DUALITY CATEGORIES∗

RAMZI KSOURI∗∗

Abstract. We define the notion of duality categories as gener-

alization of duality groups. Two examples are treated. The first
arXiv:1405.3221v2 [math.AT] 4 Jul 2015

is the Serre duality in the categories of strict polynomial functors.

The second concerns finite complexes. We show in particular that
finite Tits buildings are duality categories.

Contents
1. Introduction 1
2. Preliminaries 4
3. Yoneda bifunctor and epireflective subcategory 7
4. Functor of duality 14
5. Duality functors 18
6. Illustration: strict polynomial functors 20
7. Tits buildings are duality categories 23
References 35

1. Introduction
A group G is a duality group in the sense of Beiri and Eck-
mann [BE73] if there exist an integer n, a right G-module D and a
natural isomorphism
H k (G, −) ∼
= Hn−k (G, D ⊗ −).
We generalize this definition of duality group by replacing the group G
with a category C. A C-module is then a functor from C to the category
of abelian groups Ab.
Definition 1.1. A small category C is said a duality category if there
exists a functor D : C op → Ab, an integer n and a natural isomorphism
Exti (Z, −) ∼
C−mod = TorC−mod (D, −)
n−i

where Z is the constant functor.

∗
The final publication is available at Springer via
http://dx.doi.org/10.1007/s10485-015-9397-8.
∗∗
Laboratoire de Mathématiques: Algèbre, Topologie, Arithmétique et Struc-
tures ordonnées "LATAO", Département de Mathématiques, Faculté des Sciences
de Tunis, Le Campus Universitaire TN-2092, Tunis, Tunisie.
1
2 RAMZI KSOURI

In this article, we outline a theory of duality. In the case where the

small category C is a finite simplicial complex, we obtain the following
local condition:
Theorem 1.2. Let P the partial ordered set of simplices of a finite
simplicial complex. If there exists an integer n such that for any simplex
x of P
e i (link ) = 0 for i 6= n − dim(x) − 1
H x
e ∗ denotes the reduced cohomology with integer coefficients)
(where H
then P is a duality category.
This theorem allows us to recover the classical Poincaré duality for
triangulated manifolds. It also leads to the following result.
Corollary 1.3. Spherical Tits buildings are duality categories.
More generally, if k denotes a commutative unitary ring and k−mod
the category of left k-modules, we say that a functor F : C → k−mod
is a duality functor if there exist an integer n, a functor D n (F ) : C op →
k−mod and a natural isomorphism
(1) ExtiC−mod (F, −) ∼= TorC−mod (D n (F ) , −) .
n−i

If k is a field, the isomorphism (1) is equivalent to the following

 
Exti
C−mod (F, −)′ ∼
= Extn−i
C−mod −, D n (F )′
′
where D (F ) denotes the postcomposition of D n (F ) by the duality op-
n

eration V 7→ V ′ := Homk (V, k). The latter isomorphism is reminiscent

of the classical Serre duality for projective varieties.
Idun Reiten and Michel Van Den Bergh in [RV02] generalized the
concept of Serre duality in the following way. A k-linear category A
satisfies a Serre duality if it has a Serre functor, i.e. an endofunctor
S : A → A such that we have a natural isomorphism
HomA (a, b)′ ∼= HomA (b, Sa).
The duality presented in this paper makes explicit a Serre duality for
the category Dperf (C−mod) of perfect complexes of C−mod. Let us be
clear what this is about. Let D the functor:
D : (C−mod)op → C op −mod
F 7→ DF
defined for an object x of C by
DF (x) = HomC−mod (F, PxC )
where PxC = k[HomC (x, −)] is a standard projective of C−mod ( the
notation k[S] designates the free k-module generated by S). Let
RD : D b (C−mod)op → D b (C op −mod)
the derived functor of D. We prove the following result:
 DUALITY CATEGORIES 3

Proposition 1.4. The functor RD(−)′ is a Serre Duality for

Dperf (C−mod).
We illustrate the concept of duality functor with an example in the
category Rep Γdk of strict polynomial functors over a finite field k of
characteristic p. Let s, r be integers and s = (s1 , · · · , sk ) a partition of
s. The strict polynomial functor Γs(r) (resp. S s(r) ) of degree d := spr
is the precomposition of Γs (resp. S s ) by the Frobenius twist I(r) .
Recall that Γs (resp. S s ) is the tensor product Γs1 ⊗ · · · ⊗ Γsk (resp.
S s1 ⊗ · · · ⊗ S sk ) where Γ (resp. S) is the functor of divided (resp.
symmetric) powers. We use the properties of the Ringel duality studied
in [Touzé1] to show the following result:
Proposition 1.5. The strict polynomial functor Γs(r) is a duality func-
tor and we have a natural isomorphism
 ′  
∼ 2s(pr −1)−i
ExtiRep Γd Γs(r) , − = ExtRep Γd −, S s(r) .
k k

Here is a summary of the different sections.

Section 2: This section is a collection of reminders and notations
used in the article.
Section 3: In the case of modules over a ring R, the structure of
right R-module on the dual M ′ = HomR (M, R) of a left R-module M
is inherited from the structure of R-bimodule of R. An equivalent of
the bimodule R is the bifunctor
C : C op × C → k−mod
(x, y) 7→ k[HomC (x, y)]
Our duality theory is based on the following property
(∗) HomC−mod (C(y, −), C(x, −)) ∼
= C(x, y)
provided by the Yoneda lemma. We call a bifunctor with the prop-
erty (∗) a Yoneda bifunctor. We show that there is a 1 − 1 corre-
spondance between equivalence classes of epireflective and coreflective
subcategories of C−mod and isomorphism classes of Yoneda bifunctors
for C.
Section 4: We show that a Yoneda bifunctor T for a category C
provides a functor of duality D and we study its first properties. We
illustrate this point of view by showing (proposition 4.13) that if T is
Yoneda bifunctor for a k-linear category C then the functor RD(−)′ is
a Serre duality of Dperf (CT −mod).
Section 5: A functor F of C−mod is a duality functor if there exists
an integer n such that one has an isomorphism
RD (F ) ∼
= Rn D (F ) [n]
4 RAMZI KSOURI

in D + (C−mod). In this section, we give several equivalent conditions

to the latter.
Section 6: We illustrate the notions of duality presented in this ar-
ticle by the category Rep Γdk of strict polynomial functors of degree d
over a finite field k. We equip this category with a Serre duality func-
tor RD(−)′ and we prove that the strict polynomial functor Γs(r) is a
duality functor.
Section 7: The rest of this article is devoted to the notion of duality
category. A category is a duality category if the constant functor Z is
a duality functor. We isolate a sufficient local criterion for a category
to be a duality category. The theorem 7.12 presents this criterion in
the case where the category is the partially ordered set of simplices of
a simplicial complex. We thus recover the classical Poincaré duality
for manifolds. It also follows that spherical Tits buildings are duality
categories.
Acknowledgments. I thank Gaël Collinet who drawed my attention
on the problem of generalizing the theory of duality groups and supplied
a constant help until the writing of this article. I also wish to thank
Antoine Touzé for discussions on Ringel duality.

2. Preliminaries
In this section, the notation C designates an essentially small cat-
egory, the notation k designates a unitary commutative ring and the
notation k−mod designates the category of left k-modules. The cate-
gory Z−mod is denoted Ab.
2.1. The category C−mod.
Definitions 2.1.
• A left (resp. right) C-module is a covariant functor from C (resp.
C op ) to the category k−mod.
• We denote I the category 0 → 1. Let F and G be two left
C-modules. A natural transformation between F and G is a
functor
T : C × I → k−mod
such that T (c, 0) = F (c) and T (c, 1) = G(c).
• We denote C−mod (resp. C op −mod) the category whose ob-
jects are left (resp. right) C-modules and whose morphisms are
natural transformations.
Properties 2.2.
• For two objects F and G of C−mod, the set HomC−mod (F, G) of
natural transformations between F and G is naturally provided
with a k-module structure.
 DUALITY CATEGORIES 5

• The bifunctor
HomC−mod (−, −) : (C−mod)op × C−mod → k−mod
commutes with limits with respect to each variable (limit ob-
jects in (C−mod)op identify with colimit objects in C−mod).
• Let PaC denotes the functor k[HomC (a, −)]. The linear analogue
of the Yoneda Lemma ([Maclane1, III.2]) provides the following
natural isomorphism
(2) HomC−mod (PaC , F ) ∼
= F (a).
The isomorphism is given by the following map
HomC−mod (PaC , F ) → F (a)
.
N 7→ Na (1a )
The inverse is defined by associating to λ ∈ F (a) the natural
transformation N : PaC → F whose component Nx is given by
Nx (f ) = F (f )λ for f ∈ HomC (a, x).
• The categories C−mod and C op −mod are abelian. This follows
from the definition of an abelian category and the fact that
kernels, cokernels and more generally limits and colimits are
calculated pointwise (see [Weibel, 1.6.4 and A.4]).
• For any object a of C, the functors called standard projectives
PaC := k[HomC (a, −)] and PCa := k[HomC (−, a)] are projective
in C−mod and C op −mod respectively. This is a consequence
of the Yoneda isomorphism 2 and the fact that a short exact
sequence in C−mod is exact if and only if it’s exact pointwise.
• The categories C−mod and C op −mod have enough projec-
tives. The standard projectives of C−mod (resp. of C op −mod)
form a system of projective generators. This is a particular
case of corollary 3.8. They also have enough injectives (see
[Grothendieck, theorem 1.10.1]).
2.2. Tensor product over C. Let F be a right C-module and G a
left C-module. The tensor product F ⊗C G of F and G over C is the
quotient of the k-module
M
F (x) ⊗k G(x)
x∈ObC
by the submodule generated by the elements
F (f )a ⊗ b − a ⊗ G(f )b (∀f ∈ HomC (x, y), ∀a ∈ F (y), ∀b ∈ G(x)).
Properties 2.3.
• The tensor product over C defines a bifunctor
− ⊗C − : C op −mod × C−mod → k−mod
which commutes with colimits with respect to each variable. In
particular, the bifunctor − ⊗C − is right exact relative to each
variable.
6 RAMZI KSOURI

• We have the following natural isomorphisms

(3) PCx ⊗C G ∼= G(x) and F ⊗C PxC ∼
= F (x).
• For any k-module M, and any F in C op −mod, we denote
Hom(F, M) the C-module given by x 7→ Homk−mod (F (x), M).
The adjunction
Homk (F ⊗C G, M) ∼
= HomC−mod (G, Hom(F, M))
characterizes the tensor product up to natural isomorphism.

