Cluster-Tilted Algebras Are Gorenstein and Stably Calabi-Yau
Cluster-Tilted Algebras Are Gorenstein and Stably Calabi-Yau
Cluster-Tilted Algebras Are Gorenstein and Stably Calabi-Yau
www.elsevier.com/locate/aim
Abstract
We prove that in a 2-Calabi–Yau triangulated category, each cluster tilting subcategory is Gorenstein with
all its finitely generated projectives of injective dimension at most one. We show that the stable category
of its Cohen–Macaulay modules is 3-Calabi–Yau. We deduce in particular that cluster-tilted algebras are
Gorenstein of dimension at most one, and hereditary if they are of finite global dimension. Our results also
apply to the stable (!) endomorphism rings of maximal rigid modules of [Christof Geiß, Bernard Leclerc, Jan
Schröer, Rigid modules over preprojective algebras, arXiv: math.RT/0503324, Invent. Math., in press]. In
addition, we prove a general result about relative 3-Calabi–Yau duality over non-stable endomorphism rings.
This strengthens and generalizes the Ext-group symmetries obtained in [Christof Geiß, Bernard Leclerc, Jan
Schröer, Rigid modules over preprojective algebras, arXiv: math.RT/0503324, Invent. Math., in press] for
simple modules. Finally, we generalize the results on relative Calabi–Yau duality from 2-Calabi–Yau to
d-Calabi–Yau categories. We show how to produce many examples of d-cluster tilted algebras.
© 2006 Elsevier Inc. All rights reserved.
Keywords: Cluster algebra; Cluster category; Tilting; Gorenstein algebra; Calabi–Yau category
* Corresponding author.
E-mail addresses: keller@math.jussieu.fr (B. Keller), idunr@math.ntnu.no (I. Reiten).
0001-8708/$ – see front matter © 2006 Elsevier Inc. All rights reserved.
doi:10.1016/j.aim.2006.07.013
124 B. Keller, I. Reiten / Advances in Mathematics 211 (2007) 123–151
1. Introduction
Let k be a field and H a finite-dimensional hereditary algebra. The associated cluster cat-
egory CH was introduced in [12] and, for algebras H of Dynkin type An , in [19]. It serves
in the representation-theoretic approach to cluster algebras introduced and studied by Fomin–
Zelevinsky in a series of articles [8,21,22]. We refer to [23] for more background on cluster
algebras and to [1,2,10,11,14–18,28,33,34,46,47] for some recent developments in the study of
their links with representation theory.
The cluster category CH is the orbit category of the bounded derived category of finite-
dimensional right H -modules under the action of the unique autoequivalence F satisfying
where D denotes the duality functor Homk (?, k) and S the suspension functor of the derived
category. The functor F is τ S −1 , where τ is the Auslander–Reiten translation in the bounded de-
rived category, or alternatively, F is ΣS −2 , where Σ is the Serre functor of the bounded derived
category. The cluster category is triangulated and Calabi–Yau of CY-dimension 2, cf. [38]. We
still denote by S its suspension functor. A cluster-tilted algebra [13] is the endomorphism algebra
of a cluster tilting object T of CH ; this means that T has no self-extensions, i.e.
Hom(T , ST ) = 0,
and any direct sum T ⊕ T with an indecomposable T not occurring as a direct summand of T
does have self-extensions.
We will prove that each cluster-tilted algebra is Gorenstein in the sense of [5], cf. also [30],
and that its projective objects are of injective dimension 1 (hence its injective objects are of
projective dimension 1). We deduce that if it has finite global dimension, then it is hereditary.
In particular, if it is given as the quotient of a directed quiver algebra by an admissible ideal, then
this ideal has to vanish. We will also show that the stable category of Cohen–Macaulay modules
over a cluster-tilted algebra is Calabi–Yau of CY-dimension 3.
We work in a more general framework which also covers the stable (!) endomorphism algebras
of the maximal rigid modules of [27]. In our framework, the triangulated category CH is replaced
by an arbitrary triangulated category C with finite-dimensional Hom-spaces and which is Calabi–
Yau of CY-dimension 2, i.e. we have
D Hom(X, Y ) = Hom Y, S 2 X , X, Y ∈ C.
This holds for cluster categories (including those coming from Ext-finite hereditary abelian cat-
egories with a tilting object, cf. [12,38], and also those coming from the Ext-finite hereditary
abelian k-categories in [43,44], see [38]), for stable module categories of preprojective algebras
of Dynkin graphs (cf. [7, 3.1, 1.2], [38, 8.5]) as investigated in [27] and also for stable categories
CM(R) of Cohen–Macaulay modules over commutative complete local Gorenstein isolated sin-
gularities of dimension 3, using [4]. The tilting object is replaced by a full k-linear subcategory
T ⊂ C such that, among other conditions, we have
Hom(T , ST ) = 0
B. Keller, I. Reiten / Advances in Mathematics 211 (2007) 123–151 125
for all T , T ∈ T and where T is maximal for this property. We then consider the category
mod T of finitely presented (right) modules over T and show that it is abelian and Gorenstein,
and that the homological dimensions of the projectives and the injectives are bounded by 1. We
also prove that mod T is equivalent to the quotient of C by an ideal, thus extending a result of
[13], cf. also [49], and we generalize Theorem 4.2 of [13] to give a relationship between mod T
and mod T for ‘neighbouring’ subcategories T and T of the type we consider, where C is a
Calabi–Yau category of CY-dimension 2 whose cluster tilting subcategories have ‘no loops.’ We
give examples where T has an infinite number of isomorphism classes of indecomposables.
In Section 3, we show that the stable category of Cohen–Macaulay modules over T is Calabi–
Yau of CY-dimension 3. In the case where T is invariant under the Serre functor, this result was
first proved in [25]. It closely resembles the Ext-group symmetries proved in Proposition 6.2
of [27] for simple modules over non-stable endomorphism rings. We strengthen and generalize
these Ext-group symmetries in Section 4, where we prove a relative 3-Calabi–Yau duality over
non-stable endomorphism rings.
