<< Ļšåäūäółą’

ńņš. 42
(čē 60 ńņš.)

ĪĆĖĄĀĖÅĶČÅ

Ńėåäóžłą’ >>

from the definition of left C -objects and morphisms thereof that (T, η, µ) is a
triple and that the category of Eilenberg-Moore algebras (φ: A − C, ψ) is the
→
category of left C -objects.
Now come the heavy cannon. By the constructions of Theorem 6 and Corol-
lary 7 of Section 5.3, T factors as the top row of

(d0 )∗ d1
E /C −− −− E /C1 −− −− E /C
ā†āˆ’āˆ’ā†’
−− −− ā†āˆ’āˆ’ā†’
−− −−
d0 (d1 )∗

where the bottom row forms a right adjoint to T . Thus by Theorem 5 of Sec-
tion 3.7, T is adjoint tripleable and the category of left C -objects is the category
of coalgebras for a cotriple whose functor is G = d0 ◦ (d1 )∗ . G has a left adjoint,
so gives a left exact cotriple. Hence by Theorem 1 of Section 6.4, we have:
Theorem 2. For any category object C of a topos E , the category of left C -
objects (equivalently the category of split discrete opfibrations of C ) is a topos.
It is natural to denote the category of left C -objects by E C .
Theorem 3. Any category object functor f : C − C induces a functor f # : E C
→
− E C which is the restriction of the pullback functor f ∗ . Moreover, f # has left
→
and right adjoints.
Proof. That pulling back a split discrete opfibration produces a split discrete
opfibration is merely the statement that pulling back a pullback gives a pull-
back. The reader may alternatively define f # on left C -objects by stipulating
that f # (A , φ , ψ ) = (A, φ, ψ), where A = [(c, a ) | f c = φ a ], φ is the first
projection, and ψ(g, (c, a )) (where necessarily d0 (g) = c and φa = f c) must be
d1 (g), ψ (f (g), a )).
6.6 Categories in a Topos 225
The square
f# E C
C
E E

U
U
c c
E E /C
E /C ∗
f
commutes, where the vertical arrows are the underlying functors from triple alge-
bras. (They are in fact inclusions). Since the bottom arrow has both a left and a
right adjoint, so does the top one by Butler’s Theorems (Section 3.7, Theorem 3).

When C is the trivial category object with C = C1 = 1, then E C is E and
the induced map E − E C (where C is some category object) has left and right
→
adjoints denoted lim→ C and lim← C respectively. When E is Set they are in fact
the left and right Kan extensions. Notice that this says that, given a set function
f : Y − I, then the existence of f −1 (i) (which is a completeness property—Y
→
and I can both be infinite) depends only on the fact that in Set we have finite
limits and a power object. Thus the existence of P is a powerful hypothesis.

Exercises 6.6

1. A simplicial object in a category E is a functor S: āˆ†op − E where āˆ† is
→
the category whose objects are the finite sets {1, 2, . . . , n} for n = 0, 1, 2, . . . and
whose arrows are the order-preserving maps between these sets. Prove that the
category of category objects and functors between them in a left exact category
E is equivalent to a full subcategory of the category of simplicial objects in E .

2. Let F : D − C be a functor between categories (in Set). If A is an object of
→
C , the fiber DA over C is the subcategory of D consisting of all objects mapping
to A and all arrows mapping to 1A . Its inclusion into D is denoted JA . F is an
opfibration if for each f : A − B in C there is a functor f ∗ : DA − DB and a
→ →
∗
natural transformation θf : JA − JB ◦ f with every component lying over f , for
→
which for any arrow m: D − E in D lying over f there is a unique arrow f ∗ (D)
→
− E making
→
m EE
D
d Ā 
(Īøf )Dd Ā 
d Ā 
dĀ 
‚©
f ∗ (D)
226 6 Permanence Properties of Toposes
commute. The opfibration is split if (1A )∗ = 1(DA ) and g ∗ ◦ f ∗ = (g ◦ f )∗ whenever
g ā—¦ f is defined. It is discrete if the fibers DA are all sets—i.e., their only arrows
are identity arrows. (ā€œSplitā€ is also called ā€œsplit normal,ā€ the normal referring to
preservation of identities). A functor between opfibrations over C is a functor in
Cat/C (it does not have to preserve ∗ ).
(a) Show that, for Set, a split discrete opfibration as defined in the text is
the same as that defined here.
(b) Show that the subcategory of split opfibrations over C and functors which
commute with ∗ is equivalent to the category of functors from C to Cat and natural
transformations.
(c) Show that opfibrations have the ā€œhomotopy lifting propertyā€: If G and H
are functors from A to C , F : D − C is an opfibration, λ: G − H is a natural
→ →
transformation, and G = F ◦ G for some functor G : A − D, then there is a
→
functor H : A − D such that H = F ◦ H and a natural transformation λ : G
→
− H.
→

