betaFeedback

WorksMathematics

Auslander bounds under recollements and arrow removal

Abstract

Auslander bounds measure uniformity in the eventual vanishing of extension groups and are closely related to homological conjectures, including the Auslander-Reiten conjecture and finitistic dimension conjecture. We study global and finitistic Auslander bounds of finite-dimensional algebras under recollements and arrow removal. For a recollement induced by an idempotent ee of an algebra ΛΛ, we establish inequalities comparing the bounds of ΛΛ with those of the quotient algebra Λ/ΛeΛΛ/ΛeΛ and the corner algebra eΛeeΛe. We also establish comparison results for cleft extensions and deduce that arrow removal preserves finiteness of both the global and finitistic Auslander bounds. Finally, we construct radical-square-zero algebras with \findim(Λ_n)=0 and \mr{fab}(Λ_n)=\mr{gb}(Λ_n)=n for every n≥1n\geq1, yielding arbitrarily large gaps between finitistic Auslander bounds and finitistic dimensions.

Introduction

Auslander’s conjecture on eventual Ext vanishing asserts that, for every Artin algebra RR and every finitely generated left RR-module MM, there is a nonnegative integer bMb_M such that

Ext⁡Ri(M,N)=0for all i>bM\operatorname{Ext}^{i}_{R}(M,N)=0 \qquad\text{for all } i>b_M

whenever N∈R-modN\in R\text{-mod} satisfies Ext⁡Rj(M,N)=0\operatorname{Ext}^{j}_{R}(M,N)=0 for all sufficiently large jj; see [10]. The bound bMb_M may depend on MM, but is independent of NN. Although the conjecture fails in general [12, 20], it motivates the study of quantitative bounds on eventual Ext vanishing. The least such nonnegative integer is called the Auslander bound of MM, denoted by bR(M)b_R(M); we set bR(M)=∞b_R(M)=\infty if no such integer exists. An Artin algebra for which this vanishing property holds is said to satisfy the Auslander Condition; equivalently, every finitely generated left module has finite Auslander bound. Rings satisfying this condition have been studied in [5, 11, 16].

Let Λ\Lambda be a finite-dimensional algebra over an algebraically closed field. The global and finitistic Auslander bounds are defined, respectively, by

gb⁡(Λ)=sup⁡{bΛ(M)∣M∈Λ-mod},\operatorname{gb}(\Lambda)=\sup\{b_{\Lambda}(M)\mid M\in\Lambda\text{-mod}\},
fab⁡(Λ)=sup⁡{bΛ(M)∣M∈Λ-mod, bΛ(M)<∞}.\operatorname{fab}(\Lambda)=\sup\{b_{\Lambda}(M)\mid M\in\Lambda\text{-mod},\ b_{\Lambda}(M)<\infty\}.

Thus fab⁡(Λ)\operatorname{fab}(\Lambda) ranges only over modules with finite Auslander bound, whereas gb⁡(Λ)\operatorname{gb}(\Lambda) ranges over all modules. The Uniform Auslander Condition requires gb⁡(Λ)<∞\operatorname{gb}(\Lambda)<\infty. If Λ\Lambda satisfies the Auslander Condition, then fab⁡(Λ)=gb⁡(Λ)\operatorname{fab}(\Lambda)=\operatorname{gb}(\Lambda).

Since bΛ(M)=pd⁡ΛMb_{\Lambda}(M)=\operatorname{pd}_{\Lambda}M whenever pd⁡ΛM<∞\operatorname{pd}_{\Lambda}M<\infty, one has

findim⁡(Λ)≤fab⁡(Λ)≤gb⁡(Λ).\operatorname{findim}(\Lambda)\leq\operatorname{fab}(\Lambda)\leq\operatorname{gb}(\Lambda).

where findim⁡(Λ)=sup⁡{pd⁡ΛM∣M∈Λ-mod, pd⁡ΛM<∞}\operatorname{findim}(\Lambda)=\sup\{\operatorname{pd}_{\Lambda}M\mid M\in\Lambda\text{-mod},\ \operatorname{pd}_{\Lambda}M<\infty\} denotes the finitistic dimension of Λ\Lambda. In [21], the second author formulated the finitistic Auslander conjecture, asserting that fab⁡(Λ)<∞\operatorname{fab}(\Lambda)<\infty for every Artin algebra Λ\Lambda. By the displayed inequality, this conjecture implies the finitistic dimension conjecture. The Auslander Condition also implies the Auslander–Reiten conjecture, which asserts that a finitely generated left Λ\Lambda-module MM is projective whenever Ext⁡Λi(M,M⊕Λ)=0\operatorname{Ext}^{i}_{\Lambda}(M,M\oplus\Lambda)=0 for all i>0i>0 [5 Theorem A].

The behavior of Auslander bounds under equivalences provides a starting point for comparison results. Using syzygy complexes, the second author obtained derived-category characterizations of the finiteness of the global and finitistic Auslander bounds and proved that derived equivalences preserve both finiteness properties, as well as the Auslander Condition [22 Theorems 6.2 and 6.3]. Auslander bounds for complexes have also been studied in [15]. These results motivate comparisons of Auslander bounds under homological constructions beyond equivalences, particularly recollements and arrow removal.

Recollements, introduced by Beilinson, Bernstein and Deligne [3], relate three categories through a system of adjoint functors. In the representation theory of algebras, they have been used to compare global dimension [13], finitistic dimension [4, 17], and Gorensteinness [19]. Psaroudakis and Vitória [18] proved that a recollement whose terms are module categories is equivalent to one induced by an idempotent element.

In this paper, we study the global and finitistic Auslander bounds of algebras related by idempotent-induced recollements. For an idempotent e∈Λe\in\Lambda, we consider the recollement involving

(Λ/ΛeΛ)-mod,Λ-mod,eΛe-mod.(\Lambda/\Lambda e\Lambda)\text{-mod},\qquad\Lambda\text{-mod},\qquad e\Lambda e\text{-mod}.

We establish explicit comparisons between the Auslander bounds of Λ\Lambda and those of its quotient and corner algebras. The following theorem summarizes the resulting finiteness criteria; the explicit estimates are given in Theorems 3.4 and 3.5.

Theorem 1.1. Let ee be an idempotent element of Λ\Lambda. Then the following statements hold.

(a) Suppose that ΛeΛ\Lambda e\Lambda is a strong idempotent ideal and gldim⁡(eΛe)<∞\operatorname{gldim}(e\Lambda e)<\infty, pd⁡Λ(ΛeΛ)<∞\operatorname{pd}_{\Lambda}(\Lambda e\Lambda)<\infty, pd⁡Λop⁡(ΛeΛ)<∞\operatorname{pd}_{\Lambda^{\operatorname{op}}}(\Lambda e\Lambda)<\infty. Then

gb⁡(Λ)<∞⟺gb⁡(Λ/ΛeΛ)<∞,\operatorname{gb}(\Lambda)<\infty\quad\Longleftrightarrow\quad\operatorname{gb}(\Lambda/\Lambda e\Lambda)<\infty,
fab⁡(Λ)<∞⟺fab⁡(Λ/ΛeΛ)<∞.\operatorname{fab}(\Lambda)<\infty\quad\Longleftrightarrow\quad\operatorname{fab}(\Lambda/\Lambda e\Lambda)<\infty.

(b) Suppose that P1=P∞\mathcal{P}_{1}=\mathcal{P}_{\infty} and gldim⁡(Λ/ΛeΛ)<∞\operatorname{gldim}(\Lambda/\Lambda e\Lambda)<\infty, pd⁡Λ(ΛeΛ)<∞\operatorname{pd}_{\Lambda}(\Lambda e\Lambda)<\infty. Then

gb⁡(Λ)<∞⟺gb⁡(eΛe)<∞,\operatorname{gb}(\Lambda)<\infty\quad\Longleftrightarrow\quad\operatorname{gb}(e\Lambda e)<\infty,
fab⁡(Λ)<∞⟺fab⁡(eΛe)<∞.\operatorname{fab}(\Lambda)<\infty\quad\Longleftrightarrow\quad\operatorname{fab}(e\Lambda e)<\infty.

Here P1\mathcal{P}_{1} and P∞\mathcal{P}_{\infty} are the subcategories associated with ee defined in § 2. The two parts reduce the finiteness questions to the quotient and corner algebras, respectively. We illustrate these results with examples and obtain sharper estimates for upper triangular matrix algebras; see Proposition 3.9.

We also study the arrow-removal operation introduced by Green, Psaroudakis and Solberg [9]. Using comparison results for cleft extensions, we prove that arrow removal preserves finiteness of both Auslander bounds. The following theorem summarizes this result; the quantitative estimates are given in Theorem 4.3.

Theorem 1.2. Let Λ=kQ/I\Lambda=kQ/I, where QQ is a finite quiver and II is an admissible ideal. Suppose that ai:vei⟶vfia_i:v_{e_i}\longrightarrow v_{f_i} (1≤i≤s1\le i\le s) are arrows that do not occur in a fixed minimal generating set of II, and that

Hom⁡Λ(Λfi,Λej)=0(1≤i,j≤s).\operatorname{Hom}_{\Lambda}(\Lambda f_i,\Lambda e_j)=0 \qquad(1\le i,j\le s).

Set a‾i=ai+I\overline{a}_i=a_i+I and

Γ=Λ/∑i=1sΛa‾iΛ.\Gamma=\Lambda\Big/\sum_{i=1}^{s}\Lambda\overline{a}_i\Lambda.

Then

gb⁡(Λ)<∞⟺gb⁡(Γ)<∞,\operatorname{gb}(\Lambda)<\infty\qquad\Longleftrightarrow\qquad\operatorname{gb}(\Gamma)<\infty,
fab⁡(Λ)<∞⟺fab⁡(Γ)<∞.\operatorname{fab}(\Lambda)<\infty\qquad\Longleftrightarrow\qquad\operatorname{fab}(\Gamma)<\infty.

The paper is organized as follows. Section 2 recalls recollements, idempotent ideals, and the definitions of Auslander bounds. Section 3 proves the reduction theorems and derives sharper estimates for upper triangular matrix algebras. Section 4 establishes comparison results for cleft extensions and applies them to arrow removal. Section 5 constructs a radical-square-zero family for which fab⁡(Λ)−findim⁡(Λ)\operatorname{fab}(\Lambda)-\operatorname{findim}(\Lambda) takes arbitrarily large finite values.

Preliminaries

Unless otherwise specified, Λ\Lambda is a finite-dimensional algebra over an algebraically closed field kk, and modules are finitely generated left modules. We write Λ\Lambda-mod for the category of finitely generated left Λ\Lambda-modules. All Auslander bounds and finitistic dimensions are taken on the left. We put N={0,1,2,…}\mathbb{N}=\{0,1,2,\ldots\} and write D=Hom⁡k(−,k)D=\operatorname{Hom}_{k}(-,k) for the standard duality between finitely generated left and right Λ\Lambda-modules.

For X∈ΛX\in\Lambda-mod, we denote its projective and injective dimensions by pd⁡ΛX\operatorname{pd}_{\Lambda}X and id⁡ΛX\operatorname{id}_{\Lambda}X, respectively. We write ΩΛnX\Omega_{\Lambda}^{n}X and ΩΛ−nX\Omega_{\Lambda}^{-n}X for its nnth syzygy and cosyzygy, taken from minimal projective resolutions and minimal injective coresolutions unless otherwise specified, and set ΩΛ0X=X\Omega_{\Lambda}^{0}X=X. The full subcategory add⁡ΛX\operatorname{add}_{\Lambda}X consists of all direct summands of finite direct sums of copies of XX. For a bimodule AUB{}_{A}U_{B}, the notation pd⁡AU\operatorname{pd}_{A}U and pd⁡Bop⁡U\operatorname{pd}_{B^{\operatorname{op}}}U refers to its left AA-module and right BB-module projective dimensions, respectively.

Recollements and idempotent ideals

We recall the definition and basic properties of recollements of abelian categories.

Definition 2.1. [17 Definition 2.1] A recollement of abelian categories A\mathcal{A}, B\mathcal{B} and C\mathcal{C} is a diagram

B←i∗A→j∗Ci∗=i!↑ ↓j!=j∗A←i!A→j∗C\begin{CD} \mathcal{B} @<i^{*}<< \mathcal{A} @>j_{*}>> \mathcal{C} \\ @A{i_{*}=i_{!}}AA @. @VV{j^{!}=j^{*}}V \\ \mathcal{A} @<i^{!}<< \mathcal{A} @>j_{*}>> \mathcal{C} \end{CD}

henceforth denoted by R(A,B,C)\mathbf{R}(\mathcal{A},\mathcal{B},\mathcal{C}), satisfying the following conditions:

(RA1) (i∗,i∗,i!)(i^*,i_*,i^!) and (j!,j∗,j∗)(j_!,j^*,j_*) are adjoint triples.

(RA2) The functors i∗i_*, j!j_! and j∗j_* are fully faithful.

(RA3) Im⁡i∗=Ker⁡j∗\operatorname{Im}i_*=\operatorname{Ker}j^*.

A Serre subcategory of an abelian category is a full subcategory closed under subobjects, quotients, and extensions. The following proposition collects standard consequences of the definition.

Proposition 2.2. [7, 17, 18] The following hold for a recollement R(A,B,C)\mathbf{R}(\mathcal{A},\mathcal{B},\mathcal{C}).

(a) The functors i∗i_* and j∗j^* are exact.

(b) The compositions i∗j!i^*j_! and i!j∗i^!j_* are identically zero.

