03.12.47 · modern-geometry / homotopy

HELP and the unified Whitehead / cellular approximation theorem

shipped3 tiersLean: none

Anchor (Master): Whitehead 1949 *Combinatorial homotopy I/II* (Bull. Amer. Math. Soc. 55, originator of CW complexes, HEP, cellular approximation, equivalence detection); May *A Concise Course in Algebraic Topology* (University of Chicago Press, 1999) Ch. 10 (the HELP packaging); tom Dieck *Algebraic Topology* (EMS, 2008) Ch. 6 (relative CW theory)

Intuition Beginner

There is a recurring situation in the study of spaces built from cells. You have a big space made by gluing simple round pieces — points, line segments, disks, solid balls — one dimension at a time. Inside it sits a smaller piece you already understand. You have a map defined on the small piece that lands where you want it, and a rough map on the whole big space that lands somewhere nearby. The natural wish is to upgrade the rough map into a good one that agrees with the part you already control, without disturbing it.

May's organizing idea is that one single statement handles this wish. It is called HELP, short for the Homotopy Extension and Lifting Property. It says: when the target map you are comparing against is good enough through the dimensions of the cells involved, you can always extend the controlled part to the whole space and adjust the rough map to match, in one motion.

The payoff is striking. Two of the most quoted results about cellular spaces — that matching loop-and-sphere data forces two spaces to be the same shape, and that any map can be nudged to respect the cell structure — both fall out of this one statement in a single line each. Instead of two separate climbs up a ladder of dimensions, there is one climb, reused twice.

Visual Beginner

Picture a square diagram with four corners. The bottom edge is the small understood piece, drawn as a thick dot, with an arrow into a space labelled "good target". The top edge is the whole cellular space, drawn as a stack of growing round pieces, with an arrow into a "rough target". A wavy band between the two targets marks that they are linked by a comparison map that is good enough in low dimensions. The content of HELP is a single new diagonal arrow from the whole space into the good target, dashed to show it is the thing produced, together with a small curved arrow recording that the rough map gets adjusted to match.

The single dashed diagonal is the whole story. Everything else in cellular homotopy theory that looks like two theorems is this one diagonal, drawn in two settings.

Worked example Beginner

Here is the simplest case of the lifting idea, before any cells of positive dimension enter.

Step 1. Take the small understood piece to be a single point $a$ , sitting inside a space $X$ that has $a$ together with one extra dot $b$ glued on, with nothing connecting them yet. Take a target map that sends $a$ to a chosen spot $y_{0}$ in a space $Y$ .

Step 2. Suppose the comparison map from $Y$ into a second space $Z$ matches up the path-pieces of $Y$ with the path-pieces of $Z$ . A rough map sends the whole of $X$ into $Z$ , landing $b$ somewhere.

Step 3. Because the comparison map matches path-pieces, the spot in $Z$ where $b$ lands comes from some spot in $Y$ in the right path-piece. Pick that spot and send $b$ there. Now $X$ maps into $Y$ , agreeing with the controlled value at $a$ .

Step 4. The adjusting homotopy slides the rough image of $b$ along a path in $Z$ to the image of the chosen spot. The point $a$ never moves.

What this shows: even with no curved cells, lifting works because the comparison map controls path-pieces. The full statement just repeats this one dimension at a time, with disks in place of dots and the cofibration sliding rule in place of the path. The pattern of "match on the small piece, then extend and adjust" is the entire mechanism, and it is what gets reused to prove both headline theorems.

Check your understanding Beginner

Formal definition Intermediate+

Throughout, a relative CW pair $(X, A)$ is a space $X$ obtained from a subspace $A$ by attaching cells of increasing dimension: $A = X^{- 1} \subseteq X^{0} \subseteq X^{1} \subseteq \dots$ with $X^{n}$ formed from $X^{n - 1}$ by pushing out a coproduct of attaching maps $∐_{α} S^{n - 1} \to X^{n - 1}$ along $∐_{α} D^{n}$ , and $X = colim_{n} X^{n}$ in the weak topology. The inclusion $A ↪ X$ is a cofibration, so the pair has the homotopy extension property.

