Lecture Notes on Sheaves, Stacks, and Higher Stacks

Adrian Clough

September 16, 2016

2
Contents

Introduction - 26.8.2016 - Adrian Clough i

1 Grothendieck Topologies, and Sheaves: Definitions and the Closure Property

of Sheaves - 2.9.2016 - Neža Zager 1
1.1 Grothendieck topologies and sheaves . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.1 Presheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.2 Coverages and sheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.3 Reflexive subcategories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.4 Coverages and Grothendieck pretopologies . . . . . . . . . . . . . . . . . . 4
1.1.5 Sieves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.1.6 Local isomorphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.1.7 Local epimorphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.1.8 Examples of Grothendieck topologies . . . . . . . . . . . . . . . . . . . . . 13
1.2 Some formal properties of categories of sheaves . . . . . . . . . . . . . . . . . . . 14
1.3 The closure property of the category of sheaves on a site . . . . . . . . . . . . . . 16

2 Universal Property of Sheaves, and the Plus Construction - 9.9.2016 - Kenny

Schefers 19
2.1 Truncated morphisms and separated presheaves . . . . . . . . . . . . . . . . . . . 19
2.2 Grothendieck’s plus construction . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3 Sheafification in one step . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3.1 Canonical colimit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3.2 Hypercovers of height 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.4 The universal property of presheaves . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.5 The universal property of sheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.6 Sheaves on a topological space . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

3 Categories Fibered in Groupoids - 16.9.2016 - Sebastian Schulz 23

4 Glueing and Descent Data for Stacks - 23.9.2016 - Nina Otter 25

5 Stacks - 30.9.2016 - Jonathan Lai 27

5.1 The power of negative thinking: (−1)-stacks . . . . . . . . . . . . . . . . . . . . . 27

6 Global Model Structures on Simplicial Presheaves - 7.10.2016 - Adrian Clough 29

7 Introduction to Derived Algebraic Geometry - 18.11.2016 - Rustam Antia 31

3
A Categorical Background 33

Bibliography 35

4
Introduction - 26.8.2016 - Adrian
Clough

Sheaves

A category is a world of objects, all looking at each other. Who you are in a category is
completely determined by how you are viewed by all objects, including yourself. This is made
precise by Yoneda’s lemma. Let C be a (small) category, and denote by C
b := Cat(Cop , Set) the
category of presheaves on C.
Theorem (Yoneda’s Lemma). For any F ∈ C b and X ∈ C the map which assigns to each
x ∈ F (X) the natural transformation
( )
C(Y, X) → F (Y )
f 7→ F (f )(x)
Y ∈C

is a bijection
F (X) ∼
= C∧ (C( , X), F ),
with the inverse given by sending any natural transformation η : C( , X) ⇒ F to ηX (idX ).

As a corollary we obtain that the functor C → C b sending any object X ∈ C to the presheaf
Y 7→ C(Y, X) is fully faithful; we therefore identify objects in C by the presheaves they represent.
Now, Yoneda’s lemma tells us even more; it tells us that, in addition to the objects in C, all
other presheaves on C are determined by how they are seen by objects in C. This parallels how,
when we glue together some simple geometric objects to more complicated geometric objects,
these complicated geometric objects are completely determined by how they are seen by the
simple ones:
Example. Denote by Cart∞ the category of open subsets of Euclidean space and smooth
maps, and by Man∞ the category of smooth manifold and smooth maps, then the embed-
ding Cart∞ ,→ Man∞ induces a functor Man \∞ → Cart \∞ by precomposition; compos-
ing the Yoneda embedding Man∞ ,→ Man \∞ with Man \∞ → Cart \∞ , we obtain a functor
Man∞ → Cart \∞ , which can again be shown to be fully faithful.
Similarly, let k be a commutative ring; denote by Aff k the category of affine k-schemes, and by
Schk the category k-schemes, then by composing the Yoneda embedding Schk ,→ Sch [k with the
[k → Aff
functor Sch [k , we obtain a fully faithful functor Schk → Aff
[k . y
Further pursuing this line of thought, note that glueing just means taking certain colimits,
and that all presheaves on C are obtained as colimits of objects in C (see Theorem 1.1.1.1);

i
even better, C b is the universal cocomplete category generated under colimits by objects in C
(see Corollary 2.4.0.2). Thus, if we think of the objects in C as simple or primitive geometric
objects that we would like to glue together to more complicated ones, then it is reasonable to
assume these more complicated objects should live in C. b
Unfortunately, the category C suffers from a serious defect: Usually we already know how
b
to obtain simple objects by glueing together other simple objects; e.g. in Cart∞ an open
interval may be obtained by glueing together two smaller overlapping intervals. More generally,
when we say that we are glueing together an object U ∈ C from objects Ui → U in CU , we
` `
typically mean that U is the colimit of the diagram i,j Ui ×U Uj ⇒ i Ui (think of glueing
together an open subset of Euclidean space from an atlas, or glueing together affine schemes
from open affine subschemes). The datum specifying which such diagrams should constitute
glueing diagrams is essentially what constitutes a Grothendieck topology. In C b the diagram
i,j C( , Ui ×U Uj ) ⇒ i C( , Ui ) → C( , U ) is however no longer a colimit in all but the most
` `

trivial cases.
There are at least two ways in which we could try to remedy these defects: We could localise C b
along a suitable class of morphisms, forcing the cocones specified by the Grothendieck topology
to become colimits. Dually, we could try to restrict to the full subcategory of C b spanned by
objects to which the cocones in question (co-)look like colimits; in detail, let F be a presheaf
on C, then the diagram
a  a 
C C( , Ui ×U Uj ), F ⇔ C C( , Ui ), F ← Cb C( , U ), F


b b
i,j i

is isomorphic to Y Y
F (Ui ×U Uj ) ⇔ F (Ui ) ← F (U )
i,j i

and we could consider the full subcategory of C b consisting of presheaves F such that the above
two isomorphic diagrams are limits. This is the subcategory of sheaves on C and is denoted by
C.
e Amazingly, these two notions coincide!
The inclusion C e ,→ C
b has a left adjoint, and this is exactly the localisation functor suggested as
the first possible remedy. Mercifully, the category Ce is still cocomplete, and is moreover universal
with respect to being generated under colimits by objects in C, while keeping the colimits of the
form discussed above (see Theorem 2.5.0.1). We have thus constructed the universal “container
category” for any objects one might like to glue together from objects in C according to the
rules specified by the Grothendieck topology. More classically, e.g. in the case of schemes, one
considers more ad hoc container categories, such as locally ringed spaces. Also one can show
that iterating this process does not produce any new spaces; there is a natural way of extending
the Grothendieck topology on C to C, e and the colimits obtained from objects C e can already be
obtained as colimits of objects in C (see §1.3).

