betaFeedback

WorksMathematics

Abelian categories and double décalage

Abstract

We show that an additive category is abelian if and only if its functorial nerve is a double décalage of a pseudo-simplicial object in Cat. In this characterization, the canonical isomorphism between the image and coimage is encoded by a coherent pseudo-simplicial identity.

Introduction

Abelian categories [4, 6] are fundamental in modern mathematics, offering an abstract setting for exactness and thus enabling the systematic study of homology, derived functors, and derived categories. They provide a unifying framework for structures arising in diverse areas such as algebraic topology, representation theory, and algebraic geometry, as well as powerful tools and results for these fields.

For this reason, abelian categories have been extensively studied, leading to fundamental results such as the Freyd–Mitchell embedding theorem [4], which states that every small abelian category admits a fully faithful exact embedding into a category of modules over a ring. Considerable attention has also been devoted to the definition itself, resulting in several equivalent characterizations in terms of kernels and cokernels, normal monomorphisms and epimorphisms, and the relationship between images and coimages. This last characterization is our starting point:

An additive category CC is abelian if every morphism f:a→bf:a\to b in CC has a kernel and a cokernel, and the unique induced map

coim⁡f≅coker⁡(ker⁡f→κfa)→∃!ker⁡(b→πfcoker⁡f)≅im⁡f(1)\operatorname{coim} f \cong\operatorname{coker}\left(\ker f \xrightarrow{\kappa_f} a\right) \xrightarrow{\exists!} \ker\left(b \xrightarrow{\pi_f} \operatorname{coker} f\right) \cong\operatorname{im} f \tag*{(1)}

is invertible.

Our recent work [9] provided a new characterization in terms of lax and colax algebras over 22-monads, and the present work continues this line of research by introducing a characterization using (pseudo)simplicial language in Theorem 10.

Let us briefly explain the results and the content of this paper. In [8], categories with kernels were characterized simplicially by the equation N∙(C)≅dec⁡T(X)∙\mathbb{N}_{\bullet}(C)\cong\operatorname{dec}^{\mathsf{T}}(X)_{\bullet}, meaning that the (categorical) nerve of CC is the upper décalage of an almost strict pseudo-simplicial object X∙X_{\bullet}. In effect, this equation equips the nerve with a (coherent) family of new functors dn+1:Nn(C)→Nn−1(C)d_{n+1}:\mathbb{N}_{n}(C)\to\mathbb{N}_{n-1}(C), which determine the kernels via ker⁡=d2:N1(C)→N0(C)\ker=d_{2}:\mathbb{N}_{1}(C)\to\mathbb{N}_{0}(C).

Dually, cokernels are characterized by the equation N(C)≅dec⁡⊥X\mathbb{N}(C)\cong\operatorname{dec}_{\perp}X, which yields new functors d−1:Nn(C)→Nn−1(C)d_{-1}:\mathbb{N}_{n}(C)\to\mathbb{N}_{n-1}(C). Combining the two, we arrive at the double decalage construction dec⁡⊥T(−)\operatorname{dec}^{\mathsf{T}}_{\perp}(-) (upper and lower), which explains our title. As we show, the equation N(C)≅dec⁡⊥TX\mathbb{N}(C)\cong\operatorname{dec}^{\mathsf{T}}_{\perp}X equips the category CC with both kernels and cokernels, and the isomorphism in (1) is encoded precisely by the coherent (pseudo)simplicial identity

d−1d3≅d2d−1:N2(C)⟶N0(C).d_{-1}d_{3}\cong d_{2}d_{-1}:\mathbb{N}_{2}(C)\longrightarrow\mathbb{N}_{0}(C).

Our results emerged from the theory of operadic categories—they complete the observations made in [7]. Furthermore, the décalage construction and the pseudosimplicial setting align with the recent fundamental work [1] on simplicial characterization of the category of operadic categories, inspired by [5].

Plan of the paper. In Section 1, we recall supercoherent structures and introduce the upper, lower, and double decalage constructions. Section 2 describes the functorial nerve of a category and its basic properties. In Section 3, we recall the notions of kernels and cokernels and their simplicial characterizations in terms of upper and lower decalage, respectively. Finally, Section 4 establishes the main theorem, characterizing abelian categories by means of the double décalage construction.

Conventions. We denote by f⋅gf\cdot g the morphism gg followed by ff, that is, the composite f(g)f(g). We use the same convention for the composites of functors and functors pre- or postcomposed with a morphism. The symbol 1\mathbb{1}, with or without a subscript, denotes—depending on the context—the identity endomorphism, functor, or transformation.

Supercoherent structures, décalages, &c.

Here we recall Jardine’s supercoherent structures [3], their variants, and upper, lower and double décalage constructions.

Definition 1. A pseudo-simplicial category is a pseudofunctor X ⁣:Δop→Cat⁡X\colon\Delta^{\mathrm{op}} \to\operatorname{Cat}.

According to [3], a pseudo-simplicial category is generated by the following pieces of data, which constitute a coherent higher-dimensional analogue to simplicial identities.

Definition 2. A supercoherent structure X∙X_{\bullet} consists of categories XnX_n, n≥0n \ge0, functors

di ⁣:Xn→Xn−1,0≤i≤n, n≥1,d_i\colon X_n \to X_{n-1}, \quad0 \le i \le n,\ n \ge1,
sj ⁣:Xn→Xn+1,0≤j≤n,s_j\colon X_n \to X_{n+1}, \quad0 \le j \le n,

and natural isomorphisms

didj⟹dj−1diif i<j,(2)d_i d_j \Longrightarrow d_{j-1}d_i \quad\text{if } i < j, \tag*{(2)}
sisj⟹sj+1siif i≤j,and(3)s_i s_j \Longrightarrow s_{j+1}s_i \quad\text{if } i \le j,\quad\text{and} \tag*{(3)}
disj⟹{sj−1diif i<j,1if i=j, i=j+1,sjdi−1if i>j+1.(4)d_i s_j \Longrightarrow \begin{cases} s_{j-1}d_i & \text{if } i < j,\\ 1 & \text{if } i=j,\ i=j+1,\\ s_jd_{i-1} & \text{if } i > j+1. \end{cases} \tag*{(4)}

subject to coherence conditions (1.4.1)–(1.4.17) of [3].

Note that arrows go in the opposite direction as in [3]. In particular we assume the orientation of the top-bottom simplicial isomorphism as

αm ⁣:d0dn⟹dn−1d0,(5)\alpha_m\colon d_0d_n \Longrightarrow d_{n-1}d_0, \tag*{(5)}

which is a transformation of functors Xn→Xn−2X_n \to X_{n-2}, n≥2n \ge2. This convention is better suited for our purposes and, since all α\alpha’s are isomorphisms, the direction does not really matter.

To improve readability we shall use d⊤⊤=dnd_{\top\top}=d_n for the top face operators dn ⁣:Xn→Xn−1d_n\colon X_n \to X_{n-1} and s⊤⊤=sn−1s_{\top\top}=s_{n-1} for the top degeneracy operators sn−1 ⁣:Xn−1→Xns_{n-1}\colon X_{n-1} \to X_n, for any n≥1n \ge1. Similarly, we denote d⊤=dn−1d_{\top}=d_{n-1}, d⊥⊥=d0d_{\bot\bot}=d_0, d⊥=d1d_{\bot}=d_1, s⊤=sn−2s_{\top}=s_{n-2}, s⊥=s1s_{\bot}=s_1, s⊥⊥=s0s_{\bot\bot}=s_0.

The transformation αm\alpha_m of (5) is thus written as

αm ⁣:d⊥⊥d⊤⊤⟹d⊤⊤d⊥⊥.(6)\alpha_m\colon d_{\bot\bot}d_{\top\top} \Longrightarrow d_{\top\top}d_{\bot\bot}. \tag*{(6)}

For the promised characterization we single out those supercoherent structures, for which most of the simplicial transformations remain strict.

