betaFeedback

WorksMathematics

Auslander-Reiten relations and local finiteness in extriangulated categories

Abstract

Let C\mathcal C be a Hom-finite Krull–Schmidt extriangulated category over a commutative artin ring. Write Rel(C){\rm Rel}(\mathcal C) for the subgroup of the split Grothendieck group generated by the relations of all E\mathbb E-triangles, and AR(C){\rm AR}(\mathcal C) for the subgroup generated by those coming from Auslander–Reiten E\mathbb E-triangles. We first prove that either one-sided finiteness condition on Hom supports implies Rel(C)=AR(C){\rm Rel}(\mathcal C)={\rm AR}(\mathcal C). For incoming Hom supports, the proof uses a Jordan–Hölder filtration of the contravariant defect. The outgoing case follows by duality. This gives a new proof of the implication from local finiteness to generation by Auslander–Reiten relations. Our argument uses defects and induction on their composition length. In this way, it avoids the radical filtration argument used by Zhu and Zhuang. We then sssume that C\mathcal C has enough projectives and injectives, and prove that C\mathcal C is locally finite if and only if its projective and injective subcategories are locally finite, the Euler functions associated with Auslander–Reiten relations have finite support, stable and costable morphisms detect ordinary morphisms away from the projective and injective boundary, and Rel(C)=AR(C){\rm Rel}(\mathcal C)={\rm AR}(\mathcal C). Our results recover the classical module and triangulated cases. They also explain why the two one-sided support conditions are equivalent in the triangulated setting, but may differ in a general extriangulated category.

Introduction

Let RR be a commutative artin ring and let Λ\Lambda be an Artin RR-algebra. Butler [4] proved that finite representation type implies that the defining relations of K0(mod⁡Λ)K_0(\operatorname{mod}\Lambda) are generated by almost split sequences. Auslander [1] later proved the converse. A similar question arises for triangulated categories. Xiao and Zhu [XZ, Theorem 2.1] proved that local finiteness implies generation by Auslander–Reiten triangles. Over a field, the converse was proved by Zhou [Zh, Theorem 3.14 and Remark 3.15]. Haugland [9] obtained related criteria involving a generator or a cogenerator.

Extriangulated categories were introduced by Nakaoka and Palu [NP]. They include exact categories, triangulated categories, and extension-closed subcategories of triangulated categories. Auslander–Reiten theory for extriangulated categories was later developed by Iyama, Nakaoka and Palu [10]. For an extriangulated category C\mathcal C, Rel⁡(C)\operatorname{Rel}(\mathcal C) denotes the subgroup of K0sp(C)K_0^{\mathrm{sp}}(\mathcal C) generated by all E\mathbb{E}-triangle relations, and AR⁡(C)\operatorname{AR}(\mathcal C) denotes the subgroup generated by Auslander–Reiten E\mathbb{E}-triangle relations. Over a field, Zhu and Zhuang [ZZ, Theorems 3.12 and 4.4] proved that every locally finite extriangulated category has Auslander–Reiten E\mathbb{E}-triangles and that Rel⁡(C)=AR⁡(C)\operatorname{Rel}(\mathcal C)=\operatorname{AR}(\mathcal C). We give a new proof of these results using defects from one-sided finiteness condition on Hom supports. Defects in extriangulated categories have been studied by Ogawa [Oga] and Enomoto [6]. Our argument is closely related to the functorial method used by Enomoto [5] for exact categories. The main difference from the proof of Zhu and Zhuang is the induction process. Their proof passes to morphisms in higher powers of the radical and stops when a suitable radical power vanishes. Our proof instead removes one simple composition factor of the contravariant defect at each step. Thus the induction terminates when the composition length of the defect is exhausted. Although Zhu and Zhuang [ZZ, Theorem 4.4] assume two-sided local finiteness, the contravariant part of their argument already suggests a one-sided phenomenon. Our proof makes this point explicit. It also works over a commutative artin ring and requires no separate finiteness assumption on the bifunctor E\mathbb{E}.

The converse problem is to determine when Rel⁡(C)=AR⁡(C)\operatorname{Rel}(\mathcal C)=\operatorname{AR}(\mathcal C) implies that C\mathcal C is locally finite. The converse fails without further assumptions, see [11 Remark 2] and [ZZ, Example 4.5(4)]. To obtain a converse, two difficulties must be taken into account. For an Auslander–Reiten E\mathbb{E}-triangle, its Euler relation may contain extra contributions when an inflation is not a monomorphism or a deflation is not an epimorphism. The finite Euler support condition (EFS) (see Definition 4.2) is introduced to control this phenomenon. There exists also a second difficulty. Stable and costable morphisms do not detect morphisms that factor through projective or injective objects. The conditions (D−)(\mathrm{D}_{-}) and (D+)(\mathrm{D}_{+}) (see Definition 5.1) allow us to pass from stable and costable Hom spaces to ordinary Hom spaces.

We denote by P=Proj⁡C\mathcal P=\operatorname{Proj}\mathcal C and I=Inj⁡C\mathcal I=\operatorname{Inj}\mathcal C the full subcategories of projective and injective objects, respectively. Recall that a full additive Krull–Schmidt subcategory is locally finite if, for every indecomposable object, only finitely many indecomposable objects in the subcategory admit a nonzero morphism to or from it, up to isomorphism (see Definition 2.3 for the precise definition). It is useful to consider the two directions of local finiteness separately. The contravariant condition alone implies Rel⁡(C)=AR⁡(C)\operatorname{Rel}(\mathcal C)=\operatorname{AR}(\mathcal C) by an argument using defects. The covariant statement follows by duality. These results are proved in § 3. Together with the detection conditions, they lead to the characterization of local finiteness proved in § 5.

Theorem 1.1. (see Theorem 5.5 for details) Let C\mathcal C be a skeletally small Hom-finite Krull–

Schmidt RR-linear extriangulated category with enough projectives and enough injectives. Then the following statements are equivalent.

  1. C\mathcal C is locally finite;

  1. Both P\mathcal P and I\mathcal I are locally finite, C\mathcal C satisfies (EFS), (D−)(\mathrm D_-) and (D+)(\mathrm D_+), and Rel⁡(C)=AR⁡(C)\operatorname{Rel}(\mathcal C) = \operatorname{AR}(\mathcal C).

For the implication (1)⇒\Rightarrow(2), the one-sided defect theorem gives the equality of the relation subgroups. The other conditions follow directly from local finiteness. For the converse, syzygy and cosyzygy relations, together with Rel⁡(C)=AR⁡(C)\operatorname{Rel}(\mathcal C) = \operatorname{AR}(\mathcal C) and (EFS), show that stable and costable representable functors have finite support. The detection conditions then allow us to recover ordinary Hom spaces. The local finiteness of P\mathcal P and I\mathcal I controls the projective and injective boundary. A useful sufficient condition for the detection assumptions is that C‾\underline{\mathcal C} has a weak cogenerator and C‾\overline{\mathcal C} has a weak generator. By Lemma 5.2, we give the precise statement. The boundary assumptions are automatic when C\mathcal C has a projective generator and an injective cogenerator. In this case, P\mathcal P and I\mathcal I are of finite type by Corollary 5.8. The assumptions on the projective and injective boundaries are essential. Indeed, the category in Example 3.4 has enough projectives and injectives, satisfies (EFS) and both detection conditions, and has Rel⁡(C)=AR⁡(C)=0\operatorname{Rel}(\mathcal C) = \operatorname{AR}(\mathcal C) = 0. Nevertheless, the category is not locally finite, since its projective and injective boundaries fail to be locally finite.

The theorem recovers the classical module and triangulated cases. For mod⁡Λ\operatorname{mod}\Lambda, the semisimple module Λ/rad⁡Λ\Lambda/\operatorname{rad}\Lambda is a weak cogenerator of the stable category. The dual statement holds in the costable category. This is a special case of Enomoto [5 Proposition 4.7]. In a triangulated category, the projective and injective objects vanish, so both detection conditions are automatic. We also obtain a consequence for Frobenius extriangulated categories. Let C\mathcal C be a skeletally small Hom-finite Krull–Schmidt RR-linear Frobenius extriangulated category and assume that Rel⁡(C)=AR⁡(C)\operatorname{Rel}(\mathcal C) = \operatorname{AR}(\mathcal C). Then the same equality holds in the triangulated stable category C‾\underline{\mathcal C}, and hence C\mathcal C is locally finite. For Frobenius exact categories, Haugland [9 Lemma 3.2] proved the corresponding passage to the stable category. By Proposition 6.4, we extend this result to Frobenius extriangulated categories.

For 0-Auslander extriangulated categories whose projective and injective subcategories are of finite type, descriptions of the stable and costable quotients by module categories make the detection conditions automatic. In particular, we obtain a converse criterion for the relative extriangulated structure associated with a cluster tilting object in a Hom-finite 2-Calabi–Yau triangulated category. We also give examples beyond the exact and triangulated settings. These include a locally finite category of infinite type and a graded example whose projective and injective subcategories have infinitely many indecomposable objects.

The paper is organized as follows. In § 2, we fix the notation and recall the required facts on extriangulated categories, Grothendieck groups and local finiteness. In § 3, we use contravariant defects to prove that either one-sided local finiteness condition implies Rel⁡(C)=AR⁡(C)\operatorname{Rel}(\mathcal{C})=\operatorname{AR}(\mathcal{C}). In § 4, we introduce the Euler functions and the finite support condition. In § 5, we introduce the detection conditions and prove Theorem 1.1. In § 7, we treat the module, triangulated and Frobenius cases. We also give examples of genuinely extriangulated categories that are neither exact nor triangulated. In § 8, we establish a criterion under Serre duality.

Preliminaries

Throughout this paper, RR is a commutative artin ring and C\mathcal{C} is a skeletally small, RR-linear Krull–Schmidt category. We say that C\mathcal{C} is Hom-finite if C(X,Y)\mathcal{C}(X,Y) has finite length as an RR-module for all X,Y∈CX,Y\in\mathcal{C}, and write ℓR(−)\ell_R(-) for this length. We write ind⁡C\operatorname{ind}\mathcal{C} for a fixed set of representatives of the isomorphism classes of indecomposable objects. All subcategories are full, additive and closed under isomorphisms and direct summands. The case over a field kk is obtained by taking R=kR=k.

Extriangulated categories

We use the terminology of [NP]. An extriangulated category is a triple (C,E,s)(\mathcal{C},\mathbb{E},\mathfrak{s}) consisting of an additive category C\mathcal{C}, an additive bifunctor

E ⁣:Cop⁡×C⟶Ab⁡,\mathbb{E}\colon\mathcal{C}^{\operatorname{op}}\times\mathcal{C}\longrightarrow\operatorname{Ab},

and an additive realization s\mathfrak{s} satisfying the axioms in [NP, Definition 2.12]. A realization of δ∈E(C,A)\delta\in\mathbb{E}(C,A) will be written

A→aB→bC→δ .(1)A\xrightarrow{a}B\xrightarrow{b}C\xrightarrow{\delta}\ . \tag*{(1)}

and called an E\mathbb{E}-triangle. Associated with (1) are exact sequences

C(U,A)→C(U,a)C(U,B)→C(U,b)C(U,C)⟶E(U,A);\mathcal{C}(U,A)\xrightarrow{\mathcal{C}(U,a)}\mathcal{C}(U,B)\xrightarrow{\mathcal{C}(U,b)}\mathcal{C}(U,C)\longrightarrow\mathbb{E}(U,A);
C(C,U)→C(b,U)C(B,U)→C(a,U)C(A,U)⟶E(C,U).\mathcal{C}(C,U)\xrightarrow{\mathcal{C}(b,U)}\mathcal{C}(B,U)\xrightarrow{\mathcal{C}(a,U)}\mathcal{C}(A,U)\longrightarrow\mathbb{E}(C,U).

For morphisms f ⁣:A→A′f\colon A\to A' and g ⁣:C′→Cg\colon C'\to C, we use the standard notation

f∗δ=E(C,f)(δ)andg∗δ=E(g,A)(δ).f_*\delta=\mathbb{E}(C,f)(\delta)\quad\text{and}\quad g^*\delta=\mathbb{E}(g,A)(\delta).

An object PP is called projective if E(P,−)=0\mathbb{E}(P,-)=0, and an object II is called injective if E(−,I)=0\mathbb{E}(-,I)=0. We denote the full subcategories of projective and injective objects by P=Proj⁡C\mathcal{P}=\operatorname{Proj}\mathcal{C} and I=Inj⁡C\mathcal{I}=\operatorname{Inj}\mathcal{C}, respectively. We say that an extriangulated category C\mathcal{C} has enough projectives if, for every XX, there is an E\mathbb{E}-triangle

ΩX⟶PX⟶X⇢,\Omega X\longrightarrow P_X\longrightarrow X\overset{}{\dashrightarrow},

where PX∈PP_X\in\mathcal{P}. Dually, C\mathcal{C} has enough injectives if, for every XX, there is an E\mathbb{E}-triangle

X⟶IX⟶ΣX⇢,where IX∈I.X \longrightarrow I^X \longrightarrow\Sigma X \dashrightarrow, \qquad\text{where } I^X \in\mathcal{I}.

The objects ΩX\Omega X and ΣX\Sigma X depend on these choices. We do not require these assignments to be functorial.

Let [P][\mathcal{P}] be the ideal of morphisms which factor through a projective object. The stable category is C‾=C/[P]\underline{\mathcal{C}}=\mathcal{C}/[\mathcal{P}]. Thus

C‾(X,Y)=C(X,Y)/[P](X,Y).\underline{\mathcal{C}}(X,Y)=\mathcal{C}(X,Y)/[\mathcal{P}](X,Y).

The costable category C‾=C/[I]\overline{\mathcal{C}}=\mathcal{C}/[\mathcal{I}] is defined dually. Since an ideal quotient has the same objects as the original category, we use the same symbols for objects of C\mathcal{C} and their images in C‾\underline{\mathcal{C}} or C‾\overline{\mathcal{C}}.

Auslander–Reiten E\mathbb{E}-triangles

A morphism b:B→Cb:B\to C is right almost split if it is not a retraction and every morphism U→CU\to C which is not a retraction factors through bb. Left almost split morphisms are defined dually. An E\mathbb{E}-triangle (1) is an Auslander–Reiten E\mathbb{E}-triangle if aa is left almost split and bb is right almost split. Its end terms are indecomposable. We refer to [10] for the general Auslander–Reiten theory of extriangulated categories.

We shall use the following standard consequence of the Krull–Schmidt property. If UU is indecomposable, then rad⁡C(X,U)\operatorname{rad}_{\mathcal{C}}(X,U) consists precisely of the morphisms X→UX\to U that are not retractions. Dually, rad⁡C(U,X)\operatorname{rad}_{\mathcal{C}}(U,X) consists precisely of the morphisms U→XU\to X that are not sections.

From now on, unless otherwise stated, we assume that C:=(C,E,s)\mathcal{C}:=(\mathcal{C},\mathbb{E},\mathfrak{s}) is an extriangulated category.

