0% found this document useful (0 votes)
19 views134 pages

Sset

Notes on simplicial ests

Uploaded by

Twistor0
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
19 views134 pages

Sset

Notes on simplicial ests

Uploaded by

Twistor0
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 134

Lecture notes: Simplicial sets

Christian Rüschoff
Summer 2017
Contents

1 (Abstract) Simplicial complexes 3


1.1 Geometric realization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2 Simplicial approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Products. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.4 Collapsing subspaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

2 Simplicial sets 17
2.1 Semi-simplicial sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2 Categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.3 Simplicial sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.4 Geometric realization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.5 Adjunctions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.6 The singular nerve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.7 Isomorphisms, monomorphisms and epimorphisms . . . . . . . . . . . . . 30
2.8 Simplicial standard simplices . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.9 Limits and colimits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.10 Preservation of (co-)limits . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.11 Comma categories. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.12 Internal homomorphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.13 Ordered simplicial complexes as simplicial sets . . . . . . . . . . . . . . . 42
2.14 Homotopies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.15 Connected components . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.16 Skeleton and coskeleton . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.17 Small categories as simplicial sets . . . . . . . . . . . . . . . . . . . . . . 54

3 Abstract homotopy theory 59


3.1 Localizations of categories . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.2 Weak factorization systems . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.3 Model categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.4 Stability of the lifting property . . . . . . . . . . . . . . . . . . . . . . . . 62
3.5 Construction of weak factorization systems . . . . . . . . . . . . . . . . . 66
3.6 Lifting properties and adjunctions . . . . . . . . . . . . . . . . . . . . . . 72
3.7 Towards the standard model structure on simplicial sets . . . . . . . . . . 74
3.8 Absolute weak equivalences . . . . . . . . . . . . . . . . . . . . . . . . . . 82
3.9 Maps with fibrant codomain . . . . . . . . . . . . . . . . . . . . . . . . . 94

i
Contents

3.10 Verifying the model structure on simplicial sets using Kan’s functor Ex∞ . 99
3.11 Homotopies in model categories . . . . . . . . . . . . . . . . . . . . . . . 112
3.12 The homotopy category of a model category . . . . . . . . . . . . . . . . . 118
3.13 Characterization of weak equivalences . . . . . . . . . . . . . . . . . . . . 121
3.14 Derived functors and the comparison of model categories . . . . . . . . . . 127

ii
Introduction, motivation

Algebraic topology is the study of topological spaces by using algebraic invariants, such
as (co-)homology, the fundamental group or more generally homotopy groups. This lead
to a powerful machinery, which also became extremely useful to other fields of mathe-
matics. For example group cohomology (in this setting more often called Galois cohomol-
ogy) became an important tool in algebraic number theory. Modern algebraic geometry
is unimaginable without the study of the various forms of cohomology (Zarisiki, étale,
synthomic, fppf, fpqc, ...).
Originally the cohomology of a group G was introduced as the (singular) cohomology of
its classifying space |BG|, which is a connected space with fundamental group π1 |BG| = G
and vanishing higher homotopy groups. In algebraic number theory this group G is most
likely the Galois group of a field extension. So we are assigning to an algebraic object G
a topological space |BG|, which we are taking an algebraic invariant of.

topological
4 O
space

*
algebraic object / “construction plan” / algebraic object

This looks like a long way round and one might ask, if it is possible to define group
cohomology without taking the detour over topological spaces. The idea is to use an
abstract “construction plan” BG for the topological space |BG|.
The aim of this lecture is the study of these “construction plans”, which in our case
are given by simplicial sets. So mainly we will do algebraic topology without topology.
Apart from the given example above, simplicial sets have many applications in different
fields of mathematics. The first few lectures will be about (abstract) simplicial complexes.
Compared to simplicial sets, their geometry is much easier to understand. We point out
the difficulties that arise when trying to do homotopy theory in the context of simplicial
complexes. Their disadvantages motivate the introduction of simplicial sets, which the
main part of these lecture is about.
The theory of simplicial sets is a theory of functors, so there is no way around some
techniques provided by category theory. As category theory still enjoys the reputation of
being “abstract nonsense”, which may discourage some readers, we present the necessary
tools along the way of studying simplicial sets. We intersperse short sections of the abstract
theory, which will be applied subsequently in the context of simplicial sets. Although
using the abstract language of category theory, we try to motivate every construction

1
Contents

by its topological analog. Guided by the geometric realization, we develope the basic
constructions like homotopies and mapping cones in the context of simplicial sets.
After this topologically motivated introduction we begin to point out the main advan-
tages. Instead of working with simplicial sets, one can similarly define simplicial objects
in any other category. This is a very powerful feature that allows us to use simplicial tech-
niques in any field of mathematics. Probably the most important example is the category
of simplicial modules, which turns out to be equivalent to the category of chain complexes
via the Dold-Kan correspondence. Coming back to our example of above, when learning
group homology, one will most likely end up with studying homological algebra first.
Compared to simplicial modules, chain complexes seem much easier to handle. However
every tool in homological algebra has its analog in the simplicial context. There are even
simplicial constructions, that are not possible in the context of homological algebra (e.g.
deriving non-additive functors).
If time allows we will turn to the deeper homotopy theoretic relationship of simpli-
cial sets and topological spaces given by the Quillen equivalence. This involves abstract
homotopy theory using model categories. Like in homological algebra, one can define de-
rived functors, a technique which becomes more and more popular also in other fields of
mathematics.

2
1 (Abstract) Simplicial complexes

Recall that an n-simplex is the convex hull of a set of n + 1 points in the Rm , for some
m ≥ 0. A face of the simplex is the convex hull of a subset of these points. Moreover
a (geometric) simplicial complex is a set of simplices in some Rm , such that the
intersection of two simplices is a face of each simplex or empty. We can stretch a simplex
without changing its homeomorphic type. So topologically the exact coordinates in the
ambient space Rm are redundant, as long as we keep in mind how the simplices intersect.
As a preparation for simplicial sets we now introduce (abstract) simplicial complexes,
which can be thought of as “construction plans” for (geometric) simplicial complexes.
Compared to simplicial sets, they are more easily to understand geometrically. However
they have their disadvantages, which motivate the introduction of the more flexible notion
of simplicial sets.

Definition 1.1 (i) An (abstract) simplicial complex1 is a set X, together with a


set S(X) of subsets of X with
a) 0 < ]s < ∞, for all s ∈ S(X).2
b) {x} ∈ S(X), for all x ∈ X.
c) ∅ =
6 t ⊂ s ∈ S(X) ⇒ t ∈ S(X).
The elements of X are called vertices, the elements of S(X) simplices.
f
(ii) A simplicial map between simplicial complexes is a map X −→ Y with
f (s) ∈ S(Y ), for all s ∈ S(X).

Remark 1.2
Let X be a set.
(i) We define a simplicial complex D(X) by setting
D(X) := X, SD(X) := {{x}; x ∈ X}.
It is the smallest simplicial complex with vertices X.
We call it the discrete simplicial complex with vertices X.
1
As we will only deal with abstract simplicial complexes, we will just write “simplicial complex” for
short.
2
Most commonly the cardinality of a set S is denoted by |S|. However to avoid confusion with a later
definition, throughout this text the cardinality will be denoted by ]S.

3
Chapter 1. (Abstract) Simplicial complexes

(ii) We define another simplicial complex I(X) by setting

I(X) := X, SI(X) := {s ⊂ X; 0 < ]s < ∞}.

It is the biggest simplicial complex with vertices X.


We call it the indiscrete simplicial complex with vertices X.

Example 1.3
Let n ≥ 0.

(i) The standard combinatorial n-simplex is defined as the simplicial complex


I(n), where n := {0, ..., n}. For n = 2 the space we have in mind looks like

0 2

where the triangle in the middle is a 2-simplex.

(ii) The boundary of I(n) is given by

∂I(n) := n, S∂I(n) := {s ⊂ n; 0 < ]s < n}.

For n = 2, we can draw the same picture as in (i) without the connecting 2-simplex.

Remark 1.4
Every simplicial complex is the union3 of its finite simplicial subcomplexes.
Indeed every simplex is finite and so lies in some finite simplicial subcomplex.

1.1 Geometric realization


In this section we describe the topological space, that an (abstract) simplicial complex
stands for. We do this by assigning a topological space to an abstract simplicial complex
in a natural way. Explicitly, given an (abstract) simplicial complex X, we define
( )
X M X
|X| := a = ax · x ∈ R · x; ax ≥ 0, ax = 1, {x ∈ X; ax 6= 0} ∈ S(X) ,
x x∈X x∈X
L
where x∈X R · x denotes the real vector space of formal linear combinations over the
elements x ∈ X.

Example 1.5
Let n ≥ 0.

3
meaning that a subset of the union is a simplex, if it is a simplex in some subcomplex.

4
1.1. Geometric realization

(i) The set |I(n)| coincides with the standard (geometric) n-simplex

|∆n | := {(a0 , ..., an ) ∈ [0, 1]n+1 ; a0 + ... + an = 1}.

(ii) The set |∂I(n)| is its boundary.

(iii) More generally, for every subcomplex X ≤ I(n), we see that |X| ≤ |I(n)| is the
union of all geometric simplices in |I(n)| corresponding to simplices in X.

Up to now |X| is only a set, that we still need to put a topology on. Every simplicial
complex X with n = ]X < ∞ can be considered as a simplicial subcomplex X ≤ I(n − 1),
so it seems natural to consider |X| ⊂ |∆n−1 | ⊂ Rn = RX as a subspace of the product
space. However for infinite X, we will need a finer topology than the subspace topology
|X| ⊂ RX .

Definition 1.6
fi
Given a family of maps Fi −→ X, whose domains Fi carry a topology.
The final topology on X is the topology, for which a subset U ⊂ X is open, whenever
−1
fi (U ) ⊂ Fi is open, for all i.

Remark 1.7
fi
Suppose X carries the final topology with respect to a family of maps Fi −→ X.
Then the following holds.

(i) fi is continuous, for all i.


f
(ii) A map X −→ Y into a topological space Y is continuous, if and only if f ◦ fi is
continuous, for all i.

Remark 1.8
f
Every simplicial map X −→ Y induces a map
X X
|f | : |X| −→ |Y |, ax · x 7−→ ax · f (x).
x x

Definition 1.9
The geometric realization of a simplicial complex X is defined as the set |X| together
with the following topology.
M Y
• For ]X < ∞ we give |X| ⊂ R·x = R = RX the subspace topology of the
x∈X x∈X
product topology.
|j|
• For ]X = ∞ we give |X| the final topology with respect to all maps |F | ,−→ |X|,
j
where F ,−→ X is the inclusion of a finite subcomplex. That is U ⊂ |X| is open, iff
U ∩ |F | is open, for all finite F ⊂ X.

5
Chapter 1. (Abstract) Simplicial complexes

Proposition 1.10
f
For every simplicial map X −→ Y , the map |f | is continuous.
Proof.

1) For ]X < ∞ the map |f | is the restriction of a linear map RX −→ RY , which is


continuous.

2) For ]X = ∞, let F ⊂ X be a finite subcomplex. Then in the commutative square



|F |  / |X|
|f |
  
|f (F )|  / |Y |

the lower horizontal map is continuous by Definition of the final topology on |Y |


and Remark 1.7 (i), because the image f (F ) ⊂ Y is a finite subcomplex. Moreover
f|
the left vertical map is the realization of the restriction F −→ f (F ), which is
continuous by 1). Hence their composition is continuous, which coincides with the
other composition. By definition of the final topology on |X| and again by Remark
1.7 (ii) also the map |f | must be continuous.

1.2 Simplicial approximation


Remark 1.11
Let X, Y be simplicial complexes.
Then there are far more continuous maps |X| −→ |Y | than those coming from a sim-
plicial map X −→ Y .

The circle can be modeled by the simplicial complex ∂I(2).

0 2

We can construct an explicit homeomorphism

∼ (a1 − a0 ) + (a2 − a0 )i
h : |∂I(2)| −→ S 1 := {x ∈ C; |x| = 1}, a = (a0 , a1 , a2 ) 7−→ .
|(a1 − a0 ) + (a2 − a0 )i|

There is an isomorphism Z −→ π1 (S 1 , 1), sending n ∈ Z to the homotopy class of the
selfmap
pn : S 1 −→ S 1 , x 7−→ xn .

6
1.2. Simplicial approximation

fn
However we can only find simplicial maps ∂I(2) −→ ∂I(2), whose realizations |fn | preserve
the base point h−1 (1) = (0, 1, 0) ∈ |∂I(2)|, which are homotopic to pn , for n = −1, 0, 1.
These are given by
f−1 (k) = 2 − k, f0 (k) = 1, f1 (k) = k, 0 ≤ k ≤ 2.
However, given n ≥ 0 we can choose another simplicial complex Xn modeling S 1 , given
by
Xn = Z/3n, S(Xn ) = {{x}; x ∈ Z/3n} ∪ {{x, x + 1}; x ∈ Z/3n},
and chosing the base point e1 = (0, 1, 0, ..., 0) ∈ |Xn | ⊂ R3n we see that the map of degree
n > 0 can be modeled by the simplicial map given by the canonical quotient map
Xn = Z/3n − Z/3 = X1 ∼ = ∂I(2), [k] 7−→ [k].
[2] [3] [1]

[1] [4] //

[0] [5] [0] [2]

Question 1.12
f
Given two simplicial complexes X, Y and a continuous map |X| −→ |Y |. Can we always

find a simplicial complex X 0 and a homeomorphism |X 0 | −→ |X|, whose composition with
f is homotopic to the realization of a simplicial map?

In case the source X is a finite complex, the answer is yes. The solution lies in sub-
dividing the complex X, as we implicitly did for the circle in the example above. There
are many ways of subdividing a simplicial complex. A canonical way of doing this is the
barycentric subdivsion.

Definition 1.13
The barycentric subdivision of a simplicial complex is defined as the simplicial complex
sdB X, given by
sdB X := S(X), S(sdB X) := {{s0 ( ... ( sm }; m ≥ 0}.

Example 1.14
The barycentric subdivision of I(2) is given by
{1}

{0, 1} {1, 2}
{0, 1, 2}

{0} {0, 2} {2}

7
Chapter 1. (Abstract) Simplicial complexes

with each appearing triangle being a 2-simplex.

Lemma 1.15
For every finite simplicial complex |X| is a compact Hausdorff space.
Proof. For every simplex s ∈ S(X) we let As be the preimage of 1 under the map
X
[0, 1]X −→ R, a 7−→ ax .
x∈s

As this map is continuous and {1} ⊂ R is a closed subset, As is a closed subset of [0, 1]X . It
S the set of elements a ∈ |X| with ax = 0, for all x ∈ X\s. Hence X
is precisely by construction
|X| = s∈S(X) As is a finite union of closed subsets. So also |X| ⊂ [0, 1] is closed and
therefore compact, as [0, 1]X is compact. With [0, 1]X also |X| is Hausdorff.
2

Lemma 1.16
f
Let X −→ Y be a continuous bijection from a compact space X to a Hausdorff space Y .
Then f is a homeomorphism.
Proof. Let U ⊂ X be an open subset and x ∈ U . For every y ∈ Y \f (U ) there are open
subsets y ∈ Uy ⊂ Y and f (x) ∈ Vy ⊂ Y Swith Uy ∩ Vy = ∅, because Y is Hausdorff. As X
is compact, the open covering X = U ∪ y∈Y f −1 Uy has a finite subcovering correspondig
to elements y1 , ..., yn ∈ Y . So X\U ⊂ f −1 Uy1 ∪ ... ∪ f −1 Uyn or equivalently Y \f (U ) ⊂
Uy1 ∪ ... ∪ Uyn . Setting V := Vy1 ∩ ... ∩ Vyn , we get

V ∩ (Y \f (U )) ⊂ V ∩ (Uy1 ∪ ... ∪ Uyn ) = (V ∩ Uy1 ) ∪ ... ∪ (V ∩ Uyn ) = ∅.

In other words x ∈ V ⊂ f (U ) is an open environment of x in f (U ). So f (U ) ⊂ Y is open,


which proves that f −1 is continuous and hence f is a homeomorphism.
2

Theorem 1.17
There is a natural homeomomorphism

X X as
hX : |sdB X| −→ |X|, as · s 7−→ · x.
]s
s∈S(X) s∈S(X),
x∈s

Proof. For a ∈ |sdB X| we have as ≥ 0, for all s ∈ S(X). Hence


X as
≥ 0, x ∈ X.
]s
x∈s∈S(X)

It follows that hX (a) ∈ |X|, because we also have


X as X as X
= ]s · = as = 1.
]s ]s
s∈S(X), s∈S(X) s∈S(X)
x∈s

8
1.2. Simplicial approximation

f
By construction h is a natural map, i.e. for every simplicial map X −→ Y we get a
commutative square
hX
|sdB X| / |X|
|sdB f | |f |
 hY

|sdB Y | / |Y |.
We prove that hX is continuous by reduction on finite subcomplexes.

• For ]X < ∞, the map hX is the restriction of a linear map between finite-dimensional
real vector spaces, hence continuous.

• For ]X = ∞, let F ⊂ sdB X be a finite subcomplex. Considering F 0 := s∈F s


S
as a finite subcomplex of X, we see that F ⊂ sdB F 0 , so the canonical inclusion
j sdB (j 0 ) j0
factors as F ,−→ sdB F 0 ,−→ sdB X with F 0 ,−→ X. Using that h is natural we get
hX ◦ |sdB (j 0 )| ◦ |j| = |j 0 | ◦ hF 0 ◦ |j| and the latter is continuous by what we have just
proven. Hence hX is continuous by definition of the final topology on |sdB X|.

Next we prove that hX is bijective by constructing an inverse map. Defining ΣU :=


P
x∈U 1 · x, for U ⊂ X, every a ∈ |X| can uniquely be written as

a = a0 · Σs0 + ... + am · Σsm , ∅=


6 s0 , ..., sm ⊂ X, si ∩ sj = ∅, 0 < a0 < ... < am . (1.1)

X k
We define a map |X| −→ |sdB X| by setting
X
kX (a) := (]s0 + ... + ]sm ) · a0 · (s0 ∪ ... ∪ sm ) + (]si + ... + ]sm ) · (ai − ai−1 ) · (si ∪ ... ∪ sm ).
1≤i≤m

Every coefficient is greater than zero by assumption on ai . Moreover a ∈ |X| implies


s0 ∪ ... ∪ sm ∈ S(X) and hence {sm ( (sm−1 ∪ sm ) ( ... ( (s0 ∪ ... ∪ sm )} ∈ S(sdB X) by
definition of sdB X. Again using a ∈ |X| we see that
X
(]s0 + ... + ]sm ) · a0 + (]si + ... + ]sm ) · (ai − ai−1 ) = a0 · ]s0 + ... + am · ]sm = 1.
1≤i≤m

This shows that kX (a) ∈ |sdB X|, so kX is well-defined. We compute


X
hX kX (a) = a0 · Σ(s0 ∪ ... ∪ sm ) + (ai − ai−1 ) · Σ(si ∪ ... ∪ sm )
1≤i≤m

= a0 · Σs0 + ... + am · Σsm .

Similarly for b = b0 · t0 + ... + bm · tm ∈ |sdB X| with tm+1 := ∅ ( tm ( ... ( t0 ∈ S(X)


and bi > 0, we have
  X  b0 
b0 bm bi
hX (b) = kX · Σt0 + ... + · Σtm = + ... + · Σ(ti \ti+1 ),
]t0 ]tm 0≤i≤m
]t0 ]ti

9
Chapter 1. (Abstract) Simplicial complexes

which is an element in |X| of the form (1.1). So applying kX , we get


b0 X bi
kX hX (b) = ]t0 · · t0 + ]ti · · ti = b.
]t0 1≤i≤m
]ti

We have shown that hX is a bijection with h−1


X = kX .
It remains to check that also kX = h−1
X is continuous, which again can be reduced to
the finite case.
• For ]X < ∞, we also have ]sdB X < ∞. Hence |sdB X| and |X| are compact and
Hausdorff by Lemma 1.15, which using Lemma 1.16 implies that hX is a homeo-
morphism.
j
• For ]X = ∞ and every finite simplicial subcomplex F ,−→ X we have |j| ◦ hF =
hX ◦|sdB j|, hence h−1 −1
X ◦|j| = |sdB j|◦hF is continuous, as we have just proven. Since
X is the union of its finite simplicial subcomplexes and carries the final topology
j
for all these inclusions |F | −→ |X|, Remark 1.7 (ii) implies that also h−1
X = kX is
continuous.
2

Lemma 1.18 (Lebesgue) S


Let X be a compact metric space and X = U ∈U U be an open cover.
Then there is a number, called the Lebesgue number δ > 0, such that:
∀x ∈ X ∃Ux ∈ U : B<δ (x) ⊂ Ux .
Proof. For every x ∈ X there is an Ux ∈ U, such S that x ∈ Ux . As Ux is open, we find a
δx > 0, such that B<δx /2 (x) ⊂ Ux . Hence X = x∈X B<δx (x) is an open cover, which by
compactness of X has a finite subcover corresponding to elements x1 , ..., xn ∈ X. Then
δ := min{δx1 , ..., δxn } is the desired number. For if x ∈ X, there is a 1 ≤ j ≤ n with
x ∈ B<δxj (xj ). Moreover, for every y ∈ B<δ (x) we have

d(y, xj ) ≤ d(y, x) + d(x, xj ) < δ + δxj ≤ 2δxj ,


which implies that y ∈ B<2δxj (xj ), which by the choice of δxj is contained in Uxj .
2
Before we can prove the simplicial approximation Theorem we will need the following
key lemma. Roughly speaking it says that by iterated subdivision we can make simplices
arbitrarily small.

Lemma 1.19
Let X be a finite simplicial complex, N > 0 and a, b ∈ |sdN
B X|, such that

{x ∈ sdN N
B X; ax > 0 or bx > 0} ∈ S(sdB X).
1 N
Then khN N X
X (a) − hX (b)k1 < 2 · (1 − ]X ) , where k-k1 is the 1-norm on |X| ⊂ R , given by

kv1 · x1 + ... + vn · xn k1 = |v1 | + ... + |vn |, v1 , ..., vn ∈ R, x1 , ..., xn ∈ X.

10
1.2. Simplicial approximation

Proof. As X is finite, we can define nX := max{]s; s ∈ S(X)}. On the one hand, for
every chain of X-simplices s1 ( ... ( sn we have nX ≥ ]sn ≥ n, which proves nX ≥ nsdB X .
On the other hand, for every s = {x1 , ..., xn } ∈ S(X) with n = nX we have

nX = n = ]{{x1 } ( {x1 , x2 } ( ... ( {x1 , ..., xn }} ≤ nsdB X .

So all in all nX = nsdB X .


L h̄X L
By definition
L the map hX is the restriction of a linear map s∈sdB X R·s −→ x∈X R·x.
Let v ∈ s∈sdB X R · s with {s ∈ sdB X; vs > 0} ∈ S(sdB X). Then v is of the form

v = v1 · {x1 } + v2 · {x1 , x2 } + ... + vn · {x1 , ..., xn }, v1 , ..., vn ∈ R, {x1 , ..., xn } ∈ S(X).

Assuming v1 + .... + vn = 0, we get


v2 vn
kh̄X (v)k1 = v1 · x1 + · (x1 + x2 ) + ... + · (x1 + ... + xn )
2 n 1
 v2 vn   v2 vn  vn
= v1 + + ... + · x1 + + ... + · x2 + ... + · xn
2 n 2 n n 1
v2 vn v2 vn vn
= v1 + + ... + + + ... + + ... +
 2 n 2 n  n 
1 1 1 1 1
= 1− · v1 + − · v2 + ... + − · vn−1
n 2 n n−1 n
v2 vn vn
+ + ... + + ... +
2 n n
1 1 1 1 1
≤ 1− · |v1 | + − · |v2 | + ... + − · |vn−1 |
n 2 n n−1 n
 
1 1 1
+ · |v2 | + ... + · |vn | + ... + · |vn |
2 n n
       
1 1 1 1
= 1− · |v1 | + ... + 1 − · |vn | = 1 − · kvk1 ≤ 1 − · kvk1 ,
n n n nX
where we used v1 + ... + vn = 0 for the fourth equality and n ≤ nXLfor the last inequality.
N −1 −1 N −1
Now by assumption v := h̄X (a − b) = hN X (a) − hX (b) ∈ s∈sdB (X) R · s is of the
stated form, because X
(as − bs ) = 1 − 1 = 0.
s∈sdN
BX

Hence using nX = nsdB X = ... = nsdNB X , we can prove by induction that


 
1 −1
N N N
hX (a) − hX (b) 1 = h̄X (a − b) 1 ≤ 1 − · h̄N
X (a − b) 1
nX
 N
1
≤ ... ≤ 1 − · ka − bk1 .
nX
N
Since a, b ∈ |sdN
B X| ⊂ [0, 1]
sdB X
, we have ka − bk1 ≤ 2. Moreover nX ≤ ]X, which
concludes the proof.
2

11
Chapter 1. (Abstract) Simplicial complexes

Theorem 1.20 (Simplicial Approximation)


Let X and Y be simplicial complexes and suppose X is finite. Given a continuous map
f
between their geometric realizations |X| −→ |Y |.
g
Then there is an N ≥ 0 and a simplicial map sdN N
B X −→ Y , such that f ◦ hX and |g|
are homotopic.
Proof. The idea is to subdivide X sufficiently many times, such that the value of f alters
inside the realization of a simplex in Y , when moving between adjacent vertices in X.
Every y ∈ Y considered as an element 1 · y ∈ |Y | has a canonically defined environment
Uy := {a ∈ |Y |; ay > 0}. It is an open subset of |Y |, because Uy ∩ |F | ⊂ RF is open, for
all finite F ⊂ Y . Geometrically every element of Uy lies in the realization of a simplex
containing y. By Lemma 1.15 the space |X| is compact, as X is finite. Its topology comes
X
from the metric induced by the 1-norm S on R−1. So by Lemma 1.18 there is a Lebesgue
1 N
number δ > 0 for the open cover |X| = y∈Y f Uy . Let N > 0, such that 2·(1− ]X ) < δ.
N
For every x ∈ sdB X Lebesgue’s lemma provides an element g(x) ∈ Y such that
−1
B<δ (hN
X (x)) ⊂ f Ug(x) .
g
We claim that this defines a simplicial map sdN B X −→ Y . Let s ∈ S(sdB X) and choose
x ∈ s. For every x ∈ s we have khX (x) − hX (x0 )k1 < 2 · (1 − ]X
0 N N 1 N
) < δ by Lemma 1.19.
N N 0 −1 N
Hence hX (x) ∈ B<δ (hX (x )) ⊂ f Ug(x0 ) . In other words (f hX (x))g(x0 ) > 0 or

g(x0 ) ∈ {y ∈ Y ; f hN
X (x)y > 0} =: sx ,

0
which is a simplex in Y , because f hNX (x) ∈ |Y |. As x ∈ s was arbitrary, we have proven
that g(s) ⊂ sx ∈ S(Y ), so g(s) ∈ S(Y ).
Similarly, for every a ∈ |sdN N N
B X| we let s := {x ∈ sdB X; ax > 0} ∈ S(sdB X). Then the
same argument for a instead of x shows that g(s) ⊂ {y ∈ Y ; f hN X (a)y > 0} =: sa ∈ S(Y ),
N
which proves that |g|(a) and f hX (a) lie in the geometric realization of the same simplex
sa . It follows that

H : [0, 1] × |sdN
B X| −→ |Y |, (1 − t) · f hN
X (a) + t · |g|(a)

constitutes a well-defined homotopy from f hN


X to |g|.
2

1.3 Products
In our abstraction process of avoiding topology, we would like to have the notion of a
homotopy between two simplicial maps.

1) We need a model for the interval. This may be given by the simplicial complex
I(1), whose geometric realization is |I(1)| = |∆1 |. The Simplicial Approximation
Theorem 1.20 suggests that we may also need to consider its iterated subdivisions.

12
1.3. Products

2) We need the notion of a product of two simplicial complexes, which should be


(πX ,πY )
compatible with geometric realization, i.e. the canonical map |X×Y | −→ |X|×|Y |
should be a homeomorphism.
There is a natural product construction.

Remark 1.21
The (categorial) product X × Y of two simplicial complexes X and Y is the simplicial
complex, whose set of vertices is the cartesian product X × Y and whose simplices are

S(X × Y ) = {∅ =
6 p ⊂ s × t; s ∈ S(X), t ∈ S(Y )}.
πX Y π
The projections X ←− X × Y −→ Y are simplicial and are universal with this property.
This means that given two simplicial maps f and g there is a unique simplicial map (f, g)
fitting in the commutative diagram

Z
f g
∃!(f,g)
v  (/
Xo πX X ×Y πY Y.

Remark 1.22
For two simplicial complexes X and Y the two projections induce a natural map
(|πX |,|πY |)
|X × Y | −→ |X| × |Y |,

which is continuous by definition of the product topology.

Unfortunately this product does not have the desired properties as the following exam-
ple demonstrates.

Example 1.23
We have I(1) × I(1) = I(1 × 1), because 1 × 1 ∈ SI(1 × 1).
(i) This means we have six 1-simplices, four 2-simplices

s1,1 := {(0, 0), (0, 1), (1, 0)}, s1,0 := {(0, 0), (0, 1), (1, 1)},
s0,1 := {(0, 0), (1, 0), (1, 1)}, s0,0 := {(0, 1), (1, 0), (1, 1)},

and moreover one 3-simplex.

(0, 1) (1, 1)

(0, 0) (1, 0)

The geometric realization |I(1) × I(1)| is homeomorphic to the standard 3-simplex


|∆3 |, which is not homeomorphic to the product |I(1)| × |I(1)| = |∆1 | × |∆1 |.

13
Chapter 1. (Abstract) Simplicial complexes

(ii) However there is a simplicial subcomplex P ⊂ I(1) × I(1) with same vertices, such
that the composition |P | ,−→ |I(1) × I(1)| −→ |I(1)| × |I(1)| is a homeomorphism.
Its top dimensional simplices are s1,0 and s0,1 , which are precisely those simplices of
I(1) × I(1) being totally ordered by the product order.

(0,O 1) / (1, 1)
; O

(0, 0) / (1, 0)

We can generalize this product construction by giving simplicial complexes an orienta-


tion in form of a partial order.

Definition 1.24
An ordered simplicial complex X is a simplicial complex X together with a partial
order on its set of vertices, restricting to a total order on each of its simplices.
An ordered simplicial map between two ordered simplicial complexes X and Y is a
simplicial map X −→ Y preserving the order.

Proposition 1.25
The (categorial) product X × Y of two ordered simplicial complexes X and Y is the
subcomplex of their product of simplicial complexes, whose simplices are those being totally
ordered by the product order.
(|πX |,|πY |)
Then |X × Y | −→ |X| × |Y | is a continuous bijection.
In particular it is a homeomorphism, if X and Y are finite.
Proof. By construction the decscribed categorial product is universal for ordered simplicial
maps. To see that the map (|πX |, |πY |) is bijective, we need another description for the
set |X|.
Given two ordered simplicial complexes A, B we let Simpc (A, B) denote the set of
f
ordered simplicial maps A −→ b, which are cocontinuous. This means that f preserves
suprema, i.e. sup f (U ) = f (sup U ), for all subsets U ⊂ A. Now suppose B is finite and let
A := I(]0, 1]), that is the ordered simplicial complex, whose vertices are elements of the
half-open unit interval ]0, 1] with the canonical partial order. We define a natural bijection

φB : |B| −→ Simpc (I(]0, 1]), B)
as follows. The image of an element a = a1 · x1 + ... + an · xn ∈ |B| with x1 < ... < xn and
a1 , ..., an > 0 is the map
φB (a) :]0, 1] −→ B, ]a1 + ... + ai−1 , a1 + ... + ai ] 3 v 7−→ xi .
Vice versa, for f ∈ Simpc (I(]0, 1]), B) there are elements a1 , ..., an ∈]0, 1], such that
f (]0, 1]) = {f (a1 ) < ... < f (a1 + ... + an )} ∈ S(B). As f preserves suprema, we can find
maximal a1 , ..., an ∈]0, 1] with this property. Since f (1) = f (a1 + ... + an ), this implies
a1 + ... + an = 1, and hence a1 · f (a1 ) + ... + an · f (a1 + ... + an ) ∈ |B| is a unique preimage
for f under φB .

14
1.4. Collapsing subspaces

Now for finite X and Y we obtain a commutative square


(|πX |,|πY |)
|X × Y | / |X| × |Y |

φX×Y o φX ×φY o
 ((πX ◦-),(πY ◦-)) 
Simpc (I(]0, 1]), X × Y ) / Simpc (I(]0, 1]), X) × Simpc (I(]0, 1]), Y ).