Definition 3. A supercoherent structure U∙U_{\bullet} is called upper almost strict if U0U_0 equals the terminal category 1\mathbf{1} and all the simplicial isomorphisms of (2) are identities, except the top ones

(4b)

α⊤:d⊤⊤d⊤⊤⟹d⊤⊤d⊤\alpha_{\top}:d_{\top\top}d_{\top\top}\Longrightarrow d_{\top\top}d_{\top}

which are transformations between the functors Un→Un−2U_n \to U_{n-2}, n≥2n \ge2.

A supercoherent structure L∙L_{\bullet} is called lower almost strict if L0=1L_0=\mathbf{1} and if all the simplicial isomorphisms of (2) are identities, except the bottom ones

(4c)

α⊥:d⊥⊥d⊥⟹d⊥⊥d⊥⊥\alpha_{\bot}:d_{\bot\bot}d_{\bot}\Longrightarrow d_{\bot\bot}d_{\bot\bot}

which are transformations between the functors Ln→Ln−2L_n \to L_{n-2}, n≥2n \ge2.

A supercoherent structure X∙X_{\bullet} is called almost strict if X0=X1=1X_0=X_1=\mathbf{1}, and all the simplicial isomorphisms of (2) are identities, except three families of coherent simplicial isomorphisms (4a), (4b), and (4c).

For almost strict structures some of the coherence conditions (1.4.1)–(1.4.17) of [3] become trivial and the rest simplifies to the following two groups of equalities. The group (i)–(vii) relates 2-cells between functors Xn+3→XnX_{n+3}\to X_n, for n≥0n\ge0:

(i) α⊥d⊥⊥⋅α⊥d2=d⊥⊥α⊥⋅α⊥d⊥\alpha_{\bot}d_{\bot\bot}\cdot\alpha_{\bot}d_2=d_{\bot\bot}\alpha_{\bot}\cdot\alpha_{\bot}d_{\bot},

(ii) αmd⊥⊥⋅d⊥⊥αm⋅α⊥d⊤⊤=d⊤⊤α⊥⋅αmd⊥\alpha_m d_{\bot\bot}\cdot d_{\bot\bot}\alpha_m\cdot\alpha_{\bot}d_{\top\top}=d_{\top\top}\alpha_{\bot}\cdot\alpha_m d_{\bot},

(iii) α⊤d⊥⊥⋅d⊤⊤αm⋅αmd⊤⊤=αmd⊤⋅d⊥⊥α⊤\alpha_{\top}d_{\bot\bot}\cdot d_{\top\top}\alpha_m\cdot\alpha_m d_{\top\top}=\alpha_m d_{\top}\cdot d_{\bot\bot}\alpha_{\top},

(iv) α⊤dn−2⋅α⊤d⊤⊤=α⊤d⊤⋅d⊤⊤α⊤\alpha_{\top}d_{n-2}\cdot\alpha_{\top}d_{\top\top}=\alpha_{\top}d_{\top}\cdot d_{\top\top}\alpha_{\top},

(v) diα⊥=α⊥di+2d_i\alpha_{\bot}=\alpha_{\bot}d_{i+2}, for 1≤i≤n1\le i\le n,

(vi) diαm=αmdi+1d_i\alpha_m=\alpha_m d_{i+1}, for 1≤i≤n1\le i\le n,

(vii) diα⊤=α⊤did_i\alpha_{\top}=\alpha_{\top}d_i, for 0≤i≤n0\le i\le n.

Since X0=X1=1X_0=X_1=\mathbf{1} by assumption, for n=0,1n=0,1 the above equalities hold trivially. The group (viii)–(x) below relates 2-cells between functors Xn+1→XnX_{n+1}\to X_n, for n≥1n\ge1:

(viii) α⊥s⊥=1\alpha_{\bot}s_{\bot}=\mathbf{1}, α⊥s⊥⊥=1\alpha_{\bot}s_{\bot\bot}=\mathbf{1}, α⊥sj=sj−2α⊥\alpha_{\bot}s_j=s_{j-2}\alpha_{\bot}, for 2≤j≤n+12\le j\le n+1,

(ix) αms⊤⊤=1\alpha_m s_{\top\top}=\mathbf{1}, αms⊥⊥=1\alpha_m s_{\bot\bot}=\mathbf{1}, αmsj=sj−1αm\alpha_m s_j=s_{j-1}\alpha_m, for 1≤j≤n1\le j\le n,

(x) α⊤s⊤=1\alpha_{\top}s_{\top}=\mathbf{1}, α⊤s⊤⊤=1\alpha_{\top}s_{\top\top}=\mathbf{1}, α⊤sj=sjα⊤\alpha_{\top}s_j=s_j\alpha_{\top}, for 0≤j≤n−10\le j\le n-1.

Definition 4. The upper décalage dec⁡⊤(X)∙\operatorname{dec}^{\top}(X)_{\bullet} of a supercoherent structure X∙X_{\bullet} is the supercoherent structure with

dec⁡⊤(X)n:=Xn+1, n≥0,\operatorname{dec}^{\top}(X)_n:=X_{n+1},\ n\ge0,

and with the simplicial operators di⊤,sj⊤d_i^{\top},s_j^{\top} given by the simplicial operators of di,sjd_i,s_j of X∙X_{\bullet}, as

di⊤:dec⁡⊤(X)n=Xn+1→diXn=dec⁡⊤(X)n−1, n≥1, 0≤i≤n, andd_i^{\top}:\operatorname{dec}^{\top}(X)_n=X_{n+1}\xrightarrow{d_i}X_n=\operatorname{dec}^{\top}(X)_{n-1},\ n\ge1,\ 0\le i\le n,\ \text{and}
sj⊤:dec⁡⊤(X)n=Xn+1→sjXn+2=dec⁡⊤(X)n+1, n≥0, 0≤j≤n;s_j^{\top}:\operatorname{dec}^{\top}(X)_n=X_{n+1}\xrightarrow{s_j}X_{n+2}=\operatorname{dec}^{\top}(X)_{n+1},\ n\ge0,\ 0\le j\le n;

the top operators dn+1:Xn+1→Xnd_{n+1}:X_{n+1}\to X_n and sn+1:Xn+1→Xn+2s_{n+1}:X_{n+1}\to X_{n+2} of X∙X_{\bullet} are forgotten. The coherence isomorphisms of dec⁡⊤(X)∙\operatorname{dec}^{\top}(X)_{\bullet} are inherited from X∙X_{\bullet}.

The lower décalage dec⁡⊥(X)∙\operatorname{dec}_{\bot}(X)_{\bullet} of a supercoherent structure X∙X_{\bullet} is the supercoherent structure with

dec⁡⊥(X)n:=Xn+1, n≥0,\operatorname{dec}_{\bot}(X)_n:=X_{n+1},\ n\ge0,

and with the simplicial operators di⊥,sj⊥d_i^{\perp},s_j^{\perp} given by the simplicial operators of di+1,sj+1d_{i+1},s_{j+1} of X∙X_\bullet as

di⊥:dec⁡⊥(X)n=Xn+1→di+1Xn=dec⁡⊥(X)n−1,n≥1, 0≤i≤n,d_i^{\perp}:\operatorname{dec}_{\perp}(X)_n=X_{n+1}\xrightarrow{d_{i+1}}X_n=\operatorname{dec}_{\perp}(X)_{n-1},\quad n\geq1,\ 0\leq i\leq n,
sj⊥:dec⁡⊥(X)n=Xn+1→sj+1Xn+2=dec⁡⊥(X)n+1,n≥0, 0≤j≤n;s_j^{\perp}:\operatorname{dec}_{\perp}(X)_n=X_{n+1}\xrightarrow{s_{j+1}}X_{n+2}=\operatorname{dec}_{\perp}(X)_{n+1},\quad n\geq0,\ 0\leq j\leq n;

the bottom operators d0:Xn+1→Xnd_0:X_{n+1}\to X_n and s0:Xn+1→Xn+2s_0:X_{n+1}\to X_{n+2} of X∙X_\bullet are forgotten. The coherence isomorphisms of dec⁡⊥(X)∙\operatorname{dec}_{\perp}(X)_\bullet are inherited from X∙X_\bullet.

