PREFACE

The origin of simplicial homotopy theory coincides with the beginning of algebraic topology almost a century ago. The thread of ideas started with the work
of Poincare and continued to the middle part of the 20th century in the form
of combinatorial topology. The modern period began with the introduction of
the notion of complete semi-simplicial complex, or simplicial set, by EilenbergZilber in 1950, and evolved into a full blown homotopy theory in the work of
Kan, beginning in the 1950s, and later Quillen in the 1960s.
The theory has always been one of simplices and their incidence relations,
along with methods for constructing maps and homotopies of maps within these
constraints. As such, the methods and ideas are algebraic and combinatorial and,
despite the deep connection with the homotopy theory of topological spaces, exist completely outside any topological context. This point of view was effectively
introduced by Kan, and later encoded by Quillen in the notion of a closed model
category. Simplicial homotopy theory, and more generally the homotopy theories
associated to closed model categories, can then be interpreted as a purely algebraic enterprise, which has had substantial applications throughout homological
algebra, algebraic geometry, number theory and algebraic K-theory. The point
is that homotopy is more than the standard variational principle from topology
and analysis: homotopy theories are everywhere, along with functorial methods
of relating them.
This book is, however, not quite so cosmological in scope. The theory has
broad applications in many areas, but it has always been quite a sharp tool
within ordinary homotopy theory it is one of the fundamental sources of
positive, qualitative and structural theorems in algebraic topology. We have
concentrated on giving a modern account of the basic theory here, in a form
that could serve as a model for corresponding results in other areas.
This book is intended to fill an obvious and expanding gap in the literature.
The last major expository pieces in this area, namely [33], [67], [61] and [18],
are all more than twenty-five years old. Furthermore, none of them take into
account Quillens ideas about closed model structures, which are now part of the
foundations of the subject.
We have attempted to present an account that is as linear as possible and
inclusive within reason. We begin in Chapter I with elementary definitions and
examples of simplicial sets and the simplicial set category S, classifying objects,
Kan complexes and fibrations, and then proceed quickly through much of the
classical theory to proofs of the fundamental organizing theorems of the subject
i

ii

Preface

which appear in Section 11. These theorems assert that the category of simplicial
sets satisfies Quillens axioms for a closed model category, and that the associated
homotopy category is equivalent to that arising from topological spaces. They
are delicate but central results, and are the basis for all that follows.
Chapter I contains the definition of a closed model category. The foundations
of abstract homotopy theory, as given by Quillen, start to appear in the first
section of Chapter II. The simplicial model structure that most of the closed
model structures appearing in nature exhibit is discussed in Sections 2-7. A simplicial model structure is an enrichment of the underlying category to simplicial
sets which interacts with the closed model structure, like function spaces do for
simplicial sets; the category of simplicial sets with function spaces is a standard
example. Simplicial model categories have a singular technical advantage which
is used repeatedly, in that weak equivalences can be detected in the associated
homotopy category (Section 4). There is a detection calculus for simplicial model
structures which leads to homotopy theories for various algebraic and diagram
theoretic settings: this is given in Sections 5-7, and includes a discussion of cofibrantly generated closed model categories in Section 6 it may be heavy going
for the novice, but homotopy theories of diagrams almost characterize work in
this area over the past ten years, and are deeply implicated in much current
research. The chapter closes on a much more elementary note with a description
of Quillens non-abelian derived functor theory in Section 8, and a description of
proper closed model categories, homotopy cartesian diagrams and glueing and
coglueing lemmas in Section 9. All subsequent chapters depend on Chapters I
and II.
Chapter III is a further repository of things that are used later, although perhaps not quite so pervasively. The fundamental groupoid is defined in Chapter
I and then revisited here in Section III.1. Various equivalent formulations are
presented, and the resulting theory is powerful enough to show, for example,
that the fundamental groupoid of the classifying space of a small category is
equivalent to the free groupoid on the category, and give a quick proof of the
Van Kampen theorem. The closed model structure for simplicial abelian groups
and the Dold-Kan correspondence relating simplicial abelian groups to chain
complexes (ie. theyre effectively the same thing) are the subject of Section
2. These ideas are the basis of most applications of simplicial homotopy theory and of closed model categories in homological algebra. Section 3 contains
a proof of the Hurewicz theorem: Moore-Postnikov towers are introduced here
in a self-contained way, and then treated more formally in Chapter VII. Kans
Ex -functor is a natural, combinatorial way of replacing a simplicial set up to
weak equivalence by a Kan complex: we give updated proofs of its main properties in Section 4, involving some of the ideas from Section 1. The last section

Preface

iii

presents the Kan suspension, which appears later in Chapter V in connection

with the loop group construction.
Chapter IV discusses the homotopy theory, or more properly homotopy theories, for bisimplicial sets and bisimplicial abelian groups, with major applications. Basic examples and constructions, including homotopy colimits and the
diagonal complex, appear in the first section. Bisimplicial abelian groups, the
subject of Section 2, are effectively bicomplexes, and hence have canonical associated spectral sequences. One of the central technical results is the generalized
Eilenberg-Zilber theorem, which asserts that the diagonal and total complexes
of a bisimplicial abelian group are chain homotopy equivalent. Three different
closed model structures for bisimplicial sets, all of which talk about the same
homotopy theory, are discussed in Section 3. They are all important, and in fact
used simultaneously in the proof of the Bousfield-Friedlander theorem in Section
4, which gives a much used technical criterion for detecting fibre sequences arising from maps of bisimplicial sets. There is a small technical innovation in this
proof, in that the so-called -Kan condition is formulated in terms of certain
fibred group objects being Kan fibrations. The chapter closes in Section 4 with
proofs of Quillens Theorem B and the group completion theorem. These results are detection principles for fibre sequences and homology fibre sequences
arising from homotopy colimits, and are fundamental for algebraic K-theory and
stable homotopy theory.
From the beginning, we take the point of view that simplicial sets are usually
best viewed as set-valued contravariant functors defined on a category of
ordinal numbers. This immediately leads, for example, to an easily manipulated
notion of simplicial objects in a category C: theyre just functors op C,
so that morphisms between them become natural transformations, and so on.
Chapter II contains a detailed treatment of the question of when the category
sC of simplicial objects in C has a simplicial model structure.
Simplicial groups is one such category, and is the subject of Chapter V. We
establish, in Sections 5 and 6, the classical equivalence of homotopy theories
between simplicial groups and simplicial sets having one vertex, from a modern
perspective. The method can the be souped up to give the Dwyer-Kan equivalence between the homotopy theories of simplicial groupoids and simplicial sets
in Section 7. The techniques involve a new description of principal G-fibrations,
for simplicial groups G, as cofibrant objects in a closed model structure on the
category of G-spaces, or simplicial sets with G-action (Section 2). Then the
classifying space for G is the quotient by the G-action of any cofibrant model of
a point in the category of G-spaces (Section 3); the classical W G construction
is an example, but the proof is a bit interesting. We give a new treatment of
W G as a simplicial object of universal cocycles in Section 4; one advantage of

iv

Preface

this method is that there is a completely analogous construction for simplicial

groupoids, which is used for the results of Section 7. Our approach also depends
on a specific closed model structure for simplicial sets with one vertex, which is
given in Section 6. That same section contains a definition and proof of the main
properties of the Milnor F K-construction, which is a functor taking values in
simplicial groups that gives a model for loops suspension X of a given space
X.
The first section of Chapter V contains a discussion of skeleta in the category of
simplicial groups which is later used to show the technical (and necessary) result
that the Kan loop group functor outputs cofibrant simplicial groups. Skeleta
for simplicial sets first appear in a rather quick and dirty way in Section I.2.
Skeleta for more general categories appear in various places: we have skeleta for
simplicial groups in Chapter V, skeleta for bisimplicial sets in Section IV.3, and
then skeleta for simplicial objects in more general categories later, in Section
VII.1. In all cases, skeleta and coskeleta are left and right adjoints of truncation
functors.
Chapters VI and VII contain material on towers of fibrations, nilpotent spaces,
cosimplicial spaces and associated spectral sequences that was introduced by
Bousfield and Kan in [14].
The first section of Chapter VI describes a simplicial model structure for towers, which is used in Section 2 as a context for a formal discussion of Postnikov
towers. The Moore-Postnikov tower, in particular, is a tower of fibrations that
is functorially associated to a space X; we show, in Sections 3-5, that the fibrations appearing in the tower are homotopy pullbacks along maps, or k-invariants,
taking values in homotopy colimits of diagrams of Eilenberg-Mac Lane spaces,
which diagrams are functors defined on the fundamental groupoid of X. The
homotopy pullbacks can be easily refined if the space is nilpotent, as is done in
Section 6. The development includes an introduction of the notion of covering
system of a connected space X, which is a functor defined on the fundamental
groupoid and takes values in spaces homotopy equivalent to the covering space
of X.
Chapter VII contains a detailed treatment of the well known homotopy spectral sequence of a cosimplicial space. The method is to show that the category
of simplicial objects in the category Sop has a Reedy model structure, along
with an adequate notion of skeleta and an appropriate analogue of realization
(Sections 1-3), and then reverse all the arrows to give the usual tower of fibrations from [14] in Section 5, after a general discussion of the model category
structure of cosimplicial spaces in Section 4. The standard method of extracting a homotopy spectral sequence from a tower of fibrations (and the associated
convergence difficulty) is explained in Section 6 and then deployed in Section 7

Preface

to produce the homotopy spectral sequence of a cosimplicial space BousfieldKan p-completions appear in the examples, along with the spectral sequence for
a homotopy inverse limit.
The homotopy spectral sequence of a cosimplicial space is well known to be
fringed in the sense that the objects that appear along the diagonal in total
degree 0 are sets rather than groups. Standard homological techniques therefore
fail, and there can be substantial difficulty in analyzing the path components
of the total space. Bousfield has created an obstruction theory to attack this
problem. We give here, in Section VII.8, a special case of this theory, which
deals with the question of when elements in bidegree (0, 0) in the E2 -term lift
to path components of the total space. This particular result can be used to
give a criterion for maps between mod p cohomology objects in the category of
unstable algebras over the Steenrod algebra to lift to maps of p-completions.
Simplicial model structures return with a vengeance in Chapter VIII, in the
context of homotopy coherence. The point of view that we take is that a homotopy coherent diagram on a catgeory I in simplicial sets is a functor X : A S
which is defined on a category enriched in simplicial sets and preserves the enriched structure, subject to the object A being a resolution of I in a suitable
sense. The main results are due to Dwyer and Kan: there is a simplicial model
structure on the category of simplicial functors SA (Section 1), and a large class
of simplicial functors f : A B which are weak equivalences induce equivalences
of the homotopy categories associated to SA and SB (Section 2). Among such
weak equivalences are resolutions A I in practice, I is the category of
path components of A and each component of A is contractible. A realization
of a homotopy coherent diagram X : A S is then nothing but a diagram
Y : I S which represents X under the equivalence of homotopy categories.
This approach subsumes the standard homotopy coherence phenomena, which
are discussed in Section 3. We show how to promote some of these ideas to
notions of homotopy coherent diagrams and realizations of same in more general
simplicial model categories, including chain complexes and spectra, in the last
section.
Frequently, one wants to take a given space and produce a member of a class
of spaces for which homology isomorphisms are homotopy equivalences, without perturbing the homology. If the homology theory is mod p homology, the
p-completion works in many but not all examples. Bousfields mod p homology localization technique just works, for all spaces. The original approach to
homology localization [8] appeared in the mid 1970s, and has since been incorporated into a more general theory of f -localization. The latter means that one
constructs a minimal closed model structure in which a given map f becomes
invertible in the homotopy category in the case of homology localization the

vi

Preface

map f would be a disjoint union of maps of finite complexes which are homology isomorphisms. The theory of f -localization and the ideas underlying it are
broadly applicable, and are still undergoing frequent revision in the literature.
We present one of the recent versions of the theory here, in Sections 1-4 of Chapter IX. The methods of proof involve little more than aggressive cardinal counts
(the cogniscenti will note that there is no mention of regular cardinals): this is
where the applicability of the ideas comes from morally, if cardinality counts
are available in a model category, then it admits a theory of localization.
Some applications and related ideas are presented in the final two sections of
Chapter IX. We formally introduce homotopy inverse limits in Section 5 and show
that they can be recovered as inverse limits of fibrant objects within appropriate
(localized) closed model structures for diagram categories, and then show that
such closed model structures themselves admit f -localizations. In Section 6 we
describe Bousfields approach to localizing at a functor, and then show that it
leads to the Bousfield-Friedlander model for the stable category. We also show
that the stable category can be derived as a type of f -localization.
There are nine chapters in all; we use roman numerals to distinguish them.
Each chapter is divided into sections, plus an introduction. Results and equations
are numbered consecutively within each section. The overall referencing system
for the monograph is perhaps best illustrated with an example: Lemma 9.8 lives
in Section 9 of Chapter II it is referred to precisely this way from within
Chapter II, and as Lemma II.9.8 from outside. Similarly, the corresponding
section is called Section 9 inside Chapter II and Section II.9 from without.
Despite the length of this tome, much important material has been left out:
there is not a word about traditional simplicial complexes and the vast modern literature related to them (trees, Tits buildings, Quillens work on posets);
the Waldhausen subdivision is not mentioned; we dont discuss the HaussmanHusemoller theory of acyclic spaces or Quillens plus construction; we have
avoided all of the subtle aspects of categorical coherence theory, and there is
very little about simplicial sheaves and presheaves. All of these topics, however, are readily available in the literature we have tried to include a useful
bibiliography.
This book should be accessible to mathematicians in the second year of graduate school or beyond, and is intended to be of interest to the research worker
who wants to apply simplicial techniques, for whatever reason. We believe that
it will be a useful introduction both to the theory and the current literature.
The sensitive reader may already have correctly observed that this monograph
does not have the structure of a traditional text book. We have, for example,
declined to assign homework in the form of exercises, preferring instead to lib-

Preface

vii

erally sprinkle the text with examples and remarks that are designed to provoke
further thought. Everything here depends on the first two chapters; the remaining material often reflects the original nature of the project, which amounted to
separately written self contained tracts on defined topics. The book achieved its
current more unified state thanks to a drive to achieve consistent notation and
referencing. It nevertheless remains true that somebody with a bit of experience
in the area should be able to read each of the later chapters in isolation, and
find an essentially complete story.
Work on this monograph was partly supported by grants from the National
Science Foundation and the Natural Sciences and Engineering Research Council
of Canada, as well as a NATO Collaborative Research grant. We would like
to thank all of these agencies for their support. Work on the project involved
multiple visits by each author to the others home institution, and we would
jointly like to thank the University of Washington and the University of Western
Ontario for their hospitality.

CONTENTS
Chapter I Simplicial sets . . . . . . . . . . . . . . . . . . . . . 1
1. Basic definitions . . . . .
2. Realization . . . . . . .
3. Kan complexes . . . . . .
4. Anodyne extensions . . . .
5. Function complexes . . . .
6. Simplicial homotopy . . .
7. Simplicial homotopy groups
8. Fundamental groupoid . .
9. Categories of fibrant objects
10. Minimal fibrations . . . .
11. The closed model structure

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.

. . 3
. . 7
. 10
. 15
. 21
. 24
. 26
. 38
. 42
. 50
. 64

Chapter II Model Categories . . . . . . . . . . . . . . . . . .

1.
2.
3.
4.
5.
6.
7.
8.
9.

Homotopical algebra . . . . . . . . .
Simplicial categories . . . . . . . . .
Simplicial model categories . . . . . .
Detecting weak equivalences . . . . . .
The existence of simplicial model category
Examples of simplicial model categories .
A generalization of Theorem 5.1 . . . .
Quillens total derived functor theorem .
Homotopy cartesian diagrams . . . . .

. . . . .
. . . . .
. . . . .
. . . . .
structures
. . . . .
. . . . .
. . . . .
. . . . .

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

70
71
83
91
99
102
107
114
122
128

Chapter III Classical results and constructions . . . . . . . . . 151

1.
2.
3.
4.
5.

The fundamental groupoid, revisited .

Simplicial abelian groups: the Dold-Kan
The Hurewicz map . . . . . . . . .
The Ex functor . . . . . . . . . .
The Kan suspension . . . . . . . .

Chapter IV Bisimplicial sets

. . . . . . .
correspondence
. . . . . . .
. . . . . . .
. . . . . . .

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

152
158
177
191
199

. . . . . . . . . . . . . . . . . . 206

1. Bisimplicial sets: first properties . . . . . . . . . . . . . . . . 207

2. Bisimplicial abelian groups . . . . . . . . . . . . . . . . . . 211
2.1. The translation object . . . . . . . . . . . . . . . . . . . . 211
viii

Contents

2.2. The generalized Eilenberg-Zilber theorem

3. Closed model structures for bisimplicial sets
3.1. The Bousfield-Kan structure . . . . . .
3.2. The Reedy structure . . . . . . . . .
3.3. The Moerdijk structure . . . . . . . .
4. The Bousfield-Friedlander theorem . . . .
5. Theorem B and group completion . . . .
5.1. The Serre spectral sequence . . . . . .
5.2. Theorem B . . . . . . . . . . . . . .
5.3. The group completion theorem . . . . .

ix

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.

215
222
222
223
231
234
246
246
249
255

Chapter V Simplicial groups . . . . . . . . . . . . . . . . . . 264

1.
2.
3.
4.
5.
6.
7.

Skeleta . . . . . . . . . . . . . . . . . . .
Principal fibrations I: simplicial G-spaces . . . .
Principal fibrations II: classifications . . . . . .
Universal cocycles and W G . . . . . . . . . .
The loop group construction . . . . . . . . . .
Reduced simplicial sets, Milnors F K-construction
Simplicial groupoids . . . . . . . . . . . . .

Chapter VI The homotopy theory of towers

1.
2.
3.
4.
5.
6.

1.
2.
3.
4.
5.
6.
7.
8.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

265
274
281
285
291
303
313

. . . . . . . . . . 326

A model category structure for towers of spaces

Postnikov towers . . . . . . . . . . . . . .
Local coefficients and equivariant cohomology .
Generalities: equivariant cohomology . . . . .
On k-invariants . . . . . . . . . . . . . .
Nilpotent spaces . . . . . . . . . . . . . .

Chapter VII Cosimplicial spaces

.
.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

327
334
338
347
352
360

. . . . . . . . . . . . . . . . 364

Decomposition of simplicial objects . . . . . . . . .

Reedy model category structures . . . . . . . . . .
Geometric realization . . . . . . . . . . . . . . .
Cosimplicial spaces . . . . . . . . . . . . . . . .
The total tower of a cosimplicial space . . . . . . .
The homotopy spectral sequence of a tower of fibrations
The homotopy spectral sequence of a cosimplicial space
Obstruction theory . . . . . . . . . . . . . . . .

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

365
376
383
386
395
397
408
421

Contents

Chapter VIII Simplicial functors and homotopy coherence

1. Simplicial functors . . . . . . . . . . . . . . . . . . .
2. The Dwyer-Kan theorem . . . . . . . . . . . . . . . .
3. Homotopy coherence . . . . . . . . . . . . . . . . . .
3.1. Classical homotopy coherence . . . . . . . . . . . . .
3.2. Homotopy coherence: an expanded version . . . . . . . .
3.3. Lax functors . . . . . . . . . . . . . . . . . . . . .
3.4. The Grothendieck construction . . . . . . . . . . . . .
4. Realization theorems . . . . . . . . . . . . . . . . . .

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

437
437
444
456
456
458
461
463
465

Chapter IX Localization . . . . . . . .
1. Localization with respect to a map . . .
2. The closed model category structure . .
3. Bousfield localization . . . . . . . . .
4. Localization in simplicial model categories
5. Localization in diagram categories . . .
6. A model for the stable homotopy category

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

472
473
489
498
505
515
522

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

References . . . . . . . . . . . . . . . . . . . . . . . . . . . 534

Chapter I Simplicial sets

This chapter introduces the basic elements of the homotopy theory of simplicial sets. Technically, the purpose is twofold: to prove that the category of
simplicial sets has a homotopical structure in the sense that it admits the structure of a closed model category (Theorem 11.3), and to show that the resulting
homotopy theory is equivalent in a strong sense to the ordinary homotopy theory
of topological spaces (Theorem 11.4). Insofar as simplicial sets are algebraically
defined, and the corresponding closed model structure is combinatorial in nature,
we obtain an algebraic, combinatorial model for standard homotopy theory.
The substance of Theorem 11.3 is that we can find three classes of morphisms
within the simplicial set category S, called cofibrations, fibrations and weak
equivalences, and then demonstrate that the following list of properties is satisfied:
CM1: S is closed under all finite limits and colimits.
CM2: Suppose that the following diagram commutes in S:
g

XN

N
N
P
h N

wY

Z.
If any two of f , g and h are weak equivalences, then so is the third.
CM3: If f is a retract of g and g is a weak equivalence, fibration or cofibration,
then so is f .
CM4: Suppose that we are given a commutative solid arrow diagram

U
i

wX
i
ij

u i
V

p
u
wY

where i is a cofibration and p is a fibration. Then the dotted arrow exists,

making the diagram commute, if either i or p is also a weak equivalence.
1

I. Simplicial sets

CM5: Any map f : X Y may be factored:

(a) f = p i where p is a fibration and i is a trivial cofibration, and
(b) f = q j where q is a trivial fibration and j is a cofibration.
The fibrations in the simplicial set category are the Kan fibrations, which are
defined by a lifting property that is analogous to the notion of a Serre fibration.
The cofibrations are the monomorphisms, and the weak equivalences are morphisms which induce homotopy equivalences of CW-complexes after passage to
topological spaces. We shall begin to investigate the consequences of this list of
axioms in subsequent chapters they are the basis of a great deal of modern
homotopy theory.
Theorem 11.3 and Theorem 11.4 are due to Quillen [75], but the development
given here is different: the results are proved simultaneously, and their proofs
jointly depend fundamentally on Quillens later result that the realization of a
Kan fibration is a Serre fibration [76]. The category of simplicial sets is historically the first full algebraic model for homotopy theory to have been found, but
the verification of its closed model structure is still the most difficult proof of abstract homotopy theory. These theorems and their proofs effectively summarize
all of the classical homotopy theory of simplicial sets, as developed mostly by
Kan in the 1950s. Kans work was a natural outgrowth of the work of Eilenberg
and Mac Lane on singular homology theory, and is part of a thread of ideas that
used to be called combinatorial homotopy theory and which can be traced
back to the work of Poincare at the beginning of the twentieth century.
We give here, in the proof of the main results and the development leading to them, a comprehensive introduction to the homotopy theory of simplicial sets. Simplicial sets are defined, with examples, in Section 1, the functorial relationship with topological spaces via realization and the singular functor
is described in Section 2, and we start to describe the combinatorial homotopical structure (Kan fibrations and Kan complexes) in Section 3. We introduce the Gabriel-Zisman theory of anodyne extensions in Section 4: this is the
obstruction-theoretic machine that trivializes many potential difficulties related
to the function complexes of Section 5, the notion of simplicial homotopy in Section 6, and the discussion of simplicial homotopy groups for Kan complexes in
Section 7. The fundamental groupoid for a Kan complex is introduced in Section
8, by way of proving a major result about composition of simplicial sets maps
which induce isomorphisms in homotopy groups (Theorem 8.2). This theorem,
along with a lifting property result for maps which are simultaneously Kan fibrations and homotopy groups isomorphisms (Theorem 7.10 later strengthened
in Theorem 11.2), is used to demonstrate in Section 9 (Theorem 9.1) that the
collection of Kan complexes and maps between them satisfies the axioms for a

1. Basic definitions

category of fibrant objects in the sense of Brown [15]. This is a first axiomatic
approximation to the desired closed model structure, and is the platform on
which the relation with standard homotopy theory is constructed with the introduction of minimal fibrations in Section 10. The basic ideas there are that every
Kan fibration has a minimal model (Proposition 10.3 and Lemma 10.4), and
the Gabriel-Zisman result that minimal fibrations induce Serre fibrations after
realization (Theorem 10.9). It is then a relatively simple matter to show that
the realization of a Kan fibration is a Serre fibration (Theorem 10.10).
The main theorems are proved in the final section, but Section 10 is the heart of
the matter from a technical point of view once all the definitions and elementary
properties have been established. We have not heard of a proof of Theorem 11.3
or Theorem 11.4 that avoids minimal fibrations. The minimality concept is very
powerful wherever it appears (any rational homotopy theorist would agree), but
not much has yet been made of it from a formal point of view.
1. Basic definitions.
Let be the category of finite ordinal numbers, with order-preserving maps
between them. More precisely, the objects for consist of elements n, n 0,
where n is a string of relations
0 1 2 n
(in other words n is a totally ordered set with n + 1 elements). A morphism
: m n is an order-preserving set function, or alternatively just a functor.
We usually commit the abuse of saying that is the ordinal number category.
A simplicial set is a contravariant functor X : op Sets, where Sets is the
category of sets.
Example 1.1. There is a standard covariant functor
Top.
n7|n |

The topological standard n-simplex |n | Rn+1 is the space

n

n+1

| | = {(t0 , . . . , tn ) R

n
X

ti = 1, ti 0},

i=0

with the subspace topology. The map : |n | |m | induced by : n m

is defined by
(t0 , . . . , tm ) = (s0 , . . . , sn ),

I. Simplicial sets

where

si =

0
P

j 1 (i) tj

1 =
1 (i) 6=

One checks that 7 is indeed a functor (exercise). Let T be a topological

space. The singular set S(T ) is the simplicial set given by
n 7 hom(|n |, T ).
This is the object that gives the singular homology of the space T.
Among all of the functors m n appearing in there are special ones,
namely
di : n 1 n
j

s :n+1n

0in

(cofaces)

0jn

(codegeneracies)

where, by definition,
di (0 1 n 1) = (0 1 i 1 i + 1 n)
(ie. compose i 1 i i + 1, giving a string of arrows of length n 1 in n),
and
1
sj (0 1 n + 1) = (0 1 j
j n)
(insert the identity 1j in the j th place, giving a string of length n + 1 in n). It
is an exercise to show that these functors satisfy a list of identities as follows,
called the cosimplicial identities:

(1.2)

j i
d d = di dj1

j i
i j1

s d =ds
sj dj = 1 = sj dj+1

sj di = di1 sj

j i
s s = si sj+1

if i < j
if i < j
if i > j + 1
if i j

The maps dj , si and these relations can be viewed as a set of generators and
relations for (see [66]). Thus, in order to define a simplicial set Y, it suffices
to write down sets Yn , n 0 (sets of n-simplices) together with maps
di : Yn Yn1 ,
sj : Yn Yn+1 ,

0 i n (faces)
0 j n (degeneracies)

1. Basic definitions

satisfying the simplicial identities:

(1.3)

di dj = dj1 di

di sj = sj1 di
dj sj = 1 = dj+1 sj

d s = sj di1

i j
si sj = sj+1 si

if i < j
if i < j
if i > j + 1
if i j

This is the classical way to write down the data for a simplicial set Y .
From a simplicial set Y, one may construct a simplicial abelian group ZY (ie.
a contravariant functor op Ab), with ZYn set equal to the free abelian group
on Yn . ZY has associated to it a chain complex, called its Moore complex and
also written ZY, with

ZY0
ZY1
ZY2
...
=

n
X

and

(1)i di

i=0

in degree n. Recall that the integral singular homology groups H (X; Z) of the
space X are defined to be the homology groups of the chain complex ZSX.
Example 1.4. Suppose that C is a (small) category. The classifying space (or
nerve ) BC of C is the simplicial set with
BCn = homcat (n, C),
where homcat (n, C) denotes the set of functors from n to C. In other words an
n-simplex is a string
1

a0 a1 . . . an
of composeable arrows of length n in C.
We shall see later that there is a topological space |Y | functorially associated
to every simplicial set Y, called the realization of Y. The term classifying space
for the simplicial set BC is therefore something of an abuse one really means
that |BC| is the classifying space of C. Ultimately, however, it does not matter;
the two constructions are indistinguishable from a homotopy theoretic point of
view.

I. Simplicial sets

Example 1.5. If G is a group, then G can be identified with a category (or

groupoid) with one object and one morphism g : for each element g
of G, and so the classifying space BG of G is defined. Moreover |BG| is an
Eilenberg-Mac Lane space of the form K(G, 1), as the notation suggests; this is
now the standard construction.
Example
of finitely
associated
of QA are

1.6. Suppose that A is an exact category, like the category P(R)

generated projective modules on a ring R (see [78]). Then A has
to it a category QA. The objects of QA are those of A. The arrows
equivalence classes of diagrams

where both arrows are parts of exact sequences of A, and composition is represented by pullback. Then Ki1 (A) := i |BQA| defines the K-groups of A for
i 1; in particular i |BQP(R)| = Ki1 (R), the ith algebraic K-group of the
ring R.
Example 1.7. The standard n-simplex, simplicial n in the simplicial set category S is defined by
n = hom ( , n).
In other words, n is the contravariant functor on which is represented by n.
A map f : X Y of simplicial sets (or, more simply, a simplicial map) is
the obvious thing, namely a natural transformation of contravariant set-valued
functors defined on . S will denote the resulting category of simplicial sets and
simplicial maps.
The Yoneda Lemma implies that simplicial maps n Y classify n-simplices
of Y in the sense that there is a natural bijection
homS (n , Y )
= Yn
between the set Yn of n-simplices of Y and the set homS (n , Y ) of simplicial
maps from n to Y (see [66], or better yet, prove the assertion as an exercise).
More precisely, write n = 1n hom (n, n). Then the bijection is given by
associating the simplex (n ) Yn to each simplicial map : n Y. This
means that each simplex x Yn has associated to it a unique simplicial map
x : n Y such that x (n ) = x. One often writes x = x , since its usually
convenient to confuse the two.
n contains subcomplexes n (boundary of n ) and nk , 0 k n (k th horn,
really the cone centred on the k th vertex). n is the smallest subcomplex of

2. Realization

n containing the faces dj (n ), 0 j n of the standard simplex n . One finds

that
n
if 0 j n 1,

j
n
iterated degeneracies of elements of nk ,
j =

0 k n 1, if j n.
It is a standard convention to write 0 = , where is the unique simplicial
set which consists of the empty set in each degree. is an initial object for the
simplicial set category S.
The k th horn nk n (n 1) is the subcomplex of n which is generated by
all faces dj (n ) except the k th face dk (n ). One could represent 20 , for example,
by the picture

0N

N
N
P
2

0N

N
N
P

2 .

w2

2. Realization.
Let Top denote the category of topological spaces. To go further, we have to
get serious about the realization functor | | : S Top. There is a quick way to
construct it which uses the simplex category X of a simplicial set X. The
objects of X are the maps : n X, or simplices of X. An arrow of
X is a commutative diagram of simplicial maps
n hh

h
h
j

)
'
'
u ''
m

Observe that is induced by a unique ordinal number map : m n.

Lemma 2.1. There is an isomorphism
X
=

lim
n .

n X
in X

I. Simplicial sets

Proof: The proof is easy; it is really the observation that any functor C Sets,
which is defined on a small category C, is a colimit of representable functors.
The realization |X| of a simplicial set X is defined by the colimit
|X| =

lim
|n |.

n X
in X

in the category of topological spaces. |X| is seen to be functorial in simplicial

sets X, by using the fact that any simplicial map f : X Y induces a functor
f : X Y in the obvious way, by composition with f .
Proposition 2.2. The realization functor is left adjoint to the singular functor
in the sense that there is an isomorphism
homS (X, SY )
homTop (|X|, Y ) =
which is natural in simplicial sets X and topological spaces Y .
Proof: There are isomorphisms
homTop (|X|, Y )
= lim homTop (|n |, Y )

n
X

= lim homS (n , S(Y ))

n
X

= homS (X, SY )).

Note that S has all colimits and the realization functor | | preserves them,
since it has a right adjoint.
Proposition 2.3. |X| is a CW-complex for each simplicial set X.
Proof: Define the nth skeleton skn X of X be the subcomplex of X which is
generated by the simplices of X of degree n. Then
[
X=
skn X,
n0

and there are pushout diagrams

G

xN Xn

w skn1 X
z

Gu
xN Xn

w skn X

2. Realization

of simplicial sets, where N Xn Xn is the set of non-degenerate simplices of

degree n. In other words,
N Xn = {x Xn |x is not of the form si y for any 0 i n 1 and y Xn1 }.
The realization of n is the space |n |, since n has terminal object
1 : n n . Furthermore, one can show that there is a coequalizer
n
G
G
n2
n1 n
i=0

0i<jn

given by the relations dj di = di dj1 if i < j (exercise), and so there is a coequalizer diagram of spaces
n
G
G
|n2 |
|n1 | |n |
i=0

0i<jn
n

Thus, the induced map | | | | maps |n | onto the (n 1)-sphere

bounding |n |. It follows that |X| is a filtered colimit of spaces | skn X| where
| skn X| is obtained from | skn1 X| by attaching n-cells according to the pushout
diagram
G
w | skn1 X|
|n |
z
xN Xn
z

u
u
G
w | skn X|.
|n |
xN Xn

In particular |X| is a compactly generated Hausdorff space, and so the realization functor takes values in the category CGHaus of all such. We shall interpret
| | as such a functor. Here is the reason:
Proposition 2.4. The functor | | : S CGHaus preserves finite limits.
We wont get into the general topology involved in proving this result; a
demonstration is given in [33]. Proposition 2.4 avoids the problem that |X Y |
may not be homeomorphic to |X| |Y | in general in the ordinary category of
topological spaces, in that it implies that
|X Y |
= |X| Ke |Y |
(Kelley space product = product in CGHaus). We lose no homotopical information by working CGHaus since, for example, the definition of homotopy
groups of a CW-complex does not see the difference between Top and CGHaus.

10

I. Simplicial sets

3. Kan complexes.
Recall the presentation
G

n
G

n2

n1 n

i=0

0i<jn

of n that was mentioned in the last section. There is a similar presentation

for nk .
Lemma 3.1. The fork defined by the commutative diagram
dj1

n2

w n1

ini<j

di

ini

G u

ww

n2

0i<jn

G u
n1
i6=k

ini<j

w nk

j
d

u
inj

n2

w n1

is a coequalizer in S.
Proof: There is a coequalizer
G
n1 nk n1
i<j

n1 nk .

i 6= k
0in

But the fibre product n1 nk n1 is isomorphic to

n1 n n1
= n2
since the diagram
n2

dj1 w n 1

di

u
n1

di

dj

u
wn

3. Kan complexes

11

is a pullback in . In effect, the totally ordered set {0 . . . i . . . j . . . n} is the

intersection of the subsets {0 . . . i . . . n} and {0 . . . j . . . n} of {0 . . . n}, and this
poset is isomorphic to n 2.

The notation {0 . . . i . . . n} means that i isnt there.

Corollary 3.2. The set homS (nk , X) of simplicial set maps from nk to X is
in bijective correspondence with the set of n-tuples (y0 , . . . , yk , . . . , yn ) of (n1)simplices yi of X such that di yj = dj1 yi if i < j, and i, j 6= k.
We can now start to describe the internal homotopy theory carried by S. The
central definition is that of a fibration of simplicial sets. A map p : X Y
of simplicial sets is said to be a fibration if for every commutative diagram of
simplicial set homomorphisms

nk
wX
z

p
i

u
u
n

wY
there is a map : n X (dotted arrow) making the diagram commute. i is
the obvious inclusion of nk in n .
This requirement was called the extension condition at one time (see [58],
[67], for example), and fibrations were (and still are) called Kan fibrations. The
condition amounts to saying that if (x0 . . . x
k . . . xn ) is an n-tuple of simplices
of X such that di xj = dj1 xi if i < j, i, j 6= k, and there is an n-simplex y of
Y such that di y = p(xi ), then there is an n-simplex x of X such that di x = xi ,
i 6= k, and such that p(x) = y. It is usually better to formulate it in terms of
diagrams.
The same language may be used to describe Serre fibrations: a continuous
map of spaces f : T U is said to be a Serre fibration if the dotted arrow exists
in each commutative diagram of continuous maps

|nk |
z
u
|n |

wT

f
u
wU

12

I. Simplicial sets

making it commute. By adjointness 2.2, all such diagrams may be identified

with diagrams
nk
z

w S(T )
]
\
\
S(f )
\
u \
u
n
w S(U ),
so that f : T U is a Serre fibration if and only if S(f ) : S(T ) S(U ) is
a (Kan) fibration. This is partial motivation for the definition of fibration of
simplicial sets. The simplicial set |nk | is a strong deformation retract of |n |,
so that weve proved
Lemma 3.3. For each space X, the map S(X) is a fibration.
is different notation for the simplicial set 0 . It consists of a singleton set in
each degree, and is therefore a terminal object in the category of simplicial sets.
A fibrant simplicial set (or Kan complex) is a simplicial set Y such that the
canonical map Y is a fibration. Alternatively, Y is a Kan complex if and
only if one of the following equivalent conditions is met:
K1: Every map : nk Y may be extended to a map defined on n in the
sense that there is a commutative diagram

nk
z

u
n

wY

K2: For each n-tuple of (n1)-simplices (y0 . . . yk . . . yn ) of Y such that di yj =

dj1 yi if i < j, i, j 6= k, there is an n-simplex y such that di y = yi .
The standard examples of fibrant simplicial sets are singular complexes, as
weve seen, as well as classifying spaces BG of groups G, and simplicial groups.
A simplicial group H is a simplicial object in the category of groups; this means
that H is a contravariant functor from to the category Grp of groups. I
generally reserve the symbol e for the identities of the groups Hn , for all n 0.

3. Kan complexes

13

Lemma 3.4 (Moore). The underlying simplicial set of any simplicial group H
is fibrant.
Proof: Suppose that (x0 , . . . , xk1 , x`1 , x` , . . . , xn ), ` k + 2, is a family of
(n 1)-simplices of H which is compatible in the sense that di xj = dj1 xi for
i < j whenever the two sides of the equation are defined. Suppose that there is
an n-simplex y of H such that di y = xi for i k 1 and i `. Then the family
k1

`1

(e, . . . , e , x`1 d`1 (y 1 ), e, . . . , e)

is compatible, and di (s`2 (x`1 d`1 y 1 )y) = xi for i k 1 and i ` 1. This
is the inductive step in the proof of the Proposition.

Recall that a groupoid is a category in which every morphism is invertible.

Categories associated to groups as above are obvious examples, so that the following result specializes to the assertion that classifying spaces of groups are
Kan complexes.
Lemma 3.5. Suppose that G is a groupoid. Then BG is fibrant.
Proof: If C is a small category, then its nerve BC is a 2-coskeleton in the sense
that the set of simplicial maps f : X BC is in bijective correspondence with
commutative (truncated) diagrams
X2u

f2

w BCu 2

u
X1u

f1

u
w BCu 1

u
X0

f0

u
w BC0

in which the vertical maps are the relevant simplicial structure maps. It suffices
to prove this for X = n since X is a colimit of simplices. But any simplicial
map f : n BC can be identified with a functor f : n C, and this functor is
completely specified by its action on vertices (f0 ), and morphisms (f1 ), and the
requirement that f respects composition (f2 , and di f2 = f1 di ). Another way of
saying this is that a simplicial map X BC is completely determined by its
restriction to sk2 X.
The inclusion nk n induces an isomorphism
skn2 nk
= skn2 n .

14

I. Simplicial sets

To see this, observe that every simplex of the form di dj n , i < j, is a face of
some dr n with r 6= k: if k 6= i, j use di (dj n ), if k = i use dk (dj n ), and if k = j
use di dk n = dk1 (di n . It immediately follows that the extension problem

nk
y
u
n

w BG

is solved if n 4, for in that case sk2 nk = sk2 n .

Suppose that n = 3, and consider the extension problem

30
y

w BG

u
3

Then sk1 30 = sk1 3 and so we are entitled to write 1 : a0 a1 , 2 : a1 a2

and 3 : a2 a3 for the images under the simplicial map of the 1-simplices
0 1, 1 2 and 2 3, respectively. Write x : a1 a3 for the image of 1 3
under . Then the boundary of d0 3 maps to the graph

a1

w a2
4
43
4
7

a3

in the groupoid G under , and this graph bounds a 2-simplex of BG if and

only if x = 3 2 in G. But the images of the 2-simplices d2 3 and d1 3 under

4. Anodyne extensions

15

together determine a commutative diagram

1 h
h
k
h
a1 4

a0 4

4
42 1
6
4

3 (2 1 )

4
x 44
6

h
h

h 3
k
u h
a3

a2

in G, so that
x1 = 3 (2 1 ),
and x = 3 2 , by right cancellation. It follows that the simplicial map : 30
BG extends to 3 = sk2 3 , and the extension problem is solved.
The other cases corresponding to the inclusions 3i 3 are similar.
If n = 2, then, for example, a simplicial map : 20 BG can be identified
with a diagram
a0 [
[x
1
[
]

a1

a2

and can be extended to a 2-simplex of BG if and only if there is an arrow

2 : a1 a2 of G such that 2 1 = x. But obviously 2 = x11 does the trick.
The other cases in dimension 2 are similar.

The standard n-simplex n = Bn fails to be fibrant for n 2, precisely because

the last step in the proof of Lemma 3.5 fails in that case.
4. Anodyne extensions.
The homotopy theory of simplicial sets is based on the definition of fibration
given above. Originally, all of the work done in this area was expressed in terms
of the extension condition, and some rather grisly-looking combinatorics was
involved. The Gabriel-Zisman theory of anodyne extensions [33] gives a way to
short-circuit most of the pain.

16

I. Simplicial sets

A class M of (pointwise) monomorphisms of S is said to be saturated if the

following conditions are satisfied:
A: All isomorphisms are in M .
B: M is closed under pushout in the sense that, in a pushout square
A

wC

u
B

w B A C,

if i M then so is i (Exercise: Show that i is monic).

C: Each retract of an element of M is in M. This means that, given a diagram
1
A0

u
w A0

wA

i0

i0

u
B0

u
wB

u
w Bu 0 ,

1
if i is in M then i0 is in M .
D: M is closed under countable compositions and arbitrary direct sums,
meaning respectively that:
D1: Given

i1

i2

i3

A1 A2 A3 . . .
with ij M , the canonical map A1 lim Ai is in M .

D2: Given ij : Aj Bj in M , j I, the map

G
G
t ij :
Aj
Bj
jI

is in M .

jI

4. Anodyne extensions

17

A map p : X Y is said to have the right lifting property (RLP is the standard
acronym) with respect to a class of monomorphisms M if in every solid arrow
diagram
A

wX

u i
B

j
ii
i

p
u
wY

with i M the dotted arrow exists making the diagram commute.

Lemma 4.1. The class Mp of all monomorphisms which have the left lifting
property (LLP) with respect to a fixed simplicial map p : X Y is saturated.
Proof: (trivial) For example, we prove the axiom B. Suppose given a commutative diagram
A

wC

wX

p
u
w Y,

u
B

u
w B A C

where the square on the left is a pushout. Then there is a map : B X such
that the composite diagram
A
i

u h
B

wX
h
j
h
h

p
u
wY

commutes. But then induces the required lifting : B A C X by the

universal property of the pushout.

The saturated class MB generated by a class of monomorphisms B is the

intersection of all saturated classes containing B. One also says that MB is
the saturation of B.
Consider the following three classes of monomorphisms:
B1 := the set of all inclusions nk n , 0 k n, n > 0

18

I. Simplicial sets

B2 := the set of all inclusions

(1 n ) ({e} n ) (1 n ),

e = 0, 1

B3 := the set of all inclusions

(1 Y ) ({e} X) (1 X),
where Y X is an inclusion of simplicial sets, and e = 0, 1.
Proposition 4.2. The classes B1 , B2 and B3 have the same saturation.
MB1 is called the class of anodyne extensions.
Corollary 4.3. A fibration is a map which has the right lifting property with
respect to all anodyne extensions.
Proof of Proposition 4.2: We shall show only that MB2 = MB1 ; it is relatively easy to see that MB2 = MB3 , since a simplicial set X can be built up from
a subcomplex Y by attaching cells. To show that MB2 MB1 , observe that any
saturated set is closed under finite composition. The simplicial set n 1 has
non-degenerate simplices
hj : n+1 n 1 ,

0 j n,

where the hj may be identified with the strings of morphisms

(0, 0)

w (0, 1)

w ...

w (0, j)
u
(1, j)

w ...

w (1, n)

of length n + 1 in n 1 (anything longer must have a repeat). One can show

that there are commutative diagrams

(4.4)

u n

=
n {1} y

d0

w n+1
h0
u
w n 1

u n

=
n {0} y

dn+1

w n+1
hn
u
w n 1

4. Anodyne extensions

(4.5)

di

n
hj1

n1 1
dj+1

n
dj+1
u

n+1

hj

n
hj

w n
hj

u
di 1

19

u
w n 1

if i < j

w n+1
hj+1
u
w n 1
di

w n+1
hj

u
n
w 1

n1 1

if i > j + 1.

di1 1
Moreover dj+1 hj
/ n 1 for j 0 since it projects to n under the projection
n
1
map n . Finally, dj+1 hj is not a face of hi for j i + 1 since it has
vertex (0, j).
Let (n 1 )(i) , i 1 be the smallest subcomplex of n 1 containing
n 1 and the simplices h0 , . . . , hi . Then (n 1 )(n) = n 1 and there
is a sequence of pushouts, each having the form
n+1
i+2
z

\
(d0 hi+1 , . . . , di+2
hi+1 , . . . , dn+1 hi+1 )

w (n 1 )(i)
z
n 1 i 1

u
n+1

u
n
w ( 1 )(i+1)

hi+1

by the observation above.

To see that MB1 MB2 , suppose that k < n, and construct the functors
rk

n
n 1 n,
where i(j) = (j, 1) and rk is defined by the diagram
0
u
0

w1
u
w1

w ...
w ...

w k1
u
w k1

wk
u
wk

wk
u
w k+1

w ...
w ...

wk
u
wn

20

I. Simplicial sets

in n. Then r i = 1n , and r and i induce a retraction diagram of simplicial set

maps
nk
z

w (nk 1 ) (n {0})
z

w nk
z

u
n

u
w 1

u
w n

(apply the classifying space functor B). It follows that the inclusion nk n is
in MB2 if k < n.
Similarly, if k > 0, then the functor vk : n 1 n defined by the diagram
0
u
k

w1
u
wk

w ...
w ...

wk
u
wk

w k+1
u
w k+1

w ...
w ...

wn
u
wn

may be used to show that the inclusion nk n is a retract of

(nk 1 ) (n {1}).
Thus, nk n is in the class MB2 for all n and k.

Corollary 4.6. Suppose that i : K , L is an anodyne extension and that

Y , X is an arbitrary inclusion. Then the induced map
(K X) (L Y ) (L X)
is an anodyne extension.
Proof: The set of morphisms K 0 L0 such that
(K 0 X) (L0 Y ) (L0 X)
is anodyne is a saturated set. Consider the inclusion
(1 Y 0 ) ({e} X 0 ) (1 X 0 )

(Y 0 X 0 )

and the induced inclusion

(((1 Y 0 ) ({e} X 0 )) X) ((1 X 0 ) Y ) z

=u
1
0
((( ((Y X) (X 0 Y ))) ({e} (X 0 X))

w ((1 X 0 ) X)

u=
1 (X 0 X)

This inclusion is anodyne, and so the saturated set in question contains all
anodyne morphisms.

5. Function complexes

21

5. Function complexes.
Let X and Y be simplicial sets. The function complex Hom(X, Y ) is the
simplicial set defined by
Hom(X, Y )n = homS (X n , Y ).
If : m n is an ordinal number map, then the induced function
: Hom(X, Y )n Hom(X, Y )m
is defined by
f

(X n
Y ) 7 (X m X n
Y ).
In other words, one thinks of X n as a cosimplicial space in the obvious
way.
There is an evaluation map
ev : X Hom(X, Y ) Y
defined by (x, f ) 7 f (x, n ). To show, for example, that ev commutes with face
maps dj , one has to check that
f (1 dj )(dj x, n1 ) = dj f (x, n ).
But
f (1 dj )(dj x, n1 ) = f (dj x, dj n ) = dj f (x, n ).
More generally, ev commutes with all simplicial structure maps and is thus a
simplicial set map which is natural in X and Y .
Proposition 5.1 (Exponential Law). The function
ev : homS (K, Hom(X, Y )) homS (X K, Y ),
which is defined by sending the simplicial map g : K Hom(X, Y ) to the
composite
1g

ev

X K X Hom(X, Y ) Y,
is a bijection which is natural in K, X and Y .

22

I. Simplicial sets

Proof: The inverse of ev is the map

homS (X K, Y ) homS (K, Hom(X, Y ))
defined by sending g : X K Y to the map g : K Hom(X.Y ), where, for
x Kn , g (x) is the composite
g

1x

X n X K
Y.

The relation between function complexes and the homotopy theory of simplicial sets is given by
Proposition 5.2. Suppose that i : K , L is an inclusion of simplicial sets and
p : X Y is a fibration. Then the map
(i ,p )

Hom(L, X) Hom(K, X) Hom(K,Y ) Hom(L, Y ),

which is induced by the diagram
Hom(L, X)

u
Hom(K, X)

w Hom(L, Y )
i

u
w Hom(K, Y ),

is a fibration.
Proof: Diagrams of the form
nk
z
u
n

w Hom(L, X)
u

(i , p )

w Hom(K, X) Hom(K,Y ) Hom(L, Y )

may be identified with diagrams

(nk L) (nk K) (n K)
z
j
u
n
L

wX
p
u
wY

by the Exponential Law (Proposition 5.1). But j is an anodyne extension by

4.6, so the required lifting exists.

5. Function complexes

23

Corollary 5.3.
(1) If p : X Y is a fibration, then so is p : Hom(K, X) Hom(K, Y )
(2) If X is fibrant then the induced map i : Hom(L, X) Hom(K, X) is
a fibration.
Proof:
(1) The diagram
Hom(K, Y )

w Hom(K, Y )
u
w

is a pullback, and the following commutes:

Hom(K, X)'
NNN
N

' NNNpN
NNN

N
P
'
'
Hom(K, Y )
w Hom(K, Y )
'
1
'
'
)
u
u
Hom(, X)
w Hom(, Y )
p

for a uniquely determined choice of map Hom(K, Y ) Hom(, X).

(2) There is a commutative diagram
Hom(L, X)'
NNN
N
' NNNNN
NNN

P
'
'
Hom(K, X)
w Hom(L, )
i '
'
'
)
1
u
u
Hom(K, X)
w Hom(K, )

24

I. Simplicial sets
where the inner square is a pullback.

6. Simplicial homotopy.
Let f, g : K X be simplicial maps. We say that there is a simplicial
'

homotopy f g from f to g if there is a commutative diagram

K 0

K4

4f
44
6

1 d1

u
K u 1

wX
j
hh
hg

1 d0

K 0

The map h is called a homotopy.

Its rather important to note that the commutativity of the diagram defining
the homotopy h implies that h(x, 0) = f (x) and h(x, 1) = g(x) for all simplices
x K. We have given a definition of homotopy which is intuitively correct
elementwise it is essentially the reverse of the definition that one is usually
tempted to write down in terms of face (or coface) maps.
Suppose i : L K denotes an inclusion and that the restrictions of f and g to
L coincide. We say that there is a simplicial homotopy from f to g, (rel L) and
'

write f g (rel L), if the diagram exists above, and the following commutes
as well:
h

K u 1
i1

w Xu

L 1

wL

prL

where = f |L = g|L , and prL is projection onto the left factor (prR will denote
projection on the right). A homotopy of the form
prL

L 1 L
X
is called a constant homotopy (at ).

6. Simplicial homotopy

25

The homotopy relation may fail to be an equivalence relation in general. Consider the maps 0 , 1 : 0 n , (n 1), which classify the vertices 0 and 1,
respectively, of n . There is a simplex [0, 1] : 1 n determined by these
'

'

vertices, and so 0 1 (alternatively, 0 1). But there is no 1-simplex which

'

could give a homotopy 1 0 , since 0 1. This observation provides a second

means (see Lemma 3.5) of seeing that n not fibrant, since we can prove
Lemma 6.1. Suppose that X is a fibrant simplicial set. Then simplicial homotopy of vertices x : 0 X of X is an equivalence relation.
'

Proof: There is a homotopy x y if and only if there is a 1-simplex v of

X such that d1 v = x and d0 v = y (alternativley v = (y, x); in general the
boundary of an n-simplex is denoted by = (d0 , . . . , dn )). But then
the equation (s0 x) = (x, x) gives the reflexivity of the homotopy relation.
Suppose that v2 = (y, x) and v0 = (z, y). Then d0 v2 = d1 v0 , and so v0 and
v2 determine a map (v0 , , v2 ) : 21 X in the obvious way. Choose a lifting
21
z

(v0 , , v2 )

N
P
N
N
N
u N
2

w X.

Then
(d1 ) = (d0 d1 , d1 d1 )
= (d0 d0 , d1 d2 )
= (z, x),
and so the relation is transitive. Finally, given v2 = (y, x), set v1 = s0 x. Then
d1 v1 = d1 v2 and so v1 and v2 define a map ( , v1 , v2 ) : 20 X. Choose an
extension
20
z

( , v1 , v 2 )

N
P
N
N0
N
u N
2

w X.

26

I. Simplicial sets

Then
(d0 0 ) = (d0 d0 0 , d1 d0 0 )
= (d0 d1 0 , d0 d2 0 )
= (x, y),
and the relation is symmetric.

Corollary 6.2. Suppose X is fibrant and that L K is an inclusion of simplicial sets. Then
(a) homotopy of maps K X is an equivalence relation, and
(b) homotopy of maps K X (rel L) is an equivalence relation.
Proof: (a) is a special case of (b), with L = . But homotopy of maps K X
(rel L) corresponds to homotopy of vertices in the fibres of the Kan fibration
i : Hom(K, X) Hom(L, X)
via the Exponential Law 5.1.

7. Simplicial homotopy groups.

Let X be a fibrant simplicial set and let v X0 be a vertex of X. Define
n (X, v), n 1, to be the set of homotopy classes of maps : n X (rel
n ) for maps which fit into diagrams
u n

w Xu
v

z
n

w 0 .

One often writes v : n X for the composition

v

n 0
X.
Define 0 (X) to be the set of homotopy classes of vertices of X. 0 (X) is
the set of path components of X. The simplicial set X is said to be connected
if 0 (X) is trivial (ie. a one-element set). We shall write [] for the homotopy
class of , in all contexts.

7. Simplicial homotopy groups

27

Suppose that , : n X represent elements of n (X, v). Then the simplices

0 i n 2,

vi = v,
vn1 = , and

vn+1 =
satisfy di vj = dj1 vi if i < j and i, j 6= n, since all faces of all simplices
vi map through the vertex v. Thus, the vi determine a simplicial set map
(v0 , . . . , vn1 , , vn+1 ) : n+1
X, and there is an extension
n
n+1
n
z
u

n+1

(v0 , . . . , vn1 , , vn+1 )

w X.
'
)
'
'
''
'

'
''

Observe that
(dn ) = (d0 dn , . . . , dn1 dn , dn dn )
= (dn1 d0 , . . . , dn1 dn1 , dn dn+1 )
= (v, . . . , v),
and so dn represents an element of n (X, v).
Lemma 7.1. The homotopy class of dn (rel n ) is independent of the choices
of representatives of [] and [] and of the choice of .
'

Proof: Suppose that hn1 is a homotopy 0 (rel n ) and hn+1 is a

'

homotopy 0 (rel n ). Suppose further that

= (v, . . . , v, , dn , )
and
0 = (v, . . . , v, 0 , dn 0 , 0 ).
Then there is a map
(

n+1

(n+1
n

( 0 ,,(v,...,hn1 , ,hn+1 ))

) X

28

I. Simplicial sets

which is determined by the data. Choose an extension

n+1

)
z
u

n+1

( 0 , , (v, . . . , hn1 , , hn+1 ))

)
w
X.

(n+1
n

Then the composite

dn 1

n 1 n+1 1 X
'

is a homotopy dn dn 0 (rel n ).
It follows from Lemma 7.1 that the assignment
([], []) 7 [dn ],

where = (v, . . . , v, , dn , ),

gives a well-defined pairing

m : n (X, v) n (X, v) n (X, v).
Let e n (X, v) be the homotopy class which is represented by the constant
map
v
n 0
X.
Theorem 7.2. With these definitions, n (X, v) is a group for n 1, which is
abelian if n 2.
Proof: We shall demonstrate here that the n (X, v) are groups; the abelian
property for the higher homotopy groups will be proved later.
It is easily seen (exercise) that e = e = for any n (X, v), and that
the map n (X, v) n (X, v) induced by left multiplication by is bijective.
The result follows, then, if we can show that the multiplication in n (X, v) is
associative in general and abelian if n 2.
To see that the multiplication is associative, let x, y, z : n X represent
elements of n (X, v). Choose (n + 1)-simplices n1 , n+1 , n+2 such that
n1 = (v, . . . , v, x, dn n1 , y),
n+1 = (v, . . . , v, dn n1 , dn n+1 , z),
n+2 = (v, . . . , v, y, dn n+2 , z).

and

7. Simplicial homotopy groups

29

Then there is a map

(v,...,v,n1 , ,n+1 ,n+2 )

n+2
X
n
which extends to a map u : n+2 X. But then
(dn ) = (v, . . . , v, x, dn n+1 , dn n+2 ),
and so
([x][y])[z] = [dn n1 ][z]
= [dn n+1 ]
= [dn dn u]
= [x][dn n+1 ]
= [x]([y][z]).

In order to prove that n (X, v) abelian for n 2, it is most instructive to

show that there is a loop-space X such that n (X, v)
= n1 (X, v) and then
to show that i (X, v) is abelian for i 1. This is accomplished with a series of
definitions and lemmas, all of which well need in any case. The first step is to
construct the long exact sequence of a fibration.
Suppose that p : X Y is a Kan fibration and that F is the fibre over a
vertex Y in the sense that the square
F

wX

u
wY

p
u
0

is cartesian. Suppose that v is a vertex of F and that : n Y represents an

element of n (Y, ). Then in the diagram
( , v, . . . , v)
wX
A
C
A
A
p
A
A
u A
u
n

w Y,

n0
z

the element [d0 ] n1 (F, v) is independent of the choice of and representative of []. The resulting function
: n (Y, ) n1 (F, v)
is called the boundary map.

30

I. Simplicial sets

Lemma 7.3.
(a) The boundary map : n (Y, ) n1 (F, v) is a group homomorphism
if n > 1.
(b) The sequence
p

n (F, v) n (X, v) n (Y, )

n1 (F, v) . . .
p

. . . 1 (Y, )
0 (F ) 0 (X) 0 (Y )
is exact in the sense that kernel equals image everywhere. Moreover, there
is an action of 1 (Y, ) on 0 (F ) such that two elements of 0 (F ) have
the same image under i in 0 (X) if and only if they are in the same orbit
for the 1 (Y, )-action.
Most of the proof of Lemma 7.3 is easy, once you know
Lemma 7.4. Let : n X represent an element of n (X, v). Then [] = e if
and only if there is an (n + 1)-simplex of X such that = (v, . . . , v, ).
The proof of Lemma 7.4 is an exercise.
Proof of Lemma 7.3: (a) To see that : n (Y, ) n1 (F, v) is a homomorphism if n 2, suppose that we are given diagrams
v

n0
z

wX

i
p

u
u
n

wY

i = n 1, n, n + 1,

where the i represent elements of n (Y, ). Suppose that there is an (n + 1)simplex such that
= (, . . . , , n1 , n , n+1 ).
Then there is a commutative diagram
( , v, . . . , v, n1 , n , n+1 )
wX

u
u
n+1

w Y,

n+1
0
z

7. Simplicial homotopy groups

31

and
(d0 ) = (d0 d0 , d1 d0 , . . . , dn d0 )
= (d0 d1 , d0 d2 , . . . , d0 dn1 , d0 dn , d0 dn+1 )
= (v, . . . , v, d0 n1 , d0 n , d0 n+1 )
Thus [d0 n ] = [d0 n1 ][d0 n+1 ], and so ([n1 ][n+1 ]) = [n1 ][n+1 ] in
n1 (F, v).
(b) We shall show that the sequence
p

n (X, v) n (Y, )
n1 (F, v)
is exact; the rest of the proof is an exercise. The composite is trivial, since in
the diagram
n0
z

wX

u
u
n

p w Y
with [] n (X, v), we find that d0 = v. On the other hand, suppose that
: n Y represents a class [] such that [] = e. Choose a diagram
n0
z

wX

u
n

p
u
wY

so that [d0 ] = []. But then there is a simplex homotopy

h0

n1 1 F
giving d0 ' v (rel n ). Thus, there is a diagram
(n 1) (n 1 )
z
u
1
n

(, (h0 , v, . . . , v))

w X.

32

I. Simplicial sets

Moreover p h is a homotopy ' p (h d1 ) (rel n ).

Now for some definitions. For a Kan complex X and a vertex of X, the path
space P X is defined by requiring that the following square is a pullback.
pr

PX

w Hom(1 , X)
(d0 )

u
0

u
w Hom( , X)
= X.
0

Furthermore, the map : P X X is defined to be the composite

(d1 )

pr

P X Hom(1 , X) Hom(0 , X)
= X.
The maps (d ) are fibrations for = 0, 1, by 5.3. In particular, P X is fibrant.
Lemma 7.5. i (P X, v) is trivial for i 0 and all vertices v, and is a fibration.
Proof: d : 0 1 is an anodyne extension, and so (d0 ) has the right lifting
property with respect to all maps n n , n 0, (see the argument in 5.2).
Thus, the map P X 0 = has the right lifting property with respect to all
such maps. Any two vertices of P X are homotopic, by finding extensions

1
z
\
u \
1

w P X.
\
]
\

If : n P X represents an element of n (P X, v), then there is a commutative

diagram
n+1
z
u

n+1

(v, . . . , v, )

w P X,
5
6
5
55
5
55

7. Simplicial homotopy groups

33

and so [] = e in n (P X, v). Finally, the map sits inside the pullback diagram
w Hom(1 , X)

PX

u
X

ui
Hom(1 , X)

u=
w X X
(, 1X )

and so is a fibration since i is, by 5.3.

Define the loop space X to be the fibre of : P X X over the base point .
A simplex of X is a simplicial map f : n 1 X such that the restriction
of f to n 1 maps into . Now we can prove
Lemma 7.6. i (X, ) is abelian for i 1.
Proof: n (X, ), as a set, consists of homotopy classes of maps of the form
n u 1

w X,
j
hh
h

(n 1 ) (n 1 )
rel the boundary (n 1 ) (n 1 ). Show that n (X, ) has a second
multiplication [] ? [] (in the 1-simplex direction) such that [] is an identity
for this multiplication and that ? and the original multiplication satisfy the
interchange law
([1 ] ? [1 ])([2 ] ? [2 ]) = ([1 ][2 ]) ? ([1 ][2 ]).
It follows that [][] = [] ? [], and that both multiplications are abelian.

Corollary 7.7. Suppose that X is fibrant. Then i (X, ) is abelian if i 2.

The proof of Theorem 7.2 is now complete.
Let G be a group, and recall that the classifying space BG is fibrant, by 3.5.
BG has exactly one vertex . We can now show easily that BG is an EilenbergMac Lane space.

34

I. Simplicial sets

Proposition 7.8. There are natural isomorphisms

i (BG, )
=

G if i = 1,
0 if i 6= 1.

Proof: BG is a 2-coskeleton (see 3.5), and so i (BG, ) = 0 for i 2, by 7.4.

It is an elementary exercise to check that the identification BG1 = G induces an

isomorphism of groups 1 (BG, ) G. 0 (BG) is trivial, since BG has only

one vertex.

Suppose that f : Y X is a map between fibrant simplicial sets. f is said to

be a weak equivalence if

(7.9)

for each vertex y of Y the induced map f : i (Y, y) i (X, f y)

is an isomorphism for i 1, and

the map f : 0 (Y ) 0 (X) is a bijection.

Theorem 7.10. A map f : Y X between fibrant simplicial sets is a fibration

and a weak equivalence if and only if f has the right lifting property with respect
to all maps n n , n 0.
Proof: () The simplicial homotopy n 1 n , given by the diagram

w0

u
0

u
w1

w ...

w0

w ...

u
wn

in n, contracts n onto the vertex 0. This homotopy restricts to a homotopy

n0 1 n0 which contracts n0 onto 0.
Now suppose that were given a diagram

n
z
i
u
n

wX
p
u
w Y.

7. Simplicial homotopy groups

35

If there is a diagram

n 4
[[[
z
4 [[[
[[[
46
[[
d0 4
]
[
n 1
wX
h
z
p
i1
i
u
u
g
n 1
wY

N
P

d0 NN

u N

n
such that the lifting exists in the diagram
1
n h d w X
z
[
]
[
p
[
u [
u
n

w Y,
g d1

then the lifting exists in the original diagram D. This is a consequence of the
fact that there is a commutative diagram

z
i
u
n

d0

u
(h, )
w ( ) ( {1})
wX
6
4
4
z
44
p
4
4
4
u
u
4
n
1
w

w Yu
g
d0
n

36

I. Simplicial sets

Now, the contracting homotopy H : n0 1 n0 determines a diagram

n0 y

w n

d0
n0

u
1
u

h1

u
w Xu
(0)

d1
n0

w 0 ,

where h1 = j H. There is a diagram

(, h1 )
(n {1}) (n0 1 )
wX
4
6
4
z
4
4
44h
4
u
4
n
1

since X is fibrant. Moreover, there is a diagram
(, ph)
(n {1}) (n 1 )
wY
6
4
4
z
4
4
44 g
4
u
4
n
1

since Y is fibrant. It therefore suffices to solve the problem for diagrams
n
z
D1

(x0 , , . . . , )

wX
p

u
n

u
wY

7. Simplicial homotopy groups

37

for some vertex (= (0)) of X, since the composite diagram

n
z
i
u
n

d1

d1

w n 1
z
i1
u
w n 1

wX

u
wY

has this form. Then x0 represents an element [x0 ] of n1 (X, ) such that
p [x0 ] = e in n1 (Y, p). Thus, [x0 ] = e in n1 (X, ), and so the trivializing homotopy h0 : n1 1 X for X0 determines a homotopy
h0 = (h0 , , . . . , ) : n 1 X.
But again there is a diagram
(, ph)
(n {1}) (n 1 )
w Y,
'
)
z
'
'
''
'
' g0
u
''
n 1
so it suffices to solve the lifting problem for diagrams
n
z

wX
p

D2

u
n

u
w Y.

p is onto, so ' p (rel n ) via some homotopy h00 : n 1 Y, and

so there is a commutative diagram
n
z
u
n

d0

(, )
w (n 1 ) (n {0})
wX
5
6
5
z
5
p
55
5
5
u
u
5
n
1
w
w Y.
h00
d0

38

I. Simplicial sets

D2 is the composite of these two squares, and the lifting problem is solved.
() Suppose that p : X Y has the right lifting property with respect to
all n n , n 0. Then p has the right lifting property with respect to all
inclusions L K, and is a Kan fibration in particular. It is then easy to see
that p : 0 X 0 Y is a bijection. Also, if x X is any vertex of X and Fx
is the fibre over p(x), then Fx has the right lifting property with respect to
all n n , n 0. Then Fx is fibrant, and 0 (Fx ) = and i (Fx , x) = 0,
i 1, by the argument of Lemma 7.5. Thus, p : i (X, x) i (Y, px) is an
isomorphism for all i 1.

8. Fundamental groupoid.
Let X be a fibrant simplicial set. Provisionally, the fundamental groupoid
f X of X is a category having as objects all vertices of X. An arrow x y in
f X is a homotopy class of 1-simplices : 1 X (rel 1 ) where the diagram
0 4

x
4
6
4

d1

u
u 1

d0

wX

h
hj
hy

0
commutes. If v2 represents an arrow x y of f X and v0 represents an arrow
y z; then the composite [v0 ][v2 ] is represented by d1 , where is a 2-simplex
such that the following diagram commutes
21
z

(v0 , , v2 )

u N
2

N
P
N
N
N

w X.

The fact that this is well-defined should be clear. The identity at x is represented
by s0 x. This makes sense because, if v2 : x y and v0 : y z then s0 v0 =
(v0 , v0 , s0 y), and (s1 v2 ) = (s0 y, v2 , v2 ). The associativity is proved as it was

8. Fundamental groupoid

39

for 1 . In fact, f X(x, x) = 1 (X, x) specifies the group of homomorphisms

f X(x, x) from x to itself in f X, by definition. By solving the lifting problem
( , s0 x, v2 )
wX
C
B
B
B
B
B
u B
2
20
z

for v2 : x y, one finds a v0 : y x (namely d0 ) such that [v0 ][v2 ] = 1x .

But then [v2 ] is also epi since it has a right inverse by a similar argument. Thus
[v2 ][v0 ][v2 ] = [v2 ] implies [v2 ][v0 ] = 1y , and so f X really is a groupoid.
Now, let : n X represent an element of n (X, x) and let : 1 X
represent an element of f X(x, y). Then there is a commutative diagram
(n 1 ) (n {0})

u
u 1

h(,)

( prR , )

w X,

d0
n

and represents an element of n (X, y).

Proposition 8.1. The class [ ] is independent of the relevant choices of
representatives. Moreover, [] 7 [ ] is a group homomorphism which is functional in [], and so the assignment x 7 n (X, x) determines a functor on f X.
Proof: We shall begin by establishing independence from the choice of repre'

sentative for the class []. Suppose that G : (rel 1 ) is a homotopy of

paths from x to y. Then there is a 2-simplex of X such that
= (s0 y, , ).
Find simplices of the form h(,) and h(,) according to the recipe given above,
and let h be the composite
pr

X.
n 2 2

40

I. Simplicial sets

There is a commutative diagram

(h , ( , h(, ), h(, )))
(n 2 ) (n 20 )
w X,

1
u

n
2

since the inclusion i is anodyne. Then the composite

1d0

n 1 n 2 X
is a homotopy from to (rel n ), and so [ ] = [ ] in n (X, y).
'

Suppose that H : n 1 X gives (rel n ), and choose a

homotopy h(, ) : n 1 X as above. Let hs0 denote the composite
pr

s0

n 2 2 X.
Then there is a commutative diagram
(hs0 , (h(, ), , H))
(n 2 ) (n 21 )
w X,

n 2

for some map , since the inclusion j is anodyne. But then the simplex given by
the composite
d0

1d1

X
n n 1 n 2
is a construction for both and , so that [ ] = [ ] in n (X, y).
For the functoriality, suppose that : 1 X and : 1 X represent
elements of f X(x, y) and f X(y, z) respectively, and choose a 2-simplex such

8. Fundamental groupoid

41

that = (, d1 , ). Then [d1 ] = [] [] in f X. Choose h(,) and h(, )

according to the recipe above. Then there is a diagram
(n 2 ) (n 21 ) [[
[[[
z
pr, (h(, ), , h(, ))
[[(
[[[
[[
]
[
u
n u 2

X,

1 d1
n 1

and hence a diagram

(n 1 ) (n {0}) A
z
A

(hd1 , )
A
A
C

u
u 1

d1
n

)
'
''

w X,

'
''
'
= (d1 )
''

where hd1 is the composite

d1

pr

n 1 1 X.
The statement that is a group homomorphism is easily checked.

Theorem 8.2. Suppose that the following is a commutative triangle of simplicial set maps:
XN

N
N
P
h N

wY

Z
with X, Y, and Z fibrant. If any two of f, g, or h are weak equivalences, then so
is the third.

42

I. Simplicial sets

Proof: There is one non-trivial case, namely to show that f is a weak equivalence if g and h are. This is no problem at all for 0 . Suppose y Y is a vertex.
We must show that f : n (Y, y) n (Z, f y) is an isomorphism. y may not be
in the image of g, but there is an x X and a path : y gx since 0 (g) is
epi. But then there is a diagram

n (Y, y)

[]

u
n (Z, f y)

[f ]

w n (Y, gx) u

n (X, x).

f
h
u

w n (Z, f gx)

The maps g , h , [] , and [f ] are isomorphisms, and so both of the maps

labelled f are isomorphisms.

There are three competing definitions for the fundamental groupoid of an

arbitrary simplicial set X. The most obvious choice is the classical fundamental
groupoid |X| of the realization of X; in the notation above, this is f S|X|. Its
objects are the elements of |X|, and its morphisms are homotopy classes of paths
in |X|. The second choice is the model GP X of Gabriel and Zisman. GP X is
the free groupoid associated to the path category P X of X. The path category
has, as objects, all the vertices (elements of X0 ) of X. It is generated, as a
category, by the 1-simplices of X, subject to the relation that, for each 2-simplex
of X, the diagram
v0 [
[
]
d1 [

d2

w v1

d0

v2
commutes. The free groupoid G( X) associated to the simplex category
X is also a good model. We shall see later on, in Section 12 that |X|,
GP X and G( X) are all naturally equivalent, once we have developed the
techniques for doing so.
9. Categories of fibrant objects.
Let Sf be the full subcategory of the simplicial set category whose objects are
the Kan complexes. Sf has all finite products. We have two distinguished classes

9. Categories of fibrant objects

43

of maps in Sf , namely the fibrations (defined by the lifting property) and the
weak equivalences (defined via simplicial homotopy groups). A trivial fibration
p : X Y in Sf is defined to be a map which is both a fibration and a weak
equivalence. A path object for X Sf is a commutative diagram

XI
N
P
sNN
(d0 , d1 )
N
u
N
X
w X X

where s is a weak equivalence and (d0 , d1 ) is a fibration. The maps d0 and d1

are necessarily trivial fibrations. Any Kan complex X has a natural choice of
path object, namely the diagram

Hom(1 , X)

'
X

s'
'

)
''

u
Hom(1 , X)

u=
w X X u

(d0 , d1 )

where s is the map

(s0 )

X
= Hom(0 , X) Hom(1 , X).
(s0 ) = s is a weak equivalence; in effect, it is a right inverse for the map
(d0 )

Hom(1 , X) Hom(0 , X),

and (d0 ) is a trivial fibration, by 7.10, since it has the right lifting property
with respect to all inclusions n n , n 0. (d0 ) is isomorphic to one of
the components of the map Hom(1 , X) X X.
The following list of properties of Sf is essentially a recapitulation of things
that weve seen:

44

I. Simplicial sets

(A) Suppose given a commutative diagram

XN

N
N
P
h N

wY

Z.
If any two of f , g and h are weak equivalences, then so is the third.
(B) The composite of two fibrations is a fibration. Any isomorphism is a
fibration.
(C) Pullback diagrams of the form
Z Y X

wX
p

pr
u
Z

u
wY

exist in the case where p is a fibration. Furthermore, pr is a fibration

which is trivial if p is trivial.
(D) For any object X there is at least one path space X I .
(E) For any object X, the map X is a fibration.
Recall that we proved (A) outright in 8.2. (B) is an easy exercise. (C) holds
because fibrations and trivial fibrations are defined by lifting properties, by 7.10.
(D) was discussed above, and (E) isnt really worth mentioning.
Following K. Browns thesis [15] (where the notion was introduced), a category
C which has all finite products and has distinguished classes of maps called
fibrations and weak equivalences which together satisfy axioms (A) (E) is
called a category of fibrant objects (for a homotopy theory). Weve proved:
Theorem 9.1. Sf is a category of fibrant objects for a homotopy theory.
Other basic examples for us are the category CGHaus of compactly generated
Hausdorff spaces, and the category Top of topological spaces. In fact, more is
true. The fibrations of CGHaus are the Serre fibrations, and the weak equivalences are the weak homotopy equivalences. A map i : U V in CGHaus is
said to be a cofibration if it has the left lifting property with respect to all trivial
fibrations.

9. Categories of fibrant objects

45

Proposition 9.2. The category CGHaus and these three classes of maps satisfy the following list of axioms:
CM1: CGHaus is closed under all finite limits and colimits.
CM2: Suppose that the following diagram commutes in CGHaus:
g

XN

N
N
P
h N

wY

Z.

If any two of f , g and h are weak equivalences, then so is the third.

CM3: If f is a retract of g and g is a weak equivalence, fibration or cofibration,
then so is f .
CM4: Suppose that we are given a commutative solid arrow diagram

U
i

wX

u i
V

i
ij

p
u
wY

where i is a cofibration and p is a fibration. Then the dotted arrow exists,

making the diagram commute, if either i or p is also a weak equivalence.
CM5: Any map f : X Y may be factored:
(a) f = p i where p is a fibration and i is a trivial cofibration, and
(b) f = q j where q is a trivial fibration and j is a cofibration.
Proof: The category CGHaus has all small limits and colimits, giving CM1
(see [66, p.182]). This fact is also used to prove the factorization axioms CM5;
this is the next step.
The map p : X Y is a Serre fibration if and only if it has the right lifting
property with respect to all inclusions j : |nk | |n |. Each such j is necessarily

46

I. Simplicial sets

a cofibration. Now consider all diagrams

|nkDD |
z

wX
f

D
u
|nD |

u
w Y

and form the pushout

G
D

(D )

|nkDD |
z

wX

X0'
'

i1

G u
|nD |

u
w X1

f1

f = f0
'
'
)
w Y.

Then we obtain a factorization

f = f0 = f1 i1 ,
where i1 is a cofibration since its a pushout of such, and also a weak equivalence
since it is a pushout of a map which has a strong deformation retraction. We
repeat the process by considering all diagrams.
|nkDD |
z

w X1
f1

u
|

nD

u
w Y

and so on. Thus, we obtain a commutative diagram

X

X0 4

i1

w X1

4
4 f1
6
f0 4

i2

w X2
h

h
h f2
u h
k
Y

w ...,

9. Categories of fibrant objects

47

which induces a diagram

0

X0 [

[
]
[
f = f0 [

w lim Xi

Y
But now 0 has the left lifting property with respect to all trivial fibrations, so
its a cofibration. Moreover, 0 is a weak equivalence since any compact subset of
lim Xi lies in some finite stage Xi , and all the Xi Xi+1 are weak equivalences.

Finally, f is a fibration; in effect, for each diagram

|nk |
z
u
|n |

w lim Xi

f
u
w Y,

there is an index i and a map i making the following diagram commute:

|nk |N
z
N

i
N
N
P

j
hh

w lim Xi

h
Xi

fi

u
|n |

u
w Y.

But then
|nk |
z
u
|n |

w Xi
u
wY

fi

48

I. Simplicial sets

is one of the diagrams defining fi+1 and there is a diagram

i

|nk |
z

w Xi
A
A
A
A
D
fi
Xi+1

i 44
6

u 44
n
| |
wY

which defines the lifting.

The other lifting property is similar, using
Lemma 9.4. The map p : X Y is a trivial fibration if and only if p has the
right lifting property with respect to all inclusions |n | |n |.
The proof is an exercise.
Quillen calls this proof a small object argument [75]. CM4 is really a consequence of this argument as well. What we showed, in effect, was that any map
f : X Y has a factorization f = p i such that p is a fibration and i is a weak
equivalence which has the left lifting property with respect to all fibrations.
Suppose now that we have a diagram

U
i

j
ii

u i
V

wX
p
u
w Y,

where p is a fibration and i is a trivial cofibration. We want to construct the

dotted arrow (giving the non-trivial part of CM4). Then there is a diagram

U
i

u i
V

j
i
j
ri
i
1V

wW

u
w V,

9. Categories of fibrant objects

49

where j is a weak equivalence which has the left lifting property with respect to
all fibrations, and is a (necessarily trivial) fibration. Thus, the dotted arrow r
exists. But then i is a retract of j, and so i has the same lifting property. All of
the other axioms are trivial, and so the proof of Proposition 9.2 is complete.
A closed model category is a category C, together with three classes of maps
called cofibrations, fibrations and weak equivalences, such that the axioms CM1
CM5 are satisfied. Proposition 9.2 is the statement that CGHaus has the
structure of a closed model category.
Proposition 9.5. CGHaus is a category of fibrant objects for a homotopy
theory. In fact the subcategory of fibrant objects in any closed model category
C is a category of fibrant objects for a homotopy theory.
Proof: (E) is part of the definition. For (D), the map : X X X may
be factored
XI

P
N
sNN
(d0 , d1 )
N
u
N
X
w X X,

where s is a trivial cofibration and (d0 , d1 ) is a fibration. For (B) and (C), we
prove:
Lemma 9.6.
(a) A map f : X Y in C has the right lifting property with respect to all
cofibrations (respectively trivial cofibrations) if and only if f is a trivial
fibration (respectively fibration).
(b) U V in C has the left lifting property with respect to all fibrations
(respectively trivial fibratons) if and only if i is a trivial cofibration (respectively cofibration).
Proof: Well show that f : X Y has the right lifting property with respect
to all cofibrations if and only if f is a trivial fibration. The rest of the proof is
an exercise.

50

I. Simplicial sets

Suppose that f has the advertised lifting property, and form the diagram

1X

wX
h
j
rh
f
i
h
u h
u
V
p w Y,
where i is a cofibration, p is a trivial fibration, and r exists by the lifting property.
Then f is a retract of p and is therefore a trivial fibration. The reverse implication
is CM4.

Finally, since fibrations (respectively trivial fibrations) are those maps having
the right lifting property with respect to all trivial cofibrations (respectively
all fibrations), they are stable under composition and pullback and include all
isomorphisms, yielding (B) and (C). (A) is just CM2. This completes the
proof of Proposition 9.5.
We shall see that the category of fibrant objects structure that we have displayed for Sf is the restriction of a closed model structure on the entire simplicial
set category, as in the corollary above. This will be proved in the next section.
10. Minimal fibrations.
Minimal Kan complexes play roughly the same role in the homotopy theory of
simplicial sets as minimal models play in rational homotopy theory (there ought
to be an abstract theory of such things). Minimal Kan complexes appear as
fibres of minimal fibrations; it turns out that minimal fibrations are exactly the
right vehicle for relating the homotopy theories of S and CGHaus.
A simplicial set map q : X Y is said to be a minimal fibration if q is a
fibration, and for every diagram

n 1
z
(10.1)

u
1
n

pr

w n
u
wX
q

pr
u
n

u
wY

10. Minimal fibrations

51

the composites
n

d0 w n
1 h
wX
w
1
d

are equal. This means that, if two simplices x and y in Xn are fibrewise homotopic (rel n ), then x = y.
Note that minimal fibrations are stable under base change.
More generally, write x 'p y if there is a diagram of the form (10.1) such that
h(n 0) = x, and h(n 1) = y. The relation 'p is an equivalence relation
(exercise).
Lemma 10.2. Suppose that x and y are degenerate r-simplices of a simplicial
set X such that x = y. Then x = y.
Proof: (See also [67], p. 36.) Suppose that x = sm z and y = sn w. If m = n,
then
z = dm x = dm y = w,
and so x = y. Suppose that m < n. Then
z = dm x = dm sn w = sn1 dm w,
and so
x = sm sn1 dm w = sn sm dm w.
Thus
sm dm w = dn x = dn y = w.
Therefore x = sn w = y.
Now we can prove:

Proposition 10.3. Let p : X Y be a Kan fibration. Then p has a strong

fibre-wise deformation retract q : Z Y which is a minimal fibration.
Proof: Let Z (0) be the subcomplex of X which is generated by a choice of
vertex in each p-class, and let i(0) : Z (0) X be the canonical inclusion. There
is a map r(0) : sk0 X Z (0) which is determined by choices of representatives.
Moreover pi(0) r(0) = p|sk0 X , and j0 ' i(0) r(0) , where j0 : sk0 X X is the
obvious inclusion, via a homotopy h0 : sk0 X 1 X such that h0 (x, 0) = x
and h0 (x, 1) = r(0) (x), and h0 is constant on simplices of Z (0) . h0 can be
constructed fibrewise in the sense that p h0 is constant, by using the homotopies
implicit in the definition of 'p . The subcomplex Z (0) has a unique simplex in
each p-equivalence class that it intersects, by Lemma 10.2.

52

I. Simplicial sets

Let Z (1) be the subcomplex of X which is obtained by adjoining to Z (0) a

representative for each homotopy class of 1-simplices x such that x Z (0) and
x is not p-related to a 1-simplex of Z (0) . Again, Z (1) has a unique simplex in
each p-equivalence class that it intersects, by construction in degrees 1 and
Lemma 10.2 in degrees > 1.
Let x be a non-degenerate 1-simplex of X. Then there is a commutative diagram
(x, h0 |x )

(1 {0}) (1 1 )
z

hx

w 1
1 1
pr
L

px

wX
p
u
w Y,

by the homotopy lifting property, where the constant homotopy is chosen for hx
if x Z (1) . But then (hx (1 {1})) Z (0) and so hx (1 {1}, ) is p-related
to a unique 1-simplex r(1) (x) of Z (1) via some diagram

1 1
z
u
1
1

prL

hx (1 {1})

gx

u
wX

pr
u
1

w 1

px

u
w Y,

where gx is constant if x Z (1) , r(1) (x) = gx (1 1), and

gx (1 {0}) = hx (1 {1}).
This defines r(1) : sk1 X Z (1) .
We require a homotopy h1 : j1 'p i(1) r(1) , such that i(1) : Z (1) X and
j1 : sk1 X X are the obvious inclusions, and such that h1 is consistent with
h0 . We also require that the restriction of h1 to Z (1) be constant. This is done

10. Minimal fibrations

53

for the simplex x by constructing a commutative diagram

(s1 h0 , (gx , , hx ))
wX

x
p

u
u

1
1 2
w
w
Y,

pr
px

(1 2 ) (1 21 )
z

where the lifting x is chosen to be the composite

prL

1 2 1
X
if x Z (1) . Then h0 can be extended to the required homotopy h1 : j1 'p i(1) r(1)
by requiring that h1 |x = x (1 d1 ).
Proceeding inductively gives i : Z = lim Z (n) X and r : X Z such that

1X ' ir fibrewise, and such that q : Z Y has the minimality property. Finally,
q is a Kan fibration, since it is a retract of a Kan fibration.

Lemma 10.4. Suppose that

f

w Z0
N

q 0
q N
N
P

ZN

is a fibrewise homotopy equivalence of minimal fibrations q and q 0 . Then f is an

isomorphism of simplicial sets.
To prove Lemma 10.4, one uses:
Sublemma 10.5. Suppose that two maps
XN

N
p N
P
N

wZ

XN

N
p N
N
P

wZ

are fibrewise homotopic, where g is an isomorphism and q is minimal. Then f

is an isomorphism.

54

I. Simplicial sets

Proof of sublemma: Let the diagram

X

u
h wZ
X u 1
P
NN
N
d1
N g
N
X
d0

represent the homotopy. Suppose that f (x) = f (y) for n-simplices x and y of
X. Then inductively di x = di y, 0 i n, and so the composites
i1

x1

i1

y1

n 1 n 1 X 1
Z,
and

n 1 n 1 X 1
Z,
are equal (to a map h : n 1 Y ). Write hx for the composite homotopy
x1

n 1 X 1
Y.
Then there is a commutative diagram
((hx , hy , ), s0 h )

wZ

G
q

u
u

n
n
2

w
w Y,
pr
px = py

(n 22 ) (n 2 )
z

and the homotopy G (1 d2 ) shows that x = y. Thus, f is monic.

To see that f is epi, suppose inductively that f : Xi Zi is an isomorphism
for 0 i n 1, and let x : n Z be an n-simplex of Z. Then there is a
commutative diagram
n
z
u
n

(x0 , . . . , xn )

wX
f
u
wZ

10. Minimal fibrations

55

by the inductive assumption, and so one can find a diagram

(h|(x0 ,...,xn ) , x)
wZ

1
u
u

n
1
n

w
pr
qx w Y.

(n 1 ) (n {1})
z

Then there is a diagram

X

z
g

n
1
n

w
w Z,
h1
d1
since g is epi. The restriction of z to n is the composite g (x0 , . . . , xn ), so
that z = (x0 , . . . , xn ) since g is monic. Thus, there is a diagram
(( , h1 , hz ), s1 h|(x0 ,...,xn ) )
wZ

G
q

u
u

n
n
2
w
w Y.

pr
qx

(n 20 ) (n 2 )
z

Finally, the composite

1d0

G0

n 1 n 2 Z
is a fibrewise homotopy from f (z) to x, and so x = f (z).
Lemma 10.6. Suppose given a Kan fibration p and pullback diagrams
fi1 p

wX
p

pi
u
A

u
wY

i = 0, 1.

56

I. Simplicial sets
'

Suppose further that there is a homotopy h : f0 f1 . Then there is a fibrewise

homotopy equivalence
'

f11 p[

[
p1 [
]

w f21 p.

p2

A
Proof: Consider the diagrams of pullbacks
f1 p

w h1 p

p
u
A

wX
p

ph

u
w A 1

= 0, 1.

u
wY

Then there is a commutative diagram

f01 p

x0

d0

[
[

w h1 p
]
[

ph

u
u
[
f01 p 1
w A 1 ,
p0 1
by the homotopy lifting property. It follows that there is a diagram
1

d
w f01 p 1 '
f01 p

'

'

'

)
'

x1
p0 f11 p
w h1 p

ph
p1

u
u
A
w A 1
d1

10. Minimal fibrations

57

and hence an induced map as indicated. Similarly there are diagrams

f11 p

x1

d1

[
[

w h1 p
]
[

ph

u
u
[
f11 p 1
w A 1 ,
p1 1
and
0

d
f11 p
w f11 p 1 '

'

'

'

'
)

x0
p1 f01 p
w h1 p

ph

p0
u
u
w
A

1 .
A
0
d
Form the diagram
f01 p 20
z

( , , ( 1))

w h1 p
'
)
'
'''
ph
'
'
u
u
''
1
2
f0 p
w
A

1 ,
1
p0 s
by using the homotopy lifting property and the relations d1 = x1 = d1 ((
1)). Then there is a commutative diagram
1 d0 1
f01 p 1
w f0 p 2 '

'

'

'

'
)

x0
prL
f01 p
w h1 p
p0
u
f01 p

p0

u
wA

u
d

ph

w A 1 ,

58

I. Simplicial sets

by the simplicial identities. Then : ' 1 is a fibrewise homotopy. There

is a similar fibrewise homotopy ' 1.

Corollary 10.7. Suppose that q : Z Y is a minimal fibration, and that

fi : X Y , i = 0, 1 are homotopic simplicial maps. Then there is a commutative
diagram

=
f01 q [
[
q0 [
]

w f11 q.

q1

X
In particular, the pullbacks f01 q and f11 q are isomorphic.
Corollary 10.8. Suppose that q : Z Y is a minimal fibration with Y
connected. Suppose that F is the fibre of q over a base point of Y. Then, for
any simplex : n Y there is a commutative diagram

F n [

w 1 q.

[
]
prR [
n

Proof: Suppose that v and w are vertices of Y such that there is a 1-simplex
z of Y with z = (v, w). Then the classifying maps v : 0 Y and w : 0 Y
are homotopic, and so there is an isomorphism Fv
= Fw of fibres induced by the
homotopy. In particular, there is an isomorphism Fv
= F for any vertex v of Y.
0
n
Now let i0 : be the map that picks out the vertex 0 of n . Finally,
recall (see the proof of 7.10) that the composite
i0

n 0 n
is homotopic to the identity on n .

Theorem 10.9 (Gabriel-Zisman). Suppose that q : X Y is a minimal

fibration. Then its realization |q| : |X| |Y | is a Serre fibration.

10. Minimal fibrations

59

Proof: It is enough to suppose that Y has only finitely many non-degenerate

simplices, since the image of any continuous map |n | |Y | is contained in
some finite subcomplex of |Y |. We may also suppose that Y is connected. The
idea of the proof is to show that |q| : |X| |Y | is locally trivial with fibre |F |,
where F is the fibre over some base point of Y .
Now suppose that there is a pushout diagram

n
z
u
n

wZ
z
u
w Y,

where Z is subcomplex of Y with fewer non-degenerate simplices, and suppose

that U is an open subset of |Z| such that there is a fibrewise homeomorphism
U |Y | |X|[

[
[
]

w U |F |.

prL

U
Let U 1 = ||1 (U ) |n |. Then there is an induced fibrewise homeomorphism

U 1 |Y | |X|[

[
]
[

w U 1 |F |.

prL

U1
On the other hand the simplicial fibrewise homeomorphism
n F [

[
]
prL [

w n Y X

60

I. Simplicial sets

induces a homeomorphism
V 1 F[

[
prL [
]

w V 1 |Y | |X|

V1

over some open subset V 1 of |n | such that V 1 |n | = U 1 and U 1 is a retract

of V 1 . The map restricts to a homeomorphism
U1 F[

[
prL [
]

w U 1 |Y | |X|

U1
over U 1 . Now consider the fibrewise homeomorphism
U 1 |F |[

w U 1 |F |.

prL

[
]
prL [

U1
There is a homeomorphism
r ( 1 )
V |F |[
w V 1 |F |.

=
[

]
prL [
prL

1
V
1

which restricts to 1 over U 1 . In effect r ( 1 )(v 0 , f ) = (v 0 , (rv 0 , f )), where

1 (w, f ) = (w, (w, f )), and this definition is functorial. Thus, the fibrewise

10. Minimal fibrations

61

isomorphism
r ( 1 )
w V 1 |Y | |X|

=
[

prL [
]

V1

V 1 F[

restricts to over U 1 . It follows that there is a fibrewise homeomorphism

(V 1 U 1 U ) F [
[
prL [
]

w (V 1 U 1 U ) |Y | |X|

(V 1 U 1 U )

over the open set. V 1 U 1 U of |Y |.

The following result of Quillen [76] is the key to both the closed model structure of the simplicial set category, and the relation between simplicial homotopy
theory and ordinary homotopy theory. These results will appear in the next
section.
Theorem 10.10 (Quillen). The realization of a Kan fibration is a Serre fibration.
Proof: Let p : X Y be a Kan fibration. According to Proposition 10.3, one
can choose a commutative diagram
j

Z4

wX

4
q 44
6p

wZ
h

h
q
h
u h
k
Y

where q is a minimal fibrations, gj = 1Z and jg is fibrewise homotopic to 1X . In

view of Theorem 10.9, it clearly suffices to prove the following two results:
Lemma 10.11. g : X Z has the right lifting property with respect to all
n n , n 0.

62

I. Simplicial sets

Lemma 10.12. Suppose that g : X Z has the right lifting property with
respect to all n n , n 0. Then |g| : |X| |Z| is a Serre fibration.
Proof of Lemma 10.11: Suppose that the diagrams
X
X

d0

wX

u
X u 1

pr
u
X

u
wY

1X

w Xu

d1
X

wZ

represent the fibrewise homotopy, and suppose that the diagram

n
z
i
u
n

wX

g
u
wZ

commutes. Then there are commutative diagrams

u1

n 1
z
i1
u
n
1

n
z
i
u
n

w X 1

wX
p

prL

wX

wZ

qv

u
wY

wX

j
p

w Z q w Y,

10. Minimal fibrations

63

and hence a diagram

(h(u 1), jv)
wX

h1
p

u
u

n
1
n

w
pr
qv w Y.

(n 1 ) (n {0})
z

Let v1 be the simplex classified by the composite

d0

h1

n n 1 X.
The idea of the proof is now to show that gv1 = v. The diagram
u wX
n
z
]
[
[
i
[ v1
u [
n
commutes, and consider the composite
v1 1

n 1 X 1
X
Z.
'

Then gh(v1 1) is a homotopy gjgv1 = gv1 gv1 . Moreover, the homotopy on

the boundary is gh(u 1). It follows that there is a commutative diagram
(s0 (gh(u 1)), (gh1 , gh(v1 1), ))
wZ

u
u

n 2
w n
w Y.
pr
qv

(n 2 ) (n 22 )
z

Then the diagram

prL
n 1
w n
z
gu
i1
u
u
2
n
1 (1 d )

wZ
q

prL
u
n

qv

u
wY

64

I. Simplicial sets

commutes, and so gv1 = gjv = v by the minimality of q.

Proof of Lemma 10.12: Suppose that f : X Y has the right lifting property
with respect to all n n , n 0, and hence with respect to all inclusions of
simplicial sets. Then there is a commutative diagram

X
(1X , f )

1X

wX

h
j
rh
h

u
h
X Y pr

f
u
w Y,

and so f is a retract of the projection pr : X Y Y. But then |f | is a Serre

fibration.

This also completes the proof of Theorem 10.10.

11. The closed model structure.
The results stated and proved in this section are the culmination of all of the
hard work that we have done up to this point. We shall prove here that the
entire simplicial set category S (not just the subcategory of fibrant objects) has
a closed model structure, and that the resulting homotopy theory is equivalent
to the ordinary homotopy theory of topological spaces. These are the central
organizational theorems of simplicial homotopy theory.
Proposition 11.1. Suppose that X is a Kan complex. Then the canonical map
X : X S|X| is a weak equivalence in the sense that it induces an isomorphism
in all possible simplicial homotopy groups.
Proof: Recall that S|X| is also a Kan complex.
X induces an isomorphism in 0 : every map v : |0 | |X| factors through
the realization of a simplex || : |n | |X| and so S|X| is connected if 0 X = .
On the other hand X is a disjoint union of its path components and S| | preserves
disjoint unions, so that 0 X 0 S|X| is monic.
Suppose that we have shown that X induces an isomorphism

(X ) : i (X, x) i (S|X|, x)
for all choice of base points x X and i n. Then, using 10.10 for the pathloop fibration X P X X determined by x (see the discussion following

11. The closed model structure

65

the proof of 7.3), one finds a commutative diagram

n+1 (X, x)
=

u
n (X, x)

=
X

w n+1 (S|X|, x)

u
w n (S|X|, x),

and so were done if we can show that P X and hence S|P X| contracts onto its
base point. But there is a diagram
(x, (1P X , x))
(0 1 ) (P X 1 )
w PX
)
'
z
'
h ''
t
''
'
u
u
''
P X 1
w 0 ,

and h exists because t has the right lifting property with respect to all n n ,
n 0.

