03.12.20 · modern-geometry / homotopy

Whitehead's theorem

shipped3 tiersLean: none

Anchor (Master): Whitehead 1949 — *Combinatorial homotopy I+II* (originator papers, Bull. AMS 55); Hatcher §4.1 (full proof + Warsaw counterexample); tom Dieck *Algebraic Topology* §8

Intuition [Beginner]

Whitehead's theorem answers a basic question. Two spaces look the same to homotopy theory if their fundamental group agrees, their second homotopy group agrees, their third agrees, and so on for every dimension. Call a continuous map between two spaces a weak equivalence if it produces such matching homotopy groups in every dimension. The question is whether a weak equivalence is enough to deform the source to the target through continuous motions — that is, whether matching algebraic invariants force the spaces to be the same as homotopy types.

For arbitrary topological spaces the answer is no. There are spaces whose homotopy groups all vanish yet which are not contractible to a point. The Warsaw circle is the standard example: it has the algebra of a point but is geometrically a one-dimensional curve that wraps in on itself in a way no continuous deformation can untangle.

Whitehead's theorem says: if you restrict to spaces built by gluing cells together — the CW complexes — then matching homotopy groups is enough. A weak equivalence between CW complexes is automatically a homotopy equivalence. The cellular structure is exactly the geometric hypothesis that lets the algebra recover the topology.

Visual [Beginner]

A schematic with two CW complexes drawn as cell-attachment diagrams, one above the other, joined by a wavy arrow labelled "matching $π_{n}$ for every $n$ ". A second arrow underneath, labelled "homotopy equivalence", points the same direction with a checkmark. To one side, a small inset shows the Warsaw circle as a non-CW space where the upper arrow exists but the lower arrow does not.

The picture captures the content of the theorem: the CW hypothesis closes the gap between matching homotopy groups and matching homotopy types.

Worked example [Beginner]

Show that any two contractible CW complexes are homotopy equivalent.

Step 1. A space is contractible when every continuous loop, every continuous sphere, and so on can be shrunk to a point inside the space. Equivalently, every homotopy group vanishes: $π_{n} (X) = 0$ for every $n \geq 0$ .

Step 2. Take two contractible CW complexes $X$ and $Y$ . Pick any continuous map $f : X \to Y$ — for example a constant map sending all of $X$ to a chosen point of $Y$ .

Step 3. Compute the induced maps on homotopy groups. Both source and target have all homotopy groups equal to zero, so $f_{*}$ goes from zero to zero in every dimension. Any map between zero groups is the zero map, which is the only map and hence an isomorphism.

Step 4. Whitehead's theorem applies: $f$ induces an isomorphism on every homotopy group between two CW complexes, so $f$ is a homotopy equivalence. The two spaces are the same as homotopy types.

What this tells us: the homotopy type of a contractible CW complex is canonical — there is only one, no matter how many cells went into the construction. A point, a disk, a tree, a contractible 3-manifold, a contractible CW complex of dimension a million: same homotopy type. The cellular hypothesis is what makes this clean statement work.

Check your understanding [Beginner]

Formal definition [Intermediate+]

Let $X, Y$ be topological spaces with chosen base points $x_{0} \in X$ and $y_{0} \in Y$ . A continuous map $f : (X, x_{0}) \to (Y, y_{0})$ is a weak homotopy equivalence when for every $n \geq 0$ and every base point of $X$ , the induced map $$ f_* : \pi_n(X, x_0) \to \pi_n(Y, f(x_0)) $$ is an isomorphism. (For $n = 0$ the homotopy "groups" are pointed sets of path components and the condition is bijection of these sets together with isomorphism of fundamental groups at every base point.)

A continuous map $f : X \to Y$ is a homotopy equivalence when there exists a continuous $g : Y \to X$ such that $g \circ f ≃ id_{X}$ and $f \circ g ≃ id_{Y}$ , where $≃$ denotes homotopy of maps.

A homotopy equivalence is always a weak homotopy equivalence — homotopic maps induce the same map on homotopy groups, so a continuous inverse up to homotopy gives a group-theoretic inverse on every $π_{n}$ . The reverse implication is the content of Whitehead's theorem and requires the CW hypothesis.

A space has the homotopy type of a CW complex when there exists a CW complex $K$ and a homotopy equivalence $X ≃ K$ . Many spaces of geometric interest (smooth manifolds, polyhedra, ANRs, geometric realisations of simplicial sets) have CW homotopy type even when they are not literally CW complexes. The theorem applies to such spaces by composing with the equivalence to a CW model.

The mapping cylinder of a continuous map $f : X \to Y$ is $$ M_f = (X \times [0, 1]) \sqcup Y ,/, \sim $$ where $(x, 1) \sim f (x)$ . The inclusions $X ↪ M_{f}$ at level $0$ and $Y ↪ M_{f}$ make $M_{f}$ a thickened version of $f$ : $Y$ is a deformation retract of $M_{f}$ , and $f$ factors through the inclusion $X ↪ M_{f}$ followed by the retraction $M_{f} \to Y$ . Whenever $X$ and $Y$ are CW complexes and $f$ is cellular, $M_{f}$ inherits a CW structure with $X$ and $Y$ as subcomplexes.

Counterexamples to common slips

The CW hypothesis applies to both spaces. A weak equivalence from a CW complex to a non-CW space need not be a homotopy equivalence. The constant map from a point to the Warsaw circle induces isomorphisms on all $π_{n}$ (both sides are zero) but is not a homotopy equivalence.
A weak equivalence between two non-CW spaces can fail in either direction. The Warsaw circle is a non-CW space with $π_{n} = 0$ for every $n$ but is not contractible. Whitehead's theorem is exactly the assertion that such pathologies disappear under the CW hypothesis.
The theorem requires isomorphisms on $π_{n}$ for every base point, not just one. A non-path-connected space $X$ can have weak equivalence to $Y$ on each component while the components themselves fail to match. The base-point-by-base-point statement is the correct one.
The conclusion is homotopy equivalence, not homeomorphism. The disk $D^{2}$ and a single point are homotopy equivalent and the inclusion of the point is a homotopy equivalence, but the spaces are not homeomorphic. Whitehead's theorem makes claims at the level of homotopy types, not point-set topology.