Example. Most of the known generalisations of smooth manifolds are sheaves on Cart∞ .
In order of inclusion we have: smooth manifolds ⊂ Fréchét manifolds ⊂ Frölicher spaces ⊂
diffeological spaces ⊂ sheaves on Cart∞ . Moving to the right, the objects we consider become
less and less tangible, while the formal properties of the categories they span become increasingly
good. y

ii
Chapter 1

Grothendieck Topologies, and

Sheaves: Definitions and the Closure
Property of Sheaves - 2.9.2016 -
Neža Zager

Throughout this chapter C denotes a category, which can usually be assumed to be (essentially)
small.
In §1.1 we motivate and state various equivalent definitions of Grothendieck topologies and
(categories of) sheaves. We would like to think of C as a category of simple or primitive spaces,
which we hope to glue together to more complicated ones. As explained in the introduction,
these new spaces should be determined by how they are seen by the objects in C; by Yoneda’s
lemma this is true for objects in C,
b and the examples considered in the introduction show that
manifolds (and several generalisations thereof), and schemes embed naturally into Cart \∞ and
[k respectively. In §1.1.1 we give a more useful reformulation of how objects in C
Aff b are deter-
mined by how they are seen by objects in C. The subsections §1.1.2 and §1.1.3 then provide
the heart of §1.1: We study how to glue together objects in C from other objects which cover
them, but realise that the colimits involved in formalising this notion of glueing do not commute
with the Yoneda embedding C ,→ C. b We can attempt to fix this problem by either formally
inverting a certain class of morphisms in C,
b or we can restrict to the full subcategory C
e spanned
by objects in Cb to which the class of morphisms under consideration already look invertible. In
§1.1.3 we will see that the inclusion Ce ,→ C
b has a left adjoint, which is exactly the localisation
functor alluded to above, so that the two proposed fixes are equivalent. In §§1.1.5 - 1.1.7 we
are naturally led to consider several equivalent notions of Grothendieck topologies1 . We finish
section §1.1 with some examples.
In §1.2 we discuss two important formal properties of categories of sheaves: Suitable subcate-
gories of C
b are in bijection with Grothendieck topologies on C, and these subcategories inherit
Cartesian closedness from C. b
Finally, in §1.3 we make precise the notion that spaces that can be glued together from sheaves
on C can already be glued together from objects in C and give a proof thereof.

1
The proof that they are indeed equivalent will hopefully be added at some future point.

1
1.1 Grothendieck topologies and sheaves
1.1.1 Presheaves
Yoneda’s lemma tells us among many other things that presheaves on C are determined by how
they are seen by objects in C. We re-express this fact in terms of colimits.
Theorem 1.1.1.1. [Mac98, Th. I.7.1] Let F be a presheaf on C. Denote by CF the full subcat-
egory of CbF spanned by objects X → F with X ∈ C. The obvious cocone over CF → C b with F
at its apex is a colimit.
Note that the theorem is trivially true for any object X ∈ C because CX has a final object.

1.1.2 Coverages and sheaves

In this subsection we discuss the data needed to specify how to glue together objects in C to
other objects in C, and the failure of this data to be compatible with the Yoneda embedding.
Definition 1.1.2.1. Let U be an object in C, then a covering of U is a (small) family of
morphisms in C with U as their codomain. y
In the following we assume that the pullback of any two morphisms in a cover exists (this will
be true in the examples covered in the seminar). It is straightforward to modify the theory so
that this condition is no longer necessary (see [Joh02b, §C.2.1]), but at a first pass the key ideas
are more transparent with this extra assumption. Furthermore we assume that the coproduct
of the objects in any covering exists as well as the coproduct of the set of all pullbacks of the
morphisms in a covering.
Definition 1.1.2.2. Let U be an object in C, and let {Ui → U }i∈I be a covering of U , then
the glueing diagram2 associated to {Ui → U }i∈I is the diagram
a a
Ui ×U Uj ⇒ Ui → U,
(i,j)∈I×I i∈I

where the upper of the parallel arrows is obtained from the projection onto the first objet in
the pullback diagrams for all (i, j) ∈ I × I, and the lower arrow is obtained from the second
projection. y
In the examples Cart∞ and Aff k , if we take a covering of an open subset of Euclidean
space U or an affine k-scheme S to be literally a covering of U by open subsets or by open affine
k-subschemes respectively, then the associated glueing diagram is colimiting3 . Unfortunately
the diagram a a
C( , Ui ×U Uj ) ⇒ C( , Ui ) → C( , U ) (1.1)
| {z }
(i,j)∈I×I ∼ i∈I
=
C( ,Ui )×C( ,U ) C( ,Uj )

in C
b is no longer a colimiting diagram in all but the most trivial cases4 .
There are at least two things we could do to try to fix this: If for every U ∈ C and every
2
Warning, this terminology is nonstandard.
3
In more sophisticated categorical
`language: The pullbacks and colimits under consideration commute with
each other, so we are asserting that i Ui → U is an effective epimorphism.
4
Even if the glueing diagram in C is not a colimit, we might still want to force it to be on in C.
b

2
covering U of U we denote by CU the colimit of the glueing diagram of U in C,b then we could
try to formally invert the canonical morphisms CU → C( , U ) as we range over the coverings U.
Otherwise we can try to restrict to the subcategory of C
b spanned by those presheaves, to which
the diagrams of the form (1.1) look like colimits.

Definition 1.1.2.3. Let U be an object of C, and let {Ui → U }i∈I be a covering of U , then a
presheaf F : Cop → Set satisfies the sheaf condition for {Ui → U }i∈I if the isomorphic diagrams
 a  a 
C C( , Ui ×U Uj ), F ⇔ C C( , Ui ), F ← C
b C( , U ), F


b b
(i,j)∈I×I i∈I

and Y Y
F (Ui ×U Uj ) ⇔ F (Ui ) ← F (U ),
(i,j)∈I×I i∈I

are limiting diagrams. y

Definition 1.1.2.4. Let U be an object of C, {Ui → U }i∈I a covering of U , and F : Cop → Set a
presheaf, then a matching family for {Ui → U }i∈I of elements in F is a sequence (si ∈ F (Ui ))i∈I ,
such that for each pair (i, j) ∈ I × I we have F (Ui ×U Uj → Ui )(si ) = F (Ui ×U Uj → Uj )(sj ).
The set of matching families for {Ui → U }i∈I of elements in F is denoted by

Desc({Ui → U }i∈I , F ).

With notation as in Definition 1.1.2.4 we see that Desc({Ui → U }i∈I , F ) is simply the limit
Q Q
of the diagram i,j F (Ui ×U Uj ) ⇔ i F (Ui ), and that F satisfies the sheaf condition for
{Ui → U }i∈I iff for every matching family (si ∈ F (Ui ))i∈I there exists exactly one element
s ∈ F (U ) such that for all i ∈ I we have F (Ui → U )(s) = si .