Let $n$ be a positive integer or $\infty$ . A map $e : Y \to Z$ is an $n$ -equivalence when, at every base point $y \in Y$ , the induced map $e_{*} : π_{q} (Y, y) \to π_{q} (Z, e (y))$ is a bijection for $q < n$ and a surjection for $q = n$ , together with the corresponding statement on path components for $q = 0$ . An $\infty$ -equivalence is precisely a weak homotopy equivalence. The defining feature of an $n$ -equivalence is a lifting statement against spheres and disks of low dimension: for $q \leq n$ , a square consisting of a map $S^{q - 1} \to Y$ , a map $D^{q} \to Z$ , and a homotopy $e \circ (S^{q - 1} \to Y) ≃ (D^{q} \to Z) ∣_{S^{q - 1}}$ admits a diagonal lift $D^{q} \to Y$ filling it, compatibly with the homotopy. This compression statement against a single cell is the engine that the cellular induction feeds on.

HELP (the Homotopy Extension and Lifting Property; May Ch. 10). Let $(X, A)$ be a relative CW pair of dimension $\leq n$ (all cells of $X ∖ A$ have dimension $\leq n$ ), and let $e : Y \to Z$ be an $n$ -equivalence. Suppose given a commuting outer frame: a map $g : A \to Y$ , a map $h : X \to Z$ , and a homotopy $k : e \circ g ≃ h ∣_{A}$ . Then there exist a map $\tilde{g} : X \to Y$ extending $g$ and a homotopy $\tilde{k} : e \circ \tilde{g} ≃ h$ extending $k$ , both rel $A$ .

The statement is one diagram. The two named consequences are two readings of it. Whitehead's theorem reads it with $A = \emptyset$ and $e = id_{Y}$ for a weak equivalence between CW complexes, lifting the identity of $Y$ back across $e$ to build a homotopy inverse. Cellular approximation reads it with $e$ the skeleton inclusion $Z^{n} ↪ Z$ , an $n$ -equivalence, deforming an arbitrary map into a cellular one rel the subcomplex where it is already cellular.

Counterexamples to common slips

The dimension bound matters. If $(X, A)$ has cells of dimension exceeding $n$ and $e$ is only an $n$ -equivalence, the lift can fail at the first oversized cell, where the compression statement no longer applies. HELP for all dimensions needs $e$ a weak equivalence.
The map $e$ must be an $n$ -equivalence at every base point of $Y$ , not at one. A map that is an isomorphism on the homotopy groups of one component while mismatching another component is not an $n$ -equivalence, and the cell-by-cell lift breaks at a cell whose attaching map lands in the bad component.
HELP produces a lift only up to a homotopy rel $A$ ; it does not produce $e \circ \tilde{g} = h$ on the nose. Reading the conclusion as a strict equality rather than a homotopy is the standard error and would force $e$ to be far more than an $n$ -equivalence.
The cofibration hypothesis on $A ↪ X$ is what licenses extending each partial homotopy. Dropping it — using a pair that is not a relative CW pair — removes the extension half of the property even when the lifting half survives.

Key theorem with proof Intermediate+

Theorem (HELP; May Ch. 10). Let $(X, A)$ be a relative CW pair of dimension $\leq n$ and let $e : Y \to Z$ be an $n$ -equivalence. Given $g : A \to Y$ , $h : X \to Z$ , and a homotopy $k : e g ≃ h ∣_{A}$ , there exist $\tilde{g} : X \to Y$ extending $g$ and a homotopy $\tilde{k} : e \tilde{g} ≃ h$ extending $k$ , both rel $A$ .

Proof. Induct over the skeletal filtration $A = X^{- 1} \subseteq X^{0} \subseteq \dots$ . The inductive claim is that $g$ and $k$ extend over $X^{m}$ for each $m \leq n$ , compatibly, yielding $g_{m} : X^{m} \to Y$ and $k_{m} : e g_{m} ≃ h ∣_{X^{m}}$ rel $A$ . The base case $m = - 1$ is the given data.

