← All writingMathematics blog

The Classical Dold-Kan correspondence

A proof of the equivalence between simplicial abelian groups and non-negatively graded chain complexes, through normalization and the Eilenberg-MacLane functor.

The question which Dold-Kan answers is quite natural from the perspective of algebraic topology. Namely, if AA is a simplicial abelian group, we can try to define a chain complex by setting its group in degree nn to be AnA_n and taking the boundary map to be the alternating sum of the face operators did_i. This defines a functor

Moore ⁣:sAb⟶Ch≥0(Z)\mathrm{Moore}\colon\sab\longrightarrow\ch(\ZZ)

from simplicial abelian groups to non-negatively graded chain complexes of abelian groups. Given the resemblance of the objects in either category, one might ask whether it is an equivalence.

The answer is no because of the degeneracy operators si ⁣:An−1→Ans_i\colon A_{n-1}\to A_n, whose images make the Moore complex too large. Instead, we will take its quotient by the subcomplex generated by degeneracies. The Dold-Kan correspondence says that the resulting construction is an equivalence of categories. The proof will consist of constructing an inverse functor.

We begin by constructing this quotient and finding an isomorphic description which will help prove the theorem. Then we define the inverse in two steps: first make a semisimplicial abelian group, and then freely add degeneracies. Throughout, we work with abelian groups.

The normalized Moore complex

Definition 1. A non-negatively graded chain complex over Z\ZZ is a sequence A=A∙A=A_\bullet of abelian groups

⋯→∂n+1An→∂nAn−1⟶⋯⟶A1→∂1A0\cdots\xrightarrow{\partial_{n+1}} A_n \xrightarrow{\partial_n}A_{n-1}\longrightarrow\cdots \longrightarrow A_1\xrightarrow{\partial_1}A_0

such that ∂n∂n+1=0\partial_n\partial_{n+1}=0 for every n≥1n\geq1. A morphism of such complexes f ⁣:(A,∂)→(B,∂)f\colon(A,\partial)\to(B,\partial) is a sequence of homomorphisms fn ⁣:An→Bnf_n\colon A_n\to B_n making

An→∂nAn−1fn↓↓fn−1Bn→∂nBn−1\begin{CD} A_n @>{\partial_n}>> A_{n-1} \\ @V{f_n}VV @VV{f_{n-1}}V \\ B_n @>{\partial_n}>> B_{n-1} \end{CD}

commute for every n≥1n\geq1. We write Ch≥0(Z)\ch(\ZZ) for the corresponding category.

On the other hand, let sAb=[Δop,Ab]\sab=[\simp^\op,\ab] be the category of simplicial abelian groups. Here Δ\simp is the category of finite nonempty ordered sets [n]={0,…,n}[n]=\{0,\ldots,n\} and order-preserving maps. We write di ⁣:[n−1]↪[n]d^i\colon[n-1]\mono[n] for the injection omitting ii, and si ⁣:[n]↠[n−1]s^i\colon[n]\epi[n-1] for the surjection identifying ii and i+1i+1. Applying a simplicial object gives the face and degeneracy maps did_i and sis_i in the opposite direction.

The goal is to construct an equivalence of categories sAb≃Ch≥0(Z)\sab\simeq\ch(\ZZ). We will first set up three intimately related functors sAb→Ch≥0(Z)\sab\to\ch(\ZZ) and learn their properties.

1. The Moore complex

Let AA be a simplicial abelian group. We associate to it a chain complex (A∗,∂)(A_*,\partial) with the same abelian groups and with

∂n=∑i=0n(−1)idi.\partial_n=\sum_{i=0}^n(-1)^i d_i.

Proposition 2. The assignment A↦A∗A\mapsto A_* defines a functor sAb→Ch≥0(Z)\sab\to\ch(\ZZ).