1.1.3 Reflexive subcategories

Let B be a category, A a subcategory of B, and Σ a class of morphisms in B. The situation
considered in the last section, in which for any object X ∈ A and any morphism f : A → B in
Σ the map B(B, X) → B(A, X), g 7→ g ◦ f is a bijection will now be studied systematically:

Definition 1.1.3.1. The subcategory A is refletive if the inclusion functor A ,→ B admits a

left adjoint. The left adjoint is called the reflection of A. y

Remark 1.1.3.2. It is readily verified that a left adjoint of A ,→ B may be chosen such that it
restricts to the identity on A. y

Proposition 1.1.3.3. [Bor94a, Prop. 5.4.4] Assume A is reflective in B; denote the reflection
by R and the unit of the adjunction by η : idB ⇒ R. Assume furthermore that Σ consists
precisely of the morphisms in B which are inverted by R.

(1) For an object X ∈ B the following are equivalent:

(I) X ∈ A;
(II) for all f : A → B in Σ the map B(B, X) → B(A, X), g 7→ g ◦ f is a bijection;

3
 (III) for all B ∈ B the map B(R(B), X) → B(B, X), g 7→ g ◦ ηB is a bijection.

(2) For a morphism f in B the following are equivalent:

(I) f : A → B is in Σ;
(II) for all X ∈ A the map B(B, X) → B(A, X), g 7→ g ◦ f is a bijection.

Theorem 1.1.1.1 tells us that in a precise sense the category C b is generated by a small
category, namely C. Such categories are so-called locally presentable categories (see [Bor94b,
Def. 5.2.1]). This allows us to apply a partial converse of Theorem 1.1.3.3:

Theorem 1.1.3.4. Assume B is locally presentable and that A corresponds exactly to the objects
X such that for all f : A → B in Σ the map B(B, X) → B(A, X), g 7→ g ◦ f is a bijection,
then the subcategory A is reflective, and the reflector of A is the localisation of B along Σ.

Note on the proof. The construction of the reflector is essentially an application of the small
object argument (see [Gar09]). For a full proof put together [Bor94b, Ex. 5.5.5.b, Prop. 5.2.10]
and [Bor94a, Th. 5.4.7].

Explicitly, Theorem 1.1.3.4 tells us that if we are given a family P of coverings on the objects
in C, then restricting to the subcategory of C b spanned by presheaves which satisfy the sheaf
condition for every covering, or taking the localisation of Cb along the morphisms5 CU → C( , U )
for covering U in P produce equivalent categories.
Note that the class of morphisms in C b corresponding to Σ in Proposition 1.1.3.3 generally
contains more morphisms than the morphisms CU → C( , U ) for every covering U of U in P .

Definition 1.1.3.5. Using the same notation as above the localisation of C b along the mor-
phisms CU → C( , U ) for every covering U in P is called the P -sheafification functor, or simply
sheafification functor, when P is understood from context.
The image of any presheaf under the P -sheafification functor is referred to as its P -sheafification,
or simply sheafification. y

We will see some concrete constructions of this left adjoint Chapter 2.

1.1.4 Coverages and Grothendieck pretopologies6

Consider a morphism f : X → Y between manifolds (resp. schemes), then we can find atlases
(resp. coverings by open affine subschemes) such that f restricts to morphisms between open
subsets of Euclidean space (resp. affine schemes), and furthermore f is completely determined
by these restrictions. This is in particular true if X and Y are themselves open subsets of
Euclidean space (resp. affine schemes). That maps between spaces are determined locally in
this sense is fundamental in geometry, which is why we axiomatise this property in the following
definition7 . The notion of a Grothendieck topology given in §1.1.5 will be obtained by adding
closure axioms to the axiom defining coverages.
5
using the same notation as in the previous subsection
6
Proofs in this subsection will hopefully be added later.
7
The axiom will manifest itself in a surprising way in Theorem 1.2.0.4.

4
Definition 1.1.4.1. [Joh02b, Def. C.2.1.2.] A coverage T on C is a collection of coverings of
objects in C, (where for each object U ∈ C the collection of coverings of U in T is denoted by
TU ,) such that

(C) for each object U ∈ C and covering {fi : Ui → U }i∈I in TU we have that for any morphism
g : V → U in C there exists a covering {hj : Vj → V }j∈J in TV such that for each j ∈ J
there exists an element i ∈ I such that the map g ◦ hj factors through fi .

Assume C is equipped with a coverage T , and let g : V → U , {fi : Ui → U }i∈I ,

{hj : Vj → V }j∈J be as in Definition 1.1.4.1, then if for each j ∈ J we choose an element
i ∈ I and a factorisation of fi through g ◦ hj , we obtain by the universal property of coproducts
` `
a morphism g̃ : j∈J Vj → i∈I Ui making the bottom square in (1.2) commute. Similarly, g̃
determines a morphism g̃˜ : (j1 ,j2 )∈J×J Vj1 ×V Vj2 → (i1 ,i2 )∈I×I Ui1 ×U Ui2 making the upper
` `

squares in (1.2) commute containing only the left arrows and the right arrows of the vertical
pairs respectively. In this sense g is completely determined by g̃.

a g̃˜ a
Vj1 ×V Vj2 Ui1 ×U Ui2
(j1 ,j2 )∈J×J (i1 ,i2 )∈I×I

a g̃ a (1.2)
Vj Ui
j∈J i∈I

g
V U
Definition 1.1.4.2. Let T be a coverage on C, then a presheaf Cop is called a T -sheaf, or simply
a sheaf if T is understood from context, if F satisfies the sheaf condition for for every covering
given by T .
The subcategory of C b of T -sheaves is denoted by C
eT , or simply by C
e if T is clear from context.
y

Assume we are given coverage T on C, then any T -sheaf may satisfy the sheaf condition for
additional coverings not in T . We shall proceed to systematically investigate this matter.

Lemma 1.1.4.3. Let U be an object in C, then every presheaf on C satisfies the sheaf condition
id
for the covering of U consisting only of the identity morphism U −
→ U.

Lemma 1.1.4.4. [Joh02b, Lm. C.2.1.6.i] Assume that C is equipped with a coverage T . Let U
be an object in C, {Ui → U }i∈I a covering in TU , and {Vj → U }j∈J a covering, not necessarily
in TU , such that for each i ∈ I there exists a j ∈ J such that Ui → U factors through Vj → U ,
then any T -sheaf satisfies the sheaf condition for {Vj → U }j∈J .

Lemma 1.1.4.5. [Joh02b, Lm. C.2.1.6.i]8 Assume that C is equipped with a coverage T . Let U
be an object in C, {fi : Ui → U }i∈I a covering in TU , and for each i ∈ I let {hij : Uij → Ui }j∈Ji
be a covering in TUi , then any T -sheaf satisfies the sheaf axiom for the covering {fi ◦ hij : Uij →
U }i∈I,j∈Ji .
8
The lemma as sated in [Joh02b] is incorrect. See [Low].

5
Lemma 1.1.4.6. Assume C admits pullbacks, and that it is equipped with a coverage T . Let
U be an object in C, and {Ui → U }i∈I a covering in TU . For any morphism V → U in C any
T -sheaf satisfies the sheaf condition for the covering {V ×U Ui → V }i∈I .

Proof. Exercise.

We are now ready to give the classical definition of a Grothendieck pretopology.

