,y) : |y| ≤ 1 } is connected (because points on the curve accumulate along the vertical segment), but no path connects a point on the curve to a point on the segment. Compactness captures a different intuition: "finite" in a topological sense. A space is compact if every open cover has a finite subcover. In ℝⁿ, the Heine-Borel theorem says this is exactly closed and bounded sets. Compactness turns out to be the topological version of finiteness — and it's the key to almost every deep theorem in analysis.
Connectedness is about wholeness; compactness is about manageability. Together they define the spaces we can truly understand.

Connected vs Path-Connected

A space is connected if it's not the disjoint union of two nonempty open subsets. It's path-connected if any two points can be joined by a continuous path. Path-connected implies connected, but not vice versa (as the topologist's sine curve shows). Path components are equivalence classes under "there exists a path between them." For most nice spaces (manifolds, CW complexes), the two notions coincide.

Intermediate Value Theorem (Topological Version)
The continuous image of a connected space is connected
This is why any continuous function f: [0,1] → ℝ that takes values f(0) = a and f(1) = b must take every value between a and b.

Compactness — The Heine-Borel Theorem

In ℝⁿ, a set is compact iff it's closed and bounded. But in general topology, compactness is more subtle. The Heine-Borel theorem fails in infinite-dimensional spaces — the closed unit ball in an infinite-dimensional Hilbert space is not compact (it's not "finite enough"). A continuous function on a compact set achieves its maximum and minimum (extreme value theorem). The Bolzano-Weierstrass property (every sequence has a convergent subsequence) is equivalent to compactness in metric spaces, but not in general topological spaces, where sequential compactness and compactness diverge.

Connected
No nontrivial clopen sets
Path-connected
∀x,y ∃ path p with p(0)=x, p(1)=y
Compact
Every open cover has finite subcover
Heine-Borel
In ℝⁿ: compact ⇔ closed & bounded
✨
Compactness in Function Spaces — Arzelà–Ascoli ▶
In analysis, we often need to extract convergent subsequences of functions. The Arzelà–Ascoli theorem says a set of continuous functions on a compact set is compact (in the uniform topology) iff they're uniformly bounded and equicontinuous.

Example: The set { fₙ(x) = sin(nx) } on [0,2π] is uniformly bounded but not equicontinuous — the functions oscillate too rapidly. No subsequence converges uniformly. But { fₙ(x) = xⁿ } on [0,1] is also not equicontinuous at x=1, and indeed the limit function is discontinuous.

This theorem is the foundation of the existence of solutions to differential equations (Picard–Lindelöf) and the calculus of variations.

🔄 Continuous Functions and When Spaces Are "the Same"

Homeomorphisms — The Rubber-Sheet Equivalence
In topology, a homeomorphism is a continuous bijection with a continuous inverse. Two spaces that are homeomorphic are topologically indistinguishable — they have exactly the same topological properties. The classic joke: a topologist can't tell the difference between a coffee cup and a donut. Both have exactly one hole, so they're homeomorphic. A circle is homeomorphic to a square, an ellipse, or any simple closed curve. But a circle is NOT homeomorphic to a figure-8 (which has a point where removing it disconnects the space into 3 components, whereas removing any point from a circle gives 2 components — this is a topological invariant). The fundamental problem of topology: given two spaces, are they homeomorphic? This question drives the entire field.
To a topologist, shape is about connectivity, not geometry. Two shapes are the same if one can be deformed into the other without tearing or gluing.

Topological Invariants

A topological invariant is a property preserved under homeomorphisms. Connectedness, compactness, and the number of holes (genus) are invariants. The Euler characteristic V − E + F is an invariant for surfaces. If two spaces differ on an invariant, they cannot be homeomorphic. The search for invariants is the central activity of topology — from the fundamental group to homology to characteristic classes. The invariance of dimension theorem states that ℝᵐ and ℝⁿ are not homeomorphic for m ≠ n — a result that seems obvious but was notoriously difficult to prove.

The Circle vs the Figure-8
Consider the circle S¹ and the figure-8 space (two circles joined at a point). Remove a single point from S¹: the remainder is connected (it's an open interval). Remove the crossing point from the figure-8: the remainder has 3 connected components (two open intervals and the crossing point removal). Since removing a point changes connectivity differently, they cannot be homeomorphic. This is a point-set topological invariant.

Embeddings and Quotients

An embedding is a homeomorphism onto its image — it places one space inside another. The circle embeds in ℝ² (naturally) and in ℝ³ (as a knot, which is a whole subfield). The Möbius strip embeds in ℝ³ but not in ℝ². A quotient space identifies points of a space to create a new one — gluing the ends of [0,1] gives a circle; gluing the top and bottom edges of a square gives a cylinder; gluing both pairs of opposite edges gives a torus. Quotients are how we build all surfaces from simple pieces.