Proof. We first need ∂n−1∂n=0\partial_{n-1}\partial_n=0. From the definition,

∂n−1∂n=∑i,j(−1)i+jdjdi=∑i≤j(−1)i+jdjdi+∑i>j(−1)i+jdi−1dj=∑i≤j(−1)i+jdjdi−∑l≤k(−1)k+ldkdl=0.\begin{aligned} \partial_{n-1}\partial_n &=\sum_{i,j}(-1)^{i+j}d_jd_i \\ &=\sum_{i\leq j}(-1)^{i+j}d_jd_i +\sum_{i>j}(-1)^{i+j}d_{i-1}d_j \\ &=\sum_{i\leq j}(-1)^{i+j}d_jd_i -\sum_{l\leq k}(-1)^{k+l}d_kd_l=0. \end{aligned}

In the second step we used the simplicial identities, and in the third we set k=i−1k=i-1 and l=jl=j.

For functoriality, note that a simplicial map f ⁣:A→Bf\colon A\to B is a natural transformation, so

An→diAn−1fn↓↓fn−1Bn→diBn−1\begin{CD} A_n @>{d_i}>> A_{n-1} \\ @V{f_n}VV @VV{f_{n-1}}V \\ B_n @>{d_i}>> B_{n-1} \end{CD}

commutes. Since ∂n\partial_n is a linear combination of face maps, the same square obtained by replacing did_i with ∂n\partial_n still commutes. □

Definition 3. The chain complex A∗A_* is the Moore complex associated to the simplicial abelian group AA.

Example 4. Let XX be a topological space and Sing(X)\mathrm{Sing}(X) its singular simplicial set. Taking the free abelian group on the singular simplices in each degree gives a simplicial abelian group Z[Sing(X)]\ZZ[\mathrm{Sing}(X)]. Its Moore complex is the usual singular chain complex C∗(X;Z)C_*(X;\ZZ), whose homology is that of XX.

2. The reduced Moore complex

Recall that an nn-simplex x∈Anx\in A_n is degenerate if there is y∈An−1y\in A_{n-1} such that si(y)=xs_i(y)=x for some 0≤i≤n−10\leq i\leq n-1. Set

DAn=∑i=0n−1im⁡(si ⁣:An−1→An),DA0=0.DA_n=\sum_{i=0}^{n-1}\im(s_i\colon A_{n-1}\to A_n), \qquad DA_0=0.

We claim these groups form a subcomplex DA∗⊂A∗DA_*\subset A_*. We need to show that ∂(six)∈DAn−1\partial(s_i x)\in DA_{n-1}. Since disi=di+1si=idd_i s_i=d_{i+1}s_i=\id, the two corresponding terms cancel, and the simplicial identities give

∂(six)=∑j≠i,i+1(−1)jdjsi(x)=∑j>i+1(−1)jsidj−1(x)+∑j<i(−1)jsi−1dj(x).\begin{aligned} \partial(s_i x) &=\sum_{j\ne i,i+1}(-1)^j d_j s_i(x) \\ &=\sum_{j>i+1}(-1)^j s_i d_{j-1}(x) +\sum_{j<i}(-1)^j s_{i-1}d_j(x). \end{aligned}

Thus we have represented ∂(six)\partial(s_i x) as a sum of degenerate simplices. Empty sums are understood to be zero.

Definition 5. The reduced complex associated to a simplicial abelian group is the quotient complex

(A/DA)∗=A∗/DA∗.(A/DA)_*=A_*/DA_*.

We write q ⁣:A∗↠(A/DA)∗q\colon A_*\epi(A/DA)_* for the canonical quotient map.

This is functorial because simplicial maps preserve degenerate simplices, giving our second functor sAb→Ch≥0(Z)\sab\to\ch(\ZZ). The quotient map qq is then a natural transformation.

3. The normalized complex

Let AA be a simplicial abelian group, and set

NAn=⋂i=1nker⁡(di ⁣:An→An−1),NA0=A0.NA_n=\bigcap_{i=1}^n\ker(d_i\colon A_n\to A_{n-1}), \qquad NA_0=A_0.

We take the boundary map NAn→NAn−1NA_n\to NA_{n-1} to be d0d_0. This is well-defined because if x∈NAnx\in NA_n and 0<i≤n−10<i\leq n-1, then

did0(x)=d0di+1(x)=0.d_i d_0(x)=d_0d_{i+1}(x)=0.