Definition 1.1.4.7. Assume C admits pullbacks. A Grothendieck pretopology or simply a

pretopology τ on C is a collection of coverings of objects in C, (where for each object U ∈ C the
collection of coverings of U in τ is denoted by τU ,) such that
id
(I) For each object U ∈ C the identity morphism U −
→ U is in τU .

(Cp ) For each morphism V → U in C and each covering {Ui → U }i∈I in τU the covering
{V ×U Ui → V }i∈I is in τV .

(L) For each object U ∈ C, if {fi : Ui → U }i∈I is a covering in τU and for each i ∈ I the
covering {hij : Uij → Ui }j∈Ji is in τUi , then {fi ◦ hij : Uij → U }i∈I,j∈Ji is in τU .

By the axiom (Cp ) any grothendieck pretopology is a coverage.

Proposition 1.1.4.8. Assume C admits pullbacks. For any coverage T on C, the intersection of
all Grothendieck pretopologies containing T is again a Grothendieck pretopology with the same
sheaves as T .

Proof. The intersection τ of all Grothendieck pretopologies containing T is again a Grothendieck

pretopology, and is contained in the intersection Te of all Grothendieck topologies containing T .
As C
eT ⊆ Cee ⊆ C
T
eτ , and C
eT = Ceτ by Proposition 1.1.5.179 we have C eT = Ce e.
T

1.1.5 Sieves
In the previous section in Lemmas 1.1.4.3 - 1.1.4.5 we studied, given a coverage T on C, for which
coverings not in T , any T -sheaf nonetheless satisfies the sheaf condition. In this subsection we
will study how can add morphisms to a given covering without affecting which presheaves satisfy
the sheaf condition.

Definition 1.1.5.1. Let U be an object in C, then a covering U of U is called a sieve if for any
morphism f : V → U in U and any morphism g : W → V the composition f ◦ g : W → U is in
U. y

Any covering generates a sieve in an obvious way.

Remark 1.1.5.2. With notation as in Definition 1.1.5.1, if U is a sieve, then we could define a
simpler, “sifted” glueing diagram: Viewing U as a subcategory of CU , we can simply compose
the functors U → CU → C. If U is an open subset of Euclidean space or an affine scheme, and
U is the sieve generated by a covering by open subsets or open affine subschemes respectively,
9
We note that, although Proposition 1.1.5.17 is the next section, it does not logically depend on this propo-
sition.

6
then the colimit of the described diagram is again U . We will see in this section that we can
replace coverings by sieves, so that the discussion motivating sheaves in §1.1.4 could have been
held with sieves and sifted glueing diagrams. y

Lemma 1.1.5.3. Let U be an object in C, then sieves on U correspond bijectively to subfunctors

of C( , U ): any sieve S on U corresponds to the subfunctor S ⊆ C( , U ) which sends any object
V ∈ C to those morphisms V → U which lie in S.

Notation 1.1.5.4. For any object U ∈ C and any sieve on U , we will use the same symbol to
denote the sieve and the corresponding subfunctor of C( , U ), as well as full subcategory of CU
spanned by morphisms in S. y

Notation 1.1.5.5. Let f : V → U be a morphism in C, then for any sieve S ⊆ C( , U ) we

denote the pullback of S ,→ C( , U ) along f by f ∗ S. y

Note that for any sieve S on any object U ∈ C, the pullback of S along any morphism
f : V → U is given by all morphisms g : W → V such that f ◦ g ∈ S.

Lemma 1.1.5.6. Let U be an object in C, U a covering of U , and S the sieve generated by U,

then for any presheaf F : Cop → Set the the restriction of any element (sf ) ∈ lim(f :V →U )∈S F (V )
←−
to (sf )(f :V →U )∈U is a matching family for U, and for any matching family (tf )f ∈U , setting
tf ◦g = g(tf ) for every g : W → Domf yields a sequence in lim(f :V →U )∈S F (V ). These two maps
←−
are inverse to each other, yielding a bijection

Desc(U, F ) ∼
= lim F (V ).
←−
(f :V →U )∈S

Lemma 1.1.5.7. Let U be an object in C, U a covering of U , and S the sieve generated by U,

then any presheaf on C satisfies the sheaf condition for U iff it satisfies the sheaf condition for
S.

Proof. Let F : Cop → Set be a presheaf. By the previous lemma we have the following chain of
canonical bijections:

F (U ) ∼
= Desc(U, F ) ∼
= lim F (V ) ∼
= Desc(S, F ).
←−
(f :V →U )∈S

We thus see that we could replace coverings by sieves and define sheaves as presheaves which
take colimits to sifted glueing diagrams to limits.

Lemma 1.1.5.8. Let U be an object in C, and S a covering of U . A presheaf F : Cop → Set

satisfies the sheaf condition for S iff the map

C(C(
b , U ), F ) → C(S,
b F)

given by precomposition with S ,→ C( , U ) is a bijection.

7
 Thus F satisfies the sheaf condition for S iff every morphism S → F descends uniquely to
a morphism C( , U ) → F .

Proof. Firstly, C(C(

b , U ), F ) ∼
= F (U ) by Yoneda’s lemma. Next, we obtain the chain of isomor-
phisms
b F) ∼
C(S, = C(
b lim C( , V ), F )
−→
(V →U )∈S
∼
= lim C(C(
b , V ), F )
←−
(V →U )∈S
∼
= lim F (V ),
←−
(V →U )∈S

where the first isomorphism follows from Theorem 1.1.1.1, and the last isomorphism from
Yoneda’s lemma. Finally, we note that the square

F (U ) lim F (V )
←−
(V →U )∈S
∼
= ∼
=

C(C(
b , U ), F ) C(S,
b F)

commutes by chasing around any element s ∈ F (U ):

s (F (f )(s))f ∈S

(f 7→ F (f )(s))f ∈CU (f 7→ F (f )(s))f ∈S .

Remark 1.1.5.9. If we are given a collection P of coverings in C, the P -sheafification functor

can thus be obtained as the localisation along the morphisms in Cb given by first replacing any
cover of any object U ∈ C in P by the sieve S it generates, and then considering the canonical
morphism S ,→ C( , U ). y
The following three lemmas are analogues of Lemmas 1.1.4.3 - 1.1.4.5 respectively.

Lemma 1.1.5.10. Let U be an object in C, then every presheaf on C satisfies the sheaf condition
w.r.t. the sieve generated by the identity morphism U → U .

Lemma 1.1.5.11. [Joh02b, Lm. C.2.1.6.ii] Assume that C is equipped with a coverage T . Let
U be an object in C, and S a sieve in TU , then for any sieve S 0 on U containing S any T -sheaf
satisfies the sheaf condition for S 0 .

Proof. This is a special case of Lemma 1.1.4.4.

Lemma 1.1.5.12. [Joh02b, Lm. C.2.1.7.ii] Assume that C is equipped with a coverage T . Let
U be an object in C, let R, S be sieves on U with R in TU , and assume that f ∗ S is in TV for
every f : V → U in R, then any T -sheaf satisfies the sheaf condition for S.