3. (Categorical definition of opfibration). Let F : D − C be a functor. Let 2
→
denote the category with two objects 0 and 1 whose only nonidentity arrow u
goes from 0 to 1. Let S be the functor from Hom(2, D) to the comma category
(F, 1C ) which takes M : 2 − D to (M (0), F (M (u)), F (M (1))) (you can figure
→
out what it does to arrows of Hom(2, D), which are natural transformations).
Show that F is an opfibration if and only if S has a left adjoint which is also a
right inverse of S.


6.7 Grothendieck Topologies
A Grothendieck topology on a category is a generalization of the concept of all
open covers of all open sets in a topological space.
A sieve (called ā€œcribleā€ by some authorsā€”ā€œcribleā€ is the French word for
ā€œsieveā€) on an object A is a family of arrows with codomain A. We will use the
notation {Ai − A} for a sieve, the i varying over an unspecified index set. We
→
will follow the convention that different sieves have possibly different index sets
even if the same letter i is used, unless specifically stated otherwise.
The set of fiber products {Ai ×A Aj }, where i and j both run over the index
set of the sieve, will be used repeatedly in the sequel. If f : A − B and {Ai
→
− A} is a sieve on A, we write f |Ai for the composite of the projection Ai − A
→ →
followed by f , and f |Ai ×A Aj for the composite of Ai ×A Aj − A followed by f .
→
A sieve {Ai − A} refines a sieve {Bj − A} if every arrow in the first
→ →
factors through at least one arrow in the second.
6.7 Grothendieck Topologies 227
A Grothendieck topology in a left exact category A is a family of sieves,
called covers, with the following properties.
(i) For each object A of A , idA is a sieve.
(ii) (Stability) If {Ai − A} is a cover and B − A is an arrow, then {B ×A Ai
→ →
− B} is a cover.
→
(iii) (Composability) If {Ai − A} is a cover, and for each i, {Aij − Ai } is
→ →
a cover, then {Aij − A} is a cover.
→
A Grothendieck topology is saturated if whenever {Ai − A} is a sieve and
→
for each i, {Aij − Ai } is a sieve for which {Aij − A} (doubly indexed!) is a
→ →
cover, then {Ai − A} is a cover. It is clear that each Grothendieck topology
→
is contained in a unique saturated Grothendieck topology. It follows from (ii)
(stability) that the saturation contains all the sieves which have a refinement in
the original topology.
Refinement is also defined for topologies: one topology refines another if every
cover in the saturation of the second is in the saturation of the first.
A site is a left exact category together with a specific saturated Grothendieck
topology. A morphism of sites is a left exact functor between sites which takes
covers to covers.
Some examples of sites:
(a) The category of open sets of a fixed topological space together with all
the open covers of all open sets.
(b) Any left exact category together with all the universal regular epimor-
phisms, each regarded as a sieve containing a single arrow. (ā€œUniversalā€ or
ā€œstableā€ means preserved under pullbacks.)
(c) A sieve {fi : Ai − A} is an epimorphic family if and only if whenever
→
g = h are two maps from A to B, there is at least one index i for which g ā—¦fi = hā—¦fi .
If epimorphic families are stable under pullbacks, they form a site.
op
If C is any category, an object of SetC is called a presheaf, a terminology
used particularly when C is a site. If f : A1 − A2 in A and a ∈ A2 , then,
→
motivated by the discussion of sheaves on a topological space in Section 2.2, we
write a|A1 for F f (a). (We recommend this notation—it makes the theory much
more manageable.)
Let S = {Ai − A} be a sieve in a category C . If F : C op − Set is a presheaf,
→ →
we say that F is S-separated if F A − → F Ai is injective and that F is an
S-sheaf if
→ F Ai − → F (Ai Ć—A Aj )
FA − āˆ’ā†’
228 6 Permanence Properties of Toposes
is an equalizer. This generalizes the sheaf condition for topological spaces given
by Proposition 1, Section 2.2. Stated in terms of restrictions, F is S-separated
if whenever a, a ∈ F A with the property that for every i, a|Ai = a |Ai , then
a = a . F is an S-sheaf if in addition for every tuple of elements ai ∈ Ai with
the property that ai |Ai ×A Aj = aj |Ai ×A Aj , there is a (unique) element a ∈ F A
such that a|Ai = ai .
It is straightforward to see that a sieve S is an epimorphic family if every
representable functor Hom(−, B) is S-separated, and we say that S is a regular
or effective epimorphic family if every representable functor is an S-sheaf.
Both epimorphic and regular epimorphic families are called stable or universal
if they remain epimorphic (respectively regular epimorphic) when pulled back.
In any category, stable epimorphic families, and also stable regular epimorphic
families, form a Grothendieck topology.
Note that since every epimorphism in a topos is universal and regular by
Corollary 8 of Section 5.3 and Proposition 3 of Section 5.5, the class of all epi-
morphisms in a topos forms a Grothendieck topology on the topos. However,
epimorphic families in a topos need not be regular unless the topos has arbitrary
sums.
op
A site A determines a topology j on SetA as follows. If F is a presheaf in
op
SetA and F0 is a subpresheaf, then jF0 = jF (F0 ) is the functor whose value
at A consists of all a ∈ F A for which there is a cover {Ai − A} such that
→
a|Ai ∈ F0 Ai for all i.
op
j as constructed above is a topology on SetA .
Proposition 1.
Proof.
(i) Naturality translates into showing that if the left square below is a pullback
then so is the right one.
FE EF jFE EF
0
0