Key theorem with proof [Intermediate+]

Theorem (Whitehead's theorem; Hatcher Theorem 4.5). Let $X$ and $Y$ be CW complexes and let $f : X \to Y$ be a continuous map. If $f_ : \pi_n(X, x_0) \to \pi_n(Y, f(x_0)) $i s ani so m or p hi s m f or e v er y$ n \geq 0 $an d e v er y ba se p o in t$ x_0 \in X $, t h e n$ f$ is a homotopy equivalence.*

Proof. Replace $f$ by the inclusion $X ↪ M_{f}$ into its mapping cylinder. Since $Y$ deformation-retracts onto $M_{f}$ and the inclusion $Y ↪ M_{f}$ is a homotopy equivalence, proving that $X ↪ M_{f}$ is a homotopy equivalence is the same as proving the original $f$ is one. The inclusion $X ↪ M_{f}$ is a CW pair (after giving $M_{f}$ its natural CW structure using cellular approximation to make $f$ cellular if it was not already), and the weak-equivalence hypothesis transfers: $π_{n} (M_{f}, X)$ vanishes for every $n$ because the long exact sequence of the pair reads $$ \cdots \to \pi_n(X) \xrightarrow{i_*} \pi_n(M_f) \to \pi_n(M_f, X) \to \pi_{n-1}(X) \to \cdots $$ and $i_{*}$ is an isomorphism by hypothesis (using $π_{n} (M_{f}) ≅ π_{n} (Y)$ via the deformation retract).

It remains to show: if $(W, A)$ is a CW pair with $π_{n} (W, A) = 0$ for every $n \geq 0$ , then $A$ is a deformation retract of $W$ .

Construct the deformation retraction $r : W \to A$ skeleton by skeleton. Set $r_{0} = id_{A}$ on the $0$ -skeleton inside $A$ and define $r_{0}$ on the remaining $0$ -cells of $W$ by sending each new $0$ -cell of $W ∖ A$ to a point in $A$ (choose any point in the path component, using $π_{0} (W, A) = 0$ ). Suppose inductively that $r_{n} : W^{(n)} \to A$ has been defined together with a homotopy $H_{n} : W^{(n)} \times [0, 1] \to W$ from the inclusion $W^{(n)} ↪ W$ to a map landing in $A$ , with $H_{n}$ stationary on $A \cap W^{(n)}$ .

Extend across cells of dimension $n + 1$ . For each $(n + 1)$ -cell $e_{α}^{n + 1}$ of $W ∖ A$ with attaching map $φ_{α} : S^{n} \to W^{(n)}$ , the composite $r_{n} \circ φ_{α} : S^{n} \to A$ together with the cell's interior gives an element of $π_{n + 1} (W, A)$ . The hypothesis $π_{n + 1} (W, A) = 0$ kills this element: there is a homotopy from the cell's characteristic map (relative to $φ_{α}$ ) to a map landing entirely in $A$ . The CW pair $(W, A)$ has the homotopy extension property, so the homotopy extends to all of $W$ while remaining stationary outside the cell. This produces $r_{n + 1}$ and $H_{n + 1}$ .

Pass to the colimit: since $W$ has the weak topology and the homotopies $H_{n}$ extend coherently, the assembled map $r : W \to A$ together with the assembled homotopy $H : W \times [0, 1] \to W$ defines a deformation retraction. The composition gives the required homotopy inverse to $f$ . $□$

The proof rests on three pieces, each due to Whitehead in the 1949 Combinatorial homotopy papers: the cellular approximation theorem, the homotopy extension property for CW pairs, and the inductive obstruction-theoretic argument that uses the vanishing of relative homotopy groups to extend a partial inverse one cell at a time.

Bridge. Whitehead's theorem builds toward the entire homotopy-theoretic framework of CW complexes: the theorem is what justifies treating CW complexes as the category for homotopy theory, since their homotopy type is captured by the homotopy groups alone. The foundational reason it holds is exactly that CW pairs are cofibrations and CW complexes admit cellular approximation, so a partial homotopy inverse can be extended one cell at a time using the vanishing of relative homotopy groups as the obstruction-killer. This same skeleton-by-skeleton obstruction theory appears again in 03.12.05 (Eilenberg-MacLane spaces) when constructing $K (π, n)$ as a CW complex by attaching cells to kill unwanted homotopy groups, and again in 03.12.07 (Whitehead tower) when the tower stages are constructed by killing low-degree homotopy classes through cell attachment. The central insight is that the homotopy groups of a CW complex are not merely invariants attached to the space — they are the obstruction theory that controls map extension, and the theorem identifies the resulting equivalence relation on CW complexes with weak equivalence on homotopy groups. Putting these together, Whitehead's theorem is the bridge from algebra to topology in CW homotopy theory, and the same obstruction-theoretic mechanism appears again in 03.12.05 (Eilenberg-MacLane spaces) where the cell-attachment construction uses the same vanishing-relative-homotopy-group input.

Exercises [Intermediate+]

Exercise 3 (medium, symbolic).

Let $X = S^{2} \lor S^{4}$ (wedge of a $2$ -sphere and a $4$ -sphere) and $Y = S^{2} \times S^{4} / (S^{2} \lor S^{4})$ , the quotient that collapses the wedge subcomplex. Compute $π_{n}$ of both spaces in low dimensions and decide whether they admit a weak equivalence.

Hint

The quotient $S^{2} \times S^{4} / (S^{2} \lor S^{4}) ≅ S^{6}$ . Whitehead products give a non-zero element of $π_{5} (S^{2} \lor S^{4})$ .