Proof. This is an immediate consequence of Lemma 1.1.4.5 and 1.1.4.4.

8
Definition 1.1.5.13. A coverage on C is called sifted if all the coverings given by the coverage
are sieves. y

Convention 1.1.5.14. Let T be a sifted coverage on C, then if T is understood from context,

we will often refer to sieves given by T as covering sieves. y

Note that in a sifted coverage the axiom (C) translates into saying that the pullback of
any covering sieve contains a covering sieve. By Lemma 1.1.4.4 any sheaf satisfies the sheaf
condition for any pullback of any covering sieve, so we might as well assume that covering sieves
pull back to covering sieves.

Definition 1.1.5.15. A sifted coverage T on C is called a Grothendieck topology if

(Cs ) covering sieves pull back to covering sieves;

(M) for every object U ∈ C the sieve generated by the identity morphism U → U is in TU ;

(Ls ) let U be an object in C; if R, S are sieves on U with R in TU , and f ∗ S is in TV for every

f : V → U in R, then S is in TU .

Definition 1.1.5.16. A site is a category10 equipped with a Grothendieck topology.

Proposition 1.1.5.17. For any coverage T on C, the intersection of all Grothendieck topologies
containing T is again a Grothendieck pretopology with the same sheaves as T 11 .

Sketch of proof. It is easily verified that the intersection of all Grothendieck topologies contain-
ing T is again a Grothendieck topology, and that the maximal collection of sieves, for which
any T -sheaf satisfies the sheaf axiom, is a Grothendieck topology.

In the following sections, §§1.1.6 - 1.1.7, we give equivalent definitions for Grothendieck
topologies. We will refer to a category equipped with a Grothendieck topology or any of the
equivalent structures considered in these sections as a site. Also, more abusively, we will refer
to a category with a coverage or Grothendieck pretopology as a site.

1.1.6 Local isomorphisms

Assume C is equipped with a Grothendieck topology T , and view T as a subset of Mor(C).
By the discussion in §1.1.3 and Remark 1.1.5.9 we can view the category of T -sheaves C eT as
the localisation of C
b along T , and that the class or morphisms inverted by the left adjoint to
C
eT ,→ Ce is completely determined by CeT .

Definition 1.1.6.1. Assume C is equipped with a Grothendieck topology T , then a morphism in

C
b which is sent to an isomorphism by the T -sheafification functor is called a T -local isomorphism,
or simply a local isomorphism, when T is understood from context. y

Local isomorphisms can be described axiomatically.

10
It is often convenient to assume that this category is (essentially) small.
11
We will see in Theorem 1.2.0.3 that this topology is unique.

9
 Definition 1.1.6.2. A family of local isomorphisms I on C is a class of morphisms in C
b satisfying
the following three axioms:

(LI 1) Every isomorphism belongs to I.

(LI 2) The class I satisfies the two-out-of-three property.

(LI 3) The class I is stable under base change along morphisms with domain in C.

A morphism in I is referred to as an I-local local isomorphism, or simply a local isomorphism

when I is understood from context. y

Remark 1.1.6.3. By axioms (LI 1) and (LI 2) a family of local isomorphisms in fact forms a
category.
Also, it is true in general, that when we consider a functor C → D, the class of morphisms in
C which are sent to isomorphisms form a category, and that moreover these morphisms satisfy
(LI 1) and (LI 2). y

Theorem 1.1.6.4. Let T be a Grothendieck topology on C, then the class of morphisms in C b

sent to isomorphisms by the T -sheafification functor is a family of local isomorphisms.
Conversely, the subclass of a family of local isomorphisms on C consisting of monomorphisms
with codomain in C is a Grothendieck topology.
These two operations determine a bijection between Grothendieck topologies and families of local
isomorphisms on C.

Proof. Remains to be written.

1.1.7 Local epimorphisms

Definition 1.1.7.1. Assume C is equipped with a Grothendieck topology T , then a morphism in
C
b which is sent to an epimorphism by the T -sheafification functor is called a T -local epimorphism,
or simply a local epimorphism, when T is understood from context. y

Local epimorphisms can be described axiomatically.

Definition 1.1.7.2. A family of local epimorphisms E on C is a class of morphisms in C

b
satisfying the following four axioms:

(LE 1) Every identity morphism belongs to I.

(LE 2) The class I is closed under composition.

f g
(LE 3) Given any morphisms A − − C in C,
→B→ b if g ◦ f is a local epimorphism, then g is a local
epimorphism.

(LE 4) The class I is stable under base change along morphisms with domain in C.

A morphism in I is referred to as an I-local local isomorphism, or simply a local isomorphism

when I is understood from context. y

10
Remark 1.1.7.3. By axioms (LE 1) and (LE 2) a family of local epimorphisms in fact forms a
category.
Also, it is true in general, that when we consider a functor C → D, the class of morphisms in
C which are sent to epimorphisms form a category, and that moreover these morphisms satisfy
(LE 1) - (LE 3). y
The following is an important class of examples of local epimorphisms.

Proposition 1.1.7.4. Assume C is equipped with a coverage T , then for any object U ∈ C and
any covering {Ui → U }i∈I in TU , the map i∈I C( , Ui ) → C( , U ) a local epimorphism12 .
`

Proof. The image of i∈I C( , Ui ) → C( , U ) is the covering sieve generated by {Ui → U }i∈I ,
`

which is a local epimorphism, as it sent to an isomorphism, which is a fortiori an epimorphism,

by the sheafification functor.

We will require the following, auxiliary definition.

Definition 1.1.7.5. Assume C is equipped with a family of local epimorphisms E, then a

morphism A → B in Cb is called a local monomorphism if A → A ×B A is a local epimorphism.
y

Theorem 1.1.7.6. If C is equipped with a family of local isomorphisms, then the class of mor-
phisms f : A → B in C, b such that the image morphism13 Imf → B is a local isomorphism,
determines a family of local epimorphisms.
Conversely, if C is equipped with a family of local epimorphisms, then the class of morphisms
in C,
b which are both local epimorphism and local monomorphisms, forms a family of local iso-
morphisms.
These two operations determine a bijection between families of local isomorphisms and families
of local epimorphisms on C.

Proof. Remains to be written.

We can conclude from Theorem 1.1.6.4 and 1.1.7.5 that there is a bijection between Gro-
thendieck topologies and families of local epimorphisms on C. We give an explicit bijection in
the following theorem.

Theorem 1.1.7.7. If C is equipped with a family of local epimorphisms, then the collection of
subfunctors of representable presheaves on C which are local epimorphisms, form a Grothendieck
topology on C.
If C is equipped with a Grothendieck topology, then the collection of morphisms A → B such
that for all morphisms U → B with U ∈ C the morphism A ×B U → U is a covering sieve,
forms a family of local epimorphisms on C.
These two operations determine a bijection between families of local epimorphisms and Gro-
thendieck topologies on C.
12
Here the local epimorphisms are of course the ones given by the Grothendieck topology generated by the
coverage.
13
The image of a morphism of presheaves A → B is defined as the objectwise image of sets; it is left to the
reader to verify, that this indeed forms a subpresheaf of B.