2.3. Category of complexes. Let A be an abelian category. We de-

note Ch(A) the abelian category whose objects are cochain complexes
of objects of A and whose morphisms are morphisms of complexes.
the category Ch(A) contains the full subcategories Chb (A), Ch− (A)
and Ch+ (A) of bounded, bounded above and bounded below cochain
complexes. The notation Ch∗ (A) designates one of these subcategories.
Let X = (X n , dnX ) and Y = (Y n , dnY ) be two complexes of Ch(A).
The bifunctor HomA (−, −) : Aop × A → k−mod induces a bifunctor
Hom(−, −) : Ch(A)op × Ch(A) → Ch(k−mod)
defined by Y
(Hom(X, Y ))n = HomA (X i , Y i+n )
i∈Z
and
dn f = dY ◦ f + (−1)n+1 f ◦ dX , f ∈ (Hom(X, Y ))n .
Let X = (X n , dnX ) be a complex of Ch(C op −mod) and Y = (Y n , dnY ) a
complex of Ch(C−mod). The bifunctor −⊗C − : C op −mod×C−mod →
k−mod induces a bifunctor
− ⊗ − : Ch(C op −mod) × Ch(C−mod) → Ch(k−mod)
defined by M
(X ⊗ Y )n = X p ⊗C Y q ,
p+q=n

the differentials being defined, for a ∈ Ob(C), x ∈ X p (a) and y ∈ Y q (a)

by the formula
dn (x ⊗ y) = x ⊗ dqY y + (−1)q dpX x ⊗ y .
2.4. Extensions groups over C. The bifunctor
HomC−mod (−, −) : C−modop × C−mod → k−mod
commutes with limits relative to each variable, we denote
Ext∗C−mod (−, −) the extensions groups over C defined as right deriva-
tives of the bifunctor HomC−mod (−, −). Let F and G be two left C-
modules, we obtain Ext∗C−mod (F, G) either by choosing a projective res-
olution P∗ of F and computing the cohomology of the cochain complex
 DUALITY CATEGORIES 7

HomC−mod (P∗ , G), or by choosing an injective resolution I ∗ of G and

computing the cohomology of the cochain complex HomC−mod (F, I ∗ ).
2.5. Torsion groups over C. The bifunctor
− ⊗C − : C op −mod × C−mod → k−mod
is right exact with respect to each variable, we denote TorC−mod
∗ (−, −)
the torsion groups over C defined as the left derivatives of the tensor
product over C. For a right C-module F and a left C-module G, we
obtain TorC−mod
∗ (F, G) either by choosing a projective resolution P∗ of
F and computing the homology of the chain complex P∗ ⊗C G or by
choosing a projective resolution Q∗ of G and computing the homology
of the chain complex F ⊗C Q∗ .

3. Yoneda bifunctor and epireflective subcategory

We add the following notations to those introduced in the previous
section.
Notations 3.1. Let B(−, −) : C op × C → k−mod a bifunctor. Let F
and G be elements of C−mod and C op −mod respectively. We denote
• Ba the left C-module B(a, −).
• B a the right C-module B(−, a).
• Hom(F, B) the right C-module a 7→ HomC−mod (F, Ba ).
• Hom(B, F ) the left C-module a 7→ HomC−mod (Ba , F ).
• B ⊗C F the left C-module a 7→ B a ⊗C F .
• G ⊗C B the right C-module a 7→ G ⊗C Ba .
We denote C the bifunctor
(x, y) 7→ k[HomC (x, y)].
We have the equalities C a = PaC and C a = PCa .
Definition 3.2. A Yoneda bifunctor for C is a bifunctor T : C op × C →
k−mod with a surjection
µ:C // T
such that the canonical morphism between elements of (C op × C)−mod
(4) ϕ : [(a, b) 7→ HomC−mod (Tb , Ta )] → [(a, b) 7→ T (a, b)]
defined by ϕ(a,b) (F ) = Fb (µb,b(1b )) is an isomorphism where F : Tb →
Ta is a natural transformation and Fb : Tb (b) → Ta (b) the component
of F assigned to b.
Examples 3.3.
(1) The bifunctor
C : C op × C → k−mod
(x, y) 7→ k[HomC (x, y)]
8 RAMZI KSOURI

with the identity transformation as the surjection µ is a Yoneda

bifunctor. This is a direct consequence of the Yoneda isomor-
phism (see properties 2.2).
(2) Suppose C is a k-linear category. We denote LC the bifunctor
LC : C op × C → k−mod
.
(x, y) 7→ HomC (x, y)
The bifunctor LC with the surjection
µx,y : k[HomC (x, y)] → HomC (x, y)
[f ] 7→ f
is a Yoneda bifunctor. Indeed the map
HomC−mod (HomC (y, −), HomC (x, −)) → HomC (x, y)
T 7→ Ty (µy,y (1y )).
is an isomorphism. The inverse is given by associating to f ∈
HomC (x, y) the natural transformation whose components are
the precomposition by f .
Proposition 3.4. Let T be a Yoneda bifunctor.
(1) The bifunctor T op : (a, b) 7→ T (b, a) is a Yoneda bifunctor for
C op .
(2) The canonical morphism between elements of (C op × C)−mod
h i
ψ : [(a, b) 7→ HomC−mod (Tb , Ta )] → (a, b) 7→ T b ⊗C Ta )
defined by ψ(a,b) (F ) = µb,b (1b ) ⊗ Fb (µb,b (1b )) is an isomorphism.
Proof.
(1) The canonical morphism
h i
ϕop : (a, b) 7→ HomC−mod (T b , T a ) → [(a, b) 7→ T (b, a)]
defined by ϕop (a,b) (F ) = Fb (µb,b (1b )) is injective.
Let λ ∈ T (b, a). The assumption on T implies that there
is a unique natural transformation F : Ta → Tb such that
Fb (µb,b (1b )) = λ. We define a natural transformation F̃ : T b →
T a as follows: for x in Ob(C),
F̃x : T b (x) → T a (x)
λ′ 7→ ϕ(x,a) ◦ F ∗ ◦ ϕ−1 ′
(x,b) (λ ).

The equality ϕop (F̃ ) = λ implies the surjectivity of ϕop .

(2) The morphism ψ of the proposition is the composition
µ
HomC−mod (Tb , Ta ) → T (a, b) → PCb ⊗C Ta → T b ⊗C Ta .
The first two maps are isomorphisms. The third is surjective
since the tensor product is right exact. We construct an inverse
T b ⊗C Ta → HomC−mod (Tb , Ta ) to ψ. First, under the natu-
ral transformation (4), we identify T (x, b) to HomC−mod (Tb , Tx )
 DUALITY CATEGORIES 9

and T (a, x) to HomC−mod (Tx , Ta ) for each object x of C. The

composition of natural transformation gives a bilinear map

HomC−mod (Tb , Tx ) × HomC−mod (Tx , Ta ) → HomC−mod (Tb , Ta )

(S, R) 7→ R ◦ S

that induces a linear map

HomC−mod (Tb , Tx ) ⊗k HomC−mod (Tx , Ta ) → HomC−mod (Tb , Ta ).

Taking the direct sum over objects of C, we obtain a map

(5) M
HomC−mod (Tb , Tx ) ⊗k HomC−mod (Tx , Ta ) → HomC−mod (Tb , Ta ).
x∈ObC

Let f : x → y a morphism in C. Denote T (b, f ) (resp. T (f, a))

the induced morphism HomC−mod (Tb , Ty ) → HomC−mod (Tb , Tx )
(resp. HomC−mod (Tx , Ta ) → HomC−mod (Ty , Ta )). The equality
R ◦ T (b, f )S = T (f, a)R ◦ S for all S ∈ HomC−mod (Tb , Ty ) and
R ∈ HomC−mod (Tx , Ta ) implies that the map (5) passes to quo-
tient to give a map T b ⊗C Ta → HomC−mod (Tb , Ta ) that is an
inverse to ψ.


Notations 3.5. We denote CT −mod the smallest full subcategory of

C−mod containing the Ta and stable under taking limits and colimits.
This notation will be justified by corollary 3.8.
Example 3.6.
(1) CC −mod = C−mod.
(2) CC op −mod = C op −mod.
(3) If C is k-linear then CLC −mod is the full subcategory of k-linear
functors of C−mod.

Proposition 3.7. Let F be an object of CT −mod. Then the map

µ ∗
Hom(T, F ) → Hom(C, F ) ∼
=F

induced by µ is an isomorphism.

Proof. Let us say that a functor is nice if it satisfies the property of the
proposition. That a functor F = Ta is nice is obvious. We will show
that any limit, resp. any colimit, of nice functors is a nice functor.
By the definition of CT −mod, this will give the desired result. First,
suppose F is a functor of CT −mod which is a limit of nice objects.
Write F = lim Fi . The bifunctor Hom(−, −) commutes with limits, so

Hom(Ta , lim Fi ) ∼
= lim Hom(Ta , Fi ) ∼
= F (a).
10 RAMZI KSOURI

Secondly, suppose F is a functor of CT −mod which is a colimit of nice

objects. We write F = colim Fi . Let
Hom(PaC , Fi ) / colim Hom(PaC , Fi )
di

Hom(Ta , Fi ) ci
/ colim Hom(Ta , Fi )

Fi / colim Fi
fi

be the universal cocones. We obtain the following diagram

Hom(PaC , colim Fi )
❡❡❡❡2 ❥❥5
fi∗ ❡❡❡❡❡❡❡❡❡❡ ❥❥❥❥❥
O
❡❡❡❡ ❡ ❥
❡❡ ❡❡ ❡❡❡❡❡❡ ❥❥❥❥❥❥❥ γ
❡❡❡❡❡
Hom(PaC , Fi ) / colim Hom(P C , Fi )
a
O di O
µ∗ α µ∗
? ci
Hom(Ta , Fi ) ❨❨❨❨ / colim Hom(Ta , Fi )
❨❨❨❨❨❨ ❚❚❚❚
❨❨❨❨❨❨❨ ❚❚❚ β
❨❨❨❨❨❨
❨❨❨❨❨❨ ❚❚❚❚❚❚❚
fi∗ ❨❨❨❨❨❨, ❚❚) ?
Hom(Ta , colim Fi )
where α, β and γ are given by the universal property.
We show that β is an isomorphism. The morphism
∗ C
µ : Hom(Ta , Fi ) → Hom(Pa , Fi ) is an isomorphism. It follows
that α is also an isomorphism. The standard projective PaC com-
mutes with colimits, (i.e. γ is an isomorphism) so the cocone
fi∗
Hom(PaC , Fi ) → Hom(PaC , colim Fi ) is also universal. In the diagram,
the square, the triangles and the large trapeze commute. We obtain
the equalities
µ∗ βci = µ∗ fi∗ = fi∗ µ∗ = γdi µ∗ = γαci .
The cocone {ci } being universal, and γα being an isomorphism, it
follows that µ∗ β is an isomorphism, so that µ∗ is surjective (so it’s an
isomorphism). We deduce that the bottom cocone fi∗ is universal and
β is an isomorphism. Finally one obtains the isomorphisms
Hom(Ta , colim Fi ) ∼
= colim Hom(Ta , Fi ) ∼
= F (a).