(c) The units Id⁡B→i∗i∗\operatorname{Id}_{\mathcal{B}}\to i_*i^* and Id⁡C→j∗j!\operatorname{Id}_{\mathcal{C}}\to j^*j_! and the counits i∗i!→Id⁡Bi_*i^!\to\operatorname{Id}_{\mathcal{B}} and j∗j∗→Id⁡Cj^*j_*\to\operatorname{Id}_{\mathcal{C}} are natural isomorphisms.

(d) The image of i∗i_* is a Serre subcategory of A\mathcal{A}, and j∗j^* induces an equivalence A/Im⁡i∗≃C\mathcal{A}/\operatorname{Im}i_*\simeq\mathcal{C}.

(e) For every A∈AA\in\mathcal{A}, there exist B,B′∈BB,B'\in\mathcal{B} such that the units and counits of the adjunctions induce exact sequences

0⟶i∗(B)⟶j!j∗(A)⟶A⟶i∗i∗(A)⟶0,0\longrightarrow i_* (B)\longrightarrow j_!j^*(A)\longrightarrow A\longrightarrow i_*i^*(A)\longrightarrow0,
0⟶i∗i!(A)⟶A⟶j∗j∗(A)⟶i∗(B′)⟶0.0\longrightarrow i_*i^!(A)\longrightarrow A\longrightarrow j_*j^*(A)\longrightarrow i_*(B')\longrightarrow0.

Example 2.3. Let ee be an idempotent element of Λ\Lambda, and put I=ΛeΛI=\Lambda e\Lambda. Since Λ\Lambda is finite-dimensional, all six functors in the idempotent-induced recollement of categories of all modules [14, 17] preserve finite generation. Hence this recollement restricts to the following recollement of categories of finitely generated left modules:

Λ/I-mod→inc⁡Λ-mod→eΛ⊗Λ−eΛe-modHom⁡Λ(Λ/I,−)↑ ↓Hom⁡eΛe(eΛ,−)\begin{CD} \Lambda/I\text{-}\mathrm{mod} @>{\operatorname{inc}}>> \Lambda\text{-}\mathrm{mod} @>{e\Lambda\otimes_\Lambda-}>> e\Lambda e\text{-}\mathrm{mod}\\ @A{\operatorname{Hom}_{\Lambda}(\Lambda/I,-)}AA @. @VV{\operatorname{Hom}_{e\Lambda e}(e\Lambda,-)}V\\ \end{CD}

Strongly idempotent ideals play an important role in this paper. We now recall their definition and basic properties.

Definition 2.4. [1] A two-sided ideal II of Λ\Lambda is called a strong idempotent ideal if restriction of scalars induces isomorphisms

Ext⁡Λ/Ii(X,Y)⟶Ext⁡Λi(X,Y)(i≥0)\operatorname{Ext}^{i}_{\Lambda/I}(X,Y)\longrightarrow\operatorname{Ext}^{i}_{\Lambda}(X,Y)\qquad(i\geq0)

for all X,Y∈Λ/I-modX,Y\in\Lambda/I\text{-}\mathrm{mod}.

The following proposition gives two Ext-vanishing criteria for strong idempotence.

Proposition 2.5. [1 Proposition 1.3] Let II be a two-sided ideal of an Artin algebra Λ\Lambda. The following statements are equivalent.

(a) II is a strong idempotent ideal.

(b) Ext⁡Λi(Λ/I,Y)=0\operatorname{Ext}^{i}_{\Lambda}(\Lambda/I,Y)=0 for every Y∈Λ/I-modY\in\Lambda/I\text{-}\mathrm{mod} and every i≥1i\geq1.

(c) Ext⁡Λi(Λ/I,E)=0\operatorname{Ext}^{i}_{\Lambda}(\Lambda/I,E)=0 for every injective object EE of Λ/I-mod\Lambda/I\text{-}\mathrm{mod} and every i≥1i\geq1.

To study the relationship between Λ\Lambda and its corner algebra eΛee\Lambda e, Auslander, Platzeck, and Todorov [1] introduced the classes Pk\mathbb{P}_k and Ik\mathbb{I}_k. We now recall their definitions. Let ee be an idempotent element of Λ\Lambda. For k∈Nk \in\mathbb{N}, let Pk\mathbb{P}_k be the full subcategory of Λ\Lambda-mod consisting of modules admitting a projective resolution ⋯→P1→P0→X→0\cdots\to P_1 \to P_0 \to X \to0 with Pi∈add⁡(Λe)P_i \in\operatorname{add}(\Lambda e) for 0≤i≤k0 \le i \le k. The subcategory P∞\mathbb{P}_\infty consists of modules admitting such a resolution with every term in add⁡(Λe)\operatorname{add}(\Lambda e). Dually, Ik\mathbb{I}_k is the full subcategory of Λ\Lambda-mod consisting of modules admitting an injective coresolution whose first k+1k+1 terms belong to add⁡D(eΛ)\operatorname{add}D(e\Lambda); for I∞\mathbb{I}_\infty, every term is required to belong to this subcategory.

To relate the homological properties of Λ\Lambda to those of its quotient algebra and corner algebra, we recall the gluing sequences established in [8 Proposition 3.3]. Let I=ΛeΛI=\Lambda e\Lambda be an idempotent ideal of Λ\Lambda, and set Γ=Λ/I\Gamma=\Lambda/I. For left Λ\Lambda-modules MM and XX, we denote by τM(X)\tau_M(X) the trace of MM in XX. In particular,

τI(X)=IX=ΛeX,τΓ(X)={x∈X∣Ix=0}.\tau_I(X)=IX=\Lambda eX,\qquad\tau_\Gamma(X)=\{x\in X\mid Ix=0\}.

Thus, both X/τI(X)X/\tau_I(X) and τΓ(X)\tau_\Gamma(X) are naturally left Γ\Gamma-modules. Following [1 Section 5] and [8], we consider the perpendicular subcategories

T={X∈Λ-mod∣Ext⁡Λi(Γ,X)=0 for all i≥1},\mathcal{T}=\{X\in\Lambda\text{-mod}\mid\operatorname{Ext}^i_\Lambda(\Gamma,X)=0\text{ for all }i\ge1\},
T~={X∈Λ-mod∣Ext⁡Λi(X,DΓ)=0 for all i≥1}.\widetilde{\mathcal{T}}=\{X\in\Lambda\text{-mod}\mid\operatorname{Ext}^i_\Lambda(X,D\Gamma)=0\text{ for all }i\ge1\}.

We recall the following gluing properties of these subcategories.

Lemma 2.6. [8 Proposition 3.3] Let I=ΛeΛI=\Lambda e\Lambda be a strongly idempotent ideal of Λ\Lambda. Then the following statements hold.

(a) I∞=I0∩T\mathbb{I}_\infty=\mathbb{I}_0\cap\mathcal{T}. For every X∈TX\in\mathcal{T}, there is a gluing sequence

0⟶τΓ(X)⟶X⟶X/τΓ(X)⟶0,0\longrightarrow\tau_\Gamma(X)\longrightarrow X\longrightarrow X/\tau_\Gamma(X)\longrightarrow0,

where τΓ(X)∈Γ\tau_\Gamma(X)\in\Gamma-mod and X/τΓ(X)∈I∞X/\tau_\Gamma(X)\in\mathbb{I}_\infty.

(b) P∞=P0∩T~\mathbb{P}_\infty=\mathbb{P}_0\cap\widetilde{\mathcal{T}}. For every X∈T~X\in\widetilde{\mathcal{T}}, there is a gluing sequence

0⟶τI(X)⟶X⟶X/τI(X)⟶0,0\longrightarrow\tau_I(X)\longrightarrow X\longrightarrow X/\tau_I(X)\longrightarrow0,

where τI(X)∈P∞\tau_I(X)\in\mathbb{P}_\infty and X/τI(X)∈ΓX/\tau_I(X)\in\Gamma-mod.

Auslander bounds

We recall the definitions of the Auslander bounds used throughout this paper; see [15, 21].

Definition 2.7. For M,N∈ΛM,N\in\Lambda-mod, set

pΛ(M,N)=sup⁡{n≥0∣Ext⁡Λn(M,N)≠0}.p_\Lambda(M,N)=\sup\{n\ge0\mid\operatorname{Ext}^n_\Lambda(M,N)\ne0\}.

The Auslander bound of MM is defined by

bΛ(M)=sup⁡{pΛ(M,N)∣N∈Λ-mod, pΛ(M,N)<∞}.b_\Lambda(M)=\sup\{p_\Lambda(M,N)\mid N\in\Lambda\text{-mod},\ p_\Lambda(M,N)<\infty\}.

The global Auslander bound and the finitistic Auslander bound of Λ\Lambda are defined, respectively, by

gb⁡(Λ)=sup⁡{bΛ(M)∣M∈Λ-mod}\operatorname{gb}(\Lambda)=\sup\{b_\Lambda(M)\mid M\in\Lambda\text{-mod}\}

and

fab⁡(Λ)=sup⁡{bΛ(M)∣M∈Λ-mod, bΛ(M)<∞}.\operatorname{fab}(\Lambda)=\sup\{b_\Lambda(M)\mid M\in\Lambda\text{-mod},\ b_\Lambda(M)<\infty\}.

All suprema are taken in N∪{∞}\mathbb{N}\cup\{\infty\}, with the convention sup⁡∅=0\sup\varnothing=0.

We say that Λ\Lambda satisfies the Auslander Condition if bΛ(M)<∞b_\Lambda(M)<\infty for every M∈Λ-modM\in\Lambda\text{-mod}, and that it satisfies the Uniform Auslander Condition if gb⁡(Λ)<∞\operatorname{gb}(\Lambda)<\infty.

If pd⁡ΛM<∞\operatorname{pd}_\Lambda M<\infty, then bΛ(M)=pd⁡ΛMb_\Lambda(M)=\operatorname{pd}_\Lambda M [21 Remark 2.2(3)]. Consequently,

findim⁡(Λ)≤fab⁡(Λ)≤gb⁡(Λ).\operatorname{findim}(\Lambda)\leq\operatorname{fab}(\Lambda)\leq\operatorname{gb}(\Lambda).

Moreover, fab⁡(Λ)=gb⁡(Λ)\operatorname{fab}(\Lambda)=\operatorname{gb}(\Lambda) whenever Λ\Lambda satisfies the Auslander Condition.

In particular, finiteness of the global Auslander bound implies finiteness of the finitistic Auslander bound, but the converse fails. Indeed, the finite-dimensional commutative self-injective algebra AA constructed in [12] satisfies gb⁡(A)=∞\operatorname{gb}(A)=\infty, whereas fab⁡(A)=0\operatorname{fab}(A)=0 by [21 Proposition 3.1].

Auslander bounds under recollements

Throughout this section, we fix an idempotent e∈Λe\in\Lambda and set I=ΛeΛI=\Lambda e\Lambda. Our aim is to compare the global and finitistic Auslander bounds of Λ\Lambda with those of the quotient algebra Λ/I\Lambda/I and the corner algebra eΛee\Lambda e. We first establish comparisons of Auslander bounds for individual modules via restriction and induction functors.

Lemma 3.1. Let ee be an idempotent element of Λ\Lambda and I=ΛeΛI=\Lambda e\Lambda.

(a) If II is a strong idempotent ideal, then, for every M∈Λ/I-modM\in\Lambda/I\text{-mod},

bΛ/I(M)≤bΛ(M).b_{\Lambda/I}(M)\leq b_\Lambda(M).

(b) If P1=P∞P_1=P_\infty, then, for every M∈eΛe-modM\in e\Lambda e\text{-mod},

beΛe(M)=bΛ(Λe⊗eΛeM).b_{e\Lambda e}(M)=b_\Lambda(\Lambda e\otimes_{e\Lambda e}M).

Proof. (a) Let L∈Λ/I-modL\in\Lambda/I\text{-mod} satisfy pΛ/I(M,L)<∞p_{\Lambda/I}(M,L)<\infty. Since II is a strong idempotent ideal, Definition 2.4 gives

Ext⁡Λ/Ii(M,L)≅Ext⁡Λi(M,L)for all i≥0.\operatorname{Ext}^{i}_{\Lambda/I}(M,L)\cong\operatorname{Ext}^{i}_{\Lambda}(M,L)\qquad\text{for all }i\geq0.

Consequently, pΛ/I(M,L)=pΛ(M,L)≤bΛ(M)p_{\Lambda/I}(M,L)=p_\Lambda(M,L)\leq b_\Lambda(M). Taking the supremum over all such LL, we obtain bΛ/I(M)≤bΛ(M)b_{\Lambda/I}(M)\leq b_\Lambda(M).

(b) Let M∈eΛe-modM\in e\Lambda e\text{-mod} and set M~=Λe⊗eΛeM\widetilde{M}=\Lambda e\otimes_{e\Lambda e}M. Recall that the functor

Hom⁡Λ(Λe,−):P1⟶eΛe-mod\operatorname{Hom}_\Lambda(\Lambda e,-):P_1\longrightarrow e\Lambda e\text{-mod}

is an equivalence with quasi-inverse Λe⊗eΛe−\Lambda e\otimes_{e\Lambda e}- . Thus M~∈P1=P∞\widetilde{M}\in P_1=P_\infty. By [1 Theorem 3.2(c)], for every Y∈Λ-modY\in\Lambda\text{-mod} and every i≥0i\geq0, there are natural isomorphisms

Ext⁡Λi(M~,Y)≅Ext⁡eΛei(Hom⁡Λ(Λe,M~),Hom⁡Λ(Λe,Y))\operatorname{Ext}^{i}_{\Lambda}(\widetilde{M},Y)\cong\operatorname{Ext}^{i}_{e\Lambda e}\bigl(\operatorname{Hom}_{\Lambda}(\Lambda e,\widetilde{M}),\operatorname{Hom}_{\Lambda}(\Lambda e,Y)\bigr)
≅Ext⁡eΛei(M,eY).\cong\operatorname{Ext}^{i}_{e\Lambda e}(M,eY).