The double decalage dec⁡⊥⊤(X)∙\operatorname{dec}_{\perp}^{\top}(X)_\bullet of a supercoherent structure X∙X_\bullet is obtained by applying both upper and lower decalage to the structure X∙X_\bullet. Note that the order does not matter:

dec⁡⊤(dec⁡⊥(X))∙=dec⁡⊥⊤(X)∙=dec⁡⊥(dec⁡⊤(X))∙.\operatorname{dec}^{\top}(\operatorname{dec}_{\perp}(X))_\bullet=\operatorname{dec}_{\perp}^{\top}(X)_\bullet=\operatorname{dec}_{\perp}(\operatorname{dec}^{\top}(X))_\bullet.

The functorial nerve of a category

In this section we introduce the functorial nerve N∙(C)N_\bullet(\mathcal{C}) and investigate transformations between functors acting on its components.

Definition 5. Let n‾\underline{n} be the ordered set {0<1<⋯<n}\{0<1<\cdots<n\} considered as a category, and let Nn(C):=[n‾→C]N_n(\mathcal{C}):=[\underline{n}\to\mathcal{C}] be the nnth piece of the categorical version of the standard simplicial nerve of a category C\mathcal{C}. The latter is the category of chains of composable arrows

[f1,…,fn]:=∙→f1∙→f2⋯∙→fn−1∙→fn∙.(7)[f_1,\ldots,f_n]:=\bullet\xrightarrow{f_1}\bullet\xrightarrow{f_2}\cdots\bullet\xrightarrow{f_{n-1}}\bullet\xrightarrow{f_n}\bullet. \tag*{(7)}

The morphisms of Nn(C)N_n(\mathcal{C}) are commutative ladders in C\mathcal{C}

∙→f1′∙→f2′⋯∙→fn−1′∙→fn′∙↓φ0↓φ1↓φn−1↓φn∙→f1′′∙→f2′′⋯∙→fn−1′′∙→fn′′∙(8)\begin{array}{ccccccccc} \bullet&\xrightarrow{f'_1}&\bullet&\xrightarrow{f'_2}&\cdots&\bullet&\xrightarrow{f'_{n-1}}&\bullet&\xrightarrow{f'_n}\bullet\\ \downarrow{\varphi_0}&&\downarrow{\varphi_1}&&&&\downarrow{\varphi_{n-1}}&&\downarrow{\varphi_n}\\ \bullet&\xrightarrow{f''_1}&\bullet&\xrightarrow{f''_2}&\cdots&\bullet&\xrightarrow{f''_{n-1}}&\bullet&\xrightarrow{f''_n}\bullet \tag*{(8)} \end{array}

Each commutative ladder of length n≥2n\geq2 is determined by two overlapping ladders of length n−1n-1; namely one without φ0\varphi_0 and one without φn\varphi_n. This Segal-type condition is expressed via the isomorphism

Φn:Nn(C)→≅Nn−1(C)×Nn−2(C)d0,dn−1Nn−1(C),n≥2,(9)\Phi_n:N_n(\mathcal{C})\xrightarrow{\cong}N_{n-1}(\mathcal{C})\mathop{\times}_{N_{n-2}(\mathcal{C})}^{d_0,d_{n-1}}N_{n-1}(\mathcal{C}),\quad n\geq2, \tag*{(9)}

where the right hand side is the pullback of d0d_0 and dn−1d_{n-1}. It will be helpful to spell out that Φn\Phi_n sends a morphism XX of Nn(C)N_n(\mathcal{C}) to the pair

Φn(X):=(dnX,d0X)\Phi_n(X):=(d_nX,d_0X)

of morphisms of Nn−1(C)N_{n-1}(\mathcal{C}). Conversely, given morphisms U,VU,V of Nn−1(C)N_{n-1}(\mathcal{C}) such that d0U=dn−1Vd_0U=d_{n-1}V, there exists a unique morphism WW of Nn(C)N_n(\mathcal{C}) such that Φ(W)=(U,V)\Phi(W)=(U,V).

If we extend the definition of the nerve by declaring N−1(C)N_{-1}(\mathcal{C}) to be the terminal category, the functor in (7a) becomes

Φ1:N1(C)⟶N0(C)×N0(C)=C×C,(10)\Phi_1:N_1(\mathcal{C})\longrightarrow N_0(\mathcal{C})\times N_0(\mathcal{C})=\mathcal{C}\times\mathcal{C}, \tag*{(10)}

sending the morphism represented by the diagram

\begin{CD} \bullet@>u>> \bullet\\ @V{\varphi_0}VV @VV{\varphi_1}V \\ \bullet@>v>> \bullet \end{CD}

to the pair (φ0,φ1)(\varphi_0,\varphi_1). Notice that while Φ1\Phi_1 is no longer an isomorphism, it is still faithful.

The following statement prepares the ground for Lemma 12, which forms the main ingredient of the proof of Theorem 10.

Lemma 6. Let c∈{0,1,2}c \in\{0,1,2\} and let F={ξn}n≥0F = \{\xi_n\}_{n\geq0} be a family whose member ξn\xi_n is a natural transformation between functors Nn+2(C)→Nn(C)\mathbb{N}_{n+2}(\mathcal{C}) \to\mathbb{N}_n(\mathcal{C}) such that

di⋅ξn+1=ξn⋅di+c,0≤i≤n+1, n≥0.(11)d_i \cdot\xi_{n+1} = \xi_n \cdot d_{i+c}, \quad0 \leq i \leq n+1,\ n \geq0. \tag*{(11)}

Then

(i) the family FF is uniquely determined by ξ0\xi_0. If ξ0\xi_0 is invertible, then all higher ξn\xi_n’s are invertible too.

(ii) Any ξ0,ξ1\xi_0,\xi_1 satisfying (8) with n=0n=0 are pieces of a unique family F={ξn}n≥0F = \{\xi_n\}_{n\geq0} satisfying (8) for all $n \geq 1.

Proof. Let us start with the first part by proving that, given n≥0n \geq0, ξn\xi_n determines ξn+1\xi_{n+1}. By (7a) and by the faithfulness of Φ1\Phi_1 in (7b), it suffices to prove that ξn\xi_n determines dn+1⋅ξn+1d_{n+1} \cdot\xi_{n+1} and d0⋅ξn+1d_0 \cdot\xi_{n+1}. However, this follows from (8), since

dn+1⋅ξn+1=ξn⋅dn+1+candd0⋅ξn+1=ξn⋅dc.d_{n+1} \cdot\xi_{n+1} = \xi_n \cdot d_{n+1+c}\quad\text{and}\quad d_0 \cdot\xi_{n+1} = \xi_n \cdot d_c.

The invertibility follows from the obvious fact that ξn+1\xi_{n+1} is invertible if and only if dn⋅ξn+1d_n \cdot\xi_{n+1} and d0⋅ξn+1d_0 \cdot\xi_{n+1} are invertible. Let us prove the second part of the lemma.

Assume that we have already constructed ξn\xi_n and ξn+1\xi_{n+1} satisfying (8) with some n≥0n \geq0. By (7a), ξn+2\xi_{n+2} will be determined by transformations ω,λ\omega,\lambda of functors Nn+4(C)→Nn+1(C)\mathbb{N}_{n+4}(\mathcal{C}) \to\mathbb{N}_{n+1}(\mathcal{C}) that fulfill

d0⋅ω=dn+1⋅λ.(12)d_0 \cdot\omega= d_{n+1} \cdot\lambda. \tag*{(12)}

Such an ξn+2\xi_{n+2} will then be determined by the equalities

ω=dn+2⋅ξn+2andλ=d0⋅ξn+2.\omega= d_{n+2} \cdot\xi_{n+2}\quad\text{and}\quad\lambda= d_0 \cdot\xi_{n+2}.