Answer

The quotient $S^{2} \times S^{4} / (S^{2} \lor S^{4})$ is homeomorphic to $S^{6}$ (the top cell of the product). So $π_{n} (Y) = π_{n} (S^{6})$ which vanishes for $n < 6$ and equals $Z$ at $n = 6$ . For $X = S^{2} \lor S^{4}$ , the Hilton-Milnor theorem gives $π_{*} (X) = π_{*} (S^{2}) \oplus π_{*} (S^{4}) \oplus π_{*} (S^{5}) \oplus \dots$ where the $S^{5}$ summand comes from the Whitehead product $[ι_{2}, ι_{4}]$ . In particular, $π_{5} (X)$ contains a $Z$ from the Whitehead product; $π_{5} (Y) = 0$ . There is no weak equivalence because the homotopy groups already disagree at $n = 5$ . Rubric: full credit for the Whitehead-product detection plus the dimension count.

Exercise 4 (medium, short-answer).

Show that the inclusion of the $n$ -skeleton $X^{(n)} ↪ X$ of a CW complex is a weak equivalence on $π_{k}$ for $k < n$ but generally fails for $k \geq n$ . Conclude that, restricted to $k < n$ , the inclusion behaves as Whitehead's theorem demands.

Hint

Use cellular approximation: any map $S^{k} \to X$ is homotopic to one landing in $X^{(k)} \subseteq X^{(n)}$ when $k < n$ . The same argument controls homotopies.

Answer

By cellular approximation, every continuous map $S^{k} \to X$ is homotopic to a cellular map landing in $X^{(k)}$ , which sits inside $X^{(n)}$ when $k < n$ . Hence every element of $π_{k} (X)$ comes from $π_{k} (X^{(n)})$ , so the inclusion-induced map on $π_{k}$ is surjective. Cellular approximation applied to homotopies between maps $S^{k} \to X^{(n)}$ shows that two such maps that become homotopic in $X$ were already homotopic in $X^{(n)}$ when $k + 1 < n + 1$ — that is, when $k < n$ . So the inclusion is injective on $π_{k}$ as well. For $k \geq n$ , attaching higher-dimensional cells in $X ∖ X^{(n)}$ can change $π_{k}$ , so the isomorphism breaks. The skeleton inclusion thus realises Whitehead's hypothesis through dimension $n - 1$ but does not satisfy it overall. Rubric: full credit for the cellular-approximation argument in both directions.

Exercise 5 (medium, symbolic).

Let $f : X \to Y$ be a map between simply-connected CW complexes. Combine Whitehead's theorem with the Hurewicz theorem to show that $f$ is a homotopy equivalence if and only if $f_{*}$ is an isomorphism on every integer homology group $H_{n} (X) \to H_{n} (Y)$ .

Hint

The Hurewicz theorem says $π_{n} \to H_{n}$ is an isomorphism for the lowest non-vanishing dimension of a simply-connected space. Combine with the fact that a homotopy equivalence between simply-connected spaces induces homology isomorphisms, then run the implication backwards inductively.

Answer

If $f$ is a homotopy equivalence, $f_{*}$ is an isomorphism on every homotopy invariant including $H_{n}$ . Conversely, suppose $f_{*}$ is an isomorphism on every $H_{n}$ . The mapping cylinder $M_{f}$ has $X$ as a subcomplex with quotient $Y / X$ contractible (homologically), so the relative homology $H_{n} (M_{f}, X)$ vanishes for every $n$ by the long exact sequence. Both $X$ and $M_{f}$ are simply connected (homotopy-equivalently $Y$ is). The Hurewicz theorem applied to the pair $(M_{f}, X)$ — using simply-connectedness to make the relative form available — gives $π_{n} (M_{f}, X) ≅ H_{n} (M_{f}, X) = 0$ for every $n$ . By the long exact sequence of homotopy groups, $π_{n} (X) \to π_{n} (M_{f}) ≅ π_{n} (Y)$ is an isomorphism. Whitehead's theorem then concludes that the inclusion $X ↪ M_{f}$ , hence $f$ , is a homotopy equivalence. Rubric: full credit for the Hurewicz-then-Whitehead chain, with simply-connectedness handling.

Exercise 6 (medium, short-answer).

The Warsaw circle $W$ is the closure in $R^{2}$ of the graph of $sin (1/ x)$ for $x \in (0, 1]$ , joined by an arc to the segment ${0} \times [- 1, 1]$ . Verify that every $π_{n} (W) = 0$ and explain why $W$ is nonetheless not contractible. Use this to confirm that Whitehead's theorem requires the CW hypothesis.

Hint

The space $W$ is path-connected but not locally path-connected at the limiting segment. Continuous maps from $S^{n}$ into $W$ have small image (path-connected component analysis), but no continuous deformation can pull the limiting segment across the oscillations.

Answer

Path-connectedness of $W$ is checked via the joining arc. Any continuous $S^{n} \to W$ has compact image; compact subsets of $W$ that miss a neighbourhood of the limiting segment ${0} \times [- 1, 1]$ are contained in a contractible subset (a portion of the graph plus a piece of the arc). Compact subsets that hit the limiting segment cannot oscillate, so they are also contractible inside $W$ . Hence $π_{n} (W) = 0$ for every $n$ . However $W$ is not contractible: a deformation retract to a point would give a continuous family of paths from each point of $W$ to the chosen point, and at the limiting segment the path-connectedness fails to be uniform. Concretely, the path-component map $W \to π_{0} (W)$ is constant but not via a continuous deformation. The constant map $pt \to W$ induces isomorphisms on every $π_{n}$ (both sides vanish) but is not a homotopy equivalence. Rubric: full credit for the path-component analysis on both sides plus the failure of contractibility.

Exercise 7 (hard, short-answer).