Consequently, pΛ(M~,Y)=peΛe(M,eY)p_\Lambda(\widetilde{M},Y)=p_{e\Lambda e}(M,eY) whenever either side is finite. It follows that

bΛ(M~)≤beΛe(M).b_\Lambda(\widetilde{M})\leq b_{e\Lambda e}(M).

Conversely, let L∈eΛeL \in e\Lambda e-mod and put L~=Λe⊗eΛeL\widetilde{L}=\Lambda e\otimes_{e\Lambda e}L. Since Hom⁡Λ(Λe,L~)≃L\operatorname{Hom}_{\Lambda}(\Lambda e,\widetilde{L})\simeq L, the preceding Ext isomorphism gives Ext⁡Λi(M~,L~)≃Ext⁡eΛei(M,L)\operatorname{Ext}^{i}_{\Lambda}(\widetilde{M},\widetilde{L})\simeq\operatorname{Ext}^{i}_{e\Lambda e}(M,L) for every i≥0i\geq0. Hence, peΛe(M,L)=pΛ(M~,L~)p_{e\Lambda e}(M,L)=p_{\Lambda}(\widetilde{M},\widetilde{L}) whenever either side is finite. Therefore, beΛe(M)≤bΛ(M~)b_{e\Lambda e}(M)\leq b_{\Lambda}(\widetilde{M}). Combining the two inequalities, we conclude that beΛe(M)=bΛ(Λe⊗eΛeM)b_{e\Lambda e}(M)=b_{\Lambda}(\Lambda e\otimes_{e\Lambda e}M).

We next refine the comparison in Lemma 3.1(a) by establishing an upper bound in the opposite direction.

Proposition 3.2. Suppose that I=ΛeΛI=\Lambda e\Lambda is a strong idempotent ideal and r=pd⁡Λ(Λ/I)<∞r=\operatorname{pd}_{\Lambda}(\Lambda/I)<\infty. Then, for every N∈Λ/IN\in\Lambda/I-mod, regarded also as a left Λ\Lambda-module by restriction of scalars, one has

bΛ/I(N)≤bΛ(N)≤r+bΛ/I(N).b_{\Lambda/I}(N)\leq b_{\Lambda}(N)\leq r+b_{\Lambda/I}(N).

In particular, bΛ/I(N)<∞b_{\Lambda/I}(N)<\infty if and only if bΛ(N)<∞b_{\Lambda}(N)<\infty, and

fab⁡(Λ/I)≤fab⁡(Λ).\operatorname{fab}(\Lambda/I)\leq\operatorname{fab}(\Lambda).

Proof. The first inequality follows from Lemma 3.1(a). For the second, it suffices to consider t=bΛ/I(N)<∞t=b_{\Lambda/I}(N)<\infty. Let L∈ΛL\in\Lambda-mod satisfy pΛ(N,L)<∞p_{\Lambda}(N,L)<\infty, and put Z=ΩΛrLZ=\Omega_{\Lambda}^{r}L, with Z=LZ=L when r=0r=0. For every j≥1j\geq1, dimension shifting gives

Ext⁡Λj(Λ/I,Z)≃Ext⁡Λj+r(Λ/I,L)=0.\operatorname{Ext}^{j}_{\Lambda}(\Lambda/I,Z)\simeq\operatorname{Ext}^{j+r}_{\Lambda}(\Lambda/I,L)=0.

Thus Z∈TZ\in\mathcal{T}, and Lemma 2.6(a) yields a gluing sequence

0⟶U=τΛ/I(Z)⟶Z⟶V⟶0,0\longrightarrow U=\tau_{\Lambda/I}(Z)\longrightarrow Z\longrightarrow V\longrightarrow0,

where U∈Λ/IU\in\Lambda/I-mod and V∈I∞V\in\mathbb{I}_{\infty}. By [1 Proposition 2.6], Ext⁡Λj(N,V)=0\operatorname{Ext}^{j}_{\Lambda}(N,V)=0 for every j≥0j\geq0. The long exact Ext sequence, strong idempotence and dimension shifting therefore give

Ext⁡Λ/Ii(N,U)≃Ext⁡Λi(N,Z)≃Ext⁡Λi+r(N,L)(i≥1).\operatorname{Ext}^{i}_{\Lambda/I}(N,U)\simeq\operatorname{Ext}^{i}_{\Lambda}(N,Z)\simeq\operatorname{Ext}^{i+r}_{\Lambda}(N,L)\qquad(i\geq1).

Consequently, pΛ/I(N,U)<∞p_{\Lambda/I}(N,U)<\infty, so pΛ/I(N,U)≤tp_{\Lambda/I}(N,U)\leq t. For every k>r+tk>r+t, taking i=k−r>ti=k-r>t in the preceding isomorphisms gives Ext⁡Λk(N,L)=0\operatorname{Ext}^{k}_{\Lambda}(N,L)=0. Hence pΛ(N,L)≤r+tp_{\Lambda}(N,L)\leq r+t. Taking the supremum over all such LL proves bΛ(N)≤r+bΛ/I(N)b_{\Lambda}(N)\leq r+b_{\Lambda/I}(N). In particular, if bΛ/I(N)<∞b_{\Lambda/I}(N)<\infty, then

bΛ/I(N)≤bΛ(N)≤fab⁡(Λ).b_{\Lambda/I}(N)\leq b_{\Lambda}(N)\leq\operatorname{fab}(\Lambda).

Taking the supremum over all such NN proves the final assertion.

Lemma 3.3. Let ee be an idempotent element of Λ\Lambda and I=ΛeΛI=\Lambda e\Lambda. Suppose that pd⁡ΛI=n<∞\operatorname{pd}_{\Lambda}I=n<\infty. Then, for every M∈ΛM\in\Lambda-mod, the following statements hold.

(a) If gldim⁡(Λ/I)=l<∞\operatorname{gldim}(\Lambda/I)=l<\infty and P1=P∞\mathbb{P}_{1}=\mathbb{P}_{\infty}, then

bΛ(M)≤max⁡{beΛe(eM),n+l+2}.b_{\Lambda}(M)\leq\max\{b_{e\Lambda e}(eM),n+l+2\}.

(b) Suppose that pd⁡Λop⁡I=m<∞\operatorname{pd}_{\Lambda^{\operatorname{op}}}I=m<\infty. If gldim⁡eΛe=l<∞\operatorname{gldim}e\Lambda e=l<\infty and II is a strong idempotent ideal, then

bΛ(M)≤m+n+2+max⁡{l+1,bΛ/I(ΩΛm+1MτI(ΩΛm+1M))}.b_{\Lambda}(M)\leq m+n+2+\max\left\{l+1,b_{\Lambda/I}\left(\frac{\Omega_{\Lambda}^{m+1}M}{\tau_{I}(\Omega_{\Lambda}^{m+1}M)}\right)\right\}.

Proof. (a) Proposition 2.2(e) gives an exact sequence

0⟶i∗(B)⟶Λe⊗eΛeeM⟶M⟶i∗((Λ/I)⊗ΛM)⟶0,0 \longrightarrow i_*(B) \longrightarrow\Lambda e \otimes_{e\Lambda e} eM \longrightarrow M \longrightarrow i_*\bigl((\Lambda/I)\otimes_\Lambda M\bigr) \longrightarrow0,

where B∈Λ/IB \in\Lambda/I-mod. Since pd⁡Λ(Λ/I)≤n+1\operatorname{pd}_{\Lambda}(\Lambda/I) \le n+1 and gldim⁡(Λ/I)=l\operatorname{gldim}(\Lambda/I)=l, both outer nonzero terms have projective dimension at most n+l+1n+l+1 over Λ\Lambda. Splitting this sequence into two short exact sequences and applying the long exact Ext sequences yields

Ext⁡Λk(M,−)≃Ext⁡Λk(Λe⊗eΛeeM,−)for all k>n+l+2.\operatorname{Ext}^{k}_{\Lambda}(M,-) \simeq\operatorname{Ext}^{k}_{\Lambda}(\Lambda e\otimes_{e\Lambda e}eM,-) \qquad\text{for all } k>n+l+2.

Put r=beΛe(eM)r=b_{e\Lambda e}(eM). If r=∞r=\infty, the asserted inequality is immediate. Suppose that r<∞r<\infty. Since Λe⊗eΛeeM∈P1=P∞\Lambda e\otimes_{e\Lambda e}eM\in\mathcal{P}_1=\mathcal{P}_\infty, it follows from [1 Theorem 3.2(c)] that Ext⁡Λi(Λe⊗eΛeeM,L)≃Ext⁡eΛei(eM,eL)\operatorname{Ext}^{i}_{\Lambda}(\Lambda e\otimes_{e\Lambda e}eM,L)\simeq\operatorname{Ext}^{i}_{e\Lambda e}(eM,eL) for every L∈ΛL\in\Lambda-mod and every i≥0i\ge0. Now let L∈ΛL\in\Lambda-mod satisfy pΛ(M,L)<∞p_{\Lambda}(M,L)<\infty. Since Ext⁡Λk(M,L)≃Ext⁡Λk(Λe⊗eΛeeM,L)\operatorname{Ext}^{k}_{\Lambda}(M,L)\simeq\operatorname{Ext}^{k}_{\Lambda}(\Lambda e\otimes_{e\Lambda e}eM,L) for all k>n+l+2k>n+l+2, it follows that peΛe(eM,eL)<∞p_{e\Lambda e}(eM,eL)<\infty. Hence peΛe(eM,eL)≤rp_{e\Lambda e}(eM,eL)\le r. Therefore,

Ext⁡Λk(M,L)=0for every k>max⁡{r,n+l+2}.\operatorname{Ext}^{k}_{\Lambda}(M,L)=0 \qquad\text{for every } k>\max\{r,n+l+2\}.

Thus pΛ(M,L)≤max⁡{r,n+l+2}p_{\Lambda}(M,L)\le\max\{r,n+l+2\}. Since LL was arbitrary, we obtain

bΛ(M)≤max⁡{beΛe(eM),n+l+2}.b_{\Lambda}(M)\le\max\{b_{e\Lambda e}(eM),n+l+2\}.

(b) Set X=ΩΛm+1MτI(ΩΛm+1M)X=\dfrac{\Omega_{\Lambda}^{m+1}M}{\tau_I(\Omega_{\Lambda}^{m+1}M)} and t=bΛ/I(X)t=b_{\Lambda/I}(X). If t=∞t=\infty, the asserted inequality is immediate. Hence we may assume that t<∞t<\infty. Since pd⁡Λop⁡(Λ/I)≤m+1\operatorname{pd}_{\Lambda^{\operatorname{op}}}(\Lambda/I)\le m+1, dimension shifting gives ΩΛm+1M∈T~\Omega_{\Lambda}^{m+1}M\in\widetilde{\mathcal{T}}. Lemma 2.6(b) therefore yields a gluing sequence

0⟶τI(ΩΛm+1M)⟶ΩΛm+1M⟶X⟶00\longrightarrow\tau_I(\Omega_{\Lambda}^{m+1}M)\longrightarrow\Omega_{\Lambda}^{m+1}M\longrightarrow X\longrightarrow0

with τI(ΩΛm+1M)∈P∞\tau_I(\Omega_{\Lambda}^{m+1}M)\in\mathcal{P}_{\infty} and X∈Λ/IX\in\Lambda/I-mod. Since gldim⁡(eΛe)=l\operatorname{gldim}(e\Lambda e)=l, we obtain

pd⁡ΛτI(ΩΛm+1M)=pd⁡eΛeHom⁡Λ(Λe,τI(ΩΛm+1M))≤l\operatorname{pd}_{\Lambda}\tau_I(\Omega_{\Lambda}^{m+1}M)=\operatorname{pd}_{e\Lambda e}\operatorname{Hom}_{\Lambda}(\Lambda e,\tau_I(\Omega_{\Lambda}^{m+1}M))\le l

by [1 Corollary 3.3(a)]. Hence Ext⁡Λk(ΩΛm+1M,−)≃Ext⁡Λk(X,−)\operatorname{Ext}^{k}_{\Lambda}(\Omega_{\Lambda}^{m+1}M,-)\simeq\operatorname{Ext}^{k}_{\Lambda}(X,-) for all k>l+1k>l+1. Take any L∈ΛL\in\Lambda-mod with pΛ(M,L)<∞p_{\Lambda}(M,L)<\infty. Since pd⁡ΛI=n\operatorname{pd}_{\Lambda}I=n, the (n+1)(n+1)-th cosyzygy ΩΛ−n−1L\Omega_{\Lambda}^{-n-1}L belongs to T\mathcal{T}. By Lemma 2.6(a), there is a gluing sequence

0⟶τΛ/I(ΩΛ−n−1L)⟶ΩΛ−n−1L⟶ΩΛ−n−1LτΛ/I(ΩΛ−n−1L)⟶0,0\longrightarrow\tau_{\Lambda/I}(\Omega_{\Lambda}^{-n-1}L)\longrightarrow\Omega_{\Lambda}^{-n-1}L\longrightarrow\frac{\Omega_{\Lambda}^{-n-1}L}{\tau_{\Lambda/I}(\Omega_{\Lambda}^{-n-1}L)}\longrightarrow0,

where τΛ/I(ΩΛ−n−1L)∈Λ/I\tau_{\Lambda/I}(\Omega_{\Lambda}^{-n-1}L)\in\Lambda/I-mod and ΩΛ−n−1LτΛ/I(ΩΛ−n−1L)∈I∞\dfrac{\Omega_{\Lambda}^{-n-1}L}{\tau_{\Lambda/I}(\Omega_{\Lambda}^{-n-1}L)}\in\mathcal{I}_{\infty}. Set

U=τΛ/I(ΩΛ−n−1L)andV=ΩΛ−n−1LτΛ/I(ΩΛ−n−1L).U=\tau_{\Lambda/I}(\Omega_{\Lambda}^{-n-1}L) \qquad\text{and}\qquad V=\frac{\Omega_{\Lambda}^{-n-1}L}{\tau_{\Lambda/I}(\Omega_{\Lambda}^{-n-1}L)}.