Assume $g_{m - 1}$ and $k_{m - 1}$ are constructed. The space $X^{m}$ is obtained from $X^{m - 1}$ by attaching $m$ -cells along $φ : ∐_{α} S^{m - 1} \to X^{m - 1}$ . It suffices to extend over a single $m$ -cell, the construction being indexed over $α$ . For one cell with attaching map $φ_{α} : S^{m - 1} \to X^{m - 1}$ and characteristic map $Φ_{α} : D^{m} \to X^{m}$ , consider the boundary data: $g_{m - 1} \circ φ_{α} : S^{m - 1} \to Y$ , the restriction $h \circ Φ_{α} : D^{m} \to Z$ , and the boundary homotopy $k_{m - 1}$ pulled back along $φ_{α}$ . This is a compression square against the cell $D^{m}$ with boundary $S^{m - 1}$ , and $m \leq n$ , so the $n$ -equivalence $e$ furnishes a diagonal lift $\overset{g}{ˉ}_{α} : D^{m} \to Y$ extending $g_{m - 1} \circ φ_{α}$ , together with a homotopy $\overset{ˉ}{k}_{α} : e \overset{g}{ˉ}_{α} ≃ h Φ_{α}$ extending the pulled-back $k_{m - 1}$ .

The lifts $\overset{g}{ˉ}_{α}$ and homotopies $\overset{ˉ}{k}_{α}$ over the separate cells agree with $g_{m - 1}, k_{m - 1}$ on the attaching spheres by construction. Because $X^{m - 1} ↪ X^{m}$ is a cofibration — this is the homotopy extension property of the CW pair — the partial data assembled on $X^{m - 1}$ and on the cell interiors glue into continuous $g_{m} : X^{m} \to Y$ and $k_{m} : e g_{m} ≃ h ∣_{X^{m}}$ rel $A$ . This closes the induction at level $m$ .

Pass to the colimit. Since $X = colim_{m} X^{m}$ carries the weak topology and the $g_{m}$ , $k_{m}$ are coherent ( $g_{m} ∣_{X^{m - 1}} = g_{m - 1}$ , likewise for the homotopies), they assemble into $\tilde{g} : X \to Y$ and $\tilde{k} : e \tilde{g} ≃ h$ rel $A$ . (When $n = \infty$ the colimit runs over all $m$ ; when $n$ is finite the pair has dimension $\leq n$ , so the filtration stabilises at level $n$ .) $□$

Corollary (Whitehead's theorem). Let $f : Y \to Z$ be a weak equivalence between CW complexes. Apply HELP with $A = \emptyset$ , $(X, A) = (Z, \emptyset)$ , $e = f$ , $g$ empty, $h = id_{Z}$ , and $k$ constant: the lift $\tilde{g}$ is a map $g : Z \to Y$ with $f g ≃ id_{Z}$ . Apply HELP once more to the pair $(Z \times I, Z \times \partial I)$ to upgrade the one-sided inverse to a two-sided homotopy inverse. So $f$ is a homotopy equivalence.

Corollary (cellular approximation). Let $f : X \to Z$ be a map of CW complexes that is already cellular on a subcomplex $A$ . The skeleton inclusion $e = (Z^{n} ↪ Z)$ is an $n$ -equivalence (the relative homotopy groups $π_{q} (Z, Z^{n})$ vanish for $q \leq n$ ). Apply HELP with this $e$ , with $h = f$ and $g$ the given cellular map on $A$ , dimension by dimension over $n$ : the lift $\tilde{g}$ is a cellular map $X \to Z$ homotopic to $f$ rel $A$ .

Bridge. HELP builds toward the entire relative-CW toolkit by isolating the one induction that all of cellular homotopy theory keeps re-running, and the foundational reason it works is exactly the pairing of two prerequisites: the cofibration homotopy extension property of 02.01.08 supplies the extension half, and the compression property of an $n$ -equivalence on 03.12.01 homotopy groups supplies the lifting half, cell by cell. This is exactly the same skeleton-by-skeleton mechanism that the obstruction-theoretic proof of Whitehead in 03.12.20 runs by hand, so HELP generalises that single-purpose argument into a reusable lemma; putting these together, the two headline theorems stop being separate climbs and become one. The central insight is that "extend on the subcomplex, then lift over the next cell" is a property of the pair $(X, A)$ and the map $e$ jointly, not of either theorem, so the same diagram that proves a weak equivalence is invertible also deforms an arbitrary map onto the skeleton. This packaging appears again in 03.12.10 whenever a construction on a CW complex is built one skeleton at a time, and it is the bridge from the cofibration formalism to the equivalence-detection theorems.

Exercises Intermediate+

Exercise 4 (medium, short-answer).