If X is a Kan complex and x is any vertex of X, then it follows from Proposition 11.1 and adjointness that X induces a canonical isomorphism
n (X, x)
= n (|X|, x),

n 1,

where the group on the right is the ordinary homotopy group of the space |X|.
It follows that a map f : X Y of Kan complexes is a (simplicial) weak
equivalence if and only if the induced map |f | : |X| |Y | is a topological weak
equivalence. Thus, we are entitled to define a map f : X Y of arbitrary
simplicial sets to be a weak equivalence if the induced map |f | : |X| |Y | is a
weak equivalence of spaces. Our last major technical result leading to the closed
model structure of S is
Theorem 11.2. Suppose that g : X Y is a map between arbitrary simplicial
sets. Then g is a Kan fibration and a weak equivalence if and only if g has the
right lifting property with respect to all inclusions n n , n 0.

66

I. Simplicial sets

Proof: Suppose that g : X Y is a Kan fibration with the advertised lifting

property. We have to show that S|g| : S|X| S|Y | is a weak equivalence. Y is
an arbitrary simplicial set, so we must define 0 Y to be the set of equivalence
classes of vertices of Y for the relation generated by the vertex homotopy relation.
In other words, y ' z if and only if there is a string of vertices
y = y0 , y1 , . . . , yn = z
and a string of 1-simplices
v 1 , . . . , vn
of X such that vi = (yi1 , yi ) or vi = (yi , yi1 ) for i = 1, . . . , n. If Y is a Kan
complex, then this definition of 0 Y coincides with the old definition. Moreover,

the canonical map Y : Y S|Y | induces an isomorphism 0 Y 0 S|Y | for

all simplicial sets Y. The lifting property implies that g : 0 X 0 Y is an isomorphism, so that the induced map 0 S|X| 0 S|Y | is an isomorphism as well.
Finally, it suffices to show that the induced maps i (S|X|, x) i (S|Y |, gx) of
simplicial homotopy groups are isomorphisms for all vertices x of X and all i 1.
But Theorem 10.9 implies that the fibre of the fibration S|g| : S|X| S|Y | over
g(x) is S|Fx |, where
Fx

wX
g

u
0

gx

u
wY

is a pullback in the simplicial set category. Fx is a contractible Kan complex

(see the corresponding argument for P X in 11.1), and so S|Fx | is contractible
as well. The result then follows from a long exact sequence argument.
For the reverse implication, it suffices (see the proof of Theorem 10.10) to
assume that g : X Y is a minimal fibration and a weak equivalence and then
prove that it has the lifting property. We may also assume that Y is connected.
Consider a diagram
n
z

wX
g

u
n

u
wY

11. The closed model structure

67

and the induced diagram

n
z
u
n

w n Y X
u
w n .

It suffices to find a lifting for this last case. But there is a fibrewise isomorphism

n Y X [

[
]
[

w n F y

prL

n
by 10.8, where Fy is the fibre over some vertex y of Y. Thus, it suffices to find a
lifting of the following sort:

n
z

w Fy .
]
\
\

u \
n

But this can be done, since Fy is a Kan complex such that 0 (S|Fy |) is trivial,
and i (S|Fy |, ) = 0, i 1 for any base point , and : Fy S|Fy | is a weak
equivalence by Proposition 11.1.

A cofibration of simplicial sets is an inclusion map.

Theorem 11.3. The simplicial set category S, together with the specified classes
of Kan fibrations, cofibrations and weak equivalences, is a closed model category.
Proof: CM1 is satisfied, since S is complete and cocomplete. CM2 follows
from CM2 for CGHaus. CM3 (the retract axiom) is trivial.

68

I. Simplicial sets

To prove the factorization axiom CM5, observe that a small object argument
and the previous theorem together imply that any simplicial set map f : X Y
may be factored as:
XN

N
N
P
i N

wY

X[

[
]
j [

and

w Y,

where i is anodyne, p is a fibration, j is an inclusion, and q is a trivial fibration.

The class of inclusions i : U V of simplicial sets such that |i| : |U | |V | is a
trivial cofibration is saturated, by adjointness, and includes all nk n . Thus
all anodyne extensions are trivial cofibrations of S. To prove CM4 we must
show that the lifting (dotted arrow) exists in any commutative diagram

wX

U
i

i
u i
V

j
ii

p
u
w Y,

where p is a fibration and i is a cofibration, and either i or p is trivial. The case

where p is trivial is the previous theorem. On the other hand, if i is a weak
equivalence, then there is a diagram
j

U
i

u h
V

h
j
sh
h
1V

wZ
p
u
wV

where j is anodyne and p is a (necessarily) trivial fibration, so that s exists. But

then i is a retract of an anodyne extension, so i has the left lifting property with
respect to all fibrations (compare the proof of 9.4).

The homotopy category Ho(S) is obtained from S by formally inverting the

weak equivalences. There are several ways to do this [75], [33], [15]. One may

11. The closed model structure

69

also form the category Ho(T op) by formally inverting the weak homotopy equivalences; this category is equivalent to the category of CW-complexes and ordinary homotopy classes of maps. For the same reason (see [75]), Ho(S) is
equivalent to the category of Kan complexes and simplicial homotopy classes of
maps. The realization functor preserve weak equivalences, by definition. One
may use Theorem 10.10 (see the argument in 11.1) to show that the canonical
map : |S(Y )| Y is a weak equivalence, for any topological space Y, and
so the singular functor preserves weak equivalences as well. It follows that the
realization and singular functors induce functors
Ho(S) u

| |
S

w Ho(Top)

of the associated homotopy categories.

Theorem 11.4. The realization and singular functors induce an equivalence of
categories of Ho(S) with Ho(Top).
Proof: We have just seen that : |S(Y )| Y is a weak equivalence for all
topological spaces Y. It remains to show that : X S|X| is a weak equivalence
for all simplicial sets X. But is a weak equivalence if X is a Kan complex, by
11.1, and every simplicial set is weakly equivalent to a Kan complex by CM5.
The composite functor S| | preserves weak equivalences.

The original proof of Theorem 11.3 appears in [75], modulo some fiddling with
axioms (see [77]). Theorem 11.4 has been known in some form or other since the
late 1950s (see [67], [33], [59]).
Although it may now seem like a moot point, the function complex trick of
Proposition 5.2 was a key step in the proof of Theorem 11.3. We can now amplify
the statement of Proposition 5.2 as follows:
Proposition 11.5. The category S of simplicial sets satisfies the simplicial
model axiom
SM7: Suppose that i : U V is a cofibration and p : X Y is a fibration.
Then the induced map
(i ,p )

Hom(V, X) Hom(U, X) Hom(U,Y ) Hom(V, X)

is a fibration, which is trivial if either i or p is trivial.
Proof: Use Proposition 5.2 and Theorem 11.2.

Chapter II Model Categories

The closed model axioms have a list of basic abstract consequences, including
an expanded notion of homotopy and a Whitehead theorem. The associated
homotopy category is defined to be the result of of formally inverting the weak
equivalences within the ambient closed model category, but can be constructed
(in the CW-complex style) by taking homotopy classes of maps between objects
which are fibrant and cofibrant. These topics are presented in the first section
of this chapter.
The simplicial set category has rather more structure than just that of a closed
model category: the set hom(X, Y ) of maps between simplicial sets X and Y is
the set of vertices of the function complex Hom(X, Y ), and the collection of all
such function complexes determines a simplicial category. Weve already seen
that the function complexes satisfy an exponential law and respect cofibrations
and fibrations in a suitable sense. The existence of the function complexes and
the interaction with the closed model structure can be abstracted to a definition
of a simplicial model category, which is given in Section 2, along with various examples. Basic consequences of the additional simplicial structure are presented,
with further examples, in Sections 3 and 4.
Sections 5, 6 and 7 are about detection principles for simplicial model structures. Generally speaking, a simplicial model category structure for the category
sC of simplicial objects in a category C is induced from the simplicial model structure on the simplicial set category in the presence of an adjoint pair of functors
F : S sC : G,
(or a collection of adjoint pairs) if G has a little extra structure (such as preservation of filtered colimits) in addition to being a right adjoint this is Theorem 5.1.
In one major stream of examples, the category C is some algebraic species, such
as groups or abelian groups, and G is a forgetful functor. There is, however, an
extra technical condition on Theorem 5.1, namely that every cofibration of sC
having the left lifting property with respect to all fibrations should be a weak
equivalence. This condition can often be verified by brute force, as can be done
in the presence of a small object argument for the factorization axioms (eg.
simplicial abelian groups), but there is a deeper criterion, namely the existence
of a natural fibrant model (Lemma 6.1). The other major source of examples
has to do with G being a representable functor of the form G = hom(Z, ),
where Z is either small in the sense that homC (Z, ) respects filtered colimits,
or is a disjoint union of small objects. In this setting, Kans Ex -construction
(see Section III.4) is used to construct the natural fibrant models required by
70

1. Homotopical algebra

71

Lemma 6.1. This line of argument is generalized significantly in Section 6, at

the cost of the introduction of cofibrantly generated closed model categories and
transfinite small object arguments.
Section 8 is an apparent return to basics. We develop a criterion for a pair of
adjoint functors between closed model categories to induce adjoint functors on
the homotopy category level, known as Quillens total derived functor theorem.
Quillens result is, at the same time, a non-abelian version of the calculus of
higher direct images, and a generalization of the standard result that cohomology
is homotopy classes of maps taking values in Eilenberg-Mac Lane spaces.
The category of simplicial sets, finally, has even more structure: it is a proper
simplicial model category, which means that, in addition to everything else, weak
equivalences are preserved by pullback over fibrations and by pushout along
cofibrations. This property is discussed in Section 9. Properness is the basis
of the standard results about homotopy cartesian diagrams, as well as being
of fundamental importance in stable homotopy theory. We discuss homotopy
cartesian diagrams in the context of Gunnarssons axiomatic approach to the
glueing and coglueing lemmas [40].
1. Homotopical algebra.
Recall that a closed model category C is a category which is equipped with
three classes of morphisms, called cofibrations, fibrations and weak equivalences
which together satisfy the following axioms:
CM1: The category C is closed under all finite limits and colimits.
CM2: Suppose that the following diagram commutes in C:

XN

N
N
P
h N

wY

Z.

If any two of f , g and h are weak equivalences, then so is the third.

CM3: If f is a retract of g and g is a weak equivalence, fibration or cofibration,
then so is f .

72

II. Model Categories

CM4: Suppose that we are given a commutative solid arrow diagram

U
i

wX

u i
V

i
ij

p
u
wY

where i is a cofibration and p is a fibration. Then the dotted arrow exists,

making the diagram commute, if either i or p is also a weak equivalence.
CM5: Any map f : X Y may be factored:
(a) f = p i where p is a fibration and i is a trivial cofibration, and
(b) f = q j where q is a trivial fibration and j is a cofibration.
This set of axioms has a list of standard consequences which amplifies the
interplay between cofibrations, fibrations and weak equivalences, giving rise to
collection of abstract techniques that has been known as homotopical algebra
since Quillen introduced the term in [75]. This theory is is really an older friend
in modern dress, namely obstruction theory made axiomatic. The basic results,
along with their proofs, are sketched in this section.
We begin with the original meaning of the word closed:
Lemma 1.1. Suppose that C is a closed model category. Then we have the
following:
(1) A map i : U V of C is a cofibration if and only if i has the left lifting
property with respect to all trivial cofibrations.
(2) The map i is a trivial cofibration if and only if it has the right lifting
property with respect to all fibrations.
(3) A map p : X Y of C is a fibration if and only if it has the right lifting
property with respect to all trivial cofibrations.
(4) The map p is a trivial fibration if and only if it has the right lifting
property with respect to all cofibrations.
The point of Lemma 1.1 is that the various species of cofibrations and fibrations
determine each other via lifting properties.
Proof: We shall only give a proof of the first statement; the other proofs are
similar.

1. Homotopical algebra

73

Suppose that i is a cofibration, p is a trivial fibration, and that there is a

commutative diagram

U
(1.2)

wX
p

u
V

u
wY

Then there is a map : V X such that p = and i = , by CM4.

Conversely suppose that i : U V is a map which has the left lifting property
with respect to all trivial fibrations. By CM5, i has a factorization
j

U'

'

wW

'
q
'
)
u
V

where j is a cofibration and q is a trivial fibration. But then there is a commutative diagram
j

U
i

u h
V

h
hj

wW
q
u
V

and so i is a retract of j. CM3 then implies that i is a cofibration.

The proof of the Lemma 1.1 contains one of the standard tricks that is used to
prove that the axiom CM4 holds in a variety of settings, subject to having an
adequate proof of the factorization axiom CM5. Lemma 1.1 also immediately
implies the following:
Corollary 1.3.
(1) The classes of cofibrations and trivial cofibrations are closed under composition and pushout. Any isomorphism is a cofibration.
(2) The classes of fibrations and trivial fibrations are closed under composition and pullback. Any isomorphism is a fibration.

74

II. Model Categories

The statements in Corollary 1.3 are axioms in Quillens original definition of a

model category [75].
Quillen defines a cylinder object for an object A in a closed model category C
to be a commutative triangle
A t A4
(1.4)

4
6
4

wA

where : AtA A is the canonical fold map which is defined to be the identity
on A on each summand, i is a cofibration, and is a weak equivalence. Then a
left homotopy of maps f, g : A B is a commutative diagram
A t A4
i

u
A

4(f, g)
44
6
h

wB

where (f, g) is the map on the disjoint union which is defined by f on one
summand and g on the other, and the data consisting of
i = (i0 , i1 ) : A t A A
comes from some choice of cylinder object for A.
There are many choices of cylinder object for a given object A of a closed
model category C: any factorization of : A t A A into a cofibration followed
by a trivial fibration that one might get out of CM5 gives a cylinder object for
A. In general, the object A needs to be cofibrant for its cylinder objects to be
homotopically interesting:
Lemma 1.5.
(1) Suppose that A is a cofibrant object of a closed model category C, and that
the diagram (1.4) is a cylinder object for A. Then the maps i0 , i1 : A A
are trivial cofibrations.
(2) Left homotopy of maps A B in a closed model category C is an equivalence relation if A is cofibrant.

1. Homotopical algebra

75

Proof: Denote the initial object of C by .

For the first part, observe that the diagram

wA

u
A

inR
u
w AtA

inL

is a pushout since cofibrations are closed under pushout by Lemma 1.1, and the
unique map A is a cofibration by assumption. It follows that the inclusions
inL and inR are cofibrations, so that the compositions i0 = (i0 , i1 ) inL and
i1 = (i0 , i1 ) inR are cofibrations as well. Finally, the maps i0 and i1 are weak
equivalences by CM2, since the map is a weak equivalence.
To prove the second statement, first observe that if : A t A A t A is the
automorphism which flips summands, then the diagram

A t A4
i

4
6
4

w A,

which is constructed from (1.4) by twisting by , is a cylinder object for A. This

implies that the left homotopy relation is symmetric.
Subject to the same definitions, the map f : A B is clearly a left homotopy
from f : A B to itself, giving reflexivity.
Suppose given cylinder objects

A t A4
4

4
(i0 , i1 )
6
4
u
A
w A,

76

II. Model Categories

where = 0, 1. Form the pushout

i10

w A1

i01

u
A0

i1
u
w A

i0

Then the map

(i0 i00 ,i1 i11 )

A t A A
is a composite
i00 t1

(i0 ,i1 i11 )

A t A A0 t A A.
The map i00 t 1 is a cofibration by the first statement of the lemma, and there is
a pushout diagram
AtA

i01 t 1

w A0 t A

(i10 , i11 )

u
A1

(i0 , i1 i11 )

w A.

i1

In particular, there is a cylinder object for A

A t A4
(i0 i00 , i1 i11 )

u
A

4
6
4

w A.

It follows that if there are left homotopies h0 : A0 B from f0 to f1 and

h1 : A1 B from f1 to f2 , then there is an induced left homotopy h : A B
from f0 to f2 .

1. Homotopical algebra

77

A path object for an object B of a closed model category C is a commutative

triangle

(1.6)

P
N
sNN
p = (p0 , p1 )
N
u
N
B
w BB

where is the diagonal map, s is a weak equivalence, and p (which is given by

p0 on one factor and by p1 on the other) is a fibration.
Once again, the factorization axiom CM5 dictates that there is an ample
supply of path objects for each object of an arbitrary closed model category. If a
simplicial set X is a Kan complex, then the function complex hom(1 , X) is a
path object for X, and the function space Y I is a path object for each compactly
generated Hausdorff space Y .
There is a notion of right homotopy which corresponds to path objects: two
maps f, g : A B are said to be right homotopic if there is a diagram

B
N
P
hNN
(p0 , p1 )
N
u
N
w BB
A
(f, g)
where the map (p0 , p1 ) arises from some path object (1.6), and (f, g) is the map
which projects to f on the left hand factor and g on the right hand factor.
Lemma 1.7.
(1) Suppose that B is a fibrant object of a closed model category C, and that
is a path object for B as in (1.6). Then the maps p0 and p1 are trivial
B
fibrations.
(2) Right homotopy of maps A B in C is an equivalence relation if B is
fibrant.
Lemma 1.7 is dual to Lemma 1.5 in a precise sense. If C is a closed model
category, then its opposite C op is a closed model category whose cofibrations

78

II. Model Categories

(respectively fibrations) are the opposites of the fibrations (respectively cofibrations) in C. A map in C op is a weak equivalence for this structure if and only if
its opposite is a weak equivalence in C. Then Lemma 1.7 is an immediate consequence of the instance of Lemma 1.5 which occurs in C op . This sort of duality
is ubiquitious in the theory: observe, for example, that the two statements of
Corollary 1.3 are dual to each other.
Left and right homotopies are linked by the following result:
Proposition 1.8. Suppose that A is cofibrant. Suppose further that
A t A4
(i0 , i1 )

4(f, g)
44
6
h

wB

is a left homotopy between maps f, g : A B, and that

B
N
P
sNN
p = (p0 , p1 )
N
u
N
B
w BB

is a fixed choice of path object for B. Then there is a right homotopy

P
N
HNN
(p0 , p1 )
N
u
N
A
w B B.
(f, g)
This result has a dual, which the reader should be able to formulate independently. Proposition 1.8 and its dual together imply
Corollary 1.9. Suppose given maps f, g : A B, where A is cofibrant and
B is fibrant. Then the following are equivalent:
(1)
(2)
(3)
(4)

f
f
f
f

and g are left homotopic.

and g are right homotopic with respect to a fixed choice of path object.
and g are right homotopic.
and g are left homotopic with respect to a fixed choice of cylinder object.

1. Homotopical algebra

79

In other words, all possible definitions of homotopy of maps A B are the same
if A is cofibrant and B is fibrant.
Proof of Proposition 1.8: The map i0 is a trivial cofibration since A is
cofibrant, and (p0 , p1 ) is a fibration, so that there is a commutative diagram
sf

N
P
KNN
N

wB

(p0 , p1 )
u N
u
w BB
A
(f , h)

i0

for some choice of lifting K. Then the composite K i1 is the desired right
homotopy.

We can now, unambiguously, speak of homotopy classes of maps between objects X and Y of a closed model category C which are both fibrant and cofibrant.
We can also discuss homotopy equivalences between such objects. The classical
Whitehead Theorem asserts that any weak equivalence f : X Y of CWcomplexes is a homotopy equivalence. CW-complexes are spaces which are both
cofibrant and cofibrant. The analogue of this statement in an arbitrary closed
model category is the following:
Theorem 1.10 (Whitehead). Suppose that f : X Y is a morphism of a
closed model category C such that the objects X and Y are both fibrant and
cofibrant. Suppose also that f is a weak equivalence. Then the map f is a
homotopy equivalence.
Proof: Suppose, first of all, that f is a trivial fibration, and that

X tX
(i0 , i1 )

P
N
NN
N

wX

u N

is a cylinder object for X. Then one proves that f is a homotopy equivalence by

finding, in succession, maps and h making the following diagrams commute:

u h
Y

wX
h
j
h
h

f
u
Y

80

II. Model Categories

( f, 1)
wX
C
A
A
hA
(i0 , i1 )
f
A
A
u A
u

X
w Y.
f
X tX

Dually, if f is a trivial cofibration, then f is a homotopy equivalence.

Every weak equivalence f : X Y between cofibrant and fibrant objects has
a factorization
Xh

h
j
i h

'

wY
'
)
'p

Z
in which i is a trivial cofibration and p is a trivial fibration. The object Z is
both cofibrant and fibrant, so i and p are homotopy equivalences.

Suppose that X and Y are objects of a closed model category C which are
both cofibrant and fibrant. Quillen denotes the set of homotopy classes of maps
between such objects X and Y by (X, Y ). There is a category Ccf associated
to any closed model C: the objects are the cofibrant and fibrant objects of C,
and the morphisms from X to Y in Ccf are the elements of the set (X, Y ).
For each object X of C, use CM5 to choose, in succession, maps
iX

pX

QX X
and
jX

qX

QX RQX ,
where iX is a cofibration, pX is a trivial fibration, jX is a trivial cofibration, and
qX is a fibration. We can and will presume that X is the identity map if X is
cofibrant, and that jX is the identity map if QX is fibrant. Then RQX is an
object which is both fibrant and cofibrant, and RQX is weakly equivalent to X,
via the maps pX and jX .
Any map f : X Y lifts to a map Qf : QX QY , and then Qf extends to
a map RQf : RQX RQY . The map Qf is not canonically defined: it is any

1. Homotopical algebra

81

morphism which makes the following diagram commute:

w QY
h
j
Qf h
Y
h
u h
u
QX
wY
f X
Note, however, that any two liftings f1 , f2 : QX QY of the morphism f X
are left homotopic.
The argument for the existence of the morphism RQf : RQX RQY is dual
to the argument for the existence of Qf . If the maps f1 , f2 : QX QY are
liftings of f X and gi : RQX RQY is an extension of the map jY fi for
i = 1, 2, then f1 is left homotopic to f2 by what weve already seen, and so the
composites jY f1 and jY f2 are right homotopic, by Lemma 1.8. Observe
finally that any right homotopy between the maps jY f1 , jY f2 : QX RQY can
be extended to a right homotopy between the maps g1 , g2 : RQY RQY . It
follows that the assignment f 7 RQf is well defined up to homotopy.
The homotopy category Ho(C) associated to a closed model category C can be
defined to have the same objects as C, and with morphism sets defined by
homHo(C) (X, Y ) = (RQX, RQY ).
There is a functor
: C Ho(C)
which is the identity on objects, and sends a morphism f : X Y to the
homotopy class [RQf ] which is represented by any choice of map RQf : RQX
RQY defined as above. If f : X Y is a weak equivalence of C, then RQf :
RQX RQY is a homotopy equivalence by the Whitehead Theorem, and so
(f ) is an isomorphism of Ho(C).
This functor is universal with respect to all functors F : C D which invert
weak equivalences:
Theorem 1.11. Suppose that F : C D is a functor such that F (f ) is an
isomorphism of D for all weak equivalences f : X Y of C. Then there is a
unique functor F : Ho(C) D such that F = F .
Proof: The functor F : C D takes (left or right) homotopic maps of C
to the same map of D, since it inverts weak equivalences. It follows that, if

82

II. Model Categories

g : RQX RQY represents a morphism from X to Y in Ho(C), one can specify

a well-defined morphism F ([g]) of D by the assignment
(1.12)

F ([g]) = F (Y )F (jY )1 F (g)F (jX )F (X )1 .

This assignment plainly defines a functor F : Ho(C) D such that F = F .

Also, the morphisms (X ) and (jX ) are both represented by the identity
map on RQX, and so the composite
(Y )(jY )1 (g)(jX )(jX )1
coincides with the morphism [g] : X Y of Ho(C). The morphism F ([g]) must
therefore have the form indicated in (1.12) if the composite functor F is to
coincide with F .

Remark 1.13. One can always formally invert a class of morphisms of a

category C to get a functor : C C[1 ] which is initial among functors
F : C D which invert all members of the class of morphisms (see Schuberts
book [82]), provided that one is willing to construct C[1 ] in some higher set
theoretic universe. This means that the morphism things homC[1 ] (X, Y ) of
C[1 ] may no longer be sets. The whole point of Theorem 1.11 is that we have
found an explicit way to formally invert the class W E of weak equivalences of a
closed model category C to obtain the category Ho(C) without invoking a higher
universe. After the fact, all models of C[W E 1 ] must be isomorphic as categories
to Ho(C) on account of the universal property of the functor : C Ho(C), so
that all possible constructions have small hom sets.
Let Ccf denote the category whose objects are the cofibrant fibrant objects
of the closed model category C, and whose sets of morphisms have the form
homCcf (X, Y ) = (X, Y ).
Its obvious that the functor : C Ho(C) induces a fully faithful imbedding
: Ccf Ho(C),
and that every object of Ho(C) is isomorphic to an object which is in the image
of the functor . In other words the category Ccf of homotopy classes of maps
between cofibrant fibrant objects of C is equivalent to the homotopy category
Ho(C).
This observation specializes to several well-known phenomena. In particular,
the category of homotopy classes of maps between CW-complexes is equivalent
to the full homotopy category of topological spaces, and the homotopy category
of simplicial sets is equivalent to the category of simplicial homotopy classes of
maps between Kan complexes.

2. Simplicial categories

83

2. Simplicial categories.
A simplicial model category is, roughly speaking, a closed model category
equipped with a notion of a mapping space between any two objects. This has
to be done in such a way that it makes homotopy theoretic sense. Thus, besides
the new structure, there is an additional axiom, which is called Axiom SM7 (See
2.1 below).
The initial property one wants is the following: let S be the category of simplicial sets and let C be a model category, and suppose A C is cofibrant and
X C is fibrant. Then, the space of maps in C should be a functor to simplicial
sets
HomC : C op C S
with the property that
0 HomC (A, X)
= [A, X]C .
In addition, one would want to interpret n HomC (A, X) in C.
There are other desirable properties; for example, if A is cofibrant and X Y
a fibration in C, one would want
HomC (A, X) HomC (A, Y )
to be a fibration of spaces that is, of simplicial sets.
Before imposing the closed model category structure on C, let us make the
following definition:
Definition 2.1. A category C is a simplicial category if there is a mapping
space functor
HomC (, ) : C op C S
with the properties that for A and B objects in C
(1) HomC (A, B)0 = homC (A, B);
(2) the functor HomC (A, ) : C S has a left adjoint
A:SC
which is associative in the sense that there is a isomorphism
A (K L)
= (A K) L
natural in A C and K, L S;
(3) The functor HomC (, B) : C op S has left adjoint
homC (, B) : S C op .

84

II. Model Categories

Of course, the adjoint relationship in (3) is phrased

homS (K, HomC (A, B))
= homC (A, homC (K, B)) .
Warning: The tensor product notation goes back to Quillen, and remains for
lack of a better operator. But be aware that in this context we do not usually
have a tensor product in the sense of algebra; that is, we dont have a pairing
arising out of bilinear maps. Instead (see the next lemma), we have an adjoint
to an internal hom functor, and this is the sole justification for the notation.
Some immediate implications of the definition are the following.
Lemma 2.2.
(1) For fixed K S, there are adjoint functors
K :C C
and
homC (K, ) : C C .
(2) For all K and L in S and B in C there is a natural isomorphism
homC (K L, B)
= homC (K, homC (L, B)).
(3) For all n 0, HomC (A, B)n
= homC (A n , B).
Proof: For fixed K S and A C, the object A K represents the functor
homS (K, HomC (A, )) : C Sets.
f

A map A A0 yields a natural transformation on representing objects

f 1 : A K A0 K
and we obtain the functor K. The functor homC (K, ) is obtained similarly.
The adjointness follows from 2.1.2 and 2.1.3. This proves part 1. Part 2 then
follows from the associativity built into 2.1.2. Part 3 follows from 2.1.2 and the
fact that homS (n , X)

= Xn .
Remark: A consequence of Lemma 2.2.1 is that there is a composition pairing
of simplicial sets
HomC (A, B) HomC (B, C) HomC (A, C)

2. Simplicial categories

85

defined as follows. If f : A n B is an n-simplex of HomC (A, B) and g :

B n C is an n-simplex of HomC (B, C) then their pairing in HomC (A, C)
is the composition
f 1

1d

A n A (n n )
C.
= A n n B n
Here d : n n n is the diagonal. This pairing is associative in the obvious
sense and reduces to the composition pairing in C in simplicial degree zero. It is
also unital in the sense that if HomC (A, A) is the vertex corresponding to
the identity, then the following diagram commutes

HomC (A, B)

u
HomC (A, A) HomC (A, B)
w HomC (A, B).

There is also a diagram using the identity of B. A shorthand way of encoding

all this structure is to say that C is enriched over simplicial sets.

Another immediate consequence of the definition is the following result.

Lemma 2.3. For a simplicial category C then the following extended adjointness
isomorphisms hold:
(1) HomS (K, HomC (A, B))
= HomC (A K, B).
(2) HomS (K, HomC (A, B))
= HomC (A, hom(K, B)).
Proof: This is an easy exercise using Lemma 2.2.

Note that, in fact, Definition 2.1 implies that there are functors
:CSC
and
homC (, ) : Sop C C
satisfying 2.1.1, 2.1.2 and 2.2.1. In order to produce examples of simplicial
categories, we note the following:

86

II. Model Categories

Lemma 2.4. Let C be a category equipped with a functor

:CSC .
Suppose the following three conditions hold:
(1) For fixed K S, K : C C has a right adjoint homC (K, ).
(2) For fixed A, the functor A : S C commutes with arbitrary colimits
and A
= A.
(3) There is an isomorphism A (K L)
= (A K) L natural is A C
and K, L S.
Then C is a simplicial category with HomC (A, B) defined by:
HomC (A, B)n = homC (A n , B)
Proof: We first prove 2.1.2 holds. If K S, write K as the coequalizer in a
diagram
G
G
n q
n p K .
q

Then there is a coequalizer diagram

G
G
A nq
A np A K
q

Hence there is an equalizer diagram

G
G
homC (A K, B) homC (A (
np ), B) homC (A (
nq ), B) .
p

This, in turn, is equivalent to the assertion that the equalizer of the maps
G
G
homS (
np , homC (A, B)) homS (
nq , homC (A, B))
p

is the induced map

homS (K, homC (A, B)) homS (

np , homC (A, B))

so 2.1.2 holds. If we let homC (K, ) be adjoint to K, as guaranteed by the

hypotheses, 2.1.3 holds. Then finally, 2.1.1 is a consequence of the fact that
A = A.

2. Simplicial categories

87

We now give some examples. Needless to say S itself is a simplicial category

with, for A, B, K S
AK =AK
and (a tautology)
HomS (A, B) = HomS (A, B)
and
homS (K, B) = HomS (K, B) .
Only slightly less obvious is the following: let S denote the category of pointed
(i.e., based) simplicial sets. Then S is a simplicial category with
A K = A K+ = A K/ K
where ( )+ denote adding a disjoint basepoint
HomS (A, B)n = homS (A n+ , B)
and
homS (K, B) = HomS (K, B)
with basepoint given by the constant map
K B.
Note that HomS (A, B) S, but homS (K, B) S .
This example can be radically generalized. Suppose C is a category that is
co-complete; that is, C has all colimits. Let sC denote the simplicial objects in
C. Then if K S, we may define, for A sC, an object A K sC by
G

(A K)n =

An

kKn

F
where
denotes the coproduct in C, and if : n m is an ordinal number

map : (A K)m (A K)n is given by

G
kKm

Am

G
kKm

An

An .

kKn

The first map is induced by : Am An , the second by : Km Kn .

88

II. Model Categories

Theorem 2.5. Suppose that C is complete and complete. Then with this functor
: sC S sC, sC becomes a simplicial category with
HomsC (A, B)n = homsC (A n , B) .
Proof: This is an application of Lemma 2.3. First note that it follows from the
construction that there is a natural isomorphism
A (K L)
= (A K) L.
And one has A
= A. Thus, we need only show that, for fixed K S, the
functor K : C C has a right adjoint. To show this, one changes focus
slightly. For Y sC, define a functor
FY : C op S
by
FY (A) = homC (A, Y ) .
Then the functor C op Sets given by
A 7 HomS (K, FY (A))n = homS (K n , homC (A, Y ))
is representable. To see this, write K n as a coequalizer
G

nq

np K n

then the representing object is defined by the equalizer diagram

Y
q

Ynq

Ymq homsC (K, Y )n .

Letting the ordinal number vary yields an object homsC (K, Y ) and a natural
isomorphism of simplicial sets
(2.6)

homC (A, homsC (K, Y ))

= HomS (K, homC (A, Y )) ,

or a natural equivalence of functors

FhomSC (K,Y ) ()
= HomS (K, FY ()) .

2. Simplicial categories

89

Now the morphisms X Y in sC are in one-to-one correspondence with the

natural transformations FX FY , by the Yoneda lemma. In formulas, this
reads
homsC (X, Y )
= Nat(FX , FY ) .
Now if K S and X sC we can define a new functor
FX K : C op S
by
(FX K)(A) = FX (A) K
we will argue below that
Nat(FXK , FY )
= Nat(FX K, FY ) .
Assuming this one has:
homsC (X, homsC (K, Y ))
= Nat(FX , FhomsC (K,Y ) )

= Nat(FX , HomS (K, FY ))

by (2.6). Continuing, one has that this is isomorphic to
Nat(FX K, FY )
= Nat(FXK , FY )
= homsC (X K, Y )
so that

homsC (X, homsC (K, Y ))

= homsC (X K, Y ) .

as required. Thus we are left with

Lemma 2.7. There is an isomorphism
Nat(FXK , FY )
= Nat(FX K, FY ) .
Proof: It is easiest to show
Nat(FX K, F )
= homsC (X K, Y ) .
Given a natural transformation
: FX K FY

90

II. Model Categories

note that

(FX K)(Xn )n =

homC (Xn , Xn ) .

kKn

Thus, for each k Kn , there is a map

(1)k : Xn Yn
corresponding to the identity in the factor corresponding to k. These assemble
into a map
G
fn : (X K)n =
Xn Yn .
kKn

We leave it to the reader to verify that yields a morphism

f :X K Y
of simplicial objects, and that the assignment f yields the desired isomorphism.

Examples 2.8. One can now assemble a long list of simplicial categories: We
note in particular
1) Let C be one of the following algebraic categories: groups, abelian groups,
rings, commutative rings, modules over a ring R, algebras or commutative algebras over a commutative ring R, or Lie algebras. Then sC is a simplicial
category.
2) Let C be the graded analog of one of the categories in the previous example.
Then sC is a simplicial category. One might wish to examine the category of
graded commutative algebras over a commutative ring R, for example.
3) Let C = CA be the category of coalgebras over a field F. Then sCA is a
simplicial category.
4) Note that the hypotheses of C used Theorem 2.5 apply equally to C op . Thus
s(C op ) is also a simplicial category. But if s(C op ) is a simplicial category, so is
(s(C)op )op . But this is the category cC of cosimplicial objects in C. One must
interpret the functors , homcC (, ), etc. in light of Theorem 2.5. Thus if
K S,
Y
homcC (K, A)n =
An
kKn

and

HomcC (A, B)n = homcC (A, homcC (n , B))

and A K is defined via Theorem 2.5.

3. Simplicial model categories

91

To conclude this section, we turn to the following question: suppose given

simplicial categories C and D and a functor G : D C with left adjoint F . We
want a criterion under which the simplicial structure is preserved.
Lemma 2.9. Suppose that for all K S and A C there is a natural isomorphism F (A K)
= F (A) K. Then
(1) the adjunction extends to a natural isomorphism
HomD (F A, B)
= HomC (A, GB);
(2) for all K S and B D, there is a natural isomorphism
GhomD (K, B)
= homC (K, GB) .
Proof: Part (1) uses that HomD (F A, B)n
= homD (F A n , B). Part (2) is
an exercise in adjunctions.

We give some examples.

Examples 2.10.
1) Let G : D C have a left adjoint F . Extend this to a pair of adjoint functors
by prolongation:
G : sD sC
with adjoint F . Thus G(X)n = G(X)n , and so on. Then, in the simplicial
structure of Theorem 2.5, F (X K)
= F (X) K, since F commutes with
colimits.
2) Let C be an arbitrary simplicial category and A C. Define
G:CS
by G(B) = HomC (A, B). Then F (X) = A X and the requirement on 2.9 is
simply the formula
A (X K)
= (A X) K .
3. Simplicial model categories.
If a category C is at once a simplicial category and a closed model category,
we would like the mapping space functor to have homotopy theoretic content.
This is accomplished by imposing the following axiom.

92

II. Model Categories

3.1 Axiom SM7. Let C be a closed model category and a simplicial category.
Suppose j : A B is a cofibration and q : X Y is a fibration. Then
(j ,q )

HomC (B, X) HomC (A, X) HomC (A,Y ) HomC (B, Y )

is a fibration of simplicial sets, which is trivial if j or q is trivial.
A category satisfying this axiom will be called a simplicial model category.
The next few sections will be devoted to producing a variety of examples, but in
this section we will explore the consequences of this axiom.
Proposition 3.2. Let C be a simplicial model category and q : X Y a
fibration. Then if B is cofibrant
q : HomC (B, X) HomC (B, Y )
is a fibration in S. Similarly, if j : A B is a cofibration and X is fibrant, then
j : HomC (B, X) HomC (A, X)
is a fibration.
Proof: One sets A to be the initial object and Y to be the final object, respectively, in Axiom SM7.

In other words, HomC (, ) has entirely familiar homotopical behavior. This is

one way to regard this axiom. Another is that SM7 is a considerable strengthening of the lifting axiom CM4 of a closed model category.
Proposition 3.3. Axiom SM7 implies axiom CM4; that is, given a lifting
problem in simplicial category C satisfying SM7

wX
i
j
i q
j
i
u i
u
B
wY
with j a cofibration and q a fibration, then the dotted arrow exists if either j or
q is trivial.

3. Simplicial model categories

93

Proof: Such a square is a zero-simplex in

HomC (A, X) HomC (A,Y ) HomC (B, Y )
and a lifting is a pre-image in the zero simplices of HomC (B, X). Since trivial
fibrations are surjective, the result follows.

But more is true: Axiom SM7 implies that the lifting built in CM4 is unique
up to homotopy. To explain that, however, requires a few words about homotopy.
First we record
Proposition 3.4. Let C be a simplicial model category and j : K L a
cofibration of simplicial sets. If A C is cofibrant, then
1j :AK AL
is a cofibration in C. If X C is fibrant
j : homC (L, X) homC (K, X)
is a fibration. If j is trivial, then so are 1 j and j .
Proof: For example, one needs to show 1 j has the left lifting property with
respect to all trivial fibrations q : X Y in C. This is equivalent, by adjointness,
to show j has the left lifting property with respect to
q : HomC (A, X) HomC (A, Y )
for all trivial fibrations q. But q is a trivial fibration of simplicial sets by SM7.
The other three claims are proved similarly.

Recall the definitions of left and right homotopy from the section 1. The
following implies that if A is cofibrant, then A 1 is a model for the the
cylinder on A.
Lemma 3.5. Let C be a simplicial model category and let A C be cofibrant.
Then if q : 1 is the unique map
1 q : A 1 A
=A
is a weak equivalence. Furthermore,
d1 t d0 : A t A A 1
is a cofibration and the composite
d0 td1

1q

A t A A 1 A
is the fold map.

94

II. Model Categories

Proof: The first claim follows from Proposition 3.4, since

d1 : A
= A 0 A 1
is a weak equivalence. The second claim follows from 3.4 also since d1
equivalent to
1 j : A 1 A 1

d0 is

where j : 1 is inclusion of the boundary. For the third claim one checks
that (1 q) d1 = (1 q) d0 = 1.

Thus, if C is a simplicial model category and A C is cofibrant, then two

morphisms f, g : A X are homotopic if and only if there is a factoring
d1 t d0
w A 1
N
N
f tg
N
u N
QN H
X
AtA

(3.6)

This, too, is no surprise. As a further exercise, note that if one prefers right
homotopy for a particular application, one could require a factoring

(3.7)

homC ([1], X)
4
6
G 444
j
4
4
u
4

w X X = homC ([1], X)
A
f g

In using this formulation, one wants X to be fibrant so that j is a fibration.

To formulate the next notion, let A be cofibrant and j : A B a cofibration.
Given two maps f, g : B X so that j f = j g, we say f and g are homotopic
under A if there is a homotopy
H : B 1 X
so that H (j 1) : A 1 X is the constant homotopy on j f . That is,
h (j 1) is the composite
1q

jf

A 1 A
= A X
where q : 1 is the unique map. There is an obvious dual motion of
homotopic over Y .
The following result says that in a simplicial model category, the liftings required by axiom CM4 are unique in a strong way.

3. Simplicial model categories

95

Proposition 3.8. Let C be a simplicial model category and A a cofibrant object.

Consider a commutative square

wX
q

j
u
B

u
wY

where j is a cofibration, q is a fibration and one of j or q is trivial. Then any

two solutions f, g : B X of the lifting problem are homotopic under A and
over Y .
Proof: The commutative square is a zero-simplex in
HomC (A, X) HomC (A,Y ) HomC (B, Y )
Let s0 be the corresponding degenerate 1-simplex. Then s0 is the commutative
square
A 1

wX
q

j1
u
B 1

u
wY

where the horizontal maps are the constant homotopies. Let

f, g HomC (B, X)0 = homC (B, X)
be two solutions to the lifting problem. Thus (i , q )f = (i , q )g = . Then by
SM7, there is a 1-simplex
HomC (B, X)1
= homC (B 1 , X)
so that d1 = f, d0 = g, and (i , q ) = s0 . Then
: B 1 X
is the required homotopy.

96

II. Model Categories

We now restate a concept from Section 1. For a simplicial model category C,

we define the homotopy category Ho(C) as follows: the objects are the objects
of C and the morphisms are defined by
(3.9)

[A, X]C = homC (B, Y )/

where q : B A is a trivial fibration with B cofibrant, i : X Y is a trivial

cofibration with Y fibrant and f g if and only if f is homotopic to g. It is
an exercise that is in fact an equivalence relation, and it is a consequence of
Proposition 3.8 that [A, X]C is well-defined.
There is some ambiguity in the notation: [A, X]C depends not only on C, but
on the particular closed model category structure.
In the sequel, [ , ] means [ , ]S .
The following result gives homotopy theoretic content to the functors K
and homC (K, ).
Proposition 3.10. Let C be a simplicial model category and A and B a cofibrant and a fibrant object of C, respectively. Then
[K, HomC (A, B)]
= [A K, B]C
and

[K, HomC (A, B)]

= [A, homC (K, B)]C .

Proof: Note that HomC (A, B) is fibrant, by Proposition 3.2. Hence

[K, HomC (A, B)] = homS (K, HomC (A, B))/
where means homotopy as above. But, since
A (K 1 ) = (A K) 1
we have that
homS (K, HomC (A, B))/
= homC (A K, B)/

= [A K, B]C
where we use Proposition 3.4 to assert that A K is cofibrant.

We now concern ourselves with developing a way of recognizing when SM7

holds.

3. Simplicial model categories

97

Proposition 3.11. Let C be a closed model category and a simplicial category.

Then the axiom SM7 holds if and only if for all cofibrations i : K L in S and
cofibrations j : A B in C, the map
(j 1) (1 i) : (A L) (AK) (B K) B L .
is a cofibration which is trivial if either j or i is.
Proof: A diagram of the form
K

w HomC (B, X)

u
L

u
w HomC (A, X) HomC (A,Y ) HomC (B, Y )

is equivalent, by adjointness, to a diagram

(A L) (AK) (B K)

wX

u
BL

u
wY .

The result follows by using the fact that fibrations and cofibrations are determined by various lifting properties.

From this we deduce

Corollary 3.12. (Axiom SM7b) Let C be a closed model category and a
simplicial category. The axiom SM7 is equivalent to the requirement that for
all cofibrations j : A B in C
(A n ) (An ) (B n ) B n
is a cofibration (for n 0) that is trivial if j is, and that
(A 1 ) (A{e}) (B {e}) B 1
is the trivial cofibration for e = 0 or 1.

98

II. Model Categories

Proof: Let i : K L be a cofibration of simplicial sets. Then, since i can be

built by attaching cells to K, the first condition implies
(A L) (AK) (B K) (B L)
is a cofibration which is trivial if j is. The second condition and proposition I.4.2
(applied to B2 ) yields that (j 1) (1 i) is trivial if i is.

In the usual duality that arises in these situations, we also have

Proposition 3.13. Let C be a simplicial category and a model category and
suppose i : K L is a cofibration in S and q : X Y a fibration in C. Then
the following are equivalent:
(1) SM7,
(2) homC (L, X) homC (K, X)homC (K,Y ) homC (L, Y ) is a fibration which
is trivial if q or j is;
(3) (SM7a) homC (n , X) homC (n , X)homC (n ,Y ) homC (n , Y ) is
a fibration which is trivial if q is, and
homC (1 , X) homC (e, X) homC (e,Y ) homC (1 , Y )
is a trivial fibration for e = 0, 1.
Example 3.14. A simplicial model category structure on CGHaus We can now
show that the category CGHaus of compactly generated Hausdorff spaces is a
simplicial model category. To supply the simplicial structure let X CGHaus
and K S. Define
X K = X Ke |K|
where | | denotes the geometric realization and Ke the Kelly product, which
is the product internal to the category CGHaus. Then if X and Y are in
CGHaus, the simplicial set of maps between X and Y is given by
HomCGHaus (X, Y )n = homCGHaus (X n , Y )
regarded as a set. And the right adjoint to K is given by
homCGHaus (K, X) = F(|K|, X)
where F denotes the internal function space to CGHaus.

4. Detecting weak equivalences

99

We have seen that CGHaus is a closed model category with the usual weak
equivalences and Serre fibrations. In addition, Proposition 3.13 immediately
implies that CGHaus is a simplicial model category.
It is worth pointing out that the realization functor | | and its adjoint S()
the singular set functor pass to the level of simplicial categories. Indeed, weve
seen that if X S and K S, then
|X K| = |X| Ke |K|.
This immediately implies that if Y CGHaus, then
SF(|K|, Y ) = HomS (K, SY )
and
HomCGHaus (|X|, Y ) = HomS (X, SY ).
4. Detecting weak equivalences.
Lemma 4.1. Suppose that f : X Y is a morphism of a simplicial model
category C which induces an isomorphism in the homotopy category Ho(C). Then
f is a weak equivalence.
Proof: The map f induces an isomorphism in the homotopy category if and
only if f induces a homotopy equivalence RQX RQY of the associated cofibrant fibrant models. It suffices, therefore, to assume that X and Y are fibrant
and cofibrant, and that f : X Y is a homotopy equivalence.
Choose a factorization
j
X4
wZ
4
46 p
f 4
u
Y
where j is a trivial cofibration and p is a fibration. The map j is a homotopy
equivalence, so it suffices to show that the fibration p : Z Y of cofibrant
fibrant objects is a weak equivalence if it is a homotopy equivalence.
Choose a homotopy inverse s : Y Z for the map p, and choose a homotopy
H : Z 1 Z from ps to the identity 1Y . There is a diagram
Y

wZ
h
j
H h
p
d0
h
u
u
h
Y 1
w Y.
H

100

II. Model Categories

The composite map H d1 is a section of p, and this composite is homotopic to

s. We can therefore assume that the homotopy inverse s is a section of p.
Choose a homotopy h : Z 1 Z from sp to 1Z . The lifting exists in the
diagram
20
y

( , h, sph)

w Hom(Z, Z)
A
C
AA
p
A
u AA
u
2
w Hom(Z, Y )
s1 (ph)
since p is a fibration. Write k : Z 1 Z for the 1-simplex d0 () of
Hom(X, X). Then k is a fibrewise homotopy from sp to 1Z , meaning in particular that the following diagram commutes:
k

Z 1

wZ
p

s0

u
Z

u
w Y.

Now suppose that the diagram

wZ
p

j
u
B

u
wY

commutes, where j is a cofibration. One can form the diagram

(k( 1), s)
wZ
)
'
'
'
K
p
''
j
'
'
u
u
''
1
B
wY
s0 ( 1)

(A 1 ) (B {0})

4. Detecting weak equivalences

101

since the map j is a trivial cofibration. Now check that the map defined by
the composite
d0

B B 1 Z
satisfies p = and j = . The map p therefore has the right lifting property
with respect to all cofibrations, and is therefore a trivial fibration.

It may seem unlikely, but Lemma 4.1 is probably the the most important
technical consequence of having a simplicial model structure in hand, as opposed
to just a closed model structure. It will be used many times in the sequel. Here
is a sample application:
Lemma 4.2. Suppose that f : A B is a map between cofibrant objects in
a simplicial model category C. Then f is a weak equivalence if and only if the
induced map
f : Hom(B, Z) Hom(A, Z)
is a weak equivalence of simplicial sets for each fibrant object Z of C.
Proof: We use the fact, which appears as Lemma II.9.4 below, that a map f :
A B between cofibrant objects in a closed model category has a factorization
X
j
j hh
q
h
u
h
A
wB
f
such that j is a cofibration and the map q is left inverse to a trivial cofibration
i : B X.
If f : A B is a weak equivalence, then the map j : A X is a trivial cofibration, and hence induces a trivial fibration j : Hom(X, Z) Hom(A, Z) for
all fibrant objects Z. Similarly, the trivial cofibration i induces a trivial fibration
i , so that the map p : Hom(B, Z) Hom(X, Z) is a weak equivalence.
Suppose that the map f : Hom(B, Z) Hom(A, Z) is a weak equivalence
for all fibrant Z. To show that f is a weak equivalence, we can presume that the
objects A and B are fibrant as well as cofibrant. In effect, there is a commutative
diagram
A
u
A

wB
u
wB

102

II. Model Categories

in which the objects A and B are fibrant, and the vertical maps are trivial
cofibrations, and then one applies the functor Hom( , Z) for Z fibrant and
invokes the previous paragraph.
Finally, suppose that A and B are fibrant as well as cofibrant, and presume
that f : Hom(B, Z) Hom(A, Z) is a weak equivalence for all fibrant Z. We
can assume further that f is a cofibration, by taking a suitable factorization.
The map f : Hom(B, A) Hom(A, A) is therefore a trivial Kan fibration,
and hence surjective in all degrees, so that there is a map g : B A such that
g f = 1A . The maps f g and 1B are both preimages of the vertex f under the
trivial fibration f : Hom(B, B) Hom(A, B), so that there is a homotopy
f g ' 1B . In particular, f is a homotopy equivalence and therefore a weak
equivalence, by Lemma 4.1.

5. The existence of simplicial model category structures.

Here we concern ourselves with the following problem: Let C be a category and
sC the category of simplicial objects over C. Then, does sC have the structure of
a simplicial model category? We will assume that there is a functor G : sC S
with a left adjoint
F : S sC .
Examples include algebraic categories such as the categories of groups, abelian
groups, algebras over some ring R, commutative algebras, lie algebras, and so
on. In these cases, G is a forgetful functor. See Example 2.10.
Define a morphism f : A B is sC to be
a) a weak equivalence if Gf is a weak equivalence in S;
b) a fibration if Gf is a fibration in S;
c) a cofibration if it has the left lifting property with respect to all trivial
fibrations in sC.
A final definition is necessary before stating the result. Let {X }I be a
diagram in C. Then, assuming the category C has enough colimits, there is a
natural map
lim G(X ) G(lim X ).

This is not, in general, an isomorphism. We say that G commutes with filtered

colimits if this is an isomorphism whenever the index category I is filtered.
Theorem 5.1. Suppose C has all limits and colimits and that G commutes
with filtered colimits. Then with the notions of weak equivalence, fibration, and
cofibration defined above, sC is a closed model category provided the following
assumption on cofibrations holds: every cofibration with the left lifting property
with respect to fibrations is a weak equivalence.

5. The existence of simplicial model category structures 103

We will see that, in fact, sC is a simplicial model category with the simplicial
structure of Theorem 2.5.
The proof of Theorem 5.1 turns on the following observation. As we have seen,
a morphism f : X Y is a fibration of simplicial sets if and only if it has the
right lifting property with respect to the inclusions for all n, k
nk , n
and f is a trivial fibration if and only if it has the right lifting property with
respect to the inclusions n n of the boundary for all n. The objects
nk , n , and n are small in the following sense: the natural map
lim homS (nk , X ) homS (nk , lim X )

is an isomorphism for all filtered colimits in S. This is because nk has only

finitely many non-degenerate simplices. Similar remarks hold for n and n .
Lemma 5.2. Any morphism f : A B in sC can be factored
j

AX B
where the morphism j is a cofibration and q is a trivial fibration.
Proof: Coproducts of cofibrations are cofibrations, and given a pushout diagram
A0

w A2

j
u
wB

u
A1

in sC, then i a cofibration implies j is a cofibration, and that if X Y is a

cofibration in S, then F X F Y is a cofibration in sC. Inductively construct
objects Xn sC with the following properties:
a) One has A = X0 and there is a cofibration jn : Xn Xn+1 .
b) There are maps qn : Xn B so that qn = qn+1 jn and the diagram
A4

w Xn
4
4
4
6 qn
u
B

104

II. Model Categories

commutes, where A Xn is the composite jn1 j0 .
c) Any diagram

F m
u
F m

w Xn
qn
u
wB

can be completed to a diagram

F m

w Xn '

'

u
F m

jn

w Xn+1

qn
'
'
'
)

qn+1 w B

where the bottom morphism is .

Condition c) indicates how to construct Xn+1 given Xn . Define jn : Xn
Xn+1 by the pushout diagrams
G

F m

w Xn
jn

G u
F m

u
w Xn+1

where the coproduct is over all diagrams of the type presented in c).
Then condition c) automatically holds. Further, qn+1 : Xn+1 B is defined
and satisfies condition b) by the universal property of pushouts. Lastly, condition
a) holds by the remarks at the beginning of the proof.
Now define X = lim Xn and notice that we have a factoring

AX B

5. The existence of simplicial model category structures 105

of the original morphism. The morphism j is a cofibration since directed colimits
of cofibrations are cofibrations. We need only show q : X B is a trivial
fibration. This amounts to showing that any diagram
n

w GX

u
n

Gq
u
w GB

can be completed. But GX

= lim GXn by hypothesis on G, and the result
n
follows by the small object argument.

The same argument, but using the trivial cofibrations in S

nk , n
for n, k 0, proves the following lemma.
Lemma 5.3. Any morphism f : A B in sC can be factored
j

A
X
B
where q is a fibration and j is a cofibration which has the left lifting property
with respect to all fibrations.
Proof of Theorem 5.1: The Axioms CM1CM3 are easily checked. The
axiom CM5b is Lemma 5.2; the axiom CM5a follows from Lemma 5.3 and
the assumption on cofibrations. For axiom CM4, one half is the definition of
cofibration. For the other half, one proceeds as follows. Let
i:AB
be a trivial cofibration. Then by Lemma 5.3 we can factor the morphism i as
j

AX B
where j is a cofibration with the left lifting with respect to all fibrations, and q
is a fibration. By the hypothesis on cofibrations, j is a weak equivalence. Since
i is a weak equivalence, so is q. Hence, one can complete the diagram
A

u
B

wX
q
u
wB

106

II. Model Categories

and finds that i is a retract of j. Hence i has the left lifting property with respect
to fibrations, because j does. This completes the proof.

We next remark that, in fact, sC is a simplicial model category. For this, we

impose the simplicial structure guaranteed by Theorem 2.5. Thus if X sC and
K S, we have that
G
(A K)n =
An .
kKn

From this, one sees that if X S

F (X K)
= F (X) K.
This is because F , as a left adjoint, preserves coproducts. Thus Lemma 2.9
applies and
GhomsC (K, B)
= homS (K, GB).
Theorem 5.4. With this simplicial structure, sC becomes a simplicial model
category.
Proof: Apply Proposition 3.13.1. If j : K L is a cofibration in S and
q : X Y is a fibration in sC, the map
GhomsC (L, X) G(homsC (K, X) homsC (K,Y ) homsC (L, Y ))
is isomorphic to
homS (L, GX) homS (K, GX) hom

homS (L, GY )
S (K,GY )
by the remarks above and the fact that G, as a right adjoint, commutes with
pullbacks. Since S is a simplicial model category, the result holds.

5.5 A remark on the hypotheses. Theorem 5.1 and, by extension, Theorem 5.4 require the hypothesis that every cofibration with the left lifting property
with respect to all fibrations is, in fact, a weak equivalence. This is so Lemma 5.3
produces the factoring of a morphism as a trivial cofibration followed by a fibration. In the next section we will give some general results about when this
hypothesis holds; however, in a particular situation, one might be able to prove
directly that the factoring produced in Lemma 5.3 actually yields a trivial cofibration. Then the hypothesis on cofibrations required by these theorems holds
because any cofibration with the left lifting property with respect to all fibrations
will be a retract of a trivial cofibration. Then one need say no more.
For example, in examining the proof of Lemma 5.3 (see Lemma 5.2), one
sees that we would have a factorization of f : A B as a trivial cofibration
followed by a fibratiion provided one knew that 1.) F (nk ) F (n ) was a weak
equivalence or, more generally, that F preserved trivial cofibrations, and 2.)
trivial cofibrations in sC were closed under coproducts, pushouts, and colimits
over the natural numbers.

6. Examples of simplicial model categories

107

6. Examples of simplicial model categories.

As promised, we prove that a variety of simplicial categories satisfy the hypotheses necessary for Theorem 5.4 of the previous section to apply.
We begin with a crucial lemma.
Lemma 6.1. Assume that for every A sC there is a natural weak equivalence
A : A QA
where QA is fibrant. Then every cofibration with the left lifting property with
respect to all fibrations is a weak equivalence.
Proof: This is the argument given by Quillen, on page II.4.9 of [75]. Let
j : A B be the given cofibration. Then by hypothesis, we may factor
A

A
j

j
ii

u i
B

w QA
u
w

to get a map u : B QA so that uj = A . Then we contemplate the lifting

problem
f

w homsC (1 , QB)
6
55
5
j
q
55
5
u 5
u
g
1
B
w homsC ( , QB)
= QB QB
A

where q is induced by 1 1 , f is the composite

j

AA
B QB = homsC (, QB) homsC (1 , QB)
and
g = (B , Qj u) .
Note that f is adjoint to the constant homotopy on
B j = Qj A : A QB .

108

II. Model Categories

Then q is a fibration since

GhomsC (K, X) = homS (K, GX) ,
and S is a simplicial model category. Hence, since j is a cofibration, there exists
H : B homsC (1 , QB)
making both triangles commute. Then H is a right homotopy from
B : B QB
to Qj u, and this homotopy restricts to the constant homotopy on B j =
Qj A : A QB. In other words, we have a diagram

w QA

h
j
uh
Qj
h
u h
u
B
w QB

such that the upper triangle commutes and the lower triangle commutes up
to homotopy. Apply the functor G to this diagram. Then G preserves right
homotopies, and one checks directly on the level of homotopy groups that Gj is
a weak equivalence, which, by definition, implies j is a weak equivalence.

Example 6.2. Suppose every object of sC is fibrant. Then we may take A :

A QA to be the identity. This happens, for example, if the functor G : sC
S factors through the sub-category of simplicial groups and simplicial group
homomorphisms. Thus, Theorem 5.4 applies to
simplicial group s, simplicial abelian groups and simplicial R-modules, where G is the
forgetful functor;
(1) more generally to simplicial modules over a simplicial ring R, where G is
the forgetful functor,
(2) for a fixed commutative ring R; simplicial R-algebras, simplicial commutative R-algebras and simplicial Lie algebras over R. Again G is the
forgetful functor.

6. Examples of simplicial model categories

109

Another powerful set of examples arises by making a careful choice of the form
the functor G can take.
Recall that an object A C is small if homC (A, ) commutes with filtered
colimits. Fix a small Z C and define
(6.3)

G : sC S

by
G(X) = homsC (Z, X) .
Then G has left adjoint
FK = Z K
and G() commutes with filtered colimits. Thus, to apply Theorem 5.4, we need
to prove the existence of the natural transformation
: A QA
as in Lemma 6.1. Let
Ex : S S
be Kans Extension functor1 . Then for all K S there is a natural map
K : K Ex K
which is a weak equivalence. Furthermore, most crucially for the application
here, Ex() commutes with all limits. This is because its a right adjoint. Finally,
if Exn K is this functor applied n times and
Exn K : Exn K Exn+1 K
the induced morphism, then Ex K = lim Exn K is fibrant in S and the induced

map
K Ex K
is a trivial cofibration.
Lemma 6.4. Suppose the category C is complete and cocomplete. Fix n 0.
Then there is a functor
Q0 ()n : sC C
1

This construction is discussed in Section III.4 below.

110

II. Model Categories

so that, for all Z C, there is a natural isomorphism of sets

Ex(HomsC (Z, A))n .
homC (Z, (Q0 A)n ) =
Proof: Recall that that the functor Ex on S is right adjoint to the the subdivision functor sd. Then one has a sequence of natural isomorphisms
Ex HomsC (Z, A)n
= homS (n , Ex HomsC (Z, A))

= homS (sd n , HomsC (Z, A))

= homsC (Z sd n , A)

= homsC (Z, homsC (sd n , A))

= homC (Z, homsC (sd n , A)0 ).

The last isomorphism is due to the fact that Z is a constant simplicial object
and maps out of a constant simplicial object are completely determined by what
happens on zero simplices. Thus we can set
(Q0 A)n = homsC (sd n , A)0 .

The simplicial object Q0 A defined by

n 7 (Q0 A)n = homsC (sd n , A)0
is natural in A; that is, we obtain a functor Q0 : sC sC. Since we regard Z C
as a constant simplicial object in sC
HomsC (Z, Y )n
= homsC (Z n , Y )

= homC (Z, Yn )
one immediately has that
HomsC (Z, Q0 A)
= Ex HomSC (Z, A) .
Finally the natural transformation K : K Ex K yields a natural map
A : A Q0 A
and, by iteration, maps
A.
Qn0 A : Qn0 A Qn+1
0
Define QA = lim Qn0 A. The reader will have noticed that Q0 A and QA are

independent of Z.
Now fix a small object Z C and regard Z as a constant simplicial object
in sC. Then we define a morphism A B in sC to be a weak equivalence (or
fibration) if and only if the induced map
HomsC (Z, A) HomsC (Z, B)
is a weak equivalence (or fibration) of simplicial sets.

6. Examples of simplicial model categories

111

Proposition 6.5. If Z is small, the morphism A : A QA is a weak equivalence and QA is fibrant.

Proof: Since Z is small, we have that
HomsC (Z, QA)
= Ex HomsC (Z, A).
The morphism A is a weak equivalence if and only if
HomsC (Z, A) HomsC (Z, QA)
= Ex HomsCa (Z, A)
is a weak equivalence and QA is a fibration if and only if
HomsC (Z, QA)
= Ex HomsC (Z, A) HomsC (Z, ) =
is a fibration. Both of these facts follow from the properties of the functor
Ex ().

Corollary 6.6. Let C be a complete and cocomplete category and Z C

a small object. Then sC is a simplicial model category with A B a weak
equivalence (or fibration) if and only if
HomsC (Z, A) HomsC (Z, B)
is a weak equivalence (or fibration) of simplicial sets.
In practice one wants an intrinsic definition of weak equivalence and fibration,
in the manner of the following example.
Example 6.7. All the examples of 6.2 can be recovered from Corollary 6.6. For
example, C be the category of algebras over a commutative ring. Then C has a
single projective generator; namely A[x], the algebra on one generator. Then one
sets Z = A[x], which is evidently small, and one gets a closed model category
structure from the previous result. However, if B sC, then
HomsC (A[x], B)
=B
in the category of simplicial sets, so one recovers the same closed model category
structure as in Example 6.2.
If C is a category satisfying 5.1 with a single small projective generator, then C
is known as a category of universal algebras. Setting Z to be the generator, one
immediately gets a closed model category structure on sC from Corollary 6.6.
This is the case for all the examples of 6.2.

112

II. Model Categories

To go further, we generalize the conditions of Theorem 5.1 a little, to require

the existence of a collection of functors Gi : sC S, i I, each of which has a
left adjoint Fi : S sC. We now say that a morphism f : A B of sC is
a) a weak equivalence if Gi f is a weak equivalence of simplicial sets for all
i I;
b) a fibration if all induced maps Gi f are fibrations of S;
c) a cofibration if it has the left lifting property with respect to all trivial
cofibrations of sC.
Then Theorem 5.1 and Theorem 5.4 together have the following analogue:
Theorem 6.8. Suppose that C has all small limits and colimits and that all of
the functors Gi : C S preserve filtered colimits. Then with the notions of weak
equivalence, fibration and cofibration defined above, and if every cofibration with
the left lifting property with respect to all fibrations is a weak equivalence, then
sC is a simplicial model category.
Proof: The proof is the same as that of Theorem 5.1, except that the small
object arguments for the factorization axiom are constructed from all diagrams
of the form
Fi m

wA

u
Fi m

u
wB

Fi nk

wA

u
F i n

u
wB

and

Theorem 6.8 will be generalized significantly in the next section it is a

special case of Theorem 7.8.
Now fix a set of small objects Zi C, i I, and regard each Zi as a constant
simplicial object in sC. Then we define a morphism A B in sC to be a weak
equivalence (or fibration) if and only if the induced map
HomsC (Zi , A) HomsC (Zi , B)
is a weak equivalence (or fibration) of simplicial sets. In the case where C is
complete and cocomplete, we are still entitled to the construction of the natural
map A : A QA in sC. Furthermore, each of the objects Zi is small, so that
Proposition 6.5 holds with Z replaced by Zi , implying that the map A is a weak
equivalence and that QA is fibrant. Then an analogue of Lemma 6.1 holds for
the setup of Theorem 6.8 (with G replaced by Gi in the proof), and we obtain
the following result:

6. Examples of simplicial model categories

113

Theorem 6.9. Suppose that C is a small complete and cocomplete category,

and let Zi C, i I, be a set of small objects. Then sC is a simplicial model
category with A B a weak equivalence (respectively fibration) if and only if
the induced map
HomsC (Zi , A) HomsC (Zi , B)
is a weak equivalence (respectively fibration) for all i I
Example 6.10. Suppose that C is small complete and cocomplete, and has a set
{P } of small projective generators. Theorem 6.9 implies that C has a simplicial
model category structure, where A B is a weak equivalence (or fibration) if
HomsC (P , A) HomsC (P , B)
is a weak equivalence (or fibration) for all .
Note that the requirement that the objects P are projective generators is not
necessary for the existence of the closed model structure. However, if we also
assume that the category C has sufficiently many projectives in the sense that
there is an effective epimorphism P C with P projective for all objects C C,
then it can be shown that a morphism f : A B of sC is a weak equivalence
(respectively fibration) if every induced map
HomsC (P, A) HomsC (P, B)
arising from a projective object P C is a weak equivalence (respectively fibration) of simplicial sets. This is a result of Quillen [75, II.4], and its proof is the
origin of the stream of ideas leading to Theorem 6.9. We shall go further in this
direction in the next section.
To be more specific now, let C be the category of graded A-algebras for some
commutative ring A and let, for n 0,
Pn = A[xn ]
be the free graded algebra or an element of degree n. Then {Pn }n0 form a set
of projective generators for C. Thus sC gets a closed model category structure
and B C in sC is a weak equivalence if and only if
(B)n (C)n
is a weak equivalence of simplicial sets for all n. Here ()n denotes the elements
of degree n. This is equivalent to the following: if M is a simplicial graded
A-module, define
M = H (M, )

114

II. Model Categories

where is the alternating of the face operators. Then B C in sC is a weak

equivalence if and only if
B C
is an isomorphism of bigraded A-modules.
This formalism works for graded groups, graded abelian groups, graded Amodules, graded commutative algebras, graded Lie algebras, and so on.
Example 6.11. Let F be a field and let C = CA be the category of coalgebras
over F. Then, by [87, p.46] every coalgebra C CA is the filtered colimit of its
finite dimensional sub-coalgebras. Thus CA has a set of generators {C } where
C runs over a set of representatives for the finite dimensional coalgebras. These
are evidently small. Hence, sCA has a closed model category structure where
A B is a weak equivalence if and only if
HomsCA (C , A) HomsCA (C , B)
is a weak equivalence for all C . The significance of this example is that the C
are not necessarily projective.
7. A generalization of Theorem 5.1.
The techniques of the previous sections are very general and accessible to vast
generalization. We embark some ways on this journey here. First we expand on
what it means for an object in a category to be small. Assume for simplicity
that we are considering a category C which has all limits and colimits. We shall
use the convention that a cardinal number is the smallest ordinal number in a
given bijection class.
Fix an infinite cardinal number , and let Seq() denote the well-ordered set
of ordinals less than . Then Seq() is a category with hom(s, t) one element
if s t and empty otherwise. A -diagram in C is a functor X : Seq() C.
We will write lim Xs for the colimit. We shall say that X is a -diagram of

cofibrations if each of the transition morphisms Xs Xt is a cofibration of C.

Definition 7.1. Suppose that is an infinite cardinal. An object A C is
-small if for all -diagrams of cofibrations X in C with , the natural map
lim homC (A, Xs ) homC (A, lim Xs )

is an isomorphism. A morphism A B of C is said to be -small if the objects

A and B are both -small.

7. A generalization of Theorem 5.1

115

Example 7.2. The small objects of the previous sections were -small, where
is the first infinite cardinal. Compact topological spaces are also -small, but
this assertion requires proof.
Supposethat X : Seq() Top is a -diagram of cofibrations. Then X is a
retract of a -diagram of cofibrations X, where each of the transition morphisms
X s , X t is a relative CW-complex. In effect, set X 0 = X0 , and set X =
lim
X for limit ordinals < . Suppose given maps
s< s
rs

Xs X s Xs
with s rs = 1. Then X s+1 is defined by choosing a trivial fibration s+1 and a
relative CW-complex map js+1 : X s , X s+1 (ie. X s+1 is obtained from X s by
attaching cells, and js+1 is the corresponding inclusion), such that the following
diagram commutes:
Xs
s
u
Xs

js+1

w X s+1
s+1
u

is+1

w Xs+1 ,

where the map is+1 : Xs Xs+1 is the cofibration associated to the relation
s s + 1 by the functor X. Then there is a lifting in the diagram
Xs
is+1

js+1 rs

w X s+1

j
rs+1 hh
h
u h

Xs+1

s+1
u
w Xs+1

so that the section rs extends to a section rs+1 of the trivial fibration s+1 . The
inclusion
X0 = X 0 , lim X s

is a relative CW-complex map, and every compact subset of the colimit only
meets finitely many cells outside of X0 . Every compact subset of lim X s is

therefore contained in some subspace X s . It follows that every compact subset

of lim Xs is contained in some Xs .

116

II. Model Categories

We next produce an appropriate generalization of saturation.

Definition 7.3. A class M of morphisms in C is -saturated if it is closed
under
1) retracts: Suppose there is a commutative diagram in C

w X0

u
w Y0

wX

i0

u
Y

u
wY

with the horizontal composition the identity. Then if i0 M , then

i M.
F
F
F
2) coproducts: if each j : X Y is in M, then j : X Y
is in M;
3) pushouts: given a pushout diagram in C

wX

u
w Y,

u
B

if i is in M, then so is j.
4) colimits of -sequences: Suppose we are given a -sequence
X : Seq() C
with the following properties: a) for each successor ordinal s+1 Seq(),
the map Xs Xs+1 is in M, and b) for each limit ordinal s Seq(),
the map lim
X Xs is in M. Then
t<s t
Xs lim Xs

is in M for all s Seq().

7. A generalization of Theorem 5.1

117

Up until now we have considered only saturated clesses of morphisms with

= , the cardinality of a countable ordinal. In this case, one doesnt need the
extra care required in making the definition of what it means to be closed under
colimits.
Lemma 7.4. Let C be a closed model category. Then the class of cofibrations
and the class of trivial cofibrations are both -saturated for all .
Proof: This is an exercise using the fact that cofibrations (or trivial cofibrations) are characterized by the fact that they have the left lifting property with
respect to trivial fibrations (or fibrations).

The next step is to turn these concepts around.

Definition 7.5. Let M0 be a class of morphisms in C. Then the -saturation
of M0 is the smallest -saturated class of morphisms in C containing M0 .
We now come to the crucial axiom.
Definition 7.6. A closed model category is cofibrantly generated with respect
to a cardinal if the class of cofibrations and the class of trivial cofibrations are
the -saturations of sets of -small morphisms M0 and M1 respectively.
Remarks 7.7.
1) Suppose that and are cardinals such that . Then every -saturated
class is -saturated, because every sequence X : Seq() C can be extended
to a sequence X : Seq() C having the same colimit. It follows that the
-saturation of any set of morphisms is contained in its -saturation. Observe
also that every -small object is -small, directly from Definition 7.1. The size
of the cardinal in Definition 7.6 therefore doesnt matter, so long as it exists.
One says that the closed model category C is cofibrantly generated in cases where
the cardinal can be ignored.
2) Until now, weve taken to . Then the category of simplicial sets is cofibrantly generated, for example, by the usual small object argument. Similarly,
modulo the care required for the assertion that finite CW-complexes are -small
(Example 7.2), the category of topological spaces is cofibrantly generated with
respect to . We will see larger cases later.
3) One could require one cardinal 0 for cofibrations and 1 for trivial cofibrations. However, = max{0 , 1 } would certainly work in either case, by 1).
To give the generalization of Theorem 5.1 we establish a situation. We fix a
simplicial model category C and a simplicial category D. Suppose we have a set
of functors Gi : D C, indexed by the elements i in some set I, and suppose

118

II. Model Categories

each Gi has a left adjoint Fi which preserves the simplicial structure in the sense
that there is a natural isomorphism
Fi (X K)
= F (Xi ) K
for all X C and K S. Define a morphism f : A B in D to be a weak
equivalence (or fibration) is
Gi f : Gi A Gi B
is a weak equivalence (or fibration).
Theorem 7.8. Suppose the simplicial model category C is cofibrantly generated
with respect to a cardinal , and that
(1) all of the functors Gi commute with colimits over Seq(), and
(2) the functors Gi take the -saturation of the collection of all maps Fj A
Fj B arising from maps A B in the generating family for the cofibrations
of C and elements j of I to cofibrations of C.
Then if every cofibration in D with the left lifting property with respect to all
fibrations is a weak equivalence, D is a simplicial model category.
Proof (Outline): There are no new ideasonly minor changes from the arguments of Section 5. The major difference is in how the factorizations are
j
q
constructed. For example, to factor X Y as X Z Y where j is a cofibration which has the left lifting property with respect to all fibrations and q is a
fibration, one forms a -diagram {Zs } is D where
i) Z0 = X;
ii) if s Seq() is a limit ordinal, Zs = lim
Z and
t<s t
iii) if s + 1 is a successor ordinal, there is a pushout diagram
GG
i

Fi (A)

GGu
Fi (B)
i

w Zs

u
w Zs+1

7. A generalization of Theorem 5.1

119

where f runs over all diagrams

w Gi (Zs )

u
B

u
w Gi (Y )

where A B is in the set M1 of -small cofibrations in C whose saturation is all trivial cofibrations.

Example 7.9. Suppose that C is a cofibrantly generated simplicial model category and I is a fixed small category. Write C I for the category of functors
X : I C and natural transformations between them. There are i-section functors Gi : C I C defined by Gi X = X(i), i I, and each such Gi has a left
adjoint Fi : C C I defined by
Fi D(j) =

D.

ij in I

Say that a map X Y of C I is a pointwise cofibration if each of the maps Gi X

Gi Y is a respectively cofibration of C. If A B is a generating cofibration for
C, the induced maps Fi A(j) Fi B(j) are disjoint unions of cofibrations and
hence are cofibrations of C. The induced maps maps Fi A Fi B are therefore
pointwise cofibrations of C I . The functors Gj preserve all colimits, and so the
collection of pointwise cofibrations of C I is saturated (meaning -saturated for
some infinite cardinal similar abuses follow). The saturation of the collection
of maps Fi A Fi B therefore consists of pointwise cofibrations of C I .
A small object argument for C I produces a factorization
j
X4
wZ
4
46 q
f 4
u
Y
for an arbitrary map f : X Y of C I , with q a fibration, and for which j is in
the saturation of the collection of maps Fi C Fi D arising from the generating

120

II. Model Categories

set C D for the class of trivial cofibrations of C. But again, each induced
map Fi C(j) Fi D(j) is a trivial pointwise cofibration of C, and the j-section
functors preserve all colimits. The collection of maps of C I which are trivial
cofibrations in sections is therefore saturated, and hence contains the saturation
of the maps Fi C Fi D. It follows that the map j is a weak equivalence as well
as a cofibration. In particular, by a standard argument, every map of C I which
has the left lifting property with respect to all fibrations is a trivial cofibration.
It therefore follows from Theorem 7.8 that every diagram category C I taking
values in a cofibrantly generated simplicial model category has a simplicial model
structure for which the fibrations and weak equivalences are defined pointwise.
This result applies in particular to diagram categories TopI taking values in
topological spaces.
Heres the analog of Lemma 6.1:
Proposition 7.10. Suppose there is a functor Q : D D so that QX is fibrant
for all X and there is a natural weak equivalence X : X QX. Then every
cofibration with the left lifting property with respect to all fibrations is a weak
equivalence.
Proof: The argument is similar to that of Lemma 6.1; in particular, it begins
the same way.
The map j has the advertised lifting property, so we may form the diagrams
A

w QA

j
hh
hu

j
u h
B

u
w

and
A
j

sB j

w hom(1 , QB)

P
N
N
N

N
u N
u
B
w QB QB
(B , Qj u)

where sB j is the constant (right) homotopy on the composite

j

A
B QB.

7. A generalization of Theorem 5.1

121

The functors Gi preserve right homotopies, so the diagram

A

w QA

h
j
uh
Qj
h
u h
u
B
w QB

remains homotopy commutative after applying each of the functors Gi . It follows that the map Gi (j) is a retract of the map Gi (A ) in the homotopy category
Ho(C), and is therefore an isomorphism in Ho(C). But then a map in a simplicial model category which induces an isomorphism in the associated homotopy
category must itself be a weak equivalence: this is Lemma 4.1.

Example 7.11. The factorization axioms for a cofibrantly generated simplicial

model category C can always be proved with a possibly transfinite small object argument (see, for example, the proofs of Proposition V.6.2 and Lemma
IX.1.17). Such arguments necessarily produce factorizations which are natural
for all objects
in morphisms in C, so that there is a natural fibrant model X , X
X of C. It follows that there are natural fibrant models for the objects of any
diagram category C I taking values in C. We therefore obtain a variation of the
proof of the existence of the closed model structure for C I of Example 7.9 which
uses Proposition 7.10. This means that the requirement in Theorem 7.8 that
every cofibration which has the left lifting property with respect to all fibrations
should be a weak equivalence is not particularly severe.
Example 7.12. As an instance where the cofibrant generators are not -small,
we point out that in Section IX.3 we will take the category of simplicial sets, with
its usual simplicial structure and impose a new closed model category structure.
Let E be any homology theory and we demand that a morphism f : X Y in
S be a
1) E equivalence if E f is an isomorphism
2) E cofibration if f is a cofibration as simplicial sets
3) E fibration if f has the right lifting property with respect to all E trivial
cofibrations.
The E fibrant objects are the Bousfield local spaces. In this case the E -trivial
cofibrations are the saturation of a set of E trivial cofibrations f : A B where
B is -small with some infinite cardinal greater than the cardinality of E (pt).

122

II. Model Categories

One has functorial factorizations, so Example 7.11 can be repeated to show that
SI has a simplicial model category structure with f : X Y a weak equivalence
(or fibration) if and only if X(i) Y (i) is an E equivalence (or E fibration)
for all i.
8. Quillens total derived functor theorem.
Given two closed model categories C and D and adjoint functors between
them, we wish to know when these induce adjoint functors on the homotopy
categories. This is Quillens Total Derived Functor Theorem. We also give
criteria under which the induced adjoint functors give an equivalence of the
homotopy categories.
The ideas of this section are a generalization to non-abelian settings of an old
idea of Grothendieck which can be explained by the following example. If R is
a commutative ring and M, N are two R-modules, one might want to compute
TorR
p (M, N ), p 0. However, there is a finer invariant, namely, the chain
homotopy type of M R P where P is a projective resolution of N . One calls
the chain homotopy equivalence class of M R P by the name TorR (M, N ).
This is the total derived functor. The individual Tor terms can be recovered by
taking homology groups.
For simplicity we assume we are working with simplicial model categories,
although many of the results are true without this assumption.
Definition 8.1. Let C be a simplicial model category and A any category.
Suppose F : C A is a functor that sends weak equivalences between cofibrant
objects to isomorphisms. Define the total left derived functor
LF : Ho(C) A.
by LF (X) = F (Y ) where Y X is a trivial fibration with Y cofibrant.
It is not immediately clear that LF is defined on morphisms or a functor.
If f : X X 0 is a morphism in C and Y X and Y 0 X 0 are trivial
cofibrations with Y and Y 0 cofibrant, then there is a morphism g making the
following diagram commute:
Y
(8.2)

and we set LF (f ) = LF (g).

u
X

w Y0
u
w X0

8. Quillens total derived functor theorem

123

Lemma 8.3. The objects LF (X) and morphisms LF (f ) are independent of the
choices and LF : Ho(C) A is a functor.
Proof: First note that LF (f ) is independent of the choice of g in diagram (8.2).
This is because any two lifts g and g 0 are homotopic and one has
w F (Y 1 )

u
F (1)tF (1)
F (Y )

F (Y ) t F (Y )

FH

w F (Y 0 )

where H is the homotopy. Next, if we let f be the identity in (8.2), the same
argument implies LF (X) is independent of the choice of Y . Finally, letting
f = f1 f2 be a composite in diagram (8.2) the same argument shows LF (f1
f2 ) = LF (f1 ) LF (f2 ).

Remark 8.4. For those readers attuned to category theory we note that LF
is in fact a Kan extension in the following sense. Let : C Ho(C) be the
localization functor and
C

w Ho(C)

A
the diagram of categories. There may or may not be a functor Ho(C) A
completing the diagram; however, one can consider functors T : Ho(C) A
equipped with a natural transformation
T : T F.
The Kan extension is the final such functor T , if it exists. If R denotes this Kan
extension, final means that given any such T , there is a natural transformation
: T R so that T = R . The Kan extension is unique if it exists. To
see that it exists one applies Theorem 1, p. 233 of Mac Lanes book [66]. This
result, in this context reads as follows: one forms a category X consisting of
pairs (Z, f ) where Z C and f : X Z is a morphism in Ho(C). Then if
R(X) = lim F (Z)

exists for all X, then R exists. However, the argument of Lemma 6.3 says that
the diagram F : (X ) A has a terminal object. In fact, X has a
terminal object, namely X Y where Y X is a trivial fibration (which has
an inverse in Ho(C)) with Y cofibrant. This shows that R = LF .

124

II. Model Categories

Corollary 8.5. Let X C. If X is cofibrant, then LF (X)

= F (X). If Y X

is any weak equivalence, with Y cofibrant, then LF X = F Y .

Proof: The first statement is obvious and the second follows from

=
FY
= LF (Y ) LF (X)

since Y X is an isomorphism in Ho(C).

Example 8.6. Let C = C R be chain complexes of left modules over a ring R,

and let A = nAb be graded abelian groups. Define
F (C) = H (M R C)
for some right module M . Then
LF (C) = H (M R D)
where D C is a projective resolution of C. There is a spectral sequence
TorR
p (M, Hq C) (LF (C))p+q .
In particular, if H C = N concentrated in degree 0,
LF (C)
= TorR
(M, N ),
bringing us back to what we normally mean by derived functors.
If G : C A sends weak equivalences between fibrant objects to isomorphisms,
one also gets a total right derived functor
RG : Ho(C) A.
It is also a Kan extension, suitably interpreted: it is initial among all functors
S : Ho(C) A equipped with a natural transformation S : F S.
Now suppose we are given two simplicial model categories C and D and a
functor F : C D with a right adjoint G. The following is one version of the
total derived functor theorem:
Theorem 8.7. Suppose F preserves weak equivalences between cofibrant objects and G preserves weak equivalences between fibrant objects. Then LF :
Ho(C) Ho(D) and RG : Ho(D) Ho(C) exists and RG is right adjoint to
LF .

8. Quillens total derived functor theorem

125

Proof: That LF and RG exist is a consequence of Lemma 8.3 and its analog
for total right derived functors. We need only prove adjointness.
If X C, choose Y X a trivial fibration with Y cofibrant. Hence LF (X)
=
F (Y ). Now choose F (Y ) Z, a trivial cofibration with Z fibrant. Then
RG LF (X)
= G(Z) and one gets a unit
: X RG LF (X)
by X Y GF (Y ) G(Z).
Similarly, let A D. Choose A B a trivial cofibration with B fibrant. Then
RG(A)
= G(B). Next choose C G(B) a trivial fibration with C cofibrant.
Then LF RG(A)
= F (C) and one gets a counit : LF RG(A) A by
F (C) F GB B A.
We now wish to show
LF

RG

LF (X) LF RG LF (X) LF (X)

is an isomorphism. In evaluating RG we set A = LF (X) = F (Y ), so that
j

B = Z. Factor Y GF (Y ) G(Z) and Y C G(Z) where j is a cofibration

(so C is cofibrant) and q is a trivial fibration. Then RG is given by
Fq

F (C) F GZ Z F Y.
Furthermore there is a commutative square
Y
j

u
C

w GF (Y )
u
w G(Z)

and F j
= P LF X . Expanding the diagram gives:
F (Y )
Fj

u
F (C)

Fq

w F GF Y

w FY

u
w F GZ

u
wZ

=.

126

II. Model Categories

The line across the top is the identity and the indicated map is an isomorphism
in the homotopy category. Hence we have proved the assertion.
The other assertionthat

RG

RG(A) RG LF RG(A) RG(A)

is an isomorphismis proved similarly. The result now holds by standard arguments; e.g., [66] Thm. 2v), p. 81.

An immediate corollary used several times in the sequel is:

Corollary 8.8. Under the hypotheses of Theorem 8.7 assume further that for
X C cofibrant and A D cofibrant
X GA
is a weak equivalence if and only if its adjoint F X A is a weak equivalence.
Then LF and RG induce an adjoint equivalence of categories:
Ho(C)
= Ho(D).
Proof: We need to check that : X RG LF (X) is an isomorphism and
: LF RG(A) A is an isomorphism. Using the notation established in the
previous argument, we have a sequence of arrows that define :
X Y GZ.
Now Y is cofibrant and Z is fibrant, and F Y Z is a weak equivalence; so
Y GZ is a weak equivalence and this shows is an isomorphism. The other
argument is identical.

In practice one may not know a priori that F and G satisfy the hypotheses
of Theorem 8.7. The following result is often useful. We shall assume that the
model categories at hand are, in fact, simplicial model categories; however, it is
possible to prove the result more generally.
Lemma 8.9. Let F : C D be a functor between simplicial model categories,
and suppose F has a right adjoint G. If G preserves fibrations and trivial fibrations, then F preserves cofibrations, trivial cofibrations and weak equivalences
between cofibrant objects.

8. Quillens total derived functor theorem

127

Proof: It follows from an adjointness argument that F preserves trivial cofibrations and cofibrations; for example, suppose j : X Y is a cofibration in C.
To show F j is a cofibration, one need only solve the lifting problem
FX

wA
\
]
Fj
\ q
u \
u
FY
wB
for every trivial fibration q in D. This problem is adjoint to
X

w GA
\
]
Gq
\
u \
u
Y
w GB

which has a solution by hypothesis.

Now suppose f : X Y is a weak equivalence between cofibrant objects.
j
q
Factor f as X Z Y where j is a trivial cofibration and q a trivial fibration.
We have just shown F j is a weak equivalence. Also q is actually a homotopy
equivalence: there is a map s : Y Z so that qs = 1Y and sq ' 1Z . Here X
and Y are cofibrant. We claim that F q is a homotopy equivalence, so that it is
a weak equivalence by Lemma 4.1.
To see that F q is a homotopy equivalence, note that F (q)F (s) = 1F Y . Next
note that since Z is cofibrant, Z 1 is a cylinder object for Z and, since F
preserves trivial cofibrations F (Z 1 ) is a cylinder object for F (Z). Hence
F (s)F (q) ' 1F Z .

Remark 8.10. As usual, the previous result has an analog that reverses the
roles of F and G; namely, if F preserves cofibrations and trivial cofibrations,
then G preserves fibrations, trivial fibrations, and weak equivalences between
fibrant objects. The proof is the same, mutatis mutandis.
Example 8.11. Let I be a small category and S I the category of I diagrams.
Then S I becomes a simplicial model category, where a morphism of diagrams
X Y is a weak equivalence or fibration if and only if X(i) Y (i) is a weak
equivalence of I-diagrams. The constant functor S S I preserves fibrations
and weak equivalences, so (by Lemma 8.9), the left adjoint
F = lim : S I S

128

II. Model Categories

preserves weak equivalences among cofibrant diagrams. Hence the total left
derived functor
L lim : Ho(S I ) Ho(S)

exists. This functor is the homotopy colimit and we write L lim = holim. In
I

a certain sense, made precise by the notion of Kan extensions in Remark 8.4, is
the closest approximation to colimit that passes to the homotopy category. In
any application it is useful to have an explicit formula for holim X in terms of

the original diagram X; this is given by the coend formula

Z
holim X =

B(i I)op X(i).

These are studied in detail elsewhere in these notes.

This example, can be vastly generalized. If C is any cofibrantly generated
simplicial model category, C I becomes a simplicial model category and one gets
holim : Ho(C I ) Ho(C)

in an analogous manner.

9. Homotopy cartesian diagrams.

We return, in this last section, to concepts which are particular to the category of simplicial sets and its close relatives. The theory of homotopy cartesian
diagrams of simplicial sets is, at the same time, quite deep and essentially axiomatic. The axiomatic part of the theory is valid in arbitrary categories of
fibrant objects such as the category of Kan complexes, while the depth is implicit in the passage from the statements about Kan complexes to the category
of simplicial sets as a whole. This passage is non-trivial, even though it is completely standard, because it involves (interchangeably) either Quillens theorem
that the realization of a Kan fibration is a Serre fibration (Theorem I.10.10) or
Kans Ex construction (see III.4).
A proper closed model category C is a closed model category such that
P1 the class of weak equivalences is closed under base change by fibrations,
and
P2 the class of weak equivalences is closed under cobase change by cofibrations.

9. Homotopy cartesian diagrams

129

In plain English, axiom P1 says that, given a pullback diagram

X
u
Z

wY
p
u
wW

of C with p a fibration, if g is a weak equivalence then so is g . Dually, axiom

P2 says that, given a pushout diagram

u
C

wB
u
wD

with i a cofibration, if f is a weak equivalence then so is f .

The category of simplicial sets is a canonical example of a proper closed model
category (in fact, a proper simplicial model category). Furthermore, this is the
generic example: most useful examples of proper closed model categories inherit
their structure from simplicial sets. The assertion that the category of simplicial
sets satisfies the two axioms above requires proof, but this proof is in part a
formal consequence of the fact that every simplicial set is cofibrant and every
topological space is fibrant. The formalism itself enjoys wide applicability, and
will be summarized here, now.
A category of cofibrant objects is a category D with all finite coproducts (including an initial object ), with two classes of maps, called weak equivalences
and cofibrations, such that the following axioms are satisfied:
(A) Suppose given a commutative diagram

XN

wY
N

N
P
h N
f

130

II. Model Categories

in D. If any two of f , g and h are weak equivalences, then so is the third.

(B) The composite of two cofibrations is a cofibration. Any isomorphism is a

cofibration.
(C) Pushout diagrams of the form

wB

i
u
wD

u
C

exist in the case where i is a cofibration. Furthermore, i is a cofibration

which is trivial if i is trivial.
(D) For any object X there is at least one cylinder space X I.
(E) For any object X, the unique map X is a cofibration.
To explain, a trivial cofibration is a morphism of D which is both a cofibration
and a weak equivalence. A cylinder object X I for X is a commutative diagram

X tX
i = (i0 , i1 )

wX
[
]
[
[

u [
X I

in which i is a cofibration and is a weak equivalence, just like in the context

of a closed model category (see Section 1 above). Each of the components i of
i must therefore be a trivial cofibration.
The definition of category of cofibrant objects is dual to the definition of
category of fibrant objects given in Section I.9. All results about categories of
fibrant objects therefore imply dual results for categories of cofibrant objects, and
conversely. In particular, we immediately have the dual of one of the assertions
of Proposition I.9.5:
Proposition 9.1. The full subcategory of cofibrant objects Cc in a closed model
category C, together with the weak equivalences and cofibrations between them,
satisfies the axioms (A) - (E) for a category of cofibrant objects.

9. Homotopy cartesian diagrams

131

Remark 9.2. One likes to think that a category of cofibrant objects structure
(respectively a category of fibrant objects structure) is half of a closed model
structure. This intuition fails, however, because it neglects the power of the
axiom CM4.
Corollary 9.3.
(1) The category of simplicial sets is a category of cofibrant objects.
(2) The category of compactly generated Hausdorff spaces is a category of
fibrant objects.
Lemma 9.4. Suppose that f : A B is an arbitrary map in a category of
cofibrant objects D. Then f has a factorization f = q j, where j is a cofibration
and q is left inverse to a trivial cofibration. In particular, q is a weak equivalence.
Proof: The proof of this result is the mapping cylinder construction. Its also
dual of the classical procedure for replacing a map by a fibration.
Choose a cylinder object

AtA
(i0 , i1 )

u h
AI

wA

h
hj

for A, and form the pushout diagram

A

i0

u
AI

wB
i0
u
w B .

Then (f )i0 = f , and so there is a unique map q : B B such that q f = f

and q i0 = 1B . Then f = q (f i1 ).
The composite map f i1 is a cofibration, since the diagram
AtA

f t 1A

(i0 , i1 )

u
AI

w BtA
(i0 , f i1 )
u
w B

132

II. Model Categories

is a pushout.

Lemma 9.5. Suppose that

A

wB

u
wD

u
C

is a pushout in a category of cofibrant objects D, such that i is a cofibration and

u is a weak equivalence. Then the map u is a weak equivalence.
Proof: Trivial cofibrations are stable under pushout, so Lemma 9.4 implies
that it suffices to assume that there is a trivial cofibration j : B A such that
u j = 1B .
Form the pushout diagram

B
ju
A
iu
C

wA
j

wD

Then j is a trivial cofibration.

C be the unique map which is determined by the commutative
Let f : D
diagram
B
ju
A
iu
C

wA
i

1C

u
w C.

9. Homotopy cartesian diagrams

133

Form the prism

u

A'

w B'

'
'i
u
'
u

D
N '
N '
)
N
P'
f N

'
'

u
'
w B N '
N '
)
N
P'
f N

wD

such that the front and back faces are pushouts (ie. push out the triangle on the
left along u). Then u
is a weak equivalence, since j is a weak equivalence and
u j = 1B . It therefore suffices to show that the map f is a weak equivalence.
The bottom face
u

D
f

w B

u
C

f
u
wD

is a pushout, and the map f is a weak equivalence. The morphism

A4

j h
h
h
k

i
4
4
6

wC

is therefore a weak equivalence in the category A D, and the argument of

Lemma 9.4 says that this map has a factorization in A D of the form f = q j,
where j is a cofibration and q is left inverse to a cofibration. It follows that
pushing out along u preserves weak equivalences of A D, so that f is a weak
equivalence of D.

Corollary 9.6. The category S of simplicial sets is a proper simplicial model

category.

134

II. Model Categories

Proof: Axiom P2 is a consequence of Lemma 9.5 and Corollary 9.3. The category CGHaus of compactly generated Hausdorff spaces is a category of fibrant
objects, so the dual of Lemma 9.5 implies Axiom P1 for that category. One
infers P1 for the simplicial set category from the exactness of the realization
functor (Proposition I.2.4), the fact that the realization functor preserves fibrations (Theorem I.10.9), and the fact that the canonical map : X S|X| is a
weak equivalence for all simplicial sets X (see the proof of Theorem I.11.4).
Remark 9.7. Axiom P1 for the category of simplicial sets can alternatively be
seen by observing that Kans Ex preserves fibrations and pullbacks (Lemma
III.4.5), and preserves weak equivalences as well (Theorem III.4.6). Thus, given
a pullback diagram

X
u
Z

wY
p
u
wW

with p a fibration and g a weak equivalence, if we want to show that g is a weak

equivalence, it suffices to show that the induced map Ex g in the pullback
diagram

Ex X

Ex

Ex g

w Ex Y
Ex p

inf ty

Ex g

w Ex

is a weak equivalence. But all of the objects in this last diagram are fibrant and
the map Ex g is a weak equivalence, so the desired result follows from the dual
of Lemma 9.5.
The following result is commonly called the glueing lemma. The axiomatic
argument for it that is given here is due to Thomas Gunnarsson [40].

