03.12.10 · modern-geometry / homotopy

CW complex

shipped3 tiersLean: none

Anchor (Master): Whitehead 1949 — *Combinatorial homotopy I* (originator paper, Bull. AMS 55); Hatcher §0 and §4.1; tom Dieck *Algebraic Topology* §8

Intuition [Beginner]

A CW complex is a space built up one dimension at a time by gluing solid balls onto what is already there. Start with a discrete set of points (the 0-cells). Glue line segments (1-cells) to those points by their endpoints. Glue solid disks (2-cells) onto the 1-skeleton along their boundary circles. Glue solid 3-balls along their boundary 2-spheres. Keep going. Whatever space the gluing produces is a CW complex.

The picture is exactly how children build models from dots, sticks, and patches: every interior of every cell is unconstrained, and all the topology lives in how the boundaries get attached. A circle is one point with one arc glued by both endpoints to that point. A sphere is one point with one disk glued by collapsing its whole boundary to that point. A torus is one point, two arcs, and one square whose boundary spells $ab a^{- 1} b^{- 1}$ .

The reason this construction matters: nearly every space studied in algebraic topology can be given a CW structure, and many computations (homotopy groups, homology, cohomology) reduce to bookkeeping over the cells. The cells are the combinatorial data; the attaching maps are the geometry.

Visual [Beginner]

A schematic showing the inductive build-up: a row of dots (the 0-skeleton), arcs joining them (the 1-skeleton), disks pasted onto the resulting graph (the 2-skeleton), and an arrow indicating the process continues into higher dimensions.

Every disk's boundary circle is mapped into the previous skeleton; the interior of the disk becomes a new open cell.

Worked example [Beginner]

Build the 2-sphere $S^{2}$ as a CW complex with two cells: one 0-cell and one 2-cell.

Step 1. The 0-skeleton is one point, call it $p$ . Step 2. There are no 1-cells, so the 1-skeleton is still ${p}$ . Step 3. Take a single 2-disk $D^{2}$ . Its boundary is the circle $S^{1}$ . Glue this boundary to the 0-skeleton via the only available map: send every point of the boundary circle to $p$ . Step 4. The result is a single point with a disk attached so that its whole boundary collapses to that point. That quotient is exactly $S^{2}$ — the disk's interior becomes the 2-sphere minus the north pole, and $p$ is the north pole.

What this tells us: $S^{2}$ is captured by the data "one 0-cell, one 2-cell, with the 2-cell's boundary collapsed to a point." Two cells, four lines of description, the entire sphere. Higher spheres $S^{n}$ have the same minimal CW structure: one 0-cell and one $n$ -cell.

Check your understanding [Beginner]

Formal definition [Intermediate+]

A CW complex is a topological space $X$ together with a filtration

\emptyset = X^{(- 1)} \subseteq X^{(0)} \subseteq X^{(1)} \subseteq \dots \subseteq X^{(n)} \subseteq \dots \subseteq X

by closed subspaces, called the skeleta, such that:

The $0$ -skeleton $X^{(0)}$ is a discrete topological space.
For each $n \geq 1$ , $X^{(n)}$ is obtained from $X^{(n - 1)}$ by attaching $n$ -cells. That is, there is an indexing set $J_{n}$ and a continuous attaching map $Φ_{n} : ⨆_{α \in J_{n}} S_{α}^{n - 1} \to X^{(n - 1)}$ such that $X^{(n)} = X^{(n - 1)} \cup_{Φ_{n}} α \in J_{n} ⨆ D_{α}^{n},$ the pushout in $Top$ along $Φ_{n}$ .
$X = ⋃_{n} X^{(n)}$ as a set.
$X$ carries the weak topology: a subset $A \subseteq X$ is closed iff $A \cap X^{(n)}$ is closed in $X^{(n)}$ for every $n$ . Equivalently, $X = colim X^{(n)}$ in $Top$ .