11
 Using a family of local isomorphism we can define sheaves as those presheaves to which
the local isomorphisms look like isomorphisms. The reader may thus be lead to suspect that
given a family of local epimorphisms, sheaves correspond to those presheaves to which the local
epimorphisms look like epimorphisms. This is however not the case.
Proposition 1.1.7.8. Assume C is equipped with a family of local epimorphisms, then a presheaf
F : Cop → Set is a separated presheaf14 iff for every local epimorphism A → B the map
C(B,
b F ) → C(A,
b F ) is a monomorphism.
Sketch of proof. Assume F is a separated presheaf, and that B = U ∈ C, then A → U factors as
A  S ,→ U , where S ⊆ U is a covering sieve, and A  S is an epimorphism. By assumption
C(S,
b F ) → C(A,b F ) is a monomorphism, and C(B,
b F ) → C(S,
b F ) is a monomorphism by the
definition of epimorphisms. The general case can then be obtained by writing a general presheaf
B as a colimit of representables, and reducing to the case just described.
Conversely, if for every covering {Ui → U }i∈I , such that i∈I C( , Ui ) → C( , U ) is a local
`

epimorphism, the morphism C(C( , U ), F ) → C( i∈I C( , Ui ), F ) is a monomorphism then F is

b b `

a separated presheaf by definition.

If C is equipped with a Grothendieck topology, then every epimorphism A → B in C e is in

fact an effective15 epimorphism. It turns out that a presheaf on C is then a sheaf iff it views local
epimorphisms as effective epimorphisms16 . (So maybe the term “local effective epimorphism”
would have been a better choice, but “local epimorphism” already seems to be standard.)
Theorem 1.1.7.9. [KS06, Prop. 17.3.4.] Assume C is equipped with a family of local epimor-
phisms, then a presheaf F : Cop → Set is a sheaf iff for every local epimorphism A → B the
functor C(
b , F ) maps the sequence

A ×B A ⇒ A → B

to an equaliser diagram.
Note on proof. The “if” part is not particularly hard prove; nevertheless we refer the reader to
[KS06, Prop. 17.3.4.], as the proof is somewhat unenlightening when studying this material the
first time around.
The “only if” part of the statement is easy to verify: Let {Ui → U }i∈I be a covering in a coverage
corresponding to the family of local epimorphisms, then the coequaliser diagram corresponding
to i∈I C( , Ui ) → C( , U ) is isomorphic to
`

a a
C( , Ui ) ×C( ,U ) C( , Uj ) ⇒ C( , Ui ) → C( , U ),
(i,j)∈I×I i∈I

and a sheaf was defined precisely to be a presheaf taking such diagrams to equaliser diagrams.

14
See Definition 2.1.0.1 for the definition of a separated presheaf.
15
Recall that A → B is an effective epimorphism if it is the coequaliser of A ×B A ⇒ A.
16
Thinking back to our discussion of local isomorphisms, Proposition 1.1.3.3 gives two equivalent ways of how
to view them: As the morphisms which look like isomorphisms to sheaves, or as those morphisms which are sent
to isomorphisms by the sheafification functor. Viewing local epimorphisms as “local effective epimorphisms” we
have only discussed the first viewpoint discussed for local isomorphisms. The second viewpoint is also still valid,
which is not hard to check using that the sheafification functor for a Grothendieck topology is right exact (see
Theorem 1.2.0.1).

12
 While the discussion of this subsection, and in particular the proof of Theorem 1.1.7.9 might
suggest that local epimorphisms could be viewed as generalised coverings, we would like to make
the case that simply viewing local epimorphisms as local effective epimorphisms is more elegant.
Say we are given a map of sets X → Y , and that we would like to know, given a real valued17
function f : X → R, when there exists a unique real valued function on Y which pulls back to
f . One necessary condition is clearly that X → Y be surjective. A sufficient condition is then
that f maps any two elements in a give fibre of X → Y to the same element in R. But the set
of pairs of elements in the same fibres of X → Y is exactly X ×Y X, so f descends uniquely to
map Y → R iff X → Y is an effective morphism18 and f equalises X ×Y X ⇒ X. The point
here is that an effective epimorphism is given by a (canonical) equivalence relation; it appears
to be a fundamental feature of a nice theory (i.e. category) of spaces that all epimorphism can
be described as a being a quotient arising from an equivalence relation.

1.1.7.1 A short note maximal coverages

Assume C is equipped with a coverage. We saw in Proposition 1.1.7.4 that T -coverings {Ui →
U }i∈I furnish local epimorphisms i∈I C( , Ui ) → C( , U ). Conversely, by Theorem 1.1.7.9 we
`

see that if {Ui → U }i∈I is any covering then i∈I C( , Ui ) → C( , U ) is a local epimorphism
`

precisely when all T -sheaves satisfy the sheaf condition when {Ui → U }i∈I .
In Proposition 1.1.5.17 it was shown that any coverage can be completed to a Grothendieck
topology19 ; a more naive completion would be to try to find the maximal coverage with the
same sheaves as T , and our discussion of local epimorphisms, shows exactly that this coverage is
given by those coverings {Ui → U }i∈I such that i∈I C( , Ui ) → C( , U ) is a local epimorphism.
`

For the the case when C admits pullbacks, an axiomatic description of maximal coverages is
given in [KS06, Def. 16.1.5].

1.1.8 Examples of Grothendieck topologies

We start off with a formal class of examples before turing to some concrete examples.
Example 1.1.8.1. A coverage T on C is called subcanonical iff every representable presheaf
is a T -sheaf. It is readily verified that a coverage is subcanonical iff the associated glueing
diagrams are coequalising. Also note that the Yoneda embedding followed by T -sheafification is
fully faithful iff T is subcanonical. y
Example 1.1.8.2. The category 1 consisting of only one object and one morphism admits only
one Grothendieck topology, the trivial one. This example is not as vacuous as may originally be
thought, both conceptually and practically. The category of sheaves on 1 is canonically equivalent
to the category of sets. A set is thus precisely what you obtain by freely glueing together its points.
Also, Set is atomic in the sense that it admits no subcategories of sheaves (see Theorem 1.2.0.3),
and in topos theory, maps20 from Set to C e are thought of as “points” of C.e y
Example 1.1.8.3. Let X be a topological space, then assignment to every open subset of X the
set of coverings (in the classical sense) determines a Grothendieck pretopology on the category
OuvX of open subsets of X. y
17
Really, any non-empty target set would do.
18
Note that all epimorphisms in Set are effective.
19
In Theorem 1.2.0.3 we will see that this Grothendieck topology is unique.
20
By maps we mean geometric morphisms, which we may or may not define at some point.