Moreover, d0d0(x)=d0d1(x)=0d_0d_0(x)=d_0d_1(x)=0 implies that (NA∗,d0)(NA_*,d_0) is a chain complex. It is a subcomplex of the Moore complex: the inclusion i ⁣:NA∗→A∗i\colon NA_*\to A_* is a chain map, since

∂n(x)=d0(x)+∑i=1n(−1)idi(x)=d0(x)(x∈NAn).\partial_n(x)=d_0(x)+\sum_{i=1}^n(-1)^i d_i(x)=d_0(x) \qquad(x\in NA_n).

Proposition 6. The assignment A↦NA∗A\mapsto NA_* is functorial, and the inclusion ii is a natural transformation.

Proof. If f ⁣:A→Bf\colon A\to B is simplicial and x∈NAnx\in NA_n, then for i≥1i\geq1,

di(fn(x))=fn−1(di(x))=0.d_i(f_n(x))=f_{n-1}(d_i(x))=0.

Thus fnf_n carries NAnNA_n to NBnNB_n. Taking i=0i=0 in the same naturality identity shows that these restrictions define a chain map. Naturality of the inclusion follows directly. □

Definition 7. The normalized complex is the functor

N∗ ⁣:sAb⟶Ch≥0(Z),A⟼NA∗.N_*\colon\sab\longrightarrow\ch(\ZZ),\qquad A\longmapsto NA_*.

The Dold-Kan theorem is that N∗N_* is an equivalence. The first step in the proof will be that the reduced complex is isomorphic to the normalized complex.

Comparing the two normalizations

We will need a lemma first. For 0≤k≤n0\leq k\leq n, define

NkAn=⋂i=k+1nker⁡di,DkAn=∑i=kn−1im⁡si.N_kA_n=\bigcap_{i=k+1}^n\ker d_i, \qquad D_kA_n=\sum_{i=k}^{n-1}\im s_i.

In particular, N0An=NAnN_0A_n=NA_n and D0An=DAnD_0A_n=DA_n, whereas NnAn=AnN_nA_n=A_n and DnAn=0D_nA_n=0. The indexing is chosen so that we impose the equations dn=0,dn−1=0,…,d1=0d_n=0,d_{n-1}=0,\ldots,d_1=0 in that order.

Lemma 8 (Two exact sequences). Let AA be a simplicial abelian group, n≥1n\geq1, and 0≤k<n0\leq k<n.

  1. The sequence

    0⟶An−1/DkAn−1→skAn/Dk+1An⟶An/DkAn⟶00\longrightarrow A_{n-1}/D_kA_{n-1} \xrightarrow{s_k}A_n/D_{k+1}A_n \longrightarrow A_n/D_kA_n\longrightarrow0

    is well-defined and exact.

  2. The sequence

    0⟶NkAn−1→skNk+1An→ϕNkAn⟶00\longrightarrow N_kA_{n-1} \xrightarrow{s_k}N_{k+1}A_n \xrightarrow{\phi}N_kA_n\longrightarrow0

    is well-defined and exact, where ϕ(x)=x−skdk+1x\phi(x)=x-s_kd_{k+1}x.

Proof. For (1), the simplicial identities give, for i≥ki\geq k,

sksi=si+1sk.s_k s_i=s_{i+1}s_k.

Hence sk(DkAn−1)⊂Dk+1Ans_k(D_kA_{n-1})\subset D_{k+1}A_n, so the first map is well-defined. It is injective: if skx∈Dk+1Ans_kx\in D_{k+1}A_n, apply dkd_k. Since dksk=idd_k s_k=\id and, for i>ki>k,

dksi=si−1dk,d_k s_i=s_{i-1}d_k,

we obtain x∈DkAn−1x\in D_kA_{n-1}. The final map is a quotient map, so it is surjective, and its kernel is the image of sks_k because

DkAn=im⁡sk+Dk+1An.D_kA_n=\im s_k+D_{k+1}A_n.