In Section 5, we generalize the results obtained on relative Calabi–Yau duality to d-Calabi–
Yau categories (cf. also [47,48]). We also show how to produce many examples of d-cluster
tilting subcategories in d-cluster categories. For further examples, we refer to [34].
Let k be a field and C a triangulated k-linear category with split idempotents and suspen-
sion functor S. We suppose that all Hom-spaces of C are finite-dimensional and that C admits a
Serre functor Σ, cf. [44]. We suppose that C is Calabi–Yau of CY-dimension 2, i.e. there is an
isomorphism of triangle functors
∼
S2 −
→ Σ.
Note that we do not exclude the possibility that there might already exist an isomorphism
∼
Sd −→ Σ for d = 0 or d = 1. We fix such an isomorphism once and for all. For X, Y ∈ C and
n ∈ Z, we put
Extn (X, Y ) = C X, S n Y .
Note that condition (b) means that for each X ∈ C, there is a right T -approximation, i.e. a mor-
phism T → X with T ∈ T such that each morphism T → X with T ∈ T factors through T .
Condition (d) is self-dual (because of the Calabi–Yau property) and so are conditions (a) and (c).
Part (b) of the proposition below shows that the dual of condition (b) also holds for T .
126 B. Keller, I. Reiten / Advances in Mathematics 211 (2007) 123–151
We point out that there are examples of Calabi–Yau categories of CY-dimension 2 where
we have cluster tilting subcategories with an infinite number of isomorphism classes of inde-
composables. They arise from certain Ext-finite hereditary abelian categories with Serre duality
from [43,44]. For example, let
... ◦ ◦ ◦ ...
be the A∞ ∞ -quiver with linear orientation and H the category of its finite-dimensional represen-
tations over k. Let C = CH be the corresponding cluster category. The Auslander–Reiten quiver
of H is of the form ZA∞ . Corresponding to a section of ZA∞ with zigzag orientation, there is a
tilting class θ for H in the sense of [43], that is
Proposition.
S −1 X → T1X → T0X → X
of C with TiX in T . In any such triangle, T0X → X is a right T -approximation and X → ST1X
a left ST -approximation.
(b ) For each object X ∈ C, there is a triangle
X → ΣTX0 → ΣTX1 → SX
Proof. (a) Indeed, C is a triangulated category. Thus C admits weak kernels. Now T is con-
travariantly finite in C and therefore T also admits weak kernels. This is equivalent to the fact
that mod T is abelian.
(b) Since T is contravariantly finite in C, there is a right T -approximation T0 → X. Form a
triangle
T1 → T0 → X → ST1 .
For T ∈ T , the long exact sequence obtained by applying C(T , ?) to this triangle shows that
Ext1 (T , T1 ) vanishes. Thus T1 lies in T . Conversely, if we are given such a triangle, then T0 → X
is a right T -approximation because C(T , ST1 ) vanishes for T ∈ T , and X → ST1 is a left ST -
approximation because C(T0 , ST ) vanishes for T ∈ T .
(b ) We apply (b) to Y = S −1 X and obtain a triangle
S −2 X → T1Y → T0Y → S −1 X.
∼
Now we apply S 2 −
→ Σ to this triangle to get the required triangle
C(?, T1 ) → C(?, T0 ) → M → 0.
Form a triangle
T1 → T0 → X → ST1 .
F T1 → F T0 → F X → 0
S −1 X → T1X → T0X → X
with T1X and T0X in T . Since C(T0 , f ) = 0, the morphism f factors through ST1 .
For T ∈ T , we have
F ΣT = C(?, ΣT ) = DC(T , ?)
F S −1 ΣT = F ST = 0.
128 B. Keller, I. Reiten / Advances in Mathematics 211 (2007) 123–151
This shows that the image under F of the triangle of (b) yields a projective resolution of length 1
of F ΣT . Similarly, to obtain an injective resolution of F T , T ∈ T , we apply (b ) to X = T and
use the fact that F ST = 0. 2
Corollary.
(a) Each T -module is either of infinite projective (respectively injective) dimension or of pro-
jective (respectively injective) dimension at most 1.
(b) The category mod T is either hereditary or of infinite global dimension.
Proof. (a) Suppose that M is a T -module of finite projective dimension. Denote by ΩM the ker-
nel of an epimorphism P → M with P projective. Suppose that M is of projective dimension n.
Then Ω n M is projective. By induction on p 0, we find that Ω n−p M is of injective dimension
at most 1. In particular, M is of injective dimension at most 1. Dually, one shows the statement
on the injective dimensions. Part (b) is immediate from (a). 2
In this section, we briefly indicate how to generalize Theorem 4.2 of [13] beyond cluster cat-
egories, namely to Calabi–Yau categories of CY-dimension 2 whose cluster tilting subcategories
have ‘no loops.’ This hypothesis holds not only for cluster categories but also for the stable
categories of finite-dimensional preprojective algebras [27].
We recall some results from [33]. Let C be a Calabi–Yau category of CY-dimension 2. We
assume that for each cluster tilting subcategory T of C and for each indecomposable T of T , any
non-isomorphism f : T → T factors through an object T of T which does not contain T as a
direct summand. Now let T be a cluster tilting subcategory. Then, according to [33], the results
on cluster categories from [12] generalize literally: For any indecomposable object T of T , there
is an indecomposable T ∗ , unique up to isomorphism, such that T ∗ is not isomorphic to T and
the additive subcategory T of C with indecomposables
ind(T ) = ind(T ) \ {T } ∪ {T ∗ }
T ∗ → B → T → ST ∗ and T → B → T ∗ → ST
connecting T ∗ and T such that B and B belong to both T and T . With these notations and
assumptions, we get the following connection between mod T and mod T .
Proposition. Let ST and ST ∗ be the simple tops of C(?, T ) and C(?, T ∗ ) respectively. Then there
is an equivalence of categories
∼
→ mod T /add ST ∗ .
mod T /add ST −
The proof, left to the reader, follows the proof of Theorem 4.2 of [13] and uses the previous
proposition.
