03.12.15 · modern-geometry / homotopy

Eilenberg-Steenrod axioms

shipped3 tiersLean: none

Anchor (Master): Eilenberg-Steenrod 1952 — *Foundations of Algebraic Topology* Ch. I-III (originator monograph); Eilenberg-Steenrod 1945 *Axiomatic approach to homology theory* (Proc. Nat. Acad. Sci. 31); Hatcher §2.3 + §4.E + §4.F (Brown representability); tom Dieck *Algebraic Topology* §17

Intuition [Beginner]

By the early 1940s, mathematicians had several ways of attaching algebraic invariants to a topological space — singular chains, simplicial chains, cellular chains, Čech chains, the Vietoris construction. Each method gave a sequence of abelian groups attached to the space, and on reasonable spaces every method gave the same answer. The mystery: why do all these different recipes produce the same groups? Eilenberg and Steenrod answered the mystery by writing down a short list of properties that any reasonable assignment of groups to spaces ought to satisfy, and then proving that the list pins down the answer uniquely on a wide class of spaces.

The list is the Eilenberg-Steenrod axioms. There are seven of them. Some are easy to motivate: a homotopy between two maps should not change the induced map on the groups; a pair of spaces should give a long exact sequence linking the groups of the two pieces; cutting out a piece deep inside one of the spaces should not change the relative groups; a disjoint union of spaces should have the direct sum of the groups; the assignment should be functorial; a single point should have the integers in dimension zero and nothing else.

The payoff is a clean separation between what homology is (a list of properties) and how you compute it (a specific construction). Singular, simplicial, cellular, and Čech chains all give the same answer because they all satisfy the same axioms.

Visual [Beginner]

A schematic showing the axioms arranged around a central box labelled $h_{n}$ — the homotopy axiom feeding in on one side, the long exact sequence threading through pairs in the middle, excision on a third side, additivity on a fourth. The dimension axiom is a small inset showing $h_{n} (pt)$ equal to the integers when the index is zero and zero otherwise.

The seven axioms in one picture: the central functor $h_{n}$ takes a pair of spaces to an abelian group, and the surrounding constraints fix what the functor is allowed to be.

Worked example [Beginner]

Use the axioms to compute the homology of a two-point space $X = {p, q}$ .

Step 1. The two-point space is the disjoint union of two single-point spaces. By the additivity axiom, the groups of $X$ are the direct sum of the groups of each point.

Step 2. The dimension axiom tells us a single point has the integers in dimension zero and nothing else.

Step 3. So in dimension zero the group attached to $X$ is the integers plus the integers, which is the integers squared.

Step 4. In every other dimension the group attached to $X$ is zero plus zero, which is zero.

What this tells us: without ever defining a chain group or computing a boundary, the axioms force the answer for the two-point space. Two of the seven axioms — additivity and dimension — were enough. The same kind of reasoning, with the long exact sequence of a pair and excision added in, computes the homology of every sphere, every torus, and every CW complex.

Check your understanding [Beginner]

Exercise (easy, multiple choice).

What does the uniqueness theorem of Eilenberg and Steenrod say?

A. Every space has a unique homology
B. Two homology theories satisfying all seven axioms with the same value on a point agree on every CW pair
C. There is exactly one way to compute the homology of a sphere
D. Singular homology is the only construction that works

Hint

The theorem compares two theories on a category of well-behaved spaces.

Answer

B. Two homology theories satisfying all seven axioms with the same value on a point agree on every CW pair.

This is the deep reason the singular, simplicial, and cellular constructions give the same groups: each one verifies the axioms, each one has the integers on a point in dimension zero, and the uniqueness theorem then forces the answers to agree.

Formal definition [Intermediate+]

Let $Top_{pair}$ be the category whose objects are pairs $(X, A)$ with $A \subseteq X$ a subspace and whose morphisms are continuous maps of pairs $f : (X, A) \to (Y, B)$ (continuous $f : X \to Y$ with $f (A) \subseteq B$ ). Identify a single space $X$ with the pair $(X, \emptyset)$ . Let $GrAb$ denote the category of $Z$ -graded abelian groups.

A homology theory on $Top_{pair}$ is a sequence of functors $h_{n} : Top_{pair} \to Ab$ for $n \in Z$ together with a natural transformation $\partial_{n} : h_{n} (X, A) \to h_{n - 1} (A, \emptyset)$ for each $n$ , called the connecting morphism, satisfying the following axioms.

(1) Homotopy. If $f, g : (X, A) \to (Y, B)$ are homotopic through maps of pairs, then $h_{n} (f) = h_{n} (g)$ for every $n$ .

