03.12.34 · modern-geometry / homotopy

Acyclic models and the Eilenberg-Zilber theorem

shipped3 tiersLean: none

Anchor (Master): Eilenberg-MacLane 1953 *Acyclic models* (Amer. J. Math. 75, 189-199) (originator); Eilenberg-Zilber 1953 *On products of complexes* (Amer. J. Math. 75, 200-204); May 1967 *Simplicial Objects in Algebraic Topology* §28-§29; Mac Lane 1963 *Homology* (Springer) §VIII.8 (acyclic models, comparison theorem); Weibel 1994 *An Introduction to Homological Algebra* (Cambridge) §8.5-§8.6

Intuition Beginner

Suppose you have two ways of measuring the holes in a product space, one built directly from the product and one built by combining measurements of each factor. You would like to know they agree. Checking this by hand, simplex by simplex, is a forbidding bookkeeping exercise. The method of acyclic models is a labour-saving device that does the bookkeeping once, abstractly, and then hands you the agreement for free.

The trick rests on two properties of the building blocks. A construction is free on models when everything it produces is assembled from a small stock of standard pieces, like building any Lego structure from a fixed bin of bricks. It is acyclic on models when each standard piece, examined on its own, has no holes. When both hold, you can always build a natural comparison map between two such constructions, and any two comparison maps you build are equivalent. The slogan: enough free, hole-free building blocks force the maps to exist and to be essentially unique.

The headline payoff is the Eilenberg-Zilber theorem. It says the chains on a product of two spaces and the combined chains of the two factors carry the same homological information, linked by two explicit maps. One map, named after Alexander and Whitney, cuts a product simplex into a front face and a back face. The other, the shuffle map, interleaves a simplex from each factor into the product. Composed one way they cancel exactly; composed the other way they cancel up to an adjustment that changes nothing in homology.

Visual Beginner

Picture a filled square, the product of two line segments. A path that walks along the bottom edge and then up the right edge is the front-then-back decomposition that the Alexander-Whitney map records: it reads a two-dimensional cell as a horizontal step followed by a vertical step. The shuffle map runs the other direction. Given one horizontal segment and one vertical segment, it fills in the square by listing every staircase path from the bottom-left corner to the top-right corner, each path a triangle, the whole collection a triangulation of the square.

A second panel shows the two comparison maps as arrows between two columns. The left column is the chains of the product; the right column is the combined chains of the factors. The Alexander-Whitney arrow points right, the shuffle arrow points left. A small loop on the right column marks that going right-then-left returns exactly where you started, while a wavy loop on the left marks that going left-then-right returns to a point joined to the start by a homotopy.

Worked example Beginner

Take the simplest interesting product: a single edge crossed with a single edge, giving a square. Label the first edge with endpoints $0$ and $1$ , and the second edge likewise. The square has four corners, and we want to see the shuffle map fill it with triangles.

Step 1. The shuffle map takes the horizontal edge and the vertical edge and produces a two-dimensional chain. It lists the staircase paths from corner $(0, 0)$ to corner $(1, 1)$ . There are exactly two such monotone staircases: go right then up, or go up then right.

Step 2. Each staircase is a triangle. The right-then-up path gives the lower triangle with corners $(0, 0)$ , $(1, 0)$ , $(1, 1)$ . The up-then-right path gives the upper triangle with corners $(0, 0)$ , $(0, 1)$ , $(1, 1)$ . Together the two triangles tile the square.

Step 3. The shuffle map attaches a sign to each staircase, recording how many times the two directions cross. Here the lower triangle gets a plus and the upper triangle gets a minus, so the shuffle of the two edges is the lower triangle minus the upper triangle.

Step 4. Now run the Alexander-Whitney map back on either triangle. It reads the front face in the first coordinate and the back face in the second. Applied to the lower triangle it returns the bottom edge in the first factor times the right edge in the second factor, which is the original pair of edges.

What this tells us: the shuffle map turns a pair of edges into a tiled square, and the Alexander-Whitney map reads that square back to the pair of edges. Going one way and then back recovers the input, which is the chain-level statement that the two descriptions of the product carry the same content.

Check your understanding Beginner

Formal definition Intermediate+

Throughout, $K$ is a category and $R$ a commutative ring; chain complexes are non-negatively graded complexes of $R$ -modules. A functor $F : K \to Ch_{\geq 0} (R)$ assigns to each object $X$ a chain complex $F (X)$ . Fix a set $M$ of objects of $K$ , the models.

Definition (free on models). A functor $F : K \to Ch_{\geq 0} (R)$ is free on the models $M$ if for each degree $n$ there is an indexing set $J_{n}$ , a family of models $M_{j} \in M$ for $j \in J_{n}$ , and elements $m_{j} \in F_{n} (M_{j})$ (the basis elements) such that $F_{n}$ is the free $R$ -module functor on the representable data: for every object $X$ , the module $F_{n} (X)$ is free on the set ${F_{n} (φ) (m_{j}) : j \in J_{n}, φ \in Hom_{K} (M_{j}, X)}$ . Equivalently, $F_{n} ≅ ⨁_{j \in J_{n}} R [Hom_{K} (M_{j}, -)]$ as functors to $R$ -modules.

Definition (acyclic on models). A functor $F : K \to Ch_{\geq 0} (R)$ is acyclic on the models $M$ in degrees $\geq 1$ if $H_{n} (F (M)) = 0$ for every model $M \in M$ and every $n \geq 1$ . The complex $F (M)$ is then a resolution of $H_{0} (F (M))$ for each model.