Let us take ω:=ξn+1⋅dn+2+c\omega:= \xi_{n+1} \cdot d_{n+2+c}, λ:=ξn+1⋅dc\lambda:= \xi_{n+1} \cdot d_c. Then

d0⋅ω=d0⋅ξn+1⋅dn+2+c=ξn⋅dcdn+2+c,andd_0 \cdot\omega= d_0 \cdot\xi_{n+1} \cdot d_{n+2+c} = \xi_n \cdot d_c d_{n+2+c},\quad\text{and}
dn+1⋅λ=dn+1⋅ξn+1⋅dc=ξn⋅dn+1+cdc.d_{n+1} \cdot\lambda= d_{n+1} \cdot\xi_{n+1} \cdot d_c = \xi_n \cdot d_{n+1+c} d_c.

Since dcdn+2+c=dn+1+cdcd_c d_{n+2+c} = d_{n+1+c}d_c, (9) follows. Notice that, by construction,

dj⋅ξn+2=ξn+1⋅dj+cd_j \cdot\xi_{n+2} = \xi_{n+1} \cdot d_{j+c}

for j=0j=0 and j=n+2j=n+2. To verify this equation for the remaining 0<j<n+20<j<n+2 is an easy exercise on simplicial identities. □\square

Simplicial characterization of kernels and cokernels.

In this section we recall the simplicial characterizations of categories with kernels and cokernels in terms of upper and lower décalage, respectively.

A pointed category C\mathcal{C} is a category with a chosen object 00 which is both initial and terminal. We denote the initial map of an object a∈Ca \in\mathcal{C} by !a:0→a!^{a}: 0 \to a and the terminal map by !a:a→0!_{a}: a \to0. A pointed category has kernels if every map ff of the category has a chosen kernel

ker⁡f→κfa→fb\ker f \xrightarrow{\kappa_f} a \xrightarrow{f} b

We also assume that κ!a=!a\kappa_{!a}=!^{a} and κ!a=1a\kappa_{!_{a}}=1_{a}, for every object aa. Similarly, a pointed category has cokernels if every map ff has a chosen cokernel

a→fb→πfcoker⁡f,a \xrightarrow{f} b \xrightarrow{\pi_f} \operatorname{coker} f,

assuming that π!a=!a\pi_{!^{a}}=!_{a} and π!a=1a\pi_{!_{a}}=1_{a}, for every object aa of the category. If a pointed category has both kernels and cokernels, we additionally assume π!a=κ!a=10\pi_{!^{a}}=\kappa_{!_{a}}=1_{0}.

Theorem 7 (Theorem 26 of [8]). A pointed category C\mathcal{C} has kernels if and only if there is an isomorphism N∙(C)≅dec⁡⊤(U)∙N_{\bullet}(\mathcal{C}) \cong\operatorname{dec}^{\top}(U)_{\bullet}, where U∙U_{\bullet} is an upper almost strict supercoherent structure. The kernels are given by the values of d⊤⊤:N1(C)→N0(C)d_{\top\top}:N_{1}(\mathcal{C}) \to N_{0}(\mathcal{C}).

The following definition uses the fact that each natural transformation ξ:p⇒q\xi:p \Rightarrow q between functors p,q:A→Bp,q:\mathcal{A}\to\mathcal{B} induces a natural transformation ξop:qop⇒pop\xi^{\mathrm{op}}:q^{\mathrm{op}}\Rightarrow p^{\mathrm{op}} between opposite functors between opposite categories with components

ξaop:=(ξa)op:qop(a)→pop(a),a∈A\xi^{\mathrm{op}}_{a}:=(\xi_{a})^{\mathrm{op}}:q^{\mathrm{op}}(a)\to p^{\mathrm{op}}(a),\quad a\in\mathcal{A}

where ξa:p(a)→q(a)\xi_{a}:p(a)\to q(a) are the components of ξ\xi.

Definition 8. The opposite of an upper almost strict supercoherent structure U∙U_{\bullet} is a lower almost strict supercoherent structure (Uop)∙(U^{\mathrm{op}})_{\bullet} given by

(Uop)n:=(Un)op,(U^{\mathrm{op}})_{n}:=(U_{n})^{\mathrm{op}},

the opposite category of UnU_{n}, the operators in dimension nn are

diop:=(dn−i)op:(Un)op⟶(Un−1)op,n≥1, 0≤i≤n,d^{\mathrm{op}}_{i}:=(d_{n-i})^{\mathrm{op}}:(U_{n})^{\mathrm{op}}\longrightarrow(U_{n-1})^{\mathrm{op}},\quad n\geq1,\ 0\leq i\leq n,
sjop:=(sn−j)op:(Un)op⟶(Un+1)op,n≥0, 0≤j≤n.s^{\mathrm{op}}_{j}:=(s_{n-j})^{\mathrm{op}}:(U_{n})^{\mathrm{op}}\longrightarrow(U_{n+1})^{\mathrm{op}},\quad n\geq0,\ 0\leq j\leq n.

In particular, d⊤⊤op=(d⊥⊥)opd^{\mathrm{op}}_{\top\top}=(d_{\perp\perp})^{\mathrm{op}} and d⊥⊥op=(d⊤⊤)opd^{\mathrm{op}}_{\perp\perp}=(d_{\top\top})^{\mathrm{op}}. The coherence isomorphisms for (Uop)∙(U^{\mathrm{op}})_{\bullet} are

α⊥op:=(α⊤)op:d⊥⊥opd⊥op⟶d⊥opd⊥⊥op.\alpha^{\mathrm{op}}_{\perp}:=(\alpha_{\top})^{\mathrm{op}}:d^{\mathrm{op}}_{\perp\perp}d^{\mathrm{op}}_{\perp}\longrightarrow d^{\mathrm{op}}_{\perp}d^{\mathrm{op}}_{\perp\perp}.

Dually, the opposite of a lower almost strict supercoherent structure L∙L_{\bullet} is an upper almost strict supercoherent structure (Lop)∙(L^{\mathrm{op}})_{\bullet}. With the above definition available, we formulate a version of Theorem 7 for cokernels.

Theorem 9. A pointed category C\mathcal{C} has cokernels if and only if there exists an isomorphism

N∙(C)≅dec⁡⊥(L)∙.\mathsf{N}_{\bullet}(\mathcal{C}) \cong\operatorname{dec}_{\perp}(\mathcal{L})_{\bullet}.

with some lower almost strict supercoherent structure L∙\mathcal{L}_{\bullet}. The cokernels are given by the values of d⊥⊥:N1(C)→N0(C)d_{\perp\perp}:\mathsf{N}_{1}(\mathcal{C})\to\mathsf{N}_{0}(\mathcal{C}).