Each component of $X^{(n)} ∖ X^{(n - 1)}$ that comes from one disk $D_{α}^{n}$ is an open $n$ -cell $e_{α}^{n}$ . The composite $Φ_{α} : D_{α}^{n} ↪ ⨆ D^{n} \to X^{(n)}$ is the characteristic map of the cell $e_{α}^{n}$ ; its restriction to the boundary is the attaching map $ϕ_{α} = Φ_{α} ∣_{S^{n - 1}}$ .

The defining cellular pushout square for each $n$ is

⨆_{α} S_{α}^{n - 1} ↓ ⨆_{α} D_{α}^{n} Φ_{n} \overset{ˉ}{Φ}_{n} X^{(n - 1)} ↓ X^{(n)}

The two letters in the name "CW" record the original axioms of Whitehead 1949: C for closure-finiteness (the closure of every cell meets only finitely many cells) and W for weak topology (the colimit topology on the skeleton filtration). For finite-dimensional CW complexes both conditions are automatic; for infinite-dimensional ones they are the definitions that make the inductive limit well-behaved.

A CW pair $(X, A)$ is a CW complex $X$ together with a subcomplex $A$ , meaning $A$ is closed in $X$ and is a union of cells of $X$ .

Counterexamples to common slips

The Hawaiian earring (the union in $R^{2}$ of circles of radius $1/ n$ all tangent at the origin) is not a CW complex: any CW structure would force the wedge point to have a neighbourhood meeting only finitely many circles.
A subset $Y \subseteq X$ that is closed and union-of-cells as a set still requires the CW structure on $X$ to restrict cleanly to $Y$ (closure-finiteness for $Y$ 's own cells, weak topology) — the term "subcomplex" packages all of these.
The weak topology on $X$ may differ from any metric topology even when each $X^{(n)}$ is metrisable; infinite CW complexes are not in general metrisable.

Key theorem with proof [Intermediate+]

Theorem (homotopy extension property for CW pairs). Let $(X, A)$ be a CW pair. Then the inclusion $A ↪ X$ has the homotopy extension property: for every space $Y$ , every continuous map $f : X \to Y$ , and every homotopy $h : A \times [0, 1] \to Y$ with $h_{0} = f ∣_{A}$ , there is a homotopy $H : X \times [0, 1] \to Y$ extending both $f$ (at time $0$ ) and $h$ (on $A$ ).

Proof. Equivalently, we show that $A \times [0, 1] \cup X \times {0}$ is a retract of $X \times [0, 1]$ ; once we have a retraction $r$ , the desired extension is $H = (h \cup f) \circ r$ .

Step 1 (one cell). For a single cell attachment, $X^{(n)} = X^{(n - 1)} \cup_{ϕ} D^{n}$ , the pair $(D^{n}, S^{n - 1})$ admits an explicit retraction of $D^{n} \times [0, 1]$ onto $S^{n - 1} \times [0, 1] \cup D^{n} \times {0}$ : project radially from the point $(0, 2) \in D^{n} \times R$ above the disk's centre. The point $(x, t) \in D^{n} \times [0, 1]$ is sent along the line through $(0, 2)$ to its first intersection with the target subset. This retraction is continuous and fixes the target pointwise.

Step 2 (pass to the pushout). The retraction on each disk descends to the pushout because the radial-projection retraction restricted to $S^{n - 1} \times [0, 1]$ is the identity, so it agrees with the identity on $X^{(n - 1)} \times [0, 1]$ on the gluing locus. By the universal property of the pushout this defines a retraction $r_{n} : X^{(n)} \times [0, 1] \to X^{(n)} \times {0} \cup X^{(n - 1)} \times [0, 1]$ .

Step 3 (pass to the union). Iterate. The retractions assemble into a retraction $r : X \times [0, 1] \to X \times {0} \cup A \times [0, 1]$ on the colimit because the weak topology guarantees that a function out of $X$ is continuous iff its restriction to each skeleton is. $□$

Bridge. This builds toward the cellular approximation theorem and Whitehead's theorem stated in the next section: the HEP is the technical lemma that lets every homotopy of cellular maps be assembled cell by cell over the skeleton filtration, and the same skeleton-by-skeleton argument appears again in 03.12.05 (Eilenberg-MacLane spaces), where the inductive killing of higher homotopy is built one cell-attachment at a time. The foundational reason the whole CW machinery works is exactly this pushout-plus-colimit topology: a continuous map out of $X$ is exactly a coherent system of maps out of each cell, which generalises the analogous local-to-global passage of Seifert-van Kampen 03.12.09. Putting these together, the bridge is the recognition that a CW complex is defined as a colimit precisely so that homotopy-theoretic constructions extend across the colimit cell by cell.

Exercises [Intermediate+]

Exercise 6 (hard, short-answer).

Prove that every CW complex is locally path-connected.

Hint

Open cells are path-connected; show that every neighbourhood of a point in $X$ contains a path-connected open neighbourhood.

Answer

Let $x \in X$ lie in the open cell $e^{n}$ . The interior of $e^{n}$ is homeomorphic to an open disk, hence path-connected and open in $X^{(n)}$ . To extend to a neighbourhood in $X$ , use the inductive construction: any open $U ∋ x$ in $X^{(n)}$ extends to an open $U ∋ x$ in $X$ by attaching, at each higher dimension, a small radial mapping cylinder whose intersection with $X^{(n)}$ is $U$ itself. This extension is path-connected because each cylinder is path-connected and meets the previous level along a path-connected set. Rubric: full credit for the inductive thickening of an open cell-neighbourhood through the higher skeleta.

Exercise 7 (hard, short-answer).

Let $X$ be a CW complex. Show that any compact subset $K \subseteq X$ meets only finitely many open cells of $X$ .

Hint

Suppose for contradiction that $K$ meets infinitely many cells; pick one point from each.

Answer

Suppose $K$ meets infinitely many open cells; pick $x_{i} \in K \cap e_{α_{i}}^{n_{i}}$ , one per distinct cell. The set $S = {x_{i}}$ is countably infinite. Any subset $T \subseteq S$ is closed in $X$ : $T \cap X^{(n)}$ is finite (closure-finiteness applied to each cell touching $X^{(n)}$ contributes at most one $x_{i}$ per cell, and only finitely many cells meet any finite skeleton up to dimension $n$ — combine with the fact that compact subsets of finite CW complexes meet finitely many cells), hence closed in the simplicial / metric topology of $X^{(n)}$ . So $S$ is a closed discrete subset of $X$ . But every infinite closed discrete subset of a compact space is impossible. Contradiction. Rubric: full credit for the discrete-closed-subset argument; partial credit for sketching with reference to closure-finiteness.

Lean formalization [Intermediate+]

Mathlib does not yet ship a dedicated CW-complex API. The closest existing structure is RelativeCWComplex in the homotopy-theory development and the abstract pushout / colimit infrastructure in Mathlib.CategoryTheory.Limits. The intended skeleton-and-attaching-map definition would read schematically:

[object Promise]

This statement compiles only as a sketch; making it a usable Mathlib citizen requires the coercions, the attaching-map / characteristic-map / cell-indexing API, and the cellular-pushout naturality lemmas. See lean_mathlib_gap in the frontmatter for the concrete formalisation roadmap.

Advanced results [Master]

Theorem (cellular approximation). Let $f : X \to Y$ be a continuous map of CW complexes. Then $f$ is homotopic, via a homotopy preserving any subcomplex on which $f$ is already cellular, to a cellular map — that is, a map sending $X^{(n)}$ into $Y^{(n)}$ for every $n$ .

The proof proceeds by induction on the skeleta of $X$ , using the homotopy extension property to extend each successive deformation. The technical core is that any map $D^{n} \to Y^{(n)}$ whose image meets a cell of dimension $> n$ can be deformed off that cell, because $D^{n}$ has dimension less than the cell. Iterating gives a homotopy of $f$ to a cellular map. A full account is its own future unit; the statement is recorded here because it underlies almost every cell-counting argument in algebraic topology.