Construct a continuous map $f : S^{1} \times S^{2} \to S^{1} \lor S^{2} \lor S^{3}$ that induces isomorphisms on integer homology in every degree. Decide whether $f$ is a homotopy equivalence and identify the obstruction.

Hint

The product $S^{1} \times S^{2}$ has the same integer homology as the wedge $S^{1} \lor S^{2} \lor S^{3}$ — both have a $Z$ in degrees $0, 1, 2, 3$ . Build $f$ using the projection-like map collapsing the wedge structure of the product. The obstruction to homotopy equivalence is at $π_{1}$ .

Answer

The product $S^{1} \times S^{2}$ has cell structure with cells of dimensions $0, 1, 2, 3$ — one each — by the product CW structure. Its integer homology is $Z$ in each of those degrees. The wedge $S^{1} \lor S^{2} \lor S^{3}$ has the same integer homology by the wedge formula. A continuous $f$ inducing homology isomorphisms exists by mapping each cell to its corresponding wedge summand. However $f$ is not a homotopy equivalence: $π_{1} (S^{1} \times S^{2}) = Z$ and $π_{1} (S^{1} \lor S^{2} \lor S^{3}) = Z$ as well, but the higher homotopy groups disagree. The product has $π_{2} = Z$ from the $S^{2}$ factor and a non-zero action of $π_{1}$ on $π_{3}$ . The wedge has $π_{2}$ as a free $Z [π_{1}]$ -module on the $S^{2}$ -summand class, making $π_{2}$ a $Z [Z]$ -module of infinite rank when seen as an abelian group — completely different from $π_{2} (S^{1} \times S^{2}) = Z$ . Whitehead's hypothesis fails at $π_{2}$ and the conclusion fails accordingly. The example shows that Whitehead's integer-homology corollary (Hurewicz + Whitehead) requires simple-connectedness and breaks for spaces with non-zero $π_{1}$ . Rubric: full credit for the $π_{2}$ comparison plus the simple-connectedness caveat.

Exercise 8 (hard, short-answer).

Prove that any CW approximation $ρ : Y^{'} \to Y$ of a topological space $Y$ — that is, a CW complex $Y^{'}$ together with a weak equivalence $ρ$ — is unique up to homotopy equivalence over $Y$ . (Specifically, given two CW approximations $ρ_{1}, ρ_{2}$ , there is a homotopy equivalence $h : Y_{1}^{'} \to Y_{2}^{'}$ with $ρ_{2} \circ h ≃ ρ_{1}$ .)

Hint

Use the cellular approximation theorem to lift $ρ_{1}$ along $ρ_{2}$ to a map $h : Y_{1}^{'} \to Y_{2}^{'}$ . Then run Whitehead's theorem on $h$ .

Answer

Given CW approximations $ρ_{1} : Y_{1}^{'} \to Y$ and $ρ_{2} : Y_{2}^{'} \to Y$ , the map $ρ_{1}$ admits a lift along $ρ_{2}$ up to homotopy because $ρ_{2}$ is a weak equivalence and $Y_{1}^{'}$ is a CW complex: the obstruction to lifting a map $S^{n} \to Y$ to $Y_{2}^{'}$ is in $π_{n}$ of the homotopy fibre of $ρ_{2}$ , which vanishes by the long exact sequence and the weak-equivalence assumption. Cellular approximation makes the lift cellular. Call this lift $h : Y_{1}^{'} \to Y_{2}^{'}$ . The diagram $ρ_{2} \circ h ≃ ρ_{1}$ implies $h_{*} : π_{n} (Y_{1}^{'}) \to π_{n} (Y_{2}^{'})$ matches $ρ_{1, *}$ followed by $ρ_{2, *}^{- 1}$ , both isomorphisms, hence $h$ is a weak equivalence between CW complexes. Whitehead's theorem promotes $h$ to a homotopy equivalence. Rubric: full credit for the lifting step plus the Whitehead application.

Lean formalization [Intermediate+]

Mathlib has the abstract Top model structure and a notion of weak homotopy equivalence, but does not yet ship a CW-complex API or the CW form of Whitehead's theorem. The intended formalisation reads schematically:

[object Promise]

The proof gap is substantive. Mathlib needs a CW-complex structure with skeleton filtration, the cellular approximation theorem, the homotopy extension property witness for CW pair inclusions, the long exact sequence of homotopy groups for the pair $(M_{f}, X)$ , and the inductive obstruction-theoretic extension argument that uses vanishing relative homotopy groups to extend a partial homotopy across cells. The Warsaw circle as a sharpness counterexample would itself require a small Mathlib formalisation of the closure of the topologist's-sine-curve graph and the proof that all its homotopy groups vanish.

Advanced results [Master]

Theorem (Whitehead's theorem, sharpened with mapping cylinders). Let $f : X \to Y$ be a continuous map between CW complexes. The following are equivalent:

$f$ is a homotopy equivalence;
$f$ is a weak homotopy equivalence;
the inclusion $X ↪ M_{f}$ admits a deformation retraction;
$π_{n} (M_{f}, X) = 0$ for every $n \geq 0$ and every base point.

The four conditions are equivalent for CW pairs and provide the four standard reformulations used in different contexts. The implication (2) $\Rightarrow$ (3) is the substantive content; (1) $\Rightarrow$ (2) and (3) $\Rightarrow$ (1) are formal; (3) $\Leftrightarrow$ (4) is the long exact sequence of the pair.

Theorem (homology corollary; Hurewicz + Whitehead). Let $f : X \to Y$ be a continuous map between simply-connected CW complexes. If $f_ : H_n(X; \mathbb{Z}) \to H_n(Y; \mathbb{Z}) $i s ani so m or p hi s m f or e v er y$ n \geq 0 $, t h e n$ f$ is a homotopy equivalence.*

The corollary reduces the Whitehead-type detection of homotopy equivalence from homotopy groups (often hard to compute) to homology groups (often computable from cellular chain complexes). The simple-connectedness hypothesis is essential: $S^{1}$ and the wedge of $S^{1}$ with the dunce cap have isomorphic integer homology but different fundamental groups, so no map between them can simultaneously be a homology isomorphism and a homotopy equivalence. The corollary is the foundational tool for showing two simply-connected CW complexes are homotopy equivalent via a homology calculation.