Proof. A pointed category C\mathcal{C} has cokernels if and only if Cop\mathcal{C}^{\mathrm{op}} has kernels. By Theorem 7, this is equivalent to the isomorphism N∙(Cop)≅dec⁡⊤(U)∙\mathsf{N}_{\bullet}(\mathcal{C}^{\mathrm{op}})\cong\operatorname{dec}^{\top}(\mathcal{U})_{\bullet}, with U∙\mathcal{U}_{\bullet} an upper almost strict coherent structure. Then L∙:=(Uop)∙\mathcal{L}_{\bullet}:=(\mathcal{U}^{\mathrm{op}})_{\bullet} is a lower almost strict coherent structure and, clearly, (Lop)∙=U∙(\mathcal{L}^{\mathrm{op}})_{\bullet}=\mathcal{U}_{\bullet}. Taking into account the isomorphisms (N(C)op)∙≅N∙(Cop)(\mathsf{N}(\mathcal{C})^{\mathrm{op}})_{\bullet}\cong\mathsf{N}_{\bullet}(\mathcal{C}^{\mathrm{op}}) and dec⁡⊤(Lop)∙≅(dec⁡⊥(L)op)∙\operatorname{dec}^{\top}(\mathcal{L}^{\mathrm{op}})_{\bullet}\cong(\operatorname{dec}_{\perp}(\mathcal{L})^{\mathrm{op}})_{\bullet} we conclude that

(N(C)op)∙≅N∙(Cop)≅dec⁡⊤(U)∙=dec⁡⊤(Lop)∙≅(dec⁡⊥(L)op)∙,(\mathsf{N}(\mathcal{C})^{\mathrm{op}})_{\bullet}\cong\mathsf{N}_{\bullet}(\mathcal{C}^{\mathrm{op}})\cong\operatorname{dec}^{\top}(\mathcal{U})_{\bullet}=\operatorname{dec}^{\top}(\mathcal{L}^{\mathrm{op}})_{\bullet}\cong(\operatorname{dec}_{\perp}(\mathcal{L})^{\mathrm{op}})_{\bullet},

which gives the isomorphism (N(C)op)∙≅(dec⁡⊥(L)op)∙(\mathsf{N}(\mathcal{C})^{\mathrm{op}})_{\bullet}\cong(\operatorname{dec}_{\perp}(\mathcal{L})^{\mathrm{op}})_{\bullet}, equivalent to N∙(C)≅dec⁡⊥(L)∙\mathsf{N}_{\bullet}(\mathcal{C})\cong\operatorname{dec}_{\perp}(\mathcal{L})_{\bullet}. □

Simplicial characterization of abelian categories

This section contains Theorem 10, the main result of the paper. It combines the results of the previous section to obtain a simplicial characterization of abelian categories in terms of the double décalage construction.

Theorem 10. An additive category C\mathcal{C} is abelian if and only if dec⁡⊥⊤(X)∙≅N∙(C)\operatorname{dec}^{\top}_{\perp}(\mathcal{X})_{\bullet}\cong\mathsf{N}_{\bullet}(\mathcal{C}), where X∙\mathcal{X}_{\bullet} is an almost strict supercoherent structure of Definition 3.

In the proof of Theorem 10, we need two auxiliary lemmas. The first concerns the naturality of induced maps between kernels and between cokernels. Its proof is straightforward, and we leave it to the reader.

Lemma 11. Consider two adjacent commutative squares as in the diagram

c′→g′a′→f′b′h0↓h1↓h2↓c′′→g′′a′′→f′′b′′\begin{CD} c' @>{g'}>> a' @>{f'}>> b' \\ @V{h_0}VV @V{h_1}VV @V{h_2}VV \\ c'' @>{g''}>> a'' @>{f''}>> b'' \end{CD}

in which f′g′=0f'g'=0 and f′′g′′=0f''g''=0. The universal property of kernels induces morphisms

K:ker⁡f′→ker⁡f′′,G′:c′→ker⁡f′,andG′′:c′′→ker⁡f′′,K:\ker f'\to\ker f'',\quad G':c'\to\ker f',\quad\text{and}\quad G'':c''\to\ker f'',

and it holds that KG′=G′′h0KG'=G''h_0. Dually, the universal property of cokernels induces morphisms

C:coker⁡g′→coker⁡g′′,F′:coker⁡g′→b′,andF′′:coker⁡g′′→b′′,C:\operatorname{coker}g'\to\operatorname{coker}g'',\quad F':\operatorname{coker}g'\to b',\quad\text{and}\quad F'':\operatorname{coker}g''\to b'',

and it holds that F′′C=h2F′F''C=h_2F'.

The second auxiliary lemma is a corollary of Lemma 6. It refers to the families α⊥\alpha_{\perp}, αm\alpha_{m} and α⊤\alpha_{\top} introduced in (4a), (4b) and (4c) in § 2, satisfying the equalities listed on page 4 before Definition 4. In this lemma, we will denote, for n≥0n \ge0, the members of the families α⊥\alpha_{\perp}, αm\alpha_{m} and α⊤\alpha_{\top} which are transformations of functors Xn+4→Xn+2X_{n+4} \to X_{n+2} more specifically by α⊥n\alpha_{\perp}^{n}, αmn\alpha_{m}^{n} and α⊤n\alpha_{\top}^{n}, respectively. Notice that, since X0=X1=1X_{0} = X_{1} = 1, α⊥0\alpha_{\perp}^{0}, αm0\alpha_{m}^{0} and α⊤0\alpha_{\top}^{0} are the first nontrivial members of the corresponding families. We will also make the crucial assumption that dec⁡⊥⊤(X)∙≅N∙(C)\operatorname{dec}^{\top}_{\perp}(X)_{\bullet} \cong N_{\bullet}(C) for some category CC.

Lemma 12. The families α⊥\alpha_{\perp}, αm\alpha_{m} and α⊤\alpha_{\top} are uniquely determined by α⊥0\alpha_{\perp}^{0}, αm0\alpha_{m}^{0} and α⊤0\alpha_{\top}^{0}. The members of the families α⊥\alpha_{\perp}, αm\alpha_{m} and α⊤\alpha_{\top} are invertible if and only if the transformations α⊥0\alpha_{\perp}^{0}, αm0\alpha_{m}^{0} and α⊤0\alpha_{\top}^{0} are invertible.

The first two members α⊥i\alpha_{\perp}^{i}, αmi\alpha_{m}^{i} and α⊤i\alpha_{\top}^{i}, i=0,1i = 0,1, uniquely extend to the corresponding full families satisfying (v), (vi) and (vii), respectively. Moreover, these unique extensions automatically satisfy all the remaining conditions (i)–(x) if and only if these conditions are satisfied by the first two members of the families.

Proof. Condition (v) relates, for n≥0n \ge0, α⊥n\alpha_{\perp}^{n} with α⊥n+1\alpha_{\perp}^{n+1} via the equality

diα⊥n+1=α⊥ndi+2,1≤i≤n+2.d_{i}\alpha_{\perp}^{n+1} = \alpha_{\perp}^{n}d_{i+2}, \quad1 \le i \le n+2.

which, using the isomorphism dec⁡⊥⊤(X)∙≅N∙(C)\operatorname{dec}^{\top}_{\perp}(X)_{\bullet} \cong N_{\bullet}(C), can be rewritten as

diNα⊥n+1=α⊥ndi+2N,0≤i≤n+1.d_{i}^{N}\alpha_{\perp}^{n+1} = \alpha_{\perp}^{n}d_{i+2}^{N}, \quad0 \le i \le n+1.

where diN=di+1:Nn(C)→Nn−1(C)d_{i}^{N} = d_{i+1}: N_{n}(C) \to N_{n-1}(C), 0≤i≤n0 \le i \le n, are the simplicial operators of the nerve. Comparing this with the condition in (8), with ξn:=α⊥n\xi_{n} := \alpha_{\perp}^{n}, n≥0n \ge0, Lemma 6 shows that the family α⊥\alpha_{\perp} is uniquely determined by α⊥0\alpha_{\perp}^{0}, and that α⊥0\alpha_{\perp}^{0} together with α⊥1\alpha_{\perp}^{1} uniquely extend to the whole family α⊥\alpha_{\perp} satisfying (v) as claimed. The characterization of invertibility also follows from Lemma 6. The arguments for αm\alpha_{m} and α⊤\alpha_{\top} are similar.