(2) Long exact sequence (LES) of a pair. For each pair $(X, A)$ with inclusions $i : A ↪ X$ and $j : (X, \emptyset) ↪ (X, A)$ , the sequence

\dots \to h_{n} (A) i_{*} h_{n} (X) j_{*} h_{n} (X, A) \partial h_{n - 1} (A) \to \dots

is exact at every position.

(3) Excision. If $Z \subseteq A \subseteq X$ with $\overline{Z} \subseteq int A$ , the inclusion $(X ∖ Z, A ∖ Z) ↪ (X, A)$ induces an isomorphism $h_{n} (X ∖ Z, A ∖ Z) \sim h_{n} (X, A)$ for every $n$ .

(4) Additivity. For any disjoint union $X = ⨆_{α} X_{α}$ , the inclusions $X_{α} ↪ X$ induce an isomorphism $⨁_{α} h_{n} (X_{α}) \sim h_{n} (X)$ .

(5) Naturality. Each $h_{n}$ is functorial in maps of pairs and the connecting morphism $\partial_{n}$ is a natural transformation between the functors $(X, A) \mapsto h_{n} (X, A)$ and $(X, A) \mapsto h_{n - 1} (A)$ .

(6) Dimension. $h_{n} (pt) = 0$ for $n \neq = 0$ .

(7) Weak equivalence. A map $f : X \to Y$ inducing isomorphisms $π_{n} (f) : π_{n} (X, x_{0}) \sim π_{n} (Y, f (x_{0}))$ for all $n$ and all basepoints induces isomorphisms $h_{n} (f)$ for all $n$ .

The coefficient group of $h_{*}$ is $G = h_{0} (pt)$ . A homology theory satisfying axioms (1) through (6) is called an ordinary homology theory with coefficients $G$ . Dropping axiom (6) yields a generalised homology theory. Axiom (7) is sometimes folded into (1) for CW-restricted theories where it is automatic.

The dual notion is a cohomology theory: a contravariant family $h^{n} : Top_{pair}^{op} \to Ab$ with a connecting morphism $δ^{n} : h^{n} (A) \to h^{n + 1} (X, A)$ , satisfying the same axioms with arrows reversed. Singular cohomology is the prototype.

Counterexamples to common slips

The dimension axiom only constrains the value on a point; the additivity axiom plus the LES of a pair propagates this to determine the values on every CW complex, but the proof requires excision in an essential way.
The connecting morphism $\partial$ is part of the data of a homology theory, not a derived consequence; specifying $h_{n}$ alone without $\partial$ does not determine an Eilenberg-Steenrod theory.
The excision axiom requires the inclusion $\overline{Z} \subseteq int A$ (a strict-interior condition); weakening this to $Z \subseteq A$ fails for non-CW pairs and is the technical reason CW pairs are the natural setting for the uniqueness theorem.

Key theorem with proof [Intermediate+]

Theorem (Eilenberg-Steenrod uniqueness). Let $h_ $an d$ k_* $b e tw oor d ina r y h o m o l o g y t h eor i eso n t h ec a t e g or y o f C W p ai r ss a t i s f y in g a x i o m s (1) - (6), an d s u pp ose t h er e i s ani so m or p hi s m$ \varphi: h_0(\mathrm{pt}) \xrightarrow{\sim} k_0(\mathrm{pt}) $e x t e n d in g t o ani so m or p hi s m o f h o m o l o g y t h eor i eso n t h eo n e - p o in t s p a ce (so$ \varphi $i d e n t i f i es t h ecoe f f i c i e n t g r o u p s in t h eo n l y n o n - v ani s hin g d e g r ee) . T h e n t h er e i s ana t u r a l i so m or p hi s m o f h o m o l o g y t h eor i es$ \Phi: h_* \xrightarrow{\sim} k_* $o n t h ec a t e g or y o f C W p ai r s, e x t e n d in g$ \varphi$.*

Proof. The argument proceeds by induction on the cellular skeleton. Let $(X, A)$ be a CW pair. Filter $X$ by its skeleta $A = X^{(- 1)} \cup A \subseteq X^{(0)} \cup A \subseteq X^{(1)} \cup A \subseteq \dots$ , where $X^{(n)}$ denotes the $n$ -skeleton.