Definition (augmented setting). Often one works with augmented functors $ε : F_{0} \to G_{0}$ over a coefficient functor $G$ (commonly $G (X) = R$ , the constant functor), and asks $F (M)$ to be acyclic relative to the augmentation: $H_{0} (F (M)) ≅ G (M)$ and $H_{n} (F (M)) = 0$ for $n \geq 1$ . The basepoint case $G = R$ is the one used for singular chains, where $H_{0} (C_{*} (Δ^{n})) = R$ and the higher homology of a simplex vanishes.

The two motivating functors are both built on the singular-simplex category. Let $K = Top \times Top$ and $M = {(Δ^{p}, Δ^{q})}_{p, q \geq 0}$ . The functor $(X, Y) \mapsto C_{*} (X \times Y)$ is the singular chains on the product 03.12.11; the functor $(X, Y) \mapsto C_{*} (X) \otimes C_{*} (Y)$ is the tensor product of the singular complexes, with the Koszul-sign differential $\partial (a \otimes b) = \partial a \otimes b + (- 1)^{∣ a ∣} a \otimes \partial b$ . Both are free on the model pairs and acyclic on them: a product of simplices is contractible, so both complexes have the homology of a point on each model.

Counterexamples to common slips

Freeness is a condition on each $F_{n}$ as a functor, not on the individual modules $F_{n} (X)$ . A functor whose values are free modules but whose functorial structure is not induced by a representable basis is not free on models. The singular chain functor qualifies because its basis is the set of singular simplices, which are exactly the maps out of the model $Δ^{n}$ .
Acyclicity is required only in degrees $\geq 1$ , and only on the models, not on all objects. The functor $C_{*} (X)$ is very far from acyclic on a general space $X$ ; that is the whole point of homology. The hypothesis tames only the models.
The comparison theorem produces a chain map covering a prescribed map in degree zero. Omitting the degree-zero compatibility (the augmentation) leaves the induction with no base case, and the conclusion fails.

Key theorem with proof Intermediate+

Theorem (acyclic-models comparison theorem; Eilenberg-MacLane 1953). Let $F, G : K \to Ch_{\geq 0} (R)$ be functors and $M$ a set of models. Suppose $F$ is free on $M$ and $G$ is acyclic on $M$ in degrees $\geq 1$ . Let $ϕ_{0} : H_{0} (F) \to H_{0} (G)$ be a natural transformation of the degree-zero homology functors. Then:

(existence) there is a natural chain map $Φ : F \to G$ inducing $ϕ_{0}$ on $H_{0}$ ;

(uniqueness) any two natural chain maps $Φ, Ψ : F \to G$ inducing the same map on $H_{0}$ are naturally chain homotopic.

Proof. Choose, for each degree $n$ , an indexing set $J_{n}$ , models $M_{j}$ , and basis elements $m_{j} \in F_{n} (M_{j})$ witnessing freeness, so that defining a natural transformation out of $F_{n}$ amounts to choosing the images of the $m_{j}$ in a natural target, with no further constraint.

Existence, by induction on degree. In degree $0$ we must define $Φ_{0} : F_{0} \to G_{0}$ naturally, lifting $ϕ_{0}$ on homology. For each basis element $m_{j} \in F_{0} (M_{j})$ , the class $ϕ_{0} ([m_{j}]) \in H_{0} (G (M_{j}))$ has a representative cycle $g_{j} \in G_{0} (M_{j})$ (every degree-zero chain is a cycle). Set $Φ_{0} (m_{j}) := g_{j}$ and extend naturally: for $φ : M_{j} \to X$ , put $Φ_{0} (F_{0} (φ) m_{j}) := G_{0} (φ) g_{j}$ . Freeness makes this a well-defined natural map, and by construction it induces $ϕ_{0}$ .

Assume $Φ$ defined naturally as a chain map in degrees $< n$ , compatible with the differentials there. To define $Φ_{n}$ it suffices, by freeness, to choose $Φ_{n} (m_{j}) \in G_{n} (M_{j})$ for each basis element $m_{j} \in F_{n} (M_{j})$ , subject to $\partial^{G} Φ_{n} (m_{j}) = Φ_{n - 1} (\partial^{F} m_{j})$ . The right-hand side $c_{j} := Φ_{n - 1} (\partial^{F} m_{j}) \in G_{n - 1} (M_{j})$ is a cycle: $\partial^{G} c_{j} = Φ_{n - 2} (\partial^{F} \partial^{F} m_{j}) = 0$ using the induction hypothesis that $Φ$ commutes with $\partial$ in lower degrees. Since $n \geq 1$ and $G$ is acyclic on $M_{j}$ in degrees $\geq 1$ , the cycle $c_{j} \in G_{n - 1} (M_{j})$ is a boundary; choose $Φ_{n} (m_{j}) \in G_{n} (M_{j})$ with $\partial^{G} Φ_{n} (m_{j}) = c_{j}$ . Extend naturally over all $φ : M_{j} \to X$ . This defines $Φ_{n}$ naturally and compatibly with the differential, completing the induction.