B. Keller, I. Reiten / Advances in Mathematics 211 (2007) 123–151 129
An exact category in the sense of Quillen [42] is an additive category A endowed with a
distinguished class of sequences
i p
0→A−
→ B−
→ C → 0,
where i is a kernel of p and p a cokernel of i. We will state the axioms these sequences have to
satisfy using the terminology of [24]: The morphisms p are called deflations, the morphisms i in-
flations and the pairs (i, p) conflations. The given class of conflations has to satisfy the following
natural axioms:
p
Y Z
f f
Y Z
p
where p is a deflation.
(Ex2 ) Inflations are stable under cobase change, i.e. if i : X → Y is an inflation and g : X → X
a morphism, there is a coCartesian square
i
X Y
g g
X Y
i
where i is an inflation.
As shown in [35], these axioms are equivalent to Quillen’s and they imply that if A is small,
then there is a fully faithful functor from A into an abelian category A whose image is an
additive subcategory closed under extensions and such that a sequence of A is a conflation iff its
image is a short exact sequence of A . Conversely, one easily checks that an extension closed full
additive subcategory A of an abelian category A endowed with all conflations which become
exact sequences in A is always exact. The fundamental notions and constructions of homological
algebra, and in particular the construction of the derived category, naturally extend from abelian
to exact categories, cf. [37,41].
130 B. Keller, I. Reiten / Advances in Mathematics 211 (2007) 123–151
A Frobenius category is an exact category A which has enough injectives and enough pro-
jectives and where the class of projectives coincides with the class of injectives. In this case,
the stable category A obtained by dividing A by the ideal of morphisms factoring through a
projective-injective carries a canonical structure of triangulated category, cf. [29,31,39]. Its sus-
pension functor X → SX is constructed by choosing, for each object X of A, a conflation
X → I → SX,
where I is injective.
We refer to [5,6,30] and the references given there for the theory of Gorenstein algebras and
their modules in non-commutative and commutative algebra. In this section, we restate some
fundamental results of the theory in our setup.
Let A be a k-linear exact category with enough projectives and enough injectives. Assume
that A is Gorenstein, i.e. that the full subcategory P of the projectives is covariantly finite, the
full subcategory I of injectives is contravariantly finite and there is an integer d such that all
projectives are of injective dimension at most d and all injectives are of projective dimension at
most d. Let P (respectively I) denote the full subcategory of the projectives (respectively the
injectives) and Dfb (A) the full triangulated subcategory of the bounded derived category Db (A)
generated by the projectives (equivalently: the injectives). Call an object X of A projectively
(respectively injectively) Cohen–Macaulay if it satisfies
ExtiA (X, P ) = 0 respectively ExtiA (I, X) = 0
for all i > 0 and all projectives P (respectively injectives I ). Let CMP(A) (respectively CMI(A))
denote the full subcategories of A formed by the projectively (respectively injectively) Cohen–
Macaulay objects. Clearly, CMP(A) is an exact subcategory of A. Endowed with this exact
structure, CMP(A) is a Frobenius category and its subcategory of projective-injectives is P.
Dually, CMI(A) is a Frobenius category whose subcategory of projective-injectives is I. We
denote by CMP(A) and CMI(A) the stable categories associated with these Frobenius categories.
In analogy with a classical result on Frobenius categories (cf. [39,45]), we have the following
CMP(A) → A ← CMI(A)
A/(P) → CM(A)
B. Keller, I. Reiten / Advances in Mathematics 211 (2007) 123–151 131
admits a fully faithful left adjoint whose image is CMP(A) and the canonical functor
A/(I) → CM(A)
for all X, Y ∈ A.
We will need the following easily proved lemma (the statement of whose dual is left to the
reader).
Lemma.
(a) Suppose that all injectives of A are of projective dimension at most d. Then S d Y is injectively
Cohen–Macaulay for all Y ∈ A.
(b) Let n 1 and X, Y ∈ A. If S n Y is injectively Cohen–Macaulay, we have
∼ ∼
ExtnA (X, Y )/ ExtnA (I, Y ) −
→ A X, S n Y −→ CM(A) X, S n Y ,
Let C and T be as in Section 2. We have seen there that mod T is a Gorenstein category in
the sense of the preceding section.
Note that, by Sections 3.2 and 2.1, the theorem implies that for all T -modules X, Y , we have
a canonical isomorphism
∼
D Ext2mod T (Y, X) −
→ Ext1mod T (X, Y ),
3
α β
2 γ
4, αγ = βα = γβ = 0.
δ
0 → P1 → P3 → I2 → 0, 0 → P1 → P2 → I4 → 0
0 → P1 → I1 → I2 → 0, 0 → P2 → I1 ⊕ I4 → I2 → 0.
Thus the simple modules except S1 are injectively Cohen–Macaulay and the simple modules
except S2 are projectively Cohen–Macaulay. Therefore, we have
∼
Ext1A (Si , Sj ) −
→ D Ext2A (Sj , Si )
unless (i, j ) = (1, 2). Now dim Ext1A (Si , Sj ) equals the number of arrows from j to i and
Ext2A (Sj , Si ) is isomorphic to the space of minimal relations from i to j , cf. [9]. Thus for ex-
ample, the arrow α corresponds to the relation γβ = 0, and similarly for the arrows β and γ .
Note that no relation corresponds to the arrow δ. The stable category CM(mod A) is equivalent
to the stable category of the full subquiver with relations on 2, 3, 4. It is indeed 3-Calabi–Yau.
Consider the algebra A given by the quiver with relations:
2
α β
1 ε
3, αε = εβ = δε = εγ = 0, βα = γ δ.
δ γ
4
0 → P2 → I1 → I4 → 0, 0 → P4 → I1 → I2 → 0
0 → P4 → P3 → I2 → 0, 0 → P2 → P3 → I4 → 0.