As the product of ordered simplicial complexes is universal for cocontinuous simplicial


maps, the lower horizontal map and hence also the upper horizontal map is a bijection.
Again finiteness of X, Y and hence X × Y implies that |X × Y | and |X| × |Y | are compact
Hausdorff spaces by Lemma 1.15, so (|πX |, |πY |) is a homeomorphism by Lemma 1.16.
As before infinite X and Y may be written as the union of its finite subcomplexes and
the commutative square
(|πX |,|πY |)
|X × Y | / |X| × |Y |

∼ /
[ [
|F × G| |F | × |G|
F ⊂X, F ⊂X,
G⊂Y, G⊂Y,
finite finite

shows that (|πX |, |πY |) is a continuous bijection. In general the right equality only holds
as sets, because the topology of the union is finer than the product topology.
2

Remark 1.26
Every simplicial complex X can be ordered by choosing an arbitrary total order on its
underlying set of vertices.
(i) This does not lead to a functorial construction.
f
(ii) But given a simplicial map X −→ Y , by the order-extension principle we can
choose compatible total orders on X and Y , such that f becomes an oriented sim-
plicial map.

Remark 1.27
For every simplicial complex X, its barycentric subdivision sdB X is an ordered simplicial
complex via the inclusion order.
Every simplicial map induces an ordered simplicial map between the barycentric subdi-
visions.

1.4 Collapsing subspaces


Apart from the bad bevavior of products, simplicial complexes have another flaw. Namely
geometric realization does not commute with collapsing subspaces. We will demonstrate
this with an example.

15
Chapter 1. (Abstract) Simplicial complexes

Example 1.28
Consider the bounday subcomplex ∂I(1) ⊂ I(1), whose only simplices are the subsets of
cardinality 1.

(i) We have a homeomorphism |I(1)|/|∂I(1)| ∼ = S 1 . So if |-| would commute with quo-


tients I(1)/∂I(1) would be a very simple model of the circle.

(ii) But the quotient simplicial complex I(1)/∂I(1) consists of a single point, so it does
not model the circle.

(iii) Obviously the problem cannot be fixed by introducing orientations.

As a consequence, also the constructions of mapping cylinders and cones are getting
more complicated.

16
2 Simplicial sets

2.1 Semi-simplicial sets


We introduced the notion of an abstract simplicial complex as a “construction plan” for
a geometric simplicial complex. It was easy to imagine the geometric realization, that a
simplicial complex is standing for. But we also saw that the notion behaves badly when
it comes to forming products and collapsing subspaces.
Instead of modeling simplicial complexes, we now try to model CW-complexes. We will
do this in a “simplicial way”, meaning that we will build up the CW-complex by glueing
simplices.

Remark 2.1
Let n > 0.

(i) Then there is a homeomorphism


∼ x
|∂I(n)| = ∂|∆n | −→ H ∩ S n ∼
= S n−1 , b + x 7−→ ,
kxk
1
where b = n+1 · (1, ..., 1) ∈ |I(n)| = |∆n | is the barycenter of the standard n-simplex
n+1
and H ⊂ R is the hyperplane of elements a ∈ Rn+1 with a0 + ... + an = 0.

(ii) It extends to a homeomorphism

∼ tx
|I(n)| = |∆n | −→ H ∩ Dn+1 ∼
= Dn , b + tx 7−→ , 0 ≤ t ≤ 1.
kxk

In particular the inclusion |∂I(n)| ,−→ |I(n)| is homeomorphic to S n−1 ,−→ Dn .


S
Using this identification a CW-complex is the union of spaces C = n≥0 Cn , where C0
is a discrete set of points and Cn is obtained by glueing a set of standard n-simplices
along their boundary to Cn−1 , for each n > 0. More precisely suppose Xn is an index-set
of n-simplices we want to glue, for each n ≥ 0. We consider Xn as a discrete space. We

have a homeomorphism j0 : X0 ×|I(0)| −→ C0 . Moreover we get Cn by glueing Xn ×|I(n)|
gn
to Cn−1 and identifying the boundaries along a certain map Xn × |∂I(n)| −→ Cn−1 . This

17
Chapter 2. Simplicial sets

results in a commutative square


gn
Xn × |∂I(n)| / Cn−1 (2.1)
 _ _

 jn 
Xn × |I(n)| / Cn .

Now instead of working with arbitrary continuous maps gn , we want this map to be
“simplicial” in a certain sense. Let us take a closer look at the boundary ∂I(n).

Definition 2.2 (i) For n > 0 and 0 ≤ i ≤ n the i-th coface map is the unique
di
/ di (n − 1), i.e.
monotone injection n − 1 ,−→ n with i ∈

i k, 0 ≤ k < i,
d : n − 1 ,−→ n, k 7−→
k + 1, i ≤ k ≤ n − 1.

(ii) For 0 ≤ i ≤ n the i-th codegeneracy map is the unique monotone surjection
si
n + 1 − n with si (i) = si (i + 1), i.e.

i k, 0 ≤ k ≤ i,
s : n + 1 ,−→ n, k 7−→
k − 1, i < k ≤ n + 1.

These unorthodox names will become clear later.

Remark 2.3
Under the homeomorphism |I(n)| ∼
= |∆n | we have the following geometric description of
i i
d and s , where 0 ≤ i ≤ n.
(i) |I(di )| : |I(n − 1)| −→ |I(n)| corresponds to the inclusion of the i-th face of the
standard n-simplex.
(ii) |I(si )| : |I(n + 1)| −→ |I(n)| corresponds to the surjection identifiying the dimen-
sions i and i + 1 of the n-simplex.
In particular we also get the description
[ [
|∂I(n)| = di I(n − 1) = |di |(|I(n − 1)|).
0≤i≤n 0≤i≤n

gn
Using this description of |∂I(n)| we see that every map Xn × |∂I(n)| −→ Cn is the
union of the maps gn ◦ (id × |di |), where 0 ≤ i ≤ n. To make gn “simplicial”, we may use
jn−1
the maps Xn−1 × |I(n − 1)| −→ Cn−1 and suppose the following diagram commutes
di ×id
Xn × |I(n − 1)| / Xn−1 × |I(n − 1)|
id×|di | jn−1
 gn 
Xn × |∂I(n)| / Cn−1 ,

18
2.1. Semi-simplicial sets

d
i
with certain maps Xn −→ Xn−1 , for 0 ≤ i ≤ n. In other words we assume that the map
gn is the union of the maps jn−1 ◦ (di × id). Under this assumption an induction on n ≥ 0
shows
,
a
Cn = Xk × |I(k)| (di (x), a) ∼ (x, |di |(a)), x ∈ Xk , a ∈ |I(k − 1)|, 0 ≤ i ≤ k.
0≤k≤n

So far we did not care, wheather the maps jn−1 ◦ (di × id) are compatible to define the
map gn . For every 0 ≤ i < j ≤ n we have a commutative diagram

di /n−1
n−2
dj−1 dj
 di 
n−1 / n,

and dj di (n − 2) = di dj−1 (n − 2) = di (n − 1) ∩ dj (n − 1). In the explicit description of Cn


the map jn is the inclusion followed by the quotient map and we see that we need the
corresponding condition

di dj = dj−1 di , 0 ≤ i < j ≤ n. (2.2)

Definition 2.4
A semi-simplicial set or ∆-set is a collection of sets X = (Xn )n≥0 , together with
so-called face maps
di
Xn −→ Xn−1 , 0 ≤ i ≤ n,
satisfying (2.2).
f
A homomorphism of semi-simplicial sets X −→ Y is a collection of maps
fn
Xn −→ Yn , n ≥ 0,

such that di ◦ fn = fn−1 ◦ di , for all 0 ≤ i ≤ n.

Semi-simplicial sets were first introduced 1950 by Eilenberg-Zilber [? ], who immediately


realized that they still have a flaw. Namely, we are only able to glue an n-simplex by
identifying its boundary with a union of (n − 1)-simplices in Cn−1 . But often we also
want to identify it with lower dimensional simplices (e.g. the simplest CW-model for the
n-sphere is given by glueing an n-simplex to a single point). We can come around this
problem by adding degenerate simplices to Xn , which are formal n-simplices that in fact
stand for lower dimensional simplices. This means that some dimensions in their geometric
realizations coincide. Such degenerate simplices can be realized by also factoring out
along maps corresponding to the already introduced codegeneracy maps (cf. our explicit
description of Cn ). So similarly to the face maps we are postulating the existence of
degeneracy maps
si
Xn −→ Xn+1 , 0 ≤ i ≤ n.

19
Chapter 2. Simplicial sets

Geometrically a simplex x ∈ Xn+1 lies in the image of the map si , if and only if the
dimensions i and i + 1 in its realization coincide. Of course the presence of degeneracy
maps involves a lot more relations similar to (2.2) and finally leads to the notion of
simplicial sets. Working with all these identities is a mess, but luckily there is a much
more compact definition in the language of category theory.
The careful reader may have noticed that, for each map di from dimension n − 1 to
dimension n, we required a map di in the opposite direction from dimension n to n − 1.
The same correspondence holds for si and si . The necessary relations for di were deduced
from the relations that hold for di and si . The reader familiar to category theory will
conclude that a simplicial set must be a contravariant Set-valued functor on a certain
category, whose objects are the standard n-simplices.

2.2 Categories
Definition 2.5
A category C consists of the following data.
• A class1 of objects Obj(C) (also denoted by C) and for two objects A, B ∈ C a set
of (homo-)morphisms C(A, B) (sometimes also denoted by HomC (A, B)), whose
f
elements will also be denoted by arrows A −→ B.

• For each object A ∈ C, there is a homomorphism idA ∈ C(A) := C(A, A), called the
identity on A.

• Moreover there are composition maps

C(B, C) × C(A, B) −→ C(A, C), (g, f ) 7−→ g ◦ f,

satisfying the relations


f g h
h ◦ (g ◦ f ) = (h ◦ g) ◦ f, f ◦ idA = f = idB ◦ f, A −→ B −→ C −→ D.

F
A functor C −→ D between two categories C and D is given by the following data.
F
• A “map”2 between the classes of objects Obj(C) −→ Obj(D).

• For each pair of objects A, B ∈ C, there is a map

C(A, B) −→ D(F (A), F (B)), f 7−→ F (f ).

• It preserves the structure, i.e.


f g
F (idA ) = idF (A) , F (g ◦ f ) = F (f ) ◦ F (g), A −→ B −→ C.
1
Classes extend the notion of sets. We need this term to be able to talk about the class of all sets, which
cannot be a set itself. Further details about this problem can be found in any book about category
theory.
2
A map between classes is defined in the same way as maps between sets.

20
2.2. Categories

t F,G
A natural transformation F −→ G between two functors C −→ D is a family of
tA
morphisms (F (A) −→ G(A))A∈C inducing a commutative square
tA
F (A) / G(A)
F (f ) G(f )
 tB

F (B) / G(B),

for all X, Y ∈ C and f ∈ C(A, B).

Example 2.6
The following are categories.

(i) The category Set has the class of sets as objects and for two sets X, Y ∈ Set, the
homomorphisms Set(X, Y ) are the maps from X to Y . Composition and identities
are defined in the usual sense.

(ii) Similarly we have T op the category of topological spaces with continuous maps as
U
homomorphisms. There is a functor T op −→ Set, which assigns to a space its
underlying set. It is called a forgetful functor, because it “forgets” the topology.

(iii) Similarly we have the category Grp of groups, Ring of rings, ... All these “algebraic”
categories have an obvious forgetful functor U to the category Set.

(iv) The category Simp of simplicial complexes, whose morphisms are simplicial maps.
U
Next to the forgetful functor Simp −→ Set we already constructed the geometric
|-|
realization functor Simp −→ T op.

(v) We will also write CAT for the collection of categories and CAT(C, D) for the class
of functors from C to D, which are composed in the canonical manner.
However some attention is required here: Similarly as the collection of sets is not a
set, the collection of categories is not a class. In particular CAT is not a category
in the sense we just defined it.

Definition 2.7
A category is called small, if the class of objects is actually a set.
We call Cat the category of small categories with functors as morphisms.

Remark 2.8
Let I and C be categories and suppose I is small.
Then the class of functors CAT(I, C) becomes a category, whose morphisms are natural
transformations.

Definition 2.9
For a category C its opposite category C op is the category with the same objects and

21
Chapter 2. Simplicial sets

morphisms
C op (A, B) := C(B, A), A, B ∈ C op .
The composition is induced by the composition in C:

C op (B, C) × C op (A, B) −→ C op (A, C), (g, f ) 7−→ g ◦ f := f ◦C g.


F
A functor C op −→ D is also called a contravariant functor from C to D.

The most important example of contravariant functors are the Hom-functors. An ex-
ample appears in linear algebra, when assigning to a vector space its dual.

Proposition 2.10
Every C ∈ C ∈ CAT induces a functor

C(-, C) : C op −→ Set, B 7−→ C(B, C), C(B, B 0 ) 3 f 7−→ C(f, C) := (- ◦ f ).

The construction is natural in C, meaning that (for small C) it induces a functor

C −→ CAT(C op , Set), C 7−→ C(-, C), C(C, C 0 ) 3 f 7−→ C(-, f ) := (f ◦ -).


Proof. Let C ∈ C.

• For every B ∈ C we have C(-, C)(idB ) = idC(B,C) , because

C(-, C)(idB )(f ) = f ◦ idB = f = idC(B,C) (f ), f ∈ C(B, C).

f g
• Moreover for morphisms B −→ B 0 −→ B 00 we have by associativity

C(-, C)(g ◦ f ) = (- ◦ (g ◦ f )) = ((- ◦ g) ◦ f ) = C(-, C)(g) ◦ C(-, C)(f ).

This proves that C(-, C) is a functor.


For the second statement let f ∈ C(C, C 0 ). We need to check that f defines a natural
transformation
C(-, f ) := (- ◦ f ) : C(-, C) −→ C(-, C 0 ).
This holds, because for every g ∈ C(B, B 0 ) the diagram below commutes

C(B 0 ,f )=(f ◦-)


C(B 0 , C) / C(B 0 , C 0 )
C(g,C)=(-◦g) (-◦g)=C(g,C 0 )
 C(B,f )=(f ◦-) 
C(B, C) / C(B, C 0 ),

because again composition is associative.


2

Probably the most important Lemma in category theory is the following one by Yoneda.
It also appears in every (modern) algebraic geometry lecture.

22
2.2. Categories

Lemma 2.11 (Yoneda)


For every C ∈ C ∈ CAT and F ∈ CAT(C op , Set), there is a bijection

d : CAT(C op , Set)(C(-, C), F ) −→ F (C), t 7−→ tC (idC ).

It is natural in F and C.
Proof. Every x ∈ F (C) induces a natural transformation e(x) by setting

e(x)B : C(B, C) −→ F (B), f 7−→ F (f )(x).

It remains to check that e(x) is a natural transformation, that e and d are inverse to each
other and that d is natural in F and C. Although we will provide a full proof here, we
advise the reader to do this on his own as an exercise.
Using that F is a functor, for g ∈ C(B, B 0 ) and f ∈ C(B 0 , C) we get

e(x)B ◦ C(g, C)(f ) = e(x)B (f ◦ g) = F (f ◦ g)(x) = F (g) ◦ F (f )(x) = F (g) ◦ e(x)B 0 (f ),

which proves that e(x)B ◦ C(g, C) = F (g) ◦ e(x)B 0 , so e(x) is in fact natural. Both con-
structions are inverse to each other:
• (d ◦ e)(x) = e(x)C (idC ) = F (idC )(x) = x, for all x ∈ F (C),
• (e ◦ d)(t) = t, for all t ∈ CAT(C op , Set)(C(-, C), F ), because for all f ∈ C(B, C)
naturality of t implies

(e ◦ d)(t)B (f ) = e(d(t))B (f ) = e(tC (idC ))B (f ) = F (f )(tC (idC )) = (F (f ) ◦ tC )(idC )


= (tB ◦ C(f, C))(idC ) = tB (C(f, C)(idC )) = tB (idC ◦ f ) = tB (f ).

Next we prove that d is natural in F and C.


• The right term is a functor in F

CAT(C op , Set) −→ Set, F 7−→ F (C),


h h
sending a natural transformation F −→ F 0 to F (C) −→ C
F 0 (C). Moreover the left
term is a functor in F by Proposition 2.10. To check that d is natural in F , we need
to prove, that
dF ◦ CAT(C op , Set)(C(-, C), h) = hC ◦ dC .
But this holds, because for every t ∈ CAT(C op , Set)(C(-, C), F ), we have

dF ◦ CAT(C op , Set)(C(-, C), h)(t) = dF (h ◦ t) = (h ◦ t)C (idC ) = (hC ◦ tC )(idC )


= hC (tC (idC )) = hC (dC (t)) = hC ◦ dC (t).

• The right term is a functor in C, because F is a functor. Moreover the left term
is a functor in C by Proposition 2.10 applied to C ∈ C and then to C(-, C) ∈
CAT(C op , Set)op . To check that d is natural in C, we need to prove, that

dC ◦ CAT(C op , Set)(C(-, g), F ) = F (g) ◦ tC , g ∈ C(C, C 0 ).

23
Chapter 2. Simplicial sets

But for t ∈ CAT(C op , Set)(C(-, C), F ), we have


dC ◦ CAT(C op , Set)(C(-, g), F )(t) = dC (CAT(C op , Set)(C(-, g), F )(t)) = dC (t ◦ C(-, g))
= (t ◦ C(-, g))C (idC ) = (tC 0 ◦ C(-, g)C )(idC )
= tC 0 (C(C, g)(idC )) = tC 0 (g ◦ idC ) = tC 0 (g)
= tC 0 (C(g, C)(idC 0 )) = (tC 0 ◦ C(g, C))(idC 0 )
= (F (g) ◦ tC )(idC 0 ) = (F (g) ◦ dC )(t).

2.3 Simplicial sets


Definition 2.12
The simplex category ∆ is defined as the category of the finite totally ordered sets
n = {0 < 1 < ... < n}, for n ≥ 0, whose morphisms are the monotone maps.

Remark 2.13
Every f ∈ ∆(m, n) can be factored as
f = dip ◦ ... ◦ di1 ◦ sj1 ◦ ... ◦ sjq ,
where {i1 < ... < ip } := n\f (m) and {j1 < ... < jq } := {j ∈ m; f (j) = f (j + 1)}.
In particular ∆ is generated by the maps di and sj . It can be shown that ∆ is completely
determined by di , sj and the following cosimplicial identities
(i) dj di = di dj−1 , i < j,
 i j−1
 ds , i < j,
j i
(ii) s d = id, i = j, j + 1,
 i−1 j
d s, i > j + 1,
(iii) sj si = si sj+1 , i ≤ j.

Definition 2.14
The category of simplicial sets is defined as the functor category sSet := CAT(∆op , Set).

Remark 2.15
For a simplicial set X we will write
Xn := X(n), di := X(di ), si := X(si ), 0 ≤ i ≤ n.
For a map f ∈ ∆(m, n) we will also write f ∗ := X(f ).
(i) Remark 2.13 shows that a simplicial set is completely determined by this data satis-
fying the simplicial identities, dual to the cosimplicial identities of Remark 2.13.
(ii) Similarly the category of semi-simplicial sets is the functor category CAT(∆op
inj , Set),
where ∆inj ≤ ∆ is the subcategory with same objects but only the injective maps in
∆ as morphisms.

24
2.4. Geometric realization

2.4 Geometric realization


Definition 2.16
The geometric realization of a simplicial set X is defined as the topological space
,
a
|X| := Xn × |I(n)| (X(f )(x), a) ∼ (x, |I(f )|(a)), x ∈ Xn , a ∈ |I(m)|, f ∈ ∆(m, n),
n≥0

carrying the final topology with respect to the maps (Xn × |I(n)| −→ |X|)n≥0 .

Remark 2.17
Every f ∈ sSet(X, Y ) induces a continuous map |f | ∈ T op(|X|, |Y |).
|-|
In particular we get a functor sSet −→ T op.

Definition 2.18
Let X ∈ sSet and n ≥ 0.
Sn−1
(i) A simplex x ∈ Xn is called nondegenerate, if x ∈
/ σXn := i=0 si (Xn−1 ).

(ii) We denote by X̃n = Xn \σXn the set of nondegenerate n-simplices.

Lemma 2.19 (Eilenberg-Zilber)


Let X ∈ sSet.
Then for every x ∈ Xn there is a unique nondegenerate y ∈ Xm and a unique monotone
s
surjection n − m, such that x = s∗ (y).
s
Proof. For every x ∈ Xn we consider the set of pairs y ∈ Xm and n − m with s∗ (y) = x.
This set is non-empty, as it contains (y, s) := (x, idn ). The set of appearing dimensions
m is bounded below by 0, so we can find a pair (y, s) with minimal m. Then y must be
nondegenerate, because otherwise we would find a pair of lower dimension.
Now suppose (y 0 , s0 ) is another pair of minimal dimension m. We take a section d for s
(i.e. a ∆-morphism d with sd = id) and compute

y = (sd)∗ (y) = d∗ s∗ (y) = d∗ (x) = d∗ (s0 )∗ (y 0 ) = (s0 d)∗ (y 0 ).

Because y is nondegenerate s0 d ∈ ∆(m, m) is injective, hence s0 d = id. For every k ∈ n


we find a section d such that ds(k) = k. As observed above d is also a section for s0 and
so s0 (k) = s0 ds(k) = s(k) implying s = s0 .
2

Proposition 2.20
For every X ∈ sSet the canonical inclusion induces a continuous bijection (not a homeo-
morphism!)

a
βX : X̃n × (|I(n)|\|∂I(n)|) −→ |X|.
n≥0

25
Chapter 2. Simplicial sets

d
Proof. For (x, a) ∈ Xn × |I(n)| we let m ,−→ n be the unique monotone injection with
image
d(m) = {i ∈ n; ai 6= 0}.
Then the tuple of nonzero a-coordinates defines an element b ∈ |I(m)| with |I(d)|(b) = a.
Moreover by Eilenberg-Zilber’s Lemma 2.19 there is a unique nondegenerate y ∈ X` and
s
a unique m − `, such that d∗ (x) = s∗ (y). In the geometric realization |X| we have

(x, a) = (x, |I(d)|(b)) ∼ (d∗ (x), b) = (s∗ (y), b) ∼ (y, |I(s)|(b)).

By definition of |I(s)|, the coordinates of |I(s)|(b) are sums of coordinates of b. So bi > 0,


for all i ∈ m, implies that also |I(s)|(b)i > 0, for all i ∈ `. Equivalently |I(s)|(b) ∈
/ |∂I(`)|.
This shows that βX is surjective and uniqueness of d, y and s implies that (x, a) 7−→
(y, |I(s)|(b)) defines an inverse map for βX .
2

This shows that the n-cells in the geometric realization |X| bijectively correspond to
the nondegenerate n-simplices X̃n .

2.5 Adjunctions
Adjunctions play an important role, whenever one wants to compare two different cate-
gories. In this section we will give the definition and some basic facts.

Definition 2.21
An adjunction consists of two functors G : C −→
←− D : F and a bijection
C(F (X), Y ) = D(X, G(Y )),

which is natural in X and Y .

(i) The functor F is called the left adjoint to G.

(ii) The functor G is called the right adjoint to F .

Proposition 2.22
For an adjunction C(F (X), Y ) = D(X, G(Y )), we define

• ηX ∈ D(X, GF (X)) as the element corresponding to idF (X) ∈ C(F (X)).

• εY ∈ C(F G(Y ), Y ) as the element corresponding to idG(Y ) ∈ D(G(Y )).

Then η and ε are natural transformations and the adjunction bijection is given by

C(F (X), Y ) −→
←− D(X, G(Y )),
f 7−→ G(f ) ◦ ηX ,
εY ◦ F (g) ←−[ g.

26
2.5. Adjunctions

Proof. We denote the natural bijection of the adjunction by φ, i.e.



φX,Y : C(F (X), Y ) −→ D(X, G(Y )), f 7−→ φ(f ).
Naturality of φ can be expressed as follows:
(i) For all f ∈ C(Y, Y 0 ) and g ∈ C(F (X), Y ), we have
φX,Y 0 (f ◦ g) = (φX,Y ◦ C(X, f ))(g) = (D(X, G(f )) ◦ φX,Y 0 )(g) = G(f ) ◦ φX,Y 0 (g).

(ii) For all f ∈ C(F (X 0 ), Y ) and g ∈ D(X, X 0 ), we have


φX,Y (f ◦ F (g)) = (φX,Y ◦ C(F (g), Y ))(f ) = (D(g, G(Y )) ◦ φX 0 ,Y )(f ) = φX 0 ,Y (f ) ◦ g.

For all g ∈ D(X, X 0 ) this implies


GF (g) ◦ ηX = GF (g) ◦ φX,F (X) (idF (X) ) = φX 0 ,F (X) (F (g) ◦ idF (X) )
= φX 0 ,F (X) (idF (X 0 ) ◦ F (g)) = φX 0 ,F (X 0 ) (idF (X 0 ) ) ◦ g = ηX 0 ◦ g.
So η is natural. For all f ∈ C(F (X), Y ) we have by (i)
φX,Y (f ) = G(f ) ◦ φX,F (X) (idF (X) ) = G(f ) ◦ ηX .
In the same way one proves that ε is natural and that φ−1
X,Y = εY ◦ F (-).
2

Proposition 2.23
Given two functors G : C −→
η
ε
←− D : F and natural transformations idD −→ GF and
F G −→ idC , such that
(i) εF (X) ◦ F (ηX ) = idF (X) , for all X ∈ D,
(ii) G(εY ) ◦ ηG(Y ) = idG(Y ) , for all Y ∈ C.
Then η and ε form an adjunction

C(F (X), Y ) −→
←− D(X, G(Y )),
f 7−→ G(f ) ◦ ηX ,
εY ◦ F (g) ←−[ g.
Proof. For every f ∈ C(F (X), Y ) naturality of ε and (i) implies
εY ◦ F (G(f ) ◦ ηX ) = εY ◦ F G(f ) ◦ F (ηX ) = f ◦ εF (X) ◦ F (ηX ) = f.
Similarly for every g ∈ D(X, G(Y )) naturality of η and (ii) implies
G(εY ◦ F (g)) ◦ ηG(X) = G(εY ) ◦ GF (g) ◦ ηG(X) = G(εY ) ◦ ηY ◦ g = g.
For all f ∈ C(Y, Y 0 ) and g ∈ C(F (X), Y ), we have
G(f ◦ g) ◦ ηX = G(f ) ◦ G(g) ◦ ηX ,
which by (i) of the proof of Proposition 2.22 (i) is the same as saying that the map
G(-) ◦ ηX is natural. Similarly one checks that εY ◦ F (-) is natural. As the two maps are
inverse to each other, both are natural in X and Y .
2

27
Chapter 2. Simplicial sets

Definition 2.24
Given an adjunction C(F (X), Y ) = D(X, G(Y )).

(i) The natural transformation η is called the unit of the adjunction.

(ii) The natural transformation ε is called the counit of the adjunction.

(iii) If η and ε are both natural isomorphisms, then the adjunction is called an equiva-
lence (adjunction) between the categories C and D.

(iv) Two categories are called equivalent, if there is an equivalence adjunction between
them.

Example 2.25
Forgetful functors usually have a left adjoint:
L
(i) R-Mod( x∈X R · x, Y ) = Set(X, U (Y )).

(ii) T op(D(X), Y ) = Set(X, U (Y )),


where D(X) is the set X with the discrete topology (every subset is open).
In this setting U also has a right adjoint, i.e.

Set(U (X), Y ) = T op(X, I(Y )),

where I(X) is the set X with the indiscrete topology (∅ and X are the only open
subsets).

(iii) Similarly we have ajunctions (cf. Remark 1.2)

Simp(D(X), Y ) = Set(X, U (Y )), Set(U (X), Y ) = Simp(X, I(Y )).

(iv) Similarly we have an adjunction

Cat(D(X), Y ) = Set(X, U (Y )),

where U assigns to each small category its set of objects and D(X) is the discrete
category with objects Obj(D(X)) = X and only the identities as morphisms.

2.6 The singular nerve


|-|
It turns out that the geometric realization functor sSet −→ T op has a right adjoint,
which appears in every lecture on algebraic topology.

Definition 2.26
Let X be a topological space.

28
2.6. The singular nerve

The singular nerve of X is defined as the composite functor


U op I op |-| T op(-,X)
S(X) : ∆op −→ Setop −→ Simpop −→ T opop −→ Set.
By construction it is a simplicial set, whose n-simplices are the singular n-simplices
Sn (X) = T op(|I(n)|, X) = T op(|∆n |, X), n ≥ 0.

Remark 2.27
The singular nerve is used to define singular homology.
(i) Recall that the singular n-chains are defined as the abelian group
M
Cn (X, Z) := Z · σ.
σ∈Sn (X)

d
It becomes a chain complex by defining a differential Cn (X, Z) −→ Cn−1 (X, Z) via
X X X
d( aσ · σ) = (−1)i · aσ · (σ ◦ |I(di )|).
| {z }
σ∈Sn (X) 0≤i≤n σ
di (σ)=

(ii) The n-th singular homology group of X is defined as the n-th homology group
of the singular chain complex C∗ (X, A), i.e.
d
Hn (X, Z) = ker(Cn (X, Z) −→ Cn−1 (X, Z))/d(Cn+1 (X, Z)), n ≥ 0.

Proposition 2.28
S
The singular nerve induces a functor T op −→ sSet. Together with the geometric realiza-
tion an adjunction T op(|X|, Y ) = sSet(X, S(Y )).
Proof. Recall that the geometric realization was defined as
,
a
|X| := Xn × |I(n)| (X(f )(x), a) ∼ (x, |I(f )|(a)), x ∈ Xn , a ∈ |I(m)|, f ∈ ∆(m, n).
n≥0

For every g ∈ T op(|X|, Y ) and every n ≥ 0, we get a map


φX,Y (g)n : Xn −→ T op(|I(n)|, Y ), x 7−→ g(x, -).
By construction of the geometric realization this defines a homomorphism of simplicial
sets, because for every f ∈ ∆(m, n) and x ∈ Xn we have
φX,Y ◦ f ∗ (x) = g(f ∗ (x), -) = g(x, |I(f )|(-)) = g(x, -) ◦ |I(f )| = f ∗ ◦ φX,Y (x).
Vice versa every t ∈ sSet(X, S(Y )) glues to a map
ψX,Y (f ) : |X| −→ Y, Xn × |I(n)| 3 (x, a) 7−→ tn (x)(a),
because for every f ∈ ∆(m, n) we have
tn (f ∗ (x))(a) = (tn ◦ f ∗ )(x)(a) = (f ∗ ◦ tm )(x)(a) = (tm (x) ◦ |I(f )|)(a) = tm (x)(|I(f )|(a)).
By construction φX,Y and ψX,Y are inverse to each other and natural in X and Y .
2

29
Chapter 2. Simplicial sets

Remark 2.29
The definition of H∗ can be extended to arbitrary simplicial sets X ∈ sSet.

• The relation between simplicial sets and chain complexes will be studied in greater
detail later on.

• One might further study the map H∗ (X, Z) −→ H∗ (|X|, Z) induced by the adjunction
ηX
unit X −→ S(|X|). It can be proven, that it is a natural isomorphism. We will deduce
this from the Quillen equivalence between the model categories T op and sSet.

2.7 Isomorphisms, monomorphisms and


epimorphisms
Definition 2.30
Let C be a category and f ∈ C(B, C) a morphism.

(i) f is called a retraction, if there is a morphism g ∈ C(C, B) with f g = idC .

(ii) f is called a section, if there is a morphism g ∈ C(C, B) with gf = idB .

(iii) f is called an isomorphism, if there is a morphism g ∈ C(C, B) with gf = idB


and f g = idA .

(iv) f is called an epi(-morphism), if gf = hf implies g = h, for all g, h ∈ C(C, D).


f
Epimorphisms are denoted by arrows B − C.

(v) f is called a mono(-morphism), if f g = f h implies g = h, for all g, h ∈ C(A, B).


f
Monomorphisms are denoted by arrows B ,−→ C.

Example 2.31 (i) In the categories Set, T op, Grp, Simp the monomorphisms are pre-
cisely the injective homomorphisms.

(ii) In the category Set and Grp the epimorphisms are precisely the surjective homo-
morphisms.

(iii) Every surjective continuous map is a an epimorphism in the category of Hausdorff


spaces.
The converse is false, e.g. the inclusion of the subspace Q ,−→ R is a non-surjective
epimorphism.

(iv) Every surjective ring homomorphism is an epimorphism in the category Ring of


rings.
Again the converse is false, e.g. the homomorphism Z ,−→ Q is a non-surjective
epimorphism.

30
2.8. Simplicial standard simplices

Remark 2.32
Let C be a category.
(i) Every retraction is an epimorphism.

(ii) Every section is a monomorphism.

(iii) Every monomorphic retraction is an isomorphism.