For (2), if y∈NkAn−1y\in N_kA_{n-1} and i>k+1i>k+1, then

disky=skdi−1y=0.d_i s_k y=s_k d_{i-1}y=0.

Thus sks_k maps into Nk+1AnN_{k+1}A_n, and it is injective because dk+1sk=idd_{k+1}s_k=\id. If x∈Nk+1Anx\in N_{k+1}A_n and i>ki>k, then

didk+1x=dk+1di+1x=0,d_i d_{k+1}x=d_{k+1}d_{i+1}x=0,

so dk+1x∈NkAn−1d_{k+1}x\in N_kA_{n-1}. Subtracting skdk+1xs_kd_{k+1}x therefore preserves the equations defining Nk+1AnN_{k+1}A_n and also kills dk+1xd_{k+1}x. This proves that ϕ\phi takes values in NkAnN_kA_n.

Now ϕsk=0\phi s_k=0, while ϕ(x)=x\phi(x)=x for x∈NkAnx\in N_kA_n, so ϕ\phi is onto. Finally, if ϕ(x)=0\phi(x)=0, then x=skdk+1xx=s_kd_{k+1}x, with dk+1x∈NkAn−1d_{k+1}x\in N_kA_{n-1} as just proved. This establishes exactness in the middle. □

Theorem 9 (Normalized and degenerate simplices). For a simplicial abelian group AA, the map

NA∗⊕DA∗⟶A∗NA_*\oplus DA_*\longrightarrow A_*

induced by the two inclusions is a natural isomorphism of chain complexes. In particular, the composite

NA∗→iA∗→q(A/DA)∗NA_*\xrightarrow{i}A_*\xrightarrow{q}(A/DA)_*

is an isomorphism.

Proof. It suffices to prove the second statement. Indeed, once qiqi is an isomorphism, the map (qi)−1q ⁣:A∗→NA∗(qi)^{-1}q\colon A_*\to NA_* is a retraction of ii, and its kernel is DA∗DA_*.

We prove that for every n≥0n\geq0 and 0≤k≤n0\leq k\leq n, the composite

NkAn⟶An⟶An/DkAnN_kA_n\longrightarrow A_n\longrightarrow A_n/D_kA_n

is an isomorphism. We do this by induction on nn, and within a fixed degree by descending induction on kk.

For n=0n=0 there is nothing to prove. In any degree the case k=nk=n is the identity of AnA_n. Suppose we know the assertion in degree n−1n-1 and for k+1k+1 in degree nn. Put the two exact sequences of Lemma 8 into the diagram

0→NkAn−1→skNk+1An→ϕNkAn→0 ↓↓↓ 0→An−1/DkAn−1→skAn/Dk+1An→An/DkAn→0.\begin{CD} 0 @>>> N_kA_{n-1} @>{s_k}>> N_{k+1}A_n @>{\phi}>> N_kA_n @>>> 0 \\ @. @VVV @VVV @VVV @. \\ 0 @>>> A_{n-1}/D_kA_{n-1} @>{s_k}>> A_n/D_{k+1}A_n @>>> A_n/D_kA_n @>>> 0. \end{CD}

The vertical arrows are induced by inclusion followed by quotient. The left square commutes by the definition of how sks_k passes to the quotients. The right square commutes because

ϕ(x)=x−skdk+1x≡x(modDkAn).\phi(x)=x-s_kd_{k+1}x\equiv x\pmod{D_kA_n}.

The first nonzero vertical arrow is an isomorphism by induction on nn, and the second by induction on kk. Taking cokernels shows that the third is an isomorphism as well. At k=0k=0 this proves the desired comparison NAn≅An/DAnNA_n\cong A_n/DA_n.

All the maps used are natural in AA, and the comparison is already a chain map. Thus the resulting isomorphism is an isomorphism of chain complexes, natural in AA. □

The decomposition in Theorem 9 will be important in the proof of Dold-Kan. Before using it, however, we need to construct an inverse to N∗N_*.