Step 1. Reduction to a single cell pair $(D^{n}, S^{n - 1})$ . Each skeletal pair $(X^{(n)} \cup A, X^{(n - 1)} \cup A)$ is a disjoint union (over the $n$ -cells of $X ∖ A$ ) of pairs of the form $(D^{n}, S^{n - 1})$ , glued along their boundaries. The additivity axiom (4) reduces both $h_{n}$ and $k_{n}$ on this skeletal pair to a direct sum over cells of the values on $(D^{n}, S^{n - 1})$ ; excision (3) then identifies the relative groups $h_{n} (X^{(n)} \cup A, X^{(n - 1)} \cup A)$ with $⨁_{α} h_{n} (D^{n}, S^{n - 1})$ where the sum is over the $n$ -cells.

Step 2. Compute $h_{n} (D^{n}, S^{n - 1})$ from the axioms. The disk $D^{n}$ is contractible, so by the homotopy axiom $h_{*} (D^{n}) = h_{*} (pt)$ . The LES of the pair $(D^{n}, S^{n - 1})$ then gives, for each $n \geq 1$ ,

\dots \to h_{n} (pt) \to h_{n} (D^{n}, S^{n - 1}) \partial h_{n - 1} (S^{n - 1}) \to h_{n - 1} (pt) \to \dots

Combined with the dimension axiom $h_{q} (pt) = 0$ for $q \neq = 0$ and the additivity computation $h_{*} (pt) = G$ concentrated in degree zero, the connecting map $\partial : h_{n} (D^{n}, S^{n - 1}) \to \tilde{h}_{n - 1} (S^{n - 1})$ is an isomorphism for $n \geq 1$ . By induction on $n$ using $S^{n} = D^{n} \cup_{S^{n - 1}} D^{n}$ and Mayer-Vietoris (which follows from the axioms; see Advanced results), one finds $\tilde{h}_{n} (S^{n}) = G$ and $\tilde{h}_{q} (S^{n}) = 0$ for $q \neq = n$ . Hence $h_{n} (D^{n}, S^{n - 1}) = G$ in degree $n$ and zero elsewhere.

Step 3. Define $Φ$ cell by cell. The same axiom-based computation applies to $k_{*}$ , yielding $k_{n} (D^{n}, S^{n - 1}) = G^{'}$ in degree $n$ where $G^{'} = k_{0} (pt)$ . The isomorphism $φ : G \sim G^{'}$ propagates through the calculation in Step 2 to a canonical isomorphism $Φ_{n}^{cell} : h_{n} (D^{n}, S^{n - 1}) \sim k_{n} (D^{n}, S^{n - 1})$ for each cell pair, natural in cellular maps.

Step 4. Stitch the skeletal isomorphisms together. The LES of the skeletal pair $(X^{(n)}, X^{(n - 1)})$ for $h_{*}$ and for $k_{*}$ , together with the cell-level isomorphisms from Step 3, gives a commutative ladder of long exact sequences. The five lemma applied to the ladder propagates the isomorphism inductively up the skeleta. Naturality in maps of CW pairs follows because each step uses only the LES, additivity, excision, and homotopy invariance — all of which are natural by axiom (5).

Passing to the colimit over the skeleta — finite-dimensional CW pairs are exhausted in finitely many steps; infinite-dimensional ones use the additivity axiom on the colimit topology — yields the natural isomorphism $Φ : h_{*} (X, A) \sim k_{*} (X, A)$ . $□$

Bridge. This builds toward the entire structural picture of homology theory: the axioms reduce every computation on a CW pair to two pieces of data — the coefficient group and the cellular structure — and that reduction is exactly what makes singular, simplicial, cellular, and Čech homologies all give the same answer on the spaces where they are all defined. The same axiom-driven inductive cellular argument appears again in 03.12.05 (Eilenberg-MacLane spaces), where one builds $K (G, n)$ cell by cell to realise $G$ in a single homotopy degree, and the cellular long exact sequences used in Step 4 above are the same sequences that organise the cohomology operations on $K (G, n)$ . Putting these together, the uniqueness theorem becomes the conceptual underpinning of every Eilenberg-MacLane construction: the axioms force any homology theory with coefficient group $G$ in dimension zero to compute the same groups, so the construction of a space whose homology is $G$ in a single degree is simultaneously a construction in any of the equivalent theories. The bridge is the recognition that the axioms, the cellular computation, and the representing space are three views of one structure, and the same machinery appears again in 03.13.02 (the Leray-Serre spectral sequence) where the axioms govern the $E_{2}$ page and the cellular filtration is the one running the spectral sequence.

Exercises [Intermediate+]

Exercise 3 (medium, short-answer).

Derive the Mayer-Vietoris sequence from the Eilenberg-Steenrod axioms (in the form: for an open cover $X = U \cup V$ of a CW complex with $U, V$ subcomplexes, exhibit the long exact sequence relating $h_{*} (U \cap V)$ , $h_{*} (U) \oplus h_{*} (V)$ , and $h_{*} (X)$ ).