The next two elementary facts allow us to work in § 3 without imposing any finiteness assumption on E\mathbb{E}.

Lemma 2.1. Let A,C∈ind⁡CA,C\in\operatorname{ind}\mathcal{C} and let 0≠δ∈E(C,A)0\ne\delta\in\mathbb{E}(C,A). Suppose that

c∗δ=0for every non-retraction c:C′→C.c^*\delta=0\quad\text{for every non-retraction }c:C'\to C.

Then every realization

A→xB→yC→δA\xrightarrow{x}B\xrightarrow{y}C\xrightarrow{\delta}

is an Auslander–Reiten E\mathbb{E}-triangle.

Proof. Since δ≠0\delta\ne0, the morphism yy is not a retraction. If c:C′→Cc:C'\to C is not a retraction, then c∗δ=0c^*\delta=0, and exactness of

C(C′,B)→C(C′,y)C(C′,C)⟶E(C′,A)\mathcal{C}(C',B)\xrightarrow{\mathcal{C}(C',y)}\mathcal{C}(C',C)\longrightarrow\mathbb{E}(C',A)

shows that cc factors through yy. Thus yy is right almost split.

It remains to prove that xx is left almost split. Let a:A→A′a: A \to A' be a non-section. Assume for a contradiction that a∗δ≠0a_*\delta\ne0, and choose a realization

A′→x′B′→y′C→a∗δ.A' \xrightarrow{x'} B' \xrightarrow{y'} C \xrightarrow{a_*\delta} .

Then y′y' is not a retraction, so the hypothesis gives (y′)∗δ=0(y')^*\delta= 0. Exactness for the original E-triangle yields a morphism s:B′→Bs: B' \to B with ys=y′ys = y'. By (ET3)op^{\mathrm{op}} there is a morphism d:A′→Ad: A' \to A such that (d,s,1C)(d,s,1_C) is a morphism of E-triangles from the last triangle to the original one. Hence

δ=d∗(a∗δ)=(da)∗δ.\delta= d_*(a_*\delta) = (da)_*\delta.

Because aa is not a section, dada is not invertible. The ring End⁡C(A)\operatorname{End}_{\mathcal{C}}(A) is local, so 1A−da1_A-da is invertible. On the other hand,

(1A−da)∗δ=δ−(da)∗δ=0,(1_A-da)_*\delta= \delta-(da)_*\delta= 0,

and applying the inverse of 1A−da1_A-da gives δ=0\delta=0, a contradiction. Thus a∗δ=0a_*\delta=0. Exactness of

C(B,A′)→C(x,A′)C(A,A′)⟶E(C,A′)\mathcal{C}(B,A') \xrightarrow{\mathcal{C}(x,A')} \mathcal{C}(A,A') \longrightarrow\mathbb{E}(C,A')

shows that every non-section a:A→A′a: A \to A' factors through xx. Since δ≠0\delta\ne0, xx is not a section. Therefore xx is left almost split, and the realization is an Auslander–Reiten E-triangle. □\square

Grothendieck groups and local finiteness

The split Grothendieck group K0sp(C)K_0^{\mathrm{sp}}(\mathcal{C}) is the free abelian group on ind⁡C\operatorname{ind}\mathcal{C}. Equivalently, it is generated by symbols [X][X], for X∈CX \in\mathcal{C}, subject to [X⊕Y]=[X]+[Y][X \oplus Y]=[X]+[Y], for X,Y∈CX,Y \in\mathcal{C}. To an E-triangle δ\delta as in (1) we associate its Euler relation

[δ]=[A]+[C]−[B]∈K0sp(C).[\delta]=[A]+[C]-[B]\in K_0^{\mathrm{sp}}(\mathcal{C}).

Let Rel⁡(C)\operatorname{Rel}(\mathcal{C}) be the subgroup generated by all such relations, and let AR⁡(C)\operatorname{AR}(\mathcal{C}) be the subgroup generated by the relations of Auslander–Reiten E-triangles. Then

K0(C)=K0sp(C)/Rel⁡(C),K_0(\mathcal{C})=K_0^{\mathrm{sp}}(\mathcal{C})/\operatorname{Rel}(\mathcal{C}),

and the condition that the relations of K0(C)K_0(\mathcal{C}) are generated by Auslander–Reiten E-triangles is

AR⁡(C)=Rel⁡(C).\operatorname{AR}(\mathcal{C})=\operatorname{Rel}(\mathcal{C}).

We shall use the following comparison construction. It is a direct consequence of the extriangulated axiom (ET4).

Lemma 2.2. Let

δ:A→aB→bC→δandε:A′→xE→yC→ε\delta:\quad A \xrightarrow{a} B \xrightarrow{b} C \xrightarrow{\delta} \qquad\text{and}\qquad \varepsilon:\quad A' \xrightarrow{x} E \xrightarrow{y} C \xrightarrow{\varepsilon}

be E-triangles. Assume that ε=α∗δ\varepsilon=\alpha^*\delta for some α ⁣:A→A′\alpha\colon A\to A'. Then there is a morphism γ ⁣:B→E\gamma\colon B\to E and an E-triangle

δ′ ⁣:A⟶A′⊕B→(x,γ)E⇢\delta'\colon\quad A\longrightarrow A'\oplus B\xrightarrow{(x,\gamma)}E\mathrel{\dashrightarrow}

such that γγ=b\gamma\gamma=b. Moreover, [δ]=[ε]+[δ′][\delta]=[\varepsilon]+[\delta'] in K0sp(C)K^{\mathrm{sp}}_0(\mathcal{C}).

Proof. Put μ=ΔC∗(δ⊕ε)∈E(C,A⊕A′)\mu=\Delta_C^*(\delta\oplus\varepsilon)\in\mathbb{E}(C,A\oplus A'), where ΔC ⁣:C→C⊕C\Delta_C\colon C\to C\oplus C is the diagonal, and choose a realization A⊕A′→jM→kC→μ⇢A\oplus A'\xrightarrow{j}M\xrightarrow{k}C\xrightarrow{\mu}\mathrel{\dashrightarrow}. Let ιA,ιA′\iota_A,\iota_{A'} and pA,pA′p_A,p_{A'} be the canonical inclusions and projections. Then

(pA)∗μ=δ,(pA′)∗μ=ε.(p_A)_*\mu=\delta,\qquad(p_{A'})_*\mu=\varepsilon.

Apply (ET4) to the split E-triangle A→ιAA⊕A′→pA′A′⇢A\xrightarrow{\iota_A}A\oplus A'\xrightarrow{p_{A'}}A'\mathrel{\dashrightarrow} and the chosen realization of μ\mu. We obtain a commutative diagram

A→ιAA⊕A′→pA′A′=↓j↓x′↓A→m1M→e1′E′ k↓y′↓ C→=C(2)\begin{CD} A @>{\iota_A}>> A\oplus A' @>{p_{A'}}>> A' \\ @V{=}VV @V{j}VV @V{x'}VV \\ A @>{m_1}>> M @>{e'_1}>> E' \\ @. @V{k}VV @V{y'}VV \\ @. C @>{=}>> C \tag*{(2)} \end{CD}

in which the first row is split, the middle row and the right column are E-triangles, and the middle column is the chosen realization of μ\mu. The extension of the right column is (pA′)∗μ=ε(p_{A'})_*\mu=\varepsilon. If θ1∈E(E′,A)\theta_1\in\mathbb{E}(E',A) denotes the extension of the middle row, the compatibility in (ET4) gives

(ιA)∗θ1=(y′)∗μ.(\iota_A)_*\theta_1=(y')^*\mu.

Applying (pA)∗(p_A)_* yields θ1=(y′)∗(pA)∗μ=(y′)∗δ\theta_1=(y')^*(p_A)_*\mu=(y')^*\delta. Since A′→xE→yCA'\xrightarrow{x}E\xrightarrow{y}C realizes the same extension as the right column, we may identify E′E' with EE. After this identification we write the middle morphism as e1 ⁣:M→Ee_1\colon M\to E. Thus

A→m1M→e1E→y∗δ⇢,e1jιA′=x,ye1=k.A\xrightarrow{m_1}M\xrightarrow{e_1}E\xrightarrow{y^*\delta}\mathrel{\dashrightarrow},\qquad e_1j\iota_{A'}=x,\quad ye_1=k.

Apply (ET4) a second time, now to the split E-triangle A′→ιA′A⊕A′→pAA⇢A'\xrightarrow{\iota_{A'}}A\oplus A'\xrightarrow{p_A}A\mathrel{\dashrightarrow} and the same realization of μ\mu. This gives

A′→ιA′A⊕A′→pAA=↓j↓a↓A′→m2M→e2B k↓b↓ C→=C(3)\begin{CD} A' @>{\iota_{A'}}>> A\oplus A' @>{p_A}>> A \\ @V{=}VV @V{j}VV @V{a}VV \\ A' @>{m_2}>> M @>{e_2}>> B \\ @. @V{k}VV @V{b}VV \\ @. C @>{=}>> C \tag*{(3)} \end{CD}

where the middle row is an E-triangle. If its extension is θ2∈E(B,A′)\theta_2 \in\mathbb{E}(B,A'), the same compatibility gives (ιA′)∗θ2=b∗μ(\iota_{A'})_*\theta_2=b^*\mu, and hence θ2=b∗(pA′)∗μ=b∗ε\theta_2=b^*(p_{A'})_*\mu=b^*\varepsilon. The two diagrams show, in particular, e1m2=xe_1m_2=x, e2m1=ae_2m_1=a, ye1=be2ye_1=be_2. Since ε=α∗δ\varepsilon=\alpha_*\delta, functoriality of E\mathbb{E} gives b∗ε=b∗(α∗δ)=α∗(b∗δ)=0b^*\varepsilon=b^*(\alpha_*\delta)=\alpha_*(b^*\delta)=0. Thus the middle row of (3) is split. Choose an isomorphism φ:M⟶∼A′⊕B\varphi:M\overset{\sim}{\longrightarrow} A'\oplus B such that

φm2=(10),e2=(0,1)φ.\varphi m_2=\begin{pmatrix}1\\0\end{pmatrix},\qquad e_2=(0,1)\varphi.

Write e1φ−1=(x,γ)e_1\varphi^{-1}=(x,\gamma) for some γ:B→E\gamma:B\to E; its first component is xx because e1m2=xe_1m_2=x. The identity ye1=be2ye_1=be_2 now gives yγ=by\gamma=b. Transporting the middle row of (2) along φ\varphi gives the required E-triangle δ′\delta'. Finally,

[ε]+[δ′]=([A′]+[C]−[E])+([A]+[E]−[A′]−[B])=[A]+[C]−[B]=[δ].[\varepsilon]+[\delta']=([A']+[C]-[E])+([A]+[E]-[A']-[B])=[A]+[C]-[B]=[\delta].

For X∈CX\in\mathcal{C}, put

Supp⁡C(−,X)={U∈ind⁡C∣C(U,X)≠0},\operatorname{Supp}\mathcal{C}(-,X)=\{U\in\operatorname{ind}\mathcal{C}\mid\mathcal{C}(U,X)\ne0\},
Supp⁡C(X,−)={U∈ind⁡C∣C(X,U)≠0}.\operatorname{Supp}\mathcal{C}(X,-)=\{U\in\operatorname{ind}\mathcal{C}\mid\mathcal{C}(X,U)\ne0\}.

Recall the notion of local finiteness from [XZ, Definition 2.1] and [ZZ, Definition 3.1]. We use the same terminology for full additive Krull–Schmidt subcategories, with the Hom spaces inherited from C\mathcal{C}.

Definition 2.3. The category C\mathcal{C} is of finite type, if ind⁡C\operatorname{ind}\mathcal{C} is a finite set. The category C\mathcal{C} is locally finite if, for every indecomposable X∈CX\in\mathcal{C}, both Supp⁡C(−,X)\operatorname{Supp}\mathcal{C}(-,X) and Supp⁡C(X,−)\operatorname{Supp}\mathcal{C}(X,-) are finite. More generally, if A\mathcal{A} is a full additive Krull–Schmidt subcategory of C\mathcal{C}, we say that A\mathcal{A} is locally finite if, for every X∈ind⁡AX\in\operatorname{ind}\mathcal{A}, both sets

{U∈ind⁡A∣C(U,X)≠0}and{U∈ind⁡A∣C(X,U)≠0}\{U\in\operatorname{ind}\mathcal{A}\mid\mathcal{C}(U,X)\ne0\}\quad\text{and}\quad\{U\in\operatorname{ind}\mathcal{A}\mid\mathcal{C}(X,U)\ne0\}

are finite. Equivalently, A\mathcal{A} is locally finite as a category with the Hom spaces inherited from C\mathcal{C}.

For later use, we separate the two sides of local finiteness. We write

(LF−):∣Supp⁡C(−,X)∣<∞for every X∈ind⁡C,(\mathrm{LF}_{-}):\quad|\operatorname{Supp}\mathcal{C}(-,X)|<\infty\quad\text{for every }X\in\operatorname{ind}\mathcal{C},
(LF+):∣Supp⁡C(X,−)∣<∞for every X∈ind⁡C.(\mathrm{LF}_{+}):\quad|\operatorname{Supp}\mathcal{C}(X,-)|<\infty\quad\text{for every }X\in\operatorname{ind}\mathcal{C}.

Thus C\mathcal{C} is locally finite if and only if both (LF−)(\mathrm{LF}_{-}) and (LF+)(\mathrm{LF}_{+}) hold.

Local finiteness is weaker than finite type. A concrete example is provided by the derived category Db(kQ)D^b(kQ) of a Dynkin quiver QQ over an algebraically closed field kk. It is locally finite, but it is not of finite type.

One-sided local finiteness and Auslander–Reiten relations

No assumption that C\mathcal{C} has enough projectives or injectives is needed in this section. We now give the proof using defects announced in the introduction. Contravariant defects in extriangulated categories have been studied by Ogawa [Oga] and Enomoto [6]. We only use their elementary presentation and prove the reduction argument directly. Zhu and Zhuang [ZZ, Theorem 4.4] reduce relations by passing to morphisms in successively higher powers of the radical. Their argument terminates when a suitable radical power vanishes. Our proof uses a composition series of the contravariant defect and removes one simple factor at each step. This argument is closely related to the functorial approach used by Enomoto [5] for exact categories.

For an E\mathbb{E}-triangle

δ:A→aB→bC⇢\delta:\quad A \xrightarrow{a} B \xrightarrow{b} C \dashrightarrow

we define its contravariant defect as the additive contravariant functor

δ~:=coker⁡(C(−,B)→C(−,b)C(−,C)).\widetilde{\delta}:=\operatorname{coker}\bigl(\mathcal{C}(-,B)\xrightarrow{\mathcal{C}(-,b)}\mathcal{C}(-,C)\bigr).