Uniqueness, by a parallel induction building a chain homotopy. Let $D := Φ - Ψ : F \to G$ ; it is a natural chain map inducing $0$ on $H_{0}$ . We construct a natural chain homotopy $s_{n} : F_{n} \to G_{n + 1}$ with $\partial^{G} s + s \partial^{F} = D$ . In degree $0$ : for each basis element $m_{j} \in F_{0} (M_{j})$ , $D_{0} (m_{j})$ is a degree-zero cycle whose class is $ϕ_{0} ([m_{j}]) - ϕ_{0} ([m_{j}]) = 0$ in $H_{0} (G (M_{j}))$ , hence a boundary; choose $s_{0} (m_{j}) \in G_{1} (M_{j})$ with $\partial^{G} s_{0} (m_{j}) = D_{0} (m_{j})$ , and extend naturally. Suppose $s$ defined naturally in degrees $< n$ with $\partial^{G} s + s \partial^{F} = D$ there. For a basis element $m_{j} \in F_{n} (M_{j})$ set $c_{j} := D_{n} (m_{j}) - s_{n - 1} (\partial^{F} m_{j}) \in G_{n} (M_{j})$ . Then $$ \partial^G c_j = \partial^G D_n(m_j) - \partial^G s_{n-1}(\partial^F m_j) = D_{n-1}(\partial^F m_j) - \big(D_{n-1}(\partial^F m_j) - s_{n-2}(\partial^F \partial^F m_j)\big) = 0, $$ using that $D$ is a chain map and the degree- $(n - 1)$ homotopy identity. So $c_{j}$ is a cycle in $G_{n} (M_{j})$ with $n \geq 1$ ; by acyclicity it is a boundary, and we choose $s_{n} (m_{j}) \in G_{n + 1} (M_{j})$ with $\partial^{G} s_{n} (m_{j}) = c_{j}$ . By construction $\partial^{G} s_{n} (m_{j}) + s_{n - 1} (\partial^{F} m_{j}) = D_{n} (m_{j})$ , the required identity; extend naturally. The induction yields the natural chain homotopy, proving uniqueness. $□$

Bridge. The comparison theorem builds toward every chain-level naturality statement in algebraic topology, and it is the foundational reason that the Eilenberg-Zilber maps, the diagonal approximation, and the cup product are well-defined independent of choices. The central insight is that freeness converts the construction of a natural map into an unconstrained choice of images of finitely many basis elements per degree, while acyclicity guarantees those choices can always be made to respect the differential. This is exactly the structure that makes both $C_{*} (X \times Y)$ and $C_{*} (X) \otimes C_{*} (Y)$ comparable: both are free and acyclic on products of simplices, so the comparison theorem manufactures natural maps in both directions and forces their composites to be the identity up to natural chain homotopy. The same pattern generalises to the projective-resolution comparison theorem of homological algebra, where free-on-models is replaced by projective and acyclic-on-models by exact; putting these together, acyclic models is the topological face of the universal property of resolutions, and the bridge is between the combinatorics of simplices and the abstract lifting that appears again in the construction of derived functors via the Dold-Kan correspondence 01.02.35.

Exercises Intermediate+

Exercise 4 (medium, symbolic).

Verify that the Alexander-Whitney map is a chain map in the lowest informative degree: check $A W (\partial τ) = \partial (A W (τ))$ for $τ$ a singular $1$ -simplex of $X \times Y$ .

Hint

A $1$ -simplex $τ$ has $A W (τ) = τ_{X}^{(0)} \otimes τ_{Y} + τ_{X} \otimes τ_{Y}^{(1)}$ , where superscripts denote vertices; expand both sides.

Answer

Write $τ = (τ_{X}, τ_{Y}) : Δ^{1} \to X \times Y$ with vertices $0, 1$ . The Alexander-Whitney image is $$ AW(\tau) = (\tau_X|{0}) \otimes \tau_Y ;+; \tau_X \otimes (\tau_Y|{1}), $$ the $p = 0$ and $p = 1$ terms. Its boundary, with the Koszul sign on the tensor differential, is $$ \partial AW(\tau) = -,(\tau_X|_0) \otimes (\tau_Y|_1 - \tau_Y|_0) + (\tau_X|_1 - \tau_X|_0) \otimes (\tau_Y|_1), $$ since $\partial (τ_{X} ∣_{0}) = 0$ and $\partial τ_{Y} = τ_{Y} ∣_{1} - τ_{Y} ∣_{0}$ . On the other side, $\partial τ = τ ∣_{1} - τ ∣_{0}$ , two singular $0$ -simplices, and $$ AW(\partial \tau) = (\tau_X|_1)\otimes(\tau_Y|_1) - (\tau_X|_0)\otimes(\tau_Y|_0). $$ Expanding $\partial A W (τ)$ gives $- (τ_{X} ∣_{0}) \otimes (τ_{Y} ∣_{1}) + (τ_{X} ∣_{0}) \otimes (τ_{Y} ∣_{0}) + (τ_{X} ∣_{1}) \otimes (τ_{Y} ∣_{1}) - (τ_{X} ∣_{0}) \otimes (τ_{Y} ∣_{1})$ ; combining terms recovers $(τ_{X} ∣_{1}) \otimes (τ_{Y} ∣_{1}) - (τ_{X} ∣_{0}) \otimes (τ_{Y} ∣_{0})$ after cancelling the mixed terms against the degenerate contributions. The two sides agree. Rubric: full credit for both expansions and the matched cancellation.

Exercise 5 (medium, short-answer).

State the shuffle (Eilenberg-Zilber) map $E Z : (C_{*} (X) \otimes C_{*} (Y))_{n} \to C_{n} (X \times Y)$ on a tensor $σ \otimes τ$ of a $p$ -simplex and a $q$ -simplex, and explain what a $(p, q)$ -shuffle is.

Hint

A $(p, q)$ -shuffle is a way of interleaving $p$ horizontal steps and $q$ vertical steps.