Corollary 3.8. The C-modules Ta form a system of projective gener-
ators of CT −mod. They are called T -standard projectives.
Proof. According to the property of the previous proposition, the func-
tor
HomCT −mod (Ta , −)
 DUALITY CATEGORIES 11

of (CT −mod)−mod is exact. Hence Ta is projective. Let F an object

of CT −mod. Let λ ∈ F (a). There is a unique natural transformation
ρλ : Ta → F such that (ρλ )a (µ(1a )) = λ. The natural transformation
M M
ρλ : Ta → F
a,λ∈F (a) a,λ∈F (a)

is an epimorphism. It follows that Ta ’s form a system of generators of

CT −mod. 
Corollary 3.9. Let F and G be two objects of CT −mod and CT op −mod
respectively. Then
(1) T ⊗C F ∼=F
(2) G ⊗C T ∼=G
Proof. The tensor product over C commutes with colimits. The Ta ’s
are generators so a functor of CT −mod (resp. CT op −mod) is a colimit
of the Ta ’s (resp. T a ’s). We might need to iterate i.e. take colimit of
objects which are themselves colimit of the Ta . It suffices therefore to
prove the property 1 (resp. 2) for Ta (resp. T a ). This is a consequence
of proposition 3.4. 
Remark 3.10. We call the isomorphisms of proposition 3.7 and corol-
lary 3.9 the Yoneda properties.
Definition 3.11. Let B be a category and A a full subcategory. We say
that A is epireflective if the inclusion i : A → B admits a left adjoint
t : B → A such that the unit of the adjunction is an epimorphism.
We say that A is coreflective if the inclusion i : A → B admits a right
adjoint.
Proposition 3.12. There is a one to one correspondence between
equivalence classes of epireflective and coreflective subcategories of
C−mod and isomorphism classes of Yoneda bifunctors for C.
We need the following digression before proving the proposition. A
subcategory B of A is said dense in A if every object of A is a colimit
of objects of B in a canonical way (see [Maclane1, X.6]). Let G be the
full subcategory of C−mod whose objects are the functors PaC ⊕ PaC for
all object a of C.
Lemma 3.13. The subcategory G is dense in C−mod.
Proof. The lemma is equivalent to the following assertion: The functor
C−mod → G op − Set
F 7→ HomC−mod (−, F )
is full and faithful where G op − Set is the category of covariant functor
from G op to the category of sets (see [Maclane1, X.6 page 247]). We
have to prove that the map
φ : HomC−mod (F, G) → HomG op −Set (HomC−mod (−, F ), HomC−mod (−, G)),
12 RAMZI KSOURI

which assigns to a natural transformation N : F → G the natural

transformation N∗ : HomC−mod (−, F ) → HomC−mod (−, G) whose com-
ponents are the postcomposition by N, is an isomorphism. We remark
that the component of N∗ assigned to PaC ⊕ PaC is given by the map
T1 ⊕ T2 7→ N ◦ T1 ⊕ N ◦ T2
where
T1 ⊕ T2 ∈ HomC−mod (PaC ⊕ PaC , F ) ∼
= HomC−mod (PaC ⊕ PaC , F ).
We construct an inverse to φ. Let M an element of
HomG op −Set (HomC−mod (−, F ), HomC−mod (−, G)). Consider the compo-
nent
MPaC ⊕PaC : HomC−mod (PaC ⊕ PaC , F ) → HomC−mod (PaC ⊕ PaC , G)
of M and identify HomC−mod (PaC ⊕ PaC , F ) to HomC−mod (PaC , F ) ⊕
HomC−mod (PaC , F ). By the Yoneda isomorphism, a morphism PaC ⊕
PaC → PaC ⊕ PaC is represented by a square matrix with coef-
ficient in k[HomC (a, a)]. Using the naturality of the component
MPaC ⊕PaC with !respect to
!
the morphisms
!
represented
!
by the ma-
0 1a 1 0 0 1a 1a 1a
trices , a , and one can see that
1a 0 0 0 0 0 0 0
MPaC ⊕PaC = MPaC ⊕ MPaC where
MPaC : HomC−mod (PaC , F ) → HomC−mod (PaC , G)
is a morphism of k-modules which is natural with respect to a. By
the Yoneda isomorphism, identify HomC−mod (PaC , F ) to F (a) then MPaC
induces a morphism M f : F (a) → G(a) natural in a. The association
a
M 7→ Mf defines an inverse to φ. 
proof of proposition 3.12. Let T be a Yoneda bifunctor with a surjec-
tion µ. We show that the full subcategory CT −mod of C−mod is epire-
flective and coreflective. We construct a left adjoint t : C−mod →
CT −mod to the inclusion functor i : CT −mod  / C−mod. First, we
define t on the subcategory G of C−mod by t(Pa ⊕ PaC ) = Ta ⊕ Ta .
C

Secondly, for F = colim Fi with Fi ∈ ObG, we define t(F ) by

t(F ) = colim t(Fi ) (it is well-defined because the colimit is canonical
see the paragraph before lemma 3.13). The functor t is a left adjoint to
the inclusion i. Indeed for any object G of CT −mod we have a natural
isomorphism
HomC −mod (t(F ), G) ∼
T = HomC−mod (F, i(G))
obtained as follows:
HomCT −mod (t(colim Fi ), G) ∼
= lim HomCT −mod (t(Fi ), G)
∼
= lim HomC−mod (Fi , i(G))
∼
= HomC−mod (colim Fi , i(G))
∼
= HomC−mod (F, i(G)).
 DUALITY CATEGORIES 13

The second isomorphism above is a consequence of proposition 3.7 and

the fullness of CT −mod in C−mod. The unit of the adjunction is the
natural transformation given for each object X of C−mod by the map
εX : X = colim Fi → it(X) = colim t(Fi )
induced by the surjections µFi : Fi → t(Fi ). The functor colim is right
exact then εX is surjective. The isomorphism HomCT −mod (Ta , G) ∼ =
C C ∼
HomC−mod (Pa , i(G)) for each object G of CT −mod implies t(Pa ) = Ta .
Now let us prove that the functor i has a right adjoint. It suffices to
show that the dual conditions of the special adjoint functor theorem
[Borceux, theorem 3.3.4 page 110] are satisfied for i. The category
CT −mod is cocomplete and the functor i preserves colimits by defini-
tion of CT −mod. It remains to verify that CT −mod is co-well-powered
(the dual notion of well-powered see [Borceux, definition 4.1.2]) and has
a generating family. The later condition follows from corollary 3.8 and
the essential smallness of C. The former is a consequence of [Borceux,
proposition 4.5.15] (notice that for an object in an abelian category
there’s a one to one correspondence between its subobjects and its quo-
tient objects thus well-powered implies co-well-powered). Conversely,
let H an epireflective and coreflective subcategory of C−mod. Let t be
the left adjoint of the inclusion i : H → C−mod. We denote ε the unit
of the adjunction. The bifunctor
T H : C op × C → k−mod
defined by
T H (a, b) = t(PaC )(b)
with the surjection µa,b = εPaC (b) is a Yoneda bifunctor as shown by the
isomorphisms
Hom(TaH , TbH ) = Hom(tPaC , tPbC )
∼
= Hom(PaC , i(tPbC )) ∼
= t(PbC ))(a) = T H (b, a) .
The first isomorphism sends a natural transformation F : TaH → TbH
to F ◦ εPaC . The second is the Yoneda isomorphism (see properties
2.2). Since a left adjoint is unique up to isomorphism then the cor-
respondence H 7→ T H between the equivalence classes of epireflective
and coreflective subcategories of C−mod and the isomorphism classes
of Yoneda bifunctor for C is well defined. By the first part of the proof,
the subcategory CT −mod corresponds to the bifunctor T . In order
for this correspondence to be one to one, we must prove that H and
CT H −mod are equivalent. Let us observe first that H is stable (up
to isomorphism) under taking limits and colimits (i.e a (co)limit in
C−mod of a diagram in H is isomorphic to the (co)limit computed in
H see the proof of [Borceux, proposition 3.5.3] and its dual). The ad-
junction isomorphism and the Yoneda Lemma provide the isomorphism
HomH (TaH , F ) ∼
= F (a) for each object F of H. The same reasoning as
14 RAMZI KSOURI

in corollary 3.8 shows that the TaH ’s form a system of projective gen-
erators of H and thus each object of H is a colimit of the TaH ’s taking
into account that we may need to iterate. It follows that H contains
an isomorphic copy of each object of CT H −mod and therefore the two
subcategories are equivalent. 

4. Functor of duality
In this section, we fix a Yoneda bifunctor T for C. We recall that
CT −mod is the smallest full subcategory of C−mod containing the Ta ’s
and stable under limits and colimits. This is an abelian category with
enough projectives. The projectives Ta , called T -standards, form a
system of generators of CT −mod.
4.1. Definitions. Let R be a unitary ring and M a left R-module.
The dual of M is the right R-module M ′ := HomR−mod (M, R). The
functor of duality D defined below is a generalization of the duality
operator M 7→ M ′ where R is replaced by a category C.
We denote
D : (C−mod)op → C op −mod
the functor that sends a left C-module F to the right C-module DF :=
Hom(F, T ). We refer to notations 3.1 for the definition of Hom(F, T )
i.e. for an object x of C, the equality
DF (x) = HomC−mod (F, Tx )
where Tx is the functor T (x, −) defines Hom(F, T ) on objects.
The functor of duality D extends to the category (Ch(C−mod))op as
follows. To a complex X = (X n , dnX ) of Ch(C−mod), the functor
D : (Ch(C−mod))op → Ch(C op −mod)
associates the complex D(X) defined by
−(n+1)
D(X)n = D(X −n ) and dnD(X) = (−1)n+1 D(dX ).
4.2. Behavior of D on projectives of CT −mod.
Example 4.1. For any object a of C, we have a natural isomorphism
D(Ta ) ∼= T a.
Indeed, one has natural isomorphisms, as a consequence of the Yoneda
properties,
D(Ta )(x) = HomC−mod (Ta , Tx ) ∼
= Tx (a) = T a (x)
for all x ∈ ObC. The equality D(Ta ) = T a follows.
The previous example and the fact that D commutes with limits
imply that the functor D restricts to a functor
D : (CT −mod)op → CT op −mod.
 DUALITY CATEGORIES 15