Classification of 2-Manifolds
Every connected compact 2-manifold is homeomorphic to a sphere with g handles (genus g) or, for non-orientable surfaces, a sphere with g cross-caps
The genus g (number of holes) is the only invariant needed to classify orientable surfaces.

🌌 Deep Topology — Manifolds, the Poincaré Conjecture, and Beyond

The Poincaré Conjecture — Solved After 100 Years
In 1904, Henri Poincaré asked: if a closed 3-dimensional space has the property that every loop can be continuously shrunk to a point (it's simply connected), is it necessarily homeomorphic to the 3-sphere (the 3D analog of a sphere's surface)? This seems obvious in 2 dimensions — a simply connected closed surface is always a sphere — but in 3 dimensions, the problem stumped mathematicians for a century. Grigori Perelman proved it in 2003 using Ricci flow, a technique that smooths out curvature over time. He was awarded the $1M Clay Millennium Prize but declined it. The Poincaré conjecture was the first of the seven Millennium Problems to be solved. Its higher-dimensional analogs (n ≥ 5) were solved by Smale (1961) and Freedman (1982), leaving only dimension 3 as the hardest case.
Sometimes the most intuitive-sounding statements in mathematics are the hardest to prove — and the simplest dimensions are the most complex.

Manifolds — Local Is Not Global

An n-dimensional manifold is a topological space that locally looks like ℝⁿ: every point has a neighborhood homeomorphic to an open ball in ℝⁿ. The surface of the Earth is a 2-manifold (it locally looks flat — that's why maps work). The universe may be a 3-manifold. The sphere Sⁿ is the set of points at distance 1 from the origin in ℝⁿ⁺¹. The torus Tⁿ is the product of n circles. Manifolds are the primary objects of study in geometry and topology, and they appear everywhere from general relativity (spacetime is a 4-manifold) to robotics (the configuration space of a robot arm is a manifold).

n-Manifold
Locally ℝⁿ, Hausdorff, second-countable
Sphere Sⁿ
{ x∈ℝⁿ⁺¹ : ‖x‖ = 1 }
Torus Tⁿ
S¹ × S¹ × ... × S¹
Genus g Surface
Sphere with g handles

The Fundamental Group — Detecting Holes Algebraically

The fundamental group π₁(X) is a group whose elements are loops in X (starting and ending at a base point), considered up to continuous deformation (homotopy). The group operation is concatenation of loops. For the circle S¹, π₁(S¹) ≅ ℤ — each loop wraps around some integer number of times. For the sphere S², π₁(S²) = 0 (every loop shrinks to a point). For the torus, π₁(T²) ≅ ℤ×ℤ (loops can wind around either hole independently). The fundamental group is the first homotopy group, and it's one of the most powerful invariants. The Poincaré conjecture asks whether a simply connected (π₁ = 0) closed 3-manifold must be S³ — a question entirely about the fundamental group.

✨
Computing π₁ of a Graph ▶
A graph is a 1-dimensional CW complex. Its fundamental group is a free group — one generator for each independent cycle.

Example: A tree is contractible (π₁ = 0). A graph with one cycle (like a circle) has π₁ ≅ ℤ. The figure-8 graph (two cycles sharing a point) has π₁ ≅ F₂ — the free group on two generators. The generators are loops around each lobe, and words like ab⁻¹a²b represent paths that wind around both lobes in sequence.

Unlike the torus's ℤ×ℤ (which is abelian), free groups are non-abelian — winding around one lobe then the other is different from the reverse order. This non-commutativity captures the essential complexity of the space.

Homology and the Euler Characteristic

While homotopy groups capture the full structure of loops and higher-dimensional spheres, homology groups are easier to compute and give Abelian invariants. The nth homology group Hₙ(X) counts n-dimensional holes: H₀ counts connected components, H₁ counts 1-dimensional holes (loops), H₂ counts 2-dimensional holes (voids), and so on. The Euler characteristic χ(X) = Σ (−1)ⁿ rank(Hₙ) generalizes the formula V − E + F for polyhedra. For a surface of genus g, χ = 2 − 2g. This single number encodes the topology of the entire surface.

Gauss–Bonnet Theorem
For a compact 2-manifold, ∫_M K dA = 2π χ(M)
The total curvature of a surface is determined entirely by its topology — the Euler characteristic. Geometry and topology meet in this beautiful equation.