13
Example 1.1.8.4. The category Top admits an obvious Grothendieck pretopology where we
assign to every U ∈ C the coverings {Ui → U }i∈I such that for each i ∈ I the continuous map
Ui → X is a homeomorphism to an open subset of X and the images of all continuous maps in
{Ui → X}i∈I cover X. y

We have treated Cart∞ and Aff k as sites throughout the whole chapter, and now we finally
make it official.

Example 1.1.8.5. The assignment to every object in Cart∞ the class of coverings by open
subsets is a Grothendieck pretopology. y

Example 1.1.8.6. For a commutative ring k the category of affine k-schemes admits an abun-
dance of interesting Grothendieck pretopologies of which we describe but three, in increasing
order of fineness:

1. To any affine k-scheme X we can assign the class of coverings {Ui → X}i∈I such that for
each i ∈ I the morphism Ui → X is an open affine subscheme and the (underlying sets
of the) images of the morphisms in {Ui → X}i∈I cover (the underlying set of ) X. This
defines the Zariski pretopology on Aff k and is denoted by Zar.

2. To any affine k-scheme X we can assign the class of coverings {Ui → X}i∈I such that
for each i ∈ I the morphism Ui → X is étale, and the (underlying sets of the) images of
the morphisms in {Ui → X}i∈I cover (the underlying set of ) X. This defines the étale
pretopology on Aff k and is denoted by ét.

3. To any affine k-scheme X we can assign the class of coverings {Ui → X}i∈I such that for
each i ∈ I the morphism Ui → X is given by a faithfully flat ring homomorphism of finite
presentation, and the (underlying sets of the) images of the morphisms in {Ui → X}i∈I
cover (the underlying set of ) X. This defines the the fppf pretopology on Aff k , and is
denoted by fppf, where fppf stands for “fidèlement plat, présentation finie”.

^
It is still true that the functor Schk → (Aff k )fppf , and thus also Schk → (Aff k )ét , is fully
]
faithful, but unlike the case Schk → (Aff
]k )Zar , this is a deep theorem. y

1.2 Some formal properties of categories of sheaves

We discuss three important formal properties of categories of sheaves.

Theorem 1.2.0.1. Assume C is equipped with a collection P of coverings. If the corresponding

sheafification functor is right exact, then the collection of morphisms sent to isomorphisms
(resp. epimorphisms) by the P -sheafification functor is a family of local isomorphisms (resp.
epimorphisms); or equivalently, the family of sieves sent to isomorphisms form a Grothendieck
topology.
Conversely, if the collection of sieves sent to isomorphisms by the P -sheafification functor is
closed under pullbacks along morphisms in C, then P -sheafification is right exact.

Morally the theorem thus says that the P -sheafification functor is left exact iff P satisfies
(C), i.e. it is a coverage.

14
Discussion of proof. Assume the P -sheafification functor is right exact. If we wish to prove the
stated consequence for local isomorphisms (resp. local epimorphisms), then by Remark 1.1.6.3
(resp. 1.1.7.3), it is enough to show LI 3 (resp. LI 4). We will discuss local epimorphisms; the
statement for local isomorphisms could then be deduced from the case of local epimorphisms
using local monomorphisms (as in [SGA 4I , Th. 5.5]), or it could be proved directly by writing
presheaves as quotients of coproducts in C similarly as in the proof we are about to give (see
[Rez99, 8,9]). Let A → B be a morphisms of presheaves on C; we wish to show that if the
pullback A → B along U → B, for all U → B in CB is a local epimorphism, then A → B is
a local epimorphism. As pullbacks commute with coproducts in Set, they do so in C, b so the
`
pullback of CB U → B is given by
`
CB U ×B A A

`
CB U B.

As sheafification commutes with colimits the coproduct of a small family of local epimorphisms
is again a local epimorphism. Sheafification maps the composition of the bottom left morphisms
in the pullback to an epimorphism, so the rightmost vertical morphism has to be mapped to an
epimorphism.
Let us also briefly consider the statement for sieves. That (M) is satisfied is clear. The proof
that (L) holds is more difficult, but it is similar to (LI 3) and (LE 4) (see [Rez99, 2]). Proving
(Cs ) establishes the close link between the right exactness of P -sheafification and the axiom
(C). For any morphism f : V → U and any sieve S ⊆ U we have by assumption that

f ∗S S

V U

gets sent to a pullback diagram, where the vertical morphism to the right is an isomorphism,
and isomorphisms get pulled back to isomorphisms.
Finally, assume that P is a coverage, then we will see in Corollary 2.2.0.1 that the P -sheafification
functor is constructed using limits and filtered colimits, and is thus right exact.

Remark 1.2.0.2. From the proof of Theorem 1.2.0.1 it is easy to conclude that the reflector of
a reflective subcategory of C
b is right exact iff it preserves pullbacks. y

Theorem 1.2.0.3. [SGA 4I , Th. 5.5] The map sending any Grothendieck topology on C to its
category of sheaves determines a bijection between Grothendieck topologies on C and isomorphism-
closed subcategories of C
b admitting a right exact left adjoint21 .

Proof. By Theorem 1.1.3.4 and Theorem 1.2.0.1 isomorphism-closed reflective subcategories of

C
b with a right exactreflection are in bijection with families of local isomorphisms, which are in
bijection with Grothendieck topologies by Theorem 1.1.6.4.
21
Such adjunctions are “geometric embeddings” of topoi.

15
 We have now seen that the right exactness of sheafification is intimately related to axiom
(C). In Theorem 1.1.7.9 and the surrounding discussion we saw that right exactness is crucial
for ensuring that local epimorphisms are local effective epimorphisms. We finish this section
with another important consequence of the right exactness sheafification.

Theorem 1.2.0.4. [Joh02a, Prop. A.4.3.1] Let S ⊆ C b be a reflective subcategory, then S is

Cartesian closed with C(F, G) ∈ S for any presheaves F, G ∈ S iff the reflection of S ⊆ C
b b
preserves finite products.

Thus the axiom (C), perhaps unexpectedly, also guarantees for any collection P of coverings
that C
eP has internal hom in the simplest possible way.

1.3 The closure property of the category of sheaves on a site

Throughout this section we assume C is equipped with a Grothendieck topology T . As we
view Ce as the category of spaces that can be glued together from spaces in C according to the
rules specified by T , one should expect to be able to extend T to C e in some canonical way.
Furthermore it is natural to ask whether we can glue together spaces in C e to obtain even more
spaces. These two considerations are closely related: If we require of the extended topology to
be subcanonical, and more importantly, that sheaves on C e are still determined by how they are
seen by objects in C, then there is indeed only one extension. Moreover sheaves on C e with this
topology are representable (Theorem 1.3.0.4), which shows that C e can be viewed as a sort of
closure of C.

Using Theorem 1.1.1.1 we reformulated the fact that every presheaf is determined by how
it is seen by objects in C in terms of colimits. This natural leads to another reformulation in
terms of sieves.