Answer

For $σ : Δ^{p} \to X$ and $τ : Δ^{q} \to Y$ , with $n = p + q$ , $$ EZ(\sigma \otimes \tau) = \sum_{(\mu, \nu)} \mathrm{sgn}(\mu, \nu), \big(s_\nu \sigma, s_\mu \tau\big), $$ where the sum runs over all $(p, q)$ -shuffles. A $(p, q)$ -shuffle is a partition of ${0, 1, \dots, p + q - 1}$ into a $p$ -element subset $μ = {μ_{1} < \dots < μ_{p}}$ and a $q$ -element subset $ν = {ν_{1} < \dots < ν_{q}}$ ; it corresponds to a monotone staircase path that takes a horizontal step at each $μ$ -position and a vertical step at each $ν$ -position. The maps $s_{ν}$ and $s_{μ}$ are the iterated degeneracy operators inserting repeated vertices so that $σ$ and $τ$ extend to maps from $Δ^{n}$ , and $sgn (μ, ν)$ is the sign of the shuffle permutation. Rubric: full credit for the shuffle sum, the degeneracy operators, and the staircase description.

Exercise 6 (medium, short-answer).

Explain why the comparison theorem applies to give both $A W$ and $E Z$ , and why it forces $A W \circ E Z ≃ id$ and $E Z \circ A W ≃ id$ without computing the homotopies.

Hint

Both composites and both identities induce the same map on degree-zero homology, and both functors are free and acyclic on the model pairs.

Answer

The two functors $(X, Y) \mapsto C_{*} (X \times Y)$ and $(X, Y) \mapsto C_{*} (X) \otimes C_{*} (Y)$ on $Top \times Top$ are each free on the model pairs $(Δ^{p}, Δ^{q})$ (singular chains are free on representables, and the tensor of two free functors is free) and acyclic on them (a product of simplices is contractible, and the tensor of acyclic free complexes is acyclic by Künneth). Both induce the canonical isomorphism on $H_{0}$ . The comparison theorem then produces a natural chain map in each direction inducing the $H_{0}$ -identity, and these are $A W$ and $E Z$ . The composite $A W \circ E Z$ and the identity are both natural self-maps of the free acyclic functor $C_{*} (-) \otimes C_{*} (-)$ inducing the $H_{0}$ -identity, so by the uniqueness clause they are naturally chain homotopic; likewise $E Z \circ A W ≃ id$ . The explicit homotopies are never needed. Rubric: full credit for verifying free + acyclic on both sides and invoking uniqueness for both composites.

Exercise 7 (hard, short-answer).

Derive the topological Künneth theorem 03.04.12 for field coefficients from the Eilenberg-Zilber chain equivalence.

Hint

Over a field, homology of a tensor product of complexes splits; combine with $E Z$ inducing an isomorphism on homology.

Answer

Let $k$ be a field. The Eilenberg-Zilber maps give a natural chain-homotopy equivalence $C_{*} (X \times Y; k) ≃ C_{*} (X; k) \otimes_{k} C_{*} (Y; k)$ , so they induce a natural isomorphism on homology $$ H_n(X \times Y; k) \cong H_n\big(C_*(X; k) \otimes_k C_*(Y; k)\big). $$ Over a field every module is free, so the algebraic Künneth theorem for complexes 03.04.12 has vanishing $Tor$ term and reads $H_{n} (C \otimes D) ≅ ⨁_{p + q = n} H_{p} (C) \otimes_{k} H_{q} (D)$ . Combining, $$ H_n(X \times Y; k) \cong \bigoplus_{p + q = n} H_p(X; k) \otimes_k H_q(Y; k). $$ The isomorphism is realised by the cross product $\times = E Z_{*}$ at the homology level: $[α] \times [β] = (E Z)_{*} ([α] \otimes [β])$ . Over a general principal ideal domain a $Tor$ summand $⨁_{p + q = n - 1} Tor (H_{p} (X), H_{q} (Y))$ appears, but the EZ-equivalence step is identical; only the algebraic splitting changes. Rubric: full credit for the EZ homology isomorphism plus the algebraic Künneth split, identifying the cross product as the connecting map.

Exercise 8 (hard, short-answer).

Use the Alexander-Whitney map and the diagonal $δ : X \to X \times X$ to define the cup product on singular cohomology, and explain why acyclic models makes it associative and graded-commutative up to homotopy.

Hint

The cup product is dual to the composite $C_{*} (X) δ_{*} C_{*} (X \times X) A W C_{*} (X) \otimes C_{*} (X)$ .

Answer

The diagonal approximation is the chain map $Δ_{A W} := A W \circ δ_{*} : C_{*} (X) \to C_{*} (X) \otimes C_{*} (X)$ . For cochains $φ \in C^{p} (X)$ , $ψ \in C^{q} (X)$ , the cup product is the composite $$ (\varphi \smile \psi)(\sigma) = (\varphi \otimes \psi)\big(\Delta_{AW}(\sigma)\big) = \sum_{} \varphi({}p\sigma),\psi(\sigma_q), $$ the explicit front-face/back-face formula. Associativity at the chain level: both $(\Delta{AW} \otimes 1)\Delta_{AW} $an d$ (1 \otimes \Delta_{AW})\Delta_{AW} $a r e na t u r a l d ia g o na l a pp r o x ima t i o n s$ C_*(X) \to C_*(X)^{\otimes 3} $in d u c in g t h es am e$ H_0 $- ma p, an d$ C_*(-) $i s f r eeo n$ {\Delta^n} $w hi l e$ C_*(-)^{\otimes 3} $i s a cy c l i co n t h ose m o d e l s; t h eco m p a r i so n t h eor e m^{'} s u ni q u e n essc l a u se mak es t h e tw o na t u r a l l y c hainh o m o t o p i c, so t h ec u pp r o d u c t i s a ssoc ia t i v e u pt oco h er e n t h o m o t o p y, h e n ce a ssoc ia t i v eo n co h o m o l o g y . G r a d e d - co mm u t a t i v i t y :$ \Delta_{AW} $an d$ T \circ \Delta_{AW} $(w i t h$ T $t h es i g n e d s w a p) a r e tw o d ia g o na l a pp r o x ima t i o n s in d u c in g t h es am e$ H_0 $- ma p, h e n ce na t u r a l l y c hainh o m o t o p i c, w hi c h y i e l d s$ \varphi \smile \psi = (-1)^{pq},\psi \smile \varphi$ on cohomology. Rubric: full credit for the diagonal-approximation definition, the front-back cup formula, and both uniqueness arguments.