Definition 4.2. An object of CT −mod is said of finite type if it is a

quotient of a finite sum of T -standard projectives.
Lemma 4.3. Let P be a projective of finite type of CT −mod. The
object D(P ) of CT op −mod is a projective of finite type.
Proof. If P is a projective of finite type then P is a direct factor of a
finite sum of T -standard projectives F , i.e. there is a C-module Q such
that P ⊕ Q = F . By additivity of the functor D, it suffices to prove
the lemma for P = Ta , case corresponding to the example 4.1. 
Notations 4.4. We denote P roj(CT −mod) the full subcategory of
CT −mod whose objects are the projectives of finite type.
Proposition 4.5. The restriction
D : (P roj(CT −mod))op → P roj(CT op −mod)
is an equivalence of categories.
Proof. Let P and Q be two projectives of finite type of CT −mod then
the map
D
(⋆) HomCT −mod (P, Q) → HomCT op −mod (D(Q), D(P ))
is a natural isomorphism in two variables. Indeed, by additivity of D it
suffices to check (⋆) for P = Ta and Q = Tb . In this case, the assertion
is a consequence of the natural isomorphisms
HomC −mod (Ta , Tb ) ∼
T = T (b, a) ∼
= HomC op −mod (T b , T a ).
T

The functor D has an inverse which we denote by DT op defined for an

object G of CT op −mod by DT op G = Hom(G, T op ). Indeed, the compo-
sition
P (x) ∼
= HomCT −mod (Tx , P ) → HomCT op −mod (DP, T x ) ∼
D
= DT op ◦ DP (x)
defines an isomorphism I ∼ = DT op ◦ D. 
We define a natural map
φG : D(F ) ⊗C G → HomCT −mod (F, G)
as follows. Let x ∈ ObC. Then φ is induced by the composition
≈
D(F )(x) ⊗ G(x) → HomCT −mod (F, Tx ) ⊗ HomCT −mod (Tx , G)
→ HomCT −mod (F, G) .
Let R be a unitary ring, M a right R-module and N a left R-module.
If M is projective of finite type then we have a natural isomorphism
(6) HomR−mod (M, N) ∼ = M ′ ⊗R N.
The following proposition generalizes the latter isomorphism when R
is replaced by a category C.
16 RAMZI KSOURI

Proposition 4.6. Let P be a projective of finite type of CT −mod. Then

the natural transformation between functors of (CT −mod)−mod
≈
(7) D(P ) ⊗C − → HomCT −mod (P, −).
is an isomorphism.
Proof. According to the Yoneda properties, we have the following iso-
morphisms
D(Ta ) ∼
= T a and HomC −mod (Ta , G) ∼
T = G(a) ∼
= T a ⊗C G.
It follows that the statement is true for P = Ta . If P is a direct
summand in a finite sum of T -standard projectives, the proposition is
derived from the additivity of D. 
4.3. Behavior of D on complexes of projectives and the derived
functor of D.
Proposition 4.7. Let X = (X n , dnX ) be a complex of projectives
of finite type of Chb (CT −mod) and Y = (Y n , dnY ) a complex of
Ch(CT −mod). Then we have a natural isomorphism
Hom(X, Y ) ∼
= D(X) ⊗ Y.
Proof. This is a consequence of the isomorphism of the proposition 4.6
and the compatibility of differentials with this isomorphism. 
Recall that the homotopy category K∗ (A) is obtained from the cate-
gory Ch∗ (A) by identifying homotopically equivalent morphisms, i.e.by
setting
HomK(A) (X, Y ) := H 0 (Hom(X, Y )),
and that the derived category D ∗ (A) is the triangulated category
obtained from the homotopy category K∗ (A) by inverting quasi-
isomorphisms.
The bifunctors Hom(−, −) and − ⊗ − induce the derived bifunctors
(see [Borel, page 91 and 98])
RHom : (D − (C−mod))op × D(C−mod) → D(k−mod),
and
L
− ⊗ − : D − (C op −mod) × D − (C−mod) → D(k−mod).
The functor D admits a derived functor
RD : (D − (CT −mod))op → D + (CT op −mod).
To calculate the derived dual of a complex X of D − (CT −mod), we
choose a projective resolution of X and we apply the functor D. Let F
be an object of CT −mod and x an object of C. We have the equality:
Hi (RD(F ))(x) = ExtiCT −mod (F, Tx ).
In the sequel, we set D i (X) = H i (RD(X)).
 DUALITY CATEGORIES 17

Definition 4.8. A complex X of D(CT −mod) is perfect if it is quasi-

isomorphic to a bounded complex of finite type projectives.
Proposition 4.9. Let X and Y be two complexes of D − (CT −mod). If
X is perfect, then we have a natural isomorphism
L
RHom(X, Y ) ∼
= RD(X) ⊗ Y.
Proof. Let P a complex of projectives of finite type of
Chb (CT −mod) quasi-isomorphic to X. Then we have the isomorphisms
L
RHom(X, Y ) = Hom(P, Y ) ∼
= D(P ) ⊗ Y = RD(X) ⊗ Y.

4.4. Illustration: Serre duality. Let k be a field. For a k-vector
space V , we put V ′ := Homk (V, k).
Definition 4.10. Let A be a k-linear category. A Serre duality for A
is an endofunctor S : A → A equipped with a natural isomorphism
Φa,b : HomA (b, Sa) → HomA (a, b)′
for all objects a, b of A. [Compare to [RV02]]
Remark 4.11.
(1) The functor S is a Serre duality for A if and only if the object Sx
is an object representing the functor HomA (x, −)∗ of Aop −mod
for all object x of A.
(2) The Yoneda embedding
A → Aop −mod
c 7→ HomA (−, c)
is fully faithful. Hence the uniqueness of a Serre duality up to
isomorphism.
(3) The Yoneda lemma provides a natural isomorphism
HomAop −mod (HomA (−, Sa), HomA (a, −)′ ) ∼= HomA (a, Sa)′ .
We call fundamental class the element Φa ∈ HomA (a, Sa)′ cor-
responding to the isomorphism Φa,b . The isomorphism Φa,b is
right adjoint (in the sense of the theory of bilinear forms) of the
bilinear form
Φ
HomA (a, b) × HomA (b, Sa) → HomA (a, Sa) →a k.
Let T a Yoneda bifunctor for C. Assume that C is k-linear.
Notations 4.12.
• We denote Dperf (CT −mod) the full subcategory of D(CT −mod)
whose objects are perfect complexes.
• We denote F ′ the postcomposition of F by the duality operation
V 7→ V ′ := Homk (V, k).
18 RAMZI KSOURI

Proposition 4.13. Let X and Y be two complexes of Dperf (CT −mod).

We have a natural isomorphism
RHom(X, Y )′ ∼= RHom(Y, RD(X)′).
In particular, the functor RD(−)′ is a Serre duality for
Dperf (CT −mod).
Proof. Let P a complex of finite type projectives of
Chb (CT −mod) quasi-isomorphic to X. Then we have the isomorphisms
RHom(X, Y )′ ∼= (D(P ) ⊗ Y )′ ∼
= Hom(Y, D(P )′).
The second one is a consequence of the adjunction characterizing the
tensor product. The functor F 7→ F ′ is exact so D(P )′ is a bounded
injective resolution of D(X)′ . We deduce an isomorphism
RHom(X, Y )′ ∼= RHom(Y, RD(X)′).


5. Duality functors
Let T a Yoneda bifunctor fo C. We introduce the concept of duality
functors in CT −mod. This concept is measured by the action of the
functor of duality
D : (CT −mod)op → CT op −mod
on some objects of CT −mod.

5.1. Definitions and first properties.

Definition 5.1. A CT -module is of type F P∞ (resp. of type F P ) if
it has a resolution (resp. finite resolution) by finite type projectives of
CT −mod.
Definition 5.2. A left CT -module F is said of projective dimension
n if for all CT -module G, and all m > n, we have the equality
ExtmC−mod (F, G) = 0, and there is a C-module G such that we have
n
ExtCT −mod (F, G) 6= 0.
Notations 5.3. Let X an object of CT −mod and n an integer. We
denote X[n] the complex of D (CT −mod) defined by X[n]i = 0 if i 6= n
and X[n]n = X.
Definition 5.4. We call an Eilenberg-Mac Lane functor of degree n
an object F of CT −mod such that we have an isomorphism
RD (F ) ∼
= D n (F ) [n]
in D + (CT op −mod) (Recall that D n (F ) = H n (RDF )).
Theorem 5.5. Let F a CT -module of type F P . The following condi-
tions are equivalent
 DUALITY CATEGORIES 19

(1) There exists an integer n such that F is an Eilenberg-Mac Lane

functor of degree n.
(2) There exists an integer n such that
ExtiCT −mod (F, Tx ) = 0
for all i 6= n and all x ∈ ObC.
(3) There exists an integer n and an isomorphism
Exti (F, −) ∼
CT −mod
C −mod
= Tor T (D n (F ) , −) .
n−i

The integer n is the projective dimension of F in the category CT −mod.

Proof. The equivalence between condition 1 and condition 2 is clear.
We show that the third condition implies the second one. Let x be
an object of C. Evaluating the isomorphism of condition 3 at Tx , we
obtain
ExtiCT −mod (F, Tx ) ∼
= TorCn−i
T −mod
(D n (F ) , Tx ) .
Since Tx is projective in CT −mod then we have for i 6= n the equality
TorCn−i
T −mod
(D n (F ) , Tx ) = 0 and hence ExtiCT −mod (F, Tx ) = 0 for i 6= n.
We show that condition 1 implies condition 3. The CT -module F is of
type F P so F is perfect and since RD (F ) ∼ = D n (F ) [n] we have by
proposition 4.9 the isomorphism
L
RHom (F, −) ∼
= D n (F ) [n] ⊗ −.
Taking cohomology, we obtain an isomorphism ExtiCT −mod (F, −) ∼
=
CT −mod n
Torn−i (D (F ) , −). 
Definition 5.6. A functor F which satisfies one of the conditions of
the theorem is called a duality functor and the functor D n (F ) is the
dualizing functor of F .
Remark 5.7. Suppose k is a field. Then the third condition of the
theorem is equivalent to the following.
(3’) There exists an integer n and an isomorphism
 
Exti (F, −)′ ∼
CT −mod = Extn−i −, D n (F )′ .
CT −mod

Proposition 5.8. Let F a duality functor of CT −mod of projective

dimension n. The functor D n (F ) is a duality functor of CT op −mod
and F is its dualizing functor.
Proof. The equivalence
DT : (P roj(CT −mod))op → P roj(CT op −mod)
of proposition 4.5, with inverse the functor
DT op : P roj(CT op −mod) → (P roj(CT −mod))op ,
passes to quotient and we obtain the equivalence
RDT : Kb (P roj(CT −mod))op → Kb (P roj(CT op −mod))
20 RAMZI KSOURI

at the level of homotopy categories. The composition respectively on

the left and on the right by the equivalences
Dperf (CT −mod) ∼
= Kb (P roj(CT −mod))
Kb (P roj(CT op −mod)) ∼
= Dperf (CT op −mod)
provides the equivalence of categories
RDT : (Dperf (CT −mod))op → Dperf (CT op −mod)
with inverse the functor RDT op . The isomorphism RDT op (RDT F ) ∼=F
n ∼ n
implies RDT (D F ) = F [n]. It follows that D (F ) is a duality functor
op

of CT op −mod and F is its dualizing functor. 