For U∈ind⁡CU\in\operatorname{ind}\mathcal{C}, set

SU:=C(−,U)/rad⁡C(−,U).S_U:=\mathcal{C}(-,U)/\operatorname{rad}\mathcal{C}(-,U).

Since C\mathcal{C} is Krull–Schmidt, the functor SUS_U is simple. We shall use the radical criterion from § 2 throughout this section.

Lemma 3.1. Assume C\mathcal{C} satisfies (LF−)(\mathrm{LF}-). Then every nonprojective object U∈ind⁡CU\in\operatorname{ind}\mathcal{C} is the end term of an Auslander–Reiten E\mathbb{E}-triangle.

Proof. Fix a nonprojective U∈ind⁡CU\in\operatorname{ind}\mathcal{C}. By (LF−)(\mathrm{LF}-), there are only finitely many indecomposable objects X1,…,XtX_1,\ldots,X_t with C(Xi,U)≠0\mathcal{C}(X_i,U)\ne0. Since C\mathcal{C} is Hom-finite, each rad⁡C(Xi,U)\operatorname{rad}_{\mathcal{C}}(X_i,U) has finite length as an RR-module and is therefore finitely generated. Choose generators fijf_{ij} and let

q:Q:=⨁i,jXi⟶Uq: Q:=\bigoplus_{i,j}X_i\longrightarrow U

be the morphism whose components are the fijf_{ij}. Let h:X→Uh:X\to U be a non-retraction and decompose XX into indecomposable summands. No component of hh is a retraction. After identifying each nonzero indecomposable summand with one of the chosen representatives XiX_i, every component lies in rad⁡C(Xi,U)\operatorname{rad}_{\mathcal{C}}(X_i,U) and is an RR-linear combination of the chosen generators. Hence hh factors through qq. Thus every non-retraction to UU factors through qq.

Since UU is nonprojective, there exists an object A0∈CA_0\in\mathcal{C} such that E(U,A0)≠0\mathbb{E}(U,A_0)\ne0. Choose a nonzero extension δ0∈E(U,A0)\delta_0\in\mathbb{E}(U,A_0) and a realization

A_0\longrightarrow B_0\xrightarrow{p}U\xdashrightarrow{\delta_0}.

Since δ0≠0\delta_0\ne0, this E\mathbb{E}-triangle is nonsplit. The morphism (p,q):B0⊕Q⟶U(p,q):B_0\oplus Q\longrightarrow U is a deflation. Indeed, it is the composite

B0⊕Q→p⊕1QU⊕Q→(1U,q)U,B_0 \oplus Q \xrightarrow{p\oplus1_Q} U \oplus Q \xrightarrow{(1_U,q)} U,

where the first morphism is a deflation and the second is a split deflation. Deflations are closed under composition by [NP, Remark 2.16]. Since the E\mathbb{E}-triangle is nonsplit, pp is not a retraction, and hence p∈rad⁡C(B0,U)p \in\operatorname{rad}_{\mathcal{C}}(B_0,U). By construction, q∈rad⁡C(Q,U)q \in\operatorname{rad}_{\mathcal{C}}(Q,U). Thus (p,q)(p,q) belongs to rad⁡C(B0⊕Q,U)\operatorname{rad}_{\mathcal{C}}(B_0\oplus Q,U) and is not a retraction. Since every non-retraction to UU factors through qq, it also factors through (p,q)(p,q). Hence (p,q)(p,q) is a right almost split deflation.

Let

K⟶B0⊕Q→(p,q)U→ηK \longrightarrow B_0\oplus Q \xrightarrow{(p,q)} U \xrightarrow{\eta} {}

be the resulting nonsplit E\mathbb{E}-triangle. Then c∗η=0c^*\eta=0 for every non-retraction c:U′→Uc:U'\to U. Write K=K1⊕⋯⊕KsK=K_1\oplus\cdots\oplus K_s with the KiK_i indecomposable and choose a component ηi∈E(U,Ki)\eta_i\in\mathbb{E}(U,K_i) which is nonzero. For every non-retraction c:U′→Uc:U'\to U one has c∗ηi=0c^*\eta_i=0, because ηi\eta_i is a component of η\eta. Since KiK_i and UU are indecomposable and ηi≠0\eta_i\ne0, it follows from Lemma 2.1 that any realization of ηi\eta_i is an Auslander–Reiten E\mathbb{E}-triangle ending at UU. □

Lemma 3.2 (Defect reduction). Assume C\mathcal{C} satisfies (LF−)(\mathrm{LF}-). Let

δ:A→aB→bC⇢\delta:\quad A \xrightarrow{a} B \xrightarrow{b} C \dashrightarrow

be a nonsplit E\mathbb{E}-triangle, and let δ~↠SU\widetilde{\delta}\twoheadrightarrow S_U be a simple quotient. Then UU is nonprojective. Let

ρU:AU→xUEU→yUU⇢\rho_U:\quad A_U \xrightarrow{x_U} E_U \xrightarrow{y_U} U \dashrightarrow

be an Auslander–Reiten E\mathbb{E}-triangle ending at UU, which exists by Lemma 3.1. Then there exists an E\mathbb{E}-triangle δ′\delta' such that

[δ]=[ρU]+[δ′][\delta]=[\rho_U]+[\delta']

and there is a short exact sequence

0⟶δ~′⟶δ~⟶SU⟶0.0\longrightarrow\widetilde{\delta}'\longrightarrow\widetilde{\delta}\longrightarrow S_U\longrightarrow0.

Proof. Let π:δ~↠SU\pi:\widetilde{\delta}\twoheadrightarrow S_U be the chosen quotient. The composite

C(−,C)⟶δ~→πSU\mathcal{C}(-,C)\longrightarrow\widetilde{\delta}\xrightarrow{\pi}S_U

is nonzero. By Yoneda, it is represented by a morphism p:C→Up:C\to U whose class in

SU(C)=C(C,U)/rad⁡C(C,U)S_U(C)=\mathcal{C}(C,U)/\operatorname{rad}_{\mathcal{C}}(C,U)

is nonzero. Thus p∉rad⁡C(C,U)p\notin\operatorname{rad}_{\mathcal{C}}(C,U). By the radical criterion from § 2, the morphism pp is a retraction. Choose a section i:U→Ci:U\to C with pi=1Upi=1_U. Since pi=1Upi=1_U, the endomorphism ipip of CC is idempotent. As CC is Krull–Schmidt, this idempotent splits. Hence C≃U⊕C′C\simeq U\oplus C' for some object C′C'. After transporting δ\delta and the quotient π\pi along this isomorphism, we may write

C=U⊕C′,p=(1U,0),b=(bUb′).C = U \oplus C', \quad p = (1_U,0), \quad b = \begin{pmatrix} b_U \\ b' \end{pmatrix}.

Let qδ~:C(−,C)↠δ~q_{\widetilde{\delta}}:\mathcal{C}(-,C)\twoheadrightarrow\widetilde{\delta} be the canonical quotient, and let θp:C(−,C)→SU\theta_p:\mathcal{C}(-,C)\to S_U be the natural transformation corresponding under Yoneda to the class [p]∈SU(C)[p]\in S_U(C). Thus θp,X(c)=[pc]\theta_{p,X}(c)=[pc] for c:X→Cc:X\to C. By the choice of pp, we have πqδ~=θp\pi q_{\widetilde{\delta}}=\theta_p. Evaluating this equality at BB on the element b∈C(B,C)b\in\mathcal{C}(B,C) gives 0=(πqδ~)B(b)=[pb]=[bU]∈SU(B)0=(\pi q_{\widetilde{\delta}})_B(b)=[pb]=[b_U]\in S_U(B). Thus bU∈rad⁡C(B,U)b_U\in\operatorname{rad}_{\mathcal{C}}(B,U). In particular, bUb_U is not a retraction. Moreover, we also claim that UU is nonprojective. Indeed, if UU were projective, exactness of C(U,B)→C(U,C)→E(U,A)\mathcal{C}(U,B)\to\mathcal{C}(U,C)\to\mathbb{E}(U,A) would imply δ~(U)=0\widetilde{\delta}(U)=0, contradicting δ~(U)↠SU(U)≠0\widetilde{\delta}(U)\twoheadrightarrow S_U(U)\ne0.

Since yUy_U is right almost split, there is βU:B→EU\beta_U:B\to E_U with bU=yUβUb_U=y_U\beta_U. Take the direct sum of ρU\rho_U with the split E-triangle on C′C' and write it as

ε:AU→x‾UE:=EU⊕C′→y:=yU⊕1C′C=U⊕C′⇢,x‾U=(xU0).\varepsilon:\quad A_U\xrightarrow{\overline{x}_U}E:=E_U\oplus C'\xrightarrow{y:=y_U\oplus1_{C'}}C=U\oplus C'\dashrightarrow,\qquad\overline{x}_U=\begin{pmatrix}x_U\\0\end{pmatrix}.

Put

β=(βUb′):B→E.\beta=\begin{pmatrix}\beta_U\\b'\end{pmatrix}:B\to E.

Then b=yβb=y\beta. By (ET3)op^{\mathrm{op}}, there is a morphism α:A→AU\alpha:A\to A_U such that (α,β,1C)(\alpha,\beta,1_C) is a morphism from δ\delta to ε\varepsilon. Thus ε=α∗δ\varepsilon=\alpha_*\delta. By Lemma 2.2, we give a morphism γ:B→E\gamma:B\to E with yγ=by\gamma=b and an E-triangle

δ′:A⟶AU⊕B→(x‾U,γ)E⇢.\delta':\quad A\longrightarrow A_U\oplus B\xrightarrow{(\overline{x}_U,\gamma)}E\dashrightarrow.

Since the split summand contributes no relation, [ε]=[ρU][\varepsilon]=[\rho_U]. It follows again from Lemma 2.2 that

[δ]=[ρU]+[δ′].[\delta]=[\rho_U]+[\delta'].

For X∈CX\in\mathcal{C}, composition with yy induces

ΦX:δ~′(X)⟶δ~(X),[h]⟼[yh].\Phi_X:\widetilde{\delta}'(X)\longrightarrow\widetilde{\delta}(X),\qquad[h]\longmapsto[yh].

It is well defined because y(x‾U,γ)(u,v)=bvy(\overline{x}_U,\gamma)(u,v)=bv. If ΦX([h])=0\Phi_X([h])=0, then yh=bv=yγvyh=bv=y\gamma v for some v:X→Bv:X\to B. Hence y(h−γv)=0y(h-\gamma v)=0. Exactness of the Hom sequence for ε\varepsilon gives h−γv=x‾Uuh-\gamma v=\overline{x}_Uu for some u:X→AUu:X\to A_U, so [h]=0[h]=0. Thus ΦX\Phi_X is injective.

The projection p:C=U⊕C′→Up:C=U\oplus C'\to U induces a natural epimorphism

Ψ:δ~⟶C(−,U)/Im⁡C(−,yU).\Psi:\widetilde{\delta}\longrightarrow\mathcal{C}(-,U)/\operatorname{Im}\mathcal{C}(-,y_U).

Indeed, at an object XX it is given by ΨX([c])=[pc]\Psi_X([c])=[pc]. It is well defined because pb=bU=yUβUpb=b_U=y_U\beta_U, and it is surjective because pp is a retraction. If [c][c] lies in its kernel, write c=(cU,c′)c=(c_U,c') and choose eU:X→EUe_U:X\to E_U with cU=yUeUc_U=y_Ue_U. Then h=(eU,c′):X→Eh=(e_U,c'):X\to E satisfies yh=cyh=c, so [c][c] lies in the image of ΦX\Phi_X. The reverse inclusion is immediate. Therefore

0⟶δ~′⟶δ~→ΨC(−,U)/Im⁡C(−,yU)⟶00 \longrightarrow\widetilde{\delta}' \longrightarrow\widetilde{\delta} \xrightarrow{\Psi} \mathcal{C}(-,U)/\operatorname{Im}\mathcal{C}(-,y_U) \longrightarrow0

is exact. Since yUy_U is right almost split, Im⁡C(−,yU)=rad⁡C(−,U)\operatorname{Im}\mathcal{C}(-,y_U)=\operatorname{rad}\mathcal{C}(-,U), so the last term is SUS_U. Under this identification, Ψ\Psi is precisely the chosen quotient π\pi. In fact, if

qδ:C(−,C)↠δ~q_\delta:\mathcal{C}(-,C)\twoheadrightarrow\widetilde{\delta}

denotes the canonical quotient, then both composites in

C(−,C)→qδδ~θp↓↓πSU=SU\begin{CD} \mathcal{C}(-,C) @>{q_\delta}>> \widetilde{\delta} \\ @V{\theta_p}VV @VV{\pi}V \\ S_U @= S_U \end{CD}

are equal to θp\theta_p. Since qδq_\delta is an epimorphism, Ψ=π\Psi=\pi, and the asserted short exact sequence follows.

Theorem 3.3. Let C\mathcal{C} be a skeletally small Hom-finite Krull–Schmidt RR-linear extriangulated category. If C\mathcal{C} satisfies (LF−)(\mathrm{LF}_{-}) or (LF+)(\mathrm{LF}_{+}), then

Rel⁡(C)=AR⁡(C).\operatorname{Rel}(\mathcal{C})=\operatorname{AR}(\mathcal{C}).

Proof. Assume that (LF−)(\mathrm{LF}_{-}) holds. The case (LF+)(\mathrm{LF}_{+}) follows by applying the same argument to the opposite extriangulated category Cop\mathcal{C}^{\mathrm{op}}. An E-triangle

A⟶B⟶C⇢A \longrightarrow B \longrightarrow C \dashrightarrow

in C\mathcal{C} becomes an E-triangle C⟶B⟶A⇢C\longrightarrow B\longrightarrow A\dashrightarrow in Cop\mathcal{C}^{\mathrm{op}}, so its relation [A]+[C]−[B][A]+[C]-[B] is unchanged, and Auslander–Reiten E-triangles correspond under opposition. Hence, under the natural identification K0sp(Cop)=K0sp(C)K_0^{\mathrm{sp}}(\mathcal{C}^{\mathrm{op}})=K_0^{\mathrm{sp}}(\mathcal{C}), the relation and Auslander–Reiten subgroups correspond to Rel⁡(C)\operatorname{Rel}(\mathcal{C}) and AR⁡(C)\operatorname{AR}(\mathcal{C}), respectively. Let

δ:A⟶B⟶C⇢\delta:\quad A\longrightarrow B\longrightarrow C\dashrightarrow

be an E-triangle. Since Supp⁡C(−,C)\operatorname{Supp}\mathcal{C}(-,C) is finite and every C(X,C)\mathcal{C}(X,C) has finite RR-length, the representable functor C(−,C)\mathcal{C}(-,C) has finite length. Indeed, for a subfunctor F⊆C(−,C)F\subseteq\mathcal{C}(-,C) set

ν(F)=∑X∈Supp⁡C(−,C)ℓRF(X).\nu(F)=\sum_{X\in\operatorname{Supp}\mathcal{C}(-,C)}\ell_R F(X).