Applied Topology — TDA and Beyond

In the 21st century, topology has become a practical tool. Topological Data Analysis (TDA) uses persistent homology to study the shape of data — clustering, connectivity, and holes in point clouds at multiple scales. The mapper algorithm creates graph summaries of high-dimensional data. Topology also appears in robotics (configuration spaces), sensor networks (coverage holes), neuroscience (neural activity patterns), and materials science (the topology of atomic structures). The same invariants that classify surfaces now classify patterns in data.

"Topology is the study of qualitative properties of geometric objects. It's the branch of mathematics where a coffee cup is a donut, and a sheet of paper with a twist is a Möbius strip."
— Popular mathematics saying

🎯 Practice — Check Your Topology Intuition

These puzzles build topological intuition — no heavy math needed.

← Back to Math Map

Covering Spaces and the Classification Theorem

A covering space of a topological space X is a space \~X together with a continuous surjection p: \~X → X such that every point x ∈ X has an open neighbourhood U whose preimage p⁻¹(U) is a disjoint union of open sets each mapped homeomorphically onto U by p. The classic example is the exponential map ℝ → S¹, t ↦ e2π i t, which makes ℝ the universal cover of the circle. The group of deck transformations (homeomorphisms \~X → \~X commuting with p) is isomorphic to the fundamental group π₁(X) when \~X is simply connected. The classification theorem states that for a path‑connected, locally path‑connected, semilocally simply connected space X, there is a bijection between isomorphism classes of connected covering spaces of X and conjugacy classes of subgroups of π₁(X). The universal cover corresponds to the trivial subgroup; the cover corresponding to a subgroup H ≤ π₁(X) has fundamental group isomorphic to H. This correspondence turns algebraic problems about subgroups into geometric problems about covering spaces, and it underlies the proof that every subgroup of a free group is free (the Nielsen–Schreier theorem). In practice, constructing the cover for a given subgroup often uses the Cayley graph of the group with respect to a generating set, where vertices are cosets and edges correspond to generators.

Higher Homotopy Groups and the Hurewicz Theorem

While the fundamental group π₁ captures 1‑dimensional holes, higher homotopy groups πₙ(X) for n ≥ 2 are defined using based maps from the n‑sphere Sⁿ into X, with homotopy as equivalence. Unlike π₁, all higher homotopy groups are abelian. For spheres, πₙ(Sⁿ) ≅ ℤ, generated by the identity map, while πₖ(Sⁿ) = 0 for k < n. The Hurewicz theorem provides a bridge to homology: if X is (n‑1)‑connected (i.e., πᵢ(X) = 0 for i < n), then the Hurewicz homomorphism h: πₙ(X) → Hₙ(X) is an isomorphism, and for i = n+1 it is an epimorphism. This allows computation of the first non‑trivial homotopy group via homology, which is often easier. For example, π₂(S²) ≅ H₂(S²) ≅ ℤ, and π₃(S²) ≅ ℤ (the Hopf fibration). The theorem also implies that a simply connected space with trivial homology groups up to dimension n is contractible up to that dimension (Whitehead's theorem). In algebraic topology exams, a typical question asks to compute π₃(S²) using the long exact sequence of the Hopf fibration S¹ → S³ → S² together with the Hurewicz isomorphism for S³.

Technique
Use the Hurewicz theorem and the long exact sequence of a fibration to compute π₃(S²).
The Hopf fibration gives a fiber sequence S¹ → S³ → S². The associated long exact sequence in homotopy contains … → π₃(S¹) → π₃(S³) → π₃(S²) → π₂(S¹) → … . Since π₃(S¹) = 0 and π₂(S¹) = 0, the map π₃(S³) → π₃(S²) is an isomorphism. By Hurewicz, π₃(S³) ≅ H₃(S³) ≅ ℤ. Hence π₃(S²) ≅ ℤ.
Strategy
When a problem asks for a higher homotopy group of a sphere or a fibration, first check whether the Hurewicz theorem applies (the space is sufficiently connected). Then write down the relevant long exact sequence of the fibration; many terms vanish because spheres have trivial homotopy groups in low dimensions. The remaining isomorphism often reduces the computation to a known homology group.