Explain why the skeleton inclusion $Z^{n} ↪ Z$ of a CW complex is an $n$ -equivalence, the input cellular approximation feeds to HELP.

Hint

Compare maps and homotopies of spheres of dimension $\leq n$ using cellular approximation of the skeleton inclusion itself, or the vanishing of $π_{q} (Z, Z^{n})$ .

Answer

The relative homotopy groups $π_{q} (Z, Z^{n})$ vanish for $q \leq n$ : a map $(D^{q}, S^{q - 1}) \to (Z, Z^{n})$ with $q \leq n$ can be pushed off the cells of dimension exceeding $n$ by a transversality/general-position argument, landing in $Z^{n}$ . The long exact sequence of the pair then makes $π_{q} (Z^{n}) \to π_{q} (Z)$ an isomorphism for $q < n$ and a surjection for $q = n$ , which is the definition of an $n$ -equivalence. Rubric: full credit for the relative-group vanishing plus the long-exact-sequence conclusion.

Exercise 5 (medium, symbolic).

Let $(X, A)$ have a single $2$ -cell over $A$ , attached along $φ : S^{1} \to A$ . Let $e : Y \to Z$ be a $2$ -equivalence. Describe the compression square solved at this cell and identify, in terms of homotopy groups, the obstruction that the $2$ -equivalence kills.

Hint

The boundary data assembles a relative class in $π_{2}$ of the comparison.

Answer

The cell contributes the square with $g ∣_{A} \circ φ : S^{1} \to Y$ on the boundary, $h \circ Φ : D^{2} \to Z$ on the disk, and the boundary homotopy $k$ relating $e \circ (g φ)$ to $h Φ ∣_{S^{1}}$ . Filling the square is the same as compressing a class in $π_{2}$ of the homotopy fibre of $e$ at the relevant base point — the obstruction to lifting $h Φ$ to $Y$ rel its boundary lift. A $2$ -equivalence makes $π_{q}$ of that fibre vanish for $q \leq 1$ and act suitably at $q = 2$ , so the lift over the $2$ -cell exists. Rubric: full credit for naming the homotopy-fibre obstruction and tying its vanishing to the $2$ -equivalence.

Exercise 6 (hard, short-answer).

Use HELP to prove the relative form: if $(X, A)$ is a relative CW pair and $e : Y \to Z$ is a weak equivalence, then for any $h : X \to Z$ restricting on $A$ to a map that already lifts through $e$ , the lift extends over all of $X$ . Then deduce that any map of CW complexes is homotopic rel a chosen subcomplex to a cellular one.

Hint

Run HELP with $n = \infty$ for the first part; for the second take $e$ to be the skeleton inclusion and $A$ the subcomplex where $h$ is already cellular.

Answer

For the first part, $e$ a weak equivalence is an $\infty$ -equivalence, so HELP applies with no dimension restriction on $(X, A)$ . The data $g$ (the lift on $A$ ), $h$ , and the connecting homotopy $k$ produce $\tilde{g} : X \to Y$ extending $g$ with $e \tilde{g} ≃ h$ rel $A$ , the asserted extension of the lift. For the second part, let $f : X \to Z$ be cellular on a subcomplex $A$ . Take $e = (Z^{n} ↪ Z)$ , an $n$ -equivalence, and apply HELP over the skeleta of $X$ in increasing $n$ ; each application deforms $f$ on the next skeleton into $Z^{n}$ rel $A$ and rel the lower skeleta already handled. The colimit of these deformations is a cellular map homotopic to $f$ rel $A$ . Rubric: full credit for the $\infty$ -equivalence run plus the skeleton-by-skeleton deduction of cellular approximation rel $A$ .

Exercise 7 (hard, short-answer).

Contrast HELP with the obstruction-theoretic proof of Whitehead in 03.12.20. State precisely what each induction extends across a cell and what plays the role of the "obstruction" in each.

Hint

In the obstruction proof the input is the vanishing of a relative homotopy group; in HELP the input is a compression lift. Show they are the same datum.