Thus, the preceding gluing sequence has the form

0⟶U⟶ΩΛ−n−1L⟶V⟶0.0\longrightarrow U\longrightarrow\Omega_{\Lambda}^{-n-1}L\longrightarrow V\longrightarrow0.

Since X∈Λ/IX\in\Lambda/I-mod and V∈I∞V\in\mathcal{I}_{\infty}, it follows from [1 Proposition 2.6] that Ext⁡Λi(X,V)=0\operatorname{Ext}^{i}_{\Lambda}(X,V)=0 for all i≥0i\ge0. Applying Hom⁡Λ(X,−)\operatorname{Hom}_{\Lambda}(X,-) to the above short exact sequence therefore yields

Ext⁡Λi(X,U)≃Ext⁡Λi(X,ΩΛ−n−1L)for all i≥0.\operatorname{Ext}^{i}_{\Lambda}(X,U)\simeq\operatorname{Ext}^{i}_{\Lambda}(X,\Omega_{\Lambda}^{-n-1}L) \qquad\text{for all } i\ge0.

Since II is a strong idempotent ideal and X,U∈Λ/IX,U \in\Lambda/I-mod, Definition 2.4 gives

Ext⁡Λi(X,U)≃Ext⁡Λ/Ii(X,U)(i≥0).\operatorname{Ext}_{\Lambda}^{i}(X,U) \simeq\operatorname{Ext}_{\Lambda/I}^{i}(X,U) \qquad(i \ge0).

We next show that pΛ/I(X,U)p_{\Lambda/I}(X,U) is finite. Put T=τI(ΩΛm+1M)T=\tau_I(\Omega_{\Lambda}^{m+1}M). As shown above, pd⁡ΛT≤l\operatorname{pd}_{\Lambda}T \le l. From the gluing sequence

0⟶T⟶ΩΛm+1M⟶X⟶00 \longrightarrow T \longrightarrow\Omega_{\Lambda}^{m+1}M \longrightarrow X \longrightarrow0

we obtain Ext⁡Λi(X,ΩΛ−n−1L)≃Ext⁡Λi(ΩΛm+1M,ΩΛ−n−1L)\operatorname{Ext}_{\Lambda}^{i}(X,\Omega_{\Lambda}^{-n-1}L) \simeq\operatorname{Ext}_{\Lambda}^{i}(\Omega_{\Lambda}^{m+1}M,\Omega_{\Lambda}^{-n-1}L) for all i>l+1i>l+1. By dimension shifting, for every i≥1i \ge1,

Ext⁡Λi(ΩΛm+1M,ΩΛ−n−1L)≃Ext⁡Λi+m+n+2(M,L).\operatorname{Ext}_{\Lambda}^{i}(\Omega_{\Lambda}^{m+1}M,\Omega_{\Lambda}^{-n-1}L) \simeq\operatorname{Ext}_{\Lambda}^{i+m+n+2}(M,L).

Since pΛ(M,L)<∞p_{\Lambda}(M,L)<\infty, the right-hand side vanishes for all sufficiently large ii. Consequently,

pΛ/I(X,U)<∞.p_{\Lambda/I}(X,U)<\infty.

By the definition of tt, we obtain

pΛ/I(X,U)≤t.p_{\Lambda/I}(X,U)\le t.

It follows that

Ext⁡Λi(ΩΛm+1M,ΩΛ−n−1L)=0for all i>max⁡{l+1,t}.\operatorname{Ext}_{\Lambda}^{i}(\Omega_{\Lambda}^{m+1}M,\Omega_{\Lambda}^{-n-1}L)=0 \qquad\text{for all } i>\max\{l+1,t\}.

Another application of dimension shifting gives

Ext⁡Λq(M,L)=0for all q>m+n+2+max⁡{l+1,t}.\operatorname{Ext}_{\Lambda}^{q}(M,L)=0 \qquad\text{for all } q>m+n+2+\max\{l+1,t\}.

Therefore, pΛ(M,L)≤m+n+2+max⁡{l+1,t}p_{\Lambda}(M,L)\le m+n+2+\max\{l+1,t\}. Since LL was arbitrary, we conclude that

bΛ(M)≤m+n+2+max⁡{l+1,t}<∞.b_{\Lambda}(M)\le m+n+2+\max\{l+1,t\}<\infty.

We now combine the preceding comparisons to obtain a reduction theorem for the quotient algebra Λ/I\Lambda/I. Under suitable homological finiteness assumptions, the global and finitistic Auslander bounds of Λ\Lambda can be controlled by the corresponding bounds of the quotient. In particular, these estimates yield an equivalence of finiteness for the finitistic Auslander bounds.

Theorem 3.4. Let ee be an idempotent element of Λ\Lambda and let m,n,l∈Nm,n,l \in\mathbb{N}. Suppose that II is a strong idempotent ideal. Then the following statements hold.

(a) gb⁡(Λ/I)≤gb⁡(Λ)\operatorname{gb}(\Lambda/I)\le\operatorname{gb}(\Lambda).

(b) If gl.dim⁡Λe=l\operatorname{gl.dim}\Lambda e=l, pd⁡ΛI=n\operatorname{pd}_{\Lambda}I=n and pd⁡Λop⁡I=m\operatorname{pd}_{\Lambda^{\operatorname{op}}}I=m, then

gb⁡(Λ)≤m+n+2+max⁡{l+1,gb⁡(Λ/I)}and\operatorname{gb}(\Lambda)\le m+n+2+\max\{l+1,\operatorname{gb}(\Lambda/I)\} \qquad\text{and}
fab⁡(Λ)≤m+n+2+max⁡{l+1,fab⁡(Λ/I)}.\operatorname{fab}(\Lambda)\le m+n+2+\max\{l+1,\operatorname{fab}(\Lambda/I)\}.

Moreover, we have fab⁡(Λ/I)≤fab⁡(Λ)\operatorname{fab}(\Lambda/I)\le\operatorname{fab}(\Lambda). Consequently,

fab⁡(Λ)<∞⟺fab⁡(Λ/I)<∞.\operatorname{fab}(\Lambda)<\infty\quad\Longleftrightarrow\quad\operatorname{fab}(\Lambda/I)<\infty.

Proof. (a) Let M∈Λ/IM \in\Lambda/I-mod. By Lemma 3.1(a), bΛ/I(M)≤bΛ(M)≤gb⁡(Λ)b_{\Lambda/I}(M) \le b_{\Lambda}(M) \le\operatorname{gb}(\Lambda). Taking the supremum over all M∈Λ/IM \in\Lambda/I-mod, we obtain gb⁡(Λ/I)≤gb⁡(Λ)\operatorname{gb}(\Lambda/I) \le\operatorname{gb}(\Lambda).

(b) Suppose now that gldim⁡Λe=l<∞\operatorname{gldim}\Lambda e=l<\infty, pd⁡ΛI=n<∞\operatorname{pd}_{\Lambda}I=n<\infty, and pd⁡Λop⁡I=m<∞\operatorname{pd}_{\Lambda^{\operatorname{op}}}I=m<\infty. For every M∈ΛM\in\Lambda-mod, put

XM=ΩΛm+1MτI(ΩΛm+1M).X_M=\frac{\Omega_{\Lambda}^{m+1}M}{\tau_I(\Omega_{\Lambda}^{m+1}M)}.

By Lemma 3.3(b), bΛ(M)≤m+n+2+max⁡{l+1,bΛ/I(XM)}b_{\Lambda}(M)\le m+n+2+\max\{l+1,b_{\Lambda/I}(X_M)\}. Since bΛ/I(XM)≤gb⁡(Λ/I)b_{\Lambda/I}(X_M)\le\operatorname{gb}(\Lambda/I), we have bΛ(M)≤m+n+2+max⁡{l+1,gb⁡(Λ/I)}b_{\Lambda}(M)\le m+n+2+\max\{l+1,\operatorname{gb}(\Lambda/I)\}. Taking the supremum over all M∈ΛM\in\Lambda-mod gives

gb⁡(Λ)≤m+n+2+max⁡{l+1,gb⁡(Λ/I)}.\operatorname{gb}(\Lambda)\le m+n+2+\max\{l+1,\operatorname{gb}(\Lambda/I)\}.

We next prove the corresponding assertion for the finitistic Auslander bound. Let M∈ΛM\in\Lambda-mod satisfy bΛ(M)<∞b_{\Lambda}(M)<\infty, and write X=ΩΛm+1MτI(ΩΛm+1M)X=\frac{\Omega_{\Lambda}^{m+1}M}{\tau_I(\Omega_{\Lambda}^{m+1}M)}. We claim that bΛ/I(X)<∞b_{\Lambda/I}(X)<\infty. Indeed, let U∈Λ/IU\in\Lambda/I-mod satisfy pΛ/I(X,U)<∞p_{\Lambda/I}(X,U)<\infty. Since II is a strong idempotent ideal, Definition 2.4 gives pΛ(X,U)=pΛ/I(X,U)<∞p_{\Lambda}(X,U)=p_{\Lambda/I}(X,U)<\infty. Put T=τI(ΩΛm+1M)T=\tau_I(\Omega_{\Lambda}^{m+1}M). As in the proof of Lemma 3.3(b), one has pd⁡ΛT≤l\operatorname{pd}_{\Lambda}T\le l. Thus the exact sequence

0⟶T⟶ΩΛm+1M⟶X⟶00\longrightarrow T\longrightarrow\Omega_{\Lambda}^{m+1}M\longrightarrow X\longrightarrow0

induces isomorphisms Ext⁡Λi(X,U)≅Ext⁡Λi(ΩΛm+1M,U)≅Ext⁡Λi+m+1(M,U)\operatorname{Ext}_{\Lambda}^{i}(X,U)\cong\operatorname{Ext}_{\Lambda}^{i}(\Omega_{\Lambda}^{m+1}M,U)\cong\operatorname{Ext}_{\Lambda}^{i+m+1}(M,U) for every i>l+1i>l+1. It follows from pΛ(X,U)<∞p_{\Lambda}(X,U)<\infty that pΛ(M,U)<∞p_{\Lambda}(M,U)<\infty. Hence pΛ(M,U)≤bΛ(M)p_{\Lambda}(M,U)\le b_{\Lambda}(M). The preceding Ext isomorphisms now imply pΛ/I(X,U)≤max⁡{l+1,bΛ(M)}p_{\Lambda/I}(X,U)\le\max\{l+1,b_{\Lambda}(M)\}. Taking the supremum over all such UU, we obtain bΛ/I(X)≤max⁡{l+1,bΛ(M)}<∞b_{\Lambda/I}(X)\le\max\{l+1,b_{\Lambda}(M)\}<\infty. Therefore, bΛ/I(X)≤fab⁡(Λ/I)b_{\Lambda/I}(X)\le\operatorname{fab}(\Lambda/I). Applying Lemma 3.3(b) once more yields bΛ(M)≤m+n+2+max⁡{l+1,fab⁡(Λ/I)}b_{\Lambda}(M)\le m+n+2+\max\{l+1,\operatorname{fab}(\Lambda/I)\}. Taking the supremum over all MM satisfying bΛ(M)<∞b_{\Lambda}(M)<\infty, we conclude that

fab⁡(Λ)≤m+n+2+max⁡{l+1,fab⁡(Λ/I)}.\operatorname{fab}(\Lambda)\le m+n+2+\max\{l+1,\operatorname{fab}(\Lambda/I)\}.

Finally, the exact sequence 0→I→Λ→Λ/I→00\to I\to\Lambda\to\Lambda/I\to0 gives pd⁡Λ(Λ/I)≤n+1<∞\operatorname{pd}_{\Lambda}(\Lambda/I)\le n+1<\infty. Proposition 3.2 therefore yields fab⁡(Λ/I)≤fab⁡(Λ)\operatorname{fab}(\Lambda/I)\le\operatorname{fab}(\Lambda). Together with the upper bound just proved, this establishes the asserted equivalence of finiteness.

We next turn to the corner algebra eΛee\Lambda e and obtain the complementary reduction theorem.

Theorem 3.5. Let ee be an idempotent element of Λ\Lambda and let n,l∈Nn,l\in\mathbb{N}. Suppose that P1=P∞P_1=P_\infty. Then the following statements hold.

(a) gb⁡(eΛe)≤gb⁡(Λ)\operatorname{gb}(e\Lambda e)\le\operatorname{gb}(\Lambda) and fab⁡(eΛe)≤fab⁡(Λ)\operatorname{fab}(e\Lambda e)\le\operatorname{fab}(\Lambda).

(b) If gldim⁡(Λ/I)=l<∞\operatorname{gldim}(\Lambda/I)=l<\infty, pd⁡ΛI=n\operatorname{pd}_{\Lambda}I=n, then

gb⁡(Λ)≤max⁡{n+l+2,gb⁡(eΛe)}andfab⁡(Λ)≤max⁡{n+l+2,fab⁡(eΛe)}.\operatorname{gb}(\Lambda)\le\max\{n+l+2,\operatorname{gb}(e\Lambda e)\}\qquad\text{and}\qquad\operatorname{fab}(\Lambda)\le\max\{n+l+2,\operatorname{fab}(e\Lambda e)\}.