A strict inclusion of subfunctors strictly increases ν\nu. Indeed, if F⊊GF\subsetneq G, then F(Y)⊊G(Y)F(Y)\subsetneq G(Y) for some object YY. Decompose YY into indecomposable summands. By additivity, F(X)⊊G(X)F(X)\subsetneq G(X) for at least one indecomposable summand XX. Since G(X)≠0G(X)\ne0, such an XX belongs to Supp⁡C(−,C)\operatorname{Supp}\mathcal{C}(-,C). Dually, a strict descending chain strictly decreases ν\nu. Since

::: 0≤ν(F)≤ν(C(−,C))0 \le\nu(F) \le\nu(\mathcal{C}(-, C)), neither type of chain can be infinite. Hence C(−,C)\mathcal{C}(-, C), and therefore its quotient δ~\tilde{\delta}, has finite length. Write length⁡(F)\operatorname{length}(F) for the composition length of a functor FF of finite length.

We argue by induction on length⁡(δ~)\operatorname{length}(\tilde{\delta}). If the length is zero, then δ~=0\tilde{\delta} = 0. Evaluating at CC shows that 1C1_C factors through the deflation B→CB \to C, so δ\delta is split and [δ]=0[\delta] = 0.

Suppose length⁡(δ~)>0\operatorname{length}(\tilde{\delta}) > 0. Choose a simple quotient FF of δ~\tilde{\delta}. We claim that F≅SUF \cong S_U for some U∈ind⁡CU \in\operatorname{ind}\mathcal{C}. Choose an object on which FF is nonzero and decompose it into indecomposable summands. Since FF is additive, F(U)≠0F(U) \ne0 for some indecomposable summand UU. Choose a nonzero element x∈F(U)x \in F(U). By the Yoneda lemma, xx corresponds to a natural transformation

φ:C(−,U)⟶F,φX(f)=F(f)(x)\varphi: \mathcal{C}(-,U) \longrightarrow F,\qquad\varphi_X(f) = F(f)(x)

for f:X→Uf : X \to U. Since φU(1U)=x≠0\varphi_U(1_U) = x \ne0, the natural transformation φ\varphi is nonzero. As FF is simple, its image is a nonzero subfunctor of FF, and hence Im⁡φ=F\operatorname{Im}\varphi= F. Thus φ\varphi is an epimorphism. Let GG be its kernel. If g∈G(X)⊆C(X,U)g \in G(X) \subseteq\mathcal{C}(X,U) were a retraction, choose s:U→Xs : U \to X with gs=1Ugs = 1_U. Naturality would then give 1U∈G(U)1_U \in G(U), forcing G=C(−,U)G = \mathcal{C}(-,U), a contradiction. Hence G⊆rad⁡C(−,U)G \subseteq\operatorname{rad}\mathcal{C}(-,U). Since FF is simple, GG is a maximal proper subfunctor of C(−,U)\mathcal{C}(-,U). The radical is proper because 1U∉rad⁡C(U,U)1_U \notin\operatorname{rad}\mathcal{C}(U,U), and therefore maximality forces G=rad⁡C(−,U)G = \operatorname{rad}\mathcal{C}(-,U). Thus F≅SUF \cong S_U. By Lemma 3.2, we give

[δ]=[ρU]+[δ′][\delta] = [\rho_U] + [\delta']

and

length⁡(δ~′)=length⁡(δ~)−1.\operatorname{length}(\tilde{\delta}') = \operatorname{length}(\tilde{\delta}) - 1.

By induction, [δ′]∈AR⁡(C)[\delta'] \in\operatorname{AR}(\mathcal{C}), hence [δ]∈AR⁡(C)[\delta] \in\operatorname{AR}(\mathcal{C}). Thus Rel⁡(C)⊆AR⁡(C)\operatorname{Rel}(\mathcal{C}) \subseteq\operatorname{AR}(\mathcal{C}), and the reverse inclusion is immediate. ▫

Example 3.4. The conditions (LF−)(\mathrm{LF}_{-}) and (LF+)(\mathrm{LF}_{+}) are not equivalent for general extriangulated categories, even in the presence of enough projectives and injectives. Let kk be a field and let C\mathcal{C} be the Krull–Schmidt kk-linear additive category with indecomposable objects X1,X2,…X_1, X_2, \ldots and

C(Xi,Xj)={k,i≤j,0,i>j,\mathcal{C}(X_i,X_j) = \begin{cases} k, & i \le j,\\ 0, & i > j, \end{cases}

with the evident composition. Equip C\mathcal{C} with the split extriangulated structure. Then every object is both projective and injective. For each jj,

Supp⁡C(−,Xj)={X1,…,Xj}\operatorname{Supp}\mathcal{C}(-,X_j) = \{X_1,\ldots,X_j\}

is finite, whereas

Supp⁡C(Xi,−)={Xi,Xi+1,Xi+2,…}\operatorname{Supp}\mathcal{C}(X_i,-) = \{X_i,X_{i+1},X_{i+2},\ldots\}

is infinite. Thus (LF−-) holds but (LF++) does not. Passing to the opposite category shows that the converse implication also fails.

Euler functions

For U∈CU \in\mathcal{C}, define homomorphisms

χU−,χU+:K0sp⁡(C)⟶Z\chi_U^{-},\chi_U^{+}:K_0^{\operatorname{sp}}(\mathcal{C})\longrightarrow\mathbb{Z}

by

χU−([X])=ℓR(C(U,X)),χU+([X])=ℓR(C(X,U)).\chi_U^{-}([X])=\ell_R(\mathcal{C}(U,X)),\qquad\chi_U^{+}([X])=\ell_R(\mathcal{C}(X,U)).

For r∈K0sp⁡(C)r \in K_0^{\operatorname{sp}}(\mathcal{C}), write

Supp⁡χ−(r)={U∈ind⁡C∣χU−(r)≠0},\operatorname{Supp}\chi^{-}(r)=\{U\in\operatorname{ind}\mathcal{C}\mid\chi_U^{-}(r)\ne0\},
Supp⁡χ+(r)={U∈ind⁡C∣χU+(r)≠0}.\operatorname{Supp}\chi^{+}(r)=\{U\in\operatorname{ind}\mathcal{C}\mid\chi_U^{+}(r)\ne0\}.

The following elementary identity will be used repeatedly.

Lemma 4.1. Let A→aB→bC⇝A\xrightarrow{a}B\xrightarrow{b}C\rightsquigarrow be an E-triangle and let U∈CU\in\mathcal{C}. Then

χU−([A]+[C]−[B])=ℓR(ker⁡C(U,a))+ℓR(coker⁡C(U,b)),\chi_U^{-}([A]+[C]-[B])=\ell_R(\ker\mathcal{C}(U,a))+\ell_R(\operatorname{coker}\mathcal{C}(U,b)),
χU+([A]+[C]−[B])=ℓR(ker⁡C(b,U))+ℓR(coker⁡C(a,U)).\chi_U^{+}([A]+[C]-[B])=\ell_R(\ker\mathcal{C}(b,U))+\ell_R(\operatorname{coker}\mathcal{C}(a,U)).

In particular, both integers are nonnegative.

Proof. Set V1=C(U,A)V_1=\mathcal{C}(U,A), V2=C(U,B)V_2=\mathcal{C}(U,B) and V3=C(U,C)V_3=\mathcal{C}(U,C). Exactness at V2V_2 of the first Hom sequence displayed after (1) gives

ℓR(V1)−ℓR(V2)+ℓR(V3)=ℓR(ker⁡(V1→V2))+ℓR(coker⁡(V2→V3)).\ell_R(V_1)-\ell_R(V_2)+\ell_R(V_3)=\ell_R(\ker(V_1\to V_2))+\ell_R(\operatorname{coker}(V_2\to V_3)).

This proves the first identity. The second is proved in the same way, using exactness of the second Hom sequence associated with (1).

Definition 4.2. We say that C\mathcal{C} satisfies (EFS) if, for every Auslander–Reiten E-triangle δ\delta, both Supp⁡χ−([δ])\operatorname{Supp}\chi^{-}([\delta]) and Supp⁡χ+([δ])\operatorname{Supp}\chi^{+}([\delta]) are finite.

The property is automatic for exact categories.

Proposition 4.3. Every Hom-finite Krull–Schmidt exact category has (EFS). More precisely, if

δ:0⟶A→aB→bC⟶0\delta:\quad0\longrightarrow A\xrightarrow{a}B\xrightarrow{b}C\longrightarrow0

is an Auslander–Reiten conflation, then

Supp⁡χ−([δ])={C}andSupp⁡χ+([δ])={A}.\operatorname{Supp}\chi^{-}([\delta])=\{C\}\quad\text{and}\quad\operatorname{Supp}\chi^{+}([\delta])=\{A\}.

Here the sets are understood as sets of isomorphism classes.

Proof. Since aa is a monomorphism, C(U,a)\mathcal{C}(U,a) is injective. By Lemma 4.1, we give

χU−([δ])=ℓR(coker⁡C(U,b)).\chi_{\mathcal{U}}^{-}([\delta])=\ell_R(\operatorname{coker}\mathcal{C}(U,b)).

If UU is indecomposable and U≄CU \not\simeq C, no morphism U→CU \to C is a retraction. Since bb is right almost split, every such morphism factors through bb, and the cokernel vanishes. For U=CU=C, the identity of CC does not factor through bb, since bb is not a retraction. Hence the cokernel is nonzero. The second assertion is dual, using that bb is an epimorphism and aa is left almost split.

We also use the following criterion based on Serre duality.

Proposition 4.4. Let ERE_R be an injective cogenerator of mod⁡R\operatorname{mod} R and put D=Hom⁡R(−,ER)D=\operatorname{Hom}_R(-,E_R). Assume that C\mathcal{C} admits an RR-linear Serre functor SS, that is, an RR-linear autoequivalence together with bifunctorial isomorphisms

DC(X,Y)≃C(Y,SX).D\mathcal{C}(X,Y)\simeq\mathcal{C}(Y,SX).

Recall that DD is an exact duality on RR-modules of finite length. In particular, it preserves vanishing and composition length. For an Auslander–Reiten E-triangle δ\delta: A→aB→bC⇢A \xrightarrow{a} B \xrightarrow{b} C \dashrightarrow, one has

Supp⁡χ−([δ])⊆{C,S−1A},\operatorname{Supp}\chi^{-}([\delta])\subseteq\{C,S^{-1}A\},
Supp⁡χ+([δ])⊆{A,SC}.\operatorname{Supp}\chi^{+}([\delta])\subseteq\{A,SC\}.

Consequently, C\mathcal{C} has (EFS).

Proof. By Lemma 4.1, χU−([δ])\chi_{\mathcal{U}}^{-}([\delta]) is the sum of a kernel term and a cokernel term. The latter is ℓR(coker⁡C(U,b))\ell_R(\operatorname{coker}\mathcal{C}(U,b)), and it can be nonzero only for U≃CU\simeq C, because bb is right almost split. By Serre duality, the dual of

C(U,A)→C(U,a)C(U,B)\mathcal{C}(U,A)\xrightarrow{\mathcal{C}(U,a)}\mathcal{C}(U,B)

is identified with

C(B,SU)→C(a,SU)C(A,SU).\mathcal{C}(B,SU)\xrightarrow{\mathcal{C}(a,SU)}\mathcal{C}(A,SU).

It follows that ker⁡C(U,a)\ker\mathcal{C}(U,a) is nonzero precisely when the latter map has nonzero cokernel. Since aa is left almost split, this can happen only when SU≃ASU\simeq A, or equivalently U≃S−1AU\simeq S^{-1}A. This proves the first inclusion. The second statement follows by the dual argument. The cokernel of C(a,U)\mathcal{C}(a,U) is supported at AA. After Serre duality, the kernel of C(b,U)\mathcal{C}(b,U) is supported at SCSC.

For triangulated categories one can be more explicit. If A→B→C→A[1]A \to B \to C \to A[1] is an Auslander–Reiten triangle, the familiar exact Hom sequences give

Supp⁡χ−([δ])⊆{C,C[−1]}andSupp⁡χ+([δ])⊆{A,A[1]}.(4)\operatorname{Supp}\chi^{-}([\delta]) \subseteq\{C,C[-1]\}\quad\text{and}\quad\operatorname{Supp}\chi^{+}([\delta]) \subseteq\{A,A[1]\}. \tag*{(4)}

See, for example, [9 Lemmas 2.2 and 2.3] or [Zh, Lemma 3.11]. Thus a Hom-finite Krull–Schmidt triangulated category has (EFS) whenever the Auslander–Reiten triangles under consideration exist.

Detection conditions

Finite support of the Euler functions does not by itself control the support of representable functors. One standard way to detect nonprojective objects in the stable category is to use a weak cogenerator. An object QQ of C\mathcal C is a weak cogenerator if C(U,Q)=0\mathcal C(U,Q)=0 implies that UU is projective. Dually, an object TT of C‾\overline{\mathcal C} is a weak generator if C‾(T,U)=0\overline{\mathcal C}(T,U)=0 implies that UU is injective. For the converse theorem, we only need these detection properties objectwise. In this form, the detecting object may depend on the support under consideration.

Definition 5.1. We say that C\mathcal C satisfies (D−) if, for every X∈ind⁡CX\in\operatorname{ind}\mathcal C, there is an object QX∈CQ_X\in\mathcal C such that, for every U∈ind⁡CU\in\operatorname{ind}\mathcal C,

C(U,X)≠0⟹U∈P or C(U,QX)≠0.\mathcal C(U,X)\ne0\quad\Longrightarrow\quad U\in\mathcal P\ \text{or}\ \mathcal C(U,Q_X)\ne0.

Dually, C\mathcal C satisfies (D+) if, for every X∈ind⁡CX\in\operatorname{ind}\mathcal C, there is an object TX∈CT_X\in\mathcal C such that, for every U∈ind⁡CU\in\operatorname{ind}\mathcal C,

C(X,U)≠0⟹U∈I or C‾(TX,U)≠0.\mathcal C(X,U)\ne0\quad\Longrightarrow\quad U\in\mathcal I\ \text{or}\ \overline{\mathcal C}(T_X,U)\ne0.

Thus a single weak cogenerator or weak generator gives a uniform form of the corresponding detection condition.

Lemma 5.2. If C\mathcal C has a weak cogenerator, then C\mathcal C satisfies (D−). If C‾\overline{\mathcal C} has a weak generator, then C\mathcal C satisfies (D+).

Proof. Let QQ be a weak cogenerator of C\mathcal C. If C(U,X)≠0\mathcal C(U,X)\ne0 and UU is not projective, then C(U,Q)≠0\mathcal C(U,Q)\ne0 by definition. Thus one may take QX=QQ_X=Q for all XX, which is precisely (D−). The second assertion is dual.

The objects QXQ_X and TXT_X need not be chosen from a finite set. This is important in the triangulated setting, where local finiteness is genuinely weaker than finite type. In a module category, by contrast, a single object works for every XX.