Let us prove the last part of the lemma. Assume for instance that (i) is satisfied for the first two members of the family α⊥\alpha_{\perp}, that is

α⊥0d0⋅α⊥0d2=d0α⊥0⋅α⊥0d1.(13)\alpha_{\perp}^{0}d_{0} \mathbin{\cdot} \alpha_{\perp}^{0}d_{2} = d_{0}\alpha_{\perp}^{0} \mathbin{\cdot} \alpha_{\perp}^{0}d_{1}. \tag*{(13)}

We need to verify that

α⊥nd0⋅α⊥nd2=d0α⊥n+1⋅α⊥nd1(14)\alpha_{\perp}^{n}d_{0} \mathbin{\cdot} \alpha_{\perp}^{n}d_{2} = d_{0}\alpha_{\perp}^{n+1} \mathbin{\cdot} \alpha_{\perp}^{n}d_{1} \tag*{(14)}

holds for all n≥1n \ge1. Since (in_{n}) is an equality of transformations between functors Xn+5→Xn+2X_{n+5} \to X_{n+2}, that is between functors Nn+3(C)→Nn(C)N_{n+3}(C) \to N_{n}(C), it is enough, by (7a) if n≥2n \ge2 resp. (7b) if n=1n = 1, to compare the two sides of (in_{n}) after applying d0N=d1d_{0}^{N} = d_{1} and dnN=dn+1d_{n}^{N} = d_{n+1}. The application of d1d_{1} to the left hand side gives

d1(α⊥nd0⋅α⊥nd2)=d1α⊥nd0⋅d1α⊥nd2=α⊥n−1d3d0⋅α⊥n−1d3d2=α⊥n−1d0d4⋅α⊥n−1d2d4=(α⊥n−1d0⋅α⊥n−1d2)d4\begin{aligned} d_{1}(\alpha_{\perp}^{n}d_{0} \mathbin{\cdot} \alpha_{\perp}^{n}d_{2}) &= d_{1}\alpha_{\perp}^{n}d_{0} \mathbin{\cdot} d_{1}\alpha_{\perp}^{n}d_{2} = \alpha_{\perp}^{n-1}d_{3}d_{0} \mathbin{\cdot} \alpha_{\perp}^{n-1}d_{3}d_{2} \\ &= \alpha_{\perp}^{n-1}d_{0}d_{4} \mathbin{\cdot} \alpha_{\perp}^{n-1}d_{2}d_{4} \\ &= (\alpha_{\perp}^{n-1}d_{0} \mathbin{\cdot} \alpha_{\perp}^{n-1}d_{2})d_{4} \end{aligned}

where we used (v) and the standard simplicial identities. Analogously we calculate

d1(d0α⊥n+1⋅α⊥nd1)=d1d0α⊥n+1⋅d1α⊥nd1=(d0α⊥n⋅α⊥n−1d1)d4.d_{1}(d_{0}\alpha_{\perp}^{n+1}\cdot\alpha_{\perp}^{n}d_{1})=d_{1}d_{0}\alpha_{\perp}^{n+1}\cdot d_{1}\alpha_{\perp}^{n}d_{1}=(d_{0}\alpha_{\perp}^{n}\cdot\alpha_{\perp}^{n-1}d_{1})d_{4}.

Similarly, we verify

dn+1(α⊥nd0⋅α⊥nd2)=(α⊥n−1d0⋅α⊥n−1d2)dn+4,dn+1(d0α⊥n+1⋅α⊥nd1)=(d0α⊥n⋅α⊥n−1d1)dn+4,\begin{aligned} d_{n+1}(\alpha_{\perp}^{n}d_{0}\cdot\alpha_{\perp}^{n}d_{2})&=(\alpha_{\perp}^{n-1}d_{0}\cdot\alpha_{\perp}^{n-1}d_{2})d_{n+4},\\ d_{n+1}(d_{0}\alpha_{\perp}^{n+1}\cdot\alpha_{\perp}^{n}d_{1})&=(d_{0}\alpha_{\perp}^{n}\cdot\alpha_{\perp}^{n-1}d_{1})d_{n+4}, \end{aligned}

We see that, for n≥1n\ge1, (in−1_{n-1}) implies (in_n), so (in_n) holds inductively for all n≥0n\ge0, as claimed. The remaining items of the list can be discussed similarly. □

Proof of Theorem 10. Assume that C is abelian, in particular, that it has kernels and cokernels. Then by Theorems 7 and 9 there is an upper almost strict supercoherent structure U∙U_{\bullet} and a lower almost strict supercoherent structure L∙L_{\bullet} with dec⁡⊤(U)∙≅N∙(C)\operatorname{dec}^{\top}(U)_{\bullet}\cong N_{\bullet}(C) and dec⁡⊥(L)∙≅N∙(C)\operatorname{dec}_{\perp}(L)_{\bullet}\cong N_{\bullet}(C), such that Un+1≅Nn(C)≅Ln+1U_{n+1}\cong N_n(C)\cong L_{n+1}, n≥0n\ge0. Moreover, the simplicial operators of U∙U_{\bullet} and L∙L_{\bullet} that come from the nerve agree, and U0=L0=1U_0=L_0=1 by assumption.

We construct the required almost strict supercoherent structure X∙X_{\bullet} as follows. We define

X0=X1:=1,X_0=X_1:=1,

which already determines the simplicial operators

s0:X0→X1,d0,d1:X1→X0,andd0,d1,d2:X2→X1,s_0:X_0\to X_1,\quad d_0,d_1:X_1\to X_0,\quad\text{and}\quad d_0,d_1,d_2:X_2\to X_1,

as well as the coherence isomorphisms that involve them. The operators s0,s1:X1→X2s_0,s_1:X_1\to X_2 send the unique object of X1=1X_1=1 to the null object 0∈X2=C0\in X_2=C. For n≥2n\ge2 we put Xn:=Nn−2(C)X_n:=N_{n-2}(C), with the top and bottom simplicial operators d⊤⊤,s⊤⊤d_{\top\top},s_{\top\top} and d⊥⊥,s⊥⊥d_{\perp\perp},s_{\perp\perp} transferred from U∙U_{\bullet} and L∙L_{\bullet}, along with the accompanying coherent simplicial isomorphisms α⊤,α⊥\alpha_{\top},\alpha_{\perp}.

The last missing pieces are the mixed isomorphism αm:d⊥⊥d⊤⊤⇒d⊤⊤d⊥⊥\alpha_m:d_{\perp\perp}d_{\top\top}\Rightarrow d_{\top\top}d_{\perp\perp}. Their construction is divided into three steps. First, we construct the transformations αm\alpha_m between functors Xn→Xn−2X_n\to X_{n-2} for n=4,5n=4,5 and then we extend them to all n≥4n\ge4 using Lemma 6. In the second step we show that αm\alpha_m’s are invertible, and in the last step we verify the coherences (i)–(x) listed before Definition 4.

Step 1: Construction of αm\alpha_m. The first nontrivial αm\alpha_m is a transformation between functors X4→X2X_4\to X_2. For x=[f,g]∈N2(C)x=[f,g]\in N_2(C), the component αm(x)\alpha_m(x) is the unique morphism in X2=N0(C)=CX_2=N_0(C)=C in the top tight corner in the diagram

ker⁡gf→d⊤⊤xker⁡g→coker⁡(d⊥⊥x)κgf↓κg↓ ∃!αm(x)↑a→fb→πfcoker⁡fd⊥⊥x↑ g↓ d⊥⊥x↓ c→πgfcoker⁡gf (15)\begin{CD} \ker gf @>{d_{\top\top}x}>> \ker g @>>> \operatorname{coker}(d_{\perp\perp}x)\\ @V{\kappa_{gf}}VV @V{\kappa_g}VV @. @A{\exists!\alpha_m(x)}AA\\ a @>{f}>> b @>{\pi_f}>> \operatorname{coker}f @A{d_{\perp\perp}x}AA\\ @. @V{g}VV @. @V{d_{\perp\perp}x}VV\\ @. c @>{\pi_{gf}}>> \operatorname{coker}gf @. \tag*{(15)} \end{CD}

in which coker⁡(d⊤⊤x)=d⊥⊥d⊤⊤x\operatorname{coker}(d_{\top\top}x)=d_{\perp\perp}d_{\top\top}x, ker⁡(d⊥⊥x)=d⊤⊤d⊥⊥x\operatorname{ker}(d_{\perp\perp}x)=d_{\top\top}d_{\perp\perp}x and the dashed arrows are induced by universal properties of kernels and cokernels. Let us prove the naturality of αm\alpha_m. Namely, given a morphism x′→x′′x'\to x'' from x′=[f′,g′]x'=[f',g'] to x′′=[f′′,g′′]x''=[f'',g''], we must verify the commutativity of the diagram