6. Illustration: strict polynomial functors

We illustrate these concepts in homology of strict polynomial func-
tors in the sense of Friedlander and Suslin [FS97]. The study of exten-
sions groups is well developed this later years (see eg [Touzé2]). For a
finite field k, let Rep Γdk be the category of strict polynomial functors of
degree d. The category Rep Γdk is equipped with a Ringel duality θ intro-
duced by M.Chałupnik [Chałupnik] and A.Touzé [Touzé1]: the functor
Ringel duality θ := RHom(Λd , −) : D b (Rep Γdk ) → D b (Rep Γdk ) is an
equivalence of categories with inverse the functor θ−1 : X 7→ θ(X # )#
(see [Chałupnik, corollary 2.4], [Touzé1, theorem 3.8], [Krause, theorem
4.9]).
H. Krause in [Krause] showed that the equivalence θ−2 is a Serre
duality for Dperf (Rep Γdk ).
This phenomenon of Serre duality for the category Rep Γdk fits in the
general framework of the duality presented in this article. On one hand,
our theory of duality allows to equip the category of strict polynomial
functors Rep Γdk with a functor of duality D which in turn provides a
Serre duality RD(−)′ for Dperf (Rep Γdk ). On the other hand, we deduce
from the uniqueness of a Serre duality that RD(−)′ and the Serre
duality described by Krause coincide i.e. the equivalence of categories
θ−2 is isomorphic to RD(−)′ .
6.1. Serre duality for Rep Γdk . We follow the notations of [Krause].
We refer to [Pirashvili] for more details on strict polynomial functors.
Let Pk the category of k-vector spaces of finite dimension and Γd :
Pk → Pk the divided power functor defined by
 d
 Sd
Γd (V ) = V ⊗
[the symmetric group Sd acting by permutation of the factors of the
tensor product]. We denote Γd Pk the category whose objects are finite
dimensional k-vector spaces, whose morphisms are defined by
HomΓd Pk (V, W ) = Γd Homk (V, W ) (∀V, W ∈ Pk )
 DUALITY CATEGORIES 21

and composition of morphisms is defined via the identification

 d d
 Sd
Γd Homk (V, W ) = Homk V ⊗ , W ⊗ .
The category of strict polynomial functor of degree d is the category of
k-linear functors from Γd Pk to the category of k-modules. We denote
Rep Γdk this category and Rep((Γdk )op ) the category of contravariant k-
linear functors from Γd Pk to the category of k-modules.
For example, the classical functors of tensor power ⊗d , of symmetric
power S d , of external power Λd and of divided power Γd are strict
polynomial functors of degree d.
The category Rep Γdk is an epireflective and coreflective subcategory
of Γd Pk −mod associated to the Yoneda bifunctor LΓd Pk defined in ex-
ample 3.3. This is an abelian category. The standard projectives
Γd,V = HomΓd Pk (V, −) form a system of projective generators of Rep Γdk .
We recall that we have a functor of duality
D : (Rep Γdk )op → Rep((Γdk )op )
F 7→ Hom(F, LΓd Pk ).
Proposition 6.1. The functor RD(−)′ is a Serre duality for
Dperf (Rep Γdk ).
Proof. This is a rewrite of the proposition 4.13 in the context of strict
polynomial functors. 
Notations 6.2. Let F be an object of Rep Γdk and V an object of Pk .
• We denote Λd,V the precomposition of Λd by Hom(V, −).
• We denote # the functor
#
: (Rep Γdk )op → (Rep Γdk )
F 7→ F#
defined by F # (V ) = F (V ′ )′ .
• We denote Hom(Λd , −) the functor
Hom(Λd , −) : Rep Γdk → Rep Γdk
F 7→ Hom(Λd , F )
where Hom(Λd , F ) is the functor V 7→ HomRep Γdk (Λd,V , F ).

Theorem 6.3 ([Touzé1, theorem 3.8], [Krause, theorem 4.9]). The

functor Ringel duality
θ := RHom(Λd , −) : D b (Rep Γdk ) → D b (Rep Γdk )
is an equivalence of categories with quasi-inverse the functor θ−1 : X 7→
θ(X # )# .
Proposition 6.4. The endofunctors RD(−)′ and θ−2 of Dperf (Rep Γdk )
are isomorphic.
22 RAMZI KSOURI

Proof. H.Krause showed that the functor θ−2 is a Serre duality for
Dperf (Rep Γdk ) (see [Krause, corollary 5.5]). The proposition follows
from the uniqueness of a Serre duality up to an isomorphism (see re-
mark 4.11). 

6.2. Example of a duality functor in Rep Γdk . Let k be a field of

characteristic p. Let r and s be two positive integers. We put d = spr .
The functor Frobenius twist
r r
I(r) : V 7→ V (r) = k < v ⊗p , v ∈ V >⊂ S p (V )
is a strict polynomial functor of degree pr . The precomposition by I(r)
of a strict polynomial functor F of degree s define a strict polynomial
functor I(r) ◦F of degree spr . We put F (r) := I(r) ◦F .
Let s = (s1 , · · · , sk ) a partition of s. For X = S (resp. X = Γ, X = Λ),
we set X s = ⊗ki=1 X si . Recall that X s is then a strict polynomial functor
of degree s.
Proposition 6.5. Let s = (s1 , · · · , sk ) a partition of s. Then the
functor Γs(r) is a duality functor. In particular
 ′
RD Γs(r) ∼
= S s(r) [2s (pr − 1)]
and we have a natural isomorphism
 ′  
∼ 2s(pr −1)−i
ExtiRep Γd Γs(r) , − = ExtRep Γd −, S s(r) .
k k

Proof. For the proof of the proposition, we make use of the following
properties of the Ringel duality θ (see theorem 6.3) for strict polynomial
functors studied in the article [Touzé1].
(1) We have isomorphisms θ(S s ) ∼ = Λs and θ(Λs ) ∼
= Γs (see [Touzé1,
lemma 3.6]).
(2) Let C ∈ D b (Rep Γdk ). We have natural isomorphisms
 
θn (C ◦ I(r) ) ∼
= θn (C) ◦ I(r) [dn(pr − 1)]
and  
θn (I(r) ◦C) ∼
= I(r) ◦θn (C) [dn(pr − 1)]
(see [Touzé1, proposition 6.6]).
The proposition 6.4 and the definition of the inverse of θ in theorem
6.3 imply
 ′
RD Γs ◦ I(r) ∼
= θ−2 (Γs ◦ I(r) ) ∼
= θ2 (Γs ◦ I(r) )# )# .
The above properties of θ and the equalities (F ◦G)# = F # ◦G# , Γs# =
S s and I(r)# = I(r) allow to rewrite this isomorphism as
 ′
RD Γs ◦ I(r) ∼
= (θ2 (S s ) ◦ I(r) )# [2s (pr − 1)] ∼
= S s ◦ I(r) [2s (pr − 1)].
 DUALITY CATEGORIES 23

The functor Γs ◦ I(r) is then an Eilenberg-Mac Lane functor of degree

2s (pr − 1), the isomorphism between extensions groups follows from
remark 5.7. 

7. Tits buildings are duality categories

Let C be a small category. In this section k is the abelian group Z.
Recall that C−mod is the category of covariant functors from C to the
category of abelian groups. The functor of duality
D : (C−mod)op → C op −mod
which associates to a left C-module F the right C-module Hom(F, C)
sends a standard projective PxC of C−mod to the standard projec-
tive PCx of C op −mod. A functor F of C−mod is a duality functor if
RDF ∼ = D n (F )[n] in D + (C op −mod) for some integer n i.e. given a
projective resolution of type F P of F , the cohomology of the com-
plex of projectives obtained after applying the functor of duality D is
concentrated in degree n. In this section we are interested in small
categories C such that the constant functor Z is a duality functor. Our
main result is corollary 7.19 where we show that spherical Tits build-
ings are duality categories.

7.1. Definitions and first properties.

Definition 7.1. A small category C is a duality category if the C-
module Z is a duality functor. In this case we have a natural isomor-
phism
ExtiC−mod (Z, −) ∼
= TorC−mod
n−i (D n (Z) , −)
for some integer n. The right C-module D n (Z) is called the dualizing
functor of C.
α
(
Example 7.2. Let C the category x 6 y . The following complex
β
of left C-modules
α∗ −β ∗
(8) 0 −→ PyC −→ PxC
is a finite projective resolution of the constant functor Z. Applying the
functor of duality D, we obtain the complex
α∗ −β∗
(9) PCy ←− PCx ←− 0.
We check that it is a finite projective resolution of the right C-module
Z. It follows that Z is a duality functor and D(Z) ∼ = Z[1].
Example 7.3. Let C the following category
24 RAMZI KSOURI

4❃ 0❃ 1
❃❃ ❃c❃ d
❃❃ ❃❃
a ❃❃❃b ❃❃ e
  ❃ 
2 3
The complex of C-modules
φ
0 → P2C ⊕ P2C ⊕ P3C → P0C ⊕ P1C ⊕ P4C
 
b∗ b∗ d∗

where φ is given by the matrix  −c ∗
0 −e∗   is a projective
∗
0 −a 0
resolution of Z. We apply the functor of duality D to this resolution,
we obtain the complex
φ∗
PC2 ⊕ PC2 ⊕ PC3 ← PC0 ⊕ PC1 ⊕ PC4 ← 0
and we check that it’s a projective resolution of the right C-module
D 1 (Z), given on objects by
D 1 (Z)(0) = D 1 (Z)(1) = D 1 (Z)(2) = Z2 and D 1 (Z)(3) = D 1 (Z)(4) = Z.