c c c c
GE EG jGE EG
0
0

Suppose a ∈ jG0 A, b ∈ F A have the same image in GA. We must show
b ∈ jF0 A. Let {Ai − A} witness that a ∈ jG0 (A). In other words, a|Ai ∈ G0 Ai
→
6.7 Grothendieck Topologies 229
for all i. Now consider this pullback diagram:
E F Ai
F0 Ai



c c
E GAi
G0 Ai
We know a|Ai ∈ G0 Ai and that b|Ai ∈ F Ai for all i. They must have the
same image in GAi , so b|Ai ∈ F0 Ai . Therefore b ∈ jF0 (A).
(ii) The inflationary property follows from the fact that idA is a cover.
(iii) The monotone property is a trivial consequence of the definition.
(iv) Idempotence follows from the composition property for covers.

op
Conversely, given a topology j on SetA , we can construct a Grothendieck
topology on A as follows. Any sieve {Ai − A} determines a subfunctor R of
→
Hom(−, A) defined for an object B by letting RB be the set of f : B − A for
→
which for some i, there is a factorization B − Ai − A of f . This is the same
→ →
as saying R is the union of the images of Hom(−, Ai ) in Hom(−, A). Then we
say {Ai − A} is a covering sieve for the Grothendieck topology if R is j-dense.
→
Proposition 2. For any topology j, the definition just given produces a Grothen-
dieck topology on A .
The proof is straightforward and will be omitted.
The two constructions given above produce a one to one correspondence be-
op
tween saturated Grothendieck topologies on A and topologies on SetA . We
make no use of this fact and the proof is uninteresting, so we omit it.
op

<< Ļšåäūäółą’

ńņš. 42
(čē 60 ńņš.)

ĪĆĖĄĀĖÅĶČÅ

Ńėåäóžłą’ >>