(iv) Every epimorphic section is an isomorphism.

Remark 2.33
Let C be a category and f ∈ C(B, C) a morphism.
C(-,f )
(i) If C(-, B) −→ C(-, C) is a natural bijection, then f is a monomorphic retraction.
C(f,-)
(ii) If C(C, -) −→ C(B, -) is a natural bijection, then f is an epimorphic section.
In particular in both cases f is an isomorphism by Remark 2.32.

2.8 Simplicial standard simplices


Definition 2.34
For m ≥ 0 the simplicial standard m-simplex is defined as the simplicial set

∆m := ∆(-, m) ∈ sSet.

Proposition 2.35
There is a natural homeomorphism

em : |∆m | −→ |I(m)|, ∆m
n × |I(n)| 3 (f, a) 7−→ |I(f )|(a), m ∈ ∆.
Proof. Instead of working directly with the geometric realization we will prove this by
using category theory. Let Y ∈ T op and consider the compostion c
T op(em ,Y ) φ∆m ,Y d
T op(|I(m)|, Y ) −→ T op(|∆m |, Y ) −→ sSet(∆m , S(Y )) −→ Sm (Y ) = T op(|I(m)|, Y ),

where the map in the middle is the adjunction bijection of Proposition 2.28 and the map
on the right is Yoneda’s isomorphism of Lemma 2.11. For every g ∈ T op(|I(m)|, Y ) we
have by definition of the three maps

(d ◦ φ∆m ,Y ◦ T op(em , Y ))(g) = (d ◦ φ∆m ,Y )(g ◦ em ) = φ∆m ,Y (g ◦ em )m (idm )


= (g ◦ em )(idm , -) = g ◦ |I(idm )| = g.

This proves that c equals the identity. Since φ∆m ,Y and d are bijections, also T op(em , Y )
is a bijection. Remark 2.33 implies that em is an isomorphism in T op, which is a homeo-
morphism. By the usual arguments e is natural in m ∈ ∆.
2

31
Chapter 2. Simplicial sets

2.9 Limits and colimits


Definition 2.36
X
Let I −→ C be a functor.
i c
(i) A cone on X is a family of morphisms (C −→ X(i))i∈I , such that
X(f ) ◦ ci = cj , for all f ∈ I(i, j).
i π
(ii) A limit for X is a universal cone (lim X −→ X(i))i∈I . That is, for every other
ci c
cone (C −→ X(i))i∈I there is a unique morphism C −→ lim X, such that
π i ◦ c = ci , for all i ∈ I.
Another common notation for the limit object is lim X(i) := lim X(i) := lim X.
←− i∈I
i∈I

i c
(iii) Dually a cocone on X is a family of morphisms (X(i) −→ C)i∈I , such that
cj ◦ X(f ) = ci , for all f ∈ I(i, j).
i ι
(iv) A colimit for X is a universal cocone (X(i) −→ colim X)i∈I . That is, for every
ci c
other cocone (X(i) −→ C)i∈I there is a unique morphism colim X −→ C, such that
c ◦ ιi = ci , for all i ∈ I.
Another common notation for the colimit object is colim X(i) := colim X(i) :=
i∈I i∈I
colim X.

Remark 2.37
X
Let I −→ C be a functor.
(i) In general X may not have a limit or colimit.
(ii) If a limit/colimit of X exists, it is unique up to unique isomorphism.
X op
(iii) Considering X as a functor I op −→ C op we see that every limit/cone for X op can
be considered as a colimt/cocone for X.

There is a canonical construction for limits and colimits of Set-valued functors.

Proposition 2.38
For X ∈ CAT(I, Set) with small I ∈ Cat, the following holds.
(i) A limit of X is given by
( )
Y
lim X := x∈ X(i); X(f )(xi ) = xj , ∀f ∈ I(i, j) .
i∈I


The map lim X −→ X(i) is the projection onto the factor X(i), for each i ∈ I.

32
2.9. Limits and colimits

(ii) A colimit of X is given by


,
a
colim X := X(i) x ∼ X(f )(x), x ∈ X(i), f ∈ I(i, j).
i∈I

i ι
The map X(i) −→ colim X is the map induced by the inclusion of the disjoint
summand X(i), for each i ∈ I.
Proof.
i π
(i) By construction (lim X −→ X(i))i∈I is a limit cone for X. Given another cone
ci
(C −→ X(i))i∈I we can define

c : C −→ lim X, x 7−→ (ci (x))i∈I .

Then πi ◦ c = ci , for all i ∈ I, which also implies the uniqueness of c.


i ι
(ii) Again by construction (X(i) −→ colim X)i∈I is a colimit cone. Given another cocone
ci
(X(i) −→ C)i∈I the map

c : colim X −→ C, X(i) 3 x 7−→ ci (x),

is well-defined. Again uniqueness of c follows from the condition ci = c ◦ ιi , for all


i ∈ I.

Corollary 2.39
For X ∈ CAT(I, T op) with small I ∈ Cat, the following holds.
(i) A limit lim X is given as the Set-limit together with the initial topology, which in
this case is the subspace topology of the product topology.

(ii) A colimit colim X is given as the Set-limit together with the final topology with
respect to the inclusions.

Definition 2.40
Let C ∈ CAT and S ∈ Set. Like in Example 2.25 a family of objects (Xs )s∈S ∈ C S
corresponds to a functor

X ∈ CAT(D(S), C) = Set(S, Obj(C)) = C S .


Q
(i) s∈S Xs := lim X is called a product over the set S.
`
(ii) s∈S Xs := colim X is called a coproduct over the set S.
X
Consider the empty functor D(∅) −→ C.
(i) ∗ := lim X is called a terminal or final object in C.

33
Chapter 2. Simplicial sets

(ii) ∅ := colim X is called an initial or cofinal object in C.

Remark 2.41
From the explicit construction of limits and colimits in Set, we see that the notion of
product, coproduct, ∗ and ∅ has the usual meaning.

Remark 2.42
Let I ∈ Cat and t ∈ CAT(I, C)(X, Y ).
(i) Given limits lim X and lim Y , the universal property of lim Y applied to the cone
πi ti
(lim X −→ X(i) −→ Y (i))i∈I yields a unique lim t ∈ C(lim X, lim Y ) with πi ◦lim t =
ti ◦ πi , for all i ∈ I.
Note that the uniqueness implies:
a) lim idX = idlim X , for all X ∈ CAT(I, X).
t s
b) lim(s ◦ t) = lim s ◦ lim t, for natural transformations X −→ Y −→ Z.
X
(ii) Suppose every functor I −→ C has a constructible limit lim X. That is a canonical
construction of the limit for each functor X (like e.g. we have for the category Set)3 .
Then the universal property for limits can be expressed as an adjunction

CAT(I, C)(const X, Y ) = C(X, lim Y ),

where
• const is the functor, sending an object X to the constant functor

const X : I −→ C, i 7−→ X, I(i, j) 3 f 7−→ idX .

• lim is the functor (by (i)), sending a functor Y ∈ CAT(I, C) to its (canonical)
limit.

The subsequent Proposition shows that limits and colimits of functors targetting at
functor categories can be constructed objectwise. Together with Proposition 2.38 this
provides a general construction for limits and colimits of sSet-valued functors.

Proposition 2.43
X
Let I, J ∈ Cat be small categories and C be an arbitrary category. Let I −→ CAT(J, C)
and consider the functors

Xj : I −→ C, i 7−→ X(i)(j), j ∈ J.

Then the following holds.


3
Often it is assumed that every limit/colimit is constructible, which is implied by the axiom of choice
for classes. If you feel uncomfortable with that, note that in most cases the existence of limits/colimits
in a given category is verified by giving an explicit construction. So existence of limits is obtained by
proving the existence of constructible limits

34
2.10. Preservation of (co-)limits

(i) If Xj has a limit lim Xj , for each j ∈ J, then X has a limit in CAT(J, C), given by

L : J −→ C, j 7−→ lim Xj , J(j, j 0 ) 3 f 7−→ lim f.

(ii) If Xj has a colimit colim Xj , for each j ∈ J, then X has a colimit in CAT(J, C),
given by

L : J −→ C, j 7−→ colim Xj , J(j, j 0 ) 3 f 7−→ colim f.

Proof.
i π
(i) By Remark 2.42 L is a functor and (L −→ X(i))i∈I is a cone of functors J −→
ci
C. Given another cone of functors (C −→ X(i))i∈I , we get unique morphisms
dj i π
(lim Xj −→ C(j))j∈J being compatible with the projections (L −→ X(i))i∈I . Unique-
ness implies that (dj )j∈J defines a natural transformation d ∈ CAT(J, C)(L, C) and
that it is unique with the property πi ◦ d = ci , for all i ∈ I. We conclude that
πi
(L −→ X(i))i∈I is a limit cone for X.

(ii) This is dual to (i), i.e. by (i) we can construct a limit for

X op : I op −→ CAT(J, C)op = CAT(J op , C op ),

which then will be a colimit for X.

2.10 Preservation of (co-)limits


Definition 2.44
Let F ∈ CAT(C, D) be a functor.
X
(i) We say F preserves limits, if for each functor I −→ C having a limit lim X, the
F (πi )
cone (F (lim X) −→ F (X(i)))i∈I is a limit cone for F ◦ X.
X
(ii) We say F preserves colimits, if for each functor I −→ C having a colimit
F (ιi )
colim X, the cocone (F (X(i)) −→ F (colim X))i∈I is a colimit cocone for F ◦ X.

Remark 2.45
X F
Given two functors I −→ C −→ D.

(i) If there are limits for X and F ◦ X, then the universal property for lim(F ◦ X)
F (πi )
applied to the cone (F (lim X) −→ F (X(i)))i∈I induces a unique map

F (lim X) −→ lim(F ◦ X).

It is an isomorphism, if and only if F preserves limits.

35
Chapter 2. Simplicial sets

(ii) If there are colimits for X and F ◦ X, then the universal property for colim (F ◦ X)
F (ιi )
applied to the cocone (F (X(i)) −→ F (colim X))i∈I induces a unique map

colim (F ◦ X) −→ F (colim X).

It is an isomorphism, if and only if F preserves colimits.

Proposition 2.46
For a functor X ∈ CAT(I, C) the following holds.
pi
An arbitrary cone (L −→ X(i))i∈I for X is a limit cone for X, if and only if
(i) 
C(C,pi )
C(C, L) −→ C(C, X(i)) is a limit cone, for all C ∈ C.
i∈I

ji
(ii) An cocone (X(i) −→ L)i∈I for X is a colimit cocone for X, if and only if
 arbitrary C(j
i ,C)
C(X(i), C) −→ C(L, C) is a colimit cocone, for all C ∈ C.
i∈I

In particular hom-functors preserve (co-)limits.


Proof.
(i) Consider the canonical map
( )
Y
φC : C(C, L) −→ lim C(C, X) = c∈ C(C, X(i)); X(f ) ◦ ci = cj , ∀f ∈ I(i, j) ,
i∈I
c 7−→ (c ◦ pi )i∈I ,

where the equality on the right follows from the construction of the Set-limit in
Proposition 2.38. By definition of a cone the elements of the right object are cones
with source C. Hence by definition of a limit cone the following are equivalent:
pi
• (L −→ X(i))i∈I is a limit cone.
• The map φC is bijective, for all C ∈ C.
 C(C,pi )

• C(C, L) −→ C(C, X(i)) is a limit cone, for all C ∈ C.
i∈I

X op
(ii) This follows from (i) applied to I op −→ C op using Remark 2.37.

Corollary 2.47
For an adjunction C(F (X), Y ) = D(X, G(Y )) the following holds.
(i) G preserves limits.4
4
Preservation of limits is a strong criterion for a functor being a right adjoint. Freyd’s adjoint functor
Theorem states that almost every functor preserving limits has a left adjoint. As we will not need it
here, for a precise formulation the reader again is advised to books about category theory.

36
2.10. Preservation of (co-)limits

(ii) F preserves colimits.


Proof.
i π
(i) Given a limit cone (lim Y −→ Y (i))i∈I , Proposition 2.46 implies that
 C(F (X),πi )

C(F (X), lim Y ) −→ C(F (X), Y (i))
i∈I

is a limit cone, for all X ∈ D. Using the adjunction bijection, it is isomorphic to the
cone  
D(X,G(πi ))
D(X, G(lim Y )) −→ D(X, G(Y (i))) ,
i∈I
G(πi )
which is therefore also a limit cone, for all X ∈ D. Equivalently (G(lim Y ) −→
G(Y (i)))i∈I is a limit cone by Proposition 2.46 again.

(ii) This is dual to (i).

2
X
Given a functor J −→ CAT(I, C), where I, J ∈ Cat and C has constructible limits. Then
lim
by Remark 2.42 the limit induces a right adjoint functor CAT(I, C) −→I C and using that
limits in functor categories are computed degreewise by Proposition 2.43 we get a natural
isomorphism

lim lim X(j)(i) = lim(lim X(j)) −→ lim(lim X)(j) = lim lim X(j)(i).
i∈I j∈J I j∈J j∈J I j∈J i∈I

Dually also colimits preserve colimits. One might wonder, if also the natural transforma-
tion

colim lim X(j)(i) = colim (lim X)(j) −→ lim(colim X(j)) = lim colim X(j)(i)
j∈J i∈I j∈J I I j∈J i∈I j∈J

is an isomorphism. In general this is false for the most categories. However under certain
conditions on I and J this is true in the category Set.

Proposition 2.48
Given two small categories I, J ∈ Cat, such that

(i) I is finite, i.e. has finitely many objects and morphisms.

(ii) J is filtered, i.e.


a) For all i, j ∈ J there are morphisms i −→ k ←− j, for some object k ∈ J.
b) For all morphisms f, g ∈ J(i, j) there is a morphism h ∈ J(j, k) with hf = hg.
X
Then for every functor J −→ CAT(I, Set) the natural map below is bijective

colim lim X(j)(i) −→ lim colim X(j)(i).
j∈J i∈I i∈I j∈J

37
Chapter 2. Simplicial sets

S
Proof. First we claim that for every functor J −→ Set we have
f f0
ιj (s) = ιj 0 (s0 ) ⇐⇒ ∃j −→ k ←− j 0 : S(f )(s) = S(f 0 )(s0 ), ∀ s ∈ S(j), s0 ∈ S(j 0 ).

Using the explicit construction for the colimit given in Proposition 2.38
,
a
colim X := S(j) s ∼ S(f )(s), s ∈ S(i), f ∈ J(i, j),
i∈J

it follows that the given condition on s, s0 implies that s ∼ s0 and hence ιj (s) = ιj 0 (s0 ).
f
Moreover the condition holds for s0 = S(f )(s) with j −→ j 0 by setting f 0 = idj 0 . So
it remains to prove that the condition defines an equivalence relation. By definition the
defined relation is reflexive and symmetric. For transitivity, suppose we are given maps
f f0 g0 g 00
j −→ k ←− j 0 −→ ` ←− j 00

and s ∈ S(j), s0 ∈ S(j 0 ), s00 ∈ S(j 00 ) with S(f )(s) = S(f 0 )(s0 ) and S(g 0 )(s0 ) = S(g 00 )(s00 ).
Using that J is filtered there are morphisms

j
f
f0 
j0 /k

g0 ∃a
g 00  
j 00 / ` ∃b /m
∃c


n,

such that c(af 0 ) = c(bg 0 ) and hence

S(caf )(s) = S(ca)S(f )(s) = S(ca)S(f 0 )(s0 ) = S(caf )(s0 )


= S(cbg 0 )(s0 ) = S(cb)S(g 0 )(s0 ) = S(cb)S(g 00 )(s00 ) = S(cbg 00 )(s00 ),

which proves transivity.


Now we return to our problem of question.

• For surjectivity of the map let x ∈ limi∈I colim j∈J X(j)(i). Then using the explicit
description of limits in colimits in Set, we can consider x as a tuple
Y
x = (xi )i∈I ∈ colim X(j)(i),
j∈J
i∈I

with xi = ιji (yi ), for some yi ∈ X(ji )(i), for each i ∈ I. Using that J is filtered,
fi
by induction on ]I < ∞ we find an object j ∈ J and morphisms ji −→ j, for each
i ∈ I. So replacing each yi by its image (fi )∗ (yi ) we may assume that yi ∈ X(j)(i),

38
2.11. Comma categories

for each i ∈ I. Now for every g ∈ I(i, i0 ) we have g∗ (xi ) = xi0 and by our explicit
description of the colimits over J, there are J-morphisms f, f 0 ∈ J(j, k), such that

f∗ g∗ (yi ) = f∗0 (yi0 ).

Using that J is filtered, there is a J-morphism h ∈ J(k, `), such that hf = hf 0 and
hence
g∗ (hf )∗ (yi ) = (hf )∗ g∗ (yi ) = (hf 0 )∗ (yi0 ).
So by replacing every yi by its image (hf )∗ (yi ), we may assume that g∗ (yi ) = yi0 .
Doing this for all g ∈ Mor(i, i0 ) one after another, we may assume that (yi ) ∈
limi∈I X(j)(i) and since
x = (xi )i = (ιj yi )i
the tuple ι( yi )i ∈ colim j∈J limi∈I X(j)(i) defines a preimage for x.

• For injectivity let x, x0 ∈ colim j∈J limi∈I X(j)(i) being mapped to the same element.
Then x and x0 are represented by elements y = (yi )i ∈ limi∈I X(j)(i) and y 0 =
(yi0 )i ∈ limi∈I X(j 0 )(i) respectively. As before we may assume that j = j 0 . As x
and x0 are mapped to the same element ιj (yi ) = ιj 0 (yi0 ), for all i ∈ I. So by our
explicit description of J-colimits, there are J-morphisms fi , fi0 ∈ J(j, ki ), such that
(fi )∗ (yi ) = (fi0 )∗ (yi0 ), for all i ∈ I. Using that I is finite and J is filtered, there
is an object ` ∈ J and morphisms gi ∈ J(ki , `). Moreover we may assume that
all the maps gi fi and gi fi0 coincide. In other words there are maps h, h0 ∈ J(j, `),
such that h∗ (yi ) = (h0 )∗ (yi ), for all i ∈ I. Equivalently h∗ (y) = (h0 )∗ (y 0 ) and hence
x = ιj (y) = ιj (y 0 ) = x0 by our description of J-colimits.

2.11 Comma categories


Definition 2.49
F G
Given a two functors C −→ E ←− D, the comma category F ↓ G is the following
category.

• Its objects are tuples (C, D, α), where C ∈ C, D ∈ D and α ∈ E(F (C), G(D)).

• Its morphisms are defined as

F ↓ G((C, D, α), (C 0 , D0 , α0 )) = {(f, g) ∈ C(C, C 0 ) ×D(D, D0 ); G(g) ◦ α = α0 ◦ F (f )}

and composition is induced by that of C and D.


F
(i) If ∗ −→ E is the functor sending the only object ∗ to E, we write E/G := F ↓ G
ηX
and X for the object (∗, X, E −→ G(X)).
If moreover G is the identity functor on E = C, we write E/E := F/E and call it
the slice or under-category of E-objects under E.

39
Chapter 2. Simplicial sets

G
(ii) Dually if ∗ −→ E is the functor sending ∗ to E, we write F/E := F ↓ G and X for
εX
the object (X, ∗, F (X) −→ E).
If moreover F is the identity functor on E = C, we write E/E := F/E and call it
the coslice or over-category of E-objects over E.

Remark 2.50
Given F ∈ CAT(C, E), every f ∈ E(E, E 0 ) induces a functor
ε ε f
f∗ : F/E −→ F/E 0 , X
X = (F (X) −→ X
E) 7−→ f∗ X = (F (X) −→ E −→ E 0 ).
F/-
In particular for small C we get a functor E −→ Cat.
Moreover there is a functor
X ε
F/E −→ C, (F (X) −→ E) 7−→ X.

Lemma 2.51 (co-Yoneda Lemma)


For I ∈ Cat, we will denote the Yoneda functor of Proposition 2.10 by the same letter

I : I −→ CAT(I op , Set), i 7−→ I(-, i).

Then for X ∈ CAT(I op , Set) the the following holds.


(i) The category I/X can be described as follows:
a
Obj(I/X) = X(i), I/X(x, y) = {f ∈ I(i, j); f ∗ (y) = x}, x ∈ X(i), y ∈ X(j).
i∈I

i ε ∼
(ii) The maps I(-, i) −→ X induce a natural isomorphism colim εi : colim I(-, i) −→ X.
i∈I/X i∈I/X
op
In this case ‘natural’ means every f ∈ CAT(I , Set)(X, Y ) induces a commutative
diagram
colim i εi
colim I(-, i) / X
i∈I/X
f
colim i ιf∗ (i)
 colim i εi 
colim I(-, i) / Y.
i∈I/Y

Proof.
(i) Using Yoneda’s isomorphism

CAT(I op , Set)(I(-, i), X) = X(i), i ∈ I,


i ε
an object i = (I(-, i) −→ X) ∈ I/C corresponds to an element xi ∈ X(i), for some
i ∈ I. Moreover for f ∈ I(i, j) we have

f ∈ I/X(i, j) ⇐⇒ ε j ◦ f = εi ⇐⇒ f ∗ (xj ) = xi .

40
2.12. Internal homomorphisms

(ii) Using the description of I/X from (i), for every Y ∈ CAT(I op , Set) we get
∼ ∼
CAT(I op , Set)(colim I(-, i), Y ) −→ lim CAT(I op , Set)(I(-, i), Y ) −→ lim Y (i)
i∈I/X i∈I/X i∈I/X
Y
={y ∈ Y (i); f ∗ (yx0 ,i0 ) = yx,i = yf ∗ (x0 ),i ∀f ∈ I/X((x, i), (x0 , i0 ))}
x∈X(i),
i∈I
Y
={t ∈ Set(X(i), Y (i)); f ∗ ◦ tj = ti ◦ f ∗ , ∀f ∈ I(i, j)}
i∈I
op
=CAT(I , Set)(X, Y ).

The first isomorphism is the canonical one induced by composition with the inclusion
maps (cf. Remark 2.45), using that hom-functors preserve limits by Proposition
2.46. Next we have Yoneda’s isomorphism of Lemma 2.11. The explicit description of
limits in Set given in Proposition 2.38 gives the next equality. Rewriting the product
gives the next description, which by definition is the set of natural transformations.
We claim that composition of the upper isomorphism with the map given by compo-
sition with colim i εi : colim i∈I/X I(-, i) −→ X is the identity. Given f ∈ CAT(I op , Set)(X, Y )
the map f ◦ εi corresponding to x ∈ X(i) sends g ∈ I(j, i) to fj ◦ g ∗ (x). As Yoneda’s
isomorphism is given by evaluation at g = idi , we get fi ◦ id∗ (x) = fi (x). This
proves the claim, which implies that also composition by colim i εi is bijective, for
all Y ∈ CAT(I op , Set). So colim i εi is an isomorphism by Remark 2.33.
2

2.12 Internal homomorphisms


Remark 2.52
There is a bijection

Set(X × Y, Z) = Set(X, Set(Y, Z)), f 7−→ [x 7−→ f (x, -)],

which is natural in X, Y, Z ∈ Set.


In particular the functor (- × Y ) is left adjoint to the functor Set(Y, -).

We want to construct a similar natural bijection for the category of simplicial sets.

Definition 2.53
The simplicial set of internal homomorphisms between X, Y ∈ sSet is defined as

sSet(X, Y ) := sSet(∆• × X, Y ) : ∆op −→ Set, n 7−→ sSet(∆n × X, Y ).

Proposition 2.54
There is a bijection
sSet(X × Y, Z) = sSet(X, sSet(Y, Z)),
which is natural in X, Y, Z ∈ sSet.

41
Chapter 2. Simplicial sets


Proof. By Lemma 2.51 there is a natural isomorphism ε : colim m∈∆/X ∆m −→ X. The
natural map

δ : colim (∆m × Y ) −→ ( colim ∆m ) × Y, Y ∈ sSet
m∈∆/X m∈∆/X

is an isomorphism. Indeed by Proposition 2.43 colimits in functor categories are computed


dimensionwise and evaluated at n ∈ ∆ the corresponding map δn is a bijection. Indeed
by Remark 2.52 (- × Yn ) is a left adjoint and thus preserves colimits by Corollary 2.47.
Now the desired isomorphism is given as the composition of isomorphisms

sSet(X × Y, Z) sSet(X, sSet(Y, Z))


ε∗ o

sSet(( colim ∆m ) × Y, Z) sSet(X, sSet(∆• × Y, Z))
m∈∆/X

δ∗ o ε∗ o
 
sSet( colim (∆m × Y ), Z) sSet( colim ∆m , sSet(∆• × Y, Z))
m∈∆/X m∈∆/X

o o
 
lim sSet(∆m × Y, Z) o ∼
lim sSet(∆m , sSet(∆• × Y, Z)).
m∈∆/X m∈∆/X

Here the lower vertical maps are the natural maps, which are isomorphisms as hom-
functors preserve limits. The lower horizontal isomorphism is the Yoneda isomorphism of
Lemma 2.11. By construction all isomorphisms are natural in X, Y and Z.
2

2.13 Ordered simplicial complexes as simplicial sets


Definition 2.55
Let X ∈ Simpo be an ordered simplicial complex.
The nerve of X is defined as the composite functor
I Simpo (-,X)
BX : ∆op −→ Simpo −→ Set, n 7−→ Bn X.

B
Remark 2.56 (i) The nerve induces a functor Simpo −→ sSet.
As hom-functors preserve limits by Proposition 2.46, so does B by Proposition 2.43.

(ii) The Yoneda embedding ∆ −→ sSet coincides with the composition


I B
∆ −→ Simpo −→ sSet.

Definition 2.57
A simplicial set X ∈ sSet is called finite, if the set of nondgenerate simplices is finite.

42
2.13. Ordered simplicial complexes as simplicial sets

Remark 2.58 (i) Every simplicial set is the union of its finite simplicial subsets.
(ii) The geometric realization |X| of a finite simplicial set X is compact, since Proposi-
tion 2.20 yields a surjection
a
X̃n × |I(n)| − |X|,
n≥0

and the left space is compact being a finite disjoint union of compact spaces.

Proposition 2.59
For every ordered simplicial complex X ∈ Simpo there is a natural homeomorphism

eX : |BX| = |Simpo (I(-), X)| −→ |X|, Simpo (I(n), X) × |I(n)| 3 (f, a) 7−→ |f |(a).
Proof. Unfortunately this is not as easy as the proof for the simplicial standard simplex
given in Proposition 2.35. There are bijections

imX : Bn X = {f ∈ Simpo (I(n), X); f injective} −→ {s ∈ S(X); ]s = n + 1}, n ≥ 0,
]
f 7−→ f (n).
inducing the upper bijection in the commutative diagram of sets below5 .
`
n imX ×id
/
a a
Bn X × (|I(n)|\|∂I(n)|)
] ∼ {s ∈ S(X); ]s = n + 1} × (|I(n)|\|∂I(n)|)
n≥0 n≥0

βBX o
 eX
|BX| / |X|.
Moreover the left vertical map is the continuous bijection of Proposition 2.20 and it follows
that the lower horizontal map must be a bijection, too.
`
Now for every finite F ∈ Simpo , the set n≥0 B ] ∼
n F = S(F ) is finite and hence BF is a
finite simplicial set, so |BF | is compact and eF is a homeomorphism by Lemma 1.16. For
general X ∈ Simpo we consider the usual commutative square
colim F eF
colim |BF | / colim |F |

F ⊂X, F ⊂X,
F finite F finite
o o
 eX

|BX| / |X|.
B
As the colimit here is a union of sets containing each other6 , it is preserved by Simpo −→
|-|
sSet (this is false for arbitrary colimits!). Moreover sSet −→ T op preserves arbitrary col-
imits, as it is a left adjoint. This proves that the left vertical map is a homeomorphism. By
definition of the final topology on |X| and the description of colimits in T op of Corollary
2.39, also the right vertical map is a homeomorphism. The colimit of homeomorphism is
a homeomorphism (its inverse is the colimit of the inverse maps), so the upper horizontal
map is a homeomorphism. Finally it follows that also the lower horizontal map is one.
2
5 U
To simplify the notation we will skip the forgetful functor T op −→ Set.
6
In particular it is a filtered colimit.

43
Chapter 2. Simplicial sets

Proposition 2.60
There is a natural continuous bijection

(|πX |, |πY |) : |X × Y | −→ |X| × |Y |, X, Y ∈ sSet,

which is a homeomorphism, if X and Y are finite.


Proof. For every m, n ≥ 0 we have a commutative diagram

|(Bπ1 ,Bπ2 )| eI(m)×I(n)


|∆m × ∆n | |BI(m) × BI(n)| o ∼ |B(I(m) × I(n))| ∼
/ |I(m) × I(n)|
(|π1 |,|π2 |) (|Bπ1 |,|Bπ2 |) (|π1 |,|π2 |) o
  eI(m) ×eI(n) 
|∆m | × |∆n | |BI(m)| × |BI(n)| / |I(m)| × |I(n)|,

where the upper horizontal map in the middle is a homeomorphism, because B preserves
products by Remark 2.84 (i). The right two horizontal maps are homeomorphisms by
Proposition 2.59 and the right vertical map is a homeomorphism by Proposition 1.25. As
the diagram commutes also the vertical map in the middle and the left vertical map are
homeomorphisms.
The general statement follows from the commutative diagram of sets (not spaces!)

colim colim |∆m × ∆n | ∼ / |( colim ∆m ) × (colim ∆n )| ∼ / |X × Y |


m∈∆/X n∈∆/Y m∈∆/X n∈∆/Y

o (|πX |,|πY |)
  
colim colim |∆m | × |∆n |
∼ / |( colim ∆m )| × |(colim ∆n )|
∼ / |X| × |Y |,
m∈∆/X n∈∆/Y m∈∆/X n∈∆/Y

where the right two isomorphisms are induced by the co-Yoneda isomorphism of Lemma
2.51 and the left two horizontal maps are isomorphisms, because all appearing functors
are left adjoints (targetting to Set) and thus commute with colimits.
As in the case of simplicial complexes treated in Proposition 1.25 the lower left hori-
zontal map does not need to be a homeomorphism, for infinite X and Y .
2

Remark 2.61
For a (locally) compact space Y ∈ T op, there is an adjunction

T op(X × Y, Z) = T op(X, T op(Y, Z)),

where T op(Y, Z) is the set T op(Y, Z) together with the compact open topology. A
subset U ⊂ T op(Y, Z) is open, if it is the union of finite intersections of sets

N (C, U ) := {f ∈ T op(Y, Z); f (C) ⊂ U }, C ⊂ Y compact, U ⊂ Z open.

In particular (- × Y ) commutes with arbitrary colimits.

44
2.14. Homotopies

Corollary 2.62
We can use this to extend the proof of Proposition 2.60 to get a homeomorphism

(|πX |, |πY |) : |X × Y | −→ |X| × |Y |, X, Y ∈ sSet,

if at least one of the simplicial sets X and Y is finite.

Remark 2.63 (i) There is no space of internal homomorphisms in the category T op of


topological spaces generalizing the construction of Remark 2.61.

(ii) However there is a subcategory CGHaus ≤ T op, which has this property. It can be
shown that the geometric realization of every simplicial set is infact an object in
CGHaus.
|- |
(iii) One can show that sSet −→ CGHaus preserves finite products. Infact it even pre-
serves finite limits (i.e. limits over finite categories), which can be proven in the
same way as Propositions 1.25 and Proposition 2.60.

2.14 Homotopies
Definition 2.64 (i) A simplicial homotopy from f ∈ sSet(X, Y ) to g ∈ sSet(X, Y ),
written f ' g, is a morphism h ∈ sSet(∆1 × X, Y ) such that
h

∆0 × X = X f

d1 ×id
 
∆1 × h /
O X Y.
D
d0 ×id

∆0 × X = X g

Note that di corresponds to the inclusion of the subset {1 − i} ⊂ 1. This is why d1


corresponds to f and d0 to g.

(ii) A simplicial homotopy equivalence is a map f ∈ sSet(X, Y ), for which we


have a g ∈ sSet(Y, X) with idX ' gf and idY ' f g.

(iii) A simplicial deformation section is a section s ∈ sSet(X, Y ) having a retrac-


tion r ∈ sSet(Y, X) and a simplicial homotopy idX ' sr. It is called a strong
h
simplicial deformation section, if h fits in a commutative diagram
πY
∆1 × Y / Y
id×s s
 
∆1 × X
h / X.

We also call X a (strong) simplicial deformation retract of Y .

45
Chapter 2. Simplicial sets

(iv) A simplicial deformation retraction is a retraction r ∈ sSet(X, Y ) having a


section s ∈ sSet(Y, X) and a simplicial homotopy idY ' sr. It is called a strong
h
simplicial deformation retraction, if h fits in a commutative diagram

∆1 × X
h /X

id×r r
 
πX
∆1 × Y / Y.