9. Homotopy cartesian diagrams

135

Lemma 9.8. Suppose given a commutative cube

j1

A1 4

w B1 4

4i
1
4
4
6

fA
(9.9)

4
6
4

C1

u
A2 4

fC

4
4
6
i2 4

w D1

fB

j2

u
w B2 4

fD
4

4
6
4

u
C2

u
w D2

in a category of cofibrant objects D. Suppose further that the top and bottom
faces are pushouts, that i1 and i2 are cofibrations, and that the maps fA , fB
and fC are weak equivalences. Then fD is a weak equivalence.
Proof: It suffices to assume that the maps j1 and j2 are cofibrations. To see
this, use Lemma 9.4 to factorize j1 and j2 as cofibrations followed by weak
equivalences, and then use Lemma 9.5 to analyze the resulting map of cubes.
Form the diagram
j1

A1

C1
fA

w B1

i
1

j1
fA
u
B0

C
AA
fC
A
nB
AA
A
A

u AA
A2

i2

w D1
fC
u
D0

A
C
u AA
j2
nD
A2A
wB

A
A

u
u AA
i2
C2
w D2 .

136

II. Model Categories

by pushing out the top face along the left face of the cube (9.9). The square
w B0

A2
i2

u
w D0

u
C2

is a pushout, so the square

B0

nB

u
D0

nD

w B2
i2
u
w D2

is a pushout, and is a cofibration. The map fA is a weak equivalence, since j1 is

a cofibration and fA is a weak equivalence. Similarly, fC is a weak equivalence,
since j1 is a cofibration and fC is a weak equivalence. The map fB = nB fA
is assumed to be a weak equivalence, so it follows that nB is a weak equivalence.
Then nD is a weak equivalence, so fD = nD fC is a weak equivalence.

The dual of Lemma 9.8 is the co-glueing lemma for categories of fibrant objects:
Lemma 9.10. Suppose given a commutative cube
A1 4

w B1 4

4
4
4
6

fA
(9.11)

4p
41
6
4

u
A2 4

C1
fC
4
4
4
6

u
C2

w D1

fB
u
w B2 4

fD
4p
42
6
4

u
w D2

in a category of fibrant objects E. Suppose further that the top and bottom
squares are pullbacks, that the maps p1 and p2 are fibrations, and that the maps
fB , fC and fD are weak equivalences. Then the map fA is a weak equivalence.

9. Homotopy cartesian diagrams

137

Either of the two methods of proof of Corollary 9.6 given above (see also
Remark 9.7), together with Lemma 9.10, now yield the following expanded consequence for the full category of simplicial sets:
Corollary 9.12. Suppose given a commutative cube
X1 4
4

w Y1 4
4p
41
4
6

4
6
4

fX
u
X2 4
4

W1

fY

u
w Y2 4
4p
42
4
6

fW
4
6
4

w Z1

u
W2

fZ

u
w Z2

of morphisms in the category S of simplicial sets. Suppose further that the top
and bottom squares are pullbacks, that the maps p1 and p2 are fibrations, and
that the maps fY , fW and fZ are weak equivalences. Then the map fX is a
weak equivalence.
Corollary 9.12 is the basis for the theory of homotopy cartesian diagrams in
the category of simplicial sets. We say that a commutative square of simplicial
set maps
X

wY

u
W

f
u
wZ

(9.13)

is homotopy cartesian if for any factorization

Yh
(9.14)

h
j
i h
Y

wZ
'
)
'p
'

138

II. Model Categories

of f into a trivial cofibration i followed by a fibration p the induced map

i

X W Z Y

is a weak equivalence.
In fact (and this is the central point), for the diagram (9.13) to be homotopy
cartesian, it suffices to find only one such factorization f = p i such that the
map i is a weak equivalence. This is a consequence of the following:
Lemma 9.15. Suppose given a commutative diagram

wY

u
W

f
u
wZ

of morphisms in the category of simplicial sets, and factorizations

Yh

h
j
ij h

'

wZ
'
)
'pj

Yj

of f as trivial cofibration ij followed by a fibration pj for j = 1, 2. Then the

induced map i1 : X W Z Y1 is a weak equivalence if and only if the map
i2 : X W Z Y2 is a weak equivalence.
Proof: There is a lifting in the diagram

Y
i1

i2
h
j
h
h

u h
Y1
p1

w Y2
p2
u
w Z,

9. Homotopy cartesian diagrams

139

by the closed model axioms. Form the commutative cube

W Z Y1

w Y1 4

4p
41
4
6

u
W Z Y2

u
w Y2 4

1W

1Z
4p
42
4
6

wZ

u
W

u
w Z.

Then the map is a weak equivalence by Corollary 9.12. There is a commutative

diagram
i1 44
6
44
X

i2

W Z Y1

u
W Z Y2 ,

and the desired result follows.

Remark 9.16. The argument of Lemma 9.15 implies that the definition of homotopy cartesian diagrams can be relaxed further: the diagram (9.13) is homotopy cartesian if and only if there is a factorization (9.14) such that p is a fibration
and i is a weak equivalence, and such that the induced map i : X W Z Y
is a weak equivalence.
The way that the definition of homotopy cartesian diagrams has been phrased
so far says that the diagram
X

wY

u
W

f
u
wZ

140

II. Model Categories

is homotopy cartesian if a map induced by a factorization of the map f into a

fibration following a trivial cofibration is a weak equivalence. In fact, it doesnt
matter if we factor f or g:
Lemma 9.17. Suppose given a commutative diagram

wY
f

u
W

u
wZ

in the category of simplicial sets. Suppose also that we are given factorizations
f

Y4

4
6
i 44

wZ

W4

h
hj
p
h

g
4
6
j 44

wZ
h
hj
q
h

of f and g respectively such that i and j are trivial cofibrations and p and q are
fibrations. Then the induced map i : X W Z Y is a weak equivalence if
Z Y is a weak equivalence.
and only if the map j : X W
Proof: There is a commutative diagram

j
u

W Z Y

1i

w W Z Y
j1
u
Z Y .
wW

The map p is a fibration, so the map j 1 is a weak equivalence, and q is a

fibration, so 1 i is a weak equivalence, all by Corollary 9.12.

The coglueing lemma also has the following general consequence for homotopy
cartesian diagrams:

9. Homotopy cartesian diagrams

141

Corollary 9.18. Suppose given a commutative cube

w Y1 4
4

X1 4
4

4
4
6

4
6
4

fX
u
X2 4
4

W1

u
w Y2 4
4

fW
4
6
4

w Z1

fY

fZ
4
4
6

u
W2

u
w Z2

of morphisms in the category S of simplicial sets. Suppose further that the top
and bottom squares are homotopy cartesian diagrams, and that the maps fY ,
fW and fZ are weak equivalences. Then the map fX is a weak equivalence.
A homotopy fibre sequence of simplicial sets is a homotopy cartesian diagram
in S
j
X
wY
u

f
u
w Z.

In effect, one requires that the composite f j factor through the base point x of
Z, and that if f = p i is a factorization of f into a trivial cofibration followed
by a fibration, then the canonical map X F is a weak equivalence, where F is
the fibre of p over x. More colloquially (see also Remark 9.16), this means that
X has the homotopy type of the fibre F of any replacement of the map f by a
fibration up to weak equivalence. It is common practice to abuse notation and
say that
j

X
Y
Z
is a homotopy fibre sequence, and mean that these maps are a piece of a homotopy cartesian diagram as above. Every fibration sequence
F EB
is plainly a homotopy fibre sequence.

142

II. Model Categories

Example 9.19. Suppose that

j

X
Y
Z
is a homotopy fibre sequence, relative to a base point x of Z, and that there is
a vertex y Y such that f (y) = x. Suppose that the canonical map Y is a
weak equivalence. Then X is weakly equivalent to the loop space Z for some
(and hence any) fibrant model Z for Z. To see this, choose a trivial cofibration
where Z is a Kan complex, and use the factorization axioms to form
j : Z Z,
the commutative square
Y

w Y
p
u

w Z,

f
u
Z

where both maps labelled j are trivial cofibrations and p is a fibration. Let F
Then
denote the fibre of the fibration p over the image of the base point x in Z.
Corollary 9.18 implies that the induced map X F is a weak equivalence. Now
consider the diagram
Y Z P Z

prR

w P Z

prL
u

w Z,

where P Z is the standard path space for the Kan complex Z and the base
point x, and is the canonical fibration. Then the map y : Y is a weak
equivalence, so that the inclusion Z Y Z P Z of the fibre of the fibration
prL is a weak equivalence, by properness, as is the inclusion F Y Z P Z of
the fibre of prR . In summary, we have constructed weak equivalences
'

'

'

X F Y Z P Z Z.
This collection of ideas indicates that it makes sense to define the loop space
of a fibrant model X
for X
of a connected simplicial set X to be the loops X
the loop space of X is therefore an example of a total right derived functor,
in the sense of Section II.7.

9. Homotopy cartesian diagrams

143

Here is a clutch of results that illustrates the formal similarities between homotopy cartesian diagrams and pullbacks:
Lemma 9.20.
(1) Suppose that
X
u
Z

wY
u
wW

is a commutative diagram in the simplicial set category such that the

maps and are weak equivalences. Then the diagram is homotopy
cartesian.
(2) Suppose given a commutative diagram

X1
u
Y1

w X2
I

u
w Y2

w X3
II

u
w Y3

in S. Then
(a) if the diagrams I and II are homotopy cartesian then so is the
composite diagram I + II

X1

w X3

u
Y1

u
w Y3 ,

(b) if the diagrams I + II and II are homotopy cartesian, then I is

homotopy cartesian.

144

II. Model Categories

The proof of this lemma is left to the reader as an exercise.

The stream of results and definitions beginning with Corollary 9.12 for simplicial sets holds much more generally for arbitrary proper simplicial model categories, in view of the following result (and its dual):
Lemma 9.21. Let C be a proper simplicial model category. Consider a diagram
D1 u
u
D2 u

j1

j2

C1

w X1

u
C2

u
w X2

where j1 and j2 are cofibrations and the three vertical maps are weak equivalences. Then the map D1 C1 X1 D2 C2 X2 is a weak equivalence.
Proof: We use the proper hypothesis to reduce to the case C1 = C2 and X1 =
X2 and then argue appropriately.
First notice that there is a pushout diagram
X1

w X2

D1 C1 X1

w D1 C1 X2 .

Since X1 D1 C1 X1 is a cofibration (because j1 is) and X1 X2 is a weak

equivalence, properness implies D1 C1 X1 D1 C1 X2 is a weak equivalence.
Thus we may assume X1 = X2 = X.
Next notice that there is a pushout diagram
C1
j1

u
D1

w C2
u
w D1 C1 C2 .

Since j1 is a cofibration and C1 C2 is a weak equivalence, properness implies

D1 D1 C1 C2 is a weak equivalence. Now
D1 C1 X
= (D1 C1 C2 ) C2 X

9. Homotopy cartesian diagrams

145

and we are reduced to considering the diagram

D1 C1 C2 u

C2

u
D1 u

u
C2

wX
=

u
w X.

Note that C2 D1 C1 C2 is a cofibration and D1 C1 C2 D2 is a weak

equivalence. Thus we may assume C1 = C2 = C.
Now factor D1 D2 as
j

D1
Z
D2
where j is a trivial cofibration and q is a fibration. Then q is a trivial fibration,
and the composite C Z is a cofibration. There is a pushout diagram
D1

w D1 C X

u
Z

u
w Z C X.

Since j is a trivial cofibration, so is D1 C X Z C X. We will argue that

Z C X D2 C X is a homotopy equivalence and well be done. To see this
last claim, note that there is a left s : D2 Z solving the lifting problem
C

wZ
O q
P
O
u O
u
D2 = w D2 .
Then qs = 1D2 and sq and 1Z are both solutions to the lifting problem
C

wZ
O q
P
O
u O
u
Z q w D2

146

II. Model Categories

so sq and 1Z are homotopic under C. Thus Z C X D2 C X has a section

induced from s and is a homotopy equivalence. Now invoke Lemma 4.1.

The dual of Lemma 9.21 is an obvious generalization of the statement of

Corollary 9.12. The proofs of the statements in 9.15 9.18 and Lemma 9.20
are formal consequences of properness for the simplicial set category and Corollary 9.12, and can therefore be promoted to arbitrary proper simplicial model
categories C. Thus, we are entitled to say that a diagram
X

wY

u
W

f
u
wZ

in C is homotopy cartesian if for any factorization

f

Yh

h
j
i h

wZ
'
)
'p
'

Y
of f into a weak equivalence i followed by a fibration p the induced map
i

X W Z Y

is a weak equivalence. Then the standard calculus of homotopy cartesian diagrams in simplicial sets obtains in this extended context.
We close with a further application of categories of cofibrant objects structures.
Let C be a fixed choice of simplicial model category having an adequate supply
of colimits. Suppose that is a limit ordinal, and say that a cofibrant -sequence
in C is a functor X : Seq() C, such that all objects Xi are cofibrant, each
map Xi Xi+1 is a cofibration, and Xt = lim Xi for all limit ordinals t < .
i<t
The cofibrant -sequences, with ordinary natural transformations between them,
form a category which will be denoted by C . Say that a map f : X Y in C
is a weak equivalence if all its components f : Xi Yi are weak equivalences of
C, and say that g : A B is a cofibration of C if the maps g : Ai Bi are
cofibrations of C, as are all induced maps Bi Ai Ai+1 Bi+1 .

9. Homotopy cartesian diagrams

147

Lemma 9.22. Let C be a simplicial model category having all filtered colimits.
With these definitions, the category C of cofibrant -sequences in C satisfies the
axioms for a category of cofibrant objects.
Proof: Suppose that A B C are cofibrations of C . To show that the
composite A C is a cofibration, observe that the canonical map Ci Ai Ai+1
Ci+1 has a factorization
Ci Ai Ai+1 A

w Ci+1

j
hh

AA
C

h
Ci Bi Bi+1

and there is a pushout diagram

Bi Ai Ai+1

w Bi+1

Ci Ai Ai+1

w Ci Bi Bi+1 .

Suppose that
A
i

u
C

wB
i

u
wD

is a pushout diagram of Seq()-diagrams in C, where A, B and C are cofibrant

-sequences and the map i is a cofibration of same. We show that D is a cofibrant
-sequence and that i is a cofibration by observing that there are pushouts
Ci Ai Ai+1

w Di Bi Bi+1

Ci+1

w Di+1 ,

148

II. Model Categories

and that the maps Di Di Bi Bi+1 are cofibrations since B is a cofibrant

-sequence.
Suppose that A is a cofibrant -sequence, and let K be a simplicial set. Then
the functor A K : Seq() C is defined by (A K)i = Ai K. The functor
X 7 X K preserves cofibrations and filtered colimits of C, so that A K is
a cofibrant -sequence. Furthermore, if K L is a cofibration of S then the
induced map A K A L is a cofibration of C : the proof is an instance of
SM7. It follows that the obvious diagram

A 1
[
u
[
A 1

w A 0
[
]
[

is a candidate for the cylinder object required by the category of cofibrant objects
structure for the category C .

Lemma 9.23. Suppose that C is a simplicial model category having all filtered
colimits. Suppose that f : A B is a cofibration and a weak equivalence of C .
Then the induced map
f : lim Ai lim Bi

i<

i<

is a trivial cofibration of C.
Proof: Suppose given a diagram

lim Ai

wX

i<

u
lim Bi

i<

p
u
wY

9. Homotopy cartesian diagrams

149

where p is a trivial fibration of C. We construct a compatible family of lifts

Ai

(9.24)

wX
h
j
i h
p
fi
h
u h
u
Bi
wY

as follows:
1) Let s be the map induced by all i for i < s at limit ordinals s < .
2) Given a lifting i as in diagram (9.24), form the induced diagram

Bi Ai Ai+1
f
u

O
P
O

i+1O
O
O

wX
p
u
w Y.

Bi+1

The map f is a trivial cofibration of C, since f is a cofibration and a weak

equivalence of C , so the indicated lift i+1 exists.

Corollary 9.25. Suppose that C is a simplicial model category having all

filtered colimits, and that f : X Y is a weak equivalence of cofibrant sequences in C. Then the induced map
f : lim Xi lim Yi

i<

i<

is a weak equivalence of C.
Proof: We have it from Lemma 9.22 that C is a category of cofibrant objects,
and Lemma 9.4 says that f : X Y has a factorization f = q j, where j
is a cofibration and q is left inverse to a trivial cofibration. Then j is a trivial
cofibration since f is a weak equivalence, and so Lemma 9.23 implies that both
j and p induce weak equivalences after taking filtered colimits.

The dual assertion for Corollary 9.25 is entertaining. Suppose, again, that is
a limit ordinal and that C is a simplicial model category having enough filtered
inverse limits. Define a fibrant -tower in C to be (contravariant) functor X :

150

II. Model Categories

Seq()op C such that each Xi is a fibrant object of C, each map Xi+1 Xi

is a fibration of C, and Xt = lim Xi for all limit ordinals t < . Then the
i<t
dual of Lemma 9.22 asserts that, for pointwise weak equivalences and a suitable
definition of fibration, the category of fibrant -towers in C has a category of
fibrant objects structure. The dual of Lemma 9.23 asserts that the inverse limit
functor takes trivial fibrations of fibrant -towers to trivial fibrations of C, and
then we have
Lemma 9.26. Suppose that C is a simplicial model category having all filtered
inverse limits, and that f : X Y is a weak equivalence of fibrant -towers in
C. Then the induced map
f : lim Xi lim Yi

i<

i<

is a weak equivalence of C.
For fibrant -towers X : Seq()op S taking values in simplicial sets, one
can take a different point of view, in a different language. In that case, fibrant towers are globally fibrant Seq()op -diagrams, and inverse limits and homotopy
inverse limits coincide up to weak equivalence for globally fibrant diagrams, for
all . Homotopy inverse limits preserve weak equivalences, so inverse limits
preserve weak equivalences of fibrant -towers. Homotopy inverse limits and
homotopy theories for categories of diagrams will be discussed in Chapters 6
and 7.

Chapter III Classical results and constructions

This chapter is a rather disparate collection of stories, most of them old and
well known, but told from a modern point of view.
The first section contains several equivalent descriptions of the fundamental
groupoid, one of which (ie. left adjoint to a classifying space functor) is powerful
enough to show that the fundamental groupoid of a classifying space of a small
category is the free groupoid on that category (Corollary 1.2), as well as prove
the Van Kampen theorem (Theorem 1.4).
The second section, on simplicial abelian groups, contains a complete development of the Dold-Kan correspondence. This correspondence is an equivalence
of categories between chain complexes and simplicial abelian groups; the result
appears as Corollary 2.3. We also give an elementary description of the proper
simplicial model structure for the category of simplicial abelian groups (Theorem 2.6, Proposition 2.11, and Remark 2.12), and then use this structure to
derive the standard isomorphism
H n (X, A)
= [X, K(A, n)]
relating cohomology to homotopy classes of maps which take values in an Eilenberg-Mac Lane space (Corollary 2.17). We close Section 2 by showing that every
simplicial abelian group is non-canonically a product of Eilenberg-Mac Lane
spaces up to homotopy equivalence (Proposition 2.18).
We have included Section 3, on the Hurewicz theorem, as further evidence for
the assertion that many results in the Algebraic Topology canon have very clean
simplicial homotopy proofs. We use Postnikov towers and the Serre spectral
sequence, both of which appear here for the first time in the book and are
described more fully in later chapters. The Hurewicz homomorphism itself has
a very satisfying functorial description in this context: it is the adjunction map
X ZX from a simplicial set X to the corresponding free abelian simplicial
group ZX.
Section 4 contains a modernized treatment of Kans Ex functor. This functor
gives a combinatorial (even intuitionistic), natural way of mapping a simplicial
set into a Kan complex, via a weak equivalence. The construction is therefore preserved by left exact functors which have right adjoints these appear
throughout topos theory [38]. We have seen similar applications already in Chapter II in connection with detecting simplicial model structures on categories of
simplicial objects. The main theorem here is Theorem 4.7. The proof is an
updated version of the original: the fundamental groupoid trick in Lemma 4.2
may be new, but the heart of the matter is Lemma 4.6.
151

152

III. Classical results and constructions

There are two different suspension functors for a pointed simplicial set X,
namely the smash product S 1 X (where S 1 = 1 /1 is the simplicial circle),
and the Kan suspension X. These are homotopy equivalent but not isomorphic
constructions which have natually homeomorphic realizations. They further represent two of the standard subdivisions for the suspension of a pointed simplicial
complex. Both have their uses; in particular, the Kan suspension is more easily
related to the classifying space of a simplicial group which appears in Chapter
V, and hence to Eilenberg-Mac Lane spectra [52]. We give a full treatment of
the Kan suspension in Section 5, essentially to have it in the bank for later.
Along the way, we say formally what it means for a simplicial set to have an
extra degeneracy. This last idea has been in the folklore for a long time it
means, most succinctly, that the identity map on a simplicial set factors through
a cone.
1. The fundamental groupoid, revisited.
Recall from Section I.8 that the classical fundamental groupoid |X| of the
realization of a simplicial set X coincides with the groupoid f S|X| associated
to the singular complex S|X|. In that section, there is a remark to the effect that
this groupoid is equivalent to the free groupoids GP X and G( X) which
are associated, respectively, to the path category P X and the simplex category
X for X. This claim has the following precise form:
Theorem 1.1. The groupoids G( X), GP X and |X| are naturally equivalent as categories.
Proof: A functor f : G H between groupoids is an equivalence if and only
if
(1) the induced function f : homG (a, b) homH (f (a), f (b)) is a bijection for
every pair of objects a, b of G, and
(2) for every object c of H there is a morphism c f (a) in H.
The groupoids GP S|X| and |X| are naturally isomorphic. The 1-simplices of
S|X| are paths of |X|, and [d1 ] = [d0 ][d2 ] for every 2-simplex : |2 | |X|
of S|X|. It follows that there is a functor X : GP S|X| |X| which is
defined by sending a path to its associated homotopy class. The inverse of X
is constructed by observing that homotopic paths in |X| represent the same
element of GP S|X|.
The next step is to show that the functor GP takes weak equivalences of
simplicial sets to equivalences of groupoids. If f is a weak equivalence of S, then
f has a factorization f = q j where q is a trivial fibration and j is a trivial
cofibration. The map q is left inverse to a trivial cofibration, and every trivial

1. The fundamental groupoid, revisited

153

cofibration is a retract of a map which is a filtered colimit of pushouts of maps

of the form nk n . It suffices, therefore to show that GP takes pushouts of
maps nk n to equivalences of categories.
The induced map GP (nk ) GP (n ) is an isomorphism of groupoids if
n 2. The groupoid GP (1i ) is a strong deformation retract of GP (1 ) in the
category of groupoids for i = 0, 1. Isomorphisms and strong deformation retractions of groupoids are closed under pushout, and so GP takes weak equivalences
of simplicial sets to equivalences of groupoids as claimed. It follows in particular
that the groupoids GP X and |X| are naturally equivalent.
We can assign, to each 1-simplex x : d1 x d0 x, the morphism (d0 )1 (d1 ) of
G( X) arising from the diagram
1
0
0 4 d
0
w 1 u d
4
h
44
h
x
6
d1 x
d x
u h
kh 0
X.

This assignment defines a functor X : GP X G( X), since the following

diagram of simplicial maps commutes:
0'
'

d0

d1

w 1 u

d2
[
u
'
[
2
[ d1

d1 '

' d0 N
[
1
P
d
N

'
)
[
N
^
[
1
1

N
P

N0
d0
N d
'

0
[

0
Let vn : 0 n denote the simplicial map which picks out the last vertex n
of the ordinal number n. Then the assignment

vn

X
X 7 0 n
n

154

III. Classical results and constructions

is the object function of a functor X P X. Write X : G( X)

GP X for the induced functor on associated groupoids. Then the composite
functor X X is the identity on GP X, and the composite X X is naturally
isomorphic to the identity on G( X). The natural isomorphism is determined
by the maps
0

vn

wX

h
hj
h

vn
u h
n

of the simplex category X.

From now on, the fundamental groupoid of a simplicial set X, in any of its
forms, will be denoted by X.
A simplicial set map f : X BC associates to each n-simplex x of X a
functor f (x) : n C which is completely determined by the 1-skeleton of x and
the composition relations arising from 2-simplices of the form
x

2 n
X.
It follows that f can be identified with the graph morphism
Xu 1
u u
X0

w Mor(C)
u
u u
w Ob(C)

subject to the relations f (d1 ) = f (d0 ) f (d2 ) arising from all 2-simplices
of X. The path category P X is the category which is freely associated to the
graph
X1 u

w X0
w

and the 2-simplex relations, so that there is an adjunction isomorphism

homcat (P X, C)
= homS (X, BC).
A small category C is completely determined by its set of arrows and composition relations. It follows that the path category P BC of the nerve BC is
isomorphic to C. We therefore immediately obtain the following assertion:

1. The fundamental groupoid, revisited

155

Corollary 1.2. Let C be a small category. Then the fundamental groupoid

BC of the nerve of C is equivalent, as a category, to the free groupoid GC on
the category C.
Corollary 1.2 may be used to give a direct proof of the fact that the fundamental group 1 BQA of the nerve of the Q-construction QA on an exact category
A is isomorphic to the 0th K-group K0 A of A (see [48]).
Remark 1.3. In the proof of Theorem 1.1, observe that the composite
X

S|X| BP S|X| BGP S|X| |X|

sends a path : 1 S|X| of |X| to its homotopy class [] (rel 1 ). It follows
that the composite
S|X| BP S|X| BGP S|X|
induces isomomorphisms 1 (S|X|, x)
= 1 (BGP S|X|, x) for all choices of vertex
x S|X|. We have also seen that the functor GP takes weak equivalences of
simplicial sets to equivalences of groupoids, so that the commutativity of the
diagram
X

u
S|X|

w BP X

w BGP X

u
w BP S|X|

u
w BGP S|X|

implies that the composite

X BP X BGP X
induces isomorphisms
1 (X, x)
= 1 (BGP X, x)
= homGP X (x, x)
for any choice of vertex x X, if X is a Kan complex.
Note, in particular, that if X is connected, then there is a deformation retraction map of groupoids r : GP X homGP X (x, x) for any choice of vertex of
X. This map r determines a composite map
r

X BP X BGP X B(homGP X (x, x)),

156

III. Classical results and constructions

which induces isomorphisms

1 (X, y)
= 1 (B homGP X (x, x))
= homGP X (x, x)
for all vertices y of X. It follows that there is a map
X B(1 (X, x))
which induces isomorphisms on all fundamental groups if X is a connected Kan
complex with base point x.
Again, the fundamental groupoid construction takes weak equivalences of simplicial sets to equivalences of groupoids. It follows, in particular, that there is a
purely categorical definition of the fundamental group 1 (X, x) of an arbitrary
simplicial set X at a vertex x, given by
1 (X, x) = homX (x, x).
This definition can be used, along with the observation that the fundamental
groupoid construction = GP has a right adjoint and therefore preserves colimits, to give a rather short proof of the Van Kampen theorem:
Theorem 1.4 (Van Kampen). Suppose that
A
y
j
u
B

wX
u
wY

is a pushout diagram of simplicial sets, where the map j is a cofibration, and A,

B and X are connected. Then, for any vertex x of A, the induced diagram
1 (A, x)
j

u
1 (B, x)
is a pushout in the category of groups.

w 1 (X, x)
u
w 1 (Y, x)

1. The fundamental groupoid, revisited

157

Proof: The glueing lemma II.9.8 implies that we can presume that the map i
is also a cofibration.
The induced maps i : A X and j : A B of fundamental
groupoids are monomorphisms on objects, and so the strong deformation r :
A 1 (A, x), by suitable choice of paths in X andB, can be extended to a
strong deformation
i

B u

w X

u
1 (B, x) u

u
1 (A, x)

u
w 1 (X, x)

in a suitable diagram category in groupoids, meaning that the homotopy h :

A A1 giving the deformation extends to a homotopy
i

B u
h

u
B 1 u

u
A1

w X
h
u
w X 1

But then, if
1 (A, x)

u
1 (B, x)

w 1 (X, x)
u
wG

is a pushout in the category of groups, it is also a pushout in the category of

groupoids, and so the group G is a strong deformation retraction of the groupoid
Y , since the diagram
A
j

u
B

w X
u
w Y

is a pushout in groupoids. The group G is therefore isomorphic to 1 (Y, x).

158

III. Classical results and constructions

2. Simplicial abelian groups: the Dold-Kan correspondence.

Suppose that A is a simplicial abelian group, and define
N An =

n1
\

ker(di ) An .

i=0

The maps
(1)n dn

N An N An1
form a chain complex, on account of the simplicial identity
dn1 dn = dn1 dn1 .
Denote the corresponding chain complex by N A; this is the normalized chain
complex associated to the simplicial abelian group A. The assignment A 7 N A
is plainly a functor from the category sAb of simplicial abelian groups to the
category Ch+ of chain complexes.
The Moore complex of a simplicial abelian group A has the group An of nsimplices of A as n-chains, and has boundary : An An1 defined by
=

n
X

(1)i di : An An1 .

i=0

Of course, one has to verify that 2 = 0, but this is a consequence of the simplicial
identities. The notation A will be used for the second purpose of denoting the
Moore complex of the simplicial abelian group A this could be confusing, but
it almost never is.
Let DAn denote the subgroup of An which is generated by the degenerate
simplices. The boundary map of the Moore complex associated to A induces
a homomorphism
: An /DAn An1 /DAn1 .
The resulting chain complex will be denoted by A/D(A), meaning A modulo
degeneracies. One sees directly from the definitions that there are chain maps
i

NA
A
A/D(A),
where i is the obvious inclusion and p is the canonical projection.

2. Simplicial abelian groups: the Dold-Kan correspondence 159

Theorem 2.1. The composite
pi

N A A/D(A)
is an isomorphism of chain complexes.
Proof: Write
N j An =

j
\

ker(di ) An ,

i=0

and let Dj (An ) be the subgroup of An which is generated by the images of the
degeneracies si for i j. One shows that the composite
p

Nj An , An
An /Dj (An )
is an isomorphism for all n and j < n. Let denote this composite.
This is proved by induction on j. The case j = 0 is easy: any class [x]
An /s0 (An1 ) is represented by x s0 d0 x, and d0 (x s0 d0 x) = 0, so is onto;
if d0 x = 0 and x = s0 y, then
0 = d0 x = d0 s0 y = y,
so x = 0.
Suppose that the map
: Nk Am Am /Dk (Am )
is known to be an isomorphism if k < j, and consider the map
: Nj An An /Dj (An ).
Form the commutative diagram

Nj1 An
u

z
Nj An

w An /Dj1 (An )
uu

w An /Dj (An ).

160

III. Classical results and constructions

On account ot the displayed isomorphism, any class [x] An /Dj (An ) can be
represented by an element x Nj1 An . But then x sj dj x is in Nj An and
represents [x], so the bottom map in the diagram is onto. The simplicial
identities imply that the degeneracy sj : An1 An maps Nj1 An1 into
Nj1 An , and takes Dj1 (An1 ) to Dj1 (An ), and so there is a commutative
diagram
Nj1 An1

w An1 /Dj1 (An1 )

sj

sj
u

Nj1 An

u
w An /Dj1 (An ).

Furthermore, the sequence

sj

0 An1 /Dj1 (An1 ) An /Dj1 (An ) An /Dj (An ) 0

is exact. Thus, if (x) = 0 for some x Nj An , then x = sj y for some y
Nj1 An . But (again), dj x = 0, so that
0 = dj x = dj sj y = y,
so that x = 0, and our map is injective.

Every simplicial structure map d : An Am corresponding to a monomorphism d : m , n of ordinal numbers takes N An into N Am . In fact, such maps
are 0 unless d is of the form d = dn : n 1 n. Put a different way, suppose
given a collection of abelian group homomorphisms
: Cn Cn1 , n 0.
Associate to each ordinal number n the group Cn , and map each ordinal number
monomorphism to an abelian group homomorphisms by the rule
(
d 7

if d is not some dn , and

0
(1)n

Cn Cn1

if d = dn .

Then we get a contravariant functor on the category of ordinal number morphisms from such an assignment if and only if we started with a chain complex.

2. Simplicial abelian groups: the Dold-Kan correspondence 161

There is a simplicial abelian group whose n-simplices have the form
M
N Ak .
nk

Note that the direct sum is indexed over all ordinal number epimorphisms whose
source is n. The map
M
M
N Ar
:
N Ak
mr

nk

associated to the ordinal number map : m n is given on the summand

corresponding to : n k by the composite
d

int

N Ak N As

N Ar ,

mr

where

m s , k
is the epi-monic factorization of the composite

m
n k.
Note as well that there is a morphism of simplicial abelian groups which is
given in degree n by the map
M
:
N A k An ,
nk

which is given at the summand corresponding to : n k by the composite

N Ak , Ak An .
Proposition 2.2. The map is a natural isomorphism of simplicial abelian
groups.
Proof: An induction on the degree n starts with the observation that N A0 =
A0 , and that theres only one map from the ordinal number 0 to itself. Suppose
that is known to be an isomorphism in degrees less than n. Then any degeneracy sj x An is in the image of , because x is in the image of in degree
n 1. On the other hand, induces an isomorphism of normalized complexes,
so is epi in degree n by Theorem 2.1.

162

III. Classical results and constructions

Suppose that (x ) maps to 0 under , where x is the component of (x )

which corresponds to : n k. If k < n, then has a section d : k n,
and the component of d (x ) which corresponds to the identity on k is x . But
d (x ) = 0, so d (x ) = 0 by the inductive hypothesis, and so x = 0 in N Ak .
Thus, x = 0 for : n k with k < n. The remaining possibly non-trivial
component is x1n N An . But the restriction of to N An is the inclusion
N An , An , so that x1n = 0 as well.

Implicit in the above is a definition of a functor

: Ch+ sAb
from chain complexes to simplicial abelian groups, with
M
(C)n =
Ck
nk

for a chain complex C, and with simplicial structure maps given by the recipe
above. The following result is now obvious from the work that we have done:
Corollary 2.3 (Dold-Kan). The functors
N : sAb Ch+

and : Ch+ sAb

form an equivalence of categories.

Proof: The natural isomorphism N (A)
= A is Proposition 2.2, and the nat
ural isomorphism N (C) = C can be derived from Theorem 2.1 by collapsing
(C) by degeneracies.

There is a subcomplex Nj A of the Moore complex A, which is defined by

Tj
Nj An =

ker(di ) for n j + 2,
N An
for n j + 1.
i=0

To see that these groups form a subcomplex, one has to verify that given x
Nj An with n j + 2, then
dk (

n
X

(1)i di (x)) = 0

i=j+1

if k j. This is a consequence of the simplicial identities dk di = di1 dk that

hold for i > k.

2. Simplicial abelian groups: the Dold-Kan correspondence 163

Observe that N0 A = A, that Nj+1 A Nj A, and that N A = j0 Nj A. Let
ij denote the inclusion of Nj+1 A into Nj A.
Now define abelian group homomorphisms fj : Nj An Nj+1 An by specifying
that

x sj+1 dj+1 (x) if n j + 2, and

fj (x) =
x
if n j + 1.
One has to check that fj takes values in Nj+1 An , but this is an easy simplicial
identity argument. The simplicial identities also imply that the collection of
maps fj defines a chain map fj : Nj A Nj+1 A. It is easily seen that the
composite fj ij is the identity on the chain complex Nj+1 A.
Now define a abelian group homomorphisms tj : Nj An Nj An+1 by

tj (x) =

(1)j sj+1
0

if n j + 1, and
otherwise.

Then a little more calculating in the night shows that

1 ij fj = tj + tj : Nj An Nj An
in all degrees n, and that both sides of the equation are 0 in degrees n j + 1.
It follows that the composite abelian group homomorphisms
f0

f1

f2

fn2

An = N0 An N1 An N2 An Nn1 An = N An
define a chain map f : A N A such that f i : N A N A is the identity. The
collection of homomorphisms T : An An+1 defined by
T = i0 in2 tn1 fn2 f0 + i0 in3 tn2 fn3 f0 + + i0 t1 f0 + t0
specifies a chain homotopy i f ' 1A .
The chain maps i, f and the chain homotopy T are natural with respect to
morphisms of simplicial abelian groups A. We have proved
Theorem 2.4. The inclusion i : N A A of the normalized chain complex in
the Moore complex of a simplicial abelian group A is a chain homotopy equivalence. This equivalence is natural with respect to simplicial abelian groups
A.

164

III. Classical results and constructions

The simplicial abelian group structure on A induces an abelian group structure

on the set
n (A, 0) = [(n , n ), (A, 0)]
of homotopy classes of pairs of maps which satisfies an interchange law with
respect to the canonical group structure for the homotopy group. It follows that
the homotopy group structure and the induced abelian group structure coincide.
In particular, there is a natural isomorphism
n (A, 0)
= Hn (N A)
for n 0. Theorem 2.4 therefore immediately implies the following:
Corollary 2.5. Suppose that A is a simplicial abelian group. Then there are
isomorphisms
n (A, 0)
= Hn (N A)
= Hn (A),
where Hn (A) is the nth homology group of the Moore complex associated to A.
These isomorphisms are natural in simplicial abelian groups A.
The group A0 of vertices of A acts on the simplicial set underlying A, via the
composite
c1

A0 A A A A,
where c : A0 A is the simplicial abelian group homomorphism given by
inclusion of vertices, and + is the abelian group structure on A. It follows that
multiplication by a vertex a induces an isomorphism of homotopy groups
+a

n (A, 0) n (A, a)

for n 0. It also follows that a homomorphism f : A B of simplicial abelian

groups is a weak equivalence of the underlying Kan complexes if and only if f
induces a homology isomorphism (or quasi-isomorphism) f : N A N B of the
associated normalized chain complexes. Equivalently, f is a weak equivalence
if and only if the induced map f : A B of Moore complexes is a homology
isomorphism.
Say that a simplicial abelian group homomorphism f : A B is a weak equivalence if f is a weak equivalence of the underlying Kan complexes. We say that f
is a fibration if the underlying simplicial set map is a Kan fibration. Finally, cofibrations in the category sAb of simplicial abelian groups are morphisms which
have the left lifting property with respect to all maps which are fibrations and
weak equivalences.

2. Simplicial abelian groups: the Dold-Kan correspondence 165

Theorem 2.6. With these definitions, the category sAb of simplicial abelian
groups satisfies the axioms for a closed model category.
This result is a consequence of Theorem II.5.1 see also Remark II.6.2. One
can also argue directly as follows:
Proof: The limit axiom CM1, the weak equivalence axiom CM2 and the
retraction axiom CM3 are trivial to verify.
A map p : A B of simplicial abelian groups is a fibration if and only if
it has the right lifting property with respect to all morphism i : Znk Zn
induced by the inclusions nk n . The simplicial abelian group Znk is a
degreewise direct summand of Zn . Note as well that the each of these maps i is
a weak equivalence, because the underlying inclusions are weak equivalences and
therefore integral homology isomorphisms. A small object argument therefore
implies that any map f : C D of simplicial abelian groups has a factorization
C[
]
[
j

wD

E
such that p is a fibration and j is a morphism which has the left lifting property
with respect to all fibrations, is a weak equivalence, and is a monomorphism in
each degree. In particular, j is a cofibration and a weak equivalence, and so the
corresponding factorization axiom is verified.
Similarly, a map p : A B is a fibration and a weak equivalence if and only
if it has the right lifting property with respect to all morphisms Zn Zn
induced by the inclusions n n . It follows again by a small object argument
that any map f : C D has a factorization
C[
]
[
i

wD

E
such that q is a fibration and a weak equivalence, and such that i is a cofibration and a levelwise monomorphism. We have therefore completely verified the
factorization axiom CM5.
By standard nonsense, any map which is a cofibration and a weak equivalence
is a retract of a map of the form j in the proof of the factorization axiom, so
that has the left lifting property with respect to all fibrations. This implies
the lifting axiom CM4.

166

III. Classical results and constructions

Remark 2.7. It is a corollary of the proof of Theorem 2.6 that all cofibrations
of sAb are levelwise monomorphisms.
Lemma 2.8. Suppose that f : A B is a homomorphism of simplicial abelian
groups which is surjective in all degrees. Then f is a fibration of simplicial
abelian groups.
Proof: Suppose given a commutative diagram of simplicial set maps

nk
y
u
n

wA
f
u
w B.

Then, by the assumptions, there is a simplex An such that f () = . But

then
|nk : nk A
factors through the kernel K of f , and so |nk extends to an n-simplex x of
K in the sense that there is a commutative diagram of simplicial set maps

nk
y

|nk

wK
N
P
N
Nx
N

u N
n

Then ( x)|nk = and f ( x) = .

Lemma 2.9.
(1) A homomorphism f : A B of simplicial abelian groups is surjective (in
all degrees) if and only if the associated chain complex map f : N A N B
is surjective in all degrees.
(2) The homomorphism f : A B of simplicial abelian groups is a fibration
if and only if the induced abelian group maps f : N An N Bn are
surjective for n 1.

2. Simplicial abelian groups: the Dold-Kan correspondence 167

Proof: For (1), note that the map f : N A N B of normalized chain complexes is a retract of the map f : A B of Moore complexes. Thus, if the
simplicial abelian group homomorphism f : A B is surjective in all degrees,
then so is the associated map of normalized chain complexes.
Conversely, one shows that if f : Nj+1 A Nj+1 B is surjective in all degrees,
then the abelian group homomorphisms f : Nj An Nj Bn are surjective for all
n 0. This is proved by induction on n. Take x Nj Bn . Then x sj+1 dj+1 x
is in Nj+1 Bn and so is in the image of f : Nj+1 An Nj+1 Bn . Also, dj+1
Nj Bn1 and is therefore in the image of f : Nj An1 Nj Bn1 by the inductive
assumption. It follows that x is in the image of f : Nj An Nj Bn .
For (2), suppose that f : A B is a fibration of simplicial abelian groups.
Then the existence of solutions to the lifting problems of the form
nn
y

u
n

wA
f
u
wB

implies that f : N An N Bn is surjective for n 1.

Conversely, suppose that f : N An N Bn is surjective in non-zero degrees.
Form the diagram
A
AAA
4
AAA
AAA f
44

AAA
AAAA
6
4

C
A

wB
K(0 A, 0) K(0 B,0) B
pr

u
u
K(0 A, 0)
w K(0 B, 0),
f
where (for example) K(0 A, 0) is the constant simplicial abelian group on the
abelian group 0 A. The hypotheses imply that applying the normalization functor to the map gives a surjective chain map
: N A N K(0 A, 0) N K(0 B,0) N B
= 0 A[0] 0 B[0] N B,

168

III. Classical results and constructions

and so (1) implies that the simplicial abelian group map

: A K(0 A, 0) K(0 B,0) B
is surjective in all degrees and therefore a fibration by Lemma 2.8. The map
f : K(0 A, 0) K(0 B, 0) is a fibration, so that pr is a fibration. It follows
that f = pr is a fibration.

Corollary 2.10. The homomorphism f : A B is a trivial fibration of simplicial abelian groups if and only if the induced morphism f : N A N B of
normalized chain complexes is surjective in all degrees, with acyclic kernel.
It follows that the category Ch+ of chain complexes of abelian groups inherits
a closed model structure from the simplicial abelian group category, in which the
fibrations are the chain maps f : C D such that f is surjective in degree n for
n 1, and where the weak equivalences are the quasi-isomorphisms, or rather
the maps which induce isomorphisms in all homology groups. The cofibrations
of Ch+ are those maps which have the left lifting property with respect to all
trivial fibrations.
One can, alternatively, give a direct proof of the existence of this closed model
structure on the chain complex category Ch+ . The proof of the factorization
axioms is a small object argument which is based on some rather elementary
constructions. Explicitly, let Z[n] be the chain complex consisting of a copy of
the integers Z in degree n and 0 elsewhere, and let Zhn + 1i be the chain complex
n+1 =1

0 Z Z 0 .
Then maps Z[n] C classify n-cycles of C, and Zhn+1i is the free chain complex
on an element of degree n + 1. Write x for the generator of Z[n] in degree n
and write y for the generator of Zhn + 1i in degree n + 1. There is a canonical
map j : Z[n] Zhn + 1i which is defined by j(x) = (y). Then f : C D is a
fibration if and only if f has the right lifting property with respect to all chain
maps 0 Zhn + 1i for n 0. Further, one can show that f is a trivial fibration
if and only if f : C0 D0 is surjective and f has the right lifting property with
respect to all maps j : Z[n] Zhn + 1i, n 0.
The chain complex category Ch+ has a natural cylinder construction. For
any chain complex C, there is a chain complex C I with

CnI

Cn Cn Cn+1

if n 1, and

{(x, y, z) C0 C0 C1 |(x y) + z = 0}

if n = 0.

2. Simplicial abelian groups: the Dold-Kan correspondence 169

and with

(x, y, z) = (x, y, (1)n (x y) + z).

Then there is a commutative diagram of chain maps

N
P
N

N
N s

N
C

CI

p
u
w C C,

where p is the fibration defined by p(x, y, z) = (x, y) and s is a weak equivalence

which is defined by s(x) = (x, x, 0). It is an exercise to show that there is a
homotopy h : D C I from f to g if and only if the maps f, g : C D are
chain homotopic.
Proposition 2.11. The category of simplicial abelian groups admits a simplicial model structure.
Proof: If K is a simplicial set and A is a simplicial abelian group, then there
is a simplicial abelian group A K, which is defined by
A K = A ZK,
where (in this case) ZK denotes the free simplicial abelian group associated to
K. Equivalently, on the level of n-simplices, there is a canonical isomorphism
AK =

An ,

Kn

with simplicial structure maps induced in an obvious way from the corresponding
maps for A and the simplicial set K. Dually, the simplicial abelian group structure on A induces a simplicial abelian group structure on the simplicial function
space Hom(K, A). Finally, for simplicial abelian groups A and B, one defines
the simplicial set HomsAb (A, B) to have n-simplices given by the set (actually
abelian group) of simplicial abelian group homomorphisms A n B. Then
there are natural isomorphisms
HomsAb (A K, B)
= Hom(K, HomsAb (A, B)

= HomsAb (A, Hom(K, B)).

170

III. Classical results and constructions

The first of these isomorphisms follows from the exponential law

homsAb (A K, B)
= homS (K, HomsAb (A, B)),
which itself is a specialization of the simplicial set exponential law. The second
follows from the definition of HomsAb (A, B), together with the observation
that the simplicial set K is a colimit of its simplices. Its obvious now that
if f : A B is a fibration of simplicial abelian groups and i : K , L is a
cofibration of simplicial sets, then the induced map
(i ,f )

Hom(L, A) Hom(K, A) Hom(K,B) Hom(L, B)

is a fibration of simplicial abelian groups which is trivial if either i or f is trivial,
just by remembering that the underlying simplicial set map has the same properties. This is Quillens axiom SM7(a), whence the simplicial model structure
on sAb.

Proposition 2.11 can also be proved by appealing to Theorem II.5.4.

Remark 2.12. The category of simplicial abelian groups is also a proper simplicial model category (see Section II.8). In particular,
(1) weak equivalences are stable under pullback along fibrations, and
(2) the pushout of a weak equivalence along a cofibration is a weak equivalence.
The first requirement means that, in the pullback diagram
g

C B A

wA
p

u
C

u
w B,

if p is a fibration and g is a weak equivalence, then g is a weak equivalence.

This follows from the corresponding property for the category of simplicial sets.
The second requirement says that, given a pushout diagram
A

u
B

wC
u

w B A C,

2. Simplicial abelian groups: the Dold-Kan correspondence 171

if i is a cofibration and f is a weak equivalence, then f is a weak equivalence.
But of course i and i are monomorphisms which have the same cokernel, and so
a comparison of long exact sequences in homology shows that f is a homology
isomorphism.
The simplicial model structure on the simplicial abelian group category gives
rise to a canonical choice of cylinder for cofibrant objects A. In effect, tensoring
such an A with the commutative simplicial set diagram
1
w 0
y
[
]
[
i
[
u [
1
gives a natural diagram of simplicial abelian group homomorphisms of the form
AA
i = (d0 , d1 )

u
A 1

wA

j
hh
hs
h

The map i coincides up to isomorphism with the map

1 i : A 1 A 1 ,
while the map d0 coincides with
1 d0 : A 0 A 1 .
Since A is cofibrant, the simplicial model structure implies that 1 i is a cofibration and 1 d0 is a trivial cofibration. It follows that the map s is a weak
equivalence.
The free simplicial abelian group ZX on a simplicial set X is a cofibrant
simplicial abelian group, and there is a natural isomorphism ZXK
= Z(XK).
1
It follows that an ordinary simplicial homotopy X B taking values in a
simplicial abelian group B induces a homotopy ZX 1 B for the cylinder
ZX 1 . Since ZX is cofibrant and B is fibrant, the existence of a simplicial

172

III. Classical results and constructions

homotopy X 1 B between maps f and g is equivalent to the existence

of a chain homotopy between the induced maps f , g : ZX B of Moore
complexes: the induced maps N f , N g : N ZX N B must be chain homotopic
by formal nonsense, and so the induced maps of Moore complexes must be chain
homotopic by Theorem 2.4. There is a classical alternative method of seeing this
point, based on the following
Lemma 2.13. Any homotopy h : A 1 B from f : A B to g : A B
gives rise to a chain homotopy between the associated maps of Moore complexes.
Proof: Let j : n 1 be the ordinal number morphism such that j (i) = 0 if
and only if i j. Now, given the map h, define abelian group homomorphisms
hj : An Bn+1 , 0 j n, by specifying that
hj (a) = h(sj (a) j ).
Then, as a consequence of the simplicial identities and the relations

j1 if i j,
di (j ) =
j
if i > j,
and

si (j ) =

j+1
j

if i j,
if i > j

in the simplicial set 1 , one finds the relations

d0 h0 = f,
dn+1 hn = g,
di hj = hj1 di
if i < j,
dj+1 hj = dj+1 hj+1 ,
di hj = hj di1
if i > j + 1,
si hj = hj+1 si
if i j, and
si hj = hj si1
if i > j.
Its now straightforward to verify that the collection of alternating sums
s=

n
X

(1)i hi : An Bn+1

i=0

forms an explicit chain homotopy between the Moore complex maps f and g.

2. Simplicial abelian groups: the Dold-Kan correspondence 173

It follows that every simplicial abelian group homomorphism A1 B gives
rise to a chain homotopy of maps N A N B between the associated normalized
complexes. The converse is far from clear, unless A is cofibrant.
The following result establishes the relation between weak equivalence and
homology isomorphism:
Proposition 2.14. The free abelian simplicial group functor X 7 ZX preserves weak equivalences.
Proof: The existence of a contracting simplicial homotopy of a simplicial set
X onto a base point implies that the map X induces a weak equivalence
Z ZX, since the simplicial contracting homotopy gives rise to a homotopy
ZX 1 ZX in the simplicial abelian group category. It follows that the
canonical weak equivalences : S|n | n are homology isomorphisms in the
sense that they induce weak equivalences : ZS|n | Zn . An excision
argument and an induction on n therefore implies that the maps : S| skn X|
skn X are homology isomorphisms for any simplicial set X, so that the map
: S|X| X is a homology isomorphism as well.
Any weak equivalence f : X Y induces a homotopy equivalence of Kan
complexes S|f | : S|X| S|Y |. This map S|f | is a homology equivalence, so f
is a homology equivalence.

Write (ZX, A) to denote homotopy classes of maps between the named objects in the simplicial abelian group category, computed with respect to the
cylinder object ZX 1 .
The free abelian simplicial group functor X 7 ZX and the inclusion functor
i : sAb S both preserve weak equivalences. These functors are also adjoint.
It follows that they induce corresponding functors i : Ho(sAb) Ho(S) and
Z : Ho(S) Ho(sAb), and that these functors are adjoint.
One way of seeing this (see also Section II.7) begins with the observation that
there is a map [X, A] [ZX, A]sAb (where [ , ]sAb denotes morphisms in the
homotopy category Ho(sAb)) which is defined to be the composite

[X, A] [ZX, ZA]sAb [ZX, A]sAb ,

where the first function is induced by the free simplicial abelian group functor,
and : ZA A is one of the adjunction maps. The corresponding composite

(X, A) (ZX, ZA) (ZX, A)

174

III. Classical results and constructions

is an isomorphism, since Z(X 1 )

= ZX 1 . All sources are cofibrant and
all targets are fibrant, so there is a commutative diagram

(X, A)

w (ZX, ZA)

u
[X, A]

u
w [ZX, ZA]

w (ZX, A)

=
u
w [ZX, A]

in which all vertical maps are canonical bijections. It follows that the bottom
horizontal composite is a bijection.
Let U denote the forgetful functor sAb S. The adjointness of the induced
functors on the homotopy category level is also a consequence of a formal principle (Browns adjoint functor lemma [15, p.426], but see Section II.7): the
adjunction maps : ZU 1 and : 1 U Z induce natural transformations of
the corresponding induced functors between the homotopy categories, and these
transformations satisfy the triangle identities.
Now observe that there is a commutative diagram of isomorphisms

(ZX, A) OO
(2.15)

N
=
u
Ch (N ZX, N A) u

OOO

OOO

=
i

OOO

OOO
O
P
O

=
w Ch (ZX, A)
Ch (ZX, N A)
i

in which the map labelled by N is induced by the normalization functor, and is

well defined because ZX is cofibrant and A is fibrant, and the other labelled maps
are induced by the chain homotopy equivalences i : N A A and i : N ZX ZX.
The dotted arrow takes the homotopy class of a simplicial abelian group map
f : ZX A to the chain homotopy class which is represented by the map of
Moore complexes induced by f . In particular, we have proved
Proposition 2.16. Suppose that X is a simplicial set and A is a simplicial
abelian group. Then the group of simplicial homotopy classes (ZX, A) can
be canonically identified up to isomorphism with the group of chain homotopy
classes Ch (ZX, A) between the associated Moore complexes.
As a consequence of the proof of Proposition 2.16, we have the following:

2. Simplicial abelian groups: the Dold-Kan correspondence 175

Corollary 2.17. Suppose that X is a simplicial set and B is an abelian group.
Then there are canonical isomorphisms
[X, K(B, n)]
= H n (X, B),
for n 0.
Proof: The simplicial abelian group K(B, n) is B[n], where B[n] is the chain
complex which consists of the abelian group B concentrated in degree n. We
know that the set [X, K(B, n)] of morphisms in Ho(S) is canonically isomorphic to the set (X, K(B, n)) of simplicial homotopy classes, which in turn is
isomorphic to the set (really group) of homotopy classes
(ZX, K(B, n)) = (ZX, B[n])
in the simplicial abelian group category. But from the diagram (2.15), there is
are isomorphisms
(ZX, B[n])
= Ch (ZX, N B[n])

= Ch (ZX, B[n]),
and Ch (ZX, B[n]) is the nth cohomology group of the complex hom(ZX, B).
Suppose that C is a chain complex, let Zn denote the subgroup of n-cycles and
let Bn = (Cn+1 ) be the subgroup of boundaries in Cn . Pick an epimorphism
p : Fn Zn , where Fn is a free abelian group. Then the kernel Kn of the
composite
F n Z n Hn C
is free abelian, and the composite
p

Kn Fn
Zn
factors through a map p0 : Kn Bn . Since Kn is free abelian and the map
Cn+1 Bn is surjective, the map p0 lifts to a map p : Kn Cn+1 . Write Fn C
for the chain complex
n+1

0 Kn , Fn 0 .
Then the epimorphism Fn Hn C defines a quasi-isomorphism
qn : Fn C Hn C[n],

176

III. Classical results and constructions

while the maps labelled p define a chain map

pn : Fn C C
which induces an isomorphism Hn (Fn C)
= Hn C. It follows that there are quasiisomorphisms
M
qn M
pn
Hn C[n]
Fn C C.
n0

n0

Note that the canonical map

M

Hn C[n]

n0

Hn C[n]

n0

is an isomorphism of chain complexes.

If the chain complex C happens to be the normalized complex N A of a simplicial abelian group A, then this construction translates through the functor
into weak equivalences of simplicial abelian groups
Y
M
K(n A, n)
Fn N A A.
n0

n0

These objects are fibrant in the category of simplicial sets, and so the weak
equivalences induce homotopy equivalences of simplicial sets, proving
Proposition 2.18. Suppose that A is a simplicial abelian group. Then, as
a simplicial set, A is non-canonically homotopy equivalent to the product of
Eilenberg-Mac Lane spaces
Y
K(n A, n).
n0

The key point in the argument for Proposition 2.18 is that a subgroup of a free
abelian group is free, or at least projective. An analogous statement holds for
modules over a principal ideal domain R so the simplicial abelian group A in the
statement of the proposition can be replaced (at least) by a simplicial module
over such a ring.
This is implicit above, but there is a natural short exact sequence
p

0 K(A, n) W K(A, n)
K(A, n + 1) 0
which are constructed by applying the functor to the short exact sequence
0 A[n] Ahn + 1i A[n + 1] 0

3. The Hurewicz map

177

of chain complexes. The simplicial abelian group W K(A, n) is contractible, and

the sequence
p

K(A, n) W K(A, n)
K(A, n + 1)
is one of the standard fibre sequences which is used to construct an equivalence
K(A, n) ' K(A, n + 1) in the literature. There is one final observation about
the Eilenberg-Mac Lane spaces K(A, n) = A[n] and the fibration p which is
very commonly used:
Lemma 2.19. The map p : W K(A, n) K(A, n + 1) is a minimal fibration,
and K(A, n) is a minimal Kan complex, for all n 0.
Proof: There is a relation x 'p y (in the sense of Section I.10) if and only if
(x y) 'p 0, so it suffices to show that z 'p 0 implies that z = 0 for any simplex
z of W K(A, n). But z 'p 0 forces z to be in the fibre K(A, n), and so to show
that p is a minimal fibration we need only prove that K(A, n) is a minimal Kan
complex.
Suppose that z is an r-simplex of K(A, n). If z ' 0 in K(A, n) (rel r ),
then di z = 0 for all i, so that z is a normalized r-chain of K(A, n). By the
Dold-Kan correspondence, N K(A, n)r
= A[n]r , which group is 0 if r 6= n, so it
suffices to concentrate on the case r = n. There are identifications N K(A, n)n =
K(A, n)n = A, and the resulting map K(A, n)n n (K(A, n), 0) is an isomorphism. Thus z ' 0 (rel n ) means that z 7 0 in n (K(A, n), 0). But then
z = 0.

3. The Hurewicz map.

Suppose that X is a simplicial set. The Hurewicz map h : X ZX is alternate
notation for the adjunction homomorphism which associates to an n-simplex x
of X the corresponding generator of the free abelian group ZXn on Xn . If X
happens to be a connected Kan complex with base point , then h induces a
composite homomorphism of groups
n (X, Z),
n X n (ZX, ) n (ZX/Z, 0)
=H
(X, Z) is reduced homology of X. The map
where H
n (X, Z)
h : n X H
will denote this composite h is the Hurewicz homomorphism. The simplicial
set X will be a connected Kan complex with base point throughout this sec X to mean
tion, and all homology groups will be integral, so we shall write H

H (X, Z). For ease of notation, write

= ZX/Z .
ZX

178

III. Classical results and constructions

n X is the traditional map. To

The Hurewicz homomorphism h : n X H
n
n
n
see this, let S = / , with the obvious choice of base point. Then n X is
pointed simplicial homotopy classes of maps S n X, and the homotopy group
n)
nSn
n (ZS
=H
is a copy of the integers, canonically generated by the homotopy elemen n which
is represented by the composite map
h

n.
Sn
ZS n ZS
It is now easily seen, by drawing an appropriate commutative diagram, that if
: S n X represents an element [] n X, then
n X,
h ([]) = (n ) H
as in the standard definition.
Suppose that p : E B is a Kan fibration, where B is a simply connected Kan
complex with base point , and let F be the fibre over . There is a complete
description of the Serre spectral sequence
E2p,q = Hp (B, Hq F ) Hp+q B
given below in IV.5.1, but it is non-standard. To recover the more usual form,
filter the base B by skeleta skn B, and form the pullback diagrams
Fn E y

wE
p

p
u
skn B y

u
wB

to obtain a filtration Fn E of the total space E and fibrations p : Fn E skn B

for which 1 skn B acts trivially on the homology H F of F .
Recall that Nn B denotes the set of non-degenerate n-simplices of B. The
pushout diagram
G
n
w skn1 B
y
xNn B
y
Gu
xNn B

w skn B

3. The Hurewicz map

179

can be pulled back along p to obtain an identification of the filtration quotient

Fn E/Fn1 E with the wedge
_

p1 (n )/p1 (n ).

xNn B

The spectral sequence

Hr (n , Hs F ) Hr+s p1 (n )
of IV.5 is used, along with the fact that H F splits off H p1 (n ) to show that
there is an isomorphism
H p1 (n )
= H F H F [n 1],
and that the map induced in homology by the inclusion p1 (n ) p1 (n )
can be identified up to isomorphism with the projection
H F H F [n 1] H F
In particular, there is an isomorphism
(p1 (n )/p1 (n ))
H
= H F [n],
and hence isomorphisms
(Fn E/Fn1 E)
H
=

H F [n].

xNn B

Thus,
p+q (Fp E/Fp1 E)
H
=

Hq F,

xNp B

and its a matter of chasing simplices through the boundary map of the normalized complex N ZB Hq F to see that the E2 -term of the spectral sequence for
H E arising from the skeletal filtration for B has the form
E2p,q = Hp (B, Hq F ).
All appeals to the Serre spectral sequence in the rest of this section will be
specifically to this form.

180

III. Classical results and constructions

Now suppose that Y is a pointed Kan complex, and observe that the canonical
path object P Y is that pointed function complex
P Y = Hom (1 , Y ),
where 1 is a copy of the standard 1-simplex 1 , pointed by the vertex 1. Furthermore, the loop space Y can be identified with the complex Hom (S 1 , Y ),
and the path loop fibration for Y is the fibre sequence

d1

Hom (S , Y ) Hom (1 , Y ) Y,
where : 1 S 1 is the canonical map.
There is a canonical contracting pointed homotopy
h : 1 1 1 ,
which is defined by the picture
04
u
1

4
44
6

w1
u
w1

in the poset 1. This homotopy h induces a contracting homotopy

h

Hom (1 , Y ) 1 Hom (1 , Y )
for the path space on Y , by adjointness.
Suppose that f : X Hom (S 1 , Y ) is a pointed map, and denote the composite
f 1

X 1 Hom (1 , Y ) 1
Hom (1 , Y )
by f . Then there is a commutative diagram of pointed simplicial set maps
X
(3.1)

d1

f
u
Hom (S 1 , Y )

w X 1

u
w Hom (1 , Y )

w X S1

d1

u
w Y,

3. The Hurewicz map

181

where the indicated map f is the adjoint of f , or rather the composite

f 1

ev

X S 1 Hom (S 1 , Y ) S 1 Y.
Note as well that X 1 is a model for the pointed cone on X (a different model
for the cone is given in Section III.5 below).
Suppose that

F E
B
is a fibre sequence of pointed Kan complexes, and that B is simply connected.
The edge maps
F
u

wE

wB

1
u
wB

u
wB

and the calculation of the E2 -terms for the Serre spectral sequences for the
corresponding fibrations together imply that Enn,0 in the Serre spectral sequence
for H E is the subgroup of Hn B consisting of elements x which can be lifted
along the staircase

Hn1 F

Hn (skn B)

u
Hn (Fn E/Fn1 E) w Hn1 Fn1 E
p
u
w Hn (skn B/ skn1 B)

u
Hn B
to an element z Hn1 F , in the sense that there are elements x1 Hn (skn B)
and x2 Hn (Fn E/Fn1 E) such that x1 7 x, x1 7 p (x2 ), and z 7 (x2 ).
Furthermore, the image of such an x under the differential dn : E2n,0 E20,n1
is represented by the element z. But then comparing long exact sequences shows
that the element x2 is in the image of the map Hn (Fn E/F ) Hn (Fn E/Fn1 E).

182

III. Classical results and constructions

As a consequence, the elements x of Enn,0 can be identified with elements of

Hn (B/) which are in the image of the map p : Hn (E/F ) Hn (B/), and
for such x, if there is an element x1 Hn (E/F ) such that p (x1 ) = x, then
dn (x) is represented by (x1 ) Hn1 F . This is a classical description of the
transgression.
Suppose that Y is an n-connected pointed Kan complex, where n 1, and
consider the Serre spectral sequence for the path loop fibre sequence
d1

Y P Y Y.
i Y = 0 for i n, while H
j Y = 0 for j n 1. The point is that the
Then H
assumption on the connectivity of Kan complex Y , for example, that there is a
strong deformation retract Z of Y such that the n-skeleton skn Z is a point1 . It
follows that E2i,0 = Eii,0 and E20,i1 = Ei0,i1 for i 2n, and the only possible
non-trivial differential into or out of either group is the transgression, so that
there is an exact sequence
di

i,0
0,i1
Eii,0 Ei0,i1 E
0 E
0.

The space P Y is acyclic, so all E -terms vanish in non-zero total degree, so we

have shown that the transgression
di

Hi Y Hi1 Y
is an isomorphism for i 2n under the assumption that Y is n-connected.
The transgression di : Hi Y Hi1 Y is, at the same time, related to the
boundary map

=
i (X S 1 )
i1 X,
:H
H

in the sense of the following result:

Lemma 3.2. Suppose that f : X Y is a map of pointed simplicial sets,
where Y is an n-connected Kan complex with n 1. Then, for i 2n, there is
1

This is an old idea. People sometimes say that complexes of the form Z are n-reduced.
The construction of the deformation retraction (via an iterated homotopy extension property
argument) is one of the early applications of the Kan complex concept, and should be done as
an exercise.

3. The Hurewicz map

183

a commutative diagram of the form

i (X S 1 )
H

i1 X
wH

u
iY
H

u
i1 Y,
wH

=
di

where f : X S 1 Y is the adjoint of the map f .

Proof: Write X = X S 1 and CX = X 1 , as usual. Then the diagram
(3.1) induces a commutative diagram

i X
i (X/) u
H
=H
f

i (CX/X)
H

i1 X
wH

i (Y /) u
iY
H
=H

d1

u
i (P Y /Y )
H

i1 Y
wH

i X H
i1 X.
The top composite in the diagram is the boundary map : H
The bottom composite is the transgression, according to the discussion preceding the statement of the lemma.

Corollary 3.3. Suppose that Y is an n-connected pointed Kan complex, with

iY H
i1 Y can be identified
n 1. Then, for i 2n, the transgression di : H
with the composite

i (Y )
i ((Y ) S 1 ) H
i1 (Y ).
H
H

Proof: The adjunction map : (Y ) S 1 Y is the adjoint of the identity

map on the loop space Y .

The link between the Hurewicz homomorphism and the transgression is the
following:

184

III. Classical results and constructions

Proposition 3.4. Suppose that Y is an n-connected pointed Kan complex,

with n 1. Then, for i 2n, there is a commutative diagram

i Y

u
iY
H

w i1 (Y )
h

=
di

u
i1 (Y )
wH

Proof: The instance of the diagram (3.1) corresponding to the identity map on
Y and the naturality of the Hurewicz map together give rise to a commutative
diagram

ZY

ZY
u h Y

ZCY

u
u
h
Y u
w ZP
PY

d1
u
u
1

Z((Y ) S ) w ZY u

u
Y

Y ZY
has a factorization
The homomorphism d1 : ZP
j

ZP
B
ZY
in the category of simplicial abelian groups (or of simplicial sets your choice),
where j is a trivial cofibration and q is a fibration. Then, by comparing boundary
maps for the resulting fibrations of simplicial sets, one finds a commutative
diagram
i Y
h

w i1 (Y )

h
u

iY u = H
i ((Y ) S 1 ) = w H
i1 (Y ).
H

Now use Corollary 3.3.

3. The Hurewicz map

185

Write x y for n-simplices x and y of a simplicial set Z if x|skr n = y|skr n ,

or equivalently if the composite simplicial set maps
x

skr n n Z
and

skr n n Z
r

coincide. The relation is clearly an equivalence relation on the simplices of Z,

r
and we may form the quotient simplicial set Z(r) = Z/ and write pr : Z
Z(r). The simplicial set Z(r) is called either the rth Moore-Postnikov section of
Z, or the rth Postnikov section of Z. This construction is natural in Z: one can
show quickly that if Z is a Kan complex, then so is Z(r), and furthermore pr
induces isomorphisms j (Z, x)
= j (Z(r), x) for j r and all vertices x of Z,
and j (Z(r), x) = 0 for j > r.
Postnikov sections will be discussed more thoroughly in Chapter VI.
If X is a pointed connected Kan complex, then the object X(1) is a pointed
connected Kan complex of type K(1 X, 1), meaning that its homotopy groups
consist of 1 X in dimension one and 0 elsewhere. There is, however, a more
geometrically satisfying way to construct a space naturally having the homotopy
type of X(1) which uses the fundamental groupoid construction: we have seen
in Section III.1 that there is a map
X B(1 X)
of Kan complexes which induces an isomorphism on fundamental groups. It
follows, for example, that a connected Kan complex of type K(, 1) is weakly
equivalent to the space B.
1 (B) is isomorphic to the
The Hurewicz homomorphism h : 1 (B) H
canonical group homomorphism /[, ] from to its abelianization
this may be seen directly, or by invoking the following result:
Lemma 3.5. Suppose that Z is a Kan complex, such that the set Z0 of vertices
of Z consists of a single point. The the Hurewicz map
1 (Z)
h : 1 Z H
can be identified up to isomorphism with the canonical homomorphism
1 Z 1 Z/[1 Z, 1 Z].

186

III. Classical results and constructions

Proof: Since Z is reduced, the integral homology group H1 (Z) is the quotient
M
H1 (Z) =
Z/hd0 d1 + d2 | Z2 i.
Z1

Up to this identification, the Hurewicz map

M
h : 1 Z
Z/hd0 d1 + d2 | Z2 i
Z1

is defined by [] 7 [], for 1-simplices of Z. One then shows, by chasing

elements, that the map h is initial among all group homomorphisms f : 1 Z A
which take values in abelian groups A.

Corollary 3.6. Suppose that X is a connected pointed Kan complex. Then

1 X is isomorphic to the canonical
the Hurewicz homomorphism h : 1 X H
homomorphism
1 X 1 X/[1 X, 1 X]
from the fundamental group of X to its abelianization.
Proof: The space X has a strong deformation retract Z which is reduced. Now
apply Lemma 3.5.

Theorem 3.7 (Hurewicz). Suppose that X is an n-connected Kan complex,

i X is an
where n 1. Then the Hurewicz homomorphism h : i X H
isomorphism if i = n + 1 and an epimorphism if i = n + 2.
Proof: Suppose that F is the homotopy fibre of the map pn+1 : X X(n + 1).
Then there are commutative diagrams
n+1 X
h
u

n+1 X
H

w n+1 X(n + 1)

n+2 F

h
u
n+1 X(n + 1)
wH

h
u

n+2 F
H

w n+2 X
u

n+2 X
wH

A Serre spectral sequence argument for the fibration pn+1 shows that the map
n+1 X H
n+1 X(n + 1) is an isomorphism, since the space F is (n + 1)H
connected, and that it suffices to show
n+1 X(n + 1)
(1) that the Hurewicz homomorphism h : n+1 X(n + 1) H
is an isomorphism, and
n+2 X(n + 1) = 0.
(2) H

3. The Hurewicz map

187

If these two statements are demonstrated (for all n), then the general statement
n+1 X is an isomorphism would be true, so that the map
that h : n+1 X H
n+2 F would be an isomorphism as well. Furthermore, the
h : n+2 F H
n+2 X(n + 1) = 0 implies, via the Serre spectral sequence for
assertion that H
n+2 F H
n+2 X is an epimorphism.
pn+1 , that the map H
n+1 X(n + 1) is an isomorphism
But the claim that h : n+1 X(n + 1) H
reduces, by an inductive transgression argument involving Proposition 3.4, to
Corollary 3.6, so statement (1) is proved. Similarly, statement (2) is reduced to
3 Y = 0 for any connected Kan complex Y of type K(A, 2).
showing that H
Let Y be such a Kan complex, and pick a strong deformation retraction map
r : Y W onto a 2-reduced subcomplex. Then there is a commutative diagram

r
u
W

w ZY
r
u

w ZW

The map r is a weak equivalence of simplicial abelian groups, hence of simplicial

is 0 in degrees less than 2, as is its associated normalsets, and the complex ZW
. There is a map of chain complexes N ZW
H
2 Y [2]
ized chain complex N ZW
which induces an isomorphism in H2 . It follows that the induced composite
h

H
2 Y [2] = K(H
2 Y, 2)
Y
ZY
ZW

is a weak equivalence.
We are therefore required only to show that H3 K(A, 2) = 0 for any abelian
group A. The functor A 7 H3 K(A, 2) preserves filtered colimits, so it suffices to
presume that A is finitely generated. The functor A 7 K(A, 2) takes finite direct
sums to products of simplicial sets, and a K
unneth exact sequence argument
shows that H3 K(A B, 2) = 0 if H3 K(A, 2) = H3 K(B, 2) = 0. Finally, a few
more Serre spectral sequence arguments, combined with knowing that the circle
has type K(Z, 1) imply (successively) that H3 K(Z, 2) = 0 and H3 K(Z/n, 2) = 0
for any n.

th
There is a very easy way to form what one might call the n Postnikov section
C(n) of a chain complex C: define C(n) to be the chain complex
0 Cn /Im Cn1 C0 .

188

III. Classical results and constructions

Here, Im is the image of the boundary map : Cn+1 Cn . There is an

obvious chain map pn : C C(n); this map induces an isomorphism in Hi for
i n, and Hi C(n) is plainly trivial for i > n. There is also a natural chain map
in : Hn C[n] C(n), and this map is a homology isomorphism. We could have
used this construction in place of the retraction onto the 2-reduced complex W
in the proof of the Hurewicz Theorem. This construction is also used to prove
the following:
Corollary 3.8. Let Y be a connected Kan complex with n Y
= A, and j Y =
0 for j 6= n, where n 2. Then Y is naturally weakly equivalent to the space
K(A, n) = A[n].
The naturality in the statement of this result is with respect to maps Y Z of
connected Kan complexes having only one non-trivial homotopy group, in degree
n.
(N ZY
(n)) be the simplicial abelian group map inProof: Let pn : ZY
N Z(n)

duced by the nth Postnikov section map pn : N ZY

of the associated
normalized chain complex. Then the composite
h

pn

(N ZY
(n))
Y
ZY
is a natural weak equivalence, by the Hurewicz Theorem and the construction
of the chain map pn . But then the following maps are both natural weak equivalences
pn h
i
(n)) n
Y (N ZY
A[n] = K(A, n),
and the result is proved.

Remark 3.9. The construction in this last proof is the only known way of showing that a diagram of spaces having only one non-trivial presheaf of homotopy
groups is weakly equivalent to a diagram of spaces K(A, n) = A[n], for some
presheaf (aka. diagram) of abelian groups A.
Theorem 3.10 (Freudenthal). Suppose that X is an n-connected pointed
space, where n 0. Then the canonical map : X (X S 1 ) induces a map
i X i (X S 1 ) which is an isomorphism if i 2n and an epimorphism if
i = 2n + 1.
Proof: We shall suppose that n 1, and leave the case n = 0 for the reader.
From the characterization of the transgression for the path-loop fibre sequence
of Corollary 3.3, the map
i (X S 1 ) H
iX
: H

3. The Hurewicz map

189

is an isomorphism for i 2n. One then uses the triangle identity

S1
X S1
w (X S 1 ) S 1

1
u
X S1
iX H
i (X S 1 ) is an isomorphism if i 2n + 1. The
to infer that : H
1
space (X S ) is simply connected by assumption, so a Serre spectral sequence
argument says that the homotopy fibre F of the map : X (X S 1 ) has
i F which vanish for i 2n. But F is a simply connected
homology groups H
space, by the Hurewicz theorem together with the fact that is a homology
isomorphism in degree 2, so that F is 2n-connected (by Hurewicz again), giving
the result.

Theorem 3.10 is the classical Freudenthal suspension theorem, since the homomorphism : 1 X i (X S 1 ) is isomorphic to the suspension homomorphism i X i+1 (X S 1 ). We shall finish this section with the relative
Hurewicz theorem:
Theorem 3.11. Suppose that f : X Y is a map with homotopy fibre F and
homotopy cofibre Y /X. Suppose that F is n-connected for some n 0 and that
the total space X is simply connected. Then the homotopy fibre of the induced
map f : F S 1 Y /X is (n + 2)-connected.
Proof: The diagram
F

u
X

u
wY

induces the map f : F S 1 Y /X of associated homotopy cofibres.

The spaces X and Y are both simply connected, by the assumptions on the
map f , and so the Serre exact sequence for the map f
dn+2

Hn+2 X Hn+2 Y Hn+1 F Hn+1 X . . .

190

III. Classical results and constructions

extends to a sequence of the form

dn+3

0,n+2
Hn+3 Y En+3
Hn+2 X Hn+2 Y . . . .

The classical description of the transgression says that there is a comparison of

exact sequences

Hn+3 X/F

wH
n+2 F

uu

uu

w Hn+2 X
u=

Hn+3 Y

0,n+2
En+3

w Hn+2 X

dn+3

w Hn+2 X/F
w Hn+2 Y

w ...

uu
dn+2

w ...

Chasing elements shows that the map Hi (X/F ) Hi Y is an isomorphism for

i n + 2; it is an epimorphism for i = n + 3 because every element of Hn+3 Y is
transgressive. Comparing Puppe sequences gives a diagram

w Hn+3 (X S 1 )
u=

w ...

uu

w Hn+3 (F S 1 )
u f

w Hn+3 Y

w Hn+3 (Y /X)

w Hn+3 (X S 1 )

w ...

Hn+3 X
=u

w Hn+3 (X/F )

Hn+3 X

Then one chases some more elements to show that f : Hi (F S 1 ) Hi (Y /X)

is an epimorphism if i = n + 3 and is an isomorphism in all lower degrees. The
space Y /X is simply connected, so the homotopy fibre of f : F S 1 Y /X
must be (n + 2)-connected.

One has a right to ask why Theorem 3.11 should be called a relative Hurewicz
theorem. Heres the usual statement:
Corollary 3.12. Suppose that A is a simply connected subcomplex of X, and
that the pair (X, A) is n-connected for some n 1. Then the Hurewicz map
h

i (X, A)
i (X, A)
H
is an isomorphism if i = n + 1 and is surjective if i = n + 2.

4. The Ex functor

191

Proof: The relative homotopy group i (X, A) is the homotopy group i1 F

of the homotopy fibre F of the inclusion j : A , X, and the relative Hurewicz
map h is the composite
j

i (X/A) = H
i (X, A)
i (X, A) = i1 F
i (F S 1 ) i (X/A)
H
It is standard to say that (X, A) is n-connected and mean that the homotopy
fibre F is (n 1)-connected. The Freudenthal suspension theorem says that
the suspension map : i1 F i (F S 1 ) is an isomorphism if i = n + 1
and an epimorphism if i = n + 2 (for all n 2). Theorem 3.11 implies that
j : i (F S 1 ) i (X/A) is an isomorphism if i = n + 1 and an epi if i =
n + 2. Finally, the space X/A is n-connected so that the ordinary Hurewicz map
i (X/A) is an isomorphism if i = n + 1 and an epimorphism if
h : i (X/A) H
i = n + 2.

Rather a lot of standard homotopy theory is amenable to proof by simplicial

techniques. The reader may find it of particular interest to recast the HaussmanHusemoller treatment of acylic spaces and the Quillen plus construction [41] in
this setting. In order to achieve this, its helpful to know at the outset that
of a connected pointed Kan complex X is, generally, the
the universal cover X
homotopy fibre of the map X B(1 X) see as well the definition of covering

system in Section VI.3. This means, in particular, that the homotopy type of X
can be recovered from the pullback diagram

w E(1 X)

u
X

u
w B(1 X),

X is a principal 1 X-fibration.
so that the map : X
4. The Ex functor.
Kans Ex functor is a combinatorial construction which associates a Kan
complex Ex X to an arbitrary simplicial set X, up to natural weak equivalence.
It is constructed as an inductive limit of spaces Exn X, in such a way that the msimplices of Exn+1 X are a finite inverse limit of sets of simplices of Exn X. This
means in particular that this construction has very useful analogues in categories
other than simplicial sets. It remains interesting in its own right in the simplicial
set context, since it involves subdivision in a fundamental way.

192

III. Classical results and constructions

We give the details of this construction and establish its basic properties in
this section. It is one of the few remaining areas of simplicial homotopy theory in which the original combinatorial flavour of the subject (see the proof of
Lemma 4.7) has not been engulfed by the calculus of anodyne extensions.
Recall that the non-degenerate simplices of the standard n-simplex
n = hom ( , n)
are the monic ordinal number maps m , n. There is exactly one such monomorphism for each subset of n of cardinality m+1. It follows that the non-degenerate
simplices of n form a poset P n , ordered by the face relation, and this poset
is isomorphic to the non-empty subsets of the ordinal number n, ordered by
inclusion.
The poset has a nerve BP n . We shall write
sd n = BP n ,
and call it the subdivision of n .
Lemma 4.1. There is a homeomorphism

h : | sd n | |n |,
where h is the affine map which takes a vertex = {v0 , . . . , vk } of sd n to the
1
barycentre k+1
(v0 + + vk ) of the corresponding vertices.
In other words, | sd n | is the barycentric subdivision of |n |.
Proof: To see the co-ordinate transformation, take
0 v0 + 1 v1 + + n vn |n |,
and rewrite it as
t1 X1 + t2 X2 + . . . tr Xr ,
where 0 < t1 < t2 < < tr and Xi = vj0 + vj1 + + vjnj .
Write
r
X
Nj =
(nk + 1).
k=j

4. The Ex functor

193

Then
t1 X1 + + tr Xr
= t1 (X1 + + Xr ) + (t2 t1 )(X2 + + Xr ) + + (tr tr1 )Xr
1
1
= t1 N1 ( )(X1 + + Xr ) + (t2 t1 )N2 ( )(X2 + + Xr ) + . . .
N1
N2
1
+ (tr tr1 )Nr ( )Xr .
Nr
Note that
t1 N1 + (t2 t1 )N2 + + (tr tr1 )Nr = 1,
so that weve rewritten 0 v0 + +n vn as an affine sum of uniquely determined
barycentres.

n
Any function f : n m determines a map of posets f : P P m
via f (X) = f (X) = image of X under f . It follows that any poset morphism
: n m determines a poset map : P n P m , and hence induces a
simplicial set map : sd n sd m . This assignment is functorial, so we
obtain a cosimplicial object n 7 sd n in the category of simplicial sets.
The subdivision sd X of a simplicial set X is defined by
sd X =

lim
n

sd n ,

: X

where the colimit is indexed over the simplex category X for X.

The functor X 7 sd X is left adjoint to a functor Y 7 Ex Y , where the
simplicial set Ex Y is defined to have n-simplices given by the set of all simplicial
set maps sd n Y .
There is a natural map h : sd n n , called the last vertex map. It is
specified as a map of posets P n n by the assignment
[v0 , v1 , . . . , vk ] 7 vk ,
where [v0 , . . . , vk ] : k n is a non-degenerate simplex of n specified by i 7 vi .
There is also a map of posets g : n P n defined by g(i) = [0, 1, . . . , i]. Clearly
hg = 1 and there is a relation
[v0 , v1 , . . . , vk ] [0, 1, . . . , vk ]
in P n . It follows that the last vertex map h : sd n n is a simplicial
homotopy equivalence.
The natural maps h : sd n n induce a natural simplicial map
= h : Y Ex Y,
which is given on n-simplices by precomposition by the indicated map h.

194

III. Classical results and constructions

Lemma 4.2. The map : Y Ex Y is a 0 isomorphism, and induces a surjection on fundamental groupoids.
Proof: The map is an isomorphism on the vertex level. A 1-simplex of Ex Y
is a diagram

x
y
z
of 1-simplices of Y , and () is the diagram

s0 y

x
y y

for any 1-simplex x

y of Y (incidentally, the notation means that d0 () = y
and d1 () = x). Thus, two vertices of Ex Y are related by a string of 1-simplices
of Ex Y if and only if the corresponding vertices are related by a string on 1simplices of Y , so that is a 0 -isomorphism as claimed.
For each 1-simplex

x
y
z
of Ex Y , there is a 2-simplex of Ex Y

y
N
N

yN
y

N
y
N
Q
y4
y'
*
4
h
'
y
yh
]
[
4
4
6
'
[
h
u h
k
'
[
[
yu h
'
k
hh
'
)
'
'
[
hh
''' y
h
[
hh '
'
h
['''
wyu
z
x

Here, x is the 0th vertex v0 of , v1 = z and v2 is the top copy of y. The two lower
left 2-simplices of y are copies of s1 , the two lower right 2-simplices are copies

4. The Ex functor

195

of s1 , and the two upper 2-simplices are constant simplices of Y associated to

the vertex y. It follows that there is a relation

[()] = [x
y
z][()]
in the path category associated to Ex Y . But then

[x
y
z] = [()][()]1
in the fundamental groupoid (Ex Y ). Every generator of (Ex Y ) is therefore
in the image of the induced functor : (Y ) (Ex Y ).

Lemma 4.3. The map : Y Ex Y induces an isomorphism

: H (Y, Z) H (Ex Y, Z)
in integral homology.
Proof: The natural maps : k Ex k can be used to show that the Ex
functor preserves homotopies. It follows that Ex sd n is contractible; in effect,
the poset P n contracts onto the top non-degenerate simplex of n . It follows
that Ex sd n has the homology of a point.
The natural map : Y Ex Y induces an isomorphism 0 Y
= 0 Ex Y , by
Lemma 4.2. It follows that induces a natural isomorphism

: H0 (Y, Z) H0 (Ex Y, Z)
in the 0th integral homology group. The simplices : n Ex Y factor through
maps Ex : Ex sd n Ex Y . A standard acyclic models argument therefore
implies that there is a natural chain map : Z Ex Y ZY between the associated Moore complexes which induces the map 1 : H0 (Z Ex Y ) H0 (ZY ), and
any two natural chain maps which induce 1 are naturally chain homotopic.
Similarly, the composite natural chain map

Z Ex Y
ZY Z Ex Y
is naturally chain homotopic to the identity map, and the models Zn are used
to show that the composite

ZY Z Ex Y
ZY
is naturally chain homotopic to the identity. The map : ZY Z Ex Y is
therefore a natural chain homotopy equivalence, and so the map : Y Ex Y
is a homology isomorphism.

196

III. Classical results and constructions

Corollary 4.4. If the canonical map Y is a weak equivalence, then so is

the map Ex Y .
Proof: The fundamental groupoid Y for Y is trivial, and induces a surjection : Y Ex Y by Lemma 4.2, so that Ex Y has a trivial fundamental
groupoid as well. But Ex Y is acyclic by Lemma 4.3, so that Ex Y is weakly
equivalent to a point, by the Hurewicz Theorem (Theorem 3.7).

Lemma 4.5. The functor Ex preserves Kan fibrations.

Proof: We show that the induced map sd nk , sd n is a weak equivalence.
The simplicial set sd nk can be identified with the nerve of the poset of nondegenerate simplices of nk , and the homeomorphism

h : | sd n | |n |
restricts to a homeomorphism

| sd nk | |nk |.
It follows that sd nk , sd n is a weak equivalence.

Theorem 4.6. The natural map : Y Ex Y is a weak equivalence.

Proof: Let Yf be a fibrant model for Y in the sense that there is a weak
equivalence : Y Yf . Pick a base point y for Y , and let P Yf be path space
for Yf corresponding to the base point (y). Form the pullback diagram

1 P Yf

w P Yf

p
u
Y

p
u

w Yf ,

where p : P Yf Yf is the canonical fibration. Then is a weak equivalence

(Corollary II.9.6), so that 1 P Yf is weakly equivalent to a point, and the fibre
sequence
p

Yf 1 P Yf Y.

4. The Ex functor

197

gives rise to a comparison of fibre sequences

p

w 1 P Yf

Yf

wY

u
Ex Yf

u
w Ex

P Yf

Ex p

u
w Ex Y,

by Lemma 4.5. Computing in homotopy groups of realizations (and thereby

implicitly using the fact that the realization of a Kan fibration is a Serre fibration
Theorem I.10.10), we find a commutative diagram

1 (Y, y)

u
1 (Ex Y, y)

w 0 (Yf )

=
u
w 0 (Ex Yf ).

One uses Corollary 4.4 to see that the indicated boundary maps are isomorphisms. It follows that induces an isomorphism

: 1 (Y, y) 1 (Ex Y, y)
for all choices of base point y of Y . Inductively then, one shows that all maps

: i (Y, y) i (Ex Y, y)
are isomorphisms for all choices of base point y.

Lemma 4.7. For any map : nk Ex Y , the dotted arrow exists in the diagram
nk
y
u
n

w Ex Y

u
w Ex2 Y.

198

III. Classical results and constructions

Proof: The adjoint of the composite map

nk
Ex Y
Ex2 Y
is the composite
h

sd nk
nk
Ex Y.
It therefore suffices to show that the dotted arrow h exists in the diagram
sd nk
y

w nk

u
w Ex sd nk ,

u
sd n

making it commute, where : K Ex sd K is the counit of the adjunction

(sd, Ex), and is defined by sending a simplex : n K to the simplex of
Ex sd K given by the induced map sd : sd n sd K.
Let = (0 , . . . , q ) be a q-simplex of sd n , where the maps i : ni , n are
simplices of n . Define a function f : q n by the assignment

f (i) =

i (ni ) if i 6= dk n or n , and
k
if i = dk n or i = n .

The assignment 7 f is natural with respect to morphisms of ordinal numbers

: q0 q in the sense that f = f . There is a unique pair (X , )
consisting of a poset monic X : r , n and a surjective function : q r
such that the diagram of functions

wn
6
4

4X

<
4
r

commutes. Then a simplicial map h : sd n Ex sd nk is defined by the

assignment 7 (
(X )). Note that, whereas : q r is only a function,

5. The Kan suspension

199

it induces a poset map : P q P r and hence a simplicial map :

sd q sd r , so that the definition of the map h makes sense.

For any simplicial set X, define Ex X to be the colimit in the simplicial set
category of the string of maps
X

Ex2 X

Ex X

X Ex X Ex2 X Ex3 X . . .
Then the assignment X 7 Ex X defines a functor from the simplicial set
category to itself, which is commonly called the Ex functor . Write : X
Ex X for the canonical natural map which arises from definition of Ex X.
The results of this section imply the following:
Theorem 4.8.
(1) The canonical map : X Ex X is a weak equivalence, for any simplicial set X.
(2) For any X, the simplicial set Ex X is a Kan complex.
(3) The Ex functor preserves Kan fibrations.
Proof: The first statement is a consequence of Theorem 4.6. The second statement follows from Lemma 4.5. Statement (3) is implied by Lemma 4.7.

5. The Kan suspension.

The ordinal number map d0 : n n + 1 induces an inclusion d0 : n n+1 .
Let the vertex 0 be a base point for n+1 , and observe that any simplicial map
: n m uniquely extends to a simplicial map : n+1 m+1 , which
is pointed in the sense that (0) = 0, and such that d0 = d0 . Observe that
di = di+1 and sj = sj+1 for all i and j.
The cone CY of a simplicial set Y is the pointed simplicial set
CY = lim n+1 ,

n
Y

where the colimit is indexed in the simplex category of all simplices n Y ,

and is formed in the pointed simplicial set category. The maps d0 : n n+1
induce a natural map j : Y CY .
The cone CY on a simplicial set Y is contractible: the contracting homotopies
h : n+1 1 n+1 given by the transformations 0 i in the ordinal numbers
n + 1 glue together along the simplices of Y to give a contracting homotopy

200

III. Classical results and constructions

h : CY 1 CY onto the base point of Y , since there are commutative

diagrams
h

n+1 1
1

w n+1

m+1 1

w m+1

for any ordinal number map : n m.

Its now obvious that a simplicial set map f : Y X can be extended to a
map g : CY X in the sense that there is a commutative diagram
CY
u 4
j
Y

4g
44
6
f

wX

if and only if for each n-simplex x of Y there is an (n + 1)-simplex g(x) of X

such that
(1) d0 (g(x)) = f (x),
(2) d1 d2 dn+1 (g(x)) is some fixed vertex v of X for all simplices x of Y ,
and
(3) for all i, j 0 and all simplices x of Y , we have
di+1 (g(x)) = g(di x)

and sj+1 (g(x)) = g(sj x).

For the moment, given a simplicial set X, let X1 = 0 X and write d0 : X0

X1 for the canonical map X0 0 X.
A simplicial set X is said to have an extra degeneracy if there are functions
s1 : Xn Xn+1 for all n 1, such that, in all degrees,
(1) d0 s1 is the identity on Xn ,
(2) for all i, j 0, we have the identities
di+1 s1 (x) = s1 di (x)

and sj+1 s1 (x) = s1 sj (x).

5. The Kan suspension

201

Lemma 5.1. Suppose that a simplicial set X has an extra degeneracy. Then the
canonical map X K(0 X, 0) is a homotopy equivalence.
Proof: It suffices to assume that X is connected. Then the association x 7
s1 (x) determines an extension
CX
u 4

4s1
44
6

j
X

wX

of the identity map on X, according to the criteria given above. Also, CX

contracts onto its base point, which point maps to s1 () in X, where denotes
the unique element of 0 X.

Example 5.2. Suppose that G is a group. The translation category EG associated to the G-action G G G has the elements of G for objects, and has
morphisms of the form h : g hg. The nerve of this category is commonly also
denoted by EG. Note in particular that an n-simplex of the resulting simplicial
set EG has the form
g1

g2

gn

g0 g1 g0 . . . gn . . . g0 ,
and may therefore be identified with a string (g0 , g1 , . . . , gn ) of elements of the
group G. The simplicial set EG is plainly connected. It is also easily seen (by
thinking in terms of strings of arrows) that the assignment
(g0 , g1 , . . . , gn ) 7 (e, g0 , g1 , . . . , gn )
defines an extra degeneracy s1 : EGn EGn+1 for EG, so that EG is contractible.
If K is a pointed simplicial set, then the pointed cone C K is defined by the
pushout
C0
u

w CK
u
w C K.

202

III. Classical results and constructions

Here, the map C is induced by the inclusion of the base point : 0 K in

K.
The maps d0 : n n+1 induce a natural pointed map i : K C K. The
Kan suspension K of K is defined to be the quotient
K = C K/K.
The Kan suspension K is a reduced simplicial set, and is a concrete model
for the suspension of the associated pointed space |K|, in the sense that the realization |K| is naturally homeomorphic to the topological suspension of |K|.
The existence of this homeomorphism is one of the reasons that the Kan suspension functor : S S preserves weak equivalences one could also argue
directly from the definitions by using the cofibre sequence
K C K K
and the contracting homotopy on the cone C K.
A pointed simplicial set map : K Y consists of pointed functions
(5.3)

n : Kn Yn+1

such that
(1) d1 . . . dn+1 n (x) = , and d0 n (x) = for each x Kn , and
(2) for each ordinal number map : n m, the diagram of pointed functions

Kn

u
Km

w Yn+1
u

()

w Ym+1

commutes.
Pointed simplicial set maps of the form : C K Y have a very similar
characterization; one simply deletes the requirement that d0 n (x) = . It follows
that the pointed cone and Kan suspension functors preserve colimits of pointed
simplicial sets.

5. The Kan suspension

203

An equivalent description of C K starts from the observation that the pointed

simplicial set K is a member of a coequalizer diagram of the form
_

Km n+

Kn n+ K

n0

:nm

where, for example, Km n+ is the wedge of the pointed set of m-simplices Km ,

thought of as a discrete pointed simplicial set, with n+ , and n+ is notation
for the simplicial set n t {}, pointed by the disjoint vertex . Then C K is
defined by the coequalizer diagram
_

Km n+1

Kn n+1 C K.

n0

:nm

The set of m-simplices of n+1 is the set of ordinal number maps of the form
: m n + 1. Each such fits into a pullback diagram

wn

(d0 )nj

u
m

d0

u
w n+1

in the ordinal number category, for some uniquely determined map : j n if

1 (n) 6= . It follows that, as a pointed set, n+1
has the form
m
n+1
= nm t nm1 t t n0 t {}
m
= (n+ )m (n+ )m1 (n+ )0 ,
where the base point corresponds to the case (m) = 0. Now take another map
: k m; there is a pullback diagram of ordinal number maps

1 (j)
(5.4)
u
k

wj
(d0 )mj

u
wm

204

III. Classical results and constructions

in the case where r = 1 (j) 6= . It follows that the restriction of : n+1

m
n+1
n
k to the summand (+ )j is the map
: (n+ )j (n+ )r
if 1 (j) 6= , and is the map to the base point otherwise.
Suppose that K is a pointed simplicial set. There is a a pointed simplicial set

C (K) whose set of n-simplices is given by

C (K)n = Kn Kn1 K0 .
The map : C (K)m C (K)k associated to : k m is given on the
summand Kj by the composite

Kj Kr , C (Kk )
in the case where 1 (j) 6= , r = 1 (j), and the map is defined by the
diagram (5.4). If 1 (j) = , then the restriction of to Kj maps to the base
point.
One checks that C (K) is indeed a pointed simplicial set, and that the construction is functorial in K. Furthermore, the functor preserves colimits, so that
the diagram
_

C (Km n+ )

:nm

C (Kn n+ ) C K.

n0

is a coequalizer. On the other hand, the definitions imply that there are isomorphisms
C (Y n+ )
= Y C (n+ )

= Y n+1 ,
which are natural in the pointed sets Y and simplices n . This is enough to
prove
Lemma 5.5. There is a pointed simplicial set isomorphism
C (K)
= C (K),
which is natural in K.