In the next lemma, we relate these conditions to Grothendieck relations.

Lemma 5.3. Assume that C\mathcal C has enough projectives. For Q∈CQ\in\mathcal C, choose an E\mathbb E-triangle

ΩQ→uPQ→pQ⇢,where PQ∈P,\Omega Q \xrightarrow{u} P_Q \xrightarrow{p} Q \dashrightarrow,\qquad\text{where }P_Q\in\mathcal P,

and set

rQ‾=[ΩQ]+[Q]−[PQ]∈Rel⁡(C).r_{\overline{Q}} = [\Omega Q] + [Q] - [P_Q] \in\operatorname{Rel}(\mathcal{C}).

Then, for every U∈CU \in\mathcal{C},

χU‾(rQ‾)=ℓR(ker⁡C(U,u))+ℓR(C(U,Q)).\chi_{\overline{U}}(r_{\overline{Q}}) = \ell_R(\ker\mathcal{C}(U,u)) + \ell_R(\mathcal{C}(U,Q)).

In particular,

C(U,Q)≠0⟹χU‾(rQ‾)>0.\mathcal{C}(U,Q) \ne0 \quad\Longrightarrow\quad\chi_{\overline{U}}(r_{\overline{Q}}) > 0.

Dually, if

T→iIT→vΣT⇢,where IT∈I,T \xrightarrow{i} I^T \xrightarrow{v} \Sigma T \dashrightarrow,\qquad\text{where } I^T \in\mathcal{I},

and rT+=[T]+[ΣT]−[IT]r_T^+ = [T] + [\Sigma T] - [I^T], then

χU+(rT+)=ℓR(ker⁡C(v,U))+ℓR(C‾(T,U)).\chi_U^+(r_T^+) = \ell_R(\ker\mathcal{C}(v,U)) + \ell_R(\overline{\mathcal{C}}(T,U)).

Proof. By Lemma 4.1, we give

χU‾(rQ‾)=ℓR(ker⁡C(U,u))+ℓR(coker⁡C(U,p)).\chi_{\overline{U}}(r_{\overline{Q}}) = \ell_R(\ker\mathcal{C}(U,u)) + \ell_R(\operatorname{coker} \mathcal{C}(U,p)).

We claim that the last cokernel is C(U,Q)\mathcal{C}(U,Q). Every morphism in the image of C(U,p)\mathcal{C}(U,p) factors through a projective object. Conversely, suppose that f:U→Qf: U \to Q factors as U→P′→gQU \to P' \xrightarrow{g} Q with P′P' projective. Exactness of

C(P′,PQ)⟶C(P′,Q)⟶E(P′,ΩQ)=0\mathcal{C}(P',P_Q) \longrightarrow\mathcal{C}(P',Q) \longrightarrow\mathbb{E}(P',\Omega Q) = 0

shows that gg factors through pp. Hence ff lies in the image of C(U,p)\mathcal{C}(U,p). This proves the first formula.

For the dual statement, the image of C(IT,U)→C(T,U)\mathcal{C}(I^T,U) \to\mathcal{C}(T,U) consists exactly of the morphisms T→UT \to U which factor through an injective object. The second formula follows from the dual argument and the second identity in Lemma 4.1.

Lemma 5.4

Lemma 5.4. Let C\mathcal{C} be a locally finite Hom-finite Krull–Schmidt extriangulated category with enough projectives and injectives. Then P\mathcal{P} and I\mathcal{I} are locally finite, and C\mathcal{C} satisfies (EFS), (D−_-) and (D+_+).

Proof. The first assertion is immediate because P\mathcal{P} and I\mathcal{I} are full subcategories of C\mathcal{C}. For an Auslander–Reiten E\mathbb{E}-triangle A→B→C⇢A \to B \to C \dashrightarrow, it follows from Lemma 4.1 that

Supp⁡χ−([δ])⊆Supp⁡C(−,A)∪Supp⁡C(−,B)∪Supp⁡C(−,C),\operatorname{Supp}\chi^-([\delta]) \subseteq\operatorname{Supp}\mathcal{C}(-,A) \cup\operatorname{Supp}\mathcal{C}(-,B) \cup\operatorname{Supp}\mathcal{C}(-,C),

and the dual statement holds for χ+\chi^+. Hence (EFS) follows from local finiteness.

Fix X∈ind⁡CX \in\operatorname{ind}\mathcal{C} and let SX−\mathcal{S}_X^- be the finite set of nonprojective indecomposables UU with C(U,X)≠0\mathcal{C}(U,X) \ne0. Put QX=⨁U∈SX−UQ_X = \bigoplus_{U \in\mathcal{S}_X^-} U, with QX=0Q_X = 0 if the set is empty. For U∈SX−U \in\mathcal{S}_X^-, the canonical inclusion U→QXU \to Q_X is nonzero in CC; otherwise 1U1_U would factor through a projective object, forcing UU itself to be projective. Thus (D−_-) holds. The proof of (D+_+) is dual.

We now prove the main result of the paper.

Theorem 5.5. Let CC be a skeletally small Hom-finite Krull–Schmidt RR-linear extriangulated category with enough projectives and enough injectives. Then the following conditions are equivalent.

(1) CC is locally finite.

(2) Both P\mathcal{P} and I\mathcal{I} are locally finite, CC satisfies (EFS), (D−_-) and (D+_+), and

Rel⁡(C)=AR⁡(C).\operatorname{Rel}(C)=\operatorname{AR}(C).

Proof. Assume first that CC is locally finite. By Lemma 5.4, P\mathcal{P} and I\mathcal{I} are locally finite and CC satisfies (EFS), (D−_-) and (D+_+). Moreover, (LF−_-) holds, so Theorem 3.3 gives Rel⁡(C)=AR⁡(C)\operatorname{Rel}(C)=\operatorname{AR}(C). Thus (1) implies (2).

Conversely, assume (2) and fix X∈ind⁡CX \in\operatorname{ind} C. Choose an E\mathbb{E}-triangle

ΩX⟶PX→pXX⇢\Omega X \longrightarrow P^X \xrightarrow{p_X} X \dashrightarrow

where PX∈PP^X \in\mathcal{P}, and write PX=P1⊕⋯⊕PtP^X=P_1\oplus\cdots\oplus P_t with Pi∈ind⁡PP_i\in\operatorname{ind}\mathcal{P}. If P∈ind⁡PP\in\operatorname{ind}\mathcal{P} and 0≠f:P→X0\ne f:P\to X, projectivity gives g:P→PXg:P\to P^X with f=pXgf=p_Xg. Hence C(P,Pi)≠0\mathcal{C}(P,P_i)\ne0 for some ii, and

{P∈ind⁡P∣C(P,X)≠0}⊆⋃i=1t{P∈ind⁡P∣C(P,Pi)≠0}.\{P\in\operatorname{ind}\mathcal{P}\mid\mathcal{C}(P,X)\ne0\}\subseteq\bigcup_{i=1}^{t}\{P\in\operatorname{ind}\mathcal{P}\mid\mathcal{C}(P,P_i)\ne0\}.

The right side is finite because P\mathcal{P} is locally finite.

Choose QXQ_X as in (D−_-), and an E\mathbb{E}-triangle

ΩQX⟶PX⟶QX⇢,\Omega Q_X\longrightarrow P_X\longrightarrow Q_X\dashrightarrow,

where PX∈PP_X\in\mathcal{P}.

Set

rX−=[ΩQX]+[QX]−[PX]∈Rel⁡(C).r_X^-=[\Omega Q_X]+[Q_X]-[P_X]\in\operatorname{Rel}(C).

Since Rel⁡(C)=AR⁡(C)\operatorname{Rel}(C)=\operatorname{AR}(C), write

rX−=∑j=1maj[δj]r_X^-=\sum_{j=1}^{m}a_j[\delta_j]

with aj∈Za_j\in\mathbb{Z} and δj\delta_j Auslander–Reiten E\mathbb{E}-triangles. Then

Supp⁡χ‾−(rX−)⊆⋃j=1mSupp⁡χ‾−([δj]).\operatorname{Supp}\overline{\chi}^{-}(r_X^-)\subseteq\bigcup_{j=1}^{m}\operatorname{Supp}\overline{\chi}^{-}([\delta_j]).

which is finite by (EFS). By Lemma 5.3, it yields

{U∈ind⁡C∣C(U,QX)≠0}⊆Supp⁡χ−(rX−).\{U \in\operatorname{ind}\mathcal{C} \mid\mathcal{C}(U,Q_X) \ne0\} \subseteq\operatorname{Supp}\chi^{-}(r_X^{-}).

Using (D−)(\mathrm{D}_{-}), we obtain

Supp⁡C(−,X)⊆{P∈ind⁡P∣C(P,X)≠0}∪{U∈ind⁡C∣C‾(U,QX)≠0},\operatorname{Supp}\mathcal{C}(-,X) \subseteq\{P \in\operatorname{ind}\mathcal{P} \mid\mathcal{C}(P,X) \ne0\} \cup\{U \in\operatorname{ind}\mathcal{C} \mid\overline{\mathcal{C}}(U,Q_X) \ne0\},

so Supp⁡C(−,X)\operatorname{Supp}\mathcal{C}(-,X) is finite.

To prove finiteness of the outgoing support, choose an E-triangle

X→iXIX⟶ΣX⇢,X \xrightarrow{i_X} I^X \longrightarrow\Sigma X \dashrightarrow,

where IX∈II^X \in\mathcal{I}, and write IX=I1⊕⋯⊕IsI^X = I_1 \oplus\cdots\oplus I_s with Ij∈ind⁡II_j \in\operatorname{ind}\mathcal{I}. If I∈ind⁡II \in\operatorname{ind}\mathcal{I} and 0≠f ⁣:X→I0 \ne f\colon X \to I, injectivity of II gives a morphism h ⁣:IX→Ih\colon I^X \to I such that f=hiXf = hi_X. Hence C(Ij,I)≠0\mathcal{C}(I_j,I) \ne0 for some jj, and therefore

{I∈ind⁡I∣C(X,I)≠0}⊆⋃j=1s{I∈ind⁡I∣C(Ij,I)≠0}.\{I \in\operatorname{ind}\mathcal{I} \mid\mathcal{C}(X,I) \ne0\} \subseteq\bigcup_{j=1}^{s}\{I \in\operatorname{ind}\mathcal{I} \mid\mathcal{C}(I_j,I) \ne0\}.

The right side is finite because I\mathcal{I} is locally finite.

Choose TXT_X as in (D+)(\mathrm{D}_{+}), and choose an E-triangle

TX⟶IX⟶ΣTX⇢,T_X \longrightarrow I_X \longrightarrow\Sigma T_X \dashrightarrow,

where IX∈II_X \in\mathcal{I}. Set

rX+=[TX]+[ΣTX]−[IX]∈Rel⁡(C).r_X^{+} = [T_X] + [\Sigma T_X] - [I_X] \in\operatorname{Rel}(\mathcal{C}).

Since Rel⁡(C)=AR⁡(C)\operatorname{Rel}(\mathcal{C}) = \operatorname{AR}(\mathcal{C}), write

rX+=∑j=1nbj[εj]r_X^{+} = \sum_{j=1}^{n} b_j[\varepsilon_j]

with bj∈Zb_j \in\mathbb{Z} and εj\varepsilon_j Auslander–Reiten E-triangles. By (EFS),

Supp⁡χ+(rX+)⊆⋃j=1nSupp⁡χ+([εj])\operatorname{Supp}\chi^{+}(r_X^{+}) \subseteq\bigcup_{j=1}^{n}\operatorname{Supp}\chi^{+}([\varepsilon_j])

is finite. The second formula in Lemma 5.3 gives

{U∈ind⁡C∣C‾(TX,U)≠0}⊆Supp⁡χ+(rX+).\{U \in\operatorname{ind}\mathcal{C} \mid\overline{\mathcal{C}}(T_X,U) \ne0\} \subseteq\operatorname{Supp}\chi^{+}(r_X^{+}).

Using (D+)(\mathrm{D}_{+}), we obtain

Supp⁡C(X,−)⊆{I∈ind⁡I∣C(X,I)≠0}∪{U∈ind⁡C∣C‾(TX,U)≠0},\operatorname{Supp}\mathcal{C}(X,-) \subseteq\{I \in\operatorname{ind}\mathcal{I} \mid\mathcal{C}(X,I) \ne0\} \cup\{U \in\operatorname{ind}\mathcal{C} \mid\overline{\mathcal{C}}(T_X,U) \ne0\},

so Supp⁡C(X,−)\operatorname{Supp}\mathcal{C}(X,-) is finite. Thus C\mathcal{C} is locally finite. □

Remark 5.6. The proof separates the two sides of local finiteness. To obtain (LF−)(\mathrm{LF}_{-}), it uses enough projectives, the local finiteness of P\mathcal{P}, the finiteness of Supp⁡χ−([ρ])\operatorname{Supp}\chi^{-}([\rho]) for Auslander–Reiten E-triangles ρ\rho, the condition (D−)(\mathrm{D}_{-}), and Rel⁡(C)=AR⁡(C)\operatorname{Rel}(\mathcal{C})=\operatorname{AR}(\mathcal{C}). The dual datas give (LF+)(\mathrm{LF}_{+}). Thus the projective and injective boundary assumptions, the Euler functions, and the detection conditions enter at different stages of the argument.

The boundary assumption cannot simply be omitted. In Example 3.4, the extriangulated structure is split, so every object is projective and injective. Hence the category has enough projectives and injectives, (EFS)(\mathrm{EFS}) is vacuous, and (D−)(\mathrm{D}_{-}) and (D+)(\mathrm{D}_{+}) hold automatically. Moreover, every E-triangle is split, so

Rel⁡(C)=AR⁡(C)=0.\operatorname{Rel}(\mathcal{C})=\operatorname{AR}(\mathcal{C})=0.

Nevertheless, the category is not locally finite. In this example P=I=C\mathcal{P}=\mathcal{I}=\mathcal{C}, and these boundary subcategories are not locally finite. Thus the relation equality, even together with the Euler and detection conditions, does not by itself control morphisms through projective or injective objects.

We do not claim that the remaining assumptions are logically independent in full generality. The criteria below are useful because these assumptions arise naturally in many familiar settings. Exact structures and Serre duality give (EFS)(\mathrm{EFS}). Weak cogenerators and weak generators give the detection conditions. Projective generators and injective cogenerators ensure that the corresponding boundary subcategories are of finite type.

By Lemma 5.2, the weak detector conditions provide a useful way to verify (D−)(\mathrm{D}_{-}) and (D+)(\mathrm{D}_{+}). We first record a direct consequence of the main theorem before giving more concrete criteria.