B. Keller, I. Reiten / Advances in Mathematics 211 (2007) 123–151 133
Thus the simples S1 and S3 are both projectively and injectively Cohen–Macaulay and whenever
i, j are connected by an arrow, we have
∼
Ext1A (Si , Sj ) −
→ D Ext2A (Sj , Si ).
Since there are neither arrows nor relations between 2 and 4, we obtain a perfect correspondence
between arrows and relations: the outer arrows correspond to zero relations and the inner arrow
to the commutativity relation.
Consider the algebra A given by the quiver
2 5
1 4 6
subject to the following relations: each path of length two containing one of the outer arrows
vanishes; each of the three rhombi containing one of the inner arrows is commutative. As shown
in [26], this algebra is isomorphic to the stable endomorphism algebra of a maximal rigid module
over the preprojective algebra of type A4 . It is self-injective and thus Gorenstein. According to
the theorem, it is stably 3-Calabi–Yau, a fact which was first proved in [25] using the fact that
the projectives over A form a quadrangulated category. Note that the isomorphisms
∼
Ext1A (Si , Sj ) −
→ D Ext2A (Sj , Si )
translate into a perfect correspondence between arrows and relations: each outer arrow corre-
sponds to a zero relation and each inner arrow to a commutativity relation.
Denote by mod T the quotient of mod T by the ideal of morphisms factoring through a pro-
jective and by mod T the quotient by the ideal of morphisms factoring through an injective. It
follows from Proposition 2.1 that the functor F induces equivalences
∼ ∼
C/(T , ST ) −
→ mod T and C/(ST , ΣT ) −
→ mod T .
F S −1 M → F T1M → F T0M → F M → 0
in mod T and its middle two terms are projective. Using this sequence we see that Ext1T (F M, F L)
is isomorphic to the middle cohomology of the complex
Hom F T0M , F L → Hom F T1M , F L → Hom F S −1 M, F L . (1)
V → ΣTV0 → ΣTV1 → SV
By applying this to the three terms in the sequence (1) and using the vanishing of Ext1 between
objects of T , we obtain that E = Ext1 (F M, F L) is the intersection of the images of the maps
C S −1 M, STL1 → C S −1 M, L and C T1M , L → C S −1 M, L .
This proves the left isomorphism. The right one follows since τ is an equivalence. 2
Now let X, Y be T -modules and S the stable Cohen–Macaulay category of mod T . We will
construct a canonical isomorphism
∼
DS Y, S 2 X −→ S(X, SY ).
B. Keller, I. Reiten / Advances in Mathematics 211 (2007) 123–151 135
Since the injectives of mod T are of projective dimension at most 1 and the projectives of injec-
tive dimension at most 1, we have
HomS Y, S 2 X = Ext2mod T (Y, X)
and
HomS (X, SY ) = cok Ext1mod T (I, Y ) → Ext1mod T (X, Y ) ,
whose second and third term are injective. This is sufficient to obtain that Ext2 (F L, F M) is
isomorphic to the middle homology of the complex
1
Hom F L, F ΣTM → Hom(F L, F SM) → Hom F L, F SΣTM0 .
Ext2 (F L, F M) = ker Hom(F L, τ F M) → Hom(F L, τ I ) .
∼
D Ext2 (F L, F M) −
→ cok Ext1 (I, F L) → Ext1 (F L, F M)
and thus
∼
D Ext2 (Y, X) −
→ cok Ext1 (I, Y ) → Ext1 (X, Y ) ,
as claimed.
136 B. Keller, I. Reiten / Advances in Mathematics 211 (2007) 123–151
Let E be a k-linear Frobenius category with split idempotents. Suppose that its stable category
C = E has finite-dimensional Hom-spaces and is Calabi–Yau of CY-dimension 2. This situation
occurs in the following examples:
(1) E is the category of finite-dimensional modules over the preprojective algebra of a Dynkin
quiver as investigated in [27].
(2) E is the category of Cohen–Macaulay modules over a commutative complete local Goren-
stein isolated singularity of dimension 3.
(3) E is a Frobenius category whose stable category is triangle equivalent to the cluster category
associated with an Ext-finite hereditary category. Such Frobenius categories always exist by
[38, Section 9.9].
(4) E is a Frobenius category whose stable category is triangle equivalent to the bounded derived
category of coherent sheaves on a Calabi–Yau surface (i.e. a K3-surface). For example, one
obtains such a Frobenius category by taking the full subcategory of the exact category of left
bounded complexes of injective quasi-coherent sheaves whose homology is bounded and
coherent.
Let T ⊂ C be a cluster tilting subcategory and M ⊂ E the preimage of T under the pro-
jection functor. In particular, M contains the subcategory P of the projective-injective objects
in M. Note that T equals the quotient M of M by the ideal of morphisms factoring through a
projective-injective.
We know from Section 2 that mod M is abelian. In general, we cannot expect the category
mod M of finitely presented M-modules to be abelian (i.e. M to have weak kernels). How-
ever, the category Mod M of all right M-modules is of course abelian. Recall that the perfect
derived category per(M) is the full triangulated subcategory of the derived category of Mod M
generated by the finitely generated projective M-modules. We identify Mod M with the full
subcategory of Mod M formed by the modules vanishing on P. The following proposition is
based on the methods of [3].
Proposition.
0 → M 1 → M0 → X → 0
0 → K → M 1 → M0 → 0
∼
Z−
→ Ext1E (?, K)|M.
B. Keller, I. Reiten / Advances in Mathematics 211 (2007) 123–151 137
(c) Let Z be a finitely presented M-module. Then Z considered as an M-module lies in per(M)
and we have a canonical isomorphism
∼
D per(M)(Z, ?) −
→ per(M) ?, Z[3] .
M1 → M0 → X → SM1
p∧
M1∧ −−→ M0∧ → Z → 0
and vanishes on P. Then each morphism P → M0 with P ∈ P lifts along p. Now let q : P0 →
M0 be a deflation with projective P0 . Then we have q = pq for some q . Clearly, the morphism
M1 ⊕ P0 → M0
with components p and q is a deflation and M1 → M0 is a retract of this deflation: The retraction
is given by [1, q ] and the identity of M0 . Since E has split idempotents, it follows that M1 → M0
is a deflation. The last assertion follows from the vanishing of Ext1E (M, M1 ) for all M in M.