Theorem (CW approximation). For every topological space $Y$ , there exists a CW complex $Y^{'}$ and a weak homotopy equivalence $ρ : Y^{'} \to Y$ . The pair $(Y^{'}, ρ)$ is unique up to homotopy equivalence over $Y$ .

CW approximation extends the reach of Whitehead's theorem: any space $Y$ can be replaced by a homotopy-theoretically equivalent CW complex $Y^{'}$ for the purpose of homotopy-theoretic constructions. Together with Whitehead's theorem, this gives the foundational result that the homotopy category of CW complexes embeds fully faithfully into the homotopy category of all topological spaces, with the embedding an equivalence at the level of weak homotopy types.

Theorem (Warsaw circle as counterexample; Hatcher Example 4.36). The Warsaw circle $W$ — the closure in $R^{2}$ of ${(x, sin (1/ x)) : 0 < x \leq 1}$ joined to the segment ${0} \times [- 1, 1]$ by an arc — is path-connected with $π_{n} (W) = 0$ for every $n \geq 0$ , but $W$ is not contractible. Therefore the constant map $pt \to W$ is a weak homotopy equivalence that is not a homotopy equivalence.

The Warsaw circle was Whitehead's motivating example for restricting the theorem to CW complexes. Topologically, the failure traces to local non-CW behaviour: $W$ is not locally path-connected at the limiting segment, so the topology cannot be reconstructed from the homotopy of compact subsets. The CW hypothesis is exactly the local-topology condition that excludes this pathology.

Theorem (relative form). Let $(X, A)$ and $(Y, B)$ be CW pairs and let $f : (X, A) \to (Y, B)$ be a continuous map of pairs. If $f_ : \pi_n(X, A, x_0) \to \pi_n(Y, B, f(x_0)) $i s ani so m or p hi s m f or e v er y$ n \geq 1 $an d e v er y ba se p o in t, t h e n$ f$ is a homotopy equivalence of pairs.*

The relative form is what one needs for inductive constructions in obstruction theory and Postnikov-tower analyses. The proof parallels the absolute form, working with the pair structure throughout: the skeleton-by-skeleton extension is performed on $X ∖ A$ relative to $A$ , with the obstruction at each cell living in the relative homotopy group of the target pair.

Theorem (Whitehead's theorem and CW model categories). In the standard Quillen model structure on topological spaces, the cofibrant objects are precisely the spaces with the homotopy type of a CW complex, the weak equivalences are the weak homotopy equivalences, and Whitehead's theorem is the assertion that a weak equivalence between cofibrant objects is a homotopy equivalence.

The model-categorical packaging makes clear that Whitehead's theorem is not a topological accident but a structural feature of the abstract homotopy theory: in any model category, a weak equivalence between fibrant-cofibrant objects is a homotopy equivalence. The CW hypothesis is the cofibrancy hypothesis; every space is fibrant in the standard model structure on Top.

Synthesis. Whitehead's theorem is the foundational reason that CW complexes are the right category for homotopy theory: the homotopy type of a CW complex is captured by its homotopy groups, and a map between CW complexes is a homotopy equivalence exactly when it induces isomorphisms on every homotopy group at every base point. The central insight is that the CW hypothesis is a cofibration hypothesis in disguise — CW pair inclusions are cofibrations, and the homotopy extension property they witness is exactly what the inductive cell-by-cell extension of a partial homotopy inverse needs as input. Putting these together, the proof factors as: replace $f$ by its mapping cylinder inclusion, observe that the relative homotopy groups vanish by the weak-equivalence hypothesis, and extend the identity on $X$ to a deformation retraction of $M_{f}$ skeleton by skeleton using the homotopy extension property. The corollaries are immediate: the homology version (Hurewicz + Whitehead) reduces the detection of homotopy equivalence between simply-connected CW complexes to a homology calculation; the CW approximation theorem extends the framework to arbitrary topological spaces by replacing them with CW models; the relative form supports inductive obstruction-theoretic constructions; and the model-categorical packaging identifies the theorem with the abstract statement that weak equivalences between cofibrant-fibrant objects are homotopy equivalences.

This same skeleton-by-skeleton obstruction argument appears again in 03.12.05 (Eilenberg-MacLane spaces), where the construction of $K (π, n)$ as a CW complex starts from a model of $π_{n}$ and attaches cells of higher dimension to kill all unwanted homotopy groups, the obstruction at each step living in exactly the homotopy group being killed. The bridge to the Whitehead-tower framework of 03.12.07 is the dual construction: rather than killing higher homotopy groups, one kills lower ones by passing to a fibration whose fibre is an Eilenberg-MacLane space, and the Whitehead-tower stages are detected as homotopy equivalences via Whitehead's theorem applied to the connecting maps. Putting these together, the same obstruction-theoretic mechanism — vanishing of a relative homotopy group as the obstruction to extending a map across a cell — drives the construction of Eilenberg-MacLane spaces, Postnikov towers, Whitehead towers, and the proof of Whitehead's theorem itself. The bridge is the recognition that CW complexes carry a built-in obstruction theory, and the obstructions are precisely the homotopy groups; this is exactly the same organising principle that appears again in 03.12.07 (Whitehead tower) where stage-wise construction is governed by the same mechanism.

Full proof set [Master]

Theorem (Whitehead's theorem, full proof). Let $X, Y$ be CW complexes and $f : X \to Y$ a weak homotopy equivalence. Then $f$ is a homotopy equivalence.