Hint

Combine the LES of the pair $(X, V)$ with the excision isomorphism $h_{*} (X, V) ≅ h_{*} (U, U \cap V)$ .

Answer

By the excision axiom applied to $Z = X ∖ U$ inside $A = V$ , the inclusion $(U, U \cap V) ↪ (X, V)$ induces an isomorphism $h_{n} (U, U \cap V) ≅ h_{n} (X, V)$ . The LES of the pair $(X, V)$ then reads $\dots \to h_{n} (V) \to h_{n} (X) \to h_{n} (U, U \cap V) \to h_{n - 1} (V) \to \dots$ . Combining with the LES of the pair $(U, U \cap V)$ via the connecting-map identification and rearranging yields $\dots \to h_{n} (U \cap V) \to h_{n} (U) \oplus h_{n} (V) \to h_{n} (X) \to h_{n - 1} (U \cap V) \to \dots$ . Rubric: full credit for the excision step plus the LES manipulation.

Exercise 5 (medium, symbolic).

Show that for any ordinary homology theory $h_{*}$ with coefficient group $G$ , the relative group $h_{n} (X, x_{0})$ for a pointed space agrees with the reduced homology $\tilde{h}_{n} (X) = ker (h_{n} (X) \to h_{n} (pt))$ .

Hint

Apply the LES of the pair $(X, x_{0})$ .

Answer

The LES of $(X, x_{0})$ reads $\dots \to h_{n} (x_{0}) \to h_{n} (X) \to h_{n} (X, x_{0}) \to h_{n - 1} (x_{0}) \to \dots$ . By the dimension axiom $h_{n} (x_{0}) = 0$ for $n \geq 1$ , so $h_{n} (X) \to h_{n} (X, x_{0})$ is an isomorphism for $n \geq 1$ . In dimension zero, exactness at $h_{0} (X, x_{0})$ gives $h_{0} (X, x_{0}) = coker (h_{0} (x_{0}) \to h_{0} (X))$ , which equals $h_{0} (X) / G$ . The reduced group is the kernel of the augmentation $h_{0} (X) \to G$ , equal to the cokernel of $G \to h_{0} (X)$ when the augmentation splits (always for non-empty $X$ ). Rubric: full credit for the LES reduction.

Exercise 7 (hard, short-answer).

State precisely how the additivity axiom enters the proof of the uniqueness theorem on infinite-dimensional CW complexes, and show that without additivity the uniqueness theorem fails.

Hint

Consider what happens with an infinite wedge of spheres $⋁_{i = 1}^{\infty} S^{1}$ .

Answer

Additivity is what allows the cell-by-cell computation in the inductive step to commute with the colimit topology of an infinite-dimensional CW complex. Concretely, $h_{n} (X) = h_{n} (colim X^{(n)})$ must equal $colim h_{n} (X^{(n)})$ (filtered colimit on the right), and additivity is the axiom that turns the colimit into a direct sum. Without additivity, the wedge $W = ⋁_{i = 1}^{\infty} S^{1}$ admits two distinct theories: the standard singular theory has $h_{1} (W) = ⨁_{i = 1}^{\infty} Z$ , while a hypothetical theory using $\overset{ˇ}{C}$ ech-style products would give $\prod_{i = 1}^{\infty} Z$ , and both verify axioms (1)-(6) on finite CW complexes. Rubric: full credit for identifying the colimit step plus the wedge counterexample.

Exercise 8 (hard, short-answer).

Show that any ordinary cohomology theory $h^{*}$ with coefficient group $G$ on the homotopy category of CW complexes is naturally isomorphic to $[X, K (G, *)]$ , the homotopy classes of maps into Eilenberg-MacLane spaces.

Hint

This is Brown representability: combine the uniqueness theorem with the fact that $K (G, n)$ represents the functor $X \mapsto H^{n} (X; G)$ in singular cohomology.

Answer

Singular cohomology with coefficients $G$ satisfies $H^{n} (X; G) = [X, K (G, n)]_{*}$ for path-connected pointed CW complexes (Eilenberg-MacLane representability). The Eilenberg-Steenrod uniqueness theorem applies to cohomology theories on CW pairs, giving a natural isomorphism $h^{*} (X) ≅ H^{*} (X; G)$ when both share coefficient group $G$ . Composing yields $h^{n} (X) ≅ [X, K (G, n)]_{*}$ naturally. Rubric: full credit for combining uniqueness with Eilenberg-MacLane representability; partial credit for either piece alone.