Remark 2.65
As ∆1 is a finite simplicial set, by Corollary 2.62 every simplicial homotopy f ' g induces
h
a homotopy |f | ' |g| in the topological sense.
|h|

(i) In particular every simplicial homotopy equivalence resp. (strong) simplicial defor-
mation section/retraction induces a homotopy equivalence resp. (strong) deforma-
tion section/retraction after realization.
(ii) In contrast to homotopies between continuous maps, ‘'’ is not an equivalence rela-
tion on sSet(X, Y ), for general X, Y . It is reflexive, but need neither be transitive
nor symmetric. We will later introduce fibrant simplicial sets Y , for which ‘'’ de-
fines an equivalence relation.

Definition 2.66
Let f ∈ sSet(X, Y ).
(i) The mapping cylinder of f is defined as
d0 ×id f
M (f ) := (∆1 × X) +X Y = colim (∆1 × X ←−X ∆0 × X = X −→ Y ).

(ii) The mapping cone of f is defined as


d1 ×id ι∆1 ×X
C(f ) := ∗ +X M (f ) = colim (∗ ←− X = ∆0 × X −→X ∆1 × X −→ M (f )).

Remark 2.67
|- |
We have seen that sSet −→ T op by Corollary 2.62 behaves well with the appearing prod-
ucts. Moreover by Corollary 2.47 it also preserves arbitrary colimits, as it is a left adjoint
by Proposition 2.28. So we have natural homeomorphisms
∼ ∼
(i) |M (f )| = |(∆1 × X) +X Y | ←− |∆1 × X| +|X| |Y | −→ (|∆1 | × |X|) +|X| |Y | = M (|f |).
(ii) |C(f )| = | ∗ +X M (f )| = | ∗ | +|X| |M (f )| ∼
= ∗ +|X| M (|f |) = C(|f |).

Proposition 2.68
Every f ∈ sSet(X, Y ) has a natural factorization
d1 ι∆1 ×X (f πX )∪id
f : X −→ ∆1 × X −→ M (f ) −→ Y.
Denoting f 0 := ι∆1 ×X d1 and f 0 := (f πX ) ∪ id, the following holds.

46
2.15. Connected components

(i) f 0 is a monomorphism,
(ii) f 00 has a section ιY and there is a homotopy h ∈ sSet(∆1 × M (f ), M (f )) with
idM (f ) ' ιY f 00 inducing a commutative diagram
h

id×ιY id×f 00
∆1 × Y / ∆1 × M (f ) / ∆1 × Y
πY h πY
 ιY
 f 00 
Y / M (f ) / Y.

In particular f 00 is a strong deformation retraction and ιY is a strong deformation


section.
Proof.
(i) Monomorphisms in sSet are simply dimensionwise injections.
(ii) There is a homomorphism of simplicial sets

V : ∆1 × ∆1 −→ ∆1 , ∆1n × ∆1n 3 (s, t) 7−→ V (s, t) : [k 7−→ max{s(k), t(k)}].

inducing a commutative diagram


d0 id×f
∆1 × (∆1 × X) o ∆1 × X / ∆1 × Y
V ×id πX πY
  f 
d0
∆1 × X o X / Y,

because

(V × id)d0 (f, x) = (V × id)(1, f, x) = (1, x) = d0 πX (f, x), (f, x) ∈ ∆1n × Xn .

Taking the colimit induces a homomorphism



h : ∆1 × M (f ) ←− M (∆1 × f ) = (∆1 × ∆1 × X) +∆1 ×X (∆1 × Y ) −→ M (f ).

which by construction has all the desired properties.


2

2.15 Connected components


Definition 2.69
The set of connected components of an X ∈ sSet is defined as
 
d0
/
π0 X := colim X = colim X1 / X0 .
d1

We call X connected, if π0 X = ∗.

47
Chapter 2. Simplicial sets

Proposition 2.70 (i) A simplicial set X is connected, if and only if any two 0-simplices
x, y ∈ X0 can be connected by a chain of 1-simplices c1 , ..., cn ∈ X1 , i.e.
0d 1d 0 d 0 d 1 d
x ←−[ c1 7−→ d1 (c1 ) = d0 (c2 ) ←−[ ... ←−[ cn 7−→ d1 (cn ) = y.

In this case |X| is path-connected.



(ii) For general X ∈ sSet we get a natural bijection π0 |X| −→ π0 X.
Proof.

(i) By definition X is connected, if and only if [x] = [y] in π0 X, for all x, y ∈ X0 . By


definition of π0 X, this is the case precisely if there is a chain as described.
`
Now given (x, a), (y, b) ∈ n≥0 Xn × |I(n)| corresponding to two elements in |X|.
Supposing (x, a) ∈ Xn × |I(n)| we define

p : [0, 1] −→ |I(n)|, t 7−→ t · [0] + (1 − t) · a,

where [0] ∈ |I(n)| is the element with [0]0 = 1. This defines a path from (x, a) to
ι
(x, [0]). In the same way we construct a path q from (y, [0]) to (y, b). Let 0 −→ n
denote the canonical inclusion. Then in |X| we have

(x, [0]) = (x, ι∗ [0]) ∼ (ι∗ x, [0]).

By (i) there is a chain c1 , ..., cn ∈ X1 linking ι∗ x and ι∗ y. It corresponds to a chain



of maps c1 , ..., cn ∈ sSet(∆1 , X) −→ X1 by using Yoneda’s isomorphism. Naturality
of Yoneda’s isomorphism implies that the realization |ci | is a path linking the two
elements (d0 (ci ), [0]), (d1 (ci ), [0]) ∈ X0 × |I(0)|. So glueing |ci | ∈ T op(|∆1 |, |X|) with
p and q gives rise to a path from (x, a) to (y, b):


 p(t), t ∈ [0, 1],
|c2i+1 |(2i + 2 − t), t ∈ [2i + 1, 2i + 2],

[0, n + 2] −→ |X|, t 7−→

 |c2i |(t − 2i), t ∈ [2i, 2i + 1],
q(t − n − 1), t ∈ [n + 1, n + 2].

This proves that |X| is path-connected, i.e. π0 |X| = ∗.


X η
(ii) Realization of the unit X −→ const π0 X of the colimit adjunction of Remark 2.42
induces the continuous map

|X| −→ | const π0 X| = π0 X, Xn × |I(n)| 3 (x, a) 7−→ [ι∗ x].

The right object is the discrete space π0 X, proving that the map is surjective. Hence
applying π0 induces a natural surjection π0 |X| −→ π0 X and the argument of (i)
shows that it is also injective.

48
2.16. Skeleton and coskeleton

Remark 2.71
For X, Y ∈ sSet, we have by definition

sSet(X, Y )0 = sSet(∆1 × X, Y ).

Hence π0 sSet(X, Y ) is the set of homomorphisms sSet(X, Y ) modulo the equivalence


relation spanned by ‘'’.
This defines a category with same objects as sSet, that we will simply denote by π0 sSet.

2.16 Skeleton and coskeleton


Definition 2.72
Let X ∈ Simp and n ≥ 0.
The n-skeleton of X is the simplicial complex skn X with the same vertices, but

S(skn X) := {s ∈ S(X); ]s ≤ n + 1}.

Remark 2.73 (i) skn X ⊂ X is the subcomplex of all simplices of dimension ≤ n.


n sk
(ii) Simp −→ Simp is a functor.

(iii) ∂I(n) = skn−1 I(n).

We will later need an extension of this construction to simplicial sets.

Definition 2.74
Let n ≥ 0.
in
(i) ∆≤n ,−→ ∆ is the inclusion of the full subcategory with objects 0, ..., n.

(ii) sn Set := CAT(∆op


≤n , Set).

(iii) The n-skeleton of X ∈ sSet is defined as

skn X := colim ∗ ∆m ∈ sSet.


m∈in /(in ) X

(iv) The n-coskeleton of X ∈ sSet is defined as

coskn X := sSet(skn ∆• , X) : ∆op −→ Set, m 7−→ sSet(skn ∆m , X).

Proposition 2.75
For every n ≥ 0 there are adjunctions

sSet((in )! X, Y ) = sn Set(X, (in )∗ Y ), sn Set((in )∗ Y, Z) = sSet(Y, (in )∗ Z),

where

49
Chapter 2. Simplicial sets

• (in )! X := colim ∆m is called the left Kan extension of X along in ,


m∈∆≤n /X

• (in )∗ Y := Y ◦ in ,

• (in )∗ Z := sn Set((in )∗ ∆• , Z) is called the right Kan extension of X along in .

We have skn = (in )! (in )∗ and coskn = (in )∗ (in )∗ and get an adjunction

sSet(skn X, Y ) = sSet(X, coskn Y ).

Moreover counit of the first and unit of the second adjunction induce natural transforma-
tions
ε η
skn X = (in )! (in )∗ X −→
X
X, Y
Y −→ (in )∗ (in )∗ Y = coskn Y, X, Y ∈ sSet.
Proof. Using that Hom-functors preserve limits by Proposition 2.46 and twice Yoneda’s
isomorphism 2.11 we get the adjunction
∼ ∼
sSet((in )! X, Y ) = sSet( colim ∆m , Y ) −→ lim sSet(∆m , Y ) −→ lim Ym
m∈∆≤n /X m∈∆≤n /X m∈∆≤n /X
 

 Y 

∗ 0
= y∈ Ym ; g yx0 = yx = yg∗ x0 , g ∈ ∆≤n (x, x ) = sn Set(X, Y ).
 
 x∈Xm , 
0≤m≤n


Using the the isomorphism ε : colim m∈∆/X ∆m −→ X of the co-Yoneda Lemma 2.51, we
get the other adjunction as the composition

sn Set((in )∗ X, Y ) sSet(X, (in )∗ Y )


ε∗ o ε∗ o
 
sn Set( colim (in )∗ ∆m , Y ) sSet( colim ∆m , sn Set((in )∗ ∆• , Y ))
m∈∆/X m∈∆/X

o o
 
lim sn Set((in )∗ ∆m , Y )
∼ / lim sSet(∆m , sn Set((in )∗ ∆• , Y )),
m∈∆/X m∈∆/X

where the vertical maps are isomorphisms, because hom-functors preserve (co-)limits by
Proposition 2.46 and the lower horizontal map is Yoneda’s isomorphism of Lemma 2.11.
Moreover (in )∗ preserves (co-)limits, as colimits in functor categories are constructed di-
mensionwise by Proposition 2.43. By definition we have skn = (in )! (in )∗ and

coskn Y = sSet(skn ∆• , Y ) = sn Set((in )∗ ∆• , (in )∗ Y ) = (in )∗ (in )∗ Y, Y ∈ sSet.

50
2.16. Skeleton and coskeleton

Proposition 2.76
For X ∈ sSet there are natural cocartesian squares (meaning that the lower right object
is the colimit of the rest of the diagram)

X̃n ×  ∂∆n / skn−1 X


_

 
X̃n × ∆n / skn X,
where ∂∆n = B∂I(n) and the horizontal maps send (x, f ) to f ∗ x and the vertical maps
are the canonical maps.
Proof. By Proposition 2.46 we can equivalently prove that, for every Y ∈ sSet, applying
sSet(-, Y ) yields a cartesian square

sSet(skn X, Y ) / sSet(X̃n × ∂∆n , Y )

 
sSet(skn−1 X, Y ) / sSet(X̃n × ∆n , Y ),
which by using the adjunction bijections of Proposition 2.75 is naturally isomorphic to
the diagram

sn Set((in )∗ X, (in )∗ Y ) / Set(X̃n , Yn )

 

sn−1 Set((in−1 ) X, (in−1 ) Y ) ∗ / Set(X̃n , sSet(∂∆n , Y )).

The left vertical map sends f = (f0 , ..., fn ) to the tuple (f0 , ..., fn−1 ), while the upper
horizontal map sends f to the restriction of fn to X̃n . Now composition with the natural
surjection
a a a
d := di : ∆n−1 = BI(n − 1) − B∂I(n) = ∂∆n ,
0≤i≤n 0≤i≤n 0≤i≤n

yields an injection

Y Y
d∗ : Set(X̃n , sSet(∂∆n , Y )) ,−→ Set(X̃n , sSet(∆n−1 , Y )) −→ Set(X̃n , Yn−1 ).
i 0≤i≤n

By construction the lower horizontal map in the last diagram composed with d∗ sends f
to (fn−1 di )i , while the composition with the right vertical map sends f to (di f )i . Using
the explicit description of Set-limits of Proposition 2.38 the pullback is the set
{(f0 , ..., fn−1 , f˜n ) ∈ sn−1 Set((in−1 )∗ X, (in−1 )∗ Y ) × Set(X̃n , Yn ); di f˜n = fn−1 di 0 ≤ i ≤ n},
whose elements uniquely glue together to a natural transformation sn Set((in )∗ X, (in )∗ Y ).
Indeed since by definition [
Xn = X̃n ∪ si Xn−1 ,
0≤i≤n−1

51
Chapter 2. Simplicial sets

we can define 
si fn (x), x ∈ si Xn−1 ,
fn (x) :=
f˜n (x), x ∈ X̃n ,
which is well-defined, because (f0 , ..., fn−1 ) is a natural transformation. The glueing pro-
cess defines an inverse map to the map restriction maps induced by the upper left two
maps of the diagram. This concludes the proof.
2

Corollary 2.77
For X ∈ sSet and n ≥ 0, the counit εX : skn X ,−→ X is injective.
So skn X ⊂ X is the simplicial subset generated by all nondegenerate simplices of di-
mension ≤ n. Moreover:

(i) (skn X)m = Xm , for all 0 ≤ m ≤ n.


[
(ii) colim skn X = skn X = X.
n≥0
n≥0

(iii) skn X/skn−1 X ∼


_
= (X̃n × (∆n /∂∆n ))/(X̃n × ∗) = S n.
x∈X̃

Corollary 2.78
The geometric realization |X| of a simplicial set X ∈ sSet is a CW-complex.
Proof. The geometric realization functor is a left adjoint. So we get cocartesian squares

X̃n × |∂∆ n
| / |skn−1 X|
_

 
X̃n × |∆n | / |skn X|.

In other words |skn X| is obtained from |skn−1 X| by glueing the set of nondegenerate
simplices and [
|X| = colim |skn X| = |skn X|.
n≥0
n≥0

The following proposition shows, that the nerve functor behaves well with the two
skeleton constructions.

Proposition 2.79

skn BX ∼
= BskX , X ∈ Simp, n ≥ 0.

52
2.16. Skeleton and coskeleton

ε ιn
Proof. By naturality of the counit map skn = (in )! (in )∗ −→ idsSet the inclusion skn X ,−→
Xn induces a commutative square
εBskn X
skn Bskn X / Bskn X
skn B(ιn ) B(ιn )
 εBX 
skn BX / BX.
B
By definition of skn X and the nerve functor Simpo −→ sSet, we have an isomorphism

(in )∗ Bskn X = Simpo (I(n), skn X) −→ Simpo (I(n), X) = (in )∗ BX.

Applying the functor (in )! yields the left vertical map, which is therefore an isomorphism,
too. Moreover for the nondegenerate simplices we have

m skn X = {f ∈ Simpp (I(m), skn X); f injective} = ∅,


B^ m > n.

So in the diagram of Proposition 2.76, the left vertical map is an isomorphism. As the
diagram is a pushout square, also the right vertical map is an isomorphism and we obtain
isomorphisms
∼ ∼ ∼
εBskn X : skn Bskn X −→ skn+1 Bskn X −→ ... −→ Bskn X,

proving that also the upper horizontal map in the commutative square is an isomorphism.

Corollary 2.80

For every n ≥ 0, the simplicial standard n-simplex ∆n is n-skeletal, i.e. skn ∆n −→ ∆n .

Proof. Since skn I(n) −→ I(n), Proposition 2.79 yields a commutative square of isomor-
phisms
εBskn I(n)
skn Bskn I(n) / Bskn I(n)

 εBI(n) 
skn ∆n = skn BI(n) / BI(n) = ∆n .
ε=id
Alternatively, we can use that the category ∆/∆n has a final object f := (∆n −→ ∆n ),
inducing an isomorphism ιf and a factorization of the identity map
ιf ε n
id : ∆n −→ colimn ∆m −→

∆n .
m∈∆

Remark 2.81
Let X ∈ sSet and n ≥ 0.
Then using Corollary 2.77 we get

53
Chapter 2. Simplicial sets

(i) (coskn X)m = sSet(skn ∆m , X) = sSet(∆m , X) = Xm , 0 ≤ m ≤ n.



(ii) X −→ lim coskn X.
n≥0

ηX
The map X −→ coskn X is not surjective in general. Defining the simplicial subset
cosk0n X := ηX (X) ≤ coskn X, we get a tower

X − ... − cosk01 X − cosk00 X,

satisfying (i) and (ii), that is closely related to the Postnikov-tower. It can be shown that
for a pointed space ∗ ∈ X ∈ T op the fibre Fn of the map cosk0n S(X) − cosk0n−1 S(X)
satisfies 
∼ πn (X, ∗), m = n,
πm (|Fn |, ∗) =
1, m 6= n.
Similar properties hold for any fibrant simplicial set, that we will introduce later.

2.17 Small categories as simplicial sets


Remark 2.82
Every partially ordered set P can be considered as a category P with objects P and homo-
morphisms 
{≤}, x ≤ y,
P (x, y) :=
∅, x 6≤ y.
Indeed there is only one way to define the composition maps, which are well-defined as
‘≤’ is transitive. As it is reflexive, every object has an identity morphism.
This construction defines a functor ∆ ,−→ Cat. By construction we have

Cat(m, n) = ∆(m, n), m, n ∈ ∆,

which means that ∆ is a full subcategory of Cat.

Definition 2.83
Let C ∈ Cat be an ordered simplicial complex.
The nerve of C is defined as the composite functor
Cat(-,X)
BC : ∆op −→ Cat −→ Set, n 7−→ Bn X.

B
Remark 2.84 (i) The nerve induces a functor Cat −→ sSet.
As hom-functors preserve limits by Proposition 2.46, so does B by Proposition 2.43.

(ii) Again the Yoneda embedding ∆ −→ sSet coincides with the composition
I B
∆ −→ Cat −→ sSet.

54
2.17. Small categories as simplicial sets

Remark 2.85
For C ∈ Cat we have by construction

• B0 C = Cat(0, C) = Set(0, U (C)) = Obj(C),


a
• B1 C = Cat(1, C) = C(x0 , x1 ) =: Mor(C).
x0 ,x1 ∈C
a
• Bn C = C(x0 , x1 ) × ... × C(xn−1 , xn ), for all n ≥ 0.
x0 ,...xn ∈C

Moreover the face and degeneracy maps are given by



 (f2 , ..., fn ), i = 0,
• di (f1 , ..., fn ) = (f1 , ..., fi fi+1 , ..., fn ), 0 < i < n,
(f1 , ..., fn−1 ), i = n.

• si (f1 , ..., fn ) = (f1 , ..., fi , id, fi+1 , ..., fn ), 0 ≤ i ≤ n.

Remark 2.86
Every group G ∈ Grp (or more generally every monoid) gives rise to a category G with a
single object ∗ and homomorphisms G(∗) = G. Composition is defined by multiplication
in G and the neutral element forms the identity morphism.

(i) By definition of categories and groups, a homomorphisms f ∈ Grp(G, H) bijectively


corresponds to a functor
f
f : G −→ H, ∗ 7−→ ∗, G(∗) = G −→ H = H(∗).

In particular we can view Grp as a full subcategory of Cat.

(ii) In the context of group (co-)homology the nerve BG is better known as the bar
construction of G. Using the comparison of the homotopy categories of T op and
sSet we will see that

H∗ (G, Z) = H∗ (BG, Z) −→ H∗ (|BG|, Z),

where the first equality is by definition and the second isomorphism will be induced
ηBG
by the unit map BG −→ S|BG|.

Proposition 2.87
B
The nerve functor Cat −→ sSet is fully faithful, meaning that for all C, D ∈ Cat it
induces a bijection

B : Cat(C, D) −→ sSet(BC, BD), F 7−→ B(F ).
Proof. We define the maps

ti : 1 −→ n, 0 7−→ i − 1, 1 7−→ i, 1 ≤ i ≤ n,

55
Chapter 2. Simplicial sets

and let ti := (ti )∗ ∈ Set(Xn , X1 ) denote the corresponding map, for X ∈ sSet. Then for
all C ∈ Cat using Remark 2.85 we get

ti (f1 , ..., fn ) = fi , (f1 , ..., fn ) ∈ Bn C.

In particular, for all X ∈ sSet and g ∈ sSet(X, BC) we get

gn (x) = (t1 gn , ..., tn gn )(x) = (g1 t1 , ..., g1 tn )(x) = (g1 t1 (x), ..., g1 tn (x)), x ∈ Xn , n > 1.

So g is completely determined by g0 and g1 . Moreover it follows that every g ∈ sSet(BC, BD)


G
defines a functor C −→ D by setting

g0 : Obj(C) −→ Obj(D), g1 : Mor(C) −→ Mor(D).

Indeed:
• G(f1 ◦ f2 ) = g1 d1 (f1 , f2 ) = d1 g2 (f1 , f2 ) = d1 (g1 (f1 ), g1 (f2 )) = G(f1 ) ◦ G(f2 ),

• G(idx ) = g1 s0 (x) = s0 g0 (x) = idG(x) , x ∈ C.


We have proven that there is a bijection

Cat(C, D) −→
←− sSet(BC, BD),
F 7−→ BF,
G ←−[ g.

2
While in general it is often tedious to explicitly construct homotopies between two
homomorphisms of simplicial sets, there is an easy description, when the simplicial sets
are the nerves of categories. Like the categories Set and sSet also the category Cat has
an object of internal homomorphism.

Remark 2.88
There is a natural bijection

Cat(X × Y, Z) = Cat(X, Cat(Y, Z)), f 7−→ [x 7−→ f (x, -)],

where we recall that Cat(Y, Z) is the category of functors Y −→ Z with natural transfor-
mations as homomorphisms.

Proposition 2.89
For F0 , F1 ∈ Cat(C, D) the set of simplicial homotopies BF1 ' BF0 bijectively corresponds
h
to the set of natural transformations Cat(C, D)(F1 , F0 ).
In particular every adjunction C(F (X), Y ) = D(X, G(Y )) induces a simplicial homo-
topy equivalence
'
BF : BD −→
←− BC : BG.

56
2.17. Small categories as simplicial sets

Proof. There is a commutative diagram

sSet(B0 × BC, BD)


∼ / sSet(B(0 × C), BD) o B Cat(0 × C, D) = Cat(0, Cat(C, D))

di di di
  
sSet(B1 × BC, BD)
∼ / sSet(B(1 × C), BD) o B Cat(1 × C, D) = Cat(1, Cat(C, D)),

di
where the vertical maps are induced by the map 0 −→ 1, for i = 0, 1. The left two
horizontal maps are bijections, because B preserves products by Remark 2.84, the right
two horizontal maps are the bijections of Proposition 2.87 and the right two horizontal
maps are those of Remark 2.88. In particular using Bn = ∆n we get a bijection

{h ∈ sSet(∆1 × BD, BD); BF1 ' BF0 } = {h ∈ sSet(∆1 × BC, BD); di (h) = Fi }
h
= {t ∈ Cat(1, Cat(C, D)); di (t) = Fi }
= Cat(C, D)(F1 , F0 ).

57
3 Abstract homotopy theory

3.1 Localizations of categories


Given a category C ∈ CAT and a class of morphisms in C preserving the structure “very
well”, but not as well as isomorphisms.

Example 3.1 (i) T op and homotopy equivalences.


f |f |
(ii) sSet and maps X −→ Y , such that |X| −→ |Y | is a homotopy equivalence.
f
(iii) T op and weak homotopy equivalences, i.e. maps X −→ Y with
∼ ∼
π0 f : π0 X −→ π0 Y, πn f : πn (X, x) −→ πn (Y, f (x)), x ∈ X, n > 0.

f
(iv) Chain complexes and quasi-isomorphisms, i.e. chain maps X −→ Y with

H∗ f : H∗ X −→ H∗ Y.

We want to study the objects in C modulo such equivalences.

Definition 3.2
Let C ∈ CAT and S ⊂ Mor(C) a subclass of morphisms.
γ
Then C −→ S −1 C is called a localization of C at S, if the following holds.
(i) γ(S) ⊂ Mor(S −1 C)× , i.e. γ sends S to isomorphisms.

(ii) It is universal with respect to property (i). This means, that for all F ∈ CAT(C, D)
with F (S) ⊂ Mor(D)× , there is a unique functor inducing a commutative diagram

C
F /
<D
γ
 ∃!F̃
S −1 C.

Example 3.3
For a commutative monoid M ∈ Mon and a submonoid S ≤ M , we define a monoid

S −1 M := (M × S)/ ∼, (m, s) ∼ (m0 , s0 ) ⇐⇒ ∃t ∈ S : m + s0 + t = m0 + s + t.

59
Chapter 3. Abstract homotopy theory

Then S −1 M is generated by the images of M × 0 and 0 × S. The identity


(s, 0) + (0, s) = (s, s) ∼ (0, 0), s ∈ S,
shows that S −1 M is the monoid obtained from M = M × 0 by adding formal inverses
s−1 = (0, s) for s ∈ S.
In particular the map M −→ S −1 M , sending m to (m, 0) is a localization for M
considered as a category with one object.

Remark 3.4
In general localizations are very hard to construct and may not exist, if C is not small.

Definition 3.5
A category with weak equivalences consists of a category C ∈ CAT and a subclass of
'
morphisms wC ⊂ Mor(C), so-called weak equivalences (wes.) (written “−→”), such
that:
(i) Mor(C)× ⊂ wC.
(ii) The 2-of-3 axiom holds, i.e. for all A, B, C ∈ C and every commutative triangle
f
A / B
g
gf  
C,
if 2 of the 3 morphisms are wes., so is the third.
Its homotopy category for C, if it exists, is defined as the localization Ho(C) = (wC)−1 C.

Remark 3.6 (i) For every functor F ∈ CAT(C, D) the pair (C, F −1 Mor(D)× ) is a cat-
egory with weak equivalences.
γ
(ii) Let C ∈ CAT and suppose S ⊂ Mor(C) admits a localization C −→ S −1 C.
Then S := γ −1 Mor(S −1 C)× defines a category with weak equivalences (C, S) with
−1
Ho(C) = S C = S −1 C.

3.2 Weak factorization systems


Definition 3.7
Let C ∈ CAT and ` ∈ C(A, B), r ∈ C(C, D).
If for any commutative square
A 9/ C
∃d
` r
 
B / D
there exists a diagonal d making the diagram commutative, then one says, that

60
3.2. Weak factorization systems

(i) ` has the left lifting property (LLP) with respect to r and
(ii) r has the right lifting property (RLP) with respect to `.
For a subclass S ⊂ Mor(C) define
(i) LLP (S) := {f ∈ Mor(C); f has the LLP w.r.t. all s ∈ S},
(ii) RLP (S) := {f ∈ Mor(C); f has the RLP w.r.t. all s ∈ S}.

Definition 3.8
Let C ∈ CAT and L, R ⊂ Mor(C).
Then (L, R) is a weak factorization system on C, if
(i) Mor(C) = R ◦ L,
(ii) L = LLP (R) and R = RLP (L).

Example 3.9
The pair (Epi, Mono) forms a weak factorization system on the category Set:
f f
(i) Every map X −→ Y can be factored as X − f (X) ,−→ Y .
(ii) Every lifting problem
A
u / C
> _
∃d
e m
 
B / D
v

has a solution d sending b = e(a) to u(a). This is well-defined, because e(a) = e(a0 )
implies
mu(a) = ve(a) = ve(a0 ) = mu(a0 ),
and hence u(a) = u(a0 ), as m is injective. By construction de = u and hence mde =
mu = ve implies also md = e, as e is surjective.
Note that in this case the diagonal d is unique. Similarly one shows that (Epi, Mono) is
also a weak factorization system on Grp or R-Mod, for R ∈ Ring.

Remark 3.10
On every category C there are two trivial weak factorization systems (Mor(C)× , Mor(C))
and (Mor(C), Mor(C)× ), where Mor(C)× is the class of isomorphisms in C:
(i) Every map f ∈ C(X, Y ) can be factored as f ◦ idX (resp. idY ◦ f ).
(ii) Every lifting problem
A
u /
>C
∃d
` r
 
B / D,
v

can be solved by d := u`−1 , when ` is an isomorphism (resp. by d := r−1 v, when r


is an isomorphism).

61
Chapter 3. Abstract homotopy theory

3.3 Model categories


Definition 3.11
Let C be a finitely complete and finitely cocomplete category, i.e. there are limits and
colimits for functors I −→ C, where I ∈ Cat has only finitely many objects and morphisms.
A model structure on C consists of three subclasses wC, fib C, cof C ⊂ Mor(C), such
that:

(i) (C, wC) is a category with weak equivalences.

(ii) (cof C ∩ wC, fib C) and (cof C, wC ∩ fib C) are weak factorization systems.

The tuple (C, wC, fib C, cof C) is also called a model category.
We fix the following notation.

• The morphisms in fib C are called fibrations and written as “−_”.

• The morphisms in cof C are called cofibrations and written as “−→”.

• A (co-)fibration is called trivial, if it is also a weak equivalence.

We will show, that every model category has a homotopy category. One begins by
h
formalizing the idea of a cylinder [0, 1] × X and of homotopies [0, 1] × X −→ Y in
the context of a model category. Then the proof is quite technical, but inspired by the
constructions in the category of topological spaces.
G
Constructing model structures is often very hard. Given a right adjoint functor C −→ D
mapping into a model category D, it is often possible to construct an induced model
structure on C with wC := G−1 wD and fib C := G−1 fib D. Many important examples can
be constructed in this way.
However the remaining problem is to construct at least one archetypical model category
to begin with. This is the/a canonical model structure on the category of simplicial sets,
for which weak equivalences are maps becoming homotopy equivalences after geometric
realization. By the technique using right adjoint functors, we will get model structures
for Example 3.1 (iii) and (iv).
We will proceed by first constructing the model structure on the category of simplicial
sets. It will take us the next few lectures to achieve this goal.

3.4 Stability of the lifting property


Before coming to the construction of weak factorization systems, we will need some stabilty
properties of the lifting property.

Definition 3.12
Let C be a category.

62
3.4. Stability of the lifting property

(i) A retract of a morphism f ∈ C(A, B) is a retract of f in the category CAT(1, C) of


morphisms in C. This means a retract of f is a morphism f 0 ∈ C(A0 , B 0 ) inducing a
commutative diagam
idA0
0 /A (/
A A0
f0 f f0
  
B0 / B / B0.
6
idB 0

(ii) For C ⊂ Mor(C) we denote by retr(C) the class of retracts of morphisms in C.


(iii) A set of morphisms C ⊂ Mor(C) is closed under retracts, if retr(C) ⊂ C.

Lemma 3.13
Let M ⊂ Mor(C) be a class of morphisms in a category C ∈ CAT.
Then the following holds:
(i) LLP (M ) contains every isomorphism of C.
(ii) LLP (M ) is closed under composition.
c
0 1 c
(iii) Let Z0 −→ Z1 −→ ... be a tower of morphisms in LLP (M ).
0 ι
Then also Z0 −→ colim n≥0 Zn lies in LLP (M ).
(iv) LLP (M ) is closed under retracts.
(v) LLP (M ) is closed under pushouts along arbitrary morphisms in C.
(vi) LLP (M ) is closed under arbitrary coproducts.
By duality we have similar properties for RLP (M ).
Proof.
(i) A lifting problem
A
u /9 C
∃d
` o m∈M
 
B / D
v

is solved by d := u`−1 .
f g
(ii) Let A, B, C ∈ C and A −→ B −→ C be two morphisms in LLP (M ). For any
m
(X −→ Y ) ∈ M admitting a commutative square
A /6
=X
∃df
f

B ∃dg
m∈M

g
 
C / Y

63
Chapter 3. Abstract homotopy theory

there are morphisms df and dg (depending on df ) preserving the commutativity of


the square, because f, g ∈ LLP (M ). Hence dg is a diagonal for the outer square and
hence g ◦ f ∈ LLP (M ).
m
(iii) Given such a tower (Zn )n≥0 . For any (X −→ Y ) ∈ M and any commutative diagram

Z0
u / X
ι0 m∈M
 
colim Zn v / Y
n≥0

n d
we inductively construct maps Zn −→ X, for n ≥ 0. Set d0 = u and for any n ≥ 0
we get a diagonal dn+1 , because cn ∈ LLP (M ).

dn
Zn 5/ X
∃dn+1
cn m
 
Zn+1 / colim Zn / Y
ιn+1 n≥0 v

Since by construction dn+1 cn = dn , for all n ≥ 0, the universal property of colimits


yields a map d := colim n≥0 dn satisfying the required properties.