Answer

The obstruction proof of Whitehead replaces $f$ by a mapping-cylinder inclusion $X ↪ M_{f}$ , observes that $π_{q} (M_{f}, X)$ vanishes for all $q$ , and extends a partial deformation retraction across each cell, with the element of $π_{q} (M_{f}, X)$ attached to the cell as the obstruction that the vanishing kills. HELP extends a lift $\tilde{g}$ and a homotopy $\tilde{k}$ across each cell, with the compression square against $D^{q}$ as the datum solved; the existence of that compression lift is exactly the vanishing of the relevant relative homotopy group of the comparison map. So the two inductions extend the same kind of per-cell data and use the same per-cell input, but HELP states that input as a property of a map $e$ reusable in other settings, while the obstruction proof states it as a one-off vanishing for the specific pair $(M_{f}, X)$ . Rubric: full credit for matching the compression lift to the relative-group vanishing and noting the reusability difference.

Advanced results Master

Theorem (HELP, dimension-graded form). Let $(X, A)$ be a relative CW pair and let $e : Y \to Z$ be an $n$ -equivalence. The lifting-and-extension conclusion of HELP holds over the sub-pair $(X^{n}, A)$ generated by cells of dimension $\leq n$ , and over the cells of dimension exactly $n + 1$ it holds up to a choice obstruction in $π_{n + 1}$ of the homotopy fibre of $e$ .

The graded form is what one wants for finite obstruction calculations. When $e$ is only an $n$ -equivalence, HELP runs cleanly through dimension $n$ , and the first genuine choice appears one dimension up, where the surjectivity at the top degree leaves a coset of lifts rather than a forced one. This is the precise sense in which an $n$ -equivalence is "good through dimension $n$ ": the induction is forced below $n$ and has controlled indeterminacy at $n + 1$ .

Theorem (Whitehead via HELP, both directions). A weak equivalence $f : Y \to Z$ between CW complexes is a homotopy equivalence. A homotopy inverse is produced by applying HELP to $(Z, \emptyset)$ to get a right inverse $g$ , then to $(Y \times I, Y \times \partial I)$ to convert $g$ into a two-sided inverse.

The two-step structure is the whole proof. No mapping cylinder, no long exact sequence of a pair, no separate cellular-approximation lemma invoked from outside: the single HELP statement, read twice, gives the equivalence. This is the economy that the obstruction-theoretic route in 03.12.20 cannot match, because that route hard-codes the induction into the specific pair $(M_{f}, X)$ and cannot reuse it for cellular approximation.

Theorem (cellular approximation via HELP, relative). Any map $f : X \to Z$ of CW complexes is homotopic, rel any subcomplex on which it is already cellular, to a cellular map. The homotopy is built by applying HELP with $e$ the skeleton inclusions $Z^{n} ↪ Z$ , one $n$ at a time.

The relative clause is free in the HELP formulation because the property is stated rel $A$ from the start. In the hand-built proof one has to carry the subcomplex through the induction explicitly; here it is the subspace of the relative CW pair and the bookkeeping is automatic.

Theorem (uniqueness of CW approximation, one line). Two CW approximations of a space $W$ are homotopy equivalent over $W$ . Given weak equivalences $ρ_{1} : Y_{1} \to W$ and $ρ_{2} : Y_{2} \to W$ from CW complexes, apply HELP to $(Y_{1}, \emptyset)$ against the $\infty$ -equivalence $ρ_{2}$ with $h = ρ_{1}$ to lift $ρ_{1}$ to a map $Y_{1} \to Y_{2}$ over $W$ ; the lift is a weak equivalence between CW complexes, hence a homotopy equivalence by the previous theorem.

This corollary shows the packaging compounding: a result that takes a separate lifting argument in the obstruction-theoretic development is, in the HELP development, a single application of the same lemma against an $\infty$ -equivalence.

Theorem (model-categorical reading). In the Quillen model structure on spaces, HELP is the statement that relative CW inclusions have the left lifting property against the maps $e$ that are, in the appropriate range, acyclic fibrations — the cofibration-versus-acyclic-fibration lifting that defines the model structure, specialised to the cellular generating cofibrations and the $n$ -equivalences.

The model-categorical packaging shows HELP is not a topological accident but the cellular shadow of the lifting axiom. The relative CW pairs are the cofibrations built from the generating set ${S^{n - 1} ↪ D^{n}}$ , the $n$ -equivalences are the maps with the right lifting property against generators up to dimension $n$ , and HELP is the lift guaranteed by that orthogonality. The small-object argument is the abstract engine; the cellular induction is its hands-on instance.