Corollary 5.7. Let C\mathcal{C} be a skeletally small Hom-finite Krull–Schmidt RR-linear extriangulated category with enough projectives and enough injectives. Assume that P\mathcal{P} and I\mathcal{I} are locally finite and that C\mathcal{C} satisfies (EFS)(\mathrm{EFS}), (D−)(\mathrm{D}_{-}) and (D+)(\mathrm{D}_{+}). Then the following are equivalent:

  1. (LF−)(\mathrm{LF}_{-}) holds;

  1. (LF+)(\mathrm{LF}_{+}) holds;

  1. C\mathcal{C} is locally finite;

  1. Rel⁡(C)=AR⁡(C)\operatorname{Rel}(\mathcal{C})=\operatorname{AR}(\mathcal{C}).

Proof. Either (1) or (2) implies (4) by Theorem 3.3. Under the standing structural assumptions, (4) implies (3) by Theorem 5.5. Finally, (3) implies both (1) and (2) by definition.

Corollary 5.8 (A generator and cogenerator criterion). Let C\mathcal{C} be a skeletally small Hom-finite Krull–Schmidt RR-linear extriangulated category which has enough projectives $\operatorname{add}\mathcal{P}

and enough injectives add⁡I\operatorname{add}I. Assume that C\mathcal{C} satisfies (EFS), that C‾\overline{\mathcal{C}} has a weak cogenerator, and that C‾\overline{\mathcal{C}} has a weak generator. Then

C is of finite type⟺Rel⁡(C)=AR⁡(C).\mathcal{C}\text{ is of finite type}\quad\Longleftrightarrow\quad\operatorname{Rel}(\mathcal{C})=\operatorname{AR}(\mathcal{C}).

Proof. By assumption, C\mathcal{C} has enough projectives given by add⁡P\operatorname{add}P and enough injectives given by add⁡I\operatorname{add}I. By Lemma 5.2, (D−)(\mathrm{D}_{-}) and (D+)(\mathrm{D}_{+}) hold. Since C\mathcal{C} has projective generator PP, the implication from right to left follows from Theorem 5.5. The converse follows from Theorem 3.3.

We apply it to 0-Auslander extriangulated categories in the sense of [7 Definition 3.7].

Corollary 5.9. Let C\mathcal{C} be a skeletally small Hom-finite Krull–Schmidt RR-linear 0-Auslander extriangulated category. Assume that P=add⁡P\mathcal{P}=\operatorname{add}P and that C\mathcal{C} satisfies (EFS). Then

C is of finite type⟺Rel⁡(C)=AR⁡(C).\mathcal{C}\text{ is of finite type}\quad\Longleftrightarrow\quad\operatorname{Rel}(\mathcal{C})=\operatorname{AR}(\mathcal{C}).

Proof. A 0-Auslander extriangulated category has enough projectives and enough injectives. Since P\mathcal{P} is of finite type, I\mathcal{I} is also of finite type, they are locally finite. By [7 Proposition 3.11], the costable category C‾=C/[I]\overline{\mathcal{C}}=\mathcal{C}/[\mathcal{I}] is equivalent to mod⁡P‾\operatorname{mod}\overline{\mathcal{P}}, where P‾\overline{\mathcal{P}} is the image of P\mathcal{P} modulo the projective-injective objects. Since P\mathcal{P} is of finite type, P‾\overline{\mathcal{P}} has an additive generator P0P_{0}. Let GG be the object of C‾\overline{\mathcal{C}} corresponding to the representable functor P‾(−,P0)\overline{\mathcal{P}}(-,P_{0}). We claim that GG is a weak generator of C‾\overline{\mathcal{C}}. Indeed, suppose that X∈CX\in\mathcal{C} satisfies C‾(G,X)=0\overline{\mathcal{C}}(G,X)=0. By the above equivalence and Yoneda, we have

0=Hom⁡mod⁡P‾(P‾(−,P0),FX)≃FX(P0),0=\operatorname{Hom}_{\operatorname{mod}\overline{\mathcal{P}}}(\overline{\mathcal{P}}(-,P_{0}),F_{X})\simeq F_{X}(P_{0}),

where FXF_{X} denotes the functor corresponding to XX. Since every object of P‾\overline{\mathcal{P}} belongs to add⁡P0\operatorname{add}P_{0}, it follows that FX=0F_{X}=0. Hence XX is zero in C‾\overline{\mathcal{C}}, so 1X1_{X} factors through an injective object. Thus XX is injective, and the claim follows.

Now apply the same proposition to Cop\mathcal{C}^{\mathrm{op}}, which is again 0-Auslander. Its costable category is (C‾)op(\overline{\mathcal{C}})^{\mathrm{op}}. The projective objects of Cop\mathcal{C}^{\mathrm{op}} are precisely the injective objects of C\mathcal{C}. Hence the finite type assumption on I\mathcal{I} gives a weak generator of (C‾)op(\overline{\mathcal{C}})^{\mathrm{op}} by the same argument. Viewed in C\mathcal{C}, this object is a weak cogenerator. Indeed, if C(X,Q)=0\mathcal{C}(X,Q)=0, then XX is zero in C‾\overline{\mathcal{C}} and therefore projective.

By Lemma 5.2, we obtain (D−)(\mathrm{D}_{-}) and (D+)(\mathrm{D}_{+}). If Rel⁡(C)=AR⁡(C)\operatorname{Rel}(\mathcal{C})=\operatorname{AR}(\mathcal{C}), local finiteness follows from Theorem 5.5. Hence C\mathcal{C} is of finite type. The converse follows from Theorem 3.3.

We next give an example in which the hypotheses of Corollary 5.9 arise naturally from a relative extriangulated structure that is generally neither exact nor triangulated.

Example 5.10. Let kk be a field and let T\mathcal{T} be a skeletally small Hom-finite Krull–Schmidt kk-linear 2-Calabi–Yau triangulated category. Let TT be a cluster tilting object. We endow the underlying additive category of T\mathcal{T} with the relative extriangulated structure whose E-triangles are the triangles

X⟶Y⟶Z→ηΣXX \longrightarrow Y \longrightarrow Z \xrightarrow{\eta} \Sigma X

for which η\eta factors through an object of add⁡ΣT\operatorname{add}\Sigma T. We denote the resulting extriangulated category by TT\mathcal{T}_T.

By [7 Section 3.3.6], TT\mathcal{T}_T is a 0-Auslander extriangulated category with

Proj⁡TT=add⁡T,Inj⁡TT=add⁡Σ2T.\operatorname{Proj}\mathcal{T}_T=\operatorname{add}T,\qquad\operatorname{Inj}\mathcal{T}_T=\operatorname{add}\Sigma^2T.

Hence its projective and injective subcategories are of finite type. Since T\mathcal{T} is 2-Calabi–Yau, Σ2\Sigma^2 is a Serre functor on the underlying additive category, and Proposition 4.4 shows that TT\mathcal{T}_T satisfies (EFS). By Corollary 5.9, we give

T is of finite type⟺Rel⁡(TT)=AR⁡(TT).\mathcal{T}\text{ is of finite type}\quad\Longleftrightarrow\quad\operatorname{Rel}(\mathcal{T}_T)=\operatorname{AR}(\mathcal{T}_T).

For a Dynkin cluster category, the condition on the left is automatic. Thus this includes the familiar relative examples of finite type. To compare our result with that of Zhu and Zhuang [ZZ], note that K0(TT)K_0(\mathcal{T}_T) is precisely their relative Grothendieck group. Its defining relations come from those triangles whose connecting morphism factors through add⁡ΣT\operatorname{add}\Sigma T. Moreover, Zhu and Zhuang [ZZ, Lemma 4.13 and the proof of Theorem 4.14] showed that the Auslander–Reiten E-triangles for this relative structure are exactly the ambient Auslander–Reiten triangles satisfying the same factorization condition. Hence the equivalence above gives another proof of the cluster tilting object and 2-Calabi–Yau special case of their relative criterion [ZZ, Theorem 4.14].

Applications and examplesl cases

Module categories over Artin algebras

Let Λ\Lambda be an Artin RR-algebra and let mod⁡Λ\operatorname{mod}\Lambda be the category of finitely generated left Λ\Lambda-modules. Choose representatives S1,…,StS_1,\ldots,S_t of the simple left modules and put

SΛ=S1⊕⋯⊕St.S_\Lambda=S_1\oplus\cdots\oplus S_t.

Lemma 6.1. The object SΛS_\Lambda is a weak cogenerator of mod⁡Λ\operatorname{mod}\Lambda and a weak generator of mod⁡Λ‾\overline{\operatorname{mod}\Lambda}. Equivalently,

Hom⁡‾Λ(U,SΛ)=0⟹U is projective,\underline{\operatorname{Hom}}_\Lambda(U,S_\Lambda)=0\quad\Longrightarrow\quad U\text{ is projective},
Hom⁡‾Λ(SΛ,U)=0⟹U is injective.\overline{\operatorname{Hom}}_\Lambda(S_\Lambda,U)=0\quad\Longrightarrow\quad U\text{ is injective}.

Proof. The first assertion is [5 Proposition 4.7], since Λ/rad⁡Λ\Lambda/\operatorname{rad}\Lambda and SΛS_{\Lambda} have the same indecomposable simple summands. The second follows from the same result for Λop⁡\Lambda^{\operatorname{op}} and the standard Artin duality.

Butler [4] proved that finite representation type implies that the relations of the Grothendieck group are generated by almost split sequences. Auslander [1] proved the converse. Since mod⁡Λ\operatorname{mod}\Lambda is locally finite if and only if Λ\Lambda is of finite representation type, we obtain the following formulation.

Theorem 6.2. [4, 1] Let RR be a commutative artin ring and let Λ\Lambda be an Artin RR-algebra. The following are equivalent:

  1. Λ\Lambda is of finite representation type;

  2. mod⁡Λ\operatorname{mod}\Lambda is locally finite;

  3. Rel⁡(mod⁡Λ)=AR⁡(mod⁡Λ)\operatorname{Rel}(\operatorname{mod}\Lambda)=\operatorname{AR}(\operatorname{mod}\Lambda).

More general exact categories related to the Auslander–Butler theorem are studied in [5].

Triangulated categories

Let C\mathcal{C} be a Hom-finite Krull–Schmidt triangulated category with suspension [1][1], regarded as an extriangulated category via E(X,Y)=C(X,Y[1])\mathbb{E}(X,Y)=\mathcal{C}(X,Y[1]). Its only projective and injective object is zero, and it has enough projectives and injectives in the extriangulated sense; for example, one may use the trivial triangles

X[−1]⟶0⟶X→1XXandX⟶0⟶X[1]→1X[1]X[1].X[-1]\longrightarrow0\longrightarrow X \xrightarrow{1_X} X \quad\text{and}\quad X\longrightarrow0\longrightarrow X[1] \xrightarrow{1_{X[1]}} X[1].

Corollary 6.3. Let C\mathcal{C} be a skeletally small Hom-finite Krull–Schmidt RR-linear triangulated category. The following conditions are equivalent:

  1. ∣Supp⁡C(−,X)∣<∞\lvert\operatorname{Supp}\mathcal{C}(-,X)\rvert<\infty for every X∈ind⁡CX\in\operatorname{ind}\mathcal{C};

  2. ∣Supp⁡C(X,−)∣<∞\lvert\operatorname{Supp}\mathcal{C}(X,-)\rvert<\infty for every X∈ind⁡CX\in\operatorname{ind}\mathcal{C};

  3. C\mathcal{C} is locally finite;

  4. Rel⁡(C)=AR⁡(C)\operatorname{Rel}(\mathcal{C})=\operatorname{AR}(\mathcal{C}).

Proof. Here P=I=0\mathcal{P}=\mathcal{I}=0 and C‾=C‾=C\underline{\mathcal{C}}=\overline{\mathcal{C}}=\mathcal{C}. Thus (D−)(\mathrm{D}_{-}) and (D+)(\mathrm{D}_{+}) are automatic, while (EFS) follows from (4.1). The result now follows from Corollary 5.7.

For R=kR=k a field, the implication (3)⇒\Rightarrow(4) recovers [XZ, Theorem 2.1], while the converse (4)⇒\Rightarrow(3) recovers [Zh, Theorem 3.14 and Remark 3.15]. The equivalence of (1) and (2) shows that one-sided local finiteness is self-dual in the triangulated setting.

Frobenius extriangulated categories

Let C\mathcal{C} be a Frobenius extriangulated category and write

C‾=C/[P],P=Proj⁡C=Inj⁡C.\underline{\mathcal{C}}=\mathcal{C}/[\mathcal{P}],\qquad\mathcal{P}=\operatorname{Proj}\mathcal{C}=\operatorname{Inj}\mathcal{C}.

By [NP, Corollary 7.4 and Remark 7.5], the stable category C‾\underline{\mathcal{C}} is triangulated. It is also Krull–Schmidt. Indeed, after removing projective summands, every object of C\mathcal{C} remains a finite direct sum of indecomposable objects. The endomorphism ring of the image of a nonprojective indecomposable is a quotient of a local ring and is therefore local. For Frobenius exact categories, the corresponding statement on relation subgroups appears in Haugland [9 Lemma 3.2]. We now record the extriangulated version.

Proposition 6.4. Let C\mathcal{C} be a skeletally small Hom-finite Krull–Schmidt RR-linear Frobenius extriangulated category. If

Rel⁡(C)=AR⁡(C),\operatorname{Rel}(\mathcal{C})=\operatorname{AR}(\mathcal{C}),

then

Rel⁡(C‾)=AR⁡(C‾),\operatorname{Rel}(\underline{\mathcal{C}})=\operatorname{AR}(\underline{\mathcal{C}}),

and the triangulated category C‾\underline{\mathcal{C}} is locally finite.

Proof. Let π:C⟶C‾\pi:\mathcal{C}\longrightarrow\underline{\mathcal{C}} be the quotient functor. It induces a surjective homomorphism

π∗:K0sp(C)⟶K0sp(C‾),[X]⟼[X].\pi_*:K_0^{\mathrm{sp}}(\mathcal{C})\longrightarrow K_0^{\mathrm{sp}}(\underline{\mathcal{C}}),\qquad[X]\longmapsto[X].

By the standard Frobenius triangulation on C‾\underline{\mathcal{C}} [NP, Corollary 7.4 and Remark 7.5], the quotient functor sends every E\mathbb{E}-triangle in C\mathcal{C} to a distinguished triangle in C‾\underline{\mathcal{C}}, and, by construction, every distinguished triangle is isomorphic to one obtained in this way. For the explicit construction under (WIC), compare [NP, Definition 6.15 and Theorem 6.20]. Hence

π∗(Rel⁡(C))=Rel⁡(C‾).\pi_*(\operatorname{Rel}(\mathcal{C}))=\operatorname{Rel}(\underline{\mathcal{C}}).

By [10 Proposition 5.11], an almost split extension remains almost split in the ideal quotient by any full additive subcategory of projective-injective objects. Taking this subcategory to be P\mathcal{P}, the image of an Auslander–Reiten E\mathbb{E}-triangle in C\mathcal{C} is an Auslander–Reiten triangle in C‾\underline{\mathcal{C}}, and therefore

π∗(AR⁡(C))⊆AR⁡(C‾).\pi_*(\operatorname{AR}(\mathcal{C}))\subseteq\operatorname{AR}(\underline{\mathcal{C}}).