(iv) Consider a commutative diagram

idA

A / A0 /* A u / X
sA rA
e e0 e m∈M
 sB
 rB  
B / B0 4/ B
v / Y,
idB

where e0 ∈ LLP (M ). We want to show, that there is a diagonal d ∈ C(B, X), making
the right square commute. Since e0 ∈ LLP (M ), we have a diagonal d0 , such that
rA
A0 / A u
7/ X
∃d0
e0 m∈M
 
B0 / B / Y
rB v

commutes. Define d = d0 sB and compute

de = d0 sB e = d0 e0 sA = urA sA = u, md = md0 sB = vrB sB = v,

hence d is a diagonal of the desired form.

64
3.4. Stability of the lifting property

f
(v) Let A, B, C ∈ C and A −→ C and e ∈ LLP (M ). Consider a commutative diagram
f
A / C u / X
e ιC m∈M
  
B /B +A C / Y.
ιB v

d0
Since e ∈ LLP (M ) there is a diagonal B −→ X for the outer square, i.e. d0 ◦e = u◦f
and r ◦ d0 = v ◦ ιC . Applying the universal property of pushouts to the first identity
yields a diagonal for the right square d := d0 ∪ u ∈ C(B +X C, X):
• dιC = u by construction,
• mdιC = mu = vιC and mdιB = md0 = vιB and hence md = v by the universal
property of pushouts again.
ie
(vi) Let I ∈ Set and (Ai −→ Bi ) ∈ LLP (M ) for any i ∈ I. Let e denote the coproduct
of all ei . Consider a commutative square (on the right)
ι
Ai i / i∈I Ai /X
`

ei e m∈M
 ιi
 
/ / Y.
`
Bi i∈I Bi
i d `
For any i ∈ I there is a diagonal Bi −→ X for the outer square. Then d = i∈I di
we have d ◦ e = u is a diagonal for the right square.
2

Corollary 3.14
Let C ∈ CAT and L, R ⊂ Mor(C).
Then (L, R) is a weak factorization system, if and only if the following holds:
(i) Mor(C) = R ◦ L.
(ii) L ⊂ LLP (R).
(iii) L and R are closed under retracts.
Proof. Suppose (L, R) is a weak factorization system. Then (i) and (ii) holds, because
L = LLP (R). Furthermore LLP (R) resp. RLP (L) are closed under retractions by (iii)
of the preceding Lemma resp. its dual.
Vice versa assume the above hypotheses for (L, R). We have to show, that L = LLP (R)
and R = RLP (L). Let c ∈ C(X, Y )∩ ∈ LLP (R) be arbitrary. Then c can be factored as
` r
X −→ Z −_ Y , where ` ∈ L and r ∈ R. Since c ∈ LLP (R) we find a diagonal

X / /
`
>Z
d
c r
 _
Y Y,

65
Chapter 3. Abstract homotopy theory

showing that c is a retract of `. Hence LLP (R) ⊂ L ⊂ LLP (R) and so L = LLP (R).
Dually one gets R = RLP (L).
2

Using the preceding Corollary we may deduce Quillen’s original defintion of a closed
model category.

Corollary 3.15
Let C be a category together with three classes of morphisms wC, fib C, cof C ⊂ Mor(C).
Then C is a model category, if and only if the following holds1 .
(CM1) C is closed under finite limits and colimits.
(CM2) wC satisfies the 2-of-3 axiom.
(CM3) The classes wC, fib C, cof C are closed under retracts.
(CM4) cof C ⊂ LLP (fib C ∩ wC) and cof C ∩ wC ⊂ LLP (fib C).
(CM5) MorC = fib C ◦ (cof C ∩ wC) = (fib C ∩ wC) ◦ cof C.
Proof. Axiom (CM1) is the general hypothesis in the Definition 3.11 of model categories.
(CM2) is equivalent to (C, wC) being a category with weak equivalences. By Corollary
3.14 (cof C ∩ wC, fib C) and (cof C, wC ∩ fib C) are weak factorization systems, if and only
if (CM3) - (CM5) hold. Indeed (CM5) and (CM3) imply that (trivial) (co-)fibrations
are closed under retracts. Vice versa if trivial fibrations and cofibrations are closed under
retracts, so are weak equivalences, since they can be factored into a trivial cofibration
followed by a trivial cofibration.
2

3.5 Construction of weak factorization systems


Part of the construction of a model category is the construction of a weak factorization
system. We will present Quillen’s small object argument, which is probably the most
powerful tool for building factorization systems.

Proposition 3.16
Let C be a category with finite coproducts.
Then there is a canonical weak factorization system (retr(I), R) on C, where
ιX
(i) I is the class of inclusions X ,−→ X + Y , where X, Y ∈ C,

(ii) R is the class of retractions.

1
This is what Quillen originally called a closed model category. He skipped (CM3) in his definition of
a model category.

66
3.5. Construction of weak factorization systems

f
Proof. Every morphism X −→ Y can be factored as
f
X / Y
;

ιX
# f ∪idY
X + Y,

and f ∪ idY is a retraction with a section given by the inclusion ιY . Every lifting problem

X _ a /6 Ad
d
ιX r s
 
X +Y / B,
(ra)∪b

is solved by the morphism d = a ∪ (sb), because

dιX = a, rd = (ra) ∪ (rsb) = (ra) ∪ b.

By Lemma 3.13 LLP (R) is closed under retracts and thus also retracts of inclusions have
the left lifting property to retractions.
Using Corollary 3.14 it remains to prove, that R is closed under retracts. Suppose we
are given a commutative diagram
idA0

A0 / A /( A 0 Ad
sA rA

r0 r r0 r s
   
sB rB
B0 / B / B0, B.
6
idB 0

Define s0 = rA ssB ∈ C(B 0 , A0 ). Then we have

r0 s0 = r0 rA ssB = rB rssB = rB sB = idB 0 ,

showing that r0 is a retraction with section s0 .


2

Example 3.17
In the category Set we have (retr(I), R) = (Mono, Epi).
m
Indeed every inclusion is an injection and every injection A ,−→ B is isomorphic to the
ιA
inclusion A −→ A + (B\m(A)), so I = Mono. Using that Mono = RLP (Epi) by Example
3.9, Lemma 3.13 implies that

I = Mono = retr(Mono) = retr(I).

Moreover the axiom of choice is equivalent to R = Epi.


So we have constructed two non-trivial weak factorization systems (Epi, Mono) and
(Mono, Epi) on Set.

67
Chapter 3. Abstract homotopy theory

Definition 3.18
Let C ∈ CAT and C ⊂ Mor(C).
A C-cell complex relative to X ∈ C is an object Z ∈ C obtained by inductively
glueing C-cells to X, i.e.

(i) Z = colim n≥0 Zn , for some X = Z0 −→ Z1 −→ ....

(ii) Zn+1 is a pushout


`
i fi fi
/ Zn /
`
i Di Di Zn
ci ∈C
`
i ιi ci
`  
Ei / Zn+1 , Ei ,
i

Remark 3.19
The notion generalizes the construction of CW-complexes. More precisely in the category
T op, form the set
C := {S n ,−→ Dn ; n ≥ 0}.
Then a C-cell complex relative to X ∈ T op is a space Z obtained by inductively glueing
cells to X.
In particular a C-cell complex relative to ∅ is a usual CW-complex.

Theorem 3.20 (Quillen’s small object argument)


Let C ∈ CAT and C ⊂ Mor(C), such that

(i) C is cocomplete.

(ii) C is a set.

(iii) For every (D → E) ∈ C its domain D is ω-compact2 (also called small), i.e.
C(D, -) commutes with colimits over (N0 , ≤).
g
Then every morphism X −→ Y has a factorization
g
X /
?Y

c f
Z,

where f ∈ RLP (C) and c is a C-cell complex relative to X.


g
Proof. Given a morphism X −→ Y in C. For any n ∈ N0 we inductively construct a
sequence
c0 c1 c2
X = Z0 −→ Z1 −→ Z2 −→ ...

2
The term ω-compact refers to the smallest infinite ordinal ω. One can extend the notion of compactness
to higher ordinals κ to allow more objects appearing as a domain of maps in C.

68
3.5. Construction of weak factorization systems

fn
and morphisms Zn −→ Y for any n ≥ 0 commuting with the maps of this tower. This is
done by attaching “cells” coming from maps in C to X. We will precise this idea in the
following.
Let Z0 = X and f0 = g. For a given Zn let C/fn denote the set of all commutative
squares
D / Zn

C3c fn
 
E / Y.
c
This is a set, because C is a set and C(D, Zn ) and C(E, Y ) are sets, for all (D −→ E) ∈ C.
Define Zn+1 as the pushout

/
`
C/fn D Zn
`
C/fn ιc c =:cn
fn
`  / Zn+1

C/fn E
∃!fn+1

& 
1 Y.

Using the universal property of colimits, we get a factorization


g
X /8 Y

c:=ι0 colim n≥0 fn =:f


&
Z := colim Zn .
n≥0

To show, that f ∈ RLP (C) consider a commutative diagram

D
u /Z

C3c f
 
E
v / Y.

Because D is ω-compact we have a natural bijection



colim C(D, Zn ) −→ C(D, colim Zn ) = C(D, Z).
n≥0 n≥0

This means the map u ∈ C(D, Z) comes from a map un ∈ C(D, Zn ) with n ≥ 0, i.e.
u = ιn ◦ un . Hence fn ◦ un = (f ◦ ιn ) ◦ un = f ◦ (ιn ◦ un ) = f ◦ u = v ◦ c, meaning that
c ∈ C/fn is in the indexing set for the construction of the pushout Zn+1 . So considering

69
Chapter 3. Abstract homotopy theory

ιc
v 0 : E ,−→
`
C/fn E −→ Zn+1 we obtain a commutative diagram
u

D / Zn /8( Z
un ιn
cn
ιn+1

c
8
Zn+1 f
v0 fn+1

 &/ 
E v Y.
In particular the diagonal d := ιn+1 v 0 solves the lifting problem.
2

Corollary 3.21
Let C ∈ CAT and C ⊂ Mor(C) a set of morphisms admitting the small object argument.
Then
(i) (LLP (RLP (C)), RLP (C)) is a weak factorization system in C.
(ii) LLP (RLP (C)) is precisely the class of retracts of relative C cell complexes in C.
Proof.
(i) Clearly C ⊂ LLP (RLP (C)). By Lemma 3.13 LLP (RLP (C)) contains every iden-
tity morphism and is closed under coproducts, pushouts and directed limits. This
implies LLP (RLP (C)) contains every relative C-cell complex. Hence the small ob-
ject argument shows, that Mor(C) = RLP (C) ◦ LLP (RLP (C)). By definition every
` ∈ LLP (RLP (C)) has the LLP for any map in RLP (C). But this is also equivalent
to the fact, that every map in RLP (C) has the RLP for any map LLP (RLP (C)).
(ii) Since C ⊂ LLP (RLP (C)) and LLP (RLP (C)) by Lemma 3.13 is closed under
pushouts, filtered colimits and retractions, we have, that every retraction of a relative
C-cell complex lies in LLP (RLP (C)).
f
The other way round let (X −→ Y ) ∈ LLP (RLP (C)). By the small object argu-
ment we find a factorization
f
X /
>Y

c r
0
Y ,
where c is a relative C-cell complex and r ∈ RLP (C). Since c ∈ C ⊂ LLP (RLP (C))
we find a diagonal in the diagram
X
c / 0
>Y
∃d
f r
 
Y Y,
showing that f is a retraction of the cell complex c.

70
3.5. Construction of weak factorization systems

Proposition 3.22 (i) Every finite set F is ω-compact in Set.

(ii) Every finite simplicial set F ∈ sSet is ω-compact in sSet.


Proof. Note that the category 0 −→ 1 −→ ... is filtered.

(i) Given a sequence of sets


S0 −→ S1 −→ ...
that we consider a sequence of constant functors D(F ) −→ Set, by Proposition 2.48
we get the natural isomorphism

∼ /
Y Y
colim Sn colim Sn
n≥0 n≥0
x∈F x∈F

colim Set(F, Sn )
∼ / Set(F, colim Sn ).
n≥0 n≥0

(ii) Using Proposition 2.76 finiteness of F implies that there is an n ≥ 0, such that

εF : skn F −→ F . Moreover all the sets F0 , ..., Fn are finite. Then there is a chain of
isomorphisms

sSet(F, Y ) −→ sSet(skn F, Y ) = sSet( colim ∆m , Y )
m∈(in )/(in )∗ F
∼ ∼
−→ lim sSet(∆m , Y ) −→ lim Ym ,
m∈(in )/(in )∗ F m∈(in )/(in )∗ F

where the first isomorphism is induced by the isomorphism εF , the second holds by
Definition 2.74 of the skeleton, the third map is an isomorphism as hom-functors
preserve limits by Proposition 2.46 and the fourth one is Yoneda’s isomorphism. So
the object on the right is the limit of the functor
Y
m ∈ (in )/(in )∗ F −→ ∆op −→ Set

and the assertion again follows from Proposition 2.48, because the category (in )/(in )∗ F
is finite.

Remark 3.23
However a compact topological space does not need to be ω-compact in T op.

71
Chapter 3. Abstract homotopy theory

3.6 Lifting properties and adjunctions


Remark 3.24
Given an adjunction between two categories C and D

C(F (X), Y ) = D(X, G(Y )).

Let ` ∈ D(D, D0 ) and r ∈ C(C, C 0 ).


Then the following lifting propblems are equivalent

a / a0 /
F (D) <C D G(C)
<
d d0
F (`) r ` G(r)
   
F (D0 ) / C 0, D0 / G(C 0 ),
b b0

where a0 , b0 and d0 corresponds to a, b and d under the adjunction bijection.


In particular the two axioms for structured adjunctions are equivalent.

Proposition 3.25
Let A, B and C be categories with functors
F ⊗ G
B op × C −→ A, A × B −→ C, Aop × C −→ B.

and natural bijections

A(A, F (B, C)) = C(A ⊗ B, C) = B(B, G(A, C)), A ∈ A, B ∈ B, C ∈ C.

Let a ∈ A(A, A0 ), b ∈ B(B, B 0 ) and c ∈ C(C, C 0 ).


Then the following three lifting problems are equivalent.

u01 u1 ∪u2
A / F (B 0 , C) (A ⊗ B 0 ) +(A⊗B) (A0 ⊗ B) 4/ C
4
∃d0 ∃d
a (b∗ ,c∗ ) (a⊗id)∪(id⊗b) c
   
A0 / F (B, C) ×F (B,C 0 ) F (B 0 , C 0 ), A0 ⊗ B 0 / C 0.
(u02 ,u03 ) u3

u00
B 2
/ G(A0 , C)
3
∃d00
b (a∗ ,c∗ )
 
B0 / G(A, C) ×G(A,C 0 ) G(A0 , C 0 ).
(u00 00
1 ,u3 )

where

• u01 , u02 , u03 correspond to u1 , u2 , u3 under the first bijection and

• u001 , u002 , u003 correspond to u1 , u2 , u3 under the second bijection.

72
3.6. Lifting properties and adjunctions

In other words, we have a chain of equivalences

a ∈ LLP (b∗ , c∗ ) ⇐⇒ (b∗ , c∗ ) ∈ RLP (a)


⇐⇒ (a ⊗ id) ∪ (id ⊗ b) ∈ LLP (c) ⇐⇒ c ∈ RLP ((a ⊗ id) ∪ (id ⊗ b))
⇐⇒ (a∗ , c∗ ) ∈ RLP (b) ⇐⇒ b ∈ LLP (a∗ , c∗ )
Proof. This holds, because
• a diagonal d0 corresponds to a diagonal d under the first bijection and

• a diagonal d00 corresponds to a diagonal d under the second bijection.


Equivalently every solution of one of the lifting problems induces a solution for the other
two problems.
2

Corollary 3.26
For C ∈ CAT with arbitrary coproducts and products, there are natural bijections

Set(S, C(X, Y )) = C(S X, Y ) = C(X, Y S ), S ∈ Set, X, Y ∈ C,

where S X := s∈S X and X S := s∈S X.


` Q
Suppose moreover that C carries a weak factorization system (L, R).
Then for c ∈ C(A, B) ∩ L, m ∈ Set(S, T ) ∩ Mono and f ∈ C(C, D) ∩ R we have:
(c∗ ,f∗ )
(i) C(B, C) − C(A, C) ×C(A,D) C(B, D),
m id∪id c
(ii) (S B +S A T A −→ T
B) ∈ L,
(m∗ ,f∗ )
(iii) (C T −→ C S ×DS DT ) ∈ R.
Proof. The natural bijections follow from the universal property of (co-)products.
(i) For all such c and f surjectivity of the map (c∗ , f∗ ) is equivalent to the existence of
a diagonal d in every commutative square

A
u /
>C
∃d
c f
 
B / D.
v

Indeed each such commutative square bijectively corresponds to an element

(u, v) ∈ C(A, C) ×C(A,D) C(B, D)

by the explicit construction of limits in the category Set. Hence (u, v) has a preimage

(c∗ , f∗ ) : d 7−→ (dc, f d),

if and only if there is a solution for the lifting problem.

73
Chapter 3. Abstract homotopy theory

(ii) Using that (Mono, Epi) is a weak factorization system on Set and (L, R) one on C,
Proposition 3.25 applied to the given natural bijection yields
(c∗ , f∗ ) ∈ Epi = RLP (Mono), for all c ∈ L and f ∈ R,
⇐⇒ (c∗ , f∗ ) ∈ RLP (m), for all c ∈ L, f ∈ R and m ∈ Mono(Set),
m id
⇐⇒ ( id ∪ c) ∈ LLP (f ), for all c ∈ L, f ∈ R and m ∈ Mono(Set),
m id
⇐⇒ ( id ∪ c) ∈ LLP (R) = L, for all c ∈ L, f ∈ R and m ∈ Mono(Set).
This shows that (ii) is equivalent to (i).

(iii) Similarly one checks that (iii) is equivalent to (i).

3.7 Towards the standard model structure on


simplicial sets
Definition 3.27
Let f ∈ sSet(X, Y ).

(i) f is called a cofibration, if it is a monomorphism.

(ii) f is called a fibration, if it has the LLP with respect to all maps
c
di tc := (di ×id)∪(id×c) : ∆0 ×B+∆0 ×A ∆1 ×A −→ ∆1 ×B, A −→ B, i = 0, 1.

(iii) f is called a weak equivalence, if it induces bijections



f ∗ : π0 sSet(Y, T ) −→ π0 sSet(X, T ), T −_ ∗.

(iv) An anodyne extension is an element in LLP (fib sSet).

Remark 3.28
The definition of weak equivalences might seem a bit awkward. However:

(i) Every simplicial homotopy equivalence is a weak equivalence.

(ii) We will later see that f ∈ sSet(X, Y ) is a weak equivalence, if and only if its
geometric realization |f | is a homotopy equivalence.

Theorem 3.29
There are two weak factorization systems on the category sSet

(cof sSet, RLP (cof sSet)), (LLP (fib sSet), fib sSet).

Moreover the following holds.

74
3.7. Towards the standard model structure on simplicial sets

ε
(i) cof sSet = LLP (RLP ({∂∆n = skn−1 ∆n ,−→ ∆n ; n ≥ 0})),
di tε
(ii) fib sSet = RLP ({∆0 × ∆n +∆0 ×∂∆n ∆1 × ∂∆n −→ ∆1 × ∆n ; n ≥ 0, i = 0, 1}),
Proof. The domains of the maps
ε di tι
∂∆n ,−→ ∆n , ∆0 × ∆n +∆0 ×∂∆n ∆1 × ∂∆n −→ ∆1 × ∆n , n ≥ 0, i = 0, 1
are finite (i.e. have only finitely many nondegenerate simplices) and hence ω-compact by
Proposition 3.22. So by Corollary 3.21 we get two induced weak factorization systems.
As one of the structure class of morphisms in a weak factorization system determines the
other one, the equalities (i) and (ii) imply that these are the weak factorization systems
we are looking for.
(i) Using that (Mono, Set) is a weak factorization system on Set by Example 3.17 and
that colimits in sSet are constructed dimensionwise by Proposition 2.43, it follows
that cof sSet = MonosSet is closed under pushouts, sequential colimits and retracts
by Lemma 3.13. In particular cof sSet also contains all retracts of relative cof sSet-
cell complexes. Now the map ∂∆n ,−→ ∆n is injective and thus a cofibration, for all
n ≥ 0. This shows the inclusion
ε
L := LLP (RLP ({∂∆n ,−→ ∆n ; n ≥ 0})) ⊂ cof sSet,
because L consists of retracts of relative cell complexes by Corollary 3.21.
For the other inclusion let c ∈ sSet(A, B) ∩ cof sSet. Note first that cn restricts to
c̃n
an injection Ãn ,−→ B̃n . Indeed supposing cn (a) ∈
/ B̃n , we have cn (a) = si (b), for
some b ∈ Bn−1 and 0 ≤ i ≤ n − 1. Then
cn (si di a) = si di cn (a) = si di si (b) = si (b) = cn (a)
and injectivity of cn implies a = si di a ∈
/ Ãn .
Next we claim that there is a cocartesian square
Ãn
∆n +Ãn ∂∆n B̃n
∂∆n / skn−1 B +skn−1 A A
(c id)∪(id ε)
 
B̃n
∆n / skn B +skn A A,
for all n ≥ −1, where for the case n = −1 we define sk−1 X := ∅ for any simplicial
set X ∈ sSet.
To prove the claim we consider the commutative diagram
skn−1
O
Bo skn−1
O
A / AO

B̃n
∂∆n o Ãn
∂∆n / Ãn ∆n

 
B̃n
∆ n o Ãn
∆n Ãn
∆n ,

75
Chapter 3. Abstract homotopy theory

where the vertical maps are those of Proposition 2.76, the left horizontal maps are
induced by c̃n and the right horizontal maps are the canonical inclusions.
• By Proposition 2.76, taking colimits vertically yields a diagram

skn B ←− skn A −→ A,

of which again we may take the colimit of.


• Similarly taking colimits horizontally yields a diagram

B̃n (c id)∪(id ε)
∆n ←− Ãn ∆n +Ãn ∂∆n B̃n
∂∆n −→ skn−1 B +skn−1 A A,

of which we may take the colimit of.


Now if I is the category 2 ← 1 → 0, then the big diagram above can be considered
as a functor in
CAT(I × I, sSet) = CAT(I, CAT(I, sSet)).
By the dual version of Remark 2.42 (ii) taking the colimit is a left adjoint and
so preserves colimits by Corollary 2.47. So the construction of colimits in functor
categories given in Proposition 2.43 tells us, that we end up with the same object,
no matter if we take the colimit vertically or horizontally first. This means
 
c id
n ( id)∪( ε)
colim B̃n n Ãn n
∆ ←− ∆ +Ãn ∂∆n ∂∆ B̃n
−→ skn−1 B +skn−1 A A ∼ = skn B+skn A A,

which is an equivalent formulation of the claim.


Now as cn is injective and L is part of a weak factorization system, Corollary 3.26
implies that the left vertical map in the cocartesian square is in L. Hence also the
right vertical map is in L, which by Lemma 3.13 (v) is closed under pushouts. Next
using that pushouts commute with colimits by Remark 2.42 (ii) again, we see that
ι
0 ∼
sk−1 B+sk−1 A A −→ colim (skn B+skn A A) −→ (colim skn B)+(colim sk A) A = B+A A,
n≥0 n≥0 n
n≥0

lies in L, because L is closed under sequential colimits by Lemma 3.13 (iii). But this
map is isomorphic to c, which completes the proof that cof sSet ⊂ L.

(ii) Let f ∈ sSet(C, D). Then applying Proposition 3.25 to the natural bijection induced
by the internal homomorphisms given in Proposition 2.54, statement (i) implies the
equivalence

f ∈ fib sSet
(di ×id)∪(id×c) c
⇐⇒ f ∈ RLP (∆0 × B +∆0 ×A ∆1 × A −→ ∆1 × B), ∀A −→ B, i = 0, 1
 
((di )∗ ,f∗ )
⇐⇒ sSet(∆1 , C) −→ sSet(∆0 , C) ×sSet(∆0 ,D) sSet(∆1 , D) ∈ RLP (cof sSet)

⇐⇒ ((di )∗ , f∗ ) ∈ RLP ({∂∆n ,−→ ∆n ; n ≥ 0}), ∀i = 0, 1


⇐⇒ f ∈ RLP ({∆ × ∆ +∆0 ×∂∆n ∆ × ∂∆ −→ ∆ × ∆n ; n ≥ 0, i = 0, 1}).
0 n 1 n 1

76
3.7. Towards the standard model structure on simplicial sets

Theorem 3.30
For c ∈ sSet(A, B) and c0 ∈ sSet(A0 , B 0 ), consider the map

c t c0 := (c × id) ∪ (id ∪ c0 ) : A × B 0 +A×A0 B × A0 −→ B × B 0 .

(i) If c, c0 ∈ cof sSet, then c t c0 ∈ cof sSet.

(ii) If moreover c ∈ LLP (fib sSet) or c0 ∈ LLP (fib sSet), then also ctc0 ∈ LLP (fib sSet).
Proof.

(i) As colimits and hence pushouts are computed dimensionwise in sSet, it suffices to
check that c, c0 ∈ Mono(Set) implies c00 ∈ Mono(Set). This is Corollary 3.26 applied
to the weak factorization system (Mono, Epi) on Set.

(ii) First note that by applying Proposition 3.25 to the natural bijections

sSet(A, sSet(B, C)) = sSet(A × B, C) = sSet(B, sSet(A, C)), (3.1)

constructed in Proposition 2.54 and using the Definition 3.27 of a fibration, we get
equivalences
c t c0 ∈ LLP (fib sSet) ∀ c ∈ cof sSet, c0 ∈ LLP (fib sSet)
⇐⇒ c t c0 ∈ LLP (f ) ∀ c ∈ cof sSet, c0 ∈ LLP (fib sSet), f ∈ fib sSet
⇐⇒ (c∗ , f∗ ) ∈ RLP (c0 ) ∀ c ∈ cof sSet, c0 ∈ LLP (fib sSet), f ∈ fib sSet
⇐⇒ (c∗ , f∗ ) ∈ fib sSet ∀ c ∈ cof sSet, f ∈ fib sSet
⇐⇒ (c∗ , f∗ ) ∈ RLP (di t c00 ) ∀ c, c00 ∈ cof sSet, f ∈ fib sSet, i = 0, 1
⇐⇒ f ∈ RLP ((di t c00 ) t c) ∀ c, c00 ∈ cof sSet, f ∈ fib sSet, i = 0, 1.
Once we have shown that (di t c00 ) t c ∼
= di t (c00 t c), the last statement is equivalent
to
f ∈ RLP (di t (c00 t c)) ∀ c, c00 ∈ cof sSet, f ∈ fib sSet, i = 0, 1,
which is true by Definition 3.27 of a fibration and since c00 t c ∈ cof sSet by (i).
So it remains to prove that (di t c00 ) t c ∼
= di t (c00 t c). We consider the commutative
diagram
∆0 × BO 00 × B o ∆0 × BO 00 × A ∆0 × B 00 × A

∆0 × A00 × B o ∆0 × A00 × A / ∆0 × B 00 × A

  
∆1 × A00 × B o ∆1 × A00 × A / ∆1 × B 00 × A,

77
Chapter 3. Abstract homotopy theory

• Taking colimits vertically and using that (- × X) by Proposition 2.54 as a left


adjoint preserves colimits we obtain the diagram

(∆0 ×B 00 +∆0 ×A00 ∆1 ×A00 )×B ←− (∆0 ×A00 +∆0 ×A00 ∆1 ×A00 )×A −→ (∆1 ×B 00 )×A,

whose colimit is the domain of (di t c00 ) t c.


• Similarly taking colimits horizontally yields the diagram

∆0 ×(B 00 ×B) ← ∆0 ×(A00 ×B +A00 ×A B 00 ×A) −→ ∆1 ×(A00 ×B +A00 ×A B 00 ×A),

whose colimit is the domain of di t (c00 t c).


As in the proof of Theorem 3.29, we see that both colimits coincide and thus
(di t c00 ) t c ∼
= di t (c00 t c) as desired.

Corollary 3.31
For c ∈ sSet(A, B) and f ∈ sSet(C, D), consider the map

(c∗ , f∗ ) : sSet(B, C) −→ sSet(A, C) ×sSet(A,D) sSet(B, D).

Then the following holds.

(i) If c ∈ cof sSet and f ∈ fib sSet, then (c∗ , f∗ ) ∈ fib sSet.

(ii) If moreover c ∈ LLP (fib sSet) or f ∈ RLP (cof sSet), then (c∗ , f∗ ) ∈ RLP (cof sSet).
Proof. By Proposition 3.25 (i) applied to (3.1) assertion (i) is equivalent to Theorem 3.30
(ii). The first case of (ii) is equivalent to Theorem 3.30 (ii) and the second one to Theorem
3.30 (i).
2

Corollary 3.32

fib sSet ⊃ RLP (cof sSet), LLP (fib sSet) ⊂ cof sSet.
di
Proof. As ∆0 −→ ∆1 is injective and hence a cofibration by Definition 3.27, for every
c
cofibration C −→ D also the map di t c is a cofibration by Theorem 3.30. In other
words di t c ∈ cof sSet = LLP (RLP (cof sSet)) by Theorem 3.29 (i) or equivalently
fib sSet ⊃ RLP (cof sSet) or LLP (fib sSet) ⊂ cof sSet.
2

Proposition 3.33
For every g ∈ sSet(X, Y ) the following holds.

(i) g strong deformation retraction and fibration ⇐⇒ g ∈ RLP (cof sSet).

78
3.7. Towards the standard model structure on simplicial sets

(ii) g strong deformation section ⇒ g ∈ LLP (fib sSet).


The other implication holds, if moreover X, Y −_ ∗.
Proof.
g
(i) Suppose X −_ Y is a fibration and a strong deformation retraction, i.e. there is
an s ∈ sSet(Y, X) with gs = idY and idX ' sg with gh = gπX . To every lifting
h
problem on the left, we get a commutative square on the right
(sv)∪(h(id×u))
A
u /7 X ∆0 × B +∆0 ×A ∆1 × A 4/ X

c
∃d g ∃h0 g
d0 tc
 _  _
B / Y, /
v ∆1 × B vπB Y.

Indeed the upper horizontal map is well-defined, because


gh(id × u)d0 = gπX (id, u) = gu = sv.
Moreover the right square commutes, because
g(sv) = v, gh(id × u) = gπX (id × u) = guπA = vcπA = vπY (id × c).
As c is a cofibration, by Definition 3.27 of a fibration, there is a diagonal h0 and
d := h0 d1 solves the original lifting problem:
gd = gh0 d1 = vπY d1 = v,
dc = h0 d1 c = h0 (id × c)d1 = h(id × u)d1 = hd1 u = u.

Vice versa for every g ∈ sSet(X, Y ) ∩ RLP (cof sSet) there are liftings
id∪(sg)
∅ / X ∂∆1 × X /7 X
9
  ∃h
∃s g g
 _  _
Y Y, ∆1 × X / Y,
gπX

proving that gs = idY and idC ' sg with gh = gπX = πY (id × g). So g is a strong
h
deformation retraction. Moreover g ∈ fib sSet by Theorem 3.30.
g
(ii) Suppose X −→ Y is a strong deformation section, i.e. there is an r ∈ sSet(Y, X)
with rg = idX and idY ' gr with h(id × g) = gπY . Then in particular g is a
h
monomorphism by Remark 2.32 and hence a cofibration by Definition 3.27. Then
like in (i), to every lifting problem on the left, we get a commutative square on the
right
(ur)∪(uπX )
X u /7 C ∆0 × Y +∆0 ×X ∆1 × X 4/ C

g ∃d
f ∃h0
d0 tg f
 _  _
Y / D, / D.
v ∆1 × Y vh

79
Chapter 3. Abstract homotopy theory

Then (ur) ∪ (uπX ) is well-defined, because

uπX d0 = u = urg.

Moreover the square commutes, because

f (ur) = vgr = vhd0 , f (uπX ) = vgπX = vh(id × g).

As g is a cofibration, by Definition 3.27 of a fibration, there is a diagonal h0 and


d := h0 d1 solves the original lifting problem:

f d = f h0 d1 = vhd1 = v,
dg = h0 d1 g = h0 (id × g)d1 = uπX d1 = u.

The other implication is completely dual to (i).

Remark 3.34
By Proposition 3.33 and Remark 3.28 we have

RLP (cof sSet) ⊂ fib sSet ∩ wsSet.

The hardest part in the construction of the model structure is to see that this is infact an
equality. As a consequence we will also get the equality LLP (fib sSet) = cof sSet ∩ wsSet,
which will conclude the proof that sSet is a model category.

For checking that a given map is a fibration, the following characterization is often the
most convenient one.