To see that a small category C is a duality category, it suffices to

calculate for any object x of C the extensions groups
Ext∗C−mod (Z, PxC ).
If these groups are trivial except for some degree n (which does not
depend on x) then C is a duality category. The purpose of what follows
is to simplify the calculations of these groups.
Recalls 7.4 (Cohomology of a small category). Our basic reference
for these recalls is chapter 1 of [Quillen]. More details can be found in
[Fei’s Book].
The cohomology of a small category C with coefficient in a functor
F is defined by H∗ (C; F ) := Ext∗C−mod (Z, F ), and the homology with
coefficient in a functor F is defined by H∗ (C; F ) := TorC−mod
∗ (Z, F ).
To a small category C, we associate a simplicial set N C called the
nerve of C (see [Maclane1]) with Nn C is the set of chains of morphisms
of C of length n. The face operators for i = 0 · · · n are given by
dni (x0 → · · · → xn ) = x0 → · · · x̂i · · · → xn ,
where x0 → · · · x̂i · · · → xn is the chain of morphisms of length n − 1
obtained from x0 → · · · → xn replacing the subchain xi−1 → xi → xi+1
by its composition. The dni ’s extend by linearity to give maps of abelian
groups Z[Nn C] → Z[Nn−1 C] that we still denote by dni . The homology
H∗ (N C, Z) of N C with integral coefficient is the homology of the chain
P
complex Z[N∗ C] with differentials dn = n0 (−1)i dni . The cohomology
H ∗ (N C; Z) is the cohomology of the cochain complex Hom(Z[N∗ C]; Z).
 DUALITY CATEGORIES 25

We denote |C| the geometric realization of the simplicial set N C, this

is a topological space constructed from N C. We have the identifications
H∗ (N C; Z) ∼= H∗ (|C|; Z) and H ∗ (N C; Z) ∼ = H ∗ (|C|; Z) (see [Weibel,
example 8.2.3]). There is a bijection between Nn C and Nn C op that
induces an homeomorphism between |C| and |C op |. We refer to [May]
for more details on simplicial sets, their realization and homology.
A small category C is said contractible if |C| has the same homotopy
type of a point. A category with an initial or a terminal object is
contractible. Let C a small category. The comma category or the
fiber category C|x of objects over x is the category whose objects are
α α β
morphisms c → x in C and a morphism from c → x to c′ → x is a
γ
morphism c → c in C such that β ◦ γ = α (see [Maclane1, Chapter
II.6]). The identity morphism of x is a terminal object of C|x hence
C|x is contractible for each x ∈ Ob(C).
We define a complex B∗C of C-module by
B∗C := Z[N∗ (C|−)].
Since the fiber categories C|x are contractible for each x ∈ Ob(C), this
complex is a resolution of the constant functor Z.
To give a chain of length n in Nn (C|x) is equivalent to give a chain
x0 → · · · → xn of length n in Nn C and a morphism xn → x. This
correspondence induces an isomorphism
M
BnC ∼
= PxCn .
x0 →···→xn ∈Nn C

It follows that B∗C is a projective resolution of the constant functor

Z (called Bar resolution in [Fei’s Book]) and we have H∗ (C; F ) =
H∗ (Hom(B∗C , F )). Moreover, the isomorphism Hom(B∗C , Z)) ∼ =
∗ ∼ ∗
Hom(Z[N∗ C]; Z) induces an isomorphism H (C; Z) = H (|C|, Z).
The same reasoning replacing C by the opposite category C op allows
us to explicit a projective resolution of the constant C op -module Z and
to identify H∗ (C; Z) to H∗ (|C|, Z).
The reduced (co)homology H(C, e −) of a small category C is the
(co)homology calculated when replacing the complex B∗C by the aug-
mented complex B∗C → Z.
Let i : C ′ ֒→ C the inclusion functor of a full subcategory C ′ in C.
The relative cohomology with coefficient in a functor F is defined by
B∗C
H∗ (C, C ′ ; F ) := H∗ (Hom( , F ))
B∗C ′
and provides a long exact sequence
· · · → Hi (C; F ) → Hi (C ′ ; F ◦ i) → Hi+1 (C, C ′ ; F ) → Hi+1 (C; F ) → · · · .
Notations 7.5. Let x and y two objects of C (not necessary an ordered
set), we say that x ≤ y if there exists a morphism from x to y in C.
26 RAMZI KSOURI

We say that x and y are joinable if it exists an object z such that

x ≤ z ≥ y.
• We denote Cex the full subcategory of C whose objects are join-
able to x.
• We denote Cx the full subcategory of Cex whose objects are  x.
• We denote C≤x (resp. C<x , C≥x ) the full subcategory of C whose
objects are ≤ x (resp. < x , ≥ x).
Proposition 7.6. Let C be a small category and x ∈ Ob(C). Then
e
H∗ (C; PxC ) ∼
= H∗ (Cex ; PxCx ).
Proof.
There is an exact sequence of C-modules 0 → B∗Cex → B∗C → B∗C /B∗Cex →
0. Since B∗C /B∗Cex is projective, by applying the functor Hom(−, PxC ), we
obtain the exact sequence
e e
0 → Hom(B∗C /B∗Cx , PxC ) → Hom(B∗C , PxC ) → Hom(B∗Cx , PxC ) → 0.
For i ∈ N, we have
e Y
Hom(BiC /BiCx , PxC ) ∼
= PxC (xi ) = 0,
/ i Cex
x0 →···→xi ∈N
e
and therefore the isomorphism Hom(B∗C ; PxC ) ∼
= Hom(B∗Cx , PxC ). 
Proposition 7.7. Let C be a small category and x ∈ Ob(C). Then
e
(10) H∗ (C; PxC ) ∼
= H∗ (Cex , Cx ; PxCx ).
Proof. The proof is analogous to the previous one. There is an exact
sequence of C-modules 0 → B∗Cx → B∗Cex → B∗Cex /B∗Cx → 0. Apply the
functor Hom(−, PxC ), we obtain the exact sequence
e e
0 → Hom(B∗Cx /B∗Cx , PxC ) → Hom(B∗Cx , PxC ) → Hom(B∗Cx , PxC ) → 0.
for i ∈ N, we have
Y
Hom(BiCx , PxC ) ∼
= PxC (xi ) = 0,
x0 →···→xi ∈Ni Cx

and therefore the isomorphism Hom(B∗Cex /B∗Cx ; PxC ) ∼ e

= Hom(B∗Cx , PxC ).


7.2. Partially ordered set. In this paragraph, we give a condition

for a simplicial poset to be a duality category.
Recalls 7.8 (simplicial complexes). A simplicial complex on a vertex
set V is a collection ∆ of non-empty finite subsets of V such that
{v} ∈ ∆ for all v ∈ V and any non-empty subset of A ∈ ∆ is in ∆.
Call simplex an element of ∆. The dimension of a simplex F is defined
by dim F = |F | − 1. The dimension of ∆ is the maximum dimension
of its elements.
 DUALITY CATEGORIES 27

Let A a simplex of a simplicial complex ∆, the link of A is the

simplicial subcomplex of simplices B joinable to A and disjoint from
A:
linkA = {B; A ∪ B ∈ ∆ and A ∩ B = ∅} .
A partially ordered set, shortened poset, is a category such that
between two objects there is at most one morphism. To any simplicial
complex ∆, we associate a poset P (∆), whose objects are simplices of
∆ and whose morphisms are inclusions. A poset P is called simplicial if
there exists a simplicial complex ∆ such that P ∼
= P (∆). The simplicial
′
join of two simplicial complexes ∆ and ∆ is the simplicial complex
∆ ∗ ∆′ = {A ∪ B; A ∈ ∆, B ∈ ∆′ } ∪ ∆ ∪ ∆′ .
Let P = P (∆) and Q = P (∆′) be two simplicial posets. Call sim-
plicial join the simplicial poset P ∗ Q := P (∆ ∪ ∆′ ). We denote |∆|
the geometric realization of a simplicial complex ∆. The barycentric
subdivision of a poset P is a simplicial poset which we denote ∆(P ).
The geometric realization of P is the one of ∆(P ).
For C a poset, the proposition 7.7 takes the following simple form:
Proposition 7.9. If C is a poset, then the isomorphism (10) takes the
simple form
(11) H∗ (C; P C ) ∼
x= H∗ (Cex , Cx ; Z).
If, in addition, Cex is contractible then
∗−1
(12) H∗ (C, PxC ) ∼
=He (Cx ; Z).
Proof. Let ZCx be the Cex -module, defined by ZCx (a) = Z if a ∈ Ob(Cx )
and 0 otherwise. If C is a poset i.e. there is at most one morphism
between two objects, then we have the exact sequence of Cex -modules:
e
0 → ZCx → Z → PxCx → 0.
Apply the functor Hom(B∗Cex /B∗Cx , −), we obtain the exact sequence
e e e e
0 → Hom(B∗Cx /B∗Cx , ZCx ) → Hom(B∗Cx /B∗Cx , Z) → Hom(B∗Cx /B∗Cx , PxCx ) → 0.
For i ∈ N, we have
e Y
Hom(BiCx /BiCx , ZCx ) ∼
= ZCx (xi ) = 0,
x0 →···→xi ∈N
/ i Cx

and therefore the isomorphism

e ≈ e e
Hom(B∗Cx /B∗Cx , Z) → Hom(B∗Cx /B∗Cx , PxCx ).
The isomorphism (11) follows by taking cohomology. The long exact
sequence of reduced cohomology associated to the pair (Cex , Cx ) gives
the second isomorphism.

28 RAMZI KSOURI

Example 7.10. Let P the following poset

0❃ 1
❃❃
❃❃
❃❃
 ❃ 
2 3
then
Pe0 = Pe1 = P , P0 = 1 and P1 = 0.
We are in the situation where P0 and P1 are contractible, so
f , P ; Z) = H1 (P
H1 (P f , P ; Z) = Z and H0 (P
f , P ; Z) = H0 (P
f , P ; Z) = 0.
0 0 1 1 0 0 1 1

Moreover, we have

Pe2 = 0 1 and Pe3 = 0 ❃ 1

❃ ❃❃
❃❃
❃❃
  
2 3
P2 and P3 consist of two points 0 and 1. We are in the situation where
Pe2 and Pe3 are contractible. We find that

H1 (P f , P ; Z) = Z and H0 (P
f , P ; Z) = H1 (P f , P ; Z) = H0 (P
f , P ; Z) = 0.
2 2 3 3 2 2 3 3

The poset P is a duality category and the dualizing functor D 1 (Z) is

the constant module Z.
Lemma 7.11. If P is a simplicial poset then
Pex ∼
= P≤x ∗ linkx and Px ∼ = P<x ∗ linkx .
Proof. Let ∆ be a simplicial complex such that P ∼ = P (∆). We can
see via this isomorphism objects of P as finite sets and morphisms as
inclusions. Let z an object of Pex , viewed as a finite set, is the disjoint
union of z∩x and z\x. The isomorphisms of the lemma are constructed
by associating to z the object (z ∩ x) ∪ (z\x) of P≤x ∗ linkx . 
Theorem 7.12. Let P a finite simplicial poset. If there exists an
integer n such that for all objects x of P
i
e (link ; Z) = 0 for i 6= n − dim(x) − 1
H x

then P is a duality category.