Proof. (a) For every M∈eΛeM\in e\Lambda e-mod, Lemma 3.1(b) gives beΛe(M)=bΛ(Λe⊗eΛeM)≤gb⁡(Λ)b_{e\Lambda e}(M)=b_{\Lambda}(\Lambda e\otimes_{e\Lambda e}M)\le\operatorname{gb}(\Lambda). Taking the supremum over all M∈eΛeM\in e\Lambda e-mod gives gb⁡(eΛe)≤gb⁡(Λ)\operatorname{gb}(e\Lambda e)\le\operatorname{gb}(\Lambda).

We next prove the corresponding inequality for the finitistic Auslander bound. If beΛe(M)<∞b_{e\Lambda e}(M)<\infty, then bΛ(Λe⊗eΛeM)<∞b_{\Lambda}(\Lambda e\otimes_{e\Lambda e}M)<\infty and beΛe(M)=bΛ(Λe⊗eΛeM)≤fab⁡(Λ)b_{e\Lambda e}(M)=b_{\Lambda}(\Lambda e\otimes_{e\Lambda e}M)\le\operatorname{fab}(\Lambda). Taking the supremum over all MM satisfying beΛe(M)<∞b_{e\Lambda e}(M)<\infty gives fab⁡(eΛe)≤fab⁡(Λ)\operatorname{fab}(e\Lambda e)\le\operatorname{fab}(\Lambda).

(b) Suppose further that gldim⁡(Λ/I)=l<∞\operatorname{gldim}(\Lambda/I)=l<\infty and pd⁡ΛI=n<∞\operatorname{pd}_{\Lambda}I=n<\infty. By Lemma 3.3(a), for every M∈ΛM\in\Lambda-mod, bΛ(M)≤max⁡{n+l+2,beΛe(eM)}b_{\Lambda}(M)\le\max\{n+l+2,b_{e\Lambda e}(eM)\}. Since beΛe(eM)≤gb⁡(eΛe)b_{e\Lambda e}(eM)\le\operatorname{gb}(e\Lambda e), taking the supremum over all MM gives gb⁡(Λ)≤max⁡{n+l+2,gb⁡(eΛe)}\operatorname{gb}(\Lambda)\le\max\{n+l+2,\operatorname{gb}(e\Lambda e)\}. Finally, let M∈ΛM \in\Lambda-mod satisfy bΛ(M)<∞b_{\Lambda}(M) < \infty. We first show that beΛe(eM)<∞b_{e\Lambda e}(eM) < \infty. Let Y∈eΛeY \in e\Lambda e-mod satisfy peΛe(eM,Y)<∞p_{e\Lambda e}(eM,Y) < \infty, and put Y~=Λe⊗eΛeY\widetilde{Y}=\Lambda e\otimes_{e\Lambda e}Y. Since P1=P∞P_1=P_{\infty},

Ext⁡Λi(Λe⊗eΛeeM,Y~)≅Ext⁡eΛei(eM,Y)\operatorname{Ext}^{i}_{\Lambda}(\Lambda e\otimes_{e\Lambda e}eM,\widetilde{Y}) \cong\operatorname{Ext}^{i}_{e\Lambda e}(eM,Y)

for every i≥0i \ge0. Moreover, the proof of Lemma 3.3(a) gives

Ext⁡Λi(M,Y~)≅Ext⁡Λi(Λe⊗eΛeeM,Y~)\operatorname{Ext}^{i}_{\Lambda}(M,\widetilde{Y}) \cong\operatorname{Ext}^{i}_{\Lambda}(\Lambda e\otimes_{e\Lambda e}eM,\widetilde{Y})

for every i>n+l+2i > n+l+2. It follows from peΛe(eM,Y)<∞p_{e\Lambda e}(eM,Y) < \infty that pΛ(M,Y~)<∞p_{\Lambda}(M,\widetilde{Y}) < \infty. Therefore, pΛ(M,Y~)≤bΛ(M)p_{\Lambda}(M,\widetilde{Y}) \le b_{\Lambda}(M), and the preceding isomorphisms imply peΛe(eM,Y)≤max⁡{n+l+2,bΛ(M)}p_{e\Lambda e}(eM,Y) \le\max\{n+l+2,b_{\Lambda}(M)\}. Taking the supremum over all such YY, we obtain beΛe(eM)≤max⁡{n+l+2,bΛ(M)}<∞b_{e\Lambda e}(eM) \le\max\{n+l+2,b_{\Lambda}(M)\} < \infty. Hence beΛe(eM)≤fab⁡(eΛe)b_{e\Lambda e}(eM) \le\operatorname{fab}(e\Lambda e). Applying Lemma 3.3(a), we get bΛ(M)≤max⁡{n+l+2,fab⁡(eΛe)}b_{\Lambda}(M) \le\max\{n+l+2,\operatorname{fab}(e\Lambda e)\}. Taking the supremum over all MM for which bΛ(M)<∞b_{\Lambda}(M) < \infty gives

fab⁡(Λ)≤max⁡{n+l+2,fab⁡(eΛe)}.\operatorname{fab}(\Lambda) \le\max\{n+l+2,\operatorname{fab}(e\Lambda e)\}.

The following examples illustrate the two complementary reduction mechanisms in Theorems 3.4 and 3.5.

Example 3.6. Let Λ=kQ/⟨ab,aε,εb,ε2⟩\Lambda=kQ/\langle ab,a\varepsilon,\varepsilon b,\varepsilon^2\rangle, where QQ is the quiver

1→a2,2→b1,2↬ε1 \xrightarrow{a} 2,\qquad2 \xrightarrow{b} 1,\qquad2 \looparrowright^{\varepsilon}

Throughout this example, paths are composed from left to right; thus abab means first traversing aa and then bb. Put e=e1e=e_1 and I=ΛeΛI=\Lambda e\Lambda. Then eΛe≅ke\Lambda e\cong k, and Λ/I≅D:=k[ε]/(ε2)\Lambda/I\cong D:=k[\varepsilon]/(\varepsilon^2). Since I=⟨e1,a,b,ba⟩kI=\langle e_1,a,b,ba\rangle_k, we have ΛI≅(Λe)⊕2\Lambda I\cong(\Lambda e)^{\oplus2}, and IΛ≅(eΛ)⊕2I\Lambda\cong(e\Lambda)^{\oplus2}. Thus II is projective on both sides. By [1 Proposition 5.2], II is a strong idempotent ideal and P1=P∞P_1=P_{\infty}. We claim that gb⁡(D)=fab⁡(D)=0\operatorname{gb}(D)=\operatorname{fab}(D)=0. In fact, let S=D/(ε)S=D/(\varepsilon). Every finitely generated DD-module is a direct sum of copies of DD and SS. The periodic projective resolution

⋯→εD→εD→εD⟶S⟶0\cdots\xrightarrow{\varepsilon} D \xrightarrow{\varepsilon} D \xrightarrow{\varepsilon} D \longrightarrow S \longrightarrow0

gives Ext⁡Di(S,S)≅k\operatorname{Ext}^{i}_{D}(S,S)\cong k and Ext⁡Di(S,D)=0\operatorname{Ext}^{i}_{D}(S,D)=0 (i≥1)(i\ge1). Together with projectivity of DD, these identities show that eventual vanishing of Ext⁡Di(M,N)\operatorname{Ext}^{i}_{D}(M,N) implies vanishing for every i≥1i\ge1. Consequently,

gb⁡(D)=fab⁡(D)=0.\operatorname{gb}(D)=\operatorname{fab}(D)=0.

We can now apply Theorem 3.4(b) with m=n=l=0m=n=l=0. It gives

gb⁡(Λ)≤2+max⁡{1,gb⁡(D)}=3\operatorname{gb}(\Lambda)\le2+\max\{1,\operatorname{gb}(D)\}=3

and

fab⁡(Λ)≤2+max⁡{1,fab⁡(D)}=3.\operatorname{fab}(\Lambda)\le2+\max\{1,\operatorname{fab}(D)\}=3.

Clearly, gldim⁡Λ=∞\operatorname{gldim}\Lambda=\infty. The example therefore illustrates reduction of Auslander bounds through the quotient algebra, without finite global dimension of either Λ/I\Lambda/I or Λ\Lambda.

Example 3.7. Let Λ=kQ/⟨x2,ab⟩\Lambda= \mathsf{k}Q/\langle x^2,ab\rangle, where QQ is the quiver

xx ↺\circlearrowleft 1⇄ab21 \underset{b}{\overset{a}{\rightleftarrows}} 2.

Again, paths are composed from left to right. Put e=e1e=e_1 and I=ΛeΛI=\Lambda e\Lambda. Then eΛe≅D:=k[x]/(x2)e\Lambda e \cong D:=\mathsf{k}[x]/(x^2) and Λ/I≅k\Lambda/I\cong\mathsf{k}. Since I=⟨e1,x,a,xa,b,bx,ba,bxa⟩kI=\langle e_1,x,a,xa,b,bx,ba,bxa\rangle_{\mathsf{k}}, we have ΛI≅(Λe)⊕2{}_{\Lambda}I\cong(\Lambda e)^{\oplus2}, and IΛ≅(eΛ)⊕2I_{\Lambda}\cong(e\Lambda)^{\oplus2}. Thus II is projective on both sides. As in Example 3.6, we have P1=P∞P_1=P_{\infty} and gb⁡(D)=fab⁡(D)=0\operatorname{gb}(D)=\operatorname{fab}(D)=0. Applying Theorem 3.5(b) with n=l=0n=l=0, we have

gb⁡(Λ)≤max⁡{2,gb⁡(D)}=2andfab⁡(Λ)≤max⁡{2,fab⁡(D)}=2.\operatorname{gb}(\Lambda)\leq\max\{2,\operatorname{gb}(D)\}=2 \qquad\text{and}\qquad \operatorname{fab}(\Lambda)\leq\max\{2,\operatorname{fab}(D)\}=2.

We now show that both upper bounds are attained. Put Pi=ΛeiP_i=\Lambda e_i and Si=Pi/rad⁡PiS_i=P_i/\operatorname{rad}P_i. Right multiplication by aa gives an exact sequence

0⟶P1→⋅aP2⟶S2⟶0.0\longrightarrow P_1\xrightarrow{\cdot a}P_2\longrightarrow S_2\longrightarrow0.

Since Hom⁡Λ(P2,S1)=0\operatorname{Hom}_{\Lambda}(P_2,S_1)=0 and Hom⁡Λ(P1,S1)≅k\operatorname{Hom}_{\Lambda}(P_1,S_1)\cong\mathsf{k}, we obtain

Ext⁡Λ1(S2,S1)≅k.\operatorname{Ext}^{1}_{\Lambda}(S_2,S_1)\cong\mathsf{k}.

In particular, pd⁡ΛS2=1\operatorname{pd}_{\Lambda}S_2=1. Set X=P1/soc⁡P1X=P_1/\operatorname{soc}P_1. The exact sequence

0⟶S2⊕2⟶P1⟶X⟶00\longrightarrow S_2^{\oplus2}\longrightarrow P_1\longrightarrow X\longrightarrow0

shows that pd⁡ΛX=2\operatorname{pd}_{\Lambda}X=2, while dimension shifting gives

Ext⁡Λ2(X,S1)≅Ext⁡Λ1(S2⊕2,S1)≅k2.\operatorname{Ext}^{2}_{\Lambda}(X,S_1)\cong\operatorname{Ext}^{1}_{\Lambda}(S_2^{\oplus2},S_1)\cong\mathsf{k}^{2}.

Therefore pd⁡ΛX=bΛ(X)=2\operatorname{pd}_{\Lambda}X=b_{\Lambda}(X)=2. Combining this with the upper bounds proves

gb⁡(Λ)=fab⁡(Λ)=2.\operatorname{gb}(\Lambda)=\operatorname{fab}(\Lambda)=2.

Applying Theorems 3.4 and 3.5 to upper triangular matrix algebras, we obtain the following corollary.

Corollary 3.8. Let Λ=(AAUB0B)\Lambda=\begin{pmatrix}A&A{}_UB\\0&B\end{pmatrix}. Then

(a) max⁡{gb⁡(A),gb⁡(B)}≤gb⁡(Λ)\max\{\operatorname{gb}(A),\operatorname{gb}(B)\}\leq\operatorname{gb}(\Lambda) and fab⁡(A)≤fab⁡(Λ)\operatorname{fab}(A)\leq\operatorname{fab}(\Lambda). Moreover, if pd⁡BopU=0\operatorname{pd}_{B^{\mathrm{op}}}U=0 or gldim⁡A=l<∞\operatorname{gldim}A=l<\infty, then

max⁡{fab⁡(A),fab⁡(B)}≤fab⁡(Λ).\max\{\operatorname{fab}(A),\operatorname{fab}(B)\}\leq\operatorname{fab}(\Lambda).

(b) If gldim⁡A=l<∞\operatorname{gldim}A=l<\infty, then

gb⁡(Λ)≤max⁡{2l+3,gb⁡(B)+l+2}andfab⁡(Λ)≤max⁡{2l+3,fab⁡(B)+l+2}.\operatorname{gb}(\Lambda)\leq\max\{2l+3,\operatorname{gb}(B)+l+2\} \qquad\text{and}\qquad \operatorname{fab}(\Lambda)\leq\max\{2l+3,\operatorname{fab}(B)+l+2\}.

(c) If gldim⁡B=l<∞\operatorname{gldim}B=l<\infty, then

gb⁡(Λ)≤max⁡{2l+3,gb⁡(A)+l+2}andfab⁡(Λ)≤max⁡{2l+3,fab⁡(A)+l+2}.\operatorname{gb}(\Lambda)\leq\max\{2l+3,\operatorname{gb}(A)+l+2\} \qquad\text{and}\qquad \operatorname{fab}(\Lambda)\leq\max\{2l+3,\operatorname{fab}(A)+l+2\}.

Proof. Let e1=(1000)e_1=\begin{pmatrix}1&0\\0&0\end{pmatrix} and e2=(0001)e_2=\begin{pmatrix}0&0\\0&1\end{pmatrix}. Take e=e1e=e_1, and put I=ΛeΛI=\Lambda e\Lambda. Then eΛe≅Ae\Lambda e\cong A, Λ/I≅B\Lambda/I\cong B and I≅e1ΛI\cong e_1\Lambda. The ideal I=e1ΛI=e_1\Lambda is projective as a right Λ\Lambda-module. By [1 Proposition 5.2], it is a strong idempotent ideal and P1=P∞P_1=P_\infty for e=e1e=e_1. Theorems 3.4(a) and 3.5(a) therefore yield

max⁡{gb⁡(A),gb⁡(B)}≤gb⁡(Λ)andfab⁡(A)≤fab⁡(Λ).\max\{\operatorname{gb}(A),\operatorname{gb}(B)\}\le\operatorname{gb}(\Lambda)\qquad\text{and}\qquad\operatorname{fab}(A)\le\operatorname{fab}(\Lambda).

This proves the first part of (a). Suppose that gldim⁡A=l<∞\operatorname{gldim}A=l<\infty. We have pd⁡Λ(I)=pd⁡AU≤l\operatorname{pd}_{\Lambda}(I)=\operatorname{pd}_{A}U\le l. Since pd⁡Λop⁡I=0\operatorname{pd}_{\Lambda^{\operatorname{op}}}I=0, Theorem 3.4(b) yields

gb⁡(Λ)≤max⁡{2l+3,gb⁡(B)+l+2},\operatorname{gb}(\Lambda)\le\max\{2l+3,\operatorname{gb}(B)+l+2\},
fab⁡(Λ)≤max⁡{2l+3,fab⁡(B)+l+2}.\operatorname{fab}(\Lambda)\le\max\{2l+3,\operatorname{fab}(B)+l+2\}.

and fab⁡(B)≤fab⁡(Λ)\operatorname{fab}(B)\le\operatorname{fab}(\Lambda). This proves (b) and the second assertion of (a) when AA has finite global dimension.

Now take e=e2e=e_2, with I=ΛeΛI=\Lambda e\Lambda. Then eΛe≅Be\Lambda e\cong B, Λ/I≅A\Lambda/I\cong A and I=ΛeI=\Lambda e. As a right Λ\Lambda-module, II is the module supported on the BB-component with underlying BB-module U⊕BU\oplus B. Thus pd⁡Λop⁡(I)=pd⁡Bop⁡(U⊕B)\operatorname{pd}_{\Lambda^{\operatorname{op}}}(I)=\operatorname{pd}_{B^{\operatorname{op}}}(U\oplus B). If pd⁡Bop⁡U=0\operatorname{pd}_{B^{\operatorname{op}}}U=0, this ideal is projective on the right. By [1 Proposition 5.2], P1=P∞P_1=P_\infty for e=e2e=e_2, and Theorem 3.5(a) yields fab⁡(B)≤fab⁡(Λ)\operatorname{fab}(B)\le\operatorname{fab}(\Lambda). This proves the remaining assertion in (a). Finally, we prove (c). Since I=ΛeI=\Lambda e is projective as a left Λ\Lambda-module, we have pd⁡Λ(I)=0\operatorname{pd}_{\Lambda}(I)=0. The exact sequence of left Λ\Lambda-modules

0⟶Λe⟶Λ⟶Λ/I⟶00\longrightarrow\Lambda e\longrightarrow\Lambda\longrightarrow\Lambda/I\longrightarrow0

is a projective resolution of Λ/I\Lambda/I. For every Y∈Λ/I-modY\in\Lambda/I\text{-mod}, Hom⁡Λ(Λe,Y)≅eY=0\operatorname{Hom}_{\Lambda}(\Lambda e,Y)\cong eY=0. Hence

Ext⁡Λi(Λ/I,Y)=0for all i≥1,\operatorname{Ext}^{i}_{\Lambda}(\Lambda/I,Y)=0\qquad\text{for all }i\ge1,

so Proposition 2.5(b) shows that II is a strong idempotent ideal. Now, suppose that gldim⁡B=l<∞\operatorname{gldim}B=l<\infty. Since pd⁡Λop⁡(I)=pd⁡Bop⁡(U⊕B)≤l\operatorname{pd}_{\Lambda^{\operatorname{op}}}(I)=\operatorname{pd}_{B^{\operatorname{op}}}(U\oplus B)\le l, (c) follows from Theorem 3.4(b), with n=0n=0 and m≤lm\le l. □

The preceding corollary follows from the general recollement estimates. The triangular structure and Proposition 3.2 yield the following sharper bounds.

Proposition 3.9. Let Λ=(AAUB0B)\Lambda=\begin{pmatrix}A&AU_B\\0&B\end{pmatrix}. Then the following statements hold.

(a) Suppose that gldim⁡A=l<∞\operatorname{gldim}A=l<\infty, and put d=pd⁡AUd=\operatorname{pd}_{A}U, with d=0d=0 if U=0U=0. Then

gb⁡(Λ)≤max⁡{l+1,d+1+gb⁡(B)}≤l+1+gb⁡(B),\operatorname{gb}(\Lambda)\le\max\{l+1,d+1+\operatorname{gb}(B)\}\le l+1+\operatorname{gb}(B),
fab⁡(Λ)≤max⁡{l+1,d+1+fab⁡(B)}≤l+1+fab⁡(B).\operatorname{fab}(\Lambda)\le\max\{l+1,d+1+\operatorname{fab}(B)\}\le l+1+\operatorname{fab}(B).

(b) If gldim⁡B=l<∞\operatorname{gldim}B=l<\infty, then

gb⁡(Λ)≤l+1+gb⁡(A)andfab⁡(Λ)≤l+1+fab⁡(A).\operatorname{gb}(\Lambda)\le l+1+\operatorname{gb}(A)\qquad\text{and}\qquad\operatorname{fab}(\Lambda)\le l+1+\operatorname{fab}(A).

Consequently, in (a), finiteness of either bound for Λ\Lambda is equivalent to finiteness of the corresponding bound for BB; in (b), it is equivalent to finiteness of the corresponding bound for AA.

Proof. We write a left Λ\Lambda-module as M=(XY)ϕM=\begin{pmatrix}X\\Y\end{pmatrix}_{\phi}, where X∈AX\in A-mod, Y∈BY\in B-mod, and ϕ:U⊗BY→X\phi:U\otimes_B Y\to X is AA-linear. We omit the subscript when the structure map is clear. The functor F:AF:A-mod ⟶\longrightarrow Λ\Lambda-mod, F(X)=(X0)F(X)=\begin{pmatrix}X\\0\end{pmatrix}, is exact and preserves projectives. Applying FF to a projective resolution gives

Ext⁡Λi((X0),(X′Y′))≅Ext⁡Ai(X,X′)(i≥0).\operatorname{Ext}_{\Lambda}^{i}\left(\begin{pmatrix}X\\0\end{pmatrix},\begin{pmatrix}X'\\Y'\end{pmatrix}\right)\cong\operatorname{Ext}_{A}^{i}(X,X')\qquad(i\geq0).

Every AA-module occurs as the first component of (X′0)\begin{pmatrix}X'\\0\end{pmatrix}, so bΛ((XY))=bA(X)b_{\Lambda}\left(\begin{pmatrix}X\\Y\end{pmatrix}\right)=b_A(X). Likewise, the exact functor e2(−)e_2(-) preserves projectives, since e2Λe1=0e_2\Lambda e_1=0 and e2Λe2=Be_2\Lambda e_2=B. Applying it to a projective resolution, and using the natural Hom identification, yields

Ext⁡Λi(M,(0V))≅Ext⁡Bi(Y,V)(i≥0).\operatorname{Ext}_{\Lambda}^{i}\left(M,\begin{pmatrix}0\\V\end{pmatrix}\right)\cong\operatorname{Ext}_{B}^{i}(Y,V)\qquad(i\geq0).

In particular, bB(Y)≤bΛ(M)b_B(Y)\leq b_{\Lambda}(M).

(a) Take e=e1e=e_1, and put I=ΛeΛI=\Lambda e\Lambda. The ideal II is strong idempotent, as in the proof of Corollary 3.8. As a left Λ\Lambda-module, I=(A⊕U0)I=\begin{pmatrix}A\oplus U\\0\end{pmatrix}. Hence pd⁡ΛI≤d\operatorname{pd}_{\Lambda}I\leq d, and pd⁡ΛB≤d+1\operatorname{pd}_{\Lambda}B\leq d+1, where Λ/I≅B\Lambda/I\cong B. Proposition 3.2 gives

bΛ((0Y))≤d+1+bB(Y).b_{\Lambda}\left(\begin{pmatrix}0\\Y\end{pmatrix}\right)\leq d+1+b_B(Y).

The canonical exact sequence

0⟶(X0)⟶(XY)ϕ⟶(0Y)⟶00\longrightarrow\begin{pmatrix}X\\0\end{pmatrix}\longrightarrow\begin{pmatrix}X\\Y\end{pmatrix}_{\phi}\longrightarrow\begin{pmatrix}0\\Y\end{pmatrix}\longrightarrow0

and pd⁡Λ(X0)=pd⁡AX≤l\operatorname{pd}_{\Lambda}\begin{pmatrix}X\\0\end{pmatrix}=\operatorname{pd}_A X\leq l imply

Ext⁡Λi((0Y),L)≅Ext⁡Λi((XY)ϕ,L)(i>l+1)\operatorname{Ext}_{\Lambda}^{i}\left(\begin{pmatrix}0\\Y\end{pmatrix},L\right)\cong\operatorname{Ext}_{\Lambda}^{i}\left(\begin{pmatrix}X\\Y\end{pmatrix}_{\phi},L\right)\qquad(i>l+1)

for every left Λ\Lambda-module LL. Thus, whenever pΛ(M,L)<∞p_{\Lambda}(M,L)<\infty, one also has pΛ((0Y),L)<∞p_{\Lambda}\left(\begin{pmatrix}0\\Y\end{pmatrix},L\right)<\infty. It follows that

bΛ(M)≤max⁡{l+1,bΛ((0Y))}≤max⁡{l+1,d+1+bB(Y)}.b_{\Lambda}(M)\leq\max\left\{l+1,b_{\Lambda}\left(\begin{pmatrix}0\\Y\end{pmatrix}\right)\right\}\leq\max\{l+1,d+1+b_B(Y)\}.

Taking the supremum over all MM proves the global bound. If bΛ(M)<∞b_{\Lambda}(M)<\infty, the inequality bB(Y)≤bΛ(M)b_B(Y)\leq b_{\Lambda}(M) proved above guarantees bB(Y)<∞b_B(Y)<\infty; hence bB(Y)≤fab⁡(B)b_B(Y)\leq\operatorname{fab}(B). Taking the supremum over these MM proves the finitistic bound. The remaining inequalities follow from d≤ld\leq l.

(b) Put s=l+1s=l+1. We first show that a projective resolution of MM can be chosen so that its ssth syzygy is (Xs0)\begin{pmatrix}X_s\\0\end{pmatrix} for some AA-module XsX_s. Indeed, choose projective epimorphisms Q→YQ\to Y over BB and P→XP\to X over AA. They determine a projective epimorphism

(P0)⊕(U⊗BQQ)id⟶(XY)ϕ.\begin{pmatrix}P\\0\end{pmatrix}\oplus\begin{pmatrix}U\otimes_B Q\\Q\end{pmatrix}_{\mathrm{id}}\longrightarrow\begin{pmatrix}X\\Y\end{pmatrix}_{\phi}.

Repeating this construction along a projective BB-resolution of YY, its llth syzygy has projective second component. At the next step, choose the map on that component to be the identity. Its kernel then has zero second component, as asserted.

Write K=ΩΛsM=(Xs0)K=\Omega_{\Lambda}^{s}M=\binom{X_s}{0}. Dimension shifting gives Ext⁡Λi(K,L)≅Ext⁡Λi+s(M,L)\operatorname{Ext}_{\Lambda}^{i}(K,L)\cong\operatorname{Ext}_{\Lambda}^{i+s}(M,L), i≥1i\ge1. In particular, eventual vanishing for either pair is equivalent to eventual vanishing for the other, and

bΛ(K)≤bΛ(M),bΛ(M)≤s+bΛ(K).b_{\Lambda}(K)\le b_{\Lambda}(M),\qquad b_{\Lambda}(M)\le s+b_{\Lambda}(K).

Since bΛ(K)=bA(Xs)b_{\Lambda}(K)=b_{A}(X_s), we obtain bΛ(M)≤l+1+bA(Xs)b_{\Lambda}(M)\le l+1+b_{A}(X_s). Taking the supremum over all MM proves the global bound. If bΛ(M)<∞b_{\Lambda}(M)<\infty, then bA(Xs)=bΛ(K)<∞b_{A}(X_s)=b_{\Lambda}(K)<\infty, so the same argument with fab⁡(A)\operatorname{fab}(A) proves the finitistic bound.

Finally, the corresponding lower bounds in Corollary 3.8(a) give the asserted equivalences of finiteness.

We conclude with an example illustrating sharpness and the applicability of the estimates beyond finite global dimension.

Example 3.10. Put D=k[ε]/(ε2)D=k[\varepsilon]/(\varepsilon^2), which has infinite global dimension. By Example 3.6, gb⁡(D)=fab⁡(D)=0\operatorname{gb}(D)=\operatorname{fab}(D)=0.

Consider the two upper triangular matrix algebras

Λ1=(kkDD0D),Λ2=(DDDk0k).\Lambda_1=\begin{pmatrix} k & {}_kD_D \\ 0 & D \end{pmatrix},\qquad \Lambda_2=\begin{pmatrix} D & {}_DD_k \\ 0 & k \end{pmatrix}.

Proposition 3.9(a) and (b), respectively, give gb⁡(Λj)≤1\operatorname{gb}(\Lambda_j)\le1 and fab⁡(Λj)≤1\operatorname{fab}(\Lambda_j)\le1 for j=1,2j=1,2. To see that these estimates are exact, write

Λj=(AU0B)\Lambda_j=\begin{pmatrix} A & U \\ 0 & B \end{pmatrix}

in each case. The left AA-module UU is nonzero and projective, and there is an exact sequence

0⟶(U0)⟶Λje2=(UB)id⟶(0B)⟶0.0\longrightarrow\binom{U}{0}\longrightarrow\Lambda_j e_2=\binom{U}{B}_{\mathrm{id}}\longrightarrow\binom{0}{B}\longrightarrow0.

The first two terms are projective, but this sequence does not split: a section with second component id⁡B\operatorname{id}_B would violate the compatibility with the nonzero structure map U⊗BB→UU\otimes_B B\to U. It therefore represents a nonzero element of Ext⁡Λj1((0B),(U0))\operatorname{Ext}_{\Lambda_j}^{1}\left(\binom{0}{B},\binom{U}{0}\right). Thus bΛj((0B))=1b_{\Lambda_j}\left(\binom{0}{B}\right)=1, and

gb⁡(Λj)=fab⁡(Λj)=1(j=1,2).\operatorname{gb}(\Lambda_j)=\operatorname{fab}(\Lambda_j)=1\qquad(j=1,2).

Nevertheless, both algebras have infinite global dimension. These examples show that the new estimates apply even when the other corner and the triangular matrix algebra have infinite global dimension. They also improve the preceding corollary’s upper bound 3 to the exact value 1.

Auslander bounds and arrow removals

In this section, we investigate whether finiteness of the global and finitistic Auslander bounds is preserved under arrow removal. We first study these questions in the more general framework of cleft extensions and then apply the resulting comparison theorems to arrow removal. We begin by recalling the terminology of cleft extensions used below. We denote the faithful exact functor by ee to distinguish it from the idempotent ee used in the preceding sections.

Definition 4.1. [2 Definition 2.1] A cleft extension of an abelian category BB consists of a diagram of functors

B→iA⇄leBB \xrightarrow{i} A \underset{e}{\stackrel{l}{\rightleftarrows}} B

such that ee is faithful and exact, (l,e)(l,e) is an adjoint pair, and there is a natural isomorphism ei≅id⁡Bei\cong\operatorname{id}_B. We denote such a cleft extension by (B,A,i,e,l)(B,A,i,e,l).

Let (B,A,i,e,l)(\mathcal{B},\mathcal{A},i,e,l) be a cleft extension of categories of finitely generated modules. For t∈Nt \in\mathbb{N}, the exact functor ee is called a tt-eventually homological isomorphism if the natural maps induced by ee give isomorphisms

Ext⁡Aq(X,Y)≅Ext⁡Bq(eX,eY)(1)\operatorname{Ext}_{\mathcal{A}}^{q}(X,Y) \cong\operatorname{Ext}_{\mathcal{B}}^{q}(eX,eY) \tag*{(1)}

for all X,Y∈AX,Y \in\mathcal{A} and all q>tq > t. The invariants bb, gb⁡\operatorname{gb} and fab⁡\operatorname{fab} for these module categories are defined as in Section 2.

Proposition 4.2. Let (B,A,i,e,l)(\mathcal{B},\mathcal{A},i,e,l) be a cleft extension of categories of finitely generated modules. Suppose that ee is a tt-eventually homological isomorphism for some t∈Nt \in\mathbb{N}. Then the following statements hold.

(a) For every X∈AX \in\mathcal{A},

bA(X)≤max⁡{t,bB(eX)}andbB(eX)≤max⁡{t,bA(X)}.b_{\mathcal{A}}(X) \leq\max\{t,b_{\mathcal{B}}(eX)\} \quad\text{and} \quad b_{\mathcal{B}}(eX) \leq\max\{t,b_{\mathcal{A}}(X)\}.

Consequently, bA(X)<∞b_{\mathcal{A}}(X) < \infty if and only if bB(eX)<∞b_{\mathcal{B}}(eX) < \infty.

(b)

gb⁡(A)≤max⁡{t,gb⁡(B)},gb⁡(B)≤max⁡{t,gb⁡(A)},fab⁡(A)≤max⁡{t,fab⁡(B)},fab⁡(B)≤max⁡{t,fab⁡(A)}.\begin{aligned} \operatorname{gb}(\mathcal{A}) &\leq\max\{t,\operatorname{gb}(\mathcal{B})\}, & \operatorname{gb}(\mathcal{B}) &\leq\max\{t,\operatorname{gb}(\mathcal{A})\},\\ \operatorname{fab}(\mathcal{A}) &\leq\max\{t,\operatorname{fab}(\mathcal{B})\}, & \operatorname{fab}(\mathcal{B}) &\leq\max\{t,\operatorname{fab}(\mathcal{A})\}. \end{aligned}

Consequently, gb⁡(A)<∞\operatorname{gb}(\mathcal{A}) < \infty if and only if gb⁡(B)<∞\operatorname{gb}(\mathcal{B}) < \infty, and fab⁡(A)<∞\operatorname{fab}(\mathcal{A}) < \infty if and only if fab⁡(B)<∞\operatorname{fab}(\mathcal{B}) < \infty.

Proof. (a) Let Y∈AY \in\mathcal{A} and suppose that pA(X,Y)<∞p_{\mathcal{A}}(X,Y) < \infty. By (1), pB(eX,eY)<∞p_{\mathcal{B}}(eX,eY) < \infty, whence

pA(X,Y)≤max⁡{t,pB(eX)}.p_{\mathcal{A}}(X,Y) \leq\max\{t,p_{\mathcal{B}}(eX)\}.

Taking the supremum over all such YY proves the first inequality. Conversely, let V∈BV \in\mathcal{B} and suppose that pB(eX,V)<∞p_{\mathcal{B}}(eX,V) < \infty. Since ei≅id⁡Bei \cong\operatorname{id}_{\mathcal{B}}, the functor ee is essentially surjective. There is a Y∈AY \in\mathcal{A} with eY≅VeY \cong V. (1) implies pA(X,Y)<∞p_{\mathcal{A}}(X,Y) < \infty, and hence

pB(eX,V)≤max⁡{t,pA(X)}.p_{\mathcal{B}}(eX,V) \leq\max\{t,p_{\mathcal{A}}(X)\}.

Taking the supremum over all such VV gives the second inequality.

(b) The assertions for the global bounds follow by taking suprema in (a) and using the essential surjectivity of ee. (a) also identifies the objects having finite individual Auslander bound; taking suprema over precisely these objects proves the assertions for the finitistic bounds.

The arrow-removal operation and its homological properties have been studied in [6, 9]. We work with left modules, whereas these papers use right modules. Let QQ be a finite quiver and let Λ=kQ/I\Lambda= kQ/I be an admissible quotient of the path algebra kQkQ. Suppose that QQ contains arrows

ai:vei⟶vfi,1≤i≤s,a_i : v_{e_i} \longrightarrow v_{f_i}, \qquad1 \leq i \leq s,

which do not occur in a set of minimal generators of II and satisfy

Hom⁡Λ(Λefi,Λeej)=0for all 1≤i,j≤s.\operatorname{Hom}_{\Lambda}(\Lambda e_{f_i},\Lambda e_{e_j}) = 0 \quad\text{for all } 1 \leq i,j \leq s.

Set Γ=Λ/Λ{ai‾}i=1s\Gamma=\Lambda/\Lambda\{\overline{a_i}\}_{i=1}^{s}, where ai‾=ai+I\overline{a_i}=a_i+I. The natural projection Λ→Γ\Lambda\to\Gamma is called an arrow removal. Write π:Λ→Γ\pi:\Lambda\to\Gamma for the canonical projection. The arrow-removal construction admits an algebra section σ:Γ→Λ\sigma:\Gamma\to\Lambda satisfying πσ=id⁡Γ\pi\sigma=\operatorname{id}_{\Gamma}. In the following diagram, the Λ\Lambda-module structures on Γ\Gamma are induced by π\pi, whereas the Γ\Gamma-module structures on Λ\Lambda are induced by σ\sigma. Thus ii and ee are naturally isomorphic to restriction of scalars along π\pi and σ\sigma, respectively. The arrow-removal construction gives the following diagram of functors:

q=ΓΓ⊗Λ−l=ΛΛΓ⊗Γ−Γ-mod → i=Hom⁡Γ(ΓΓΛ,−)  Λ-mod → e=Hom⁡Λ(ΛΛΓ,−)  Γ-modp=Hom⁡Λ(ΛΓΓ,−)r=Hom⁡Γ(ΓΛ,−)\begin{gathered} q=\Gamma_{\Gamma}\otimes_{\Lambda}- \qquad\qquad\qquad l={}_{\Lambda}\Lambda_{\Gamma}\otimes_{\Gamma}-\\ \Gamma\text{-mod}\ \xrightarrow[\ i=\operatorname{Hom}_{\Gamma}(\Gamma_{\Gamma}\Lambda,-)\ ]{}\ \Lambda\text{-mod}\ \xrightarrow[\ e=\operatorname{Hom}_{\Lambda}({}_{\Lambda}\Lambda_{\Gamma},-)\ ]{}\ \Gamma\text{-mod}\\ p=\operatorname{Hom}_{\Lambda}({}_{\Lambda}\Gamma_{\Gamma},-)\qquad\qquad r=\operatorname{Hom}_{\Gamma}(\Gamma_{\Lambda},-) \end{gathered}

We now show that the associated module categories form a cleft extension satisfying the hypotheses of Proposition 4.2. To apply the right-module results of [6, 9], we pass to the opposite algebras. The corresponding arrows in the opposite quiver are aiop:vfi→veia_i^{\mathrm{op}}:v_{f_i}\to v_{e_i}, and our hypothesis gives

Hom⁡mod-Λop(fiΛop,ejΛop)≅fiΛej=0(1≤i,j≤s).\operatorname{Hom}_{\mathrm{mod}\text{-}\Lambda^{\mathrm{op}}}(f_i\Lambda^{\mathrm{op}},e_j\Lambda^{\mathrm{op}})\cong f_i\Lambda e_j=0\qquad(1\leq i,j\leq s).

Moreover, these arrows do not occur in the corresponding minimal generators of the opposite relation ideal. Thus Λop→Γop\Lambda^{\mathrm{op}}\to\Gamma^{\mathrm{op}} satisfies the arrow-removal hypotheses. Applying [9 Proposition 4.6] to this arrow removal and identifying left modules with right modules over the opposite algebras, we obtain a cleft extension

(Γ-mod,Λ-mod,i,e,l).(\Gamma\text{-mod},\Lambda\text{-mod},i,e,l).

Furthermore, ll and rr are exact, and ee preserves projective modules. By [6 Corollary 3.3], applied in the same way, ee is a 1-eventually homological isomorphism. The following theorem now follows from Proposition 4.2(b).

Theorem 4.3. For an arrow removal Λ→Γ\Lambda\to\Gamma as above,

gb⁡(Λ)≤max⁡{1,gb⁡(Γ)},gb⁡(Γ)≤max⁡{1,gb⁡(Λ)},\operatorname{gb}(\Lambda)\leq\max\{1,\operatorname{gb}(\Gamma)\},\qquad \operatorname{gb}(\Gamma)\leq\max\{1,\operatorname{gb}(\Lambda)\},
fab⁡(Λ)≤max⁡{1,fab⁡(Γ)},fab⁡(Γ)≤max⁡{1,fab⁡(Λ)}.\operatorname{fab}(\Lambda)\leq\max\{1,\operatorname{fab}(\Gamma)\},\qquad \operatorname{fab}(\Gamma)\leq\max\{1,\operatorname{fab}(\Lambda)\}.

Remark 4.4. The Uniform Auslander Condition, as well as finiteness of the finitistic Auslander bound, is invariant under arrow removal. If either member of one of the two pairs of invariants is greater than 1, then the two members of that pair are equal.

The following example illustrates the role of the constant 1 in Theorem 4.3. Although arrow removal preserves finiteness of both Auslander bounds, it need not preserve their values when they are at most 1. In this example, both bounds decrease from 1 to 0 after removing a single arrow, while both algebras have infinite global dimension.

Example 4.5. Let QQ be the quiver

x↺1→a2\begin{array}{cc} x\circlearrowleft& \qquad1\xrightarrow{a}2 \end{array}

and put Λ=kQ/⟨x2⟩\Lambda=kQ/\langle x^2\rangle. Paths are composed from left to right. Removing the arrow aa gives

Γ=Λ/Λa‾Λ≅D×k,D=k[x]/(x2).\Gamma=\Lambda/\Lambda\overline{a}\Lambda\cong D\times k,\qquad D=k[x]/(x^2).

Then

gb⁡(Γ)=fab⁡(Γ)=0,gb⁡(Λ)=fab⁡(Λ)=1.\operatorname{gb}(\Gamma)=\operatorname{fab}(\Gamma)=0,\qquad \operatorname{gb}(\Lambda)=\operatorname{fab}(\Lambda)=1.

Thus the constant 1 in Theorem 4.3 cannot in general be omitted.

Proof. The arrow aa does not occur in the minimal generating set {x2}\{x^2\} of the relation ideal. Moreover, Hom⁡Λ(Λe2,Λe1)≅e2Λe1=0\operatorname{Hom}_{\Lambda}(\Lambda e_2,\Lambda e_1)\cong e_2\Lambda e_1=0, since there is no path from vertex 2 to vertex 1. Hence Λ→Γ\Lambda\to\Gamma is an arrow removal satisfying the stated hypotheses.