Proposition 3.35
[
fib sSet = RLP {Λnk := di ∆n−1 ,−→ ∆n ; 0 ≤ k ≤ n}.
0≤i≤n,
i6=k

Proof. First note, that Λnk = BAnk , where Ank ≤ I(n) is the ordered simplicial subcomplex
with same vertices and simplices

S(Ank ) = {s ∈ SI(n); i ∈
/ s for some i 6= k}.

Moreover given an ordered simplicial complex X ∈ Simpo with an ordered subcomplex


c
U ,−→ X, we define the ordered subcomplex

U (i) := (di I(0) × X) ∪ (I(1) × U ) ≤ I(1) × X, i = 0, 1.

Then we have

Bdi t Bc : ∆0 × X +∆0 ×BU ∆1 × BX = BU (i) ,−→ B(I(1) × X) = ∆1 × BX.

80
3.7. Towards the standard model structure on simplicial sets

Indeed as Bdi t Bc is injective by Theorem 3.30 the left object is isomorphic to its image
in ∆1 × BX, which is exactly BU (i) by construction.
Moreover by definition of the product complex the simplices of U (i) are the nonempty
subsets

t := {(0, x1 ), ..., (0, xj ), (1, xj+1 ), ..., (1, xm )} ⊂ I(1) × X, 0 ≤ j ≤ m,

such that

• in the case i = 0, we have πX (t) ∈ S(X) and πX (t) ∈ S(U ), whenever j < m.

• in the case i = 1, we have πX (t) ∈ S(X) and πX (t) ∈ S(U ), whenever j > 0.

Next we define commutative diagrams of ordered simplicial complexes


id

 &
Ank 
s| r|
/ Ank (i) // Ank
_  _ _

   
I(n)  / //
s r
I(1) × I(n) I(n).
8

id

• For 0 ≤ k < n, we let i = 1 and define s and r by



min{x, k}, a = 0,
s(x) := (1, x), r(a, x) :=
y, a = 1.

To see that r(Ank (1)) ⊂ Ank , note that


– for j < m, we have r(t) ⊂ πX (t) ∪ {k} ∈ S(Ank ), since πX (t) ∈ S(Ank ).
/ r(t) ∈ S(Ank ).
– for j = m, we have r(t) ⊂ k and hence n ∈

• Similarly, for 0 < k ≤ n, we let i = 0 and define s and r by



x, a = 0,
s(x) := (0, x), r(a, x) :=
max{x, k}, a = 1.

To see that r(Ank (0)) ⊂ Ank , note that


– for j > 0, we have r(t) ⊂ πX (t) ∪ {k} ∈ S(Ank ), since πX (t) ∈ S(Ank ).
/ r(t) ∈ S(Ank ).
– for j = 0, we have r(t) ⊂ {k, ..., n} and hence 0 ∈

Applying the nerve of the diagram yields (Λnk ,−→ ∆n ) ∈ LLP (fib sSet), for all 0 ≤ k ≤ n,
as the latter is closed under retracts by Lemma 3.13. Equivalently we have

RLP ({Λnk ,−→ ∆n ; 0 ≤ k ≤ n}) ⊃ RLP (fib sSet).

81
Chapter 3. Abstract homotopy theory

c
For the other inclusion consider the inclusion U := ∂I(n) ,−→ I(n). Then every simplex
in S(I(1) × I(n)) not contained in S(U (1)) is of the form

tk := {(0, 0), ..., (0, k − 1), (1, k), ...(1, n)}, t0k := tk ∪ {(0, k)}, 0 < k ≤ n.

Defining ∂I(n) ≤ U (1, k) ≤ I(1) × I(n) as the intermediate subcomplex with

S(U (1, k)) := S(U (1)) ∪ {t1 , t01 , ..., tk , t0k }, 0 ≤ k ≤ n,

we get isomorphisms of ordered simplicial complexes



∼ (0, j), j < k,
aj : An+1
k −→ U (1, k − 1) ∩ t0k , j 7−→
(1, j), j ≥ k.

So U (1, k) is obtained by glueing I(t0k ) ∼


= I(n) to U (1, k − 1) along Ank . In other words
there is a cocartesian square
Bak
Λnk / BU (1, k − 1)
_

 
∆n / BU (1, k).

It follows that

(d1 t Bc) : BU (1) = BU (1, 0) ,−→ ... ,−→ BU (1, n) = B(I(1) × I(n))
[
is contained in LLP (RLP ({Λnk := di ∆n−1 ,−→ ∆n ; 0 ≤ k ≤ n}), which by Lemma
0≤i≤n,
i6=k
3.13 is closed under pushouts and composition.
By similar arguments we see that it also contains d0 t Bc, which by using (ii) proves
that
RLP ({Λnk ,−→ ∆n ; 0 ≤ k ≤ n}) ⊂ RLP (fib sSet).

3.8 Absolute weak equivalences


For the key step in the construction of the model structure we will need another class of
morphisms, which behave very well under many operations. Infact it will turn out that
absolute weak equivalences are precisely the weak equivalences. However we will not know
this until the construction of the model structure, for which the two notions are needed
for.

82
3.8. Absolute weak equivalences

Definition 3.36
An absolute weak equivalence is an a ∈ sSet(A, B), such that for every commutative
square as on the left, there is a diagonal in the induced right one
A
u / X A
u
4/ X
∃d∪h
a f a0 :=d1 f
 _  _
B / Y, M (a) = B +A (∆1 × A) / Y.
v v∪(f uπA )

In other words, there is a diagonal d ∈ sSet(B, X) in the square


A
u /
9X
'
h
a f
 ∃d _
B / Y,
v

such that the lower triangle strictly commutes and the upper triangle commutes up to a
simplicial homotopy u ' da satisfying f h = f uπA .
h

Proposition 3.37
Every absolute weak equivalence is a weak equivalence.
t
Proof. Let a ∈ sSet(A, B) be an absolute weak equivalence and T −_ ∗.
• For every u ∈ sSet(A, T ), the left square below commutes. So by definition there is
a diagonal in the right square
A u / T A u /8 T
d∪h
a t a0 t
 _  _
B / ∗, M (a) / ∗.
v v∪(f uπA )

It follows that u ' da, which proves surjectivity


h

a∗ = (- ◦ a) : π0 sSet(B, T ) − π0 sSet(A, T ).

• For all v0 , v1 ∈ sSet(B, T ) with v1 a ' v0 a, we let h0 ∈ sSet(A, sSet(∆1 , T )) be the


h
map corresponding to h under the natural bijection of Proposition 2.54
sSet(∆1 × A, T ) = sSet(A, sSet(∆1 , T )).
Then the left square below commutes. So by definition there is a diagonal in the
right square
h0 / h0 /
A sSet(∆1 , T ) A sSet(∆1 , T )
6
k0 ∪H
a a0
 _  _
B / T × T = sSet(∂∆ , T ) 1
M (a) / sSet(∂∆1 , T ).
v:=(v0 ,v1 ) v∪(f uπA )

It follows that v1 ' v0 , where k ∈ sSet(∆1 × B, T ) is the map corresponding to k 0 .


k

83
Chapter 3. Abstract homotopy theory

The subsequent proposition shows that the class of absolute weak equivalences satisfies
some weakened 2-of-3 property.

Proposition 3.38
a b
For two maps of simplicial sets A −→ B −→ C the following implications hold.
(i) a, ba absolute weak equivalences ⇒ b absolute weak equivalence.
(ii) a, b absolute weak equivalences ⇒ ba absolute weak equivalence.
Proof.
(i) For every commutative diagram as on the left, there is a diagonal in the right square,
as ba is an absolute weak equivalence by assumption.

A
a / B
u / X A
u /7 X
∃d∪h
b f (ba)0 f
_ _
%  
ba
C / Y. M (ba) / Y.
v
v∪(f uπA )

In particular we have dba ' ua. Like in the proof of Proposition 3.37 (ii) we construct
h
a simplicial homotopy db ' u. To that aim consider the commutative square
k

h0 / sSet(∆1 , X)
A
a (ε∗ ,f∗ )
 _
B / sSet(∂∆1 , X) ×sSet(∂∆1 ,Y ) sSet(∆1 , Y ),
V :=((db,u),s0 v)

where the right vertical map is a fibration by Corollary 3.31 and the map h0 corre-
sponds to the homotopy h under the natural bijection of Proposition 2.54. So using
that also a is an absolute weak equivalence we obtain a diagonal in the square
h0 /
A 3
sSet(∆1 , X)
k0 ∪H
a0 (ε∗ ,f∗ )
 _
M (a) / sSet(∂∆1 , X) × sSet(∆1 , Y ),
V ∪(f uπA ) sSet(∂∆1 ,Y )

and we define k as the simplicial homotopy corresponding to k 0 . By construction


d ∪ k defines a diagonal in the square

B
u /
<X
d∪k
b0 f
 _
M (b) / Y,
v

which finishes the proof, that b is an absolute weak equivalence.

84
3.8. Absolute weak equivalences

(ii) For every commutative square as on the left, there are diagonals in the middle and
on the right

A u / X A u /8 X B d /8 X
a ∃d∪h ∃d0 ∪h0
a0 f b0 f
  _  _
B f M (a) / Y M (b) / Y.
(vb)∪(f uπA ) v∪(f uπA )
b
 _
C / Y
v

In particular we have uda ' d0 ba and like in the proof for transitivity of the
h h0 (id×a)
simplicial homotopy relation of Proposition 3.46, we obtain a composite simplicial
homotopy u ' db0 a defining a diagonal in the square
k

7/ X
u
A
d∪k
(ab)0 f
 _
M (ba) / Y.
v

Proposition 3.39
Absolute weak equivalences are stable under pushouts along cofibrations.
Proof. Given a commutative square as on the left

A / / /
c u
C X
a ιC f
  _
B /B +A C / Y,
v∪w

where a is an absolute weak equivalence, we have to prove that also ιC is an absolute weak
equivalence. So given a commutative square as on the right, we use that a is an absolute
weak equivalence to obtain a diagonal in the commutative square

A uc /8 X
∃d∪h
a0 f
 _
M (a) v∪(f ucπ / Y.
A)

By construction of h we get a well-defined map (uc) ∪ h rendering the diagram below


commutative
(uc)∪h
M (c) 7/ X
∃k
d1 tc f
 _
∆1 × C / Y.
wπC

85
Chapter 3. Abstract homotopy theory

As c is a cofibration, we have d1 t c ∈ LLP (fib sSet) by Theorem 3.30 (ii). Hence there is
a diagonal k as depicted. As the first two small squares below are cocartesian by definition

c / d1 / ∆1 d1 / id×c /
A C ×C A ∆1 × A ∆1 × C
a ιC a
  (ιC )0    
B
ιB
/B +A C / M (ιC ), B a0 / M (a) ιB +(id×c) / M (ιC ),

so is their composite, which coincides with the composite of the other two squares. As
also the third small square is cocartesian, the same holds for the fourth small square. So
by construction D := (d ∪ h) ∪ k defines a diagonal in the diagram below

5/
u
C X
D
(ιC )0 f
 _
M (ιC ) / Y,
(v∪w)∪(f uπB+A C )

which concludes the proof that ιC is an absolute weak equivalence.


2

From the following technically complicated Lemma we will derive more useful properties
of absolute weak equivalences.

Lemma 3.40
Given a commutative diagram
A / c / B
a b
 
c0
A0 / / B0,
where a and b are absolute weak equivalences.
Then for every commutative square as on the left and every diagonal d ∪ h, there is is
a diagonal d00 ∪ h00 rendering the right diagram commutative
u / c /B u /
B X A :4 X
b0
b f a0 f
d∪h ∃d00 ∪h00
 _   _
B0 / Y, M (a) / M (b) / Y,
v C:=c0 +(id×c) v∪(f uπB )

Proof. As b is an absolute weak equivalence, there is a diagonal in

B
u /8 X
∃d0 ∪h0
b0 f
 _
M (b) / Y.
v∪(f uπB )

86
3.8. Absolute weak equivalences

However this does not guarantee that (d0 ∪ h0 ) ◦ C = d ∪ h. The strategy is to construct
a deformation d00 ∪ h00 of d0 ∪ h0 restricting to d ∪ h along C. The construction of d00 ∪ h00
is quite technical and requires to apply the lifting property 5 more times. For i = 0, 1 we
define the maps
ιi := ε t di : Ui := ∂∆1 × ∆1 +∂∆1 ×∆0 ∆1 × ∆0 ,−→ ∆1 × ∆1 ,
which for i = 0 e.g. corresponds to the inclusion
(0,O 1) (1,O 1) (0,O 1) / (1,O 1) (3.2)
 / “solid”

(0, 0) / (1, 0) (0, 0) / (1, 0).

In particular ιi maps Ui isomorphically onto the simplicial subset


(d1 ∆0 × ∆1 ) ∪ (d0 ∆0 × ∆1 ) ∪ (∆1 × di ∆1 ) ∼
= ∆1 ∪ ∆1 ∪ ∆1 . (3.3)
Recall that by Theorem 3.30 (ii) we have ιi ∈ LLP (fib sSet).
1) In the commutative diagram
h∪(h0 (id×c))∪(ucπA )
U1 × A /5 X
∃H
ι1 ×id f
 _
∆1 × ∆1 × A / Y,
vc0 aπA =f ucπA

the upper horizontal map is defined using the description (3.3). Since ιi ∈ LLP (fib sSet),
we also have ιi × idA ∈ LLP (fib sSet) by Theorem 3.30 applied to the cofibration
∅−→ A. So there is a diagonal H as depicted.
To clarify the situation we will introduce another representation for the lifting prob-
lem. Using the picture (3.2) we may represent the left vertical map of the diagram
by the picture
AO AO AO
A /A
O
 /
A A A A A

A / A A / A.
A A

Moreover we may represent the upper horizontal map by the left diagram below and
the lifting property translates as there is an extension H represented by the right
diagram

da d0 bc =O d0 c0 a da k:= / d 0 c0 a
O O O
 /
h h0 (id×c) h ∃H h0 (id×c)

uc / uc uc / uc,
ucπA ucπA

87
Chapter 3. Abstract homotopy theory

2) Next we use that a is an absolute weak equivalence for the commutative square

A k / sSet(∆1 , X)
a (ε∗ ,f∗ )
 _
A0 / sSet(∂∆1 , X) ×sSet(∂∆1 ,Y ) sSet(∆1 , Y )
((d,d0 c0 ),s0 vc0 )

to get a diagonal k 0 ∪K in the corresponding square (use Remark 3.24 for Proposition
2.54)
(d∪(d0 c0 ))∪k
∂∆1 × M (a) +∂∆1 ×A ∆1 × A 2/ X
k0 ∪K
a f
 _
∆1 × M (a) / M (a) / Y.
πM (a) (vc0 )∪(f kπA )

Translating into our notation the left vertical map corresponds to the picture
A0 /
AO 0 AO 0 AO 0 AO 0
 /
A A A A A

A / A A / A,
A A

and the lifting property can be described by the existence of an extension


∃k0 / d 0 c0
dO d0O c0 dO O
 /
daπA d0 c0 aπA daπA ∃K d0 c0 aπA

da / d 0 c0 a da / d0 c0 a.
k k

3) Using that c0 is a cofibration and the Definition 3.27 of a fibration there is a diagonal
in the commutative square
d0 ∪k0
∆0 × B 0 +∆0 ×A0 ∆1 × A0 4/ X
∃k00
d1 tc0 f
 _
∆1 × B 0 / Y.
vπB 0

Again the left vertical map is represented by


A0 / B0  / B0 /
A0 B0 B0,

and the lifting property by


k0 / d0  / ∃k00 / d0 .
d d00 := k 00 d1

88
3.8. Absolute weak equivalences

4) Using that c is a cofibration, so is the following map by Theorem 3.30 again

BO
B / BO BO
B / BO
 /
B A B B B B

B / B B / B,
A B

and thus there is an extension


k00 (id×b) k00 (id×b)
d00O b / d0 b d00O b / d0 b
O O
 /
d00 bπB K d0 bπB d00 bπB ∃L dbπB

d00 b / d0 b d00 b / d0 b.
k `:=

Here it should be noted that the left map is well-defined, because by construction
• k 00 (id × b)(id × c) = k 00 (id × c0 )(id × a) = k 0 (id × a) = K(id × d0 ),
• d00 bπB (id × c) = d00 bcπA = d00 c0 aπA = daπA = K(d1 × id),
• d0 bπB (id × c) = d0 bcπA = d0 c0 aπA = K(d0 × id).
5) Finally, using that c is a cofibration again, so is the following map by Theorem 3.30
again
BO
B /B BO
B /B
O O
 /
A A B B B B

B / B B / B,
B B

and thus there is an extension

d00O b
` / d0 b d00O b
` / d0 b
O O
 /
d00 bπB H h0 h00 := ∃M h0

u /u u / u.
uπB uπB

Again the left map is well-defined, because by construction we have


• `(id × c) = k = H(id × d0 ),
• uπB (id × c) = H(id × d1 ),
• h0 (id × c) = J(d0 × id).
By construction d00 ∪ h00 has the desired property.
2
The next two properties are essential for showing that a certain natural map between
two functors targetting to sSet is an absolute weak equivalence.

89
Chapter 3. Abstract homotopy theory

Theorem 3.41 (i) For every commutative diagram of simplicial sets


g
Y o o /
c
X Z
y x z
  g0

c0
Y0 o o X0 / Z0
if x, y, z are absolute weak equivalences, then taking the colimit horizontally induces
an absolute weak equivalence
y +x z : Y +X Z −→ Y +X 0 Z 0 .

(ii) For every map of sequences of cofibrations of simplicial sets


c0
X0 / / X1 / c1 / ...

x0 x1
 c00 c1  0
X00 / / X10 / / ...

if xn is an absolute weak equivalence, for all n ≥ 0, then taking the colimit horizon-
tally induces an absolute weak equivalence
x := colim xn : X := colim Xn −→ colim Xn0 =: X 0 .
n≥0 n≥0 n≥0

Proof.
(i) For every commutative square as on the left, there is a diagonal DZ on the right,
since z is an absolute weak equivalence
g
Y +X Z t∪u / C X / Z u /8 C
∃DZ
y+x z f x0 z0 f
 _   _
Y 0 +X 0 Z 0 / D, M (x) / M (z) / D.
v∪w G:=g 0 +(id×g) w∪(f uπZ )

Setting DX := DZ G, we also get a diagonal for the outer square on the right. By
Lemma 3.40 we also obtain a diagonal DY in the diagram
X
c / Y
u /49 C
y0
x0 DX
f
∃DY
  _
M (x) / M (y) / D,
c0 +(id×c) v∪(f uπY )

So by construction DY ∪ DZ defines a diagonal for the square

Y +X Z t∪u /4 C
DY ∪DZ
(y+x z)0 f
 _
M (y +x z) / D,
(v∪w)∪(f (t∪u)πY +X Z )

which proves that y +x z is an absolute weak equivalence.

90
3.8. Absolute weak equivalences

(ii) For every commutative square as on the left, there is a diagonal D0 on the right,
since x0 is an absolute weak equivalence
uι0
X
u / C X0 5/ C
∃D0
x f x00 f
 _  _
X0 / D, M (x0 ) / D.
v (vι0 )∪(f uι0 πX0 )

By induction on n ≥ 0, Lemma 3.40 provides a diagonal Dn+1 in the diagram


cn uιn+1
Xn / Xn+1 5/2 C
x0n+1
x0n f
Dn ∃Dn+1
  _
M (xn ) / M (xn+1 ) / D,
c0n +(id×cn ) (vιn+1 )∪(f uιn+1 πXn+1 )

Then by construction colim n≥0 Dn defines a diagonal in the diagram

X u /4 C
colim n Dn
x0 f
 _
M (x) / D,
v∪(f uπX )

proving that x is an absolute weak equivalence.

Proposition 3.42
The class wa sSet of absolute weak equivalences has the following properties.

(i) wa sSet is closed under pushouts along cofibrations.

(ii) wa sSet is closed under arbitrary coproducts.

Absolute weak equivalences are stable under pushouts along cofibrations.


Proof.
c a
(i) Given a sequence of maps C ←− A −→ B, where a is an absolute weak equivalence.
Then applying Theorem 3.41 (i) to the commutative diagram

Co c o A A
a

Co o /
c a
A B
C ι
yields that also C −→ B +A C is an absolute weak equivalence.

91
Chapter 3. Abstract homotopy theory

i a
(ii) Given a family of absolute weak equivalences (Ai −→ Bi )i∈I indexed by some set
I ∈ Set. Setting a a a
a := ai : A := Ai −→ Bi =: B,
i∈I i i

and using that coproducts commute pushouts, we get a natural isomorphism


`
i Ai A
a0i
`
i a0
 
∼ / M (a).
`
i M (ai )

So for every commutative square as on the left, there are diagonals in the middle
square, which glue to a diagonal in the right square
uιi
A
u / C Ai 6/ C A
u
6/ C
`
∃Di D:= i Di
a f a0i f a f
 _  _  _
B /D / M (a) / D
v M (ai ) (vιi )∪(f uιi πAi )
D (vιi )∪(f uιi πAi )

Proposition 3.43
Let wa sSet denote the class of absolute weak equivalences. Then
LLP (fib sSet) = cof sSet ∩ wa sSet.
Proof. Let c ∈ sSet(A, B) ∩ LLP (fib sSet). Then for every fibration f ∈ sSet(C, D) there
is a diagonal in the left square, providing a diagonal in the right square

A
u /9 C A
u /8 C
 d∪(uπA )
∃d
c f c0 f
 _  _
B / D, M (c) / D,
v v∪(f uπX )

proving that c is an absolute weak equivalence. Moreover c ∈ cof sSet by Corollary 3.32,
so we get the inclusion
LLP (fib sSet) ⊂ cof sSet ∩ wa sSet. (3.4)
c
Vice versa suppose X −→ Y is an absolute weak equivalence. As (LLP (fib sSet), fib sSet)
c0 f
is a weak factorization system, we may factor c as X −→ X 0 −_ Y , where c0 ∈
LLP (fib sSet) and f ∈ fib sSet. Proposition 3.38 implies that f is an absolute weak
equivalence, because c0 is one by (3.4) and f c0 = c is one by assumption. Solving the
lifting problem
X0 8X
0
'
h
f f
 ∃s _
Y Y

92
3.8. Absolute weak equivalences

proves that f is a strong deformation retraction and hence f ∈ RLP (cof sSet) by Propo-
sition 3.33. So we can also find a diagonal in

c0 / 0
X X
>

∃d
c f
 
Y Y,

which shows that c is a retract of c0 and hence c ∈ LLP (fib sSet), because the latter is
closed under retracts by Lemma 3.13.
2

If we knew that every absolute weak equivalence was a weak equivalence Proposition
3.43 would comparably easily imply that sSet is a model category with the three classes
of maps.

Corollary 3.44
f g
For X −→ Y −→ Z the following implication holds.

f, gf ∈ LLP (fib sSet), g ∈ cof sSet ⇒ g ∈ LLP (fib sSet).


Proof. By Proposition 3.43 the maps f and gf are absolute weak equivalences, hence
g is an absolute weak equivalence by Propostion 3.38. So g ∈ cof sSet ∩ wa sSet =
LLP (fib sSet) by using Proposition 3.43 again.
2

Combined with Proposition 3.43 the subsequent lemma will be used later to show
that the endofunctor of barycentric subdivision on the category sSet preserves anodyne
extensions.

Lemma 3.45
Let C, D ∈ Cat be small categories with a terminal object. Then:
f
(i) The nerve of every functor 0 −→ C is an anodyne extension.
f
(ii) The nerve of every functor C −→ D is an absolute weak equivalence.
Proof.
g
(i) Let 0 −→ C be the functor sending 0 to the terminal object ∗ ∈ C. Moreover let
r
C −→ 0 be the unique functor sending every object to 0. Then rg = id0 and by the
universal property of the terminal object there are unique C-morphisms
ηx
x −→ ∗ = gr(x), x ∈ C.

In other words the functors r and g form an adjunction

0(r(x), y) = C(x, g(y)).

93
Chapter 3. Abstract homotopy theory

So Proposition 2.89 implies that Bg and Br form a simplicial homotopy equivalence.


As 0 consists of a single object and morphism, it follows that Bg is infact a strong
deformation section and thus Bg ∈ LLP (fib sSet) by Proposition 3.43. This proves
the assertion, for f = g.
If f 6= g we define the functor

h : 1 −→ C, 0 7−→ f (0), 1 7−→ ∗,

sending the map 0 −→ 1 to the unique map ηf (0) . Taking nerves we get a commu-
tative diagram
∆0 B0 Bf

d1 Bd1
  
∆O 1 B(1)
Bh / BC.
M
O
d0 Bd0

∆0 B0 Bg

Since f 6= g, the map Bh is injective and hence a cofibration by Definition 3.27. So


Bg, d0 ∈ LLP (fib sSet) implies Bh ∈ LLP (fib sSet) by Corollary 3.44. Moreover
d1 , Bh ∈ LLP (fib sSet) implies Bf ∈ LLP (fib sSet) by Lemma 3.13.
g
(ii) Choose an arbitrary functor 0 −→ C, e.g. the functor g constructed in the proof of
(i). Then the functors g and f g satisfy the hypothesis in (i) and so Bg, Bf ◦ Bg ∈
LLP (fib sSet) ⊂ wa sSet by Propostion 3.43. So the 2-of-3 property for absolute
weak equivalences proven in Proposition 3.38 implies that also Bf is an absolute
weak equivalence.

3.9 Maps with fibrant codomain


In general simplicial homotopy does not define an equivalence relation on the set of ho-
momorphisms between two simplicial sets. But it does once the target simplicial set is
fibrant, as the following proposition demonstrates.

Proposition 3.46
t
Let Y −_ T and e, f, g ∈ sSet(X, Y ) with s = te = tf = tg.

(i) f ' f ,
f πX

(ii) f ' g, th = sπX ⇒ g '0 f, sh0 = sπX ,


h h

(iii) e ' f ' g, sh = sk = sπX ⇒ e ' g, s` = sπX .


h k `

94
3.9. Maps with fibrant codomain

In particular simplicial homotopy “'” defines an equivalence relation on sSet(X, Y ), if


Y −_ ∗.
Proof. (i) holds by construction, because f πX di = f , for i = 0, 1. For the rest of the proof
we will again use the anodyne extension
ιi := ε t di : Ui := ∂∆1 × ∆1 +∂∆1 ×∆0 ∆1 × ∆0 ,−→ ∆1 × ∆1
and the description

ιi : Ui −→ (d1 ∆0 × ∆1 ) ∪ (d0 ∆0 × ∆1 ) ∪ (∆1 × di ∆1 ) ∼
= ∆1 ∪ ∆1 ∪ ∆1
like in the proof of Lemma 3.40.
Under the hypothesis of (ii) there is a solution for the lifting problem
(gπX )∪h∪(gπX )
Ui × X 5/ Y
∃H
ι0 ×id t
 _
∆1 × ∆1 × X / T,
sπX

which in the notation of Lemma 3.40 corresponds to the picture


gπX gπX
gO / gO gO / gO
 /
gπX h gπX ∃H h

g f g / f,
h0 :=

Applying Theorem 3.30 for the cofibration ∅−→ X implies that ι0 × id ∈ LLP (fib sSet)
and hence there is a diagonal H. By construction we get a simplicial homotopy g '0 f .
h
Similarly for e ' f ' g there is a diagonal in the commutative square
h k

h∪(gπX )∪k
Ui × X /4 Y
∃H
ι0 ×id t
 _
∆1 × ∆1 × X / T.
sπX

which in the notation of Lemma 3.40 corresponds to the picture

fO
k /g fO
k /g
O O
 /
h gπX h ∃H gπX

e g e / g,
`:=

and by construction e ' g.


` 2
Transitivity and symmetric of the simplicial homotopy relation is not the only property
that holds, when the target is fibrant.

95
Chapter 3. Abstract homotopy theory

Proposition 3.47
f
For X −_ Y −_ ∗ the following are equivalent.

(i) f is a weak equivalence.

(ii) f is a homotopy equivalence.

(iii) f is a strong deformation retraction.


Proof. We begin by assuming (i), i.e. f is a weak equivalence.

• Since Y −_ ∗, the map π0 sSet(X, Y ) − sSet(Y, Y ) is surjective. So there is a


g
Y −→ X with idX ' gf .
h

• Since X −_ ∗, the map π0 sSet(X, X) ,−→ sSet(Y, X) is injective. So f ' f gf


fh
implies idY ' f g.
k

This proves, that f is a homotopy equivalence.


g
Next we assume (ii), i.e. that there is a map Y −→ X and simplicial homotopies
idX ' gf and idY ' f g. Using that ∅−→ Y is a cofibration there is a solution for the
h k
lifting problem
g
∆0 × Y /
;X
∃k0
d0 ×id f
 _
∆1 × Y / Y.
k

Setting s := k 0 d1 we get

f s = f k 0 d1 = kd1 = idY , idX ' gf, sf ' gf.


h k0 (id×f )

So by Proposition 3.46 there is a simplicial homotopy idX '0 sf . Using that ∅−→ X is a
h
cofibration there is a solution for the lifting problem

h0 ∪(sf h0 )∪(sf πX )
U0 × X 5/ X
∃H
ι0 ×id f
 _
∆1 × ∆1 × X / Y,
f πX (id×h0 )

which in the notation of Lemma 3.40 corresponds to the picture


sf πX sf πX
sfO / sfO sfO / sf
O
 /
h0 sf h0 h0 ∃H sf h0

idX sf idY / sf.


h00 :=

96
3.9. Maps with fibrant codomain

By construction we have

idY '00 sf, f h0 = f H(id × d1 ) = f πX (id × (h0 d1 )) = f πX ,


h

so f is a strong deformation retraction.


Finally every simplicial homotopy equivalence is a weak equivalence by Remark 3.28,
which shows that (iii) implies (i).
2

If we knew the preceding proposition for arbitrary fibrations (without assuming that
the codomain is fibrant), this would immediately give the desired model structure on
sSet. Indeed it yields the following characterization for anodyne extensions with fibrant
codomain, which we (for general codomains) could only verify (yet) for absolute weak
equivalences instead of weak equivalences in Proposition 3.33. The key idea in the veri-
fication of the model structure on sSet is to find a well-behaved method of replacing a
map by one having a fibrant codomain.

Corollary 3.48
The following implications are valid.

(i) If c ∈ cof sSet ∩ wsSet has a fibrant codomain, then c ∈ LLP (fib sSet).

(ii) if f ∈ fib sSet has a fibrant codomain, then f ∈ RLP (cof sSet ∩ wsSet).
Proof.

(i) The proof is similar to that of Proposition 3.43. Using that (LLP (fib sSet), fib sSet)
c0 f
is a weak factorization system, we may factor c as X −→ X 0 −_ Y , where c0 ∈
LLP (fib sSet) and f ∈ fib sSet. Then f ∈ wsSet by the 2-of-3 axiom, because
f c0 = c ∈ wsSet by assumption and c0 ∈ wa sSet ⊂ wsSet by Proposition 3.43 and
Proposition 3.37. So f is a strong deformation retraction by Proposition 3.47 and
hence f ∈ RLP (cof sSet) by Proposition 3.33. Again we can find a diagonal in

c0 / 0
X X
>

∃d
c f
 
Y Y,

which shows that c is a retract of c0 and hence c ∈ LLP (fib sSet), because the latter
is closed under retracts by Lemma 3.13.

(ii) Given a lifting problem

A
u / X
9

∃d
c ' f
 _
v /  ,2
B Y ∗,

97
Chapter 3. Abstract homotopy theory

c0
we may factor B −→ ∗ as B −→ B 0 −_ ∗ with c0 ∈ LLP (fib sSet). So there is a
diagonal in the square
v /
B >Y
∃v 0
c0
 _
B0 / ∗.

Moreover c0 c ∈ cof sSet ∩ wsSet and hence c0 c ∈ LLP (fib sSet) by (i), because
B 0 −_ ∗. So we can find a diagonal in the square

A
u /X
>
∃d0
c0 c f
 _
v0 /
B0 Y.
Then d := d0 c0 solves the first lifting problem, because by construction

f d = f d0 c0 = v 0 c0 = v, dc = d0 c0 c = u.

We also need the following “fibrerd version” of the dual of Proposition 3.47.

Proposition 3.49
Given a t-fibered homotopy equivalence

X / /
c
Y

y >
tc t
%
T
d
meaning that there is a map Y −→ X satisfying

tcd = t, idX ' dc, tch = tcπX , idY ' cd, tk = tπY .
h k

Then c is a strong deformation section.