Synthesis. HELP is the foundational reason that cellular homotopy theory needs only one induction: the central insight is that "extend on the subcomplex via the cofibration property, then lift over the next cell via the $n$ -equivalence" is a property of the pair $(X, A)$ and the comparison map $e$ jointly, so the same diagram does double duty. Whitehead's theorem is exactly this diagram read with an empty subcomplex and an identity target, and cellular approximation is exactly this diagram read with a skeleton inclusion as the comparison map; putting these together, two theorems that the obstruction-theoretic route of 03.12.20 proves by two bespoke inductions become two specialisations of one lemma. The packaging generalises further at no extra cost: the uniqueness of CW approximation is the same lemma against an $\infty$ -equivalence, and the dimension-graded form is the same lemma with the indeterminacy at degree $n + 1$ made explicit, which is what obstruction theory needs. This is dual to the per-pair vanishing statements of the hand-built proof — where that route attaches a relative-homotopy obstruction to the specific cylinder pair, HELP states the identical per-cell datum as reusable structure on the map $e$ . The bridge is the recognition that the cofibration formalism of 02.01.08 and the $n$ -equivalence formalism of 03.12.01 fit together into a single lifting property, and that property, not either theorem, is the primitive object; this is exactly the organising principle that recurs through the construction of 03.12.10 CW complexes one skeleton at a time, and it is May's signature move.

Full proof set Master

Proposition 1 (compression lemma for an $n$ -equivalence). Let $e : Y \to Z$ be an $n$ -equivalence and $q \leq n$ . Given $a : S^{q - 1} \to Y$ , $b : D^{q} \to Z$ with $b ∣_{S^{q - 1}} = e \circ a$ , there is a lift $\overset{a}{ˉ} : D^{q} \to Y$ extending $a$ with $e \circ \overset{a}{ˉ} ≃ b$ rel $S^{q - 1}$ .

Proof. Form the homotopy fibre $F$ of $e$ over the base point $b (*)$ , with its fibration $F \to Y \to Z$ . The pair $(b, a)$ together with the constant homotopy assembles a relative class in $π_{q} (Z, Y)$ via the long exact sequence $$ \cdots \to \pi_q(Y) \xrightarrow{e_*} \pi_q(Z) \to \pi_q(Z, Y) \to \pi_{q-1}(Y) \xrightarrow{e_*} \pi_{q-1}(Z) \to \cdots . $$ For $q \leq n$ , the map $e_{*}$ is surjective at degree $q$ and bijective below, so $π_{q} (Z, Y) = 0$ . The vanishing of this relative group is precisely the existence of the diagonal lift $\overset{a}{ˉ}$ filling the square $(a, b)$ , with the homotopy $e \overset{a}{ˉ} ≃ b$ rel $S^{q - 1}$ recording the nullhomotopy of the relative class. $□$

Proposition 2 (the inductive step of HELP is well-posed). In the skeletal induction proving HELP, the per-cell lifts and homotopies glue to a continuous map on each skeleton.

Proof. Fix the skeleton $X^{m}$ , formed from $X^{m - 1}$ by attaching $m$ -cells along $φ : ∐_{α} S^{m - 1} \to X^{m - 1}$ through the pushout $$ \begin{array}{ccc} \coprod_\alpha S^{m-1} & \longrightarrow & X^{m-1} \ \downarrow & & \downarrow \ \coprod_\alpha D^m & \longrightarrow & X^m . \end{array} $$ On $X^{m - 1}$ the maps $g_{m - 1}, k_{m - 1}$ are already defined; on each $D^{m}$ Proposition 1 supplies $\overset{g}{ˉ}_{α}, \overset{ˉ}{k}_{α}$ agreeing with $g_{m - 1} \circ φ_{α}$ on the boundary sphere. The universal property of the pushout assembles a continuous map out of $X^{m}$ from data on $X^{m - 1}$ and on $∐_{α} D^{m}$ that agree on $∐_{α} S^{m - 1}$ , which is exactly the agreement just arranged. The homotopies glue identically. Because $X^{m - 1} ↪ X^{m}$ is a cofibration, the assembled homotopy $k_{m}$ extends consistently rel $A$ . $□$

Proposition 3 (Whitehead's theorem as a corollary). A weak equivalence $f : Y \to Z$ between CW complexes is a homotopy equivalence.

