SUBDIVISIONS OF SMALL CATEGORIES
NOTES FOR REU BY J.P. MAY
Let A be a (small) category. For example, monoids (sets with associative and
unital products) can be identied with categories with a single object. Analogously,
posets can be identied with those categories A with at most one arrow between
any two objects by dening x y if there is an arrow x y between the objects
x and y of A.
In particular, we write [n] for the poset {0 < 1 < n}. If we think of [n]
as a nite space, then a continuous map f : [m] [n] is a monotonic, or non-
decreasing, function, f(i) f(j) if i j. We can equally well regard such a function
as a functor [m] [n]. We are perilously close to dening the fundamental notion
of a simplicial object in a possibly large category C, so lets do so.
Let denote the category of posets [n] and monotonic maps between them.
This category is generated by certain canonical monotonic maps. We have the
face map
i
: [n] [n + 1] which is the monomorphism that misses i. That
is
i
(j) = j if j < i and
i
(j) = j + 1 if j i. We also have the surjection
i
: [n + 1] [n] that hits i twice. That is,
i
(j) = j if j i and
i
(j) = j 1 if
j > i. Every morphism in is a composite of these morphisms, and they satisfy
certain easily determined identities. A simplicial object in C is a contravariant
functor C. In detail, it is a sequence of objects C
n
C together with face
maps d
i
: C
n
C
n1
and degeneracy maps s
i
: C
n
C
n+1
, 0 i n, such
that
d
i
d
j
= d
j1
d
i
if i < j
d
i
s
j
=
s
j1
d
i
if i < j
id if i = j or i = j + 1
s
j
d
i1
if i > j + 1.
s
i
s
j
= s
j+1
s
i
if i j.
When C = Set is the category of sets, we obtain sSet, the category of simplicial
sets. For example, if S is a topological space, we obtain the simplicial set SX such
that S
n
X is the set of all continuous maps from the standard topological n-simplex
n
to X. Explicitly,
n
= {(t
0
, . . ., t
n
)|0 t
i
1,
t
i
= 1} R
n+1
.
We have the face maps
i
:
n1
n
, 0 i n,
specied by
i
(t
0
, . . ., t
n1
) = (t
0
, . . ., t
i1
, 0, t
i
, . . ., t
n1
)
and degeneracy maps
i
:
n+1
n
, 0 i n,
Date: Summer, 2008.
1
2 NOTES FOR REU BY J.P. MAY
specied by
i
(t
0
, . . ., t
n+1
) = (t
0
, . . ., t
i1
, t
i
+ t
i+1
, t
i+2
, . . ., t
n+1
).
Precomposing with these maps, we obtain the maps d
i
and s
i
that make SX into a
simplicial set. It has long been known that we can use simplicial sets pretty much
interchangeably with topological spaces when studying homotopy theory.
For a simplicial set K, we dene a space |K|, called the geometric realization
of K, as follows. As a set
|X| =
n0
(K
n
n
)/(),
where the equivalence relation is generated by
(k,
i
u) (d
i
(k), u) for k K
n
and u
n1
and
(k,
i
v) (s
i
(k), v) for K K
n
and v
n+1
.
Topologize |K| by giving
|K|
n
0qn
(K
q
q
)/()
the quotient topology and then giving |K| the topology of the union, so that a
subset is closed if it intersects each |K|
n
in a closed subset. Write |k, u| for points
of |K|. Say that (k, u) is nondegenerate if k K
n
is not of the form s
i
j for any i
and any j K
n1
and if u
n
is an interior point. Every (k, u) is equivalent to
one and only one nondegerate point.
Dene : |SX| X by
|f, u| = f(u) for f :
n
X and u
n
.
It is a fact that is a weak homotopy equivalence for every space X, although we
shall not prove that. There is also a map : K S|K| of simplicial sets specied
by (k)(u) = |k, u| for k K
n
and u
n
. Again, as we also shall not prove,
|| : |K| |S|K|| is a homotopy equivalence.
There is a neat relationship between | | and S. They are left and right adjoint
functors, meaning that there is a bijection, natural in both variables, between
morphism sets:
Top(|K|, X)
= sSet(K, SX).
It is specied by letting f correspond to g if f(|k, u|) = g(k)(u).