coker⁡(d⊤⊤x′)→αm(x′)ker⁡(d⊥⊥x′)↓↓coker⁡(d⊤⊤x′′)→αm(x′′)ker⁡(d⊥⊥x′′)(16)\begin{CD} \operatorname{coker}(d_{\top\top}x') @>{\alpha_m(x')}>> \operatorname{ker}(d_{\perp\perp}x')\\ @VVV @VVV\\ \operatorname{coker}(d_{\top\top}x'') @>{\alpha_m(x'')}>> \operatorname{ker}(d_{\perp\perp}x'') \tag*{(16)} \end{CD}

in which the vertical arrows are the induced natural morphisms. To this end, we notice the commutativity of the square

ker⁡g′→φ′ker⁡(d⊥⊥x′)↓↓ker⁡g′′→φ′′ker⁡(d⊥⊥x′′)\begin{CD} \operatorname{ker}g' @>{\varphi'}>> \operatorname{ker}(d_{\perp\perp}x')\\ @VVV @VVV\\ \operatorname{ker}g'' @>{\varphi''}>> \operatorname{ker}(d_{\perp\perp}x'') \end{CD}

which follows from the kernel version of Lemma 11 applied to

ker⁡g′→πf′κg′coker⁡f′→d⊥⊥x′coker⁡g′f′↓↓↓ker⁡g′′→πf′′κg′′coker⁡f′′→d⊥⊥x′′coker⁡g′′f′′.\begin{CD} \operatorname{ker}g' @>{\pi_{f'}\kappa_{g'}}>> \operatorname{coker}f' @>{d_{\perp\perp}x'}>> \operatorname{coker}g'f'\\ @VVV @VVV @VVV\\ \operatorname{ker}g'' @>{\pi_{f''}\kappa_{g''}}>> \operatorname{coker}f'' @>{d_{\perp\perp}x''}>> \operatorname{coker}g''f''. \end{CD}

The commutativity of (11) in turn follows from the cokernel version of Lemma 11 applied to

ker⁡g′f′→d⊤⊤x′ker⁡g′→φ′ker⁡(d⊥⊥x′)↓↓↓ker⁡g′′f′′→d⊤⊤x′′ker⁡g′′→φ′′ker⁡(d⊥⊥x′′).\begin{CD} \operatorname{ker}g'f' @>{d_{\top\top}x'}>> \operatorname{ker}g' @>{\varphi'}>> \operatorname{ker}(d_{\perp\perp}x')\\ @VVV @VVV @VVV\\ \operatorname{ker}g''f'' @>{d_{\top\top}x''}>> \operatorname{ker}g'' @>{\varphi''}>> \operatorname{ker}(d_{\perp\perp}x''). \end{CD}

For n=5n=5, let y∈X5=N3(C)y\in X_5=\mathcal{N}_3(\mathcal{C}), y=A→fB→gC→hDy=A\xrightarrow{f}B\xrightarrow{g}C\xrightarrow{h}D. Equality (vi) before Definition 4 gives d1αm(y)=αm(d2y)d_1\alpha_m(y)=\alpha_m(d_2y) and d2αm(y)=αm(d3y)d_2\alpha_m(y)=\alpha_m(d_3y), so αm(y)\alpha_m(y) is represented by the diagram

Q→d0d5(y)Rαm(d2y)↓↓αm(d3y)O→d4d0(y)P(17)\begin{CD} Q @>{d_0d_5(y)}>> R\\ @V{\alpha_m(d_2y)}VV @VV{\alpha_m(d_3y)}V\\ O @>{d_4d_0(y)}>> P \tag*{(17)} \end{CD}

with some objects O,P,Q,R∈CO,P,Q,R\in\mathcal{C}. Notice that the operators d2d_2 and d3d_3 of XX are the shifted operators d1Nd^{\mathcal{N}}_1 and d2Nd^{\mathcal{N}}_2 of the nerve. To show that the square in (17) commutes, we draw in Figure 1 the diagram describing the construction of αm(y)\alpha_m(y). In that diagram, E→F→G=d⊥⊥yE\to F\to G=d_{\perp\perp}y, and H→I→J=d⊤⊤yH\to I\to J=d_{\top\top}y. Applying the kernel version of Lemma 11 to the diagram

Construction of $\alpha_m(y)$

Figure 1. Construction of αm(y)\alpha_m(y).

I→κhgB→πfE→d0d3yG↓↓↓∥J→κhC→πgfF→d0d2yG\begin{CD} I @>{\kappa_{hg}}>> B @>{\pi_f}>> E @>{d_0d_3y}>> G\\ @VVV @VVV @VVV @||\\ J @>{\kappa_h}>> C @>{\pi_{gf}}>> F @>{d_0d_2y}>> G \end{CD}

in which I→JI \to J is the induced morphism between kernels, and E→FE \to F is the induced morphism between cokernels, we verify that the square (IJOP)(I JOP) commutes. Finally, the cokernel version of Lemma 11 applied to

H→d5d3yI→φ′O∥↓↓H→d5d2yJ→φ′′P\begin{CD} H @>{d_5d_3y}>> I @>{\varphi'}>> O \\ @| @VVV @VVV \\ H @>{d_5d_2y}>> J @>{\varphi''}>> P \end{CD}

gives the required commutativity of (12). The naturality of αm\alpha_m on X5X_5 follows from the fact that it is determined by αm(d2)\alpha_m(d_2) and αm(d3)\alpha_m(d_3), cf. (12), and that αm\alpha_m is natural on X4X_4 as we already established.

We have constructed the two initial terms of the family αm\alpha_m. Lemma 12 tells us that they extend uniquely to the full family satisfying (vi). It remains to prove that the families α⊥\alpha_\perp, αm\alpha_m and α⊤\alpha_\top together satisfy all conditions (i)–(x). By Lemma 12 again, it will suffice to check that they are satisfied by the first two members of the families. This will be done in Step 3 below.

Step 2: Invertibility of αm\alpha_m. By Lemma 12, it suffices to show that αm(x)\alpha_m(x) is an invertible morphism for any x=[h,g]∈N2(C)x=[h,g]\in N_2(\mathcal{C}). If C\mathcal{C} is an abelian category of modules over a ring RR, this was already observed in [7 Proposition 26]. The morphisms hZh_Z, gSg^S, and ϱ\varrho in diagram (25) of [7] are precisely morphisms d⊤⊤(x)d_{\top\top}(x), d⊥⊥(x)d_{\perp\perp}(x), and the component αm(x)\alpha_m(x), respectively.

If C\mathcal{C} is arbitrary abelian, we consider its small abelian subcategory generated by the diagram S→hT→gZS \xrightarrow{h} T \xrightarrow{g} Z. Freyd–Mitchell’s embedding theorem [4 Theorem 4.44 & page 104] gives an exact fully faithful embedding of this subcategory into the category of modules, proving invertibility of ϱ\varrho in the general case.

Step 3: Coherence. We need to verify equalities (ii), (iii) and (ix) only. Indeed, equalities (i), (v) and (viii) hold because LL is lower almost strict, while (iv), (vii) and (x) hold because UU is upper almost strict. Moreover, since the higher αm\alpha_m’s are inductively constructed from the lower ones, equality (vi) follows from the second part of Lemma 12.

Condition (ii) is dual to (iii), so it suffices to prove one of them, say (iii). By Lemma 12 again, it is enough to verify (iii) for the first two members of the families, where it is an equality of transformations of functors X5→X2X_5 \to X_2, that is, of functors N3(C)→N0(C)\mathbb{N}_3(\mathcal{C}) \to\mathbb{N}_0(\mathcal{C}). Let therefore

y=a→fb→gc→hd∈N3(C).y = a \xrightarrow{f} b \xrightarrow{g} c \xrightarrow{h} d \in\mathbb{N}_3(\mathcal{C}).