Theorem (Whitehead 1949). Let $f : X \to Y$ be a continuous map of CW complexes inducing an isomorphism $f_: \pi_n(X, x_0) \to \pi_n(Y, f(x_0)) $f or a l l$ n \geq 0 $an d e v er y ba se p o in t . T h e n$ f$ is a homotopy equivalence.*

The hypothesis "isomorphism on all $π_{n}$ " is called a weak homotopy equivalence. Whitehead's theorem says that for CW complexes the weak and the strong notion coincide. The proof builds the homotopy inverse cell by cell: an inductive construction of a map $g : Y \to X$ using the homotopy lifting along the skeleta of $Y$ , with the obstruction to extending each step lying in $π_{n} (X)$ , which by hypothesis is matched isomorphically with $π_{n} (Y)$ and so vanishes.

Theorem (HEP for CW pairs is a cofibration). The inclusion $A ↪ X$ of any CW subcomplex is a Hurewicz cofibration. This is the Master-tier rephrasing of the HEP from the Intermediate-tier theorem: stating it in the language of model categories, every CW pair is a cofibrant pair, and the inclusion is what model-categorically witnesses the HEP. The Quillen model structure on $Top$ takes weak equivalences = weak homotopy equivalences and cofibrations generated by the disk-sphere inclusions $S^{n - 1} ↪ D^{n}$ ; CW complexes are exactly the cell complexes of this model structure.

Examples and applications.

Spheres. $S^{n}$ has CW structure with one 0-cell and one $n$ -cell ( $n \geq 1$ ); the attaching map is the constant map collapsing $S^{n - 1}$ to the basepoint. Alternative structure: regard $S^{n}$ as the equator $S^{n - 1}$ with two $n$ -cells attached (two hemispheres).
Real projective space. $RP^{n}$ has one cell in each dimension $0, 1, \dots, n$ , with attaching maps the antipodal quotient $S^{k - 1} \to RP^{k - 1}$ .
Complex projective space. $CP^{n}$ has one cell in each even dimension $0, 2, \dots, 2 n$ , with attaching maps the Hopf-type quotient $S^{2 k - 1} \to CP^{k - 1}$ .
Lie groups. Every compact Lie group $G$ admits a CW structure (by Bott's analysis using Morse theory on the loop space, or directly via a cell decomposition adapted to a maximal torus and the Schubert decomposition of the flag variety $G / T$ ). The classifying space $B G$ is most naturally constructed as a CW complex.
Classifying spaces. $B U (n)$ has a CW structure with one cell for each Schubert symbol; its cohomology ring is the polynomial ring in the universal Chern classes.

Synthesis. Cellular structures generalise the polyhedral / simplicial-complex picture from combinatorial topology while keeping the attaching-map flexibility needed to realise spaces that are not triangulable in any natural way. The central insight is that the cellular pushout is exactly what makes a CW complex a colimit-type object: a map out of $X$ is a coherent system of maps out of each disk, and a homotopy out of $X$ is a coherent system of homotopies. This is dual to the cellular cochain complex, where a cocycle is a coherent system of values on each cell. The foundational reason the cellular machinery — chain complex, approximation theorem, Whitehead's theorem — operates so cleanly is that the colimit structure of $X$ pushes through every homotopy-theoretic functor that respects pushouts. Putting these together, the bridge is the identification of the CW category with the cell-complex category of the Quillen model structure on $Top$ : the CW complexes are cofibrant; weak equivalences between them are homotopy equivalences (Whitehead); every space is weakly equivalent to one (CW approximation). The same colimit-cell pattern appears again in simplicial sets, where the cells are non-degenerate simplices and the attaching maps are the face / degeneracy operators, and identifies the homotopy theory of CW complexes with that of Kan complexes — one of the structural facts organising modern algebraic topology.