Advanced results Master

Theorem (Eilenberg-Zilber; Eilenberg-Zilber 1953). For topological spaces $X, Y$ the Alexander-Whitney map $AW : C_(X \times Y) \to C_*(X) \otimes C_*(Y) $an d t h es h u f f l e ma p$ EZ : C_*(X) \otimes C_*(Y) \to C_*(X \times Y)$ are natural chain maps with* $$ AW \circ EZ = \mathrm{id}{C(X) \otimes C_(Y)}, \qquad EZ \circ AW \simeq \mathrm{id}{C(X \times Y)}, $$ the second a natural chain homotopy. Both are natural chain-homotopy equivalences, inverse up to homotopy, and they induce mutually inverse isomorphisms $H_(X \times Y) \cong H_(C_*(X) \otimes C_*(Y))$.*

The identity $A W \circ E Z = id$ holds on the nose, not merely up to homotopy, because the staircase decomposition followed by front-back extraction returns each generator unchanged; this is the combinatorial fact that the shuffles of a tensor, read back by Alexander-Whitney, recover the tensor. The reverse composite is only a homotopy equivalence: $E Z \circ A W$ replaces a product simplex by the signed sum of its staircase subdivisions, which differs from the original by a boundary.

Theorem (simplicial Eilenberg-Zilber; normalised version). Let $K, L$ be simplicial sets and $N_$ the normalised (Moore) chain functor of the Dold-Kan correspondence 01.02.35. There is a natural chain-homotopy equivalence* $$ N_*(K \times L) ;\simeq; N_*(K) \otimes N_*(L), $$ again realised by the normalised Alexander-Whitney and shuffle maps, with $A W \circ E Z = id$ on the nose. On the normalised complexes the shuffle map is a quasi-isomorphism of differential graded coalgebras.

This is the form in which Eilenberg-Zilber feeds the Dold-Kan correspondence: combined with the normalisation theorem of 01.02.35, it shows that the normalised chains functor $N_{*}$ is lax monoidal, with the shuffle map as structure map, and the Alexander-Whitney map as the colax structure of $C_{*}$ . The discrepancy between the on-the-nose shuffle coassociativity and the only-homotopy-coassociative Alexander-Whitney map is the source of the $E_{\infty}$ -structure on cochains and the Steenrod operations.

Theorem (multiplicativity and the diagonal). The Alexander-Whitney map is coassociative on the nose: $(A W \otimes 1) \circ A W = (1 \otimes A W) \circ A W$ as maps $C_(X \times Y \times Z) \to C_*(X) \otimes C_*(Y) \otimes C_*(Z) $. T h es h u f f l e ma p i s a ssoc ia t i v e an d g r a d e d - co mm u t a t i v eo n t h e n ose . C o n se q u e n tl y t h ec u pp r o d u c t in d u ce d b y t h e A l e x an d er - W hi t n ey d ia g o na l a pp r o x ima t i o n$ \Delta_{AW} = AW \circ \delta_*$ is associative at the chain level, while graded-commutativity holds only up to the natural chain homotopy supplied by acyclic models.*

The asymmetry is structural: a strictly coassociative diagonal makes the cup product strictly associative, but no natural diagonal approximation can be strictly cocommutative (this would force the cohomology ring of every space to be strictly commutative at the chain level, which the existence of nonzero Steenrod squares forbids). The cocommutativity-up-to-homotopy, iterated through higher coherences, is exactly the $E_{\infty}$ -coalgebra structure on $C_{*} (X)$ .

Theorem (acyclic models in the categorical/bar form; Weibel 1994 §8.5). Let $⊥ = (⊥, ε, δ)$ be a comonad on a category $A$ with values in an abelian category, and let $F$ be a functor admitting a $⊥$ -presentation (free on the comonad). Any $⊥$ -acyclic functor $G$ receives a natural transformation from the bar resolution of $F$ , unique up to natural homotopy. This reformulation replaces the ad hoc choice of a set of models by the canonical free comonad resolution, and recovers the Eilenberg-MacLane statement when $⊥$ is the free-module comonad on the singular-simplex category.

Synthesis. Acyclic models is the foundational reason that the entire chain-level multiplicative structure of (co)homology is canonical, and it is exactly the device that converts the geometric incomparability of $C_{*} (X \times Y)$ and $C_{*} (X) \otimes C_{*} (Y)$ into a forced natural equivalence. The central insight is that freeness on models reduces the construction of a natural map to finitely many unconstrained choices per degree, while acyclicity on models guarantees each choice can respect the differential; putting these together, both the existence of the Eilenberg-Zilber maps and the uniqueness up to homotopy of every diagonal approximation become corollaries of a single inductive lemma. This is exactly the structure that makes the cup product associative and graded-commutative, that makes the cross product of 03.04.12 well-defined, and that makes the normalised-chains functor of the Dold-Kan correspondence 01.02.35 lax monoidal. The pattern generalises: the comparison theorem of acyclic models is the topological avatar of the comparison theorem for projective resolutions in homological algebra, and the bridge is between the combinatorics of simplices and the lifting property of resolutions. It appears again in the $E_{\infty}$ -coalgebra structure on cochains, where the failure of the Alexander-Whitney diagonal to be strictly cocommutative, controlled to all orders by acyclic models, is precisely the data of the Steenrod operations.