Step 1: Reduce to a CW pair. Replace $f$ by the inclusion $i : X ↪ M_{f}$ into the mapping cylinder $M_{f} = (X \times [0, 1]) ⊔ Y / ((x, 1) \sim f (x))$ . The natural projection $π : M_{f} \to Y$ is a deformation retract: the homotopy $H_{t} (x, s) = (x, (1 - t) s + t)$ on $X \times [0, 1]$ , extended to be the identity on $Y$ , deformation-retracts $M_{f}$ onto $Y \subseteq M_{f}$ . Hence $i$ is a homotopy equivalence if and only if $π \circ i = f$ is one. By cellular approximation (Hatcher Theorem 4.8), $f$ may be replaced by a cellular map; the resulting $M_{f}$ inherits a CW structure with $X$ and $Y$ as subcomplexes.

Step 2: Translate to vanishing relative homotopy. The long exact sequence of the pair $(M_{f}, X)$ reads $$ \cdots \to \pi_n(X, x_0) \xrightarrow{i_*} \pi_n(M_f, x_0) \to \pi_n(M_f, X, x_0) \to \pi_{n-1}(X, x_0) \to \cdots. $$ Since $π : M_{f} \to Y$ is a homotopy equivalence, $π_{*}$ identifies $π_{n} (M_{f}, x_{0})$ with $π_{n} (Y, f (x_{0}))$ , and the composite $π_{*} \circ i_{*} = f_{*}$ is an isomorphism by hypothesis. Hence $i_{*}$ is an isomorphism for every $n$ , and the long exact sequence forces $π_{n} (M_{f}, X, x_{0}) = 0$ for every $n \geq 0$ and every base point.

Step 3: Construct the deformation retraction inductively. It suffices to prove: if $(W, A)$ is a CW pair with $π_{n} (W, A, a) = 0$ for every $n \geq 0$ and every base point $a \in A$ , then $A$ is a deformation retract of $W$ . Apply this with $(W, A) = (M_{f}, X)$ to conclude $X$ is a deformation retract of $M_{f}$ , hence $i : X ↪ M_{f}$ is a homotopy equivalence.

Construct the retraction $r : W \to A$ together with a homotopy $H : W \times [0, 1] \to W$ from $id_{W}$ to $ι \circ r$ (where $ι : A ↪ W$ ), with $H$ stationary on $A$ , by induction on the skeleton. Set $r$ on $A$ to be the identity and on $W^{(0)} ∖ A$ to send each $0$ -cell to a point in $A$ in the same path component (using $π_{0} (W, A) = 0$ to find such a point). The homotopy $H_{0}$ on $W^{(0)}$ is the path homotopy realising this connection.

Suppose inductively that $r_{n} : W^{(n)} \to A$ and $H_{n} : W^{(n)} \times [0, 1] \to W$ have been constructed, with $H_{n}$ a homotopy from the inclusion to $ι \circ r_{n}$ , stationary on $A \cap W^{(n)}$ . For each $(n + 1)$ -cell $e_{α}^{n + 1}$ of $W ∖ A$ with characteristic map $Φ_{α} : (D^{n + 1}, S^{n}) \to (W^{(n + 1)}, W^{(n)})$ , the composite $$ S^n \xrightarrow{\varphi_\alpha} W^{(n)} \xrightarrow{(r_n, H_n(-, 1))} A $$ together with the cell's interior gives a relative homotopy class $[Φ_{α}, r_{n} \circ φ_{α}] \in π_{n + 1} (W, A, a_{α})$ for some base point $a_{α} \in A$ . The hypothesis $π_{n + 1} (W, A) = 0$ kills this class: there is a homotopy $G_{α} : D^{n + 1} \times [0, 1] \to W$ from $Φ_{α}$ to a map $D^{n + 1} \to A$ , with the boundary homotopy on $S^{n}$ matching $H_{n}$ on $φ_{α} (S^{n})$ .

The CW pair $(W, W^{(n)})$ has the homotopy extension property (Hatcher Proposition 0.16): the partial homotopy $G_{α}$ on each $(n + 1)$ -cell, together with $H_{n}$ on $W^{(n)}$ , extends to a homotopy on all of $W^{(n + 1)}$ . Define $r_{n + 1}$ as the time- $1$ value of this extension, restricted to land in $A$ , and $H_{n + 1}$ as the extended homotopy. Stationarity on $A \cap W^{(n + 1)}$ is preserved by the construction.

Step 4: Pass to the colimit. The CW topology on $W$ is the weak topology with respect to the skeleton filtration: a set is closed in $W$ iff its intersection with every $W^{(n)}$ is closed. The maps $r_{n}$ and homotopies $H_{n}$ are coherent across the filtration ( $r_{n + 1} ∣_{W^{(n)}} = r_{n}$ , similarly for $H$ ), so they assemble into continuous maps $r : W \to A$ and $H : W \times [0, 1] \to W$ . The map $H$ is a deformation retraction of $W$ onto $A$ . $□$

Theorem (homology corollary), proof. Let $f : X \to Y$ be a map between simply-connected CW complexes inducing isomorphisms on all integer homology. The mapping cylinder $M_{f}$ is simply connected (both $X$ and $Y$ are, and $M_{f}$ deformation-retracts to $Y$ ). The long exact sequence of the pair $(M_{f}, X)$ in homology, combined with $f_{*}$ being an iso on each $H_{n}$ , gives $H_{n} (M_{f}, X) = 0$ for every $n$ . The relative Hurewicz theorem 03.12.19 applied to the simply-connected pair $(M_{f}, X)$ — both spaces simply connected, the relative homology vanishing — gives $π_{n} (M_{f}, X) = 0$ for every $n \geq 1$ . The case $n = 0$ is automatic since $X ↪ M_{f}$ is a path-connected inclusion. Whitehead's theorem (Step 3 above, applied to $(W, A) = (M_{f}, X)$ ) concludes $X ↪ M_{f}$ is a homotopy equivalence, hence so is $f$ . $□$