Full proof set [Master]

Theorem (cellular approximation), proof sketch. Let $f : X \to Y$ be continuous between CW complexes. Inductively assume $f ∣_{X^{(n - 1)}}$ has been homotoped to land in $Y^{(n - 1)}$ . Each cell $e_{α}^{n} \subseteq X$ has characteristic map $Φ_{α} : D^{n} \to X$ . The image $f (Φ_{α} (D^{n}))$ is compact, hence by Exercise 7 meets only finitely many open cells of $Y$ . If any of those cells has dimension $> n$ , the standard transversality / piecewise-linear approximation theorem (or the explicit smooth-approximation argument when $f$ is smooth) deforms $f \circ Φ_{α}$ off the higher-dimensional cell by a homotopy supported on a small ball: a compact $n$ -dimensional image cannot fill a cell of dimension $> n$ , so a generic position avoids the cell entirely. Iterate over the finitely many higher-dimensional cells. Use HEP to extend the resulting homotopy from the cell back to $X$ . Iterate over $α$ then over $n$ . The colimit structure of $X$ ensures the inductive homotopies assemble. $□$

Theorem (Whitehead), proof sketch. Suppose $f : X \to Y$ is a weak equivalence of CW complexes. Construct a homotopy inverse $g : Y \to X$ skeleton by skeleton.

Step 0. Define $g$ on $Y^{(0)}$ by sending each 0-cell $y_{α}$ to any point in the homotopy class corresponding to $y_{α}$ under $π_{0} (f)^{- 1}$ , then choose a path in $Y$ from $f (g (y_{α}))$ to $y_{α}$ to record the homotopy from $f \circ g$ to $id_{Y}$ on $Y^{(0)}$ .

Step n, inductive step. Suppose $g$ is defined on $Y^{(n - 1)}$ together with a homotopy $h_{n - 1} : f \circ g ≃ id$ on $Y^{(n - 1)}$ . For each $n$ -cell $e_{β}^{n}$ of $Y$ with attaching map $ϕ_{β} : S^{n - 1} \to Y^{(n - 1)}$ , the composite $g \circ ϕ_{β} : S^{n - 1} \to X$ represents an element of $π_{n - 1} (X)$ . Its image under $f_{*}$ is $f g ϕ_{β} ≃ ϕ_{β}$ via $h_{n - 1}$ , which extends over $D^{n}$ (the cell itself), so $f_{*} [g ϕ_{β}] = 0$ . Since $f_{*}$ is an isomorphism, $[g ϕ_{β}] = 0$ in $π_{n - 1} (X)$ , which means $g \circ ϕ_{β}$ extends to a map $D^{n} \to X$ . This extension is the value of $g$ on the new $n$ -cell. Use HEP on $Y$ to extend the homotopy $h_{n - 1}$ across this cell.

Pass to the colimit. The result is $g : Y \to X$ with $f \circ g ≃ id_{Y}$ . By a symmetric argument $g$ is also a weak equivalence, and the same construction gives $f \circ h ≃ id_{X}$ for some $h$ , whence $f$ is a homotopy equivalence. $□$

Theorem (HEP for CW pairs), full proof. Done in the Intermediate-tier theorem above, using the radial-projection retraction of $D^{n} \times [0, 1]$ onto $S^{n - 1} \times [0, 1] \cup D^{n} \times {0}$ combined with the colimit topology.

Stated without proof — see Hatcher §4.1 ^{[Hatcher §4.1]} and tom Dieck §8 ^{[tom Dieck §8]}. The CW approximation theorem (every space is weakly equivalent to a CW complex), the simplicial approximation theorem in the special case of finite simplicial complexes, and the dual / cellular cochain identification of cellular cohomology with singular cohomology.