(c) Clearly, each representable M-module is finitely presented as an M-module. Since a
cokernel of a morphism between finitely presented modules is finitely presented, the module Z
considered as an M-module is finitely presented. So we can choose a conflation
0 → K → M 1 → M0 → 0
0 → M3 → M2 → K → 0
0 → M3 → M 2 → M 1 → M 0 → 0
0 → M3 → M2 → K → 0
which sends f to the map g → φ(f g). Clearly this is a morphism of cohomological functors
defined on the triangulated category per(M). To check that it is an isomorphism, it suffices to
check that its evaluation at all shifts S i M ∧ , i ∈ Z, of representable functors M ∧ is an isomor-
phism. Now indeed, we have
per(M) S i M ∧ , P [3] = per(M) S i M ∧ , Z[3] = Ext3−i M ∧ , Z .
Ext1E (M, K)
for i = 3. On the other hand, to compute per(M)(P , S i M ∧ ), we have to compute the homology
of the complex
and
using that Ext1E (M0 , M) = 0 = Ext1E (M2 , M), we get that the homology is 0 except at E(M3 , M),
where it is Ext1E (K, M0 ). Now we know that the canonical linear form on Ext2E (K, K) yields a
canonical isomorphism
∼
Ext1E (M, K) −
→ D Ext1E (K, M)
B. Keller, I. Reiten / Advances in Mathematics 211 (2007) 123–151 139
The following corollary generalizes Proposition 6.2 of [27]. The fact that mod M is abelian
and of global dimension 3 is a special case of the results of [32].
Corollary. Suppose that E is abelian. Then mod M is abelian of global dimension at most 3 and
for each X ∈ mod M and each Y ∈ mod M, we have canonical isomorphisms
∼
→ D Ext3−i
Extimod M (X, Y ) − mod M
(Y, X), i ∈ Z.
M1∧ → M0∧ → X → 0
0 → K → M1 → M0
0 → M3 → M2 → K → 0
using part (a) of the proposition above. By splicing the two, we obtain a complex
0 → M3 → M2 → M1 → M0 → 0,
whose image under the Yoneda functor is a projective resolution of length at most 3 of X. This
shows that M admits weak kernels (hence mod M is abelian) and that mod M is of global
dimension at most 3. Thus the perfect derived category coincides with the bounded derived cat-
egory of mod M. Now the claim about the extension groups is obvious from the proposition.
For the last assertion, we choose X to be a non-zero M-module. Then Ext3mod M (X, X) is non-
zero. 2
Let k be a field and C a triangulated k-linear category with split idempotents and suspension
functor S. We suppose that all Hom-spaces of C are finite-dimensional and that C admits a Serre
functor Σ, cf. [44]. Let d 1 be an integer. We suppose that C is Calabi–Yau of CY-dimension d,
i.e. there is an isomorphism of triangle functors
∼
Sd −
→ Σ.
140 B. Keller, I. Reiten / Advances in Mathematics 211 (2007) 123–151
Note that (a)–(c) are self-dual and so is (d) (by the Calabi–Yau property). For d = 1, we have
T = C by condition (d). This definition is slightly different from that in [47] but most probably
equivalent in the context of [loc.cit.]. As in Section 2, one proves that mod T , the category of
finitely presented right T -modules, is abelian. Let
F : C → mod T
be the functor which sends X to the restriction of C(?, X) to T . For classes U , V of objects of T ,
we denote by U ∗ V the full subcategory of all objects X of C appearing in a triangle
U → X → V → SU.
Lemma. Suppose that d 2. For each finitely presented module M ∈ mod T , there is a triangle
T0 → T1 → X → ST0
U/(ST ) → mod T ,
Proof. Since T has split idempotents, the functor F induces an equivalence from T to the cat-
egory of projectives of mod T . Now let P1 → P0 → M → 0 be a projective presentation of M.
Choose a morphism T1 → T0 of T whose image under F is P1 → P0 . Define X by the triangle
T1 → T0 → X → ST1 .
One can use Proposition 5.6 below (cf. also [20,47,48]) to produce the following examples:
Consider the algebra A given by the quiver
1 2
3 4
with the relations given by all paths of length 2. Then A is 3-cluster tilted of type A4 . It is self-
injective hence Gorenstein. It is not hard to check that its stable category is 4-Calabi–Yau. More
generally, a finite-dimensional algebra of radical square 0 whose quiver is an oriented cycle with
d + 1 vertices is d-cluster-tilted of type Ad+1 , self-injective and stably (d + 1)-Calabi–Yau.
Let A be the algebra given by the quiver
◦ ◦
◦ ◦ ◦
◦ ◦
subject to the relations αβ = 0 for all composable arrows α, β which point in different directions
(the composition of the long skew arrows in non-zero). Then A is 3-cluster tilted of type A7 .
One can show that it is Gorenstein of dimension 1 and that its stable Cohen–Macaulay category
is 3-Calabi–Yau.
5.3. A counterexample
The following example due to Osamu Iyama shows that part (d) of Proposition 2.1 does not
generalize from the 2-dimensional to the d-dimensional case: It is an example of a 3-cluster
tilting subcategory of a 3-Calabi–Yau triangulated category which is not Gorenstein.
Let k be a field of characteristic = 3 and ω a primitive third root of 1. Let S = kJt, x, y, zK
and let the generator g of G = Z/3Z act on S by
1 2
subject to all commutativity relations, where, between any two vertices, the arrows pointing
counterclockwise are labelled t and x and the arrows pointing clockwise y and z. The vertex 0
corresponds to the indecomposable projective S0 , which vanishes in the stable category C. There-
fore the full subcategory ind(T ) of C is given by the quiver
1 2
subject to the relations given by all paths of length 2. It is easy to see that the free modules
T (?, Si ), i = 1, 2, are of infinite injective dimension. Thus, T is not Gorenstein.