Theorem (CW approximation), proof. Construct $Y^{'}$ inductively. Start with a discrete set of points $Y^{' (0)}$ mapping bijectively onto the path components of $Y$ . Suppose $Y^{' (n)}$ has been built with a map $ρ_{n} : Y^{' (n)} \to Y$ inducing surjections on $π_{k}$ for $k \leq n$ and isomorphisms for $k < n$ . To kill the kernel of $π_{n} (Y^{' (n)}) \to π_{n} (Y)$ , attach $(n + 1)$ -cells along representatives of each generator of the kernel, extending $ρ_{n}$ to send each cell into $Y$ via the null-homotopy that exists because the class died in $π_{n} (Y)$ . To make $ρ$ surjective on $π_{n + 1}$ , attach $(n + 1)$ -cells of the form $S^{n + 1} \to Y^{'}$ for each generator of $π_{n + 1} (Y)$ , with attaching map mapping into $Y^{' (n)}$ via the existing $ρ_{n}$ -pullback. The result $Y^{' (n + 1)}$ together with the extended $ρ_{n + 1}$ has the inductive properties shifted up by one. Pass to the colimit $Y^{'} = colim Y^{' (n)}$ with the weak topology; the assembled $ρ : Y^{'} \to Y$ induces isomorphisms on every $π_{n}$ , hence is a weak equivalence.

Uniqueness up to homotopy equivalence over $Y$ : given two CW approximations $ρ_{1} : Y_{1}^{'} \to Y$ and $ρ_{2} : Y_{2}^{'} \to Y$ , lift $ρ_{1}$ along $ρ_{2}$ to a map $h : Y_{1}^{'} \to Y_{2}^{'}$ with $ρ_{2} \circ h ≃ ρ_{1}$ . The lift exists because $ρ_{2}$ is a weak equivalence and $Y_{1}^{'}$ is a CW complex (the obstruction to lifting at each cell vanishes by the weak-equivalence hypothesis). Then $h_{*} = ρ_{2, *}^{- 1} \circ ρ_{1, *}$ on every $π_{n}$ , an isomorphism. By Whitehead's theorem, $h$ is a homotopy equivalence. $□$

Theorem (Warsaw circle), proof of properties. The Warsaw circle $W = \overline{Γ} \cup α$ , where $Γ = {(x, sin (1/ x)) : 0 < x \leq 1}$ and $α$ is a continuous arc from a point of $\overline{Γ}$ to $(0, 0)$ avoiding the limiting segment $L = {0} \times [- 1, 1]$ except at its endpoint. Path-connectedness uses $α$ . To compute $π_{n} (W)$ , observe that any continuous $S^{n} \to W$ has compact image; compact subsets of $W$ either avoid $L$ entirely (in which case they sit in a contractible portion of $\overline{Γ} \cup α$ ) or contain a portion of $L$ , in which case the map factors through a contractible neighbourhood of that portion (the limiting segment is itself contractible, and the only way to reach $L$ from $\overline{Γ}$ is through $α$ ). In either case the map is null-homotopic. Hence $π_{n} (W) = 0$ for every $n \geq 0$ .

Non-contractibility: a deformation retract $W \to pt$ would give a continuous family of paths $γ_{w} : [0, 1] \to W$ with $γ_{w} (0) = w$ and $γ_{w} (1)$ a fixed point. As $w$ approaches $L$ along $\overline{Γ}$ , the paths $γ_{w}$ must travel through $α$ to reach the basepoint, but the lengths of these paths grow without bound (the oscillations force the path to traverse most of $\overline{Γ}$ ). Continuous dependence on $w$ fails at $L$ . Hence no such deformation retract exists. $□$

Theorem (relative form), proof. Mirror the absolute proof, with $W = X$ (or $M_{f}$ in the cylinder reduction), $A =$ subcomplex containing $A$ from $(X, A)$ glued to the cylinder structure. The skeleton-by-skeleton extension uses the relative homotopy extension property for triples $(X, A, B)$ where $B \subset A \subset X$ is a CW triple, and the obstruction at each cell lives in the relative homotopy group of the target pair. The vanishing of these obstructions is the hypothesis. $□$

Theorem (model-categorical packaging), proof sketch. In the Quillen model structure on Top, the weak equivalences are the weak homotopy equivalences and the cofibrations are the relative CW inclusions (more precisely, the retracts of relative cell complexes). The cofibrant objects are exactly the spaces with the homotopy type of a CW complex. Whitehead's theorem is the standard model-categorical statement: between cofibrant-fibrant objects, weak equivalences are homotopy equivalences. The proof in this generality (Hovey Model Categories Theorem 1.2.10) abstracts the inductive cell-by-cell extension into the small-object argument. The CW form is the concrete instance with explicit cells. $□$

Connections [Master]