Lean formalization [Intermediate+]

Mathlib does not yet ship a packaged Eilenberg-Steenrod-axioms structure. The intended definition would package the seven axioms as fields of a typeclass:

[object Promise]

The proof gap is real and substantive on every line. The structure itself depends on a careful definition of TopPair with the right interaction between subspace topology and excision; the singular instance requires the full prism/chain-homotopy proof of homotopy invariance plus the small-simplices argument for excision; the uniqueness theorem requires the cellular skeleton induction and the five lemma at scale. None of these are formalised in Mathlib as of this writing.

Advanced results [Master]

Theorem (Mayer-Vietoris from the axioms). Let $h_ $b e ah o m o l o g y t h eor y s a t i s f y in g t h e E i l e nb er g - S t ee n r o d a x i o m so na c a t e g or y o f p ai r sco n t ainin g$ X = U \cup V $w i t h$ U, V $o p e n s u b s p a ceso f a C W co m pl e x$ X$. There is a natural long exact sequence*

\dots \to h_{n} (U \cap V) \to h_{n} (U) \oplus h_{n} (V) \to h_{n} (X) \to h_{n - 1} (U \cap V) \to \dots

The argument combines the LES of the pair $(X, V)$ with excision $(X, V) ≅ (U, U \cap V)$ as in Exercise 3 above. Mayer-Vietoris is a consequence of the axioms, not an independent axiom — a remark exploited in the original Eilenberg-Steenrod monograph to streamline computations.

Theorem (CW computation). For any ordinary homology theory $h_ $w i t h coe f f i c i e n t g r o u p$ G $an d an y C W co m pl e x$ X $, t h ece l l u l a r c hain co m pl e x$ C^{\mathrm{cell}}n(X) = h_n(X^{(n)}, X^{(n-1)}) $t o g e t h er w i t h t h e b o u n d a r y d e f in e d b y t h eco nn ec t in g m or p hi s m o f t h e t r i pl e$ (X^{(n)}, X^{(n-1)}, X^{(n-2)}) $co m p u t es$ h(X)$.

The cellular chain complex is determined entirely by the axioms — no construction-specific input is needed. This is the deep reason cellular homology of a CW complex agrees with singular, simplicial, and Čech homologies, and it is the form of the uniqueness theorem most useful in computations.

Theorem (Eilenberg-Steenrod 1952, Theorem III.10.1). On the category of CW pairs, two homology theories $h_ $an d$ k_* $s a t i s f y in g a x i o m s (1) - (6) an d a d mi tt in g ani so m or p hi s m o n t h ecoe f f i c i e n t g r o u p$ h_0(\mathrm{pt}) \cong k_0(\mathrm{pt})$ are naturally isomorphic. Equivalently, ordinary homology theory on CW pairs is determined up to natural isomorphism by its coefficient group.*

Theorem (generalised cohomology theories, Brown representability). Dropping the dimension axiom (6) yields a generalised (co)homology theory. On the homotopy category of pointed CW complexes, every contravariant functor $h^n: \mathrm{Ho}(\mathbf{CW}_)^{\mathrm{op}} \to \mathbf{Set}* $s a t i s f y in g t h e w e d g e a x i o m (a dd i t i v i t y f or w e d g es) an d t h e M a y er - V i e t or i s a x i o mi sr e p r ese n t ab l e : t h er ee x i s t s a p o in t e d C W co m pl e x$ E_n $w i t h$ h^n(X) = [X, E_n] $, na t u r a l in$ X$. (Brown 1962.) The sequence of representing spaces $E_{*}$ assembles into a spectrum, and every generalised cohomology theory arises from a spectrum in this way.

Theorem (singular = simplicial = cellular = Čech on CW pairs). On the category of CW pairs (or more generally on triangulable spaces) the singular, simplicial, cellular, and Čech homology theories are naturally isomorphic.

The argument is uniform: each construction satisfies the Eilenberg-Steenrod axioms with coefficient group $Z$ , so the uniqueness theorem applies. The pre-axiomatic proofs of agreement (e.g., Lefschetz's direct chain-level comparison between simplicial and singular) are subsumed into a single short corollary.

Examples of generalised theories.

$K$ -theory $K^{*} (X)$ , built from isomorphism classes of complex vector bundles on $X$ , satisfies all the axioms except dimension; $K^{0} (pt) = Z$ and $K^{- 2} (pt) = Z$ (Bott periodicity), so dimension fails. Represented by the spectrum $K U$ .
Bordism $Ω_{*}^{O} (X)$ and oriented bordism $Ω_{*}^{SO} (X)$ , classes of singular manifolds in $X$ modulo bordism. $Ω_{n}^{O} (pt)$ is non-zero in many degrees (Thom 1954), so dimension fails. Represented by Thom spectra.
Stable homotopy $π_{*}^{s} (X) = [X, S]_{*}$ , where $S$ is the sphere spectrum. The sphere spectrum's coefficient ring $π_{*}^{s} (S^{0})$ has been computed only in low degrees and is the subject of much current research.