The Eilenberg-MacLane functor

To prove Dold-Kan, we will explicitly describe a functor

K ⁣:Ch≥0(Z)⟶sAbK\colon\ch(\ZZ)\longrightarrow\sab

inverse to the normalized complex construction. Its construction factors through the category of semisimplicial abelian groups.

Definition 10. Write Δinj⊂Δ\simp_{\mathrm{inj}}\subset\simp for the wide subcategory consisting of injective order-preserving maps. The category of semisimplicial abelian groups is

ssAb=[Δinjop,Ab].\ssab=[\simp_{\mathrm{inj}}^\op,\ab].

Thus a semisimplicial abelian group has face maps satisfying the face identities, but no degeneracy maps are specified.

Let (A,∂)(A,\partial) be a chain complex. There is an associated semisimplicial abelian group Ks(A)K^s(A) defined as follows:

  1. Ks(A)n=AnK^s(A)_n=A_n.
  2. An identity injection induces the identity homomorphism.
  3. For a nonidentity injection j ⁣:[m]↪[n]j\colon[m]\mono[n], the map Ks(A)(j) ⁣:An→AmK^s(A)(j)\colon A_n\to A_m is zero unless m=n−1m=n-1 and j=d0j=d^0, in which case it is ∂n\partial_n.

To see that Ks(A)K^s(A) is a presheaf, consider a composite

[l]→j[m]→i[n].[l]\xrightarrow{j}[m]\xrightarrow{i}[n].

We must check Ks(A)(ij)=Ks(A)(j)∘Ks(A)(i)K^s(A)(ij)=K^s(A)(j)\circ K^s(A)(i). If one of the injections is an identity, this is immediate. Otherwise the composite has codimension at least two, so its associated map is zero. The composite of the two associated maps is also zero, except possibly when both injections are d0d^0; in that case it is ∂n−1∂n=0\partial_{n-1}\partial_n=0.

This construction is functorial in the chain complex: a chain map commutes with the nonzero face map because it commutes with ∂\partial, and commutes with all the other face maps because those are zero.

Definition 11. The functor

Ks ⁣:Ch≥0(Z)⟶ssAbK^s\colon\ch(\ZZ)\longrightarrow\ssab

just constructed is the semisimplicial Eilenberg-MacLane functor.

It remains to provide the second half of the composite Ch≥0(Z)→ssAb→sAb\ch(\ZZ)\to\ssab\to\sab. Recall the following elementary result about Δ\simp.

Lemma 12 (Epimonic factorization). Every map u ⁣:[m]→[n]u\colon[m]\to[n] in Δ\simp has a unique factorization

[m]→π[k]→i[n][m]\xrightarrow{\pi}[k]\xrightarrow{i}[n]

in which π\pi is surjective and ii is injective. Moreover, π\pi is a split epimorphism.

Proof. Factor uu through its image, which is uniquely order-isomorphic to [k][k] for some kk. The image has cardinality k+1k+1, so no other [k][k] can work; order preservation then determines both maps. An order-preserving section of π\pi is obtained by choosing the least element in each fiber. □

Proposition 13 (Adding degeneracies). There is a canonical functor E ⁣:ssAb→sAbE\colon\ssab\to\sab defined in degree nn by

E(F)n=⨁ϕ ⁣:[n]↠[k]Fk.E(F)_n=\bigoplus_{\phi\colon[n]\epi[k]}F_k.

The sum is taken over all surjections in Δ\simp with source [n][n].

Proof. Let u ⁣:[n]→[n′]u\colon[n]\to[n'] be a map in Δ\simp. To define u∗ ⁣:E(F)n′→E(F)nu^*\colon E(F)_{n'}\to E(F)_n, we define it on the summand associated to each surjection ϕ′ ⁣:[n′]↠[k′]\phi'\colon[n']\epi[k']. Factor ϕ′u\phi'u into a surjection followed by an injection, obtaining the unique commutative square

[n]→u[n′]ϕ↓↓ϕ′[k]→v[k′].\begin{CD} [n] @>{u}>> [n'] \\ @V{\phi}VV @VV{\phi'}V \\ [k] @>{v}>> [k']. \end{CD}

Here ϕ\phi and ϕ′\phi' are surjective, and vv is injective. On the ϕ′\phi'-summand, define u∗u^* to be

Fk′→F(v)FkF_{k'}\xrightarrow{F(v)}F_k

with image in the ϕ\phi-summand of E(F)nE(F)_n; all its other components are zero.