There is also a construction that assigns a simplicial set K
s
to a simplicial
complex K. The idea is to allow repeated elements in the sets of simplices. It is
easiest to dene K (X)
s
for a poset X, and then the construction gives a functor,
but one can use any arbitrarily chosen ordering of the set of vertices to apply the
denition more generally. One lets the set K (X)
s
n
of n-simplices be the set of
sequences x
0
x
1
x
n
. Deleting x
i
gives d
i
, and repeating x
i
gives s
i
. The
geometric realization gives the geometric realization |K (X)
s
|. When X is nite
this is homeomorphic to any choice of geometric realization as we dened it earlier,
but the denition |K (X)| = |K (X)
s
| works in general and gives a functor of X.
With this as background, we turn to the homotopy theory of small categories
A. We construct a simplicial set NA called the nerve of A. Regarding [n] as a
category, we dene the set N
n
A of n-simplices to be the set of functors [n] A.
Regarding a monotonic function f : [m] [n] as a functor, precomposition with
f gives us the required contravariant functoriality on . The denition should
SUBDIVISIONS OF SMALL CATEGORIES 3
look very similar to the denition of the total singular functor S from spaces to
simplicial sets. It gives us a functor N from Cat, the category of small categories
and functors between them, to simplicial sets. We dene BA = |NA|. This is
called the classifying space of the category A. When G is a group regarded as a
category with a single object, BG is called the classsifying space of the group G. The
space BG is often written as K(G, 1) and called an Eilenberg-Mac Lane space. It
is characterized (up to homotopy type) as a connected space with
1
(K(G, 1)) = G
and with all higher homotopy groups
q
(K(G, 1)) = 0. These are fundamentally
important constructions in topology and its applications.
The nerve functor N is accompanied by a functor
1
: sSets Cat, called the
fundamental category functor. It is left adjoint to N, meaning that
Cat(
1
K, A)
= sSet(K, NA).
This means that it is conceptually sensible, but it does not have good homotopical
properties. For a simplicial set K, the objects of the category
1
K are the vertices
(= 0-simplices) of K. To construct the morphisms, one starts by thinking of the
1-simplices y as maps d
1
y d
0
y. One forms all words (formal composites) that
make sense, that is, whose targets and sources match up. Then one imposes the
relations on morphisms determined by
s
0
x = id
x
for x K
0
and d
1
z = d
0
z d
2
z for z K
2
.
This makes good sense since if K = NA, then a 0-simplex is an object x of A, a
1-simplex y is a map d
1
y d
0
y and s
0
x = id
x
, and a 2-simplex z is given by a pair
of composable morphisms d
2
z and d
0
z together with their composite d
1
z. Therefore
there is a natural map
1
NA A that is the identity on zero simplices and is
induced by the identity on 1-simplices. In fact, it is an isomorphism of categories:
it is the identity on objects, and it presents the category in terms of generators
given by the morphism sets modulo relations determined by the category axioms.
For the adjunction, a functor F :
1
K A is constructed from a map of sim-
plicial sets g : K NA by letting F be the unique functor that agrees with g
on objects (= 0-simplices) and equivalence classes of morphisms (= 1-simplices).
Applying the adjunction to the identity map of
1
K, we obtain a natural map
: K N
1
K. Backing up, we dene the standard simplicial n-simplex [n] to
be the simplicial set whose q-simplices are the monotonic functions : [q] [n];
precomposition with monotonic functions : [p] [q] gives the required con-
travariant functoriality on . The nondegenerate q-simplices in [n] are the
monomorphisms (= strictly monotonic functions) [q] [n], and there is one for
each subset of [n] of cardinality q +1. We may identify the set of all non-degenerate
simplices with the poset of non-empty subsets of the set [n] of n + 1 elements,
ordered by inclusion. In other words, [n] = (K ([n])
s
is the simplicial set de-
termined by the simplicial complex K ([n]). A monotonic function : [m] [n]
gives a map : [m] [n] of simplicial sets that sends : [q] [m] to .
Thus [] is a covariant functor from to simplicial sets.