5. The Kan suspension

205

The composite
i

K
C (K)
= C (K),
in degree n is the inclusion of the wedge summand Kn . The canonical map
i : K C (K) is therefore an inclusion. Collapsing by K in each degree also
gives a nice description of the Kan suspension K: the set of (n + 1)-simplices
of K is given by the wedge sum
Kn+1 = Kn Kn1 K0 .
Its also a worthy exercise to show that the maps n : Kn Kn+1 corresponding to the identity map K K under the association (5.3) are inclusions
Kn Kn Kn1 K0 of wedge summands.

Chapter IV Bisimplicial sets

This chapter is a basic exposition of the homotopy theory of bisimplicial sets
and bisimplicial abelian groups.
A bisimplicial set can be viewed either as a simplicial object in the category of
simplicial sets or a contravariant functor on the product of two copies of
ordinal number category : both points of view are constantly exploited. Similar
considerations apply to bisimplicial objects in any category, and to bisimplicial
abelian groups in particular.
Categories of bisimplicial objects are ubiquitous sources of spectral sequence
constructions. In many contexts, bisimplicial sets and bisimplicial abelian groups
function as analogs of projective resolutions for homotopy theoretic objects. The
Serre spectral sequence is one of the original examples: pullbacks over simplices
of the base of a map p : E B form a bisimplicial resolution of the total space.
Then every bisimplicial set has canonically associated bisimplicial abelian groups
and hence bicomplexes, and so a spectral sequence (5.3) drops out. If the map
p happens to be a fibration, the resulting spectral sequence is the Serre spectral
sequence (5.5). It is not much of a conceptual leap from this construction to the
notion of a homology fibre sequence, which is the basis for the group completion
theorem (Theorem 5.15), but then these ideas are essentially non-abelian, so
that the theory can be pushed to give the basic detection principle for homotopy
cartesian diagrams (Lemma 5.7) that is the basis of proof for Quillens Theorem
B (Theorem 5.6). Group completion and Theorem B are fundamental tools for
algebraic K-theory. This collection of results appears in Section 5.
We begin in Section 1 with some definitions and examples of bisimplicial sets
and abelian groups, which examples include all homotopy colimits. Section 2
contains a discussion of the basic features of bisimplicial abelian groups, including homotopy colimit objects and the generalized Eilenberg-Zilber theorem
(Theorem 2.5). This theorem asserts that the two standard ways of extracting
a chain complex from a bisimplicial abelian group, namely the chain complex
associated to the diagonal simplicial abelian group and the total complex, are
naturally chain homotopy equivalent.
A description of the formal homotopy theory of bisimplicial sets is given in
Section 3. This homotopy theory is a little complicated, because there are closed
model structures associated to multiple definitions of weak equivalence for these
objects. The diagonal functor creates an external notion one says that a
map of bisimplicial sets is a diagonal weak equivalence if it induces a weak
equivalence of associated diagonal simplicial sets. There are also two internal
descriptions of weak equivalence, corresponding to viewing a bisimplicial set as
a diagram in its vertical or horizontal simplicial sets. In particular, we say that
206

1. Bisimplicial sets: first properties

207

a bisimplicial set map is a pointwise (or vertical) weak equivalence if each ofbisimplicial set
the induced maps of vertical simplicial sets is a weak equivalence. We discussexponential law
closed model structures associated to all of these definitions. Diagonal weakbisimplex
equivalences are the objects of study in the Moerdijk structure, whereas pointwise
weak equivalences figure into two different structures, namely the BousfieldKan structure in which the fibrations are also defined pointwise, and the Reedy
structure where the cofibrations are defined pointwise. All of these theories are
useful, and they are used jointly in the applications that follow, but this is by no
means the end of the story: there is a further notion of E2 -weak equivalence due
to Dwyer, Kan and Stover [28], [29] and a corresponding closed model structure
that is not discussed here.
We confine ourselves here to applications of the homotopy theory of bisimplicial sets that involve detection of cartesian squares of bisimplicial set morphisms
that become homotopy cartesian after applying the diagonal functor. There are
two extant non-trivial techniques. One of these is the circle of ideas related
to the Serre spectral sequence in Section 5, which has already been discussed.
The other is dealt with in Section 4, and arises from the Reedy closed model
structure, in the presence of the -Kan condition (Theorem 4.9 of Bousfield and
Friedlander). The -Kan condition is satisfied widely in nature, in particular
for all pointwise connected bisimplicial sets; it is best expressed by saying that
the canonical maps from the homotopy group objects fibred over the simplicial
set of vertical vertices of a bisimplicial set to the vertices are Kan fibrations. The
Bousfield-Friedlander theorem leads to a spectral sequence (see (4.14)) for the
homotopy groups of the diagonal of a pointwise connected bisimplicial set. This
spectral sequence is the origin of the definition of E2 -weak equivalence that is
referred to above.
1. Bisimplicial sets: first properties.
A bisimplicial set X is a simplicial object in the category of simplicial sets, or
equivalently a functor X : op S where is the ordinal number category and
S denotes the category of simplicial sets as before. Write S 2 for the category of
bisimplicial sets.
A bisimplicial set X can also be viewed as a functor
X : op op S,
(or as a contravariant functor on the category ), by the categorical exponential law. From this point of view, the data for X consists of sets X(m, n)
with appropriately defined functions between them. The set X(m, n) will often
be called the set of bisimplices of X of bidegree (m, n), or the (m, n)-bisimplices
of X. We shall also say that a bisimplex x X(m, n) has horizontal degree m
and vertical degree n.

208

IV. Bisimplicial sets

Example 1.1. Any simplicial set-valued functor Z : I S gives rise to ahomotopy colimit
category, translation
bisimplicial set with (m, n)-bisimplices
classifying bisimple
G
(1.2)
Z(i0 )n .
i0 i1 im

Note that the indexing is over simplices of degree m in the nerve BI of the
category I, or equivalently over strings of arrows of length m in I.
This bisimplicial set is often called the homotopy colimit of the functor Z, and
is denoted by holim Z.
I
The easiest way to see that holim Z is in fact a bisimplicial set is to observe
I
that each of the n-simplex functors Zn gives rise to a translation category EZn
having objects (i, x) with i an object of I and x Zn (i), and with morphisms
: (i, x) (j, y) where : i j is a morphism of I such that Zn ()(x) = y.
Then the set of m-simplices of the nerve BEZn is the set displayed in (1.2).
Furthermore, the data is simplicial in n, so the simplicial object BEZ = holim Z
I
is a bisimplicial set.
Example 1.3. The bisimplicial set k,l is the simplicial object in S which is
composed of the simplicial sets
G
l ,

n
k

where l is the standard l-simplex in S and the disjoint union is indexed over
morphisms : n k in the ordinal number category . The bisimplicial set
k,l classifies bisimplices of bidegree (k, l) in the sense that there is a simplex
G
k,l
l ,

k
k

which is a classifying (k, l)-simplex in the sense that the bisimplices x X(k, l)
in a bisimplicial set X are in one to one correspondence with maps x : k,l X
such that (k,l ) = x. Specifically, the classifying bisimplex k,l is the copy of the
classifying l-simplex l ll in the summand corresponding to the identity map
1 : k k.
It follows that k,l is the contravariant functor on which is represented
by the object (k, l).
Example 1.4. Suppose that K and L are simplicial sets. Then there is a
with (m, n)-bisimplices specified by
bisimplicial set K L

K L(m,
n) = Km Ln .

1. Bisimplicial sets: first properties

209

will be called the external product of K with L. Noteexternal product

The bisimplicial set K L
that the bisimplicial set k,l may be alternatively described as the externaldiagonal simplicial
vertical, simplicial s
l.
product k
The diagonal simplicial set d(X) of a bisimplicial set X has n-simplices given
by
d(X)n = X(n, n).
It can also be viewed as the composite functor

op op op S,
where is the diagonal functor.
Think of the bisimplicial set X as a simplicial object in the simplicial set
category by defining the vertical simplicial sets Xn = X(n, ). Any morphism
: m n gives rise to a diagram
Xn m
(1.5)

w X n n

u
Xm m .

The collection of all such maps therefore determines a pair of maps

G
G
Xn m
Xn n ,
n

m
n

by letting the restriction of the displayed maps to the summand corresponding

to be 1 and 1, respectively. There are also simplicial set maps
n : Xn n d(X)
defined on r-simplices by

n (x, r
n) = (x).
Here, x is an r-simplex of Xn , and : Xn Xr is a simplicial structure map
of X, so that (x) is an r-simplex of X(r, ) and is therefore an r-simplex of
d(X). The collection of the maps n together determine a simplicial set map
G
:
Xn n d(X).
n

210

IV. Bisimplicial sets

Exercise 1.6. Show that the resulting diagram

G
G

Xn m
Xn n
d(X)
n

m
n

is a coequalizer in the category of simplicial sets.

The exercise implies that the diagonal simplicial set d(X) is a coend in the
category of simplicial sets for the data given by all diagrams of the form (1.5).
The diagonal simplicial set d(X) has a natural filtration by subobjects d(X)(p) ,
p 0, where
G
d(X)(p) = image of (
Xn n ) in d(X).
0np

The degenerate part (with respect to the horizontal simplicial structure) of Xp+1
is filtered by subobjects
[
si (Xp ) Xp+1 .
s[r] Xp =
0ir

It follows from the simplicial identities that there are pushout diagrams
sr+1

s[r] Xp1
y

w s[r] Xp
y

(1.7)
u
Xp

u
sr+1

w s[r+1] Xp ,

and
(s[p] Xp p+1 ) (Xp+1 p+1 )
y

w d(X)(p)
y

(1.8)
u
Xp+1 p+1

u
(p+1)

w d(X)

Diagrams (1.7) and (1.8) and the glueing lemma (Lemma II.9.8) are the basis
of an inductive argument leading to the proof of

2. Bisimplicial abelian groups

211

Proposition 1.9. Suppose that f : X Y is a map of bisimplicial sets which

is a pointwise weak equivalence in the sense that all of the maps f : Xn Yn are
weak equivalences of simplicial sets. Then the induced map f : d(X) d(Y )
of associated diagonal simplicial sets is a weak equivalence.
2. Bisimplicial abelian groups.
We collect here the basic facts about bisimplicial abelian groups, in two subsections.
The first of these is effectively about homotopy colimit constructions in the
category of simplicial abelian groups. Any functor A : I sAb taking values in
simplicial abelian groups has an associated bisimplicial abelian group which we
call translation object, and is formed by analogy with the homotopy colimit of a
diagram of simplicial sets. We derive the basic technical result that translation
object for the diagram A is weakly equivalent to the simplicial abelian group
A(t) if the index category I has a terminal object t.
When we say that a bisimplicial abelian group B is weakly equivalent to a
simplicial abelian group C, we mean that the diagonal simplicial abelian group
d(B) is weakly equivalent to C within the simplicial abelian group category. One
could alternatively interpret B as a bicomplex and C as a chain complex, and
then ask for a weak equivalence between the chain complexes Tot(B) and C. The
generalized Eilenberg-Zilber theorem says that Tot(B) and the chain complex
d(B) are naturally chain homotopy equivalent, so in fact there is no distinction
between the two approaches to defining such weak equivalences. This is the
subject of the second subsection. We further show that the standard spectral
sequence
E2p,q = Hp (Hq A ) Hp+q (d(A))
for a bisimplicial abelian group A can be derived by methods which are completely internal to the simplicial abelian group category.
2.1. The translation object.
This section contains technical results concerning a simplicial abelian groupvalued analogue of the homotopy colimit construction for diagrams of abelian
groups, called the translation object. This construction is of fundamental importance for the description of the Serre spectral sequence that appears in a
subsequent section of this chapter. More generally, it appears canonically in any
homology spectral sequence arising from a homotopy colimit of a diagram of
simplicial sets.
Supppose that A : I Ab is an abelian group valued functor, where I is a

212

IV. Bisimplicial sets

small category. There is a simplicial abelian group EA, with

EAn =

translation object

A(0) ,

:nI

where the direct sum is indexed by the n-simplices of the nerve BI of the index
category I. The abelian group homomorphism : EAn EAm induced by an
ordinal number map : m n is specified by requiring that all diagrams

A(0)

w A(0)

in

M u

A(0)

:nI

M u
w

in
A(0)

:mI

commute, where the homomorphism A(0) A(0) is induced by the relation

0 (0) in the ordinal number n. The simplicial abelian group EA is called the
translation object associated to the functor A.
Note that EA is not the nerve of a category, even though its definition is
analogous to that of the nerve of a translation category for a set-valued functor.
Suppose given functors
F

J I Ab,
where J and I are small categories. The functor F induces a simplicial abelian
group homomorphism F : E(AF ) EA, which is defined on n-simplices in
such a way that the diagram

AF(0)
in
M u
AF(0)
:nJ

commutes.

w AF (0)
M u
:nI

inF
A(0)

2. Bisimplicial abelian groups

213

Any natural transformation : A B of functors I Ab determines a

morphism : EA EB of simplicial abelian groups. This morphism is
defined on n-simplices by the requirement that the following diagram commutes:

A(0)
in
M u

A(0)

:nI

w B(0)

in
u
M
w
B(0) .
:nI

Now consider a functor B : I 1 Ab, and let d1 : I I 1 and d0 :

I I 1 be defined, respectively, by d1 (i) = (i, 0) and d0 (i) = (i, 1). The
maps (i, 0) (i, 1) in I 1 induce a natural transformation : Bd1 Bd0 of
functors I Ab, and hence induce a simplicial abelian group homomorphism
: EBd1 EBd0 . In general, the group of n-simplices of the simplicial abelian
group EA 1 can be identifed with the direct sum
M

A(0) .

(,):nI1

In the case at hand, one finds a canonical map h : EBd1 1 EB, which is
defined on n-simplices by the requirement that the diagram

B((0),0)
in(,)
M

w B((0),(0))

(,):nI1

B((0),0)

in(,)
B((0),(0))

(,):nI1

commutes, where the top horizontal map is induced by the morphism ((0), 0)
((0), (0)) in I 1. One can now check that there is a commutative diagram

214

IV. Bisimplicial sets

of simplicial abelian group homomorphisms

EBd1 [[
d
(2.1)

[[[ d1

[[
[[

[[
[
]

EBd1u 1

w EB.

h
hj
hd0

d0
EBd

h
1

w EBd

Lemma 2.2. Let A : I Ab be an abelian group valued functor on a small

category I, and suppose that I has a terminal object t. Then there is a canonical weak equivalence EA K(At , 0), which is specified on n-simplices by the
homomorphism
M
A(0) At
:nI

given on the summand corresponding to : n I by the map A(0) At

induced by the unique morphism (0) t of the index category I.
Proof: Let t also denote the composite functor
I {t} I.
Then the discrete category {t} is a strong deformation retract of the category I
in the sense that there is a commutative diagram of functors

u
H w I.
I u 1
P
NN
0
N
d
N
N t
1

2. Bisimplicial abelian groups

215

Now, EAHd1 = EA, so that (2.1) can be used to show there is a commutativebisimplicial abelian
diagram of simplicial abelian group homomorphisms
EA [[
1

u
1
EA
u

[[[
[1[[
[[

[
]
[

w EAH

w EA.
H [
[
]
[
[ t
[

d0
EA

w EAt

Note that t is the identity map on the simplicial abelian group EAt, so that
the Moore complex of EA is chain homotopy equivalent to the Moore complex of
EAt. The map EA K(At , 0) factors as the weak equivalence : EA EAt ,
followed by the map EAt K(At , 0), which map is defined on n-simplices by
the codiagonal map
M

At At .
:nI

This last homomorphism is a weak equivalence, because the space BI is contractible.

Remark 2.3. The argument for Lemma 2.2 fails in the case where the index
category has an initial object.
2.2. The generalized Eilenberg-Zilber theorem.
A bisimplicial abelian group A is a simplicial object in the category of simplicial
abelian groups, or equivalently a functor of the form
A : op op Ab,
where Ab denotes the category of abelian groups, as before. Subject to the
latter description, the simplicial abelian group A(n, ) will be referred as the
vertical simplicial abelian group in horizontal degree n associated to A, while
the object A(, m) is the horizontal simplicial abelian group in vertical degree
m. The category of bisimplicial abelian groups and natural transformations
between them will be denoted by s2 Ab.
It is often convenient to write An = A(n, ) for the vertical simplicial abelian
group in horizontal degree n. The simplicial abelian group morphism An Am

216

IV. Bisimplicial sets

associated to an ordinal number map : m n will sometimes be denoted byMoore bicomplex

h : this morphism is given on k-simplices by the abelian group homomorphismweak equivalence, p
(, 1) : A(n, k) A(m, k). We shall also occasionally write v = (1, ) :generalized Eilenbe
A(n, k) A(n, p) for the vertical structure maps associated to : p k.
The diagonal simplicial abelian group d(A) associated to the bisimplicial abelian group A has n-simplices given by d(A)n = A(n, n). The association A 7
d(A) defines a functor d : s2 Ab sAb. The diagonal functor is plainly exact.
Furthermore, if B is a simplicial abelian group and K(B, 0) is the associated
horizontally (or vertically) constant bisimplicial abelian group, then there is a
natural isomorphism d(K(B, 0))
= B.
The Moore bicomplex for a bisimplicial abelian group A is the bicomplex having
(p, q)-chains A(p, q), horizontal boundary
h =

p
X

(1)i di : A(p, q) A(p 1, q),

i=0

and vertical boundary

v =

q
X

(1)p+j dj : A(p, q) A(p, q 1).

j=0

We shall also write A for the Moore bicomplex of a bisimplicial abelian group
A, and Tot A will denote the associated total complex. Write also An = A(n, )
for the Moore complex in horizontal degree n. Then filtering the bicomplex A
in the horizontal direction gives a spectral sequence
(2.4)

E2p,q = Hp (Hq A ) Hp+q (Tot A).

It follows that, if the bisimplicial abelian group map f : A B is a pointwise

weak equivalence in the sense that all of the maps f : An Bn , n 0, of vertical
simplicial abelian group are weak equivalences, then f induces an homology
isomorphism f : Tot A Tot B of the associated total complexes. Of course,
the meaning of vertical and horizontal are in the eyes of the beholder, so it
follows immediately that any map of bisimplicial abelian groups which consists
of weak equivalences on the horizontal simplicial abelian group level must again
induce a homology isomorphism of total complexes. One can, alternatively,
make an argument with the second spectral sequence for the homology of Tot A
(constructed by filtering in the Moore bicomplex in the vertical direction).
The generalized Eilenberg-Zilber theorem of Dold and Puppe [20] asserts the
following:

2. Bisimplicial abelian groups

217

Theorem 2.5. The chain complexes d(A) and Tot A are chain homotopy equivalent. This equivalence is natural with respect to morphisms of bisimplicial
abelian groups A.
Proof: Suppose that K and L are simplicial sets. The usual Eilenberg-Zilber
theorem asserts that there are natural chain maps
f : Z(K L) Tot(ZK ZL),
and
g : Tot(ZK ZL) Z(K L),
and that there are natural chain homotopies f g ' 1 and gf ' 1. Specializing K
and L to the standard simplices gives bicosimplicial chain maps
f : Z(p q ) Tot(Zp Zq ),
and

g : Tot(Zp Zq ) Z(p q ),

as well as bicosimplicial chain homotopies f g ' 1 and gf ' 1.

Observe that p q = d(p,q ), where p,q is the bisimplicial set represented
by the pair of ordinal numbers (p, q). It follows that the chain complex Z(p
q ) can be identified up to natural isomorphism with d(Zp,q ). Note as well
that, up to natural isomorphism, Tot(Zp Zq ) is the total complex of the
bisimplicial abelian group Zp,q . Every bisimplicial abelian group A sits in a
functorial exact sequence
M
M
Zr,s
Zp,q A 0.
(r,s)(p,q)

(p,q)

The functors Tot and d are both right exact and preserve direct sums, so the
chain maps
f : d(Zp,q ) Tot Zp,q
uniquely extend to a natural chain map
f : d(A) Tot A.
which is natural in bisimplicial abelian groups A. Similarly, the chain maps
g : Tot Zp,q d(Zp,q )
induce a natural chain map
g : Tot A d(A).
The same argument implies that the bicosimplicial chain homotopies f g ' 1 and
gf ' 1 extend uniquely to chain homotopies which are natural in bisimplicial
abelian groups.

218

IV. Bisimplicial sets

Remark 2.6. The maps f and g in the proof of Theorem 2.5 can be preciselyAlexander-Whitney
specified as the unique extensions of the classical Alexander-Whitneys and shuffleshuffle map
maps, respectively. The definitions will not be written down here (see [20], [64,
pp.241243]).
The underlying acyclic models argument for the Eilenberg-Zilber theorem is
somewhat less than conceptual, so that the usual approach of using the spectral
sequence (2.4) and the generalized Eilenberg-Zilber theorem to construct the
standard spectral sequence
E2p,q = p (q A ) p+q d(A)
is rather indirect. We can now give an alternative construction. The basic trick
is to arrange for some independent means of showing the following:
Lemma 2.7. A pointwise weak equivalence f : A B of bisimplicial abelian
groups induces a weak equivalence f : d(A) d(B) of the associated diagonal
complexes.
Proof: There is a bisimplicial abelian group given in vertical degree m by the
simplicial abelian group
M
A(nk , m).
mn0 nk

This simplicial abelian group is the translation object associated to the functor
A(, m) : (m )op Ab which is defined by associating to the object m n
the abelian group A(n, m). The category (m )op has a terminal object,
namely the identity map 1 : m m, so Lemma 2.2 implies that the canonical
simplicial abelian group map EA(, m) K(A(m, m), 0) is a weak equivalence.
It follows that the Moore complex for EA(, m) is canonically weakly equivalent
to the chain complex A(m, m)[0] consisting of the group A(m, m) concentrated
in degree 0. The morphism of bicomplexes which is achieved by letting m vary
therefore induces a natural weak equivalence of chain complexes Tot EA(, )
d(A). On the other hand, the vertical simplicial abelian group of EA(, ) in
horizontal degree k has the form
M
n0 A(nk , ).
n0 nk

It follows that any pointwise equivalence f : A B induces a homology isomorphism f : Tot EA(, ) Tot EB(, ), and hence a weak equivalence
f : d(A) d(B).

2. Bisimplicial abelian groups

219

Define the horizontal normalization Nh A of a bisimplicial abelian group Ahorizontal normaliz

to be the simplicial chain complex whose n-simplices are given by the chain
complex Nh An = N A(, n). The bisimplicial abelian group A can be recovered
from the simplicial chain complex Nh A by applying the functor in all vertical degrees. The simplicial chain complex Nh A can be filtered: one defines a
simplicial chain complex Fp Nh A as a chain complex object by specifying

Fp Nh Ai =

Nh Ai
0

if i p, and
if i > p.

Now, Fp1 Nh A Fp Nh A, with quotient Nh Ap [p], which can be thought of

as the simplicial chain complex which is the simplicial abelian group Nh Ap in
chain degree p and is 0 in other chain degrees. Applying the functor to the
corresponding short exact sequence gives a short exact sequence of bisimplicial
abelian groups
0 Fp1 Nh A Fp Nh A Nh Ap [p] 0.
It follows that the bisimplicial abelian groups Fp Nh A filter A. In vertical degree
n, we have an identification
Nh Ap [p](, n) = K(N A(, n)p , p),
since, in general, the Eilenberg-Mac Lane space K(B, n) can be identified with
the simplicial abelian group B[n] which arises by applying the functor to the
chain complex B[n] which consists of B concentrated in degree n.
Lemma 2.8. There is an isomorphism
n Nh Ap
= N (n A )p .
Proof: Write

Nhj Ap = ji=0 ker(dhi ) Ap

for 0 j p 1. We show that the canonical map

n Nhj Ap N j (n A )p

(2.9)

is an isomorphism for all j.

The degree 0 case is shown by observing that there is a short exact sequence
0

Nh0 Ap

dh
0

Ap Ap1 0

220

IV. Bisimplicial sets

which is split by sh0 : Ap1 Ap , so that the map n Nh0 Ap N 0 (n A )p is an

isomorphism. Furthermore, the induced map n Nh0 Ap n Ap is a monomorphism.
Assume that the map (2.9) is an isomorphism, and that the induced map
n Njh Ap n Ap
is monic. Consider the pullback diagram of simplicial abelian groups
Nhj+1 Ap
y

w ker(dhj+1 )

u
w Ap ,

Nhj Ap
and form the pushout
Nhj+1 Ap
y

w ker(dhj+1 )

u
Nhj Ap
w

Nhj Ap

u
+ ker(dhj+1 ).

The inclusion map Nhj+1 Ap Nhj Ap is split by the map x 7 x shj+1 dhj+1 (x),
and so a comparison of long exact sequences shows that the induced diagram of
abelian group homomorphisms
n Nhj+1 Ap
y
u
n Nhj Ap
w

w n ker(dhj+1 )

n (Nhj Ap

u
+ ker(dhj+1 ))

is a pushout. But the inclusion ker(dhj+1 ) Ap is a split monomorphism, so that

the induced map n ker(dhj+1 ) n Ap is monic, and the map n Nhj Ap n Ap
is monic by the inductive assumption. It follows that the induced map
n (Nhj Ap + ker(dhj+1 )) n Ap

2. Bisimplicial abelian groups

221

is a monomorphism, and that the canonical sequence

0 n Nhj+1 Ap n Nhj Ap n ker(dhj+1 ) n Ap
is exact.
The alternative method for constructing a spectral sequence
(2.10)

E2p,q = p (q A ) p+q d(A)

for a bisimplicial abelian group A is now clear. The filtration Fp Nh A for the
bisimplicial abelian group A gives rise to short exact sequences
0 Fp1 Nh A Fp Nh A Nh Ap [p] 0
Let Nh Ap hpi be simplicial chain complex
p

p1

0 Nh Ap [p]
Nh Ap [p] 0
which is non-trivial in horizontal degrees p and p 1 only. Then Nh Ap hpi
is horizontally acyclic, and there is a short exact sequence of simplicial chain
complexes
0 Nh Ap [p] Nh Ap hpi Nh Ap [p 1] 0.
It therefore follows from Lemma 2.8 that there are natural isomorphisms
(2.11)

p+q Nh Ap [p]
= q Nh Ap
= N (q A )p .

Furthermore, if we define a spectral sequence by letting

E1p,q = p+q Nh Ap [p],
then the differential d1 : E1p,q E1p1,q is identified by the isomorphisms (2.11)
with the standard differential
: N (q A )p N (q A )p1
in the normalized complex for the simplicial abelian group q A . It follows that
there are canonical isomorphisms
E2p,q
= p (q A ),
and the construction of the spectral sequence (2.10) is complete.

222

IV. Bisimplicial sets

3. Closed model structures for bisimplicial sets.

closed model struct
2
There are three closed model structures on the category S of bisimplicial setsweak equivalence, p
fibration, pointwise
that will be discussed in this section, namely
(1) the Bousfield-Kan structure, in which a fibration is a pointwise fibration
and a weak equivalence is a pointwise weak equivalence,
(2) the Reedy structure, in which a cofibration is a pointwise cofibration (aka.
inclusion) of bisimplicial sets, and a weak equivalence is a pointwise weak
equivalence, and
(3) the Moerdijk structure, in which a fibration (respectively weak equivalence) is a map f which induces a fibration (respectively weak equivalence)
d(f ) of associated diagonal simplicial sets.
In all of the above, a map f : X Y is said to be a pointwise weak equivalence
(respectively fibration, cofibration) if each of the simplicial set maps f : Xn Yn
is a weak equivalence (respectively fibration, cofibration). This idea was partially
introduced at the end of Section 1.
3.1. The Bousfield-Kan structure.
The Bousfield-Kan closed model structure on the category S2 is a special case
of a closed model structure introduced by Bousfield and Kan for the category
SI of I-diagrams of simplicial sets arising from a small category I and natural
transformations between such. The special case in question corresponds to letting
I be the opposite category op of the ordinal number category.
A map f : X Y is defined to be a weak equivalence (respectively fibration) in
the Bousfield-Kan structure on SI if each induced simplicial set map f : X(i)
Y (i) (in sections) is a weak equivalence (respectively fibration). One says that
such maps are pointwise weak equivalences (respectively pointwise fibrations).
Cofibrations are defined by the left lifting property with respect to pointwise
trivial fibrations, suitably defined.
The closed model axioms for the Bousfield-Kan structure on SI are verified in
Example II.7.11. They can also be seen directly by using a small object argument
based on the observation that a map f : X Y of SI is a pointwise fibration if
it has the right lifting property with respect to all maps
Fi nk Fi n ,

i I,

and f is a pointwise trivial fibration if it has the right lifting property with
respect to all induced maps
Fi n Fi n ,

i I.

3. Closed model structures for bisimplicial sets

223

Here, we need to know that K 7 Fi K is the left adjoint of the i-sections functorclosed model struct
n-truncated simplic
X 7 X(i), and explicitly
G
Fi K(j) =
K
ij

defines the functor Fi K at j I.

3.2. The Reedy structure.
First, a word or two about skeleta and coskeleta of simplicial sets the
skeleton construction for more general simplicial objects will be discussed in
Section V.1.
Write n for the full subcategory of the ordinal numbers on the objects
0, . . . , n. An n-truncated simplicial set Y is a functor Y : op
n Sets. Let
Sn denote the category of n-truncated simplicial sets.
Every simplicial set X : op Sets gives rise to an n-truncated simplicial set
in X by composition with the inclusion functor in : n . The n-truncation
functor X 7 in X has a left adjoint Y 7 in Y and a right adjoint Z 7 i!n Z, and
these adjoints are defined by left and right Kan extension respectively. Explicitly,
the set in Ym of m-simplices of in Y is defined by
in Ym =

lim

Yk ,

lim

Zk ,

mkn

while

i!n Zm =

nkm

where the indicated morphisms in both cases are in the ordinal number category
.
Exercise 3.1. Show that the canonical maps
: Y in in Y
and

: in i!n Z Z

are isomorphisms.
There is a canonical map of simplicial sets : in Z i!n Z and a commutative
diagram
Zm

()

w Zu k

in Zn+1

w i!n Zn+1

224

IV. Bisimplicial sets

for each composite

k
n+1
m
in the ordinal number category with k, m n. It follows that an extension of
the n-truncated simplicial set Z to an (n + 1)-truncated simplicial set consists
precisely of a factorization

y
in Zn+1 A
(3.2)

A
AC
A

matching object
fibration, Reedy
n-coskeleton *a{c}
n-skeleton *a{s}

w Zn+1
u
i!n Zn+1

of the function : in Zn+1 i!n Zn+1 . The indicated map is a monomorphism,

because in Zn+1 must be the degenerate part of Zn+1 on account of the universal
property implicit in diagram (1.7).
All of the foregoing is completely natural, and gives corresponding results
for diagrams of simplicial sets. In particular, an extension of an n-truncated
bisimplicial set Z : op
n S to an (n + 1)-truncated bisimplicial set consists
of a factorization of the canonical simplicial set map : in Zn+1 i!n Zn+1 of
the form (3.2). Of course, the extended object consists of the simplicial sets
Z0 , . . . Zn , Zn+1 .
In the literature, the simplicial set in i!n1 i(n1) X arising from a bisimplicial
set X is denoted by Mn X, and is called the nth matching object for X [28], [29].
A map p : X Y of bisimplicial sets is said to be Reedy fibration if
(1) the map p : X0 Y0 is a Kan fibration of simplicial sets, and
(2) each of the induced maps p : Xn Yn Mn Y Mn X is a Kan fibration
for n > 0.
It is common to write
coskn X = i!n in X.
and
skn X = in in X.
From this point of view, the canonical maps in the second condition in the

3. Closed model structures for bisimplicial sets

225

definition are induced by the commutative diagram of simplicial set maps

Xn

w coskn1 Xn

u
Yn

u
w coskn1 Yn

which arises from the naturality of the adjunction maps. In particular, a bisimplicial set X is Reedy fibrant if the simplicial set X0 is a Kan complex and each
of the maps Xn coskn1 Xn , n > 0, is a Kan fibration.
Lemma 3.3.
(1) Suppose that a map p : X Y is a Reedy fibration. Then p has the
right lifting property with respect to all maps of bisimplicial sets which
are pointwise cofibrations and pointwise weak equivalences.
(2) Suppose that p : X Y is a Reedy fibration such that each of the
fibrations
Xn Yn coskn1 Yn coskn1 Xn
is also a weak equivalence. Then p has the right lifting property with
respect to all maps which are pointwise cofibrations.
Proof: We ll prove the first assertion. The second is similar.
Suppose given a commutative diagram
wX

u
V

u
wY

of bisimplicial set maps in which the map i is a pointwise cofibration and a

pointwise weak equivalence. Suppose inductively that there is a commutative
diagram of n-truncated bisimplicial set maps of the form
in U
i
u

in V

w in X
h

h
j
h
h

p
u
w in Y

226

IV. Bisimplicial sets

Then there is an induced solid arrow diagram of simplicial set maps

skn Vn+1 skn Vn+1 Un+1

i

w Xn+1

p
u

w Yn+1 coskn Yn+1 coskn Xn+1

Vn+1

The map i is a trivial cofibration, since the functor U 7 skn Un+1 takes maps
which are pointwise weak equivalences to weak equivalences of simplicial sets via
diagram (1.7), and the diagram of simplicial set maps

skn Un+1 y

w Un+1

u
skn Vn+1 y

w Vn+1

is a pullback. The map p is a Kan fibration by assumption, so the indicated

dotted arrow exists.

Corollary 3.4. Suppose that p : X Y is a Reedy fibration such that each

of the fibrations
Xn Yn coskn1 Yn coskn1 Xn
is also a weak equivalence. Then p is a pointwise weak equivalence.
Proof: The map p has the right lifting property with respect to the cofibrations
Fn n Fn n , so that p is a pointwise trivial fibration.

Lemma 3.5. Suppose that f : X Y is a map of bisimplicial sets. Then f has

factorizations
Z
p

P
N
iN

N
f
X'
wY
'
)
]
[
[q
j
W,

3. Closed model structures for bisimplicial sets

227

where
(1) the map i is a pointwise cofibration and a pointwise weak equivalence and
p is a Reedy fibration, and
(2) the map j is a pointwise cofibration and q is a Reedy fibration and a
pointwise weak equivalence.
Proof: Well prove the second claim. The first has a similar argument.
It suffices to find a factorization f = q j, where j is a pointwise cofibration
and q meets the conditions of Corollary 3.4.
Suppose, inductively, that weve found a factorization
j
in X 4
wW
4
4
q
6
f 4
u
in Y
in the category of n-truncated bisimplicial sets, such that j is a pointwise cofibration, and such that the maps q : W0 Y0 and
q : Wm Ym i!m1 Ym i!m1 Wm ,

0<mn

are trivial fibrations of simplicial sets. The commutative diagram

in Xn+1N
j

w Xn+1N
N

u
Nf

in Wn+1 N
N
NP
q
in Yn+1

w i!n Xn+1N
N

Nf
N
N
NP
w Yn+1

N
u
Nf
w i!n Wn+1 N
N
NP
q
w i!n Yn+1

induces a diagram
in Wn+1
u
Yn+1 i!n Yn+1

w in Wn+1 in Xn+1 Xn+1

u
!
!
in Wn+1
w in Wn+1 .

228

IV. Bisimplicial sets

Choose a factorization
j0

in Wn+1 in Xn+1 Xn+1 A

A
AA
C

w Wn+1
q0
u

Yn+1 i!n Yn+1 i!n Wn+1

such that j 0 is a cofibration and q 0 is a trivial fibration of simplicial sets. Then
the desired factorization of f at level n + 1 is given by the maps
j 0 j

q q 0

Xn+1 Wn+1 Yn+1 .

Note that the map
j : Xn+1 in Wn+1 in Xn+1 Xn+1
is a cofibration.

The matching space Mn X = coskn1 Xn is a special case of a construction

which associates a simplicial set MK X to each pair consisting of a simplicial set
K and a bisimplicial set X. Explicitly, the p-simplices of MK X are defined to
be a set of simplicial set morphisms by setting
MK Xp = homS (K, X(, p)).
The simplicial set MK X is a matching space for K in the bisimplicial set X.
Subject to the tacit indentification
Xn = X(n, ),
the bisimplicial set coskn1 X has (m, p)-bisimplices specified by
coskn1 X(m, p) = (coskn1 X(, p))m .
It follows that the p-simplices of the simplicial set coskn1 Xn have the form
coskn1 X(n, p) = (coskn1 X(, p))n
= homS (skn1 n , X(, p))
= homS (n , X(, p))
= Mn Xp ,

3. Closed model structures for bisimplicial sets

229

so that Mn X is naturally isomorphic to Mn X = coskn1 Xn . The functor

K 7 MK X is right adjoint to the functor S S2 which is defined by L 7 K L.

Suppose given integers s0 , . . . , sr such that 0 s0 < s1 < < sr n,
and let n hs0 , . . . , sr i be the subcomplex of n which is generated by the
simplices dsj n , j = 0, . . . , r. Then the simplicial identities di dj = dj1 di for
i < j imply that the complex n hs0 , . . . , sr i can be inductively constructed by
pushout diagrams or the form
sr 1
n1 hs0 , . . . , sr1 i d
w n hs0 , . . . , sr1 i
y
y

(3.6)
u

n1

u
w hs0 , . . . , sr i.
n

dsr

If k 6= sj for j = 0, . . . , r, so that n hs0 , . . . , sr i is a subcomplex of nk , then k is

a vertex of all of the generating simplices of the object n1 hs0 , . . . , sr1 i in the
diagram (3.6). It follows that this copy of n1 hs0 , . . . , sr1 i is a subcomplex
n1 , where q = k if k < sr and q = k 1 if k > sr .
of n1
q
Following Bousfield and Friedlander, we shall write
Mn(s0 ,...,sr ) X = Mn hs0 ,...,sr i X.
(0,1,...,n)

Then, in particular, Mn

X is yet another notation for coskn1 Xn .

Lemma 3.7. Suppose that the map p : X Y is a Reedy fibration and a

pointwise weak equivalence. Then p has the right lifting property with respect
to all pointwise cofibrations.
Proof: We show that each of the fibrations
Xm Ym coskm1 Ym coskm1 Xm
is a weak equivalence, and then apply Lemma 3.3.
There are canoncial simplicial set morphisms
(0,...,k)

Xn+1 Yn+1 M (0,...,k) Y Mn+1

n+1

which generalize the map

Xn+1 Yn+1 coskn Yn+1 coskn Xn+1 .

230

IV. Bisimplicial sets

The idea of proof is to show that each of the maps

(0,...,k)

Xn+1 Yn+1 M (0,...,k) Y Mn+1

n+1

is a trivial fibration of simplicial sets, by induction on n.

There is a pullback diagram
(0,...,k+1)

Yn+1 M (0,...,k+1) Y Mn+1

w Xn

n+1

u
u

(0,...,k)
Yn+1 M (0,...,k) Y Mn+1 X
n+1

w Yn M (0,...,k) Y Mn(0,...,k) X,
n

and so each of the vertical maps in the diagram

Xn+1

(0,...,n+1)

w Yn+1 M (0,...,n+1) Y Mn+1

n+1

u
(0,...,n)

Yn+1 M (0,...,n) Y Mn+1

n+1

u
..
.
u

Yn+1 M (0)

n+1

(0)

Mn+1 X

u
w Yn+1
is a trivial fibration by the inductive hypothesis. The map p is a weak equivalence, so that all of the intermediate maps
(0,...,k)

Xn+1 Yn+1 M (0,...,k) Y Mn+1

n+1

are weak equivalences as well.

Weve now done all the work that goes into the proof of

3. Closed model structures for bisimplicial sets

231

Theorem 3.8. The category S2 of bisimplicial sets, together with the classes ofsimplicial presheave
pointwise weak equivalences, pointwise cofibrations and Reedy fibrations, satis-closed model struct
fibration, diagonal
fies the axioms for a closed model category.

weak equivalence, d

Proof: The factorization axiom CM5 is Lemma 3.5, and the lifting axiomcofibration, Moerdij
CM4 is Lemma 3.3 together with Lemma 3.7.

Remark 3.9. There is a completely different approach to proving Theorem 3.8,

which involves a closed model structure for a certain category of presheaves of
simplicial sets [46], [51] (ie. for the chaotic topology on the ordinal number
category ). The global fibrations for that theory coincide with the Reedy
fibrations: seeing this requires having both the Reedy structure and the closed
model structure for the simplicial presheaf category in hand. The applications
of the Reedy structure given in the next section depend on the matching space
description of Reedy fibrations, so theres nothing to be gained by introducing
simplicial presheaf homotopy theory here.
3.3. The Moerdijk structure.
A map f : X Y is said to be a diagonal fibration (respectively diagonal
weak equivalence) if the induced map f : d(X) d(Y ) of associated diagonal
simplicial sets is a Kan fibration (respectively weak equivalence). A Moerdijk
cofibration is a map which has the left lifting property with respect to all maps
which are diagonal fibrations and diagonal weak equivalences.
The diagonal functor d : S2 S has a left adjoint d : S S2 , which is
completely determined by the requirements
(1) d (n ) = n,n , and
(2) d preserves colimits.
This follows from the fact that every simplicial set is a colimit of its simplices.
It implies that the bisimplicial set d (nk ) is the subcomplex of n,n which is
generated by the bisimplices (di n , di n ) for i 6= k, whereas, d (n ) is the
subcomplex of n,n which is generated by all (di n , di n ).
Alternatively, the set of bisimplices d (nk )(r, s) can be characterized as the set
of all pairs of ordinal number maps ( : r n, : s n) such that the images
of the functions and miss some common element i, where i 6= k. Put a
different way, the simplicial set of bisimplices d (nk )(, ) can be identified with
the subcomplex C of nk which is generated by the faces di n which contain the
s-simplex . This observation is the heart of the proof of
Lemma 3.10. The inclusion map d (nk ) d (n ) = n,n is a diagonal weak
equivalence of bisimplicial sets.

232

IV. Bisimplicial sets

Proof: The projection map

( : r n, : s n) 7
defines a map of bisimplicial sets
G
n
k

n
k

which in turn induces a map of simplicial sets

pr : d(d (nk )) nk
after applying the diagonal functor. The complex C of n is covered by subcomplexes isomorphic to n1 , each of which contains the vertex k. It follows
that the contracting homotopy
(x, t) 7 tx + (1 t)vk
of the affine simplex |n | onto the vertex vk corresponding to the element k n
restricts to a contracting homotopy
|C | I |C |.
The map pr is therefore a weak equivalence of simplicial sets, by Proposition 1.9,
so d(d (nk )) is contractible.

Remark 3.11. In the proof of Lemma 3.10, the complex C is a subcomplex of

nk of the form
C = n hs0 , . . . , sr i.
One can alternatively give a combinatorial argument for the contractibility of
this complex by making an inductive argument based on the existence of the
pushout diagram (3.6).
The diagonal functor d also has a right adjoint d : S S2 : the bisimplicial
set d K which is associated to the simplicial set K by this functor is defined by
d Kp,q = homS (p q , K).
Lemma 3.12. The functor d : S S2 takes fibrations to diagonal fibrations.

3. Closed model structures for bisimplicial sets

233

Proof: Suppose that p : X Y is a Kan fibration, and suppose that there is

a commutative diagram of simplicial set maps

nk

w d(d X)
i
ij

i
u i
n

d(d p)

w d(d Y )

Then the indicated lifting exists if and only if the lifting exists in the diagram

d(d nk )
d(d j)
u

wX

d(d )

p
u
w Y,

and Lemma 3.10 says that the inclusion d(d j) is a trivial cofibration.

Now suppose that the map i : U V of bisimplicial sets has the left lifting
property with respect to all diagonal fibrations. Then the lifting exists in all
diagrams
U
i

i
u i
V

w d X
j
ii
u

d p

w d Y

where p : X Y is a Kan fibration, so that the induced simplicial set map i :

d(U ) d(V ) has the left lifting property with respect to all Kan fibrations. In
particular, the map i is a diagonal weak equivalence. In view of Theorem II.5.1,
this implies the following:
Theorem 3.13. The category S2 of bisimplicial sets, together with the classes
of Moerdijk cofibrations, diagonal fibrations and diagonal weak equivalences,
satisfies the axioms for a closed model category.

234

IV. Bisimplicial sets

An alternative proof of this result can be given by using a small object argumentgroup object
(as was done originally in [73]), based on the observation that a map p : X
Y of bisimplicial sets is a diagonal fibration (respectively a diagonal fibration
and a diagonal weak equivalence) if and only if f has the right lifting property
with respect to all maps d (nk ) n,n (respectively with respect to all maps
d (n ) n,n ). One of the outcomes of the small object argument is the
assertion that every Moerdijk cofibration is a monomorphism of bisimplicial sets.
4. The Bousfield-Friedlander theorem.
Suppose that X is a pointwise fibrant bisimplicial set. The simplicial set
G

n (Xm , x)

xX(m,0)

and the obvious simplicial map

m X :

n (Xm , x)

xX(m,0)

= X(m, 0)

xX(m,0)

together form a group object in the category S X(, 0) of simplicial sets over
the vertex simplicial set X(, 0). This group object is abelian if m > 1.
Recall that a vertex v MK X is a simplicial set morphism v : K X(, 0).
The set of morphisms
MK (m X, v) = homSX(,0) (v, m X)
therefore has a group structure for m 1, which is abelian for m 2.
Suppose that i 7 K(i) defines an I-diagram K : I S in the category of
simplicial sets, and let
v : lim K(i) X(, 0)

iI

be a map of simplicial sets. Let v(i) : K(i) X(, 0) be the composite

ini

K(i) lim K(i)

X(, 0)

iI

of v with the canonical map ini of the colimiting cone. Then

v = lim vi

iI

4. The Bousfield-Friedlander theorem

235

in the category S X(, 0) of simplicial sets over X(, 0), and so there is an -Kan condition *
vertical, path comp
isomorphism
homSX(,0) (v, m X)
= lim homSX(,0) (vi , m X).

iI

It follows that there is an isomorphism

Mlim K(i) (m X, v)
= lim MK(i) (m X, vi ).

i
iI
In particular, the group Mn (m X, v) is canonically isomorphic to the group
m (Xn , v), where the map v : n X(, 0) is identified with a vertex v
X(n, 0). Any such vertex v : n X(, 0) restricts to a composite map
v

nk n
X(, 0),
which will also be denoted by dv. It follows in particular that the corresponding
group Mnk (m X, dv) fits into an equalizer diagram
(4.1)

Mnk (m X, dv)

m (Xn1 , di v)

i6=k

m (Xn2 , di dj v),

i<j;i,j6=k

where the parallel pair of arrows is defined by the simplicial identities di dj =

dj1 di .
A pointwise fibrant bisimplicial set X is said to satisfy the -Kan condition
if the map
d : m (Xn , v) Mnk (m X, dv)
induced by restriction along the inclusion nk n is a surjective group homomorphism for all m 1 and all n, k that make sense. The -Kan condition for
X is equivalent to the requirement that all of the structure maps
m X X(, 0)
for the homotopy group objects m X, m 1, are Kan fibrations, on account of
the description of Mnk (m X, dv) given in (4.1).
Suppose that Y is an arbitrary bisimplicial set, and write 0 Y for the simplicial
set having n-simplices 0 Yn = 0 Y (n, ). This is the simplicial set of vertical
path components of the bisimplicial set Y . There is a canonical simplicial set
map Y (, 0) 0 Y .

236

IV. Bisimplicial sets

Lemma 4.2.
(1) A pointwise fibrant bisimplicial set X satisfies the -Kan condition if all
of the vertical simplicial sets Xn = X(n, ) are path connected.
(2) Suppose that f : X Y is a pointwise weak equivalence of pointwise
fibrant bisimplicial sets. Then X satisfies the -Kan condition if and
only if Y satisfies the -Kan condition.
Proof: In the case of statement (1), there is a path from a given vertex
x X(n, ) to a horizontally degenerate vertex s(y), where y X(0, ) and so
the action the corresponding morphism [] of the fundamental groupoid for Xn
induces an isomorphism of maps

m (Xn , x)
[]
=
u
m (Xn , s(y))

w Mnk (m X, x)

= []
u
n
w Mk (m X, s(y)),

and the simplicial group n 7 m (Xn , s(y)) is a Kan complex.

To prove statement (2), note first of all that the map f : X Y is a pointwise
weak equivalence of pointwise fibrant bisimplicial sets if and only if
(a) the induced map f : 0 X 0 Y is an isomorphism of simplicial sets,
and
(b) the induced simplicial set diagrams

m X
(4.3)

u
X(, 0)

w m Y
u
w Y (, 0)

are pullbacks for m 1.

Kan fibrations are stable under pullback, so if f is a pointwise weak equivalence
and Y satisfies the -Kan condition, then X satisfies the -Kan condition.

4. The Bousfield-Friedlander theorem

237

Suppose that X satisfies the -Kan condition, and that there is a diagram fibrant model, point

-Kan condition *

nk
y
u
n

w m Y
u
w Y (, 0)

The simplicial set map f : 0 X 0 Y is an isomorphism, so that the vertex

v Y (n, ) is homotopic to a vertex f (w) for some vertex w X(n, ). The
argument for statement (1) implies that the -Kan condition for the vertex v is
equivalent to the -Kan condition for the vertex f (w), so we can replace v by
f (w). But now the diagram (4.3) is a pullback, by assumption, and so there is
a commutative diagram
nk
y

w m X
\
]
\

w m Y

u \
n
w

u
w X(, 0)

u
w Y (, 0)

where f = . Then X satisfies the -Kan condition, by assumption, so

that the indicated lifting exists, and the -Kan condition for the vertex f (w) is
verified.

A pointwise fibrant model of a bisimplicial set X consists of a pointwise weak

equivalence j : X Z, where Z is a pointwise fibrant bisimplicial set.
We shall say that an arbitrary bisimplicial set X satisfies the -Kan condition if for any pointwise fibrant model j : X Z of X, the pointwise fibrant
object Z satisfies the -Kan condition in the sense described above. Lemma 4.2
says that the -Kan condition for pointwise fibrant bisimplicial sets is an invariant of pointwise weak equivalence, so (by appropriate manipulation of the
Bousfield-Kan closed model structure for bisimplicial sets) it suffices to find only
one pointwise fibrant model j : X Z which satisfies the -Kan condition.
The -Kan condition for arbitrary bisimplicial sets is also an invariant of weak
equivalence.
Suppose that f : X Y is a Reedy fibration. Let i : K L be an inclusion
of simplicial sets, and observe that all induced bisimplicial set maps
n ) (K
nk ) , L
n
(K
n ) (L
k

238

IV. Bisimplicial sets

are pointwise trivial cofibrations of bisimplicial sets. The functor X 7 MK X is

, so it follows that the simplicial set map
right adjoint to the functor Y 7 K Y
(f ,i )

(4.4)

ML X ML Y ML X MK X

which is jointly induced by the Reedy fibration f and the inclusion i is a Kan
fibration.
Lemma 4.5. Suppose X is a Reedy fibrant bisimplicial set X that satisfies the
-Kan condition. Take a vertex x X(n, ). Then there is a canonical isomorphism
m (Mnk X, dx)
= Mnk (m X, dx).
There is also an isomorphism
0 (Mnk X)
= Mnk (0 X).
Proof: Choose integers 0 s0 < s1 < < sr n with si 6= k, and recall
that
Mn(s0 ,...,sr ) X = Mn hs0 ,...,sr i X
is a subcomplex of Mnk X. Write
Mn(s0 ,...,sr ) (m X, dx) = Mn hs0 ,...,sr i (m X, dx),
(s0 ,...,sr )

and let Mn

0 X denote the set

Mn hs0 ,...,sr i (0 X) = homS (n hs0 , . . . , sr i, 0 X).

The pushout diagram (3.6) induces a pullback diagram

(4.6)

Mn(s0 ,...,sr ) X

w Xn1

Mn(s0 ,...,sr1 ) X

(s ,...,sr1 )

0
w Mn1

Then the map d is an instance of the fibration (4.4), and inductively the canonical
map
m (Mn(s0 ,...,sr1 ) X, dx) Mn(s0 ,...,sr1 ) (m X, dx)

4. The Bousfield-Friedlander theorem

239

is an isomorphism. The map

(s ,...,sr1 )

0
d : m (Xn1 , dsr x) Mn1

(m X, dx)

is surjective for all m 1, since X satisfies the -Kan condition. It follows that
the induced map
(s ,...,sr1 )

0
d : m (Xn1 , x) m (Mn1

X, dx)

is surjective for all m 1, and so the commutative square

m (Mn(s0 ,...,sr ) X, dx)

w m (Xn1 , dsr x)

m (Mn(s0 ,...,sr1 ) X, dx)

(s ,...,sr1 )

0
w m (Mn1

X, ddsr x)

(s0 ,...,sr )

of abelian group homomorphisms is a pullback. The group m (Mn

therefore has the required form.
The map
(s ,...,sr1 )

0
d : 1 (Xn1 , x) 1 (Mn1

X, dx)

X, dx)

is surjective for all choices of base point x Xn1 , and so all inclusions Fx ,
Xn1 of fibres over dx induce injections 0 Fx 0 Xn1 . It follows that applying
the path component functor 0 to all diagrams of the from (4.6) gives pullback
diagrams of sets. This is what is required to show inductively that the canonical
maps
0 (Mn(s0 ,...sr ) X) Mn(s0 ,...,sr ) 0 X
are bijections.

Lemma 4.7. Suppose that X and Y are Reedy fibrant bisimplicial sets which
satisfy the -Kan condition, and that the bisimplicial set map f : X Y is
a Reedy fibration. Suppose further that the induced simplicial set map f :
0 X 0 Y of vertical path components is a Kan fibration. Then the map f is
a horizontal pointwise Kan fibration.

240

IV. Bisimplicial sets

To understand the meaning of the word horizontal in the statement of

Lemma 4.7, note that every Reedy fibration is a pointwise Kan fibration, since
m
the maps Fn m
k Fn are trivial cofibrations. The assumptions of the lemma
therefore imply that the simplicial sets X(n, ) and Y (n, ) are Kan complexes,
and that the simplicial set maps f : X(n, ) Y (n, ) are Kan fibrations.
The lemma asserts that under the stated conditions the simplicial set maps
f : X(, m) Y (, m) are Kan fibrations as well.
Proof: We shall prove the lemma by showing that each canonical map
d : Xm Ym Mm Y Mm
X
k
k

is a surjective simplicial set map. The map d is an instance of (4.4), hence a

Kan fibration, so it suffices to show that d induces a surjective function
d : 0 Xm 0 (Ym Mm Y Mm
X)
k
k

in path components.
From the previous result, the induced map
0 Xm 0 (Ym Mn Y Mnk X)
k

can be identified up to isomorphism with the map

0 Xn 0 Yn Mn 0 Y Mnk 0 X,
k

and the latter is surjective on account of the assumption that the induced map
0 X 0 Y in vertical path components is a Kan fibration.

Recall that the bisimplicial set Fn K is canonically isomorphic to the object

n
n.
K, and that the classifying bisimplex m,n is a copy of m
Lemma 4.8. Suppose that f : X Y is a Reedy fibration and a horizontal
pointwise Kan fibration in the sense that each of the maps f : X(, n) Y (, n)
is a Kan fibration. Then f is a diagonal fibration.
Proof: The cofibration
d nk d n = n,n
factors as a composite of two maps
n n
n.
d nk nk

4. The Bousfield-Friedlander theorem

241

The first map is a pointwise trivial cofibration, by the proof of Lemma 3.10. Thehomotopy cartesian
second has the left lifting property with respect to all horizontal pointwise Kan
fibrations.

A commutative square
X

wY

u
Z

u
wW

of bisimplicial set maps is said to be pointwise homotopy cartesian if each of the

induced squares of simplicial set maps
Xn

w Yn

u
Zn

u
w Wn

is homotopy cartesian, for n 0.

Theorem 4.9 (Bousfield-Friedlander). Suppose given a pointwise homotopy cartesian square
X

wY

u
Z

p
u
wW

in the category of bisimplicial sets such that Y and W satisfy the -Kan condition. Suppose further that the induced map p : 0 Y 0 W of vertical path
components is a Kan fibration. Then the associated commutative square
d(X)

w d(Y )

u
d(Z)

d(p)
u
w d(W )

of diagonal simplicial sets is homotopy cartesian.

242

IV. Bisimplicial sets

Proof: Construct a diagram

Y

jY

w Y0

p
u
W

p0

jW

u
w W0

in which jY and jW are pointwise trivial cofibrations, W 0 is Reedy fibrant, and

p0 is a Reedy fibration. Then the composite square
X

jY

wY
I

u
Z

p
u
wW

w Y0
p0

II

u
w W0

jW

is pointwise homotopy cartesian by Lemma II.9.20, and the bisimplicial sets Y 0

and W 0 satisfy the -Kan condition by Lemma 4.2. The Reedy fibrant objects
Y 0 and W 0 are both pointwise fibrant.
Applying the diagonal functor gives a composite diagram in the simplicial set
category
d(X)
(4.10)

u
d(Z)

w d(Y )
I

p
u
w d(W )

jY

w d(Y 0 )
p0

II
jW

u
w d(W 0 ),

in which the induced maps jY and jW are weak equivalences by Proposition 1.9.
Thus, in order to demonstrate that the square
d(X)
u
d(Z)

w d(Y )
I

p
u
w d(W )

4. The Bousfield-Friedlander theorem

243

is homotopy cartesian, it suffices, by Lemma II.9.20, to show that the compositepointwise k-connect
square (4.10) is homotopy cartesian.
The map p0 is a diagonal fibration, by Lemma 4.7 and Lemma 4.8, and the
induced bisimplicial set map
X Z W 0 Y 0
is a pointwise weak equivalence, since the Reedy fibration p0 is a pointwise fibration. It follows that the induced map
d(X) d(Z) d(W 0 ) d(Y 0 )
of diagonal simplicial sets is a weak equivalence.

Corollary 4.11. Suppose that a pointed bisimplicial set X is pointwise fibrant

and pointwise connected. Then there is a weak equivalence
d(X) ' d(X).
Proof: We have tacitly chosen a base point x X(0, 0) for all of the vertical
simplicial sets X(n, ) in our assumption that X is pointed. The loop object
X can then be characterized as the pointed bisimplicial set having vertical
simplicial sets X(n, ). There is a corresponding path space P X, which is the
pointed bisimplicial set with vertical simplicial sets P X(n, ), and a pointwise
fibre sequence
p

X P X
X.
The map p induces an isomorphism 0 P X
= 0 X of vertical path component
simplicial sets. The bisimplicial sets P X and X both satisfy the -Kan condition, and so applying the diagonal functor gives a homotopy fibre sequence
d(p)

d(X) d(P X) d(X)

in the category of pointed simplicial sets, by Theorem 4.9. The simplicial set
d(P X) is contractible, by Proposition 1.9.

A bisimplicial set X is said to be pointwise k-connected if each of the associated

simplicial sets Xn is k-connected, for n 0. A pointwise connected (or 0connected) bisimplicial set can also be characterized as a bisimplicial set X such
that the associated vertical path component simplicial set 0 X is a copy of the
terminal object .

244

IV. Bisimplicial sets

Lemma 4.12. Suppose that X is a bisimplicial set which is pointwise connected.

Then the diagonal simplicial set d(X) is connected.
Proof: There is a coequalizer diagram
0 X(1, )

d0
d1

ww 0 X(1, )

w 0 d(X),

where d0 and d1 are induced by the horizontal face maps d0 , d1 : X(1, )

X(0, ).

Proposition 4.13. Suppose that X is a pointed bisimplicial set which is pointwise k-connected. The the diagonal d(X) is a k-connected simplicial set.
where X
is pointwise
Proof: Choose a pointwise weak equivalence i : X X,

fibrant. Then d(i) : d(X) d(X) is a weak equivalence. We may therefore

presume that X is pointwise fibrant.
The space d(X) is connected, by Lemma 4.12. Also, the bisimplical set X is
pointwise (k 1)-connected, and there are isomorphisms
j d(X)
= j1 d(X),
for j 1, by Corollary 4.11. This does it, by induction on k.

Proposition 4.13 admits a vast generalization, in the sense that it is a consequence of a very general spectral sequence calculation. The existence of the
spectral sequence in question is an easy consequence of Theorem 4.9, modulo a
few technical observations.
First of all, if G is a simplicial group, then there is a bisimplicial set BG
whose vertical simplicial in horizontal degree n is the classifying space BGn of
the group Gn of n-simplices of G. Similarly, the translation categories of the
various groups Gn can be collected together to form a bisimplicial set EG and a
canonical map : EG BG. These bisimplicial sets are pointwise fibrant and
connected in each horizontal degree, and the map is a pointwise fibration. It
therefore follows from Theorem 4.9 that there is an induced fibre sequence

G d(EG) d(BG).
Furthermore, the bisimplicial set EG consists of contractible simplicial sets EGn ,
so that the associated diagonal d(EG) is contractible, by Proposition 1.9. It
follows that there are natural isomorphisms
n d(BG)
= n1 G

4. The Bousfield-Friedlander theorem

245

for n 1. The space d(BG) is connected, by Lemma 4.12.

Bousfield-Kan spect
Suppose that X is a pointwise fibrant and pointed bisimplicial set such that
each of the vertical simplicial sets Xn is an Eilenberg-Mac Lane space of the
form K(m , m), for some fixed number m 2. Then Corollary III.3.8 implies
that there is a pointwise weak equivalence of bisimplicial sets of the form
X (m X[m]),
where m X[m] denotes the chain complex of simplicial groups, concentrated in
chain degree m. Then, by Theorem 4.9 or via chain complex arguments, one
sees that there are natural isomorphisms
j d(X)
= jm (m X)
for j 0.
Suppose finally that X is a pointwise fibrant bisimplicial set which is pointed
and pointwise connected. Then the Postnikov tower construction applied to each
of the vertical simplicial sets Xn induces a tower of pointwise fibrations
X(n) X(n 1) X(1) X(0),
such that the the fibre Fn of the map X(n) X(n1) is a diagram of EilenbergMac Lane spaces of the form K(n X, n). Each of the pointwise fibre sequences
Fn X(n) X(n 1)
induces a fibre sequence of associated diagonal simplicial sets, by Theorem 4.9.
The resulting long exact sequences

j+1 d(X(n 1))

j d(Fn ) j d(X(n)) j d(X(n 1)) . . .
determine an exact couple which gives rise to a convergent spectral sequence
with
(4.14)

s (t X) s+t d(X),
E2s,t = s+t Ft =

t + s 0.

This spectral sequence is due to Bousfield and Friedlander [12, p.122]. It is a

reindexed example of the Bousfield-Kan spectral sequence for a tower of fibrations, which will be discussed at more length in Section VII.3. Convergence
for such spectral sequences is usually an issue, but it follows in this case from
Proposition 4.13, which implies that the map X X(n) induces isomorphisms
j d(X)
= j d(X(n))
for j n.

246

IV. Bisimplicial sets

5. Theorem B and group completion.

translation category
The stream of ideas leading to Quillens Theorem B begins with the most
general formulation of the Serre spectral sequence.
5.1. The Serre spectral sequence.
Suppose that p : E B is an arbitrary map of simplicial sets, and recall that
B denotes the simplex category for B. There is a functor p1 : B S
taking values in the simplicial set category which is defined by associating to the
simplex : n B the simplicial set p1 (), where p1 () is defined by the
pullback diagram
p1 ()

wE
p

u
n

u
w B.

The homotopy colimit holim p1 arising from the functor p1 : B S

has vertical simplicial set in horizontal degree n given by

G
p1 (0 ).
0 n

Note that this disjoint union is indexed by strings of arrows of length n in the
simplex category B, and that these strings form the set of n-simplices of
the nerve B( B).
The simplicial set B is a colimit of its simplices in the simplicial set category,
and pulling back along p : E B is right exact, so that the maps : p1 ()
E induce an isomorphism of simplicial sets
lim p1 ()
= E.

In particular, the set Em of m-simplices of E may be identified with the set of

path components of the translation category Ep1
m arising from the functor
7 p1 ()m .
The objects of this category are pairs (, x), where : n B is a simplex of
B and x p1 ()m , and a morphism : (, x) (, y) consists of a morphism
: in B such that (x) = y. The nerve BEp1
m for this translation

5. Theorem B and group completion

247

category coincides with the horizontal simplicial set in holim p1 appearing in

vertical degree m.
1
Write Ep1
m,x for the path component of the translation category Epm corresponding to a simplex x of Em . This component Ep1
m,x is the full subcategory
1
of Epm on objects of the form (, y), where (y) = x. In particular, y must
have the form y = (, x) in
p1 () = n B E.
One sees, now, that the object (p(x), (m , x)) is initial in the category Ep1
m,x , so
1
that BEpm,x is contractible, and the simplicial set map
BEp1
m Em

(5.1)

is a weak equivalence. We have taken the liberty of identifying the set Em with
the corresponding constant simplicial set K(Em , 0); by further abuse, the maps
(5.1) are the horizontal components of a bisimplicial set map
holim p1 E,

which is a weak equivalence in each vertical degree. Proposition 1.9 therefore

implies the following:
Lemma 5.2. Suppose that p : E B is a map of simplicial sets. Then the
canonical bisimplicial set map
holim p1 E

induces a weak equivalence

d(holim p1 ) ' E

of simplicial sets.
Now let A be an abelian group, and consider the bisimplicial abelian group
Z(holim p1 ) A. The diagonal of this object is the simplicial abelian group

Z(d(holim p1 )) A, and the weak equivalence d(holim p1 ) E of Lemma 5.2

induces a weak equivalence of simplicial groups

Z(holim p1 ) A Z(E) A,

248

IV. Bisimplicial sets

by universal coefficients and the fact that the free abelian group functor pre-Grothendieck spect
serves weak equivalences (see Lemma III.2.14). Note as well that the bisim-Serre spectral seque
plicial abelian group Z(holim p1 ) A can be identified with the translation

object associated to the functor defined on the simplex category B by

7 Z(p1 ()) A (Section II.4). It follows, by the results of Section 2, that
there is a spectral sequence
(5.3)

E2p,q = p EHq (Z(p1 ) A) Hp+q (E, A).

In other words, E2p,q is the pth homotopy group of the translation object for the
abelian group valued functor Hq (Z(p1 ) A) : B Ab defined by
(5.4)

7 Hq (Z(p1 ()) A) = Hq (p1 (), A).

The spectral sequence (5.3) is sometimes called the Grothendieck spectral sequence, and is defined for any simplicial set map p : E B. This spectral
sequence specializes to the Serre spectral sequence in the case where the map
p : E B is a fibration. When p is a fibration, any map : in the
simplex category B induces a weak equivalence : p1 () p1 ( ), and
hence induces isomorphisms

Hq (p1 (), A) Hq (p1 ( ), A)

for all q 0. It follows (Theorem III.1.1) that the functors (5.4) factor through
functors
Hq (Z(p1 ) A) : (B) = G( B) Ab,
which are defined on the fundamental groupoid (B) of the space B (see Theorem III.1.1).
If in addition B is simply connected, and F is the fibre over a choice of base
point for B, then the functor Hq (Z(p1 ) A) is naturally isomorphic to the
constant functor 7 Hq (F, A) on the simplex category for B, and so there is a
natural isomorphism
E2p,q = p EHq (Z(p1 ) A)
= Hp (B( B), Hq (F, A)).
The assertion that there is a natural weak equivalence d(holim p1 ) E can be

specialized to the case of the identity map B B, implying that the bisimplicial
set map
G
m0 B
0 n

5. Theorem B and group completion

249

(where j : mj B are simplices of B) induces a weak equivalence of diagonalTheorem B

simplicial sets. One also knows that the canonical bisimplicial set map
G

m 0

0 n

0 n

is a pointwise weak equivalence, and hence induces a weak equivalence

d(

m0 ) B( B).

0 n

It follows that there are isomorphisms

Hp (B( B), Hq (F, A))
= Hp (B, Hq (F, A)),
p 0, and we obtain the standard form of the Serre spectral sequence
(5.5)

E2p,q = Hp (B, Hq (F, A)) Hp+q (E, A).

5.2. Theorem B.
Quillens Theorem B is the following:
Theorem 5.6. Suppose that F : C D is a functor between small categories
such that for every morphism : y y 0 of D the induced simplicial set map
: B(y 0 F ) B(y F ) is a weak equivalence. Then, for every object y of
D, the commutative diagram

B(y F )

w BC

u
B(y D)

F
u
w BD

of simplicial set maps is homotopy cartesian.

Here, one should recall that the objects of the comma category y F are pairs
(, x), where x is an object of C and : y F (x) is a morphism of D. A

250

IV. Bisimplicial sets

morphism : (0 , x0 ) (1 , x1 ) of y F is a morphism : x0 x1 of C such

that the diagram
0 44
6
44
y

F (x0 )
F ()
u
F (x1 )

commutes in the category D.

Theorem B has important applications in algebraic K-theory. In some sense,
however, one of the steps in its proof is even more important, this being the
following result:
Lemma 5.7. Suppose that X : I S is a simplicial set valued functor which is
defined on a small category I. Suppose further that the induced simplicial set
map X() : X(i) X(j) is a weak equivalence for each morphism : i j
of the index category I. Then, for each object j of I the pullback diagram of
simplicial sets
X(j)

w d(holim X)

(5.8)

u
w BI

is homotopy cartesian.
Proof: The diagram (5.8) is obtained by applying the diagonal functor to the
following pullback diagram of bisimplicial sets:
X(j)

G
w

X(i0 )

i0 in

Gu
i0 in

5. Theorem B and group completion

251

In particular, the canonical map : d(holim X) BI is the map which is

associated to the natural transformation of functors on I which is given by the

simplicial set maps X(i) .
The strategy of proof is to find a factorization
j

h
j
i h

'

w BI
'
)
'p

of the inclusion of the vertex j in BI such that j is a trivial cofibration, p is a

fibration, and such that the induced map X(j) U BI d(holim X) is a weak

equivalence.
Pulling back along the map : d(holim X) BI preserves colimits in the

category S BI of simplicial set maps K BI. The small object argument

therefore implies that it suffices to show that any diagram

d(holim X)

(5.9)
nk y

w n

u
w BI

induces a weak equivalence

(5.10)

i : nk BI d(holim X) n BI d(holim X).

The simplex in the diagram (5.9) is a functor : n I, and the space

n BI d(holim X) coincides up to isomorphism with the diagonal of the homo
topy colimit holim X associated to the composite functor

n
I S.

252

IV. Bisimplicial sets

Furthermore, the map i in (5.10) can be identified with the diagonal of the map
i in the following pullback diagram of bisimplicial sets:
G

X(k0 )

G
w

k0 kr
n
k

X(k0 )

k0 kr
n

Gu

k0 kr
n
k

Gu
w

k0 kr
n

The initial object 0 n determines a natural transformation : X(0) X,

where X(0) denotes the constant functor at the object of the same name, and
there is an induced diagram of bisimplicial set maps
G

X(0)

k0 kr
n
k

k0 kr
n
k

X(0)

k0 kr
n

u
X(k0 )

X(k0 ).

k0 kr
n

The vertical maps induce weak equivalences of associated diagonal simplicial

sets, by Proposition 1.9, and the diagonal of the top horizontal map is the weak
equivalence
i 1 : nk X(0) n X(0).
It follows that the map i of (5.10) is a weak equivalence.

There is a homology version of Lemma 5.7 for every homology theory h which
satisfies the wedge axiom. Here is a specimen statement:
Lemma 5.11. Suppose that X : I S is a simplicial set valued functor which is
defined on a small category I, and that A is an abelian group. Suppose further
that the induced simplicial set map X() : X(i) X(j) induces an isomorphism

5. Theorem B and group completion

253

H (X(i), A)
= H (X(j), A) for each morphism : i j of the index category
I. Then, for each object j of I the pullback diagram of simplicial sets
X(j)

w d(holim X)

u
w BI

is homology cartesian in the sense that the corresponding map X(j) Fj from
X(j) to the homotopy fibre Fj over j induces an isomorphism H (X(j), A)
=
H (Fj , A).
The proof of Lemma 5.11 is a spectral sequence argument which follows the basic
outline of the proof of Lemma 5.7.
Proof of Theorem 5.6: The functor y 7 B(y F ) determines a contravariant simplicial set valued functor D S, with homotopy colimit having (m, n)bisimplices
G
B(y0 F )m
yn y0

This set of bisimplices can also be identified with the set of all strings of arrows
in D of the form
yn y0 F (x0 ) F (xm ).
The degenerate simplices
1

y
y
...
y
determine a commutative diagram of bisimplicial set maps
B(y F )m

B(y0 F )m

w BCm

yn y0

F
(5.12)

u
B(y D)m
w

u
B(y0 D)m

G
yn y0

'
u

'
Gu
yn y0

u
w BDm

254

IV. Bisimplicial sets

The bisimplicial set map

G

(5.13)

B(y0 F )m BCm

yn y0

is an alternate way of representing the forgetful map

G
G

x0 xm

yn ...y0 F (x0 )F (xm )

corresponding to the functor F , and can also be identified with the map
G
G
B(F (x0 ) D)op
.
x0 xm

x0 xm

The category (F (x0 ) D)op has a terminal object, so the map Q in (5.13)
induces a weak equivalence of associated diagonals, by Proposition 1.9. The
bisimplicial set map
G
Q
B(y0 D)m BDm
yn y0

is an instance of the map in (5.13), corresponding to the case where F is the

identity functor on the category D, so it induces a weak equivalence of associated
diagonal simplicial sets as well. The categories y D and y0 D have initial
objects, so Proposition 1.9 implies that the indicated maps in the diagram (5.12)
induce weak equivalences of associated diagonals.
Thus (see Lemma II.9.20), to show that the simplicial set diagram
B(y F )

w BC

u
B(y D)

F
u
w BD

is homotopy cartesian, it suffices to see that the bisimplicial set diagram

B(y F )m

B(y F )m

yn y0

Gu
yn y0

5. Theorem B and group completion

255

induces a homotopy cartesian diagram of the associated diagonal simplicial sets.group completion
This is a consequence of Lemma 5.7.

5.3. The group completion theorem.

Suppose that M is a simplicial monoid, and that X is a simplicial set with an
M -action M X X. There is a Borel construction for this action, namely a
bisimplicial set EM M X having vertical simplicial set in horizontal degree n
given by M n X. The fastest way to convince yourself that this thing actually
exists is to observe that the action M X X is composed of actions Mn Xn
Xn of monoids of n-simplices on the corresponding sets Xn . These actions admit
Borel constructions EMn Mn Xn (or nerves of translation categories), and this
construction is natural in n.
The canonical maps : EMn Mn Xn BMn are also natural in n, and
therefore define a map
: EM M X BM
of bisimplicial sets, which is given in horizontal degree n by the projection
M n X M n .
There is an obvious pullback diagram of bisimplicial set maps
X

w EM M X

u
w BM

(5.14)

The group completion theorem gives a criterion for this diagram to be homology
cartesian.
Theorem 5.15 (Group Completion). Suppose that M X X is an action
of a simplicial monoid M on a simplicial set X, and let A be an abelian group.
Suppose further that the action of each vertex v of M induces an isomorphism
v : H (X, A)
= H (X, A). Then the diagram (5.14) is homology cartesian in
the sense that the map X F to the homotopy fibre of the simplicial set map
d() induces an isomorphism in homology with coefficients in A.
Theorem 5.15 is used, in the main, to analyze the output of infinite loop space
machines. It implies, for example, that each connected component of the 0th
space of the -spectrum corresponding to the sphere spectrum is a copy of the
space B+
obtained by applying Quillens plus construction to the classifying
space of the infinite symmetric group [6], [83]. Here is another typical calculation:

256

IV. Bisimplicial sets

Example 5.16. Suppose that R is a ring with identity. Then matrix addition
induces a simplicial monoid structure on the simplicial set
G

M (R) =

BGln (R).

n0

In particular, right multiplication by the vertex e = BGl1 (R)0 (and all of

its degeneracies) induces a simplicial set map
? e : M (R) M (R),
which restricts, on the nth summand, to the map BGln (R) BGln+1 (R) which
is induced by the canonical inclusion Gln (R) , Gln+1 (R) defined by A 7 AI,
where I denotes the identity element of Gl1 (R). The filtered colimit of the system
?e

?e

M (R) M (R) . . .
can be identified up to isomorphism with the simplicial set
X(R) =

BGl(R).

The simplicial set X(R) has an obvious left M (R) action, and Theorem 5.15
implies that the diagram

X(R)

w EM (R) M (R) X(R)

u
w BM (R)

is homology cartesian. In effect, left multiplication by a vertex n M (R) just

shifts the vertices of X(R): the summand corresponding to r Z is taken to
the summand corresponding to n + r. The map induced on the rth summand
itself is the simplicial set map BGl(R) BGl(R) which is induced by the group
homomorphism In ? : Gl(R) Gl(R) defined by A 7 In A, where In is
the n n identity matrix. As such, the group homomorphism In ? is a filtered
colimit of group homomorphisms Glm (R) Glm+n (R). The key point is that

5. Theorem B and group completion

257

In A is conjugate, via a suitable choice of permutation matrix, to A In in

Glm+n (R). It follows that (vertical) components of the comparison map
H (BGlm (R), Z)
(In ?)

can

w H (BGlm+1 (R), Z)
(In ?)

can

w ...

H (BGlm+n (R), Z) can w H (BGlm+n+1 (R), Z) can w . . .

coincide with morphisms induced by canonical inclusions, and so the group homomorphism In ? induces the identity map on H (BGl(R), Z).
The space arising from the bisimplicial set EM (R)M (R) X(R) is contractible,
since it is a filtered colimit of objects of the form EM (R) M (R) M (R). It
follows that X(R) has the homology of the loop space d(BM (R)), and that
the component d(BM (R))0 of 0 Z is an H-space having the homology of
BGl(R). This component d(BM (R))0 must therefore be a copy of Quillens
space BGl(R)+ .
Finally (without going into a lot of details), the monoidal structure on M (R)
is abelian up to coherent isomorphism, so effectively one is entitled to form a
collection of connected objects
BM (R), BBM (R), B 3 M (R), . . .
such that B n+1 M (R) is a delooping of B n M (R) for all n, just like one could
do if M (R) happened to be a simplicial abelian group. The list of spaces corresponding to
BM (R), BM (R), BBM (R), B 3 M (R), . . .
is the algebraic K-theory spectrum for the ring R; we have used the group
completion theorem to identify its 0th term.
There are several proofs of the group completion theorem in the literature:
[48], [49], [73], [70]. The most elementary of these, and the one that will be
given here, involves an analogue of the construction leading to the Serre spectral
sequence, for maps of bisimplicial sets.
Suppose that f : X Y is a map of bisimplicial sets, and consider all bisimplices : r,s Y of Y . Form the pullback diagram
f 1 ()

wX

u
wY

f
r,s

258

IV. Bisimplicial sets

in the category of bisimplicial sets. The bisimplices m,n Y of Y are thecategory, of bisimpl
objects of the category 2 Y , called the category of bisimplices of Y . A
morphism of this category is a commutative diagram of bisimplicial set
maps
r,s hh
u
m,n

h
h
j

)
'
'
''

Y.

One sees immediately that the assignment 7 f 1 () defines a functor

f 1 : 2 Y S2 .
Lemma 5.17. Suppose that f : X Y is a map of bisimplicial sets. Then the
corresponding map
G
f 1 (0 ) X
0 r

B(2 Y )

of trisimplicial sets induces a weak equivalence of associated diagonal simplicial

sets.
Proof: The bisimplicial set Y is a colimit of its bisimplices, and so X is a
colimit of their pullbacks, giving rise to a coequalizer
G

f 1 (0 )

0 1

f 1 () Y.

Also, the component of the simplicial set

G

f 1 (0 )(m, n)

0 r

corresponding to each bisimplex x X(m, n) is contractible, since its the nerve

of a category having an initial object. It follows that diagonalizing the trisimplicial set map
G
f 1 (0 )(m, n) X(m, n)
0 r

B(2 Y )

5. Theorem B and group completion

259

first with respect to the variables r and m gives a bisimplicial set map
G

trisimplicial sets
n-fold simplicial set

f 1 (0 )(r, n) X(r, n)

0 r

B(2 Y )

which is a weak equivalence of simplicial sets in each vertical degree n, by Proposition 1.9. This same result then implies that the simplicial set map
G

f 1 (0 )(r, r) X(r, r)

0 r

B(2 Y )

is a weak equivalence.

Remark 5.18. There is a paradigm in the proof of Lemma 5.17 for the manipulation of trisimplicial sets. The diagonal of a trisimplicial set X is the simplicial
set whose set of n-simplices is the set X(n, n, n). This simplicial set defines the
homotopy type arising from X, and it can be formed by iterating the diagonal
construction for bisimplicial sets in three different ways. One picks the most convenient iteration for the problem at hand. Similar considerations apply, more
generally, to n-fold simplicial sets
Consider the diagram of trisimplicial set maps
f 1 (v)

G
w

f 1 (0 )

'

wX

'

u
wY

0 r

0,0

G
w

m0 ,n0

0 r

'
u

'
v

Gu

0 r

where v : 0,0 Y is a vertex of Y . Observe that = 0,0 is the terminal

object in the bisimplicial set category. The labelled horizontal maps induce
weak equivalences of the corresponding diagonal simplicial sets by Lemma 5.17,

260

IV. Bisimplicial sets

while the corresponding vertical maps induce weak equivalences of diagonals by

iterated application of Proposition 1.9. It follows from Lemma 5.11 that the
pullback diagram
f 1 (v)

wX

u
wY

of bisimplicial sets induces a homology cartesian diagram of associated diagonals

if each morphism of bisimplices
r0 ,s0 hh
0
h
j
h
u

r1 ,s1

)
'
'
''1

Y.

induces an isomorphism

H (f 1 (0 )) H (f 1 (1 )).
By this, it is meant that there should be an induced isomorphism

H (d(f 1 (0 )), A) H (d(f 1 (1 )), A),

relative to some choice of coefficient group A. Lets agree to suppress mention
of the diagonals and the coefficient groups in the rest of this section.
Proof of Theorem 5.15: We will show that each morphism

(5.19)

r,s hh
h
j
h
(1 , 2 )
BM.
'
)
'
u ''
k,`

5. Theorem B and group completion

261

induces an isomorphism

(1 , 2 ) : H ( 1 ( )) H ( 1 ()),
where : EM M X BM is the canonical map.
Recall that, in horizontal degree m, k,`
m may be identified with the simplicial
set
G
` .
mk

There is a pullback diagram of simplicial sets

G

w M m X

(` X)

mk

pr
Gu

wM

mk

where the map on the left is a disjoint union of projections. Each ordinal number
map : n m induces a simplicial map
G

(` X)

mk

(` X),

nk

according to the horizontal structure of the bisimplicial set 1 (). There is a

commutative diagram
`

w ` X

in
G
mk

u
(` X)

u
`

in

( X),

nk

where is defined on the simplex level by

(, x) 7 (, m (, x)).

262

IV. Bisimplicial sets

Here, m is the composite simplicial set map

1

` X M X X,
where m is the action of M on X and is some `-simplex of M . The assumptions
on the action m therefore imply that is a homology isomorphism.
One of the spectral sequences for the homology of the diagonal of the bisimplicial set
G
(` X)
mk

has E2 -term
H (

A()),

:mk

where A is a contravariant functor on the simplices of k : the group A() is a

copy of Hq (` X). We have just seen that every morphism

m hh
h
h
j

k
'
)
'
'
u '
n
induces an isomorphism : A()
= A(). The morphism 1k is terminal in the
k
simplex category for , so there is a natural isomorphism which is defined by
diagrams
A(1k )

u
A(1k )

w A()

u
w A().

Lemma 2.2 says that there are isomorphisms

Hi (

:mk

A())
=

A(1k ) if i = 0,
0
if i > 0.

5. Theorem B and group completion

263

In particular, the inclusion

inv

X `

(X ` )

0k

corresponding to a vertex v of k induces an isomorphism

Hq (X ` )
= Hq ( 1 ())
Finally, observe that every bisimplex map of the form (5.19) induces a commutative diagram of simplicial set maps
X s
inw
u
G
(X s )
0r

1 2

(1 , 2 )

w X `
G

in1 w

(X ` ),

0k

where w is a vertex of r . It follows that the map (1 , 2 ) is a homology

isomorphism.

Chapter V Simplicial groups

This is a somewhat complex chapter on the homotopy theory of simplicial
groups and groupoids, divided into seven sections. Many ideas are involved.
Here is a thumbnail outline:
Section 1, Skeleta: Skeleta for simplicial sets were introduced briefly in Chapter I,
and then discussed more fully in the context of the Reedy closed model structure
for bisimplicial sets in Section IV.3.2. Skeleta are most precisely described as
Kan extensions of truncated simplicial sets. The current section gives a general
description of such Kan extensions in a more general category C, followed by a
particular application to a description of the skeleta of almost free morphisms of
simplicial groups. The presentation of this theory is loosely based on the ArtinMazur treatment of hypercovers of simplicial schemes [3], but the main result
for applications that appear in later sections is Proposition 1.9. This result is
used to show in Section 5 that the loop group construction outputs cofibrant
simplicial groups.
Section 2, Principal Fibrations I: Simplicial G-spaces: The main result of this
section asserts that the category SG of simplicial sets admitting an action by a
fixed simplicial group G admits a closed model structure: Theorem 2.3. Principal G-fibrations in the classical sense may then be identified with cofibrant
objects of SG , by Corollary 2.10, and an equivariant map between two such objects is an isomorphism if and only if it induces an isomorphism of coinvariants
(Lemma 2.9).
Section 3, Principal Fibrations II: Classifications: This section shows (Theorem 3.9) that isomorphism classes of principal G-fibrations p : E B can be
classified by homotopy classes of maps B BG, where BG = EG/G, and EG
is an arbitrary cofibrant object of SG admitting a trivial fibration EG , all
with respect to the the closed model structure for SG of Section 2.
Section 4, Universal cocycles and W G: It is shown here that the classical model
W G for the classifying object BG of Section 3 can be constructed as a simplicial
set of cocycles taking values in the simplicial group G. This leads to global
descriptions of the simplicial structure maps for W G, as well as for the Gbundles associated to simplicial set maps X W G. The total space W G for
the (canonical) bundle associated to the identity map on W G is contractible
(Lemma 4.6).
Section 5, The loop group construction: The functor G 7 W G has a left adjoint
X 7 GX, defined on reduced simplicial sets X (Lemma 5.3). The simplicial
264

1. Skeleta

265

group GX is the loop group of the reduced simplicial set X, in the sense that
the total space of the bundle associated to the adjunction map X W GX is
contractible: this is Theorem 5.10. The proof of this theorem is a modernized
version of the Kans original geometric proof, in that it involves a reinterpretation
of the loop group GX as an object constructed from equivalence classes of loops.
Section 6, Reduced simplicial sets, Milnors F K-construction: This section gives
a closed model structure for the category S0 of reduced simplicial sets. This
structure is used to show (in conjunction with the results of Section 1) that the
loop group functor preserves cofibrations and weak equivalences, and that W
preserves fibrations and weak equivalences (Proposition 6.3). In particular, the
loop group functor and the functor W together induce an equivalence between
the homotopy categories associated to the categories of reduced simplicial sets
and simplicial groups (Corollary 6.4). Furthermore, any space of the form W G is
a Kan complex (Corollary 6.8); this is the last piece of the proof of the assertion
that W G is a classifying space for the simplicial group G, as defined in Section 3.
Milnors F K-construction is a simplicial group which gives a fibrant model for
the space K: Theorem 6.15 asserts that F K is a copy of G(K), by which
is meant the loop group of the Kan suspension of K. The Kan suspension was
introduced in Section III.5.
Section 7, Simplicial groupoids: The main result of Section 5, which leads to the
equivalence of homotopy theories between reduced simplicial sets and simplicial
groups of Section 6, fails badly for non-reduced simplicial sets. We can nevertheless recover an analogous statement for the full category of simplicial sets if we
replace simplicial groups by simplicial groupoids, by a series of results of Dwyer
and Kan. This theory is presented in this section. There is a closed model structure on the category sGd of simplicial groupoids (Theorem 7.6) whose associated
homotopy category is equivalent to that of the full simplicial set category (Corollary 7.11). The classifying object and loop group functors extend, respectively,
to functors W : sGd S and G : S sGd; the object W A associated to a
simplicial groupoid A is a simplicial set of cocycles in a way that engulfs the
corresponding object for simplicial groups, and the extended functor G is its left
adjoint.
1. Skeleta.
Suppose that C is a category having all finite coproducts, and let sC denote
the category of simplicial objects in C. Recall that simplicial objects in C are
contravariant functors of the form op C, defined on the ordinal number
category .
The ordinal number category contains a full subcategory n , defined on the

266

V. Simplicial groups

objects m with 0 m n. Any simplicial object X : op C restricts to

a contravariant functor in X : op
n C, called the n-truncation of X. More
generally, an n-truncated simplicial object in C is a contravariant functor Y :
op
n C, and the category of such objects (functors and natural transformations
between them) will be denoted by sn C.
The n-truncation functor sC sn C defined by X 7 in X has a left adjoint

in : sn C sC, on account of the exactness assumption on the category C.

Explicitly, the theory of left Kan extensions dictates that, for an n-truncated
object Y , in Ym should be defined by
in Ym =

lim
Yi .

m
i, in
As the notation indicates, the colimit is defined on the finite category whose objects are ordinal number morphisms : m i with i n, and whose morphisms
: are commutative diagrams

i
6
44
44

u

j
in the ordinal number category. The simplicial structure map : in Ym in Yk
is defined on the index category level in the obvious way, by precomposition with
the morphism : k m.
The functor Y 7 in Y is left adjoint to the n-truncation functor: this can be
seen by invoking the theory of Kan extensions, or directly.
If m n, then the index category of arrows m i, i n, has an initial
object, namely 1m : m m, so that the canonical map
in1m

Ym

lim
Yi

m
i, in
is an isomorphism by formal nonsense. Furthermore, maps of this form in C are
the components of the adjunction map

Y
in in Y,
so that this map is an isomorphism of sn C.
The objects in Ym , m > n, require further analysis. The general statement
that is of the most use is the following:

1. Skeleta

267

Lemma 1.1. There is a coequalizer diagram

G

n
G

Yn1

i<j

Yn
in Yn+1 ,

i=0

where the maps in the coequalizer are defined by the commutativity of the following diagram:

Yn1

si

w Yn A

G u
Yn1
i<j

Yn

w in Yn+1 .

i=0

ini<j
Yn1

n
G

insj
A
A
C
A

inj

ini<j

ini
w Yn

sj1

Proof: Write D for the category of ordinal number morphisms : n + 1 j,

j n. Suppose that t : n + 1 i is an ordinal number epimorphism, where
i n, and write Dt for the category of ordinal number morphisms : n + 1 j,
j n, which factor through t. Then Dt has an initial object, namely t, so that
the canonical map int induces an isomorphism
int

Yi

lim
Yj

n+1
j Dt

Furthermore, if t has a factorization

n + 1

t
4
m

wi
4
6
4s

268

V. Simplicial groups

where r and s are ordinal number epimorphisms, the inclusion Dt Dr induces

a morphism s of colimits which fits into a commutative diagram
s

Yi
int
=
u
lim
Yj

n+1j Dt

w Ym

= inr
u
lim
Yj .

n+1j Dr

Write Dj for the category Dsj , 0 j n.

For i < j, the diagram

n+1
(1.2)

sj

wn
si

si

u
n

u
w n1

sj1

is a pushout in the ordinal number category: this is checked by fiddling with

simplicial identities. Now, suppose given a collection of maps
fj :

lim
Yi X,

n+1
i Dj

0 j n, such that the diagrams

lim

Yi

sj1
w

n+1i Dt

(1.3)

n+1i Dsi

si
u
lim

n+1i Dsj

lim

Yi

fj

fi
u
wX

Yi

1. Skeleta

269

commute, where t = si sj = sj1 si . Let : n + 1 k be an object of D. Then

Di for some i, and we define a morphism f : Yk X to be the composite
in

Yk

lim

fi

Yk X.

n+1k Di

Its easily seen, using the pushout diagram (1.2) and the commutativity conditions (1.3), that the definition of f is independent of i. The collection of maps
f , D, determine a unique map
f :

lim

Yk

n+1k D

which restricts to the maps fi , for 0 i n, and the lemma is proved.

Write skn Y = in in Y , and write : skn Y Y for the counit of the adjunction. skn Y is the n-skeleton of Y .
Lemma 1.4. Let Y be a simplicial object in the category C, and suppose that
there is a morphism f : N Yn+1 such that the canonical map : skn Y Y
and f together induce an isomorphism
skn Yn+1

(,f )

N Yn+1 .

Then an extension of a map g : skn Y Z to a map g 0 : skn+1 Y Z corresponds to a map g : N Zn+1 such that di g = gdi f for 0 i n + 1.
Proof: Given such a map g, define a map
G
g 0 : Yn+1
= skn Yn+1 N Zn+1
by g 0 = (g, g). In effect, we are looking to extend a map g : in Y in Z to
a map g 0 : i(n+1) Y i(n+1) Z. The truncated map g 0 will be the map g 0 in
degree n + 1 and will coincide with the map g in degrees below n + 1, once we
show that g 0 respects simplicial identities in the sense that the following diagram
commutes:
Yn+1
u

u
Ym

g0

w Zn+1
u

u
w Zm

270

V. Simplicial groups

for all ordinal number maps : m n + 1 and : n + 1 m, where m < n+1.

The canonical map : skn Y Y consists of isomorphisms

skn Yi Yi

in degrees i n, so that : Ym Yn+1 factors through the map : skn Yn+1

Yn+1 ; the restriction of g 0 to skn Yn+1 is a piece of a simplicial map, so that g 0
respects . The map factors through some face map di , so its enough to
show that g 0 respects the face maps, but this is automatic on skn Yn+1 and is an
assumption on g.
The converse is obvious.

Lemma 1.5. Supppose that i : A B is a morphism of sn+1 C which is an

isomorphism in degrees j n. Suppose further that there is a morphism f :
N Bn+1 such that the maps i and f together determine an isomorphism
An+1

(i,f )

N Bn+1 .

Suppose that g : A Z is a morphism of sn+1 C. Then extensions

g

A
i

wZ

j
hh
hg0

u h
B

of the morphism g to morphisms g 0 : B Z are in one to one correspondence

with morphisms g : N Zn+1 of C such that di g = gdi f for 0 i n.
Proof: This lemma is an abstraction of the previous result. The proof is the
same.

A morphism j : G H of simplicial groups is said to be almost free if there is

a contravariant set-valued functor X defined on the epimorphisms of the ordinal
number category such that there are isomorphisms
n

Gn F (Xn ) Hn

which
(1) are compatible with the map j in the sense that n inGn = jn for all n,
and

1. Skeleta

271

(2) respect the functorial structure of X in the sense that the diagram

Gn F (Xn )

t F (t )

w Hn

u
Gm F (Xm )

u
w Hm

commutes for every ordinal number epimorphism t : m n.

The n-skeleton skn (j) of the simplicial group homomorphism j : G H is
defined by the pushout diagram

skn G

u
G

w skn H
u
w skn (j)

in the category of groups. There are maps skn (j) skn+1 (j) and morphisms
skn (i) H such that the diagrams
G
A

A
A
D
w skn+1 (j)
skn (j)

N
N
Q
H

G'

'

and

w skn (j)

'
)
'
j '
u
H

commute, and such that j : G H is a filtered colimit of the maps G skn (j)
in the category of simplicial groups under G. The maps skn (j)i Hi are
group isomorphisms for i n, so the map skn (j) skn+1 (j) consists of group
isomorphisms in degrees up to n.