Let d be an integer 1. Let E be a k-linear Frobenius category with split idempotents. Sup-
pose that its stable category C = E has finite-dimensional Hom-spaces and is Calabi–Yau of
CY-dimension d. Let T ⊂ C be a d-cluster tilting subcategory and M ⊂ E the preimage of
T under the projection functor. In particular, M contains the subcategory P of the projective-
injective objects in E. Note that T equals the quotient M of M by the ideal of morphisms
factoring through a projective-injective.
We call a complex M-acyclic if it is acyclic as a complex over E (i.e. obtained by splicing
conflations of E) and its image under the functor X → E(?, X)|M is exact.
Theorem.
0 → Md−1 → Md−2 → · · · → M0 → X → 0
with all Mi in M.
(b) For each finitely presented M-module Z, there is an M-acyclic complex
0 → Md+1 → Md → · · · → M1 → M0 → 0
∼ ∼
Z−
→ cok E(?, M1 ) → E(?, M0 ) −→ Ext1E (?, Z1 ),
(c) Let Z be a finitely presented M-module. Then Z considered as an M-module lies in per(M)
and we have a canonical isomorphism
∼
per(M) ?, Z[d + 1] −
→ D per(M)(Z, ?).
0 → Z0 → M0 → X → 0.
0 → Zi → Mi → Zi−1 → 0
0 → Zd−1 → Md−2 → · · · → M0 → X → 0.
0 → Z1 → M1 → M0 → 0
∼
Since Z −
→ Ext1E (?, Z1 ), we have to compute
cok Ext1E (Md , Z1 ) → Ext1E (Md+1 , Z1 ) .
Put X = SZ1 . We have Ext1E (?, Z1 ) = E(?, SZ1 ) = E(?, X) and so we have to compute
cok E(Md , X) → E(Md+1 , X) .
We now apply part (c) of Proposition 5.5 to the image in E of the complex
0 → Md+1 → Md → · · · → Md
144 B. Keller, I. Reiten / Advances in Mathematics 211 (2007) 123–151
vanishes if i = d + 1 and equals E(M, X) for i = d + 1. On the other hand, we have to compute
the group
per(M) P , M ∧ [i] .
But Ext1E (Zi−1 , M) = E(Zi−2 , SM) vanishes because Zi−2 is an iterate extension of objects in
S −(i−2) T , . . . , S −1 T , T . 2
We work with the notations and assumptions of Section 5.1 and assume moreover that d 2.
Let Y be an object of C. Let T0 → Y be a right T -approximation of Y . We define an object Z0
by the triangle
Z0 → T0 → Y → SZ0 .
Z1 → T1 → Z0 → SZ1 .
Zi → Ti → Zi−1 → SZi
B. Keller, I. Reiten / Advances in Mathematics 211 (2007) 123–151 145
for 1 < i d − 2. By the proposition below, the object Zd−2 belongs to T . We put Td−1 = Zd−2 .
The triangles
Zi → Ti → Zi−1 → SZi
yield morphisms
Put T -mod = mod(T op ). Let F : C → mod T and G : C → T -mod be the functors which take
an object X of C to C(?, X)|T respectively C(X, ?)|T . The following proposition is related to
Theorem 2.1 of [33].
Td−1 → Td−2 → · · · → T1 → T0 → 0
0 → Td−1 → Td−2 → · · · → T1 → T0
Proof. (a) From the construction of the Zi , we get that, for T ∈ T , we have
Ext1 (T , Zi ) = 0
∼ ∼ ∼
→ Extj −1 (T , Zi−1 ) −
Extj (T , Zi ) − → ··· −
→ Ext1 (T , Zi−j +1 ) = 0
for all 2 j i + 1. Thus we have Extj (T , Zd−2 ) = 0 for 1 j d − 1 and Zd−1 indeed
belongs to T .
(b) This follows readily from the construction.
(c) We consider the triangle
For T ∈ T , the long exact sequence obtained by applying C(?, T ) combined with the vanishing
of C(S −1 Td−2 , T ) shows that we have the isomorphism
∼
→ C S −1 Zd−3 , T .
cok C(Td−2 , T ) → C(Td−1 , T ) −
the sequence
C(Td−3 , T ) → C(Zd−3 , T )
it suffices to show that C(S −1 Zd−4 , T ) vanishes. This is clear because the object Zd−4 is an
iterated extension of objects in
S −(d−2) T , S −(d−3) T , . . . , S −1 T , T . 2
B. Keller, I. Reiten / Advances in Mathematics 211 (2007) 123–151 147
has been considered in [38,47,48]. Let π : D → C be the projection functor. Let T ⊂ D be a tilting
subcategory, i.e. a full subcategory whose objects form a set of generators for the triangulated
category D and such that
D T , Si T = 0
for all T , T ∈ T and all i = 0. It follows that there is a triangle equivalence, cf. [36],
Db (mod T ) → Db (H)
which takes the projective module T ∧ = T (?, T ) in mod T to T , for each T ∈ T . We assume
that T is locally bounded, i.e. for each indecomposable T , there are only finitely many indecom-
posables T in T such that T (T , T ) = 0 or T (T , T ) = 0. We denote the Nakayama functor of
mod T by ν. Recall from Section 3.7 of [24] that it is the unique functor endowed with isomor-
phisms
D Hom(P , M) = Hom(M, νP )
for all finitely generated projectives P and all finitely presented T -modules M. It follows that
∼
if we view a projective P as an object of Db (mod T ), we have ΣP − → νP . The category of
finitely generated projectives is functorially finite in D (mod T ): Indeed, if M is an object of
b
Proposition. Assume that the objects of T have their homology concentrated in degrees i with
−(d − 2) i 0. Then π(T ) is a d-cluster tilting subcategory in C and for all objects T , T
of T , we have a functorial isomorphism compatible with compositions
π(T ) π(T ), π(T ) = T (T , T ) ⊕ ExtdT νT ∧ , T ∧ .