CW complex 03.12.10. Whitehead's theorem is the structural reason the CW framework dominates homotopy theory. Every CW complex $X$ has its homotopy type determined by the sequence $π_{n} (X)$ together with the action of $π_{1}$ on higher groups and the $k$ -invariants from the Postnikov tower. The theorem is what licenses treating "the CW complex" as a homotopy-theoretic object: any two CW complexes inducing the same homotopy-group data are the same as homotopy types. Without Whitehead's theorem, the CW category would be one model among many; with it, the CW category is canonically equivalent to the homotopy category of weak homotopy types.
Homotopy and homotopy group 03.12.01. The theorem identifies the homotopy groups as the complete homotopy invariant for CW complexes. Computing $π_{n} (X)$ for a CW complex $X$ is therefore the central problem of CW homotopy theory: solve it for every $n$ and the homotopy type is determined. The corollary direction — that a homotopy equivalence induces isomorphisms on every homotopy group — is formal; the theorem's content is the converse, which makes the homotopy groups a complete rather than merely necessary invariant.
Hurewicz theorem 03.12.19. The Hurewicz theorem and Whitehead's theorem together produce the homology version: a map between simply-connected CW complexes inducing isomorphisms on every integer homology group is a homotopy equivalence. The Hurewicz theorem reduces homology iso to homotopy iso under simple-connectedness; Whitehead's theorem converts homotopy iso to homotopy equivalence. The combined statement is the foundational tool for showing two simply-connected CW complexes are homotopy equivalent through homology calculations alone — far more practical than direct comparison of homotopy groups.
Eilenberg-MacLane spaces 03.12.05. The Eilenberg-MacLane space $K (π, n)$ is characterised by having $π_{n} = π$ and all other homotopy groups zero. Whitehead's theorem makes this characterisation rigid: any two CW complexes with these homotopy-group data are homotopy equivalent. The construction of $K (π, n)$ proceeds by attaching cells to a model of $π_{n}$ to kill higher homotopy groups, with the same skeleton-by-skeleton obstruction argument that drives the proof of Whitehead's theorem. The resulting characterising property — uniqueness up to homotopy equivalence — is exactly Whitehead's theorem applied.
Whitehead tower 03.12.07. The Whitehead tower of a space $X$ is the sequence of fibrations $\dots \to X ⟨ n ⟩ \to X ⟨ n - 1 ⟩ \to \dots \to X$ where $X ⟨ n ⟩$ is $n$ -connected. The tower is well-defined up to homotopy equivalence at each stage by Whitehead's theorem: the construction kills successive homotopy groups, and the resulting space is determined up to homotopy equivalence by its homotopy-group profile. The theorem is also what justifies passing freely between models of the same Whitehead-tower stage built by different cell-attachment schemes.
CW approximation and model categories 03.12.10. Every topological space admits a CW approximation — a weak equivalence from a CW complex — unique up to homotopy equivalence by Whitehead's theorem. This extends the CW framework to all of Top: the homotopy category of CW complexes is equivalent to the homotopy category of all spaces under weak equivalences. The theorem is the bridge that makes the abstract Quillen model structure on Top concretely computable through CW models.

Historical & philosophical context [Master]

J.H.C. Whitehead introduced both CW complexes and the theorem that bears his name in the 1949 papers Combinatorial homotopy I and Combinatorial homotopy II (Bull. Amer. Math. Soc. 55, 213-245 and 453-496) ^[pending]. The two papers together produced the modern framework for combinatorial-cellular topology: CW complexes as the basic objects (paper I), the homotopy extension property and the cellular approximation theorem (paper I), and the equivalence-detection theorem now called Whitehead's theorem (paper II). The framework was a deliberate response to the technical difficulties of working with general topological spaces in homotopy theory; Whitehead's introductory remarks in paper I cite the Warsaw circle and similar pathologies as the motivating examples for restricting attention to cellular spaces.

Whitehead's theorem in paper II was stated and proved for CW complexes specifically. The skeleton-by-skeleton inductive argument using the homotopy extension property was Whitehead's; the formulation in terms of the mapping cylinder (which simplifies the bookkeeping) and the relative form of the theorem were standard within a few years through Steenrod's seminars and the Eilenberg-Steenrod axiomatic treatment in Foundations of Algebraic Topology (Princeton 1952) ^[pending].

The conceptual impact of the theorem ran in two directions. Within algebraic topology, it justified the CW framework as the category for computational homotopy theory: every theorem about homotopy types could be stated and proved for CW complexes first, then transported to general spaces by CW approximation. Within abstract homotopy theory, the theorem became the prototype for the modern model-categorical formulation: a weak equivalence between cofibrant-fibrant objects is a homotopy equivalence, with CW complexes playing the role of cofibrant-fibrant in the standard Quillen model structure on Top. Quillen's Homotopical Algebra (Springer LNM 43, 1967) ^[pending] made this packaging explicit, and Hovey's Model Categories (AMS 1999) ^[pending] consolidated the modern presentation.

The Warsaw-circle counterexample and the continued sharpness of the CW hypothesis remain part of the standard pedagogical narrative — Hatcher's Example 4.36 reproduces it directly from Whitehead's original paper II. The theorem's role as the bridge between weak equivalence and homotopy equivalence is what makes the homotopy category of CW complexes the natural setting for algebraic topology.

Bibliography [Master]

[object Promise]

Prerequisites

03.12.10
03.12.01
03.12.19

Used in

03.12.05
03.12.07

Tier anchors

beginner: Hatcher *Algebraic Topology* §4.1 informal opening on weak vs strong equivalence
intermediate: Hatcher §4.1 (Theorem 4.5 + cellular approximation); May *A Concise Course in Algebraic Topology* Ch. 10
master: Whitehead 1949 — *Combinatorial homotopy I+II* (originator papers, Bull. AMS 55); Hatcher §4.1 (full proof + Warsaw counterexample); tom Dieck *Algebraic Topology* §8

References

TODO_REF
Hatcher — Algebraic Topology · §4.1 Theorem 4.5 (Whitehead's theorem), Proposition 4.13 (cellular approximation), Example 4.36 (Warsaw circle counterexample)
TODO_REF
Whitehead — Combinatorial homotopy I · Bull. AMS 55 (1949) 213-245, originator paper for CW complexes and HEP
TODO_REF
Whitehead — Combinatorial homotopy II · Bull. AMS 55 (1949) 453-496, originator paper for the CW form of the theorem
TODO_REF
May — A Concise Course in Algebraic Topology · Ch. 10 (CW complexes), Ch. 13 (the Whitehead theorem and CW approximation)
TODO_REF
tom Dieck — Algebraic Topology · §8 (CW complexes, cellular approximation, Whitehead's theorem)
TODO_REF
Spanier — Algebraic Topology · Ch. VII §6 weak homotopy equivalences, Ch. VIII obstruction theory

Reviewer

TBD

Estimated time

beginner: 12m
intermediate: 30m
master: 60m