It remains to check composition. For maps [m]→u[n]→w[p][m]\xrightarrow{u}[n]\xrightarrow{w}[p] and a surjection θ ⁣:[p]↠[l]\theta\colon[p]\epi[l], perform the factorization twice:

[m]→u[n]→w[p]ψ↓ϕ↓↓θ[j]→a[k]→b[l].\begin{CD} [m] @>{u}>> [n] @>{w}>> [p] \\ @V{\psi}VV @V{\phi}VV @VV{\theta}V \\ [j] @>{a}>> [k] @>{b}>> [l]. \end{CD}

The vertical maps are surjections and the bottom maps are injections. By uniqueness, the outer rectangle is the factorization of θwu\theta wu. Hence the two successive maps on the θ\theta-summand are F(a)F(b)=F(ba)F(a)F(b)=F(ba), exactly the map prescribed for (wu)∗(wu)^*. Identities are respected for the same reason.

Finally, for a semisimplicial map η ⁣:F→G\eta\colon F\to G, define

E(η)n=⨁ϕ ⁣:[n]↠[k]ηk.E(\eta)_n=\bigoplus_{\phi\colon[n]\epi[k]}\eta_k.

Naturality of η\eta with respect to the injection vv in the first square shows that this commutes with every u∗u^*. Thus E(η)E(\eta) is a simplicial map, and the construction respects identities and composition. □

Definition 14. The composite

K ⁣:Ch≥0(Z)→KsssAb→EsAbK\colon\ch(\ZZ)\xrightarrow{K^s}\ssab\xrightarrow{E}\sab

is the Eilenberg-MacLane functor. In particular,

K(A)n=⨁ϕ ⁣:[n]↠[k]Ak.K(A)_n=\bigoplus_{\phi\colon[n]\epi[k]}A_k.

The Dold-Kan correspondence says that KK is both a left and a right inverse to N∗N_*, up to natural isomorphism. We can now prove one direction.

Theorem 15 (Normalizing the Eilenberg-MacLane construction). For a chain complex AA, inclusion of the identity summand gives a natural isomorphism

A→∼N∗K(A).A\xrightarrow{\sim}N_*K(A).

Proof. The only surjection [n]→[n][n]\to[n] is the identity, so the definition of K(A)nK(A)_n contains AnA_n as a distinguished direct summand. First we show that this summand lies in N(KA)nN(KA)_n.

For the identity summand, the factorization used to define the face did_i is simply

[n−1]→di[n]id↓↓id[n−1]→di[n].\begin{CD} [n-1] @>{d^i}>> [n] \\ @V{\id}VV @VV{\id}V \\ [n-1] @>{d^i}>> [n]. \end{CD}

Thus did_i on AnA_n is Ks(A)(di)K^s(A)(d^i), with image in the identity summand of degree n−1n-1. It is zero for i≥1i\geq1 and is ∂n\partial_n for i=0i=0.

Next we identify the degenerate part of K(A)nK(A)_n. A degeneracy sis_i takes the summand indexed by ψ ⁣:[n−1]↠[k]\psi\colon[n-1]\epi[k] identically to the summand indexed by ψsi ⁣:[n]↠[k]\psi s^i\colon[n]\epi[k]. Therefore its image lies in the sum of the nonidentity summands. Conversely, every nonidentity surjection ϕ ⁣:[n]↠[k]\phi\colon[n]\epi[k] identifies some adjacent pair i,i+1i,i+1, and hence factors as ϕ=ψsi\phi=\psi s^i. Its entire summand is in the image of sis_i. Consequently,

D(KA)n=⨁ϕ:[n]↠[k]k<nAk,K(A)n=An⊕D(KA)n.D(KA)_n=\bigoplus_{\substack{\phi\colon[n]\epi[k]\\k<n}}A_k, \qquad K(A)_n=A_n\oplus D(KA)_n.