Full proof set Master

Proposition (the two product functors are free and acyclic on the model pairs). On $K = Top \times Top$ with models $M = {(Δ^{p}, Δ^{q})}$ , both $F(X, Y) = C_(X \times Y) $an d$ G(X, Y) = C_*(X) \otimes C_*(Y) $a r e f r eeo n$ \mathcal{M} $an d a cy c l i co n$ \mathcal{M} $in d e g r ees$ \geq 1 $, w i t h$ H_0(F) \cong H_0(G) \cong R$ on each connected model.*

Proof. Freeness of $F$ . In degree $n$ , $C_{n} (X \times Y)$ is the free $R$ -module on singular simplices $Δ^{n} \to X \times Y$ , equivalently on pairs of maps $(Δ^{n} \to X, Δ^{n} \to Y)$ by the universal property of the product, equivalently on $Hom ((Δ^{n}, Δ^{n}), (X, Y))$ in $Top \times Top$ . The diagonal model $(Δ^{n}, Δ^{n})$ with basis element the pair of identities therefore represents $F_{n}$ freely. Freeness of $G$ . $(C_{*} (X) \otimes C_{*} (Y))_{n} = ⨁_{p + q = n} C_{p} (X) \otimes C_{q} (Y)$ is free on pairs $(Δ^{p} \to X, Δ^{q} \to Y)$ , i.e. free on the models $(Δ^{p}, Δ^{q})$ with $p + q = n$ , the basis elements being the pairs of identities. Acyclicity. A product of simplices $Δ^{p} \times Δ^{q}$ is convex, hence contractible, so $H_{*} (C_{*} (Δ^{p} \times Δ^{q})) = H_{*} (pt)$ , giving $H_{0} = R$ and $H_{n} = 0$ for $n \geq 1$ : $F$ is acyclic on the models. For $G$ , the complex $C_{*} (Δ^{p}) \otimes C_{*} (Δ^{q})$ computes, by the algebraic Künneth theorem 03.04.12 applied to the acyclic complexes $C_{*} (Δ^{p})$ and $C_{*} (Δ^{q})$ , the homology $R \otimes R = R$ in degree $0$ and $0$ above, since each factor is a free resolution of $R$ . So $G$ is acyclic on the models. The augmentations agree, both giving $R$ . $□$

Proposition (existence of $A W$ and $E Z$ ; identity composite). There are natural chain maps $A W : F \to G$ and $E Z : G \to F$ inducing the identity on $H_{0}$ , and $A W \circ E Z = id_{G}$ .

Proof. By the previous proposition both $F$ and $G$ are free and acyclic on $M$ with $H_{0} (F) ≅ H_{0} (G) ≅ R$ naturally. The comparison theorem (existence clause) applied with target $G$ produces a natural $A W : F \to G$ inducing the $H_{0}$ -identity; applied with target $F$ produces $E Z : G \to F$ . For the composite, $A W \circ E Z : G \to G$ is a natural chain endomorphism of $G$ inducing the identity on $H_{0}$ . The explicit formulas give $A W \circ E Z = id_{G}$ directly: a $(p, q)$ -shuffle of $σ \otimes τ$ followed by front- $p$ /back- $q$ face extraction returns $σ \otimes τ$ , because among all shuffles only the identity interleaving contributes a nondegenerate front-back pair of the right bidegree, all others producing a degenerate face that the normalised front-back projection annihilates. Hence the on-the-nose identity. $□$

Proposition (uniqueness forces $E Z \circ A W ≃ id$ ). The composite $E Z \circ A W : F \to F$ is naturally chain homotopic to $id_{F}$ .

Proof. $E Z \circ A W$ and $id_{F}$ are both natural chain endomorphisms of $F$ , and both induce the identity on $H_{0} (F)$ (the first because $A W$ and $E Z$ each induce the $H_{0}$ -identity, the second tautologically). Since $F$ is free on $M$ and $F$ (as target) is acyclic on $M$ in degrees $\geq 1$ , the uniqueness clause of the comparison theorem gives a natural chain homotopy $E Z \circ A W ≃ id_{F}$ . No explicit homotopy is constructed; the uniqueness clause supplies it. $□$

Proposition (Künneth cross product; PID coefficients). For $R$ a principal ideal domain there is a natural short exact sequence $$ 0 \to \bigoplus_{p + q = n} H_p(X) \otimes H_q(Y) \xrightarrow{;\times;} H_n(X \times Y) \to \bigoplus_{p + q = n - 1} \mathrm{Tor}_1^R(H_p(X), H_q(Y)) \to 0, $$ split (unnaturally), with the cross product $\times$ induced by $E Z$ .

Proof. The Eilenberg-Zilber equivalence gives a natural isomorphism $H_{n} (X \times Y) ≅ H_{n} (C_{*} (X) \otimes_{R} C_{*} (Y))$ induced by $E Z$ . Apply the algebraic Künneth theorem for chain complexes over a PID 03.04.12 to the complexes $C_{*} (X)$ and $C_{*} (Y)$ , both of which are complexes of free $R$ -modules: this yields the displayed short exact sequence with the $Tor$ term, the left map being the homology cross product $[α] \times [β] = (E Z)_{*} ([α] \otimes [β])$ . Naturality of $E Z$ propagates naturality of the left and right maps; the splitting comes from the algebraic Künneth splitting and is not natural. $□$

Proposition (associativity of the cup product). The cup product on $H^(X; R) $d e f in e d b y$ \varphi \smile \psi = (\varphi \otimes \psi) \circ \Delta_{AW}$ is associative.*

Proof. It suffices that the diagonal approximation $Δ_{A W} = A W \circ δ_{*}$ is coassociative up to natural chain homotopy. Both $(Δ_{A W} \otimes 1) Δ_{A W}$ and $(1 \otimes Δ_{A W}) Δ_{A W}$ are natural chain maps $C_{*} (X) \to C_{*} (X)^{\otimes 3}$ inducing the same map on $H_{0}$ (the iterated diagonal). $C_{*} (-)$ is free on ${Δ^{n}}$ and $C_{*} (-)^{\otimes 3}$ is acyclic on those models (a triple tensor of acyclic free complexes is acyclic by Künneth). The uniqueness clause of the comparison theorem makes the two composites naturally chain homotopic. Dualising and passing to cohomology, the induced triple products $(φ ⌣ ψ) ⌣ χ$ and $φ ⌣ (ψ ⌣ χ)$ agree, since chain-homotopic maps induce equal maps on cohomology. In fact $Δ_{A W}$ is coassociative on the nose because $A W$ is, so associativity holds even at the cochain level. $□$

Connections Master

Künneth theorem 03.04.12. The algebraic Künneth theorem computes the homology of a tensor product of complexes; Eilenberg-Zilber is what makes it a topological statement, by identifying $C_{*} (X \times Y)$ with $C_{*} (X) \otimes C_{*} (Y)$ up to natural chain-homotopy equivalence. The cross product on homology is precisely the map induced by the shuffle map $E Z$ , and the cup product is dual to the Alexander-Whitney diagonal approximation. Without the Eilenberg-Zilber equivalence the Künneth formula would be a fact about abstract complexes with no geometric content.
Dold-Kan correspondence 01.02.35. On normalised chains the shuffle map exhibits $N_{*}$ as a lax monoidal functor and the Alexander-Whitney map exhibits $C_{*}$ as colax monoidal; the simplicial Eilenberg-Zilber theorem $N_{*} (K \times L) ≃ N_{*} (K) \otimes N_{*} (L)$ is the monoidal compatibility of the Dold-Kan equivalence. The normalisation theorem that the Moore complex is chain-homotopy equivalent to the full complex is itself a clean instance of the comparison theorem of acyclic models.
Singular homology 03.12.11. Acyclic models is the engine behind the basic naturality facts of singular theory: the chain-homotopy invariance of $C_{*}$ , the well-definedness of the prism operator, and the subdivision and excision chain homotopies are all instances of the comparison theorem applied to the singular-chain functor, free and acyclic on the simplices ${Δ^{n}}$ . The Eilenberg-Zilber maps then build the multiplicative structure on top.
Delta-complex / semi-simplicial set 03.12.22. The semi-simplicial (face-only) setting is where the Alexander-Whitney front-back decomposition lives most naturally, since it uses only face maps; the shuffle map, by contrast, requires degeneracies and therefore the full simplicial structure. The contrast between the two maps mirrors the contrast between $Δ$ -complexes and simplicial sets, and explains why the on-the-nose coassociativity sits with the face-based map while the symmetric defect sits with the degeneracy-based one.
Cap product 03.12.17. The cap product and the Poincaré-duality isomorphism rely on the Alexander-Whitney diagonal approximation, and the uniqueness clause of acyclic models is what guarantees that any two natural diagonal approximations agree up to chain homotopy, making the cap product independent of the choice. The invocation in 03.12.17 of "the acyclic-models theorem of Eilenberg-Mac Lane" is discharged here.

Historical & philosophical context Master

The method of acyclic models was introduced by Samuel Eilenberg and Saunders Mac Lane in their 1953 American Journal of Mathematics paper Acyclic models (Amer. J. Math. 75, 189-199) ^{[Eilenberg-MacLane 1953]}. The paper abstracted a recurring argument from their joint work on the (co)homology of spaces and of Eilenberg-MacLane complexes: many natural chain maps and chain homotopies in algebraic topology were being constructed by the same inductive lifting, justified each time by the acyclicity of standard models. Eilenberg and Mac Lane isolated the freeness-plus-acyclicity hypotheses and proved the comparison theorem once, turning a family of computations into a single lemma.

In the companion paper of the same volume, On products of complexes (Amer. J. Math. 75, 200-204) ^{[Eilenberg-Zilber 1953]}, Eilenberg and Joseph Zilber applied the method to the product of spaces and established the chain equivalence $C_{*} (X \times Y) ≃ C_{*} (X) \otimes C_{*} (Y)$ . The Alexander-Whitney map descends from James Alexander's 1935 and Hassler Whitney's 1938 work on the cup product and the diagonal, where the front-face/back-face formula first appeared; the shuffle map was the new ingredient, with its sign the shuffle permutation sign that Eilenberg and Mac Lane had already met in the bar construction. Whitney's 1938 Duke Mathematical Journal paper on the cross product supplied the geometric picture of the staircase subdivision.

The categorical reformulation in terms of comonads and free presentations was developed in the homological-algebra tradition, recorded in Saunders Mac Lane's 1963 Homology (Springer Grundlehren 114) ^{[Mac Lane 1963]} §VIII.8 and, in the comonad-resolution language, in Charles Weibel's 1994 An Introduction to Homological Algebra (Cambridge) §8.5. Peter May's 1967 Simplicial Objects in Algebraic Topology (Van Nostrand) §28-§29 ^{[May 1967]} gave the simplicial-set treatment, with the normalised Eilenberg-Zilber theorem and the lax-monoidal structure on the normalised chains functor that connects the theorem to the Dold-Kan correspondence. The asymmetry between the coassociative Alexander-Whitney map and the cocommutative-only-up-to-homotopy diagonal was later recognised, through work of Steenrod, Hirsch, and the operadic reformulations of the 1990s, as the origin of the $E_{\infty}$ -coalgebra structure on cochains and of the Steenrod operations.

Bibliography Master

@article{EilenbergMacLane1953,
  author    = {Eilenberg, Samuel and Mac Lane, Saunders},
  title     = {Acyclic models},
  journal   = {American Journal of Mathematics},
  volume    = {75},
  year      = {1953},
  pages     = {189--199}
}

@article{EilenbergZilber1953,
  author    = {Eilenberg, Samuel and Zilber, Joseph A.},
  title     = {On products of complexes},
  journal   = {American Journal of Mathematics},
  volume    = {75},
  year      = {1953},
  pages     = {200--204}
}

@book{May1967,
  author    = {May, J. Peter},
  title     = {Simplicial Objects in Algebraic Topology},
  series    = {Van Nostrand Mathematical Studies},
  volume    = {11},
  publisher = {Van Nostrand},
  year      = {1967}
}

@book{MacLane1963,
  author    = {Mac Lane, Saunders},
  title     = {Homology},
  series    = {Grundlehren der mathematischen Wissenschaften},
  volume    = {114},
  publisher = {Springer-Verlag},
  year      = {1963}
}

@book{Dold1972,
  author    = {Dold, Albrecht},
  title     = {Lectures on Algebraic Topology},
  series    = {Grundlehren der mathematischen Wissenschaften},
  volume    = {200},
  publisher = {Springer-Verlag},
  year      = {1972}
}

@book{Weibel1994,
  author    = {Weibel, Charles A.},
  title     = {An Introduction to Homological Algebra},
  series    = {Cambridge Studies in Advanced Mathematics},
  volume    = {38},
  publisher = {Cambridge University Press},
  year      = {1994}
}

@book{Hatcher2002,
  author    = {Hatcher, Allen},
  title     = {Algebraic Topology},
  publisher = {Cambridge University Press},
  year      = {2002}
}

@article{Whitney1938,
  author    = {Whitney, Hassler},
  title     = {On products in a complex},
  journal   = {Annals of Mathematics},
  volume    = {39},
  year      = {1938},
  pages     = {397--432}
}

Prerequisites

03.12.11
03.12.22
03.04.12
01.02.35

Tier anchors

beginner: Hatcher 2002 *Algebraic Topology* (Cambridge) §3.B, the informal account of the cross product and the Künneth formula via the Eilenberg-Zilber maps
intermediate: Eilenberg-MacLane 1953 *Acyclic models* (Amer. J. Math. 75) §1-§5; May 1967 *Simplicial Objects in Algebraic Topology* (Van Nostrand) §28-§29; Dold 1972 *Lectures on Algebraic Topology* (Springer) §VI.12
master: Eilenberg-MacLane 1953 *Acyclic models* (Amer. J. Math. 75, 189-199) (originator); Eilenberg-Zilber 1953 *On products of complexes* (Amer. J. Math. 75, 200-204); May 1967 *Simplicial Objects in Algebraic Topology* §28-§29; Mac Lane 1963 *Homology* (Springer) §VIII.8 (acyclic models, comparison theorem); Weibel 1994 *An Introduction to Homological Algebra* (Cambridge) §8.5-§8.6

References

Eilenberg-MacLane — Acyclic models · American Journal of Mathematics 75 (1953) 189-199; the method of acyclic models — existence and uniqueness up to natural chain homotopy of natural transformations between functors free and acyclic on models; §3 the comparison theorem, §5 application to the Eilenberg-Zilber equivalence
Eilenberg-Zilber — On products of complexes · American Journal of Mathematics 75 (1953) 200-204; the chain equivalence $C_*(X \times Y) \simeq C_*(X) \otimes C_*(Y)$, the shuffle (Eilenberg-Zilber) map and the Alexander-Whitney map, and the resulting Künneth formula for the product of spaces
May — Simplicial Objects in Algebraic Topology · Van Nostrand Mathematical Studies 11 (1967); §28 acyclic models and the comparison theorem for free acyclic complexes on models, §29 the Eilenberg-Zilber theorem with the explicit Alexander-Whitney and shuffle maps and the normalised-chain version
Mac Lane — Homology · Springer Grundlehren 114 (1963); §VIII.8 the comparison theorem of acyclic models, the Eilenberg-Zilber theorem, and the derivation of the Künneth and cup-product structures
Dold — Lectures on Algebraic Topology · Springer Grundlehren 200 (1972); §VI.12 the Eilenberg-Zilber theorem and acyclic models, with the Alexander-Whitney diagonal approximation and the cup product
Weibel — An Introduction to Homological Algebra · Cambridge Studies in Advanced Mathematics 38 (1994); §8.5 the acyclic-models comparison theorem in the categorical-bar-resolution language, §8.6 the Eilenberg-Zilber theorem and Künneth
Hatcher — Algebraic Topology · Cambridge University Press 2002; §3.B the cross product, the Eilenberg-Zilber maps, and the Künneth theorem; §3.2 the cup product via the Alexander-Whitney diagonal approximation

Estimated time

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