Synthesis. The Eilenberg-Steenrod axioms are the structural fact one returns to whenever one wants to know what a homology theory is: a list of properties stable under all the operations of homotopy theory (suspension, products, fibrations, cofibrations) and complete enough to pin down the answer on CW pairs once the coefficient group is specified. The central structural insight is the separation between the content of the theory (the seven axioms) and any specific construction of a representative; this separation is what makes the singular, simplicial, cellular, and Čech constructions all give the same groups, and it is what makes the modern spectrum-level identification of homology theories with $H Z$ -modules possible. The bridge is the recognition that the dimension axiom is the discriminant between ordinary and generalised theories: imposing it pins down singular homology up to the choice of coefficient ring; dropping it opens the door to $K$ -theory, bordism, and stable homotopy.

This builds toward the spectrum-level perspective of 03.12.04 (the spectrum unit), where the Eilenberg-MacLane spectrum $H Z$ realises ordinary homology as a generalised theory associated to a particular spectrum, and the same axiomatic machinery applied to other spectra (the $K$ -theory spectrum, Thom spectra) yields the corresponding generalised theories. The same axiomatic pattern appears again in 03.12.05 (Eilenberg-MacLane spaces), where the construction of $K (G, n)$ depends only on the axioms and the coefficient group, and putting these together identifies ordinary cohomology with homotopy classes of maps into Eilenberg-MacLane spaces — Brown representability. The downstream consequence is 03.13.02 (the Leray-Serre spectral sequence), where the axioms determine the $E_{2}$ page of the spectral sequence of a fibration entirely from the homology of the base and fibre, and the spectral sequence then converges to the homology of the total space; the uniqueness theorem is what guarantees the $E_{2}$ identification is independent of the homology construction one uses.

Full proof set [Master]

Theorem (Mayer-Vietoris from the axioms), proof. Apply the excision axiom (3) to $Z = X ∖ U \subseteq A = V$ . The hypothesis $\overline{Z} \subseteq int A$ is equivalent to $X = U \cup V$ with $U, V$ open. The conclusion is $h_{n} (U, U \cap V) ≅ h_{n} (X, V)$ naturally in maps of triples.

The LES of the pair $(X, V)$ reads

\dots \to h_{n} (V) \to h_{n} (X) \to h_{n} (X, V) \to h_{n - 1} (V) \to \dots

Substituting the excision identification gives

\dots \to h_{n} (V) \to h_{n} (X) \to h_{n} (U, U \cap V) \to h_{n - 1} (V) \to \dots

The LES of the pair $(U, U \cap V)$ reads

\dots \to h_{n} (U \cap V) \to h_{n} (U) \to h_{n} (U, U \cap V) \to h_{n - 1} (U \cap V) \to \dots

A small categorical computation (the Barratt-Whitehead lemma, equivalently a diagram chase on the two LES's together with naturality of the connecting morphism) splices these into the Mayer-Vietoris sequence

\dots \to h_{n} (U \cap V) \to h_{n} (U) \oplus h_{n} (V) \to h_{n} (X) \to h_{n - 1} (U \cap V) \to \dots

with maps $(i_{*}, j_{*})$ on the inclusion piece and $k_{*} - ℓ_{*}$ on the difference piece, where $i : U \cap V ↪ U$ , $j : U \cap V ↪ V$ , $k : U ↪ X$ , $ℓ : V ↪ X$ . $□$

Theorem (CW computation), proof. Filter $X$ by skeleta $X^{(- 1)} = \emptyset \subseteq X^{(0)} \subseteq X^{(1)} \subseteq \dots$ . The triple $(X^{(n)}, X^{(n - 1)}, X^{(n - 2)})$ gives a long exact sequence

\dots \to h_{n} (X^{(n - 1)}, X^{(n - 2)}) \to h_{n} (X^{(n)}, X^{(n - 2)}) \to h_{n} (X^{(n)}, X^{(n - 1)}) \partial h_{n - 1} (X^{(n - 1)}, X^{(n - 2)}) \to \dots

Define the cellular chain group $C_{n}^{cell} (X) = h_{n} (X^{(n)}, X^{(n - 1)})$ , which equals $⨁_{α} h_{n} (D^{n}, S^{n - 1}) = ⨁_{α} G$ (a direct sum over the $n$ -cells, by additivity plus the Step 2 calculation in the proof of the uniqueness theorem). Define the cellular boundary $d_{n} : C_{n}^{cell} (X) \to C_{n - 1}^{cell} (X)$ as the composite

h_{n} (X^{(n)}, X^{(n - 1)}) \partial h_{n - 1} (X^{(n - 1)}) \to h_{n - 1} (X^{(n - 1)}, X^{(n - 2)})

The relation $d_{n - 1} \circ d_{n} = 0$ follows from $\partial \circ$ inclusion $\circ \partial = 0$ (the LES gives $\partial$ -then-inclusion-then- $\partial$ is the connecting map of the triple composed with itself, which factors through $h_{n - 2} (X^{(n - 1)}, X^{(n - 1)}) = 0$ ).

A diagram chase using exactness of the LES of pairs $(X^{(n)}, X^{(n - 1)})$ at every $n$ identifies the homology $H_{n} (C^{cell} (X))$ with $h_{n} (X)$ , naturally in cellular maps. The required passage from finite-dimensional to infinite-dimensional CW complexes uses additivity to commute homology with the colimit topology. $□$

Theorem (Eilenberg-Steenrod uniqueness, full proof). Combine the Step 1-4 outline in Key theorem with proof with the cellular computation above. The resulting natural transformation $Φ : h_{*} \to k_{*}$ is constructed degree by degree at the cellular level, propagated to all CW complexes by the five lemma applied to the LES ladder, and verified natural by chasing the connecting morphism axiom. The full argument occupies §III.10 of Eilenberg-Steenrod 1952; the modern compact version is in May Ch. 13. $□$

Stated without proof — see Brown 1962 ^[pending]. Brown representability: every contravariant functor $h : Ho (CW_{*})^{op} \to Set_{*}$ satisfying the wedge and Mayer-Vietoris axioms is representable by a pointed CW complex $E$ . The dual covariant statement on countable CW complexes is due to Heller and Adams.

Stated without proof — see Switzer 1975 ^[pending], May 1999 ^[pending]. The bijection between generalised cohomology theories on $Ho (CW_{*})$ and spectra in the modern sense (sequential, $Ω$ -spectra, or symmetric spectra), with morphisms of theories corresponding to maps of spectra in the stable homotopy category.

Connections [Master]

Eilenberg-MacLane spaces 03.12.05. The Eilenberg-Steenrod axioms identify ordinary cohomology with homotopy classes of maps into Eilenberg-MacLane spaces. Concretely, the uniqueness theorem combined with the singular-cohomology computation $H^{n} (K (G, n); G) = G$ produces a natural isomorphism $h^{n} (X) ≅ [X, K (G, n)]_{*}$ for any ordinary cohomology theory with coefficient group $G$ . The Eilenberg-MacLane construction itself uses the axioms in reverse: the cell-by-cell construction of $K (G, n)$ is verified to produce the right homotopy groups via the cellular homology computation, which is itself an instance of the axiomatic CW computation theorem.
Spectrum 03.12.04. A generalised (co)homology theory on the category of CW complexes is the same data as a spectrum, by Brown representability. The Eilenberg-MacLane spectrum $H Z$ represents singular homology; the $K$ -theory spectrum $K U$ represents complex topological $K$ -theory; the Thom spectrum $M O$ represents unoriented bordism. The Eilenberg-Steenrod axioms are exactly the conditions a spectrum must satisfy; dropping the dimension axiom corresponds to dropping the Eilenberg-MacLane condition $π_{n} (E) = 0$ for $n \neq = 0$ on the representing spectrum.
Leray-Serre spectral sequence 03.13.02. The Leray-Serre spectral sequence for a fibration $F \to E \to B$ has $E_{2}^{p, q} = H^{p} (B; H^{q} (F))$ converging to $H^{p + q} (E)$ , where $H^{*}$ is any ordinary cohomology theory satisfying the axioms. The $E_{2}$ page is determined by the axioms applied to the cellular filtration of $B$ , and the convergence statement is independent of which ordinary cohomology theory one uses — a structural consequence of the uniqueness theorem.
Singular homology 03.12.11. Singular homology is the prototype ordinary homology theory; the seven axioms are exactly the properties of the singular chain complex viewed externally. The chain-level construction is one specific recipe; the axioms describe what any recipe must produce.
Excision theorem 03.12.14. Excision is axiom (3); it is the single axiom that requires substantive geometric input on the singular side (the small-simplices / barycentric-subdivision argument), and it is the axiom that makes the cellular reduction work. The uniqueness theorem is impossible without excision.
Cellular homology 03.12.13. Cellular homology is the computational consequence of the axioms applied to a CW complex: the cellular chain complex $C_{n}^{cell} (X) = h_{n} (X^{(n)}, X^{(n - 1)})$ computes $h_{*} (X)$ for any ordinary homology theory, and the agreement of cellular homology with singular homology is then an instance of the uniqueness theorem.
CW complex 03.12.10. The natural setting for the uniqueness theorem is the category of CW pairs. CW complexes are the spaces where the cellular reduction works cleanly; on more general spaces the uniqueness theorem fails (different theories can disagree on, e.g., the Hawaiian earring or solenoids).

Historical & philosophical context [Master]

The axioms were introduced by Samuel Eilenberg and Norman Steenrod in Axiomatic approach to homology theory (Proc. Nat. Acad. Sci. 31 (1945) 117-120) ^[pending], a four-page note announcing the project. The full development came in their 1952 monograph Foundations of Algebraic Topology (Princeton University Press) ^[pending], which laid out the seven axioms, established the uniqueness theorem on CW pairs (Theorem III.10.1), and verified that singular, simplicial, Čech, and Vietoris homology each satisfy the axioms with $Z$ coefficients. Before 1945 the agreement between these constructions had been established case by case, often with awkward technical hypotheses on the spaces involved. The axiomatic treatment replaced a patchwork of comparison theorems with a single clean uniqueness statement.

The shift in perspective matched a broader move in mid-century mathematics from constructive to axiomatic descriptions of mathematical objects. Bourbaki's Algèbre (1942-) treated algebraic structures axiomatically; Mac Lane and Eilenberg's 1945 paper on natural transformations introduced the language that made the Eilenberg-Steenrod axioms expressible as a statement about functors between categories. The 1952 monograph is one of the early major applications of categorical thinking to a substantial body of pre-existing mathematics: every axiom is a statement about a functor or a natural transformation, and the uniqueness theorem is a statement about the category of homology theories.

The generalised theories came later. In 1958-1962 Atiyah and Hirzebruch developed topological $K$ -theory as an axiomatic theory in the Eilenberg-Steenrod style minus the dimension axiom; their 1961 paper Vector bundles and homogeneous spaces established that complex vector bundles satisfy the axioms (1)-(5) and (7) but fail (6). Brown's 1962 paper Cohomology theories (Ann. Math. 75 (1962) 467-484) ^[pending] proved that any contravariant functor on pointed CW complexes satisfying the wedge and Mayer-Vietoris axioms is representable, identifying generalised cohomology theories with sequences of representing spaces — what would soon be called spectra (Lima 1959, Whitehead 1962, Boardman 1965). By the late 1960s the spectrum-level perspective had absorbed the Eilenberg-Steenrod axiomatic framework: an ordinary homology theory is a module over $H Z$ , and a generalised homology theory is a module over its representing spectrum. The 1952 axiomatic statement remains the foundation of the modern stable homotopy category.

Bibliography [Master]

[object Promise]

Prerequisites

03.12.11
03.12.14

Used in

03.12.05
03.13.02

Tier anchors

beginner: Hatcher *Algebraic Topology* §2.3 informal opening; the axioms as a wish-list for what 'homology' should mean
intermediate: Hatcher §2.3; May *A Concise Course in Algebraic Topology* Ch. 13
master: Eilenberg-Steenrod 1952 — *Foundations of Algebraic Topology* Ch. I-III (originator monograph); Eilenberg-Steenrod 1945 *Axiomatic approach to homology theory* (Proc. Nat. Acad. Sci. 31); Hatcher §2.3 + §4.E + §4.F (Brown representability); tom Dieck *Algebraic Topology* §17

References

TODO_REF
Eilenberg-Steenrod — Foundations of Algebraic Topology · Princeton 1952, Ch. I-III the seven axioms and the uniqueness theorem on CW pairs
TODO_REF
Eilenberg-Steenrod — Axiomatic approach to homology theory · Proc. Nat. Acad. Sci. USA 31 (1945) 117-120, originator note
TODO_REF
Hatcher — Algebraic Topology · §2.3 axioms; §4.E generalised cohomology; §4.F Brown representability
TODO_REF
May — A Concise Course in Algebraic Topology · Ch. 13 axiomatic treatment, derivation of consequences from the axioms
TODO_REF
tom Dieck — Algebraic Topology · §17 axiomatic homology, generalised cohomology theories, spectra
TODO_REF
Brown — Cohomology theories · Ann. Math. 75 (1962) 467-484, representability of cohomology functors

Reviewer

TBD

Estimated time

beginner: 15m
intermediate: 35m
master: 70m