Cohomology Rings and the Cup Product

Cohomology groups Hⁿ(X; R) with coefficients in a ring R carry a graded‑commutative product called the cup product, ∪: Hᵖ(X) × Hᑫ(X) → Hᵖ⁺ᑫ(X). This turns the direct sum ⊕ₙ Hⁿ(X) into a ring, the cohomology ring H*(X). The cup product distinguishes spaces that have identical homology groups but different ring structures; for instance, the torus T² and the wedge S¹ ∨ S¹ ∨ S² have the same homology groups but different cup products (the torus has a non‑zero product in degree 2, the wedge does not). The Künneth formula for cohomology expresses H*(X × Y) as the tensor product of the cohomology rings of X and Y, up to a Tor term that vanishes over a field. Poincaré duality for a closed oriented n‑manifold M states that the cap product with the fundamental class [M] gives an isomorphism Hᵏ(M) ≅ Hₙ₋ₖ(M). In terms of cup product, this means the pairing Hᵏ(M) × Hⁿ⁻ᵏ(M) → Hⁿ(M) ≅ R is non‑degenerate. The ring structure is also essential for characteristic classes: Stiefel‑Whitney, Chern, and Pontryagin classes live in cohomology rings and their products encode geometric information about vector bundles. In examinations, a common task is to compute the cohomology ring of a product space such as ℂP² × S² using the known rings H*(ℂP²) ≅ ℤ[x]/(x³) (|x| = 2) and H*(S²) ≅ ℤ[y]/(y²) (|y| = 2), then apply the Künneth theorem to obtain H*(ℂP² × S²) ≅ ℤ[x, y]/(x³, y²).

Morse Theory and Critical Points

Morse theory relates the topology of a smooth manifold M to the critical points of a smooth function f: M → ℝ. A critical point p is non‑degenerate if the Hessian matrix of second derivatives at p is invertible; its index is the number of negative eigenvalues. The Morse lemma states that near a non‑degenerate critical point of index λ, there are coordinates in which f(x) = f(p) − x₁² − … − x_λ² + xλ+1² + … + x_n². The fundamental theorem of Morse theory says that as one passes a critical value, the sublevel set Mᵃ = f⁻¹(−∞, a] changes by attaching a λ‑cell. Consequently, M has the homotopy type of a CW complex with one cell of dimension λ for each critical point of index λ. The Morse inequalities relate the number of critical points of each index to the Betti numbers: for each k, the number of index‑k critical points is at least the k‑th Betti number bₖ. A perfect Morse function achieves equality, giving a minimal cell decomposition. The classic example is the height function on the torus standing vertically, which has four critical points (minimum, two saddles, maximum) matching the Betti numbers b₀ = 1, b₁ = 2, b₂ = 1. Morse theory also underlies the proof of the h‑cobordism theorem and the classification of high‑dimensional manifolds. In competitive exams, a typical question asks to determine the minimal number of critical points of a smooth function on a given manifold using the Morse inequalities and the Euler characteristic χ = Σ (−1)ᵏ bₖ.

Fixed Point Theorems: Brouwer, Lefschetz, and Applications

The Brouwer fixed point theorem states that every continuous map f: Dⁿ → Dⁿ from the closed n‑ball to itself has a fixed point. The proof uses homology: if f had no fixed point, one could construct a retraction r: Dⁿ → Sⁿ⁻¹, which would induce a surjection on Hₙ₋₁, contradicting Hₙ₋₁(Dⁿ) = 0. The Lefschetz fixed point theorem generalises this to compact polyhedra (or more generally compact ENRs): for a continuous map f: X → X, the Lefschetz number L(f) = Σ (−1)ᵏ tr(f*: Hₖ(X; ℚ) → Hₖ(X; ℚ)). If L(f) ≠ 0, then f has a fixed point. For the identity map, L(id) = χ(X), so any space with non‑zero Euler characteristic forces a fixed point for every map homotopic to the identity. The Lefschetz‑Hopf theorem refines this by expressing L(f) as a sum of indices of isolated fixed points. Applications include the Borsuk‑Ulam theorem (every continuous map Sⁿ → ℝⁿ identifies a pair of antipodal points) and the existence of equilibria in game theory (Nash equilibrium via Kakutani's fixed point theorem, a set‑valued generalisation). In examinations, a standard problem is to compute the Lefschetz number of a map on the torus T² induced by a linear map A ∈ GL(2, ℤ) on the universal cover ℝ²; the trace on H₁ is tr(A), on H₀ and H₂ it is 1, so L(f) = 2 − tr(A) + det(A). If this number is non‑zero, the map must have a fixed point.

  • Confusing the classification of covering spaces with the classification of subgroups of the fundamental group; the correspondence is up to conjugacy, not equality.
  • Assuming higher homotopy groups are non‑abelian; they are always abelian for n ≥ 2.
  • Neglecting the graded‑commutativity sign rule in cup product calculations, leading to incorrect ring structures.
  • Applying Morse inequalities without checking that the function is Morse (all critical points non‑degenerate).
  • Forgetting that the Lefschetz number uses rational coefficients; torsion in homology does not contribute.
  • Misidentifying the index of a critical point in Morse theory; the index equals the number of negative eigenvalues of the Hessian, not the dimension of the unstable manifold.
  • Overlooking that the Brouwer fixed point theorem fails for non‑compact or non‑convex domains, a common trap in applied problems.