The hypothesis gives

Rel⁡(C‾)=π∗(Rel⁡(C))=π∗(AR⁡(C))⊆AR⁡(C‾)⊆Rel⁡(C‾),\operatorname{Rel}(\underline{\mathcal{C}})=\pi_*(\operatorname{Rel}(\mathcal{C}))=\pi_*(\operatorname{AR}(\mathcal{C}))\subseteq\operatorname{AR}(\underline{\mathcal{C}})\subseteq\operatorname{Rel}(\underline{\mathcal{C}}),

so Rel⁡(C‾)=AR⁡(C‾)\operatorname{Rel}(\underline{\mathcal{C}})=\operatorname{AR}(\underline{\mathcal{C}}). The stable category is skeletally small and Hom-finite, and it is Krull–Schmidt as noted above. By Corollary 6.3, we show that C‾\underline{\mathcal{C}} is locally finite. □\square

Notice that no finiteness assumption on the indecomposable projective objects of C\mathcal{C} is needed here. The conclusion concerns the stable category C\mathcal{C}, where the projective and injective boundary has vanished.

The examples below illustrate several features of Theorem 5.5 that do not appear in the classical cases. We verify local finiteness and the existence of enough projectives and injectives directly. The remaining conditions then follow from the implication (1)⇒(2)(1)\Rightarrow(2) of Theorem 5.5. These examples include categories of infinite type, genuinely extriangulated structures, and cases in which the projective and injective subcategories contain infinitely many indecomposable objects.

Example 6.5 (Two-term projective complexes). Let kk be a field and let QQ be a Dynkin quiver. Put Λ=kQ\Lambda= kQ and consider

C=K[−1,0](proj⁡Λ)⊆Kb(proj⁡Λ)\mathcal{C} = K^{[-1,0]}(\operatorname{proj}\Lambda) \subseteq K^{b}(\operatorname{proj}\Lambda)

with the extriangulated structure induced from the ambient triangulated category. Categories of complexes of fixed size were studied in [3]. In the present homotopy–extriangulated setting, it follows from [7 Section 3.3.3] that this is a reduced 0-Auslander extriangulated category. In particular, it has enough projectives and injectives. Its projective objects are the stalk complexes concentrated in degree 0, while its injective objects are the stalk complexes concentrated in degree −1-1.

Since Λ\Lambda is hereditary, every object XX of C\mathcal{C}, viewed in Db(mod⁡Λ)D^{b}(\operatorname{mod}\Lambda), is isomorphic to

H−1(X)[1]⊕H0(X),H^{-1}(X)[1] \oplus H^{0}(X),

and H−1(X)H^{-1}(X) is projective. See, for example, [8 Chapter I, Section 5]. Under the fully faithful equivalence Kb(proj⁡Λ)≃Db(mod⁡Λ)K^{b}(\operatorname{proj}\Lambda) \simeq D^{b}(\operatorname{mod}\Lambda), the module H0(X)H^{0}(X) is represented by its projective resolution. This resolution is concentrated in degrees −1-1 and 0. Thus the indecomposable objects of C\mathcal{C} are represented by indecomposable projective modules in degree −1-1 or by projective resolutions of indecomposable modules. Since QQ is Dynkin, there are only finitely many such objects. Hence C\mathcal{C} is of finite type and therefore locally finite.

The induced extriangulated structure is neither exact nor triangulated. For a nonzero projective module PP, the triangle

P⟶0⟶P[1]⟶P[1]P \longrightarrow0 \longrightarrow P[1] \longrightarrow P[1]

has all three terms in K[−1,0](proj⁡Λ)K^{[-1,0]}(\operatorname{proj}\Lambda). Hence P→0P \to0 is an inflation in the induced extriangulated structure. However, it is not a monomorphism. Thus the structure is not exact. It is not triangulated either because it has nonzero projective objects.

The preceding example is of finite type. We next give an example of infinite type which is neither exact nor triangulated.

Let kk be a field and let

Qn:1⟶2⟶⋯⟶n,n≥3,Q_n:\quad1\longrightarrow2\longrightarrow\cdots\longrightarrow n,\quad n\geq3,

and set Λn=kQn\Lambda_n=kQ_n. Let Bn\mathcal{B}_n be the full subcategory of mod⁡Λn\operatorname{mod}\Lambda_n consisting of modules whose component at the vertex 11 is zero. Thus Bn\mathcal{B}_n is a Serre subcategory, equivalent to the module category of the linearly oriented quiver on the vertices 2,…,n2,\ldots,n. Define

Xn={X∈Db(mod⁡Λn)∣Hr(X)∈Bn for every r≠0}.\mathcal{X}_n=\{X\in D^{\mathrm{b}}(\operatorname{mod}\Lambda_n)\mid H^r(X)\in\mathcal{B}_n\text{ for every }r\neq0\}.

Proposition 6.6. The category Xn\mathcal{X}_n is an extension-closed Hom-finite Krull–Schmidt subcategory of Db(mod⁡Λn)D^{\mathrm{b}}(\operatorname{mod}\Lambda_n), hence it is extriangulated. It has the following properties.

(i) Xn\mathcal{X}_n is locally finite but not of finite type;

(ii) Xn\mathcal{X}_n has enough projectives and injectives, and

Proj⁡Xn=add⁡P1,Inj⁡Xn=add⁡S1,\operatorname{Proj}\mathcal{X}_n=\operatorname{add}P_1,\qquad\operatorname{Inj}\mathcal{X}_n=\operatorname{add}S_1,

where P1P_1 is the indecomposable projective at the vertex 11 and S1S_1 is the simple module at the vertex 11;

(iii) Xn\mathcal{X}_n is neither exact nor triangulated.

Proof. Since Bn\mathcal{B}_n is a Serre subcategory, the long exact cohomology sequence associated with a triangle shows that Xn\mathcal{X}_n is extension closed. Indeed, if X,Z∈XnX,Z\in\mathcal{X}_n and X→Y→Z→X[1]X\to Y\to Z\to X[1] is a triangle, then for r≠0r\neq0 there is a short exact sequence

0⟶coker⁡(Hr−1(Z)⟶Hr(X))⟶Hr(Y)⟶ker⁡(Hr(Z)⟶Hr+1(X))⟶0.0\longrightarrow\operatorname{coker}\left(H^{r-1}(Z)\longrightarrow H^r(X)\right)\longrightarrow H^r(Y)\longrightarrow\operatorname{ker}\left(H^r(Z)\longrightarrow H^{r+1}(X)\right)\longrightarrow0.

Both end terms belong to Bn\mathcal{B}_n, and hence so does Hr(Y)H^r(Y). Thus Xn\mathcal{X}_n inherits an extriangulated structure from Db(mod⁡Λn)D^{\mathrm{b}}(\operatorname{mod}\Lambda_n) by [NP, Remark 2.18].

Write Mi,jM_{i,j} for the interval module supported on the vertices i,i+1,…,ji,i+1,\ldots,j; these are precisely the indecomposable Λn\Lambda_n-modules, see [2]. Since Λn\Lambda_n is hereditary, every indecomposable object of Db(mod⁡Λn)D^{\mathrm{b}}(\operatorname{mod}\Lambda_n) is a shift of an indecomposable module; see [8 Chapter I, Section 5]. Hence

ind⁡Xn={Mi,j∣1≤i≤j≤n}∪{Mi,j[r]∣2≤i≤j≤n, r≠0}.\operatorname{ind}\mathcal{X}_n=\{M_{i,j}\mid1\leq i\leq j\leq n\}\cup\{M_{i,j}[r]\mid2\leq i\leq j\leq n,\ r\neq0\}.

In particular, the objects S2[r]S_2[r], r∈Zr\in\mathbb{Z}, are pairwise nonisomorphic, so Xn\mathcal{X}_n is not of finite type. Local finiteness can be checked directly. For indecomposable modules M,NM,N and integers r,sr,s,

Hom⁡Db(mod⁡Λn)(M[r],N[s])≃Ext⁡Λns−r(M,N),\operatorname{Hom}_{D^{\mathrm{b}}(\operatorname{mod}\Lambda_n)}(M[r],N[s])\simeq\operatorname{Ext}^{s-r}_{\Lambda_n}(M,N),

and the right-hand side can be nonzero only for s−r=0s-r=0 or 11, since Λn\Lambda_n is hereditary. There are only finitely many interval modules Mi,jM_{i,j}. Hence, for any fixed indecomposable object of Xn\mathcal{X}_n, only finitely many indecomposable objects of Xn\mathcal{X}_n can occur in either Hom support. Thus Xn\mathcal{X}_n is locally finite.

We determine the projective objects. Put P1=M1,nP_1=M_{1,n}. If N∈BnN\in\mathcal{B}_n, then

Hom⁡Λn(P1,N)=0.\operatorname{Hom}_{\Lambda_n}(P_1,N)=0.

Let YY be an indecomposable object of Xn\mathcal{X}_n. If YY is a module concentrated in degree 00, then

Hom⁡Db(mod⁡Λn)(P1,Y[1])=Ext⁡Λn1(P1,Y)=0.\operatorname{Hom}_{D^{b}(\operatorname{mod}\Lambda_n)}(P_1,Y[1])=\operatorname{Ext}^{1}_{\Lambda_n}(P_1,Y)=0.

If Y=N[r]Y=N[r] with N∈BnN\in\mathcal{B}_n and r≠0r\ne0, then projectivity of P1P_1 and the hereditary property show that Hom⁡(P1,N[r+1])\operatorname{Hom}(P_1,N[r+1]) can be nonzero only when r=−1r=-1, in which case it is Hom⁡Λn(P1,N)=0\operatorname{Hom}_{\Lambda_n}(P_1,N)=0. Therefore

Hom⁡Db(mod⁡Λn)(P1,Y[1])=0(Y∈Xn),\operatorname{Hom}_{D^{b}(\operatorname{mod}\Lambda_n)}(P_1,Y[1])=0\quad(Y\in\mathcal{X}_n),

so P1P_1 is projective in Xn\mathcal{X}_n. Every nonzero object N[r]N[r] with N∈BnN\in\mathcal{B}_n is not projective, since

E⁡Xn(N[r],N[r−1])=End⁡Db(Λn)(N[r])≠0.\operatorname{E}_{\mathcal{X}_n}(N[r],N[r-1])=\operatorname{End}_{D^{b}(\Lambda_n)}(N[r])\ne0.

For 1≤j<n1\le j<n, the nonsplit exact sequence

0⟶Mj+1,n⟶P1⟶M1,j⟶00\longrightarrow M_{j+1,n}\longrightarrow P_1\longrightarrow M_{1,j}\longrightarrow0

shows that M1,jM_{1,j} is not projective. Hence Proj⁡Xn=add⁡P1\operatorname{Proj}\mathcal{X}_n=\operatorname{add}P_1.

Dually, put S1=M1,1S_1=M_{1,1}. The module S1S_1 is injective over Λn\Lambda_n, and Hom⁡Λn(N,S1)=0\operatorname{Hom}_{\Lambda_n}(N,S_1)=0 for N∈BnN\in\mathcal{B}_n. If YY is a module concentrated in degree 00, then Hom⁡(Y,S1[1])=Ext⁡Λn1(Y,S1)=0\operatorname{Hom}(Y,S_1[1])=\operatorname{Ext}^{1}_{\Lambda_n}(Y,S_1)=0. If Y=N[r]Y=N[r] with N∈BnN\in\mathcal{B}_n and r≠0r\ne0, injectivity of S1S_1 and the hereditary property show that Hom⁡(N[r],S1[1])\operatorname{Hom}(N[r],S_1[1]) can be nonzero only when r=1r=1, in which case it is Hom⁡Λn(N,S1)=0\operatorname{Hom}_{\Lambda_n}(N,S_1)=0. Thus S1S_1 is injective in Xn\mathcal{X}_n. Every nonzero N[r]N[r] with N∈BnN\in\mathcal{B}_n is not injective because

E⁡Xn(N[r+1],N[r])=End⁡Db(Λn)(N[r+1])≠0.\operatorname{E}_{\mathcal{X}_n}(N[r+1],N[r])=\operatorname{End}_{D^{b}(\Lambda_n)}(N[r+1])\ne0.

If j>1j>1, then

E⁡Xn(Sj[1],M1,j)=Hom⁡Λn(Sj,M1,j)≠0,\operatorname{E}_{\mathcal{X}_n}(S_j[1],M_{1,j})=\operatorname{Hom}_{\Lambda_n}(S_j,M_{1,j})\ne0,

so M1,jM_{1,j} is not injective. Thus Inj⁡Xn=add⁡S1\operatorname{Inj}\mathcal{X}_n=\operatorname{add}S_1.

We finally check that there are enough projectives and injectives. For N∈BnN\in\mathcal{B}_n and r∈Zr\in\mathbb{Z}, the trivial triangle

N[r−1]⟶0⟶N[r]→1N[r]N[r-1]\longrightarrow0\longrightarrow N[r]\xrightarrow{1}N[r]

provides a projective approximation. For M1,jM_{1,j} one uses the triangle induced by

0⟶Mj+1,n⟶P1⟶M1,j⟶0,0\longrightarrow M_{j+1,n}\longrightarrow P_1\longrightarrow M_{1,j}\longrightarrow0,

with the convention Mn+1,n=0M_{n+1,n}=0. Thus Xn\mathcal{X}_n has enough projectives. Similarly, the trivial triangle gives an injective approximation for N[r]N[r], while the exact sequence

0⟶M2,j⟶M1,j⟶S1⟶00 \longrightarrow M_{2,j} \longrightarrow M_{1,j} \longrightarrow S_1 \longrightarrow0

yields, after rotation, a triangle

M1,j⟶S1⟶M2,j[1]⟶M1,j[1],M_{1,j} \longrightarrow S_1 \longrightarrow M_{2,j}[1] \longrightarrow M_{1,j}[1],

with the convention M2,1=0M_{2,1}=0. Hence Xn\mathcal{X}_n has enough injectives.

The category is not triangulated because it has the nonzero projective object P1P_1. It is not exact either. For example, consider the rotated triangle

M1,2⟶S1⟶S2[1]⟶M1,2[1]M_{1,2} \longrightarrow S_1 \longrightarrow S_2[1] \longrightarrow M_{1,2}[1]

in Xn\mathcal{X}_n. The morphism M1,2⟶S1M_{1,2} \longrightarrow S_1 is an inflation, but it is not a monomorphism.