By Theorem 9, we also have

K(A)n=N(KA)n⊕D(KA)n.K(A)_n=N(KA)_n\oplus D(KA)_n.

Since An⊂N(KA)nA_n\subset N(KA)_n, these two decompositions imply N(KA)n=AnN(KA)_n=A_n as subgroups of K(A)nK(A)_n. Explicitly, if x=a+dx=a+d belongs to N(KA)nN(KA)_n, then d=x−ad=x-a lies in N(KA)n∩D(KA)n=0N(KA)_n\cap D(KA)_n=0.

We already checked that d0d_0 on this summand is ∂n\partial_n, so these identifications respect the differentials. A chain map acts on each identity summand by its degree-nn component, which proves naturality. □

Finishing Dold-Kan

We now complete the proof of the classical correspondence.

Theorem 16 (Dold-Kan correspondence). The functor

N∗ ⁣:sAb⟶Ch≥0(Z)N_*\colon\sab\longrightarrow\ch(\ZZ)

is an equivalence of categories, with inverse given by KK.

We have already seen one direction in Theorem 15. To finish, we construct a natural isomorphism

μ ⁣:K∘N∗⟶IdsAb.\mu\colon K\circ N_*\longrightarrow\mathrm{Id}_{\sab}.

For a simplicial abelian group AA, define

μ(A)n ⁣:K(NA∗)n=⨁ϕ ⁣:[n]↠[k]NAk⟶An\mu(A)_n\colon K(NA_*)_n =\bigoplus_{\phi\colon[n]\epi[k]}NA_k\longrightarrow A_n

on the ϕ\phi-summand by the composite

NAk↪Ak→ϕ∗An.NA_k\mono A_k\xrightarrow{\phi^*}A_n.

Proposition 17. The maps μ(A)n\mu(A)_n define a natural transformation KN∗→IdsAbKN_*\to\mathrm{Id}_{\sab}.

Proof. First consider an injection v ⁣:[k]↪[k′]v\colon[k]\mono[k']. On NAk′NA_{k'}, the induced map v∗v^* agrees with the map prescribed by Ks(NA∗)K^s(NA_*):

  • If vv is the identity, both maps are the identity.
  • If v=d0v=d^0, both maps are the differential d0d_0.
  • Otherwise vv omits some j≥1j\geq1, so it factors through dj ⁣:[k′−1]↪[k′]d^j\colon[k'-1]\mono[k']. Its induced map therefore factors through djd_j, which vanishes on NAk′NA_{k'}.

For the last point, an injection omitting at least two elements necessarily omits a positive element, so this also covers all injections of codimension at least two.

Now let u ⁣:[n]→[n′]u\colon[n]\to[n'], and factor ϕ′u=vϕ\phi'u=v\phi as in Proposition 13. On the ϕ′\phi'-summand, applying the simplicial map of KN∗AKN_*A and then μ(A)n\mu(A)_n gives

NAk′→v∗NAk↪Ak→ϕ∗An.NA_{k'}\xrightarrow{v^*}NA_k \mono A_k\xrightarrow{\phi^*}A_n.

By contravariance this is (vϕ)∗=(ϕ′u)∗=u∗(ϕ′)∗(v\phi)^*=(\phi'u)^*=u^*(\phi')^*, which is exactly what we obtain by applying μ(A)n′\mu(A)_{n'} and then u∗u^*. Thus μ(A)\mu(A) is simplicial.

If f ⁣:A→Bf\colon A\to B is a simplicial map, then fnϕ∗=ϕ∗fkf_n\phi^*=\phi^*f_k, and fkf_k carries NAkNA_k to NBkNB_k. This is precisely naturality of μ\mu on each summand. □

Proving Theorem 16 now consists of showing that μ(A)n\mu(A)_n is bijective in every degree. Surjectivity is easier. By Theorem 9, we have An=NAn⊕DAnA_n=NA_n\oplus DA_n. The identity summand of K(NA∗)nK(NA_*)_n maps onto NAnNA_n. Inductively, every element of An−1A_{n-1} is in the image of μ(A)n−1\mu(A)_{n-1}, and compatibility with degeneracies then puts every si(An−1)s_i(A_{n-1}) in the image of μ(A)n\mu(A)_n. These subgroups generate DAnDA_n, proving surjectivity. The induction starts in degree zero, where μ(A)0\mu(A)_0 is the identity.