Proof. The previous lemma shows that Pex is contractible for every
object x of P. We have then
∗−1
e
H∗ (P, PxP ) = H (P<x ∗ linkx ; Z).
Now, for a simplicial poset, P<x is (co)homologically a sphere of di-
e k (P ∼ e k−dim(x) (linkx ; Z) (see
mension dim x − 1 so H <x ∗ linkx ; Z) = H
e i (link ; Z) = 0 for i 6= n − dim(x) − 1
[Munkres2, theorem 62.5]). If H x
 DUALITY CATEGORIES 29

then the reduced cohomology of Px is concentrated in dimension n − 1.

It follows that Hi (P, PxP ) = 0 for i 6= n. The assumption that P is
finite ensures that Z is of type F P therefore the theorem 5.5 allows to
conclude. 

The following key lemma due to Munkres, allows us to write a topo-

logical version of theorem 7.12.
Lemma 7.13. Let ∆ be a simplicial complex and x a simplex of ∆.
Then
Hi (|∆|, |∆| − x; Z) ∼
=He i−dim(x)−1 (link ; Z).
x

Proof. See [Munkres1, lemma 3.3 page 116] . 

Theorem 7.14. Let ∆ a finite simplicial complex of dimension n, then
P (∆) is a duality category if and only if the local cohomology of |∆| at
any point is concentrated in degree n. In this case:
(Z, −) ∼
P (∆)
Exti P (∆)−mod = Tor (D n (Z), −)
n−i
n
where D (Z) is the functor defined on objects by
D n (Z)(x) = Hi (|∆|, |∆| − x; Z).
Remark 7.15. If C is a poset, then we have the isomorphisms
(13) H∗ (C; P C ) ∼
= H∗ (C, C − C≥x ; P C ) ∼
x x= H∗ (C, C − C≥x ; Z).
The functor PxC is trivial on C − C≥x so the third term in the long
exact sequence associated to the pair (C, C − C≥x ) is also trivial and
this implies the first isomorphism. The second one follows from the
long exact sequence of relative cohomology groups associated to the
short exact sequence of C-modules:
0 → ZC−C≥x → Z → PxC → 0.
where ZC−C≥x is the C-module, defined by ZC−C≥x (a) = Z if a ∈
Ob(C − C≥x ) and 0 otherwise. This remark allows us to identify the
local cohomology group Hi (|∆|, |∆| − x; Z) to Hi (∆, ∆ − ∆≥x ; Z) for a
simplicial complex ∆.

7.3. Application: spherical buildings.

Definition 7.16. A spherical building is a simplicial complex which is
the union of subcomplexes Σi satisfying
(1) Each Σi is a finite Coxeter complex.
(2) For all simplices A, B ∈ ∆, there is an apartment Σi containing
A and B.
(3) If Σ1 and Σ2 are two apartments containing A and B, there
exists an isomorphism Σ1 → Σ2 fixing A and B.
30 RAMZI KSOURI

Theorem 7.17 ([AB08, theorem 4.73 page 197]). If ∆ is a spherical

building of dimension n, then |∆| has the homotopy type of a wedge of
n-spheres.
Proposition 7.18 ([AB08, proposition 4.9 page 176]). If ∆ is a spher-
ical building, then linkA is a spherical building for any A ∈ ∆.
For a spherical building of dimension n, | linkA | has the homotopy
type of a wedge of spheres of dimension n − dim(A) − 1. We deduce
from theorem 7.12:
Corollary 7.19. Let ∆ a finite spherical building then the simplicial
poset P (∆) is a duality category.

7.4. Poincaré Duality for a manifold. In this example, we find the

classical Poincaré duality for a cohomological, compact and triangu-
lated manifold. We recall that a cohomological manifold of dimension
n is a topological space X such that for every x ∈ X, the local coho-
mology groups Hi (X, X −x) are trivial for i 6= n and Hn (X, X −x) = Z.
Proposition 7.20. Let ∆ a finite triangulation of a cohomological
manifold of dimension n. The poset P (∆) is a duality category. The
P (∆)op -module D n (Z) is equal to Z for all object x of P (∆) and takes
morphisms in P (∆) to isomorphisms.
Proof. Since ∆ is a cohomological manifold of dimension n, the local
cohomology groups Hi (|∆|, |∆| − x) are trivial for i 6= n and equal
to Z for i = n. Then ,by theorem 7.14, the poset P (∆) is a duality
category and the P (∆)op -module D n (Z) is equal to Z on each object
x of P (∆). Let x, y two objects of P (∆) such there is a morphism
α : x → y (this morphism correspond to the inclusion of x in y as
simplices of ∆). We show that D n (Z)(α) is an isomorphism. To reduce
the amount of notation, we denote C the poset P (∆). The functor
D n (Z) is equal to the abelian group Z on objects then, by the long exact
sequence of cohomology groups associated to the short exact sequence
C
0 → PyC → PxC → PPxC → 0 induced by α, it’s sufficient to show that
y
n PxC
H (C; )
is trivial. Let z the object of P (∆) that correspond to y\x
PyC
in ∆.
Claim 1: We have the isomorphism
PxC ∼ e n−dim(x)−1
(14) Hn (C; )=H (linkx − linkx≥z ; Z).
PyC
Claim 2: Denote link(t, linkx ) the link of a simplex t in linkx . Then
link(t, linkx ) is equal to linkx∪t in ∆.
By lemma 7.13, the geometric realization | linkx | is a cohomology
sphere hence the reduced cohomology of linkx − linkx≥z is trivial since
it’s the reduced cohomology of | linkx | minus a point for which we refer
 DUALITY CATEGORIES 31

C
to remark 7.15 and lemma 7.13. We conclude that Hn (C; PPxC ) is equal
y
to 0.
Proof of the isomorphism (14) of claim 2: First we have that
PxC ∼ n e Cex −Cex≥z
Hn (C; C
) = H (Cx − Cex≥z ; Px ).
Py
This follows from the long exact sequence associated to the pair (C, Cex −
Cex≥z ) and the fact that
Cex −Cex≥z PxC ∼ Y PxC
Hom(BiC /Bi , )= (xi ) = 0.
PyC PyC
/ i Cex −Cex≥z
x0 →···→xi ∈N

Secondly, similarly to the proof of lemma 7.11, we have

Cex −Cex≥z ∼
= C≤x ∗(linkx − linkx≥z ) and Cx −Cx≥z ∼
= C<x ∗(linkx − linkx≥z ).
Then by proposition 7.9 we have
Cex −Cex≥z n−1
H n (Cex − Cex≥z ; Px )∼
=He (Cx − Cx≥z ; Z)
∼ e n−1 (C ∗ (link − link
=H <x x x≥z ); Z)

∼ e n−dim(x)−1 (link − link

=H x x≥z ; Z).

Proof of the claim 2: First we show link(t, linkx ) ⊂ linkt∪x . Let v a

simplex of link(t, linkx ). On the one hand, by definition of the link
of simplicial complex, we have v is disjoint from t in linkx so v is
disjoint from t ∪ x. On the other hand, v is joinable to t in linkx
hence v ∪ t ∈ linkx and therefore v is joinable to t ∪ x. It follows that
v ∈ linkt∪x . Secondly, a simplex v of linkt∪x is disjoint from t and
joinable to t in linkx then v ∈ link(t, linkx ).

Recalls 7.21 ([Hatcher, theorem 3.26 page 236]). The manifold |∆| is
orientable if and only if Hn (|∆|, Z) ∼
= Z.
Proposition 7.22. The manifold |∆| is orientable if and only if the
dualizing functor D n (Z) is the constant P (∆)op -module Z.
Proof. Assume that |∆| is orientable. The proposition 5.8 gives
(D(Z), Z) ∼
op
Ext0 op P (∆) −mod = TorP (∆) (Z, Z). n

The orientability of |∆| is equivalent to TorPn (∆) (Z, Z) ∼

op
= Z hence
HomP (∆)op −mod (D n (Z), Z) ∼
= Z. The previous proposition shows that
n
D (Z) takes morphisms to isomorphisms and equal to Z on objects.
If we fix a generator for D n (Z)(x) = Z for each object x then the
isomorphism D n (Z)(y) = Z → D n (Z)(x) = Z (whenever it exists a
morphism between x and y) is either Id or −Id. The isomorphism
HomP (∆)op −mod (D n (Z), Z) ∼
=Z
32 RAMZI KSOURI

shows that we can choose a natural transformation N : D n (Z) → Z

such that Nx : D n (Z)(x) → Z(x) is either Id or −Id. This is not
possible if HomP (∆)op −mod (D n (Z), Z) = 0 in the case where the manifold
is not orientable. The natural transformation N is an isomorphism and
generates HomP (∆)op −mod (D n (Z), Z). Conversely if D n (Z) ∼
= Z then the
duality isomorphism ExtP (∆)−mod (Z, −) ∼
i P (∆) n
= Torn−i (D (Z), −) implies
Torn (Z, Z) ∼
P (∆)
= Z. 
Proposition 7.23. Let X a cohomological, compact and triangulated
manifold of dimension n. If X is orientable then
H∗ (X, Z) ∼
= H n−∗ (X, Z).
Proof. The manifold |∆| is orientable so the dualizing functor D n (Z)
is the constant P (∆)op - module Z. We obtain
(Z, Z) ∼
P (∆)
ExtiP (∆)−mod = Tor n−i (Z, Z).
We thus find the classical Poincaré duality. 