272

V. Simplicial groups

Write DXn for the degenerate part of Xn+1 . This subset can be described (as
usual) as the union of the images of the functions si : Xn Xn+1 , 0 i n.
For i < j the diagram of group homomorphisms

Hn1
(1.6)

si

w Hn
sj

sj1
u
Hn

u
si

w Hn+1

is a pullback, by manipulating the simplicial identities. Pullback diagrams are

closed under retraction, so the diagram of group homomorphisms

F (Xn1 )
(1.7)

si

sj1
u
F (Xn )

si

w F (Xn )
sj
u
w F (Xn+1 )

is also a pullback. All the homomorphisms in (1.7) are monomorphisms (since

they are retracts of such), so an argument on reduced words shows that (1.7)
restricts on generators to a pullback

Xn1
(1.8)

si

w Xn
sj

sj1
u
Xn

u
si

w Xn+1

in the set category. It follows that the degenerate part DXn of the set Xn+1 can
be defined by a coequalizer
G
i<j

Xn1

n
G
i=0

Xn ,
DXn

1. Skeleta

273

such as one would expect if X were part of the data for a simplicial set, in which
case DXn would be a copy of skn Xn+1 .
It therefore follows from Lemma 1.1 that the diagram of group homomorphisms
skn Gn+1

w skn Hn+1

u
w skn (i)n+1

Gn+1

can be identified up to canonical isomorphism with the diagram

skn Gn+1

w skn Gn+1 F (DXn )

u
w Gn+1 F (DXn ).

Gn+1

The map skn (i)n+1 skn+1 (i)n+1 can therefore be identified up to isomorphism
with the monomorphism
Gn+1 F (DXn ) Gn+1 F (Xn+1 )
which is induced by the inclusion DXn Xn+1 .
Let N Xn+1 = Xn+1 DXn be the non-degenerate part of Xn+1 . The truncation at level n + 1 of the map skn (j) skn+1 (j) is an isomorphism in degrees
up to n, and is one of the components of an isomorphism
skn (j)n+1 F (N Xn+1 )
= skn+1 (j)n+1 .
in degree n + 1.
Proposition 1.9. Suppose that j : G H is an almost free simplicial group
homomorphism, with H generated over G by the functor X as described above.
Let N Xn+1 be the non-degenerate part of Xn+1 . Then there is pushout diagram
of simplicial groups of the form

F (n+1 )

w skn (j)

u
F (n+1 )

xN Xn+1

xN Xn+1

w skn+1 (j)

274

V. Simplicial groups

for each n 1.
Corollary 1.10. Any almost free simplicial group homomorphism j : G H
is a cofibration of simplicial groups.
Proof of Proposition 1.9: Any commutative diagram

xN Xn+1

F (n+1 )

w skn (j)

xN Xn+1

u
F (n+1 )

u
wK

of simplicial group homomorphisms uniquely determines an morphism of (n+1)truncations : i(n+1) skn+1 (j) i(n+1) K which makes the following diagram
commute:

i(n+1) G
A

jA
A

A
A
D
A
i(n+1) H u
=

i
(n+1) (|G )
i(n+1) skn (j)N
N

P
i(n+1) NN
u
w i(n+1) K
i(n+1) skn+1 (j)

The homomorphisms |G : G K and therefore uniquely determine a morphism : skn+1 (j) K.

2. Principal fibrations I: simplicial G-spaces.

A principal fibration is one in which the fibre is a simplicial group acting in
a particular way on the total space. They will be defined completely below and
we will classify them, but it simplifies the discussion considerably if we discuss
more general actions first.

2. Principal fibrations I: simplicial G-spaces

275

Definition 2.1. Let G be a simplicial group and X a simplicial set. Then G

acts on X if there is a morphism of simplicial sets
:GX X
so that the following diagrams commute:

GGX

w GX

m1

u
GX

u
wX

and
X
1
X

iu
GX w X
where m is the multiplication in G and i(X) = (e, X).
In other words, at each level, Xn is a Gn -set and the actions are compatible
with the face and degeneracy operators.
Let SG be the category of simplicial sets with G-action, hereinafter known as
G-spaces. Note that SG is a simplicial category. Indeed, if K S, then K can
be given the trivial G-action. Then for X SG set
(2.2.1)

X K =X K

with diagonal action,

(2.2.2)

homSG (K, X) = HomS (K, X)

with action in the target, and for X and Y in SG ,

(2.2.3)

HomSG (X, Y )n = homSG (X n , Y ).

Then the preliminary result is:

276

V. Simplicial groups

Theorem 2.3. There is a simplicial model category structure on SG such that

f : X Y is
1) a weak equivalence if and only if f is a weak equivalence in S;
2) a fibration if and only if f is a fibration in S; and
3) a cofibration if and only if f has the left lifting property with respect to
all trivial fibrations.
Proof: The forgetful functor SG S has a left adjoint given by
X 7 G X.
Thus we can apply Theorem II.7.8 once we show that every cofibration having
the left lifting property with respect to all fibrations is a weak equivalence. Every
j
q
morphism X Y can be factored as X Z X where q is a fibration and j
is obtained by setting Z = lim Zn with Z0 = X and Xn defined by a pushout
n
diagram
G

G nk

in u
G n

w Zn1
jn
u
w Zn

where runs over all diagrams in SG

G nk
in u
G

w Zn1
u
w Y.

Since in is a trivial cofibration in S, we have that jn a trivial cofibration in S

(and also in SG ). So j : X Z is a trivial cofibration in S (and SG ).

A crucial structural fact about SG is the following:

Lemma 2.4. Let f : X Y be a cofibration in SG . Then f is an inclusion and
at each level Yk f (Xk ) is a free Gk -set.

2. Principal fibrations I: simplicial G-spaces

277

Proof: Every cofibration is a retract of a cofibration j : X Z where Z =

lim Zn and Zn is defined recursively by setting Z0 = X and defining Zn by a

pushout diagram
G
G n
w Zn1

u
u

G n

jn

w Zn .

So it is sufficient to prove the result for these more specialized cofibrations. Now
each jn is an inclusion, so j : X Z is an inclusion. Also, at each level, we have
a formula for k simplices
(Zn )k (Zn1 )k = (q G n )k (q G n )k
is free. Hence
(Z)k (Xk ) =

[
(Zn )k (Zn1 )k
n

is free.

For X SG , let X/G be the quotient space by the G-action. Let q : X X/G
be the quotient map. If X SG is cofibrant this map has special properties.
Lemma 2.5. Let X SG have the property that Xn is a free Gn set for all n.
Let x (X/G)n be an n-simplex. If fx : n X represents x, define Fx by the
pullback diagram
Fx
u
n

fx

wX
uq
w X/G.

Then for every z X so that q(z) = x, there is an isomorphsim in SG

z : G n Fx
so that the following diagram commutes

G n z w F x
2
u
u
n
n

= w .

278

V. Simplicial groups

Proof: First note that there is a natural G-action on Fx so that Fx X is a

morphism of G-spaces. Fix z Xn so that q(z) = x. Now every element of n
can be written uniquely as cn where cn nn is the canonical n-simplex and
: m n is an ordinal number map. Define z by the formula, for g Gm :
z (g, cn ) = ( cn , g z).
One must check this is a simplicial G-map. Having done so, diagram (1.5.1)
obviously commutes, so we need only check z is a bijection.
To see z is onto, for fixed (a, b) Fx one has fx a = q(b). We can write
a = cn for some , so
fx a = fx cn = x = q z
so b is in the same orbit as z, as required.
To see z is one-to-one, suppose
( cn , g z) = ( cn , h z).
Then = and, hence, g z = h z. By Lemma 5.4, the action is level-wise
free, so g = h.

Corollary 2.6. Let X SG have the property that each Xn is a free Gn set.
The quotient map q : X X/G is a fibration in S. It is a minimal fibration if
G is minimal as a Kan complex.
Proof: Consider a lifting problem

nk
wX
y
O
P q
O
u O
u
n

w X/G.
This is equivalent to a lifting problem

nk
w Fx
y
P
O
u O
u
n
= w n .

2. Principal fibrations I: simplicial G-spaces

279

By Lemma 2.5, this is equivalent to a lifting problem

nk
w G n
y
C j
B
u BB
u
n
n

= w .
Because G is fibrant in S (Lemma I.3.4), j is a fibration, so the problem has
a solution. If G is minimal, the lifting has the requisite uniqueness property to
make q a minimal fibration (see Section I.10).

Lemma 2.7. Let X SG have the property that each Xn is a free Gn set. Then
X = lim Xn where X1 = and for each n 0 there is a pushout diagram

n G

w Xn1

u
w Xn

where runs over the non-degenerate n-simplices of X/G.

Proof: Define Xn by the pullback diagram
Xn
u
skn (X/G)

wX
u
w X/G.

Then X1 = and lim Xn = X. Also, since there is a pushout diagram

w skn1 (X/G)

Gu

u
w skn (X/G)

280

V. Simplicial groups

there is a diagram
G

F ()|

w Xn1

(2.8)

G u
F ()

u
w Xn ,

where F () is defined to be the pullback along . But Lemma 2.5 rewrites

F ()
= n G and, hence, F ()|n
= n G.
Thus we need only show that the diagram (2.8) is a pushout diagram. To do
this let Y be the pushout. Then Y is a free G-space (since Xn1 is) and there
is a map of G-spaces Y Xn . Since taking orbits is a left adjoint it preserves
pushouts; therefore, Y /G Xn /G is an isomorphism and the result follows from
the next lemma.

One can alternatively finish the argument for Lemma 2.7 by observing that
pullbacks and pushouts in the category of G-spaces are formed by the corresponding construction in the underlying simplicial set category, and pullbacks
preserve pushouts.
Lemma 2.9. Suppose given a morphism in SG f : Y X so that
1) Xn is a free Gn set for all n
2) the induced map Y /G X/G is an isomorphism
then f is an isomorphism.
Proof: This is a variation on the proof of the 5-lemma. To show f is onto,
choose z X. Let qX : X X/G and qY : Y Y /G be the quotient maps.
Then there is a w Y /G so that (f /G)(w) = qX (z). Let y Y be so that
qY (y) = w. Then there is a g G so that gf (y) = f (gy) = z. To show f is
one-to-one suppose f (y1 ) = f (y2 ). Then qX f (y1 ) = qX f (y2 ) so qY (y1 ) = qY (y2 )
or there is a g G so that gy1 = y2 . Then
gf (y1 ) = f (y2 ) = f (y1 )
Since X is free at each level, g = e, so y1 = y2 .

Corollary 2.10. An object X SG is cofibrant if and only if Xn is a free Gn

set for all n.

3. Principal fibrations II: classifications

281

Proof: One implication is Lemma 2.4. The other is a consequence of Lemma

2.7.

3. Principal fibrations II: classifications.

In this section we will define and classify principal fibrations. Let G be a fixed
simplicial group.
Definition 3.1. A principal fibration f : E B is a fibration in SG so that
1) B has trivial G-action;
2) E is a cofibrant G-space; and
3) the induced map E/G B is an isomorphism.
Put another way, f : E B is isomorphic to a quotient map
q : X X/G
where X SG is cofibrant. Such a map q is automatically a fibration by Corollary 2.6. Cofibrant objects can be recognized by Corollary 2.10, and Lemma 2.5
should be regarded as a local triviality condition. Finally, there is a diagram

GE
f u

wE
uf

B w B
=
where is the action; such diagrams figure in the topological definition of principal fibration.
In the same vein, it is quite common to say that a principal G-fibration is a
G-bundle .
Definition 3.2. Two principal fibrations f1 : E1 B and f2 : E2 B will
be called isomorphic if there is an isomorphism g : E1 E2 of G-spaces making
the diagram commute
E1 4
6
4
f1

w E2
h
kf2
h

B
Remark 3.3. By Lemma 2.9 it is sufficient to construct a G-equivariant map
g : E1 E2 making the diagram commute. Then g is automatically an isomorphism.

282

V. Simplicial groups

Let P FG (B) be the set of isomorphism class of principal fibrations over B.

The purpose of this section is to classify this set.
To begin with, note that P FG () is a contravariant functor. If q : E B is a
principal fibration and f : B 0 B is any map of spaces, and if q 0 : E(f ) B 0
is defined by the pullback diagram

E(f )
q0 u
B0

wE
q
u
w B,

then f 0 is a principal fibration. Indeed

E(g) = {(b, e) B 0 E | f (b) = q(e)}
has G action given by g(b, e) = (b, ge). Then parts 1) and 3) of Definition 3.1
are obvious and part 2) follows from Corollary 2.10.
But, in fact, P FG () is a homotopy functor. Recall that two maps f0 , f1 :
0
B B are simplicially homotopic if there is a diagram

B0

B 0 N d t d w B 0 1
N

N
P
f0 t f1

Lemma 3.4. If f0 and f1 are simplicially homotopic, P FG (f0 ) = P FG (f1 ).

Proof: It is sufficient to show that given q : E B a principal fibration, the
pullbacks E(f0 ) B 0 and E(f1 ) B 0 are isomorphic. For this it is sufficient
to consider the universal example: given a principal fibration E B 1 , the
pullbacks E(d0 ) B and E(d1 ) B are isomorphic. For this consider the
lifting problem in SG
E(d0 )

C
B
BB

d0 u
0

E(d )

wE

u
w B 1 .

3. Principal fibrations II: classifications

283

Since E(d0 ) is cofibrant in SG , d0 is a trivial cofibration, so the lifting exists and

by Lemma 1.8 defines an isomorphism of principal fibrations

E(d0 ) 1 A
C
A

wE

B 1
Pulling back this diagram along d1 gives the desired isomorphism.
A similar sort of argument proves the following lemma:

Lemma 3.5. Let B S be contractible. Then any principal fibration over B is

isomorphic to 2 : G B B.
Proof: The isomorphism is given by lifting in the diagram (in SG ).
G
wE
P u
j u OO
GB w B
2

Here j is induced by any basepoint B; since G is cofibrant in SG , j is a

trivial cofibration in SG .

We can now define the classifying object for principal fibrations.

Definition 3.6. Let EG SG be any cofibrant object so that the unique
map EG is a fibration and a weak equivalence. Let BG = EG/G and
q : EG BG the resulting principal fibration.
Note that EG is unique up to equivariant homotopy equivalence, so q : EG
BG is unique up to homotopy equivalence.
In other words we require more than that EG be a free contractible G-space;
EG must also be fibrant. The extra condition is important for the proof of
Theorem 3.9 below. It also makes the following result true.
Lemma 3.7. The space BG is fibrant as a simplicial set.
Proof: First note that given any map f : nk BG, then the pullback E(f )
nk of EG BG fits into a diagram of principal fibrations

nk G A =
A
C
nk .

w E(f )

284

V. Simplicial groups

Thus there is a lifting of f to a map g : nk EG. Now suppose given an

extension problem
nk

w BG
P .
O
O

u O
n

Then, since EG is fibrant there is a factoring

nk

w EG
h
j
u hh
n
h

and the composite n EG BG solves the original extension problem.

Note that the same argument proves that if E SG is cofibrant and fibrant,
the resulting principal fibration E E/G has fibrant base. This claim and
Lemma 3.7 are alternatively consequences of the following:
Exercise 3.8. Suppose that p : X Y is a surjective fibration of simplicial
sets, and that X is fibrant. Show that Y is fibrant.
We now come to the main result.
Theorem 3.9. For all spaces B S, the map
: [B, BG] P FG (B)
sending the class [f ] [B, BG] to the pullback of EG BG along f is a
bijection.
Here, [B, BG] denotes morphisms in the homotopy category Ho(S) from B to
BG. The space BG is fibrant, so this morphism set cat be identified with the
set of simplicial homotopy classes of maps from B to BG.
Proof: Note that is well-defined by Lemma 3.4. To prove the result we
construct an inverse. If q : E B is a principal fibration, there is a lifting in
the diagram in SG

(3.10)

w EG
j
h
u h
u
w
E

4. Universal cocycles and W G

285

since E is cofibrant and EG is fibrant, and this lifting is unique up to equivariant

homotopy. Let f : B BG be the quotient map. Define : P FG (B)
[B, BG], by sending q : E B to the class of f .
Note that if E(f ) is the pullback of f , there is a diagram
EA
A
C

w E(f )

so Lemma 2.9 implies = 1. On the other hand, given a representative g :

B BG of a homotopy class in [B, BG], the map g 0 in the diagram
E(g)
u
B

g0

w EG

u
g w BG

makes the diagram (3.10) commute, so by the homotopy uniqueness of liftings

= 1.

4. Universal cocycles and W G.

In the previous sections, we took a simplicial group G and assigned to it a
homotopy type BG; that is, the space BG depended on a choice EG of a fibrant,
cofibrant contractible G-space. In this section we give a natural, canonical choice
for EG and BG called, respectively, W G and W G. The spaces W G and W G
are classically defined by letting W G be the simplicial set with
W Gn = Gn Gn1 G0
and
di (gn , gn1 , . . . g0 )

(di gn , di1 gn1 , . . . , d0 gni gni1 , gni2 , ...g0 )

=
(dn gn , dn1 gn1 , . . . d1 g1 )

i < n,
i = n.

si (gn , g(n1) . . . , g0 ) = (si gn , si1 gn1 . . . s0 gn1 , e, gni1 , . . . g0 )

where e is always the unit. Note that W G becomes a G-space if we define
G W G W G by:
(h, (gn , gn1 , . . . , g0 ) (hg, gn1 , . . . g0 ).
Then W G is the quotient of W G by the left G-action; write q = qG : W G W G
for the quotient map.

286

V. Simplicial groups

Lemma 4.1. The map q : W G W G is a fibration.

Proof: This follows from Corollary 1.6 since (W G)n is a free Gn set.

The functor G 7 W G takes values in the category S0 of reduced simplicial

sets, where a reduced simplicial set is a simplicial set having only one vertex.
The salient deeper feature of the functor W : sGr S0 is that it has a left
adjoint G : S0 sGr, called the loop group functor, such that the canonical
maps G(W G) G and X W (GX) are weak equivalences for all simplicial
groups G and reduced simplicial sets X. A demonstration of these assertions
will occupy this section and the following two. These results are originally due
to Kan, and have been known since the late 1950s. The original proofs were
calculational we recast them in modern terms here. Kans original geometric
insights survive and are perhaps sharpened, in the presence of the introduction
of a closed model structure for reduced simplicial sets and a theory of simplicial
cocycles.
A segment of an ordinal number n is an ordinal number monomorphism
n j , n which is defined by i 7 i + j. This map can also be variously
characterized as the unique monomorphism n j , n which takes 0 to j, or as
the map (d0 )j . This map will also be denoted by [j, n], as a means of identifying
its image. There is a commutative diagram of ordinal number maps
n k hh[k, n]
hh
j

n
)
'
'
u ''
[j, n]
nj
if and only if j k. The map is uniquely determined and must be a segment
map if it exists: its the map (d0 )kj . Thus, we obtain a poset Seg(n) of segments
of the ordinal number n. This poset is plainly isomorphic to the poset opposite
to the ordinal n.
Suppose that G is a simplicial group. An n-cocycle f : Seg(n) G associates
to each relation : [k, n] [j, n] in Seg(n) an element f ( ) Gnk , such that
the following conditions hold:
(1) f (1j ) = e Gnj , where 1j is the identity relation [j, n] [j, n],

(2) for any composeable pair of relations [l, n]

[k, n]
[j, n], there is an
equation
(f ( ))f () = f ( ).

4. Universal cocycles and W G

287

Any ordinal number map : r s has a unique factorization

ws

N
P
N
0 (0)
N [(0), s] = (d )
s (0)

where is an ordinal number map such that (0) = 0. It follows that any
relation : [k, m] [j, m] in Seg(m) induces a commutative diagram of ordinal
number maps
k

m k

(4.2)

[k, m]

w n (k)

mj

[
[
[j,
m]
[
^
u [
m

w n (j)
N

N
N
N
u N
Q [(j), n]
wn

[(k), n]

where the maps j and k take 0 to 0. Given an n-cocyle f : Seg(n) G, define,

for each relation : [k, m] [j, m] in Seg(m), an element (f )( ) Gmk by
(f )( ) = k (f ( )).
Its not hard to see now that the collection of all such elements (f )( ) defines
an m-cocycle (f ) : Seg(m) G, and that the assignment 7 is contravariantly functorial in ordinal maps . We have therefore constructed a simplicial set
whose n-simplices are the n-cocycles Seg(n) G, and whose simplicial structure
maps are the induced maps .
This simplicial set of G-cocycles is W G . This claim is checked by chasing the
definition through faces and degeneracies, while keeping in mind the observation
that an n-cocycle f : Seg(n) G is completely determined by the string of
relations
(4.3)

n2

n1

[n, n] [n 1, n] . . . [1, n] [0, n],

288

V. Simplicial groups

and the corresponding element

(f (n1 ), f (n2 ), . . . , f (0 )) Gn1 Gn2 G0 .
Of course, each i is an instance of the map d0 .
The identification of the simplicial set of G-cocycles with W G leads to a
global description of the simplicial structure of W G. Suppose that : m n
is an ordinal number map, and let
g = (gn1 , gn2 , . . . , g0 )
be an element of Gn1 Gn2 G0 . Let Fg be the cocycle Seg(n) G
associated to the n-tuple g. Then, subject to the notation appearing in diagram
(4.2), we have the relation

(gn1 , gn2 , . . . , g0 ) = (1 Fg (m1 ), 2 Fg (m2 ), . . . , m

Fg (0 )),

where mi = (d0 )(i)(i1) is the induced relation [(i), n] [(i 1), n] of

Seg(n).
A simplicial map f : X W G, from this point of view, assigns to each nsimplex x a cocycle f (x) : Seg(n) G, such that for each ordinal number map
: m n and each map : [k, m] [j, m] in Seg(m) there is a relation
k f (x)( ) = f ( (x))( ).
Any element j n determines a unique diagram
nu [[ [0, n] = 1
[[[
]
[j, n]
n
)
'
'
'
' [j, n]
nj
and hence unambiguously gives rise to elements
f (x)([j, n]) Gnj .
Observe further that if j k and : [k, n] [j, n] denotes the corresponding
relation in Seg(n), then the cocyle condition for the composite

[j,n]

[k, n]
[j, n] [0, n]

4. Universal cocycles and W G

289

can be rephrased as the relation

(f (x)([j, n])) = f (x)([k, n])f (x)( )1 .
Now, given a map (cocycle) f : X W G, and an ordinal number map
: m n, there is an induced function
: Gn Xn Gm Xm ,
which is defined by
(4.4)

(g, x) 7 ( (g)0 (f (x)([(0), n])), (x)),

where 0 : m n (0) is the unique ordinal number map such that [(0), n]
0 = .
Lemma 4.5. The maps defined in (4.4) are functorial in ordinal number maps
.
Proof: Suppose given ordinal number maps

k
m
n,
and form the diagram

k hh
0
u

hh
hh
hh
h
j

m (0)

[(0), m]

(0)

u
n ((0))

w m
u

[(0), m]

w n (0)

[(0), n]

wn

in the ordinal number category. In order to show that (g, x) = () (g, x)

in Gk Xk , we must show that

0 (f (x)([(0), n]))0 (f ( (x))([(0), m])) = 0 (0)

(f (x)([((0)), n]))

290

V. Simplicial groups

in Gk . But

0 = 0 (0)
[(0), m] ,

and
[(0), m] (f (x)[(0), n]) = f (x)([(0), n])(f (x)([(0), m] ))1
by the cocycle condition. Finally,

(f (x)([(0), m] )) = f ( (x))([(0), m]),

(0)

since f is a simplicial map. The desired result follows.

The simplicial set constructed in Lemma 4.5 from the map f : X W G will
be denoted by Xf . The projection maps Gn Xn Xn define a simplicial
map : Xf X, and this map obviously has the structure of a G-bundle.
This is a natural construction: if h : Y X is a simplicial set map, then the
maps Gn Yn Gn Xn defined by (g, y) 7 (g, h(y)) define a G-equivariant
simplicial set map h : Yf h Xf such that the diagram

Yf h

w Xf

u
Y

u
wX

commutes. Furthermore, this diagram is easily seen to be a pullback.

The simplicial set W G1 associated to the identity map 1 : W G W G is W G,
and the G-bundle : W G W G is called the canonical G-bundle.
Lemma 4.6. W G is contractible.
Proof: Suppose given an element (gn , (gn1 , . . . , g0 )) W Gn . Then the (n +
2)-tuple (e, (gn , gn1 , . . . g0 )) defines an element of W Gn+1 , in such a way that
the following diagram of simplicial set maps commutes:

n+1
u [[
0

[[[ (e, (g , g
n n1 , . . . , g0 ))
[[[
[[[
]
[

(gn , (gn1 , . . . , g0 ))

w WG

5. The loop group construction

291

commutes. Furthermore, if : m n is an ordinal number map, and :

m + 1 n + 1 is the unique map such that (0) = 0 and d0 = d0 , then
(e, (gn , gn1 , . . . , g0 )) = (e, (gn , (gn1 , . . . , g0 )),
as is easily checked. It follows that the simplices (e, gn , . . . , g0 ) define an extra
degeneracy on W G in the sense of Section III.5, and so Lemma III.5.1 implies
that W G is contractible.

Remark 4.7. Every principal G-fibration p : Y X is isomorphic to a principal

fibration Xf X for some map f : X W G. In effect, let denote the
subcategory of the category consisting of all ordinal number morphisms :
m n such that (0) = 0. Then the map p restricts to a natural transformation
p : Y | X| , and this transformation has a section : X| X|
in the category of contravariant functors on , essentially since the simplicial
map p is a surjective Kan fibration. Classically, the map is called a pseudo
cross-section for the bundle p. The pseudo cross-section defines Gn -equivariant
isomorphisms
n : Gn Xn
= Yn
given by (g, x) 7 g (x). If : n k n j is a morphism of Seg(n) then
((dj0 x)) = fx ( )( dk0 x)
for some unique element fx ( ) Gnk . The elements fx ( ) define a cocycle
fx : Seg(n) G for each simplex x of X, and the collection of cocycles fx , x X,
defines a simplicial map f : X W G such that Y is G-equivariantly isomorphic
to Xf over X via the maps n . The classical approach to the classification of
principal G-bundles is based on this construction, albeit not in these terms.
5. The loop group construction.
Suppose that f : X W G is a simplicial set map, and let x Xn be an nsimplex of X. Recall that the associated cocycle f (x) : Seg(n) G is completely
determined by the group elements
f (x)(d0 : (d0 )k+1 (d0 )k ).
On the other hand,
f (x)(d0 : (d0 )k+1 (d0 )k ) = f (dk0 (x))(d0 : d0 1nk ).
It follows that the simplicial map f : X W G is determined by the elements
f (x)(d0 ) = f (x)(d0 1n ) Gn1 ,

292

V. Simplicial groups

for x Xn , n 1. Note in particular that f (s0 x)(d0 ) = e Gn1 .

Turning this around, suppose that x Xn+1 , and the ordinal number map
: m n has the factorization
n (0)
j
0 hh
h
h
m

u
w n,

where 0 (0) = 0 and is a segment map, and suppose that d0 : d0 1n+1 is

the inclusion in Seg(n + 1). Then
f (x)(d0 ) = f (x)(d0 )(f (d0 (x))( ))1 .
by the cocycle condition for f (x), and so
f (x)(d0 ) = 0 (f (x)(d0 ))
= 0 (f (x)(d0 ))0 (f (d0 (x))( ))1
= f ( (x))(d0 )(f ((c) (d0 (x)))(d0 ))1 ,
where : m + 1 n + 1 is defined by

(i) =
(i 1) + 1

if i = 0, and
if i 1,

and c : m + 1 n is the ordinal number map defined by (c)(0) = 0 and

(c)(i) = (i 1) for i 1. Observe that c = s0 .

Define a group GXn = F (Xn+1 )/s0 F (Xn ) for n 0, where F (Y ) denotes the
free group on a set Y . Note that GXn may also be described as the free group
on the set Xn+1 s0 Xn .
Given an ordinal number map : m n, define a group homomorphism
: GXn GXm on generators [x], x Xn+1 by specifying
(5.1)

([x]) = [ (x)][(c) (d0 (x))]1 .

If : k m is an ordinal number map which is composable with , then the

relations
(c) d0 (x) = (c) d0 (x)
= (c) d0 (c) d0 (x)

5. The loop group construction

and

293

(c) d0 (x) = (c()) d0 (x)

together imply that ([x]) = () ([x]) for all x Xn+1 , so that we have a
simplicial group, called the loop group of X, which will be denoted GX. This
construction is plainly functorial in simplicial sets X.
Each n-simplex x X gives rise to a string of elements
x]) GXn1 GXn2 GX0 ,
([x], [d0 x], [d20 x], . . . , [dn1
0
which together determine a cocycle Fx : Seg(n) GX. Suppose that : m n
is an ordinal number map such that (0) = 0. The game is now to obtain a
recognizable formula for [ x], in terms of the simplicial structure of GX.
Obviously, if (1) = (0), then [ x] = e GXm1 . Suppose that (1) > 0.
Then there is a commutative diagram of ordinal number maps
m1

w n (1)
0 (1)1
(d
)

d0
u
m

'n 1

u
wn

''
'
*' d0

If = (d0 )(1)1 1 , then = , and so

(1)1

[ (x)] = (1 d0

[x])[f () (d0 x)],

where f () is defined by f () = c. We have f ()(0) = 0 by construction, and

there is a commutative diagram
m1

w n (1)
(d0 )(1)1

d0

u
m

f ()

u
w n 1,

so an inductive argument on the exponent (1)1 implies that there is a relation

(1)2

[f () (d0 x)] = (1 d0

(1)1

[d0 x]) . . . (1 [d0

(x)]).

294

V. Simplicial groups

It follows that
(1)1

(5.2) [ (x)] = (1 d0

(1)2

[x])(1 d0

(1)1

[d0 x]) . . . (1 [d0

(x)]) = 1 (Fx ( )).

Lemma 5.3.
(a) The assignment
x 7 ([x], [d0 x], [d20 x], . . . , [d0n1 x])
defines a natural simplicial map : X W GX.
(b) The map is one of the canonical homomorphisms for an adjunction
homsGr (GX, H)
= homS (X, W H),
where sGr denotes the category of simplicial groups.
Proof:
(a) Suppose that : m n is an ordinal number map, and recall the decomposition of (4.2). It will suit us to observe once again that the map [j, m]
is the composite (d0 )j , and that = (d0 )(k)(j) . Note in particular that
= (d0 )(0) 0 , and recall that 0 (0) = 0. It is also clear that there is a commutative diagram
Xn

w W GXn

(0)

d0

(0)

d0

Xn(0)

u
w W GXn(0)

Let Fx be the cocycle Seg(n) GX associated to the element

([x], [d0 x], [d20 x], . . . , [dn1
x]).
0
Then, for x Xn ,
(0)

0 ([d0

(0)+1

x], [d0

x], . . . , [d0n1 x]) = (1 Fx (m1 ), . . . , m

Fx (0 ))
(0)

= ([0 d0

(1)

x], [1 d0

(m1)

x], . . . , [m1
d0

= ([ x], [d0 x], . . . , [dm1

x]),
0

x])

5. The loop group construction

295

where mi = (d0 )(i)(i1) as before, and this by repeated application of the

formula (5.2). In particular, is a simplicial set map. The naturality is obvious.
(b) Suppose that f : X W H is a simplicial set map, where H is a simplicial
group. Recall that the cocycle f (x) : Seg(n) H can be identified with the
element
(f (x)(d0 ), f (d0 x)(d0 ), . . . , f (dn1
x)(d0 )) Hn1 Hn2 H0 .
0
The simplicial structure for GX given by the formula (5.1) implies that f :
X W H induces a simplicial group map f : GX H which is specified on
generators by f ([x]) = fx (d0 ). It follows that the function
homsGr (GX, H) homS (X, W H)
defined by g 7 (W g) is surjective. Furthermore, any map f : X W H is
uniquely specified by the elements f (x)(d0 ), and hence by the simplicial group
homomorphism f .

Remark 5.4. Any simplicial group homomorphism f : G H induces a f equivariant morphism of associated principal fibrations of the form
f

G
u
WG
u
WG

Wf

Wf

wH
u
w WH
u
w W H,

as can be seen directly from the definitions. The canonical map : X W GX

induces a morphism
GX

GX

u
X

u
w W GX

u
X

u
w W GX

296

V. Simplicial groups

of GX-bundles. It follows that, for any simplicial group homomorphism f :

GX H, the map f and its adjoint f = W f fit into a morphism of bundles
GX

wH

u
X

u
w WH

u
X

u
w W H.

Suppose now that the simplicial set X is reduced in the sense that it has only
one vertex. A closed n-loop of length 2k in X is defined to be a string
(x2k , x2k1 , . . . , x2 , x1 )
of (n + 1)-simplices xj of X such that d0 x2i1 = d0 x2i for 0 i k. Define an
equivalence relation on loops by requiring that
(x2k , . . . , x1 ) (x2k , . . . , xi+2 , xi1 , . . . , x1 )
if xi = xi+1 . Let
hx2k , . . . , x1 i
denote the equivalence class of the loop (x2k , . . . , x1 ). Write G0 Xn for the set of
equivalence classes of n-loops under the relation . Loops may be concatenated,
giving G0 Xn the structure of a group having identity represented by the empty nloop. Any ordinal number morphism : m n induces a group homomorphism
: G0 Xn G0 Xm ,
which is defined by the assignment
hx2k , . . . , x2 , x1 i 7 h x2k , . . . x2 , x1 i.
The corresponding simplicial group will be denote by G0 X. This construction is
clearly functorial with respect to morphisms of reduced simplicial sets.
There is a homomorphism
n : G0 Xn GXn

5. The loop group construction

297

which is defined by
n hx2k , x2k1 , . . . , x2 , x1 i = [x2k ][x2k1 ]1 . . . [x2 ][x1 ]1 .
Observe that
([x2i ][x2i1 ]1 ) = [ (x2i )][(c) d0 (x2i )]1 [(c) d0 (x2i1 )][ (x2i1 ]1
= [ (x2i )][ (x2i1 ]1 ,
so that
([x2k ][x2k1 ]1 . . . [x2 ][x1 ]1 ) = [ x2k ][ x2k1 ]1 . . . [ x2 ][ x1 ]1 .
The homomorphisms n : G0 Xn GXn , taken together, therefore define a
simplicial group homomorphism : G0 X GX.
Lemma 5.5. The homomorphism : G0 X GX is an isomorphism of simplicial
groups which is natural with respect to morphisms of reduced simplicial sets X.
Proof: The homomorphism n : G0 Xn GXn has a section, which is defined
on generators by
[x] 7 hx, s0 d0 xi,
and elements of the form hx, s0 d0 xi generate G0 Xn .

Again, let X be a reduced simplicial set. The set E 0 Xn consists of equivalence

classes of strings of (n + 1)-simplices
(x2k , . . . , x1 , x0 )
with d0 x2i = d0 x2i1 , i 1, subject an equivalence relation generated by relations if the form
(x2k , . . . , x0 ) (x2k , . . . , xi+2 , xi1 , . . . , x0 )
if xi = xi+1 . We shall write hx2k , . . . , x0 i for the equivalence class containing
the element (x2k , . . . , x0 ). Any ordinal number map : m n determines a
function : E 0 Xn E 0 Xm , which is defined by
hx2k , . . . , x0 i = h (x2k ), . . . , (x0 )i,
and so we obtain a simplicial set E 0 X. Concatenation induces a left action
G0 X E 0 X E 0 X of the simplicial group G0 X on E 0 X.

298

V. Simplicial groups

There is a function
0n : E 0 Xn GXn Xn
which is defined by
0n hx2k , . . . , x1 , x0 i = ([x2k ][x2k1 ]1 . . . [x2 ][x1 ]1 [x0 ], d0 x0 ).
The function 0n is n -equivariant, and so
0n (1
n (g)hs0 xi) = (g, x)
for any (g, x) GXn Xn , and 0n is surjective. There is an equation
hx2k , . . . , x0 i = hx2k , . . . , x0 , s0 d0 (x0 )ihs0 d0 (x0 )i
for every element of E 0 Xn , so that E 0 Xn consists of G0 Xn -orbits of elements
hs0 xi. The function 0n preserves orbits and n is a bijection, so that 0n is
injective as well.
The set GXn Xn is the set of n-simplices of the GX-bundle X which is
associated to the natural map : X W GX. If : m n is an ordinal
number map, then the associated simplicial structure map in X has the form
([x2k ] . . . [x1 ]1 [x0 ], d0 x0 )
= ([ (x2k )] . . . [ (x1 )]1 [ (x0 )][(c) (d0 x0 )]1 ((x)([(0), n])), d0 (x0 ))
0

since d0 (x0 ) = (d0 x0 ). But

[(c) (d0 x0 ))] = 0 ((x)([(0), n]),
by equation (5.2). The bijections 0n therefore define a -equivariant simplicial
map, and so we have proved
Lemma 5.6. There is a -equivariant isomorphism
0 : E 0 X X .
This isomorphism is natural with respect to maps of reduced simplicial sets.

5. The loop group construction

299

There is a simplicial set E 00 X whose n-simplices consist of the strings of (n+1)simplices (x2k , . . . , x0 ) of X as above, and with simplicial structure maps defined
by
(x2k , . . . , x0 ) = ( x2k , . . . , x0 )
for : m n. Observe that E 0 X = E 00 X/ .
Given this description of the simplicial structure maps in E 00 X, the best way to
think of the members of an n-simplex is as a string (x2k , . . . , x0 ) of cones on their
0th faces, with the obvious incidence relations. A homotopy n 1 E 00 X
can therefore be identified with a string
(h2k , . . . , h1 , h0 ),
where
(1) hi : C(n 1 ) X is a map defined on the cone C(n 1 ) on the
simplicial set n 1 , and
(2)
h2i |n 1 = h2i1 |n 1
for 1 i k.
We shall say that maps of the form C(n 1 ) Y are cone homotopies.
Examples of such include the following:
(1) The canonical contracting homotopy

0
u
0

w0
u
w1

w
w

w0
u
w n+1

of n+1 onto the vertex 0 induces a map C(n 1 ) n+1 which is

jointly specified by the vertex 0 and the restricted homotopy
0
u
1

w0
u
w2

w
w

w0
u
w n + 1.

This map is a contracting cone homotopy.

300

V. Simplicial groups

(2) The vertex 0 and the constant homotopy

1
u
1

w2
u
w2

w n+1
u
w n + 1.

jointly specify a constant cone homotopy C(n 1 ) n+1 .

In both of these cases, its helpful to know that the cone CBP on the nerve
BP of a poset P can be identified with the nerve of the cone poset CP which
is obtained from P by adjoining a disjoint initial object. Furthermore, a poset
map : P Q can be extended to a map CP Q by mapping the initial
object of CP to some common lower bound of the objects in the image of , if
such a lower bound exists.
(E 0 X, Z) = 0.
Lemma 5.7. E 0 X is acyclic in the sense that H
Proof: Both the contracting and constant cone homotopies defined above are
natural in n in the sense that the diagram

C(m 1 )

w m+1

C( 1)
u
C(n 1 )

u
h

w n+1

commutes for each ordinal number map : m n, where h denotes one of the
two types. It follows that there is a homotopy from the identity map on E 00 X
to the map E 00 X E 00 X defined by
(x2k , . . . , x1 , x0 ) 7 (x2k , . . . , x1 , ),
and that this homotopy can be defined on the level of simplices by strings of
cone homotopies
(h(x2k ), . . . , h(x1 ), h(x0 )),
where h(x0 ) is contracting on d0 x0 , and all other h(xi ) are constant. This
homotopy, when composed with the canonical map E 00 X E 0 X, determines a

5. The loop group construction

301

chain homotopy S from the induced map ZE 00 X ZE 0 X to the map ZE 00 X

ZE 0 X which is induced by the simplicial set map defined by
(x2k , . . . , x1 , x0 ) 7 hx2k , . . . , x1 , i.
For each element (x2k , . . . , x1 , x0 ), the chain S(x2k , . . . , x1 , x0 ) is an alternating
sum of the simplices comprising the homotopy (h(x2k ), . . . , h(x1 ), h(x0 )). It
follows in particular, that if xi = xi+1 for some i 1, then the corresponding
adjacent simplices of the components of S(x2k , . . . , x1 , x0 ) are also equal.
It also follows that there is a chain homotopy defined by
(x2k , . . . , x1 , x0 ) 7 S(x2k , . . . , x1 , x0 ) S(x2k , . . . , x1 , x1 ),
and that this is a chain homotopy from the chain map induced by the canonical
map E 00 X E 0 X to the chain map induced by the simplicial set map
(x2k , . . . , x1 , x0 ) 7 hx2k , . . . , x3 , x2 i
This construction can be iterated, to produce a chain homotopy H defined by
k1
X

(x2k , . . . , x0 ) 7 (

(S(x2k , . . . , x2i+1 , x2i ) S(x2k , . . . , x2i+1 , x2i+1 ))) + S(x2k )

i=0

from the chain map ZE 00 X ZE 0 X to the chain map induced by the simplicial
set map E 00 X E 0 X which takes all simplices to the base point . One can
show that
H(x2k , . . . , x0 ) = H(x2k , . . . xi+2 , xi1 , . . . , x0 )
if xi = xi+1 . It follows that H induces a contracting chain homotopy on the
complex ZE 0 X.

Lemma 5.8. E 0 X is simply connected.

Proof: Following Lemma 5.6, we shall do a fundamental groupoid calculation
in X
= E 0 X.
The boundary of the 1-simplex (s0 g, x) in X has the form
(s0 g, x) = ((g[x], ), (g, )).
There is an oriented graph T (X) (hence a simplicial set) having vertices coinciding with the elements of GX0 and with edges x : g gx for x X1 {}. There
is plainly a simplicial set map T (X) X which is the identity on vertices and

302

V. Simplicial groups

sends each edge x : g gx to the 1-simplex (s0 g, x). This map induces a map
of fundamental groupoids
T (X) X
which is surjective on objects. A reduced word argument shows that T (X) is
contractible, hence has trivial fundamental groupoid, so we end up showing that
X is simply connected if we show that the 1-simplices (s0 g, x) generate the
fundamental groupoid X .
There are boundary relations
(s1 g, s0 x) = (d0 (s1 g, s0 x), d1 (s1 g, s0 x), d2 (s1 g, s0 x))
= (s0 d0 g, x), (g, x), (g, ))
and in the same notation,
(s0 g, y) = ((g[y], d0 y), (g, d1 y), (s0 d1 g, d2 y)).
The upshot is that there are commuting diagrams in X of the form
4

(s0 d1 g, d2 y)

w 4
h
4

(g[y], x)

w
4
h
j
h
4 (g[y], d0 y) h
4(g[y], )
h(s d (g[y]), x)
h
4
(g, d1 y) 46
4
6
4
h 0 0
h
h
k
h
u

(s0 d0 (g[y]), d0 y)
and
(s0 d1 (g[y]1 ), d2 y)

4
u

(g[y]1 , x)

w
h
j
h
h
h(s0 d0 (g[y]1 ), x)

4 (g[y]1 , d y) h
4(g[y]1 , )
1 h
4
4
(g, d0 y) 4
h
44
6
4
6
h
kh
h
h
u

(s0 d0 (g[y]1 ), d1 y)

It follows that any generator (g[y], x) (respectively (g[y]1 , x) of X can be

replaced by a generator (g, d1 y) (respectively (g, d0 y)) of X up to multiplication by elements of T (X). In particular, any generator (h, x) of X can be
replaced up to multiplication by elements of T (X) by a generator (h0 , x0 ) such
that h0 has strictly smaller word length as an element of the free group GX1 . An
induction on word length therefore shows that the groupoid X is generated
by the image of T (X).

6. Reduced simplicial sets, Milnors F K-construction

303

Remark 5.9. The object T (X) in the proof of Lemma 5.8 is the Serre tree
associated to the generating set X1 {} of the free group GX0 . See p.16 and
p.26 of [84].
We have proved
Theorem 5.10. Suppose that X is a reduced simplicial set. Then the total
space X of the principal GX-fibration X X is contractible.
Corollary 5.11. There are weak equivalences
'

'

GX X X P X X,
which are natural with respect to morphisms of reduced Kan complexes X.
6. Reduced simplicial sets, Milnors F K-construction.
The proof of Theorem 5.10 depends on an explicit geometric model for the
space X , and the construction of this model uses the assumption that the simplicial set X is reduced. There is no such restriction on the loop group functor:
GY is defined for all simplicial sets Y . The geometric model for X can be
expanded to more general simplicial sets (see Kans paper), but Theorem 5.10
fails badly in the non-reduced case: the loop group G(1 ) on the simplex 1 is
easily seen to be the constant simplicial group on the free group Z on one letter,
which is manifestly not contractible. This sort of example forces us (for the time
being see Section 8) to restrict our attention to spaces with one vertex.
We now turn to the model category aspects of the loop group and W functors.
Lemma 6.1. Let f : X Y be a cofibration of simplicial sets. Then Gf :
GX GY is a cofibration of simplicial groups. In particular, for all simplicial
sets X, GX is a cofibrant simplicial group.
Proof: This result is a consequence of Corollary 1.10.
Note that since s0 Xn Xn+1 there is an isomorphism of groups
F (Xn+1 s0 Xn ).
GXn =
Furthermore, for all i 0, the map si+1 : Xn Xn+1 restricts to a map
si+1 : Xn s0 Xn1 Xn+1 s0 Xn
since si+1 s0 X = s0 si X. Hence there is a diagram
GXn1
si
u
GXn

= w F (X s X )
n
0 n
F si+1

= w F (X
n+1 s0 Xn )

304

V. Simplicial groups

and GX is almost free, hence cofibrant. For the general case, if X Y is a

level-wise inclusion
Yn+1 s0 Yn = (Xn+1 s0 Xn ) Zn+1
where Zn+1 = Yn+1 (Xn+1 s0 Yn ). Thus
GYn
= GXn F Zn+1
where the denotes the free product. Now si+1 : Yn Yn+1 restricts to a map
si+1 : Zn Zn+1 and, hence, the inclusion GX GY is almost-free and a
cofibration.

As a result of Theorem 5.10 and Lemma 6.1 one sees that G preserves cofibrations and weak equivalences between spaces with one vertex. This suggests
that the proper domain category for G at least from a model category point
of view is the category S0 of simplicial sets with one vertex. Our next project
then is to give that category a closed model structure.
Proposition 6.2. The category S0 has a closed model category structure where
a morphism f : X Y is a
1) a weak equivalence if it is a weak equivalence as simplicial sets;
2) a cofibration if it is a cofibration as simplicial sets; and
3) a fibration if it has the right lifting property with respect to all trivial
cofibrations.
The proof is at the end of the section, after we explore some consequences.
Proposition 6.3.
1) The functor G : S0 sGr preserves cofibrations and weak equivalences.
2) The functor W : sGr S0 preserves fibrations and weak equivalences.
3) Let X S0 and G sGr. Then a morphism f : GX G is a weak equivalence in sGr if only if the adjoint f : X W G is a weak equivalence
in S0 .
Proof: Part 1) follows from Lemma 6.1 and Theorem 5.10. For part 2) notice
that since W is right adjoint to a functor which preserves trivial cofibrations, it
preserves fibrations. The clause about weak equivalences follows from Lemma 4.6
Finally, part 3), follows from Remark 5.4, Lemma 4.6 and Theorem 5.10.

6. Reduced simplicial sets, Milnors F K-construction

305

Corollary 6.4. Let Ho(S0 ) and Ho(sGr) denote the homotopy categories.
Then the functors G and W induce an equivalence of categories
Ho(S0 )
= Ho(sGr).
Proof: Proposition 6.3 implies that the natural maps : GW H H and
: X W GX are weak equivalences for all simplicial groups H and reduced
simplicial sets X.

Remark 6.5. If Ho(S)c Ho(S) is the full sub-category of the usual homotopy
category with objects the connected spaces, then the inclusion H0 (S0 ) H0 (S)c
is an equivalence of categories. To see this, it is sufficient to prove if X is
connected there is a Y weakly equivalent to X with a single vertex. One way is
to choose a weak equivalence X Z with Z fibrant and then let Y Z be a
minimal subcomplex weakly equivalent to Z.
We next relate the fibrations in S0 to the fibrations in S.
Lemma 6.6. Let f : X Y be a fibration in S0 . Then f is a fibration in S if
and only if f has the right lifting property with respect to
S 1 = 1 /1 .
Proof: First suppose f is a fibration in S. Consider a lifting problem

(6.7)

0
y
i
d
u
1

wX

u
u
w S1
w Y.

Since f is a fibration in S, there is a map g : 1 X solving the lifting problem

for the outer rectangle. Since X has one vertex g factors through the quotient
map,
g

1 1 / sk0 1 = S 1 X
and g solves the original lifting problem. Here sk0 denotes the zero skeleton.
Now suppose f has the stipulated lifting property. Then one must solve all
lifting problems
nk
y
u (
n

wX
(
)
(

u
w Y.

306

V. Simplicial groups

If n = 0, this is trivial since X0

= Y0 . If n > 1, this diagram can be expanded to

The map

nk
y

w nk / sk0 nk

u
n

(
u
n
n
w / sk0

wX
(
)
(

f
u
w Y.

nk / sk0 nk n / sk0 n

is still a trivial cofibration, now in S0 . So the lift exists. If n = 1, the expanded

diagram is an instance of diagram (6.7), and the lift exists by hypothesis.

Corollary 6.8. Let X S0 be fibrant in S0 , then X is fibrant in S. In

particular, if G sGr, then W G is fibrant in S.
Proof: The first clause follows from the previous lemma. For the second, note
that every object of sGr is fibrant. Since W : sGr S0 preserves fibrations,
W G is fibrant in S0 .

Corollary 6.9. Let f : X Y be a fibration in S0 between fibrant spaces.

Then f is a fibration in S if and only if
f : 1 X 1 Y
is onto. In particular, if G H is a fibration of simplicial groups, W G W H
is a fibration of simplicial sets if and only if 0 G 0 H is onto.
Proof: Consider a lifting problem

wX

u
u
1
S
w Y.
This can be solved up to homotopy; that is there is a diagram
S1

d0

d1
u
w S 1 1+

wX
u
h w Y.

6. Reduced simplicial sets, Milnors F K-construction

307

where h d0 = . But d1 : S 1 S 1 1+ is a trivial cofibration in S0 so the

homotopy h can be lifted to e
h : S 1 1+ X and e
h d0 solves the original
lifting problem.
For the second part of the corollary, note that Proposition 3.5 implies 1 W G
=
0 G.

We now produce the model category structure promised for S0 . The following
lemma sets the stage. If X is a simplicial set, let #X denote the cardinality of
the non-degenerate simplices in X. Let be the first infinite cardinal.
Lemma 6.10.
1) Let A B be a cofibration in S and x Bk a k-simplex. Then there is
a subspace C B so that #C < and x C.
2) Let A B be a trivial cofibration in S and x Bk a k-simplex. Then
there is a subspace D B so that #D , x D and A D D is a
trivial cofibration.
Proof: Part 1) is a reformulation of the statement that every simplicial set
is the filtered colimit of its finite subspaces. For part 2) we will construct an
expanding sequence of subspaces
D1 D2 B
so that x D1 , #Dn and
p (|Dn |, |Dn A|) p (|Dn+1 |, |Dn+1 A|)
S
is the zero map. Then we can set D = Dn .
n

To get D1 , simply choose a finite subspace D1 B with x D1 . Now suppose

Dq , q n, have been constructed and satisfy the above properties. Let
(|Dn |, |Dn A|).
Since maps to zero under
(|Dn |, |Dn A|) (|B|, |A|)
there must be a subspace D B, such that #D < and so that maps to
zero under

Set Dn+1

(|Dn |, |Dn A|) (|Dn D |, (Dn D ) A|).

S
= Dn ( D ).

308

V. Simplicial groups

Remark 6.11. The relative homotopy groups (|B|, |A|) for a cofibration i :
A B of simplicial sets are defined to be the homotopy groups of the homotopy
fibre of the realized map i : |A| , |B|, up to a dimension shift. The realization
of a Kan fibration is a Serre fibration (Theorem I.10.10), so it follows that these
groups coincide up to isomorphism with the simplicial homotopy groups Fi
of any choice of homotopy fibre Fi in the simplicial set category. One can use
Kans Ex functor along with an analog of the classical method of replacing a
continuous map by a fibration to give a rigid construction of the Kan complex
Fi which satisfies the property that the assignment i 7 Fi preserves filtered
colimits in the maps i. The argument for part 2) of Lemma 6.10 can therefore
be made completely combinatorial. This observation becomes quite important
in contexts where preservation of functoriality is vital see [38].
Lemma 6.12. A morphism f : X Y in S0 is a fibration if and only if it has
the right lifting property with respect to all trivial cofibrations C D in S0
with #D .
Proof: Consider a lifting problem
a wX

j
f

u
u
wY
B
A

where j is a trivial cofibration. We solve this by a Zorns Lemma argument.

Consider the set of pairs (Z, g) where A Z B, A Z is a weak equivalence
and g is a solution to the restricted lifting problem
a wX

u
u
Z
w Y.
A

Partially order by setting (Z, g) < (Z 0 , g 0 ) if Z Z 0 and g 0 extends g. Since

(A, a) , is not empty and any chain
(Z1 , g1 ) < (Z2 , g2 ) <

6. Reduced simplicial sets, Milnors F K-construction

309

in has an upper bound, namely (Zn , gn ). Thus satisfies the hypotheses of

Zorns lemma and has a maximal element (B0 , g0 ). Suppose B0 6= B. Consider
the diagram
B0
i

u
B

g0

wX
f
u
w Y.

Then i is a trivial cofibration. Choose x B with x 6 B0 . By Lemma 6.10.2

there is a subspace D B with x D, #D and B0 D D a trivial
cofibration. The restricted lifting problem
B0 B

wX
i
j
i
f
i
u i
u
D
wY

has a solution, by hypotheses. Thus g0 can be extended over B0 D. This

contradicts the maximality of (B0 , g0 ). Hence B0 = B.

Remark 6.13. The proofs of Lemma 6.10 and Lemma 6.12 are actually standard
moves. The same circle of ideas appears in the arguments for the closed model
structures underlying both the Bousfield homology localization theories [8], [9]
and the homotopy theory of simplicial presheaves [46], [51], [38]. We shall return
to this topic in Chapter IX.
The Proof of Proposition 6.2: Axioms CM1CM3 for a closed model
category are easy in this case. Also, the trivial cofibration-fibration part of
CM4 is the definition of fibration. We next prove the factorization axiom CM5
holds, then return to finish CM4.
Let f : X Y be a morphism in S0 . To factor f as a cofibration followed
by a trivial fibration, use the usual small object argument with pushout along
j
q
cofibrations A B in S0 with #B < to factor f as X Z Y where j
is cofibration and q is a map with the right lifting property with respect to all
cofibrations A B with #B < . The evident variant on the Zorns lemma
argument given in the proof of Lemma 6.12 using 6.10.1 implies q has the right
lifting property with respect to all cofibrations in S0 . Hence q is a fibration. We

310

V. Simplicial groups

claim it is a weak equivalence and, in fact, a trivial fibration in S. To see this

consider a lifting problem
n
y

wZ

u
u
n

w Y.

If n = 0 this has a solution, since Z0

= Y0 . If n > 0, this extends to a diagram
n
y

w n / sk0 (n )

wZ

w n / sk0 n

u
n

u
w Y.

Since n 1, sk(n ) = sk0 (n ), so j is a cofibration between finite complexes

in S0 and the lift exists.
Return to f : X Y in S0 . To factor f as a trivial cofibration followed by a
fibration, we use a transfinite small object argument.
We follow the convention that a cardinal number is the smallest ordinal number
within a given bijection class; we further interpret a cardinal number as a poset
consisting of strictly smaller ordinal numbers, and hence as a category. Choose
a cardinal number such that > 2 .
Take the map f : X Y , and define a functor X : S0 and a natural
transformation fs : X(s) Y such that
(1) X(0) = X,
(2) X(t) = lim
X(s) for all limit ordinals t < , and
s<t
(3) the map X(s) X(s + 1) is defined by the pushout diagram
G

AD

(D )

w X(s)

tiD
Gu
BD
D

u
w X(s + 1)

6. Reduced simplicial sets, Milnors F K-construction

311

where the index D refers to a set of representatives for all diagrams

D

AD
iD

u
BD

w X(s)
u
wY

fs

such that iD : AD BD is a trivial cofibration in S0 with #BD .

Then there is a factorization
X4

i0

w X
4
4
f 4
6 f
u
Y

for the map f , where X = lim X(s), and i0 : X = X(0) X is the canonical
s
map into the colimit. Lemma 6.12. A pushout along a trivial cofibration in S0 is a
trivial cofibration in S0 because the same is true in S, so i0 is a trivial cofibration.
Also, any map A X must factor through one of the canonical maps is :
X(s) X if #A , for otherwise A would have too many subobjects on
account of the size of . It follows that the map f : X Y is a fibration of
S0 . This finishes CM5.
To prove CM4 we must show any trivial fibration f : X Y in S0 has the
right lifting property with respect to all cofibrations. However, we factored f as
a composite
j
q
X Z Y
where j is a cofibration and q is a trivial fibration with the right lifting property
with respect to all cofibrations. Now j is a trivial cofibration, since f is a weak
equivalence. Thus there is a lifting in
= wX

j
f

u
u
Z q wY
X

312

V. Simplicial groups

since f is a fibration. This shows f is a retract of q and has the requisite lifting
property, since q does.

As an artifact of the proof we have:

Lemma 6.14. A morphism f : X Y in S0 is a trivial fibration in S0 if and
only if it is a trivial fibration in S.
The Milnor F K construction associates to a pointed simplicial set K the
simplicial group F K, which is given in degree n by
F Kn = F (Kn {}),
so that F Kn is the free group on the set Kn {}. This construction obviously
gives a functor from pointed simplicial sets to simplicial groups. The group F K
is also a loop group:
Theorem 6.15. There is a natural isomorphism
G(K)
= F K,
for pointed simplicial sets K.
Proof: Recall that K denotes the Kan suspension of K. The group of nsimplices of G(K) is defined to be the quotient
G(K)n = F (Kn+1 )/F (s0 Kn ).
The map s0 : Kn Kn+1 can be identified with the wedge summand inclusion
Kn1 K0 , Kn Kn1 K0 ,
so that the composite group homomorphism
n

F (Kn ) F (Kn+1 ) F (Kn+1 )/F (s0 Kn )

can be identified via an isomorphism
(6.16)

F (Kn+1 )/F (s0 Kn )

= F Kn

with the quotient map

F (Kn ) F (Kn )/F ()
= F Kn .

7. Simplicial groupoids

313

Recall that for : m n, the map : G(K)n G(K)m is specified on

generators [x] by
([x]) = [ (x)][(c) (d0 (x))]1 .
But then
([n (x)]) = [ (n (x))][(c) (d0 (n (x)))]1
= [ (n (x))]
= [m ( (x))],
since d0 (n (x)) = . It follows that the isomorphisms (6.16) respect the simplicial
structure maps.

The proof of Theorem 6.15 is easy enough, but this result has important
consequences:
Corollary 6.17.
(1) The Milnor F K construction takes weak equivalences of pointed simplicial
sets to weak equivalences of simplicial groups.
(2) The simplicial group F K is a natural fibrant model for K, in the
category of pointed simplicial sets.
Proof: The first assertion is proved by observing that the Kan suspension
functor preserves weak equivalences; the loop group construction has the same
property by Theorem 5.10 (see also Section II.8).
Let K Y be a fibrant model for K in the category of reduced simplicial
sets. Then Y is a Kan complex which is weakly equivalent to K, so that Y
is a model for K. The loop group functor preserves weak equivalences, so
that the induced map G(K) GY is a weak equivalence of simplicial groups.
Finally, we know that GY is weakly equivalent to Y , so that G(K) and hence
F K is a model for K.

7. Simplicial groupoids.
A simplicial groupoid G, for our purposes, is a simplicial object in the category
of groupoids whose simplicial set of objects is discrete. In other words, G consists
of small groupoids Gn , n 0 with a functor : Gm Gn for each ordinal
number map : n m, such that all sets of objects Ob(Gn ) coincide with a
fixed set Ob(G), and all functors induce the identity function on Ob(G). Of
course, 7 is also contravariantly functorial in ordinal number maps . The
set of morphisms from x to y in Gn will be denoted by Gn (x, y), and there is

314

V. Simplicial groups

a simplicial set G(x, y) whose n-simplices are the morphism set Gn (x, y) in the
groupoid Gn . We shall denote the category of simplicial groupoids by sGd.
The free groupoid G(X) on a graph X has the same set of objects as X, and
has morphisms consisting of reduced words in arrows of X and their inverses.
There is a canonical graph morphism : X G(X) which is the identity on
objects, and takes an arrow to the reduced word represented by the string
consisting of alone. Any graph morphism f : X H taking values in a
groupoid H extends uniquely to a functor f : G(X) H, in the sense that the
following diagram commutes:

u
G(X)

wH

j
hh
h f
h

There is a similar construction of a free groupoid GC on a category C, which

has been used without comment until now. The groupoid GC is obtained by the
free groupoid on the graph underlying the category C by killing the normal subgroupoid generated by the composition relations of C and the strings associated
to the identity morphisms of C (see also Sections I.8 and III.1). The category
of groupoids obviously has all small coproducts, given by disjoint unions. This
category also has pushouts, which are actually pushouts in the category of small
categories, so the category of groupoids is cocomplete. Note that filtered colimits
are formed in the category of groupoids just as filtered colimits on the object and
morphism levels. The initial object in the category of groupoids has an empty
set of morphisms and an empty set of objects and is denoted by .
It is also completely straightforward to show that the category of simplicial
groupoids has all inverse limits.
Dwyer and Kan define [25], for every simplicial set X, a groupoid F 0 X having
object set {0, 1}, such that the set of n-simplices Xn is identified with a set of
arrows from 0 to 1, and such that F 0 Xn is the free groupoid on the resulting
graph.
The groupoid F 0 K is morally the same thing as the Milnor construction, for
pointed simplicial sets K. If x denotes the base point of K, then there is a
homomorphism of simplicial groups
g : F K F 0 K(0, 0)

7. Simplicial groupoids

315

which is defined on generators y Kn {x} by y 7 x1 y. Also, regarding F K

as a simplicial groupoid with one object, we see that there is a map of simplicial
groupoids
f : F 0K F K
defined by sending x to e in all degrees and such that y Kn {x} maps to the
arrow y. The collection of all products y 1 z, y, z Kn , generates F 0 K(0, 0) in
degree n, and so it follows that the composite simplicial group homomorphism
f

F 0 K(0, 0)
FK
F 0 K(0, 0)
is the identity. The composite
g

FK
F 0 K(0, 0)
FK
sends y Kn to x1 y = y F Kn , so the homomorphism g is an isomorphism.
Lemma 7.1. Suppose that K is a pointed simplicial set. Then the simplicial
sets F 0 K(a, b), a, b {0, 1}, are all isomorphic to the Milnor F K construction.
Proof: The base point x of K determines an isomorphism x : 0 1 in the
groupoid F 0 Kn for all n 0. Composition and precomposition with x therefore
determines a commutative diagram of simplicial set isomorphisms

(7.2)

F 0 K(0, 0)
u
=
x
F 0 K(1, 0)

=
x

w F 0 K(0, 1)
u

= x
w F 0 K(1, 1),

and of course weve seen that F 0 K(0, 0)

= F K.

Corollary 7.3. A weak equivalence f : X Y of simplicial sets induces weak

equivalences f : F 0 X(a, b) F 0 Y (a, b) for all objects a, b {0, 1}.
Proof: We can suppose that X is non-empty. Pick a base point x in X, and
observe that the diagram (7.2) is natural in pointed simplicial set maps, as is
the isomorphism F 0 X(0, 0)
= F X. Weve already seen that the Milnor F X
construction preserves weak equivalences.

316

V. Simplicial groups

For an ordinary groupoid H, its standard to write 0 H for the set of path
components of H. By this, one means that
0 H = Ob(H)/ ,
where there is a relations x y between two objects of H if and only if there
is a morphism x y in H. This is plainly an equivalence relation since H is a
groupoid, but more generally 0 H is the specialization of a notion of the set of
path components 0 C for a small category C.
If now G is a simplicial groupoid, its easily seen that all of the simplicial
structure functors : Gn Gm induce isomorphisms 0 Gn
= 0 Gm . We shall
therefore refer to 0 G0 as the set of path components of the simplicial groupoid
G, and denote it by 0 G.
A map f : G H of simplicial groupoids is said to be a weak equivalence of
sGd if
(1) the morphism f induces an isomorphism 0 G
= 0 H, and
(2) each induced map f : G(x, x) H(f (x), f (x)), x Ob(G) is a weak
equivalence of simplicial groups (or of simplicial sets).
Corollary 7.3 says that the functor F 0 : S sGd takes weak equivalences of
simplicial sets to weak equivalences of simplicial groupoids.
A map g : H K of simplicial groupoids is said to be a fibration if
(1) the morphism g has the path lifting property in the sense for every object x
of H and morphism : g(x) y of the groupoid K0 , there is a morphism

: x z of H0 such that g(
) = , and
(2) each induced map g : H(x, x) K(g(x), g(x)), x Ob(H), is a fibration
of simplicial groups (or of simplicial sets).
According to this definiton, every simplicial groupoid G is fibrant, since the map
G which takes values in the terminal simplicial groupoid is a fibration.
A cofibration of simplicial groupoids is defined to be a map which has the left
lifting property with respect to all morphisms of sGd which are both fibrations
and weak equivalences.
Several comments are in order. First of all, picking a representative x [x]
for each [x] 0 G determines a map of simplicial groupoids
G
i:
G(x, x) G
[x]0 G

which is plainly
a weak equivalence. But more is true, in that the simplicial
F
groupoid [x]0 G G(x, x) is a deformation retract of G in the usual groupoidtheoretic sense. To see this, pick morphisms y : y x in G0 for each y [x] and

7. Simplicial groupoids

317

for each [x] 0 G, such that x = 1x for all the fixed choices of representatives x
of the various path components x. Then there is a simplicial groupoid morphism
G

r:G

G(x, x),

[x]0 G

which is defined by conjugation by the paths y , in that r(y) = x if and only

if y [x] for all objects y of G, and r : G(y, z) G(x, x) is the map sending
: y z to the composite z y1 G(x, x) for all y, z [x], and for each
[x] 0 G. The morphisms y also determine a groupoid homotopy
h:GI G
where I denotes the free groupoid on the ordinal number (category) 1. This
homotopy is from the identity on G to the composite ir, and is given by the
obvious conjugation picture.
The choices of the paths which define the retraction map r are non-canonical,
and are certainly not natural with respect to morphisms of simplicial groupoids,
except in certain useful isolated cases (see the proof of Lemma 7.5 below). The
choices of the representatives x for path components [x] are non-canonical as
well, but are more nearly natural, in that for every weak equivalence f : G H
of simplicial groupoids, choices of representatives can be made (in G) such that
there is a commutative diagram
G

G(x, x)
w

[x]0 G

(7.4)

H(f (x), f (x))

[x]0 G

u
G

i
u
w H.

of weak equivalences of simplicial groupoids. We can also prove the following:

Lemma 7.5. Suppose that the morphism j : A B of simplicial groupoids has
the left lifting property with respect to all fibrations. Then the map j is a weak
equivalence.
Proof: The initial object in the category of simplicial groupoids is the category
having no objects and no arrows. The path lifting property for the canonical

318

V. Simplicial groups

map G is vacuously satisfied, and, similarly, all simplicial set maps X

are Kan fibrations, so that each map G is a fibration of simplicial groupoids.
If A = , then the lifting exists in the diagram

w
\
]
\
\
u \
u
B
= w B,
forcing B = , and so j is a weak equivalence in this case.
In general, there is a commutative diagram
A
j
u h
B

=
j
hh
r
h

wA
u
w

since A is fibrant. In particular, j induces an injection i : Ob(A) , Ob(B), as

well as an injection i : 0 A 0 B. Let C be the subobject of B, given by the
union of path components containing objects of A. If C 6= B, then B = C t C 0
and the map j : A B is a disjoint union of maps of the form A C and
C 0 , each of which has the left lifting property with respect to all fibrations.
It follows that C 0 = , and so j induces an isomorphism on 0 .
Recall that j induces a monomorphism on objects. It follows that, by choosing
paths : y x for a fixed list of representatives x for elements [x] 0 A, first
in A and then for the remaining objects in B, we can construct a commutative
diagram of simplicial groupoid maps
G

A(x, x)

wA

[x]0 A

G
[x]0 A

A(x, x)

[x]0 A

j
u
B(j(x), j(x))

G
w

u
wB

G
[x]0 A

j
u
B(j(x), j(x))

7. Simplicial groupoids

319

such that the top and bottom horizontal composites are the identity. The restricted map
G
G
j:
A(x, x)
B(j(x), j(x))
[x]0 A

[x]0 A

therefore has the right lifting property with respect to all fibrations, and so each
of the maps j : A(x, x) B(j(x), j(x)) has the same lifting property, and is
therefore a trivial cofibration of simplicial groups.

Theorem 7.6. With these definitions, the category sGd of simplicial groupoids
satisfies the axioms for a closed model category.
Proof: Only the factorization axiom has an interesting proof. Write F 0 0 to
denote the discrete simplicial groupoid on the object set {0, 1}, and write F 0 00 to
denote the terminal groupoid . Then a map of simplicial groupoids f : G H
is a fibration if and only if it has the right lifting property with respect to all
morphisms F 0 nk F 0 n , 0 k n, and f is a trivial fibration (aka. fibration
and weak equivalence) if and only if it has the right lifting property with respect
to all morphisms F 0 n F 0 n , n 0, and the morphism (compare
[25]). We can therefore use a small object argument to show that every simplicial
groupoid morphism g : K L has factorizations

)
i''
'
Kh

Nh

hp
h
j

h
j
j h

'

wL
'
)
'q

M
where p is a fibration and i has the left lifting property with respect to all
fibrations, and q is a trivial fibration and j is a cofibration. Lemma 7.5 implies
that i is a weak equivalence.
The proof of the lifting axiom CM4 is a standard consequence of the proof
of the factorization axiom: any map which is both a cofibration and a weak
equivalence (ie. a trivial cofibration) is a retract of a map which has the left
lifting property with respect to all fibrations, and therefore has that same lifting
property.

320

V. Simplicial groups

There is a simplicial set W G for a simplicial groupoid G that is defined

by analogy with and extends the corresponding object for a simplicial group.
Explicitly, suppose that G is a simplicial groupoid. An n-cocycle X : Seg(n)
G associates to each object [k, n] an object Xk of G, and assigns to each relation
: [j, n] [k, n] in Seg(n) a morphism X( ) : Xj Xk in Gnj , such that the
following conditions hold:
(1) X(1j ) = 1Xj Gnj , where 1j is the identity relation [j, n] [j, n],

(2) for any composeable pair of relations [l, n]

[k, n]
[j, n], there is a
commutative diagram
X()
Xl 4
w Xk
4

4
6 X( )
X( ) 4
u
Xj
in the groupoid Gnj .
Suppose that : m n is an ordinal number map. As before, induces
a functor : Seg(m) Seg(n), which is defined by sending the morphism
: [k, m] [j, m] to the morphism : [(k), n] [(j), n]. Composing the
n-cocycle X : Seg(n) G with gives a cocycle X : Seg(m) G, defined
for each relation : [k, m] [j, m] in Seg(m), (and in the notation of (4.2)) by
the morphism
X( ) = k (X( )) : X(k) X(j) .
of Gmk The assignment 7 is contravariantly functorial in ordinal maps .
We have therefore constructed a simplicial set whose n-simplices are the ncocycles Seg(n) G, and whose simplicial structure maps are the induced
maps . This simplicial set of G-cocycles is W G. In particular, an n-cocycle
X : Seg(n) G is completely determined by the string of relations
0

n2

n1

[n, n] [n 1, n] . . . [1, n] [0, n],

and the corresponding maps
X(0 )

X(1 )

X(n1 )

Xn Xn1 Xn2 X1 X0 .

7. Simplicial groupoids

321

Each i is an instance of the map d0 , and X(i ) is a morphism of the groupoid

Gi . Note, in particular, that the ith vertex of the cocycle X : Seg(n) G is the
object Xi of G: this means that Xi can be identified with the cocycle i X,
where i : 0 n.
Suppose that : m n is an ordinal number map, and let g denote the string
of morphisms
g0

gn1

g1

Xn Xn1 Xn2 X1 X0
in G, with gi a morphism of Gi . Let Xg be the cocycle Seg(n) G associated to
the n-tuple g. Then, subject to the notation appearing in diagram (4.2), Xg
is the string

m1
Xg (1 )

m
Xg (0 )

1 Xg (m1 )

X(m) X(m1) X(m2) X(1) X(0) .

Its easily seen that this definition specializes to the cocycle definition of W G in
the case where G is a simplicial group.
A simplicial map f : X W G assigns to each n-simplex x a cocycle f (x) :
Seg(n) G, such that for each ordinal number map : m n and each map
: [k, m] [j, m] in Seg(m) there is a relation
k f (x)( ) = f ( (x))( ).
Furthermore, f (x) is determined by the string of maps
f (x)(0 )

f (x)(n1 )

f (x)(1 )

f (xn ) f (xn1 ) f (xn2 ) f (x1 ) f (x0 ),

in G, where xi is the ith vertex of x, and ni is the map ni = d0 : [i, n]
[i1, n] of Seg(n). By the simplicial relations, f (x)(ni ) = f (di1
0 (x))(ni ), so
that the simplicial map f : X W G is completely determined by the morphisms
f (x)(d0 = n1 : [1, n] n) : f (x1 ) f (x0 )
in Gn1 , x X. In alternate notation then, the cocycle f (x) is completely
determined by the string of morphisms
x)(d0 )
f (dn1
0

f (d0 x)(d0 )

f (x)(d0 )

f (xn ) f (xn1 ) . . . f (x1 ) f (x0 )

in G.

322

V. Simplicial groups

The morphism f (s0 x)(d0 ) is the identity on f (x0 ). We now can define a
groupoid GXn to be the free groupoid on generators x : x1 x0 , where x
Xn+1 , subject to the relations s0 x = 1x0 , x Xn . The objects of this groupoid
are just the vertices of X. Following the description of the loop group from
a previous section, we can define a functor : GXn GXm for each ordinal
number morphism : m n by specifying that is the identity on objects, and
is defined on generators [x], x Xn+1 , by requiring that the following diagram
commutes:
[(c) d0 x]
x(0)+1
w x1

[x]

[ (x)]
u
x0 ,
or rather that
[x] = [ (x)][(c) d0 x]1 .
One checks, as before, that this assignment is functorial in ordinal number morphisms , so that the groupoids GXn , n 0, and the functors form a simplicial
groupoid GX, which we call the loop groupoid for X.
Any n-simplex x of the simplicial set X determines a string of morphisms
[dn1
x]
0

[d0 x]

[x]

xn xn1
. . . x1 x0
in GX, which together determine a cocycle (x) : Seg(n) GX in the simplicial
groupoid GX. The calculations leading to Lemma 5.3 also imply the following:
Lemma 7.7.
(a) The assignment x 7 (x) defines a natural simplicial map : X
W GX.
(b) The map is one of the canonical homomorphisms for an adjunction
homsGd (GX, H)
= homS (X, W H),
where sGd denotes the category of simplicial groupoids.
Heres the homotopy theoretic content of these functors:

7. Simplicial groupoids

323

Theorem 7.8.
(1) The functor G : S sGd preserves cofibrations and weak equivalences.
(2) The functor W : sGd S preserves fibrations and weak equivalences.
(3) A map K W X S is a weak equivalence if and only if its adjoint
GK X sGd is a weak equivalence.
Proof: The heart of the matter for this proof is statement (2). We begin by
showing that W preserves weak equivalences.
Suppose that A is a simplicial groupoid, and choose a representative x for each
[x] 0 A. Recall that the inclusion
G

i:

A(x, x) A

[x]0 A

is a homotopy equivalence of simplicial groupoids in the sense that there is a

groupoid map
G
r:A
A(x, x)
[x]0 A

which is determined by paths, such that ri is the identity and such that the
paths defining r determine a groupoid homotopy
h:AI A
from the identity on A to the composite morphism ir. The object I is the
constant simplicial groupoid associated to the groupoid having two objects 0, 1
and exactly one morphism a b for any a, b {0, 1}. One sees that W I = BI
and that W preserves products. It follows that the groupoid homotopy h induces
a homotopy of simplicial sets from the identity on W A to the composite map
W i W r, and so W i is a weak equivalence. If f : A B is a weak equivalence
of simplicial groupoids, then f induces an isomorphism 0 A
= 0 B, and there
is a commutative diagram of simplicial groupoid maps
G

A(x, x)

x0 A

B(f (x), f (x))

x0 A

u
A

i
u
wB

324

V. Simplicial groups

in which the vertical maps are homotopy equivalences. To see that W f is a weak
equivalence, it therefore suffices to show that W takes the top horizontal map
to a weak equivalence. But W preserves disjoint unions, and then one uses the
corresponding result for simplicial groups (ie. Proposition 6.3).
To show that W preserves fibrations, we have to show that a lifting exists for
all diagrams
nk
y

(7.9)
u
n

w WA
Wf
u
w W B,

given that f : A B is a fibration of sGd. We can assume that A and B are

connected. The lifting problem is solved by the path lifting property for f if
n = 1.
Otherwise, take a fixed x A0 and choose paths i : yi x in A0 , where yi
is the image of the ith vertex in nk . Note that the vertices of nk coincide with
those of n , since n 2. These paths, along with their images in the groupoid
B0 determine cocycle homotopies from the diagram (7.9) to a diagram
nk
y

w W A(x, x)
C
B
B
B
Wf
B
B
u B
u
n
w W B(f (x), f (x)).
0

(7.10)

More explicitly, if the simplicial set map is determined by the string of

morphisms
gn1

gn2

g0

f (yn ) f (yn1 ) f (yn2 ) f (y1 ) f (y0 )

in B, then the cocycle homotopy from to 0 is the diagram
f (yn )
f (n )

u
f (x)

gn1

w f (yn1 )

gn2

w ...

g1

f (n1 )
hn1

u
w f (x)

w f (y1 )

g0

f (1 )
hn2

w ...

h1

u
w f (x)

h0

w f (y0 )
f (0 )
u
w f (x)

7. Simplicial groupoids

325

where hi = f (i )gi f (i+1 )1 , and 0 is defined by the string of morphisms hi .

The cocycle 0 is a cocycle conjugate of , in an obvious sense.
The indicated lift exists in the diagram (7.10), because the simplicial set map
W f : A(x, x) B(f (x), f (x)) satisfies the lifting property for n 2 (see the
proof of Lemma 6.6). The required lift for the diagram (7.9) is cocycle conjugate
to .
We have therefore proved statement (2) of the theorem. An adjointness argument now implies that the functor G preserves cofibrations and trivial cofibrations. Every weak equivalence K L of simplicial sets can be factored as
a trivial cofibration, followed by a trivial fibration, and every trivial fibration
in S is left inverse to a trivial cofibration. It follows that G preserves weak
equivalences, giving statement (1).
Statement (3) is proved by showing that the unit and counit of the adjunction
are both weak equivalences. Let A be a simplicial groupoid. To show that the
counit : GW A A is a weak equivalence, we form the diagram

GW (

A(x, x))

x0 A

A(x, x)

x0 A

GW i

u
GW A

' i
u
w A,

where we note that GW i is a weak equivalence by statements (1) and (2). The
functors G and W both preserve disjoint unions, so its enough to show that the
simplicial group map : GW A(x, x) A(x, x) is a weak equivalence, but this
is the traditional result for simplicial groups (Proposition 6.3; see also Corollary 6.4).
Let K be a simplicial set. To show that the unit : K W GK is a weak
equivalence, it suffices to assume that K is a reduced Kan complex, by statements
(1) and (2). Now apply Proposition 6.3.

Corollary 7.11. The functors G and W induce an equivalence of homotopy

categories
Ho(sGd) ' Ho(S).

Chapter VI The homotopy theory of towers

In principle, the Postnikov tower construction gives a method of breaking up
a space X into a collection of spaces X(n), n 0, such that X(n) carries the
homotopy groups of X up to level n, along with a tower of fibrations X(n)
X(n 1) each of which gives a calculus of adding the nth homotopy group of X
to X(n 1) to create the space X(n). Then, in practice but sometimes with a
little luck, the space X can be recovered as the inverse limit (or more properly
a homotopy inverse limit) of the tower. This gives one of the central methods
for recovering qualitative and calculational properties of X from the EilenbergMac Lane spaces corresponding to its homotopy groups and the associated kinvariants.
Weve already seen an application, namely the proof of the Hurewicz theorem
in Section III.3. The theory of Postnikov towers and related constructions is
the central object of study of this chapter. We begin in Section 1 by describing
an approach to constructing homotopy inverse limits of towers: a closed model
structure of towers of spaces is introduced, for which the homotopy inverse limit
of a tower is the inverse limit of an associated fibrant model. This is a special
case of a much more general concept which is more fully described in Chapter
IX. Section 2 contains a formal introduction to Postnikov towers, both for spaces
and maps. The construction is the standard one for simplicial sets, which is due
to Moore.
In many applications, say in rational homotopy theory or more generally in
the theory of R-completions and localizations, one is presented with a suite of
results which says
1) Eilenberg-Mac Lane spaces have property P,
2) If X Y is a principal K(A, n)-fibration and Y has property P, then so
does X, and
3) If X1 X0 is a tower of fibrations such that all Xi have
property P, then so does lim Xi .

This gives a method of inferring that a certain class of spaces X has property
P which involves crawling up a Postnikov tower or some refinement thereof,
provided that all of the fibrations in the tower are principal fibrations, at least
up to homotopy equivalence. There is a good class of spaces for which this yoga
works, namely the nilpotent spaces, and the method for establishing it involves
a careful analysis of k-invariants.
326

1. A model category structure for towers of spaces

327

For us, a k-invariant is a map

kn : X(n 1) holim K(n X, n + 1)

for a connected space X taking values in a homotopy colimit arising from the
action of the fundamental groupoid on the homotopy group n X. This map
is implicitly fibred over the classifying space B of the fundamental groupoid
, and can be interpreted as a representative of a class in a suitably defined
equivariant cohomology group Hn+1 (X(n 1), n X). The main point overall
(Proposition 5.1) is that the fibration X(n) X(n 1) in the Postnikov tower
for X sits in a homotopy cartesian diagram
X(n)

w holim L(n X, n + 1)

p
u
X(n 1)

u
kn

w holim K(n X, n + 1)

where the map p is induced by a natural (hence equivariant) contractible covering L(n X, n + 1) K(n X, n + 1) of the space K(n X, n + 1). From this, its
pretty much immediate (Corollary 5.3) that the fibration X(n) X(n 1)
is homotopy equivalent to a principal K(n , n)-fibration if the fundamental
group acts trivially on n X. Similarly, if X is nilpotent then the covering
L(n X, n + 1) K(n , n + 1) has a finite refinement by fibrations that induce principal fibrations after taking homotopy colimit (Proposition 6.1), giving
the refined Postnikov tower for a nilpotent space.
This stream of results occupies the last four sections of this chapter. The
main techniques involve relating homotopy classes of maps [X, holim K(A, n)]

fibred over the classifying space B of a groupoid to equivariant cohomology

in various forms (Theorems 3.10, 3.11) this and some general nonsense about
equivariant homotopy theory is the subject of Sections 3 and 4. The formal introduction of k-invariants and the proofs of their main properties occupy Section
5, while the application to nilpotent spaces appears in Section 6.
1. A model category structure for towers of spaces.
The purpose of this section is to introduce the structure of a simplicial model
category on tow(S). This will have implications for Postnikov systems as well
as allowing us to define homotopy inverse limits for towers of spaces.

328

VI. The homotopy theory of towers

Let C be a category. Then tow(C) is the category of towers in C. Thus an

object in tow(C) is a diagram in C
Xn Xn1 X1 X0
and a morphism in tow(C) is a morphism of diagrams; that is, a commutative
ladder. It will be convenient to write X for a tower {Xn }
First notice that if C is a simplicial category, then tow(C) is a simplicial
category. Indeed if X tow(C) and K S, let
X K = {Xn K}
and
hom(K, X) = {homC (K, Xn )}.
Then tow(C) is a simplicial category with
Hom(X, Y)n = hom(X n , Y).
The subscript in the last equation is the simplicial degree.
Definition 1.1. Define a morphism f : X Y in tow(S) to be
1) a weak equivalence if fn : Xn Yn is a weak equivalence for all n 0;
2) a cofibration if fn : Xn Yn is a cofibration for all n 0; and
3) a fibration if f0 : X0 Y0 is a fibration and for all n 1, the induced
map
Xn Yn Yn1 Xn1
is a fibration.
A useful preliminary lemma is:
Lemma 1.2. Let q : X Y be a fibration in tow(C). Then for all n 0,
qn : Xn Yn is a fibration.
Proof: This is true if n = 0. If it is true for n 1, contemplate the induced
diagram
p

Xn n w Yn Yn1 Xn1

qn

u qn

Yn

w Xn1
u

qn1

w Yn1

The morphism pn is a fibration by hypothesis; the morphism q n is a fibration

because the pullback of a fibration is a fibration. So qn is a fibration.

1. A model category structure for towers of spaces

329

Proposition 1.3. With these definitions, tow(C) becomes a simplicial model

category.
Proof: Axioms CM1CM3 are obvious. Suppose given a lifting problem in
tow(C)
wX
uq
wY

A
u
B

with j a cofibration and q a fibration. If j is also a weak equivalence, one can

recursively solve the lifting problem
An
w Xn
(
)
(
ju
u
((
Bn
w Yn Yn1 Xn1

(1.4)

to solve the lifting problem in tow(C). If q is a weak equivalence, the pullback

diagram
w Xn1

Yn Yn1 Xn1

u qn1

0
qn

u
Yn

w Yn1

shows qn0 is a trivial fibration, since qn1 is a fibration and a weak equivalence.
Hence Xn Yn Yn1 Xn1 is a trivial fibration and we can again recursively
solve the lifting problem of (1.4). This proves CM4.
To prove CM5 fix a morphism f : X Y. To factor f as a cofibration
followed by a trivial fibration, proceed inductively as follows. First, factor f0 :
X0 Y0 as a cofibration followed by a trivial fibration
j0

q0

X0 Z0 Y0 .
Then, having factored through level n 1, consider the induced maps
Xn

pn

w Zn1 Yn1 Yn
u
Zn1

q n1

qn1

w Yn
u

w Yn1 .

330

VI. The homotopy theory of towers

Since qn1 is a trivial fibration so is q n1 . Factor pn as

jn

0
qn

Xn Zn Zn1 Yn1 Yn
where jn is a cofibration and qn0 is a trivial fibration. The other factoring is
similar, but easier.
Finally, SM7 follows from SM7b, which is obvious in this case.

Remark 1.5.
1) If every object of C is cofibrant, then every object of tow(C) is cofibrant.
This applies, for example, if C = S or S .
2) The fibrant objects of tow(C) are the ones where X0 is fibrant in C and
every qn : Xn Xn1 is a fibration.
Remark 1.6. The model structure of Proposition 1.3 extends easily to bigger
towers. Suppose that is a limit ordinal, and identify it with a poset. A tower is a simplicial set-valued functor X : op S which is contravariant on
. Say that a map f : X Y between -towers is a cofibration (respectively
weak equivalence) if the maps f : Xs Ys are cofibrations (respectively weak
equivalences) for s < . A map g : Z W is a fibration if the following
conditions hold:
1) the map g : Z0 W0 is a fibration,
2) for all ordinals s < the induced map Zs+1 Ws+1 Ws Zs is a fibration,
and
3) for all limit ordinals < the map
Z W (lim Ws ) (lim Zs )

s<
s<

is a fibration.
With these definitions, the category of -towers satisfies the conditions for a
simplicial model category. This model structure could be arrived at in a different way, by using the model structure for op -diagrams of simplicial sets which
appears in Section IX.5 below, but that method does not produce the completely
explicit description of fibration that you see here.
An extremely flexible notion is the following idea of a homotopy inverse limit.

1. A model category structure for towers of spaces

331

Definition 1.7. Let X tow(C). Choose a weak equivalence X Y where

Y is fibrant in tow(C). Then
holim X = lim Y.

As usual, holim X is well-defined and functorial up to homotopy; indeed,

holim() is the total right derived functor of lim(). Notice that lim() : tow(C)

C is right adjoint to the constant tower functor, which preserves cofibrations and
weak equivalences, so lim() preserves fibrations and trivial fibrations. Further
more if X X0 is a weak equivalence, holim X holim X0 is a weak equiv

alence. It is this invariance property that justifies the name homotopy inverse
limit.
Exercise 1.8 (Homotopy pullbacks). Let I be the category with objects 1,
2, and 12 and non-identity morphisms as follows:
1 12 2.
If C is a category, let C I denote the resulting diagram category. An object in C I
is a diagram
Y2
u
w Y12

Y1

in C. Suppose C is a simplicial model category. Then C I becomes a simplicial

model category with, for K S,

Y1

Y2
u
K =
w Y12
Y1 K

Y2 K
u
w Y12 K.

The techniques of this section can be adapted to prove the following:

Theorem 1.9. The category C I is a simplicial model category with a morphism
X1
u
Y1

f1

w X12 u
u
w Y12 u

f12

X2
u f2
Y2

332

VI. The homotopy theory of towers

1) a weak equivalence if f1 , f2 , and f12 are;
2) a cofibration if f1 , f2 , and f12 are; and
3) a fibration if f12 : X12 Y12 is a fibration and for i = 1, 2 the induced
maps
Xi X12 Y12 Yi
are fibrations.

A crucial lemma is:

Lemma 1.10. If a morphism f : X Y in C I is a trivial fibration, then the
induced maps
Xi X12 Y12 Yi
are trivial fibrations.
Note the fibrant objects in C I are those diagrams
Y1 Y12 Y2
with Y12 fibrant and Yi Y12 a fibration.
If X C I is a diagram, define the homotopy pullback by taking the actual
pullback of a fibrant replacement. This has the usual homotopy invariance properties, much as after Definition 1.7. Prove that if C is a proper model category
and
Y1 Y12 Y2
is a diagram with Y12 fibrant and one of Y1 Y12 or Y2 Y12 a fibration, then
Y1 Y12 Y2 is weakly equivalent to the homotopy pullback.
Here is another description of the homotopy inverse limit that is often useful
for computations. Fix a simplicial model category C and write, for X C,
D : homC (1 , X) homC (1 , X)
= X X.
Then define, for a tower X = {Xn } tow(C), the object T (X) C by the
pullback diagram
T (X)
w

hom(1 , Xn )

(1.11)

Yu
n

Xn

(1,q)

Q
Y
n

Dn

Xn Xn

1. A model category structure for towers of spaces

333

where (1, q) is the product of the maps

Y
1qn
Xn Xn Xn Xn1 .
n

Lemma 1.12. Let X = {Xn } tow(C) be a tower so that each Xn is fibrant in

C. Then there is a weak equivalence
T (X) ' holim X.

Proof: The functor T : tow(C) C has a left adjoint defined as follows. Note
that specifying a map f : Y T (X) is equivalent to specifying a sequence of
maps gn : Y Xn and right homotopies
Y hom(1 , Xn )
between gn and qn+1 gn+1 . Define [0, ) S to be the simplicial half-line; thus
[0, ) has non-degenerate 1-simplices [n, n + 1], 0 n < , and d0 [n 1, n] =
d1 [n, n + 1]. Let [n, ) [0, ) be the evident sub-complex. Then the left
adjoint to T is given by the functor F ():
F (Y ) = {Y [n, )}.
Since this left adjoint preserves cofibrations and weak equivalences among cofibrant objects, F () has a total left derived functor LF . But the unique maps
[n, ) induces a weak equivalence F (Y ) {Y }n0 from F (Y ) to the constant tower, so LF : Ho(C) Ho(tow(C)) is the total left derived functor of the
constant tower functor.
The functor T preserves fibrations, since it is right adjoint to a functor that
preserves trivial cofibrations. T also preserves weak equivalences between objects
of the form X = {Xn } where Xn is fibrant for each n. This is because
Y
Y
Y
Dn :
hom(1 , Xn )
Xn Xn
n

will be a fibration, so we may apply the definition and homotopy invariance

property of homotopy pullback given in Exercise 1.8. Thus T has a total right
derived functor RT : Ho(tow(C)) Ho(C), right adjoint to LF . However,
LF is the total left derived functor of the constant diagram functor; hence, by
uniqueness of adjoints
RT
= holim : Ho(tow(C)) Ho(C).

Since T preserves weak equivalences among objects X = {Xn } with all Xn

fibrant one has for such X, holim X ' RT (X) ' T (X).

334

VI. The homotopy theory of towers

2. Postnikov towers.
This section presents basic facts about Postnikov towers, including Moores
construction. The material on k-invariants, which uses cohomology with twisted
coefficients is presented in Section 4.
Definition 2.1. Let X be a space. A Postnikov tower {Xn } for X is a tower
of spaces
q1

q0

X2 X1 X0
equipped with maps in : X Xn so that qn in = in1 : X Xn1 and so that
for all vertices v X, i Xn = 0 for i > n and

(in ) : i X i Xn
for i n.
The purpose of the next few sections is to construct such and to prove a
uniqueness theorem.
F
Remark 2.2. If X is not connected we may write X = X as the disjoint
union of its components.
If {(X )n } is a Postnikov tower for X , then, by
F
setting Xn = (X )n we create a Postnikov tower for X. Conversely, if {Xn }
is a Postnikov tower for X and v X X is a vertex of X , then if (X )n Xn
is the component of in (v), {(X )n } is a Postnikov tower for X . Hence without
loss of generality, we may assume X is connected.
Remark 2.3. We also may as well assume X is fibrant. For if X Y is a weak
equivalence with Y fibrant, then a Postnikov tower for Y is a Postnikov tower
for X.
We now give a specific model for the Postnikov tower, due to Moore. For this
reason it is called the Moore-Postnikov tower. It is pleasing to note that it is
functorial in fibrant X.
Definition 2.4. Let X be a fibrant simplicial set. Define, for each integer n 0,
an equivalence relation n on the simplices of X as follows: two q-simplices
f, g : q X
are equivalent if

f = g : skn q X;

that is, the classifying maps f and g agree on the n-skeleton. Define X(n) =
X/n .

2. Postnikov towers

335

Then there are evident maps qn : X(n) X(n 1) and in : X X(n)

yielding a map of towers : {X} {X(n)}. The principal result of this section
is
Theorem 2.5. The tower {X(n)} is a Postnikov tower for X, and it is a tower
of fibrations. Furthermore, the evident map
X lim X(n)

is an isomorphism.
Note that it is a simple exercise that {X(n)} is a tower of fibrations. Thus it
suffices to prove
Lemma 2.6. For any choice of base point in X, k X(n) = 0 for k > n and
(in ) : k X k X(n) is an isomorphism for k n.
Before proving this, let us recall a combinatorial definition of k X, with X
pointed and fibrant. An element in k X can be represented by a pointed map
f : S k = k /k X
Two such maps yield the same element in k X if they are related by a pointed
homotopy
H : S k 1+ X.
Proof of Lemma 2.6: First notice that k X(n) = 0 for k > n. This is because
any representative of an element in k X(n)
f : S k X(n)
has the property that the composite
f

k S k
X(n)
is the constant map, by the definition of n . Next, let E(n) X be the fiber
of the projection X X(n) at some vertex v. Then E(n) consists of those
simplices f : q X so that
f : skn q X
is constant. In particular E(n) is fibrant and E(n)f = {v} for k n, where v
is the chosen base point. Hence k E(n) = 0 for k n. The result now follows
from the long exact sequence of the fibration
E(n) X X(n).

336

VI. The homotopy theory of towers

The complex E(n) X of this proof is called, by Moore, the n-th Eilenberg
subcomplex of X. It depends on the choice of base point. It is worth pointing
out that X(0) has contractible components; however, it is not necessarily true
that these componets have a single vertex. Indeed X(0)0 = X0 , so that, for
example, X(0) is a one-point space if and only if X0 = {v}.
Corollary 2.7. Let X be fibrant and connected, and v X a vertex. Let
v X(n) be the image of v under X X(n) and K(n) the fiber at v of the
projection qn : X(n) X(n 1). Then there is a weak equivalence
K(n) K(n X, n).
Proof: This follows from the long exact sequence of the fibration sequence
K(n) X(n) X(n 1).

In particular, since X is assumed to be connected,

K(1) X(1)
is a weak equivalence, so there is a weak equivalence
X(1) ' K(1 X, 1) = B1 X.
The following is also worth noting:
Proposition 2.8. Suppose X is connected, fibrant, and minimal. Then qn :
X(n) X(n 1) is a minimal fibration and X(0) = {v} where v X is the
unique vertex. Furthermore K(n) is a minimal complex and there are isomorphisms, n 1,

and

K(n)
= K(n X, n)
K(1)
= B1 X.

Next, let us remark that there is a relative version of the Moore construction.
Definition 2.9. Let f : X B be a morphism of simplicial sets. Then a
Postnikov tower for f is a tower of space {Xn } equipped with a map
i : {X} {Xn }
from the constant tower and a map
p : {Xn } {B}

2. Postnikov towers

337

to the constant tower so that

in

pn

1) for all n, the composite X X(n) B is f .