For injectivity, the same exact sequences used in normalization allow us to keep track of all possible relations among the degenerate summands.

Lemma 18 (Normalization detects injectivity). Let f ⁣:B→Af\colon B\to A be a simplicial homomorphism. If Nf ⁣:NB∗→NA∗Nf\colon NB_*\to NA_* is injective in every degree, then ff is injective in every degree.

Proof. We induct on the degree nn. In degree zero the assertion is the assumption, since NB0=B0NB_0=B_0. Suppose fn−1f_{n-1} is injective. For each 0≤k<n0\leq k<n, Lemma 8 gives a commutative diagram with exact rows:

0→NkBn−1→skNk+1Bn→ϕNkBn→0 fn−1↓fn↓↓fn 0→NkAn−1→skNk+1An→ϕNkAn→0.\begin{CD} 0 @>>> N_kB_{n-1} @>{s_k}>> N_{k+1}B_n @>{\phi}>> N_kB_n @>>> 0 \\ @. @V{f_{n-1}}VV @V{f_n}VV @VV{f_n}V @. \\ 0 @>>> N_kA_{n-1} @>{s_k}>> N_{k+1}A_n @>{\phi}>> N_kA_n @>>> 0. \end{CD}

The left vertical map is injective by induction on nn. The right vertical map is injective when k=0k=0 by the hypothesis on NfNf.

If the right vertical map is injective, so is the middle one: if x∈Nk+1Bnx\in N_{k+1}B_n maps to zero, its image ϕ(x)∈NkBn\phi(x)\in N_kB_n maps to zero, and hence ϕ(x)=0\phi(x)=0. Exactness gives x=skyx=s_k y with y∈NkBn−1y\in N_kB_{n-1}. Then skfn−1(y)=fn(x)=0s_k f_{n-1}(y)=f_n(x)=0. Since sks_k and fn−1f_{n-1} are injective, y=0y=0, and hence x=0x=0.

Starting with k=0k=0 and increasing kk, we conclude that fnf_n is injective on every NkBnN_kB_n, up to NnBn=BnN_nB_n=B_n. This completes the induction. □

Proof (Proof of the Dold-Kan correspondence). Apply Theorem 15 to the chain complex NA∗NA_*. It identifies

N∗K(NA∗)≅NA∗N_*K(NA_*)\cong NA_*

by the identity summands. On these summands, μ(A)\mu(A) is simply the inclusion NAn↪AnNA_n\mono A_n. Therefore, after normalization, Nμ(A)N\mu(A) is the identity of NA∗NA_* under the displayed identification. In particular, it is injective in every degree.

By Lemma 18, μ(A)\mu(A) is levelwise injective. We have already shown that it is levelwise surjective. Thus it is an isomorphism of simplicial abelian groups. The construction is natural in AA, so

KN∗≅IdsAb.KN_*\cong\mathrm{Id}_{\sab}.

Together with N∗K≅IdCh≥0(Z)N_*K\cong\mathrm{Id}_{\ch(\ZZ)} from Theorem 15, this proves the theorem. □

The correspondence gives, in particular, the canonical decomposition

⨁ϕ ⁣:[n]↠[k]NAk→∼An,(xϕ)ϕ⟼∑ϕϕ∗(xϕ).\bigoplus_{\phi\colon[n]\epi[k]}NA_k\xrightarrow{\sim}A_n, \qquad (x_\phi)_\phi\longmapsto\sum_\phi\phi^*(x_\phi).

Thus the normalized complex retains precisely the information needed to recover the entire simplicial abelian group, including its degeneracies.

For more details and the extension beyond abelian groups, see the paper version, Extending the Dold-Kan correspondence.

Back to the blog →