The n-skeleton K
n
of a simplicial set K is the subsimplicial set generated by the
q-simplices for all q n. Visibly,
1
K depends only on the 2-skeleton K
2
. Therefore
the inclusion K
2
K of simplicial sets induces an isomorphism of categories
1
K
2
1
K for any K. In particular,
1
takes the inclusion : [n] [n]
of the boundary of the n-simplex to the identity functor when n > 2. Thus
1
loses homotopical information: upon realization, || is equivalent to the inclusion
4 NOTES FOR REU BY J.P. MAY
S
n1
D
n
. What is amazing is that this extreme loss of information disappears
after subdividing twice. This is something I am trying to better understand myself.
The reader will nd it easy to believe that there is a subdivision functor on
simplicial sets that generalizes the subdivision functor Sd on simplicial complexes
in the sense that (SdK)
s
= Sd(K
s
) for a simplicial complex K. This allows one
to dene a subdivision functor on categories by setting SdA =
1
SdNA. One can
iterate subdivision, forming functors Sd
2
on both simplicial sets and categories.
What is mind blowing at rst is that the iterated subdivision Sd
2
A is actually a
poset whose classifying space BSd
2
A is homotopy equivalent to BA. I will explain
at least the construction in a slow way to try to make the idea transparent.
However, before heading for that, let us summarize a schematic and technically
oversimplied global picture of all of the big categories we are constructing and com-
paring by functors. There is an interesting picture of lots of kinds of mathematics
that come together with a focus on simplicial sets.
Spaces
S
1
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
Groups
K(,1)
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
i
Simp. Sets
||
Sd
2
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
1
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
Categories
Sd
2
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
N
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
Simp. Complexes
i
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
X
.u
u
u
u
u
u
u
u
u
u
u
u
u
u
u
u
u
u
u
Posets
i
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
T
0
Aspaces
u
u
u
u
u
u
u
u
u
u
u
u
u
u
u
u
u
u
u
Our earlier talks focused on nite spaces, but the basic theory generalizes with
the niteness removed, provided we understand simplicial complexes to mean ab-
stract simplicial complexes. As noted above, we didnt dene geometric realization
in general earlier, but we have done so now. The equivalence of posets with T
0
Alexandro spaces and the constructions K and X that we worked out in detail
for nite spaces work in exactly the same way when we no longer restrict ourselves
to the nite case. The functors i in the diagram are thought of as inclusions of cat-
egories. We have dened all of the categories and functors exhibited in the diagram
except for Sd
2
.
As very often, there are two ways to dene subdivision, a conceptual one and a
concrete combinatorial one. Which one prefers is a matter of taste. We will start
conceptually and then come back down to earth.
For a set C and a simplicial set L, one can form a new simplicial set C L
by letting (C L)
q
= C L
q
, and similarly letting the faces and degeneracies be
SUBDIVISIONS OF SMALL CATEGORIES 5
induced by those of L. A simplicial set K can be reconstructed from the disjoint
union over n of the simplicial sets K
n
[n] for n 0 by taking equivalence classes
under the equivalence relation generated by
(1) (
(k), ) (k,
())
for k K
n
, [m]
q
, and : [m] [n]. Here
(k) K
m
is given by the fact
that K is a contravariant functor from to sets and
() [n]
q
is given by
the fact that [] is a covariant functor from to simplicial sets. The simplicial
structure is induced from the simplicial structure on the [n]. The point is that an
arbitrary pair (k, ) in K
n
[n]
q
is equivalent to the pair ((k),
q
) in K
q
[q]
q
,
where
q
: [q] [q] is the identity map viewed as a canonical q-simplex in [q],
and : [q] [n] is viewed as a morphism of , so that =
(
q
). Identifying
equivalence classes of q-simplices with elements of K
q
in this faction, we nd that
the faces and degeneracies agree. Indeed, for : [p] [q],
p
=
q
and
(k,
(
q
)) = (k,
(
p
)) (
(k),
p
).
We dene Sd[n] = (K
[n])
s
. That is, we take the simplicial set associated to
the barycentric subdivision of the simplicial complex K [n], where we again regard
[n] as a poset. Just like [], this gives a covariant functor Sd[] from to
simplicial sets. We use it in exactly the same way as above to construct SdK as a
quotient of the disjoint union of the K [n] Sd[n] by the equivalence relation of
the same form as (1). This is obviously sensible! We then dene SdA =
1
SdNA.
MORE TO COME.