Equality (iii) evaluated at yy appears as the equality

S→T→U=S→W→V→U(18)S \to T \to U = S \to W \to V \to U \tag*{(18)}

of the composites of morphisms of C\mathcal{C} in the diagram in Figure 2. The arrow N→SN \to S is a cokernel map, hence an epimorphism, and the arrow U→EU \to E is a kernel map, hence a monomorphism. Therefore, (13) is equivalent to

Diagram of the coherence

Figure 2. The coherence.

N→S→T→U→E=N→S→W→V→U→E,N \to S \to T \to U \to E = N \to S \to W \to V \to U \to E,

which can be verified by the standard diagram chase. Notice that all other regions of the diagram commute. Thus (iii), and hence also (ii), have been verified.

By Lemma 12, it is enough to prove the equalities in (ix) only when they compare transformations of functors X3→X2X_3 \to X_2, that is, functors N1(C)→N0(C)\mathbb{N}_1(\mathcal{C}) \to\mathbb{N}_0(\mathcal{C}). Let us evaluate, for example, αms⊥⊥=1\alpha_m s_{\perp\perp}=1 at f:a→b∈N1(C)f:a\to b\in\mathbb{N}_1(\mathrm{C}). Diagram (10) defining αmx\alpha_m x for x=s⊥⊥f=[!a,f]x=s_{\perp\perp}f=[!^a,f] becomes

0→d⊤⊤s⊥⊥fker⁡f=ker⁡f→αms⊥⊥fker⁡f↓κf↓κf↓0→!aa=a!b↓f↓f↓b=b\begin{CD} 0 @>{d_{\top\top}s_{\perp\perp}f}>> \ker f @= \ker f @>{\alpha_m s_{\perp\perp}f}>> \ker f\\ @VVV @V{\kappa_f}VV @V{\kappa_f}VV\\ 0 @>{!^a}>> a @= a\\ @V{!^b}VV @V{f}VV @V{f}VV\\ b @= b \end{CD}

so clearly αms⊥⊥f=1\alpha_m s_{\perp\perp}f=1 as required. The remaining equalities in (ix) can be verified analogously.

This concludes the proof that, for an abelian category C\mathrm{C}, N∙(C)≅dec⁡⊥⊤(X)∙\mathbb{N}_{\bullet}(\mathrm{C})\cong\operatorname{dec}^{\top}_{\perp}(\mathrm{X})_{\bullet} for some almost strict supercoherent structure X∙\mathrm{X}_{\bullet}.

Opposite implication. Assume that dec⁡⊥⊤(X)∙≅N∙(C)\operatorname{dec}^{\top}_{\perp}(\mathrm{X})_{\bullet}\cong\mathbb{N}_{\bullet}(\mathrm{C}) for an almost strict supercoherent structure X∙\mathrm{X}_{\bullet}. Since then N∙(C)≅dec⁡⊥⊤(dec⁡⊥(X))∙\mathbb{N}_{\bullet}(\mathrm{C})\cong\operatorname{dec}^{\top}_{\perp}(\operatorname{dec}_{\perp}(\mathrm{X}))_{\bullet}, Theorem 7 implies that the category C\mathrm{C} has kernels. Dually, Theorem 9 implies the existence of cokernels. It remains to establish the isomorphism (1). To this end, we use the factorization

ιf:=a→1a⊕fa⊕b→0+1bb(19)\iota_f:=a\xrightarrow{1_a\oplus f}a\oplus b\xrightarrow{0+1_b}b \tag*{(19)}

regarded as an object of X4=N2(C)\mathrm{X}_4=\mathbb{N}_2(\mathrm{C}), and the associated diagram (10) with x=ιfx=\iota_f. As in the proof of [9 Theorem 24], we observe that this diagram holds for an arbitrary choice of concrete (co)kernels in C\mathrm{C}, which leads to

ker⁡f→κfa→coker⁡(κf)κf↓1a⊕0↓αm(ιf)↓a→1a⊕fa⊕b→f−1bbf↓0+1b↓−πf↓b→πfcoker⁡f\begin{CD} \ker f @>{\kappa_f}>> a @>>> \operatorname{coker}(\kappa_f)\\ @V{\kappa_f}VV @V{1_a\oplus0}VV @V{\alpha_m(\iota_f)}VV\\ a @>{1_a\oplus f}>> a\oplus b @>{f-1_b}>> b\\ @V{f}VV @V{0+1_b}VV @V{-\pi_f}VV\\ b @>{\pi_f}>> \operatorname{coker}f \end{CD}

in which the isomorphism αm(ιf)\alpha_m(\iota_f) comes from the supercoherent structure. This concludes the proof of Theorem 10. □\square

The additivity of the category C\mathrm{C} was used in the second part of the proof of Theorem 10 merely to construct the factorization in (14). Factorizations of the same type exist under weaker assumptions, formulated in

Definition 13. A category with kernels and cokernels has a KCKC-factorization if every morphism f:a→bf:a\to b can be written as f=emf=em with a monomorphism mm and an epimorphism ee, such that there are isomorphisms ξ:a→≅ker⁡(e)\xi:a\xrightarrow{\cong}\ker(e) and ζ:coker⁡(m)→≅b\zeta:\operatorname{coker}(m)\xrightarrow{\cong}b satisfying mκf=κeξκfm\kappa_f=\kappa_e\xi\kappa_f and πfe=πfζπm\pi_f e=\pi_f\zeta\pi_m.

Factorization (14) is clearly of the type required by the above definition. Recall from [2 S. 1.1.] that a Puppe-exact category is a pointed category with kernels and cokernels such that the canonical map coker⁡(κf)→ker⁡(πf)\operatorname{coker}(\kappa_f) \to\operatorname{ker}(\pi_f) is an isomorphism for every morphism ff. Our theory provides the following characterization.

Theorem 14. A category C\mathcal{C} with a KCKC-factorization is Puppe-exact if and only if there is an isomorphism dec⁡⊥⊤(X)∙≅N∙(C)\operatorname{dec}^{\top}_{\perp}(X)_\bullet\cong N_\bullet(\mathcal{C}) for some almost strict supercoherent structure X∙X_\bullet.

References

Email address: [email protected]; [email protected]

INSTITUTE OF MATHEMATICS, THE CZECH ACADEMY OF SCIENCES, ŽITNA 25, 110 00 PRAHA, CZECH REPUBLIC

References

  1. [1]M. Batanin, J. Kock, and M. Weber. Operadic categories as (pseudo)-simplicial groupoids. Preprint arXiv:2606.15671, 2026.arxiv.org/abs/2606.15671
  2. [2]F. Borceux and M. Grandis. Jordan-Hölder, modularity and distributivity in non-commutative algebra. Journal of Pure and Applied Algebra, 208:665–689, 2007.DOI
  3. [3]J.F. Jardine. Supercoherence. J. Pure Appl. Algebra, 75, 103–194, 1991.DOI
  4. [4]P. Freyd. Abelian Categories. Harper & Row, New York, 1964.
  5. [5]R. Garner, J. Kock and M. Weber. Operadic categories and décalage. Adv. Math., 377:107440, 2021.DOI
  6. [6]S. Mac Lane. Categories for the Working Mathematician. Springer-Verlag, New York, 1971. Graduate Texts in Mathematics, Vol. 5.
  7. [7]M. Markl. Bivariant operadic categories. Preprint arXiv:2402.12963, 2024.arxiv.org/abs/2402.12963
  8. [8]M. Markl and D. Trnka. Kernels, lax algebras, décalage, and supercoherence. Preprint arXiv:2601.20322, 2026.arxiv.org/abs/2601.20322
  9. [9]M. Markl and D. Trnka. Lax distributivity and a characterization of abelian categories. Preprint arXiv:2607.25740, 2026.arxiv.org/abs/2607.25740

Paper details

vlnc:W_9zvgwr4y
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents