Introduction
Tilting theory plays a central role in the representation theory of algebras by relating module categories through torsion pairs [8 , 6 , 7 , 13 ] . It has developed into a general method for constructing and studying equivalences of categories. The derived Morita theory associated with tilting modules, complexes and bundles has important applications in the representation theory of finite groups and algebraic groups, algebraic geometry and topology; cf. [29 , 5 ] . The development of cluster tilting theory has further provided categorical models for cluster algebras and their mutation combinatorics [9 ] . Silting complexes were introduced in [19 ] as a generalization of tilting complexes. Derived Morita theory [17 ] also applies to silting complexes, yielding equivalences with derived categories of differential graded endomorphism algebras.
The relation between 2-term silting complexes and torsion pairs was studied in [14 , 2 ] . Building on this work, the classical tilting theorem was generalized to 2-term silting complexes in [10 ] . Let A A A be a finite-dimensional algebra over a field k k k , and let mod A \operatorname{mod} A mod A denote the category of finitely generated right A A A -modules. For a 2-term silting complex P P P over A A A , let B = End D b ( mod A ) ( P ) B=\operatorname{End}_{D^b(\operatorname{mod} A)}(P) B = End D b ( mod A ) ( P ) . The theorem gives torsion pairs ( T ( P ) , F ( P ) ) (\mathcal{T}(P),\mathcal{F}(P)) ( T ( P ) , F ( P )) in mod A \operatorname{mod} A mod A and ( X ( P ) , Y ( P ) ) (\mathcal{X}(P),\mathcal{Y}(P)) ( X ( P ) , Y ( P )) in mod B \operatorname{mod} B mod B , together with equivalences
T ( P ) ≃ Y ( P ) and F ( P ) ≃ X ( P ) . \mathcal{T}(P)\simeq\mathcal{Y}(P)\quad\text{and}\quad\mathcal{F}(P)\simeq\mathcal{X}(P). T ( P ) ≃ Y ( P ) and F ( P ) ≃ X ( P ) .
A 2-term silting complex Q Q Q over B B B is also constructed, which induces the torsion pair ( X ( P ) , Y ( P ) ) (\mathcal{X}(P),\mathcal{Y}(P)) ( X ( P ) , Y ( P )) . Moreover, there exists an algebra homomorphism A → End D b ( mod B ) ( Q ) A\to\operatorname{End}_{D^b(\operatorname{mod} B)}(Q) A → End D b ( mod B ) ( Q ) which is an isomorphism if and only if P P P is tilting. One key point in this theory is to realize mod B \operatorname{mod} B mod B as the heart of the bounded t t t -structure on D b ( mod A ) D^b(\operatorname{mod} A) D b ( mod A ) induced by P P P . This realization works for arbitrary silting complexes [20 ] . To generalize the silting theorem, one also needs counterparts of the two torsion pairs and the second silting complex. A dg approach to the silting theorem was developed in [34 ] . This approach also yields a partial generalization to higher-term silting objects, giving equivalences between subcategories defined by cohomological vanishing; see [34 ] , Section 4.5. The question of whether the corresponding second object is silting after passing to the ordinary endomorphism algebra was left open.
Extended module categories provide a natural setting for a higher-term silting theorem. The correspondence between 2-term silting complexes and functorially finite torsion pairs generalizes to ( d + 1 ) (d+1) ( d + 1 ) -term silting objects and functorially finite s s s -torsion pairs in d d d -extended module categories [11 Z, 33 , 15 ] . These categories also admit versions of Happel-Reiten-Smalø tilting [1 Z] and Auslander-Reiten theory [Z, MP, Ru]. Related developments concern semibricks and wide subcategories [12 ] , Auslander correspondence [23 ] , higher cluster tilting [24 ] , and the structure of extended hearts [22 P, MNO] .
In this paper, we extend the silting theorem of Buan and Zhou to ( d + 1 ) (d+1) ( d + 1 ) -term silting objects by using truncated dg endomorphism algebras and extended module categories. More precisely, let d ≥ 1 d \ge1 d ≥ 1 , and let A A A be a proper d d d -truncated non-positive differential graded (or dg for short) k k k -algebra, i.e., a non-positive dg algebra whose total cohomology is finite-dimensional and concentrated in [ − d + 1 , 0 ] [-d+1,0] [ − d + 1 , 0 ] . Let D f d ( A ) D_{\mathrm{fd}}(A) D fd ( A ) be the full subcategory of D ( A ) D(A) D ( A ) consisting of objects with finite-dimensional total cohomology. The d d d -extended module category of A A A is
d - mod A : = { X ∈ D f d ( A ) ∣ H i ( X ) = 0 for all i ∉ [ − d + 1 , 0 ] } . d\text{-}\operatorname{mod} A := \{X \in D_{\mathrm{fd}}(A) \mid H^{i}(X) = 0 \text{ for all } i \notin[-d+1,0]\}. d - mod A := { X ∈ D fd ( A ) ∣ H i ( X ) = 0 for all i ∈ / [ − d + 1 , 0 ]} .
Let P \mathbf{P} P be a ( d + 1 ) (d+1) ( d + 1 ) -term silting object in the perfect category per A \operatorname{per} A per A of A A A , i.e., P ∈ add A ∗ add A [ 1 ] ∗ ⋯ ∗ add A [ d ] \mathbf{P} \in\operatorname{add} A * \operatorname{add} A[1] * \cdots* \operatorname{add} A[d] P ∈ add A ∗ add A [ 1 ] ∗ ⋯ ∗ add A [ d ] , Hom D ( A ) ( P , P [ i ] ) = 0 \operatorname{Hom}_{D(A)}(\mathbf{P},\mathbf{P}[i])=0 Hom D ( A ) ( P , P [ i ]) = 0 for all i > 0 i>0 i > 0 , and thick ( P ) = per A \operatorname{thick}(\mathbf{P})=\operatorname{per} A thick ( P ) = per A . The full subcategories
T ( P ) = { M ∈ d - mod A ∣ Hom D ( A ) ( P , M [ i ] ) = 0 for all i > 0 } \mathcal{T}(\mathbf{P}) = \{M \in d\text{-}\operatorname{mod} A \mid\operatorname{Hom}_{D(A)}(\mathbf{P},M[i])=0 \text{ for all } i>0\} T ( P ) = { M ∈ d - mod A ∣ Hom D ( A ) ( P , M [ i ]) = 0 for all i > 0 }
and
F ( P ) = { M ∈ d - mod A ∣ Hom D ( A ) ( P , M [ i ] ) = 0 for all i ≤ 0 } \mathcal{F}(\mathbf{P}) = \{M \in d\text{-}\operatorname{mod} A \mid\operatorname{Hom}_{D(A)}(\mathbf{P},M[i])=0 \text{ for all } i \le0\} F ( P ) = { M ∈ d - mod A ∣ Hom D ( A ) ( P , M [ i ]) = 0 for all i ≤ 0 }
form an s s s -torsion pair in d - mod A d\text{-}\operatorname{mod} A d - mod A [15 Corollary 4.1] ; see also [11 Z] for the case that A A A is an ordinary algebra. Moreover, ( F ( P ) [ d ] , T ( P ) ) (\mathcal{F}(\mathbf{P})[d],\mathcal{T}(\mathbf{P})) ( F ( P ) [ d ] , T ( P )) is an s s s -torsion pair in the d d d -extended heart D f d ( P ) [ − d + 1 , 0 ] D_{\mathrm{fd}}(\mathbf{P})^{[-d+1,0]} D fd ( P ) [ − d + 1 , 0 ] of the bounded t t t -structure induced by P \mathbf{P} P .
The d d d -truncated dg endomorphism algebra of P \mathbf{P} P is defined to be
B = d - End A ( P ) = σ ≥ − d + 1 σ ≤ 0 ( End A ( P ) ) . B = d\text{-}\operatorname{End}_{A}(\mathbf{P}) = \sigma^{\ge-d+1}\sigma^{\le0}\bigl(\operatorname{End}_{A}(\mathbf{P})\bigr). B = d - End A ( P ) = σ ≥ − d + 1 σ ≤ 0 ( End A ( P ) ) .
where End A ( P ) \operatorname{End}_{A}(\mathbf{P}) End A ( P ) is the dg endomorphism algebra of a K K K -projective dg A A A -module representing P \mathbf{P} P , and the truncations σ ≥ − d + 1 \sigma^{\ge-d+1} σ ≥ − d + 1 and σ ≤ 0 \sigma^{\le0} σ ≤ 0 are canonical. For d = 1 d=1 d = 1 , B B B is isomorphic to the ordinary endomorphism algebra of P \mathbf{P} P .
The following theorem summarizes the main results (Theorem 4.2 , Corollary 4.3, Proposition 4.4, and Theorems 4.6 , 5.2 and 5.3 ) of the paper.
Theorem 1.1. Let A A A be a proper d d d -truncated non-positive dg algebra, and let P \mathbf{P} P be a ( d + 1 ) (d+1) ( d + 1 ) -term silting object in per A \operatorname{per} A per A . Let B = d - End A ( P ) B=d\text{-}\operatorname{End}_{A}(\mathbf{P}) B = d - End A ( P ) .
(1) There exists an equivalence
Θ P : D f d ( P ) [ − d + 1 , 0 ] ⟶ d - mod B , \Theta_{\mathbf{P}}:D_{\mathrm{fd}}(\mathbf{P})^{[-d+1,0]} \longrightarrow d\text{-}\operatorname{mod} B, Θ P : D fd ( P ) [ − d + 1 , 0 ] ⟶ d - mod B , which preserves and reflects triangles whose first three terms belong to the respective categories. Moreover, for any objects M , N ∈ D f d ( P ) [ − d + 1 , 0 ] M,N \in D_{\mathrm{fd}}(\mathbf{P})^{[-d+1,0]} M , N ∈ D fd ( P ) [ − d + 1 , 0 ] , there exist natural isomorphisms
Hom D ( A ) ( M , N [ i ] ) ≅ Hom D ( B ) ( Θ P ( M ) , Θ P ( N ) [ i ] ) , i ≤ 1. \operatorname{Hom}_{D(A)}(M,N[i]) \cong\operatorname{Hom}_{D(B)}(\Theta_{\mathbf{P}}(M),\Theta_{\mathbf{P}}(N)[i]), \qquad i \le1. Hom D ( A ) ( M , N [ i ]) ≅ Hom D ( B ) ( Θ P ( M ) , Θ P ( N ) [ i ]) , i ≤ 1. (2) Let X ( P ) = Θ P ( F ( P ) [ d ] ) \mathcal{X}(\mathbf{P})=\Theta_{\mathbf{P}}(\mathcal{F}(\mathbf{P})[d]) X ( P ) = Θ P ( F ( P ) [ d ]) and Y ( P ) = Θ P ( T ( P ) ) \mathcal{Y}(\mathbf{P})=\Theta_{\mathbf{P}}(\mathcal{T}(\mathbf{P})) Y ( P ) = Θ P ( T ( P )) . Then ( X ( P ) , Y ( P ) ) (\mathcal{X}(\mathbf{P}),\mathcal{Y}(\mathbf{P})) ( X ( P ) , Y ( P )) is an s s s -torsion pair in d - mod B d\text{-}\operatorname{mod} B d - mod B , with equivalences
T ( P ) ≃ Y ( P ) and F ( P ) ≃ X ( P ) . \mathcal{T}(\mathbf{P}) \simeq\mathcal{Y}(\mathbf{P}) \quad\text{and}\quad\mathcal{F}(\mathbf{P}) \simeq\mathcal{X}(\mathbf{P}). T ( P ) ≃ Y ( P ) and F ( P ) ≃ X ( P ) .
(3) Let Γ = σ ≤ 0 End A ( P ) \Gamma= \sigma^{\leq0}\operatorname{End}_{A}(\mathbf{P}) Γ = σ ≤ 0 End A ( P ) be the non-positive truncation of End A ( P ) \operatorname{End}_{A}(\mathbf{P}) End A ( P ) . Then
Q : = RHom A ( P , A [ d ] ) ⊗ Γ L B (1) \mathbf{Q} := \operatorname{RHom}_{A}(\mathbf{P}, A[d]) \mathbin{\otimes_{\Gamma}^{\mathbf{L}}} B \tag*{(1)} Q := RHom A ( P , A [ d ]) ⊗ Γ L B (1)
is a ( d + 1 ) (d+1) ( d + 1 ) -term silting object in per B \operatorname{per} B per B , whose induced s s s -torsion pair ( T ( Q ) , F ( Q ) ) (\mathcal{T}(\mathbf{Q}),\mathcal{F}(\mathbf{Q})) ( T ( Q ) , F ( Q )) satisfies T ( Q ) = X ( P ) \mathcal{T}(\mathbf{Q})=\mathcal{X}(\mathbf{P}) T ( Q ) = X ( P ) and F ( Q ) = Y ( P ) \mathcal{F}(\mathbf{Q})=\mathcal{Y}(\mathbf{P}) F ( Q ) = Y ( P ) .
(4) Let C = d - End B ( Q ) C=d\text{-}\operatorname{End}_{B}(\mathbf{Q}) C = d - End B ( Q ) , using a K K K -projective dg A A A -B B B -bimodule representing Q \mathbf{Q} Q . There exists a dg algebra homomorphism
ϑ : A ⟶ C , \vartheta:A\longrightarrow C, ϑ : A ⟶ C ,
which is a quasi-isomorphism if and only if Hom per A ( P , P [ i ] ) = 0 \operatorname{Hom}_{\operatorname{per} A}(\mathbf{P},\mathbf{P}[i])=0 Hom per A ( P , P [ i ]) = 0 for all i ≤ − d i\leq-d i ≤ − d .
(5) Let ϑ ∗ : d - mod C ⟶ d - mod A \vartheta_*:d\text{-}\operatorname{mod} C\longrightarrow d\text{-}\operatorname{mod} A ϑ ∗ : d - mod C ⟶ d - mod A be the restriction functor induced by ϑ \vartheta ϑ . Then ϑ ∗ \vartheta_* ϑ ∗ restricts to equivalences X ( Q ) → ∼ T ( P ) \mathcal{X}(\mathbf{Q})\xrightarrow{\sim}\mathcal{T}(\mathbf{P}) X ( Q ) ∼ T ( P ) and Y ( Q ) → ∼ F ( P ) \mathcal{Y}(\mathbf{Q})\xrightarrow{\sim}\mathcal{F}(\mathbf{P}) Y ( Q ) ∼ F ( P ) .
When d = 1 d=1 d = 1 and A A A is an ordinary finite-dimensional algebra, this theorem recovers the silting theorem of Buan and Zhou. In that case, the functor ϑ ∗ \vartheta_* ϑ ∗ is fully faithful. For d > 1 d>1 d > 1 , this may fail; see Example 5.4.
If ( T ( P ) , F ( P ) ) (\mathcal{T}(\mathbf{P}),\mathcal{F}(\mathbf{P})) ( T ( P ) , F ( P )) is split, then ϑ \vartheta ϑ is a quasi-isomorphism. More precisely, Proposition 7.3 bounds the negative self-extensions of P \mathbf{P} P in terms of the cohomological range of A A A . In particular, if A A A is an ordinary finite-dimensional algebra and ( T ( P ) , F ( P ) ) (\mathcal{T}(\mathbf{P}),\mathcal{F}(\mathbf{P})) ( T ( P ) , F ( P )) is split, then P \mathbf{P} P is tilting, generalizing [10 Proposition 5.7] .
We apply the higher Brenner–Butler theorem to Auslander–Reiten theory in d - mod B d\text{-}\operatorname{mod} B d - mod B . We describe the connecting almost split triangles and, when ( X ( P ) , Y ( P ) ) (\mathcal{X}(\mathbf{P}),\mathcal{Y}(\mathbf{P})) ( X ( P ) , Y ( P )) is split, determine all almost split triangles in d - mod B d\text{-}\operatorname{mod} B d - mod B ; see Propositions 6.5 and 7.6 .
The paper is organized as follows. § 2 recalls preliminaries on dg algebras, derived categories, adjunctions and extended module categories. § 3 discusses s s s -torsion pairs and almost split triangles. In § 4 , we prove the higher Brenner–Butler theorem and construct the silting object Q \mathbf{Q} Q . § 5 studies the d d d -truncated dg endomorphism algebra C C C of Q \mathbf{Q} Q and the dg algebra morphism ϑ : A → C \vartheta:A\to C ϑ : A → C . § 6 applies the theorem to Auslander–Reiten theory. § 7 is devoted to separating and splitting silting objects.
Convention
Throughout this paper, k \mathbf{k} k is a field, and d d d is a positive integer. We write D = Hom k ( − , k ) \mathbb{D}=\operatorname{Hom}_{\mathbf{k}}(-,\mathbf{k}) D = Hom k ( − , k ) for the k \mathbf{k} k -linear dual. Any subcategory of a category is assumed to be full and closed under isomorphisms. We use X ∈ X X\in\mathcal{X} X ∈ X to denote that X X X is an object in a category X \mathcal{X} X , and use X ⊆ Y \mathcal{X}\subseteq\mathcal{Y} X ⊆ Y to denote that X \mathcal{X} X is a subcategory of Y \mathcal{Y} Y . For an object X X X of an additive category, we denote by add X \operatorname{add}X add X the full subcategory consisting of direct summands of finite direct sums of copies of X X X . For any two morphisms f : Y → Z f:Y\to Z f : Y → Z and g : X → Y g:X\to Y g : X → Y , we denote by f ∘ g : X → Z f\circ g:X\to Z f ∘ g : X → Z their composition.
Let Y \mathcal{Y} Y be a category. For any X ⊆ Y \mathcal{X}\subseteq\mathcal{Y} X ⊆ Y and any Y ∈ Y Y\in\mathcal{Y} Y ∈ Y , a right X \mathcal{X} X -approximation of Y Y Y is a morphism f : X → Y f:X\to Y f : X → Y with X ∈ X X\in\mathcal{X} X ∈ X and such that any morphism from an object X ′ X' X ′ in X \mathcal{X} X to Y Y Y factors through f f f . A left X \mathcal{X} X -approximation of Y Y Y is defined dually. A subcategory X \mathcal{X} X of Y \mathcal{Y} Y is called contravariantly (resp. covariantly) finite in Y \mathcal{Y} Y if any object in Y \mathcal{Y} Y has a right (resp. left) X \mathcal{X} X -approximation. A subcategory X \mathcal{X} X of Y \mathcal{Y} Y is called functorially finite if it is both contravariantly finite and covariantly finite. A morphism f : X → Y f:X\to Y f : X → Y is called right minimal if every morphism g : X → X g:X\to X g : X → X such that f ∘ g = f f\circ g=f f ∘ g = f is an isomorphism. A left minimal morphism is defined dually.
The shift functor in a triangulated category is denoted by [ 1 ] [1] [ 1 ] . For any two subcategories X \mathcal{X} X and Y \mathcal{Y} Y of a triangulated category, we denote by X ∗ Y \mathcal{X}*\mathcal{Y} X ∗ Y the full subcategory consisting of objects Z Z Z such that there is a triangle
X ⟶ Z ⟶ Y ⟶ X [ 1 ] , X\longrightarrow Z\longrightarrow Y\longrightarrow X[1], X ⟶ Z ⟶ Y ⟶ X [ 1 ] ,
with X ∈ X X \in\mathcal{X} X ∈ X and Y ∈ Y Y \in\mathcal{Y} Y ∈ Y . We use Hom ( X , Y ) = 0 \operatorname{Hom}(\mathcal{X},\mathcal{Y})=0 Hom ( X , Y ) = 0 to denote Hom ( X , Y ) = 0 \operatorname{Hom}(X,Y)=0 Hom ( X , Y ) = 0 for any X ∈ X X \in\mathcal{X} X ∈ X and Y ∈ Y Y \in\mathcal{Y} Y ∈ Y . A subcategory X \mathcal{X} X of a triangulated category is called closed under extensions if X ∗ X ⊆ X \mathcal{X} * \mathcal{X} \subseteq\mathcal{X} X ∗ X ⊆ X .
Let X \mathcal{X} X be an extension-closed subcategory of a triangulated category D \mathcal{D} D . A triangle in X \mathcal{X} X means a triangle in D \mathcal{D} D whose first three terms belong to X \mathcal{X} X . An object X X X in X \mathcal{X} X is called Ext-projective (resp. Ext-injective) if Hom D ( X , Y [ 1 ] ) = 0 \operatorname{Hom}_{\mathcal{D}}(X,Y[1])=0 Hom D ( X , Y [ 1 ]) = 0 (resp. Hom D ( Y , X [ 1 ] ) = 0 \operatorname{Hom}_{\mathcal{D}}(Y,X[1])=0 Hom D ( Y , X [ 1 ]) = 0 ) for all Y ∈ X Y \in\mathcal{X} Y ∈ X . We denote by proj X \operatorname{proj}\mathcal{X} proj X and inj X \operatorname{inj}\mathcal{X} inj X the subcategories of X \mathcal{X} X consisting of Ext-projective and Ext-injective objects in X \mathcal{X} X , respectively. We say that X \mathcal{X} X has enough projectives (resp. injectives) if for any object X ∈ X X \in\mathcal{X} X ∈ X , there exists a triangle in X \mathcal{X} X :
Y → P → X → Y [ 1 ] (resp. X → I → Z → X [ 1 ] ) , Y \to P \to X \to Y[1]\quad\text{(resp. }X \to I \to Z \to X[1]\text{)}, Y → P → X → Y [ 1 ] (resp. X → I → Z → X [ 1 ] ) ,
with P ∈ proj X P \in\operatorname{proj}\mathcal{X} P ∈ proj X (resp. I ∈ inj X I \in\operatorname{inj}\mathcal{X} I ∈ inj X ).
Preliminaries
Differential graded algebras, derived categories and adjunctions
We recall some basic notions and results on differential graded (dg) algebras, derived categories and adjunctions; cf. [17 , 18 , 21 , 30 , 35 ] .
All dg algebras and dg modules are defined over k k k and are Z \mathbb{Z} Z -graded, with differentials of degree 1 1 1 . Their underlying dg k k k -modules are complexes of k k k -vector spaces. For n ∈ Z n \in\mathbb{Z} n ∈ Z , we denote by σ ≤ n \sigma^{\leq n} σ ≤ n and σ ≥ n \sigma^{\geq n} σ ≥ n the canonical truncations of complexes of k k k -vector spaces; cf. [Z, Appendix A]. We write H i ( X ) H^i(X) H i ( X ) for the i i i -th cohomology of a complex of k k k -vector spaces X X X . Thus, we have
σ ≥ i ( σ ≤ i ( X ) ) ≅ H i ( X ) [ − i ] ≅ σ ≤ i ( σ ≥ i ( X ) ) . \sigma^{\geq i}(\sigma^{\leq i}(X)) \cong H^i(X)[-i] \cong\sigma^{\leq i}(\sigma^{\geq i}(X)). σ ≥ i ( σ ≤ i ( X )) ≅ H i ( X ) [ − i ] ≅ σ ≤ i ( σ ≥ i ( X )) .
Unless otherwise stated, all dg modules are right dg modules. We write R M S {}_{R}M_{S} R M S for a dg bimodule with the left R R R -action and right S S S -action. We omit action subscripts already determined by the ambient category or the subscript of Hom \operatorname{Hom} Hom or ⊗ \otimes ⊗ .
Let Λ \Lambda Λ be a dg algebra. For dg Λ \Lambda Λ -modules X , Y X,Y X , Y , we denote the dg Hom complex by H o m Λ ( X , Y ) \mathcal{H}om_{\Lambda}(X,Y) H o m Λ ( X , Y ) and the dg endomorphism algebra by E n d Λ ( X ) = H o m Λ ( X , X ) \mathcal{E}nd_{\Lambda}(X)=\mathcal{H}om_{\Lambda}(X,X) E n d Λ ( X ) = H o m Λ ( X , X ) , with multiplication given by composition. We denote by D ( Λ ) D(\Lambda) D ( Λ ) the derived category of right dg Λ \Lambda Λ -modules, by per Λ \operatorname{per}\Lambda per Λ its perfect derived category, and by D f d ( Λ ) D_{\mathrm{fd}}(\Lambda) D fd ( Λ ) the full subcategory of objects with finite-dimensional total cohomology.
A dg Λ \Lambda Λ -module P P P is K K K -projective if H o m Λ ( P , N ) \mathcal{H}om_{\Lambda}(P,N) H o m Λ ( P , N ) is acyclic for every acyclic dg Λ \Lambda Λ -module N N N . A K K K -projective resolution of a dg Λ \Lambda Λ -module X X X is a quasi-isomorphism P X → X P_X \to X P X → X where P X P_X P X is K K K -projective. Every dg module admits such a resolution. For dg Λ \Lambda Λ -modules X , N X,N X , N , the derived Hom RHom Λ ( X , N ) ∈ D ( k ) \operatorname{RHom}_{\Lambda}(X,N) \in D(k) RHom Λ ( X , N ) ∈ D ( k ) is represented by H o m Λ ( P X , N ) \mathcal{H}om_{\Lambda}(P_X,N) H o m Λ ( P X , N ) , where P X → X P_X \to X P X → X is a K K K -projective resolution. For every n ∈ Z n \in\mathbb{Z} n ∈ Z , we have
H n RHom Λ ( X , N ) ≅ Hom D ( Λ ) ( X , N [ n ] ) . H^n\operatorname{RHom}_{\Lambda}(X,N) \cong\operatorname{Hom}_{D(\Lambda)}(X,N[n]). H n RHom Λ ( X , N ) ≅ Hom D ( Λ ) ( X , N [ n ]) .
Let Λ ′ \Lambda' Λ ′ be another dg algebra and let Λ T Λ ′ {}_{\Lambda}T_{\Lambda'} Λ T Λ ′ be a dg Λ \Lambda Λ -Λ ′ \Lambda' Λ ′ -bimodule. For a dg Λ \Lambda Λ -module X X X , the derived tensor product X ⊗ Λ L T ∈ D ( Λ ′ ) X \otimes_{\Lambda}^{\mathbb{L}} T \in D(\Lambda') X ⊗ Λ L T ∈ D ( Λ ′ ) is represented by P X ⊗ Λ T P_X \otimes_{\Lambda} T P X ⊗ Λ T . If, moreover, X X X is a dg R R R -Λ \Lambda Λ -bimodule for some dg algebra R R R , we choose a K K K -projective resolution P X → X P_X \to X P X → X in the category of dg R o p ⊗ k Λ R^{\mathrm{op}} \otimes_k \Lambda R op ⊗ k Λ -modules. Since P X P_X P X remains K K K -projective as a right Λ \Lambda Λ -module, the above constructions then give RHom Λ ( X , N ) ∈ D ( R ) \operatorname{RHom}_{\Lambda}(X,N) \in D(R) RHom Λ ( X , N ) ∈ D ( R ) and X ⊗ Λ L T ∈ D ( R o p ⊗ k Λ ′ ) X \otimes_{\Lambda}^{\mathbb{L}} T \in D(R^{\mathrm{op}} \otimes_k \Lambda') X ⊗ Λ L T ∈ D ( R op ⊗ k Λ ′ ) . More generally, if N N N is a dg S S S -Λ \Lambda Λ -bimodule for some dg algebra S S S , then RHom Λ ( X , N ) \operatorname{RHom}_{\Lambda}(X,N) RHom Λ ( X , N ) belongs to D ( S o p ⊗ k R ) D(S^{\mathrm{op}} \otimes_k R) D ( S op ⊗ k R ) .
For M ∈ D ( Λ ) M \in D(\Lambda) M ∈ D ( Λ ) and N ∈ D ( Λ ′ ) N \in D(\Lambda') N ∈ D ( Λ ′ ) , there is a natural isomorphism in D ( k ) D(k) D ( k ) :
RHom Λ ′ ( M ⊗ L Λ T , N ) ≅ RHom Λ ( M , RHom Λ ′ ( Λ ′ T , N ) ) . (2) \operatorname{RHom}_{\Lambda'}\left(M \mathbin{\overset{\mathbb{L}}{\otimes}_{\Lambda}} T,N\right) \cong\operatorname{RHom}_{\Lambda}\left(M,\operatorname{RHom}_{\Lambda'}(\Lambda' T,N)\right). \tag*{(2)} RHom Λ ′ ( M ⊗ L Λ T , N ) ≅ RHom Λ ( M , RHom Λ ′ ( Λ ′ T , N ) ) . (2)
Taking H 0 H^0 H 0 provides the adjunction − ⊗ L Λ T ⊣ RHom Λ ′ ( Λ ′ T , − ) -\mathbin{\overset{\mathbb{L}}{\otimes}_{\Lambda}} T \dashv\operatorname{RHom}_{\Lambda'}(\Lambda' T,-) − ⊗ L Λ T ⊣ RHom Λ ′ ( Λ ′ T , − ) between D ( Λ ) D(\Lambda) D ( Λ ) and D ( Λ ′ ) D(\Lambda') D ( Λ ′ ) . Its unit and counit are
η M : M ⟶ RHom Λ ′ ( Λ ′ T , M ⊗ L Λ T ) , (3) \eta_M:M \longrightarrow\operatorname{RHom}_{\Lambda'}\left(\Lambda' T,M\mathbin{\overset{\mathbb{L}}{\otimes}_{\Lambda}}T\right), \tag*{(3)} η M : M ⟶ RHom Λ ′ ( Λ ′ T , M ⊗ L Λ T ) , (3)
ε N : RHom Λ ′ ( Λ ′ T , N ) ⊗ L Λ T ⟶ N (4) \varepsilon_N:\operatorname{RHom}_{\Lambda'}(\Lambda' T,N)\mathbin{\overset{\mathbb{L}}{\otimes}_{\Lambda}}T \longrightarrow N \tag*{(4)} ε N : RHom Λ ′ ( Λ ′ T , N ) ⊗ L Λ T ⟶ N (4)
in D ( Λ ) D(\Lambda) D ( Λ ) and D ( Λ ′ ) D(\Lambda') D ( Λ ′ ) , respectively.
More generally, if R M Λ R M_{\Lambda} R M Λ and S N Λ ′ S N_{\Lambda'} S N Λ ′ are dg bimodules, then (2.1) is an isomorphism in D ( S op ⊗ k R ) D(S^{\operatorname{op}}\otimes_k R) D ( S op ⊗ k R ) . The unit and counit in (2.2) and (2.3) are morphisms in D ( R op ⊗ k Λ ) D(R^{\operatorname{op}}\otimes_k\Lambda) D ( R op ⊗ k Λ ) and D ( S op ⊗ k Λ ′ ) D(S^{\operatorname{op}}\otimes_k\Lambda') D ( S op ⊗ k Λ ′ ) , respectively. Whenever this adjunction is an equivalence, the morphisms in (2.2) and (2.3) are isomorphisms.
For a dg algebra homomorphism f : Λ → Λ ′ f:\Lambda\to\Lambda' f : Λ → Λ ′ , the right Λ ′ \Lambda' Λ ′ -module Λ ′ \Lambda' Λ ′ becomes a dg Λ \Lambda Λ -Λ ′ \Lambda' Λ ′ -bimodule Λ Λ Λ ′ ′ {}_{\Lambda}\Lambda_{\Lambda'}' Λ Λ Λ ′ ′ , with left Λ \Lambda Λ -action induced by f f f . We denote by
f ∗ : D ( Λ ′ ) ⟶ D ( Λ ) f_*:D(\Lambda')\longrightarrow D(\Lambda) f ∗ : D ( Λ ′ ) ⟶ D ( Λ )
the restriction functor along f f f , and denote by
f ∗ : D ( Λ ) ⟶ D ( Λ ′ ) f^*:D(\Lambda)\longrightarrow D(\Lambda') f ∗ : D ( Λ ) ⟶ D ( Λ ′ )
the functor given by f ∗ ( M ) = M ⊗ L Λ Λ ′ f^*(M)=M\mathbin{\overset{\mathbb{L}}{\otimes}_{\Lambda}}\Lambda' f ∗ ( M ) = M ⊗ L Λ Λ ′ . The morphism f f f is a quasi-isomorphism if and only if f ∗ f_* f ∗ is an equivalence.
For any N ∈ D ( Λ ′ ) N\in D(\Lambda') N ∈ D ( Λ ′ ) , there exists a natural isomorphism in D ( Λ ) D(\Lambda) D ( Λ ) :
RHom Λ ′ ( Λ ′ , N ) ≅ f ∗ ( N ) . (5) \operatorname{RHom}_{\Lambda'}(\Lambda',N)\cong f_*(N). \tag*{(5)} RHom Λ ′ ( Λ ′ , N ) ≅ f ∗ ( N ) . (5)
Hence, we obtain the adjunction
f ∗ ⊣ f ∗ . f^* \dashv f_*. f ∗ ⊣ f ∗ .
In this case, using (2.4), the unit and counit in (2.2) and (2.3) become
η M : M ≅ M ⊗ L Λ Λ ′ → 1 M ⊗ L f f ∗ f ∗ ( M ) , (6) \eta_M:M\cong M\mathbin{\overset{\mathbb{L}}{\otimes}_{\Lambda}}\Lambda'\xrightarrow{1_M\mathbin{\overset{\mathbb{L}}{\otimes}f}}f_*f^*(M), \tag*{(6)} η M : M ≅ M ⊗ L Λ Λ ′ 1 M ⊗ L f f ∗ f ∗ ( M ) , (6)
ε N : f ∗ f ∗ ( N ) ⟶ N , (7) \varepsilon_N:f^*f_*(N)\longrightarrow N, \tag*{(7)} ε N : f ∗ f ∗ ( N ) ⟶ N , (7)
and the natural isomorphism given by this adjunction is
Hom D ( Λ ′ ) ( f ∗ ( M ) , N ) → ≅ Hom D ( Λ ) ( M , f ∗ ( N ) ) \operatorname{Hom}_{D(\Lambda')}(f^*(M),N)\xrightarrow{\ \cong\ }\operatorname{Hom}_{D(\Lambda)}(M,f_*(N)) Hom D ( Λ ′ ) ( f ∗ ( M ) , N ) ≅ Hom D ( Λ ) ( M , f ∗ ( N ))
h ⟼ f ∗ ( h ) ∘ η M . (8) h\longmapsto f_*(h)\circ\eta_M. \tag*{(8)} h ⟼ f ∗ ( h ) ∘ η M . (8)
For the case M = Λ M=\Lambda M = Λ , under f ∗ ( Λ ) ≅ Λ ′ f^*(\Lambda)\cong\Lambda' f ∗ ( Λ ) ≅ Λ ′ , we have
η Λ = f : Λ ⟶ f ∗ ( Λ ′ ) . (9) \eta_{\Lambda}=f:\Lambda\longrightarrow f_*(\Lambda'). \tag*{(9)} η Λ = f : Λ ⟶ f ∗ ( Λ ′ ) . (9)
A dg algebra Λ \Lambda Λ is proper if its total cohomology is finite dimensional. It is non-positive if Λ i = 0 \Lambda^i=0 Λ i = 0 for i > 0 i>0 i > 0 , and connective if H i ( Λ ) = 0 H^i(\Lambda)=0 H i ( Λ ) = 0 for i > 0 i>0 i > 0 . Following [23 Definition 1.1] , it is d-truncated if H i ( Λ ) = 0 H^i(\Lambda)=0 H i ( Λ ) = 0 for i ≤ − d i\leq-d i ≤ − d . Thus, a connective d d d -truncated dg algebra has cohomology concentrated in [ − d + 1 , 0 ] [-d+1,0] [ − d + 1 , 0 ] .
If Λ \Lambda Λ is non-positive, then D ( Λ ) D(\Lambda) D ( Λ ) has the canonical t t t -structure ( D ( Λ ) ≤ 0 , D ( Λ ) ≥ 0 ) (D(\Lambda)^{\leq0},D(\Lambda)^{\geq0}) ( D ( Λ ) ≤ 0 , D ( Λ ) ≥ 0 ) , where, for n ∈ Z n\in\mathbb{Z} n ∈ Z ,
D ( Λ ) ≤ n = { X ∈ D ( Λ ) ∣ H i ( X ) = 0 for all i > n } , D(\Lambda)^{\leq n}=\{X\in D(\Lambda)\mid H^i(X)=0\text{ for all }i>n\}, D ( Λ ) ≤ n = { X ∈ D ( Λ ) ∣ H i ( X ) = 0 for all i > n } ,
D ( Λ ) ≥ n = { X ∈ D ( Λ ) ∣ H i ( X ) = 0 for all i < n } . D(\Lambda)^{\geq n}=\{X\in D(\Lambda)\mid H^i(X)=0\text{ for all }i<n\}. D ( Λ ) ≥ n = { X ∈ D ( Λ ) ∣ H i ( X ) = 0 for all i < n } .
This t t t -structure restricts to a bounded t t t -structure on D f d ( Λ ) D_{\mathrm{fd}}(\Lambda) D fd ( Λ ) . The canonical truncations above induce the truncation functors σ Λ ≤ n \sigma_{\Lambda}^{\leq n} σ Λ ≤ n and σ Λ ≥ n \sigma_{\Lambda}^{\geq n} σ Λ ≥ n on D ( Λ ) D(\Lambda) D ( Λ ) . For each X ∈ D ( Λ ) X \in D(\Lambda) X ∈ D ( Λ ) , they give a triangle in D ( Λ ) D(\Lambda) D ( Λ )
σ Λ ≤ n ( X ) → X → σ Λ ≥ n + 1 ( X ) → σ Λ ≤ n ( X ) [ 1 ] . \sigma_{\Lambda}^{\leq n}(X) \to X \to\sigma_{\Lambda}^{\geq n+1}(X) \to\sigma_{\Lambda}^{\leq n}(X)[1]. σ Λ ≤ n ( X ) → X → σ Λ ≥ n + 1 ( X ) → σ Λ ≤ n ( X ) [ 1 ] .
For a morphism f : Λ → Λ ′ f:\Lambda\to\Lambda' f : Λ → Λ ′ of non-positive dg algebras, restriction preserves cohomology, is standard t t t -exact, commutes with the canonical truncations, and reflects isomorphisms.
Extended hearts and extended module categories
Let ( D ≤ 0 , D ≥ 0 ) (\mathcal{D}^{\leq0},\mathcal{D}^{\geq0}) ( D ≤ 0 , D ≥ 0 ) be a bounded t t t -structure on a triangulated category D \mathcal{D} D . Its d d d -extended heart is
D [ − d + 1 , 0 ] : = D ≥ − d + 1 ∩ D ≤ 0 . \mathcal{D}^{[-d+1,0]} := \mathcal{D}^{\geq-d+1} \cap\mathcal{D}^{\leq0}. D [ − d + 1 , 0 ] := D ≥ − d + 1 ∩ D ≤ 0 .
It is closed under extensions, and hence an extriangulated category in the sense of [27 ] . When d = 1 d=1 d = 1 , the d d d -extended heart coincides with the classical heart.
For a non-positive dg algebra Λ \Lambda Λ , its d d d -extended module category is the d d d -extended heart of the standard t t t -structure on D f d ( Λ ) D_{\mathrm{fd}}(\Lambda) D fd ( Λ ) :
d - mod Λ : = D f d ( Λ ) [ − d + 1 , 0 ] = { X ∈ D f d ( Λ ) ∣ H i ( X ) = 0 ∀ i ∉ [ − d + 1 , 0 ] } . d\text{-}\operatorname{mod}\Lambda:= D_{\mathrm{fd}}(\Lambda)^{[-d+1,0]} = \{X \in D_{\mathrm{fd}}(\Lambda) \mid H^{i}(X)=0\ \forall i \notin[-d+1,0]\}. d - mod Λ := D fd ( Λ ) [ − d + 1 , 0 ] = { X ∈ D fd ( Λ ) ∣ H i ( X ) = 0 ∀ i ∈ / [ − d + 1 , 0 ]} .
If Λ \Lambda Λ is proper, this category is Hom-finite, idempotent complete and Krull–Schmidt (see [26 ] ).
We compare the d d d -extended module categories of a non-positive dg algebra and its d d d -truncated version. This is also mentioned in [23 Remark 1.10] .
Proposition 2.1. Let Λ \Lambda Λ be a proper non-positive dg algebra. Let Λ ′ = σ Λ ≥ − d + 1 ( Λ ) \Lambda'=\sigma_{\Lambda}^{\geq-d+1}(\Lambda) Λ ′ = σ Λ ≥ − d + 1 ( Λ ) and let f : Λ → Λ ′ f:\Lambda\to\Lambda' f : Λ → Λ ′ be the canonical projection. Then the restriction functor f ∗ : D ( Λ ′ ) → D ( Λ ) f_*:D(\Lambda') \to D(\Lambda) f ∗ : D ( Λ ′ ) → D ( Λ ) induces an equivalence
f ∗ : d - mod Λ ′ → ∼ d - mod Λ , f_*:d\text{-}\operatorname{mod}\Lambda' \xrightarrow{\sim} d\text{-}\operatorname{mod}\Lambda, f ∗ : d - mod Λ ′ ∼ d - mod Λ , with quasi-inverse X ↦ σ Λ ′ ≤ 0 RHom Λ ( Λ Λ ′ ′ , X ) X \mapsto\sigma_{\Lambda'}^{\leq0}\operatorname{RHom}_{\Lambda}(\Lambda'_{\Lambda'},X) X ↦ σ Λ ′ ≤ 0 RHom Λ ( Λ Λ ′ ′ , X ) , where Λ ′ \Lambda' Λ ′ is regarded as a right dg Λ \Lambda Λ -module via f f f . Moreover, for all X , Y ∈ d - mod Λ ′ X,Y \in d\text{-}\operatorname{mod}\Lambda' X , Y ∈ d - mod Λ ′ , we have natural isomorphisms
Hom D ( Λ ′ ) ( X , Y [ i ] ) ≅ Hom D ( Λ ) ( f ∗ X , f ∗ Y [ i ] ) , i ≤ 1. (10) \operatorname{Hom}_{D(\Lambda')}(X,Y[i]) \cong\operatorname{Hom}_{D(\Lambda)}(f_*X,f_*Y[i]), \qquad i \leq1. \tag*{(10)} Hom D ( Λ ′ ) ( X , Y [ i ]) ≅ Hom D ( Λ ) ( f ∗ X , f ∗ Y [ i ]) , i ≤ 1. (10)
Proof. Let f ! = RHom Λ ( Λ Λ Λ ′ ′ , − ) : D ( Λ ) → D ( Λ ′ ) f^!=\operatorname{RHom}_{\Lambda}({}_{\Lambda}\Lambda'_{\Lambda'},-):D(\Lambda) \to D(\Lambda') f ! = RHom Λ ( Λ Λ Λ ′ ′ , − ) : D ( Λ ) → D ( Λ ′ ) be the right adjoint of f ∗ f_* f ∗ . For any X ∈ d - mod Λ X \in d\text{-}\operatorname{mod}\Lambda X ∈ d - mod Λ , we have f ! ( X ) ∈ D ( Λ ′ ) ≥ − d + 1 f^!(X) \in D(\Lambda')^{\geq-d+1} f ! ( X ) ∈ D ( Λ ′ ) ≥ − d + 1 . Moreover, the vector space
H j ( f ! ( X ) ) ≅ Hom D ( Λ ′ ) ( Λ ′ , f ! ( X ) [ j ] ) ≅ Hom D ( Λ ) ( f ∗ Λ ′ , X [ j ] ) H^j(f^!(X)) \cong\operatorname{Hom}_{D(\Lambda')}(\Lambda',f^!(X)[j]) \cong\operatorname{Hom}_{D(\Lambda)}(f_*\Lambda',X[j]) H j ( f ! ( X )) ≅ Hom D ( Λ ′ ) ( Λ ′ , f ! ( X ) [ j ]) ≅ Hom D ( Λ ) ( f ∗ Λ ′ , X [ j ])
is finite-dimensional. Hence, σ Λ ′ ≤ 0 f ! ( X ) ∈ d - mod Λ ′ \sigma_{\Lambda'}^{\leq0}f^!(X) \in d\text{-}\operatorname{mod}\Lambda' σ Λ ′ ≤ 0 f ! ( X ) ∈ d - mod Λ ′ . Applying RHom Λ ( − , X ) \operatorname{RHom}_{\Lambda}(-,X) RHom Λ ( − , X ) to the exact sequence of dg Λ \Lambda Λ -Λ \Lambda Λ -bimodules 0 → ker f → Λ → f Λ ′ → 0 0 \to\ker f \to\Lambda\xrightarrow{f} \Lambda' \to0 0 → ker f → Λ f Λ ′ → 0 gives a triangle in D ( Λ ) D(\Lambda) D ( Λ ) :
f ∗ f ! ( X ) → α X X → RHom Λ ( ker f , X ) → ( f ∗ f ! ( X ) ) [ 1 ] . (11) f_*f^!(X) \xrightarrow{\alpha_X} X \to\operatorname{RHom}_{\Lambda}(\ker f,X) \to(f_*f^!(X))[1]. \tag*{(11)} f ∗ f ! ( X ) α X X → RHom Λ ( ker f , X ) → ( f ∗ f ! ( X )) [ 1 ] . (11)
where α \alpha α is the counit of f ∗ ⊣ f ! f_* \dashv f^! f ∗ ⊣ f ! . Since ker f ∈ D ( Λ ) ≤ − d \ker f \in D(\Lambda)^{\leq-d} ker f ∈ D ( Λ ) ≤ − d and X ∈ D ( Λ ) ≥ − d + 1 X \in D(\Lambda)^{\geq-d+1} X ∈ D ( Λ ) ≥ − d + 1 , we have H j RHom Λ ( ker f , X ) ≅ Hom D ( Λ ) ( ker f , X [ j ] ) = 0 H^j\operatorname{RHom}_{\Lambda}(\ker f,X) \cong\operatorname{Hom}_{D(\Lambda)}(\ker f,X[j])=0 H j RHom Λ ( ker f , X ) ≅ Hom D ( Λ ) ( ker f , X [ j ]) = 0 for all j ≤ 0 j \leq0 j ≤ 0 . Consequently, α X \alpha_X α X induces isomorphisms in all degrees j ≤ 0 j \leq0 j ≤ 0 . It follows that there are natural isomorphisms
f ∗ ( σ Λ ′ ≤ 0 f ! ( X ) ) ≅ σ Λ ≤ 0 ( f ∗ f ! ( X ) ) → ≅ σ Λ ≤ 0 ( α X ) X , f_*\left(\sigma_{\Lambda'}^{\leq0}f^!(X)\right) \cong\sigma_{\Lambda}^{\leq0}\left(f_*f^!(X)\right) \xrightarrow[\cong]{\sigma_{\Lambda}^{\leq0}(\alpha_X)} X, f ∗ ( σ Λ ′ ≤ 0 f ! ( X ) ) ≅ σ Λ ≤ 0 ( f ∗ f ! ( X ) ) σ Λ ≤ 0 ( α X ) ≅ X ,
which implies the natural isomorphism between f ∗ ∘ σ Λ ′ ≤ 0 ∘ f ! f_* \circ\sigma_{\Lambda'}^{\leq0} \circ f^! f ∗ ∘ σ Λ ′ ≤ 0 ∘ f ! and the identity functor on d - mod Λ d\text{-}\operatorname{mod}\Lambda d - mod Λ . In particular, f ∗ f_* f ∗ is dense.
Since H 1 ( X ) = 0 H^1(X)=0 H 1 ( X ) = 0 , the triangle (11) also implies H 1 ( f ∗ f ! ( X ) ) = 0 H^1(f_*f^!(X))=0 H 1 ( f ∗ f ! ( X )) = 0 and hence H 1 ( α X ) H^1(\alpha_X) H 1 ( α X ) is also an isomorphism. For V ∈ d - mod Λ ′ V \in d\text{-}\operatorname{mod}\Lambda' V ∈ d - mod Λ ′ , let β V : V → f ! f ∗ ( V ) \beta_V: V \to f^!f_*(V) β V : V → f ! f ∗ ( V ) be the unit. The adjunction identity
α f ∗ ( V ) ∘ f ∗ ( β V ) = 1 f ∗ ( V ) \alpha_{f_*(V)} \circ f_*(\beta_V) = 1_{f_*(V)} α f ∗ ( V ) ∘ f ∗ ( β V ) = 1 f ∗ ( V )
and the isomorphism σ Λ ′ ≤ 1 ( α f ∗ ( V ) ) \sigma_{\Lambda'}^{\leq1}(\alpha_{f_*(V)}) σ Λ ′ ≤ 1 ( α f ∗ ( V ) ) imply that σ Λ ′ ≤ 1 ( β V ) \sigma_{\Lambda'}^{\leq1}(\beta_V) σ Λ ′ ≤ 1 ( β V ) is an isomorphism. Consequently, we have cone ( β V ) ∈ D ( Λ ′ ) ≥ 2 \operatorname{cone}(\beta_V) \in D(\Lambda')^{\geq2} cone ( β V ) ∈ D ( Λ ′ ) ≥ 2 . Therefore, for any U ∈ d - mod Λ ′ U \in d\text{-}\operatorname{mod}\Lambda' U ∈ d - mod Λ ′ and i ≤ 1 i \leq1 i ≤ 1 , we have natural isomorphisms
Hom D ( Λ ′ ) ( U , V [ i ] ) → ≅ β V [ i ] ∘ ( − ) Hom D ( Λ ′ ) ( U , f ! f ∗ ( V ) [ i ] ) ≅ Hom D ( Λ ) ( f ∗ ( U ) , f ∗ ( V ) [ i ] ) . \operatorname{Hom}_{D(\Lambda')}(U,V[i]) \xrightarrow[\cong]{\beta_V[i]\circ(-)} \operatorname{Hom}_{D(\Lambda')}(U,f^!f_*(V)[i]) \cong\operatorname{Hom}_{D(\Lambda)}(f_*(U),f_*(V)[i]). Hom D ( Λ ′ ) ( U , V [ i ]) β V [ i ] ∘ ( − ) ≅ Hom D ( Λ ′ ) ( U , f ! f ∗ ( V ) [ i ]) ≅ Hom D ( Λ ) ( f ∗ ( U ) , f ∗ ( V ) [ i ]) .
The composition of these two isomorphisms sends g : U → V [ i ] g: U \to V[i] g : U → V [ i ] to α f ∗ ( V ) [ i ] ∘ f ∗ ( β V [ i ] ∘ g ) = f ∗ ( g ) \alpha_{f_*(V)}[i]\circ f_*(\beta_V[i]\circ g)=f_*(g) α f ∗ ( V ) [ i ] ∘ f ∗ ( β V [ i ] ∘ g ) = f ∗ ( g ) . Hence, we obtain the required natural isomorphisms. Taking i = 0 i=0 i = 0 shows that f ∗ f_* f ∗ is fully faithful on d - mod Λ ′ d\text{-}\operatorname{mod}\Lambda' d - mod Λ ′ . Hence, f ∗ f_* f ∗ is an equivalence, with quasi-inverse σ Λ ′ ≤ 0 ∘ f ! \sigma_{\Lambda'}^{\leq0}\circ f^! σ Λ ′ ≤ 0 ∘ f ! . □
We next consider restriction along morphisms between d d d -truncated dg algebras.
Proposition 2.2. Let Λ \Lambda Λ and Λ ′ \Lambda' Λ ′ be proper d d d -truncated non-positive dg algebras, and let f : Λ → Λ ′ f:\Lambda\to\Lambda' f : Λ → Λ ′ be a dg algebra homomorphism. Then f f f is a quasi-isomorphism if and only if f ∗ f_* f ∗ restricts to an equivalence:
f ∗ : d - mod Λ ′ → d - mod Λ . f_*:d\text{-}\operatorname{mod}\Lambda' \to d\text{-}\operatorname{mod}\Lambda. f ∗ : d - mod Λ ′ → d - mod Λ.
Proof. Suppose that f f f is a quasi-isomorphism. Then f ∗ : D ( Λ ′ ) → D ( Λ ) f_*:D(\Lambda')\to D(\Lambda) f ∗ : D ( Λ ′ ) → D ( Λ ) is an equivalence. Since f ∗ f_* f ∗ preserves cohomology, it restricts to the required equivalence.
Conversely, suppose that f ∗ : d - mod Λ ′ → d - mod Λ f_*:d\text{-}\operatorname{mod}\Lambda'\to d\text{-}\operatorname{mod}\Lambda f ∗ : d - mod Λ ′ → d - mod Λ is an equivalence. By (9) , we have η Λ = f : Λ → f ∗ ( Λ ′ ) \eta_\Lambda=f:\Lambda\to f_*(\Lambda') η Λ = f : Λ → f ∗ ( Λ ′ ) under f ∗ ( Λ ) ≅ Λ ′ f_*(\Lambda)\cong\Lambda' f ∗ ( Λ ) ≅ Λ ′ . For any object Y ∈ d - mod Λ ′ Y\in d\text{-}\operatorname{mod}\Lambda' Y ∈ d - mod Λ ′ , by (8) , there exists a natural isomorphism
Hom D ( Λ ′ ) ( Λ ′ , Y ) → ≅ h ↦ f ∗ ( h ) ∘ f Hom D ( Λ ) ( Λ , f ∗ ( Y ) ) . \operatorname{Hom}_{D(\Lambda')}(\Lambda',Y)\xrightarrow[\cong]{h\mapsto f_*(h)\circ f}\operatorname{Hom}_{D(\Lambda)}(\Lambda,f_*(Y)). Hom D ( Λ ′ ) ( Λ ′ , Y ) h ↦ f ∗ ( h ) ∘ f ≅ Hom D ( Λ ) ( Λ , f ∗ ( Y )) .
Since f ∗ f_* f ∗ is fully faithful, this induces a natural isomorphism
Hom D ( Λ ) ( f ∗ ( Λ ′ ) , f ∗ ( Y ) ) → ≅ g ↦ g ∘ f Hom D ( Λ ) ( Λ , f ∗ ( Y ) ) . \operatorname{Hom}_{D(\Lambda)}(f_*(\Lambda'),f_*(Y))\xrightarrow[\cong]{g\mapsto g\circ f}\operatorname{Hom}_{D(\Lambda)}(\Lambda,f_*(Y)). Hom D ( Λ ) ( f ∗ ( Λ ′ ) , f ∗ ( Y )) g ↦ g ∘ f ≅ Hom D ( Λ ) ( Λ , f ∗ ( Y )) .
Since f ∗ f_* f ∗ is dense, we can replace f ∗ ( Y ) f_*(Y) f ∗ ( Y ) in the above natural isomorphism by any object in d - mod Λ d\text{-}\operatorname{mod}\Lambda d - mod Λ . Since Λ \Lambda Λ and Λ ′ \Lambda' Λ ′ are proper and d d d -truncated, we have Λ ∈ d - mod Λ \Lambda\in d\text{-}\operatorname{mod}\Lambda Λ ∈ d - mod Λ and Λ ′ ∈ d - mod Λ ′ \Lambda'\in d\text{-}\operatorname{mod}\Lambda' Λ ′ ∈ d - mod Λ ′ . Thus, by the Yoneda Lemma, f : Λ → f ∗ ( Λ ′ ) f:\Lambda\to f_*(\Lambda') f : Λ → f ∗ ( Λ ′ ) is an isomorphism in d - mod Λ d\text{-}\operatorname{mod}\Lambda d - mod Λ . Hence f f f is a quasi-isomorphism. □
For the remainder of this subsection, let Λ \Lambda Λ be a proper d d d -truncated non-positive dg algebra. By [26 Corollary 2.3] , d - mod Λ d\text{-}\operatorname{mod}\Lambda d - mod Λ has enough projectives and injectives, with
proj d - mod Λ = add Λ and inj d - mod Λ = add ( D ( Λ ) [ d − 1 ] ) . (12) \operatorname{proj}d\text{-}\operatorname{mod}\Lambda=\operatorname{add}\Lambda\quad\text{and}\quad\operatorname{inj}d\text{-}\operatorname{mod}\Lambda=\operatorname{add}\bigl(D(\Lambda)[d-1]\bigr). \tag*{(12)} proj d - mod Λ = add Λ and inj d - mod Λ = add ( D ( Λ ) [ d − 1 ] ) . (12)
By taking cohomology, we may relate the extended module category d - mod Λ d\text{-}\operatorname{mod}\Lambda d - mod Λ with the category mod H 0 ( Λ ) \operatorname{mod}H^0(\Lambda) mod H 0 ( Λ ) of finitely generated right H 0 ( Λ ) H^0(\Lambda) H 0 ( Λ ) -modules as follows.
There exist equivalences
H 0 : add Λ → ∼ proj H 0 ( Λ ) , (13) H^0:\operatorname{add}\Lambda\xrightarrow{\sim}\operatorname{proj}H^0(\Lambda), \tag*{(13)} H 0 : add Λ ∼ proj H 0 ( Λ ) , (13)
H − d + 1 : add ( D Λ [ d − 1 ] ) → ∼ inj H 0 ( Λ ) , (14) H^{-d+1}:\operatorname{add}\bigl(D\Lambda[d-1]\bigr)\xrightarrow{\sim}\operatorname{inj}H^0(\Lambda), \tag*{(14)} H − d + 1 : add ( D Λ [ d − 1 ] ) ∼ inj H 0 ( Λ ) , (14)
where proj H 0 ( Λ ) \operatorname{proj}H^0(\Lambda) proj H 0 ( Λ ) (resp., inj H 0 ( Λ ) \operatorname{inj}H^0(\Lambda) inj H 0 ( Λ ) ) denotes the subcategory of mod H 0 ( Λ ) \operatorname{mod}H^0(\Lambda) mod H 0 ( Λ ) consisting of projective (resp. injective) objects.
(2) For any morphism u : X → Y u:X\to Y u : X → Y in d d d -mod Λ \Lambda Λ , we have equivalences
cone ( u ) ∈ d -mod Λ ⟺ H − d + 1 ( u ) is injective , (15) \operatorname{cone}(u)\in d\text{-mod }\Lambda\Longleftrightarrow H^{-d+1}(u)\text{ is injective}, \tag*{(15)} cone ( u ) ∈ d -mod Λ ⟺ H − d + 1 ( u ) is injective , (15)
cone ( u ) [ − 1 ] ∈ d -mod Λ ⟺ H 0 ( u ) is surjective . (16) \operatorname{cone}(u)[-1]\in d\text{-mod }\Lambda\Longleftrightarrow H^{0}(u)\text{ is surjective}. \tag*{(16)} cone ( u ) [ − 1 ] ∈ d -mod Λ ⟺ H 0 ( u ) is surjective . (16)
Cf. also [26 Remark 1.1] .
(3) For any X ∈ d X\in d X ∈ d -mod Λ \Lambda Λ and any S ∈ mod H 0 ( Λ ) S\in\operatorname{mod}H^{0}(\Lambda) S ∈ mod H 0 ( Λ ) , regarding S S S as a dg Λ \Lambda Λ -module via Λ → H 0 ( Λ ) \Lambda\to H^{0}(\Lambda) Λ → H 0 ( Λ ) , we have natural isomorphisms
Hom D ( Λ ) ( X , S ) ≅ Hom H 0 ( Λ ) ( H 0 ( X ) , S ) , (17) \operatorname{Hom}_{D(\Lambda)}(X,S)\cong\operatorname{Hom}_{H^{0}(\Lambda)}(H^{0}(X),S), \tag*{(17)} Hom D ( Λ ) ( X , S ) ≅ Hom H 0 ( Λ ) ( H 0 ( X ) , S ) , (17)
Hom D ( Λ ) ( S [ d − 1 ] , X ) ≅ Hom H 0 ( Λ ) ( S , H − d + 1 ( X ) ) . (18) \operatorname{Hom}_{D(\Lambda)}(S[d-1],X)\cong\operatorname{Hom}_{H^{0}(\Lambda)}(S,H^{-d+1}(X)). \tag*{(18)} Hom D ( Λ ) ( S [ d − 1 ] , X ) ≅ Hom H 0 ( Λ ) ( S , H − d + 1 ( X )) . (18)
Definition 2.3 ([26 Definition 3.6] ). For any object X ∈ d X\in d X ∈ d -mod Λ \Lambda Λ , we define
top X = top H 0 ( Λ ) H 0 ( X ) and soc X = ( soc H 0 ( Λ ) H − d + 1 ( X ) ) [ d − 1 ] , \operatorname{top}X=\operatorname{top}_{H^{0}(\Lambda)}H^{0}(X)\quad\text{and}\quad\operatorname{soc}X=\left(\operatorname{soc}_{H^{0}(\Lambda)}H^{-d+1}(X)\right)[d-1], top X = top H 0 ( Λ ) H 0 ( X ) and soc X = ( soc H 0 ( Λ ) H − d + 1 ( X ) ) [ d − 1 ] ,
where the modules on the right are regarded as dg Λ \Lambda Λ -modules via Λ → H 0 ( Λ ) \Lambda\to H^{0}(\Lambda) Λ → H 0 ( Λ ) . Define rad X \operatorname{rad}X rad X and X / soc X X/\operatorname{soc}X X / soc X by the following triangles:
rad X → X → top X → ( rad X ) [ 1 ] , \operatorname{rad}X\to X\to\operatorname{top}X\to(\operatorname{rad}X)[1], rad X → X → top X → ( rad X ) [ 1 ] ,
soc X → X → X / soc X → ( soc X ) [ 1 ] . \operatorname{soc}X\to X\to X/\operatorname{soc}X\to(\operatorname{soc}X)[1]. soc X → X → X / soc X → ( soc X ) [ 1 ] .
Both rad X \operatorname{rad}X rad X and X / soc X X/\operatorname{soc}X X / soc X belong to d d d -mod Λ \Lambda Λ by the equivalences (15) and (16) .
s-torsion pairs and almost split triangles
In this section, we study basic properties of s s s -torsion pairs and almost split triangles in extended module categories. We begin in the more general setting of extended hearts.
Let D [ − d + 1 , 0 ] \mathcal{D}^{[-d+1,0]} D [ − d + 1 , 0 ] be the d d d -extended heart of a bounded t t t -structure on a triangulated category D \mathcal{D} D . An s s s -torsion pair ([1 Definition 3.1] ) in D [ − d + 1 , 0 ] \mathcal{D}^{[-d+1,0]} D [ − d + 1 , 0 ] is a pair ( T , F ) (\mathcal{T},\mathcal{F}) ( T , F ) of full subcategories closed under isomorphisms such that
Hom D ( T , F ) = 0 \operatorname{Hom}_{\mathcal{D}}(\mathcal{T},\mathcal{F})=0 Hom D ( T , F ) = 0 ,
D [ − d + 1 , 0 ] = T ∗ F \mathcal{D}^{[-d+1,0]}=\mathcal{T}*\mathcal{F} D [ − d + 1 , 0 ] = T ∗ F , and
Hom D ( T , F [ − 1 ] ) = 0 \operatorname{Hom}_{\mathcal{D}}(\mathcal{T},\mathcal{F}[-1])=0 Hom D ( T , F [ − 1 ]) = 0 .
The last condition implies Hom D ( T , F [ i ] ) = 0 \operatorname{Hom}_{\mathcal{D}}(\mathcal{T},\mathcal{F}[i])=0 Hom D ( T , F [ i ]) = 0 for all i < 0 i<0 i < 0 by [Z, Corollary 1.10]. For any s s s -torsion pair ( T , F ) (\mathcal{T},\mathcal{F}) ( T , F ) in D [ − d + 1 , 0 ] \mathcal{D}^{[-d+1,0]} D [ − d + 1 , 0 ] , we have
T = { X ∈ D [ − d + 1 , 0 ] ∣ Hom D ( X , F ) = 0 } , \mathcal{T}=\{X\in\mathcal{D}^{[-d+1,0]}\mid\operatorname{Hom}_{\mathcal{D}}(X,\mathcal{F})=0\}, T = { X ∈ D [ − d + 1 , 0 ] ∣ Hom D ( X , F ) = 0 } ,
F = { X ∈ D [ − d + 1 , 0 ] ∣ Hom D ( T , X ) = 0 } . \mathcal{F}=\{X\in\mathcal{D}^{[-d+1,0]}\mid\operatorname{Hom}_{\mathcal{D}}(\mathcal{T},X)=0\}. F = { X ∈ D [ − d + 1 , 0 ] ∣ Hom D ( T , X ) = 0 } .
Moreover, any object X ∈ D [ − d + 1 , 0 ] X\in\mathcal{D}^{[-d+1,0]} X ∈ D [ − d + 1 , 0 ] admits a triangle
t X → X → f X → ( t X ) [ 1 ] , (19) tX\to X\to fX\to(tX)[1], \tag*{(19)} tX → X → f X → ( tX ) [ 1 ] , (19)
with t X ∈ T tX\in\mathcal{T} tX ∈ T and f X ∈ F fX\in\mathcal{F} f X ∈ F . This triangle is unique up to isomorphism ([1 Proposition 3.7] ), and is called the canonical triangle of X X X with respect to the s s s -torsion pair ( T , F ) (\mathcal{T},\mathcal{F}) ( T , F ) . These triangles define additive functors
t : D [ − d + 1 , 0 ] → T and f : D [ − d + 1 , 0 ] → F , (20) t:\mathcal{D}^{[-d+1,0]}\to\mathcal{T}\quad\text{and}\quad f:\mathcal{D}^{[-d+1,0]}\to\mathcal{F}, \tag*{(20)} t : D [ − d + 1 , 0 ] → T and f : D [ − d + 1 , 0 ] → F , (20)
which are called torsion and torsion-free functors with respect to the s s s -torsion pair ( T , F ) (\mathcal{T},\mathcal{F}) ( T , F ) .
Let A A A be a proper d d d -truncated non-positive dg algebra.
Lemma 3.1. Let ( T , F ) (\mathcal{T},\mathcal{F}) ( T , F ) be an s s s -torsion pair in d d d -mod A A A . Then
inj T = add t ( D A [ d − 1 ] ) and proj F = add f A . \operatorname{inj}\mathcal{T}=\operatorname{add}t(DA[d-1])\quad\text{and}\quad\operatorname{proj}\mathcal{F}=\operatorname{add}fA. inj T = add t ( D A [ d − 1 ]) and proj F = add f A . Moreover, T \mathcal{T} T has enough Ext-injectives and F \mathcal{F} F has enough Ext-projectives.
Proof. We only show the assertions for T \mathcal{T} T , because those for F \mathcal{F} F can be shown dually.
For any I ∈ add ( D A [ d − 1 ] ) I\in\operatorname{add}(DA[d-1]) I ∈ add ( D A [ d − 1 ]) and T ∈ T T\in\mathcal{T} T ∈ T , applying Hom D ( A ) ( T , − ) \operatorname{Hom}_{D(A)}(T,-) Hom D ( A ) ( T , − ) to the canonical triangle of I I I with respect to ( T , F ) (\mathcal{T},\mathcal{F}) ( T , F ) yields an exact sequence
Hom D ( A ) ( T , f I ) ⟶ Hom D ( A ) ( T , t I [ 1 ] ) ⟶ Hom D ( A ) ( T , I [ 1 ] ) , \operatorname{Hom}_{D(A)}(T,fI)\longrightarrow\operatorname{Hom}_{D(A)}(T,tI[1])\longrightarrow\operatorname{Hom}_{D(A)}(T,I[1]), Hom D ( A ) ( T , f I ) ⟶ Hom D ( A ) ( T , t I [ 1 ]) ⟶ Hom D ( A ) ( T , I [ 1 ]) ,
where the first term vanishes since T ∈ T T\in\mathcal{T} T ∈ T and f I ∈ F fI\in\mathcal{F} f I ∈ F , and the last term vanishes, since by [2.11], I ∈ inj d I\in\operatorname{inj}d I ∈ inj d -mod A A A . Hence, we have Hom D ( A ) ( T , t I [ 1 ] ) = 0 \operatorname{Hom}_{D(A)}(T,tI[1])=0 Hom D ( A ) ( T , t I [ 1 ]) = 0 , which implies t I ∈ inj T tI\in\operatorname{inj}\mathcal{T} t I ∈ inj T .
For any X ∈ T X\in\mathcal{T} X ∈ T , since d d d -mod A A A has enough injectives add ( D A [ d − 1 ] ) \operatorname{add}(DA[d-1]) add ( D A [ d − 1 ]) , there exists I 0 ∈ add ( D A [ d − 1 ] ) I_0\in\operatorname{add}(DA[d-1]) I 0 ∈ add ( D A [ d − 1 ]) and a triangle X → α I 0 ⟶ L ⟶ X [ 1 ] X\xrightarrow{\alpha}I_0\longrightarrow L\longrightarrow X[1] X α I 0 ⟶ L ⟶ X [ 1 ] with L ∈ d L\in d L ∈ d -mod A A A . Consider the canonical triangle of I 0 I_0 I 0 with respect to ( T , F ) (\mathcal{T},\mathcal{F}) ( T , F ) :
X → α I 0 → L → X [ 1 ] α ′ ↓ γ ↓ t I 0 → β I 0 → f I 0 → t I 0 [ 1 ] \begin{CD}
X @>{\alpha}>> I_0 @>>> L @>>> X[1]\\
@V{\alpha'}VV @V{\gamma}VV\\
tI_0 @>{\beta}>> I_0 @>>> fI_0 @>>> tI_0[1]
\end{CD} X α ′ ↓ ⏐ t I 0 α β I 0 γ ↓ ⏐ I 0 L f I 0 X [ 1 ] t I 0 [ 1 ]
We have γ ∘ α = 0 \gamma\circ\alpha=0 γ ∘ α = 0 by X ∈ T X\in\mathcal{T} X ∈ T and f ( I 0 ) ∈ F f(I_0)\in\mathcal{F} f ( I 0 ) ∈ F . So there exists a morphism α ′ : X → t I 0 \alpha':X\to tI_0 α ′ : X → t I 0 such that α = β ∘ α ′ \alpha=\beta\circ\alpha' α = β ∘ α ′ . The injectivity of H − d + 1 ( α ) H^{-d+1}(\alpha) H − d + 1 ( α ) implies that of H − d + 1 ( α ′ ) H^{-d+1}(\alpha') H − d + 1 ( α ′ ) . Hence, by [2.14], cone ( α ′ ) ∈ d \operatorname{cone}(\alpha')\in d cone ( α ′ ) ∈ d -mod A A A . Since both X X X and t I 0 tI_0 t I 0 belong to T \mathcal{T} T , by [12 Lemma 3.9] , it follows that cone ( α ′ ) ∈ T \operatorname{cone}(\alpha')\in\mathcal{T} cone ( α ′ ) ∈ T . Consequently, we obtain a triangle
X → α ′ t I 0 ⟶ cone ( α ′ ) ⟶ X [ 1 ] X\xrightarrow{\alpha'}tI_0\longrightarrow\operatorname{cone}(\alpha')\longrightarrow X[1] X α ′ t I 0 ⟶ cone ( α ′ ) ⟶ X [ 1 ]
in T \mathcal{T} T with middle term Ext-injective, which implies that T \mathcal{T} T has enough injectives. If X ∈ inj T X\in\operatorname{inj}\mathcal{T} X ∈ inj T , this triangle splits. Thus, we have inj T = add t ( D A [ d − 1 ] ) \operatorname{inj}\mathcal{T}=\operatorname{add}t(DA[d-1]) inj T = add t ( D A [ d − 1 ]) . □
An object P ∈ per A P\in\operatorname{per}A P ∈ per A is called silting if Hom per A ( P , P [ i ] ) = 0 \operatorname{Hom}_{\operatorname{per}A}(P,P[i])=0 Hom per A ( P , P [ i ]) = 0 for all i > 0 i>0 i > 0 , and thick ( P ) = per A \operatorname{thick}(P)=\operatorname{per}A thick ( P ) = per A . It is called tilting if, in addition, Hom D ( A ) ( P , P [ i ] ) = 0 \operatorname{Hom}_{D(A)}(P,P[i])=0 Hom D ( A ) ( P , P [ i ]) = 0 for all i < 0 i<0 i < 0 . An object of per A \operatorname{per}A per A is called ( d + 1 ) (d+1) ( d + 1 ) -term if it belongs to add A ∗ add A [ 1 ] ∗ ⋯ ∗ add A [ d ] \operatorname{add}A*\operatorname{add}A[1]*\cdots*\operatorname{add}A[d] add A ∗ add A [ 1 ] ∗ ⋯ ∗ add A [ d ] .
Let P P P be a ( d + 1 ) (d+1) ( d + 1 ) -term silting object in per A \operatorname{per}A per A . We will use the following consequence of [3 Lemma 3.7] .
Lemma 3.2. We have A [ d ] ∈ add P ∗ add P [ 1 ] ∗ ⋯ ∗ add P [ d ] A[d]\in\operatorname{add}P*\operatorname{add}P[1]*\cdots*\operatorname{add}P[d] A [ d ] ∈ add P ∗ add P [ 1 ] ∗ ⋯ ∗ add P [ d ] .
By [33 Proposition 2.1] , the subcategories
D f d ( P ) ≤ 0 = { X ∈ D f d ( A ) ∣ Hom D ( A ) ( P , X [ i ] ) = 0 for i > 0 } , D_{\mathrm{fd}}(P)^{\leq0}=\{X\in D_{\mathrm{fd}}(A)\mid\operatorname{Hom}_{D(A)}(P,X[i])=0\text{ for }i>0\}, D fd ( P ) ≤ 0 = { X ∈ D fd ( A ) ∣ Hom D ( A ) ( P , X [ i ]) = 0 for i > 0 } ,
D f d ( P ) ≥ 0 = { X ∈ D f d ( A ) ∣ Hom D ( A ) ( P , X [ i ] ) = 0 for i < 0 } D_{\mathrm{fd}}(P)^{\geq0}=\{X\in D_{\mathrm{fd}}(A)\mid\operatorname{Hom}_{D(A)}(P,X[i])=0\text{ for }i<0\} D fd ( P ) ≥ 0 = { X ∈ D fd ( A ) ∣ Hom D ( A ) ( P , X [ i ]) = 0 for i < 0 }
form a bounded t t t -structure, and we have
D f d ( A ) ≤ − d ⊆ D f d ( P ) ≤ 0 ⊆ D f d ( A ) ≤ 0 , D_{\mathrm{fd}}(A)^{\leq-d}\subseteq D_{\mathrm{fd}}(P)^{\leq0}\subseteq D_{\mathrm{fd}}(A)^{\leq0}, D fd ( A ) ≤ − d ⊆ D fd ( P ) ≤ 0 ⊆ D fd ( A ) ≤ 0 ,
D f d ( A ) ≥ 1 ⊆ D f d ( P ) ≥ 1 ⊆ D f d ( A ) ≥ − d + 1 . (21) D_{\mathrm{fd}}(A)^{\geq1}\subseteq D_{\mathrm{fd}}(P)^{\geq1}\subseteq D_{\mathrm{fd}}(A)^{\geq-d+1}. \tag*{(21)} D fd ( A ) ≥ 1 ⊆ D fd ( P ) ≥ 1 ⊆ D fd ( A ) ≥ − d + 1 . (21)
Then the d d d -extended heart of this t t t -structure is
D f d ( P ) [ − d + 1 , 0 ] = D f d ( P ) ≤ 0 ∩ D f d ( P ) ≥ − d + 1 = { X ∈ D f d ( A ) ∣ Hom D ( A ) ( P , X [ i ] ) = 0 for i ∉ [ − d + 1 , 0 ] } . \begin{aligned}
D_{\mathrm{fd}}(P)^{[-d+1,0]}=D_{\mathrm{fd}}(P)^{\leq0}\cap D_{\mathrm{fd}}(P)^{\geq-d+1} \\
&=\{X\in D_{\mathrm{fd}}(A)\mid\operatorname{Hom}_{D(A)}(P,X[i])=0\text{ for }i\notin[-d+1,0]\}.
\end{aligned} D fd ( P ) [ − d + 1 , 0 ] = D fd ( P ) ≤ 0 ∩ D fd ( P ) ≥ − d + 1 = { X ∈ D fd ( A ) ∣ Hom D ( A ) ( P , X [ i ]) = 0 for i ∈ / [ − d + 1 , 0 ]} .
The subcategories
T ( P ) = D f d ( P ) ≤ 0 ∩ d - mod A and F ( P ) = D f d ( P ) ≥ 1 ∩ d - mod A \mathcal{T}(\mathbf{P})=D_{\mathrm{fd}}(\mathbf{P})^{\leq0}\cap d\text{-}\operatorname{mod} A\quad\text{and}\quad\mathcal{F}(\mathbf{P})=D_{\mathrm{fd}}(\mathbf{P})^{\geq1}\cap d\text{-}\operatorname{mod} A T ( P ) = D fd ( P ) ≤ 0 ∩ d - mod A and F ( P ) = D fd ( P ) ≥ 1 ∩ d - mod A
form a functorially finite s s s -torsion pair in d - mod A d\text{-}\operatorname{mod} A d - mod A ; see [15 Corollary 4.1] . It is straightforward to check that
D f d ( P ) [ − d + 1 , 0 ] ∩ d - mod A = D f d ( P ) [ − d + 1 , 0 ] ∩ D ( A ) ≥ − d + 1 = T ( P ) , D_{\mathrm{fd}}(\mathbf{P})^{[-d+1,0]}\cap d\text{-}\operatorname{mod} A=D_{\mathrm{fd}}(\mathbf{P})^{[-d+1,0]}\cap D(A)^{\geq-d+1}=\mathcal{T}(\mathbf{P}), D fd ( P ) [ − d + 1 , 0 ] ∩ d - mod A = D fd ( P ) [ − d + 1 , 0 ] ∩ D ( A ) ≥ − d + 1 = T ( P ) ,
D f d ( P ) [ − d + 1 , 0 ] ∩ d - mod A [ d ] = D f d ( P ) [ − d + 1 , 0 ] ∩ D ( A ) ≤ − d = F ( P ) [ d ] . D_{\mathrm{fd}}(\mathbf{P})^{[-d+1,0]}\cap d\text{-}\operatorname{mod} A[d]=D_{\mathrm{fd}}(\mathbf{P})^{[-d+1,0]}\cap D(A)^{\leq-d}=\mathcal{F}(\mathbf{P})[d]. D fd ( P ) [ − d + 1 , 0 ] ∩ d - mod A [ d ] = D fd ( P ) [ − d + 1 , 0 ] ∩ D ( A ) ≤ − d = F ( P ) [ d ] .
We denote by t P \mathsf{t}_{\mathbf{P}} t P and f P \mathsf{f}_{\mathbf{P}} f P the torsion and torsion-free functors with respect to the s s s -torsion pair ( T ( P ) , F ( P ) ) (\mathcal{T}(\mathbf{P}),\mathcal{F}(\mathbf{P})) ( T ( P ) , F ( P )) ; see (20) .
The Nakayama functor ν A = − ⊗ A L D A \nu_A=-\otimes_A^{\mathbf{L}}DA ν A = − ⊗ A L D A restricts to an equivalence add A ≃ add D A \operatorname{add} A\simeq\operatorname{add} DA add A ≃ add D A . For P ∈ per A P\in\operatorname{per} A P ∈ per A and M ∈ D f d ( A ) M\in D_{\mathrm{fd}}(A) M ∈ D fd ( A ) , there are natural isomorphisms
D Hom D ( A ) ( P , M ) ≅ Hom D ( A ) ( M , ν A P ) . (22) \mathbb{D}\operatorname{Hom}_{D(A)}(P,M)\cong\operatorname{Hom}_{D(A)}(M,\nu_A P). \tag*{(22)} D Hom D ( A ) ( P , M ) ≅ Hom D ( A ) ( M , ν A P ) . (22)
The following proposition gives the dg version of [Z, Proposition 2.14 and Corollary 4.13]. For an object X X X , let ∣ X ∣ |X| ∣ X ∣ denote the number of isomorphism classes of its indecomposable direct summands.
Proposition 3.3. The following hold.
(1) Both T ( P ) \mathcal{T}(\mathbf{P}) T ( P ) and F ( P ) \mathcal{F}(\mathbf{P}) F ( P ) have enough E x t Ext E x t -projectives and E x t Ext E x t -injectives, and
proj T ( P ) = add σ A ≥ − d + 1 ( P ) , inj T ( P ) = add t P ( D A [ d − 1 ] ) , \operatorname{proj}\mathcal{T}(\mathbf{P})=\operatorname{add}\sigma_A^{\geq-d+1}(\mathbf{P}),\qquad\operatorname{inj}\mathcal{T}(\mathbf{P})=\operatorname{add}\mathsf{t}_{\mathbf{P}}(DA[d-1]), proj T ( P ) = add σ A ≥ − d + 1 ( P ) , inj T ( P ) = add t P ( D A [ d − 1 ]) , proj F ( P ) = add f P ( A ) , inj F ( P ) = add σ A ≤ 0 ( ν A P [ − 1 ] ) . \operatorname{proj}\mathcal{F}(\mathbf{P})=\operatorname{add}\mathsf{f}_{\mathbf{P}}(A),\qquad\operatorname{inj}\mathcal{F}(\mathbf{P})=\operatorname{add}\sigma_A^{\leq0}(\nu_A\mathbf{P}[-1]). proj F ( P ) = add f P ( A ) , inj F ( P ) = add σ A ≤ 0 ( ν A P [ − 1 ]) . (2) For any indecomposable P ∈ add A P\in\operatorname{add} A P ∈ add A , f P ( P ) \mathsf{f}_{\mathbf{P}}(P) f P ( P ) is either zero or indecomposable, and for indecomposable P , P ′ ∈ add A P,P'\in\operatorname{add} A P , P ′ ∈ add A , if f P ( P ) ≅ f P ( P ′ ) ≠ 0 \mathsf{f}_{\mathbf{P}}(P)\cong\mathsf{f}_{\mathbf{P}}(P')\ne0 f P ( P ) ≅ f P ( P ′ ) = 0 then P ≅ P ′ P\cong P' P ≅ P ′ . For any indecomposable I ∈ add D A [ d − 1 ] I\in\operatorname{add} DA[d-1] I ∈ add D A [ d − 1 ] , t P ( I ) \mathsf{t}_{\mathbf{P}}(I) t P ( I ) is either zero or indecomposable, and for indecomposable I , I ′ ∈ add D A [ d − 1 ] I,I'\in\operatorname{add} DA[d-1] I , I ′ ∈ add D A [ d − 1 ] , if t P ( I ) ≅ t P ( I ′ ) ≠ 0 \mathsf{t}_{\mathbf{P}}(I)\cong\mathsf{t}_{\mathbf{P}}(I')\ne0 t P ( I ) ≅ t P ( I ′ ) = 0 then I ≅ I ′ I\cong I' I ≅ I ′ .
(3) For any indecomposable P ∈ add A P\in\operatorname{add} A P ∈ add A , we have f P ( P ) = 0 \mathsf{f}_{\mathbf{P}}(P)=0 f P ( P ) = 0 if and only if P ∈ add P P\in\operatorname{add}\mathbf{P} P ∈ add P ; t P ( ν A ( P ) [ d − 1 ] ) = 0 \mathsf{t}_{\mathbf{P}}(\nu_A(P)[d-1])=0 t P ( ν A ( P ) [ d − 1 ]) = 0 if and only if P [ d ] ∈ add P P[d]\in\operatorname{add}\mathbf{P} P [ d ] ∈ add P . In particular, at least one of f P ( P ) \mathsf{f}_{\mathbf{P}}(P) f P ( P ) and t P ( ν A ( P ) [ d − 1 ] ) \mathsf{t}_{\mathbf{P}}(\nu_A(P)[d-1]) t P ( ν A ( P ) [ d − 1 ]) is not zero.
(4) Let P 1 , … , P n P_1,\ldots,P_n P 1 , … , P n form a complete set of non-isomorphic indecomposable objects in add A \operatorname{add} A add A . Then
∣ σ A ≥ − d + 1 P ∣ = ∣ t P ( D A [ d − 1 ] ) ∣ = n − # { i ∣ P i [ d ] ∈ add P } , |\sigma_A^{\geq-d+1}\mathbf{P}|=|\mathsf{t}_{\mathbf{P}}(DA[d-1])|=n-\#\{i\mid P_i[d]\in\operatorname{add}\mathbf{P}\}, ∣ σ A ≥ − d + 1 P ∣ = ∣ t P ( D A [ d − 1 ]) ∣ = n − # { i ∣ P i [ d ] ∈ add P } , ∣ σ A ≤ 0 ( ν A ( P ) [ − 1 ] ) ∣ = ∣ f P ( A ) ∣ = n − # { i ∣ P i ∈ add P } . |\sigma_A^{\leq0}(\nu_A(\mathbf{P})[-1])|=|\mathsf{f}_{\mathbf{P}}(A)|=n-\#\{i\mid P_i\in\operatorname{add}\mathbf{P}\}. ∣ σ A ≤ 0 ( ν A ( P ) [ − 1 ]) ∣ = ∣ f P ( A ) ∣ = n − # { i ∣ P i ∈ add P } . In particular, for each of T ( P ) \mathcal{T}(\mathbf{P}) T ( P ) and F ( P ) \mathcal{F}(\mathbf{P}) F ( P ) , the numbers of isomorphism classes of indecomposable E x t Ext E x t -projectives and E x t Ext E x t -injectives are equal.
Proof. We prove the assertions concerning T ( P ) \mathcal{T}(\mathbf{P}) T ( P ) ; those concerning F ( P ) \mathcal{F}(\mathbf{P}) F ( P ) follow dually. For (1), applying [33 Proposition 2.12(a), (c)] to our setting yields
proj T ( P ) = add σ A ≥ − d + 1 ( P ) , \operatorname{proj}\mathcal{T}(\mathbf{P})=\operatorname{add}\sigma_A^{\geq-d+1}(\mathbf{P}), proj T ( P ) = add σ A ≥ − d + 1 ( P ) ,
and shows that T ( P ) \mathcal{T}(\mathbf{P}) T ( P ) has enough E x t Ext E x t -projectives. The assertions concerning its E x t Ext E x t -injectives follow from Lemma 3.1.
For (2), let I , I ′ ∈ add ( D A [ d − 1 ] ) I,I'\in\operatorname{add}(DA[d-1]) I , I ′ ∈ add ( D A [ d − 1 ]) . Applying Hom D ( A ) ( − , I ′ ) \operatorname{Hom}_{D(A)}(-,I') Hom D ( A ) ( − , I ′ ) to the canonical triangle t P I → I → f P I → t P I [ 1 ] \mathsf{t}_{\mathbf{P}}I\to I\to\mathsf{f}_{\mathbf{P}}I\to\mathsf{t}_{\mathbf{P}}I[1] t P I → I → f P I → t P I [ 1 ] of I I I with respect to ( T ( P ) , F ( P ) ) (\mathcal{T}(\mathbf{P}),\mathcal{F}(\mathbf{P})) ( T ( P ) , F ( P )) yields an exact sequence
Hom D ( A ) ( I , I ′ ) → Hom D ( A ) ( t P I , I ′ ) → Hom D ( A ) ( f P I , I ′ [ 1 ] ) = 0. (23) \operatorname{Hom}_{D(A)}(I,I')\to\operatorname{Hom}_{D(A)}(\mathsf{t}_{\mathbf{P}}I,I')\to\operatorname{Hom}_{D(A)}(\mathsf{f}_{\mathbf{P}}I,I'[1])=0. \tag*{(23)} Hom D ( A ) ( I , I ′ ) → Hom D ( A ) ( t P I , I ′ ) → Hom D ( A ) ( f P I , I ′ [ 1 ]) = 0. (23)
where the last term vanishes since I ′ I' I ′ is Ext-injective in d d d -mod A A A . Applying the functor Hom D ( A ) ( t P I , − ) \operatorname{Hom}_{D(A)}(t_P I,-) Hom D ( A ) ( t P I , − ) to the canonical triangle t P I ′ → I ′ → f P I ′ → t P I ′ [ 1 ] t_P I' \to I' \to f_P I' \to t_P I'[1] t P I ′ → I ′ → f P I ′ → t P I ′ [ 1 ] of I ′ I' I ′ yields an exact sequence
0 = Hom D ( A ) ( t P I , f P I ′ [ − 1 ] ) → Hom D ( A ) ( t P I , t P I ′ ) → Hom D ( A ) ( t P I , I ′ ) → Hom D ( A ) ( t P I , f P I ′ ) = 0. \begin{aligned}
0=\operatorname{Hom}_{D(A)}(t_P I,f_P I'[-1])\to\operatorname{Hom}_{D(A)}(t_P I,t_P I')\to\operatorname{Hom}_{D(A)}(t_P I,I') \\
&\to\operatorname{Hom}_{D(A)}(t_P I,f_P I')=0.
\end{aligned} 0 = Hom D ( A ) ( t P I , f P I ′ [ − 1 ]) → Hom D ( A ) ( t P I , t P I ′ ) → Hom D ( A ) ( t P I , I ′ ) → Hom D ( A ) ( t P I , f P I ′ ) = 0.
where the first and last terms vanish since ( T ( P ) , F ( P ) ) (\mathcal{T}(P),\mathcal{F}(P)) ( T ( P ) , F ( P )) is an s s s -torsion pair. Combining with the surjective map in (23) , this implies that t P t_P t P induces a surjective map
Hom D ( A ) ( I , I ′ ) → Hom D ( A ) ( t P I , t P I ′ ) → 0. \operatorname{Hom}_{D(A)}(I,I')\to\operatorname{Hom}_{D(A)}(t_P I,t_P I')\to0. Hom D ( A ) ( I , I ′ ) → Hom D ( A ) ( t P I , t P I ′ ) → 0.
It follows that, if I I I is indecomposable, then End D ( A ) ( t P I ) \operatorname{End}_{D(A)}(t_P I) End D ( A ) ( t P I ) is a quotient of the local algebra End D ( A ) ( I ) \operatorname{End}_{D(A)}(I) End D ( A ) ( I ) and hence is either zero or local. Thus t P I t_P I t P I is zero or indecomposable. Moreover, if I I I and I ′ I' I ′ are indecomposable and t P ( I ) ≅ t P ( I ′ ) ≠ 0 t_P(I)\cong t_P(I')\ne0 t P ( I ) ≅ t P ( I ′ ) = 0 , then there exist morphisms u : I → I ′ u:I\to I' u : I → I ′ and v : I ′ → I v:I'\to I v : I ′ → I such that t P ( v ) ∘ t P ( u ) = 1 t P ( I ) t_P(v)\circ t_P(u)=1_{t_P(I)} t P ( v ) ∘ t P ( u ) = 1 t P ( I ) . It follows that v ∘ u v\circ u v ∘ u is not nilpotent. Since End D ( A ) ( I ) \operatorname{End}_{D(A)}(I) End D ( A ) ( I ) is local and finite-dimensional, v ∘ u v\circ u v ∘ u is invertible. Thus u u u is a section, and the indecomposability of I ′ I' I ′ implies that I ≅ I ′ I\cong I' I ≅ I ′ .
For (3), let P ∈ add A P\in\operatorname{add} A P ∈ add A be indecomposable. By definition, t P ( ν A ( P ) [ d − 1 ] ) = 0 t_P(\nu_A(P)[d-1])=0 t P ( ν A ( P ) [ d − 1 ]) = 0 if and only if ν A ( P ) [ d − 1 ] ∈ F ( P ) \nu_A(P)[d-1]\in\mathcal{F}(P) ν A ( P ) [ d − 1 ] ∈ F ( P ) , while by (22) , the latter is equivalent to Hom D ( A ) ( P [ d ] , P [ i ] ) = 0 \operatorname{Hom}_{D(A)}(P[d],P[i])=0 Hom D ( A ) ( P [ d ] , P [ i ]) = 0 for all i > 0 i>0 i > 0 . Since P ∈ add A ∗ ⋯ ∗ add A [ d ] P\in\operatorname{add} A*\cdots*\operatorname{add} A[d] P ∈ add A ∗ ⋯ ∗ add A [ d ] , we also have Hom D ( A ) ( P , P [ d + i ] ) = 0 \operatorname{Hom}_{D(A)}(P,P[d+i])=0 Hom D ( A ) ( P , P [ d + i ]) = 0 for all i > 0 i>0 i > 0 . Hence, this is equivalent to P ⊕ P [ d ] P\oplus P[d] P ⊕ P [ d ] being silting, and therefore to P [ d ] ∈ add P P[d]\in\operatorname{add}P P [ d ] ∈ add P by [4 Theorem 2.18] .
If both f P ( P ) f_P(P) f P ( P ) and t P ( ν A ( P ) [ d − 1 ] ) t_P(\nu_A(P)[d-1]) t P ( ν A ( P ) [ d − 1 ]) are zero, then both P P P and P [ d ] P[d] P [ d ] belong to add P \operatorname{add}P add P , contradicting Hom D ( A ) ( P , P [ d ] ) = 0 \operatorname{Hom}_{D(A)}(P,P[d])=0 Hom D ( A ) ( P , P [ d ]) = 0 .
For (4), we have ∣ P ∣ = ∣ A ∣ = n |P|=|A|=n ∣ P ∣ = ∣ A ∣ = n by [4 Corollary 2.28] . Let r = # { i ∣ P i [ d ] ∈ add P } r=\#\{i\mid P_i[d]\in\operatorname{add}P\} r = # { i ∣ P i [ d ] ∈ add P } . By the equivalence induced by truncation in [26 Theorem 2.5] , the indecomposable direct summands of P P P not belonging to add A [ d ] \operatorname{add} A[d] add A [ d ] correspond bijectively, up to isomorphism, to the indecomposable direct summands of σ A ≥ − d + 1 ( P ) \sigma_A^{\geq-d+1}(P) σ A ≥ − d + 1 ( P ) . Consequently, ∣ σ A ≥ − d + 1 ( P ) ∣ = n − r |\sigma_A^{\geq-d+1}(P)|=n-r ∣ σ A ≥ − d + 1 ( P ) ∣ = n − r . By (2) and (3), the non-zero objects t P ( ν A ( P i ) [ d − 1 ] ) t_P(\nu_A(P_i)[d-1]) t P ( ν A ( P i ) [ d − 1 ]) are indecomposable and pairwise non-isomorphic, and exactly r r r of these n n n objects vanish. Thus ∣ t P ( D A [ d − 1 ] ) ∣ = n − r |t_P(DA[d-1])|=n-r ∣ t P ( D A [ d − 1 ]) ∣ = n − r , as required. □
We next recall some facts about almost split triangles. Let E ⊆ d \mathcal{E}\subseteq d E ⊆ d -mod A A A be an extension-closed additive subcategory closed under direct summands. A non-split triangle
X → f M → g Z → X [ 1 ] X\xrightarrow{f}M\xrightarrow{g}Z\to X[1] X f M g Z → X [ 1 ]
in E \mathcal{E} E is called almost split in E \mathcal{E} E starting at an indecomposable object X X X and ending at an indecomposable object Z Z Z if, for every U ∈ E U\in\mathcal{E} U ∈ E , every morphism X → U X\to U X → U which is not a section factors through f f f , and every morphism U → Z U\to Z U → Z which is not a retraction factors through g g g . We say that E \mathcal{E} E has Auslander–Reiten triangles for any indecomposable object Z Z Z which is not Ext-projective (resp. Ext-injective) in E \mathcal{E} E , there exists an almost split triangle in E \mathcal{E} E ending (resp. starting) at it.
By [26 Theorem 1.2 and Corollary 1.3] , d d d -mod A A A has Auslander–Reiten triangles. More precisely, for any indecomposable object Z Z Z of d d d -mod A A A which is not Ext-projective, there exists an almost split triangle in d d d -mod A A A :
τ [ d ] A ( Z ) → Y → Z → τ [ d ] A ( Z ) [ 1 ] . \tau^{A}_{[d]}(Z)\to Y\to Z\to\tau^{A}_{[d]}(Z)[1]. τ [ d ] A ( Z ) → Y → Z → τ [ d ] A ( Z ) [ 1 ] .
For any indecomposable object X X X of d d d -mod A A A which is not Ext-injective, there exists an almost split triangle in d d d -mod A A A :
X → W → ( τ [ d ] A ) − 1 X → X [ 1 ] . X\to W\to(\tau^{A}_{[d]})^{-1}X\to X[1]. X → W → ( τ [ d ] A ) − 1 X → X [ 1 ] .
Here, for any object Z ∈ d Z \in d Z ∈ d -mod A A A , the object τ [ d ] A ( Z ) \tau_{[d]}^{A}(Z) τ [ d ] A ( Z ) is defined up to isomorphism as follows: by [26 Theorem 2.5] , there exists a ( d + 1 ) (d+1) ( d + 1 ) -term object R ∈ per A R \in\operatorname{per} A R ∈ per A having no non-zero direct summand in add A [ d ] \operatorname{add} A[d] add A [ d ] such that Z ≅ σ A ≥ − d + 1 R Z \cong\sigma_{A}^{\geq-d+1}R Z ≅ σ A ≥ − d + 1 R . Then
τ [ d ] A ( Z ) : = σ A ≤ 0 ( ν A R [ − 1 ] ) . (24) \tau_{[d]}^{A}(Z):=\sigma_{A}^{\leq0}(\nu_{A}R[-1]). \tag*{(24)} τ [ d ] A ( Z ) := σ A ≤ 0 ( ν A R [ − 1 ]) . (24)
The object ( τ [ d ] A ) − 1 X (\tau_{[d]}^{A})^{-1}X ( τ [ d ] A ) − 1 X for X ∈ d X \in d X ∈ d -mod A A A is defined dually.
The following is the analogue of [32 Lemma 0.1] in extended module categories.
Lemma 3.4. Let ( T , F ) (\mathcal{T},\mathcal{F}) ( T , F ) be an s s s -torsion pair in d d d -mod A A A .
(1) An indecomposable object X X X in F \mathcal{F} F is Ext-injective in F \mathcal{F} F if and only if either X ∈ inj d X \in\operatorname{inj} d X ∈ inj d -mod A A A , or X ∉ inj d X \notin\operatorname{inj} d X ∈ / inj d -mod A A A and ( τ [ d ] A ) − 1 X ∈ T (\tau_{[d]}^{A})^{-1}X \in\mathcal{T} ( τ [ d ] A ) − 1 X ∈ T .
(2) An indecomposable object Z Z Z in T \mathcal{T} T is Ext-projective in T \mathcal{T} T if and only if either Z ∈ proj d Z \in\operatorname{proj} d Z ∈ proj d -mod A A A , or Z ∉ proj d Z \notin\operatorname{proj} d Z ∈ / proj d -mod A A A and τ [ d ] A ( Z ) ∈ F \tau_{[d]}^{A}(Z) \in\mathcal{F} τ [ d ] A ( Z ) ∈ F .
In particular, if X → Y → Z → X [ 1 ] X \to Y \to Z \to X[1] X → Y → Z → X [ 1 ] is an almost split triangle in d d d -mod A A A , with X ∈ F X \in\mathcal{F} X ∈ F and Z ∈ T Z \in\mathcal{T} Z ∈ T , then X ∈ inj F X \in\operatorname{inj}\mathcal{F} X ∈ inj F and Z ∈ proj T Z \in\operatorname{proj}\mathcal{T} Z ∈ proj T .
Proof. We prove (1); assertion (2) is dual. The case X ∈ inj d X \in\operatorname{inj} d X ∈ inj d -mod A A A is trivial. Hence, we assume X ∉ inj d X \notin\operatorname{inj} d X ∈ / inj d -mod A A A . Let Z = ( τ [ d ] A ) − 1 X Z=(\tau_{[d]}^{A})^{-1}X Z = ( τ [ d ] A ) − 1 X , and let
X → f W → g Z → δ X [ 1 ] X \xrightarrow{f} W \xrightarrow{g} Z \xrightarrow{\delta} X[1] X f W g Z δ X [ 1 ]
be the almost split triangle in d d d -mod A A A starting at X X X .
The “if” part: suppose Z ∈ T Z \in\mathcal{T} Z ∈ T . For any ε ∈ Hom D ( A ) ( F , X [ 1 ] ) \varepsilon\in\operatorname{Hom}_{D(A)}(\mathcal{F},X[1]) ε ∈ Hom D ( A ) ( F , X [ 1 ]) with F ∈ F F \in\mathcal{F} F ∈ F , if ε ≠ 0 \varepsilon\ne0 ε = 0 , complete it to a triangle X → a E → F → ε X [ 1 ] X \xrightarrow{a} E \to F \xrightarrow{\varepsilon} X[1] X a E → F ε X [ 1 ] in d d d -mod A A A . Since a a a is not a section, there exists a morphism b : W → E b\colon W \to E b : W → E such that a = b ∘ f a=b\circ f a = b ∘ f . Consequently, there exists a morphism c : Z → F c\colon Z \to F c : Z → F such that the following diagram commutes:
X → f W → g Z → δ X [ 1 ] ∥ ↓ b ↓ c ↓ ∥ ↓ X → a E → F → ε X [ 1 ] \begin{CD}
X @>{f}>> W @>{g}>> Z @>{\delta}>> X[1]\\
@V{\Vert}VV @V{b}VV @V{c}VV @V{\Vert}VV\\
X @>{a}>> E @>>> F @>{\varepsilon}>> X[1]
\end{CD} X ∥ ↓ ⏐ X f a W b ↓ ⏐ E g Z c ↓ ⏐ F δ ε X [ 1 ] ∥ ↓ ⏐ X [ 1 ]
Since Z ∈ T Z \in\mathcal{T} Z ∈ T and F ∈ F F \in\mathcal{F} F ∈ F , we have c = 0 c=0 c = 0 , which implies δ = ε ∘ c = 0 \delta=\varepsilon\circ c=0 δ = ε ∘ c = 0 , a contradiction. Thus, ε = 0 \varepsilon=0 ε = 0 . Hence, X ∈ inj F X \in\operatorname{inj}\mathcal{F} X ∈ inj F .
The “only if” part: suppose that X ∈ inj F X \in\operatorname{inj}\mathcal{F} X ∈ inj F and Z ∉ T Z \notin\mathcal{T} Z ∈ / T . Then, in the canonical triangle of Z Z Z with respect to ( T , F ) (\mathcal{T},\mathcal{F}) ( T , F )
t Z → α Z → β f Z → ( t Z ) [ 1 ] , tZ \xrightarrow{\alpha} Z \xrightarrow{\beta} fZ \to(tZ)[1], tZ α Z β f Z → ( tZ ) [ 1 ] ,
the morphism α \alpha α is not a retraction. It therefore factors through g g g , which implies that δ ∘ α = 0 \delta\circ\alpha=0 δ ∘ α = 0 . Hence, δ \delta δ factors through β \beta β . However, since f Z ∈ F fZ \in\mathcal{F} f Z ∈ F and X ∈ inj F X \in\operatorname{inj}\mathcal{F} X ∈ inj F , we have Hom D ( A ) ( f Z , X [ 1 ] ) = 0 \operatorname{Hom}_{D(A)}(fZ,X[1])=0 Hom D ( A ) ( f Z , X [ 1 ]) = 0 . Therefore, we have δ = 0 \delta=0 δ = 0 , a contradiction.
The final assertion follows from (1) and (2). □
Higher Brenner-Butler theorem
Let A A A be a proper d d d -truncated non-positive dg algebra. Let P P P be a ( d + 1 ) (d+1) ( d + 1 ) -term silting object in per A \operatorname{per} A per A . In this section, we prove a higher Brenner-Butler theorem for the s s s -torsion pair ( T ( P ) , F ( P ) ) (\mathcal{T}(P),\mathcal{F}(P)) ( T ( P ) , F ( P )) in d d d -mod A A A .
Fix a K K K -projective resolution of P \mathbf{P} P , which is still denoted by P \mathbf{P} P . Let End A ( P ) \operatorname{End}_{A}(\mathbf{P}) End A ( P ) be the dg endomorphism algebra of P \mathbf{P} P . It is connective since P \mathbf{P} P is silting, and is proper since A A A is proper and P ∈ per A \mathbf{P} \in\operatorname{per} A P ∈ per A . Let
Γ = σ ≤ 0 ( End A ( P ) ) \Gamma=\sigma^{\leq0}\bigl(\operatorname{End}_{A}(\mathbf{P})\bigr) Γ = σ ≤ 0 ( End A ( P ) )
be the non-positive truncation of End A ( P ) \operatorname{End}_{A}(\mathbf{P}) End A ( P ) . Thus the inclusion
j P : Γ ⟶ End A ( P ) \mathsf{j}_{\mathbf{P}}\colon\Gamma\longrightarrow\operatorname{End}_{A}(\mathbf{P}) j P : Γ ⟶ End A ( P )
is a quasi-isomorphism of dg algebras. This inclusion gives P \mathbf{P} P a structure of dg left Γ \Gamma Γ -module, and the functor
RHom A ( Γ P , − ) : D ( A ) ⟶ D ( Γ ) \operatorname{RHom}_{A}(\Gamma\mathbf{P},-)\colon D(A)\longrightarrow D(\Gamma) RHom A ( Γ P , − ) : D ( A ) ⟶ D ( Γ )
is a triangle equivalence, whose quasi-inverse is given by
− ⊗ Γ L P : D ( Γ ) ⟶ D ( A ) . -\mathbin{\mathop{\otimes}^{\mathbf{L}}_{\Gamma}}\mathbf{P}\colon D(\Gamma)\longrightarrow D(A). − ⊗ Γ L P : D ( Γ ) ⟶ D ( A ) .
See [17 Lemma 6.1(a)] and [18 Lemma 3.10] . These restrict to the equivalences
per A ≃ per Γ and D f d ( A ) ≃ D f d ( Γ ) . \operatorname{per} A\simeq\operatorname{per}\Gamma
\quad\text{and}\quad
D_{\mathrm{fd}}(A)\simeq D_{\mathrm{fd}}(\Gamma). per A ≃ per Γ and D fd ( A ) ≃ D fd ( Γ ) .
Recall that the d d d -extended heart of the t t t -structure induced by P \mathbf{P} P is
D f d ( P ) [ − d + 1 , 0 ] = { X ∈ D f d ( A ) ∣ Hom D ( A ) ( P , X [ i ] ) = 0 for i ∉ [ − d + 1 , 0 ] } . D_{\mathrm{fd}}(\mathbf{P})^{[-d+1,0]}
=
\{X\in D_{\mathrm{fd}}(A)\mid\operatorname{Hom}_{D(A)}(\mathbf{P},X[i])=0\text{ for }i\notin[-d+1,0]\}. D fd ( P ) [ − d + 1 , 0 ] = { X ∈ D fd ( A ) ∣ Hom D ( A ) ( P , X [ i ]) = 0 for i ∈ / [ − d + 1 , 0 ]} .
Then the above mutually inverse functors RHom A ( Γ P , − ) \operatorname{RHom}_{A}(\Gamma\mathbf{P},-) RHom A ( Γ P , − ) and − ⊗ Γ L P -\mathbin{\mathop{\otimes}^{\mathbf{L}}_{\Gamma}}\mathbf{P} − ⊗ Γ L P restrict to equivalences between D f d ( P ) [ − d + 1 , 0 ] D_{\mathrm{fd}}(\mathbf{P})^{[-d+1,0]} D fd ( P ) [ − d + 1 , 0 ] and d d d -mod Γ \Gamma Γ .
Definition 4.1. The d d d -truncated dg endomorphism algebra of P \mathbf{P} P is defined to be
d - End A ( P ) : = σ ≥ − d + 1 ( Γ ) . d\text{-}\operatorname{End}_{A}(\mathbf{P}):=\sigma^{\geq-d+1}(\Gamma). d - End A ( P ) := σ ≥ − d + 1 ( Γ ) .
The d d d -truncated dg endomorphism algebra d d d -End A ( P ) \operatorname{End}_{A}(\mathbf{P}) End A ( P ) is a proper d d d -truncated non-positive dg algebra. In what follows, we write B B B for d d d -End A ( P ) \operatorname{End}_{A}(\mathbf{P}) End A ( P ) . Up to quasi-isomorphism, Γ \Gamma Γ and B B B do not depend on the choice of the K K K -projective resolution of P \mathbf{P} P .
We write
p = p P : Γ ↠ B p=p_{\mathbf{P}}\colon\Gamma\twoheadrightarrow B p = p P : Γ ↠ B
for the canonical quotient morphism. Let p ∗ : D ( B ) ⟶ D ( Γ ) p_{*}\colon D(B)\longrightarrow D(\Gamma) p ∗ : D ( B ) ⟶ D ( Γ ) be the restriction functor along p p p , and p ∗ : = − ⊗ Γ L B : D ( Γ ) ⟶ D ( B ) p^{*}:=-\mathbin{\mathop{\otimes}^{\mathbf{L}}_{\Gamma}}B\colon D(\Gamma)\longrightarrow D(B) p ∗ := − ⊗ Γ L B : D ( Γ ) ⟶ D ( B ) be its left adjoint. Composing the functors p ∗ p_{*} p ∗ and − ⊗ Γ L P -\mathbin{\mathop{\otimes}^{\mathbf{L}}_{\Gamma}}\mathbf{P} − ⊗ Γ L P , we obtain a triangle functor
real P : D ( B ) ⟶ D ( A ) \operatorname{real}_{\mathbf{P}}\colon D(B)\longrightarrow D(A) real P : D ( B ) ⟶ D ( A )
N ⟼ p ∗ ( N ) ⊗ Γ L P , (25) N\longmapsto p_{*}(N)\mathbin{\mathop{\otimes}^{\mathbf{L}}_{\Gamma}\mathbf{P},} \tag*{(25)} N ⟼ p ∗ ( N ) ⊗ Γ L P , (25)
which restricts to a triangle functor D f d ( B ) ⟶ D f d ( A ) D_{\mathrm{fd}}(B)\longrightarrow D_{\mathrm{fd}}(A) D fd ( B ) ⟶ D fd ( A ) . By the adjunction p ∗ ⊣ p ∗ p^{*}\dashv p_{*} p ∗ ⊣ p ∗ and the Hom-tensor adjunction, we have the adjunction
p ∗ ∘ RHom A ( Γ P , − ) ⊣ real P . p^{*}\circ\operatorname{RHom}_{A}(\Gamma\mathbf{P},-)\dashv\operatorname{real}_{\mathbf{P}}. p ∗ ∘ RHom A ( Γ P , − ) ⊣ real P .
One can realize the d d d -extended module category d d d -mod B B B via the d d d -extended heart D f d ( P ) [ − d + 1 , 0 ] D_{\mathrm{fd}}(\mathbf{P})^{[-d+1,0]} D fd ( P ) [ − d + 1 , 0 ] as follows.
Theorem 4.2. The functor
Θ P : D f d ( P ) [ − d + 1 , 0 ] ⟶ d -mod B \Theta_{\mathbf{P}}\colon D_{\mathrm{fd}}(\mathbf{P})^{[-d+1,0]}\longrightarrow d\text{-mod }B Θ P : D fd ( P ) [ − d + 1 , 0 ] ⟶ d -mod B M ⟼ σ B ≤ 0 RHom A ( B ⊗ Γ L P , M ) M\longmapsto\sigma_{B}^{\leq0}\operatorname{RHom}_{A}\left(B\mathbin{\mathop{\otimes}^{\mathbf{L}}_{\Gamma}}\mathbf{P},M\right) M ⟼ σ B ≤ 0 RHom A ( B ⊗ Γ L P , M )
is an equivalence, whose quasi-inverse is given by real P \operatorname{real}_{\mathbf{P}} real P . The equivalence preserves and reflects triangles whose first three terms belong to the respective extended hearts. Moreover, for any U , V ∈ d - mod B U,V\in d\text{-}\operatorname{mod} B U , V ∈ d - mod B , the functor real P \operatorname{real}_{\mathbf{P}} real P induces natural isomorphisms
Hom D ( B ) ( U , V [ i ] ) ≅ Hom D ( A ) ( real P ( U ) , real P ( V ) [ i ] ) , i ≤ 1. (26) \operatorname{Hom}_{D(B)}(U,V[i])\cong\operatorname{Hom}_{D(A)}(\operatorname{real}_{\mathbf{P}}(U),\operatorname{real}_{\mathbf{P}}(V)[i]),\qquad i\leq1. \tag*{(26)} Hom D ( B ) ( U , V [ i ]) ≅ Hom D ( A ) ( real P ( U ) , real P ( V ) [ i ]) , i ≤ 1. (26)
Proof. This follows directly from Proposition 2.1 and the equivalence between the categories D f d ( P ) [ − d + 1 , 0 ] D_{\mathrm{fd}}(\mathbf{P})^{[-d+1,0]} D fd ( P ) [ − d + 1 , 0 ] and d - mod Γ d\text{-}\operatorname{mod}\Gamma d - mod Γ , noting that there exists a natural isomorphism
RHom A ( B B ⊗ L Γ P , M ) ≅ RHom Γ ( B B , RHom A ( Γ P , M ) ) . \operatorname{RHom}_{A}(B B\mathop{\otimes^{\mathbf{L}}}_{\Gamma}\mathbf{P},M)\cong\operatorname{RHom}_{\Gamma}(B B,\operatorname{RHom}_{A}(\Gamma\mathbf{P},M)). RHom A ( B B ⊗ L Γ P , M ) ≅ RHom Γ ( B B , RHom A ( Γ P , M )) .
□
We consider the following full subcategories of d - mod B d\text{-}\operatorname{mod} B d - mod B :
X ( P ) : = Θ P ( F ( P ) [ d ] ) and Y ( P ) : = Θ P ( T ( P ) ) . \mathcal{X}(\mathbf{P}):=\Theta_{\mathbf{P}}(\mathcal{F}(\mathbf{P})[d])\quad\text{and}\quad\mathcal{Y}(\mathbf{P}):=\Theta_{\mathbf{P}}(\mathcal{T}(\mathbf{P})). X ( P ) := Θ P ( F ( P ) [ d ]) and Y ( P ) := Θ P ( T ( P )) .
Corollary 4.3. The pair ( X ( P ) , Y ( P ) ) (\mathcal{X}(\mathbf{P}),\mathcal{Y}(\mathbf{P})) ( X ( P ) , Y ( P )) is an s s s -torsion pair in d - mod B d\text{-}\operatorname{mod} B d - mod B . There are equivalences T ( P ) ≃ Y ( P ) \mathcal{T}(\mathbf{P})\simeq\mathcal{Y}(\mathbf{P}) T ( P ) ≃ Y ( P ) and F ( P ) ≃ X ( P ) \mathcal{F}(\mathbf{P})\simeq\mathcal{X}(\mathbf{P}) F ( P ) ≃ X ( P ) .
Proof. Since ( T ( P ) , F ( P ) ) (\mathcal{T}(\mathbf{P}),\mathcal{F}(\mathbf{P})) ( T ( P ) , F ( P )) is an s s s -torsion pair in d - mod A d\text{-}\operatorname{mod} A d - mod A , by [Z, Theorem 1.12], the pair ( F ( P ) [ d ] , T ( P ) ) (\mathcal{F}(\mathbf{P})[d],\mathcal{T}(\mathbf{P})) ( F ( P ) [ d ] , T ( P )) is an s s s -torsion pair in the d d d -extended heart D f d ( P ) [ − d + 1 , 0 ] D_{\mathrm{fd}}(\mathbf{P})^{[-d+1,0]} D fd ( P ) [ − d + 1 , 0 ] . Thus, the required assertion follows from Theorem 4.2 . □
Let Q ~ = RHom A ( Γ P , A A [ d ] ) ∈ D ( A o p ⊗ k Γ ) \widetilde{Q}=\operatorname{RHom}_{A}(\Gamma\mathbf{P},A A[d])\in D(A^{\mathrm{op}}\mathbin{\otimes_k}\Gamma) Q = RHom A ( Γ P , AA [ d ]) ∈ D ( A op ⊗ k Γ ) . We define
Q = p ∗ ( RHom A ( Γ P , A [ d ] ) ) ∈ D ( A o p ⊗ k B ) . (27) Q=p^*\bigl(\operatorname{RHom}_{A}(\Gamma\mathbf{P},A[d])\bigr)\in D(A^{\mathrm{op}}\mathbin{\otimes_k}B). \tag*{(27)} Q = p ∗ ( RHom A ( Γ P , A [ d ]) ) ∈ D ( A op ⊗ k B ) . (27)
Forgetting the left A A A -module structure, we have Q ∈ D ( B ) Q\in D(B) Q ∈ D ( B ) .
Proposition 4.4. The object Q Q Q is a ( d + 1 ) (d+1) ( d + 1 ) -term silting object in per B \operatorname{per} B per B .
Proof. Since A [ d ] A[d] A [ d ] is a silting object in per A \operatorname{per} A per A , and by Lemma 3.2 , A [ d ] ∈ add P ∗ add P [ 1 ] ∗ ⋯ ∗ add P [ d ] A[d]\in\operatorname{add}\mathbf{P}*\operatorname{add}\mathbf{P}[1]*\cdots*\operatorname{add}\mathbf{P}[d] A [ d ] ∈ add P ∗ add P [ 1 ] ∗ ⋯ ∗ add P [ d ] , applying the triangle equivalence RHom A ( Γ P , − ) \operatorname{RHom}_{A}(\Gamma\mathbf{P},-) RHom A ( Γ P , − ) , we obtain that Q ~ \widetilde{Q} Q is a silting object in per Γ \operatorname{per}\Gamma per Γ , and
Q ~ ∈ add Γ ∗ add Γ [ 1 ] ∗ ⋯ ∗ add Γ [ d ] ⊆ D ( Γ ) ≤ 0 . (28) \widetilde{Q}\in\operatorname{add}\Gamma*\operatorname{add}\Gamma[1]*\cdots*\operatorname{add}\Gamma[d]\subseteq D(\Gamma)^{\leq0}. \tag*{(28)} Q ∈ add Γ ∗ add Γ [ 1 ] ∗ ⋯ ∗ add Γ [ d ] ⊆ D ( Γ ) ≤ 0 . (28)
Applying the functor p ∗ p^* p ∗ to (28) gives Q ∈ add B ∗ add B [ 1 ] ∗ ⋯ ∗ add B [ d ] Q\in\operatorname{add}B*\operatorname{add}B[1]*\cdots*\operatorname{add}B[d] Q ∈ add B ∗ add B [ 1 ] ∗ ⋯ ∗ add B [ d ] . Hence Q Q Q is a ( d + 1 ) (d+1) ( d + 1 ) -term object in per B \operatorname{per} B per B .
Applying the functor Q ~ ⊗ L Γ − \widetilde{Q}\mathop{\otimes^{\mathbf{L}}}_{\Gamma}- Q ⊗ L Γ − to the canonical exact sequence of dg Γ \Gamma Γ -Γ \Gamma Γ -bimodules 0 → ker p → Γ → p B → 0 0\to\ker p\to\Gamma\xrightarrow{p}B\to0 0 → ker p → Γ p B → 0 gives a triangle in D ( Γ ) D(\Gamma) D ( Γ ) :
Z ⟶ Q ~ → η Q ~ p ∗ Q ⟶ Z [ 1 ] , (29) Z\longrightarrow\widetilde{Q}\xrightarrow{\eta_{\widetilde{Q}}}p_*Q\longrightarrow Z[1], \tag*{(29)} Z ⟶ Q η Q p ∗ Q ⟶ Z [ 1 ] , (29)
where η \eta η is the unit of the adjunction p ∗ ⊣ p ∗ p^*\dashv p_* p ∗ ⊣ p ∗ (cf. (6) ), and Z = Q ~ ⊗ L Γ ker p Z=\widetilde{Q}\mathop{\otimes^{\mathbf{L}}}_{\Gamma}\ker p Z = Q ⊗ L Γ ker p . Applying − ⊗ L Γ ker p -\mathop{\otimes^{\mathbf{L}}}_{\Gamma}\ker p − ⊗ L Γ ker p to (28) , we obtain
Z ∈ add ( ker p ) ∗ add ( ( ker p ) [ 1 ] ) ∗ ⋯ ∗ add ( ( ker p ) [ d ] ) ⊆ D ( Γ ) ≤ − d . Z\in\operatorname{add}(\ker p)*\operatorname{add}((\ker p)[1])*\cdots*\operatorname{add}((\ker p)[d])\subseteq D(\Gamma)^{\leq-d}. Z ∈ add ( ker p ) ∗ add (( ker p ) [ 1 ]) ∗ ⋯ ∗ add (( ker p ) [ d ]) ⊆ D ( Γ ) ≤ − d .
Applying the functor Hom D ( Γ ) ( Q ~ , − ) \operatorname{Hom}_{D(\Gamma)}(\widetilde{Q},-) Hom D ( Γ ) ( Q , − ) to triangle (29) gives a long exact sequence
⋯ ⟶ Hom D ( Γ ) ( Q ~ , Q ~ [ i ] ) ⟶ Hom D ( Γ ) ( Q ~ , p ∗ Q [ i ] ) ⟶ Hom D ( Γ ) ( Q ~ , Z [ i + 1 ] ) ⟶ ⋯ . \cdots\longrightarrow\operatorname{Hom}_{D(\Gamma)}(\widetilde{Q},\widetilde{Q}[i])\longrightarrow\operatorname{Hom}_{D(\Gamma)}(\widetilde{Q},p_*Q[i])\longrightarrow\operatorname{Hom}_{D(\Gamma)}(\widetilde{Q},Z[i+1])\longrightarrow\cdots. ⋯ ⟶ Hom D ( Γ ) ( Q , Q [ i ]) ⟶ Hom D ( Γ ) ( Q , p ∗ Q [ i ]) ⟶ Hom D ( Γ ) ( Q , Z [ i + 1 ]) ⟶ ⋯ .
When i ≥ 1 i\geq1 i ≥ 1 , the first term is zero because Q ~ \widetilde{Q} Q is silting, and the last term is zero because Z ∈ D ( Γ ) ≤ − d Z\in D(\Gamma)^{\leq-d} Z ∈ D ( Γ ) ≤ − d and Q ~ ∈ add Γ ∗ add Γ [ 1 ] ∗ ⋯ ∗ add Γ [ d ] \widetilde{Q}\in\operatorname{add}\Gamma*\operatorname{add}\Gamma[1]*\cdots*\operatorname{add}\Gamma[d] Q ∈ add Γ ∗ add Γ [ 1 ] ∗ ⋯ ∗ add Γ [ d ] . Hence we have
Hom D ( B ) ( Q , Q [ i ] ) ≅ Hom D ( Γ ) ( Q ~ , p ∗ Q [ i ] ) = 0 for all i > 0. \operatorname{Hom}_{D(B)}(Q,Q[i])\cong\operatorname{Hom}_{D(\Gamma)}(\widetilde{Q},p_*Q[i])=0\quad\text{for all }i>0. Hom D ( B ) ( Q , Q [ i ]) ≅ Hom D ( Γ ) ( Q , p ∗ Q [ i ]) = 0 for all i > 0.
Finally, since Q ~ \widetilde{Q} Q is silting in per Γ \operatorname{per}\Gamma per Γ , we have Γ ∈ thick ( Q ~ ) \Gamma\in\operatorname{thick}(\widetilde{Q}) Γ ∈ thick ( Q ) . Since p ∗ p^{*} p ∗ is a triangle functor, we have B = p ∗ ( Γ ) ∈ thick ( Q ) B=p^{*}(\Gamma)\in\operatorname{thick}(Q) B = p ∗ ( Γ ) ∈ thick ( Q ) . Hence Q Q Q is silting in per B \operatorname{per}B per B .
Using the left A A A -module structure of Q Q Q , we obtain an alternative realization of the functor real P \operatorname{real}_{P} real P .
Lemma 4.5. There is a natural isomorphism of triangle functors
real P ≅ RHom B ( A Q , − ) [ d ] : D ( B ) → D ( A ) . (30) \operatorname{real}_{P}\cong\operatorname{RHom}_{B}({}_{A}Q,-)[d]\colon D(B)\to D(A). \tag*{(30)} real P ≅ RHom B ( A Q , − ) [ d ] : D ( B ) → D ( A ) . (30) Proof. Let N ∈ D ( B ) N\in D(B) N ∈ D ( B ) . Since RHom A ( Γ P , − ) \operatorname{RHom}_{A}(\Gamma P,-) RHom A ( Γ P , − ) and − ⊗ Γ L P -\mathbin{\mathop{\otimes}\limits^{\mathrm L}_{\Gamma}}P − Γ ⊗ L P are equivalences, the unit (3) at p ∗ N p_{*}N p ∗ N and the counit (4) at A ( A [ d ] ) A(A[d]) A ( A [ d ]) give natural isomorphisms
p ∗ N ≅ RHom A ( Γ P , real P ( N ) ) and Q ~ ⊗ Γ L P ≅ A ( A [ d ] ) . p_{*}N\cong\operatorname{RHom}_{A}(\Gamma P,\operatorname{real}_{P}(N))\quad\text{and}\quad\widetilde{Q}\mathbin{\mathop{\otimes}\limits^{\mathrm L}_{\Gamma}}P\cong A(A[d]). p ∗ N ≅ RHom A ( Γ P , real P ( N )) and Q Γ ⊗ L P ≅ A ( A [ d ]) . Thus we obtain natural isomorphisms in D ( A ) D(A) D ( A ) :
RHom B ( A Q , N ) ≅ RHom Γ ( A Q ~ , p ∗ N ) ≅ RHom Γ ( A Q ~ , RHom A ( Γ P , real P ( N ) ) ) ≅ RHom A ( Q ~ ⊗ Γ L P , real P ( N ) ) ≅ RHom A ( A ( A [ d ] ) , real P ( N ) ) . (31) \begin{aligned}
\operatorname{RHom}_{B}({}_{A}Q,N)&\cong\operatorname{RHom}_{\Gamma}({}_{A}\widetilde{Q},p_{*}N)\\
&\cong\operatorname{RHom}_{\Gamma}({}_{A}\widetilde{Q},\operatorname{RHom}_{A}(\Gamma P,\operatorname{real}_{P}(N)))\\
&\cong\operatorname{RHom}_{A}(\widetilde{Q}\mathbin{\mathop{\otimes}\limits^{\mathrm L}_{\Gamma}}P,\operatorname{real}_{P}(N))\\
&\cong\operatorname{RHom}_{A}(A(A[d]),\operatorname{real}_{P}(N)). \tag*{(31)}
\end{aligned} RHom B ( A Q , N ) ≅ RHom Γ ( A Q , p ∗ N ) ≅ RHom Γ ( A Q , RHom A ( Γ P , real P ( N ))) ≅ RHom A ( Q Γ ⊗ L P , real P ( N )) ≅ RHom A ( A ( A [ d ]) , real P ( N )) . (31) where the first is given by the adjunction p ∗ ⊣ p ∗ p^{*}\dashv p_{*} p ∗ ⊣ p ∗ and the third by the derived tensor-Hom adjunction. Finally, there exists a canonical natural isomorphism from the last term to real P ( N ) [ − d ] \operatorname{real}_{P}(N)[-d] real P ( N ) [ − d ] . This proves the assertion. □
Since the object Q Q Q is a ( d + 1 ) (d+1) ( d + 1 ) -term silting object in per B \operatorname{per}B per B , it induces an s s s -torsion pair ( T ( Q ) , F ( Q ) ) (\mathcal{T}(Q),\mathcal{F}(Q)) ( T ( Q ) , F ( Q )) in d - mod B d\text{-}\operatorname{mod}B d - mod B , where
T ( Q ) = { X ∈ d - mod B ∣ Hom D ( B ) ( Q , X [ i ] ) = 0 for i > 0 } , \mathcal{T}(Q)=\{X\in d\text{-}\operatorname{mod}B\mid\operatorname{Hom}_{D(B)}(Q,X[i])=0\text{ for }i>0\}, T ( Q ) = { X ∈ d - mod B ∣ Hom D ( B ) ( Q , X [ i ]) = 0 for i > 0 } ,
F ( Q ) = { X ∈ d - mod B ∣ Hom D ( B ) ( Q , X [ i ] ) = 0 for i ≤ 0 } . \mathcal{F}(Q)=\{X\in d\text{-}\operatorname{mod}B\mid\operatorname{Hom}_{D(B)}(Q,X[i])=0\text{ for }i\leq0\}. F ( Q ) = { X ∈ d - mod B ∣ Hom D ( B ) ( Q , X [ i ]) = 0 for i ≤ 0 } .
Theorem 4.6. We have X ( P ) = T ( Q ) \mathcal{X}(P)=\mathcal{T}(Q) X ( P ) = T ( Q ) and Y ( P ) = F ( Q ) \mathcal{Y}(P)=\mathcal{F}(Q) Y ( P ) = F ( Q ) . In particular, the s s s -torsion pair ( X ( P ) , Y ( P ) ) (\mathcal{X}(P),\mathcal{Y}(P)) ( X ( P ) , Y ( P )) is functorially finite. Moreover, there exist two pairs of mutually quasi-inverse equivalences:
T ( P ) ⇄ σ B ≤ 0 RHom A ( B ⊗ Γ L P , − ) real P ≅ RHom B ( A Q , − ) [ d ] F ( Q ) , \mathcal{T}(P)\ \underset{\operatorname{real}_{P}\cong\operatorname{RHom}_{B}({}_{A}Q,-)[d]}{\overset{\sigma_{B}^{\leq0}\operatorname{RHom}_{A}(B\mathbin{\mathop{\otimes}\limits^{\mathrm L}_{\Gamma}}P,-)}{\rightleftarrows}}\ \mathcal{F}(Q), T ( P ) real P ≅ RHom B ( A Q , − ) [ d ] ⇄ σ B ≤ 0 RHom A ( B Γ ⊗ L P , − ) F ( Q ) , F ( P ) ⇄ σ B ≤ 0 RHom A ( B ⊗ Γ L P , − [ d ] ) real P [ − d ] ≅ RHom B ( A Q , − ) T ( Q ) . \mathcal{F}(P)\ \underset{\operatorname{real}_{P}[-d]\cong\operatorname{RHom}_{B}({}_{A}Q,-)}{\overset{\sigma_{B}^{\leq0}\operatorname{RHom}_{A}(B\mathbin{\mathop{\otimes}\limits^{\mathrm L}_{\Gamma}}P,-[d])}{\rightleftarrows}}\ \mathcal{T}(Q). F ( P ) real P [ − d ] ≅ RHom B ( A Q , − ) ⇄ σ B ≤ 0 RHom A ( B Γ ⊗ L P , − [ d ]) T ( Q ) . Proof. Let N ∈ d - mod B N\in d\text{-}\operatorname{mod}B N ∈ d - mod B . For any i ∈ Z i\in\mathbb{Z} i ∈ Z , by (31) , we have natural isomorphisms
Hom D ( B ) ( Q , N [ i ] ) ≅ Hom D ( A ) ( A [ d ] , real P ( N ) [ i ] ) , i ∈ Z . \operatorname{Hom}_{D(B)}(Q,N[i])\cong\operatorname{Hom}_{D(A)}(A[d],\operatorname{real}_{P}(N)[i]),\qquad i\in\mathbb{Z}. Hom D ( B ) ( Q , N [ i ]) ≅ Hom D ( A ) ( A [ d ] , real P ( N ) [ i ]) , i ∈ Z . This implies that N ∈ T ( Q ) N\in\mathcal{T}(Q) N ∈ T ( Q ) if and only if real P ( N ) ∈ D ( A ) ≤ − d \operatorname{real}_{P}(N)\in D(A)^{\leq-d} real P ( N ) ∈ D ( A ) ≤ − d . By Theorem 4.2 , the latter is equivalent to real P ( N ) ∈ D ( A ) ≤ − d ∩ D f d ( P ) [ − d + 1 , 0 ] = F ( P ) [ d ] \operatorname{real}_{P}(N)\in D(A)^{\leq-d}\cap D_{\mathrm{fd}}(P)[-d+1,0]=\mathcal{F}(P)[d] real P ( N ) ∈ D ( A ) ≤ − d ∩ D fd ( P ) [ − d + 1 , 0 ] = F ( P ) [ d ] . Therefore, F ( P ) [ d ] = real P ( T ( Q ) ) \mathcal{F}(P)[d]=\operatorname{real}_{P}(\mathcal{T}(Q)) F ( P ) [ d ] = real P ( T ( Q )) . Applying the functor Θ P \Theta_{P} Θ P and using Theorem 4.2 , we obtain X ( P ) = T ( Q ) \mathcal{X}(P)=\mathcal{T}(Q) X ( P ) = T ( Q ) . Similarly, we have Y ( P ) = F ( Q ) \mathcal{Y}(P)=\mathcal{F}(Q) Y ( P ) = F ( Q ) .
By Theorem 4.2 , the functors displayed from T ( P ) \mathcal{T}(P) T ( P ) to F ( Q ) \mathcal{F}(Q) F ( Q ) and from F ( P ) \mathcal{F}(P) F ( P ) to T ( Q ) \mathcal{T}(Q) T ( Q ) are equivalences, whose quasi-inverses are the restrictions of real P \operatorname{real}_{P} real P and real P ( − ) [ − d ] \operatorname{real}_{P}(-)[-d] real P ( − ) [ − d ] , respectively. By Lemma 4.5 , these quasi-inverses are naturally isomorphic to the restrictions of RHom B ( A Q , − ) [ d ] \operatorname{RHom}_{B}({}_{A}Q,-)[d] RHom B ( A Q , − ) [ d ] and RHom B ( A Q , − ) \operatorname{RHom}_{B}({}_{A}Q,-) RHom B ( A Q , − ) , respectively. □
The iterated endomorphism algebra
In the previous section, starting from a proper d d d -truncated non-positive dg algebra A A A and a ( d + 1 ) (d+1) ( d + 1 ) -term silting object P P P in per A \operatorname{per} A per A , we constructed a proper d d d -truncated non-positive dg algebra B = d - End A ( P ) B=d\text{-}\operatorname{End}_A(P) B = d - End A ( P ) and an object Q ∈ D ( A op ⊗ k B ) Q\in D(A^{\operatorname{op}}\otimes_k B) Q ∈ D ( A op ⊗ k B ) , which, regarded as an object in D ( B ) D(B) D ( B ) , is a ( d + 1 ) (d+1) ( d + 1 ) -term silting object in per B \operatorname{per} B per B . Now, we repeat the process with Q Q Q . Fix a K K K -projective resolution of Q Q Q in the category of dg A A A -B B B -bimodules, which is still denoted by Q Q Q . Then Q Q Q is K K K -projective as a right B B B -module (see § 2.1 ). Let
Δ = σ ≤ 0 ( End B ( Q ) ) \Delta=\sigma^{\leq0}(\operatorname{End}_B(Q)) Δ = σ ≤ 0 ( End B ( Q ))
be the non-positive truncation of End B ( Q ) \operatorname{End}_B(Q) End B ( Q ) , and let
C = d - End B ( Q ) = σ ≥ − d + 1 ( Δ ) C=d\text{-}\operatorname{End}_B(Q)=\sigma^{\geq-d+1}(\Delta) C = d - End B ( Q ) = σ ≥ − d + 1 ( Δ )
be the d d d -truncated dg endomorphism algebra of Q Q Q . Write
j Q : Δ ⟶ End B ( Q ) and p Q : Δ ⟶ C j_Q:\Delta\longrightarrow\operatorname{End}_B(Q)\quad\text{and}\quad p_Q:\Delta\longrightarrow C j Q : Δ ⟶ End B ( Q ) and p Q : Δ ⟶ C
for the canonical inclusion and projection, respectively. Then C C C is a proper d d d -truncated non-positive dg algebra and j Q j_Q j Q is a quasi-isomorphism.
The left dg A A A -module structure of Q Q Q gives a dg algebra homomorphism
ϑ ^ : A ⟶ End B ( Q ) . \widehat{\vartheta}:A\longrightarrow\operatorname{End}_B(Q). ϑ : A ⟶ End B ( Q ) .
Since A A A is non-positive, there exists a unique morphism α : A ⟶ Δ \alpha:A\longrightarrow\Delta α : A ⟶ Δ such that j Q ∘ α = ϑ ^ j_Q\circ\alpha=\widehat{\vartheta} j Q ∘ α = ϑ . We define
ϑ : = p Q ∘ α : A ⟶ C . \vartheta:=p_Q\circ\alpha:A\longrightarrow C. ϑ := p Q ∘ α : A ⟶ C .
See the following commutative diagram.
A → ϑ C ϑ ^ ↙ ↘ α End B ( Q ) ← j Q Δ → p Q C \begin{array}{ccccc}
& A & \xrightarrow{\vartheta} & C & \\
{}^{\widehat{\vartheta}}\swarrow& & \searrow^{\alpha} & & \\
\operatorname{End}_B(Q) & \xleftarrow{j_Q} & \Delta& \xrightarrow{p_Q} & C
\end{array} ϑ ↙ End B ( Q ) A j Q ϑ ↘ α Δ C p Q C
We next describe the restriction functor ϑ ∗ : D ( C ) ⟶ D ( A ) \vartheta_*:D(C)\longrightarrow D(A) ϑ ∗ : D ( C ) ⟶ D ( A ) . The inclusion j Q j_Q j Q gives the dg bimodule Δ Q B {}_{\Delta}Q_B Δ Q B . Then we have a triangle functor
real Q : D ( C ) ⟶ D ( B ) Y ⟼ ( p Q ) ∗ ( Y ) ⊗ Δ L Q . \begin{aligned}
\operatorname{real}_Q:D(C)&\longrightarrow D(B)\\
Y&\longmapsto(p_Q)_*(Y)\mathbin{\mathop{\otimes}^{\mathbf L}_{\Delta}}Q.
\end{aligned} real Q : D ( C ) Y ⟶ D ( B ) ⟼ ( p Q ) ∗ ( Y ) ⊗ Δ L Q .
Lemma 5.1. There exists a natural isomorphism of triangle functors
ϑ ∗ ≅ real P ( real Q ( − ) ) [ − d ] : D ( C ) ⟶ D ( A ) . \vartheta_*\cong\operatorname{real}_P(\operatorname{real}_Q(-))[-d]:D(C)\longrightarrow D(A). ϑ ∗ ≅ real P ( real Q ( − )) [ − d ] : D ( C ) ⟶ D ( A ) .
Proof. For any Y ∈ D ( C ) Y\in D(C) Y ∈ D ( C ) , since − ⊗ Δ L Q : D ( Δ ) ⟶ D ( B ) -\mathbin{\mathop{\otimes}^{\mathbf L}_{\Delta}}Q:D(\Delta)\longrightarrow D(B) − ⊗ Δ L Q : D ( Δ ) ⟶ D ( B ) is a triangle equivalence, the unit of the derived tensor-Hom adjunction is a natural isomorphism in D ( Δ ) D(\Delta) D ( Δ ) :
( p Q ) ∗ ( Y ) ≅ RHom B ( Δ Q , ( p Q ) ∗ ( Y ) ⊗ Δ L Q ) = RHom B ( Δ Q , real Q ( Y ) ) . (p_Q)_*(Y)\cong\operatorname{RHom}_B({}_{\Delta}Q,(p_Q)_*(Y)\mathbin{\mathop{\otimes}^{\mathbf L}_{\Delta}}Q)=\operatorname{RHom}_B({}_{\Delta}Q,\operatorname{real}_Q(Y)). ( p Q ) ∗ ( Y ) ≅ RHom B ( Δ Q , ( p Q ) ∗ ( Y ) ⊗ Δ L Q ) = RHom B ( Δ Q , real Q ( Y )) .
Therefore, we obtain the following natural isomorphisms in D ( A ) D(A) D ( A ) :
ϑ ∗ Y = α ∗ ( p Q ) ∗ Y ≅ α ∗ RHom B ( Δ Q , real Q ( Y ) ) ≅ RHom B ( A Q , real Q ( Y ) ) \vartheta_*Y=\alpha_*(p_Q)_*Y\cong\alpha_*\operatorname{RHom}_B({}_{\Delta}Q,\operatorname{real}_Q(Y))\cong\operatorname{RHom}_B({}_A Q,\operatorname{real}_Q(Y)) ϑ ∗ Y = α ∗ ( p Q ) ∗ Y ≅ α ∗ RHom B ( Δ Q , real Q ( Y )) ≅ RHom B ( A Q , real Q ( Y ))
≅ real P ( real Q ( Y ) ) [ − d ] , \cong\operatorname{real}_P(\operatorname{real}_Q(Y))[-d], ≅ real P ( real Q ( Y )) [ − d ] ,
where the second isomorphism follows from j Q ∘ α = ϑ ^ j_Q\circ\alpha=\widehat{\vartheta} j Q ∘ α = ϑ , and the last isomorphism follows from Lemma 4.5 . □
We now characterize when ϑ \vartheta ϑ is a quasi-isomorphism.
Theorem 5.2. The following conditions are equivalent:
ϑ : A → C \vartheta\colon A \to C ϑ : A → C is a quasi-isomorphism.
p : Γ → B p\colon\Gamma\to B p : Γ → B is a quasi-isomorphism.
Hom per A ( P , P [ i ] ) = 0 \operatorname{Hom}_{\operatorname{per} A}(\mathbf{P},\mathbf{P}[i])=0 Hom per A ( P , P [ i ]) = 0 for every i ≤ − d i\le-d i ≤ − d , i.e., Γ \Gamma Γ is d d d -truncated.
ϑ ^ : A → End B ( Q ) \widehat{\vartheta}\colon A \to\operatorname{End}_{B}(\mathbf{Q}) ϑ : A → End B ( Q ) is a quasi-isomorphism.
The restriction functor ϑ ∗ : d - mod C → d - mod A \vartheta_{*}\colon d\text{-}\operatorname{mod} C \to d\text{-}\operatorname{mod} A ϑ ∗ : d - mod C → d - mod A is an equivalence.
Proof. Applying Proposition 2.2 gives (1) ⇔ \Leftrightarrow ⇔ (5). The equivalence (2) ⇔ \Leftrightarrow ⇔ (3) follows from B = σ ≥ − d + 1 ( Γ ) B=\sigma^{\geq-d+1}(\Gamma) B = σ ≥ − d + 1 ( Γ ) and H i ( Γ ) ≅ Hom per A ( P , P [ i ] ) H^{i}(\Gamma)\cong\operatorname{Hom}_{\operatorname{per} A}(\mathbf{P},\mathbf{P}[i]) H i ( Γ ) ≅ Hom per A ( P , P [ i ]) . We prove (2) ⇒ \Rightarrow ⇒ (4) ⇒ \Rightarrow ⇒ (1) ⇒ \Rightarrow ⇒ (2).
(2) ⇒ \Rightarrow ⇒ (4): By (2), the functor p ∗ : D ( Γ ) → D ( B ) p^{*}\colon D(\Gamma)\to D(B) p ∗ : D ( Γ ) → D ( B ) is an equivalence. Let ι : Q → p ∗ RHom A ( Γ P , A [ d ] ) \iota\colon\mathbf{Q}\to p^{*}\operatorname{RHom}_{A}(\Gamma\mathbf{P},A[d]) ι : Q → p ∗ RHom A ( Γ P , A [ d ]) be the canonical isomorphism in D ( A o p ⊗ k B ) D(A^{\mathrm{op}}\otimes_{k}B) D ( A op ⊗ k B ) . For every n ∈ Z n\in\mathbb{Z} n ∈ Z and every closed a ∈ A n a\in A^{n} a ∈ A n , let λ a : A [ d ] → A [ d + n ] \lambda_{a}\colon A[d]\to A[d+n] λ a : A [ d ] → A [ d + n ] be the morphism induced by the left action of a a a on A [ d ] A[d] A [ d ] . Then we have the following commutative diagram in D ( B ) D(B) D ( B ) :
Q → ι p ∗ RHom A ( Γ P , A [ d ] ) ϑ ^ ( a ) ↓ ↓ p ∗ RHom A ( Γ P , λ a ) Q [ n ] → ι [ n ] p ∗ RHom A ( Γ P , A [ d + n ] ) \begin{CD}
\mathbf{Q} @>{\iota}>> p^{*}\operatorname{RHom}_{A}(\Gamma\mathbf{P},A[d]) \\
@V{\widehat{\vartheta}(a)}VV @VV{p^{*}\operatorname{RHom}_{A}(\Gamma\mathbf{P},\lambda_{a})}V \\
\mathbf{Q}[n] @>{\iota[n]}>> p^{*}\operatorname{RHom}_{A}(\Gamma\mathbf{P},A[d+n])
\end{CD} Q ϑ ( a ) ↓ ⏐ Q [ n ] ι ι [ n ] p ∗ RHom A ( Γ P , A [ d ]) ↓ ⏐ p ∗ RHom A ( Γ P , λ a ) p ∗ RHom A ( Γ P , A [ d + n ])
Under the identifications
H n ( A ) ≅ Hom D ( A ) ( A [ d ] , A [ d + n ] ) and H n End B ( Q ) ≅ Hom D ( B ) ( Q , Q [ n ] ) , H^{n}(A)\cong\operatorname{Hom}_{D(A)}(A[d],A[d+n])\quad\text{and}\quad H^{n}\operatorname{End}_{B}(\mathbf{Q})\cong\operatorname{Hom}_{D(B)}(\mathbf{Q},\mathbf{Q}[n]), H n ( A ) ≅ Hom D ( A ) ( A [ d ] , A [ d + n ]) and H n End B ( Q ) ≅ Hom D ( B ) ( Q , Q [ n ]) ,
the map H n ( ϑ ^ ) H^{n}(\widehat{\vartheta}) H n ( ϑ ) is induced by p ∗ RHom A ( Γ P , − ) p^{*}\operatorname{RHom}_{A}(\Gamma\mathbf{P},-) p ∗ RHom A ( Γ P , − ) . Since this functor is an equivalence, the map H n ( ϑ ^ ) H^{n}(\widehat{\vartheta}) H n ( ϑ ) is an isomorphism for every n ∈ Z n\in\mathbb{Z} n ∈ Z . Thus ϑ ^ \widehat{\vartheta} ϑ is a quasi-isomorphism.
(4) ⇒ \Rightarrow ⇒ (1): Since j Q j_{\mathbf{Q}} j Q is a quasi-isomorphism and j Q ∘ α = ϑ ^ j_{\mathbf{Q}}\circ\alpha=\widehat{\vartheta} j Q ∘ α = ϑ , condition (4) implies that α \alpha α is a quasi-isomorphism. It follows that Δ \Delta Δ is d d d -truncated. Thus p Q p_{\mathbf{Q}} p Q is a quasi-isomorphism, and so is ϑ \vartheta ϑ .
(1) ⇒ \Rightarrow ⇒ (2): By (1), ϑ ∗ : D ( C ) → D ( A ) \vartheta_{*}\colon D(C)\to D(A) ϑ ∗ : D ( C ) → D ( A ) is an equivalence. Therefore, by Lemma 5.1 , real P : D ( B ) → D ( A ) \operatorname{real}_{\mathbf{P}}\colon D(B)\to D(A) real P : D ( B ) → D ( A ) is dense. By (25) , since − ⊗ Γ L P -\otimes_{\Gamma}^{\mathbb{L}}\mathbf{P} − ⊗ Γ L P is an equivalence, the functor p ∗ : D ( B ) → D ( Γ ) p^{*}\colon D(B)\to D(\Gamma) p ∗ : D ( B ) → D ( Γ ) is also dense. Hence, there exist X ∈ D ( B ) X\in D(B) X ∈ D ( B ) and an isomorphism u : Γ → ∼ p ∗ X u\colon\Gamma\xrightarrow{\sim}p^{*}X u : Γ ∼ p ∗ X in D ( Γ ) D(\Gamma) D ( Γ ) . By (8) , there exists an isomorphism:
Hom D ( B ) ( p ∗ Γ , X ) → ∼ Hom D ( Γ ) ( Γ , p ∗ X ) \operatorname{Hom}_{D(B)}(p^{*}\Gamma,X)\xrightarrow{\sim}\operatorname{Hom}_{D(\Gamma)}(\Gamma,p_{*}X) Hom D ( B ) ( p ∗ Γ , X ) ∼ Hom D ( Γ ) ( Γ , p ∗ X )
h ⟼ p ∗ ( h ) ∘ η Γ h\longmapsto p_{*}(h)\circ\eta_{\Gamma} h ⟼ p ∗ ( h ) ∘ η Γ
where η \eta η is the unit of the adjunction p ∗ ⊣ p ∗ p^{*}\dashv p_{*} p ∗ ⊣ p ∗ . It follows that the morphism
− ∘ η Γ : Hom D ( Γ ) ( p ∗ p ∗ Γ , p ∗ X ) → Hom D ( Γ ) ( Γ , p ∗ X ) -\circ\eta_{\Gamma}\colon\operatorname{Hom}_{D(\Gamma)}(p^{*}p^{*}\Gamma,p^{*}X)\to\operatorname{Hom}_{D(\Gamma)}(\Gamma,p^{*}X) − ∘ η Γ : Hom D ( Γ ) ( p ∗ p ∗ Γ , p ∗ X ) → Hom D ( Γ ) ( Γ , p ∗ X )
is surjective. Thus, the isomorphism u u u factors through p p p , which implies that η Γ : Γ → p ∗ p ∗ Γ ≅ p ∗ B \eta_{\Gamma}\colon\Gamma\to p_{*}p^{*}\Gamma\cong p_{*}B η Γ : Γ → p ∗ p ∗ Γ ≅ p ∗ B is a section. Thus Γ \Gamma Γ is a direct summand of p ∗ B p_{*}B p ∗ B . Therefore, due to p ∗ B ∈ D ( Γ ) ≥ − d + 1 p_{*}B\in D(\Gamma)^{\geq-d+1} p ∗ B ∈ D ( Γ ) ≥ − d + 1 , we have Γ ∈ D ( Γ ) ≥ − d + 1 \Gamma\in D(\Gamma)^{\geq-d+1} Γ ∈ D ( Γ ) ≥ − d + 1 . Hence p : Γ → B p\colon\Gamma\to B p : Γ → B is a quasi-isomorphism. □ \square □
Applying Corollary 4.3 to Q \mathbf{Q} Q , we obtain an s s s -torsion pair ( X ( Q ) , Y ( Q ) ) (\mathcal{X}(\mathbf{Q}),\mathcal{Y}(\mathbf{Q})) ( X ( Q ) , Y ( Q )) in d - mod C d\text{-}\operatorname{mod} C d - mod C , where
X ( Q ) = Θ Q ( F ( Q ) [ d ] ) and Y ( Q ) = Θ Q ( T ( Q ) ) . \mathcal{X}(\mathbf{Q})=\Theta_{\mathbf{Q}}(\mathcal{F}(\mathbf{Q})[d])\quad\text{and}\quad\mathcal{Y}(\mathbf{Q})=\Theta_{\mathbf{Q}}(\mathcal{T}(\mathbf{Q})). X ( Q ) = Θ Q ( F ( Q ) [ d ]) and Y ( Q ) = Θ Q ( T ( Q )) .
Theorem 5.3. The functor ϑ ∗ : d - mod C → d - mod A \vartheta_{*}\colon d\text{-}\operatorname{mod} C\to d\text{-}\operatorname{mod} A ϑ ∗ : d - mod C → d - mod A induces equivalences
X ( Q ) → ∼ T ( P ) and Y ( Q ) → ∼ F ( P ) . \mathcal{X}(\mathbf{Q})\xrightarrow{\sim}\mathcal{T}(\mathbf{P})\quad\text{and}\quad\mathcal{Y}(\mathbf{Q})\xrightarrow{\sim}\mathcal{F}(\mathbf{P}). X ( Q ) ∼ T ( P ) and Y ( Q ) ∼ F ( P ) .
Proof. Applying Theorem 4.6 to P \mathbf{P} P and Q \mathbf{Q} Q , we obtain equivalences
X ( Q ) → real Q ( − ) [ − d ] F ( Q ) = Y ( P ) → real P T ( P ) , \mathcal{X}(\mathbf{Q}) \xrightarrow{\operatorname{real}_{\mathbf{Q}}(-)[-d]} \mathcal{F}(\mathbf{Q}) = \mathcal{Y}(\mathbf{P}) \xrightarrow{\operatorname{real}_{\mathbf{P}}} \mathcal{T}(\mathbf{P}), X ( Q ) real Q ( − ) [ − d ] F ( Q ) = Y ( P ) real P T ( P ) ,
Y ( Q ) → real Q T ( Q ) = X ( P ) → real P ( − ) [ − d ] F ( P ) . \mathcal{Y}(\mathbf{Q}) \xrightarrow{\operatorname{real}_{\mathbf{Q}}} \mathcal{T}(\mathbf{Q}) = \mathcal{X}(\mathbf{P}) \xrightarrow{\operatorname{real}_{\mathbf{P}}(-)[-d]} \mathcal{F}(\mathbf{P}). Y ( Q ) real Q T ( Q ) = X ( P ) real P ( − ) [ − d ] F ( P ) .
Both compositions are naturally isomorphic to real P ( real Q ( − ) ) [ − d ] \operatorname{real}_{\mathbf{P}}\left(\operatorname{real}_{\mathbf{Q}}(-)\right)[-d] real P ( real Q ( − ) ) [ − d ] and hence to ϑ ∗ \vartheta_* ϑ ∗ by Lemma 5.1 . □
This result recovers [10 Theorem 4.4] for the case d = 1 d=1 d = 1 . Note that, in that case, ϑ ∗ : d -mod C → d -mod A \vartheta_*:d\text{-mod }C\to d\text{-mod }A ϑ ∗ : d -mod C → d -mod A is always fully faithful. However, this is not true for general d d d . See the following example.
Example 5.4. Let d = 2 d=2 d = 2 , and A = k Q A=kQ A = k Q , where Q Q Q is the quiver
1 → α 2 → β 3 1 \xrightarrow{\alpha} 2 \xrightarrow{\beta} 3 1 α 2 β 3
with ∣ α ∣ = ∣ β ∣ = 0 |\alpha|=|\beta|=0 ∣ α ∣ = ∣ β ∣ = 0 . Let P i A P_i^A P i A (resp. I i A I_i^A I i A and S i A S_i^A S i A ) denote the indecomposable projective (resp. indecomposable injective and simple) right A A A -module corresponding to vertex i i i . We use analogous notation for algebras B B B and C C C below. Let
P = P 3 A ⊕ P 2 A [ 1 ] ⊕ P 1 A [ 2 ] . \mathbf{P}=P_3^A\oplus P_2^A[1]\oplus P_1^A[2]. P = P 3 A ⊕ P 2 A [ 1 ] ⊕ P 1 A [ 2 ] .
It is straightforward to check that P \mathbf{P} P is a 3-term silting object in per A \operatorname{per} A per A , and the induced s s s -torsion pair ( T ( P ) , F ( P ) ) (\mathcal{T}(\mathbf{P}),\mathcal{F}(\mathbf{P})) ( T ( P ) , F ( P )) in 2 -mod A 2\text{-mod }A 2 -mod A is given by
T ( P ) = add ( P 3 A ⊕ P 3 A [ 1 ] ⊕ P 2 A [ 1 ] ⊕ S 2 A [ 1 ] ) , \mathcal{T}(\mathbf{P})=\operatorname{add}\left(P_3^A\oplus P_3^A[1]\oplus P_2^A[1]\oplus S_2^A[1]\right), T ( P ) = add ( P 3 A ⊕ P 3 A [ 1 ] ⊕ P 2 A [ 1 ] ⊕ S 2 A [ 1 ] ) ,
F ( P ) = add ( S 2 A ⊕ I 2 A ⊕ I 1 A ⊕ I 1 A [ 1 ] ) . \mathcal{F}(\mathbf{P})=\operatorname{add}\left(S_2^A\oplus I_2^A\oplus I_1^A\oplus I_1^A[1]\right). F ( P ) = add ( S 2 A ⊕ I 2 A ⊕ I 1 A ⊕ I 1 A [ 1 ] ) .
See Figure 1 , where red and blue shaded regions indicate the torsion and torsion-free classes of the displayed s s s -torsion pair, respectively.
Figure 1. ( T ( P ) , F ( P ) ) (\mathcal{T}(\mathbf{P}),\mathcal{F}(\mathbf{P})) ( T ( P ) , F ( P )) in 2 2 2 -mod A A A .
Label the primitive idempotents of Γ = End A ( P ) \Gamma=\operatorname{End}_A(\mathbf{P}) Γ = End A ( P ) so that the vertices 1 , 2 , 3 1,2,3 1 , 2 , 3 correspond to P 1 A [ 2 ] P_1^A[2] P 1 A [ 2 ] , P 2 A [ 1 ] P_2^A[1] P 2 A [ 1 ] , P 3 A P_3^A P 3 A , respectively. Then Γ \Gamma Γ has zero differential and
Γ ≅ k ( 1 → a 2 → b 3 ) \Gamma\cong k\left(1\xrightarrow{a}2\xrightarrow{b}3\right) Γ ≅ k ( 1 a 2 b 3 )
as a graded dg algebra, where ∣ a ∣ = ∣ b ∣ = − 1 |a|=|b|=-1 ∣ a ∣ = ∣ b ∣ = − 1 . The path a b ab ab has degree − 2 -2 − 2 , so
B = σ ≥ − 1 ( Γ ) ≅ k ( 1 → a 2 → b 3 ) / ( a b ) , B=\sigma^{\geq-1}(\Gamma)\cong k\left(1\xrightarrow{a}2\xrightarrow{b}3\right)/(ab), B = σ ≥ − 1 ( Γ ) ≅ k ( 1 a 2 b 3 ) / ( ab ) ,
again with zero differential. The induced 3-term silting object in per B \operatorname{per} B per B is
Q ≅ P 1 B ⊕ P 2 B [ 1 ] ⊕ P 3 B [ 2 ] , \mathbf{Q}\cong P_1^B\oplus P_2^B[1]\oplus P_3^B[2], Q ≅ P 1 B ⊕ P 2 B [ 1 ] ⊕ P 3 B [ 2 ] ,
and the induced s s s -torsion pair ( T ( Q ) , F ( Q ) ) = ( X ( P ) , Y ( P ) ) (\mathcal{T}(\mathbf{Q}),\mathcal{F}(\mathbf{Q}))=(\mathcal{X}(\mathbf{P}),\mathcal{Y}(\mathbf{P})) ( T ( Q ) , F ( Q )) = ( X ( P ) , Y ( P )) is given by
T ( Q ) = add ( S 2 B [ 1 ] ⊕ S 1 B ⊕ S 1 B [ 1 ] ⊕ P 1 B ) , \mathcal{T}(\mathbf{Q})=\operatorname{add}\left(S_{2}^{B}[1]\oplus S_{1}^{B}\oplus S_{1}^{B}[1]\oplus P_{1}^{B}\right), T ( Q ) = add ( S 2 B [ 1 ] ⊕ S 1 B ⊕ S 1 B [ 1 ] ⊕ P 1 B ) ,
F ( Q ) = add ( P 3 B ⊕ P 3 B [ 1 ] ⊕ S 2 B ⊕ P 2 B ) . \mathcal{F}(\mathbf{Q})=\operatorname{add}\left(P_{3}^{B}\oplus P_{3}^{B}[1]\oplus S_{2}^{B}\oplus P_{2}^{B}\right). F ( Q ) = add ( P 3 B ⊕ P 3 B [ 1 ] ⊕ S 2 B ⊕ P 2 B ) .
See Figure 2 .
Figure 2. ( T ( Q ) , F ( Q ) ) (\mathcal{T}(\mathbf{Q}),\mathcal{F}(\mathbf{Q})) ( T ( Q ) , F ( Q )) in 2-mod B B B .
The d d d -truncated dg endomorphism algebra C C C of Q \mathbf{Q} Q is quasi-isomorphic to
k ( 1 → α 2 → β 3 ) / ( α β ) , \mathbf{k}(1\xrightarrow{\alpha}2\xrightarrow{\beta}3)/(\alpha\beta), k ( 1 α 2 β 3 ) / ( α β ) ,
where ∣ α ∣ = ∣ β ∣ = 0 |\alpha|=|\beta|=0 ∣ α ∣ = ∣ β ∣ = 0 . Under this quasi-isomorphism, the morphism ϑ : A → C \vartheta:A\to C ϑ : A → C is the canonical quotient. Moreover, the induced s s s -torsion pair ( X ( Q ) , Y ( Q ) ) (\mathcal{X}(\mathbf{Q}),\mathcal{Y}(\mathbf{Q})) ( X ( Q ) , Y ( Q )) is given by
X ( Q ) = add ( S 3 C ⊕ S 3 C [ 1 ] ⊕ P 2 C [ 1 ] ⊕ S 2 C [ 1 ] ) , \mathcal{X}(\mathbf{Q})=\operatorname{add}\left(S_{3}^{C}\oplus S_{3}^{C}[1]\oplus P_{2}^{C}[1]\oplus S_{2}^{C}[1]\right), X ( Q ) = add ( S 3 C ⊕ S 3 C [ 1 ] ⊕ P 2 C [ 1 ] ⊕ S 2 C [ 1 ] ) ,
Y ( Q ) = add ( S 1 C ⊕ S 1 C [ 1 ] ⊕ S 2 C ⊕ P 1 C ) . \mathcal{Y}(\mathbf{Q})=\operatorname{add}\left(S_{1}^{C}\oplus S_{1}^{C}[1]\oplus S_{2}^{C}\oplus P_{1}^{C}\right). Y ( Q ) = add ( S 1 C ⊕ S 1 C [ 1 ] ⊕ S 2 C ⊕ P 1 C ) .
See Figure 3 , where X = ( P 2 C → α P 1 C ) X=(P_{2}^{C}\xrightarrow{\alpha}P_{1}^{C}) X = ( P 2 C α P 1 C ) is concentrated in degrees − 1 , 0 -1,0 − 1 , 0 .
Figure 3. ( X ( Q ) , Y ( Q ) ) (\mathcal{X}(\mathbf{Q}),\mathcal{Y}(\mathbf{Q})) ( X ( Q ) , Y ( Q )) in 2-mod C C C .
Although the functor ϑ ∗ \vartheta_{*} ϑ ∗ restricts to equivalences between X ( Q ) \mathcal{X}(\mathbf{Q}) X ( Q ) and T ( P ) \mathcal{T}(\mathbf{P}) T ( P ) and between Y ( Q ) \mathcal{Y}(\mathbf{Q}) Y ( Q ) and F ( P ) \mathcal{F}(\mathbf{P}) F ( P ) by Theorem 5.3 , it does not restrict to a fully faithful functor from 2-mod C C C to 2-mod A A A in this example. This is because the image of X X X under ϑ ∗ \vartheta_{*} ϑ ∗ is decomposable: ϑ ∗ X ≅ P 3 A [ 1 ] ⊕ I 1 A \vartheta_{*}X\cong P_{3}^{A}[1]\oplus I_{1}^{A} ϑ ∗ X ≅ P 3 A [ 1 ] ⊕ I 1 A . To see this, consider the canonical triangle of X X X with respect to the s s s -torsion pair ( X ( Q ) , Y ( Q ) ) (\mathcal{X}(\mathbf{Q}),\mathcal{Y}(\mathbf{Q})) ( X ( Q ) , Y ( Q )) :
P 3 C [ 1 ] ⟶ X ⟶ S 1 C . P_{3}^{C}[1]\longrightarrow X\longrightarrow S_{1}^{C}. P 3 C [ 1 ] ⟶ X ⟶ S 1 C .
Its image under ϑ ∗ \vartheta_{*} ϑ ∗ is P 3 A [ 1 ] ⟶ ϑ ∗ X ⟶ I 1 A = S 1 A P_{3}^{A}[1]\longrightarrow\vartheta_{*}X\longrightarrow I_{1}^{A}=S_{1}^{A} P 3 A [ 1 ] ⟶ ϑ ∗ X ⟶ I 1 A = S 1 A , which is split.
Auslander-Reiten theory
In this section, we apply the higher Brenner-Butler theorem to study the AR theory of d d d -mod B B B , where B = d - End A ( P ) B=d\text{-}\operatorname{End}_{A}(\mathbf{P}) B = d - End A ( P ) .
For an indecomposable P ∈ add A P\in\operatorname{add} A P ∈ add A , we write I P = ν A ( P ) [ d − 1 ] ∈ inj d I_P=\nu_A(P)[d-1]\in\operatorname{inj}d I P = ν A ( P ) [ d − 1 ] ∈ inj d -mod A A A . Recall that t P \mathbf{t}_P t P and f P \mathbf{f}_P f P denote torsion and torsion-free functors respectively, with respect to the s s s -torsion pair ( T ( P ) , F ( P ) ) (\mathcal{T}(P),\mathcal{F}(P)) ( T ( P ) , F ( P )) .
Lemma 6.1. For any indecomposable P ∈ add A P\in\operatorname{add} A P ∈ add A , at least one of Θ P ( ( f P P ) [ d ] ) \Theta_P((\mathbf{f}_P P)[d]) Θ P (( f P P ) [ d ]) and Θ P ( t P I P ) \Theta_P(\mathbf{t}_P I_P) Θ P ( t P I P ) is not zero. Moreover, any non-zero object among them is indecomposable.
Proof. By Theorem 4.2 , Θ P \Theta_P Θ P is an equivalence. Therefore, this lemma follows from Proposition 3.3 (2) and (3). □
For an indecomposable P ∈ add A P\in\operatorname{add} A P ∈ add A , we define Q ~ P = RHom A ( Γ P , P [ d ] ) \widetilde{Q}_P=\operatorname{RHom}_{A}(\Gamma_P,P[d]) Q P = RHom A ( Γ P , P [ d ]) and Q P = p ∗ ( Q ~ P ) Q_P=p^*(\widetilde{Q}_P) Q P = p ∗ ( Q P ) . They are direct summands of Q ~ \widetilde{Q} Q and Q Q Q , respectively.
Lemma 6.2. For any indecomposable P ∈ add A P\in\operatorname{add} A P ∈ add A , Q P Q_P Q P is indecomposable, and
σ B ‾ ≥ − d + 1 ( Q P ) ≅ Θ P ( ( f P P ) [ d ] ) and σ B ‾ ≤ 0 ( ν B ( Q P ) [ − 1 ] ) ≅ Θ P ( t P I P ) . (32) \sigma_{\overline{B}}^{\geq-d+1}(Q_P)\cong\Theta_P((\mathbf{f}_P P)[d])\quad\text{and}\quad\sigma_{\overline{B}}^{\leq0}(\nu_B(Q_P)[-1])\cong\Theta_P(\mathbf{t}_P I_P). \tag*{(32)} σ B ≥ − d + 1 ( Q P ) ≅ Θ P (( f P P ) [ d ]) and σ B ≤ 0 ( ν B ( Q P ) [ − 1 ]) ≅ Θ P ( t P I P ) . (32)
Proof. For N ∈ d N\in d N ∈ d -mod B B B , taking cohomology in (31) and then the direct summands corresponding to P P P gives natural isomorphisms
Hom D ( B ) ( Q P , N [ i ] ) ≅ Hom D ( A ) ( P [ d ] , real P ( N ) [ i ] ) , i ∈ Z . (33) \operatorname{Hom}_{D(B)}(Q_P,N[i])\cong\operatorname{Hom}_{D(A)}(P[d],\operatorname{real}_P(N)[i]),\qquad i\in\mathbb{Z}. \tag*{(33)} Hom D ( B ) ( Q P , N [ i ]) ≅ Hom D ( A ) ( P [ d ] , real P ( N ) [ i ]) , i ∈ Z . (33)
Applying Hom D ( A ) ( − , real P ( N ) ) \operatorname{Hom}_{D(A)}(-,\operatorname{real}_P(N)) Hom D ( A ) ( − , real P ( N )) to the canonical triangle t P P → P → f P P → t P P [ 1 ] \mathbf{t}_P P\to P\to\mathbf{f}_P P\to\mathbf{t}_P P[1] t P P → P → f P P → t P P [ 1 ] of P P P with respect to ( T ( P ) , F ( P ) ) (\mathcal{T}(P),\mathcal{F}(P)) ( T ( P ) , F ( P )) yields an exact sequence
0 = Hom D ( A ) ( ( t P P ) [ d + 1 ] , real P ( N ) ) → Hom D ( A ) ( ( f P P ) [ d ] , real P ( N ) ) → Hom D ( A ) ( P [ d ] , real P ( N ) ) → Hom D ( A ) ( ( t P P ) [ d ] , real P ( N ) ) = 0 , 0=\operatorname{Hom}_{D(A)}((\mathbf{t}_P P)[d+1],\operatorname{real}_P(N))\to\operatorname{Hom}_{D(A)}((\mathbf{f}_P P)[d],\operatorname{real}_P(N))
\to\operatorname{Hom}_{D(A)}(P[d],\operatorname{real}_P(N))\to\operatorname{Hom}_{D(A)}((\mathbf{t}_P P)[d],\operatorname{real}_P(N))=0, 0 = Hom D ( A ) (( t P P ) [ d + 1 ] , real P ( N )) → Hom D ( A ) (( f P P ) [ d ] , real P ( N )) → Hom D ( A ) ( P [ d ] , real P ( N )) → Hom D ( A ) (( t P P ) [ d ] , real P ( N )) = 0 ,
where the first and last terms vanish since t P P ∈ T ( P ) ⊆ D f d ( P ) ≤ 0 \mathbf{t}_P P\in\mathcal{T}(P)\subseteq D_{\mathrm{fd}}(P)^{\leq0} t P P ∈ T ( P ) ⊆ D fd ( P ) ≤ 0 and real P ( N ) ∈ D f d ( P ) [ − d + 1 , 0 ] ⊆ D f d ( P ) ≥ − d + 1 \operatorname{real}_P(N)\in D_{\mathrm{fd}}(P)^{[-d+1,0]}\subseteq D_{\mathrm{fd}}(P)^{\geq-d+1} real P ( N ) ∈ D fd ( P ) [ − d + 1 , 0 ] ⊆ D fd ( P ) ≥ − d + 1 . Consequently, we obtain natural isomorphisms
Hom D ( B ) ( σ B ‾ ≥ − d + 1 ( Q P ) , N ) ≅ Hom D ( B ) ( Q P , N ) ≅ Hom D ( A ) ( P [ d ] , real P ( N ) ) \operatorname{Hom}_{D(B)}(\sigma_{\overline{B}}^{\geq-d+1}(Q_P),N)\cong\operatorname{Hom}_{D(B)}(Q_P,N)
\cong\operatorname{Hom}_{D(A)}(P[d],\operatorname{real}_P(N)) Hom D ( B ) ( σ B ≥ − d + 1 ( Q P ) , N ) ≅ Hom D ( B ) ( Q P , N ) ≅ Hom D ( A ) ( P [ d ] , real P ( N ))
≅ Hom D ( A ) ( ( f P P ) [ d ] , real P ( N ) ) ≅ Hom D ( B ) ( Θ P ( ( f P P ) [ d ] ) , N ) , \cong\operatorname{Hom}_{D(A)}((\mathbf{f}_P P)[d],\operatorname{real}_P(N))
\cong\operatorname{Hom}_{D(B)}(\Theta_P((\mathbf{f}_P P)[d]),N), ≅ Hom D ( A ) (( f P P ) [ d ] , real P ( N )) ≅ Hom D ( B ) ( Θ P (( f P P ) [ d ]) , N ) ,
where the first isomorphism is given by canonical truncation of Q P ∈ D ( B ) ≤ 0 Q_P\in D(B)^{\leq0} Q P ∈ D ( B ) ≤ 0 . Then the Yoneda Lemma shows the first isomorphism in (32) .
Next, applying Hom D ( A ) ( real P ( N ) , − ) \operatorname{Hom}_{D(A)}(\operatorname{real}_P(N),-) Hom D ( A ) ( real P ( N ) , − ) to the canonical triangle t P I P → I P → f P I P → t P I P [ 1 ] \mathbf{t}_P I_P\to I_P\to\mathbf{f}_P I_P\to\mathbf{t}_P I_P[1] t P I P → I P → f P I P → t P I P [ 1 ] of I P I_P I P with respect to ( T ( P ) , F ( P ) ) (\mathcal{T}(P),\mathcal{F}(P)) ( T ( P ) , F ( P )) yields an exact sequence
0 = Hom D ( A ) ( real P ( N ) , f P I P [ − 1 ] ) → Hom D ( A ) ( real P ( N ) , t P I P ) → Hom D ( A ) ( real P ( N ) , I P ) → Hom D ( A ) ( real P ( N ) , f P I P ) = 0 , 0=\operatorname{Hom}_{D(A)}(\operatorname{real}_P(N),\mathbf{f}_P I_P[-1])\to\operatorname{Hom}_{D(A)}(\operatorname{real}_P(N),\mathbf{t}_P I_P)
\to\operatorname{Hom}_{D(A)}(\operatorname{real}_P(N),I_P)\to\operatorname{Hom}_{D(A)}(\operatorname{real}_P(N),\mathbf{f}_P I_P)=0, 0 = Hom D ( A ) ( real P ( N ) , f P I P [ − 1 ]) → Hom D ( A ) ( real P ( N ) , t P I P ) → Hom D ( A ) ( real P ( N ) , I P ) → Hom D ( A ) ( real P ( N ) , f P I P ) = 0 ,
where the first and last terms vanish since f P I P ∈ F ( P ) ⊆ D f d ( P ) ≥ 1 \mathbf{f}_P I_P\in\mathcal{F}(P)\subseteq D_{\mathrm{fd}}(P)^{\geq1} f P I P ∈ F ( P ) ⊆ D fd ( P ) ≥ 1 and real P ( N ) ∈ D f d ( P ) [ − d + 1 , 0 ] ⊆ D f d ( P ) ≤ 0 \operatorname{real}_P(N)\in D_{\mathrm{fd}}(P)^{[-d+1,0]}\subseteq D_{\mathrm{fd}}(P)^{\leq0} real P ( N ) ∈ D fd ( P ) [ − d + 1 , 0 ] ⊆ D fd ( P ) ≤ 0 . Consequently, we obtain natural isomorphisms
Hom D ( B ) ( N , σ B ‾ ≤ 0 ( ν B ( Q P ) [ − 1 ] ) ) ≅ Hom D ( B ) ( N , ν B ( Q P ) [ − 1 ] ) ≅ D Hom D ( B ) ( Q P , N [ 1 ] ) \operatorname{Hom}_{D(B)}(N,\sigma_{\overline{B}}^{\leq0}(\nu_B(Q_P)[-1]))\cong\operatorname{Hom}_{D(B)}(N,\nu_B(Q_P)[-1])
\cong\mathbb{D}\operatorname{Hom}_{D(B)}(Q_P,N[1]) Hom D ( B ) ( N , σ B ≤ 0 ( ν B ( Q P ) [ − 1 ])) ≅ Hom D ( B ) ( N , ν B ( Q P ) [ − 1 ]) ≅ D Hom D ( B ) ( Q P , N [ 1 ])
≅ D Hom D ( A ) ( P [ d ] , real P ( N ) [ 1 ] ) ≅ Hom D ( A ) ( real P ( N ) [ 1 ] , ν A ( P [ d ] ) ) \cong\mathbb{D}\operatorname{Hom}_{D(A)}(P[d],\operatorname{real}_P(N)[1])
\cong\operatorname{Hom}_{D(A)}(\operatorname{real}_P(N)[1],\nu_A(P[d])) ≅ D Hom D ( A ) ( P [ d ] , real P ( N ) [ 1 ]) ≅ Hom D ( A ) ( real P ( N ) [ 1 ] , ν A ( P [ d ]))
≅ Hom D ( A ) ( real P ( N ) , t P I P ) ≅ Hom D ( B ) ( N , Θ P ( t P I P ) ) , \cong\operatorname{Hom}_{D(A)}(\operatorname{real}_P(N),\mathbf{t}_P I_P)
\cong\operatorname{Hom}_{D(B)}(N,\Theta_P(\mathbf{t}_P I_P)), ≅ Hom D ( A ) ( real P ( N ) , t P I P ) ≅ Hom D ( B ) ( N , Θ P ( t P I P )) ,
where the first isomorphism is given by canonical truncation of ν B ( Q P ) [ − 1 ] ∈ D ( B ) ≥ − d + 1 \nu_B(Q_P)[-1]\in D(B)^{\geq-d+1} ν B ( Q P ) [ − 1 ] ∈ D ( B ) ≥ − d + 1 . Then the Yoneda Lemma shows the second isomorphism in (32) .
By Lemma 6.1, the isomorphisms in (32) imply that at least one of σ B ‾ ≥ − d + 1 ( Q P ) \sigma_{\overline{B}}^{\geq-d+1}(\mathbf{Q}_P) σ B ≥ − d + 1 ( Q P ) and σ B ‾ < 0 ( ν B ( Q P ) [ − 1 ] ) \sigma_{\overline{B}}^{<0}(\nu_B(\mathbf{Q}_P)[-1]) σ B < 0 ( ν B ( Q P ) [ − 1 ]) is not zero. It follows that Q P ≠ 0 \mathbf{Q}_P \ne0 Q P = 0 . Taking the direct summand corresponding to P P P in (29) , we obtain a triangle
Z P → Q ~ P → p ∗ Q P → Z P [ 1 ] , Z_P \to\widetilde{Q}_P \to p_*\mathbf{Q}_P \to Z_P[1], Z P → Q P → p ∗ Q P → Z P [ 1 ] ,
with Z P ∈ D ( Γ ) ≤ − d Z_P \in D(\Gamma)^{\leq-d} Z P ∈ D ( Γ ) ≤ − d . By Q ~ P ∈ add Γ ∗ ⋯ ∗ add Γ [ d ] \widetilde{Q}_P \in\operatorname{add}\Gamma* \cdots* \operatorname{add}\Gamma[d] Q P ∈ add Γ ∗ ⋯ ∗ add Γ [ d ] , we have Hom D ( Γ ) ( Q ~ P , Z P [ 1 ] ) = 0 \operatorname{Hom}_{D(\Gamma)}(\widetilde{Q}_P,Z_P[1])=0 Hom D ( Γ ) ( Q P , Z P [ 1 ]) = 0 . Applying Hom D ( Γ ) ( Q ~ P , − ) \operatorname{Hom}_{D(\Gamma)}(\widetilde{Q}_P,-) Hom D ( Γ ) ( Q P , − ) to the above triangle and using the adjunction p ∗ ⊣ p ∗ p^* \dashv p_* p ∗ ⊣ p ∗ gives a surjective algebra homomorphism
End D ( A ) ( P ) ≅ End D ( Γ ) ( Q ~ P ) ↠ End D ( B ) ( Q P ) . \operatorname{End}_{D(A)}(P) \cong\operatorname{End}_{D(\Gamma)}(\widetilde{Q}_P) \twoheadrightarrow\operatorname{End}_{D(B)}(\mathbf{Q}_P). End D ( A ) ( P ) ≅ End D ( Γ ) ( Q P ) ↠ End D ( B ) ( Q P ) .
Since End D ( A ) ( P ) \operatorname{End}_{D(A)}(P) End D ( A ) ( P ) is local and Q P ≠ 0 \mathbf{Q}_P \ne0 Q P = 0 , the object Q P \mathbf{Q}_P Q P is indecomposable. □
An almost split triangle in d d d -mod B B B is called connecting if its left term belongs to Y ( P ) \mathcal{Y}(P) Y ( P ) and its right term belongs to X ( P ) \mathcal{X}(P) X ( P ) .
Lemma 6.3. If Y → E → X → Y [ 1 ] Y \to E \to X \to Y[1] Y → E → X → Y [ 1 ] is a connecting triangle, then there exists an indecomposable P ∈ add A P \in\operatorname{add} A P ∈ add A such that Y ≅ Θ P ( t P I P ) Y \cong\Theta_P(t_P I_P) Y ≅ Θ P ( t P I P ) .
Proof. By Lemma 3.4, Y ∈ inj Y ( P ) Y \in\operatorname{inj}\mathcal{Y}(P) Y ∈ inj Y ( P ) . By Theorem 4.2 and Corollary 4.3, there exists an indecomposable Z ∈ inj T ( P ) Z \in\operatorname{inj}\mathcal{T}(P) Z ∈ inj T ( P ) such that Y ≅ Θ P ( Z ) Y \cong\Theta_P(Z) Y ≅ Θ P ( Z ) . By Proposition 3.3, inj T ( P ) = add t P ( D A [ d − 1 ] ) \operatorname{inj}\mathcal{T}(P)=\operatorname{add}t_P(DA[d-1]) inj T ( P ) = add t P ( D A [ d − 1 ]) . Thus, the assertion follows. □
We now compute the Auslander-Reiten translations of the end terms of connecting triangles.
Lemma 6.4. If Θ P ( ( f P P ) [ d ] ) ≠ 0 \Theta_P((\mathfrak{f}_P P)[d]) \ne0 Θ P (( f P P ) [ d ]) = 0 , then
τ [ d ] B Θ P ( ( f P P ) [ d ] ) ≅ Θ P ( t P I P ) . \tau_{[d]}^B\Theta_P((\mathfrak{f}_P P)[d]) \cong\Theta_P(t_P I_P). τ [ d ] B Θ P (( f P P ) [ d ]) ≅ Θ P ( t P I P ) . If Θ P ( t P I P ) ≠ 0 \Theta_P(t_P I_P) \ne0 Θ P ( t P I P ) = 0 , then
( τ [ d ] B ) − 1 Θ P ( t P I P ) ≅ Θ P ( ( f P P ) [ d ] ) . (\tau_{[d]}^B)^{-1}\Theta_P(t_P I_P) \cong\Theta_P((\mathfrak{f}_P P)[d]). ( τ [ d ] B ) − 1 Θ P ( t P I P ) ≅ Θ P (( f P P ) [ d ]) . In particular, if Θ P ( ( f P P ) [ d ] ) = 0 \Theta_P((\mathfrak{f}_P P)[d])=0 Θ P (( f P P ) [ d ]) = 0 , then the object Θ P ( t P I P ) \Theta_P(t_P I_P) Θ P ( t P I P ) is Ext-injective in d d d -mod B B B ; if Θ P ( t P I P ) = 0 \Theta_P(t_P I_P)=0 Θ P ( t P I P ) = 0 , then the object Θ P ( ( f P P ) [ d ] ) \Theta_P((\mathfrak{f}_P P)[d]) Θ P (( f P P ) [ d ]) is Ext-projective in d d d -mod B B B .
Proof. Assume that Θ P ( ( f P P ) [ d ] ) ≠ 0 \Theta_P((\mathfrak{f}_P P)[d]) \ne0 Θ P (( f P P ) [ d ]) = 0 . By (32) , σ B ‾ ≥ − d + 1 ( Q P ) ≠ 0 \sigma_{\overline{B}}^{\geq-d+1}(\mathbf{Q}_P) \ne0 σ B ≥ − d + 1 ( Q P ) = 0 . Since Q P \mathbf{Q}_P Q P is indecomposable by Lemma 6.2, it has no non-zero direct summand in add B [ d ] \operatorname{add}B[d] add B [ d ] . Thus we have
τ [ d ] B ( Θ P ( ( f P P ) [ d ] ) ) ≅ (6.1) τ [ d ] B ( σ B ‾ ≥ − d + 1 ( Q P ) ) ≅ (3.6) σ B ‾ < 0 ( ν B ( Q P ) [ − 1 ] ) ≅ (6.1) Θ P ( t P I P ) . \tau_{[d]}^B\bigl(\Theta_P((\mathfrak{f}_P P)[d])\bigr)
\mathrel{\overset{\text{(6.1)}}{\cong}}
\tau_{[d]}^B\bigl(\sigma_{\overline{B}}^{\geq-d+1}(\mathbf{Q}_P)\bigr)
\mathrel{\overset{\text{(3.6)}}{\cong}}
\sigma_{\overline{B}}^{<0}\bigl(\nu_B(\mathbf{Q}_P)[-1]\bigr)
\mathrel{\overset{\text{(6.1)}}{\cong}}
\Theta_P(t_P I_P). τ [ d ] B ( Θ P (( f P P ) [ d ]) ) ≅ (6.1) τ [ d ] B ( σ B ≥ − d + 1 ( Q P ) ) ≅ (3.6) σ B < 0 ( ν B ( Q P ) [ − 1 ] ) ≅ (6.1) Θ P ( t P I P ) .
The other isomorphism can be shown dually.
If Θ P ( t P I P ) = 0 \Theta_P(t_P I_P)=0 Θ P ( t P I P ) = 0 , then Θ P ( ( f P P ) [ d ] ) ≠ 0 \Theta_P((\mathfrak{f}_P P)[d])\ne0 Θ P (( f P P ) [ d ]) = 0 . Hence, by the first isomorphism in this lemma, we have τ [ d ] B Θ P ( ( f P P ) [ d ] ) = 0 \tau_{[d]}^B\Theta_P((\mathfrak{f}_P P)[d])=0 τ [ d ] B Θ P (( f P P ) [ d ]) = 0 , which implies that Θ P ( ( f P P ) [ d ] ) \Theta_P((\mathfrak{f}_P P)[d]) Θ P (( f P P ) [ d ]) is Ext-projective in d d d -mod B B B . The assertion for the case Θ P ( ( f P P ) [ d ] ) = 0 \Theta_P((\mathfrak{f}_P P)[d])=0 Θ P (( f P P ) [ d ]) = 0 can be shown dually. □
By Lemmas 6.3 and 6.4, together with Proposition 3.3 (3), the connecting almost split triangles in d d d -mod B B B are precisely those of the form
Θ P ( t P I P ) → l P E P → r P Θ P ( ( f P P ) [ d ] ) → δ P Θ P ( t P I P ) [ 1 ] , P , P [ d ] ∉ add P . (34) \Theta_P(t_P I_P) \xrightarrow{l_P} E_P \xrightarrow{r_P} \Theta_P((\mathfrak{f}_P P)[d]) \xrightarrow{\delta_P} \Theta_P(t_P I_P)[1], \qquad P,P[d]\notin\operatorname{add}P. \tag*{(34)} Θ P ( t P I P ) l P E P r P Θ P (( f P P ) [ d ]) δ P Θ P ( t P I P ) [ 1 ] , P , P [ d ] ∈ / add P . (34)
Proposition 6.5. Let P ∈ add A P \in\operatorname{add}A P ∈ add A be indecomposable with $P,P[d]\notin\operatorname{add}P. The canonical triangle of the middle term E P E_P E P in (34) with respect to ( X ( P ) , Y ( P ) ) (\mathcal{X}(P),\mathcal{Y}(P)) ( X ( P ) , Y ( P )) is
Θ P ( ( rad f P P ) [ d ] ) → E P → Θ P ( t P I P / soc t P I P ) → Θ P ( ( rad f P P ) [ d ] ) [ 1 ] . \Theta_P((\operatorname{rad}\mathfrak{f}_P P)[d]) \to E_P \to\Theta_P(t_P I_P/\operatorname{soc}t_P I_P) \to\Theta_P((\operatorname{rad}\mathfrak{f}_P P)[d])[1]. Θ P (( rad f P P ) [ d ]) → E P → Θ P ( t P I P / soc t P I P ) → Θ P (( rad f P P ) [ d ]) [ 1 ] .
Proof. We adapt the proof of [10 Corollary 5.3] with suitable modifications.
Let S = top H 0 ( A ) H 0 ( P ) S=\operatorname{top}_{H^{0}(A)}H^{0}(P) S = top H 0 ( A ) H 0 ( P ) , regarded as a dg A A A -module via A → H 0 ( A ) A\to H^{0}(A) A → H 0 ( A ) . Since H 0 ( P ) H^{0}(P) H 0 ( P ) is an indecomposable projective H 0 ( A ) H^{0}(A) H 0 ( A ) -module by (13) , S S S is a simple H 0 ( A ) H^{0}(A) H 0 ( A ) -module. Since P , P [ d ] ∉ add P P,P[d]\notin\operatorname{add}P P , P [ d ] ∈ / add P , by Proposition 3.3 (3), both f P P \mathsf{f}_{P}P f P P and t P I P \mathsf{t}_{P}I_{P} t P I P are non-zero. We claim that
top f P P ≅ S and soc t P I P ≅ S [ d − 1 ] . \operatorname{top}\mathsf{f}_{P}P\cong S\quad\text{and}\quad\operatorname{soc}\mathsf{t}_{P}I_{P}\cong S[d-1]. top f P P ≅ S and soc t P I P ≅ S [ d − 1 ] .
Consider the canonical triangle of P P P with respect to ( T ( P ) , F ( P ) ) (\mathcal{T}(P),\mathcal{F}(P)) ( T ( P ) , F ( P )) , after a rotation:
f P P [ − 1 ] ⟶ t P P ⟶ P ⟶ f P P . \mathsf{f}_{P}P[-1]\longrightarrow\mathsf{t}_{P}P\longrightarrow P\longrightarrow\mathsf{f}_{P}P. f P P [ − 1 ] ⟶ t P P ⟶ P ⟶ f P P .
Taking H 0 H^{0} H 0 gives a surjection H 0 ( P ) → H 0 ( f P P ) H^{0}(P)\to H^{0}(\mathsf{f}_{P}P) H 0 ( P ) → H 0 ( f P P ) . If H 0 ( f P P ) = 0 H^{0}(\mathsf{f}_{P}P)=0 H 0 ( f P P ) = 0 , then f P P [ − 1 ] ∈ d - mod A \mathsf{f}_{P}P[-1]\in d\text{-}\operatorname{mod}A f P P [ − 1 ] ∈ d - mod A . Since P ∈ proj d - mod A P\in\operatorname{proj}d\text{-}\operatorname{mod}A P ∈ proj d - mod A , it follows that the above triangle splits. Thus, P ∈ T ( P ) P\in\mathcal{T}(P) P ∈ T ( P ) and hence f P P = 0 \mathsf{f}_{P}P=0 f P P = 0 , a contradiction. Thus H 0 ( f P P ) H^{0}(\mathsf{f}_{P}P) H 0 ( f P P ) is a non-zero quotient of the indecomposable projective H 0 ( P ) H^{0}(P) H 0 ( P ) , so its top is S S S . Dually, H − d + 1 ( t P I P ) H^{-d+1}(\mathsf{t}_{P}I_{P}) H − d + 1 ( t P I P ) is a non-zero submodule of H − d + 1 ( I P ) H^{-d+1}(I_{P}) H − d + 1 ( I P ) , whose socle is S S S by (22) . This proves the claim.
Consider the following two triangles in d - mod A d\text{-}\operatorname{mod}A d - mod A (see Definition 2.3):
rad f P P → α f P P → β S → γ ( rad f P P ) [ 1 ] , \operatorname{rad}\mathsf{f}_{P}P\xrightarrow{\alpha}\mathsf{f}_{P}P\xrightarrow{\beta}S\xrightarrow{\gamma}(\operatorname{rad}\mathsf{f}_{P}P)[1], rad f P P α f P P β S γ ( rad f P P ) [ 1 ] ,
S [ d − 1 ] → f t P I P → g t P I P / soc t P I P → h S [ d ] . S[d-1]\xrightarrow{f}\mathsf{t}_{P}I_{P}\xrightarrow{g}\mathsf{t}_{P}I_{P}/\operatorname{soc}\mathsf{t}_{P}I_{P}\xrightarrow{h}S[d]. S [ d − 1 ] f t P I P g t P I P / soc t P I P h S [ d ] .
Since S [ − 1 ] ∈ D f d ( A ) ≥ 1 ⊆ D f d ( P ) ≥ 1 S[-1]\in D_{\mathrm{fd}}(A)^{\geq1}\subseteq D_{\mathrm{fd}}(P)^{\geq1} S [ − 1 ] ∈ D fd ( A ) ≥ 1 ⊆ D fd ( P ) ≥ 1 and S [ d ] ∈ D f d ( A ) ≤ − d ⊆ D f d ( P ) ≤ 0 S[d]\in D_{\mathrm{fd}}(A)^{\leq-d}\subseteq D_{\mathrm{fd}}(P)^{\leq0} S [ d ] ∈ D fd ( A ) ≤ − d ⊆ D fd ( P ) ≤ 0 by (21) , applying Hom D ( A ) ( P , − ) \operatorname{Hom}_{D(A)}(P,-) Hom D ( A ) ( P , − ) to these two triangles shows that rad f P P ∈ F ( P ) \operatorname{rad}\mathsf{f}_{P}P\in\mathcal{F}(P) rad f P P ∈ F ( P ) and t P I P / soc t P I P ∈ T ( P ) \mathsf{t}_{P}I_{P}/\operatorname{soc}\mathsf{t}_{P}I_{P}\in\mathcal{T}(P) t P I P / soc t P I P ∈ T ( P ) .
Since β ≠ 0 \beta\ne0 β = 0 , the morphism α \alpha α is not a retraction. Hence Θ P ( α [ d ] ) \Theta_{P}(\alpha[d]) Θ P ( α [ d ]) factors through the morphism r P r_{P} r P in (34) . Consequently, δ P ∘ Θ P ( α [ d ] ) = 0 \delta_{P}\circ\Theta_{P}(\alpha[d])=0 δ P ∘ Θ P ( α [ d ]) = 0 . Applying real P \operatorname{real}_{P} real P , by Theorem 4.2 , we have real P ( δ P ) ∘ α [ d ] = 0 \operatorname{real}_{P}(\delta_{P})\circ\alpha[d]=0 real P ( δ P ) ∘ α [ d ] = 0 . Hence, there exists a morphism θ : S [ d − 1 ] → t P I P \theta:S[d-1]\to\mathsf{t}_{P}I_{P} θ : S [ d − 1 ] → t P I P such that real P ( δ P ) = θ [ 1 ] ∘ β [ d ] \operatorname{real}_{P}(\delta_{P})=\theta[1]\circ\beta[d] real P ( δ P ) = θ [ 1 ] ∘ β [ d ] ; see the following commutative diagram.
( rad f P P ) [ d ] → α [ d ] ( f P P ) [ d ] → β [ d ] S [ d ] → γ [ d ] ( rad f P P ) [ d + 1 ] ↓ real P ( δ P ) ↘ θ [ 1 ] t P I P [ 1 ] \begin{aligned}
(\operatorname{rad}\mathsf{f}_{P}P)[d]&\xrightarrow{\alpha[d]}(\mathsf{f}_{P}P)[d]\xrightarrow{\beta[d]}S[d]\xrightarrow{\gamma[d]}(\operatorname{rad}\mathsf{f}_{P}P)[d+1]\\
&\hspace{4.4cm}\downarrow\operatorname{real}_{P}(\delta_{P})\qquad\qquad\searrow\theta[1]\\
&\hspace{3.8cm}\mathsf{t}_{P}I_{P}[1]
\end{aligned} ( rad f P P ) [ d ] α [ d ] ( f P P ) [ d ] β [ d ] S [ d ] γ [ d ] ( rad f P P ) [ d + 1 ] ↓ real P ( δ P ) ↘ θ [ 1 ] t P I P [ 1 ]
Since real P ( δ P ) ≠ 0 \operatorname{real}_{P}(\delta_{P})\ne0 real P ( δ P ) = 0 by (26) , we have θ ≠ 0 \theta\ne0 θ = 0 . Then by (18) and soc t P I P ≅ S [ d − 1 ] \operatorname{soc}\mathsf{t}_{P}I_{P}\cong S[d-1] soc t P I P ≅ S [ d − 1 ] , there exists an automorphism u u u of S S S such that θ = f ∘ ( u [ d − 1 ] ) \theta=f\circ(u[d-1]) θ = f ∘ ( u [ d − 1 ]) . Consequently, we have cone θ ≅ t P I P / soc t P I P \operatorname{cone}\theta\cong\mathsf{t}_{P}I_{P}/\operatorname{soc}\mathsf{t}_{P}I_{P} cone θ ≅ t P I P / soc t P I P . Applying the octahedral axiom to θ ∘ β [ d − 1 ] = real P ( δ P ) [ − 1 ] \theta\circ\beta[d-1]=\operatorname{real}_{P}(\delta_{P})[-1] θ ∘ β [ d − 1 ] = real P ( δ P ) [ − 1 ] , we obtain the following commutative diagram of triangles:
( f P P ) [ d − 1 ] → β [ d − 1 ] ( f P P ) [ d − 1 ] ↓ ↓ real P ( δ P ) [ − 1 ] ( t P I P / soc t P I P ) [ − 1 ] ⟶ S [ d − 1 ] → θ t P I P ⟶ t P I P / soc t P I P ∥ ↓ ↓ ∥ ( t P I P / soc t P I P ) [ − 1 ] ⟶ ( rad f P P ) [ d ] ⟶ real P ( E P ) ⟶ t P I P / soc t P I P ↓ ↓ ( f P P ) [ d ] = ( f P P ) [ d ] \begin{aligned}
(\mathsf{f}_{P}P)[d-1]&\xrightarrow{\beta[d-1]}(\mathsf{f}_{P}P)[d-1]\\
\downarrow&&\downarrow\operatorname{real}_{P}(\delta_{P})[-1]\\
(\mathsf{t}_{P}I_{P}/\operatorname{soc}\mathsf{t}_{P}I_{P})[-1]&\longrightarrow S[d-1]\xrightarrow{\theta}\mathsf{t}_{P}I_{P}\longrightarrow\mathsf{t}_{P}I_{P}/\operatorname{soc}\mathsf{t}_{P}I_{P}\\
\Vert&&\downarrow\qquad\qquad\downarrow\qquad\qquad\Vert\\
(\mathsf{t}_{P}I_{P}/\operatorname{soc}\mathsf{t}_{P}I_{P})[-1]&\longrightarrow(\operatorname{rad}\mathsf{f}_{P}P)[d]\longrightarrow\operatorname{real}_{P}(E_{P})\longrightarrow\mathsf{t}_{P}I_{P}/\operatorname{soc}\mathsf{t}_{P}I_{P}\\
&&\downarrow\qquad\qquad\downarrow\\
&&(\mathsf{f}_{P}P)[d]\xlongequal{\hspace{1cm}}(\mathsf{f}_{P}P)[d]
\end{aligned} ( f P P ) [ d − 1 ] ↓ ( t P I P / soc t P I P ) [ − 1 ] ∥ ( t P I P / soc t P I P ) [ − 1 ] β [ d − 1 ] ( f P P ) [ d − 1 ] ⟶ S [ d − 1 ] θ t P I P ⟶ t P I P / soc t P I P ⟶ ( rad f P P ) [ d ] ⟶ real P ( E P ) ⟶ t P I P / soc t P I P ↓ real P ( δ P ) [ − 1 ] ↓ ↓ ∥ ↓ ↓ ( f P P ) [ d ] ( f P P ) [ d ]
Thus, by Theorem 4.2 , applying Θ P \Theta_{P} Θ P to the third row yields the required triangle, where Θ P ( ( rad f P P ) [ d ] ) ∈ X ( P ) \Theta_{P}((\operatorname{rad}\mathsf{f}_{P}P)[d])\in\mathcal{X}(P) Θ P (( rad f P P ) [ d ]) ∈ X ( P ) and Θ P ( t P I P / soc t P I P ) ∈ Y ( P ) \Theta_{P}(\mathsf{t}_{P}I_{P}/\operatorname{soc}\mathsf{t}_{P}I_{P})\in\mathcal{Y}(P) Θ P ( t P I P / soc t P I P ) ∈ Y ( P ) .
Separating and splitting silting objects
We generalize the notions of separating and splitting 2-term silting complexes to the ( d + 1 ) (d+1) ( d + 1 ) -term setting.
An s s s -torsion pair ( T , F ) (\mathcal{T},\mathcal{F}) ( T , F ) in the d d d -extended module category of a proper d d d -truncated non-positive dg algebra is called split if Hom ( F , T [ 1 ] ) = 0 \operatorname{Hom}(\mathcal{F},\mathcal{T}[1])=0 Hom ( F , T [ 1 ]) = 0 . This is also equivalent to every indecomposable object in the extended module category belonging to T \mathcal{T} T or F \mathcal{F} F .
Definition 7.1. Let A A A be a proper d d d -truncated non-positive dg algebra, and let P P P be a ( d + 1 ) (d+1) ( d + 1 ) -term silting object in per A \operatorname{per} A per A . Let B = d - End A ( P ) B=d\text{-}\operatorname{End}_A(P) B = d - End A ( P ) be the d d d -truncated dg endomorphism algebra of P P P . Then
P P P is called separating if the s s s -torsion pair ( T ( P ) , F ( P ) ) (\mathcal{T}(P),\mathcal{F}(P)) ( T ( P ) , F ( P )) in d - mod A d\text{-}\operatorname{mod} A d - mod A is split, and
P P P is called splitting if the s s s -torsion pair ( X ( P ) , Y ( P ) ) (\mathcal{X}(P),\mathcal{Y}(P)) ( X ( P ) , Y ( P )) in d - mod B d\text{-}\operatorname{mod} B d - mod B is split.
The following lemma is useful.
Lemma 7.2. Let A A A be a proper d d d -truncated non-positive dg algebra, and let P P P be a ( d + 1 ) (d+1) ( d + 1 ) -term silting object in per A \operatorname{per} A per A . There are triangles in per A \operatorname{per} A per A :
X r + 1 → g r + 1 A r → f r X r → X r + 1 [ 1 ] , 0 ≤ r ≤ d , (35) X_{r+1} \xrightarrow{g_{r+1}} A_r \xrightarrow{f_r} X_r \to X_{r+1}[1], \qquad0 \leq r \leq d, \tag*{(35)} X r + 1 g r + 1 A r f r X r → X r + 1 [ 1 ] , 0 ≤ r ≤ d , (35) where X 0 = P X_0=P X 0 = P , X d + 1 = 0 X_{d+1}=0 X d + 1 = 0 , A r ∈ add A A_r \in\operatorname{add} A A r ∈ add A and each morphism f r f_r f r is a minimal right add A \operatorname{add} A add A -approximation. Moreover, for every simple right H 0 ( A ) H^0(A) H 0 ( A ) -module S S S , regarded as a dg A A A -module via the canonical projection A → H 0 ( A ) A \to H^0(A) A → H 0 ( A ) , there are isomorphisms
Hom D ( A ) ( P , S [ r ] ) ≅ Hom D ( A ) ( A r , S ) , 0 ≤ r ≤ d . (36) \operatorname{Hom}_{D(A)}(P,S[r]) \cong\operatorname{Hom}_{D(A)}(A_r,S), \qquad0 \leq r \leq d. \tag*{(36)} Hom D ( A ) ( P , S [ r ]) ≅ Hom D ( A ) ( A r , S ) , 0 ≤ r ≤ d . (36)
Proof. The proofs of [12 Lemmas 8.1, 8.2 and Proposition 8.3] apply after interchanging the roles of A A A and P P P and replacing minimal left approximations by minimal right approximations.
Separating silting objects can be shown to satisfy the equivalent conditions in Theorem 5.2 .
Proposition 7.3. Let A A A be a proper d d d -truncated non-positive dg algebra that is not acyclic, and let P P P be a separating ( d + 1 ) (d+1) ( d + 1 ) -term silting object in per A \operatorname{per} A per A . Then
Hom per A ( P , P [ i ] ) = 0 for all i < − a , \operatorname{Hom}_{\operatorname{per} A}(P,P[i])=0 \qquad\text{for all } i<-a, Hom per A ( P , P [ i ]) = 0 for all i < − a , where a a a is the largest non-negative integer such that H − a ( A ) ≠ 0 H^{-a}(A)\ne0 H − a ( A ) = 0 . In particular, the equivalent conditions in Theorem 5.2 hold.
Proof. For any 1 ≤ r ≤ d 1 \leq r \leq d 1 ≤ r ≤ d , let l r = max { 0 , r − a − 1 } l_r=\max\{0,r-a-1\} l r = max { 0 , r − a − 1 } . Then
0 ≤ l r ≤ d − 1 − a and 1 ≤ r − l r ≤ a + 1. 0 \leq l_r \leq d-1-a \quad\text{and} \quad1 \leq r-l_r \leq a+1. 0 ≤ l r ≤ d − 1 − a and 1 ≤ r − l r ≤ a + 1.
It follows that A r [ l r ] ∈ D f d ( A ) [ − a − l r , − l r ] ⊆ d - mod A A_r[l_r]\in D_{\mathrm{fd}}(A)^{[-a-l_r,-l_r]}\subseteq d\text{-}\operatorname{mod} A A r [ l r ] ∈ D fd ( A ) [ − a − l r , − l r ] ⊆ d - mod A for all r r r . We claim that A r [ l r ] ∈ F ( P ) A_r[l_r]\in\mathcal{F}(P) A r [ l r ] ∈ F ( P ) for all r r r . Indeed, let E E E be an indecomposable summand of A r A_r A r . By (13) , H 0 ( E ) H^0(E) H 0 ( E ) is an indecomposable projective H 0 ( A ) H^0(A) H 0 ( A ) -module. Let S S S be the simple top of H 0 ( E ) H^0(E) H 0 ( E ) . We regard S S S as a dg A A A -module via the canonical morphism A → H 0 ( A ) A\to H^0(A) A → H 0 ( A ) . Then there exists a non-zero morphism from E E E to S S S in D f d ( A ) D_{\mathrm{fd}}(A) D fd ( A ) , which implies Hom D f d ( A ) ( A r , S ) ≠ 0 \operatorname{Hom}_{D_{\mathrm{fd}}(A)}(A_r,S)\ne0 Hom D fd ( A ) ( A r , S ) = 0 . Note that both E [ l r ] E[l_r] E [ l r ] and S [ l r ] S[l_r] S [ l r ] are in d - mod A d\text{-}\operatorname{mod} A d - mod A . Hence, by (36) in Lemma 7.2, we have
Hom D f d ( A ) ( P , S [ l r ] [ r − l r ] ) = Hom D f d ( A ) ( P , S [ r ] ) ≠ 0. \operatorname{Hom}_{D_{\mathrm{fd}}(A)}(P,S[l_r][r-l_r])=\operatorname{Hom}_{D_{\mathrm{fd}}(A)}(P,S[r])\ne0. Hom D fd ( A ) ( P , S [ l r ] [ r − l r ]) = Hom D fd ( A ) ( P , S [ r ]) = 0.
Since r − l r ≥ 1 r-l_{r} \ge1 r − l r ≥ 1 , this implies S [ l r ] ∉ T ( P ) S[l_{r}] \notin\mathcal{T}(\mathbf{P}) S [ l r ] ∈ / T ( P ) . Since ( T ( P ) , F ( P ) ) (\mathcal{T}(\mathbf{P}),\mathcal{F}(\mathbf{P})) ( T ( P ) , F ( P )) is split, we have S [ l r ] ∈ F ( P ) S[l_{r}] \in\mathcal{F}(\mathbf{P}) S [ l r ] ∈ F ( P ) . Since Hom D f d ( A ) ( E [ l r ] , S [ l r ] ) ≅ Hom D f d ( A ) ( E , S ) ≠ 0 \operatorname{Hom}_{\mathcal{D}_{\mathrm{fd}}(A)}(E[l_{r}],S[l_{r}]) \cong\operatorname{Hom}_{\mathcal{D}_{\mathrm{fd}}(A)}(E,S) \ne0 Hom D fd ( A ) ( E [ l r ] , S [ l r ]) ≅ Hom D fd ( A ) ( E , S ) = 0 , we have E [ l r ] ∉ T ( P ) E[l_{r}] \notin\mathcal{T}(\mathbf{P}) E [ l r ] ∈ / T ( P ) . Since E [ l r ] E[l_{r}] E [ l r ] is indecomposable, we have E [ l r ] ∈ F ( P ) E[l_{r}] \in\mathcal{F}(\mathbf{P}) E [ l r ] ∈ F ( P ) , which proves the claim.
Fix a negative integer i < − a i < -a i < − a . For any 1 ≤ r ≤ d 1 \le r \le d 1 ≤ r ≤ d , since A r [ l r ] ∈ F ( P ) A_{r}[l_{r}] \in\mathcal{F}(\mathbf{P}) A r [ l r ] ∈ F ( P ) and r + i − l r ≤ a + 1 + i ≤ 0 r+i-l_{r} \le a+1+i \le0 r + i − l r ≤ a + 1 + i ≤ 0 , we have
Hom D f d ( A ) ( P , A r [ r + i ] ) = Hom D f d ( A ) ( P , A r [ l r ] [ r + i − l r ] ) = 0. \operatorname{Hom}_{\mathcal{D}_{\mathrm{fd}}(A)}(\mathbf{P},A_{r}[r+i])=\operatorname{Hom}_{\mathcal{D}_{\mathrm{fd}}(A)}(\mathbf{P},A_{r}[l_{r}][r+i-l_{r}])=0. Hom D fd ( A ) ( P , A r [ r + i ]) = Hom D fd ( A ) ( P , A r [ l r ] [ r + i − l r ]) = 0.
For the case r = 0 r=0 r = 0 , since A ∈ D f d ( A ) [ − a , 0 ] A \in\mathcal{D}_{\mathrm{fd}}(A)^{[-a,0]} A ∈ D fd ( A ) [ − a , 0 ] , we have A 0 [ i ] ∈ D f d ( A ) ≥ 1 A_{0}[i] \in\mathcal{D}_{\mathrm{fd}}(A)^{\ge1} A 0 [ i ] ∈ D fd ( A ) ≥ 1 . Since P ∈ D f d ( A ) ≤ 0 \mathbf{P} \in\mathcal{D}_{\mathrm{fd}}(A)^{\le0} P ∈ D fd ( A ) ≤ 0 , we have Hom D f d ( A ) ( P , A 0 [ i ] ) = 0 \operatorname{Hom}_{\mathcal{D}_{\mathrm{fd}}(A)}(\mathbf{P},A_{0}[i])=0 Hom D fd ( A ) ( P , A 0 [ i ]) = 0 . Therefore, we have Hom D f d ( A ) ( P , A r [ r + i ] ) = 0 \operatorname{Hom}_{\mathcal{D}_{\mathrm{fd}}(A)}(\mathbf{P},A_{r}[r+i])=0 Hom D fd ( A ) ( P , A r [ r + i ]) = 0 for all 0 ≤ r ≤ d 0 \le r \le d 0 ≤ r ≤ d .
Applying the functor Hom D f d ( A ) ( P , − ) \operatorname{Hom}_{\mathcal{D}_{\mathrm{fd}}(A)}(\mathbf{P},-) Hom D fd ( A ) ( P , − ) to the triangles in (35) yields an exact sequence
0 = Hom D f d ( A ) ( P , A r [ r + i ] ) → Hom D f d ( A ) ( P , X r [ r + i ] ) → Hom D f d ( A ) ( P , X r + 1 [ r + i + 1 ] ) . 0=\operatorname{Hom}_{\mathcal{D}_{\mathrm{fd}}(A)}(\mathbf{P},A_{r}[r+i]) \to\operatorname{Hom}_{\mathcal{D}_{\mathrm{fd}}(A)}(\mathbf{P},X_{r}[r+i]) \to\operatorname{Hom}_{\mathcal{D}_{\mathrm{fd}}(A)}(\mathbf{P},X_{r+1}[r+i+1]). 0 = Hom D fd ( A ) ( P , A r [ r + i ]) → Hom D fd ( A ) ( P , X r [ r + i ]) → Hom D fd ( A ) ( P , X r + 1 [ r + i + 1 ]) .
Starting with X d + 1 = 0 X_{d+1}=0 X d + 1 = 0 , descending induction gives Hom D f d ( A ) ( P , X r [ r + i ] ) = 0 \operatorname{Hom}_{\mathcal{D}_{\mathrm{fd}}(A)}(\mathbf{P},X_{r}[r+i])=0 Hom D fd ( A ) ( P , X r [ r + i ]) = 0 for all 0 ≤ r ≤ d 0 \le r \le d 0 ≤ r ≤ d . Since X 0 = P X_{0}=\mathbf{P} X 0 = P , this completes the proof. □
An immediate corollary of the above result is the following, which generalizes [10 Proposition 5.7] .
Corollary 7.4. If A A A is an ordinary finite-dimensional algebra, then any separating ( d + 1 ) (d+1) ( d + 1 ) -term silting object in per A \operatorname{per} A per A is tilting.
For splitting silting objects, we have the following criterion; compare [10 Lemma 5.5] .
Lemma 7.5. Let A A A be a proper d d d -truncated non-positive dg algebra, and let P \mathbf{P} P be a ( d + 1 ) (d+1) ( d + 1 ) -term silting object in per A \operatorname{per} A per A . Then P \mathbf{P} P is splitting if and only if
Hom D ( A ) ( T ( P ) , F ( P ) [ d + 1 ] ) = 0. \operatorname{Hom}_{\mathcal{D}(A)}(\mathcal{T}(\mathbf{P}),\mathcal{F}(\mathbf{P})[d+1])=0. Hom D ( A ) ( T ( P ) , F ( P ) [ d + 1 ]) = 0.
Proof. This follows from the definition and the isomorphism (26) in Theorem 4.2 for i = 1 i=1 i = 1 . □
If A A A is quasi-isomorphic to an ordinary finite-dimensional hereditary algebra, then we have Hom D ( A ) ( d - mod A , d - mod A [ d + 1 ] ) = 0 \operatorname{Hom}_{\mathcal{D}(A)}(d\text{-}\operatorname{mod} A,d\text{-}\operatorname{mod} A[d+1])=0 Hom D ( A ) ( d - mod A , d - mod A [ d + 1 ]) = 0 . Hence, in this case, by Lemma 7.5 , every ( d + 1 ) (d+1) ( d + 1 ) -term silting object is splitting.
Since both T ( P ) \mathcal{T}(\mathbf{P}) T ( P ) and F ( P ) \mathcal{F}(\mathbf{P}) F ( P ) are functorially finite, by [16 Theorem 5.14] , they also have almost split triangles.
Proposition 7.6. Assume that P \mathbf{P} P is splitting. Then every almost split triangle in d - mod B d\text{-}\operatorname{mod} B d - mod B has its first three terms in either X ( P ) \mathcal{X}(\mathbf{P}) X ( P ) or Y ( P ) \mathcal{Y}(\mathbf{P}) Y ( P ) , or else it is of the form
Θ P ( t P I P ) → Θ P ( ( rad f P P ) [ d ] ) ⊕ Θ P ( t P I P / soc t P I P ) → Θ P ( ( f P P ) [ d ] ) → Θ P ( t P I P ) [ 1 ] , \Theta_{\mathbf{P}}(t_{\mathbf{P}}I_{\mathbf{P}}) \to\Theta_{\mathbf{P}}((\operatorname{rad} f_{\mathbf{P}}P)[d]) \oplus\Theta_{\mathbf{P}}(t_{\mathbf{P}}I_{\mathbf{P}}/\operatorname{soc} t_{\mathbf{P}}I_{\mathbf{P}}) \to\Theta_{\mathbf{P}}((f_{\mathbf{P}}P)[d]) \to\Theta_{\mathbf{P}}(t_{\mathbf{P}}I_{\mathbf{P}})[1], Θ P ( t P I P ) → Θ P (( rad f P P ) [ d ]) ⊕ Θ P ( t P I P / soc t P I P ) → Θ P (( f P P ) [ d ]) → Θ P ( t P I P ) [ 1 ] , where P P P is an indecomposable object in add A \operatorname{add} A add A with P , P [ d ] ∉ add P P,P[d]\notin\operatorname{add}\mathbf{P} P , P [ d ] ∈ / add P . Moreover, the functor Θ P \Theta_{\mathbf{P}} Θ P (resp. Θ P ( − [ d ] ) \Theta_{\mathbf{P}}(-[d]) Θ P ( − [ d ]) ) identifies the almost split triangles in T ( P ) \mathcal{T}(\mathbf{P}) T ( P ) (resp. F ( P ) \mathcal{F}(\mathbf{P}) F ( P ) ) with those in d - mod B d\text{-}\operatorname{mod} B d - mod B whose first three terms lie in Y ( P ) \mathcal{Y}(\mathbf{P}) Y ( P ) (resp. X ( P ) \mathcal{X}(\mathbf{P}) X ( P ) ).
Proof. Since the pair ( X ( P ) , Y ( P ) ) (\mathcal{X}(\mathbf{P}),\mathcal{Y}(\mathbf{P})) ( X ( P ) , Y ( P )) is split, every indecomposable object in d - mod B d\text{-}\operatorname{mod} B d - mod B belongs to X ( P ) \mathcal{X}(\mathbf{P}) X ( P ) or Y ( P ) \mathcal{Y}(\mathbf{P}) Y ( P ) .
The first statement follows from Lemma 6.4 and Proposition 6.5. To show the second statement, by Theorem 4.2 , it suffices to prove that any almost split triangle in X ( P ) \mathcal{X}(\mathbf{P}) X ( P ) (resp. Y ( P ) \mathcal{Y}(\mathbf{P}) Y ( P ) ) is also an almost split triangle in d - mod B d\text{-}\operatorname{mod} B d - mod B . We only show this for X ( P ) \mathcal{X}(\mathbf{P}) X ( P ) , since the case of Y ( P ) \mathcal{Y}(\mathbf{P}) Y ( P ) is dual. Let X → α Y → Z → X [ 1 ] X \xrightarrow{\alpha} Y \to Z \to X[1] X α Y → Z → X [ 1 ] be an almost split triangle in X ( P ) \mathcal{X}(\mathbf{P}) X ( P ) . Let f : X → M f:X\to M f : X → M be a non-section morphism in d - mod B d\text{-}\operatorname{mod} B d - mod B . Write M = M 1 ⊕ M 2 M=M_{1}\oplus M_{2} M = M 1 ⊕ M 2 with M 1 ∈ X ( P ) M_{1}\in\mathcal{X}(\mathbf{P}) M 1 ∈ X ( P ) and M 2 ∈ Y ( P ) M_{2}\in\mathcal{Y}(\mathbf{P}) M 2 ∈ Y ( P ) . Then f f f has two components f 1 : X → M 1 f_{1}:X\to M_{1} f 1 : X → M 1 and f 2 : X → M 2 f_{2}:X\to M_{2} f 2 : X → M 2 , where f 1 f_{1} f 1 is not a section and f 2 f_2 f 2 has to be zero since Hom D ( B ) ( X ( P ) , Y ( P ) ) = 0 \operatorname{Hom}_{D(B)}(\mathcal{X}(P),\mathcal{Y}(P)) = 0 Hom D ( B ) ( X ( P ) , Y ( P )) = 0 . Thus, f f f factors through α \alpha α . Since Z Z Z is indecomposable, by [16 Theorem 2.9] , X ⟶ Y ⟶ Z ⟶ X [ 1 ] X \longrightarrow Y \longrightarrow Z \longrightarrow X[1] X ⟶ Y ⟶ Z ⟶ X [ 1 ] is an almost split triangle in d d d -mod B B B .
Use of AI: The authors found a proof of Proposition 7.3 with AI assistance and wrote it up themselves. AI tools were also used to correct grammar and spelling.
References
LC: School of Mathematical Sciences, Beijing Normal University, 100875 Beijing, China Email address: 202631130047@mail.bnu.edu.cn
ZY: School of Mathematical Sciences, Beijing Normal University, 100875 Beijing, China Email address: yuzhoumath@gmail.com