Proof. Ignoring the fact that in general we have T 6= ∗, the arguments of the proof are
exactly the same as the proof of the implication (ii) ⇒ (iii) in Proposition 3.47. First
there is a solution for the lifting problem

∆0 × Y +∆0 ×X ∆1 × X
d∪k /6 X
∃k0
d0 tc tc
 _
∆1 × Y / T.
tπY

Setting r := k 0 d1 we get

rc = k 0 d1 c = k 0 (id × c)d1 = kd1 = idX , idY ' cd, cr '0 cd.


h ck

98
3.10. Verifying the model structure on simplicial sets using Kan’s functor Ex∞

Using the hypothesis on h and k Proposition 3.46 yields a homotopy


idY '0 cd, th0 = tπY .
h

Moreover there is a diagonal H in


(h0 ∪(h0 (id×(cr)))∪crπY )∪(πY (id×h0 )(id×id×c))
U0 × Y +U0 ×X ∆1 × ∆1 × X 1/ Y
∃H
ι0 tc t
 _
∆1 × ∆1 × Y /T
tπY

corresponding to the picture


crπX crπY
crO / crO crO / crO
 /
h0 crh0 h0 ∃H crh0

idY cr idY / cr.


h00 :=

By construction we have
idX '00 cr, h00 (id × c) = cπX ,
h

meaning that c is a strong deformation section.


2

3.10 Verifying the model structure on simplicial


sets using Kan’s functor Ex∞
As mentioned before, for the key step in the construction of the model structure on sSet,
'
we need a well-behaved functorial fibrant replacement X −→ Q(X) −_ ∗, for X ∈ sSet.
One example of such a functor is Kan’s functor Ex∞ , which is closely related to barycentric
subdivision.

Definition 3.50
The barycentric subdivision sdB X of a simplicial set X ∈ sSet is defined as
sdB X := colim BsdB I(m),
m∈∆/X

where BsdB I(n) is the nerve of the barycentric subdivision of the simplicial complex I(n).

Remark 3.51
Using Yoneda’s Lemma 2.11, we get an adjunction
sSet(sdB X, Y ) = sSet(X, ExY ),
where ExY := sSet(sdB ƥ , Y ).

99
Chapter 3. Abstract homotopy theory

As the following proposition shows, this extends the notion of the barycentric subdivi-
sion for simplicial complexes of Definition 1.13.

Proposition 3.52
There is a natural isomorphism

fC := colim BsdB (εm ) : sdB BC = colim BsdB I(m) −→ B(sdB C), C ∈ Simpo .
m∈∆/BC m∈∆/BC

Proof. Suppose first, that n := ]C < ∞. Then C = s1 ∪ ... ∪ sm with si ∈ S(C) maximal.
We will prove that fC is an isomorphism by induction on m ≥ 1.

• For m = 1 there is a (unique) isomorphism g : I(n) −→ C. In particular g ∈ ∆/BC
forms a final object, so

ιg : BsdB I(n) −→ colim BsdB (εm ) = sdB BC.
m∈∆/BC

As also fC ιg = BsdB (g) is an isomorphism, it follows that fC is an isomorphism.

• For m > 1 define the ordered subcomplex C 0 := s1 ∪ ... ∪ sm−1 ≤ C. Then there is
are canonical cartesian squares of ordered simplicial complexes
 
C 0 ∩ _ sm  / sm
_ sdB C 0  ∩
_ sm
 / sdB s _ m

     
C0 / C sdB C 0  / sdB C.
B
As the nerve functor Simpo −→ sSet preserves limits, we get induced cartesian
squares of simplicial sets
 
B(C 0 ∩ _ sm )  / Bsm
_ BsdB (C 0_ ∩ sm )  / BsdB _ sm

     
BC 0  / BC. BsdB C 0  / BsdB C.

Using C = C 0 ∪ sm and hence sdB C = sdB C 0 ∪ sdB sm , we get BC = BC 0 ∪ Bsm and


BsdB C = BsdB C 0 ∪ BsdB sm . It follows that the two squares are also cocartesian.
sdB
Using that sSet −→ sSet is a left adjoint by Remark 3.51, it commutes with colimits
by Corollary 2.47, so also sdB applied to the left square is a pushout square. By the
induction hypothesis for C 0 = s1 ∪...∪sm−1 and C 0 ∩sm = (s1 ∩sm )∪...∪(sm−1 ∩sm ),
the vertical maps in the diagram

sdB BC 0 o sdB B(C 0 ∩ sm ) / sdB Bsm

fC 0 o fC 0 ∩sm o fsm o
  
BsdB C 0 o BsdB (C 0 ∩ sm ) / BsdB sm

are isomorphisms. So taking the colimit horizontally yields the isomorphism fC :



sdB BC −→ BsdB C.

100
3.10. Verifying the model structure on simplicial sets using Kan’s functor Ex∞

For ]C = ∞ there is a comutative square


colim F fF
colim sdB BF / colim BsdB F

F ≤C, F ≤C,
finite finite
o o
 fC 
sdB BC / BsdB C,

where the vertical maps are isomorphisms, because sdB and B commute with the filtered
colimit, which is just a union in this case. As the square commutes, fC must be an
isomorphism.
2

Corollary 3.53
The natural homeomorphism of Theorem 1.17

hC : |sdB C| −→ |C|, C ∈ Simpo
extends to a natural homeomorphism
hC
|sdB C| / |C|
∼ O
o
colim hI(m)
colim |BsdB I(n)| / colim |BI(n)|,

m∈∆/C m∈∆/C

where the isomorphism on the right is induced by the canonical one from the co-Yoneda
Lemma 2.51.

Having developed the language of category, it is easy to see that the barycentric subdi-
vision functor on the category of simplicial complexes factors over the category of partially
ordered sets:

Remark 3.54
Every partially ordered set (P, ≤) induces an ordered simplicial complex T (P ) with vertices
P and simplices
ST (P ) = {s ⊂ P ; 0 < ]s < ∞, “≤” restricts to a total order on s}.
Then by construction, we have
(i) BP = Cat(-, P ) = Simp(-, T (P )) = BT (P ),
(ii) sdB X = T S(X), for all X ∈ Simp.
In particular we have
sdB X := colim BsdB I(n) = colim BSI(n),
m∈∆/X m∈∆/X

where BSI(n) is the nerve of the partially ordered set of simplices in I(n).

101
Chapter 3. Abstract homotopy theory

Using that partially ordered sets can be considered as categories, for which we can
more easily construct homotopy equivalences by using Proposition 2.89, the preceding
characterization of the subdivision functor is quite useful for actual proofs.

Proposition 3.55
The following holds.
(i) sdB cof sSet ⊂ cof sSet, Ex(RLP (cof sSet)) ⊂ RLP (cof sSet),
(ii) Ex(fib sSet) ⊂ fib sSet.
Proof.
(i) There are equivalences
sdB c ∈ cof sSet, ∀ c ∈ cof sSet
⇐⇒ sdB c ∈ LLP (f ), ∀ c ∈ cof sSet, f ∈ RLP (cof sSet)
⇐⇒ Exf ∈ RLP (c), ∀ c ∈ cof sSet, f ∈ RLP (cof sSet)
ε
⇐⇒ Exf ∈ RLP (∂∆n ,−→ ∆n ), ∀ f ∈ RLP (cof sSet), n≥0
ε
⇐⇒ sdB (∂∆n ,−→ ∆n ) ∈ LLP (f ), ∀ f ∈ RLP (cof sSet), n≥0
ε
⇐⇒ sdB (∂∆n ,−→ ∆n ) ∈ cof sSet, ∀ n ≥ 0,
where the first equivalence holds, because (cof sSet, RLP (cof sSet)) is a weak fac-
torization system by Theorem 3.29 and hence cof sSet = LLP (RLP (cof sSet)). The
second equivalence follows from Remark 3.24 applied to the adjunction (sdB , Ex) of
Remark 3.51. The thirs equivalence holds, because
ε
RLP (cof sSet) = RLP {∂∆n ,−→ ∆n ; n ≥ 0}
by Theorem 3.29 again. The last two equivalences are similar.
For checking the last assertion, by using the commutative diagram
f∂I(n)
sdB ∂∆n sdB B∂I(n) / BsdB ∂I(n)

sdB ε sdB ε BsdB (ε)


  fI(n) 
sdB ∆n sdB BI(n) / BsdB I(n)

it suffices to verify that the right vertical map is injective. But this is true since
∂I(n) ,−→ I(n) and hence sdB I(n) ,−→ sdB I(n) is injective.
(ii) By Theorem 3.29 we have
di tε
fib sSet = RLP {∆0 × ∆n +∆0 ×∂∆n ∆1 × ∂∆n −→ ∆1 × ∆n ; n ≥ 0, i = 0, 1}.
So by the same arguments as in (i), it suffices to check that
sdB (di t ε) ∈ LLP (fib sSet), ∀ n ≥ 0, i = 0, 1.
However this is much more difficult than the case in (i) and will be obtained in
several steps.

102
3.10. Verifying the model structure on simplicial sets using Kan’s functor Ex∞

j
a) First we prove that sdB BI(m ,−→ n) ∈ LLP (fib sSet), for every injection
j
m ,−→ n. Using Proposition 3.52 and Remark 3.54 we have a natural isomor-
phism

sdB BI(m) −→ BsdB I(m) = BT SI(m) = BSI(m), m ≥ 0,

and the partially ordered set SI(m) considered as a category has a terminal
object m. So Lemma 3.45 implies that sdB BI(j) ∈ wa sSet. Since sdB BI(j) ∈
cof sSet by (i), Proposition 3.43 implies that sdB BI(j) ∈ LLP (fib sSet).
di ×id
b) Next we prove that sdB B(I(0) × C −→ I(1) × C) ∈ LLP (fib sSet), where
i = 0, 1 and C is an ordered simplicial complex with n := ]C < ∞. Then
C = s1 ∪ ... ∪ sm with si ∈ S(C) maximal. We prove that di × id is an absolute
weak equivalence by induction on m ≥ 1.

• For m = 1 there is an isomorphism g : I(n) −→ C, so we may assume
C = I(n). We only prove the case i = 0 and proceed similarly as in the
proof of Proposition 3.35 by constructing intermediate subcomplexes

I(0) × I(n) = U0 ,−→ U1 ,−→ ... ,−→ Un = I(1) × I(n).

More precisely we define

Uk+1 := Uk ∪ tk , tk = {(0, 0), ..., (0, k), (1, k), ..., (1, n)}, 1 ≤ k ≤ n.

Then Uk ∩ tk = t0k = {(0, 0), ..., (0, k − 1), (1, k), ..., (1, n)} ∼
= I(n) and we
get a cartesian square as on the left

I(n) / Uk BI(n) / BUk


dk dk =:jk+1
   
I(n + 1) / Uk+1 , BI(n + 1) / BUk+1 .

Hence also the right square is cartesian and by definition of Uk+1 also
cocartesian. The right square stays cocartesian after applying the left ad-
joint functor sdB . Since sdB dk ∈ LLP (fib sSet) by b), Lemma 3.13 implies
that also (BUk −→ BUk+1 ) ∈ LLP (fib sSet) and thus sdB (di × id) =
sdB jn ◦ ... ◦ sdB j1 ∈ LLP (fib sSet).
• For m > 1 we consider the ordered subcomplex C 0 := s1 ∪ ... ∪ sm−1 ≤ C
again. Like in the proof of Proposition 3.52 we get a commutative diagram

sdB B(I(0) × C 0 ) o o sdB B(I(0) × (C 0 ∩ sm )) / / B(I(1) × sm )


sdB B(di ×id) sdB B(di ×id) sdB B(di ×id)
  
sdB B(I(1) × C 0 ) o o B(I(1) × (C 0 ∩ sm )) / / B(I(1) × sm ),

in which the vertical maps are absolute weak equivalences by the induction
hypothesis. As all the horizontal maps are cofibrations by (i), Theorem 3.41

103
Chapter 3. Abstract homotopy theory

sdB B(di ×id


(i) implies that the horizontal colimit sdB B(I(0)×C) −→ B(I(1)×C)
also is an absolute weak equivalence. As it is injective and thus a cofibra-
tion, Proposition 3.43 implies that is in LLP (fib sSet).
c) Now in the commutative diagram

sdB (di ×id)


sdB (∆0 × ∂∆n ) / sdB (∆1 × ∂∆n )
sdB (id×ε) sdB (id×ε)
 
sdB (∆0 × ∆n ) / sdB (∆0 × ∆n +∆0 ×∂∆n ∆1 × ∂∆n )
di tε
+ 
sdB (di ×id) / sdB (∆1 × ∆n )

every map is injective by Theorem 3.30 and (i). Using the isomorphisms

∆m × ∂∆n = B(I(m) × ∂I(n)), ∆m × ∆n = B(I(m) × I(n)), m = 0, 1,

we have shown that the upper horizontal map and the lower curved map is
in LLP (fib sSet). As the square is cocartesian, Lemma 3.13 implies that the
lower horizontal map is in LLP (fib sSet). So Corollary 3.44 implies that di tε ∈
LLP (fib sSet).

For our purposes much more important than the map h is the following map a.

Proposition 3.56
The natural so-called last vertex map of partially ordered sets

V : SI(m) − m, s 7−→ max s, m≥0

induces a natural surjective absolute weak equivalence


aX
sdB X // XO
'

colim m BV /
colim BsdB I(n) colim BSI(n) ∼ colim B(n).
m∈∆/X m∈∆/X m∈∆/X

Again the isomorphism on the right is induced by the canonical one from the co-Yoneda
Lemma 2.51.
Proof. The map V is a retraction, an order-preserving section is given by

W : m −→ SI(m), k 7−→ k.

It follows that also BV is a retraction and thus an epimorphism. As epimorphisms are


stable under colimits, it follows that aX is an epimorphism.

104
3.10. Verifying the model structure on simplicial sets using Kan’s functor Ex∞

It remains to prove that aX is an absolute weak equivalence, which is a lenghty induction


using our theory of absolute weak equivalence established in section 3.8. First we assume
X = BC, where C is an ordered simplicial complex with n := ]C < ∞. Then C =
s1 ∪ ... ∪ sm with si ∈ S(C) maximal. We prove that aBC is an absolute weak equivalence
by induction on m ≥ 1.

• For m = 1 there is an isomorphism g : I(n) −→ C. Using Remark 3.54 we get a
commutative diagram
fI(n) sdB B(g)
BSI(n) BsdB I(n) o ∼ sdB BI(n) ∼
/ sdB BC
BV aBI(n) aBC
  B(g) 
B(n) BI(n) / BC,

so it suffices to prove that the left vertical map is an absolute weak equivalence.
But considering the partially ordered sets SI(n) and n as categories, the objects
n ∈ SI(n) and n ∈ n are terminal. So Lemma 3.45 implies that BV is an anodyne
extension.

• For m > 1 we consider the ordered subcomplex C 0 := s1 ∪ ... ∪ sm−1 ≤ C again. Like
in the proof of Proposition 3.52 we get a commutative diagram

sdB BC 0 o o sdB B(C 0 ∩ sm ) / / Bsm


aBC 0 aB(C 0 ∩sm ) aBsm
  
BC 0 o o B(C 0 ∩ sm ) / / Bsm ,

in which the vertical maps are absolute weak equivalences by the induction hypoth-
aBC
esis. Moreover taking the colimit horizontally yields the map sdB BC −→ BC. As
all the horizontal maps are injective and hence cofibrations, Theorem 3.41 (i) im-
plies that also aBC is an absolute weak equivalence. This concludes the proof for the
induction step.

Now let X ∈ sSet be an arbitrary simplicial set. By induction on n ≥ 0 we will show that
askn X
sdB skn X −→ skn X is an absolute weak equivalence.

• For n = 0 the simplicial set sk0 X is a set of points. As sdB BI(0) −→ BsdB I(0) =
ask0 X
BI(0) it follows that sdB sk0 X −→ sk0 X is an isomorphism.

• For n > 0, recall that by Proposition 2.76 we have cocartesian squares

X̃n
B∂I(n) = X̃n × ∂∆n / skn−1 X
_

 
X̃n
BI(n) = X̃n × ∆n / skn X.

105
Chapter 3. Abstract homotopy theory

Using that sdB is a left adjoint and therefore commutes with pushouts and coprod-
uct, taking colimits horizontally in the commutative diagram
X̃n
sdB BI(n) o o X̃n sdB B∂I(n) / sdB skn−1 X
X̃n X̃n askn−1 X
aBI(n) aB∂I(n)
  
X̃n
BI(n) o o X̃n B∂I(n) / skn−1 X
n ask X
yields the map sdB skn X −→ skn X. Using that absolute weak equivalences are
closed under arbitrary coproducts by Proposition 3.42, the left two vertical maps
are absolute weak equivalences, because ∂I(n) ≤ I(n) are finite simplicial complexes.
It follows that also askn X is an absolute weak equivalence by Theorem 3.41, which
proves the induction step.
Finally we apply Theorem 3.41 to the map of sequences
sdB sk0 X / / sk1 X / / sk2 X / / ...
ask0 X ask1 X ask2 X
  
sk0 X / / sk1 X / / sk2 X / / ...
a
X
to see that also sdB X −→ X is an absolute weak equivalence.
2
Infact we are more interested in the functor Ex right adjoint to sdB . An colimit over its
iteration will provide the desired fibrant replacement functor Ex∞ , which was introduced
by Daniel Kan during his foundational studies of simplicial sets.

Definition 3.57
For every X ∈ sSet, we define
b b b 2
Ex∞ X := colim (X −→
X ExX
ExX −→ Ex2 X −→
Ex X
...),
Xb X a
where X −→ ExX is the map corresponding to sdB −→ X under the adjunction 3.51.

Proposition 3.58
For every X ∈ sSet, we have Ex∞ X −_ ∗.
Proof. Using the description of Proposition 3.35
[
fib sSet = RLP {Λnk := di ∆n−1 ,−→ ∆n ; 0 ≤ k ≤ n}
0≤i≤n,
i6=k

we need to prove that there is a solution for every lifting problem

Λnk
u / Ex; ∞ X
_
∃d
 
∆n / ∗.

106
3.10. Verifying the model structure on simplicial sets using Kan’s functor Ex∞

As Λnk is a finite simplicial set, Proposition 3.22 implies that u factors as


v ι
Λnk −→ Exm X −→
m
Ex∞ X,
for some m ≥ 1. So setting Y := Exm X it suffices to construct a map d in every diagram
bExY
Λnk
v / ExY /3 Ex2 Y
_

 ∃d

∆ n / ∗
By applying Remark 3.24 twice and using that Ex∗ = sSet(sdB ∆• , ∗) = ∗, we can equiv-
alently solve one of the lifting problems below
sdB aΛn
v0 /Y bY
/3 / v0
sd2B Λ 3/
n n k
sdB Λ
 k ExY  k sdB Λnk Y
_ _ 7
∃d
 ∃d
  ∃d0
/ 
sdB ∆ n
∗ sd2B ∆n / ∗

Recall that in the proof of Proposition 3.35 we introduced the ordered subcomplex Ank ≤
I(n), which after applying the nerve functor yields the simplicial subset Λnk ≤ ∆n . Now
by construction of a and f there is a commutative square
fI(n)
BSI(n) BT SI(n) BsdB I(n) o ∼ sdB BI(n) sdB ∆n
BV BV aBI(n) a∆n
   
B(n) BI(n) BI(n) ∆n .
V V
So by naturality of a and f the map T SI(n) −→ I(n) restricts to a map T SI(Ank ) −→ Ank
and there are natural isomorphisms
aΛn
sdB ∆n o ? _ sdB Λn k
/ Λnk
k
fI(n) o fAn o
k
 
BT SI(n) o ? _ BT S(An ) BV / B(An ).
k k

So applying sdB and using f again it suffices to solve the lifting problem of simplicial
complexes (the map d need and will not be order-preserving!)

T S(A n V / Ank
 k) _ ;

 ∃d
T SI(n).

But this is possible by setting



max s, s ∈ S(Ank ),
d : T SI(n) −→ Ank , s 7−→
{k}, s∈/ S(Ank ).

107
Chapter 3. Abstract homotopy theory

/ S(Ank ). Then
Indeed for s := {s0 ( ... ( sm } ∈ ST SI(n), let j be maximal with sj ∈
there is an 0 ≤ i ≤ n, i 6= k with i ∈
/ sj and hence
/ {max s0 , ..., max sj , {k}} ⊃ d(s) ∈ S(Ank ).
i∈

2
Finally by using all the good properties of the functor Ex∞ we can prove the difficult
inclusion fib sSet ⊂ RLP (cof sSet ∩ wsSet) in the construction of the model structure.

Proposition 3.59
For every fibration f : X −_ Y the following holds.
(i) Ex∞ f ∈ fib sSet.
(bX ,f )
(ii) X −→ ExX ×ExY Y is a strong deformation section.
(ιX ,f )
(iii) (X −→ Ex∞ X ×Ex∞ Y Y ) ∈ LLP (fib sSet).
(iv) f ∈ RLP (cof sSet ∩ wsSet).
Proof.
(i) Using Theorem 3.29 we need to solve every lifting problem
∆0 × ∆n +∆0 ×∂∆n ∆1 × ∂∆n
u / Exm X
4
∃d
di tε Exm f
 
∆1 × ∆n / Exm Y,
v

for i = 0, 1 and m = ∞. As the two simplicial sets on the left are finite and hence
ω-compact by Proposition 3.22, the map u and v factor over Exm X and Exm Y
respectively, for some m < ∞. So we may assume that m < ∞. But then there is a
solution d, because Exm f is a fibration by m-fold application of Proposition 3.55.
(ii) Using that the natural map a epimorphic by Proposition 3.56, it follows that
(a∆• ) ∗
bX : X ∼
= sSet(∆• , X) ,−→ sSet(sdB ∆• , X) = ExX
is injective and hence a cofibration by Definition 3.27. It follows that also the map
c := (bX , f ) : X ,−→ ExX×ExY =: X 0
is injective, because πY (bX , f ) = bX . So in the commutative diagram
aX
sdB X / X (

sdB c
  c
sdB X 0 / 0
sdB X +sdB X X

aX 0 ∪c
'  0
aX 0 0X,

108
3.10. Verifying the model structure on simplicial sets using Kan’s functor Ex∞

the maps c and sdB c are cofibrations by Definition 3.27 and Proposition 3.55 re-
spectively. As aX and aX 0 are absolute weak equivalences by Proposition 3.56, also
aX 0 ∪ c is an absolute weak equivalence by the 2-of-3 property for absolute weak
equivalences proven in Proposition 3.38.
Now the commutative diagram on the left induces a commutative square in the
middle, for which there is a diagonal and homotopies h, k as depicted below
0
πExX ∪id
X X sdB X 0 +sdB X X /8 X
'
h∪k
c (bX ,f ) aX 0 ∪c f
∃d
 (πExX ,πY )   _
X0 / ExX ×ExY Y /
= X0 πY Y,
0
where πExX is the map corresponding to πExX under the adjunction bijection

sSet(sdB X 0 , X) = sSet(X 0 , ExX)

of Remark 3.51. Defining k 0 ∈ sSet(∆1 , X 0 , ExX) as the map similarly corresponding


to
(a 1 ,sdB (πX 0 )) k
sdB (∆1 × X 0 ) ∆ −→ ∆1 × sdB X 0 −→ X,
we can define k 00 := (k 0 , πY ) : ∆1 × X 0 −→ ExX ×ExY Y = X 0 and get

idX ' dc, f h = f πX , idX 0 '00 cd, f dk 00 = πY k 00 = πY .


h k

As f is a fibration and hence also Exf by Proposition 3.55, so is also the left vertical
map in the cartesian square
πExX
ExX ×ExY Y / ExX
πY Exf
 _
bY
Y / ExY.

because fib sSet is stable under pullbacks by the dual of Lemma 3.13. In particular
the map
c=(bX ,f ) πY
X −→ X 0 = ExX ×ExY Y −_ ∗
is a πY -fibrered homotopy equivalence and hence (bX , f ) ∈ LLP (fib sSet) by Propo-
sition 3.49.

(iii) For m ≥ 0 the lower horizontal map is a fibration by m-fold appliation of Proposition
3.55
πY
Exm X ×Exm Y Y /Y

πExm X
 
Exm f  ,2 m
Exm X Ex Y.
As fib sSet is stable under pullbacks by the dual of Lemma 3.13, it follows that also
the upper horizontal map is a fibration. So applying (ii) yields that the left vertical

109
Chapter 3. Abstract homotopy theory

map in the commutative square below is a strong deformation retraction and hence
in LLP (fib sSet) by Proposition 3.33
bExm X ×id
Exm X ×Exm Y Y / Exm+1 Y ×O Exm+1 Y Y
(b,πY ) (πExm+1 X ,πY )
 (Ex(πExm X ),Ex(πY ))×id
Ex(Exm X ×Exm Y Y ) ×ExY Y / (Exm+1 X ×Exm+1 Y ExY ) ×ExY Y.

• As Ex is a right adjoint and therefore commutes with limits, the map (Ex(πExm X ), Ex(πY ))
and hence the lower horizontal map is an isomorphism.
• The right vertical map is an isomorphism, because the two small squares

(Exm+1 X ×Exm+1 Y ExY ) ×ExY Y / Exm+1 X ×Exm+1 Y ExY / Exm+1 X


πExY Exm+1 f
  
bY bExm Y ◦...◦bExY
Y / ExY / Exm+1 Y

are cartesian and hence also the outer square is cartesian.


It follows that the upper horizontal map is in LLP (fib sSet) and by Lemma 3.13
the same holds for the map
ι (bX ,f ) bExX ×id ∼
0
(ι0 , f ) : X −→ colim (X −→ ExX ×ExY Y −→ ...) −→ Ex∞ X ×Ex∞ Y Y.
Here the last map is an isomorphism, because filtered colimits commute with finite
limits and hence pullbacks by Proposition 2.48.
(iv) In the commutative diagram
X 6 X
∃d
(ι0 ,f ) f
 _
πY

Ex X ×Ex∞ Y Y / Y
πEx∞ X ι0
 Ex∞ f 
 ,2 ∞  ,2
Ex∞ X Ex Y ∗
the left vertical map is in LLP (fib sSet). So there is a diagonal d, which proves
that f is a retract of πY . The lower left horizontal map is a fibration by (i), while
the lower right horizontal map is a fibration by Proposition 3.58. So Corollary 3.48
implies that Ex∞ f ∈ RLP (cof sSet ∩ wSet). It follows that also πY and its retract
f is in RLP (cof sSet ∩ wSet), which is stable under pullbacks and retracts by the
dual of Lemma 3.13.
2

Remark 3.60
By the structure of the proof of Proposition 3.59 (iv), we see that one can prove the
ηX
inclusion fib sSet ⊂ RLP (cof sSet ∩ wsSet) by using any natural transformation X −→
ι0
Q(X) instead of X −→ Ex∞ X having the following properties.

110
3.10. Verifying the model structure on simplicial sets using Kan’s functor Ex∞

(i) Q(X) −_ ∗, for all X ∈ sSet,

(ii) Q(fib sSet) ⊂ fib sSet,


(ηX ,f )
(iii) (X −→ Q(X) ×Q(Y ) Y ) ∈ LLP (fib sSet).
X η
By using different techniques one can show that also the unit X −→ S|X| of the adjunction
T op(|X|, Y ) = sSet(X, S(Y )) has all these properties.

After having shown the difficult inclusion fib sSet ⊂ RLP (cof sSet ∩ wsSet) in Propo-
sition 3.59 (iv), it is now comparably easy to deduce the model structure:

Theorem 3.61
The category sSet is a model category with weak equivalences, fibrations and cofibrations
defined as in Definition 3.27.
Proof. For every T −_ ∗ Remark 3.6 implies that the inverse image WT of Mor(Setop )×
under the functor
π0 sSet(-,T )
sSet −→ π0 sSet −→ Setop
satisfies the 2-of-3 axiom. Hence the same holds for the class
\
wsSet = WT .
T −_∗

Recall that by Theorem 3.29 there are two weak factorization systems

(cof sSet, RLP (cof sSet)), (LLP (fib sSet), fib sSet),

and it remains to prove that

RLP (cof sSet) = fib sSet ∩ wsSet, LLP (fib sSet) = cof sSet ∩ wsSet.

• We begin with the second equality. By Proposition 3.59 we have

fib sSet ⊂ RLP (cof sSet ∩ wsSet).

Combining this with Proposition 3.43 and Proposition 3.37 yields

LLP (fib sSet) = cof sSet ∩ wa sSet ⊂ cof sSet ∩ wsSet


⊂ LLP (RLP (cof sSet ∩ wsSet) ⊂ LLP (fib sSet),

so all these classes are equal.

• For the first equality note that in Remark 3.34 we have already checked that

RLP (cof sSet) ⊂ fib sSet ∩ wsSet. (3.5)

So let f ∈ fib sSet ∩ wsSet. As (cof sSet, RLP (cof sSet)) is a weak factoriza-
c g
tion system, the map f can be factored as X −→ Z −→ Y , where c ∈ cof sSet

111
Chapter 3. Abstract homotopy theory

and g ∈ RLP (cof sSet). As f = gc and g are weak equivalences by assump-


tion and (3.5) respectively, also c is a weak equivalence by the 2-of-3 property.
So c ∈ cof sSet ∩ wsSet = LLP (fib sSet) by (i) and hence there is a diagonal d in
the commutative square
X ;X 
∃d
c f
 _
g
Z / Y.

In particular f is a retract of g and thus also f ∈ RLP (cof sSet), as this class of
maps is closed under retracts by the dual of Lemma 3.13.
2

3.11 Homotopies in model categories


The construction of the homotopy category of a model category is obtained by using the
terms of abstract homotopies and homotopy equivalences.

Definition 3.62
Let C be a model category, ∅ an initial and ∗ a terminal object in C.
(i) A cylinder object for X ∈ C is an object I · X ∈ C together with morphisms
X +X
 idX ∪idX =:∇
i=i0 ∪i1

I ·X /' X.
'

Every object has a cylinder object given by any factorization of the fold map ∇
provided by the model structrue on C.
(ii) Two morphisms f, g ∈ C(X, Y ) are called left homotopic via h, short f h ' g, if
i '
there is a map h ∈ C(I · X, Y ), for some cylinder object X + X −→ I · X −→ X,
giving rise to a commutative diagram
X +X
f ∪g  ∇
i
w  '/
Y o h
I ·X ' X.

(iii) Dually a path object for Y ∈ C is an object Y I ∈ C together with morphisms

Y
' /YI

p
∆ ' _
Y × Y.
Every object has a path object given by any factorization of the diagonal map ∆
provided by the model structrue on C.

112
3.11. Homotopies in model categories

(iv) Two morphisms f, g ∈ C(X, Y ) are called right homotopic via h, short f 'h g, if
' p
there is a map h ∈ C(X, Y I ) for some path object Y −→ Y I −_ Y × Y , giving rise
to a commutative diagram

Y
' / YI o
h
X
p
∆ ' _ w (f,g)
Y × Y.

Example 3.63
In the model category sSet the following holds.

(i) For every X ∈ sSet there is a canonical cylinder object, given by

X + X = ∂∆1 × X
 idX ∪idX =:∇
i=d0 ∪d1

' )/
∆1 × X πX X.

Indeed i is injective and hence a cofibration, while πX is a strong deformation re-


traction and thus a weak equivalence.

(ii) For every Y ∈ sSet with Y −_ ∗ there is a canonical path object, given by
s0
Y = sSet(∆0 , Y ) / sSet(∆1 , Y )
'

p:=(d0 ,d1 )
∆ + _
sSet(∂∆1 , Y ) = Y × Y.

Indeed p = (d0 ∪ d1 )∗ is a fibration by Corollary 3.31, since Y −_ ∗, while s0 is a


strong deformation section and thus a weak equivalence.

In contrast to the simplicial theory developed before, the term of homotopies is now
slightly more relaxed, meaning that the cylinder/path objects may vary. This freedom
allows the proof of the following Lemma, which we know does not hold in this generality
for the term “simplicial homotopy”.

Lemma 3.64
Let ∅−→ X with a cylinder object
i0 ∪i1 X r
X + X −→ I · X −→ X.

Then the following holds.


ιj
(i) ij : X −→ X + X−→ I · X is a trivial cofibration, for j = 0, 1.

113
Chapter 3. Abstract homotopy theory

i1 ∪i0 '
(ii) X + X −→ I · X −→ X is a cylinder object for X.
To distinguish it from I · X we denote it by I 0 · X = I · X and call it inverse
cylinder object.
i0 ∪i0
X r
(iii) Let X + X −→ J · X −→ X be another cylinder object for X.
(ι0 ◦i0 )∪(ι1 ◦i1 ) r ∪r
Then there is a composed cylinder object X + X −→ (I ∗ J) · X X−→X X
defined by the pushout square

X
i1
i0

X / I ·X
i1 ιI
i0
 ιJ
 rX
X / J ·X / (I ∗ J) · X
rX ∪rX

% 
rX 0 X.

(iv) Left homotopy is an equivalence relation on C(X, Y ) for every Y ∈ C.