2) for any choice of vertex of X, the map (in ) : k X k X(n) is an
isomorphism for k n;
3) for any choice of vertex of X, the map (pn ) : k X(n) k B is an
isomorphism for k > n + 1.
4) for any choice of vertex v of X, there is an exact sequence

0 n+1 X(n) n+1 B

n F
where Fv is the homotopy fiber of f at v and is the connecting homomorphism in the long exact sequence of the homotopy fibration Fv X B.
Definition 2.10. Suppose f : X B is a fibration and B is fibrant. Define
an equivalence relation n as simplices
, : q X
by saying n if and only if
1) f = f and
2) f = g : skn q X.
Then there are maps in : X X(n) and pn : X(n) B and the appropriate
generalization of Theorem 2.5 is:
Theorem 2.11. The tower {X(n)} is a Postnikov tower for f , and it is a tower
of fibrations. Furthermore, the evident map
X holim X(n)

is an isomorphism.
Proof: Again we need only check the statements about homotopy groups.
Choose a vertex in X and let F (n) be the fiber of the fibration X(n) B. Then
a moments thought shows the notation F (n) is unambiguous F (n) really is
the n-th stage in the Moore-Postnikov tower of the fiber F of f : X B. The
result thus follows from Lemma 2.6 and the long-exact sequence of the fibration
F (n) X(n) B.

Note that we have used

338

VI. The homotopy theory of towers

Lemma 2.12. Let f : X B be a fibration with be fibrant and fiber F , for

some vertex of B. Then the fiber of pn : X(n) B is the n-th stage F (n) in
the Moore-Postnikov tower for F .
Corollary 2.13. If F is connected p0 : X(0) B is a weak equivalence.
3. Local coefficients and equivariant cohomology.
The purpose of this section is to demonstrate that cohomology with local
coefficients is representable in an appropriate homotopy category, and to identify
the homotopy type of the representing object. We develop a calculation scheme
in terms of equivariant cohomology on the universal cover in Section 4.
Let X be a fibrant space (meaning fibrant simplicial set) and = f X its
fundamental groupoid (see Sections I.8, III.1). There is a canonical simplicial
map : X B which associates to the simplex : n X the string
(0) (1) (n)
in arising from the image of the string
0 1 n
in the 1-skeleton of n .
We shall, more generally, consider the category of spaces S B over B. The
objects of this category are the simplicial set maps Y B, and the morphisms
are commutative diagrams of simplicial set maps
Yh

h
h
j

'
'
*

wZ
'

B
It is trivial to observe the following:
Lemma 3.1. Suppose that X is an object of a closed model category C. Then
the category C X has a closed model structure, in which a morphism
Z[
[
]

wY

X
is a weak equivalence, fibration, or cofibration if the same is true for the map
f : Z Y of C.

3. Local coefficients and equivariant cohomology

339

In particular, the category S B inherits a closed model category from

the category of simplicial sets. Furthermore, the fibrant objects in S B are
fibrations Y B. At this level of generality, the following result gives us an
good selection of fibrations and fibrant objects (compare with Lemma IV.5.7):
Lemma 3.2. Suppose that p : Z Y is a natural transformation of functors
S such that each map p : Zv Yv is a fibration. Then the induced map
p : holim Z holim Y

is a fibration. If all fibrations pv are minimal then p is minimal.

Proof: Consider a commutative solid arrow diagram

nk

w holim Z

i
ij

(3.3)

u i
n

p
u
w holim Y.

We need to show that the dotted arrow exists, making the diagram commute.
The pullback of the canonical map : holim Y B along the composite

holim Y
B
n

is the homotopy colimit of the functor () : n . The category is a

groupoid, so that there is a natural isomorphism of functors on the category n
of the form
X(()(0)) 1 w X(()(0)) 1 w . . .

=u
=u
X(()(0))
w X(()(1))
w ...

1 w X(()(0))

u=
w X(()(n)).

It follows that the induced map n B holim Y n is isomorphic over

n to n X(()(0)) n . Similarly, the map n B holim Z n

340

VI. The homotopy theory of towers

is isomorphic over n to the projection n Z(()(0)) n . The original

diagram (3.3) therefore factors (uniquely) through a diagram of the form

nk

w n Z(()(0))

j
ii

u i
n

1p
u
w n Y (()(0))

The map 1 p is a fibration since p is a fibration by assumption, so the dotted

arrow exists. Observe further that if all fibrations p : Zv Yv are minimal, then
1 p is minimal, and so p is minimal.

Corollary 3.4. Suppose that K : S is a functor such that Kv is fibrant

for each v, then the canonical map : holim K B is a fibration.

A local coefficient system is a functor A : Ab. We shall write Ax for A(x).

Note there is nothing special about abelian groups. If C is a category, a local
system of objects in C is simply a functor C. In particular, the functors of
Lemma 3.2 are local systems of simplicial sets.
The cohomology of X with local coefficients in A can be defined via cochains.
The group Cn (X, A) of n-cochains with coefficients in A is defined to be the
collection of commutative diagrams

Xn

w holim An

N
N
N
Q

Bn
or rather Cn (X, A) is the collection of all assignments x 7 (x) A(x)(0) ,
x Xn .
All calculations in such groups of cochains depend (initially) on the following
observation:
Lemma 3.5. Suppose that : Y B is a second space over B, : m n
is an ordinal number map, and that f : Yn Xm is a function such that the

3. Local coefficients and equivariant cohomology

341

diagram
f

Yn

u
Bn

w Xm

u
w Bm

commutes. Then, given any cochain : Xm holim Am , there is a unique

cochain f () : Yn holim An such that the diagram

Yn

f ()

w holim An

u
Xm

u
w holim Am

commutes.
Proof: Recall that the simplicial structure map : holim An holim Am

is defined on the component A(0) of

G
holim An =
A(0)

(0)(1)(n)

corresponding to the simplex : n by the composite

(0,(0))

in

A(0) A((0)) holim Am ,

where (0, (0)) denotes the composite of the morphisms

(0) (1) ((0))
in . For y Yn , f ()(y) is therefore the unique element in A(y)(0) which
maps to (f (y)) A(f (y))(0) = A (y)(0) = A(y)(0) under the isomorphism
(y)(0,(0))

A(y)(0) A(y)(0) .

342

VI. The homotopy theory of towers

Lemma 3.5 gives a cosimplicial object C (X, A) write H (X, A) for the
corresponding cohomology groups.
Suppose that y is an object of the groupoid . Recall that the groupoid
y has for objects all morphisms : x y, and has morphisms given by
commutative diagrams in of the form

xh

h
j
h

w x0
'
' 0
'
*

Write Ey for the nerve of the category y. The functor y which

forgets y induces a simplicial set map y : Ey B, whereas composition with
: y z induces a functor : y z which commutes with forgetful
functors. It follows that there is a functor S defined by y 7 Ey which
is fibred over B. Note that each of the categories y has a terminal object,
namely the identity 1y : Y y, so that all spaces Ey are contractible. Observe
finally that the maps y : Ey B are Kan fibrations.
Suppose that : X B is a simplicial set map, and define a collection of
y by forming pullbacks
spaces X

y
X

w Ey
y

u
X

u
w B

: S. We shall call X
the covering system
In this way, we define a functor X
for ; the construction specializes to a -diagram of covering spaces for X in the
case where is the fundamental groupoid of X and : X B is the canonical
map.
Write S for the category of diagrams X : S taking values in simplicial
sets. There is plainly a functor holim : S S B defined by taking

defines a functor
homotopy colimits. The covering system construction X 7 X
S B S . These two functors are adjoint:

3. Local coefficients and equivariant cohomology

343

Lemma 3.6. There is a natural bijection

Y)
homS (X,
= homSB (X, holim Y ).

y has n-simplices (X
y )n
Proof: For each object y , the simplicial set X
consisting of all pairs of the form (x, ), where x Xn and : (x)(0) y is a
y )n (X
y )m associated to
morphism of . The simplicial structure map : (X
: m n is defined by
(x, ) = ( (x), (x)(0, (0))1 ).
There is a natural bijection
(3.7)

n , A)
homSets (X
= homSetsBn (Xn , holim An )

in each degree n. In effect, if A : Sets is an arbitrary set valued functor,

then a natural transformation
y )n Ay y ,
gy : (X
is completely determined by its effect on the elements (x, 1(x)(0) ), x Xn ,
and g(x, 1(x)(0) ) A(x)(0) , so the assignment x 7 g(x, 1(x)(0) ) determines an
element of homSetsBn (Xn , holim An ).

One checks to see that the isomorphisms (3.7) assemble to give the desired
adjunction on the simplicial set level.

If Y : S is a -diagram of simplicial sets and A : Ab is a local

coefficient system, there is an obvious way to form a cochain complex (aka.
cosimplicial abelian group) hom (Y, A), having n-cochains given by the group
hom (Yn , A) of all natural transformations from Yn to A.
Corollary 3.8. There is a natural isomorphism of cosimplicial abelian groups
A)
C (X, A)
= hom (X,
for all spaces X over B and all local coefficient systems A.
Proof: Use the calculation in the proof of Lemma 3.5 to show that the natural
bijections (3.7) from the proof of Lemma 3.6 preserve cosimplicial structure.

344

VI. The homotopy theory of towers

The category S has a simplicial model structure for which a map f : Z W

is a weak equivalence (respectively fibration) if and only if each of the components f : Zv Wv , v , is a weak equivalence (respectively fibration) of
simplicial sets. This is the Bousfield-Kan structure for -diagrams see Example II.7.9. The homotopy colimit construction takes weak equivalences of S
to weak equivalences of S B (Proposition IV.1.9), whereas the covering system functor S B S preserves weak equivalences by the coglueing lemma
(Lemma II.9.12). There are therefore induced functors
Ho(S ) Ho(S B)
which are adjoint, and so the natural isomorphism of Lemma 3.6 extends to a
natural bijection
Y ]S
[X,
= [X, holim Y ]SB

on the homotopy category level.

To go further, we need to know the following:

Lemma 3.9. For any object : X B, the corresponding covering system X

is a cofibrant object of S .
Proof: Suppose that the groupoid is connected, and pick an object v .
Write = hom (v, v), and pick a morphism x : x v in for each object x,
such that v = 1v .
There is an inclusion functor i : which identifies the group with the
group of automorphisms of v in . There is also a functor r : which is
defined by sending : x y to the composite y x1 : v v, and the functors
r and i are the component functors of an equivalence of the (connected) groupoid
with the group : r i = 1 , and the isomorphisms x : x v determine a
natural isomorphism 1
= i r.
The functors i and r induce functors i : S S and r : S S , just by
precomposition. The composite i r is the identity functor on S , while the
isomorphisms x : Xv Xx define a natural isomorphism r i (X)
= X, so
that S and S are equivalent categories.
v is a cofibrant
The functor i reflects cofibrations, so it suffices to show that X
-space. To see this, observe that the functor r induces a commutative diagram
Ev

w E

u
B

u
w B,

3. Local coefficients and equivariant cohomology

345

and this diagram is a pullback. But then there is a -equivariant isomorphism

(Ev )n
= Bn ; in other words, Ev is a free -set in each degree, so it must
be a cofibrant -space, by Corollary V.2.9.

n
We can now relate elements in the cohomology groups H (X, A) arising from
fibred spaces X B and local coefficient systems A : Ab to morphisms
in some homotopy category. The method is to use a -equivariant version of the
Dold-Kan correspondence (Section III.2), in conjunction with Lemma 3.9.
First of all, recall from Corollary 3.8 that Hn (X, A) is the nth cohomology
A) which is defined in degree n by the
group of the cochain complex hom (X,
n , A) of natural transformations X
n A. It follows that H n (X, A)
set hom (X

A[n]) of chain homotopy classes

is canonically isomorphic to the group (Z(X),
of morphisms of -diagrams of chain complexes from the -diagram of Moore
to A[n], where A[n] is the -diagram of chain complexes which
chains Z(X)
consists of a copy of A concentrated in degree n. The normalized chain complex
is naturally chain homotopy equivalent to Z(X),
so that there is an
N Z(X)
isomorphism
i
A[n]) (N Z(X),
A[n])
(Z(X),

, Z(X).

which is induced by the inclusion i : N Z(X)

But now there are
isomorphisms
K(A, n))
A[n])
(N Z(X),
= (Z(X),

K(A, n))
= (X,

K(A, n)] ,
= [X,
where the first isomorphism is induced by the Dold-Kan correspondence, the
second isomorphism relates naive homotopy classes in -diagrams of simplicial
abelian groups to naive homotopy classes in -diagrams of simplicial sets, and the
last isomorphism relating naive homotopy classes to morphisms in the homotopy
is cofibrant and the Eilenbergcategory of -spaces results from the fact that X
Mac Lane object K(A, n) is fibrant. Subject to chasing an explicit cochain
through these identifications (which is left to the reader), we have proved:
Theorem 3.10. Suppose that X B is a space over B, and let A : Ab
be a local coefficient system. Then there is an isomorphism

K(A, n)] Hn (X, A),

[X,
K(A, n) to
which is defined by sending a class represented by a map f : X
the class represented by the cocycle fn : Xn holim An .

346

VI. The homotopy theory of towers

The equivalence of categories i : S S : r induces an equivalence of

associated homotopy categories, giving an isomorphism
i

K(A, n)] [X
v , K(Av , n)] ,
[X,

while the identifications leading to Theorem 3.10 are preserved by r as well

as being valid in the category of -diagrams. This leads to an identification of
H (X, A) with ordinary -equivariant cohomology:
Theorem 3.11. Suppose that X and A are as in the statement of Theorem
3.10, suppose that is a connected groupoid and let be the group hom (v, v)
of an object v . Then restriction to -spaces determines a commutative
diagram of isomorphisms

K(A, n)]
[X,
=
i
u
v , K(Av , n)]
[X

w Hn (X, A)

= i
u
v , Av ).
w Hn (X

v is a copy of the universal cover

Observe, from the proof of Lemma 3.9, that X
of X if the underlying map : X B is the canonical map arising from the
fundamental groupoid of X.
4. Generalities: equivariant cohomology.
There is a notion of reduced cohomology with local coefficients. Suppose
= f X is the fundamental groupoid and the map X B has a section
s : B X. Such a section is a map from the terminal object to X in S B,
which is the natural notion of base point. For example, we could take
X = holim K(A, n)

where s() = (0, ) A pointed map in S

X holim K(A, n)

is a map commuting with sections.

4. Generalities: equivariant cohomology

347

In gross generality, if C is a simplicial model category, then the category C of

pointed objects in C (meaning morphisms t X, where t is terminal) satisfies the
axioms for a closed model category, where a map of C is a fibration, cofibration
or weak equivalence if and only if it is so as a map of C (compare Lemma 3.1).
A reduced cochain Cn (X, A) is a cochain so that ((x)) = 0 for all
e n (X, A) is the group of such there is a split short exact sequence of
x B; if C

cosimplicial abelian groups

e (X, A) C (X, A) C (B, A) 0

0C

and a split short exact sequence of graded groups

e (X, A) H (X, A) H (B, A) 0.
0H
Of course, if X is connected, then for any vertex v X
H (B, A)
= H (1 (X, v), Av ),
by Theorem 3.11.
The following is left as an exercise for anyone who has read the proof of
Theorem 3.10. Let S B denote the category of pointed objects over B.
Lemma 4.1. For X S B there are natural isomorphisms
e n (X; A).
[X, holim K(A, n)]SB
=H

Now suppose X Y that is, X is a subspace of Y and = f X

= f Y :
This would happen for example, if X and Y have the same 2-skeleton. Define
B X Y by the push-out diagram in S B

X
u
B

wY
u
w B X Y.

Then B X Y S B and we define

e (B X Y ; A).
H (Y, X; A) = H

348

VI. The homotopy theory of towers

For a different point of view, observe that the covering system construction
f for B consists
preserves cofibrations and pushouts, and the covering system B
of contractible spaces, so that the covering system for BX Y is equivalent to the
of the inclusion X
, Y in the category of -diagrams.
homotopy cofibre Y /X
It follows that there is an isomorphism
K(A, n)] .
Hn (Y, X; A)
= [Y /X,
Proposition 4.2. There is a long exact sequence
Hn (Y, X; A) Hn (Y ; A) Hn (X; A) Hn+1 (Y, X; A)
Proof: One easily checks
e (B X Y ; A) C (Y ; A) C (X; A) 0
0C
is exact.
Alternatively, one could use Theorem 3.10, Corollary 4.1, and the Puppe sequence. For this, let C denote the loop space functor in any simplicial model
category C. Then Lemma 3.2 implies that there is a natural weak equivalence
SB (holim K(A, n)) ' holim K(A, n 1).

We shall now write down some variants of a spectral sequence (Proposition

4.5) which computes equivariant cohomology.
Fix a discrete group , and observe that, if A is a Z module then for all
n 0, K(A, n) is a fibrant object in the model structure for S of Section V.1.
Let X S be cofibrant, and suppose that n 0. We have already seen that
there is a natural isomorphism
Hn (X, A)
= [X, K(A, n)]
in the proof of Theorem 3.11. If X is not cofibrant we make the following:
Definition 4.3. For arbitrary -spaces X, define Hn (X, A) by setting
Hn (X, A) = [X, K(A, n)] .
The moral is that if X is cofibrant, then Hn (X, A) has a cochain description;
otherwise, take a weak equivalence Y X and Y cofibrant in S , so that
[X, K(A, n)]
= [Y, K(A, n)]
= Hn (Y, A).
While less important for our applications, this also tells one how to define
homology.

4. Generalities: equivariant cohomology

349

Definition 4.4. Let A be a Z module, X S and Y X a weak equivalence

with Y cofibrant. Then define
Hn (X, A) = n (ZY Z A).
The standard model category arguments (cf. Section II.1) show that H (X, A)
is independent of the choice of Y .
For calculational purposes one has the following result. If X S , then
H X = H (X, Z) is a graded Z module.
Proposition 4.5. There is a first quadrant cohomology spectral sequence
ExtpZ (Hq X, A) Hp+q (X, A).
Proof: We may assume X is cofibrant and, indeed, that Xn is a free -set
for all n, by Lemma V.2.4. Let P () = { P1 () P0 () ()} denote
a functorial projective resolution for example the bar resolution of Z
modules. We require each Ps () to be exact. Form the double complex
homZ (P (ZX), A) = {homZ (Pp (ZXq ), A}.
Filtering by degree in p we get a spectral sequence with
E1p,q = homZ (Pp (Hq X), A)
so that E2 is as required. To determine what the spectral sequence abuts to,
filter by degree in p, whence
E1p,q = ExtpZ (ZXq , A).
But ZXq is a free Z module, so E1p,q = 0 if p > 0 and
E10,q = homZ (ZXq , A).
Hence the spectral sequence abuts to E20, = H (X, A), as required.

Proposition 4.5 is a special case of a result that holds in great generality, in

the context of homotopy theories of simplicial sheaves and presheaves [45], [46].
We shall confine ourselves here to displaying a few of its close relatives.
An argument similar to the one given in Proposition 4.5 yields a first quadrant
homology spectral sequence
(4.6)

TorZ
p (Hq X, A) Hp+q (X, A).

350

VI. The homotopy theory of towers

There are further refinements. For example, if A is an F module where F is a

field one has
ExtpF (Hq (X; F), A) Hp+q (X; A).

(4.7)

Example 4.8. Let X = be a point. Then H X = Z concentrated in degree 0

and
p
Hp (; A)
= ExtZ (Z, A) = H q (, A).
Note that unlike ordinary homology, H (; A) need not split off of H (X; A).

The reader is invited to calculate HZ/2Z

(S n , Z) where S n has the antipodal
point action and Z the trivial action. Use the next observation.
Example 4.9. If A is a trivial module, and X is cofibrant,
Hn (X, A)
= H n (X/, A).
This follows from the definition.
There is a relative version of this cohomology. In fact, let S denote pointed
-spaces. For all Z modules A, K(A, n) has an evident base point.
Definition 4.10. For X S , let
e n (X; A) = [X, K(A, n)]S .
H

Hence one factors X as

Y X where i is a cofibration in S and
e n (X; A) = 0 HomS (Y, K(A, n)).
H

Proposition 4.11.
1) For X S there is a split natural short exact sequence
e (X; A) H (X; A) H (; A) 0.
0H
2) There is a first quadrant cohomology spectral sequence
e q X, A) H
e p+q (X; A).
ExtpZ (H
We leave the proof of this, and of the homology analogs, as an exercise.

5. On k-invariants

351

Definition 4.12. Let X Y be an inclusion in S . Define relative cohomology

by
e n (Y /X; A).
Hn (Y, X; A) = H
Proposition 4.13. There is a natural long exact sequence
Hn (Y, X; A) Hn (Y ; A) Hn (X; A) Hn+1 (Y, X; A) .
Proof: Consider the diagram in S

U
'u
X

wV
u'
wY
i

where U X is a weak equivalence with U cofibrant, U

V Y factors
U X Y as a cofibration followed by a weak equivalence. Then V /U is
e (Y /X)
a cofibration and H
= H (V /U ), so
e n (Y /X; A)
e n (V /U, A).
H
=H
Now use the chain description of the various cohomology groups.

5. On k-invariants.
This section is devoted to k-invariants their definition and properties. We
will end with a discussion of the uniqueness of Postnikov towers.
Let X be a connected fibrant space and X = {Xn }n0 a Postnikov tower
for X. Let f Xn1 be the fundamental groupoid for Xn1 and n the local
coefficient system on Xn1 obtained as follows: factor Xn Xn1 as
i

Xn
Y
Xn1
with i a weak equivalence and q a fibration. Then if v Xn1 a vertex, let
Fv = q 1 (v) and
n (v) = n (Fv ).
If f Xn = f Xn1 , as in the case of the Moore-Postnikov tower, there is obviously no ambiguity in n (v), since the fiber has a single vertex. In general,
however, the fiber is simply connected, and for any two choice of vertices x and

352

VI. The homotopy theory of towers

y in Fv there is a canonical isomorphism n (Fv , x)

= n (Fv , y), so we identify
the two groups via this isomorphism and obtain n (Fv ).
Let p : L(, n+1) K(, n+1) be a fixed functorial fibration with L(, n+1)
contractible. Setting L(, n + 1) = W K(, n) would certainly suffice this has
the advantage that the corresponding fibration p : L(, n + 1) K(, n + 1) is
minimal (Lemma III.2.19).
The cohomology class kn in the statement of the following result is known as
the nth k-invariant. follows:
Proposition 5.1. There is a cohomology class kn Hn (Xn1 , n ) and a homotopy pullback diagram over B
Xn

w holim L(n , n + 1)

u
Xn1

kn

u
w holim K(n , n + 1).

This is a consequence of the more general result Theorem 5.9 below. For now,
we state some consequences.
Corollary 5.2. Suppose X is pointed with chosen base point v. Let 1 =
1 (X, v). Then there is a homotopy pullback diagram of over B1
Xn

w holim L(n (v), n + 1)

u
Xn1

u
kn

w holim K(n (v), n + 1).

Proof: This result is equivalent to Proposition 5.1, since there is a homotopy

cartesian square
holim L(n , n + 1)

'

u
holim K(n , n + 1)

w holim L(n (v), n + 1)

u
'

w holim K(n (v), n + 1).

5. On k-invariants

353

Corollary 5.3. If X is simple and pointed, with chosen base point v. Then
there is a homotopy pullback diagram
Xn

w L(n (v), n + 1)

u
w K(n (v), n + 1).

Xn1

Proof: Again, let 1 = 1 (X, v) and consider the homotopy pullback diagram
that appears in Corollary 5.2. Then
holim K(n (v), n + 1) = K(n (v), n + 1) B1

as spaces over B1 and similarly for holim L(n (v), and the diagram
1
prL

L(n (v), n + 1) B1
p1
u
K(n (v), n + 1) B1

prL

w L(n (v), n + 1)
p
u
w K(n (v), n + 1)

is homotopy cartesian.

To prove Proposition 5.1, we pass to essentials. Let f : Y X be a map of

spaces over B, where is the fundamental groupoid of Y . If we factor f as
j

Y
Z
X
where j is a cofibration and q is a weak equivalence, then for each vertex v of Y
we get a relative homotopy group
n (f, v) = n (Z, Y, v) = n1 (Fv , v),
where Fv is the fibre of q over v, or in other words the homotopy fibre of Y X.
This gives a local system n (f ) over B and, if is equivalent to the fundamental
groupoid of X, there is a long exact sequence of local systems
f

n+1 (f ) n (Y ) n (X) n (f ) .

354

VI. The homotopy theory of towers

Lemma 5.4. Suppose f : Y X is a morphism of spaces over B, where is

the fundamental groupoid of Y . Suppose n 2 and for all choices of base point
of Y , the map f : k (Y, v) k (X, f v) is an isomorphism for k < n and a
surjection for k = n. Then for any local system of abelian groups A over B,
there is an isomorphism

f : Hk (X, A) Hk (Y, A),

k<n

and an exact sequence

f

0 Hn (X, A) Hn (Y, A) hom (n+1 (f ), A)

d

Hn+1 (X, A) Hn+1 (Y, A).

The sequence is natural in maps f satisfying the hypotheses.
Proof: The hypotheses on homotopy groups in degree 0 and degree 1 imply
that is equivalent to the fundamental groupoid on X. Hence, the long exact
sequence of local systems above exists, and the groups H (X, A) are defined.
We may assume that f : Y X is a cofibration over B and use the long
exact sequence of 4.2:
Hk (X, A) Hk (Y, A) Hk+1 (X, Y, A) Hk+1 (X, A) .
Note that the results of Section 4 imply
k (B Y X, A)
k (Z, Av )
Hk (Y, X, A)
=H
=H
where Z is the universal cover of B Y X. The latter group may be computed
k (Z, Av ) = 0 for k n and
using Proposition 4.11.2, and one gets H
n+1 Z, Av )
n+1 (Z, Av )
H
= homZ (H

= homZ (n+1 (f, v)), Av )

= hom (n+1 (f ), A),

by appropriate use of the relative Hurewicz theorem (Theorem III.3.12).

In particular, in Lemma 5.4, one can set A = n+1 (f ) and define the kinvariant of f to be
(5.5)

k(f ) = d(1n+1 (f ) ) Hn+1 (X, n+1 (f )).

5. On k-invariants

355

Remark 5.6. Given the identifications that have been made, the k-invariant
k(f ) is also represented by the composite
X/
Y X/
Y (n + 1) ' K(n+1 (X/
Y ), n + 1)
X
= K(n+1 (f ), n + 1)
Y (n + 1) is meant
in the homotopy category of -diagrams. The object X/
Y this
to denote the (n + 1)st Postnikov section of a fibrant model of X/
is sensible, because the Postnikov section construction is functorial. The relative Hurewicz isomorphism is also functorial, so that there is an isomorphism
Y )
n+1 (X/
= n+1 (f ) of local coefficient systems.
One possible way to build maps f : Y X satisfying the hypotheses of
Lemma 5.4 is to kill a cohomology class by the following method. Let X be
space with fundamental groupoid and let A be a local coefficient system. Fix
x Hn+1 (X, A) with n 2 and form the pullback diagram
Y

w holim L(A, n + 1)

(5.7)

u
X

u
w holim K(A, n + 1)

where represents x. By construction, n+1 (p)

= A and we have the following
result.
Lemma 5.8. There is an equality k(p) = x Hn+1 (X, A).
Proof: By naturality, we may take to the identity, so that f is the projection
q : holim L(A, n + 1) holim K(A, n + 1)

and x is the universal class. Of course, q is not a projection, but the zero section
s : B holim L(A, n + 1) is a weak equivalence, and we may calculate using

the composite
q s : B holim K(A, n + 1).

Then B B holim K(A, n + 1) has as universal cover

Z = holim K(A, n + 1) ' K(Av , n + 1).

356

VI. The homotopy theory of towers

Thus
n+1 (Z, Av )
H
= hom(Av , Av )
= hom (A, A)
and
d : hom (A, A) Hn+1 (holim K(A, n + 1), A)

is the standard isomorphism taking 1A to the universal class.

We point out that in the diagram (5.7) the object holim K(A, n) is a group

object over B and it acts on the space Y over B; that is, there is a map
: holim K(A, n) B Y Y

over B satisfying the usual associativity and unital conditions. Notice further
that this action descends to the trivial action on X in the sense that the following
diagram commutes
holim K(A, n) B Y

wY

p2

u
X

u
wX

where p2 (a, b) = p(b).

Now suppose we are given f : Y X satisfying the hypotheses of Lemma 5.4.
Then we may form the k-invariant k(f ) as in (5.5), and the pullback diagram
with A = n+1 (f ),
Z
p

u
X

w holim L(A, n + 1)

u
w
holim
K(A,
n + 1).
k(f )

Notice that we have confused the cohomology class k(f ) with a map representing it. This abuse of notation pervades the literature, and is suggestive: the
homotopy type of Z does not depend on the choice of a representative for k(f ).

5. On k-invariants

357

Theorem 5.9. The map f : Y X lifts to a map g : Y Z inducing an

isomorphism
g : n+1 (f ) n+1 (p).
Proof: Since f k(f ) is in the image of the composite
f

hom (A, A)
Hn+1 (X, A) Hn+1 (Y, A)
the composite k(f ) f is null-homotopic, and hence lifts to holim L(A, n + 1).

Choose a lifting, and let h : Y Z be the induced map. We wish to modify h,

if necessary, to a homotopy equivalence.
By the naturality clause of Lemma 5.4, there is a diagram of exact sequences

w Hn (X, A)
=

u
n
w H (X, A)

w Hn (Z, A)
h

u
n
w H (Y, A)

w hom (A, A)

dZ

h]

u
w hom (A, A)

dY

w Hn+1 (X, A)
u=
w Hn+1 (X, A).

Note that dY (1A ) = k(f ) by definition and dZ (1A ) = k(f ) by Lemma 4.5.
Furthermore h] (1A ) = h where we write h for the composite
A = n+1 (f ) n+1 (p)
= A.
Since dY (1A h ) = k(f ) k(f ) = 0, there is a class x Hn (Y, A) mapping to
1A h . Let g : Y Z be the composite

xh

Y holim K(A, n) B Z
Z.

Remarks 5.10.
1) Note that the map g : Y Z of Theorem 5.9 is not unique, but may be
modified by any element of H n (X, A).
2) The proof of Theorem 5.9 can also be carried out entirely within the diagram category, and writing it down that way would be a good exercise.
Now let X be space and let {Xn } be any Postnikov tower for X. Let qn :
Xn Xn1 be the projection. Then k (qn ) = 0 k 6= n+1 and n+1 (qn ) = n X.
Thus Proposition 5.1 follows immediately from Theorem 5.9. More generally, let
f : X B be any map and {Xn } a Postnikov tower for f (see Definition 2.9).

358

VI. The homotopy theory of towers

Again, let qn : Xn Xn1 be the projection. Then k (qn ) = 0 for k 6= n + 1

and
n+1 (qn ) = n (F )
by the relative Hurewicz theorem, and one obtains a homotopy pullback square
Xn

w holim L(n , n + 1)

Xn1

kn

w holim K(n F, n + 1).

We now examine the consequences for Postnikov towers. Let X be a fibrant

connected space and {Xn } a Postnikov tower for X. Let n = n X regarded as
a local coefficient system as Xn1 . Then there is a homotopy cartesian square,
with = f Xn1
Xn
qn

(5.11)

w holim L(n , n + 1)

Xn1

kn

u p
w holim K(n , n + 1)

and Xn is weakly equivalent to the pullback. More is true:

Corollary 5.12. If the tower map qn : Xn Xn1 is a minimal fibration
then the diagram (5.11) is a pullback.
Proof: The map
p : holim L(n , n + 1) holim K(n , n + 1)

is a minimal fibration (Lemma 3.2), and so the induced map

k
Xn n w Xn1 holim K(n ,n+1) holim L(n , n + 1)

qn

u
Xn1

5. On k-invariants

359

is a weak equivalence (and hence a homotopy equivalence) of minimal fibrations

over Xn1 . Lemma I.10.4 implies that the map kn is an isomorphism.

This applies, for example, to the Moore-Postnikov tower {X(n)}. In that case
= f X = f Xn for all n. If X is minimal, then qn : X(n) X(n 1) is a
minimal fibration.
Corollary 5.13. Let X be a connected fibrant minimal space. Then for all
n 2 there is a pullback diagram

X(n)
qn

w holim L(n , n + 1)

X(n 1)

kn

u
w holim K(n , n + 1)

where = 1 X. If the fundamental group 1 (X) acts trivially on n (X), then

the map qn : X(n) X(n 1) is a principal K(n , n)-fibration.
Proof: If X is minimal and connected it has a single vertex, so
holim K(n , n + 1) = holim K(n , n + 1),

etc. If 1 acts trivially on n , then

p : holim L(n , n + 1) holim K(n , n + 1)

can be identified up to isomorphism with the map

p 1 : L(n , n + 1) B K(n , n + 1)
as in the proof of Corollary 5.3, and p 1 is a principal K(n , n)-fibration.

Theorem 5.14. Let X be a connected fibrant space. Then any two Postnikov
towers for X are weakly equivalent as towers under X.
Proof: Choose a minimal subcomplex X0 X which is a weak equivalence and
let v X0 be the vertex. Choose a retraction g : X X0 . Then the MoorePostnikov tower {X0 (n)} is a Postnikov tower under X. We will show that any
Postnikov tower {Xn } under X is weakly equivalent (under X) to {X0 (n)}.

360

VI. The homotopy theory of towers

Inductively define tower maps fn : Xn X0 (n) as follows. Let = 1 (X, v).

Then for n = 1 choose a weak equivalence X1 X0 (1) = B. Suppose Xk
X0 (k) have been defined, and are compatible and weak equivalences for k < n.
Then Theorem 5.9 implies that there is a diagram
X0 (n)
j
h
h
u
h
Xn
w X0 (n 1)

w holim L(n , n + 1)

fn

u
w holim K(n , n + 1)

where n = n (X, v), and fn is a weak equivalence. Thus we have a weak

equivalence of towers f : {Xn } {X0 (n)}. To modify this is into a weak
equivalence of towers under X, consider the induced map
f : holim Xn lim X0 (n) = X0 .

The canonical map X holim Xn is a weak equivalence, and the induced map

f : X X0 is a weak equivalence. There is an isomorphism : X0 X0 so

that
f

X0
X0
X
is g. Since the Moore-Postnikov tower is natural we get an induced isomorphism
of towers : {X0 (n)} {X0 (n)} and the composite
f : {Xn } {X0 (n)}
is a weak equivalence of towers under X.

6. Nilpotent spaces.
We now describe how to refine a Postnikov tower for a nilpotent space. A
group G is nilpotent if the lower central series eventually stabilizes at the trivial
group. Thus, if we define Fn G G by F0 G = G and Fn G = [Fn1 G, G], we
are asking that there be an integer k so that Fk G = {e}. If G is a group, a
G-module M is nilpotent if there is finite filtration of M by G-modules
0 = Fk M Fk1 M F1 M M
so that G acts trivially on the successive quotients. A simplicial set X is nilpotent
if X is connected, 1 X is a nilpotent group and n X is nilpotent 1 X module
for n 2.

6. Nilpotent spaces

361

Proposition 6.1. Let X be a nilpotent space and {Xn } a Postnikov tower for
X. Each of the maps qn : Xn Xn1 can be refined to a finite composition
Xn = Yk Yk1 Y1 Y0 = Xn1
so that each of the maps Yi Yi1 fits into a homotopy pullback square
Yi

w L(Ai , n + 1)

Yi1

w K(Ai , n + 1)

for some abelian group Ai

Proof: We begin with the case n = 1. Then X1 X0 is weakly equivalent to
B1 where 1 = 1 X. Let {Fi } be the lower central series of 1 X and let
Gi = 1 /Fi . Let k be an integer so that Fk = {e}. Then there is a tower
B1 = BGk BGk1 BG1 .
Since
{e} Fi1 /Fi Gi Gi1 {e}
is a central extension, there is homotopy pullback diagram, with Ai = Fi1 /Fi ,
BGi

w L(Ai , 2)

u
BGi1

u
w K(Ai , 2).

In effect, the map BGi1 K(Ai , 2) classifies the principal BAi -fibration
BGi BGi1 .
Next we assume that n 2. Let {Fi } be a filtration of n X by 1 modules so
that each of the successive quotients is a trivial 1 module. We define Yi by the
homotopy pullback diagram
Yi

w holim K(Fi , n + 1)

Xn1

w holim K(n X, n + 1)

362

VI. The homotopy theory of towers

where the bottom map is the k-invariant. Since there is a homotopy pullback
diagram
holim K(Fi , n + 1)

w holim L(Fi1 /Fi , n + 1)

u
holim K(Fi1 , n + 1)

w holim K(Fi1 /Fi , n + 1).

there is a homotopy pullback diagram

Yi

w holim L(Fi1 /Fi , n + 1)

Yi1

w holim K(Fi1 /Fi , n + 1).

Since 1 acts trivially on Fi1 /Fi , one can finish the proof by arguing as in
Proposition 5.3.

The previous result has the following consequence.

Corollary 6.2. Let X be a nilpotent space. Then there is a tower of fibrations
{Zj } so that X ' lim Zj and each of the maps Zj Zj1 fits into a homotopy

pullback square
Zj

w L(Aj , nj + 1)

u
w K(Aj , nj + 1)

Zj1

where each Aj is an abelian group and {nj } is a non-decreasing sequence of

positive integers so that lim nj =

Further variations on this result are possible. For example, one can remove
the hypothesis that X be connected, or consider nilpotent fibrations. It is also
possible to construct this refined Postnikov tower by a method similar to the
construction of the Moore-Postnikov tower. See Bousfield and Kan [14].

Chapter VII Cosimplicial spaces

One of the most commonly used tools in homotopy theory is the homotopy
spectral sequence of a cosimplicial space. This first appeared in the work of
Bousfield and Kan [14] and has been further analyzed by Bousfield [10]. Two
of the standard examples include the Bousfield-Kan spectral sequence an
unstable Adams spectral sequence that arose before the general example [7]
and the spectral sequence for computing the homotopy groups of the homotopy
inverse limit of a diagram of pointed spaces. One of the main purposes of this
chapter is to define and discuss this kind of spectral sequence. We will do much
more however; for example, we will give a detailed analysis of the total tower of
a cosimplicial space, which is the basic object from which the spectral sequence
is derived.
We begin with a discussion of skeleta in a very general context. This is the
purpose of Section 1, and the main results are Propositions 1.8 and 1.13, which
discuss to what extent a simplicial object in a category C with enough colimits can
be built by attaching cells. One application is a characterization of cofibrations
in the kind of model category considered in Section II.5. See Example 1.14.
If the category C happens to itself be a closed model category, then the category
sC inherits a closed model category structure in which a morphism X Y in
sC is a weak equivalence in and only if Xn Yn is a weak equivalence for all
n 0. This is the Reedy model category structure on sC and one of the main
results here is that there is a geometric realization functor
| | : sC C
which preserves weak equivalences between Reedy cofibrant objects. See Proposition 3.6. The Reedy model category structure is discussed in detail in Section
2.
In Section 4, this theory is specialized to the case of cosimplicial spaces cS;
that is, to the case of the opposite category to the category of simplicial objects
in Sop . The resulting model category structure on cS probably the original
example of a Reedy model category structure is the one discussed by Bousfield
and Kan, and it has some unusual features. In particular, one can give a simple
characterization of cofibrations. This is done in Proposition 4.16.
If x cS, the total space of Tot(X) of X is the geometric realization of X;
that is,
Tot() = | |op : cS = (sS)op (Sop )op .
As a result, if X cS is fibrant, Tot(X) is the inverse limit of a tower of
fibrations. If X is pointed, the tower is pointed, and there is spectral sequence in
364

1. Decomposition of simplicial objects

365

homotopy groups. This is the spectral sequence of the cosimplicial space. There
are several issues: defining the spectral sequence in low degrees in homotopy,
where one might not have groups, or abelian groups; the question of whether the
spectral sequence converges to Tot(X) or not; and what the E2 term is. This
all occupies Sections 5, 6, and 7. The main examples are discussed at the end of
Section 7.
Finally, there is a possibility that Tot(X) might be empty its an inverse
limit, and inverse limits can be empty. To analyze this question, and to address
the calculation of 0 Tot(X) in general, Bousfield has developed an extensive
obstruction theory in the reference cited above. We give a small, but very useful,
example of this theory in Section 8. The main results are Propositions 8.3 and
8.4.
1. Decomposition of simplicial objects.
Let C be a category with all limits and colimits. The purpose of this section
is to analyze how simplicial objects are constructed out of smaller components.
We will use this inductive argument in later sections.
op
We begin with skeleta. The category sC is the functor category C . Let
in : n be the inclusion of the full subcategory with objects k, k n, and
op
let sn C = C n . There is a restriction function in : sC sn C which simply
forgets the k-simplices, k > n. This restriction functor has a left adjoint given
by
(1.1)

(in X)m = lim Xk = lim Xk

mk
m n

and the colimit is over morphisms m k in with k n. This is an example

of a left Kan extension. Since every morphism m k in can be factored
uniquely as

k
m
k0
where is a surjection and is one-to-one, the surjections m k, k n, are
cofinal in the category of all morphisms used to define the colimit of (1.1) and
we get
(1.2)

(in X)m = lim Xk

+
m n

where + is the subcategory with the same objects but only surjections as
morphisms and +
n = n + .

366

VII. Cosimplicial spaces

If X sC we define the nth skeleton of X by the formula

(1.3)

skn X = in in X.

There are natural maps skm X skn X, m n, and skn X X. In light of

(1.2), (skn X)m = Xm for m n It is clear that there is a natural isomorphism

lim skn X X.

Here is an example. Because C has limits and colimits, sC has a canonical

structure as a simplicial category. (See Section II.2). In particular if X sC
and K S, then
G
(X K)n =
Xn .
Kn

It is now a straightforward exercise to prove

Proposition 1.4. If X sC is constant and K S, then there is a natural
isomorphism
skn X skn K
= skn (X K).
To explain how skn X is built from skn1 X we define the nth latching object
Ln X of X by the formula
(1.5)

Ln X = (skn1 X)n

= lim Xk .

+
n n1

If Z C, we may regard Z as a constant object in sC and there is an adjoint

isomorphism
homC (Z, Xn )
= homsC (Z n , X)
for all n 0. This immediately supplies maps in sC
Ln X n skn1 X
and
Xn n skn X.

1. Decomposition of simplicial objects

367

Furthermore, by Proposition 1.4, skn1 (Xn n ) = Xn skn1 n = Xn n

and we obtain a diagram

(1.6)

Ln X n

w L n X n

u
Xn n

u
w skn1 X.

Proposition 1.7. For all X sC there is a natural pushout diagram, n 0,

Xn n Ln Xn Ln X n

w skn1 X

u
Xn n

u
w skn X

Proof: In light of Proposition 1.4, if we apply in in = skn () to this diagram

we obtain an isomorphic diagram. Since in : sn C sC is a left adjoint, we need
only show this is a pushout diagram in degrees less than or equal to n.
In degrees m < n, the map (Ln X n )m (Ln X n )m is an isomorphism,
so
(Xn n Ln Xn Ln X n )m (Xn m )m
is an isomorphism, and the assertion is that (skn1 X)m
= (skn X)m , which is
true: both are isomorphic to Xm . In degree n, the left vertical map is isomorphic
to
G
G
Xn t Ln X
Xn t Xn
(n )m

(n )m

so the assertion is that the right vertical map is isomorphic to the natural map
Ln X Xn . This is true by the definition of Ln X.

Morphisms also have skeletal fibrations. If f : A X is a morphism in sC,

A
A
define skA
n X by setting sk1 = A and, for n 0, defining skn by the pushout
diagram
skn A
skn f

wA

skn X
w

u
skA
n

X.

The adjunct to Proposition 1.7 is the next result, which is proved in an identical
manner. Let Ln (f ) = (skA
n1 X)n = An Ln A Ln X.

368

VII. Cosimplicial spaces

Proposition 1.8. For all morphisms A X in sC there is a pushout diagram

Xn n Ln (f )n Ln (f ) n

w skA
n1 X

u
Xn n

w skA
n X.

There is a situation under which the pushout diagrams of Proposition 1.7 and
1.8 simplify considerably. This we now explain.
If I is a small category, let I be the discrete category with the same objects
as I, but no non-identity morphism. The left adjoint r to the restriction functor

r : C I C I has a very simple form

(r Z)j =

Zi .

ij

We call such a diagram I-free. More generally, a morphism f : A X in C I is

I-free if there is an I-free object X 0 C I and an isomorphism under A of f with
the inclusion of the summand A A t X 0 . This implies that there is an object

{Zi } C I so that
G
Xj
Zi ).
= Aj t (
ij

Definition 1.9. An object X sC is degeneracy free if the underlying degeneracy diagram is free. That is, if + is the subcategory with same objects but
op
only surjective morphisms, then X regarded as an object in C + is op
+ -free. A
op
morphism A X is degeneracy free if, when regarded as a morphism in C +
it is op
+ free.

If X is degeneracy free, then there is a sequence {Zn }n0 of objects in C so

that
G
Xn
Zm
=
:nm

where runs over the epimorphisms in . A degeneracy free map A X yields

a similar decomposition:
(1.10)

Xn
= An t

G
:nm

Zm .

1. Decomposition of simplicial objects

369

We say A X is degeneracy free on {Zm }. Although we have not made this

explicit before, notice an object X is I-free if and only if the morphism X
from the initial object in I-free.
The following result says that degeneracy free maps are closed under a variety
of operations.
Lemma 1.11.
1) LetFf : A X be
of maps so that f is degeneracy free on {Zn }.
F a set

Then f is free on { Zn }.
2) Suppose f : A X is degeneracy free on {Zn } and
A
f

u
X

wB
u
wY

is a pushout diagram. Then g is degeneracy free on {Zn }.

3) Let A0 A1 A2 be a sequence of morphisms so that Aj1
F j Aj is
j
degeneracy free on {Zn }. Then A0 lim Aj is degeneracy free on { j Zn }.

4) Let F : C D be a functor that preserves coproducts. If f : A X in sC is

degeneracy free on {Zn }, then F f : F A F X is degeneracy free on {F Zn }.
Lemma 1.12. A morphism f : A X is degeneracy free if and only if there are
objects Zn and maps Zn Xn so that the induced map
(An Ln A Ln X) t Zn Xn
is an isomorphism.
Proof: If f is degeneracy free on {Zn }, the decomposition follows from the
formulas (1.5) and (1.10). To prove the converse, fix the given isomorphisms.
op
Then Proposition 1.8 implies there is a pushout diagram in C +
Zn n
u
Z n n

w skA
n1 X
w

u
skA
n

X.

However, the morphism of simplicial n n is degeneracy free on the

A
canonical n-simplex in (n )n ; hence, skA
n1 X skn X is degeneracy free on Zn
in degree n. The result now follows from Lemma RefVII.1.11.3; indeed A X
is degeneracy free on {Zn }.

370

VII. Cosimplicial spaces

Proposition 1.13. Suppose A X is degeneracy-free on {Zn } in sC. Then

for all n 0 there is a pushout diagram,
Zn n

w skA
n1 X

u
Z n n

u
w skA
n X.

Proof: Combine Proposition 1.8 and Lemma 1.12.

Notice that one can interpret this result as saying skA

n X is obtained from
A
skn1 X by attaching n-cells.
Example 1.14. Lemmas 1.11 and 1.12 provide any number of examples of degeneracy free morphisms. For example, a cofibration in simplicial sets is degeneracy free, by Lemma 1.12. Also, consider a category C equipped with a
functor G : C Sets with a left adjoint F and satisfying the hypotheses of
Theorem II.5.1. Define a morphism f : A X in sC to be free (this terminology
is from Quillen) if there is a sequence of sets {Zn } so that f is degeneracy free
on {F Zn }. Then a morphism in sC is a cofibration if and only if it is a retract
of a free map. To see this, note that the small object argument of Lemma II.5.2,
coupled with Lemma 1.11 factors any morphism A B as
j

A
X
Y
where j is a free map and a cofibration and q is a trivial fibration. Thus any
cofibration is a retract of a free map. Conversely, Proposition 1.13 and Proposition II.3.4 imply any free map is a cofibration. Similar remarks apply to the
model categories supplied in Theorems II.6.8 and II.7.8. In the latter case one
must generalize Lemma 1.11.3 to longer colimits.
Dual to the notion of skeleta is the notion of coskeleta. The theory is analogous,
and we give only an outline. The restriction functor in : sC sn C has right
adjoint i!n with
(i!n X)m = lim Xk ,

km

with the limit over all morphism k m in with k n. Equally, one can take
the limit over morphisms k m which are injections. The composite gives the
nth coskeleton functor:
(1.15)

coskn X = i!n in X.

1. Decomposition of simplicial objects

371

More generally, if f : X B is a morphism in sC, let coskB

1 X = B and let
B
coskn X be defined by the pullback
coskB
n X

w coskn X

u
B

u
w coskn B.

(1.16)

B
Then there are maps coskB
n X coskn1 X and
lim coskB X.
X=
n

Note that if B = , the terminal object, then coskB

n X = coskn X.
th
For X sC, define the n matching object Hn X by the formula
(1.17)

Mn X = (coskn1 X)n = lim Xk

kn

where k n runs over all morphisms (or all monomorphisms) in with k < n.
B
To fit the map coskB
n X coskn1 X into a pullback diagram, we make some
definitions. Let n : sC C be the functor n X = Xn . This is a restriction
functor between diagram categories and has a right adjoint !n . The functor
Ln : sC C assigning each simplicial object the latching object Ln X also has a
right adjoint, which we will call !n . To see this, see Lemma 1.23 below, or note
that we may write
Ln X = (skn1 X)n = n in1 i(n1) X
where i(n1) : sC sn1 C is the restriction functor. Hence
!n Z = i!n1 i(n1) !n Z = coskn1 (!n Z).
The natural map s : Ln X X gives a natural transformation !n !n . The
reader is invited to prove the following result.
Proposition 1.18. Let f : X B be a morphism in sC. Then for all n 0
there is a pullback diagram
coskB
n X

w !n Xn

u
coskB
n1

u
X

where Mn (f ) = Bn Mn B Mn X.

!n Mn (f )

!n Mn (f ) !n Xn

372

VII. Cosimplicial spaces

We have never encountered the analog of degeneracy-free morphisms (face

free?) and dont include an exposition here.
The latching and matching object functors Ln and Mn are examples of a much
more general sort of functor, which we now introduce and analyze. We begin
with generalized matching objects.
Proposition 1.19. Let K S be a simplicial set. Then the functor
K : C sC
has right adjoint MK . For fixed X sC, the assignment K 7 MK X induces a
functor Sop C which has a left adjoint.
Proof: If Z sC is constant, homsC (Z, Y ) = homC (Z, Y0 ). Hence we may set
MK X = (X K )0 , since
homsC (Z K, X)
= homsC (Z, X K ).
Furthermore
homC (Z, MK X)
= homsC (Z K, X)
= homS (K, HomsC (Z, X)).

Since homsC (Z n , X)
= homC (Z, Xn ), Mn X
= Xn . More generally, we
can give a decomposition of MK X as follows. Let K be the simplex category
of K. Thus an object of K is a morphism : n K in S and a morphism
in K is a triangle in S
n

wK
[
]
[

u [
.
m

If we define a functor FK : K S by
FK ( : n K) = n
then

lim FK
= K.

Since MK X sends colimits in the K variable to limits we have

(1.20)

MK X
= lim MFK X
= lim X
op
op
K

1. Decomposition of simplicial objects

373

where we write X : K op C to be
X( : n K) = Xn .
Example 1.21. Let : k n be any morphism in S. Then factors
uniquely as a composition
0

k m
n
where 0 is a surjection and is an injection. Thus if K n is any subcomplex, the full subcategory K0 K with objects
: m K
with an injection is cofinal. Equivalently K0op K op is final; hence
MK X
= lim X.
op
K0

In particular Mn X
X where is an injection and k < n.
= lim
:kn k
Thus Mn X
= Mn X. Note that the inclusion map n n induces the
projection
Xn Mn X.
To generalize the latching objects we appropriate the formulation of matching
objects presented in (1.20). Let J be a small category and F : J op . For
op
X sC = C , define the generalized latching object to be
(1.22)

LJ X = lim X F

= lim XF (j) .

We are primarily interested in sub-categories J of op . We write X J for X F .

Lemma 1.23. For fixed F : J op , the functor LJ : sC C has a right
adjoint and, hence, preserves colimits.

374

VII. Cosimplicial spaces

Proof: For homC (LJ X, Z)

= homC J (X F, Z) where Z is regarded as a constant diagram. But

homC J (X J , Z) = homsC (X, F ! Z)

where F ! is the right Kan extension functor.

This restriction to sub-categories of op is practical, but perhaps gratuitous.

An example of a generalized latching object is the following. Let On be the
category with objects the morphisms : n m with surjective and m < n.
The morphisms in On are commutative triangles in under n. Define F :
Onop op by F ( : n m) = m. Then there is a natural isomorphism
Ln X
= LOnop X.
In the next section we will need a decomposition of Ln X. To accomplish this
define sub-categories Mn,k On , 0 k n, to be the full sub-category of
surjections : n m, m < n, with (k) < k. Define, for X sC
X.
Ln,k X = LMop
n,k
Then Ln,0 X = is the initial object in C (since Mn,0 is empty) and Ln,n X =
Ln X, since Mn,n = On .
Also define M(k) On to be the full subcategory of surjections : n m,
m < n, with (k) = (k + 1). Notice that sk : n n 1 is the initial object of
M(k). Hence for X sC
LM(k)op X
= Xn1 .
The reader can verify the following statements about these subcategories.
Lemma 1.24.
1) Mn,k and M(k) are subcategories of Mn,k1 , and every object in Mn,k+1 is
in Mn,k or M(k) (or both).
2) There is an isomorphism of categories
sk : Mn1,k M(k) Mn,k
sending to sk .
3) If is an object of Mn,k (or M(k)) and is a morphism in Mn,k+1 ,
then is a morphism in Mn,k (or M(k)).

1. Decomposition of simplicial objects

375

The following example, illustrating the case n = 3, k = 2, might be helpful.

In the following diagram the symbol 012 near a dot () indicates the object
0 1 2
s s s in the appropriate category. The unlabeled arrows indicated composition
with si for some i; for example 0 02 means s0 7 s1 s0 = s0 s2 .
0
'

1
'
'
' [
''
[
''
)
[ )
u [
^
02
01 '
12
'
''
)u

012

01

M2,2 = O2

M3,2

0
4

1
2
'

' [
4hh
[
'
4
4
6
'
) u
u h
h
h
k
01 Ak
h
12
02
A

C u
A

h
h
u
k
h
12
02

012

012

M(2)

M3,3 = O3

Proposition 1.25. Let X sC. Then there is a pushout diagram in C

Ln1,k X

w Xn1

u
Ln,k X

u
w Ln,k+1 X.

376

VII. Cosimplicial spaces

Proof: Lemma 1.24 implies that

Mn1,k X

w M(k)

Mn,k X

w Mn,k+1

is a pushout in the category of small categories. The same is then true of the
opposite categories. The result follows.

2. Reedy model category structures.

Let C be a closed model category. Using the considerations of the previous
section, this structure can be promoted to a closed model category structure on
the category sC of simplicial objects in C that is particularly useful for dealing
with geometric realization. The results of this section are a recapitulation of the
highly influential, but unpublished paper of C.L. Reedy [80].
In the following definition, let L0 X and M0 X denote the initial and final object
of D respectively.
Definition 2.1. A morphism f : X Y in sC is a
1) Reedy weak equivalence if f : Xn Yn is a weak equivalence for all
n 0;
2) a Reedy fibration if
Xn Yn Mn Y Mn X
is a fibration for all n 0;
3) a Reedy cofibration if
Xn Ln X Ln Y Yn
is a cofibration for all n 0.
The main result is that this defines a model category structure on sC. Before
proving this we give the following lemma.

2. Reedy model category structures

377

Lemma 2.2. A morphism f : X Y in sC is a

1) Reedy trivial fibration if and only if
Xn Yn Mn Y Mn X
is a trivial fibration for n 0;
2) a Reedy trivial cofibration if and only if
Xn Ln X Ln Y Yn
is a trivial cofibration for all n 0.
Proof: Let n,k = d0 n1 dk n1 n , 1 k n. Then
n,1 = and n,n = n . There are pushout diagrams 1 k n 1

n1,k

w n1

n,k

u
w

n,k+1

Taking matching objects yields a natural pullback square

Mn,k+1 X

w Xn1

Mn,k X

w Mn1,k X

where we have written Mn,k X for Mn,k X. This yields a pullback square

(2.3)

Yn Mn,k+1 Y Mn,k+1 X

w Xn1

Yn Mn,k Y Mn,k X

w Yn1 Mn1,k Y Mn1,k X.

378

VII. Cosimplicial spaces

Assume f : X Y is a Reedy fibration. Then X0 Y0 is a fibration and

this begins an induction where the induction hypothesis is that
(2.4)

Xn1 Yn1 Mn1,k Y Mn1,k X

is a fibration. To complete the inductive step, one uses (2.3) to show

(2.5)

Yn Mn,k+1 Y Mn,k+1 X Yn Mn,k Y Mn,k X

is a fibration for all k, 1 k n 1. Since composites of fibrations are

fibrations and Xn Yn Mn Y Mn X is a fibration, we close the loop.
Now suppose f : X Y is a Reedy trivial fibration. Then, inductively, each
of the maps of (2.4) is a trivial fibration. For the inductive step, use (2.3) to
show each of the maps (2.5) is a trivial fibration. Then the axiom CM2 of
closed model categories and the fact that Xn Yn a trivial fibration finishes
the argument. In particular Xn Yn Mn Y Mn X is a trivial fibration for all n.
Conversely, suppose Xn Xn Mn Y Mn X is a trivial fibration. Then one
again runs the same sort of argument to conclude Xn Yn is a trivial fibration.
The argument for part 2) is similar, using the pushout diagram

Ln1,k X

w Xn1

Ln,k X

w Ln,k+1 X

of Proposition 1.25.

We break the proof of the verification of the Reedy model category structure
into several steps.
Lemma 2.6. The Reedy structure on sC satisfies the lifting axiom CM4.
Proof: Suppose we are given a lifting problem in sC

A
j

u
B

wX
u
wY

2. Reedy model category structures

379

where j is cofibration, q is a fibration and either j or q is a weak equivalence.

We inductively solve the lifting problems
skA
n1 B

wX

u
wY

skA
n

for n 0. Because of the pushout diagram of Proposition 1.8 it is sufficient to

solve the lifting problems, with Ln (j) = An Ln A Ln B
Ln (j) n Ln (j)n Bn n

wX

u
Bn n

u
w Y.

By an adjunction argument this is equivalent to the lifting problem

An Ln A Ln B

w Xn

u
Bn

u
w Yn Mn Y Mn X.

This is solvable by Lemma 2.2.

For the proof of the factorization axiom we need to know how much data we
need to build a simplicial object.
Lemma 2.7. Let X sC. Then skn X is determined up to natural isomorphism
by the following natural data: skn1 X, Xn , and maps Ln X Xn Mn X so
that the composite Ln X Mn X is the canonical map.
Proof: The given map Xn Mn X is adjoint to a map Xn n X which
factors uniquely through skn1 X. The data listed thus yields the pushout square
Ln X n Ln Xn Xn n

w skn1 X

u
Xn n

u
w skn X

380

VII. Cosimplicial spaces

of Proposition 1.7.

op
This can be restricted: let sn C be the functor category C n and X sn C.
Let r X sn1 C be the restriction and r : sn1 C sn C the left Kan extension
functor. Hence (r r X)n = Ln X.
Lemma 2.7 immediately implies

Lemma 2.8. Let X sn C. Then X is determined up to natural isomorphism

by the following natural data: r X, Xn , and maps Ln X Xn Mn X so that
the composite Ln X Mn X is the canonical map.
Proof: Let in : sC sn C be the restriction and in : sn C sC the left adjoint.
Then in in
= 1 and in in = skn . Thus
in X
= skn in X
is determined up to natural isomorphism by skn1 in X
= in1 r X (where in1 :
sn1 C sC is left adjoint to restriction), Xn and maps Ln in X Xn Mn in X
that compose to the canonical map Ln in X Mn in X. But Ln X
= Ln in X and

Mn X = Mn in X.
Lemma 2.9. The Reedy structure on sC satisfies the factorization axiom CM5.
Proof: Let us do the trivial cofibration-fibration factorization (compare the
proof of Lemma IV.3.5).
Let X Y be a morphism in sC and let in X in Y be the induced
morphism in sn C. For each n 0 we construct a factorization
in X Z(n) in Y
in sn C with the property that restricted to sn1 C we get the factorization
i(n1) X Z(n 1) i(n1) Y .
j

For n = 0, simply factor X0 Y0 and X0

Z(0)
Y0 where j is a trivial
cofibration in C and q is a fibration in C. Suppose the factorization in sn1 C
has been constructed. Let r : sn C sn1 C be the restriction functor and
r , r! : sn1 C sn C be respectively, the left and right Kan extension functors.
Thus
r i(n1) X = i(n1) skn1 X and r! i(n1) X = in coskn1 X.

2. Reedy model category structures

381

Furthermore, the morphism r r Z(n 1) Z(n 1) yields a map r Z(n 1)

r! Z(n 1) so that the following diagram commutes
r i(n1) X

w r Z(n 1)

w r! Z(n 1)

w r! i(n1) Y
u

u
in X

w in Y.

At level n this yields a map (write Ln Z = Ln Z(n 1), etc.)

Xn Ln X Ln Z Yn Mn Y Mn Z.
Factor this map as
(2.10)

Xn Ln X Ln Z
Zn
Yn Mn Y Mn Z

where j is a trivial cofibration and q is a fibration. The morphisms j and q yield

diagrams
Ln X

w Xn

w Mn X

u
w Zn

Ln Z
u

w Mn Z

u
w Yn

Ln Y

u
w Mn Y

such that all the horizontal maps are induced from the canonical natural transformation Ln Mn . Hence Lemma 2.8 produces
in X Z(n) in Y.
Finally, define Z sC by Zk = Z(n)k , k n. There is a factoring
j

X
Z
Y
and using (2.10) j is a trivial cofibration by Lemma 2.2 and q is a fibration by
definition.
The other factorization is similar.

We now state

382

VII. Cosimplicial spaces

Theorem 2.11. With the definitions of Reedy weak equivalence, cofibration,

and fibration given in 2.1, sC is a closed model category.
Proof: The axioms CM1CM3 are easy and Lemmas 2.6 and 2.9 prove CM4
and CM5 respectively.

Next suppose C is in fact a simplicial model category. For K S and Y, Z C

write ZK, Z K , and HomC (Y, Z) for the tensor object, the exponential object,
and the mapping space. We use this notation to distinguish it from Z K
defined by the simplicial structure.
The category sC inherits a simplicial structure. If X sC and K S, then
XK and X K are defined level-wise
(XK)n = Xn K and (X K )n = XnK .
The mapping space is defined by the usual formula
HomsC (X, Y )n
= homsC (Xn , Y ).
We call this the internal structure on sC.
Corollary 2.12. With this internal simplicial structure on sC, the Reedy
model category structure is a simplicial model category.
Proof: We claim Mn (Y K )
= Mn (Y )K . For there is a sequence of natural
isomorphisms, Z C,
homC (Z, Mn (Y K ))
= homsC (Z n , Y K )

= homsC ((Z n )K, Y )

= homsC ((ZK) n , Y )

= homsC (Z(Mn Y )K ).
= homsC (XK, Mn Y )
The isomorphism (Z n )K
= (ZK) n follows by a level-wise calculation. The result follows from the claim using Proposition II.3.13.

One can ask if sC in the Reedy model category is a simplicial model category
in the standard simplicial structure obtained by Quillens method (as in the
previous section). The answer is no; for if Z C is cofibrant in C, then
1 d0 : Z 0 Z 1
is a Reedy cofibration, but not, in general, a Reedy weak equivalence. However
we do have

3. Geometric realization

383

Proposition 2.13.
1) Let f : X Y be a Reedy cofibration in sC and j : K L a cofibration in
S. Then
X L XK Y K Y L
is a Reedy cofibration which is a Reedy weak equivalence if f is a Reedy weak
equivalence.
2) Suppose f : X Y is a Reedy fibration in sC and j : K L is a cofibration
in S. Then
X L X K Y K Y L
is a Reedy fibration which is a Reedy weak equivalence if f is a Reedy weak
equivalence.
Proof: By an adjunction argument these two statements are equivalent. Let
K S. Define Ln K Kn to be the degenerate n-simplices; that is
Ln K = lim Kk

:nk

where runs over the surjective morphisms. Then if X sC

G
Ln (X K)
Ln X.
=
xLn K

We leave the rest of the proof as an exercise.

3. Geometric realization.
Suppose C is a simplicial category and X sC. Then the geometric realization
|X| C is defined by the coequalizer diagram
(3.1)

d0

d1

n0

Xm n

:nm

Xn n |X|

where runs over the morphisms of and d0 and d1 or the factor associated to
: n m are respectively
Xm n

w Xn n
w

Xn n

n0

Xm n

w Xm m
w

G
n0

Xn n .

384

VII. Cosimplicial spaces

This is the obvious generalization of the geometric realization of Chapters I and

III. Note that |X| is a coend:
Z
|X| =

X,

where denotes the covariant functor n 7 n on . We discuss the homotopical properties of |X|.
First note that | | : sC C has a right adjoint
(3.2)

Y 7 Y = {Y }.

If we give sC the internal (or level-wise) simplicial structure induced from C, it

follows immediately that if X sC and K S, then
(3.3)

|XK|
= |X|K.

Indeed, homsC (XK, Y )

= homsC (X, (Y K ) ).
Now assume C is a simplicial model category. Endow sC with the Reedy model
category structure. By Corollary 2.12, this is a simplicial model category in the
internal simplicial structure.
Lemma 3.4. The functor () : C sC preserves fibrations and trivial fibrations.
Proof: The first point to be proved is this: if K S and Y C, then
MK (Y )
= Y K.

(3.5)

Let K be the simplex category of K (cf. Example 1.21). Then

MK (Y )
= lim (Y )n
op
K

lim
n
n

= lim Y
= Y K
= Y K.
op
K

Now let X Y be a fibration in C. Then

(X )n (Y )n Mn (Y ) Mn (X )
is isomorphic to

X Y Y n X

and the result follows from Lemma 2.2 and SM7 for C.

3. Geometric realization

385

Proposition 3.6. The geometric realization functor | | : sC C preserves

cofibrations, trivial cofibrations and weak equivalences between Reedy cofibrant
objects.
Proof: Use Lemma 3.4 and Lemma II.8.9.

Implicit in the proof of Lemma 3.4, is the assertion that if Z sC is constant

and K S then there is a natural isomorphism
(3.7)

|Z K|
= ZK.

Indeed, using the isomorphism (3.5), we have

homC (|Z K|, Y )
= homC (Z, Y K ),
and the assertion follows. Therefore, for X sC Reedy cofibrant, the realization
comes with a natural skeletal filtration. Define
skn |X| = | skn X|.
Then Proposition 1.7 and the natural isomorphism of (3.7) yield that there are
natural pushout squares

(3.8)

Xn n Ln Xn Ln n

w skn1 |X|

u
Xn n

u
w skn |X|.

This is because | |, as a left adjoint commutes with all colimits. If X is cofibrant, Proposition 3.6 implies that each of the maps skn1 |X| skn |X| is a
cofibration. Furthermore, again since | | commutes with colimits
(3.9)

lim skn |X|

= |X|.

Finally if X happened to s-free on same set {Zn } of objects in C, then (3.8) can
be refined (as in Proposition 1.13) to a pushout diagram

(3.10)

Zn n

w skn1 |X|

u
Zn n

u
w skn |X|.

386

VII. Cosimplicial spaces

The object skn |X| can also be described as a coend. Let skn be the functor
from to S with
m 7 skn m .
Proposition 3.11. Let X sC. Then there is a natural isomorphism
skn |X|
=

Z
X skn

and this isomorphism is compatible with the skeletal filtrations of source and
target.
Proof: There is a sequence of natural isomorphisms, where
in : sC sn C
is the restriction functor
homC (skn |X|, Y )
= homsC (skn X, Y )

= homsC (i in X, Y )
n

= homsC (X, i!n in Y )

Now i!n in Y
= coskn (Y )
= Y skn . Hence
homC (skn |X|, Y )
= homsC (X, Y skn )
Z

X skn , Y ).
= homC (

4. Cosimplicial spaces.
The language and technology of the previous three sections can be used to
give a discussion of the homotopy theory of cosimplicial spaces; that is, of the
category cS = S of functors from the ordinal number category to simplicial
sets. We go through some of the details and give some examples. It turns out
that cofibrations in cS have a very simple characterization; we close the section
with a proof of this fact.
We begin with two important examples.

4. Cosimplicial spaces

387

Example 4.1. Let R = Fp , the prime field with p > 0, or let R be a subring
of the rationals. The forgetful functor from R-modules to sets has left adjoint
X 7 RX, where RX is the free R-module on X. These functors prolong to
an adjoint pair between simplicial R-modules and simplicial sets. By abuse of
notation we write
R:SS
for the composite of these two functors. Then R is the functor underlying a
cotriple on S and, if X S,
RX
= H (X; R).
Let T : S S be any cotriple on S with natural structure maps : X T X
and : T 2 X T X. If X S is any object, there is a natural augmented
cosimplicial space
X T X
with (T X)n = T n+1 X and
di = T i T : (T X)n (T X)n+1
si = T i T ni : (T X)n+1 (T X)n .
The augmentation is given by : X T X = (T X)0 ; note that
d0 = d1 : X (T X)1 .
In particular, if we let T = R : S S we get an augmented cosimplicial space
X R X
with the property that di , i 1, and si , i 0, are all morphisms of simplicial Rmodules. Furthermore, if we apply R one more time, the augmented cosimplicial
R-module
RX R(R X)
has a cosimplicial contraction; hence

H (R X; R)
=
p

H (X; R),
0,

p=0
p > 0.

The object X R X is a variation on the Bousfield-Kan R-resolution of X.

388

VII. Cosimplicial spaces

Example 4.2. Let J be a small category and SJ the category of J-diagrams in S.

Let J be the category with some objects as J but no non-identity morphisms
J is J made discrete. There is an inclusion functor J J, hence a restriction
functor

r : SJ SJ .
The functor r has a right adjoint r! given by right Kan extension; in formulas
Y
(r! X)j =
Xi
ji

where the product is over morphisms in J with source j. Let

T = r! r : SJ SJ .
Then T is the functor of a cotriple on SJ , and if Y SJ , there is a natural
cosimplicial object in SJ
Y T Y.

(4.3)

This cosimplicial object has the property that the underlying J diagram has
a cosimplicial contraction. Put another way, for each j J, the augmented
cosimplicial space
Yj (T Y )j
has a cosimplicial contraction. We can apply the functor lim () to (4.3) to
J
obtain an augmented cosimplicial space
lim Y lim T Y.

(4.4)

Note that lim (T Y )n can be easily computed because

J
Y
lim r! X
Xj
=

where j runs over the objects on J. This last follows from the isomorphisms
homS (Z, lim r! X)
= homSJ (Z, r! X)

= homSJ (Z, X)
=

homS (Z, Xj )

j
s

where Z S is regarded as a constant diagram in SJ or SJ .

The cosimplicial object (4.4) is used in analyzing the homotopy inverse limit
of Y SJ see Section IX.5.

4. Cosimplicial spaces

389

With these examples in hand, we begin to analyze the homotopy theory of

cS. Since S is a simplicial model category, so is the opposite category Sop . Thus
the category (s(Sop ))op = cS acquires a Reedy model category structure as in
section 2 of this chapter. We take some care with the definitions as the opposite
category device can be confusing.
To begin with, cS is a simplicial category: if K S and X cS define XK
and X K in cS by
(4.5)

(XK)n = X n K

and
(4.6)

(X K )n = (X n )K .

One often writes X K for XK; however, we arein this chapterreserving

the tensor product notation for the external operation on sC. The mapping
spaces functor is then
(4.7)

HomcS (X, Y )n
= homcS (Xn , Y ).

There are also latching and matching objects. At this point the literature goes
to pieces. The matching objects in cS defined in [BK] are the latching objects
in s(Sop ) as defined in Section 1. Since we would hope the reader will turn to
the work of Bousfield and Kan as needed, we will be explicit.
Let X cS. The nth matching space M n X S is the (n + 1)st latching object
Ln+1 X of X s(Sop ). Specifically,
(4.8)

M nX
=

lim

Xk

:n+1k

where : n k runs over the surjections in with k n. Thus n 1. There

are natural maps s : X n+1 M n X and f : X Y in cS is a fibration if and
only if
(4.9)

X n+1 Y n+1 M n Y M n X

is a fibration for n 1.
The reader should be aware of 1) the shift in indices M n X = Ln+1 X and 2)
the superscript versus the subscript: M n X 6= Mn X.

390

VII. Cosimplicial spaces

Similarly, there are latching objects. If X cS, Ln X = Mn+1 X, the matching

spaces of X s(Sop ); thus
Ln X =

lim

Xk

:kn+1

where runs over the injections in . A morphism X Y in cS is a cofibration

if and only if
(4.10)

X n+1 Ln X Ln Y Y n+1

is a cofibration (that is, inclusion) of simplicial sets.

Finally, we define a morphism X Y in cS to be a weak equivalence if
n
X Y n is a weak equivalence for all n 1. Theorem 2.11 and Corollary 2.12
now imply the following result.
Proposition 4.11. With the definitions above, the category cS is a simplicial
model category.
We can give a simple characterization of cofibrations in cS, and, along the
way, show that cS is cofibrantly generated. First, let us define a set of specific
cofibrations.
The functors from cS S
n : X 7 X n , n 0
and

n : X 7 M n X, n 1

all have left adjoints, given by variations on left Kan extension. Indeed, the
adjoint to n is given by the formula
(n Y )k =

:nk

where runs over all morphisms in with source n. This is a left Kan extension.
The adjoint to n is slightly more complicated: if J is the category with objects
surjections n + 1 k in , and r : J the functor sending n + 1 k to k,
then the left adjoint n to n is characterized by
homcS (n Z, X)
= homS J (Z, r X)

= homS (Z, Mn X)

4. Cosimplicial spaces

391

where Z SJ is the constant diagram. Thus n Z is a left Kan extension of a

constant diagram. Note that the natural transformation
Xn Mn1 X
gives a natural transformation
n1 Z n Z.
Define morphisms in cS as follows:
(4.12)

n m

n1 m

n1 m

m
n

im
n

m
n

= n m ,

for n 0, m 0, and

m,k
jn
mk
m
m

m
(4.13)

= n k n1 m
n1
= n m ,
k
n
n
for n 0, 0 k m, m 1.
Lemma 4.14. A morphism f : X Y in cS is a fibration if and only if it
has the right lifting property with respect to the morphisms jnm,k of (4.13). A
morphism in cS is a trivial fibration if and only if it has the right lifting property
with respect to the morphisms im
n of (4.12).
Proof: We prove the trivial fibration case; the other is similar. A lifting problem

m

wX
n
u
m

u
wY

is equivalent, by adjointness to a lifting problem

m

w Xn

u
m

w Yn M n1 Y M n1 X.

Lemma 2.2.2 (remember M n1 X = Ln X, where X s(Sop )) implies that f :

X Y is a trivial fibration if and only if s : Xn Yn M n1 Y M n1 X is a
trivial fibration for all n. The result follows.

392

VII. Cosimplicial spaces

Proposition 4.15. The simplicial model category structure on cS is cofibrantly

generated: the morphisms im
n of (4.12) generate the cofibrations and the morm,k
phisms jn of (4.13) generate the trivial cofibrations.
Proof: In light of Lemma 4.14, the small object argument now applies.

We can use this result to characterize cofibrations. If X cC is a cosimplicial

object in any category, define the maximal augmentation aX by the equalizer
diagram in C
d0

aX X 0 X 1 .
d1

Let d0 : aX X 0 be the natural map.

Proposition 4.16. A morphism f : X Y in cS is a cofibration if and only if
X n Y n is a cofibration in S for all n 0 and the induced map aX aY of
maximal augmentations is an isomorphism.
We give the proof below, after some technical preliminaries.
Let 1 be the augmented ordinal number category; this has objects
n = {0, 1, . . . , n 1}, n 1,
(1 = ) and ordering preserving maps. An augmented cosimplicial object in C
is a functor X : 1 C. We write d0 : X 1 X 0 for the unique map. Note
that the maximal augmentation extends any cosimplicial object to an augmented
cosimplicial object.
Lemma 4.17. Let X be an augmented cosimplicial set, and
n

Z =X

n
[

di X n1

i=0

If d0 : X 1 X 0 isomorphic to the inclusion of the maximal augmentation,

then
G
Zk Xn

with : k n running over all injections, 1 k n, is an isomorphism.

Proof: Out of any cosimplicial set X we may construct a simplicial set Y
without d0 as follows: Yn = X n and
di = sni : Yn Yn1 , 1 i n
si = dni : Yn Yn+1 , 0 i n.

4. Cosimplicial spaces

393

Notice that this construction does not use dn : X n1 X n . Let Zn Yn be

the non-degenerate simplices:
Zn = Yn

si Yn1 .

The standard argument for simplicial sets (see Example 1.14) shows that
G

Yn
=

Zk

:nk

where runs over the surjections in . Unraveling the definitions shows that our
claim will follow if we can show that dn : X n1 X n induces an isomorphism

Z n t Z n1 Zn or, equivalently, an isomorphism

Z n1 Zn dn Xn1 .
First note that dn does induce an injection
dn : Z n1 Zn dn Xn1 .

(4.18)

To show this it is sufficient to show that if y Z n1 , then dn y Zn ; that is, if

n1
n1
S i
S i
y 6
d Xn2 , then dn y 6
d Xn1 . The contrapositive of this statement
i=0

reads: if dn y

n1
S
i=0

i=0

di Xn1 , then y

i < n. If i < n 1, then

n1
S
i=0

di Xn2 . So assume dn y = di z with

z = si dn y = dn1 si y

so dn y = di z = di dn1 si y = dn di si y and y = di si y. If i = n 1 and n > 1, then

y = sn1 dn y = sn1 dn1 z = z
so dn y = dn1 y; hence
y = sn2 dn1 y = sn2 dn y = dn1 sn2 y.
If n = 1 and i = n 1, we have d1 y = d0 z, hence y = z; since d0 : X 1 X 0 is
the inclusion of the maximal augmentation, y = d0 w for some w This is where
the hypothesis is used.

394

VII. Cosimplicial spaces

We now must show that dn , as in (4.18) is onto. If n = 0, Z0 = X0 and the

n1
S i
d Xn1 ,
result is clear. If n 1, we need to show that if x = dn y and x 6
then y 6

n1
S

i=0

di X n2 . The contrapositive of this statement is: if x = dn y and

i=0

y = di w, i n 1, then x = dj z for j n 1. But

x = dn y = dn di w = di dn1 w.

We can now prove Proposition 4.16:

Proof of 4.16: For the purposes of this argument, we say a morphism f : X
Y in cS has Property C if X n Y n is a cofibration in S for all n 0 and aX
=
aY . We leave it as an exercise to show that the class of morphisms satisfying
Property C is closed under isomorphisms, coproducts, retracts, cobase change,
and colimits over ordinal numbers. Only the statement about cobase
change

m
m
is non-trivial. Furthermore, the generating cofibrations n : n m
n
have Property C. Hence Proposition 4.15 implies all cofibrations have Property
C.
For the converse, suppose f : X Y has Property C. Referring to Lemma
4.17, write Z n (Xm ) for Z n obtained from the cosimplicial set of m simplices Xk .
Then, Lemma 4.17 implies
(L0 X)m
= Z 0 (Xm ) t Z 1 (Xm ) t Z 0 (Xm ) t Z 1 (Xm )
and if n > 1,
(Ln1 X)m
=

Z k (Xm )

:kn
n
Ymn is
with running over injections with 1 k < n. Since f : Xm
one-to-one, Z k () is natural in f . Since Z 1 (Xm )
= Z 1 (Ym ),

[X n Ln1 X Ln1 Y ]m
= Z n (Xm ) t

Z k (Ym )

:kn

for all n 0. Again runs over injections k, 1 k < n. The result follows.
Proposition 4.16 makes it very easy to decide when an object of cS is cofibrant;
for example, a constant object on a non-empty simplicial set is not cofibrant,
but the standard simplices n form a cosimplicial space {n }n = which is
cofibrant.

5. The total tower of a cosimplicial space

395

5. The total tower of a cosimplicial space.

The results on geometric realization from Section 3, interpreted in the context
of cosimplicial spaces, leads to the study of the total space of a cosimplicial space.
The purpose of this section is to work through these results.
If X cS
= (s(Sop ))op , we can define the geometric realization |X|
op op
(S ) = S; we will write Tot(X) for this object. Using the coequalizer (3.1),
suitably interpreted, we find that Tot(X) fits into an equalizer diagram
(5.1)

Tot(X)

Y
n0

Xn

Xm
.

:nm

If = {n } cS is the cosimplicial space composed of the standard simplices,

(5.1) immediately implies there is a natural isomorphism in S
(5.2)

Tot(X)
= HomsC (, X).

Since cS is cofibrant, SM7 for cS implies that Tot() preserves weak equivalences between fibrant objects; alternatively, this fact is a consequence of Proposition 3.6.
The skeletal filtration on the geometric realization (as in (3.8) and (3.9) writes
Tot(X) as the inverse limit of a tower. We define Totn (X) = skn |X| (Sop )op ;
arguing as in (5.1) and (5.2) and using Proposition 3.11, we have
(5.3)

Totn (X)
= HomcS (skn , X)

and the natural projection Totn (X) Totn1 (X) is induced by the inclusion
skn1 skn . Note
(5.4)

Tot(X)
= lim Totn (X).

By Proposition 4.16, skn1 skn is a cofibration in cS, hence, if X cS is

fibrant, SM7 for cS implies
(5.5)

{Totn (X)}n0

is a tower of fibrations. This fact equally follows from Proposition 3.6 This is
the total tower of X.
The diagrams (3.8) and (3.10) tell us how Totn (X) is built from Totn1 (X).
Recalling that for X cS, M n1 X
= Ln X, X s(Sop ), (3.8) implies that for

396

VII. Cosimplicial spaces

all X cS there is a pullback diagram

n

w Xn

Totn (X)
(5.6)
u
Totn1 (X)

n
Xn

n X
Mn1

(Mn1 X) .

Note Tot0 (X)

= X 0 . If X is fibrant, the vertical maps in (5.6) are fibrations.
Finally, if the cosimplicial object X cS is codegeneracy-free in the sense that
the underlying codegeneracy diagram is the right Kan extension of a discrete
diagram {Z n }, then (3.10) implies that there is a pullback diagram in S.

Totn X
(5.7)
u
Totn1 X

w (Z n )
u
w (Z )

n n

The rest of the section is devoted to showing that our two examples 4.1 and 4.2
are codegeneracy free in this sense.
Example 5.8. The category of cosimplicial abelian groups is equivalent to the
category (s(Abop ))op . Since Ab is abelian, so is Abop ; hence the normalization
functor N : s(Abop ) C Abop defines an equivalence of categories. See Section
III.2. In particular every object A s(Abop ) is naturally degeneracy free on
{N As }. An examination of the proof shows that d0 plays no role in the decomposition of A s(Abop ); specifically if nAbop denotes the category with objects
{Xn }n0 , Xn Abop , then normalization and its adjoint define an equivalence
of categories
N : s0 (Ab)op nAbop .
Here s0 C denotes the category of functors op
0 C where 0 is the subcategory with some objects and morphisms restricted (0) = 0. These are simplicial objects without d0 .
If c+ Ab denotes the cosimplicial abelian groups with without d0 , these remarks
supply an equivalence categories
N : c+ Ab nAb

6. The homotopy spectral sequence of a tower of fibrations397

sending A to N A with

(N A) =

n1
\

ker{si : An An1 }.

s=0

In particular any object A c+ Ab is naturally codegeneracy free on N A.

Now let X S and R X the Bousfield-Kan R-resolution of Example 4.1. Then
R X cS is not a cosimplicial simplicial abelian group; however, when restricted
an object R X c+ S is has a natural structure as an object in c+ (sAb). Since
products in sAb are products in S, we have that R X is naturally codegeneracy
free on N R X. Then (5.7) gives the existence of a pullback diagram
Totn R X

w Hom(n , N Rn X)

Totn1 R X

w Hom( , N Rn X).

Example 5.9. Consider the cosimplicial object T X cSJ constructed in Example 4.2. Examining the definitions we have
Y
(T X)n = T n+1 (X)j =
Xin .
ji0 in

It follows that T X is naturally codegeneracy free as an object in cSJ on Z n (X)

where
Y
Z n (X) =
Xin
ji0 i1 in

where it it+1 is not the identity for any t, 0 t n 1. It follows that

lim T X at cS is also codegeneracy free, since lim preserves products.
J

6. The homotopy spectral sequence of a tower of fibrations.

Given a tower of pointed fibrations, the exact sequences in homotopy give
rise to a spectral sequence. There are a number of problems with this spectral
sequence. First, there is the problem that 0 and 1 are not generally abelian
groups; second, the spectral sequence need not converge to anything homotopical;
third, the spectral sequence is fringed in the sense that E can contain extra
elements; and, fourth, the E2 term may not have a sensible description in terms
of homological algebra. We analyze these difficulties in turn. The first two

398

VII. Cosimplicial spaces

difficulties are addressed in this section, and the last in the next. The fringing
effect will be dealt with in Section 8.
So let

w Xu n

pn

in

w Xn1
u

in1

Fn

p1

w Xu 1

w Xu 0
=

i1

Fn1

F1

F0

be a tower of pointed fibrations, with Fn the fiber of pn . Applying homotopy to

each of the fiber sequences, yields

w X
u n

w Xn1
u
i

k
ii
Fn

ii
k

w
i

w X
u 1

Fn1

w X0
i
ki
i
F1
F0

where each of the dotted maps is of degree 1. This gives an exact couple and
hence a spectral sequence. We will be explicit because 1 is not abelian and
because
1 Xn1 0 Fn 0 Xn 0 Xn1
is exact only in the sense of pointed sets, with the additional proviso that the
action of 1 Xn1 on 0 Fn extends to an injection of sets
0 Fn /1 Xn1 0 Xn .
Note that 0 Xn 0 Xn1 need not be onto, although the pre-image of the
basepoint is 0 Fn /1 Xn1 . (Here and throughout, if a group G acts on a set Z,
Z/G denotes the orbits.)
The spectral sequence is now defined as follows. Let
(6.1)

i Xn(r) = Im(i Xn+r i Xn ) i Xn ,

and for t s 0, let

(6.2)

Zrs,t = Ker(ts Fs ts Xs /ts Xs(r1) ).

6. The homotopy spectral sequence of a tower of fibrations399

Then let
(6.3)

Brs,t = Ker(ts+1 Xs1 ts+1 Xsr1 ).

Then we set
(6.4)

Ers,t = Zrs,t /Brs,t

where this formula must be interpreted: if t s 1 and j : i+1 Xn1 i Fn

is the connecting homomorphism, j Brs,,t Zrs,t is a normal subgroup and Ers,t
is the quotient group, and if t s = 0, Brs,t acts on Zrs,t and Ers,t is the set of
orbits. Define differentials for t s 1
(6.5)

dr : Ers,t Ers+r,t+r1

by the composite map

(6.6)

Ers,t ts Xs(r1) Ers+r,t+r1

where the first map is induced by noticing that if x Zrs,t ts Fs , then

(r1)
(r1)
we
and the second by noticing that if y is in ts Xs
(is ) x ts Xs
may choose z s1 Xs+r1 mapping to y and
j z Ers+r,t+r1
is independent of that choice. Notice that no differential is defined on Ers,s and,
indeed, that Ers,t is not defined for t s < 0.
The following result is left as an exercise:
Lemma 6.7.
1) Ers,t is a pointed set if t s 0, a group if t s 1, and an abelian group if
t s 2;
2) The function dr : Ers,t Ess+r,t+r1 is a homomorphism if t s 2 and a
map of pointed sets if t s 1.
3) The image of dr : Ers,t Ers+r,t+r1 is in the center of Ers+r,t+r1 if t s 2
s,t
and if t s 1, Er+1
= Ker(dr )/Im(dr ).
4) The map dr : Ers,s+1 Ers+r,s+r extends to an action of the source on the
s,s
target and Er+1
Ers,s /Ersr,sr+1 .
Proof: All these facts follow from the properties of the long exact sequence in
homotopy of a fibration.

400

VII. Cosimplicial spaces

s,s
Remark 6.8. The fact that Er+1
Ers,s /Ersr,sr+1 and is not necessarily
equal to what is meant by saying the spectral sequence is fringed. The failure of
equality arises from the fact that 0 Xn 0 Xn1 need not be onto.

The collection {Ers,t ; dr }r1 is the homotopy spectral sequence of the tower of
fibrations. It is natural in the tower and we may write {Ers,t X}. We now begin
to address what it might converge to.
s,t
Notice that if r > s, Er+1
Ers,t . Define
(6.9)

s,t
E
= lim Ers,t =

r>s

Ers,t .

r>s

s,t
One hopes that E
may have something to do with the homotopy groups of
lim Xn . For this, we must first decide how to compute (lim Xn ) in terms of

the tower of groups { Xn }. This is done by a lim1 exact sequence, once we

have defined lim1 for non-abelian groups.

First recall that lim1 for a tower of abelian groups A = {An }n0 is defined

by the exact sequence

0 lim An

Y
n

An

Y
n

An lim1 An 0

where (an ) = (an f (an+1 )), with f : An+1 An the maps in the tower. The
following is an easy exercise in homological algebra.
Lemma 6.10. The functor lim1 : tow(Ab) Ab is characterized up to natural

isomorphism by:
1) If 0 A B C 0 is a short exact of towers of abelian groups, then
there is a six-term exact sequence
0 lim A lim B lim C lim1 A lim1 B lim1 C 0.

2) If A is any tower so that An+1 An is surjective for all n, then lim1 A = 0.

We give a similar statement for non-abelian groups. QLet G = {Gn } be a

towerQof not necessarily abelian groups. Then the group n Gn acts on the set
Z = n Gn by (gn ) (xn ) = (gn xn f (gn+1 )1 ). Then the stabilizer subgroup of
e Z is lim Gn and we define lim1 Gn to be the pointed set of orbits. If G is a

tower of abelian groups these definitions agree with those above.

6. The homotopy spectral sequence of a tower of fibrations401

Lemma 6.11. The functor lim1 : tow(Gr) pointed sets is characterized up

to natural isomorphism by
1) If G1 G2 G2 is short exact sequence of towers of groups then
there is a six term sequence
lim G1 lim G2 lim G3 lim1 G1 lim1 G2 lim1 G3

which is exact as groups at the first three terms, as pointed sets at the last three
terms, and lim G3 acts on lim1 G1 and there is an induced injection

lim1 G1 / lim G3 lim1 G2 .

2) If G is any tower so that Gn+1 Gn is surjective for all n, then lim1 Gn = .

Proof: The proof is the same as in the abelian case.

Remark 6.12. It is good to have examples where lim1 is non-zero. Let B be

any group and let B also be the constant tower on that group. let
T An B be a
descending (An+1 An ) sequence of normal subgroups so that n An = e. Let
A = {An } be the resulting tower. Then one gets a short exact sequence
(6.13)

B lim(B/An ) lim1 A

so that lim1 A = if and only if B is complete in the topology defined by the

An . For example, if B = Z and An = pn Z for some prime p, (6.13) becomes

0 Z Zp Zp /Z 0
whereLZp is the group ofLp-adic integers, which is uncountable. One could take
B = s1 Z and An = sn Z B. Then (6.13) becomes
0

M
s1

Y
s1

Y
s1

Z/

Z 0.

s1

Now let X = {Xn } be a tower of pointed fibrations.

Proposition 6.14. For every t 0 there is a sequence
lim1 t+1 Xn t (lim Xn ) lim t Xn

which is exact as groups if t 1 and as pointed sets if t = 0.

402

VII. Cosimplicial spaces

Proof: Consider the description of the homotopy inverse limit given in Section
IX.5. Then one has a homotopy fibration sequence
Y
n

Xn holim Xn

Xn .

A simple calculation with the homotopy exact sequence of this fibration gives the
result except possibly for the claim that 0 holim Xn lim 0 Xn is onto. But it

is an easy exercise to show that for a tower of fibrations 0 lim Xn lim 0 Xn

is onto.

We now return to computing lim Xn by the homotopy spectral sequence.

s,t
s,t
Because of Ers,t given in (6.1)(6.4) and E
in (6.9) we can think of E
as
an equivalence class of elements x ts Fs so that for each r 0 there is an
element yr ts Xs+r mapping to i x ts Xs . (Notice we do not say yr
maps to yr1 ; see Example 6.17.2 below.) If z ts lim Xn maps to zero in

ts Xs1 , then one gets such an equivalence class. To make this precise, let
Gs i X = Im(i X i Xs )
and
(6.15)

es,t
= ker{Gs ts Xs Gs ts Xs1 }.

If y es,t
, let x ts Fs be any element so that i x = y.
Lemma 6.16.
s,st
1) The assignment y 7 x induces a well-defined monomorphism es,t
E .
2) The inclusion maps Gs X Xs induce an isomorphism

lim Gs X lim Xs .

Proof: This is an exercise in unraveling the definitions.

Notice that any element of lim1 X maps to zero in all Gs X, so will not

s,t
s,t
be detected in any es,t
. It also turns out that e E need not be an
isomorphism.
Here are two examples of the potential difficulties.

6. The homotopy spectral sequence of a tower of fibrations403

Example 6.17.
1) Consider the tower of abelian groups
p

Z Z Z Z
where p Z is a prime number. The resulting tower of Eilenberg-MacLane
spaces
p

K(Z, n) K(Z, n) K(Z, n)

can be transformed into a tower of fibrations. One easily calculates that E10,n = Z
and E1s,s+n1 = Z/pZ and E1s,t = 0 otherwise. Next Er0,n = pr1 Z Z = E10,n
s,t
and dr : Er0,n Err,r+n1 = Z/pZ is onto.. Thus E
= 0 for all t s 0.
However
n1 lim Xs = Zp /Z
= lim 1 (pr Z) 6= 0.

Note that in this case lim1 Er0,n 6= 0.

2) Consider the tower of abelian groups, where a is the addition map,

M
s3

M
s2

Z
Z

s1

and let {Xn } be the resulting tower of fibrations obtained by applying K(, n)
and the converting tower of fibrations. Now
n1 lim Xm
= lim1 n Xm
=

Y
s1

Z/

s1

and k lim Xm = 0 otherwise. Then

s = 0, t = n
Z
L
s,t
Z s = 1, t = n + 1
E1 =
i1
Z
s 2, t = s + n 1
s,t
0,n
= 0 otherwise.
= Z and E
and dr : Er1,n+1 Err+1,n+r is onto. Thus E
1
0,n
0,n
1,n+1
Thus e 6= E . Note also lim Er
6= 0.

The following definition now seems appropriate.

404

VII. Cosimplicial spaces

Definition 6.18. The spectral sequence of the pointed tower of fibrations {Xn }
converges completely if
1) lim1 i Xn = for i 1, and

s,t
2) es,t
= E for all (s, t) with t s 1.
When 6.18.2 holds, the spectral sequence effectively computes lim Xn , sub
ject to the fringing effect when t s = 0, and when 6.18.1 holds, we have
(lim Xn )
= lim Xn . One would like to be able to decide when one has com

plete convergence from knowledge of the spectral sequence, rather than using
knowledge of lim Xn , which Definition 6.18 demands after all one is trying

to compute lim Xn . The next lemma is the best possible result. Compare the

examples of 6.17.
Lemma 6.19 (Complete Convergence Lemma). The spectral sequence of
the pointed tower of fibrations {Xn } converges completely if and only if
lim 1 Ers,t = ,

t s 1.

This will be proved below, after some examples and preliminaries.

Corollary 6.20. Suppose that for each integer n 0 there are only finitely
many s, so that E2s,s+n 6= . Then the spectral sequence converges completely.
More generally we have
Corollary 6.21. Suppose that for each integer n 0 and each integer s, there
is an integer N so that
s,s+n
s,s+n
E
.
= EN
Then the spectral sequence converges completely.
Proof: For lim1 Ers,t = . See 6.22.1 below.

The phenomenon encountered in Corollary 6.21 is called Mittag-Leffler convergence. The wider implications are explored in [14, p.264].
As a preliminary to proving the complete convergence lemma, we discuss de(r)
rived towers of group. Suppose {Gn } is a tower of groups. Define Gn =
(r)
Im{Gn+r Gn }. We already encountered Xs = ( Xs )(r) in (6.1)(6.4).
There are maps
(r+1)
(r)
G(r)
n Gn1 Gn1

6. The homotopy spectral sequence of a tower of fibrations405

and one has two new towers of groups
{G(r)
n }n0

{G(r)
n }r0 .