Connections [Master]

Eilenberg-MacLane spaces 03.12.05. The standard construction of $K (π, n)$ proceeds by attaching cells dimension by dimension to kill all homotopy groups except $π_{n} = π$ . The construction is cellular by design and uses Whitehead's theorem to identify any two such spaces as homotopy equivalent.
Whitehead tower 03.12.07. The successive killing of low homotopy groups is performed by attaching cells to fix the homotopy type one degree at a time; the resulting tower is a tower of CW complexes connected by cellular fibrations.
Seifert-van Kampen theorem 03.12.09. Computing $π_{1}$ of a CW complex reduces to reading off the 2-skeleton: 1-cells give generators, 2-cells give relations via their attaching words. Higher cells (dimension $\geq 3$ ) leave $π_{1}$ unchanged because their attaching spheres $S^{n - 1}$ are simply connected for $n \geq 3$ .
Covering space 03.12.02. Every covering $p : X \to X$ of a CW complex inherits a canonical CW structure with one cell of $X$ for each pair (cell of $X$ , sheet of the covering); cellular cohomology with local coefficients makes this lift functorial.
Cellular approximation theorem (planned future unit). The formal companion to the statement quoted in this unit: every continuous map between CW complexes is homotopic to a cellular one; this is the bridge between the topological category and the combinatorial cell-counting category.
Homotopy and homotopy group 03.12.01. Whitehead's theorem identifies weak homotopy equivalences with homotopy equivalences in the CW world, which is exactly what makes CW complexes the natural objects of study for $π_{n}$ .

Historical & philosophical context [Master]

J. H. C. Whitehead introduced CW complexes in his 1949 paper Combinatorial homotopy I (Bull. AMS 55, 213-245) ^{[Whitehead 1949]}. The construction was Whitehead's response to the rigidity of simplicial complexes: a sphere $S^{n}$ requires at least $n + 2$ vertices to be triangulated as a simplicial complex, but only two cells (one 0-cell and one $n$ -cell) to be presented as a CW complex. The economy and the flexibility of attaching by arbitrary maps rather than by face / degeneracy operators made CW complexes the dominant model of "well-behaved space" in algebraic topology. Whitehead also proved the theorem now bearing his name in the same paper.

The companion paper Combinatorial homotopy II (1949) developed the algebraic apparatus (crossed complexes, the formalism that later flowered in Brown-Higgins) needed to track the homotopy type of CW complexes algebraically. Whitehead's larger programme was to make the homotopy category accessible to combinatorial methods; CW complexes were the carrier objects, weak equivalences the morphisms, and what is now called the Whitehead theorem the structural fact connecting the two.

The model-category-theoretic incarnation came with Quillen's Homotopical Algebra (Lecture Notes in Mathematics 43, 1967), where CW complexes are the cofibrant objects of the standard model structure on $Top$ , weak homotopy equivalences are the weak equivalences, and Hurewicz fibrations are the fibrations. Quillen's framework retroactively justified Whitehead's choice: the cellular pushout is the small-object argument's basic input, and the inductive cell-attachment construction is exactly the transfinite composition of pushouts of generating cofibrations.

Bibliography [Master]

[object Promise]

Prerequisites

02.01.01
02.01.06
03.12.01

Used in

03.12.05
03.12.07

Tier anchors

beginner: Hatcher *Algebraic Topology* §0; 3Blue1Brown — cell-attachment intuition
intermediate: Hatcher §0; May *A Concise Course in Algebraic Topology* Ch. 10
master: Whitehead 1949 — *Combinatorial homotopy I* (originator paper, Bull. AMS 55); Hatcher §0 and §4.1; tom Dieck *Algebraic Topology* §8

References

TODO_REF
Hatcher — Algebraic Topology · §0 (CW complexes), §4.1 (cellular approximation, Whitehead theorem)
TODO_REF
May — A Concise Course in Algebraic Topology · Ch. 10 CW complexes
TODO_REF
Whitehead — Combinatorial homotopy I · Bull. AMS 55 (1949) 213-245, originator paper for CW complexes
TODO_REF
tom Dieck — Algebraic Topology · §8 CW-complexes
TODO_REF
Brown — Topology and Groupoids · §4 identification topology and adjunction spaces

Reviewer

TBD

Estimated time

beginner: 15m
intermediate: 40m
master: 75m