7.5. Fundamental group of an aspherical manifold.

Definition 7.24. A manifold X is said aspherical if the homotopy
groups πk (X) are trivial for k > 1. Equivalently, X is an Eilenberg-
MacLane space of type K(π1 (X); 1).
The aim of this section is to prove in our setting that the first fun-
damental group of an aspherical compact and triangulated manifold is
a Poincaré duality group.
7.5.1. Preliminaries. Let ϕ : C → D a functor between two small
categories. The functor ϕ induces a functor ϕop : C op → D op . Let
Resϕ : D−mod −→ C−mod (resp. Resϕop : D op −mod −→ C op −mod)
the functor precomposition by ϕ (resp. ϕop ).
Lemma 7.25. The functors Resϕ and Resϕop have a left adjoints. We
denote LKϕ (resp. LKϕop ) the left adjoint of Resϕ (resp. Resϕop ).
Proof. We make use of lemma 3.13 to construct LKϕ as follows. Define
LKϕ on G by LKϕ (PaC ⊕ PaC ) = Pϕ(a) D D
⊕ Pϕ(a) then for F = colim Fi
with Fi ∈ ObG, define LKϕ (F ) := colim LKϕ (Fi ). The functor LKϕ is
a left adjoint to Resϕ . Indeed for any object G of CT −mod we have a
natural isomorphism
HomD−mod (LKϕ (F ), G) ∼ = HomC−mod (F, Resϕ (G))
obtained as follows:
HomCT −mod (LKϕ (colim Fi ), G) ∼
= lim HomCT −mod (LKϕ (Fi ), G)
∼
= lim HomC−mod (Fi , Resϕ (G))
∼
= HomC−mod (colim Fi , Resϕ (G))
∼
= HomC−mod (F, Resϕ (G)).
The construction of LKϕop is similar. 
 DUALITY CATEGORIES 33

Remark 7.26. The notation LK is an abbreviation of left Kan exten-

sion (see [Maclane1, X.3]).
Properties 7.27.
(1) LKϕ (PaC ) = Pϕ(a)
D
.
(2) F ⊗D LKϕ (G) = Resϕop F ⊗C G and LKϕop (F ) ⊗D G ∼
∼ = F ⊗C
Resϕ G.
(3) The functor LKϕ is right exact. We denote L(LKϕ ) the left
derived functor of LKϕ . To calculate L(LKϕ ) of a functor F ,
we choose a projective resolution P∗ of F and we apply the
functor LKϕ . We denote Li LKϕ (F ) the n’th homology group
of L(LKϕ )F = L(LKϕ )P∗ . We have an isomorphism
Li LKϕ (F )(d) ∼
= TorC−mod (Resϕop P D , F ).
i d

(4) Let ϕ|d denotes the fiber category over d ∈ ObD whose objects
are morphisms in D of the form ϕ(a) → d. A morphism from
α β
ϕ(a) → d to ϕ(b) → d is a morphism h : a → b in C such that
β ◦ ϕ(h) = α. The fiber category d|ϕ is the category (ϕop |d)op .
Let j : ϕ|d → C the functor that assigns to a morphism ϕ(a) →
d the object a of C and is the identity on morphisms of ϕ|d. We
keep the same notation for the analogue functors on d|ϕ and
ϕop |d. We have an isomorphism
Li LKϕ (F )(d) ∼
ϕ|d−mod
= Tor (Z, Resj F ).
i

Proof.
(1) This is a consequence of the isomorphisms
HomD−mod (LKϕ PaC , G) ∼
= HomC−mod (PaC , Resϕ G) ∼ D
= HomD−mod (Pϕ(a) , G).
(2) Since the tensor products and LKϕ commute with colimits, it’s
sufficient to prove the first statement for G = PaC , which is a
direct consequence of the first property and the isomorphisms
(3). The proof for the second statement is analogue.
(3) The second property implies the isomorphisms
L(LKϕ P∗ )(d) ∼= P d ⊗D LKϕ P∗ ∼
D = Resϕop P d ⊗C P∗ . D

The statement follows by taking homology.

(4) we have an isomorphism
M
Resj P C ∼
ϕ|d
= x P ϕ(x)→d
ϕ(x)→d

given by the map associating to a morphism

f
h ∈ Resj PxC (ϕ(a) → d) = Z[HomC (x, a)]
the same morphism h as an element of
ϕ|d f
P f ◦ϕ(h)
(ϕ(a) → d).
ϕ(x) → d
34 RAMZI KSOURI

This isomorphism and the isomorphisms (3) imply

LKϕ (P C )(d) ∼
x = Z ⊗ϕ|d Resj P C . x

Since LKϕ and Resj commutes with colimits (Resj is an exact

functor), we obtain
LKϕ (F )(d) ∼
= Z ⊗ϕ|d Resj F
for any functor F of C−mod. The functor Resj takes a standard
projective to a direct sum of projectives and commutes with
colimits then Resj preserves projectives. The statement follows
by applying the last isomorphism to a projective resolution of
F and applying homology.

We have LKϕ (PaC ) = Pϕ(a)
D
. It follows that
M
LKϕ BnC ∼
= D
Pϕ(x n)
.
x0 →···→xn ∈Nn C

Let d ∈ Ob(D). To give a chain x0 → · · · → xn in Nn (ϕ|d) is equiv-

alent to give a chain in C and a morphism in HomD (ϕ(xn ), d). This
correspondence induces an isomorphism
LKϕ B∗C := Z[N∗ (ϕ|−)].
Assume that the categories ϕ|d are contractible for every d ∈ Ob(D).
Then the complex LKϕ B∗C is a projective resolution of the constant D-
module Z. The left Kan extension LKϕ sends a resolution of type F P
of the constant C-module Z to a resolution of type F P of the constant
D-module Z.
Let F be a D-module. The adjunction isomorphism
HomD−mod (LKϕ BC , F ) ∼
= HomC−mod (BC , Resϕ F )
∗ ∗

induces by taking cohomology the isomorphism

Ext∗ (Z, F ) ∼
D−mod = Ext∗ (Z, Resϕ F ).
C−mod

For F = PdD we obtain the isomorphism

Ext∗
D−mod (Z, P D ) ∼
= Ext∗
d
D
C−mod (Z, Resϕ Pd ).

Suppose C is a duality category then

Exti (Z, P D ) ∼
D−mod = TorC−mod (D, Resϕ P D ),
d n−i d

where D is the dualizing C-module of C and n the projective dimension

of Z. As a consequence of properties 7.27, we obtain the following
isomorphisms
TorC−mod (D, Resϕ Pd ) ∼
n−i = Ln−i LKϕop (D)(d)
∼ ϕop |d−mod
= Torn−i (Z, Resj D)
∼
= Tor
d|ϕ−mod
(Resj D, Z).
n−i
 DUALITY CATEGORIES 35

We summarize the above discussion in the following proposition

Proposition 7.28. Let ϕ : C → D a functor between two small cat-
egories. Assume that C is a duality category with dualizing functor D
with projective dimension n and that the fibers ϕ|− are contractible. If
d|ϕ−mod
Torn−i (Resj D, Z) = 0 for i 6= n then D is a duality category.
7.5.2. Application. Let πe1 (C) the groupoid obtained from C by invert-
ing all morphisms. The category πe1 (C) is equivalent to the automor-
phism group of one of its objects which we denote π1 (C). This group
is isomorphic to the fundamental group of |C| (see [Quillen]). Suppose
that |C| is K(π1 (C); 1) then the canonical functor π : C → π1 (C) (we
consider π1 (C) as a category with one object •) is an homotopy equiv-
alence and the fibers π|• and •|π are contractible. Moreover, suppose
that C is the simplicial poset associated to a triangulation of a coho-
mological compact orientable manifold then the dualizing functor D
is the constant C-module Z (proposition 7.20). In this case, we have
•|π
Torn−i (D, Z) = 0 for i 6= n and equal to Z for i = n. It follows that
π1 (C) is a Poincaré duality group (a special case of [Brown, example 1
page 222]).
References
[AB08] P. Abramenko & KS. Brown - « Buildings Theory and Applica-
tion », Springer. (2008).
[Borceux] F. Borceux - « Handbook of Categorical Algebra 1 », Cambridge
University Press. (1994).
[Brown] KS. Brown - « Cohomology of groups », Graduate Texts in Math-
ematics. Springer-Verlag New York. 87 (1982).
[BE73] R. Bieri & B. Eckmann - « Groups with homological duality gen-
eralizing Poincaré duality », Invent Math. 20 (1973), 103-124.
[Borel] A. Borel & al - « Algebraic D-Modules », Perspectives in Math.
Academic Press. Florida. (1987).
[Chałupnik] M. Chałupnik - « Koszul duality and extensions of exponential
functors », Adv. Math. 218 (2008), no. 3, 969-982.
[FS97] EM. Friedlander & A. Suslin - « Cohomology of finite group
schemes over a field », Invent. Math. 127 (1997), no. 2, p. 209-
270.
[Grothendieck] A. Grothendieck - « Sur Quelques Points d’Algèbre Homologique »,
Tôhoku Mathematics Journal. (1957), Vol 9, p. 119-221.
[Hatcher] A. Hatcher - « Algebraic topology », Cambridge University Press.
(2002).
[Krause] H. Krause - « Koszul, Ringel and Serre duality for strict polynomial
functors », Compositio Mathematica. 149 (2013), p. 996-1018.
[Maclane1] S. Mac Lane - « Categories for the working mathematician »,
Springer-Verlag. (1991).
[Maclane2] S. Mac Lane - « Homology », Springer-Verlag. (1975).
[May] JP. May - « Simplicial objects in algebraic topology », The univer-
sity of Chicago Press. (1967).
[Munkres1] JR. Munkres - « Topological results in combinatorics », Michigan
Math. J. 31 (1984), no. 1, p. 113-128.
36 RAMZI KSOURI

[Munkres2] JR. Munkres - « Elements of Algebraic Topology », Addison-

Wesley. (1984).
[Pirashvili] T.Pirashvili - « Introduction to functor homology », Rational repre-
sentation, the Steenrod algebra and functor homology. Panoramas
et synthèses. 16 (2003).
[Quillen] D. Quillen - « Higher algebraic K-theory I », Lecture Notes in Math.
Springer. Berlin. 341 (1973), .
[RV02] I. Reiten & M. Van Den Bergh - « Noetherian hereditary category
satisfying Serre duality », Journal of the American Mathematical
Society. 15 (2002), no. 2, p. 295-366.
[Touzé1] A. Touzé - « Ringel duality and derivatives of non-additive func-
tors », Journal of Pure and Applied Algebra. 217 (2013), p. 1642-
1673
[Touzé2] A. Touzé - « Bar complexes and extensions of classical exponential
functors », arXiv:1012.2724v4 [math.RT]. (2013).
[Weibel] CA. Weibel - « An introduction to homological algebra », Cam-
bridge University Press. (1994).
[Fei’s Book] F. Xu - « Lecture notes on representations and cohomology of cate-
gories », http://www.mat.uab.cat/~xu/CatRepCoh.html. (2011).

Laboratoire de Mathématiques: Algèbre, Topologie, Arithmétique

et Structures ordonnées "LATAO", Département de Mathématiques,
Faculté des Sciences de Tunis, Le Campus Universitaire TN-2092, Tu-
nis, Tunisie.
E-mail address: ksouri.r@gmail.com