We first calculate the bounds of Γ\Gamma. By Example 3.6, gb⁡(D)=fab⁡(D)=0\operatorname{gb}(D)=\operatorname{fab}(D)=0. Since Ext over a product algebra is computed componentwise and k\mathbb{k} is semisimple, we obtain

gb⁡(Γ)=fab⁡(Γ)=0.\operatorname{gb}(\Gamma)=\operatorname{fab}(\Gamma)=0.

Theorem 4.3 therefore yields gb⁡(Λ)≤1\operatorname{gb}(\Lambda)\leq1, fab⁡(Λ)≤1\operatorname{fab}(\Lambda)\leq1. To prove equality, put Pi=ΛeiP_i=\Lambda e_i and Si=Pi/rad⁡PiS_i=P_i/\operatorname{rad}P_i. Right multiplication by aa gives an exact sequence

0⟶P1→⋅aP2⟶S2⟶0.0\longrightarrow P_1\xrightarrow{\cdot a}P_2\longrightarrow S_2\longrightarrow0.

Applying Hom⁡Λ(−,S1)\operatorname{Hom}_{\Lambda}(-,S_1) and using Hom⁡Λ(P2,S1)=0\operatorname{Hom}_{\Lambda}(P_2,S_1)=0 and Hom⁡Λ(P1,S1)≅k\operatorname{Hom}_{\Lambda}(P_1,S_1)\cong\mathbb{k}, we obtain

Ext⁡Λ1(S2,S1)≅k.\operatorname{Ext}^{1}_{\Lambda}(S_2,S_1)\cong\mathbb{k}.

Consequently,

pd⁡ΛS2=bΛ(S2)=1.\operatorname{pd}_{\Lambda}S_2=b_{\Lambda}(S_2)=1.

Together with the upper bounds, this proves

gb⁡(Λ)=fab⁡(Λ)=1.\operatorname{gb}(\Lambda)=\operatorname{fab}(\Lambda)=1.

Finally, both algebras have infinite global dimension. Thus the example also illustrates the arrow-removal result beyond finite global dimension.

An arbitrarily large gap between fab⁡(Λ)\operatorname{fab}(\Lambda) and findim⁡(Λ)\operatorname{findim}(\Lambda)

In this section, we give a family with zero finitistic dimension and arbitrarily large finite Auslander bounds.

Fix n≥1n\geq1. Let QnQ_n be the quiver

n←αnn−1⟵⋯⟵1←α10n\xleftarrow{\alpha_n}n-1\longleftarrow\cdots\longleftarrow1\xleftarrow{\alpha_1}0

with a loop ε\varepsilon at 00. Let JJ be the arrow ideal of kQn\mathbb{k}Q_n and set Λn=kQn/J2\Lambda_n=\mathbb{k}Q_n/J^2. For each vertex ii, let SiS_i be the corresponding simple left module and Pi=ΛneiP_i=\Lambda_n e_i its projective cover. Since J2=0J^2=0, the radicals of the indecomposable projectives are semisimple, with

rad⁡P0≅S0,rad⁡Pi≅Si−1(1≤i≤n).(2)\operatorname{rad}P_0\cong S_0,\qquad\operatorname{rad}P_i\cong S_{i-1}\quad(1\leq i\leq n). \tag*{(2)}

In particular,

ΩS0≅S0,ΩSi≅Si−1(1≤i≤n).(3)\Omega S_0\cong S_0,\qquad\Omega S_i\cong S_{i-1}\quad(1\leq i\leq n). \tag*{(3)}

All simple left Λn\Lambda_n-modules consequently have infinite projective dimension. Clearly, findim⁡Λn=0\operatorname{findim}\Lambda_n=0. In the following, we prove that

fab⁡(Λn)=gb⁡(Λn)=n.\operatorname{fab}(\Lambda_n)=\operatorname{gb}(\Lambda_n)=n.

Lemma 5.1. For every 0≤i≤n0\leq i\leq n, bΛn(Si)=ib_{\Lambda_n}(S_i)=i.

Proof. Since ΩS0≅S0\Omega S_0 \cong S_0, dimension shifting gives

Ext⁡Λnj+1(S0,N)≅Ext⁡Λnj(S0,N)(j≥1)\operatorname{Ext}_{\Lambda_n}^{j+1}(S_0,N) \cong\operatorname{Ext}_{\Lambda_n}^{j}(S_0,N) \qquad(j \ge1)

for every finitely generated NN. If these groups eventually vanish, they all vanish in positive degrees. Hence bΛn(S0)=0b_{\Lambda_n}(S_0)=0.

For i≥1i \ge1, ΩiSi≅S0\Omega^i S_i \cong S_0 by (3). Consequently, if Ext⁡Λnj(Si,N)=0\operatorname{Ext}_{\Lambda_n}^{j}(S_i,N)=0 for all sufficiently large jj, then dimension shifting and the preceding paragraph show Ext⁡Λnj(Si,N)=0\operatorname{Ext}_{\Lambda_n}^{j}(S_i,N)=0 (j>i)(j > i). It follows that bΛn(Si)≤ib_{\Lambda_n}(S_i) \le i.

To prove the reverse inequality, take N=P0N=P_0. The exact sequence

0⟶S0⟶P0⟶S0⟶00 \longrightarrow S_0 \longrightarrow P_0 \longrightarrow S_0 \longrightarrow0

shows that Ext⁡Λn1(S0,P0)=0\operatorname{Ext}_{\Lambda_n}^{1}(S_0,P_0)=0 since the restriction map Hom⁡Λn(P0,P0)⟶Hom⁡Λn(S0,P0)\operatorname{Hom}_{\Lambda_n}(P_0,P_0) \longrightarrow\operatorname{Hom}_{\Lambda_n}(S_0,P_0) is surjective. Periodicity of S0S_0 then yields

Ext⁡Λnj(S0,P0)=0(j≥1).(4)\operatorname{Ext}_{\Lambda_n}^{j}(S_0,P_0)=0 \qquad(j \ge1). \tag*{(4)}

For S1S_1, the projective cover sequence is

0⟶S0⟶P1⟶S1⟶0.0 \longrightarrow S_0 \longrightarrow P_1 \longrightarrow S_1 \longrightarrow0.

Now Hom⁡Λn(P1,P0)=0\operatorname{Hom}_{\Lambda_n}(P_1,P_0)=0, whereas Hom⁡Λn(S0,P0)≠0\operatorname{Hom}_{\Lambda_n}(S_0,P_0)\ne0. Therefore

Ext⁡Λn1(S1,P0)≠0,Ext⁡Λn>1(S1,P0)=0,\operatorname{Ext}_{\Lambda_n}^{1}(S_1,P_0)\ne0,\qquad\operatorname{Ext}_{\Lambda_n}^{>1}(S_1,P_0)=0,

where the second assertion follows from (4). For arbitrary i≥1i \ge1, since Ωi−1Si≅S1\Omega^{i-1}S_i \cong S_1, dimension shifting gives

Ext⁡Λni(Si,P0)≅Ext⁡Λn1(S1,P0)≠0,Ext⁡Λn>i(Si,P0)=0.\operatorname{Ext}_{\Lambda_n}^{i}(S_i,P_0)\cong\operatorname{Ext}_{\Lambda_n}^{1}(S_1,P_0)\ne0,\qquad\operatorname{Ext}_{\Lambda_n}^{>i}(S_i,P_0)=0.

Thus the last nonzero extension group for (Si,P0)(S_i,P_0) occurs in degree ii, proving bΛn(Si)≥ib_{\Lambda_n}(S_i)\ge i and hence equality.

Proposition 5.2. fab⁡(Λn)=gb⁡(Λn)=n\operatorname{fab}(\Lambda_n)=\operatorname{gb}(\Lambda_n)=n.

Proof. Let M∈ΛnM\in\Lambda_n-mod be nonprojective. Then ΩM≅⨁i=0n−1Siai\Omega M\cong\bigoplus_{i=0}^{n-1}S_i^{a_i} for suitable nonnegative integers aia_i. If Ext⁡Λn>0(M,N)=0\operatorname{Ext}_{\Lambda_n}^{>0}(M,N)=0, then Ext⁡Λn>0(Si,N)=0\operatorname{Ext}_{\Lambda_n}^{>0}(S_i,N)=0 for every ii with ai>0a_i>0. Lemma 5.1 implies Ext⁡Λn>i(Si,N)=0\operatorname{Ext}_{\Lambda_n}^{>i}(S_i,N)=0 for such ii. Dimension shifting therefore gives

Ext⁡Λnj(M,N)≅Ext⁡Λnj−1(ΩM,N)=0(j>n).\operatorname{Ext}_{\Lambda_n}^{j}(M,N)\cong\operatorname{Ext}_{\Lambda_n}^{j-1}(\Omega M,N)=0 \qquad(j>n).

Hence bΛn(M)≤nb_{\Lambda_n}(M)\le n. Thus gb⁡(Λn)≤n\operatorname{gb}(\Lambda_n)\le n. On the other hand, Lemma 5.1 gives bΛn(Sn)=nb_{\Lambda_n}(S_n)=n. Therefore

n≤fab⁡(Λn)≤gb⁡(Λn)≤n,n\le\operatorname{fab}(\Lambda_n)\le\operatorname{gb}(\Lambda_n)\le n,

which proves the result.

Acknowledgements

Kaili Wu is supported by the Natural Science Foundation of Jiangsu Province of China (Grant No. BK20240657) and the National Natural Science Foundation of China (Grant No. 12526627).

Jiaqun Wei is supported by the National Natural Science Foundation of China (Grant Nos. 12571042, 12271248) and the Natural Science Foundation of Zhejiang Province (Grant No. LZ25A010002).

Conflict of Interest

The authors declare that they have no conflict of interest.

References

  1. [1]M. Auslander, M. I. Platz(e)ck and G. Todorov, Homological theory of idempotent ideals, Trans. Am. Math. Soc. 332 (2) (1992): 667–692.DOI
  2. [2]A. Beligiannis, Cleft extensions of abelian categories and applications to ring theory, Comm. Algebra 28 (2000): 4503–4546.
  3. [3]A. A. Beilinson, J. Bernstein and P. Deligne, Faisceaux pervers, Soc. Math. France, vol. 100 (1982).
  4. [4]H. Chen and C. Xi, Recollements of derived categories III: finitistic dimensions, J. Lond. Math. Soc. 95 (2) (2017): 633–658.arxiv.org/abs/1405.5090
  5. [5]L. W. Christensen and H. Holm, Algebras that satisfy Auslander’s condition on vanishing of cohomology, Math. Z. 265 (1) (2010): 21–40.arxiv.org/abs/0801.0401
  6. [6]K. Erdmann, C. Psaroudakis and Ø. Solberg, Homological invariants of the arrow removal operation, Represent. Theory 26 (2022): 370–387.arxiv.org/abs/2108.04891
  7. [7]V. Franjou and T. Pirashvili, Comparison of abelian categories recollements. Doc. Math. 9 (2004): 41–56.DOI
  8. [8]M. A. Gatica, M. Lanzilotta and M. I. Platzeck, Idempotent ideals and the Igusa-Todorov functions. Algebr. Represent. Theory. 20(2017):275–287.
  9. [9]E. L. Green, C. Psaroudakis and Ø. Solberg, Reduction techniques for the finitistic dimension, Trans. Am. Math. Soc. 374 (2021): 6839–6879.arxiv.org/abs/1808.03564
  10. [10]D. Happel, Homological conjectures in representation theory of finite-dimensional algebras, Sherbrooke Lecture Notes Series, Université de Sherbrooke, 1991.
  11. [11]C. Huneke and D. A. Jorgensen, Symmetry in the vanishing of Ext over Gorenstein rings, Math. Scand. 93 (2) (2003): 161–184.
  12. [12]D. A. Jørgensen and L. M. Sega, Nonvanishing cohomology and classes of Gorenstein rings, Adv. Math. 188 (2) (2004):470–490.
  13. [13]S. König, Tilting complexes, perpendicular categories and recollements of derived categories of rings, J. Pure Appl. Algebra. 73 (1991): 211–232.
  14. [14]N. J. Kuhn, The generic representation theory of finite fields: a survey of basic structure, in: Infinite Length Modules. Trends in Mathematics. Birkhäuser, Basel. (2000): 193–212.
  15. [15]A. J. S. Levins, The Auslander bound for complexes, Arch. Math. 126(2026): 29–39.arxiv.org/abs/2402.06130
  16. [16]I. Mori, Symmetry in the vanishing of Ext over stably symmetric algebras, J. Algebra 310 (2) (2007): 708–729.
  17. [17]C. Psaroudakis, Homological theory of recollements of abelian categories, J. Algebra 398 (2014): 63–110.DOI
  18. [18]C. Psaroudakis and J. Vitória, Recollements of module categories, Appl. Categ. Struct. 22 (2014): 579–593.
  19. [19]Y. Qin, and Y. Han, Reducing homological conjectures by n-recollements, Algebr. Represent. Theory 19 (2016): 377–395.arxiv.org/abs/1410.3223
  20. [20]S. O. Smalø, Local limitations of the Ext functor do not exist, Bull. London Math. Soc. 38 (1) (2006): 97–98.DOI
  21. [21]J. Wei, Auslander bounds and homological conjectures, Rev. Mat. Iberoam. 27 (3) (2011): 871–884.arxiv.org/abs/0802.1085
  22. [22]J. Wei, Derived invariance by syzygy complexes, Math. Proc. Cambridge Philos. Soc. 164 (2018), no. 2, 325–343.DOI

Paper details

vlnc:W_w9atxfpf
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