Proof. Apply HELP to $(X, A) = (Z, \emptyset)$ with $e = f$ (an $\infty$ -equivalence), $h = id_{Z}$ , and the empty $g$ , $k$ . This produces $g : Z \to Y$ with $f g ≃ id_{Z}$ . Now apply HELP to the relative CW pair $(Y \times I, Y \times \partial I)$ against $e = f$ , with the map on $Y \times \partial I$ given by $id_{Y}$ at one end and $g f$ at the other, and $h$ the homotopy $f ≃ f g f$ obtained by composing $f g ≃ id_{Z}$ with $f$ . The lift is a homotopy $g f ≃ id_{Y}$ rel nothing, so $g$ is a two-sided homotopy inverse and $f$ is a homotopy equivalence. $□$

Proposition 4 (cellular approximation as a corollary). Any map $f : X \to Z$ of CW complexes is homotopic rel a subcomplex $A$ on which it is cellular to a cellular map.

Proof. For each $n$ , the inclusion $e_{n} : Z^{n} ↪ Z$ is an $n$ -equivalence (Proposition 1's input, since $π_{q} (Z, Z^{n}) = 0$ for $q \leq n$ by general position). Apply HELP to the pair $(X^{n}, A \cup X^{n - 1})$ against $e_{n}$ , with $h = f ∣_{X^{n}}$ and $g$ the cellular map already built on the lower skeleton and on $A$ . The lift deforms $f ∣_{X^{n}}$ into $Z^{n}$ rel the controlled subspace, making it cellular on $X^{n}$ . Running over increasing $n$ and concatenating the deformations — coherent because each is rel the previous skeleton — yields a cellular map homotopic to $f$ rel $A$ , using the weak topology on $X$ to pass to the colimit of homotopies. $□$

Connections Master

CW complex 03.12.10. HELP is the operational heart of the CW formalism. Every construction that builds or compares maps on a CW complex by working up the skeleton filtration — attaching cells to kill or create homotopy classes, extending a partial map, comparing two cellular models — is an instance of the one induction HELP isolates. The skeleton filtration and the weak topology of 03.12.10 are exactly the data the colimit step consumes, and the cofibration of each skeleton inclusion is the extension half of the lemma.
Whitehead's theorem 03.12.20. This unit and 03.12.20 prove the same theorem by deliberately contrasted routes. The obstruction-theoretic proof in 03.12.20 replaces the map by a mapping cylinder and extends a deformation retraction across cells, attaching a relative homotopy class $π_{n} (M_{f}, X)$ to each cell as the obstruction killed by the weak-equivalence hypothesis. HELP states that identical per-cell datum as a compression property of the comparison map, reusable beyond Whitehead. The two are the same induction; HELP is the packaged, the obstruction proof the unpacked, version.
Homotopy and homotopy group 03.12.01. The notion of an $n$ -equivalence is phrased entirely in the homotopy groups of 03.12.01, and the lifting half of HELP is the compression statement that an $n$ -equivalence's induced maps on those groups make available through dimension $n$ . The dimension-graded form of HELP is what turns the abstract iso/surjection bookkeeping on $π_{q}$ into a concrete forced-then-indeterminate cell-by-cell extension.
Cofibration and homotopy extension property 02.01.08. The extension half of HELP is exactly the homotopy extension property of 02.01.08 applied to each skeleton inclusion. Without the cofibration structure, the per-cell lifts could not be glued back to the previously built map, and the induction would collapse. The two prerequisites — cofibration HEP and $n$ -equivalence lifting — are the two halves whose union is HELP, and the lemma is precisely their pairing into one diagram.

Historical & philosophical context Master

J.H.C. Whitehead created the entire substrate of this unit in the 1949 papers Combinatorial homotopy I and Combinatorial homotopy II (Bull. Amer. Math. Soc. 55, 213-245 and 453-496) ^{[Whitehead 1949]}. Paper I introduced CW complexes, the homotopy extension property, and the cellular approximation theorem; paper II proved the equivalence-detection theorem now bearing his name. In Whitehead's own development these were separate results with separate inductions, each climbing the skeleton filtration by hand. The motivation throughout was to escape the pathologies of general spaces — the Warsaw circle and its kin — by restricting to spaces assembled from cells, where homotopy-theoretic constructions could be controlled dimension by dimension.