Lemma 1.3.0.1. Let F : Cop → Set be a presheaf, and let S ⊆ C(

b , F ) be the sieve generated
∼
by CF , then lim(G→F )∈S G = F .
−→
Proof. Let (G → Z)(G→F )∈S be a cocone, then this cocone restricts to a cocone over CF , so we
obtain a unique morphism F → Z of cocones over CF by Theorem 1.1.1.1. It is thus enough to
show that this morphism extends to a morphism of cocones over S. Thus let G → F be in S,
then by assumption there exists an object U ∈ C such that G → F factors through U , and we
obtain the diagram
F Z

G,
where the three triangles having a U as a corner commute by assumption, so that the outer
triangle must then also commute.

Definition 1.3.0.2. [Joh02b, Def. C.2.2.1] A full subcategory D ⊆ C is called dense if every
object U ∈ C has a covering sieve generated by objects in D. y

16
Theorem 1.3.0.3. [Joh02b, Th. C.2.2.3] Let D ⊆ C be a small, dense subcategory, then the
restriction functor C
e→D
e is an equivalence of categories.

As an immediate corollary we obtain:

Theorem 1.3.0.4. Assume T is subcanonical. If there exists a topology on C

e which restricts
to T on C, and contains all sieves generated by CF for all F ∈ C,
b then any sheaf on C e is
representable.

So if an extension of T with the desired properties exists, then it must be the canonical
topology; by assumption it is subcanonical, and if it is coarser than the canonical topology,
by Theorem 1.2.0.3 there are guaranteed to be sheaves on C, e which are note representable,
contradicting Theorem 1.3.0.4.
It turns out the canonical on C
e does have the desired properties, in fact we have the following
remarkable result:

Theorem 1.3.0.5. [Joh02b, Th. 2.2.8] A category E is equivalent to the category of sheaves on
a site iff it has a small dense subcategory for the canonical topology on E, and every sheaf on E
is representable.

To get an idea of just how canonical the canonical topology on a category E equivalent
to a category of sheaves on a site is, one should dwell upon the fact that no matter which
subcanonical22 site D one chooses in order to realise E as D,
e the topology on D is the restriction
of the canonical topology on E23 .
We finish by showing that canonical topology on C e can be described in an very natural way. If
C is sufficiently nice, and T is subcanonical, then for every T -covering {Ui → U }i∈I the glueing
diagram exists, and we see that i∈I Ui → U is an effective epimorphism. The sheaves on C
`

are then those presheaves to which the morphisms i∈I C( , Ui ) → C( , U ) look like effective
`

epimorphisms, and in C e are genuine effective epimorphisms. In fact the coverings ` Ui → U

i∈I
for which T -sheaves satisfy the sheaf condition are characterised as those coverings such that
i∈I C( , Ui ) → C( , U ) is an effective epimorphism in C. There is now an obvious extension of
` e
T to a coverage on C: e

Theorem 1.3.0.6. [Joh02b, Ex. C.2.1.12.e] The canonical topology on C e is given by the coverage
T such that a (small) family of morphisms {Gi → F }i∈I is a covering in TF0 iff iff i∈I Gi → F
0
`

is an effective epimorphism.

22
Which exists by Theorem 1.3.0.5.
23
This might explain why there is a great deal of literature on canonical topologies on categories of sheaves.

17
18
Bibliography

[Bor94a] Francis Borceux. Handbook of categorical algebra. 1, volume 50 of Encyclopedia of

Mathematics and its Applications. Cambridge University Press, Cambridge, 1994. Basic
category theory.

[Bor94b] Francis Borceux. Handbook of categorical algebra. 2, volume 51 of Encyclopedia of

Mathematics and its Applications. Cambridge University Press, Cambridge, 1994. Cat-
egories and structures.

[Gar09] Richard Garner. Understanding the small object argument. Appl. Categ. Structures,
17(3):247–285, 2009.

[Gro95] Alexander Grothendieck. Technique de descente et théorèmes d’existence en géometrie

algébrique. I. Généralités. Descente par morphismes fidèlement plats. In Séminaire
Bourbaki, Vol. 5, pages Exp. No. 190, 299–327. Soc. Math. France, Paris, 1995.

[Hol08] Sharon Hollander. A homotopy theory for stacks. Israel J. Math., 163:93–124, 2008.

[Jar01] J. F. Jardine. Stacks and the homotopy theory of simplicial sheaves. Homology Homo-
topy Appl., 3(2):361–384, 2001. Equivariant stable homotopy theory and related areas
(Stanford, CA, 2000).

[Jar15] John F. Jardine. Local homotopy theory. Springer Monographs in Mathematics.

Springer, New York, 2015.

[Joh02a] Peter T. Johnstone. Sketches of an elephant: a topos theory compendium. Vol. 1,

volume 43 of Oxford Logic Guides. The Clarendon Press, Oxford University Press, New
York, 2002.

[Joh02b] Peter T. Johnstone. Sketches of an elephant: a topos theory compendium. Vol. 2,

volume 44 of Oxford Logic Guides. The Clarendon Press, Oxford University Press,
Oxford, 2002.

[KS06] Masaki Kashiwara and Pierre Schapira. Categories and Sheaves, volume 332 of
Grundlehren der mathematischen Wissenschaften. Springer-Verlag Berlin Heidelberg,
2006.

[Low] Zhen-Lin Low. Elephant: how do i prove lemma 2.1.7, section c2.1? Mathematics Stack
Exchange. URL:http://math.stackexchange.com/q/358709 (version: 2013-04-11).

[Lur09] Jacob Lurie. Higher Topos Theory. Princeton University Press, 2009.

19
[Mac98] Saunders Mac Lane. Categories for the working mathematician, volume 5 of Graduate
Texts in Mathematics. Springer Verlag New York, Inc., second edition, 1998.

[NS11] Thomas Nikolaus and Christoph Schweigert. Equivariance in higher geometry. Adv.
Math., 226(4):3367–3408, 2011.

[Rez99] Charles Rezk. A note on sheaf theory. A note available at http://www.math.uiuc.

edu/~rezk/phil-sheaves.dvi, 1999.

[SGA 4I ] M. Artin, A. Grothendieck, and J.L. Verdier, editors. Théorie des topos et cohomologie
étale des schémas. Tome 1: Théorie des topos. Lecture Notes in Mathematics, Vol.
269. Springer-Verlag, Berlin, 1972. Séminaire de Géométrie Algébrique du Bois-Marie
1963–1964 (SGA 4), Dirigé par M. Artin, A. Grothendieck, et J. L. Verdier. Avec la
collaboration de N. Bourbaki, P. Deligne et B. Saint-Donat.

[Sta14] Alexandru E. Stanculescu. Stacks and sheaves of categories as fibrant objects, I. Theory
Appl. Categ., 29:No. 24, 654–695, 2014.

[Str04] Ross Street. Categorical and combinatorial aspects of descent theory. Appl. Categ.
Structures, 12(5-6):537–576, 2004.

[Yuh07] Sergio A. Yuhjtman. Bilı́mites 2-cofiltrantes de topos. PhD thesis, Universidad de

Bueno Aires, December 2007. Available online at http://cms.dm.uba.ar/academico/
carreras/licenciatura/tesis/yuhjtman.pdf.

20