Lemma 6.22.
(r)
1) The inclusions Gn Gn induce isomorphisms

lim G(r)
= lim Gn
n

lim1 G(r)
= lim1 Gn .
n

and

2) There is a natural isomorphism

lim lim G(r)

= lim Gn
n

and a natural short exact sequence

lim1 lim G(r)
lim1 Gn lim lim1 G(r)
.
n

n
n

(r)

Proof: For 1) note that the functor {Gn }n0 7 {Gn }n0 is exact and takes
(n)
tower of surjections to surjections. Since lim Gn
= lim Gn by inspection, the
n
n
result follows from the characterization of lim1 given in Lemma 6.11: the functor

{Gn } lim1 G(r)

n
satisfies the requirements there.
For 2), note that if the groups were abelian, what we would have is a degenerate
form of a composite functor spectral sequence. In the non-abelian case, it is
better to proceed topologically. For fixed r define a tower Yr = {Yr,n } with
(r)
Yr,n = BGn . There are maps of towers Yr+1 Yr ; convert the resulting tower
of towers into a tower of fibrations in tow(S ):
Z2 Z1 Z0 .
Thus Y0 Z0 is a weak equivalence in tow(S ) with Z0 fibrant (i.e., a tower
of fibrations) and Yr Zr Zr1 factors Yr Yr1 Zr1 as a weak
equivalence followed by a fibrations. Then
lim Z2 lim Z1 lim Z0

406

VII. Cosimplicial spaces

is a tower of fibrations and for all r, 1 lim Zr

= lim Gn and 0 lim Zr
=
n
n
n
1
lim Gr , by part 1). Hence
n
1 (lim lim Zr,n )
= lim Gn

0 (lim lim Zr,n )

= lim1 Gn .

Now one takes limits in the other direction: for fixed n, {Zr,n }r0 is a tower
fibration, so
1 lim Zr,n
= lim G(r)

n
r

0 lim Zr,n

= lim1 G(r)
n
r

Now we calculate lim lim Zr,n and use that limits commute to finish the
n r
result.

Proof of the Complete Convergence Lemma 6.19: We prove only that

if lim1 Ers,t = , then we have complete convergence. The reverse implication is

r
left as an exerciseit is a matter of reversing the arguments.
For each pair of positive integers r n, let us define Knr by the exact sequence
of groups
(rn)

Knr ts Xs+n

(6.23)

ts Xs(r) .

If n = 1 and r > s, Knr = Ers+1,t+1 ; therefore, if r > s + n, there is a composition

series for Knr with all of the horizontal maps surjections
Knr
u

r+1
w Kn1
u

Ers+n,t+n

s+n1,t+n1
Er+1

w K1r+n1
s+1,t+1
Er+n1

An induction argument shows lim1 Knr = and that each of the maps lim Knr
r
r
r+1
lim Kn1
is onto. Taking the limit over r in (6.23) yields a short exact sequence
r
(rn)

lim Knr lim ts Xs+n

lim ts Xs(r) .

6. The homotopy spectral sequence of a tower of fibrations407

Now we take the limit over n. By the previous lemma
(rn)

lim lim ts Xs+n

n

= lim ts Xn ,

and since lim1 vanishes on towers of surjections, we get a short exact sequence

(6.24)

lim lim Knr lim ts Xn lim ts Xs(r)

and an injection
(rn)

lim1 lim ts Xs+n

(6.25)

lim1 ts Xs(r) .

(r)

In particular, (6.24) yields that the natural map lim ts Xn lim ts Xs is

n
r
(r)
s,t

a surjection. Thus Gs ts lim Xn = lim ts Xs and, by induction, es,t

= E .

r
(r)
Also, since n does not appear in lim1 ts Xs , this term is trivial, so we
n
(rn)
(rn)
have that the term lim1 lim ts Xs+n = . If lim lim1 ts Xs+n = , the
n r
n r
previous lemma implies lim1 ts Xn = and well be done. So well close by

(rn)
showing lim1 ts Xs+n = .
r
(r)
As with Knr , there is a composition series for ts Xs : if r > s
ts Xs(r)
u
s,t
Er+1

(r+1)

w ts Xs1
u

s1,t1
Er+2

(r+s)

w ts X0
u

0,ts
Er+s+1

(r)

A similar inductive argument as with Knr , shows lim1 ts Xs = .

r
It is worth pointing out that even without convergence one has a comparison
result.
Proposition 6.26. Suppose f : X = {Xn } Y = {Yn } is a map of towers of
pointed fibrations and suppose there is a (finite) N 1 so that
s,t
s,t
f : EN
X EN
Y

is an isomorphism for all (s, t). If Ers,s X = for all s, then lim Xn lim Yn is

a weak equivalence of connected spaces.

408

VII. Cosimplicial spaces

Proof: Note that f : Ers,t X Ers,t Y is an isomorphism for all r N . Then

induction on n and the exact sequence
(r1)

Ers+t,t+1 ts Xs+1
(n)

ts Xs(r)

(r)

(see (6.23)) implies k Xn k Yn is an isomorphism for all r N 1. Finally,

(r)
(r)
(r)
(r)
the hypotheses on Ers,s imply 0 Xn = = 0 Yn and 1 Xn 1 Xn1 is
onto. Thus the result follows from Lemmas 6.22.1, 6.11, and Proposition 6.14.
7. The homotopy spectral sequence of a cosimplicial space.
If X is a pointed cosimplicial space, then the tower {Totn X}n0 has a homotopy spectral sequence. In this section, we work out the E1 and E2 terms of the
spectral sequence, and give an extended examplethe original Bousfield-Kan
spectral sequence, which is a kind of unstable Adams spectral sequence.
Fix a fibrant pointed cosimplicial space X, and form the tower {Totn X}n0 ,
where Totn X = HomcS (skn , X). The constant map is the basepoint of
Totn X and by the diagram (5.6), the fiber of Totn X Totn1 X is n N n X,
where N n X is the fiber of s : X n M n1 X. Thus the E1 term of the associate
spectral sequence is, for t s 0,
E1s,t = ts s N s X
= t N s X.
If G is a cosimplicial (not necessarily abelian) group, define
s

N G=

j1
\

ker{si : Gs Gs1 }.

i=0

If G is a cosimplicial abelian group, then N G = {N s G, (1)i di } becomes a

cochain complex. Define N 0 X = 0 X 0 .
Lemma 7.1. For a fibrant pointed cosimplicial space X, there is a natural isomorphism
E1s,t
= t N s X
= N s t X, t s 0
Proof: First assume t 2. The cases t = 0, 1 will be handled at the end. We
use the pullback square
M n1,k+1 X
(7.2)
M

w X n1

n1,k

n2,k

wM

7. The homotopy spectral sequence of a cosimplicial space 409

of Proposition 1.25, and a double induction on the following two statements.
Notice that M n,k is a functor defined as cC where C is any category with finite
limits.
H(n, k): The natural map t M n1,k X Mn1,k t X is an isomorphism.
I(n, k): Let N n,k X be the fiber of s : X n Mn1,k X. Then
0 t N n,k X t X n t Mn1,k X 0
is short exact.
The result follows from setting n = k = s in I(n, k) and H(n, k).
First notice that I(0, 0) is trivial since M 1,0 X = . Also H(n, 0) and I(n, 0)
are trivial since M n1,0 X = and M n1,0 t X = 0.
Next, H(n, k), H(n 1, k), and I(n 1, k) yield H(n, k + 1), since t () applied
to (7.2) yields a pullback diagram

t M n1,k+1 X

w t Xn1

n1,k

t X

w Mn2,k t X.

Finally, H(n, k) I(n, k) because t X n M n1,k t X is surjective. Indeed,

as in Example 5.8,
M
M n1,k t X
N m t X
=
:nm

where runs over the objects of Mn,k .

For the cases t = 0 and 1 of Lemma 7.1, note that if t = 0, s = 0, so one
need only check 0 N 0 X = 0 X 0 = N 0 X which is true since N 0 X = X 0 . If
t = 1, s = 0 or 1. If s = 0, one has 1 N 0 X = 1 X 0 as required. If s = 1,
s0

1 N 1 X 1 X 1 1 X 0 is exact, since s0 d0 = 1 and X 0 = M 0 X.

We use this calculation as the input for the calculation of the E2 term. In
fact, we wish to claim that E2s,t = s t X, but we need to take some care with
the definitions, especially for t = 1.
If A is a cosimplicial abelian group, we can make A into a cochain complex

A2
A1
A0

410

VII. Cosimplicial spaces

with = (1)i di . Then s A = H s (A). This is equivalent to computing

H s N A, where N A is the normalized cochain complex with
N A = {N s A, (1)i di }.
One has
s A = H s A
= H s N A.

(7.3)

If G is a cosimplicial (not necessarily abelian) group, one still has 0 G and 1 G.

The group 0 G is the equalizer of d0 , d1 :
d0

0 G G0 G1 .

(7.4)

d1

The pointed set 1 G is defined by the formula

1 G = Z 1 G/G0

(7.5)

where Z 1 G is the pointed set of cocycles

Z 1 G = {x G1 |(d2 x)(d1 x)1 (d0 x) = e}
and G0 acts on Z 1 G by
g x = d0 g x(d1 g)1 .
Notice that as G is a cosimplicial abelian group, the two definitions of 1 G agree.
One can also normalize in the non-abelian case; indeed, the situation is already
normalized in the sense that if x Z 1 G, then x N 1 G because
e = s0 [(d2 x)(d1 x)1 (d0 x)] = d1 s0 x
so
e = s0 d1 s0 x = s0 x.
Thus N 1 G Z 1 G = Z 1 G and
(7.6)

1 G = (N 1 G Z 1 G)/G0 .

If X is a cosimplicial set we define 0 X to be the equalizer of d0 , d1 : X 0 X 1 .

7. The homotopy spectral sequence of a cosimplicial space 411

Proposition 7.7. Let X be a pointed fibrant cosimplicial space. Then the
E2 term of the spectral sequence of the tower of fibrations {Totn X} can be
calculated by natural isomorphism
E2s,t
= s t X,

t s 0.

Because of this result, one often writes

s t X ts Tot X
for the homotopy spectral sequence arising from a pointed cosimplicial space.
The double arrow is not meant to prejudice anyone for or against convergence
of the spectral sequence; for this, one must appeal to the results of Section 6.
Before proving Proposition 7.7, we line up notation and ideas. We are interested in the tower {Totn X}, where the tower maps
Tots X = HomcS (sks , X) HomcS (sks1 , X) = Tots1 X
are induced by the inclusion sks1 sks . Thus the fiber at the constant
map is the space of pointed maps
HomcS (sks / sks1 , X).
If f : sks / sks1 X is a pointed map one gets a diagram
s /s
u

w Xs
u
w Ms X

and hence a map of spaces

HomcS (sks / sks1 , X) Hom (s /s , Ns X).
Diagram (5.6) that this map is an isomorphism. To calculate E2s,t = Z2s,t /B2s,t
(see (6.1)(6.4)) we calculate Z2s,t , B2s,t , and the action of B2s,t . These are given
in Lemmas 7.8, 7.14, and 7.15 below. Proposition 7.7 will follow.
Recall that for t 2, we may define
=

s+1
X
i=0

di : t N s X t N s+1 X.

412

VII. Cosimplicial spaces

This defines the sub-group ker() t N s X, t 2. If t = 1, define

ker() 1 N 1 X
by

ker() = Z 1 1 N X

(see (7.4)(7.5)). This is a pointed set. If t = 0 or 1 define Ker() t N 0 X

t X 0 to be the equalizer of
d0 , d1 : t X 0 t X 1 .
If t = 1, ker() is a group; if t = 0, ker() is a pointed set.
Lemma 7.8. For all t s 0, there is a natural isomorphism
Z2s,t
= ker() t N s X.
Proof: First take s 1. According to (6.2),
f ts HomcS (sks / sks1 , X)
is in Z2s,t if and only if there is a lift

Tots+1 X

f
ts
S
w HomcS (sks / sks1 , X)

u
w Tots X.

Here we have confused f with a representative. We can do this as all objects are
fibrant. Thus, by adjointness, f Z2s,t if and only if there is an extension (with
+ denoting a disjoint basepoint)
S ts (sks )+

ts

(sks+1 )+

w S ts (sks / sks1 )
wX

7. The homotopy spectral sequence of a cosimplicial space 413

in the category of pointed cosimplicial spaces. Here g is adjoint to f . By Diagram
(5.6) and adjointnetss, f Z2s,t if and only if there is a lift
S ts (s+1 )+
(7.9)
S

ts

s+1
w S ts (s+1 / sks1 s+1 )

w X

(s+1 )+

w Ms X

in the category of pointed spaces. Let : s /s s /s be the identity.

Then the collapse map s+1 s+1 / sks1 s1 represents, when we choose
0 s+1 as the basepoint,
s+1
X

(1)i di s (s+1 / sks1 s+1 )

if s + 1 > 2

i=0

and

[d2 ][d1 ]1 [d0 ] 1 (2 / sk0 2 )

if s + 1 = 2.

Thus if the original map f was adjoint to t N s X, the morphism across the
top of (7.9) represents (if s + 1 > 2) and [d2 ][d1 ]1 [d0 ] (if s + 1 = 2).
Thus f Z2s,t if and only if the morphism across the top is null-homotopic; that
is, if and only if ker().
If s = 0, one is asking whether there is a lift
Tot1 X
6
5
5
5
u
55
S t f w Tot0 X = X 0
or, again by (5.6) and adjointness, whether the following diagram can be completed
S t (1 )+

w X1
P
O
O
u
u
O
t
1
S ( )+
w X0 .

414

VII. Cosimplicial spaces

This happens if and only if d0 f = d1 f t X 1 .

s,t
s,t
In order to calculate the boundaries B2 and the quotient E2 , we must be
explicit about the action of the fundamental group on the components of the
fiber. This requires a short digression.
Let C be any simplicial model category. Suppose A is pointed cofibrant and
A B is a pointed cofibration. Then if B/A is define the pushout diagram
A

wB

u
w B/A,

the object B/A is pointed and there is a pointed fibration sequence. If A C is

pointed cofibrant and K is a pointed simplicial set
A K = A K/A K.
If X C is pointed and fibrant, there is a pointed fibration sequence
Hom (B/A, X) Hom (B, X) Hom (A, X)
where we take the constant to be the basepoint. We describe the action of
t Hom (A, X)
= [A S t , X] on ts Hom (B/A, X)
= [B/A S t1 , X] .
One model for the pointed cone on A is C 1 A = A 1+ /A {1}+ . For another
model let [0, 2] S be the simplicial set with two non-degenerate one simplices
[0, 1] and [1, 2] and d0 [0, 1] = {1} = d1 [1, 2]. (We can think of 1 as [0, 1].)
Then C 2 A = A [0, 2]+ /A {2}+ is another model for the pointed cone on A.
Think of A C 1 A or A C 2 A as the subspace A {0}+ . Then there are weak
equivalences
B A C 2 A ' B/A ' B A C 1 A
and an isomorphism C 1 A/A
= A S 1 . Thus one gets a map
(7.10)

: B A C 2 A B t C 2 A/A {1}+ = A S 1 B A C 1 A

If t Hom (A, X) and t1 Hom (B/A, X) are represented by f : A

S t X and g : (B A C 1 A)S t1 X respectively t1 Hom (B/A, X)
is represented by
1

f g

(7.11) (B A C 2 A) S t1 (A S 1 ) S t1 (B A C 1 A) S t1 X.

7. The homotopy spectral sequence of a cosimplicial space 415

If t 2, t1 Hom (B/A, X) is a group, the action is given by = (j )
where j is the connecting homomorphism of the long exact sequence defined by
j = e.
If A B is not a pointed cofibration, we still have a fibration sequence
Hom (B/A, X) Hom(B, X) Hom(A, X).
We can calculate the action t Hom(A, X) on t1 Hom (B/A, X) via the device of the disjoint basepoint. Define A+ = A t . Then Hom(A, X)
=
t
Hom (A+ , X), so if t Hom(A, X) is represented by f : A+ S X
and t1 Hom(B/A, X) is represented by g : (B A C 1 A) S t1 X, then
is represented by
f g

(7.12) (B A C 2 A) S t1 (A+ S 1 ) S t1 (B A C 1 A) S t1 X.
The disjoint basepoint is crucial in the following.
Example 7.13. Let X S be pointed and fibrant and consider the cofibration
1 1 . Then, the associated fiber sequence is
X Hom(1 , X) X X
and the action of t (X X)
= t X t X on t1 X
= t X is given by
(1 , 2 ) = 1 21 .

We are now in a position to calculate with B2s,t . Recall from (6.1)(6.4), that
B2s,t = Ker{ts+1 Tots1 X ts+1 Tots2 X}, and, in order to calculate
E2s,t we need to know the action of B2s,t on
Z2s,t ts s1 N s1 X.
We separate out the cases s = 1 and s > 1.
Lemma 7.14. If s = 1, B2s,t = t Tot0 X
= t X and if t X and
1 1
1

t1 N X = t N X, then
= (d0 )(d1 )1 .
Proof: One is concerned with the fibration sequence in cS
Hom (sk1 / sk0 , X) Hom(sk1 , X) Hom(sk0 , X).

416

VII. Cosimplicial spaces

Thus if t Hom(sk0 , X)
= t X and t1 Hom (sk1 / sk0 , X)
=
1 1
t1 N X, then is represented by (see (7.12))

(sk1 sk0 C 2 sk0 ) S t1

u
(sk0 )+ S 1 S t1 (sk1 sk0 C 1 sk0 ) S t1

w X.

Specializing to cosimplicial degree 1, we have

1

1 C S

t1

( )+ S S

t1

( 1 C )S

t1

1 1

X 1 .

However 1 = d0 d1 : (1 )+ S t X 1 and 1 is a representative for

t1 N 1 X. Hence Example 7.13 implies the result.

Lemma 7.15. If s > 1, then

i ts+1 s1 N s1 X = B2s,t ts+1 Tots1 X
P
and j B2s,t
= Im( (1)i di ) ts s N s X
= t N s X.
Proof: The equality s1 N s1 X = B2s, follows from the long exact of
the fibration
s1 N s1 X Tots1 X Tots2 X.
For the calculation of j : Tots1 X 1 s N s X we are concerned with
the fibration sequence
Hom (sks / sks1 , X) Hom(sks , X) Hom(sks1 , X).
If ts+1 Hom(sks1 , X), then j is the composite (where f represents
)
f

X.
(sks sks1 C 2 sks1 ) S ts (sks1 + ) S ts+1
At cosimplicial degree s, this yields
(7.16)

fs

s s C 2 s S ts (s+ ) S ts+1 X s .

7. The homotopy spectral sequence of a cosimplicial space 417

If B2s,t then f s factors as
w Xs

(s+ ) S ts+1
(7.17)
(s+ / sks2

s
ts+1
+ ) S

Now s s C 2 s ' s /s ' S s , s ' S s1 and if i : s1 s1 is

the identity then the collapse map
s s / sks2 s
represents (1)i di i. Combining these facts with (7.16) and (7.17) yields the
result.

Example 7.18. Let X be a pointed space and let X Fp X be the BousfieldKan resolution of X. Then the total space Tot(Fp X) = Xp is called the
Bousfield-Kan p-completion of X, and Xp may be calculated by the spectral sequence
s t F X ts Xp .
The point of this example is to rewrite the E2 term in terms of homological
algebra. We assume the reader is familiar with the category CA of unstable
coalgebras over the Steenrod algebra. See [71, Sec. 1]. If X is a space, H X
CA. Let CA be the category of augmented (or pointed) coalgebras in CA; thus
an object in CA is an unstable coalgebra C over the Steenrod algebra equipped
with a morphism of such coalgebras Fp C. The homology of a pointed space
is in CA .
First notice that if V is any simplicial vector space, then, choosing 0 V as
the basepoint,
(7.19)

t V
= homCA (H S t , H V ).

One way
L to see this is to note that there is a (non-canonical) weak equivalence
V ' K(Z/p, n ) for some set of non-negative integers {n }, and to use the
calculations of Serre and Cartan on H K(Z/p, n ). Now the fact that V is a
simplicial abelian group implies (7.19) holds for any choice of basepoint. Since
a choice of basepoint for X induces a basepoint for Fp X,
(7.20)

s t Fp X
= s homCA (H S t , H Fp X).

418

VII. Cosimplicial spaces

Now, the forgetful functor CA nFp to the category of graded vector spaces
has a right adjoint G. The functor G is characterized by the fact that it commutes
with products and filtered colimits, and by the fact that if W nFp is dimension
1 concentrated in degree n, then
(7.21)

G(W )
= H K(Z/p, n).

(Note that G is the functor of a cotriple on nFp and CA is the category of

coalgebras over G.)
Let G : CA CA be the composite of G and the forgetful functor. Then G
is a triple on CA, so if C CA, we have an augmented cosimplicial object (see
Example 4.1).
(7.22)

C G C.

This is a resolution in the sense that as an augmented cosimplicial graded vector

space it has a retraction, so G C
= C via the augmentation. Thus we define,
for C CA ,
(7.23)

ExtsCA (H S t , C) = s homCA (H S t , G C).

Now if X is a space, H FX
= G(H X). This follows from (7.19), (7.21) and the
properties of G. Thus combining (7.20) with (7.23) we have
s t Fp X
= ExtCA (H S t , H X)
and the homotopy spectral sequence of Fp X is the Bousfield-Kan spectral sequence
ExtsCA (H S t , H X) ts Xp .
It is an unstable Adams spectral sequence.
The reader might object that, for pointed coalgebras, one should define the
Ext groups differently. If C CA , define the coaugmentation ideal by
JC = coker{Fp C}.
Then J : CA nFp has a right adjoint G with G V = GV augmented by
G(0)
= Fp G(V ). Let G = G J : CA CA and one might demand that
the Ext object be
s homCA (H S t , G C).
However, a bicomplex argument with
homCA (H S t , G G C)
shows that this definition agrees with the previous one (7.23); furthermore, (7.23)
offers additional flexibility with basepoints.

7. The homotopy spectral sequence of a cosimplicial space 419

We finish this section by constructing a spectral sequence for computing the
homotopy groups of a pointed homotopy inverse limit.
We begin with a discussion of the derived functors of limit on I diagrams of
abelian groups. Let AbI be the category of these diagrams. Then the functor
Gi : AbI Ab with Gi A = Ai has both a left and a right adjoint; in particular,
the right adjoint is given by
(7.24)

(Ti B)j =

B.

ji

If B Ab is injective, then Ti B is injective in AbI and this implies AbI has

enough injectives;
indeed, if we embed each i in an injective abelian group Bi ,
Q
then A i Ti Bi is an embedding of A into an injective in AbI . We define the
functors lims = lims to be the right derived functors of lim . Since limit is left

I
I
exact, lim = lim0 . The standard observation is that one need not go all the way

to injectives to compute the derived functors. As with sheaf cohomology one can
resolve by flabby objects rather than injective ones. The crucial result is:
Lemma 7.25. Let B be an abelian group. Then for all i we have lim Ti B
=B

s
and lim Ti B = 0 if s > 0.

Proof: To prove lim Ti B

= B, let A AbI be a constant diagram on an

abelian group A. Then

homAb (A, lim Ti B)
= homAbI (A, Ti B)
= homAb (A, B).

To obtain the vanishing of the higher derived functors, let B J be an

injective resolution of B. Then (see (7.24)), Ti B Ti J is an injective resolution
of Ti B. Applying lim to this resolution, one recovers B J , which is acyclic

by construction.

We use this to build a cosimplicial resolution of A AbI suitable for computing lims . For our category I let dI be the associated discrete category; i.e.,

dI has the same objects as I but no non-identity morphisms. There is a functor

dI I, inducing a forgetful functor AbI AbdI . The category AbdI has
objects indexed families of abelian groups A = {Ai }iI . This forgetful functor
has a right adjoint given by
Y
TA =
Ti Ai .
i

420

VII. Cosimplicial spaces

We also write T : AbI AbI be the composite functor. This is the same
functor T as in Example 4.2. Then T is the functor of a triple on AbI and one,
therefore, obtains a natural augmented cosimplicial object in AbI
A T A.

(7.26)

If one forgets to AbdI this cosimplicial object has a retraction and so is acyclic.
This implies it is acyclic in AbI and Lemma 7.25 implies:
Lemma 7.27. For all A AbI there is a natural isomorphism
lims A
= s lim T A.

By analogy we use this construction to define lim1 G where G is an I diagram

of non-abelian groups. The reader will have noticed that there is nothing special
about abelian group in the construction of the cosimplicial object A T A of
(7.26). Indeed, if C is any category and C C I is an I diagram, we obtain
a cosimplicial object C T C. In particular if G is an I-diagram of (not
necessarily abelian) groups, one has a cosimplicial I-diagram of groups
G T G
and we define
lim1 G = 1 lim T G.

(7.28)

Thus lim1 G is a pointed set and is a group of cycles modulo a group of bound
aries. See (7.5). At this point we have potentially two different definitions of
lim1 G for a tower of groups, one from (7.28) and the other from Lemma 6.11;
n
this difficulty will be resolved by Proposition 7.30 below.
Now let X SI be an I-diagram of pointed spaces. We wish to calculate with
holim X. We use the cosimplicial space T X of Example 4.2. We will see in

Section IX.5 that Tot T X is naturally weakly equivalent holim X. Note that if
I
t 0,
t T X
= T t X.
Hence the spectral sequence
s t T X = ts Tot T X

8. Obstruction theory

421

is naturally isomorphic to a spectral sequence

lims t X = ts holim X.

(7.29)

This follows from Lemma 7.25 and a simple calculation in the case t = s = 0.
Finally, we show that our two potentially different definitions of lim1 Gn agree

up to natural isomorphism.
For the next few paragraphs, let {Gn } be a tower of groups. We define three
functors. First, there is lim1 Gn , as in Lemma 6.7. Then there is lim1 Gn , which

is a temporary name for lim1 Gn as defined in (7.28). Finally, define a functor

by
{Gn } 7 0 holim BGn .

Proposition 7.30. There are natural isomorphisms

0 holim BGn
= lim1 Gn .
= lim1 Gn

Proof: We will show 0 holim BGn

= lim1 Gn and 0 holim BGn
= lim1 Gn .

The second of these isomorphisms was already proved in Proposition 6.14. For
the first, we have that
0 holim BGn
= 0 Tot lim T {BGn }.

By (5.7) and Example 5.9 we have an equality

0 Tot lim T {BGn } = 0 Tot1 lim T {BGn }.

Q
Since 0 Tot0 lim T {BGn } = 0 n BGn = , we have

0 Tot lim T {BGn } = E21,1

= lim1 Gn

by (6.1)(6.4).

8. Obstruction theory.
The purpose of this section is to develop a small amount of the theory of
Bousfields paper [10], and to discuss to some extent the meaning of the elements
on the fringe of the spectral sequence of a cosimplicial space. We close the section
with an extended example intended to make the theory concrete.

422

VII. Cosimplicial spaces

In order to make the exposition flow more smoothly we will make the following
assumption: we have a fibrant cosimplicial space X with the property that for all
n 0 and every choice of basepoint v X n , Whitehead products in (X n , v)
vanish. Thus 1 (X n , v) is abelian and the action of 1 (X n , v) on m (X n , v),
m 1, is trivial. We do not assume 0 (X n , v) = , as one of our main examples
will be X = Hom(Y, R Z), where Y S, and R Z is the R-resolution of Z S,
as in Example 4.1.
We wish to address the following question: when is an element x 0 0 X in
the image of 0 Tot X 0 Tot0 X = 0 X. We will try to lift x in successive
stages up the Tot tower. The following three results are our goal.
Proposition 8.1. An element x 0 X 0 = 0 Tot0 X lifts to 0 Tot1 X if and
only if x 0 0 X.
An element x 0 0 X determines elements xn 0 X n by
xn = din di0 X
where dit : 0 X t 0 X t+1 is any of the coface operators. We write t (X, x)
for the resulting cosimplicial group. Recall
s

N t (X, x) =

s1
\

ker{si : t (X s , xs ) t (X s1 , xs1 )}

i=0

and Z s t (X, s) is the group of cocycles in degree s in N t (X, x).

Proposition 8.2. Let n 1, y 0 Totn X, and x 0 0 X the image of y
under the projection 0 Totn X 0 0 X 0 Tot0 X. Then there is an element
(y) Z n+1 n (X, x)
so that y lifts to 0 Totn+1 X if and only if n (x) = 0.
Next recall s t (X, x) = H s N t (X, x).
Proposition 8.3. Let n 1 and suppose z 0 Totn1 X lifts to 0 Totn X.
Let x 0 0 X be the image of z under the projection 0 Totn1 X 0 0 X
0 Tot0 X. Then there is an element
(z) n+1 n (X, x)
so that z lifts to 0 Totn+1 X if and only if (z) = 0.

8. Obstruction theory

423

We will prove these three results in order. Before starting we write down the
obvious corollary. Recall our assumptions on X made in the second paragraph
of this section.
Proposition 8.4. Let x 0 0 X, and suppose n+1 n (X, x) = 0 for n 1.
Then x lifts to 0 Tot X.
Proof: This follows from Propositions 8.1 and 8.3 since
0 Tot(X) lim 0 Totn X

is surjective.

We begin the proofs of Proposition 8.1 to 8.3. To pass from 0 Totn X to

0 Totn1 X we will have to investigate the pullback square
w Hom(n+1 , X n+1 )

Totn+1 X
u

Totn X

w Hom(

n+1

, M X) Hom(n+1 ,M n X) Hom(n+1 , X n+1 ).

Let us make an abbreviation

Hom(in+1 , sn+1 )
for the lower right-hand corner of this square. The following technical lemma is
crucial to calculations. Let
I0 Totn X = Im{0 Totn X 0 Hom(in+1 , sn+1 )}.
Lemma 8.5. The composition of the map I0 Totn X 0 Hom(in+1 , sn+1 )
with
0 Hom(in+1 , sn+1 )
u
0 Hom(
is injective.

n+1

, M X) 0 Hom(n+1 ,M n X) 0 Hom(n+1 , X n+1 )

424

VII. Cosimplicial spaces

Proof: Take f 0 Hom(in+1 , sn+1 ), and let f1 Hom(n+1 , X n+1 ), g

Hom(n+1 , M n X), and f0 Hom(n+1 , M n X) be the induced vertices. If
we can show
1 (Hom(n+1 , X n+1 ), f1 ) 1 (Hom(n+1 , M n X); f0 )
is onto for all f I0 Totn X, the result will follows from the Barratt-Puppe
sequence.
Let v0 = dn+1 d1 cn (n+1 )0 be the basepoint, then let x1 = f1 (v0 ) and
x0 = f0 (v0 ) be the basepoints of X n+1 and M n X
Evaluating at the basepoint gives a diagram where the vertical maps are fibrations and the horizontal maps give a fiber sequence
Hom (n+1 , X n+1 )

w Hom(n+1 , X n+1 )

u
Hom (

w X n+1

n+1

, M X)

w Hom(

n+1

u
n

, M X)

w M X.

The means pointed maps. This leads to short exact sequences in homotopy,
m 1,
(8.6)
0

u
m (Hom (

n+1

,X

n+1

); f1 )

Hom(n+1 ,s)

u
w m (Hom (

n+1

u
m (Hom(

n+1

, M n X); f0 )

u
,X

n+1

); f1 )

w m (Hom(

n+1

, M n X); f0 )

u
m (X n+1

u
w m (M X, x0 )

u
0

u
0

The right sequence is exact because of the existence of g : n+1 M n X. Thus

f0 is contractible and the constant map section M n X Hom(n+1 , M n X)
of p actually shows the right is split exact. The left sequence is short exact

8. Obstruction theory

425

because Whitehead products vanish in X n+1 : : m /m X n+1 lifts, up to

homotopy, to Hom(n+1 , X n+1 ) with basepoint f1 if and only if the following
factors
m /m n+1

f1

w X n+1 .

m /m n+1
To finish the proof we use that f (which determines f1 , f0 , x1 , and x0 ) arises
from an element of y 0 Totn X. By projection, y determines an element
x 0 0 X and hence a unique element zk 0 X k , via any composition of
coface maps, and, as above, we write m (X, x) for the resulting cosimplicial
abelian group. Now zn+1 is the class of x1 in 0 X n+1 and the argument of
Lemma 7.1 implies
m (M n X, x0 )
= M n m (X, z).
It follows that, in diagram (8.6), that s is surjective. Also, since f0 is homotopic
to the constant map, if we let f 1 : n+1 X n+1 be a lift of f0 which is
homotopic to the constant map, then
Hom(n+1 , s) : m Hom (n+1 , X n+1 ); f 1 )
= m+n (X; x1 )
M n m+n (X, x)
= m Hom (n+1 , Mn X)
is onto. Now f 1 need not be homotopic to the constant map. However, if we let
N n+1 X be the fiber of s : X n+1 M n X at X0 , then Hom (n+1 , N n+1 X)
acts on Hom (n+1 , X n+1 ) over Hom (n+1 , M n X). Since sf1 = f0 , there
is a vertex v Hom (n+1 , N n+1 ) so that operation by v which is a homotopy
equivalence) induces a diagram

Hom (n+1 , X n+1 )N

N
P
N

w Hom (n+1 , X n+1 )

Hom (n+1 , M n X)

and so that f 1 = vf1 . Thus Hom(n+1 , s) , as in (8.6), is surjective, and the

result follows.

426

VII. Cosimplicial spaces

Proof of Proposition 8.1: There is, by (5.6), a pullback diagram (since

Tot0 X = X 0 )

Hence

Tot1 X

w Hom(1 , X 1 )

u
X0

u
w Hom(i1 , s1 ).

0 Tot1 X 0 X 0 0 Hom(i1 ,s1 ) 0 Hom(1 , X 1 )

is surjective. By Lemma 8.5, the target of this map on 0 is isomorphic to the

pullback of
0 X 1
0 X 0

w 0 X 0 0 X 0 0 X 0 0 X 1 0 X 1

where p and q are respectively

p(x) = (x, d0 x, d1 x)
q(y) = (s0 y, y, y).
But this pullback is exactly 0 0 X.

Before proving Proposition 8.2, we define the obstruction cocycle (y)

n+1
Z
n (X, x). We will use heavily that all Whitehead products vanish in X s ,
0, so that we neednt worry about the difference between unbased and based
maps in computing homotopy groups. Components do matter, however.
The element y 0 Totn (X) is determined by a map
f : skn X
and, hence, determines a unique homotopy class of maps
f n+1 : n+1 = skn n+1 X n+1 .
Then
(8.7)

(y) = [f n+1 ] n (X n+1 , x).

8. Obstruction theory

427

Since there are commutative diagrams

n+1

f n+1

w X n+1

si

u
n

si

u
w Xn

we have that (y) N n+1 n (X, x). Since there are commutative diagrams

n+1

f n+1

di

u
skn n+2

f n+2

w X n+1
u

di

w X n+2

and since n (X n+1 , x) is abelian for n 1, we will have (y) Z n+1 n (X, x) if
0=

n+1
X

(1)i di an Hn (skn+2 n+2 ).

i=0

Here an Hn (n+1 ) is the generator. But this is a standard calculation.

In the next argument we will need the following technical lemma. Recall that
X is fibrant in cS and all Whitehead products vanish.
Lemma 8.8. Let x 0 0 X and b X n any vertex with [b] = x 0 X n . Let
s : X n M n1 X be the canonical map and v = s(b). Then for k 1,
k (M n1 X, v)
= M n1 k (X, x).
Furthermore, if N n (X, x) is the fiber of s at v. Then for k 1,
k (N n (X, x), b)
= N n n (X, x).
Proof: The argument of Lemma 7.1 goes through verbatim, provided one is
explicit about basepoints. Notice the argument extends to 1 because of the
assumption that 1 (X, x) is abelian.

428

VII. Cosimplicial spaces

Proof of Proposition 8.2: If y lifts to 0 Totn+1 X, then there is a factoring

skn
u

w X.

h
j
h
h

skn+1
In particular, fn+1 factors as n+1 n+1 X n+1 and (y) = 0.
The converse is a little more difficult. We use the pullback square
Totn+1 X

w Hom(n+1 , X n+1 )

u
Totn X

u
w Hom(in+1 , sn+1 ).

Then
0 Totn+1 X 0 Totn X 0 Hom(in+1 ,sn+1 ) 0 Hom(n+1 , X n+1 )
is surjective. By Lemma 8.5, the target of this map is isomorphic to the pullback
of
0 Hom(n+1 , X)
(8.9)
u
0 Totn X

0 Hom(n+1 ,M n X)

0 Hom(

p
n+1
,X n+1 )
n+1 ,M n X) 0 Hom(

We need to show that k(y) is in the image of p.

To calculate k(y), first note that for any (fibrant) space Y
G
0 Hom(n+1 , Y )
nf r (Y, x),
=
x0 Y

where nf r (Y, x) denotes (unbased) homotopy classes of maps n+1 Y landing in the component x. If y Y is a vertex with [y] = x, then there is a
quotient
n (Y, y) nf r (Y, x)

8. Obstruction theory

429

which is an isomorphism if Whitehead products vanish in (Y, y). This later

will hold if Y = X n+1 . If x 0 Mn X is a component arising from an element
of 0 0 X it also holds as (with [y] = x)
n (Mn X, y) = M n n (X, x), n 1.
For each x 0 X n+1 , let
Kx = { n (X, y) | (sn+1 ) = 0 nf r (Mn X, f (y))}.
Then, since 0 Hom(n+1 , M n X)
= 0 M n X
0 Hom(n+1 , M n X) 0 Hom(n+1 M n X) 0 Hom(n+1 , X n+1 )
G

Kx .
=
x0 X n+1

Now let us calculate k(y), where k is as in (8.9). Since y is a map

f : skn X n+1
we have k(y) is the class of f n+1 : skn n+1 X n+1 ; that is
G

k(y) = (y) N n+1 n (X, x) = Kx

x0

The image of
p : 0 X n+1

G
x0

Kx .

X n+1

Kx

X n+1

since x 0 X n+1 to 0 Kx . Hence k(y) is in the image of p if (and only if)

(y) = 0.

To prove Proposition 8.3, we first define the obstruction class

(z) n+1 n (X, x).
In fact, let y 0 Totn X be any lift of z and let
(8.10)

(z) = [(y)] n+1 n (X, x)

where (y) Z n+1 n (X, x) is the obstruction cocycle of Proposition 8.2, defined
in (8.7).

430

VII. Cosimplicial spaces

Lemma 8.11. The class (z) is independent of the choice of lift y 0 Totn X.
Proof: Let y0 and y1 be two lifts of z. Then we choose representative
f0 , f1 : skn X

of y0 , y1 respectively. Since f0 skn1 and f1 skn1 both represent z, they are

homotopic; since skn1 skn is a cofibration we may, by altering f1 by a
homotopy, assume

f0 skn1 = f1 skn1 .
Thus we obtain a map
f0 t f1 : skn skn1 skn X.
In cosimplicial degree n
[skn skn1 skn ]n
= n n n
which is a simplicial model for S n . Let a1 , a0 be the two obvious n-simplices of
n n n and let the class [a1 a0 ] Hn (n n n ) define the orientation.
Let d(f0 , f1 ) n (X n , x) be the resulting class.
In degree n 1,
[skn skn1 skn ]n1
= n1 ,
so

d(f0 , f1 ) N n n (X, x).

In degree n + 1
[skn skn1 skn ]n+1
= n+1 skn1 n+1 n+1 .
Let it : n+1 [skn skn1 skn ]n+1 be the two obvious inclusions and
[i1 ] [i0 ] H n (skn skn1 skn ) the resulting homology class. Then
[q1 a0 ] = [i1 ] [i0 ].
Since (f1 t f0 ) it = (yt ), it follows from naturality that
d(f0 , f1 ) = (y1 ) (y0 ).

To complete the proof of Proposition 8.3, we must argue that if (z) = 0 there
exists a lift y of z to 0 Totn X so that (y) = 0. This follows from the next
result.

8. Obstruction theory

431

Lemma 8.12. Let z 0 Totn1 X and y0 0 Totn X a left of z. Let

N n n (X, x), where x 0 0 X is the image of z. Then there exists y1
1 Totn X a left of z so that
(y1 ) = (y0 ) + .
Remark 8.13. Note that Proposition 8.3 immediately follows, for if (z) = 0,
(y) = for some , and we may choose y1 so that
(y1 ) = (y) + ()
and apply Proposition 8.2.

To prove 8.12 we use a sequence of lemmas. Let

inn : nn nn

be the cofibration of (4.13). Then nn = pn n , so

homcS ( nn , X)
= homS (n , X n ).

Let n : nn skn be adjoint to the identity. Also, since

nn = n n n1 n n1 n

we have that homcS ( nn , X) is isomorphic to the pullback
homS (n , M n1 X) homS (n ,M n1 ,X) homS (n , X n ).

Let n : nn skn1 be adjoint to the element (sn , 1n ) of
homS (n , M n1 skn1 ) homS (n ,M n1 skn1 ) homS (n , n ).
Here we are using that
M n1 skn1
= M n1 skn
and sn : n M n1 skn is the canonical map. The following diagram commutes

(8.14)

in
n

u n
n

w skn1
u

w skn .

Lemma 8.15. The diagram (8.14) is a pushout diagram.

432

VII. Cosimplicial spaces

Proof: It suffices to show that after applying homsC (, X), X sC, one obtains
a natural pullback diagram. This follows from (5.6).

Now let ij : skn skn skn1 skn be the inclusion of the left functor.
Let jn : nn skn skn1 skn be adjoint to the inclusion n n n
n of the right factor.
Lemma 8.16. There is a pushout square

n in
n

w skn

in
n

u n

jn

i0

w skn skn1 skn .

Proof: This follows from Lemma 8.15 and the formula

(B2 A B1 ) B1 (B1 A B2 ) = B2 A (B1 A B2 ).

This brings us to our final technical argument.

Proof of Lemma 8.12: Let f0 : skn X represent y0 and let
N n n (X, x) n (X n , x).
Then

f0 : (skn )n = n X n

can be extended to a map

g : (skn skn1 skn )n = n n n X n
representing simply because
0 f0 : 0 n = {pt} 0 X n
n
n
n
hits x. Let
of the right hand factor and
ni1 : n be the inclusion
let h : n X be adjoint to g i1 : n X n . The claim is that we may
choose h so that the following diagram commutes

(8.17)

n in
n

in
n

f0

u n
n

w skn

u
w X.

8. Obstruction theory

433

Assuming this for the moment, we would then have, by Lemma 8.16, a map
g : skn skn1 skn X
extending g. Then if f1 is the composite
g

i1

skn skn skn1 skn X,

the argument for Lemma 8.11 and the fact that g represents finishes the
argument; in fact, if y1 = fj 0 Totn X, then
d(f0 , f1 ) =
and d(f0 , f1 ) = (y1 ) (y0 ).
To prove the claim about diagram (8.17) we compute in
hom(n , M n1 X) hom(n ,M n1 X) hom(n , X n ).
In this pullback, f0 inn n is the pair (, f0n ) where is the composite
M n1 f0

n
M n1 skn M n1 X
and f0n is short-hand for the composite
n

f0n

(skn ) X n .
On the other hand h inn is the pair (, g n i1 ) where is the composite
g

i1

Xn
M n1 X
n n n n
and g i1 denotes the composite

i1

n n n n n
Xn
since f0 = g i1 when restricted to the right hand factors; the left hand factors
are the issue. Thus we need to choose g which provides a lifting in the following
diagram
f0

i0

u
n skn1 n n

n
w X

sS

w M n1 skn

M n1 f0

u
w M n1 X.

434

VII. Cosimplicial spaces

That is, there is a representative

N n (X, x) 0 Hom(n skn1 n n , X n )
that maps to (f0 , M n1 f0 (s S)) in
0 (Hom(n , X) Hom(n ,M n1 X) Hom(n skn1 n n , M n1 X)).
Of course [f0n ] = x 0 Hom(n , X n ) = 0 X n .
Thus we need to find a representative g of
N n (X, x) 0 Hom(n n n , X n )
that maps to the class of (f0 , M n1 f0 (s S)) in
0 (Hom(n , X n )Hom(n ,M n1 X) Hom(n n n , M n1 X))
G

n (M n1 X, s(b)).
=
[b]0 X n

Now [f0 ] = x 0 Hom(n , X n ) = 0 X n and

[M n1 f0 (s S)]] = 0 n (M n1 X, x)

n (M n1 X, s(b)).

[b]0 X n

By Lemma 8.8,

n (M n1 X, x)
= M n1 n (X, x).

Since N n n (X, x) is exactly the kernel of

s : n (X n , x) M n1 n (X, x)
such a g can be found.

Now we come to the example. Let Y S be a space and let Fp Y be the

Bousfield-Kan resolution of Y , as in Example 4.1. Given another space X we
get a cosimplicial space Hom(X, Fp Y ) with
Tot Hom(X, Fp Y )
= Hom(X, Tot(Fp Y ))

= Hom(X, Yp )

8. Obstruction theory

435

where Yp is the Bousfield-Kan completion of Y . See Example 7.18. We describe

the sets s t Hom(X, Fp Y ).
Let H = H (, Fp ) and let K be the category of unstable algebras over the
Steenrod algebra. Then for any space Z there is a Hurewicz homomorphism
(8.18)

0 Hom(X, Z) homK (H Z, H Y )

send f to f . If f Hom(X, Z) is a chosen basepoint, and : S t Hom(X, Z)

represents a class in t Hom(X, Z) with this basepoint, then is adjoint to a
map : S t X Z that fits into a diagram
S t u X

wZ
[
]
[
[ f

X.
The morphism : H Z H (S t X)
= H S t H X can be decomposed
(x) = 1 f (x) + xt (x)
where xt H t S t is a chosen generator. Because is a morphism in K, the
induced map of graded vector spaces
: H Z t H X
is a morphism of unstable modules over the Steenrod algebra and a derivation
over f :
(xy) = (1)t|x| f x(y) + (x)f y.
We write DerK (H Z, t H X; f ) or simply DerK (H Z, t H X) for the vector
space of such derivations. Hence we get a Hurewicz map
(8.19)

t Hom(X, Z) DerK (H Z, t H X; f ).

If Z is a simplicial Fp vector space and Z and X are of finite type in the sense
that H n Z and H n X are finite dimensional for all n, then the Hurewicz maps
of (8.18) and (8.19) are isomorphisms. Therefore, if X and Y are of finite type
then
(8.20)

0 0 Hom(X, Fp Y )
= homK (H Y, H X)

436

VII. Cosimplicial spaces

and for any homK (H Y, H X)

(8.21)

s t Hom(X, Fp Y )
= s DerK (H Fp Y, t X; ).

In the last equation we have written for any composition of face operators
H (Fsp Y ) H Y

w H X.

The category sK of simplicial objects in in K has a simplicial model category

structure, as in Example II.6.2. Also the objects H K(Z/p, n) form a set of
projective generators for K. If we regard H Y as a constant object in sK, then
the augmentation
H Fp Y H Y
is a weak equivalence in sK and H Fp Y is cofibrant indeed, almost-free as in
Example 1.14. This last assertion follows form Example 5.8. Thus
DerK (H Fp Y, t X; )
is a model for the total derived functor of DerK (, t H X; ) applied to H Y .
Thus we may as well write
(8.22)

s DerK (H Fp Y, t X; ) = Rs DerK (H Y, t X; ).

This is an example of an Andre-Quillen cohomology and a great deal of work

has gone into learning how to compute this object in special cases. For a general
viewpoint one might look at [35]. However, the most successful application of
this and related techniques is [62].
Combining (8.22) with (8.20) and (8.21) and appealing to Proposition 8.4 we
have
Proposition 8.23. Suppose X and Y are spaces of finite type. A morphism
: H Y H X in K can be lifted to a map X Yp if
Rs+1 DerK (H Y, s X; ) = 0
for s 1.
The techniques of Lanness paper [62], cited above, show that if X = B(Z/p)n
is the classifying space of an elementary abelian p-group and A K is any
unstable algebra, then
Rs+1 DerK (A, t X; ) = 0
for all homK (A, H B(Z/p)n ) and all s 0, t 0.

Chapter VIII Simplicial functors and homotopy coherence

Suppose that A is a simplicial category. The main objects of study in this
chapter are the functors X : A S taking values in simplicial sets, and which
respect the simplicial structure of A. In applications, the simplicial category A
is typically a resolution of a category I, and the simplicial functor X describes
a homotopy coherent diagram. The main result of this chapter (due to Dwyer
and Kan, Theorem 2.13 below) is a generalization of the assertion that simplicial
functors of the form X : A S are equivalent to diagrams of the form I S in
the case where A is a resolution of the category I. The proof of this theorem uses
simplicial model structures for categories of simplicial functors, given in Section
1, and then the result itself is proved in Section 2.
The Dwyer-Kan theorem immediately leads to realization theorems for homotopy coherent diagrams in cases where the simplicial categories A model homotopy coherence phenomena. A realization of a homotopy coherent diagram
X : A S is a functor Y : I S which is weakly equivalent to X in a strong
sense. Insofar as the information arising from X typically consists of simplicial set maps X() : X(i) X(j), one for each morphism of I which only
respect the composition laws of I up to some system of higher homotopies, a
realization Y is a replacement of X, up to weak equivalence, by a collection
of maps Y () : Y (i) Y (j) which define a functor on the nose. Diagrams
of spaces which are not quite functorial are really very common: the machines
which produce the algebraic K-theory spaces, for example, are not functors (on
scheme categories in particular), but they have homotopy coherent output for
categorical reasons [49].
Approaches to homotopy coherence for simplicial set diagrams arising from
some specific resolution constructions are discussed in Section 3; traditional homotopy coherence (in the sense of [90]) is one of the examples.
But more generally, we take the point of view that a homotopy coherent diagram on a fixed index category I is a simplicial functor X : A S defined on any
simplicial resolution A of I. We can further ask for realization results concerning
homotopy coherent diagrams A M taking values in more general simplicial
model categories M. This is the subject of Section 4. We derive, in particular,
realization theorems for homotopy coherent diagrams taking values in pointed
simplicial sets, spectra and simplicial abelian groups (aka. chain complexes).
1. Simplicial functors.
We shall take the point of view throughout this chapter that a simplicial
category A is a simplicial object in the category of categories having a discrete
simplicial class of objects and satisfying the conditions of Definition 2.1 of Chap437

438

VIII. Simplicial functors and homotopy coherence

ter II. This means, in particular that the objects A K and hom(K, A) exist for
every (finite) simplicial set K and every object A of A, and that these objects
satisfy the exponential laws. The full simplicial set category S with the function
complexes Hom(X, Y ) is the canonical example of a simplicial category. In all
that follows, we shall assume that the simplicial category A has a small set of
objects.
The simplicial set of morphisms from A to B in a simplicial category A is
denoted by Hom(A, B); from this point of view, the corresponding set of nsimplices Hom(A, B)n is the set of morphisms from A to B in the category at
level n. Any morphism : B C in Hom(B, C)0 induces a simplicial set map
: Hom(A, B) Hom(A, C),
which one understands to be composition with , and which can be defined to
be the composite simplicial set map
1

0
Hom(A,B)
Hom(A,C).
=Hom(A,B) Hom(A,B)Hom(B,C)

Similarly, any morphism : A B induces a simplicial set map

: Hom(B, C) Hom(A, C),
that one thinks of as precomposition. Composition and precomposition respect
composition in and in the traditional sense.
A simplicial functor f : A B is a morphism of simplicial categories. This
means that f consists of a function f : Ob(A) Ob(B) and simplicial set maps
f : Hom(A, B) Hom(f (A), f (B)) which respect identities and composition
at all levels.
A natural transformation : f g of simplicial functors f, g : A B consists
of morphisms
A : f (A) g(A)
in hom(f (A), g(A)) = Hom(f (A), g(A))0 , one for each object A of A, such that
the following diagram of simplicial set maps commutes
Hom(A, B)

g
u
Hom(g(A), g(B))

w Hom(f (A), f (B))

B

u
w Hom(f (A), g(B))

1. Simplicial functors

439

for each pair of objects A, B of A. Notice that this is just another way of saying
that the various degeneracies of the morphisms A are natural transformations
between the functors induced by f and g at all levels. The collection of all
simplicial functors from A to B and all natural transformations between them
form a category, which we shall denote by B A . Write Nat(f, g) for the set of all
natural transformations from f to g.
The category SA of simplicial functors taking values in the simplicial set category S are of particular interest, and can be given a much more explicit description. Suppose that An denotes the small category at level n within the
simplicial category A. Then a simplicial functor X : A S consists of a function X : Ob(A) Ob(S) as before, together with a collection of simplicial set
maps X() : X(A) n X(B), one for each morphism : A B in An ,
n 0, such that
(1) the simplicial set map X() is the composite
1n

X()1

X()

X(A) n X(A) n n X(B) n X(C)

for each composable string of morphisms

A
B
C
in An ,
(2) the simplicial set map X(1A ) associated to the identity on A An is the
projection X(A) n X(A), and
(3) for each ordinal number map : m n and each morphism : A B
of An , the following diagram commutes:

X(A) m A
1
u

X(A) n

A X( ())
AA
C

X()

w X(B).

Then, from this point of view, a natural transformation : X Y of simplicial functors taking values in simplicial sets consists of simplicial set maps

440

VIII. Simplicial functors and homotopy coherence

A : X(A) Y (A), one for each A Ob(A), such that the diagram
X(A) n
A 1

X()

Y (A) n

Y ()

w X(B)
B
u
w Y (B)

commutes for each morphism : A B of An , and for all n 0.

Now its easy to see that the category SA is the category at level 0 of a
simplicial category. Given a simplicial functor X : A S and a simplicial set
K, there is a simplicial functor K X which assigns to each A Ob(A) the
simplicial set K X(A) and to each morphism : A B in An the simplicial
set map
1X()

K X(A) n K X(B).
Dually, there is a simplicial functor hom(K, X) which associates to each A
Ob(A) the simplicial set hom(K, X(A)), and to each morphism : A B of An
the simplicial set map hom(K, X(A)) n hom(K, X(B)) which is defined
to be the adjoint of the composite
ev1

X()

K hom(K, X(A)) n X(A) n X(B),

where ev : K hom(K, X(A)) X(A) is the standard evaluation map. It
follows immediately that the collection of all such evaluation maps induces a
natural bijection
Nat(K Y, X)
= Nat(Y, hom(K, X)).
There is also, plainly, a simplicial set hom(Y, X) whose set of n-simplices is
the set Nat(n Y, X). Finally, since the simplicial set K is a colimit of its
simplices, there is a natural bijection (exponential law)
Nat(K Y, X)
(1.1)
hom(K, Hom(Y, X)) =
relating morphisms in the simplicial set category to natural transformations of
simplicial functors.
The representable simplicial functor Hom(A, ) is the simplicial functor which
associates to each B Ob(A) the simplicial set Hom(A, B), and associates to
each morphism : B C of An the composite simplicial set map
1

Hom(A, B) n Hom(A, B) Hom(B, C)

Hom(A, C).
It is central to observe that the representable simplicial functors in SA satisfy
the Yoneda lemma:

1. Simplicial functors

441

Lemma 1.2. Suppose that A is an object of a simplicial category A, and that

X : A S is a simplicial functor. Then there is a natural isomorphism of
simplicial sets
Hom(Hom(A, ), X)
= X(A).
Proof: The natural transformations : Hom(A, ) X are in one to one
correspondence with the vertices of X(A). In effect, given a morphism : A B
of Am , there is a commutative diagram
Hom(A, A) m

A 1

u
Hom(A, B)

w X(A) m
X()
u
w X(B),

so that
B () = X()(A (1A ) m ),
where m = 1m m
m is the classifying simplex. But 1A = s(1A,0 ) where 1A,0
denotes the identity on A in A0 . Thus, is completely determined by the vertex
A (1A,0 ) of X(A).
It follows that there are natural bijections
Nat(n Hom(A, ), X)
= Nat(Hom(A, ), hom(n , X))

= hom(n , X(A))0

= X(A)n
and that these bijections respect the simplicial structure.

n
A natural transformation : Hom(A, ) X is completely determined
by the n-simplex (n , 1A ) of the simplicial set X(A): this is a corollary of the
proof of Lemma 1.2.
The functor S S defined by K 7 K n preserves all small colimits of
simplicial sets and takes limits to limits fibred over n . It follows immediately
that the category SA of simplicial functors from A to S is complete and cocomplete, and that all limits and colimits are formed pointwise.
There is an analogue of the simplex category for each simplicial functor X
A
S . Write A X for the category whose objects are the transformations
n Hom(A, ) X,

442

VIII. Simplicial functors and homotopy coherence

and whose objects are all commutative triangles of transformations of the form
n Hom(A, )

4
6
44

u
4
Hom(B, )
m

The objects of this category are simplices of sections of X, and the morphisms
form its structural data. It follows that X is a colimit of its simplices in SA in
the sense that there is a natural isomorphism
(1.3)

X
=

n Hom(A, ).

lim

n Hom(A, )X

Say that a map f : X Y of SA is a fibration (respectively weak equivalence)

if the component maps f : X(A) Y (A) are fibrations (respectively weak
equivalences) of simplicial sets, for all objects A of A. Such maps will be called
pointwise fibrations and pointwise weak equivalences, respectively. A cofibration
of SA is a map which has the left lifting property with respect to all trivial
fibrations.
Proposition 1.4. With these definitions, the category SA of simplicial setvalued simplicial functors defined on a small simplicial category A satisfies the
axioms for a simplicial model category.
Proof: CM1 is a consequence of the completeness and cocompleteness of SA ,
as described above. The weak equivalence axiom CM2 and the retract axiom
CM3 are both trivial consequences of the corresponding axioms for simplicial
sets.
Note that a map f : X Y of SA is a fibration if and only if f has the right
lifting property with respect to all maps
i1

nk Hom(A, ) n Hom(A, ),
and f is a trivial fibration if and only if it has the right lifting property with
respect to all maps
i0 1

n Hom(A, ) n Hom(A, ),

1. Simplicial functors

443

where i : nk , n and i0 : n , n denote the respective canonical inclusions. Both statements follow from the exponential law (1.1) and Lemma 1.2.
The map i 1 is a weak equivalence.
We may therefore apply standard small object arguments to prove the factorization axiom CM5 for SA . In particular, any map f : X Y has a
factorization
j

X4

4
46
f 4

wZ
p
u
Y,

where p is a fibration and j is a filtered colimit of maps of the form jn : Xn

Xn+1 , each of which is defined by a pushout diagram
G

nkrr Hom(Ar , )

w Xn

F
r

jn

(i 1)

nr

u
Hom(Ar , )

u
w Xn+1 .

It follows that each of the maps jn is a pointwise weak equivalence, and has the
left lifting property with respect to all fibrations, and so the map j has the same
properties, giving one of the factorizations required by CM5 in particular. The
other factorization has a similar construction.
The axiom CM4 is a standard consequence of the method of proof of the
factorization axiom CM5, in that any map f : X Y which is a cofibration
and a pointwise weak equivalence has a factorization f = p j, where p is a
fibration and where j is a pointwise weak equivalence and has the right lifting
property with respect to all fibrations. In particular, p is a pointwise weak
equivalence, and so the indicated lifting exists in the diagram
X

wZ
\
]
\
p
f
\
u \
u
w Y.
Y
1G

444

VIII. Simplicial functors and homotopy coherence

The map f is therefore a retract of j, and therefore has the left lifting property
with respect to all fibrations.
The axiom SM7 is a consequence of the corresponding statement for simplicial
sets.

Remark 1.5. Proposition 1.4 is a special case of a result of Dwyer and Kan
[24]. The proof given here is much simpler.
Remark 1.6. The Bousfield-Kan closed model structure for I-diagrams of simplicial sets given in Example II.7.11 and Section IV.3.1 is a special case of Proposition 1.4. Every category I having all coproducts and products is a simplicial
category, for which Hom(X, Y ) is the discrete simplicial set on the set of morphisms hom(X, Y ) from X to Y in I. The required isomorphisms
hom(X K, Y )
= homS (K, Hom(X, Y ))
= hom(X, hom(K, Y ))
F
respectively force X K to be Q
a copy of the coproduct 0 K X, and dictate
that hom(K, Y ) is the product 0 K Y . In the language of Section IV.3.1, the
simplicial set K Hom(A, ) is a copy of the Kan extension FA K.
2. The Dwyer-Kan theorem.
A simplicial functor X : A S can alternatively be described as a rule
which associates to each object A a simplicial set X(A), and to each morphism
: A B of An a function : X(A)n X(B)n on the n-simplex level such
that
(1) the assignment 7 is functorial in morphisms of An , for all n 0,
and
(2) for each morphism : A B of An and each ordinal number map
: m n, the following diagram commutes:
X(A)n

u
X(A)m

w X(B)n

( )

u
w X(B)m .

Given a simplicial functor X, with induced maps : X(A) n X(B)

as defined above for : A B, the corresponding morphism : X(A)n
X(B)n is the assignment x 7 (x, n ). Conversely, given a system of maps
: X(A)n X(B)n having the properties described in (1) and (2) above, one
defines a map : X(A) n X(B) on m-simplices (x, ) by the assignment
(x, ) 7 ( ) (x). These two assignments are inverse to each other.

2. The Dwyer-Kan theorem

445

Example 2.1. Suppose that C is an object of a simplicial category A. Then, according to this new description, the representable simplicial functor Hom(C, ) :
A S associates to each object A the simplicial set Hom(C, A) and to each
morphism : A B of An the function

Hom(C, A)n Hom(C, B)n

defined by composition with in An . Its easier to make sense of this if one
recalls that Hom(C, B)n is precisely the set of morphisms from C to B in the
category An .
From this new point of view, a natural transformation of simplicial functors
f : X Y is a collection of simplicial set maps f : X(A) Y (A) such that, for
each morphism : A B of An the diagram of functions

X(A)n

u
Y (A)n

w X(B)n
f
u
w Y (B)n

commutes.
Suppose that f : A B and X : A S are simplicial functors and that B
is an object of the simplicial category B, then the bisimplicial object f X(B)
defined by
f X(B)n,m
G
=
X(A)0,m Hom(A0 , A1 )m Hom(An1 , An )m Hom(f An , B)m
The horizontal simplicial set f X(B),m is the nerve of the translation category
associated to the composite functor
Q

Xm

f B Am Sets,
and each morphism : B C of Bm determines a simplicial set map
: f X(B),m f X(C),m .

446

VIII. Simplicial functors and homotopy coherence

On the other hand, if : k m is an ordinal number map, then there is a

commutative diagram of simplicial set maps

f X(B),m
(2.2)

w f X(C),m

f X(B),k

( )

w f X(C),k .

It follows that applying the diagonal simplicial set functor gives a simplicial
functor B S defined by (n, B) 7 f X(B)n,n , which will also be denoted by
f X. The simplicial functor f X is the homotopy left Kan extension of X along
the simplicial functor f .
Weve already noted that the simplicial set f X(B),m is the nerve of a translation category. Furthermore, the simplicial set diagram (2.2) is induced by a
commutative diagram of functors. Suppose, for example, that the simplicial
functor X is the representable functor Hom(A, ). Then the corresponding simplicial set f Hom(A, )(B),m can be identified with the nerve of the translation
category associated to the composite contravariant functor
Q

Hom( ,B)m

A Am Am
Bm Sets.
The identity morphism on A is initial in A Am , so the category
f Hom(A, )(B),m
contracts canonically onto the discrete subcategory on the set of objects
Hom(f (A), B)m .
In this way, we obtain maps
(
s
Hom(f (A), B)m
f Hom(A, )(B),m ,
(2.3)
r
f Hom(A, )(B),m
Hom(f (A), B)m
such that rs = 1, as well as a natural transformation
H

f Hom(A, )(B),m 1 f Hom(A, )(B),m

from the composite functor sr to the identity. All of this data is natural in B,
and simplicial, and so we have proved

2. The Dwyer-Kan theorem

447

Lemma 2.4. Suppose that f : A B is a simplicial functor, and let A be an

object of A. Then the simplicial functor Hom(f (A), ) is a strong deformation
retract of f Hom(A, ).
We shall also need
Lemma 2.5. The simplicial functors Hom(f (A), ) and f Hom(A, ) are cofibrant.
Proof: The assertion about Hom(f (A), ) is a consequence of the Yoneda
lemma and the observation that any trivial fibration p : X Y of SB consists of maps p : X(B) Y (B) which are trivial fibrations of simplicial sets and
are therefore surjective on vertices.
The object f Hom(A, ) may also be interpreted as a simplicial object in the
category SB of simplicial functors defined on B.
Suppose, for the moment, that Z is an arbitrary simplicial object in SB . Then
Z consists of simplicial functors Zn , n 0, and simplicial structure maps :
Zn Zm in SB corresponding to ordinal number maps : m n. The object
Z is, in other words, a type of diagram taking values in bisimplicial sets. The
standard constructions on bisimplicial sets apply to Z. There is, in particular, an
associated diagonal simplicial functor d(Z) which is defined by d(Z)n = Zn,n , and
d(Z) is a co-end in the simplicial functor category for the diagrams of simplicial
functor maps
Z n m

w Z n n

u
Z m n

arising from ordinal number maps : m n. This means that there are
simplicial functor maps n : Zn n d(Z) defined in sections by (x, ) 7
(x), and these maps assemble to give the universal arrow in a coequalizer
diagram
G
G
Z n m
Zn d(Z).
:mn

Finally, d(Z) has a filtration d(Z)(p) d(Z)(p+1) . . . , where d(Z)(p) is generated by the images of the maps r , 0 r p. We are interested in proving that
the simplicial functor d(Z) is cofibrant, so the key, for us, is the existence of the

448

VIII. Simplicial functors and homotopy coherence

pushout diagram
(s[p] Zp p+1 ) (Zp+1 p+1 )
y

w d(Z)(p+1)
y

u
Zp+1 p+1

u
w d(Z)

(p+1)

in the category SB . Here, the inclusion on the left is induced by the inclusion
s[p] Zp , Zp+1 of the horizontally degenerate part in Zp+1 . The central observation is that the simplicial model structure on SB implies that d(Z) is cofibrant
if each of the inclusions s[p] Zp , Zp+1 is a cofibration.
The object at level n for the simplicial object f Hom(A, ) is the disjoint
union
G
Hom(A, A0 ) . . . Hom(An1 , An ) Hom(f (An ), ),
which may in turn be written as a disjoint union
G

B(A A)C
n Hom(f (C), ),

COb(B)

where B(A A)C

n indicates the simplicial set of strings
A A0 An1 An = C,
ending at C. The horizontal degeneracies preserve this decomposition, and so
the inclusion of the degenerate part in
G

B(A A)C
n+1 Hom(f (C), ),

COb(B)

can be identified with the map

(2.6)

F
COb(B)

DB(AA)C
n+1 Hom(f (C), ),

F
COb(B)

B(AA)C
n+1 Hom(f (C), ),

which is induced by the simplicial set inclusions

C
DB(A A)C
n+1 B(A A)n+1 .

2. The Dwyer-Kan theorem

449

Each of the simplicial functors Hom(f (C), ) is cofibrant, so the map (2.6) is
a cofibration of simplicial functors, and the simplicial functor f Hom(A, ) is
cofibrant, as claimed.

Suppose now that Y : B S is a simplicial functor. Then a simplicial functor

f Y : A S is specified at an object A of A by
f Y (A) = Hom(f Hom(A, ), Y ).
In other words, an n-simplex of f (Y )(A), or rather a morphism
n Hom(A, ) f Y
is defined to be a morphism
n f Hom(A, ) Y.
There is an isomorphism
n f Hom(A, )
= f (n Hom(A, )),
by formal nonsense. The resulting isomorphisms
Nat(n Hom(A, ), f Y )
= Nat(n f Hom(A, ), Y )

= Nat(f (n Hom(A, )), Y )

are natural with respect to maps
m Hom(B, ) n Hom(A, ).
Furthermore, every simplicial functor X : A S is a colimit of its simplices,
as in (1.3), and the functor f preserves colimits. It follows that there is an
adjunction isomorphism
(2.7)

Nat(f X, Y )
= Nat(X, f Y ).

The simplicial functor f : A B also induces a functor f : SB SA , which

is defined by f Y (A) = Y (f (A)). The notation is supposed to remind one of a
direct image functor.
Note in particular, that the adjoint of a n-simplex
x

n Hom(A, )
f Y

450

VIII. Simplicial functors and homotopy coherence

is the composite
x

f (n Hom(A, ))
Y
= n f Hom(A, )
It follows that the canonical map
: n Hom(A, ) f f (n Hom(A, ))
is the map which sends the classifying simplex (n , 1A ) to the canonical isomorphism

=
n f Hom(A, ) f (n Hom(A, )),
since must be adjoint to the identity on f (n Hom(A, )). Thus, if x
is an n-simplex of X(A), then (x) f f X(A) is the simplex defined by the
composite
(2.8)

f (x)

=
f Hom(A, ) f (n Hom(A, )) f X.
n

The adjunction (2.7) is easily promoted to a natural isomorphism of simplicial

sets
Hom(f X, Y )
= Hom(X, f Y ).

(2.9)

We have therefore proved the first part of the following result:

Proposition 2.10. Suppose that f : A B is a functor between small simplicial categories. Then
(1) The functors
f : SA SB : f
are simplicially adjoint in the sense that there is a natural isomorphism
of simplicial sets
Hom(f X, Y )
= Hom(X, f Y ).
(2) the functor f preserves weak equivalences and fibrations.
(3) the functor f preserves weak equivalences and cofibrations.

2. The Dwyer-Kan theorem

451

Proof: To prove Part (2), observe that f Hom(A, ) is cofibrant for every
object A A, by Lemma 2.5, so that every map
nk f Hom(A, ) n f Hom(A, )
is a trivial cofibration on account of the simplicial model structure of the category
SB . It follows that the functor f preserves fibrations.
The retraction map
r : f Hom(A, ) Hom(f (A), )
is natural in A, and therefore induces a map
r : f Y f Y
which is natural in Y SB . It follows that f preserves weak equivalences, if it
can be shown that the map r is a natural weak equivalence.
Recall that the map r has a section
s : Hom(f (A), ) f Hom(A, ),
and that there is a homotopy
H : f Hom(A, ) 1 f Hom(A, )
from the composite sr to the identity. The induced map
H : Hom(f Hom(A, ), Y ) Hom(f Hom(A, ) 1 , Y )
can be composed with the canonical map
Hom(f Hom(A, ) 1 , Y ) hom(1 , Hom(f Hom(A, ), Y ))
to give a map which is adjoint to a simplicial homotopy
Hom(f Hom(A, ), Y ) 1 Hom(f Hom(A, ), Y )
from the identity map to s r . This implies that r is a pointwise weak equivalence, as required.
For Part (3), the functor f preserves weak equivalences as a consequence
of its definition as a type of homotopy colimit. Also, f is left adjoint to the
functor f , and it is a consequence of Part (2) that the latter preserves trivial fibrations. Standard closed model category tricks therefore imply that the functor
f preserves cofibrations.

452

VIII. Simplicial functors and homotopy coherence

Corollary 2.11. The functor f : SB SA induces an equivalence

f : Ho(SB ) Ho(SA )
of homotopy categories if and only if the adjunction maps
: X f f X

and

: f f Y Y

are weak equivalences.

Proof:
The functors f and f induce an adjoint pair of functors
f : Ho(SA ) Ho(SB ) : f
between the associated homotopy categories, by Proposition 2.10. If the induced
functor f is an equivalence of categories, then it is part of an adjoint equivalence
(see [66, p.91]). The adjoint to f of that adjoint equivalence is therefore naturally isomorphic to f in Ho(SB ), and so the adjunction maps and induce
isomorphisms on the homotopy category level. It follows that and are weak
equivalences.
The converse is clear.

There is a simplicial map i : X(A) f X(f (A)) that is defined on the

bisimplicial set level to be the map
G
X(A)
X(A0 )Hom(A0 , A1 ) Hom(An1 , An )Hom(f (An ), f (A))
by sending an n-simplex x to the element (x, 1A , . . . , 1A , 1f (A) ) in the summand
X(A) Hom(A, A) Hom(A, A) Hom(f (A), f (A))
corresponding to Ai = A for 0 i n. It follows from the description of the
adjunction map given in (2.8) that i is the composite of the map

X(A)
f f X(A) = Hom(f Hom(A, ), f X)
with the map
s

Hom(f Hom(A, ), f X) Hom(Hom(f (A), ), f X)

= f X(f (A))
induced by precomposition with the section s : Hom(f (A), ) f Hom(A, )
given in (2.3). Neither s nor i is natural in A.

2. The Dwyer-Kan theorem

453

Corollary 2.12. The functor f : SB SA induces an equivalence

f : Ho(SB ) Ho(SA )
of homotopy categories if and only if the following conditions are satisfied:
(1) for every simplicial functor X SA and every object A A the map
i : X(A) f X(f (A)) = f f X(A)
is a weak equivalence,
(2) a map : Y1 Y2 of SB is a weak equivalence if and only if the induced
map f : f Y1 f Y2 is a weak equivalence of SA .
Proof: If f is an equivalence of homotopy categories, then (2) holds, by
Lemma II4.1. Furthermore, the canonical map map is a weak equivalence
by Corollary 2.11. It follows from the proof of Proposition 2.10 that s is a weak
equivalence in general, and so the map i is a weak equivalence as claimed.
Conversely, if conditions (1) and (2) are satisfied, then is a weak equivalence,
and then f () is a weak equivalence on account of a triangle identity. The
conditions for Corollary 2.11 are therefore met, since f () must also be a weak
equivalence.

Dwyer and Kan say that a functor f : A B between small simplicial categories is a weak r-equivalence if two conditions hold:
(a) for every pair of objects A1 , A2 of A, the functor f induces a weak equivalence of simplicial sets
Hom(A1 , A2 ) Hom(f (A1 ), f (A2 )),
(b) every object in the category of components 0 B is a retract of an object
in the image of 0 f .
Every functor f : A B which satisfies condition (a) and is surjective on
objects is a weak r-equivalence. Most examples of weak r-equivalences that are
encountered in nature have this form.
The following is the main result of this chapter:
Theorem 2.13 (Dwyer-Kan). Suppose that f : A B is a functor between
small simplicial categories. Then the induced functor
f : Ho(SB ) Ho(SA )
of homotopy categories is an equivalence of categories if and only if the functor
f is a weak r-equivalence.

454

VIII. Simplicial functors and homotopy coherence

Proof: Suppose that the functor f is a weak r-equivalence. We shall verify the
conditions of Corollary 2.12.
The instance of i corresponding to the identity functor 1A on A has the form
i : X(A) 1A X(A).
This map is a weak equivalence since 1A induces an equivalence Ho(SA )
Ho(SA ), and it fits into a commutative diagram
1A X(A)

'
)
i ''
''
f
X(A)

u
i

f X(f (A)),
where the indicated map f is induced by a bisimplicial set map given on summands by the simplicial set maps
X(A0 ) Hom(An , A) X(A0 ) Hom(f (An ), f (A))
defined by 1 1 f . This map f is a weak equivalence, by assumption,
and so the map i : X(A) f X(f (A)) is a weak equivalence as well.
Suppose that : Y1 Y2 is a map of SB such that f : f Y1 f Y2 is a
weak equivalence of SA . Then the induced simplicial set maps : Y1 (f (A))
Y2 (f (A)) are weak equivalences for all objects A A. The assumption that f
is a weak r-equivalence means, in part, that for each B B there is an object
A A and maps j : B f (A) and q : f (A) B in B such that qj maps to the
identity in 0 B. It follows that, for each Y SB , there is a path of simplicial
homotopies from the composite simplicial set map
Y (j)

Y (q)

Y (B) Y (f (A)) Y (B)

to the identity on Y (B). It follows that this composite is a weak equivalence for
any Y . In particular, in the diagram
Y1 (B)

Y1 (j)

w Y1 (f (A))

u
Y2 (B)

Y1 (q)

u
Y2 (j)

w Y2 (f (B))

w Y1 (B)

Y2 (q)

u
w Y2 (B)

2. The Dwyer-Kan theorem

455

both horizontal composites are weak equivalences, so that : Y1 (B) Y2 (B) is

a weak equivalence as well.
Conversely, suppose that f : A B induces the indicated equivalence of
homotopy categories. The following diagram commutes:
Hom(B, A) A i
w f Hom(B, )(A)
A
AC
r
f A
u
Hom(f (B), f (A))
The map i is a weak equivalence by assumption, and r is the weak equivalence
of Lemma 2.4, and so the condition (a) for f to be weak r-equivalence is verified.
The simplicial functor f f Hom(B, ) is the colimit
f Hom(A, ),

lim

x:f Hom(A, )n Hom(B, )

and the canonical map

: f f Hom(B, ) Hom(B, )
is the unique map having components given by the morphisms
x : f Hom(A, ) n Hom(B, )
in the definition of the colimit. The natural map is a weak equivalence by
assumption, and the path component functor commutes with colimits, so there
is a map
f Hom(A, ) n Hom(B, )
such that 1B Hom(B, B)0 lifts to a vertex of f Hom(A, B) n up to
(iterated) homotopy. The map
s1

Hom(f (A), B) n f Hom(A, B) n

is a weak equivalence, so that 1B is in the image of the composite
s1

Hom(f (A), B) n f Hom(A, B) n Hom(B, B)

up to homotopy. This composite classifies an n-simplex of Hom(B, f (A)),
and one verifies that this maps sends a 0-simplex (, v) of Hom(f (A), B) n
maps to the composite v ().

456

VIII. Simplicial functors and homotopy coherence

3. Homotopy coherence.
Homotopy coherence theory is more a collection of phenomena than theorems.
The whole point of the theory is to determine when a diagram which commutes
up to a system of higher homotopies can be replaced by a diagram which commutes on the nose. To this end, the game is either to recognize when systems of
higher homotopies can be suitably defined to serve as input for the Dwyer-Kan
theorem (Theorem 2.13), or to avoid that result altogether, as one does (preferably) with lax functors. Various examples of these phenomena will be described
here. This material pertains to homotopy coherent diagrams of simplicial sets,
or spaces. A discussion of homotopy coherence phenomena for other selected
simplicial model categories appears in Section 4.
3.1. Classical homotopy coherence.
Suppose that I is a small category, and write U I for the underlying directed
graph of I. The graph U I is pointed in the sense that there is a distinguished
element, namely 1a , in the set I(a, a) of arrows from a to itself, for all objects
a I. There is a free category F X associated to each pointed directed graph X,
which has the same objects as X, and all finite composable strings of non-identity
arrows in X as morphisms. Composition in F X is given by concatenation. The
free category functor F is left adjoint to the underlying graph functor U , with
canonical maps : 1 U F and : F U 1. The canonical maps can be
used, along with various iterations of the composite functor F U to define a
simplicial category F I. The category Fn I of n-simplices of F I has the form
Fn I = (F U )n+1 I, and the faces and degeneracies of F I are defined by

di = (F U )i1 (F U )ni ,
sj = (F U )j F U (F U )nj .

and

One can appeal to the dual of results of [66, p.134,171] to see that this definition
works, or check the simplicial identities directly. The diagram

F U F U (I)

d0
d1

ww F U (I) w I

is a coequalizer in the category of small categories. It follows that, for any pair
of objects a, b of I, the diagram of functions between morphism sets

F U F U (I)(a, b)

d0
d1

ww F U (I)(a, b) w I(a, b)

3. Homotopy coherence

457

is a coequalizer. It can be shown directly that the underlying simplicial graph

U F I has an extra degeneracy (see Section III.5), given by the functors
s1 = U (F U )n (I) U (F U )n+1 (I).
Now, U F (I)(a, b) = F (I)(a, b) as a simplicial set, so that F (I)(a, b) has an
extra degeneracy. The simplicial set map
F I(a, b) K(I(a, b), 0)
induced by is therefore a weak equivalence by Lemma III.5.1, so we have
constructed a weak r-equivalence
: F I K(I, 0)
of simplicial categories called simplicial free resolution of the category I. It is a
special case of a construction that is quite ubiquitous.
The simplicial free resolution F I of a category I is the traditional basis of the
definition of homotopy coherence. Explicitly, a homotopy coherent I-diagram is
classically defined to be a simplicial functor X : F I S.
Homotopy coherent diagrams are notoriously difficult to interpret, much less
construct. Intuitively, X : F I S associates a simplicial map : X(a)
X(b) to each non-identity morphism : a b of I ( is the image of the
string () F U (I)). Given : b c in I, its not the case that () =
, but rather that the two maps are homotopic in a precise way, via the
homotopy X((, )) : X(a) 1 X(c) which is the image of the arrow
((, )) F U F U (I). Given yet another I-morphism : c d, one sees that
the associativity relationship () = in I in the sense that the morphism
(((), (, ))) of F U F U F U (I) determines a map X(a) 2 X(d), which is
a higher homotopy that has the homotopies
X((, ))
()
w ()

X((), (, ))

X((, , ))
u

appearing in its 1-skeleton. The higher, or iterated, associativity relation
(()) =

458

VIII. Simplicial functors and homotopy coherence

corresponds to the 3-simplex (((()), ((), (, )))) of F I. The homotopy coherent diagram X is determined by all higher associativities or iterated bracketings
of strings n n1 1 of morphisms in I.
Homotopy coherent diagrams are very rarely constructed from scratch, although there are obstruction theoretic techniques for doing so [27]. They nevertheless appear quite naturally, usually as the output of large categorical machines
more will be said about this below.
Most of the point of having a homotopy coherent diagram in hand is that
one can immediately replace it up to pointwise weak equivalence by a diagram
that commutes on the nose. This is a consequence of the following realization
theorem:
Theorem 3.1 (Realization). Suppose that X : F I S is a homotopy
coherent diagram in the category of simplicial sets. Then X is naturally pointwise
weakly equivalent to a diagram Y , for some ordinary I-diagram Y : I S.
Proof: This result follows easily from the Dwyer-Kan theorem 2.13 and its
proof: the simplicial functor : F I K(I, 0) is a weak r-equivalence, so there
are natural pointwise weak equivalences
i

X
X X.
Take Y to be X.

3.2. Homotopy coherence: an expanded version.

The classical definition of homotopy coherence is interesting and complicated,
but not flexible enough for all applications. It does not, for example, take into
account diagrams where images of identity maps could wiggle away from actual
identities up to controlled homotopy The realization result, Theorem 3.1, is also
obviously just a special case of a much broader statement.
Define a resolution of a category I to be a simplicial functor : A K(I, 0)
such the is a weak r-equivalence. A homotopy coherent diagram on the category
I shall henceforth be defined to be a simplicial functor X : A S, where
: A K(I, 0) is a resolution of I. We now have an expanded form of the
realization theorem:
Theorem 3.2. Suppose that X : A S is a homotopy coherent diagram which
is defined with respect to some realization : A K(I, 0) of I. Then X is
naturally pointwise equivalent to a diagram of the form Y , for some I-diagram
Y : I S.
Proof: The proof is a copy of the proof of Theorem 3.1.

3. Homotopy coherence

459

The rest of this section will be taken up with a description of a natural resolution BIs K(I, 0) of the category I which is different from the simplicial free
resolution of the previous section. Homotopy coherent diagrams X : BIs S
provide for variance of morphisms induced by the identities of I up to homotopy.
Let a and b be objects of the small category I. There is a category, denoted
Is (a, b), whose objects are the functors of the form : n I, with (0) = a and
(n) = b. The morphisms of Is (a, b) are commutative diagrams
n0

6
44
u 44
1
n1

where 0 and 1 are objects of Is (a, b) and : n0 n1 is an ordinal number

map which is end point preserving in the sense that (0) = 0 and (n0 ) = n1 .
Suppose that : n I is an object of Is (a, b) and : m I is an object
of Is (b, c). The poset join n m may be identified up to isomorphism with the
ordinal number n + m, in such a way that the inclusion n , n m is identified
with the ordinal number map n n + m defined by j 7 j + n for 0 j m.
The objects and together determine a functor : n + m I, defined by

(j),
if 0 j n,
(j) =
(j n), if n j m + n.
This is plainly the object level description of a functor

Is (a, b) Is (b, c)
Is (a, c),
which is called the join functor. This operation is associative, and has a two-sided
identity in each Is (a, a) given by the object a : 0 I.
We have constructed a category Is which has the same objects as I, and is
enriched in the category cat of small categories. Applying the nerve construction
to each of the categories Is (a, b) gives simplicial sets BIs (a, b), which then form
the morphism objects for a simplicial category BIs , which has the same objects
as I. These constructions are obviously natural: a functor f : I J induces a
functor f : Is Js of categories enriched in cat, and hence determines a functor
f : BIs BJs of of simplicial categories.

460

VIII. Simplicial functors and homotopy coherence

The category Is (a, b) has the form

Is (a, b) =

Is (a, b)f ,

f :ab
in I

where Is (a, b)f is the subcategory of those strings : n Is whose composite

is f : a b. The category Is (a, b)f has an initial object, namely the functor
f : 1 Is which is determined by f, so that, on the simplicial set level, we have
a decomposition
G
BIs (a, b) =
BIs (a, b)f ,
f :ab
in I

of BIs (a, b) into connected components, each of which is contractible. It follows

that the path component functor determines a resolution BIs K(I, 0) of the
category I.
A homotopy coherent diagram X : BIs S exhibits the standard homotopy
coherence phenomena. Suppose that
1

a0 a1 a2
is a composable pair of morphisms of I, and let these morphisms canonically
determine a functor : 2 I, which in turn is a 0-simplex of BIs (a0 , a2 ).
Then X associates simplicial set morphisms (1 ) : X(a0 ) X(a1 ) and (2 ) :
X(a1 ) X(a2 ) to the morphisms (aka. 0-simplices) 1 and 2 respectively, and
associates the composite simplicial set map
(2 ) (1 ) : X(a0 ) X(a2 )
to the simplex , since X takes joins to composites. There is a 1-simplex of
BIs (a0 , a2 ), defined by the picture

2u

d1
1

Is ,
6
4
4
442 1

3. Homotopy coherence

461

where 1 2 is the unique endpoint preserving ordinal number map, and this
1-simplex is mapped to a homotopy X() : X(a0 )1 X(a2 ) from (2 1 ) to
the composite (2 ) (1 ) . In the same way, the associativity relation 3 (2 1 ) =
3 2 1 gives rise to a 2-simplex
d1

14

d2

w2

(3 2 ,1 )
4
3 2 1 4
44
6

w3
h

h
h
(3 ,2 ,1 )

h
h
u h
k
I

and hence to a higher homotopy X(a0 ) 2 X(a3 ).

If a is an object of I, then the identity map 1a : a a defines a 0-simplex 1a :
1 I, and hence gets mapped to the simplicial set map (1a ) : X(a) X(a),
whereas the 0-simplex a : 0 I is an identity for BIs , and is therefore sent to
the identity 1X(a) on X(a). There is a 1-simplex of BIs (a, a), of the form
0u
a

6
44
441a

which is then mapped to a homotopy X() : X(a) 1 X() from (1a ) to

1X(a) .
3.3. Lax functors.
A lax functor F : I cat associates a category F (a) to each object a of I,
and associates a functor : F (a) F (b) to each morphism : a b, in such
a way that there are natural transformations
(, ) : ()

and a : (1a ) 1F (a) ,

which together satisfy the cocycle conditions

()
(3.3)

(, )

(, )

u
()

(, )

w ()
(, )
u
w

462

VIII. Simplicial functors and homotopy coherence

1F (a) u

(1a )
u

(3.4)

(1b )
u

(, 1a )

w 1F (b)

(1b , )

(1a )

(1b )

The category cat of categories is enriched in categories. Its morphism categories Hom(A, B) have the functors from A to B as objects with natural transformations as morphisms. The composition functor
Hom(A, B) Hom(B, C) Hom(A, C)
is defined on morphisms as follows: given natural transformations 1 and 2
F1
A

F2
w

G1

G2

C
w

one defines the composition 2 1 of 1 and 2 to be the diagonal transformation

appearing in the commutative diagram
F2 1

F2 F1 4
2 F1

G2 F1

4
4
6

G2 1

w F2 G1
2 G1
u
w G2 G1 .

There is a one to one correspondence between functors

Is cat
and lax functors on I. In particular, given a lax functor F as above, associate to
each string n I in I of the form
1

a0 a1 . . . an

3. Homotopy coherence

463

the composite functor

(n ) . . . (1 ) : F (a0 ) F (an ).
The requisite functors
Is (a0 , an ) Hom(F (a0 ), F (an ))
are defined on morphisms of Is (a0 , an ) by first making a definition on cofaces and
codegeneracies, which are completely determined up to join by the transformations and , and then by showing that the relevant cosimplicial identities are
satisfied. The non-trivial cosimplicial identities amount to the cocycle conditions
which appear in the definition of the lax functor F .
The object cat has associated to it a simplicial category Bcat in the obvious way: one uses the nerves BHom(A, C) of the categories of morphisms
Hom(A, C) appearing in cat. Note that BHom(A, C) is canonically isomorphic to the function complex Hom(BA, BC). The collection of such canonical
isomorphisms respects the composition laws of Bcat and S and hence determine
a functor Bcat S. It follows in particular that any lax functor F : Is cat
determines a homotopy coherent diagram
BF

BIs Bcat S
on I.
3.4. The Grothendieck construction.
Homotopy coherent diagrams arising from lax functors are most often realized (or rectified) by using a category theoretic method that is known as the
Grothendieck construction in place of a result like Theorem 3.2. Suppose that
F : I cat is a lax functor. The Grothendieck construction associated to F is a
category LF whose set of objects consists of all pairs (a, x), where a is an object
of I and x is an object of the category F (a). A morphism (, f ) : (a, x) (b, y)
is a pair consisting of a morphism : a b of the base category I and a
morphism (x) y of the category F (b). The composite of (, x) with the
morphism (, g) : (b, y) (c, z) of LF is defined to be the map (, g f ), where
g f is the composite
(,)

(f )

() (x) (x) (y)

z
of F (c). The identity morphism on an element (a, x) of LF is the morphism
(1a , a ). The associativity of the composition operation in LF and the fact that

464

VIII. Simplicial functors and homotopy coherence

the morphisms (1a , x) are two-sided identities are, respectively, consequences of

the cocycle conditions (3.3) and (3.4).
Projection onto the first variable, for both objects and morphisms, defines a
canonical functor F : LF I. The comma category F c associated to an
object c of I has all morphisms : F (a, x) c for objects. There is a functor
fc : F c F (c) which associates the object (c) F (I) to the object
: F (a, x) c. Also, there is a functor gc : F (c) F c which associates
the object 1c : F (c, x) c to x F (c). The functor fc is left adjoint to gc , as
can be seen by observing that the commutative diagram

F (a, x)

(, h)

)
'
'
u
'' 1
c
F (c, y)

is uniquely determined by the map h : (x) y of F (c).

Any morphism : c d induces a functor : F c F d in an obvious
way, and this assignment determines a functor F ? : I cat. The diagram of
functors
F (c)

gc

u
F (c)

w F c
u

gd

w F d

commutes up to canonically determined natural transformation.

Generally, any functor f : D E that has a left or right adjoint induces a
homotopy equivalence f : BD BE of the associated nerves. This is a result
of the fact that any natural transformation of functors gives rise to a homotopy
of the respective induced maps of simplicial sets. The Grothendieck construction
therefore gives rise to homotopy equivalences
'

gc : BF (c) B(F c)

4. Realization theorems

465

such that the diagrams

BF (c)

gc

u
BF (c)

w B(F c)
u

gd

w B(F d)

commute up to homotopy. In other words, the I-diagram c 7 B(F c) is a

realization of the homotopy coherent diagram c 7 BF (c).
4. Realization theorems.
Suppose that : A K(I, 0) is a resolution of a small category I, and let
X : A S be a homotopy coherent diagram on I. A realization of X is a
simplicial set valued functor Y : I S such that there is a pointwise weak
equivalence X ' Y in the simplicial functor category SA . Theorem 3.2 says
that any homotopy coherent diagram X has a realization, and that the weak
equivalence in SA can be chosen canonically.
Given an arbitrary simplicial model category M and a resolution : A
K(I, 0), it is certainly sensible to say that a homotopy coherent diagram on I is
a simplicial functor X : A M. One analogously defines a realization of X to be
a functor Y : I M such that there is a pointwise weak equivalence X ' Y .
The purpose of this section is to show that all homotopy coherent diagrams in
M admit realizations if the simplicial model category M has an adequate notion
of homotopy colimit. This is done by giving a proof of Theorem 3.2 which does
not depend on the Yoneda lemma 1.2. The proof is achieved by thinking about
the functor
in a different way; we lose the functor
and along with it any
notion of an equivalence of homotopy categories associated to the categories SA
and SI . We begin with a weakened version of the Dwyer-Kan theorem.
Theorem 4.1. Suppose that a simplicial functor f : A B is homotopically
full and faithful in the sense that all induced simplicial set maps
f : Hom(A, B) Hom(f (A), f (B))
are weak equivalences, and let X : A S be a simplicial functor. Then X is
naturally pointwise weakly equivalent to the simplicial functor f f X.

466

VIII. Simplicial functors and homotopy coherence

Proof: The assumption on f implies that the map of simplicial objects in SB

defined by
G

X(A0 ) Hom(A0 , A1 ) Hom(An , A)

(A0 ,...,An )

1 f

u
X(A0 ) Hom(A0 , A1 ) Hom(f (An ), f (A))

G
(A0 ,...,An )

is a levelwise weak equivalence, which is natural in A A. The simplicial object

G

X(A0 ) Hom(A0 , A1 ) Hom(An , A)

(A0 ,...,An )

is, in n-simplices, the nerve of the translation category associated to the composite functor
Q

Xm

Am A Am Sets,
and the identity element 1A is terminal in Am A. It follows (see the development around (2.3)) that there is a canonical weak equivalence induced by the
bisimplicial set map
r:

X(A0 ) Hom(A0 , A1 ) Hom(An , A) X(A)

(A0 ,...,An )

which is induced on the translation category level by functors

(x, : A0 A) 7 (x).
We therefore have natural pointwise weak equivalences
'

f f X 1 1 X X.
'

Suppose now that f : A B is a simplicial functor, and that Y : A M

is a simplicial functor taking values in a simplicial model category M. We are

4. Realization theorems

467

entitled to an analogue of the functor f ; in particular, we define f Y (B) to be

the simplicial object having n-simplices
G

Y (A0 ) Hom(A0 , A1 ) Hom(f (An ), B).

(A0 ,...,An )

In this way, we define a functor

f : MA S(MB ),
where S(MB ) denotes the category of simplicial objects in MB . We are also
entitled to the following analogues of the maps appearing in the proof of Theorem 4.1 in the category S(MB ):
(4.2)
G
r
Y (A0 ) Hom(A0 , A1 ) Hom(An , A)
w Y (A)
(A0 ,...,An )

u
Y (A0 ) Hom(A0 , A1 ) Hom(f (An ), f (A)).

(A0 ,...,An )

The trick, either for a given simplicial model category or for a particular class
of objects of the simplicial functor category MA , is to find a realization functor
S(M) M which takes the maps in this diagram to weak equivalences of MB .
The naive realization d(Z) of a simplicial object Z in M is the coend, given
by the coequalizer
G

Z n m

mn

Zn n d(Z).

n0

The object d(Z) has a filtration

d(Z)(n) d(Z)(n+1) . . . ,
where d(Z)(n) is the epimorphic image of the restricted map
G
pn

Zp p d(Z).

468

VIII. Simplicial functors and homotopy coherence

There are pushout diagrams

(s[n] Zn n+1 ) (Zn+1 n+1 )

w d(Z)(n)

u
Zn+1 n+1

u
w d(Z)(n+1) ,

in M, where the map j is canonically induced by the monomorphism j :

s[n] Zn , Zn+1 given by taking the epimorphic image of the map
s:

Zn Zn+1

0in

which applies the ith degeneracy si on the ith summand. The monomorphism
d(Z)(n) , d(Z)(n+1) is a cofibration of M if j : s[n] Zn Zn+1 is a cofibration.
Say that a simplicial object Z in M is diagonally cofibrant if
(1) Z0 is a cofibrant object of M, and
(2) each morphism j : s[n] Zn Zn+1 is a cofibration.
If Z is a diagonally cofibrant simplicial object of M in this sense, then
d(Z)(0) = Z0
is cofibrant and all of the maps d(Z)(n) , d(Z)(n+1) are cofibrations, and so the
realization d(Z) is a cofibrant object of M. It also follows that if f : Z W
is a map of diagonally cofibrant simplicial objects which is a levelwise weak
equivalence in the sense that all of the maps f : Zn Wn are weak equivalences
of M, then the induced map f : d(Z) d(W ) is a weak equivalence as well.
Theorem 4.3. Suppose that f : A B is a simplicial functor which is homotopically full and faithful. Suppose that Y : A M is a simplicial functor
taking values in simplicial model category M, such that Y (A) is a cofibrant object of M for all objects A A. Then the morphisms f and r induce pointwise
weak equivalences
f

'

'

f (df Y )
= d(f f Y ) d(1 1 Y ) d(Y ) = Y.

4. Realization theorems

469

Proof: The assumptions imply that the simplicial object f Y (B) is diagonally
cofibrant, for all B B. This is seen by observing that the object
G
Y (A0 ) Hom(A0 , A1 ) Hom(f (An ), B)
(A0 ,...,An )

can be rewritten in the form

G

0
Y (A0 ) B(f B)A
n ,

A0

where

0
B(f B)A
n = Hom(A0 , A1 ) Hom(f (An ), B)

means strings of length n in the simplicial nerve B(f B) which begin at A0 .