Note that the proposition does not produce all d-cluster tilting subcategories of C. For exam-
ple, as one checks easily, the non-connected algebra k × k is 3-cluster tilted of type A2 .
Proof. Put F = Σ −1 S d . For n ∈ Z, denote by Dn the full subcategory of D formed by the
objects X with H i (X) = 0 for i > n and by Dn the full subcategory formed by the Y with
148 B. Keller, I. Reiten / Advances in Mathematics 211 (2007) 123–151
H i (Y ) = 0 for i < n. Then D0 and D0 are the aisles of the natural t-structure on D. Thus
the subcategory D1 is the right perpendicular subcategory of D0 and we have in particular
Hom(X, Y ) = 0 if X ∈ D0 and Y ∈ D1 . Since H is hereditary, we also have Hom(X, Y ) = 0 if
X ∈ D0 and Y ∈ D−2 . We have F D0 ⊂ D−(d−1) . Indeed, for X ∈ D0 and Y ∈ D−(d−2) ,
we have
D(F X, Y ) = D Σ −1 S d X, Y = DD Y, S d X = 0
Since we have
Since we have
for 0 < i < d. By re-examining the above computation, we also find that D(T , F n T ) vanishes
for each integer n = 0, 1 and that we have
D(T , F T ) = D T , Σ −1 S d T = D ΣT , S d T = per(T ) ΣT ∧ , S d T ∧ = ExtdT νT ∧ , T ∧ .
for all 0 < i < d and all T ∈ T , then π(X) belongs to π(T ). We may assume that X is indecom-
posable and belongs to
D0 ∩ D−(d−1) .
Let U0 , U0 be the aisles obtained as the images of the natural aisles in Db (mod T ) under the
∼
triangle equivalence Db (mod T ) −→ D associated with T . We claim that we have
D0 ⊂ U(d−1) .
Indeed, for Y ∈ D0 and i d, we have S −i T ∈ D2 so that D(S −i T , Y ) = 0 and this implies
the claim. Since X lies in D0 it follows that X ∈ U(d−1) . Now the assumption on X yields that
D S −i T , X = 0
D0 ⊂ U0 .
Indeed, for Y ∈ D0 and i > 0, we have D(S i T , Y ) = 0 since S i T ∈ D<0 . This implies the claim.
Since X lies in D−(d−1) , it follows that X lies in U−(d−1) . Finally, for T ∈ T and 0 < i < d,
we have
0 = D T , S i F X = D T , S i−d ΣX = DD S i−d X, T = DD X, S d−i T .
Thus we have D(X, S i T ) = 0 for 0 < i < d. But since X lies in D−(d−1) and T in D0 we
also have D(X, S i T ) = 0 for all i d. Thus X is left orthogonal to U−1 . Therefore, the ob-
ject Y ∈ Db (mod T ) corresponding to X via the triangle equivalence associated with T is in
b (mod T ) and left orthogonal to D b
D0 −1 (mod T ). Since mod T is of finite global dimension,
this implies that Y is a projective T -module. So X lies in T . 2
Acknowledgments
This article has grown out of discussions which the authors had during the meeting ‘Interac-
tions between non-commutative algebra and algebraic geometry’ held at the BIRS in Banff in
September 2005. We are grateful to the organizers of the meeting and in particular to Colin In-
galls and James Zhang. We thank Osamu Iyama for pointing out a mistake in a previous version
of this article and for allowing us to include his Example 5.3. We are grateful to Xiao-Wu Chen
for pointing out missing conditions in Section 3.2.
References
[1] Ibrahim Assem, Thomas Brüstle, Ralf Schiffler, Gordana Todorov, Cluster categories and duplicated algebras,
arXiv: math.RT/0509501.
[2] Ibrahim Assem, Thomas Brüstle, Ralf Schiffler, Cluster-tilted algebras as trivial extensions, arXiv: math.RT/
0601537.
[3] Maurice Auslander, Coherent functors, in: Proc. Conf. Categorical Algebra, La Jolla, CA, 1965, Springer, New
York, 1966, pp. 189–231.
150 B. Keller, I. Reiten / Advances in Mathematics 211 (2007) 123–151
[4] Maurice Auslander, Functors and morphisms determined by objects, in: Representation Theory of Algebras, Proc.
Conf., Temple Univ., Philadelphia, PA, 1976, in: Lecture Notes in Pure and Appl. Math., vol. 37, Dekker, New
York, 1978, pp. 1–244.
[5] Maurice Auslander, Idun Reiten, Applications of contravariantly finite subcategories, Adv. Math. 86 (1) (1991)
111–152.
[6] Maurice Auslander, Idun Reiten, Cohen–Macaulay and Gorenstein Artin algebras, in: Representation Theory of
Finite Groups and Finite-Dimensional Algebras, Bielefeld, 1991, in: Progr. Math., vol. 95, Birkhäuser, Basel, 1991,
pp. 221–245.
[7] Maurice Auslander, Idun Reiten, DTr-periodic modules and functors, in: Representation Theory of Algebras, Co-
coyoc, 1994, in: CMS Conf. Proc., vol. 18, Amer. Math. Soc., Providence, RI, 1996, pp. 39–50.
[8] Arkady Berenstein, Sergey Fomin, Andrei Zelevinsky, Cluster algebras. III. Upper bounds and double Bruhat cells,
Duke Math. J. 126 (1) (2005) 1–52.
[9] Klaus Bongartz, Algebras and quadratic forms, J. London Math. Soc. (2) 28 (3) (1983) 461–469.
[10] Aslak Bakke Buan, Idun Reiten, Cluster algebras associated with extended Dynkin quivers, Int. Math. Res.
Not. 12804 (2006) 1–10.
[11] Aslak Bakke Buan, Idun Reiten, From tilted to cluster-tilted algebras of Dynkin type, arXiv: math.RT/0510445.
[12] Aslak Bakke Buan, Robert J. Marsh, Markus Reineke, Idun Reiten, Gordana Todorov, Tilting theory and cluster
combinatorics, Adv. Math. 204 (2) (2006) 572–618.