Example 6.7. Let kk be a field and let Λ=k[ε]/(ε2)\Lambda=k[\varepsilon]/(\varepsilon^2) be the algebra of dual numbers, regarded as a Z\mathbb{Z}-graded algebra with ε\varepsilon homogeneous of degree 1. Let C=grmod⁡Λ\mathcal{C}=\operatorname{grmod}\Lambda be the category of finite-dimensional Z\mathbb{Z}-graded left Λ\Lambda-modules with morphisms of degree zero. Then C\mathcal{C} is a Hom-finite Krull–Schmidt exact category which is locally finite but not of finite type. Moreover,

∣ind⁡Proj⁡C∣=∣ind⁡Inj⁡C∣=∞,|\operatorname{ind}\operatorname{Proj}\mathcal{C}|=|\operatorname{ind}\operatorname{Inj}\mathcal{C}|=\infty,

and C\mathcal{C} satisfies all the hypotheses of Theorem 5.5(2).

Proof. Put S=Λ/(ε)S=\Lambda/(\varepsilon). For a graded module

M=⨁i∈ZMi,M=\bigoplus_{i\in\mathbb{Z}}M_i,

multiplication by ε\varepsilon gives kk-linear maps Mi→εMi+1M_i \xrightarrow{\varepsilon} M_{i+1}. Since ε2=0\varepsilon^2=0, the composite of any two consecutive maps is zero. Thus MM, together with multiplication by ε\varepsilon, determines a bounded complex of finite-dimensional kk-vector spaces

⋯⟶Mi−1→εMi→εMi+1⟶⋯ .\cdots\longrightarrow M_{i-1} \xrightarrow{\varepsilon} M_i \xrightarrow{\varepsilon} M_{i+1} \longrightarrow\cdots.

Conversely, every such bounded complex defines a finite-dimensional graded Λ\Lambda-module by letting ε\varepsilon act as the differential. Moreover, morphisms of degree zero correspond precisely to morphisms of complexes.

Every bounded complex of finite-dimensional vector spaces is isomorphic to a direct sum of stalk complexes and two-term contractible complexes. Indeed, for each ii one may choose a decomposition

Mi=Bi⊕Hi⊕CiM_i=B_i\oplus H_i\oplus C_i

such that

ker⁡(ε:Mi→Mi+1)=Bi⊕Hi,ε:Ci⟶∼Bi+1.\ker(\varepsilon: M_i \to M_{i+1}) = B_i \oplus H_i,\qquad\varepsilon: C_i \overset{\sim}{\longrightarrow} B_{i+1}.

It follows that the indecomposable objects of C\mathcal{C} are precisely

S⟨n⟩andΛ⟨n⟩,n∈Z,S\langle n\rangle\quad\text{and}\quad\Lambda\langle n\rangle,\quad n \in\mathbb{Z},

where ⟨n⟩\langle n\rangle denotes the grading shift. In particular, the objects S⟨n⟩S\langle n\rangle, n∈Zn \in\mathbb{Z}, are pairwise nonisomorphic, and hence C\mathcal{C} is not of finite type.

We next prove local finiteness. The module S⟨n⟩S\langle n\rangle is concentrated in one degree, whereas Λ⟨n⟩\Lambda\langle n\rangle is concentrated in two consecutive degrees. A morphism in C\mathcal{C} has degree zero. Hence, for a fixed indecomposable object XX, a nonzero morphism Y⟶XY \longrightarrow X or X⟶YX \longrightarrow Y is possible only if the graded supports of XX and YY intersect. Since there are only two types of indecomposable objects for each grading shift, this leaves only finitely many possibilities for YY. Therefore both

Supp⁡C(−,X)andSupp⁡C(X,−)\operatorname{Supp}\mathcal{C}(-,X)\quad\text{and}\quad\operatorname{Supp}\mathcal{C}(X,-)

are finite for every X∈ind⁡CX \in\operatorname{ind}\mathcal{C}. Thus C\mathcal{C} is locally finite.

We now determine the projective and injective objects. For each n∈Zn \in\mathbb{Z}, set

Sn=S⟨n⟩,Pn=Λ⟨n⟩.S_n=S\langle n\rangle,\qquad P_n=\Lambda\langle n\rangle.

Since PnP_n is a grading shift of the free module Λ\Lambda, it is projective. Moreover, there is a nonsplit short exact sequence

0⟶Sn+1⟶Pn⟶Sn⟶0.0 \longrightarrow S_{n+1} \longrightarrow P_n \longrightarrow S_n \longrightarrow0.

In particular, SnS_n is not projective. Hence ind⁡Proj⁡C={Pn∣n∈Z}\operatorname{ind}\operatorname{Proj}\mathcal{C}=\{P_n\mid n\in\mathbb{Z}\}. On the other hand, Λ\Lambda is a graded Frobenius algebra. Indeed, the linear functional

λ:Λ⟶k,λ(1)=0,λ(ε)=1,\lambda:\Lambda\longrightarrow k,\qquad\lambda(1)=0,\quad\lambda(\varepsilon)=1,

defines an associative bilinear form Λ×Λ⟶k\Lambda\times\Lambda\longrightarrow k, (a,b)⟼λ(ab)(a,b)\longmapsto\lambda(ab), which is nondegenerate. Thus DΛD\Lambda is isomorphic, as a graded Λ\Lambda-module, to a grading shift of Λ\Lambda. Consequently the grading shifts PnP_n are injective as well as projective. Since

0⟶Sn⟶Pn−1⟶Sn−1⟶00 \longrightarrow S_n \longrightarrow P_{n-1} \longrightarrow S_{n-1} \longrightarrow0

is nonsplit, SnS_n is not injective. Therefore ind⁡Inj⁡C={Pn∣n∈Z}\operatorname{ind}\operatorname{Inj}\mathcal{C}=\{P_n\mid n\in\mathbb{Z}\}. In particular,

∣ind⁡Proj⁡C∣=∣ind⁡Inj⁡C∣=∞.\lvert\operatorname{ind}\operatorname{Proj}\mathcal{C}\rvert=\lvert\operatorname{ind}\operatorname{Inj}\mathcal{C}\rvert=\infty.

The preceding short exact sequences also show directly that C\mathcal{C} has enough projectives and enough injectives.

For completeness, we verify the equality of relation subgroups directly. For every n∈Zn \in\mathbb{Z}, the short exact sequence

0⟶Sn+1⟶Pn⟶Sn⟶00 \longrightarrow S_{n+1} \longrightarrow P_n \longrightarrow S_n \longrightarrow0

is an Auslander–Reiten sequence. To see this, note first that the epimorphism Pn→SnP_n \to S_n is right almost split. Indeed, using the above classification of indecomposables, a nonzero morphism from an indecomposable object to SnS_n which is not an isomorphism can only come from PnP_n, and hence factors through Pn→SnP_n \to S_n. Dually, Sn+1→PnS_{n+1} \to P_n is left almost split.

Since CC is a length category and its simple objects are precisely the SnS_n, one has

K0(C)≃⨁n∈ZZ[Sn].K_0(C) \simeq\bigoplus_{n \in\mathbb{Z}} \mathbb{Z}[S_n].

On the other hand,

K0sp(C)=⨁n∈Z(Z[Sn]⊕Z[Pn]).K_0^{\mathrm{sp}}(C) = \bigoplus_{n \in\mathbb{Z}} \left(\mathbb{Z}[S_n] \oplus\mathbb{Z}[P_n]\right).

The above short exact sequence gives

[Pn]=[Sn]+[Sn+1] in K0(C).[P_n] = [S_n] + [S_{n+1}] \text{ in } K_0(C).

Therefore the kernel of the canonical homomorphism

K0sp(C)⟶K0(C)K_0^{\mathrm{sp}}(C) \longrightarrow K_0(C)

is generated by the elements

[Sn]+[Sn+1]−[Pn],n∈Z.[S_n] + [S_{n+1}] - [P_n], \quad n \in\mathbb{Z}.

These are exactly the relations of the Auslander–Reiten sequences above. Consequently,

Rel⁡(C)=AR⁡(C).\operatorname{Rel}(C) = \operatorname{AR}(C).

Finally, CC is abelian and hence exact, and it is Hom-finite and Krull–Schmidt. Since CC is locally finite and has enough projectives and injectives, Lemma 5.4 shows that Proj⁡C\operatorname{Proj} C and Inj⁡C\operatorname{Inj} C are locally finite and that CC satisfies (EFS), (D−)(\mathrm{D}_{-}) and (D+)(\mathrm{D}_{+}). Thus every hypothesis of Theorem 5.5 is satisfied.

A criterion using Serre duality

Ordinary Serre duality interchanges the two Hom support conditions. Thus, once one side is known to be finite, no separate argument is needed for the opposite projective or injective boundary.

Theorem 7.1. Let C\mathcal{C} be a skeletally small Hom-finite Krull–Schmidt RR-linear extriangulated category admitting a Serre functor SS, and assume that

Rel⁡(C)=AR⁡(C).\operatorname{Rel}(\mathcal{C})=\operatorname{AR}(\mathcal{C}).

Suppose that one of the following two sets of conditions holds:

  1. C\mathcal{C} has enough projectives, P\mathcal{P} is locally finite, and C\mathcal{C} satisfies (D−)(\mathrm{D}_{-});

  2. C\mathcal{C} has enough injectives, I\mathcal{I} is locally finite, and C\mathcal{C} satisfies (D+)(\mathrm{D}_{+}).

Then C\mathcal{C} is locally finite.

Proof. We prove the assertion under (1), the proof under (2) is dual. By Proposition 4.4, we show that Supp⁡χ−([δ])\operatorname{Supp}\chi^{-}([\delta]) is finite for every Auslander–Reiten E\mathbb{E}-triangle δ\delta. Fix X∈ind⁡CX\in\operatorname{ind}\mathcal{C}. The projective boundary argument in the proof of Theorem 5.5, using enough projectives and the local finiteness of P\mathcal{P}, shows that

{P∈ind⁡P∣C(P,X)≠0}\{P\in\operatorname{ind}\mathcal{P}\mid\mathcal{C}(P,X)\ne0\}

is finite. Choose QXQ_X as in (D−)(\mathrm{D}_{-}) and an E\mathbb{E}-triangle

ΩQX⟶PX⟶QX⇢,where PX∈P.\Omega Q_X\longrightarrow P_X\longrightarrow Q_X\overset{}{\dashrightarrow}, \qquad\text{where }P_X\in\mathcal{P}.

Since its relation belongs to Rel⁡(C)=AR⁡(C)\operatorname{Rel}(\mathcal{C})=\operatorname{AR}(\mathcal{C}), it is a finite integral combination of Auslander–Reiten relations. By Proposition 4.4 and Lemma 5.3, we imply that {U∈ind⁡C∣C(U,QX)≠0}\{U\in\operatorname{ind}\mathcal{C}\mid\mathcal{C}(U,Q_X)\ne0\} is finite. Condition (D−)(\mathrm{D}_{-}) now gives ∣Supp⁡C(−,X)∣<∞|\operatorname{Supp}\mathcal{C}(-,X)|<\infty.

Since DD is a duality on finite-length RR-modules, it reflects the zero module. Serre duality therefore yields, for indecomposable XX and UU,

C(X,U)≠0  ⟺  C(U,SX)≠0.\mathcal{C}(X,U)\ne0\iff\mathcal{C}(U,SX)\ne0.

Consequently Supp⁡C(X,−)=Supp⁡C(−,SX)\operatorname{Supp}\mathcal{C}(X,-)=\operatorname{Supp}\mathcal{C}(-,SX), as sets of isomorphism classes. Since SS is an autoequivalence, SXSX is indecomposable, and the right side is finite by the first part of the proof. Thus C\mathcal{C} is locally finite. □\square

The Serre functor in Theorem 7.1 is defined on C\mathcal{C} itself. It should not be confused with Auslander–Reiten–Serre duality for extriangulated categories. The latter is formulated on the stable and costable categories, see [10]. In particular, an identity

DE(X,Y)≃C(Y,TX)D\mathbb{E}(X,Y)\simeq\mathcal{C}(Y,TX)

with TT an autoequivalence of C\mathcal{C} forces every projective and every injective object to be zero. Indeed, if PP is projective, then 0=DE(P,TP)≃End⁡(TP)0=D\mathbb{E}(P,TP)\simeq\operatorname{End}(TP). Hence TP=0TP=0, and therefore P=0P=0. The injective case is dual.

Statement on AI ChatGPT was used only for language editing and improving the clarity of the manuscript. All mathematical results, arguments, and proofs were developed and verified by the authors.

Data Availability Data sharing not applicable to this article as no datasets were generated or analysed during the current study.

Conflict of Interests The authors declare that they have no conflicts of interest to this work.

Panyue Zhou

School of Mathematics and Statistics, Changsha University of Science and Technology, 410114 Changsha, Hunan, P. R. China

Email: [email protected]

Bin Zhu

Department of Mathematical Sciences, Tsinghua University, 100084 Beijing, P. R. China

Email: [email protected].

References

  1. [1][Au] M. Auslander, Relations for Grothendieck groups of Artin algebras, Proc. Amer. Math. Soc. 91 (1984), no. 3, 336–340.DOI
  2. [2][ARS] M. Auslander, I. Reiten, S. O. Smalø, Representation Theory of Artin Algebras, Cambridge Studies in Advanced Mathematics, vol. 36, Cambridge University Press, Cambridge, 1995.DOI
  3. [3][BSSZ] R. Bautista, M. J. Souto Salorio, and R. Zuazua, Almost split sequences for complexes of fixed size, J. Algebra 287 (2005), no. 1, 140–168.DOI
  4. [4][Bu] M. C. R. Butler, Grothendieck groups and almost split sequences, in Integral Representations and Applications (Oberwolfach, 1980), Lecture Notes in Math., vol. 882, Springer, Berlin–New York, 1981, pp. 357–368.DOI
  5. [5][En] H. Enomoto, Relations for Grothendieck groups and representation-finiteness, J. Algebra 539 (2019), 152–176.DOI
  6. [6][En2] H. Enomoto, Classifying substructures of extriangulated categories via Serre subcategories, Appl. Categ. Structures 29 (2021), no. 6, 1005–1018.DOI
  7. [7][GNP] M. Gorsky, H. Nakaoka, and Y. Palu, Hereditary extriangulated categories: silting objects, mutation, negative extensions, preprint, arXiv:2303.07134.arxiv.org/abs/2303.07134
  8. [8][Ha] D. Happel, Triangulated Categories in the Representation Theory of Finite Dimensional Algebras, London Math. Soc. Lecture Note Ser., vol. 119, Cambridge University Press, Cambridge, 1988.
  9. [9][Hau] J. Haugland, Auslander–Reiten triangles and Grothendieck groups of triangulated categories, Algebr. Represent. Theory 25 (2022), 1379–1387.arxiv.org/abs/1904.02506
  10. [10][INP] O. Iyama, H. Nakaoka, Y. Palu, Auslander–Reiten theory in extriangulated categories, Trans. Amer. Math. Soc. Ser. B 11 (2024), 248–305.arxiv.org/abs/1805.03776
  11. [11][MMP] E. N. Marcos, H. Merklen, M. I. Platzeck, The Grothendieck group of the category of modules of finite projective dimension over certain weakly triangular algebras, Comm. Algebra 28 (2000), no. 3, 1387–1404.DOI

Paper details

vlnc:W_udhqtckb
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