Dual assertions hold for path objects and right homotopies.
Proof.
(i) We have a pushout diagram
∅ / X
ι0
 ι1

X / X + X.
Because X is cofibrant and cofibrations are preserved by pushouts by Lemma 3.13,
ι0 , ι1 is a cofibration. As compositions preserve cofibrations i0 , i1 are cofibrations,
too.
For j = 0, 1 the map ij is a weak equivalence, because of the 2-of-3 axiom for weak
equivalences.
X +O X / /I
6 ·X
ij
ιj '
idX 
X / X.
'

(ii) This is by symmetry of the axiom for cylinder objects.


0 i
(iii) By (i) the morphism X −→ I · X s a trivial cofibration. Since trivial cofibrations are
ιJ
stable under pushouts, the morphism J ·X −→ (I ∗J)·X is also a trivial cofibration.
r ∪r
It follows from the 2-of-3 axiom, that (I ∗ J) · X X−→X X is a weak equivalence,
rX
because J · X −→ X is a weak equivalence.

114
3.11. Homotopies in model categories

We have
i0 +i1 ιJ ∪ιI
X + X −→ (J · X) + (I · X) −→ X,
where the first morphism is a cofibration, because i0 and i1 are cofibrations by (i)
and cofibrations are stable under coproducts. The second morphism is a cofibration,
because we have a pushout diagram
i1 +i0
X +X / (I · X) + (J · X)

i0 ∪i1 ιI ∪ιJ
 ιJ

J(X) / (I ∗ J) · X,

where the left vertical morphism is a cofibration by definition and hence the right
vertical morphism is a cofibration, because cofibrations are stable under pushouts.
'
(iv) Let f ∈ C(X, Y ). Let rX : I · X −→ X be the canonical map. Then there is the
constant homotopy f f rX ' f .

f ∪f
X +X
 ∇
 "
I ·X
rX
/' X
f
/ Y.

Now let f, g ∈ C(X, Y ) with f h ' g, for some h ∈ C(I · X, Y ). Then I · X with
ij = i1−j , for j = 0, 1, is also a cylinder object and the same morphism h0 = h
defines an inverse homotopy g h0 ' f with respect to its inverse cylinder object.
Finally if e, f, g ∈ C(X, Y ) with e k ' f h ' g, where h ∈ C(I · X, Y ) and k ∈
C(J · X, Y ). Then there is a canonical composition of homotopies h ∗ k, such
that e h∗k ' g with respect to the cylinder object (I ∗ J) · X of (iii). More precisely
h ∗ k is defined as the natural pushout morphism in the diagram

X

' i1
i0

X / / I ·X
 
i1 ' ' ι1
i0 ι0
  h
X / ' / J · X / ' / (I ∗ J) · X
h∗k

% 
k 0 Y.

Note that the outer paths commute, since by hypothesis hi1 = f = ki0 .

It was crucial in the construction of the model structure on simplicial sets, that we have
a “Pushout-Product” property as given Theorem 3.30. It enabled us to deform homotopies

115
Chapter 3. Abstract homotopy theory

in the desired way. Unfortunately we do not have a similar property in an arbitrary model
category. Instead deformations of homotopies and other construction of higher homotopies
are established by the interplay of cylinder and path objects. This will be demonstrated
in the following proposition.

Proposition 3.65
f,g ' p
Let ∅−→ X −→ Y and Y −→ Y I −_ Y × Y be a fixed path object.
If f, g are left homotopic, then f, g are right homotopic w.r.t. Y I .
ιj
Proof. Let ij : X ,−→ X + X−→ I · X denote the inclusions for j = 0, 1. Since X is
cofibrant, i0 is a trivial cofibration. Consider the diagram
f
X / Y '
2/ YI

∃K p
i0 '
 _
I ·X / X × (I · X) / Y × Y.
(can,id) f ×h

The upper right composition is equal to ∆◦f , the lower left one is (f ×h)◦(idX ×i0 )◦∆ =
(f ×f )◦∆ = ∆◦f . Hence the square commutes and the diagonal exists. Because h◦i1 = g
we obtain a right homotopy
K◦i1
Y ' / YI o X
p
∆ ' _ w (f,g)
Y × Y.

Corollary 3.66
Let f, g ∈ C(X, Y ), where X is cofibrant and Y is fibrant. Then the following is equivalent:

(i) f, g are left homotopic,

(ii) f, g are left homotopic w.r.t. a fixed cylinder I · X,

(iii) f, g are right homotopic,

(iv) f, g are right homotopic w.r.t. a fixed path object Y I .

Lemma 3.67
f
a / / b /
Let C be a model category and consider morphisms A X / Y B.
g

(i) f h ' g ⇒ bf bh ' bg.

116
3.11. Homotopies in model categories

(ii) f h ' g ⇒ f a ha ' ga, for some ha , if Y is fibrant or for ha = h(I · a) if we


have a map I · a rending the following diagram commutative

A+A / /I /
i '
·A A
a+a ∃I·a a
  
X +X /i /I ·X ' / X.

(iii) f 'h g ⇒ f a 'ha ga.

(iv) f 'h g ⇒ bf 'hb bg, for some hb , if X is cofibrant or for hb = bI h if we have


bI
a map Y I −→ B I rendering the following diagram commutative
' / p  ,2
Y YI Y ×Y
b ∃bI b×b
  
' / p  ,2
B BI B ×B
Proof.
(i) By assumption there is a cylinder I · X and an h, such that f h ' g. Then bf bh ' bg
is a left homotopy:
(bf )∪(bg)
X +X
idX ∪idX =∇  f ∪g

w  '/ #
Xo '
I ·X h
Y b / B.

(ii) Consider a homotopy

X +X
idX ∪idX =∇  f ∪g

w 
Xo
rX
I ·X h /' Y.
'

Then rX can be factored as on the left and we find a diagonal in the right
rX
I · X$ / I ·X h /
' 7 7A X <Y

' ' h0
c '
c
$ 0
rX  _
I · X 0. I · X0 / ∗.

' '
By construction we have f h0 ' g, because X + X −→ I · X −→ I · X 0 is another
cylinder object for X. Furthermore we find a diagonal

A+A
a+a / X +X / / I · X0
4
 I·a 0
' rX
 _
I ·A / A a / X.

117
Chapter 3. Abstract homotopy theory

Then ha = h(I · a) is the desired homotopy

A+A
a+a / X +X
idA ∪idA =∇   f ∪g

w   '/
h0
Ao /
' I·a
I ·A I · X0 Y.

The last two statements are dual.


2

3.12 The homotopy category of a model category


As the title suggests, the goal of this section is the construction of the derived category
of an arbitrary model category. Before we do that let us record the following abstract
Whitehead Theorem, which partly has strong similarity to Proposition 3.33 in the con-
text of simplicial sets. It is an abstract version of the classical Whitehead Theorem for
topological spaces, stating that each map between CW-complexes inducing isomorphisms
on homotopy groups is infact a homotopy equivalence.
Interestingly in the abstract setting of an arbitrary model category its proof is much
easier than the other implication that every homotopy equivalence is a weak equivalence,
which in the context of topological spaces or simplicial sets is trivial. The proof of the
latter will be given in the subsequent section.

Theorem 3.68 (Abstract Whitehead Theorem)


Let C be a model category.
'
(i) Let f : Z −_ Y with Y cofibrant and I · Z is a fixed cylinder object for Z.
Then f is a strong deformation retraction with respect to I · Z.
'
(ii) Let c : X −→ Z with X fibrant and Z I is a fixed cylinder object for Z.
Then c is a strong deformation section with respect to Z I .
(iii) Every weak equivalence between bifibrant objects is a homotopy equivalence.
Proof.
(i) Using that fib C ∩ wC = RLP (cof C) the proof is the same as Proposition 3.33 in
the context of simplicial sets. Because f is a trivial fibration and Y is cofibrant, we
can find a section s for f and an h, such that the following diagrams commute
(sf )∪idZ
∅ /9 Z Z +Z 7/ Z
  ∃h
∃s
' f ' f
 _  _
Y Y, ∇ I ·Z 7/ Y.
f
'

Z

118
3.12. The homotopy category of a model category

Hence f s = idY and sf h ' idZ with f h being trivial, meaning that the lower two
triangles commute. This means f is a strong deformation retraction.

(ii) This is dual to (i).


'
(iii) Let X, Y ∈ C bifibrant and a : X −→ Y a weak equivalence. Then a has a factor-
ization
3/ 5> Y
a
X % '
'
c f
%
Z,
where f is a trivial fibration and c is a cofibration. By the 2-of-3 axiom c is trivial,
too. Since X is cofibrant, Y fibrant and fibrations and cofibrations are closed under
composition Z is bifibrant. By (i) and (ii) we find a section s for f with sf h ' idZ ,
for some left homotopy h, and a retraction r for c with cr 'k idZ for some right
homotopy k. Using Lemma 3.67, we see that crs 'ks s. Define b := rs. By Corollary
3.66 there is a k 0 , such that crs k0 ' s and again by the preceding Lemma we get
ab = f crs f k0 ' f s = idY . Similarly we find a h0 such that ba = rsf c h0 c ' rc = idX .
Thus a and b define a homotopy equivalence.

The homotopy category of a model category will be constructed via the quotient cate-
gory of the subcategory of objects ∅−→ X −_ ∗. For this purpose we need the notion of
a bifibrant replacement of an object.

Definition 3.69
Let C be a model category and X ∈ C.

(i) A fibrant replacement of X is an object F (X) in a factorization


'
X −→ F (X) −_ ∗.

For f ∈ C(X, Y ) and a chosen fibrant replacement F (Y ) of Y , we let F (X) and


F (f ) be given by a factorization

f
X / Y

' '
 F (f )  
,2 F (Y )  ,2
F (X) ∗.

(ii) A cofibrant replacement of X is an object C(X) in a factorization


'
∅−→ C(X) −→ X.

119
Chapter 3. Abstract homotopy theory

For c ∈ C(X, Y ) and a chosen cofibrant replacement C(X) of X, we let C(Y ) and
C(c) be given by a factorization
C(c)
∅ / C(X) / / C(Y )

' '
 _
X c / Y.

(iii) A bifibrant replacement of X is a cofibrant replacement CF (X) of a fibrant


replacement F (X) of X.

Lemma 3.70
Let C be a model category, f0 , f1 ∈ C(X, Y ) with liftings

∅ / /
6 C(Y ) i = 0, 1.
 C(fi )
' qY
 _
'  ,2 /
C(X) qX X fi
Y.

If Y is fibrant, then f0 'h f1 implies C(f0 ) k ' C(f1 ), for some right homotopy k.
Proof. By Lemma 3.67 f0 'h f1 implies qY C(f0 ) = f0 qX 'hqX f1 qX = qY C(f1 ), and thus
qY C(f0 ) ` ' qY C(f1 ) by the dual of Proposition 3.65, since Y is fibrant. This homotopy
can be lifted as
C(f0 )∪C(f1 )
C(X) + CF (X) /4 CY
 k
i0 ∪i1 ' qY
 _
I · C(X) / Y.
`

Theorem 3.71
Let C be a model category.
Then C has a homotopy category Ho(C), whose objects are the same as C and

Ho(C)(X, Y ) := C(CF (X), CF (Y ))/ ', for all X, Y ∈ C,

for chosen bifibrant replacements CF (X) of X and CF (Y ) of Y .


Proof. For every X ∈ C, there is a bifibrant replacement, given by factorizations

X

'

 ,2
∅ / / CF (X) '
F (X)

_
∗,

120
3.13. Characterization of weak equivalences

where we chose F (X) = X, if X is fibrant, and CF (X) = F (X), if X and hence F (X) is
cofibrant.
Let X, Y ∈ C and f ∈ C(X, Y ). Then the lifting property induces a morphism CF (f ) ∈
C(CF (X), CF (Y )) between some chosen bifibrant replacements of X and Y :
f
X / Y /
' / F (Y ) ∅ / / CF (Y )
 6  4
F (f ) CF (f )
' '
 _  _
 ,2   ,2 /
F (X) ∗. CF (X) ' F (X) F (Y ).
F (f )

We have to check that CF (f ) is well-defined up to homotopy. Suppose f0 , f1 ∈ C(F (X), F (Y ))


are two liftings of f , meaning that the left squares commute. Then there exists a homotopy
h as in the right square
f f
X / Y X / Y /
' / F (Y )
' /3 F (Y )I
  
' ' h
' (p0 ,p1 )
 f0   _
/ / F (Y ) × F (Y ).
F (X) / F (Y ) F (X)
f1 (f0 ,f1 )

It follows that f0 'h f1 and thus C(f0 ) k ' C(f1 ) by the preceding Lemma, since F (Y )
is fibrant. Similarly one proves that the lifting CF (f ) of the map F (f ) is unique up to
homotopy.
It follows that Ho(C) is a well-defined category with the composition induced by the
γ
composition in C. Moreover the assignment f 7−→ CF (f ) defines a functor C −→ Ho(C).
Furthermore, if f ∈ C(X, Y ) is a weak equivalence, then CF (f ) is a homotopy equivalence
by the Whitehead Theorem and so γ(f ) is an isomorphism.
G
To check the universal property let C −→ D be a functor, which sends weak equivalences
to isomorphisms. Suppose f, g ∈ C(X, Y ) with X and Y bifibrant and f h ' g, where
i0 ∪i1 X r
X + X −→ I · X −→ X is a cylinder object. Since rX is a weak equivalence G(rX ) is an
isomorphism, so G(rX )G(i0 ) = G(idX ) = G(rX )G(i1 ) implies G(i0 ) = G(i1 ). Hence
G(f ) = G(hi0 ) = G(h)G(i0 ) = G(h)G(i1 ) = G(hi1 ) = G(g),
γ G0
proving that G factors as C −→ Ho(C) −→ D. By definition of Ho(C) the functor G0
is uniquely determined on objects. Since γ is full on bifibrant objects, G0 is uniquely
determined on the full subcategory of Ho(C) of objects, which are bifibrant in C. But for
every other object in Ho(C) there is a formal isomorphism to an object in this category,
proving that G0 is uniquely determined on Ho(C).
2

3.13 Characterization of weak equivalences


The goal of this section is to prove that in every model category the weak equivalences are
exactly those morphisms, which become isomorphisms in the homotopy category. This is
a often very useful characterization of weak equivalences.

121
Chapter 3. Abstract homotopy theory

Lemma 3.72
Let C be a model category and f0 , f1 ∈ C(X, Y ) with fibrant Y .
Then f0 'h f1 implies γ(f0 ) = γ(f1 ).
Proof. By construction of the homotopy category Ho(C) we have to check that the bifi-
idY
brant replacements of f0 and f1 are homotopic. As Y is fibrant, we have Y −→ Y = F (Y ).
It follows that the maps f0 and f1 can be lifted to fibrant replacments f0 and f˜1 as in the
˜
square on the left, and the homotopy h can be lifted to a homotopy h̃ as on the right.
fi
X / h / I
 <Y X
 :Y
f˜i h̃
iX ' iX ' (p0 ,p1 )
 _  _
F (X) / ∗, F (X) / Y × Y.
(f˜0 ,f˜1 )

Now Lemma 3.70 implies that C(f˜0 ) is homotopic to C(f˜1 ) and thus

γ(f0 ) = γ(f˜0 iX ) = γ(f˜0 )γ(iX ) = γ(f˜1 )γ(iX ) = γ(f˜1 iX ) = γ(f1 ).

Lemma 3.73
Let C be a model category. Let a, b : X −→ Y with a h ' b, for some h ∈ C(I · X, Y ).
Then h0 ∗ h H ' arX , where h0 ∗ h is the composed homotopy of h with its inverse h0 ,
obtained as the pushout morphism in the diagram

X

' i0
i1

X / / I ·X
 
i1 ' ' ι1
i0
 ι0
 h
X / ' /I ·X / '
/ I0 ·X
h0 ∗h

# 
h 0 Y.

The homotopy H is meant as a commutative diagram

(I 0 · X) +(X+X) (I 0 · X)
0 ∪r 0
rX X
 (h0 ∗h)∪arX
0
c

v (/
X rl 
f H
' X̃ Y,

0 '
where rX : I 0 ·X −→ X is the canonical map and rX
0 0
∪rX = f c is an arbitrary factorization
of the given type.

122
3.13. Characterization of weak equivalences

Proof. We choose a path object for Y as depicted in the left diagram and find a diagonal
in the right square
sY sY
Y / '
/ YI X a / Y 5/ YI

k
(p0 ,p1 ) i0 ' (p0 ,p1 )
∆=(idY ,idY ) # _  _
Y × Y, I ·X / Y × Y.
(arX ,h)

Let k = k 0 be the inverse homotopy and define a morphism

k̄ = k ∪ k 0 : I · X 0 = I · X +X I · X −→ Y I ,

which fits in the diagram


sY
X +X a∪a / Y 2/ YI

i0 ∪i1 (p0 ,p1 )
 
I0 · X / Y × Y.
0 ,h0 ∗h)
(arX

Now as c is a cofibration and p1 a trivial fibration we get a diagonal in the square


0 )
k̄∪(sY arX
(I 0 · X) +(X+X) (I 0 · X) /4 YI
 K
c ' p0
 _
X̃ / X / Y.
f a

Then H = p1 K has the desired property, because by construction


0 0
Hc = p1 Kc = p1 (k̄ ∪ (sY arX )) = (p1 k̄) ∪ (p1 sY arX ) = (h0 ∗ h) ∪ (arX
0
).

The subsequent proposition has strong similarity to the simplicial analog Proposition
3.47.

Proposition 3.74 (Quillen)


f
Let C be a model category and X −_ Y , where X, Y are bifibrant.
Then the following is equivalent.

(i) f is a weak equivalence.

(ii) f is a strong deformation retraction.

(iii) γ(f ) is an isomorphism in Ho(C).

123
Chapter 3. Abstract homotopy theory

Proof. The implication (i) ⇒ (ii) was proven in the abstract Whitehead Theorem 3.68.
Y s (p0 ,p1 )
Vice versa let tf h ' idX and choose a path object Y −→ Y I −_ Y × Y . Next choose a
factorization
(∆,sY f ) f
X / (X × I /
0 4< X)Y ×Y Y , X Y sY
' (∆,sY f )
sX ((p0 ,p1 ),f I ) '
X I (X × X) ×Y ×Y Y I / YI
∆ (p0 ,p1 )
(  _
X ×X /Y × Y.
f ×f

Since fibrations are stable under pullbacks and composition we have a fibration
((p0 ,p1 ),f I )
XI −_ (X × X) ×Y ×Y Y I −_ X × X
sX (p0 ,p1 )
and thus X −→ X I −_ X × X is a path object for X. Moreover we find a diagonal
sX
X /4 XI

H
i1 ' ((p0 ,p1 ),f I )
 _
I ·X / (X × X) ×Y ×Y Y I ,
((h,rX ),sY f rX )

because the square commutes, since

(f × f )(h, rX )i1 = (f hi1 , f rX i1 ) = (f rX i1 , f rX i1 ) = (f, f ) = ∆f = (p0 , p1 )sY f rX i1 .

Let k = Hi0 : X −→ X I . Then by construction

p0 k = p0 Hi0 = hi0 = tf, p1 k = p1 Hi0 = p1 rX = idX , f I k = f I Hi0 = sY f rX i0 = sY f.

In particular tf 'k idX . Now suppose we have a lifting problem as on the left, then we
find a diagonal as on the right

A
u /7 X A
ku
6/ XI
 
d D
c f c (p0 ,f I )
   
B / Y, B / X ×Y Y I .
v (tv,sY v)

It follows that d = p1 D solves the left problem, because

dc = p1 Dc = p1 ku = u, f d = f p1 D = p1 f I D = p1 sY v = v.

Hence f has the right lifting property with respect to all cofibration, meaning that it is a
trivial fibration and thus in particular a weak equivalence.

124
3.13. Characterization of weak equivalences

To see that (ii) implies (iii), it suffices to note, that for any homotopy equivalence f
with homotopy inverse t, the morphism γ(f ) is an isomorphism with inverse γ(t). The
other implication is similar to the proof of the implication (ii) ⇒ (iii) in Proposition 3.47.
However the deformation of the homotopy is slightly more difficult, because we have to
work with path objects.
Supposing γ(f ) is an isomorphism, then f is a homotopy equivalence with homotopy
inverse g, i.e. we have a left homotopy f g h ' idY , for some h ∈ C(I ·Y, Y ). By the homotopy
lifting property of f , there is a diagonal
g
Y /
 <X

i0 ' f
 _
I ·Y
h / Y.

Note that i0 is a cofibration by Lemma 3.64 (i), because Y is cofibrant. For t = h̃i1 we have
by construction f t = f h̃i1 = hi1 = idY and g h̃ ' t. By Lemma 3.67 (ii) tf is homotopic to
gf , since X is fibrant. But gf is homotopic to idX , so by composition we get a homotopy
idX k ' tf with k ∈ C(I 00 · X, X). We compose it with the homotopy tf k 0 , i.e.
X

' i0
i1

X / / I 00 ·X
 
i1 ' ' ι1
i0
ι0
 tf k
X / ' / I 00 · X / ' / I 0 · X
tf k0 ∗k
# 
k 0 Y.

We get two induced diagrams, where the right one comes from the homotopy f k 0 ∗f k H ' f rX
we get from the preceding Lemma.
7 X (I 0 · X) +(X+X) (I 0 · X)
tf k0 ∗k rX ∪rX  (f k0 ∗f k)∪f rX
f c
 
/
I0 · X
f k0 ∗f k
Y X lr u  ' X̃ H
/) Y.

From the right diagram we get a diagonal K in the diagram


n tf k0 ∗k
ι0
I0 · X /
<X

w
K
(I 0 · X) +(X+X) (I 0 · X) ' f
'
 _
c 0 X̃ / Y.
H

It follows, that h = Kcι1 is the desired homotopy tf h ' idX fibred over f , because

0 ki0 = idX , j = 0,
hιj = Kcι1 ιj = Kcι0 ιj = (tf k ∗ k)ιj =
tf ki0 = tf, j = 1,

125
Chapter 3. Abstract homotopy theory

and furthermore we have f h = f Kcι1 = Hcι1 = f rX by construction.


2

Using Proposition 3.74 we can deduce the following two useful corollaries characterizing
weak equivalences.

Corollary 3.75
g
Let C be a model category and X −→ Y , where X, Y are bifibrant.
Then the following is equivalent.

(i) g is a weak equivalence.

(ii) g is a homotopy equivalence.

(iii) γ(g) is an isomorphism in Ho(C).


Proof. By the Whitehead Theorem (i) implies (ii). Furthermore (ii) implies (iii) by con-
struction of the homotopy category in Theorem 3.71. Finally suppose (iii) holds. We can
c f
factor g as X −→ Z −_ Y , where c is a weak equivalence. Thus by the implications
we have shown γ(c) is an isomorphism in Ho(C). It follows that γ(f ) = γ(g)γ(c)−1 is
an isomorphism and hence f is a weak equivalence by the preceding Proposition. Hence
g = f c is a weak equivalence, proving (i).
2

Corollary 3.76
g
Let C be a model category and X −→ Y . Then the following is equivalent.

(i) g is a weak equivalence.

(ii) γ(g) is an isomorphism in Ho(C).


Proof. A bifibrant replacement of g induces a commutative diagram
g
X / Y
' '
 F (g) 
F (X) / F (Y )
O O
' '
CF (g)
CF (X) / CF (Y ).

Using the preceding Corollary and the 2-of-3 axiom for weak equivalences resp. for mor-
phisms in C being mapped to isomorphisms in Ho(C) by γ, the following holds.
g is a weak equivalence ⇐⇒ F (g) is a weak equivalence
⇐⇒ CF (g) is a weak equivalence ⇐⇒ γCF (g) is an isomorphism
⇐⇒ γF (g) is an isomorphism ⇐⇒ γ(g) is an isomorphism.
2

126
3.14. Derived functors and the comparison of model categories

3.14 Derived functors and the comparison of model


categories
Like in the context of chain complexes, we are able to define left and right derived functors
for functors on a model category.

Definition 3.77
G
Let C be a model category, D an arbitrary category and C −→ D.

(i) The left derived functor LG is defined as the right Kan extension of G along
γ LG
C −→ Ho(C), it it exists. That is a functor Ho(C) −→ D with a natural transforma-
ε
tion LG ◦ γ −→ G, such that LG is a terminal object in the comma category
-◦γ
γ ∗ /G, γ ∗ := (- ◦ γ) : CAT(Ho(C), D) −→ CAT(C, D).

(ii) The right derived functor RG is defined as the left Kan extension of G
γ RG
along C −→ Ho(C), if it exists. That is a functor Ho(C) −→ D with a natural
η
transformation G −→ RG ◦ γ, such that RG is an initial object in the comma
category G/γ ∗ .

A useful tool in many applications is the following Lemma by Ken Brown.

Lemma 3.78 (K. Brown)


In every model category C the following holds.
'
(i) Every weak equivalence w : X −→ Y between cofibrant X, Y ∈ C can be factored as

3/ 5> Y
w
X % ' 
' '

%
Z, o '

' '
meaning that Y −→ Z −_ Y is the identity on Y .
'
(ii) Every weak equivalence w : X −→ Y between fibrant X, Y ∈ C can be factored as

3/ 5> Y
w
XS[ %
o '
' '

'
%
Z,
' '
meaning that X −→ Z −_ X is the identity on X.
Proof. Since X and Y are cofibrant and cofibrations are stable under pushouts, we have
'
X−→ X + Y ←−Y . Take a factorization X + Y −→ Z −_ Y and let X−→ X + Y −→ Z

127
Chapter 3. Abstract homotopy theory

be the composition. This is a trivial cofibration by the 2-of-3 axiom. Moreover Y −→
'
X + Y −→ Z is a section for Z −_ Y and so is also a trivial cofibration.
Statement (ii) is dual to (i).
2

Theorem 3.79
Let C be a model category and G ∈ CAT(C, D).

(i) Suppose G maps trivial cofibrations between cofibrants to isomorphisms.


Then G exists and is given by
'
εG : LG(X) = GC(X) −→ G(X), ∅−→ C(X) −→ X.

(ii) Suppose G maps trivial fibrations between fibrants to isomorphisms.


Then RG exists and is given by
'
ηG : G(X) −→ GF (X) = RG(X), X −→ F (X) −_ ∗.
'
Proof. Let ∅−→ C(X) −_ X be a chosen cofibrant replacement, for every X ∈ C, where
idX
we set C(X) −→ X, if X ∈ C is already cofibrant. Dually to as was seen in the proof of
Theorem 3.71 any two liftings f0 , f1 ∈ C(C(X), C(Y )) of a morphism f ∈ C(X, Y ) are
left homotopic via some homotopy
f0 ∪f1
C(X) + C(X) /3 C(Y )
h
i0 ∪i1 '
 _
/  ,2 /
I · C(X) rC(X) C(X) ' X f
Y,

i0 ∪i1 '
where C(X) + C(X) −→ I · C(X) −_ C(X) is a cylinder object for C(X). Since C(X)
i0
is cofibrant so is ∅−→ C(X) −→ I · C(X) by Lemma 3.64 (i). By K. Brown’s Lemma G
maps weak equivalences between cofibrant objects to isomorphisms and thus G(rC(X) ) is
an isomorphism. Hence

G(rC(X) )G(i0 ) = G(rC(X) i1 ) = G(idC(X) ) = G(rC(X) i1 ) = G(rC(X) )G(i1 )

implies G(i0 ) = G(i1 ) and thus

G(a) = G(hi0 ) = G(h)G(i0 ) = G(h)G(i1 ) = G(hi1 ) = G(b).


GC
In particular we get a well-defined functor C −→ D. Since f ∈ C(X, Y ) is a weak
equivalence, if and only if a lifting F (f ) ∈ C(F (X), F (Y )) is one by the 2-of-3 ax-
iom, the functor GC maps weak equivalences to isomorphisms and we get a unique
γ LG
factorization GC : C −→ Ho(C) −→ D by the universal property of the homotopy
'
category Ho(C). By construction pX : C(X) −_ X defines a natural transformation

128
3.14. Derived functors and the comparison of model categories

εG = G(p) : γ ∗ L(G) = GC −→ G. We have to check the universal property of right Kan


f
extensions, meaning that for all natural γ ∗ G0 −→ G there is a natural f 0 rendering the
diagram below commutative
f
γ ∗ G0 / G0
<G
γ ∗ (f 0 ) ∃!f 0
εG
 
γ ∗ LG LG.

For X ∈ C consider the commutative diagram

γ ∗ G0 (pX )
γ ∗ (G0 )(C(X)) / γ ∗ (G0 )(X)

fC(X) fX
 εG

γ ∗ LG(X) = GC(X) / G(X),

where the upper horizontal morphism is an isomorphism, since γ maps weak equivalences
to isomorphisms. It follows that f 0 = G0 (fC )G0 (p)−1 is the unique morphism we are looking
for. Statement (ii) is dual to (i).
2

We will also introduce the notion of total derived functors between model categories,
providing the most important tool for comparing different model categories and their
homotopy categories.

Definition 3.80
G
Let C and D be model categories and C −→ D.

(i) The total left derived functor of G is defined as LG = L(γG).

(ii) The total right derived functor of G is defined as RG = R(γG).

Theorem 3.81 (Quillen’s adjoint functor theorem)


An adjunction between model categories C and D

C(E(X), Y ) = D(X, G(Y )),

subject to the following (by Remark 3.24 equivalent) hypotheses.

• E preserves trivial cofibrations between cofibrants and cofibrations.

• G preserves trivial fibrations between fibrants and fibrations.

Then the following holds.

(i) The total derived funtors of E and G exist and induce an adjunction

Ho(C)(LE(X), Y ) = Ho(D)(X, RG(Y )).

129
Chapter 3. Abstract homotopy theory

(ii) (LE, RG) form an equivalence of categories, if and only if



a) X −→ GE(X) −→ GF E(X) is a weak equivalence, for all cofibrant X ∈ D,
Y ε
b) ECG(Y ) −→ EG(Y ) −→ Y is a weak equivalence, for all fibrant Y ∈ C.
Proof. The existence of LE and RG follows from Theorem 3.79 and its dual. Let F resp.
C denote a chosen fibrant resp. cofibrant replacement in the particular model category.
Let X ∈ D and Y ∈ C. We have a chain of natural bijections

Ho(C)(LE(X), Y ) = Ho(C)(γEC(X), Y ) = C(CF EC(X), CF (Y ))/ ∼


= C(EC(X), F (Y ))/ ∼= D(C(X), GF (Y ))/ ∼
= D(CF (X), CF GF (Y ))/ ∼= Ho(D)(X, γGF (Y ))/ ∼= Ho(D)(X, RG(Y )),

where the first one is by definition and the second one by construction of the homotopy
category in Theorem 3.71. The third one is induced by composition with the (co-)fibrant
replacement morphisms using Corollary 3.66 and Lemma 3.67. The middle one is induced
by the adjunction and the arguments for the rest are dual to those given before.
By construction of the adjunction between the homotopy categories we see that its unit
resp. counit is an isomorphism, if and only if the conditions a) and b) hold.
2

Remark 3.82
By Brown’s Lemma the hypotheses of Theorem 3.81 are satisfied, if one of the following
equivalent conditions holds:

• E preserves cofibrations and trivial cofibrations.

• G preserves fibrations and trivial fibrations.

In applications it is often more convieniant to check these conditions.

Definition 3.83
Let C and D be model categories.

(i) A Quillen adjunction between C and D is an adjunction

C(E(X), Y ) = D(X, G(Y ))

subject to the following (by Remark 3.24 equivalent) hypotheses.


• E preserves trivial cofibrations and cofibrations.
• G preserves trivial fibrations and fibrations.

(ii) A Quillen equivalence is a Quillen adjunction inducing an equivalence between


the homotopy categories.

130
3.14. Derived functors and the comparison of model categories

Example 3.84
Using the geometric realization and singular nerve adjunction of Proposition 2.28

T op(|X|, Y ) = sSet(X, S(Y ))

one can also construct a model structure on the category of topological spaces, induced
by the model structure on simplicial sets. More precisely for a map of topological spaces
f ∈ T op(X, Y ) we define

• f is a weak equivalence, if S(f ) is a weak equivalence,

• f is a fibration, if S(f ) is a fibration,

• f is a cofibraiton, if f ∈ LLP (fib T op ∩ wT op).

Then by Theorem 3.81 and the subsequent Remark the total derived functors exist and
form an adjunction
Ho(T op)(L|X|, Y ) = Ho(X, RS(Y )).
One can moreover prove that also |-| preserves fibrations, which is quite unusual for a
left adjoint in a Quillen adjunction and therefore not so easy. It follows that the Quillen
adjunction is infact a Quillen equivalence.
Moreover one can prove that a continuous map f is a weak equivalence if and only if
∼ ∼
π0 f : π0 X −→ π0 Y, πn f : πn (X, x) −→ πn (Y, f (x)), x ∈ X, n > 0.

131

You might also like