[13] Aslak Bakke Buan, Robert J. Marsh, Idun Reiten, Cluster-tilted algebras, Trans. Amer. Math. Soc., in press.
[14] Aslak Bakke Buan, Robert J. Marsh, Idun Reiten, Cluster-tilted algebras of finite representation type, arXiv:
math.RT/0509198, J. Algebra, in press.
[15] Aslak Bakke Buan, Robert J. Marsh, Idun Reiten, Gordana Todorov, Clusters and seeds in acyclic cluster algebras,
arXiv: math.RT/0510359, Proc. Amer. Math. Soc., in press.
[16] Aslak Bakke Buan, Idun Reiten, Ahmet Seven, Tame concealed algebras and cluster quivers of minimal infinite
type, arXiv: math.RT/0512137.
[17] Philippe Caldero, Bernhard Keller, From triangulated categories to cluster algebras, arXiv: math.RT/0506018.
[18] Philippe Caldero, Bernhard Keller, From triangulated categories to cluster algebras II, arXiv: math.RT/0510251.
[19] Philippe Caldero, Frédéric Chapoton, Ralf Schiffler, Quivers with relations arising from clusters (An case), arXiv:
math.RT/0401316.
[20] Sergey Fomin, Nathan Reading, Generalized cluster complexes and Coxeter combinatorics, arXiv: math.CO/
0505085.
[21] Sergey Fomin, Andrei Zelevinsky, Cluster algebras. I. Foundations, J. Amer. Math. Soc. 15 (2) (2002) 497–529
(electronic).
[22] Sergey Fomin, Andrei Zelevinsky, Cluster algebras. II. Finite type classification, Invent. Math. 154 (1) (2003) 63–
121.
[23] Sergey Fomin, Andrei Zelevinsky, Cluster algebras: Notes for the CDM-03 conference, in: Current Developments
in Mathematics, 2003, Int. Press, Somerville, MA, 2003, pp. 1–34.
[24] P. Gabriel, A.V. Roiter, Representations of Finite-Dimensional Algebras, Encyclopaedia Math. Sci., vol. 73,
Springer, 1992.
[25] Christof Geiß, Bernhard Keller, Quadrangulated categories, Oberwolfach talk, February 2005.
[26] Christof Geiß, Bernard Leclerc, Jan Schröer, Auslander algebras and initial seeds for cluster algebras, arXiv:
math.RT/0506405.
[27] Christof Geiß, Bernard Leclerc, Jan Schröer, Rigid modules over preprojective algebras, arXiv: math.RT/0503324,
Invent. Math., in press.
[28] Christof Geiß, Bernard Leclerc, Jan Schröer, Semicanonical bases and preprojective algebras II: A multiplication
formula, arXiv: math.RT/0509483.
[29] Dieter Happel, On the derived category of a finite-dimensional algebra, Comment. Math. Helv. 62 (3) (1987) 339–
389.
[30] Dieter Happel, On Gorenstein algebras, in: Representation Theory of Finite Groups and Finite-Dimensional Alge-
bras, Bielefeld, 1991, in: Progr. Math., vol. 95, Birkhäuser, Basel, 1991, pp. 389–404.
[31] Alex Heller, Stable homotopy categories, Bull. Amer. Math. Soc. 74 (1968) 28–63.
[32] Osamu Iyama, Higher dimensional Auslander–Reiten theory on maximal orthogonal subcategories, arXiv: math.RT/
0411631, Adv. Math., in press.
[33] Osamu Iyama, Maximal orthogonal subcategories of triangulated categories satisfying Serre duality, Oberwolfach
Report 6, 2005.
[34] Osamu Iyama, Idun Reiten, Fomin–Zelevinsky mutation and tilting modules over Calabi–Yau algebras, arXiv:
math.RT/0605136.
B. Keller, I. Reiten / Advances in Mathematics 211 (2007) 123–151 151
[35] Bernhard Keller, Chain complexes and stable categories, Manuscripta Math. 67 (4) (1990) 379–417.
[36] Bernhard Keller, Derived categories and universal problems, Comm. Algebra 19 (1991) 699–747.
[37] Bernhard Keller, Derived categories and their uses, in: Handbook of Algebra, vol. 1, North-Holland, Amsterdam,
1996, pp. 671–701.
[38] Bernhard Keller, On triangulated orbit categories, Doc. Math. 10 (2005) 551–581.
[39] Bernhard Keller, Dieter Vossieck, Sous les catégories dérivées, C. R. Acad. Sci. Paris Sér. I Math. 305 (6) (1987)
225–228.
[40] Henning Krause, The stable derived category of a Noetherian scheme, Compos. Math. 141 (5) (2005) 1128–1162.
[41] Amnon Neeman, The derived category of an exact category, J. Algebra 135 (1990) 388–394.
[42] Daniel Quillen, Higher algebraic K-theory. I, in: Algebraic K-Theory, I: Higher K-Theories, Proc. Conf., Battelle
Memorial Inst., Seattle, WA, 1972, in: Lecture Notes in Math., vol. 341, Springer, 1973, pp. 85–147.
[43] Idun Reiten, Helmut Lenzing, Hereditary noetherian categories of positive Euler characteristic, Math. Z., in press.
[44] Idun Reiten, Michel Van den Bergh, Noetherian hereditary abelian categories satisfying Serre duality, J. Amer.
Math. Soc. 15 (2) (2002) 295–366 (electronic).
[45] Jeremy Rickard, Derived categories and stable equivalence, J. Pure Appl. Algebra 61 (1989) 303–317.
[46] Claus Michael Ringel, Some remarks concerning tilting modules and tilted algebras. Origin. Relevance. Future, in:
Lidia Angeleri-Hügel, Dieter Happel, Henning Krause (Eds.), Handbook of Tilting Theory.
[47] Hugh Thomas, Defining an m-cluster category, arXiv: math.RT/0607173.
[48] Anette Wraalsen, Ph.D. Thesis, in preparation.
[49] Bin Zhu, Tilting theory in cluster categories of hereditary abelian categories, preprint, 2005.