The consolidation into a single lemma is J.P. May's, in A Concise Course in Algebraic Topology (University of Chicago Press, 1999) Ch. 10 ^{[May 1999]}. May observed that the cell-by-cell induction underlying Whitehead's theorem and the one underlying cellular approximation are not merely similar but identical once phrased as a lifting-and-extension property of a relative CW pair against an $n$ -equivalence. Naming this property HELP and proving it once, May derives both classical theorems as immediate corollaries — the empty-subcomplex reading for Whitehead, the skeleton-inclusion reading for cellular approximation — and gets the uniqueness of CW approximation almost for free. The move is characteristic of May's expository philosophy: find the single statement of maximal reuse and let the named theorems become its specialisations.

The deeper significance is structural. Whitehead's hand-built inductions were eventually recognised as the cellular shadow of the lifting axioms of a Quillen model category, where cofibrations lift against acyclic fibrations. HELP is the concrete instance: relative CW inclusions are the cofibrations generated by the sphere-disk inclusions, $n$ -equivalences are the maps orthogonal to those generators through dimension $n$ , and the small-object argument is the abstract form of the cellular induction. Reading the 1949 papers and Ch. 10 together traces the arc from Whitehead's combinatorial bookkeeping to the abstract homotopy theory that subsumes it, with HELP as the hinge between the cellular and the model-categorical pictures.

Bibliography Master

@book{MayConcise,
  author    = {May, J. Peter},
  title     = {A Concise Course in Algebraic Topology},
  publisher = {University of Chicago Press},
  year      = {1999}
}

@article{WhiteheadCH1,
  author  = {Whitehead, J. H. C.},
  title   = {Combinatorial homotopy. I},
  journal = {Bull. Amer. Math. Soc.},
  volume  = {55},
  year    = {1949},
  pages   = {213--245}
}

@article{WhiteheadCH2,
  author  = {Whitehead, J. H. C.},
  title   = {Combinatorial homotopy. II},
  journal = {Bull. Amer. Math. Soc.},
  volume  = {55},
  year    = {1949},
  pages   = {453--496}
}

@book{HatcherAlgebraicTopology,
  author    = {Hatcher, Allen},
  title     = {Algebraic Topology},
  publisher = {Cambridge University Press},
  year      = {2002}
}

@book{tomDieckAlgebraicTopology,
  author    = {tom Dieck, Tammo},
  title     = {Algebraic Topology},
  publisher = {European Mathematical Society},
  year      = {2008}
}

Prerequisites

03.12.10
03.12.20
03.12.01
02.01.08

Tier anchors

beginner: May *A Concise Course in Algebraic Topology* (University of Chicago Press, 1999) Ch. 10 informal opening — one lifting picture that does the work of two named theorems
intermediate: May *A Concise Course in Algebraic Topology* (University of Chicago Press, 1999) Ch. 10 (the statement and inductive proof of HELP from the cofibration HEP and n-equivalence lifting; corollaries Whitehead + cellular approximation)
master: Whitehead 1949 *Combinatorial homotopy I/II* (Bull. Amer. Math. Soc. 55, originator of CW complexes, HEP, cellular approximation, equivalence detection); May *A Concise Course in Algebraic Topology* (University of Chicago Press, 1999) Ch. 10 (the HELP packaging); tom Dieck *Algebraic Topology* (EMS, 2008) Ch. 6 (relative CW theory)

References

May — A Concise Course in Algebraic Topology · Ch. 10 (HELP, the unified statement; corollaries: Whitehead's theorem and cellular approximation)
Whitehead — Combinatorial homotopy I · Bull. Amer. Math. Soc. 55 (1949) 213-245, originator paper for CW complexes, HEP, cellular approximation
Whitehead — Combinatorial homotopy II · Bull. Amer. Math. Soc. 55 (1949) 453-496, originator paper for the CW form of the equivalence-detection theorem
Hatcher — Algebraic Topology · §4.1 (Whitehead's theorem and the compression lemma — the obstruction-theoretic route contrasted here)
tom Dieck — Algebraic Topology · Ch. 6 (relative CW complexes, HEP, the lifting/extension formalism)

Estimated time

beginner: 18m
intermediate: 45m
master: 85m