This decomposition is preserved by all degeneracies, and so we have
G
0
X(A0 ) DB(f B)A
s[n] f (B) =
n+1 ,
A0
0
where DB(f B)A
n+1 denotes the degenerate strings of length n + 1 which begin
at A0 . Each map

A0
0
Y (A0 ) DB(f B)A
n+1 Y (A0 ) B(f B)n+1
A0
0
induced by the simplicial set inclusion DB(f B)A
n+1 B(f B)n+1 is a
cofibration, since Y (A0 ) is cofibrant, and the claim is verified.
The maps

Y (A0 ) Hom(A0 , A1 ) Hom(An , A)

(A0 ,...,An )

1 f

u
Y (A0 ) Hom(A0 , A1 ) Hom(f (An ), f (A))

(A0 ,...,An )

are weak equivalences since all Y (A0 ) are cofibrant and f is a weak equivalence
by assumption. It follows that the map
f : d(1
1 Y (A)) d(f f (f (A)))

470

VIII. Simplicial functors and homotopy coherence

is a weak equivalence, since the simplicial objects at issue are diagonally cofibrant.
The retraction map r : 1
1 Y (A) Y (A) has a section s : Y (A) 1 1 Y (A),
along with an associated homotopy
h : 1
1 Y (A) 1 1 1 Y (A)
1 Y (A). Applying the coend functor to h gives a
from sr to the identity on 1
homotopy
d(h) : d(1
1 Y (A)) 1 d(1 1 Y (A)),
from s r to the identity, again since 1 1 Y (A) is diagonally cofibrant. It follows
from Lemma II.4.1 that r is a weak equivalence of M.

All pointed simplicial sets are cofibrant, so Theorem 4.3 immediately implies
Corollary 4.4. Suppose that a simplicial functor f : A B is homotopically
full and faithful, and let X : A S be a simplicial functor taking values in
the category S of pointed simplicial sets. Then X is naturally pointwise weakly
equivalent to the simplicial functor f d(f X).
Spectra, and the Bousfield-Friedlander model for the stable category, are formally discussed in Section IX.6 below. Within that model, it is not true that
all spectra are cofibrant, but homotopy colimits in the category of spectra are
constructed levelwise within the category of pointed simplicial sets, giving
Corollary 4.5. Suppose that a simplicial functor f : A B is homotopically
full and faithful, and let X : A Spt be a simplicial functor taking values
in the category Spt of spectra. Then X is naturally pointwise strictly weakly
equivalent to the simplicial functor f d(f X).
A simplicial functor X : A Spt taking values in the category of spectra can
be identified with a spectrum object in the category SA
of pointed simplicial set
valued simplicial functors. The object X therefore consists of simplicial functors
Xn : A S and pointed transformations S 1 Xn Xn+1 . A map X Y in
SptA is a pointwise strict weak equivalence if all of the maps Xn (A) Yn (A)
are weak equivalences of pointed simplicial sets.
Despite all the huffing and puffing in its proof, Theorem 4.3 is a formal result
which may not always be the best tool. To illustrate, suppose that Y : A sAb
is a simplicial functor taking values in the category of simplicial abelian groups,
and consider the maps of bisimplicial abelian groups corresponding to the diagram (4.2). One uses spectral sequence arguments and the generalized EilenbergZilber theorem (Theorem IV.2.5) to see that, if f : A B is homotopically faithful, then the maps f and r induce weak equivalences of the associated diagonal
simplicial abelian groups, proving

4. Realization theorems

471

Lemma 4.6. Suppose that a simplicial functor f : A B is homotopically full

and faithful, and let Y : A sAb be a simplicial functor taking values in the
category sAb of simplicial abelian groups. Then Y is naturally pointwise weakly
equivalent to the simplicial functor f d(f Y ).
Here, d is the diagonal functor. Note that the simplicial abelian groups Y (A) do
not have to be cofibrant.
The Dold-Kan equivalence (Corollary III.2.3) relating the categories sAb of
simplicial abelian groups and the category Ch+ of ordinary chain complexes,
gives a simplicial model structure to the chain complex category. It also immediately implies the following result:
Corollary 4.7. Suppose that a simplicial functor f : A B is homotopically
full and faithful, and let Y : A Ch+ be a simplicial functor taking values in
the category Ch+ of ordinary chain complexes. Then Y is naturally pointwise
weakly equivalent to the simplicial functor f d(f Y ).
In this case, d(f Y ) is the normalized chain complex N d(f (Y )), where Y
is the diagram of simplicial abelian groups associated to the diagram of chain
complexes Y .

Chapter IX Localization
Localization is more a way of life than any one specific result. For example,
under this rubric one can include Bousfield localization with respect to a homology theory, localization with respect to a map as pioneered by Bousfield,
Dror-Farjoun and elaborated on by many others, and even the formation of the
stable homotopy category. We will touch on all three of these subjects, but we
also have another purpose. There is a body of extremely useful techniques that
we will explore and expand on. These have come to be known as Bousfield factorization, which is a kind of trivial cofibration-fibration factorization necessary
for producing localizations, and the Bousfield-Smith cardinality argument. This
latter technique arises when one is confronted with a situation where a fibration
is defined to be a map which has the right lifting property with respect to some
class of maps. However, for certain arguments one needs to know it is sufficient
to check that the map has the right lifting property with respect to a set of
maps. We explain both Bousfield factorization and the cardinality argument
and explore the implications in a variety of contexts. In particular, we explore
localization in diagram categories, with an eye towards producing a model for the
stable homotopy category, and we produce a simplicial model category structure
on categories of diagrams that will be useful in a later discussion of homotopy
inverse limits.
The concept of localization probably has its roots in the notion of a Serre class
of abelian groups and the Whitehead Theorem mod a Serre class [86, 9.6]. This
result is still useful and prevalent so prevalent, in fact, that it is often used
without reference. The idea of localizing a space with respect to a homology
theory appeared in Sullivans work on the Adams conjecture [85], where there
is an explicit localization of a simply connected space with respect to ordinary
homology with Z[1/p] coefficients. Bousfield and Kan [14] gave the first categorical definition of localization with respect to homology theory and provided a
localization for nilpotent spaces with respect to H (, R), where R = Fp for some
prime p or R a subring of the rationals. Their technique was the R-completion
of space, recapitulated in Section 3 below for the case R = Fp . It was Bousfield
himself who introduced model category theoretic techniques to provide the localization of any space with respect to an arbitrary homology theory. His paper
[8] has been enormously influential, as much for the methods as for the results,
and its hard to overestimate its impact. For example, the concept of localization
with respect to a map and the construction of its existence, which appears in the
work for Dror-Farjoun [22] and [23] is directly influenced by Bousfields ideas.
About the time Dror-Farjouns papers were first circulating, a whole group of
people began to explore these ideas, both in homotopy theory and in related
472

1. Localization with respect to a map

473

algebraic subjects. The paper by Cascuberta [17] is a useful survey. One should
also mention the important paper of Bousfield [11], which uses similar techniques
for its basic constructions. The longest and most general work in this vein, a
work that includes an exposition of the localization model category in an arbitrary cellular model category is that of Hirschhorn [42], available at this writing
over the Internet. The notion of a cellular model category is one way of axiomatizing the structure necessary to make Bousfields arguments work, based on the
concept of cell complexes and inclusions of sub-complexes. We give, in Section 4,
a slightly different list of hypotheses along these lines. Both systems of axioms
work in any example that we know.
We emphasize, however, that Bousfields ideas had influence outside of the
area of homotopy localization. For example, Jeff Smith realized very early on
that one could use these constructions to put a model category structure on the
category of small diagrams of simplicial sets, so that homotopy inverse limits can
be computed as total right derived functors of inverse limit. This never made it
into print, and we go through the arguments in Section 5. Beyond this, there is
the second authors work on the homotopy theory of simplicial presheaves [46],
see also [38] as well as Joyals result for simplicial sheaves [53]. In the context
of the present discussion that work can be interpreted as follows: the category
of presheaves on a Groethendieck site is a category of diagrams and there is a
closed model category structure obtained by localizing with respect to a class of
cofibrations determined by the topology of the underlying site.
1. Localization with respect to a map.
This section is an exposition of a technique due to Bousfield for defining localization with respect to a map in a simplicial model category. We explain the
technique for the category S of simplicial sets. In Section 4 we will examine how
this technique extends to other simplicial model categories.
In the category S of simplicial sets we fix a cofibration f : A B.
Definition 1.1. A space Z S is f -local if Z is fibrant and
f : Hom(B, Z) Hom(A, Z)
is a weak equivalence.
Remarks 1.2.
1) Because f is a cofibration, f is a fibration. Hence we could equally require
that f be a trivial fibration.
2) The hypothesis that f be a cofibration is innocuous. Indeed, if we drop the
hypothesis that f be a cofibration, we have the following observation. Factor f

474

IX. Localization

as

f0

A B0
B
where f0 is a cofibration and q is a trivial fibration. Then f is a weak equivalence
if and only if f0 is a weak equivalence, because q is left inverse to a trivial
cofibration.
3) If F is a set of cofibrations f : A B we could define a space Z to be
F-local if it is fibrant and f is a weak equivalence for all f F. However, Z
would be F-local if and only if Z were f -local where f = tf . Hence we would
achieve no greater generality.
We expand on the notion of f -local:
Definition 1.3. A map q : X Y in S is an f -injective if q is a fibration and
(q , f ) : Hom(B, X) Hom(B, Y ) Hom(A,Y ) Hom(A, X)
is a trivial fibration.
Since this map is a fibration by SM7, we are only requiring it to be weak
equivalence. In light of Remark 1.2.1, Z is f -local if and only if the unique map
Z is an f -injective.
It is convenient to have a recognition principal for f -injectives. First, if j :
C D and q : X Y are maps in S, let us write
D(j, q) = Hom(D, Y ) Hom(C,Y ) Hom(C, X).
Note that D(j, q) is a space of diagrams; indeed an n-simplex is a commutative
diagram
C n
j1

u
D n

wX
q

u
w Y.

If L is a simplicial set, let #L denote the cardinality of the set of non-degenerate

simplices in L.
Lemma 1.4. A morphism q : X Y in S is an f -injective if and only if it is a
fibration and has the right lifting property with respect to all maps
A L AK B K B L
where K L is a cofibration in S with #L finite.

1. Localization with respect to a map

475

Proof: A map in S is a trivial fibration if and only if it has the right lifting
property with respect to all cofibrations K L with #L finite. Now use an
adjunction argument.

Note that one could specialize to the cases

K = n , n = L.
Lemma 1.5. If q : X Y is an f -injective and K S, then
q : Hom(K, X) Hom(K, Y )
is an f -injective.
Proof: Since q is an f -injective
Hom(B, X) D(f, q)
is a trivial fibration. Hence
Hom(K, Hom(B, X)) Hom(K, D(f, q))
is a trivial fibration. But this is isomorphic to
Hom(B, Hom(K, X)) D(f, q ).

From this is follows that the f -local spaces and f -injectives defined above,
with the usual notion of weak equivalence, form a category of fibrant objects
for a homotopy theory. This is direct from the definitions, using the previous
two lemmas. In particular the mapping object is supplied by, for f -local Z,
Hom(1 , Z). This uses Lemma 1.4.
We now let Locf be the resulting homotopy category of f -local spaces obtained
as the full subcategory of Ho(S) with the f -local spaces. Since every object of
S is cofibrant and every f -local space if fibrant, this is the equivalent to the
category of f -local spaces and homotopy classes of maps. We will examine the
inclusion functor
Locf Ho(S)
Definition 1.6. A localization with respect to f is a functor
Lf : Ho(S) Ho(S)
equipped with a natural transformation X : X Lf X so that
1) Lf X Locf and the restricted functor Lf : Ho(S) Locf is left adjoint
to the inclusion; and
2) for all X, the two morphisms Lf X , Lf X : Lf X Lf Lf X are equal, to
an isomorphism.

476

IX. Localization

If it exists, such a localization will be unique up to isomorphism in the homotopy category Ho(S). The existence follows from Bousfields factorization,
embodied in Proposition 1.8 below. Before stating this we need a definition of a
class of cofibrations which behave much like a class of trivial cofibrations.
Definition 1.7. A cofibration j : C D is an f -cofibration if the map
(q , j ) : Hom(D, X)
Hom(D, Y ) Hom(C,Y ) Hom(C, X)
is a trivial fibration for all f -injectives q : X Y . Let Cf be the class of
f -cofibrations.
Proposition 1.8 (Bousfield factorization). Every morphism g : X Y
may be factored
j

X
Ef
Y
where j is an f -cofibration followed by an f -injective. Furthermore, this factorization is natural in the morphism g.
We prove this below after some further preliminaries and proving the existence
of the localization demanded by Definition 1.6.
First some basic properties of f -cofibrations.
Lemma 1.9.
1) The morphism f : A B is an f -cofibration.
2) Any trivial cofibration is an f -cofibration.
3) Suppose i : C D and j : D E are cofibrations and i is an f cofibration. Then if either of j or ji is an f -cofibration so is the other.
Proof: Only part 3) needs comment. We use the following fact: if one has a
diagram
q

X[

wY

[
]
[

u
Z

and a morphism W Z, then there is a pullback square

W Z X

wX
q

u
W Z Y

u
w Y.

1. Localization with respect to a map

477

Hence if q is a trivial fibration so is W Z X W Z Y . Applying this remark

to the diagram
Hom(D, X)

w Hom(D, Y ) Hom(C,Y ) Hom(C, X)

u
Hom(D, Y )

and the map Hom(E, Y ) Hom(D, Y ) we have that if i is an f -cofibration

and q : X Y is an f -injective
Hom(E, Y ) Hom(D,Y ) Hom(D, X) Hom(E, Y ) Hom(C,Y ) Hom(C, X)
is a trivial fibration. Now consider the composite
Hom(E, X)

w Hom(E, Y ) Hom(D,Y ) Hom(D, X)

u
Hom(E, Y ) Hom(C,Y ) Hom(C, X).

If j is an f -cofibration, the first map is a trivial fibration, so the composite is

and ji is an f -cofibration. The converse equally applies to show that if ji is an
f -cofibration so is j.

More constructive properties of f -cofibrations are given in the next result.

Lemma 1.10.
1) Any retract of an f -cofibration is an f -cofibration. Any coproduct of
f -cofibrations is an f -cofibration.
2) Given a pushout diagram with j an f -cofibration
C
j

u
D

w C0
j0

u
w D0 ,

478

IX. Localization
then j 0 is an f -cofibration.
3) If j : C D is an f -cofibration and K L is any cofibration, then
C L CK D K D L
is an f -cofibration.

Proof: Again only part 3) needs comment. First note that if K = , we are
asserting C L DL is an f -cofibration. This requires that for all f -injectives
q:XY
Hom(D L, X) Hom(D L, Y ) Hom(CL,Y ) Hom(C L, X)
be a trivial fibration. But this is isomorphic to

Hom(D, Hom(L, X))

u
Hom(D, Hom(L, Y )) Hom(C,Hom(L,Y )) Hom(C, Hom(L, X))

and Hom(L, X) Hom(L, Y ) is an f -injective by Lemma 1.5.

For the general case consider the diagram

C K

w C L

u
DK

w C L CK D K
j

u
D L.
Part 2) of this lemma says i is a f -cofibration. Since ji is a f -cofibration, the
previous lemma says j is.

Next is the crucial lifting result.

1. Localization with respect to a map

479

Proposition 1.11. Given a lifting problem

C

wX

q
j

u
u
D
wY
where j is a f -cofibration and q is an f -injective, the lift exists and is unique up
to homotopy under C and over Y .
Proof: Consider the trivial fibration
Hom(D, X) Hom(D, Y ) Hom(C,Y ) Hom(C, X) = D(j, q).
A diagram as above is a 0-simplex in the target. Since a trivial fibration is
surjective such a lift exists. The required homotopy is adjoint to a lift in
1

w Hom(D, X)
P
O
O
O
O
u O
u
1

w D(j, q).

The next few paragraphs are devoted to the existence of localization. Define
a functor L : S S and a natural transformation j : X LX, by taking the
Bousfield factorization of X as
j

X
LX

where j is an f -cofibration and q is a f -injective. This is a functor since the

factorization is natural. Then LX is f -local. The previous result implies that j
has the following universal property: given a diagram
X

wZ

u
LX

with Z f -local, then the dotted arrow exists and is unique up to homotopy. If g
is a f -cofibration, g is a homotopy equivalence.
The first result is that L() passes to a functor on the homotopy category.

480

IX. Localization

Lemma 1.12. If : X Y is a weak equivalence, then L : LX LY is a

weak equivalence.
Proof: First of all, L takes trivial cofibrations to homotopy equivalences. In effect, if i : X Y is a trivial cofibration, then i is an f -cofibration by Lemma 1.9,
and so the composites
jX

X LX

and

jY i

X LY

are two factorizations of X as an f -injective following an f -cofibration. It

follows from Proposition 1.11 that Li : LX LY is a homotopy equivalence.
More generally, any weak equivalence f : X Y has a factorization f = p i,
where i is a trivial cofibration and p is left inverse to a trivial cofibration. Thus,
Lp is a weak equivalence, as is the map Lf .

Let Lf : Ho(S) Ho(S) be the functor induced by L, and : X Lf X be

induced in Ho(S) by the map jX : X LX.
Proposition 1.13. The functor Lf : Ho(S) Ho(S) is an f -localization.
Proof: Since LX is f -local, Lf restricts to a functor
Lf : Ho(S) Locf .
By the universal property of LX, Lf is left adjoint to the inclusion.
If j : X Y is an f -cofibration, then any choice of extension Lj : LX
LY is a homotopy equivalence, by an obvious extension of the argument giving
Lemma 1.12. It follow that LjX is a homotopy equivalence, for any X. Also,
the map jX : X LX and the composite
jX

jLX

X LX L2 X
are f -cofibrations taking values in f -local spaces, and the diagram
jLX
w LX
w L2 X
)
'
'
'
''
jX
'
'

u ''
LX
X

jX

commutes, where could be either LjX or jLX . It follows that the maps LjX
and jLX are homotopic.

1. Localization with respect to a map

481

We now turn to the existence of Bousfield factorizations. To start, we must

expand on Lemmas 1.9 and 1.10 and supply one more construction that preserves
f -cofibrations, namely certain types of directed colimits.
Fix an infinite cardinal number and let Seq() denote the well-ordered set
of ordinals less than . Then Seq() is a category with hom(s, t) having one
element of s t and empty otherwise.
Lemma 1.14. Let C S be a diagram of spaces over Seq(). If
1) for each successor ordinal s < , Cs1 Cs is an f -cofibration, and
2) for each limit ordinal t < , lim Cs = Ct ,

then the map

C0 lim Cs

s<

is an f -cofibration.
Proof: If K is a simplicial set, denote by #K the cardinality of the member
of non-degenerate simplices of K. Notice that D E is an f -cofibration if and
only if every lifting problem
D L DK E K
u
EL

wX
B
C
B

BB

u
wY

can be solved, where q is an f -injective and K L is a cofibration in S with

#L finite. The strategy is this: we will use transfinite induction and solve the
successive lifting problems
C0 L C0 K Cs K
(1.15)
u
Cs L

B
BB

wX
B
C
BB

u
wY

in a compatible way, meaning that if s < t, the solution of the lifting problem
for t will restrict to the solution for s. Then taking the colimit over s < will
solve the problem for C0 lim
C .
s< s

482

IX. Localization

So regard the lifting problem for s given in (1.15). If s is a successor ordinal,

we can complete the diagram
C0 L C0 K Cs1 K
u
Cs1 L

w C0 L C0 K Cs K

wX
B
C
B
BB
B
u
u
B
w Cs1 L Cs1 K Cs K
wY

since we have a solution for s 1 and the left square is a pushout. Thus we need
to solve a lifting problem
Cs1 L Cs1 K Cs K
wX
C
B
BB
B
B
u
u
B
Cs L
w Y.
But this is possible since Cs1 Cs is an f -cofibration. Note that the constructed lift satisfies the compatibility requirement spelled out above.
If s is a limit ordinal, we have constructed compatible lifts, t < s,
C0 L C0 K Ct K
A

wX
A
C
AA

u
AA
Ct L

u
w Y.

Taking the colimit over t yields a solution to the lifting problem for C0
lim
C = Cs compatible with all previous lifts.

t<s t
We now come to the factorization result. It turns on the following construction.
Main Construction 1.16.
Let g : X Y be a map in S, and let I be the set of morphisms
A L AK B K B L,

1. Localization with respect to a map

483

where K L runs over representatives for isomorphism classes of cofibrations

of simplicial sets so that #L is finite. Write j : C D for a typical element
in I.
Define a factorization
j1
g1
X E1 X Y
of g by the pushout diagram
G

C D(j , g)

wX

I
j1

u
w E1

D D(j , g)

where
D(j , g) = Hom(C , X) Hom(C ,Y ) Hom(D , Y )
is the space of commutative squares and is induced by evaluation. Evaluation
also defines a map D D(j , g) Y and g1 : E1 Y is induced by the
universal property of pushouts.
Note that since every morphism in I is an f -cofibration, Lemma 1.10 implies
j1 is an f -cofibration. Equally important, note that if one has any diagram
C

wX

u
D

u
wY

where j I, then factors through E1 . This is because such a diagram is a

0-simplex of D(j , g). Finally, note this construction is natural in g.

Proposition 1.17. Bousfield factorization exists.

Proof: This is a simple transfinite variation on the small object argument.
Pick an infinite cardinal > #B, and choose an infinite cardinal > 2 .
Note that #C < for all . Fix a map g : X Y . Using recursion, we
construct a diagram {Es } over Seq(), with E0 = X and so that there is a map
{gs } : {Es } {Y } to the constant diagram on Y .

484

IX. Localization

For s = 0, let g0 X = E0 Y . If s + 1 is a successor ordinal, let Es+1 be

defined by applying the Main Construction 1.16 to gs : Es Y to obtain a
factorization
(gs )1

Es E1 Es Y.
Then Es+1 = E1 Es and gs+1 = (gs )1 . If s is a limit ordinal, let Es = lim
E
t<s t
and gs the induced map. This defines {Es }.
Define Ef = lim
E . Then g : X Y factors as
s< s
j

X
Ef
Y
where j is an f -cofibration by Lemma 1.14. To see that q is an f -injective, it is
sufficient to show that any lifting problem
C

w Ef
] g
j
\\
u \
u
D w Y
with j I can be solved. Since 2#C < ,
hom(C , Ef )
= lim hom(C , Es ),

for otherwise C has too many subobjects. Thus, there is an s < , and a
factorization of
C Es Ef .
Then the main construction implies that factors through Es+1 and the result
follows.

The constructive part of this argument allows us to identify the class of f cofibrations in another way. Let be a cardinal, and recall from Definition II.7.3
that a class M of morphisms in a cocomplete category C is -saturated if
1) coproducts and retracts of morphisms in M are in M
2) given a pushout diagram
C
j

u
D

w C0
j0

u
w D0

1. Localization with respect to a map

485

with j in M, then j 0 is in M.
3) If C : Seq() C is a diagram over , and
a) for each successor ordinal s, Cs1 Cs is in M and
b) for each limit ordinal s, lim
C
= Cs ,
t<s t
then C0 lim
C is in M.
t< t
Corollary 1.18. The class of f -cofibrations in S is the -saturation of the
morphisms
A L AK B K B L
with K L a cofibration in S and #L < , and trivial cofibrations K L
with #L < , and is an infinite cardinal greater than 2 , where is an infinite
cardinal larger than #B.
Proof: Let Cf be this saturation. Then every morphism in Cf is an f -cofibration since the class of f -cofibrations is -saturated. Also if g : X Y is
factored
j
q
X
Z
Y
as in the proof of 1.17, then j Cf . If g is any f -cofibration, then there is a
lifting
X

wZ
q

u
u
Y
wY
since q is an f -injective. This lifting shows g is a retract of j, hence in Cf .

We end this section with a sequence of technical lemmas on the properties of

the functor L(). The purpose here is to have in place the structure necessary to
prove the existence of the model category structure in the next section.
Recall that L(X) = lim
E (X) where Es+1 (X) is obtained from Es =
s< s
Es (X) by a pushout diagram
G

C Hom(C , Es )

w Es (X)

u
D Hom(C , Es )

j J

(1.19)

G
j J

w Es+1 (X)

486

IX. Localization

Of course E0 (X) = X and if s is a limit ordinal, Es (X) = lim

E (X).
t<s t
Lemma 1.20. Let be a cardinal larger than 2 , where is an infinite cardinal
larger than #B. Let X : Seq() S be -diagram. Suppose for each s < ,
Xs Xs+1 is a cofibration, and suppose for each limit ordinal Xs = lim
X.
t<s t
Then the natural map
lim L(Xs ) L(lim Xs )

s<

s<

is an isomorphism.
Proof: It is sufficient, since colimits commute, to show this statement holds
for each of the functors Es (). For the same reason, it is sufficient to show, that
if this statement holds for Et () it holds for Et+1 (). Then it will automatically
hold for limit ordinals. Now, since 2#C < for all , the natural map
lim hom(C n , Et (Xs )) hom(C n , lim Et Xs )

s<

s<

= hom(C n , Et (lim Xs ))

s<

is an isomorphism in each degree n. It follows that the simplicial map

lim Hom(C , Et (Xs )) Hom(C , Et (lim Xs ))

s<

s<

is an isomorphism. Now use the diagram (1.19) and the fact that colimits commute.

For the next result we need a Reedy Lemma. Suppose we have two pushout
squares
A1

w B1

A2

w B2

u
C1

u
w D1

u
C2

u
w D2

and a map from the first square to the second.

Lemma 1.21. If B1 B2 is a cofibration and the induced map C1 A1 A2 C2
is a fibration, then the map D1 D2 is a cofibration.
Proof: The hypotheses are what is required to show D1 D2 has the left
lifting property with respect to all trivial fibrations.

1. Localization with respect to a map

487

Lemma 1.22. The functor L() preserves cofibrations.

Proof: Suppose that X Y is a cofibration, and presume that the induced
map Es X Es Y is a cofibration. We will show that the map Es+1 X Es+1 Y
is a cofibration.
The diagram 1.20 for X maps to the corresponding diagram for Y . The
diagram
G

C Hom(C , Es X)

G
w

D Hom(C , Es X)

u
C Hom(C , Es Y )
w

u
D Hom(C , Es Y )

is a disjoint union of diagrams of the form

C K

j 1

1i

u
C L

j 1

w D K
1i
u
w D L

which are induced by cofibrations j : C D and i : K L. Each induced

map
C L C K D K D L
is plainly a cofibration. Now use Lemma 1.21.

The next result is a variation on Lemma 1.20.

Lemma 1.23. Let be a cardinal number with > . For X S, let {Xj } be
the filtered system of sub-complexes Xj X with #Xj . Then the natural
map
lim L(Xj ) L(X)

is an isomorphism.

488

IX. Localization

Proof: The proof is the same as that for Lemma 1.20, once one notices that
lim Hom(C , Xj ) Hom(C , X)

is an isomorphism on account of the size of C .

Next, we have a counting argument.

Lemma 1.24. Let be any cardinal such that . If X S has #X 2 ,

then #L(X) 2 .
Proof: We show that for each ordinal s < , #Es (X) 2 . This is done by
induction: it is true for E0 (X) = X; if s is a limit ordinal and it is true for
Et (X), t < s, then it is true for Es (X). Finally, if it is true for s it is true for
s + 1 by the diagram (1.19) and because
#(D Hom(C , Es (X)) (2 ) = 2 = 2 .

Because it is a colimit, one would not expect the functor L to preserve inverse
limits, we do have the following result. Recall that L preserves cofibrations and
the cofibrations are inclusions.
Lemma 1.25. Let X be a simplicial set and C, D X two sub-simplicial sets.
Then L(C D) = L(C) L(D).
Proof: First suppose that we can show that for every ordinal s there is an
equality Es+1 (C D) = Es+1 (C) Es+1 (D). Then, because Es () Es+1 (),
an inductive argument shows that we have the following equalities for any limit
ordinal s
[
[
[
[
Es (C D) =
[Et (C D)] =
[Et (C) Et (D)] = [ Et (C)] [ Et (D)].
t<s

t<s

t<s

t<s

The last of these equalities uses that we have a nested sequence of inclusions,
and the last listed object is Es (C) Es (D).
Now suppose that we can show E1 (C D) = E1 (C) E1 (D). Then another
inductive argument implies
Es+1 (C D) = E1 (Es (C D)) = E1 (Es (C) Es (D)) = Es+1 (C) Es+1 (D).
Finally, to see that E1 (C D) = E1 (C) E1 (D), note that E1 Xn has the form
G
G
E1 Xn = (
(D C )n )
Xn

in each simplicial degree n.

2. The closed model category structure

489

2. The closed model category structure.

We have presented Bousfield factorization as if it were a trivial cofibrationfibration factorization for some appropriate closed model category structure on
simplicial sets. This is the case, but it requires some care to make this statement
precise.
Fix a cofibration f : A B and note that the previous section gives a functor
L : S S equipped with a natural f -cofibration
j : X LX.
The functor L preserves weak equivalences and induces the localization function
Lf on the homotopy category. We now define a morphism g : C D in S to be
an f -local equivalence if the induced map
g : Hom(D, X) Hom(C, X)
is a weak equivalence for all f -local spaces X.
Lemma 2.1.
1) Any f -cofibration is an f -local equivalence.
2) A morphism g : C D is an f -local equivalence if and only if Lg : LC
LD is a weak equivalence.
Proof: Part 1) is a consequence of Definition 1.7 with q the unique morphism
X . For part 2) consider the induced diagram, with X f -local
Hom(LD, X)

Lg

u
Hom(LC, X)

w Hom(D, X)
g

u
w Hom(C, X).

By part 1) both maps labeled j are weak equivalences. If Lg is a weak equivalence, then Lg and, hence, g are weak equivalences, so g is a f -local equivalence.
Conversely, if g is a weak equivalence, so is Lg and
Lg : [LD, X] [LC, X]
is a bijection for all f -local X. This implies Lg is a homotopy equivalence. To see
this, set X = LC, then there is a map h : LD LC so that Lg h = hLg ' 1LC .
Then set X = LD and compute
Lg (Lg h) = Lg h Lg ' Lg = 1LD Lg = Lg (1LD )
so Lg h ' 1LD .

490

IX. Localization

Definition 2.2. We define a morphism g : C D to be

1) an f -local cofibration if it is a cofibration; and
2) an f -local fibration if it has the right lifting property with respect to all
cofibrations which are also f -local equivalences.
The main result of this section is:
Theorem 2.3. With its usual simplicial structure and the above definitions of
f -local equivalence, cofibrations, and f -local fibration, the category S becomes
a simplicial model category.
Remark 2.4. If every f -injective were an f -local fibration, this would be easy.
However, this need not be the case as the following example shows. Let f :
n+1 n+1 be the inclusion and consider the fibration q : W K(Z, n 1)
K(Z, n). Note that q is an f -injective; that is,
Hom(n+1 , W K(Z, n 1)) D(f, q)
is a trivial fibration. This is equivalent to the assertion that D(f, q) is contractible. For this, there is a pullback diagram
w Hom(n+1 , K(Z, n))

D(f, q)
Hom(

n+1

u
, W K(Z, n 1))

u
w Hom(

n+1

, K(Z, n))

so that the fibre of the induced fibration D(f, q) Hom(n , W K(Z, n 1))
is the pointed function complex
hom (n+1 /n+1 , K(Z, n)) ' n+1 K(Z, n) ' .
Now let n+1 be any vertex. Then n+1 is a trivial cofibration and
n+1 n+1 is an f -local equivalence (by Lemma 2.1.1 and Lemma 1.9.1) so
n+1 is an f -local equivalence. But if i : n+1 K(Z, n) is non-trivial
in homotopy, there is no solution to the lifting problem

w W K(Z, n 1)
]
q
\\
u
u
\
n+1
w K(Z, n),
i
so q is an f -injective which is not an f -local fibration. We will have more to say
about f -local fibrations after the proof of Theorem 2.3.

2. The closed model category structure

491

We start the proof of Theorem 2.3 with the following lemmas.

Lemma 2.5. A morphism in S is at once an f -local fibration and an f -local
equivalence if and only if it is a trivial fibration.
Proof: If q : X Y is a trivial fibration it has the right lifting property with
respect to all cofibrations. Furthermore Lq is a weak equivalence by Lemma 1.12.
So Lemma 2.1 implies q is a f -local fibration and an f -local equivalence.
For the converse, fix q : X Y which is an f -local fibration and an f -local
equivalence. Factor q as
i

X
Z
Y,
where i is a cofibration and p is a trivial fibration. Then Lemmas 1.12 and 2.1
imply i is an f -local equivalence, so there is a solution to the lifting problem

wX
\
]
q
\
i
u \
u
Z p w Y.
Thus q is a retract of p and a trivial fibration.

Lemma 2.6. Any morphism g : X Y in S can be factored as

j

X
Z
Y
where q is an f -local fibration and j is at once a cofibration and an f -local
equivalence.
This is the heart of the matter and will be proved below. We also record:
Lemma 2.7.
1) The class of morphisms which are at once cofibrations and f -local equivalences is closed under pushouts and colimits over ordinal numbers.
2) Let j : C D be at once a cofibration and an f -local equivalence, and
let K L be any cofibration, then
D K CK C L D L
is a cofibration and an f -local equivalence.

492

IX. Localization

Proof: Observe that j : C D is a cofibration and an f -local weak equivalence

if and only if the induced map
j : Hom(D, X) Hom(C, X)
is a trivial fibration for all f -local spaces X. Part 1) follows immediately.
For part 2), it suffices to show that the map j 1 : C K D K is a
cofibration and an f -local weak equivalence if the same is true for j : C D.
But the map
(j1)

Hom(D K, X) Hom(C K, X)
is isomorphic to
j

Hom(K, Hom(D, X)) Hom(K, Hom(C, X)),

which is a trivial fibration if X is f -local and j is a cofibration and f -local weak
equivalence.

Proof of Theorem 2.3: The trivial cofibration-fibration factorization is

Lemma 2.6. The other part of CM5 follows from Lemma 2.5 and the axioms
for S in its usual closed model category structure. The remaining closed model
axioms are automatic. Finally, SM7 follows from Lemma 2.7.2.

We now must prove Lemma 2.6. This argument that we shall give is a sequence
of ideas originally due to Bousfield, and then later modified by by J. Smith and
Hirschhorn. The argument that we shall give here is the iteration presented in
[38]. The central device is the following:
Lemma 2.8. Suppose that is an infinite cardinal, and suppose given a diagram
of simplicial set maps

Ay

X
y
j
u
wY

where #A and the inclusion j : X , Y is a weak equivalence. Then

there is a subcomplex B X containing A such that #B and the inclusion
jB : B Y , B is a weak equivalence.
In the language of [38], Lemma 2.8 is called the bounded cofibration property for simplicial set cofibrations. To effect the proof, we need the following
(fundamental) method of converting a map to a fibration:

2. The closed model category structure

493

Lemma 2.9. Suppose that f : X Y is a map of simplicial sets. Then there is

a diagram
f

Xh

h
j
f h

wY
'
)
'f
'

Zf
such that is a fibration and is a weak equivalence. Furthermore, this factorization preserves filtered colimits in f .
Proof: Let f : X Y be any simplicial set map, and use Kans Ex construction (see III.4) to form the diagram

X
(2.10)

w Ex X

u
Y

u
w Ex Y

Note that the Ex construction commutes with filtered colimits. The next step
is to use the dual of Lemma II.9.4 to convert the map f : Ex X Ex Y into
a fibration according to the standard classical method (which works because g is
a map between Kan complexes). In the current context, one forms the diagram

Ex X Ex Y Hom(1 , Ex Y )

w Hom(1 , Ex Y )

prL
u

w Ex Y

d0

Ex X

d1

w Ex Y

Write = d1 f, and observe that prL has a section which is induced by

the map 1 0 . Then = f , the map is a fibration, and is a
section of the trivial fibration prL and is therefore a weak equivalence. Write

494

IX. Localization

Zf = Ex X Ex Y Hom(1 , Ex Y ). Finally, one forms the pullback diagram

Y Ex Y Zf
(2.11)

w Zf

u
Y

u
Y

w Ex Y

Write Zf = Y Ex Y Zf . The factorization = f determines a map from

diagram (2.10) to diagram (2.11), and hence there is an induced map f : X
Zf such that f = f f . Furthermore, is a fibration so that f is a fibration
and the map is a weak equivalence by properness of the closed model structure
for the simplicial set category S (Corollary II.9.6). It follows that the map f is
a weak equivalence.
Note that all constructions here are natural in f ane either involve finite limits
or filtered colimits, all of which commute with filtered colimits in f .

Proof of Lemma 2.8: Say that a subcomplex B of Y is -bounded if #B

. In the language of Lemma 2.9, we must show that there is a -bounded
subcomplex B Y containing A such that the induced fibration jB is trivial.
Note further that ZjB is -bounded if B is -bounded.
Write A = B0 , and consider all diagrams of the form

n
y
(2.12)

u ((
n

w ZjB0
)
((
((
(
((
w B0

w Zj
j
u
w Y.

Observe that the lifting always exists, because j is a trivial fibration by

assumption. The simplicial set Y is a filtered colimit of its -bounded subcomplexes B so that Zj = lim ZjB , and there are at most such (solid arrow)
B
diagrams. It follows that there is a -bounded subcomplex B1 of Y such that
all such solid arrow lifting problems have solutions in ZjB1 . Repeat this construction countably many times to form a sequence of -bounded subcomplexes
A = B0 B1 B2 . . . , so that all lifting problems of the form (2.12) over Bi

2. The closed model category structure

495

are solved over Bi+1 . Then B = i0 Bi is -bounded, and each lifting problem
n
y

w Zj B
jB

u
n

u
wB

factors through a corresponding lifting problem over some Bi and is therefore

solved over Bi+1 .

Lemma 2.13. Let = 2 , where > 2 and > #B are the choices of cardinals
appearing in Proposition 1.17 Let j : X Y be at once a cofibration and an
f -trivial fibration, and suppose that A is a -bounded subcomplex of Y . Then
there is a -bounded sub-complex B Y so that A B and B X , B is an
f -local equivalence.
Proof: We inductively define a chain of -bounded sub-objects A = A0 A1
A2 Y over , and a chain of sub-objects
L(A) = L(A0 ) X1 L(A1 ) X2 L(A2 ) L(Y ),
also over , with the property that
L(X) Xs Xs
is a weak equivalence. Then we set B = lim
A and, by Lemmas 1.20 and
s< s
1.25,
L(X B) = L(X) L(B) = lim L(X) Xs

s<

lim Xs
= L(B)

s<

is a weak equivalence as required.

The objects As and Xs are defined recursively. Suppose s + 1 is a successor
ordinal and As has been defined. Then, since As is -bounded, LAs is -bounded
by Lemma 1.24. Then Lemma 2.8 implies that there is a -bounded sub-object
Xs+1 L(Y ) so that L(As ) Xs+1 and L(X) Xs+1 Xs+1 is a weak
equivalence. Since L(Y ) = lim L(Yj ) where Yj Y runs over the -bounded
j

496

IX. Localization

sub-objects of Y (Lemma 1.23), there is a -bounded sub-object A0s+1 so that

Xs+1 L(A0s+1 ). Let As+1 = As A0s+1 . Finally, suppose s is a limit ordinal.
Then set Xs = lim
L(At )
X . The object Xs is -bounded and
= lim
t<s
t<s t
L(X) Xs Xs is a weak equivalence. Choose A0s Y so that A0s is -bounded
and Xs L(A0s ) and set As = lim
A A0s .

t<s t
Lemma 2.14. Suppose that is the cardinal chosen in Lemma 2.13. A map
q : X Y in S is an f -local fibration if and only if it has the right lifting
property with respect of all morphisms j : C D which are at once cofibrations
and f -local equivalences and so that #D .
Proof: The only if implication is clear. For the reverse implication we use a
Zorns Lemma argument. Consider an arbitrary lifting problem

wX
\
]
\ q
u \
u
D
wY

where j is any cofibration which is also an f -local equivalence. We must complete

the dotted arrow. Define to be the set of pairs (C 0 , g) where C C 0 D and
C C 0 is a f -local equivalence, and g solves the lifting problem

wX

]
\
g\

u \
C0

u
w Y.

Define (C 0 , g) < (C 00 , h) if C 0 C 00 and h|C 0 = g. Then satisfies the hypotheses of Zorns Lemma and, thus, has a maximal element (C0 , g0 ). We show
C0 = D. Consider the new lifting problem

C0

g0

wX
\ q
]
\
u \
u
w Y.
D

2. The closed model category structure

497

If C0 6= D, choose x D so that x 6 C0 . By the previous lemma, there is a

-bounded subobject D0 D so that x D0 and C0 D0 D0 is an f -local
equivalence. By hypothesis, the restricted lifting problem
g0

C0

wX
i
j
i q
u
u
i
C0 D0
wY
has a solution. Lemma 2.7.2 implies C0 C0 D0 is an f -local equivalence, so
we have a contradiction to the maximality of (C0 , g0 ). Thus C0 = D and the
proof is complete.

The proof of Lemma 2.6:

We use a Bousfield factorization. Let J be a set of maps containing one
representative for each isomorphism class of cofibrations j : C D which are
f -local equivalences and so that #D . Define
G
G
G

D=D
C
j : C =
=
J

and factor g : X Y as
i

X
E
Y
where i is the -cofibration of Proposition 1.17 and q is a -injective.
The class of -cofibrations is the saturation of the class of all cofibrations
K D
L
D
C L
CK

induced by and all inclusions K , L of finite simplicial sets, by Corollary 1.18.

Each such map is an f -local equivalence, by Lemma 2.6.2, as well as a cofibration. The class of morphisms which are cofibrations and f -local equivalences is
saturated, by Lemma 2.6.1. It follows that all -cofibrations are f -local equivalences. The definition of -injective (see 1.3) implies q has the right lifting
property with respect to and, hence, with respect to all j : C D of J.
Hence the result follows from Lemma 2.14.

One can also make an argument for Lemma 2.6 directly from Lemma 2.14, by
means of a transfinite small object argument.
The reader who has gotten this far will have noticed that we used Bousfield factorization twice: once to produce the functor L() and once to prove
Lemma 2.6. Now, local objects are produced using L(X) and fibrant objects by
using Lemma 2.6 to factor X as a trivial cofibration followed by a fibration;
therefore, they might be different. However, we have:

498

IX. Localization

Proposition 2.15. A space X S is f -local if and only if it is fibrant in the

f -local model category structure.
Proof: Suppose that j : C D is a cofibration and an f -local equivalence. in
particular that the induced map
j : Hom(D, X) Hom(C, X)
is a trivial fibration for all f -local spaces X. The map j is surjective in degree
0, so that the lifting problem
C

wX
j
ii

u i
D

u
w

can always be solved. Every f -local space is therefore fibrant in the f -local closed
model structure.
Now suppose that X is fibrant. The map f : A B is an f -local cofibration
by Lemma 1.9, and is therefore an f -local equivalence by Lemma 2.1 as well as
a cofibration. Each induced map
A n An B n B n
is an f -local equivalence and a cofibration, by Lemma 2.6, so that the map
f : Hom(B, X) Hom(A, X)
is a trivial Kan fibration. In particular, the space X is f -local.

3. Bousfield localization.
Let S be the category of simplicial sets and E a generalized homology theory
which satisfies the limit axiom in the sense that it preserves filtered colimits. A
space Z S is E -local if and only if Z is fibrant, and any diagram
X
g

u
Y

wZ

3. Bousfield localization

499

with E g and isomorphism can be completed uniquely up to homotopy. The

E -localization of a space X is a map : X Z with Z E -local and E an
isomorphism. One easily checks that if such a localization exists, it is unique up
to homotopy. The existence hinges on the existence of an appropriate Bousfield
factorization and, hence, is best described under the rubric of the previous two
sections.
Definiton 3.1. A morphism g : X Y in S is said to be
1) an E -equivalence if E g is an isomorphism,
2) an E -cofibration if it is a cofibration in S, and
3) an E -fibration if it has the right lifting property with respect to all E
trivial fibrations.
Then one has
Theorem 3.2. With these definitions and its usual simplicial structure, S becomes a simplicial model category.
This will be proved below.
Corollary 3.3.
1) A space Z S is E -local if and only if it is E -fibrant.
2) Every space X S has an E -localization.
Proof: For 1) first assume Z is E fibrant and factor any E -equivalence X
Y as
q
i
X
W
Y
where q is a trivial fibration and i is a cofibration. Then i will be an E -trivial
cofibration, so Hom(W, Z) Hom(X, Z) is a weak equivalence. The trivial
fibration q has a section : Y W which is a trivial cofibration cofibration, so
Hom(Y, Z) Hom(W, Z) is a weak equivalence. It follows that Z is E -local.
Conversely, if Z is E -local and X Y is an E -trivial cofibration, then we
need to solve any lifting problem

X
u
Y

wZ

500

IX. Localization

Since Z is E local it can be solved up to homotopy. Since Z is fibrant, the original lifting problem can be solved by appropriate use of the homotopy extension
property.
For 2), simply factor the unique map X as
j

X
Z

where j is a trivial E -cofibration and q is an E -fibration. (Compare the proof

of Proposition 1.13.)

Corollary 3.4. Let j : X Z be an E -localization of X. Then the map j is

up to homotopy, both the terminal E -equivalence out of X and the initial map
to an E -local space.
Proof: Both clauses follow from the definition of localization.

The difficulty in proving Theorem 3.3 arises in verifying the trivial cofibration-fibration factoring axiom. This is done by using Bousfield factorization
for a particular map f : A B in S. The next three lemmas construct and
identify the map f . Choose an infinite cardinal greater
F than the cardinality of
E that is, greater than the cardinality of the set n En .
Lemma 3.5. Let g : C D be a cofibration and an E equivalence. Let x D
be a simplex. Then there is a sub-simplicial set D0 D so that x D, #B ,
and
C D0 D0
is an E equivalence.
Proof: We use that for any simplicial set Z, the natural map
lim E Z E Z

is an isomorphism, where Z Z runs over the sub-simplicial sets with finitely

many non-degenerate simplices.
We recursively define a sequence (over the natural numbers) of sub-simplicial
sets
K0 K1 K2 D
with the properties that x K0 , #Ks , and
e (Kn /Kn C) E
e (Kn+1 /Kn+1 C)
E

3. Bousfield localization

501

S
is the zero map. Then we can set D0 = Kn .
For K0 , choose any sub-simplicial set with #K0 finite and x K0 . Having
e (Kn /Kn C) there is a
defined Kn , produce Kn+1 as follows. For each y E
e ((Kn /Kn C) Zy ).
finite sub-complex Zy D/X so that y maps to zero in E
Choose a finite sub-complex Yy Y that maps onto Zy and let Kn+1 = Kn
S

y Yy .
Lemma 3.6. A morphism q : Z W in S is an E fibration if and only if it has
the right lifting property with respect to all E -trivial cofibrations X Y with
#Y .
Proof: Use the same Zorns lemma argument as for Lemma 2.14.

The proof of Theorem 3.3: As in Lemma 2.5, a morphism is at once an

E fibration and an E -equivalence if and only if it is a trivial fibration. Then
all but the trivial cofibration-fibration factorization axiom follow immediately.
For the final factorizations, let J be a set of E -trivial cofibrations j : C D
containing one representative of each isomorphism class of E trivial cofibrations
j : C D with #D . Let
G
G
G
f=
j :
C
D
with the coproduct over J. Fix a morphism X Y in S and let
X Ef Y
be the Bousfield Factorization with respect to f . Then the Mayer-Vietoris sequence and the Main Construction 1.16 implies X Ef is an E trivial cofibration and, using Lemma 3.6 and arguing as in Proposition 1.17, one sees Ef Y
is an E -fibration.
To prove SM7 we know it is sufficient to show that if C D is an E trivial
cofibration and K L is any cofibration. Then
D K CK C L D L
is a trivial E -cofibration. Now use the Mayer-Vietoris sequence.

We give a brief review of the homotopical underpinnings of the Bousfield-Kan

p-completion of spaces. The main source is of course [14], but many of the ideas
here are also examined in [37].
Let tow be the category of towers X = {Xn } of simplicial sets with morphisms
the pro-maps
hom(X, Y ) = lim lim homS (Xk , Yn ).

n

502

IX. Localization

Thus a morphism X Y is an equivalence class of tower maps, meaning a

commutative ladder of the form

w Xks+1

w Xks

w Xk2

w Xk1

w Yns+1

w Yns

u
w Yn2

w Yn1

where the horizontal maps are induced from the tower projections of X and Y
and lim ks = = lim ns .

There is a simplicial model category structure on tow where a pro-map f :

X Y is a weak equivalence if the induced map
f : lim H (Xn , Fp ) lim H (Yn , Fp )

is an isomorphism, and f : X Y is a cofibration if lim Xn lim Yn is an

injection of simplicial sets. If X S is a space, the Bousfield-Kan p-completion

of X is defined as follows: regard X tow as a constant tower, choose a weak
equivalence X Y with Y fibrant and set
Xp = lim Yn .

Note that Y is actually a tower of fibrations. There is a map : X Xp called

the completion map, and X is p-complete if this map is a weak equivalence.
The class of p-complete spaces is closed under homotopy inverse limits and any
p-complete space is H (, Fp ) local. Note also that if X Y is an H (, Fp )
isomorphism, then Xp ' Yp .
If X Fp X is the Bousfield-Kan resolution of X (see Example VII.4.1) then
the total tower of this cosimplicial space {Totn Fp X} (see Section VII.5) is a
fibrant model for X in tow by [21], hence
Xp ' Tot(Fp X).
This is the original definition of the completion. See also Example VII.7.18.
To analyze the homotopy type of Xp we introduce and discuss the p-completion
functor on abelian groups. If A is an abelian group define
Ap = lim(Z/pn Z A).

3. Bousfield localization

503

A group is p-complete if the map A Ap is an isomorphism. We write Abp

for the category of p-complete groups. Then completion is left adjoint to the
inclusion functor Abp Ab. Note that the functor A 7 Ap is neither left nor
right exact, which implies that Abp is not an abelian sub-category. Nonetheless,
completion has left derived functors Ls ()p and one has that Ls (A)p = 0 for
s > 1, a short exact sequence
0 lim 1 Tor(Z/pn Z, A) L0 (A)p Ap 0,

and a natural isomorphism

L1 (A)p
= lim Tor(Z/pn Z, A).

See [39]. For example, if Z/p Z = lim Z/pn Z, then L0 (Z/p Z)p = 0 and

L1 (Z/p Z)p
= Zp , the p-adic numbers. Since
Tor(Z/pn Z, A)
= Hom(Z/pn Z, A)
one has
L1 (A)p
= lim Hom(Z/pn Z, A)
= Hom(Z/p Z, A)

from which is follows that

L0 (A)p
= Ext(Z/p Z, A).
The smallest abelian sub-category of Ab containing the p-complete groups is the
category of groups so that
Hom(Z[1/p], A) = 0 = Ext(Z[1/p], A).
These are known variously as Ext p complete groups , p-cotorsion groups , or
weakly p-complete groups. See Section 4 of [44].
The functors of p-completion and its derived functors can be extended to the
class of nilpotent groups, which are those groups from which the lower central
series in eventually zero. See [14].
Now we consider nilpotent spaces X. These are connected simplicial sets X
for which 1 X is a nilpotent group and for which 1 X acts nilpotently on n for
all n. See Section VI.6 for complete definitions. The main result is the following.

504

IX. Localization

Theorem 3.7. Let X be a nilpotent space. Then

1) H (X, Fp ) H (Xp , Fp ) is an isomorphism and Xp is the H (, Fp ) localization of X;
2) for all n 1 there is short exact sequence
0 L0 (n X)p n Xp L1 (n1 X)p 0.
This sequence is split, but not naturally.
Here is an outline of the proof. First one shows the result for K(A, n) where
A is an abelian group and n 1. If
0 F1 F0 A 0
is a free resolution of A, let K be the simplicial abelian group with normalized
chain complex
0 F1 F0 0
with F0 in degree n. Then K is weakly equivalent to K(A, n) and the simplicial
abelian group Kp is a model for K(A, n)p .
Next is X is nilpotent, we have, by the results of Section VI.6, a refined
Postnikov tower for X; in particular, X may be written as X ' lim Xk where
k
the Xk fit into a tower and each successive stage is built by a pullback diagram

Xk

w W K(A, nk )

u
w K(A, nk + 1),

Xk1

and lim nk = . Thus one can proceed inductively using the nilpotent fiber
k
lemma of Bousfield and Kan [14]. The necessary corollary of that result which
is needed here is the following.
Proposition 3.8. Let K(A, n) E B be a principal fibration, and suppose
n 1. Then
K(A, n)p Ep Bp
is a fibration sequence up to homotopy.

4. Localization in simplicial model categories

505

This can be proved by arguing directly that it holds for

K(A, n) W K(A, n) K(A, n + 1)
and then using the Serre spectral sequence. Specifically, if we define E 0 by the
pullback diagram
E0

w W K(A, n)p

u
Bp

u
w K(A, n)p ,

there is a map E E 0 which is an H (, Fp ) isomorphism by the Serre spectral

sequence. Since E 0 is p-complete the class of p-complete spaces is closed under
homotopy inverse limits we have Ep ' E 0 and the result follows.
4. Localization in simplicial model categories.
In this section we explore localization in a more general context; namely, we
examine localization with respect to a cofibration between cofibrant objects in
a simplicial model category. It is possible to provide axioms so that localization
exists, and to provide axioms to say when the model category structure can
be localized, as we did with simplicial sets in Section 2. All of these axioms
are specific and concrete except one, and that is the one which, in the case of
simplicial sets, required the Bousfield-Smith cardinality argument. In the next
section we give a lengthy example we apply the technology of this section to
diagram categories.
Fix a simplicial model category C and let f : A B be a cofibration between
cofibrant objects. The following exactly mimics Definition 1.3.
Definition 4.1. A map q : X Y in C is an f -injective if q is a fibration and
Hom(B, X) Hom(B, Y ) Hom(A,Y ) Hom(A, X)
is a trivial fibration. An object X C is f -local if the unique map X to the
terminal object is an f -injective.
Corresponding to f -injectives we have f -cofibrations, as in Definition 1.7.
Definition 4.2. A cofibration j : C D is an f -cofibration if the induced
fibration
Hom(D, X) Hom(C, X) Hom(C,Y ) Hom(D, Y )
is a weak equivalence for all f -injectives X Y . Let Cf be the class of f cofibrations.

506

IX. Localization

We now work through some of the results of Section 1 to see what must be
required of C in order that localization exist.
Lemma 4.3.
1) A morphism q : X Y in C is an f -injective if and only if it is a fibration
and has the right lifting property with respect to all maps
AL

G
AK

BK BL

where K L is a cofibration in S and #L is finite.

2) If q : X Y is an f -injective and K S, then
homC (K, X) homC (K, Y )
is an f -injective.
The proofs are the same and this gives the flavor of how one must modify the
statements of the previous section. Then Lemmas 1.9, 1.10, and 1.14 go through
with the obvious changes.
The existence of Bousfield factorization, however, requires a smallness assumption. We codify this in the following definition, which is a modification of
Bousfields notion of s-definite .
Definition 4.4. Let be a cardinal number. A -cofibration sequence in C is
a diagram C C over Seq() with the properties that
1) for each successor ordinal s < , Cs1 Cs is a cofibration
2) for each limit ordinal s < , lim
C
= Cs .
t<s t
Remark 4.5. A cofibrant -sequence in the sense of Section II.9 is a -cofibration sequence C for which every object Cs is cofibrant.
Definition 4.6. An object A C is cofibration definite if there is a cardinal
such that for every cardinal and every -cofibration sequence C , the
natural map
(4.7)

lim homC (A, Cs ) homC (A, lim Cs )

s<

is a bijection.

s<

4. Localization in simplicial model categories

507

Examples 4.8. In the following simplicial model categories, every object is

cofibration-definite:
1) Simplicial sets S and pointed simplicial sets S .
2) Simplicial groups, simplicial abelian groups, simplicial algebras, simplicial lie
algebras, and so on. The same applies to the graded analogs.
3) Compactly generated Hausdorff spaces. Notice that the map (4.7) is not, in
this category, an isomorphism forS
an arbitrary colimit. However, cofibrations are

closed inclusions so lim

C = s Cs . Thus we need only take greater than
s< s
the cardinality 2#A of the power set of A.
4) The category S I of I diagrams in simplicial sets S. Here I is small category.
Proposition 4.9 (Bousfield factorization). Let C be a simplicial model
category satisfying the additional hypotheses that
1) every object of C is cofibration definite; and
2) there is a set F of trivial cofibrations so that a morphism in C is a fibration
if and only if it has the right lifting property with respect to all morphisms
in F.
Then any morphism g : X Y in C can be factored naturally as
j

X
Z
Y.
where j is an f -cofibration and q is an f -injective.
The proof follows the Main Construction 1.16 and Proposition 1.17. To choose
, notice that for all L S with #L finite and C D in F, the functors
homC (B L, ) and homC (D, ) will commute with -cofibration sequences for
some large enough . Note than in all the examples given in 4.8, the closed
model category structure is cofibrantly generated (see Definition II.6.6) which
implies 4.9.2.
In the presence of Bousfield factorization, one also has localization, provided
one takes a little care. Let Locf Ho(C) be the full sub-category with objects the f -local objects. An f -localization will be an idempotent functor Lf :
Ho(C) Ho(C) that restricts to a functor Lf : Ho(C) Locf left adjoint to
the inclusion. If it exists it is unique.
Proposition 4.10. If Bousfield factorization holds in C, f -localization exists.
Proof: If every object in C is cofibrant so that the notion of homotopy makes
sense, we can proceed as in 1.111.13. If not, we must modify the procedure
somewhat.

508

IX. Localization

If X C, choose a trivial fibration Y X with Y cofibrant. Y is unique up

to unique homotopy equivalence over X. Define LY by factoring Y as
j

Y
LY

where j is an f -cofibration and q is an f -fibration. Then Lf X = LY . The reader

is invited to modify 1.111.13 to complete the proof.
Notice that if C has natural factorizations (see Section II.5), then LY can be
chosen naturally in X.

We now turn to the model category structure. As in the previous proof we

have a functor L : C C and a natural f -cofibration X LX so that LX is
f -local although LX is not a model for the f -localization of X unless X is
cofibrant. But see Lemma 4.18 below.
Lemma 4.11. Suppose that for all cofibrant C C and all cofibrations U V
of C, the function homC (C, U ) homC (D, V ) is an injection. Then the functor
L : C C preserves cofibrations.
Proof: When recapitulating the proof of Lemma 1.22, one finds that one needs
to know that for all j : C D in J, the map of simplicial sets
HomC (C , Es X) HomC (D , Es Y )
is a cofibration; that is, we need it to be an inclusion. This is equivalent to the
assertion that for all n,
homC (C n , Es X) homC (C n , Es Y )
is an injection. Since, by induction, Es X Es Y is a cofibration and C n
is cofibrant, this follows from the hypothesis.

Remark 4.12. The hypothesis of Lemma 4.11 is relatively innocuous. It is

satisfied, for example, by all the simplicial model categories listed in Example 4.8.
Definition 4.13. A morphism C D between cofibrant objects is an f -local
equivalence if HomC (D, X) HomC (C, X) is a weak equivalence for all f -local
objects X of C. More generally, a morphism g : C D is an f -local equivalence
if there is a diagram
C0
(4.14)

g0

u
C

w C0
q

u
wD

where p and q are weak equivalences, C0 and D0 are cofibrant, and g0 : C0 D0

is a f -local equivalence.

4. Localization in simplicial model categories

509

A diagram such as (4.14) will be called a cofibrant approximation to g.

Remark 4.15. If g0 : C0 D0 is an f -local equivalence for some cofibrant
approximation to g, it is for all. To see this, choose a cofibrant approximation
C1
u
C

g1

w D1
u
wD

with D1 D a trivial fibration and C1 C D D1 a trivial fibration. Then

any cofibrant approximation (4.14) lifts to a diagram
C0
u
C1

g0

g1

w D0
u
w D1

and, hence, for an f -local X C there is a diagram

HomC (D1 , X)
u
HomC (D0 , X)

g1

g0

w HomC (C1 , X)
u
w HomC (D1 , X).

The vertical maps are weak equivalences by SM7 and the claim follows.
Wed now like to define f -local cofibrations in C to be cofibrations in C, and
f -local fibrations to be those morphisms with the appropriate lifting property.
Without a further hypothesis, however, we cannot prove a crucial lemma; namely,
it will not be clear that the class of morphisms which are at once cofibrations
and f -local equivalences is closed under pushouts. Thus:
Definition 4.16. A closed model category C is rigid on cofibrations if
1) given any diagram
D1 u
u
D2 u

j1

j2

C1

w X1

u
C2

u
w X2

510

IX. Localization
with j1 and j2 cofibrations and all vertical maps weak equivalences, then
the induced map on pushouts D1 C1 X1 D2 C2 X2 is a weak equivalence.
2) Let be an ordinal number and let X Y be a morphism of diagrams
over Seq() such that for each s < , Xs Ys is a weak equivalence, the
maps Xs Xs+1 , Ys Ys+1 are cofibrations and if s is a limit ordinal,
lim
X = Xs and lim
Y = Ys . Then the induced map
t<s t
t<s t
lim Xs lim Ys

s<

s<

is a weak equivalence.
Before exploring the implications of this definition for localized closed model
categories, we give some examples.
Examples 4.17.
1) Any closed model category in which every object is cofibrant is rigid on cofibrations (Lemma II.9.8, Corollary II.9.25).
2) Lemma II.9.21 says that any proper closed model category satisfies 4.16.1.
In many cases, 4.16.2 is a triviality. Thus, for example, compactly generated
Hausdorff spaces in the usual structure is rigid on cofibrations. And so are the
categories of simplicial algebras, simplicial commutative algebras, simplicial Lie
algebras, simplicial groups, simplicial abelian groups and their graded analogs.
3) Let C be a closed model category which is rigid on cofibrations and let C I
be the category of diagrams in C for some small category I. Suppose C I has a
closed model category structure where a morphism of diagrams X Y is a
weak equivalence if and only if Xi Yi is a weak equivalence for all i I, and
if a morphism X Y is a cofibration, then Xi Yi is a cofibration. (This
applies to SI in its usual structure (Example II.6.9), for example.) Then C I is
rigid on cofibrations. This follows from the fact that colimits in C I are formed
level-wise; that is, for any diagram of objects X in C I ,
(lim X )i
= lim(Xi ).

The crucial implication of this property of the rigidity hypothesis is the following.
Lemma 4.18. Assume the hypotheses of Proposition 4.9, and suppose the simplicial closed model category C is rigid on cofibrations. Then the functor L :
C C preserves all weak equivalences.

4. Localization in simplicial model categories

511

Proof: In advance we know only that L preserves weak equivalences between

cofibrant objects. Fix any weak equivalence X Y . We will modify the proof
of Proposition 1.17 to produce a diagram
X
u
Y

w L0 (X)
u
w L (Y )
0

where X and Y are f -cofibrations and L0 (X) L0 (Y ) is a weak equivalence

between f -local objects. The result will follow.
As in the proof of Proposition 1.17 we will produce, for each ordinal number s,
0
an object Es0 X, functorial in X, so that there is an f -cofibration Es0 X Es+1
X,
if s is a limit ordinal, then
lim Et0 X = Es0 X,

t<s

is f -local. Then L0 (X) = Es0 X for s large enough.

and, for s large enough,
However, we will modify the construction so that Es0 X Es0 Y is a weak equivalence for all s.
As before, set E00 X = X. Then E00 X E00 Y is a weak equivalence. Suppose
Es0 X has been defined and Es0 X Es0 Y is a weak equivalence. Since C is
cofibrantly generated, there is a natural square
Es0 X

Es0 X

w EsX

Es0 Y

w EsY

where the horizontal morphisms are trivial cofibrations and E s X and E s Y are
fibrant. Now apply the Main Construction 1.16 to E s X to define Es+1 X; thus,
there is a pushout diagram
G

C Hom(C , E s X)

w EsX

u
D Hom(C , E s X)

u
0
X.
w Es+1

G
I

512

IX. Localization

Now, since E s X E s Y is a weak equivalence between fibrant objects and C

is cofibrant,
C Hom(C , E s X) C Hom(D , E s Y )
is a weak equivalence. Similarly with D in place of C . Thus Condition 4.16.1
0
0
implies Es+1
X Es+1
Y is a weak equivalence. Since any trivial cofibration is
0
an f -cofibration, one has as in the proof of Proposition 1.17, that Es0 X Es+1
Y
is an f -cofibration.
If s is a limit ordinal, define Es0 X = lim
E 0 X. Then Condition 4.16.2 implies
t<s t
Es0 X Es0 Y is a weak equivalence. The remainder of the proof is accomplished
as in Proposition 1.17.

Corollary 4.19. Assume the hypotheses of Lemma 4.18. Then a morphism

C D in C is an f -local equivalence if and only if L(C) L(D) is a weak
equivalence.
Proof: If C and D are cofibrant, this can be proved as in Lemma 2.1.2. For
the general case, g : C D is an f -local equivalence if and only if g0 : C0 D0
is an f -local equivalence for some cofibrant approximation for g. Now consider
the diagram
L(C0 )
u
L(C)

Lg0

Lg

w L(D0 )
u
w L(D).

By Lemma 4.18, the vertical maps are weak equivalences; hence Lg0 is a weak
equivalence if and only if Lg is a weak equivalence.

Corollary 4.20. Assume the hypotheses of Lemma 4.18.

1) The class of morphisms in C which are at once cofibrations and f -local
equivalences is closed under pushouts and colimits over ordinal numbers.
2) Let j : C D be a morphism in C that is at once a cofibration and an
f -local equivalence, and K L be any cofibration in S. Then if L()
preserves cofibrations
D K CK C L D L
is a cofibration and an f -local equivalence.

4. Localization in simplicial model categories

513

Proof: Consider the maps

j

D
CX
where j is a cofibration and an f -local equivalence. We need to show that the
cofibration X D C X is an f -local equivalence. If D, C, and X are cofibrant,
then one can use the proof of Lemma 2.7.1. If not, find a diagram

D0 u
u
Du

j0

C0

w X0

u
C

u
wX

where the vertical maps are weak equivalences, j0 is a cofibration and D0 , C0 ,

and X0 are cofibrant. Then j0 is a f -local equivalence by Lemma 4.18 and
Corollary 4.19. Hence X0 D0 C0 X0 is an f -local equivalence. By 4.16.1
there is a diagram
X0

w D0 C0 X0

u
X

u
w D C X

with the vertical maps weak equivalences. Hence X D C X is an f -local

equivalence.
A similar argument using 4.16.2 handles the case of colimits over ordinal numbers. Part 2) follows from part 1) exactly as in the proof of Lemma 2.7.2.

Hypotheses/Definition 4.21. We now recapitulate. We have a simplicial

model category C such that
1) C is cofibrantly generated;
2) every object of C is cofibration definite;
3) C is rigid on cofibrations.
Then Lemma 4.18, Corollary 4.19 and Corollary 4.20.1 hold. If, in addition, we
have
4) for all cofibrant objects C C and all cofibrations U V , the functor
homC (C, U ) hom(C, V ) is an injection,

514

IX. Localization

then L() preserves cofibrations and Corollary 4.20.2 holds. Examples of C satisfying hypotheses 1)4) are given in 4.17.
We now define a morphism in C to be an f -local cofibration if it is a cofibration
and and f -local fibration if it has the right lifting property with respect to all
morphisms which are at once f -local equivalences and cofibrations. The notions
of f -local equivalence, f -local cofibration, and f -local fibration together will be
called the f -local structure.
We now add one more hypothesis of a less concrete nature:
5) there is a set J of morphisms j : C D in C so that
i) every element of J is a cofibration and an f -local equivalence; and
ii) a morphism X Y in C is an f -local fibration if and only if it has
the right lifting property with respect to all elements of J.
Proposition 4.22. Let C be a simplicial model category satisfying 4.21.14.
Then with the f -local structure defined above and the same simplicial structure,
C becomes a simplicial model category provided
() every morphism X Y can be factored
j

X
Z
Y
where q is an f -local fibration and j is at once a cofibration and an f -local
equivalence.
Proof: This goes exactly as in the proof of Theorem 2.3. The analog of
Lemma 2.5 is easy in the presence of Corollary 4.19 (use the same proof as in 2.5)
and we have assumed Lemma 2.6. Axiom SM7 follows from Corollary 4.20.2.
In order to apply this result, we will need to verify the factorization hypothesis.
In practice this is done as in the proof of Lemma 2.6. To be specific, we have
Proposition 4.23. Suppose that C is a simplicial model category satisfying
4.21.13 and 4.21.5. Then every morphism X Y can be factored
j

X
Z
Y
where q is an f -local fibration and j is at once a cofibration and an f -local
equivalence.

5. Localization in diagram categories

515

5. Localization in diagram categories.

Suppose that I is a simplicial category, and form the category SI of simplicial functors, as in Section VIII.1. This means in particular that I has a set
of objects, and the morphism objects between objects i and j of I consist of
simplicial sets Hom(i, j), so that a simplicial functor X : I S consists of a
simplicial set-valued function i 7 XI defined on Ob(I) and simplicial set maps
X : Hom(i, j) Hom(Xi , Xj ) which respect composition and identities in
the obvious sense. Recall that a morphism f : X Y of SI is a collection
of simplicial set maps f : Xi Yi , i Ob(I), which behaves like a natural
transformation in each simplicial degree. Recall further that there is a natural
simplicial structure on SI for which X K is defined for X S I and simplicial
sets K by the formula
(X K)i = Xi K,
and for which Hom(X, Y ) has n-simplices of the form::
Hom(X, Y )n = homSI (X n , Y ).
It has been shown (in various places: Remark VIII.1.6, Section IV.3.1, Example II.7.11) that there is a simplicial model structure on SI for which a map
f : X Y is a fibration (respectively weak equivalence) if all the component
maps f : Xi Yi are fibrations (respectively weak equivalences) of simplicial
sets. This result, which we often call the pointwise fibration structure is one of
the key components of all our discussions of homotopy coherence, but a sensible
discussion of localization in this context requires the introduction of a different
structure:
Proposition 5.1 (The pointwise cofibration structure). With the simplicial structure defined above, SI has a simplicial model category structure
where a morphism f : X Y is
1) a weak equivalence if each fi : Xi Yi is a weak equivalence;
2) a cofibration if each fi : Xi Yi is a cofibration (that is, an injection),
and
3) a fibration if it has the right lifting property with respect to all trivial
cofibrations.
It helps to know that the functor
(5.2)

Gi : SI S

516

IX. Localization

given by Gi X = Xi has both a left and right adjoint. The left adjoint is given
by
(5.3)

(Li Y )(j) = Y HomI (i, j)

and the right adjoint given by

(5.4)

(Ri Y )(j) = homC (HomI (j, i), Y ).

Note
that in the trivial case
F
Q where HomI (i, j) = homI (i, j) one has (Li Y )(j) =
ij Y and (Ri Y )(j) =
ji Y .
Proof of Proposition 5.1: Assume first that we can prove CM5. Then
CM1CM3 are clear and we need only prove the cofibration-trivial fibration
half of CM4 to get a closed model category. For this consider a lifting problem
in SI
A

wX
\
]
\ p
i
u \
u
B
wY
where i is a cofibration and p is a trivial fibration. Form the diagram
X

u
X A B

wX
\
]
\ p
u
\
q
wZ
wY

where we have used CM5 to factor X tA B Y as a cofibration j by a trivial

fibration q. Then q(j) = p is a weak equivalence and q is a weak equivalence,
so ji is a trivial cofibration and the indicated lift exists. The composite B
X tA B Z X solves the original lifting problem. Finally, Axiom SM7
follows from the corresponding property for simplicial sets.
This leaves CM5. For this we use Bousfield factorization twice. The argument
is completed in Lemma 5.6, once we prove a preliminary lemma.

F
I
If X S let #X be the cardinality of the simplices of i Xi.
Lemma 5.5. There is a cardinal number so that a morphism q : X Y is
a fibration in the level-wise cofibration structure if and only if it has the right
lifting property with respect to all trivial cofibrations j : A B with #B .

5. Localization in diagram categories

517

Proof:
Let be a fixed infinite cardinal greater than the cardinality of
F
(i,j) HomI (i, j), where (i, j) runs over all pairs of objects in I. The claim
is that if j : A B is any trivial cofibration, an x Bi is a simplex, there
is a sub-object C B so that x C, #C , and A C C is a trivial
cofibration. Then the Zorns lemma argument of Lemma 2.14 applies to finish
the result.
To prove the claim, one constructs objects Cn B so that x C0 , #C ,
Cn Cn+1 and k (|Cn |, |Cn A|) k (|Cn+1 |, |Cn+1 A|) is trivial for all k 0
(Note that S
0 (P, Q) is the quotient set 0 P/0 Q.) and all choices of basepoint.
Then C = n Cn . To get Cn , choose C0 to be any sub-object with x C0 and
#C0 . This is possible; choose x C0 (i) Bi where C0 (i) has only finitely
non-degenerate simplices and let C0 be the image of the induced map Li C0 (i)
B where Li is as in (5.3). Having produced Cn , proceed as follows. For each i and
each k and any choice of basepoint in Cn (i) A, let z k (|Cn (i)|, |Cn (i) A|).
(i)
There is a sub-complex Dz B(i) with finite many-degenerate simplices so
that z is trivial in k (|Cn (i) Dz (i)|, |(Cn (i) Dz (i)) A(i)|). This is because
A B is a trivial
cofibration. Let Dz = image of Li Dz (i) B and let
S
Cn+1 = Cn ( z Dz ).

Lemma 5.6. Axiom CM5 holds for the level-wise cofibration structure on SI .
Proof: For the trivial cofibration-fibration half of this axiom, note that trivial cofibrations are closed under pushouts and colimits over ordinal numbers.
Thus we need only perform the Bousfield factorization with respect to
G G
G
f=
j:
A
B
J

where J runs over a set of isomorphism representatives for all trivial cofibrations
j : A B with #B . The argument can be completed as in Proposition 1.17.
For the cofibration-trivial fibration half of this axiom, we make the following
observation: a morphism q : X Y is a trivial fibration if and only if it is a
fibration and qi : Xi Y i is a trivial fibration for all i. This follows from the
fact that the functors Li : S SI preserve trivial cofibrations. Hence q : X Y
is a trivial fibration if and only if it has the right lifting property with respect
to the set of maps J 0 of the form A Fi n . Since
n , we
F can perform
F #LiF
the Bousfield factorization with respect to g = J 0 j : J 0 C J 0 D to finish
the argument.

I
The level-wise cofibration structure on S is particularly suited to localization,
even though the exact nature of the fibrations is mysterious. We take some
paragraphs to explain.
Fix a level-wise cofibration f : A B in SI .

518

IX. Localization

Proposition 5.7. Localization with respect to f exists in SI ; that is, there is a

functor L : SI SI and a natural f -cofibration X LX so that LX is f -local
and induces a localization functor
Lf : Ho(SI ) Locf .
Proof: This follows from Propositions 4.9 and 4.10.

Note that the analogous statement holds in the level-wise fibration structure,
although one must take some care with Proposition 4.9 as not every object is
cofibrant.
Proposition 5.8. The functor L : SI SI has the following properties:
1) if C D is a level-wise cofibration, so is LC LD
2) there is a cardinal so that for all C SI
lim L(C ) L(C)

is an isomorphism where C runs over the sub-objects of C with #C .

3) If C1 , C2 are sub-objects of some D SI , then L(C1 C2 )
= L(C1 )
L(C2 ).
Proof: Part 1) follows from Lemma 4.11, Part 2) is proved as in Lemma 1.23,
and Part 3) as in Lemma 1.25.

Proposition 5.9. The category SI is rigid on level-wise cofibrations, and the

functor L : SI SI preserves all weak equivalences.
Proof: The category SI is proper since the simplicial set category is proper, so
we may apply Lemmas II.9.21 and 4.18.

The analogous statement holds for the level-wise fibration structure. Wed now
like to produce the localization model category structure. We define a morphism
q : X Y in SI to be an f -local fibration if it has the right lifting property with
respect to all level-wise cofibrations j : C D which are f -local equivalences.
Proposition 5.10. There is an infinite cardinal so that a morphism in SI is
an f -local fibration if and only if it has the right lifting property with respect
to all level-wise cofibrations j : C D which are f -local equivalences and for
which #D .
Proof: In the presence of Proposition 5.8, the Bousfield-Smith cardinality argument pertains. See Lemmas 2.132.14.

The next result says that the level-wise cofibrations structure localizes. We
have a level-wise cofibration f : A B in SI .

5. Localization in diagram categories

519

Proposition 5.11. The category SI , with its simplicial structure, has an f local simplicial category structure where a morphism is a
1) weak equivalence if it is an f -local equivalence;
2) cofibration if it is a level-wise cofibration;
3) fibration if it is an f -local fibration.
Proof: Using Proposition 5.10, apply Propositions 4.22 and 4.23.

A consequence of Proposition 5.11 is that one can also localize the level-wise
fibration closed model category structure on SI . For this one would require a
cofibration f : A B between cofibrant objects in this structure. We append
the adjective modified to all the constructions; therefore, we have modified f local equivalences and modified f -local fibrations. As noted above, the analogs of
Propositions 5.7 and 5.9 hold. Note that a cofibration in the level-wise fibration
structure is a level-wise cofibration.
Lemma 5.12.
1) If an object X in SI is f -local, it is modified f -local.
2) If a morphism C D is cofibration in the level-wise fibration structure
and a modified f -local equivalence, it is a level-wise cofibration and an f -local
equivalence.
3) If a morphism X Y is an f -local fibration, it is a modified f -local fibration.
Proof: Part 1) follows from the definition of f -local and the fact that a fibration in the level-wise cofibration structure is a level-wise fibration. Part 2) is a
consequence of Part 1) and the definition of f -local equivalence. Part 3) follows
from Part 2) by a lifting argument.

We now have the following consequence of Proposition 5.11.

Corollary 5.13. The category SI , with its simplicial structure has an f -local
simplicial category structure, where a morphism is
1) a weak equivalence if it is an f -local equivalence;
2) a cofibration if it is a cofibration in the level-wise fibration structure;
3) a fibration if it is a modified f -local fibration.
Proof: We apply Proposition 4.22. To verify the trivial cofibration-fibration
factorization we use Proposition 5.11. Fix a morphism: X Y . Factor g as
j

X
Z
Y

520

IX. Localization

where j is a level-wise cofibration and an f -local equivalence, and q is an f -local

fibration in the sense of Proposition 5.11. Now factor j as
p

X
Z0
Y
where p is a level-wise trivial fibration and i is a cofibration in the level-wise
fibration structure. Then j = pi is an f -local equivalence, so i is an f -local
equivalence. Hence we need only verify qp : Z 0 Y is a modified f -local fibration. However, any level-wise trivial fibration will be a modified f -local fibration,
because it will have the right lifting property with respect to all cofibrations in
the level-wise fibration structure. Finally q is a modified f -local fibration by
Lemma 5.12.3.

We close this section with some comments on homotopy inverse limits. Let I
be a discrete category. The constant diagram functor S SI sending X S
to the diagram Xi = X and all the maps the identity sends cofibrations to
level-wise cofibrations and preserves weak equivalences. Thus the inverse limit
functor
lim = lim : SI S

I

preserves weak equivalences between fibrant objects (in the level-wise cofibration
structure) and hence there is a total right derived functor
holim = R lim : Ho(SI ) Ho(S).

Thus if X SI one chooses a level-wise trivial cofibration X Y to a fibrant

object. Then lim Y is a model for holim X.
I
I
Observe that there is a much more concrete description in the case of towers
see Definition VI.1.7.
Proposition 5.14. Let f : A B be a cofibration in S and suppose X SI
is an I-diagram so that Xi is f -local for all i I. Then holim X is f -local.
I
Proof: We work with the level-wise cofibration structure on SI and the localization in SI provided by Proposition 5.7.
Let G : SI S be the functor Gi Z = Zi and let Li and Ri be its right and
left adjoints respectively. Let c : S SI be the constant diagram functor. Let
g be the coproduct in SI of the maps
Li f : Li A Li B

iI

5. Localization in diagram categories

521

and
cf : cA cB.
Then g is a level-wise cofibration. Choose a g-cofibration X Y with Y glocal. Then we claim that for each i, Yi is f -local and the maps Xi Yi are
f -cofibrations. If this is the case holim X = lim Y , since Y is fibrant and lim Y

is f -local since
f : Hom(B, lim Y ) Hom(A, lim Y )

is isomorphic
cf : Hom(cB, Y ) Hom(cA, Y )
which is a weak equivalence.
To see that each Yi is local, note that
f : Hom(B, Yi ) Hom(A, Yi )
is isomorphic to
Li f : Hom(Li B, Y ) Hom(Li A, Y )
which is a weak equivalence.
To see that Xi Yi is an f -cofibration, we prove that if E D is an f injective, then Ri E Ri D is a g-injective. Then
Hom(Yi , E) Hom(Yi , D) Hom(Xi ,D) Hom(Xi , E)
is isomorphic to
Hom(Y, Ri E) Hom(Y, Ri D) Hom(BX,Ri D) Hom(X, Ri E)
which will be a weak equivalence.
Now consider
(5.15) Hom(Lj B, Ri E) Hom(Lj B, Ri D) Hom(Lj A,Ri D) Hom(Lj A, Ri E).
This is isomorphic to
(5.16) Hom((Lj B)i , E) Hom((Lj B)i , D) Hom((Lj A)i ,D) Hom((Lj A)i , E).
But (Lj A)i (Lj A)i is the map
G
G
G
f:
A
B
ji

ji

which is an f -cofibration. So (5.16) is a weak equivalence and, hence (5.15) is a

weak equivalence. The same argument shows
Hom(cB, Ri E) Hom(cB, Ri D) Hom(cA,Ri D) Hom(cA, Ri E)
is a weak equivalence, and we have proved Ri E Ri D is a g-injective.

522

IX. Localization

6. A model for the stable homotopy category.

We have so far concentrated on the concept of localization at a cofibration
or set of cofibrations. This is not the only method for constructing localization
theories: there are also theories which arise from classes of cofibrations satisfying
certain axioms this is the approach taken in [38], and it completely subsumes
the material presented here in cases where the ambient closed model category
admits cardinality arguments. There is another, older approach, which involves
localizing at a sufficiently well-behaved functor Q : C C from a proper simplicial model category to itself. This technique is due to Bousfield and Friedlander
[12], and is the basis for their approach to constructing the stable homotopy
category. We shall present this method here.
Suppose that C is a proper closed (simplicial) model category, and let Q :
C C be a functor. Suppose further that there is a natural transformation
X : X Q(X) from the identity functor on C to Q. Say that a map f : X Y
is a Q-weak equivalence if the induced map Q(f ) : Q(X) Q(Y ) is a weak
equivalence of C, and say that a map is a Q-fibration if it has the right lifting
property with respect to all cofibrations of C which are Q-weak equivalences.
We shall require that the functor Q and the natural map together satisfy
the following list of properties:
A1 The functor Q preserves weak equivalences.
A2 The maps QX , Q(X ) : Q(X) Q2 (X) are weak equivalences of C.
A3 The class of Q-weak equivalences is closed under pullback along Q-fibrations: given maps
g

B
Y
X
with p a Q-fibration and g a Q-weak equivalence, the induced map g :
B Y X X is a Q-weak equivalence. Dually, the class of Q-weak
equivalences is closed under pushout along cofibrations of C.
The first major result of this section is the following:
Theorem 6.1. Suppose that C is a proper closed model category, Q : C C is
a functor, and X : X Q(X), x Ob(C), is a natural transformation, all satisfying the properties A1A3 above. Then there is a closed model structure on
C for which the weak equivalences are the Q-weak equivalences, the cofibrations
are the cofibrations of C, and the fibrations are the Q-fibrations.
We shall refer to the closed model structure of Theorem 6.1 as the Q-structure
on the category C. Note first that this result implies that there is an idempotent
functor G : Ho(C) Ho(C) in the sense of Adams, and hence an associated
categorical localization theory (see [8, p.135]):

6. A model for the stable homotopy category

523

Corollary 6.2. Suppose that the conditions for Theorem 6.1 hold. Then there
is a functor G : Ho(C) Ho(C) and a natural transformation j : 1Ho(C) G
such that jGX = G(jX ), and jGX is an isomorphism of Ho(C) for all X.
Proof: It suffices to work on the level of objects of C which are fibrant and
cofibrant for the original closed model structure on C, so that the morphisms of
Ho(C) can be identified with homotopy classes of maps.
Choose a map jX : X GX such that jX is a cofibration and a Q-equivalence,
and GX is Q-fibrant. Every trivial cofibration of C is a Q-equivalence, by A1,
so GX is fibrant in the original structure on C.
If f : X Y is a map such that Y is Q-fibrant, then there is an extension
f : GX Y such that f jX = f . Furthermore, the choice of f is independent up to homotopy of the homotopy class of f . It follows that any map
f : X Y determines an extension Gf : GX GY such that Gf jX = jY f ,
and this construction determines a functor G : Ho(C) Ho(C), and a natural
transformation j : 1Ho(C) G. The maps jGX and GjX are two extensions
of a common map X G2 X, so they coincide up to homotopy, and the map
GjX : GX G2 X is a Q-equivalence between objects which are Q-fibrant and
cofibrant, so GjX is a homotopy equivalence.

The proof of Theorem 6.1 will be given in a series of lemmas.

Lemma 6.3. Assume the conditions of Theorem 6.1. Then a map f : X Y is
a fibration and a weak equivalence of C if and only if it is a Q-equivalence and a
Q-fibration.
Proof: If f is a fibration and a weak equivalence of C, it is a Q-equivalence by
A1 and has the right lifting property with respect to all cofibrations, so its a
Q-fibration.
Suppose that f is a Q-equivalence and a Q-fibration, and take a factorization
f = p i, where where i : X Z is a cofibration and p : Z Y is a trivial
fibration in C. Then i is a Q-equivalence by A1, so the lifting exists in the
diagram
= wX
]
[
f
i [
u [
u
Z p wY

and so f is a retract of p.

Lemma 6.4. Suppose that f : X Y is a fibration of C, and that X : X QX

and Y : Y QY are weak equivalences of C. Then f is a Q-fibration.

524

IX. Localization

Proof: Start with a lifting problem

wX
]
\
f
i
\
u \
u
B
wY

where i is a cofibration and a Q-equivalence. We have to show that the dotted

arrow exists.
There is a diagram
Q

QA
Qi

6
44

44 p
j
Z

u
QB

w QX

6
44
u 44p
W

Qf
u
w QY

where j and j are trivial cofibrations of C and p and p are fibrations. To

see this, find the factorization Q = p j , and then factorize the induced map
QA W QY QX as a trivial cofibration j followed by fibration. There is an
induced diagram
A
i

u
B

w Z QX X

w W QY Y

wX
u
wY

such that the top horizontal composite is and the bottom composite is .
We will show that the conditions on f and i imply that the map is a weak
equivalence of C. This suffices, for then has a factorization = q j where q

6. A model for the stable homotopy category

525

is a trivial fibration and j is a trivial cofibration, and there is a diagram

w Z QX X
wX
A
C
A
A
j
u AA
f
i
U
A
C
AA
q
A
u A
u
u
B
w W QY Y
wY

so that is the desired lift.

There is a final diagram

QA

Qi

u
QB

wZu

pr

u
w W u pr

Z QX X
u

W QY Y

The map Qi is a weak equivalence since i is a Q-equivalence, so that is a weak

equivalence. The maps X and Y are weak equivalences by assumption and
p and p are fibrations, so the maps labelled pr are weak equivalences by the
properness assumption on C. It follows that is a weak equivalence of C.

Lemma 6.5. Any map f : QX QY has a factorization f = p i where

p : Z QY is a Q-fibration and i : QX Z is a cofibration and a Qequivalence.
Proof: Take a factorization f = p i where p : Z QY is a fibration and
i : QX Z is a cofibration and a weak equivalence. Applying the functor Q to
this diagram, one sees that QX is a weak equivalence by A2, and i is a weak
equivalence so that Qi is a weak equivalence by A1. It follows that Z : Z QZ
is a weak equivalence, as is QY , so that p is a Q-fibration by Lemma 6.4.

Lemma 6.6. Any map f : X Y of C has a factorization f = q j where

q : Z Y is a Q-fibration and j : X Z is a cofibration and a Q-equivalence.

526

IX. Localization

Proof: Take the factorization f = p i of Lemma 6.5 and pull Y : Y QY

along p to give a diagram

u
QX

w Z QY Y

u
wZ

wY
Y
u
w QY,

where p i = f . Then is a Q-equivalence by A2 and A3, as are both X

and i. It follows that i is a Q-equivalence.
Now factorize i in C as i = j, where j is a cofibration and is a trivial
fibration. Then is a Q-fibration by Lemma 6.3, so that q = p is a Qfibration.

Proof of Theorem 6.1: The closed model axioms CM1CM3 are trivial
to verify. One part of the factorization axiom is Lemma 6.6. The other is a
consequence of Lemma 6.3, as is CM4.

Remark 6.7. Observe that if C has a simplicial model structure in addition to

the conditions of Theorem 6.1, then the Q-structure on C is a simplicial model
structure, essentially for free. Axiom SM7 is a consequence of the fact that the
two structures have the same cofibrations, and every trivial cofibration of C is a
Q-equivalence.
Similarly, the full statement of A3 (weve only used the fibration part of it up
to now) implies that the Q-structure is proper.
Heres the second major result:
Theorem 6.8. Suppose that the proper closed model category C and the functor
Q together satisfy the conditions for Theorem 6.1. Then a map f : X Y is a
Q-fibration if and only if it is a fibration and the square

X
(6.9)

u
Y

is a homotopy cartesian diagram of C.

w QX
Qf
u
w QY

6. A model for the stable homotopy category

527

Proof: Suppose that the diagram (6.9) is homotopy cartesian and that f : X
Y is a fibration of C. We will show that the map f is a retract of a Q-fibration.
Factorize Qf as Qf = p i, where p : Z QY is a fibration of C and
i : QX Z is a trivial cofibration. Then, as in the proof of Lemma 6.5, the
map p is a Q-fibration, so that the induced map p : Y QY Z Y is a Qfibration. The induced map i : X Y QY Z is a weak equivalence of C, since
C is proper, and so it has a factorization i = j, where j : X W is a trivial
cofibration and : W Y QY Z of C is a trivial fibration of C. Then the
composite p : W Y is a Q-fibration by Lemma 6.3, and the lifting exists in
the diagram
= wX
]
[
j
f
[
u [
u
W p wY
X

since f is a fibration. It follows that f is a retract of p .

For the converse, suppose that f : X Y is a Q-fibration. We show that f is
a retract of a fibration g for which the diagram
Z
g
u
W

(6.10)

w QZ
Qg
u
w QW

is homotopy cartesian in C.
Observe that the class of maps g which are fibrations of C and for which the
diagram (6.10) is homotopy cartesian is closed under composition, and includes
all trivial fibrations of C by A1. It suffices therefore, with respect to the construction giving the first part of the proof, to show that the map p is a candidate
for one of these maps g.
The component square diagrams in

Y QY Z
u
Y

wZ
p

w QZ

Qp
u
u
w QY
w Q2 Y
QY

528

IX. Localization

are homotopy cartesian since Z and QY are weak equivalences (see the proof of
Lemma 6.5), so that the composite square is homotopy cartesian. This composite
coincides with the composite of the squares
Y QY Z
(6.11)

u
Y

w Q(Y QY Z)

Qp
Y

u
w QY

QY

w QZ
Qp
u
w QQY

The map is a Q-equivalence by A3, so that Q is a weak equivalence, as is

QY . It follows that the square on the left in (6.11) is homotopy cartesian, but
this is what we had to prove.

Corollary 6.12. Suppose that the proper closed model category C and the
functor Q together satisfy the conditions for Theorem 6.1. Then an object X of
C is Q-fibrant if and only if it is fibrant in C and the map X : X QX is a
weak equivalence of C.
The principal application of this theory is the Bousfield-Friedlander construction of the stable homotopy category.
A spectrum X (or rather a spectrum object in simplicial sets) consists of
pointed simplicial sets X n , n 0, together with pointed simplicial set maps
: S 1 X n X n+1 , which we call bonding maps. Here, S 1 = 1 /1 is the
simplicial circle. A map of spectra f : X Y consists of pointed simplicial set
maps f : X n Y n , n 0, which respect structure in the sense that all diagrams
S1 X n

S1 f

u
1
S Yn

w X n+1
u

w Y n+1

commute. The resulting category of spectra will be denoted by Spt.

A map f : X Y of spectra is said to be a strict weak equivalence (respectively
strict fibration) if all of the component maps f : X n Y n are weak equivalences
(respectively fibrations) of simplicial sets. A map j : A B of spectra is said
to be a cofibration if the following two conditions are satisfied:
(1) the map i : A0 B 0 is a cofibration (or monomorphism) of simplicial
sets, and

6. A model for the stable homotopy category

529

(2) each induced map

S 1 B n S 1 An An+1 B n+1
is a cofibration of simplicial sets.
Proposition 6.13. With the definitions of strict fibration, strict weak equivalence, and cofibration given above, the category Spt satisfies the axioms for a
proper simplicial model category.
It follows from the definition of cofibration that a spectrum X is cofibrant
if and only if the bonding maps S 1 X n X n+1 are all cofibrations. Notice
further that if i : A B is a cofibration of spectra, then all of the maps
i : An B n are cofibrations of simplicial sets, but the converse may not be
true. We usually say that a map j : C D of spectra for which all the
maps j : C n Dn are cofibrations is a pointwise cofibration. In the same
terminology, strict fibrations are pointwise fibrations and strict weak equivalences
are pointwise weak equivalences.
Remark 6.14. To see the simplicial structure for Spt a little more clearly, note
that if X is a spectrum and K is a pointed simplicial set, then there is a spectrum
X K with (X K)n = X n K, and having bonding maps of the form
K

S 1 X n K X n+1 K.
Then, for an arbitrary simplicial set L and a spectrum X, X L = X L+ ,
where L+ = L t is L with a disjoint base point attached. Dually, the pointed
function complex spectrum hom (K, X) (denoted by homS (K, X) in Section
II.2) is the spectrum with
hom (K, X)n = hom (K, X n ),
with bonding map S 1 hom (K, X n ) hom (K, X n+1 ) adjoint to the composite
1

S 1 ev

S hom (K, X ) K S 1 X n
X n+1
It follows that the evaluation maps ev : hom (K, X n ) K X n determine a
map of spectra ev : hom (K, X) K X.
Given a spectrum X, there is a spectrum X having (X)n = S 1 X n and
with bonding maps
S 1

S 1 S 1 X n S 1 X n+1 .

530

IX. Localization

This spectrum X is not the suspension object X S 1 arising from the simplicial
structure on Spt: the two differ by a twist of circle smash factors this fact
is an avatar of one of the standard dangerous bends in the foundations of stable
homotopy theory. We say that X is the fake suspension spectrum of X.
Similarly, there is a fake loop spectrum Y , which does not coincide with the
function complex object hom (S 1 , Y ). In effect, (Y )n = hom (S 1 , Y n ), but
the bonding map is the composite

ev

S 1 hom (S 1 , Y n ) hom (S 1 , Y n ) S 1 Y n hom (S 1 , Y n+1 ),

where : Y n hom (S 1 , Y n+1 ) is adjoint to the composite

Y n S 1 S 1 Y n
Y n+1 .

The fake loop construction is right adjoint to the fake suspension.

The closed model structure determined by Proposition 6.13 is usually called
the strict closed model structure on the category of spectra. The associated
homotopy category is not yet the stable category. We stabilize by localizing at
a suitable functor Q.
The construction of the functor Q : Spt Spt requires a natural strictly
fibrant model. There are both geometric and combinatorial ways of producing
such an object, corresponding to the two standard methods of functorially replacing simplicial set by a Kan complex. For the geometric method, observe that
there is a natural map
S|K| S|L| S|K L|
for each pair of pointed simplicial sets K and L. Given a spectrum X, the
bonding maps : S 1 X n X n+1 therefore induce composites
S||

S 1 S|X n | S|S 1 | S|S n | S|S 1 X n | S|X n+1 |,

so that applying realization and singular functors levelwise to a spectrum X
determines a spectrum S|X|. Its not hard to see that the canonical weak equivalences : X n S|X n | collectively determine a natural strict equivalence of
spectra : X S|X|. Of course, S|X| is strictly fibrant. Kans Ex construction can be goaded into performing a similar service: the natural composite
1

K Ex L Ex K Ex L
= Ex (K L)

6. A model for the stable homotopy category

531

induces a natural map

: K Ex L Ex (K L)
which, in the presence of a spectrum X determines composites
Ex

S 1 Ex X n Ex (S 1 X n ) Ex X n+1 .
Furthermore, the canonical maps : X n Ex X n together determine a natural strict equivalence : X Ex X, taking values in a strictly fibrant spectrum. Observe that Ex X and S|X| are naturally strictly equivalent.
The indices in spectra can be shifted at will: if X is a spectrum and n is an
integer, there is a spectrum X[n] with

X[n] =

X n+k

if n + k 0, and

if n + k < 0.

Shifting indices is functorial, and is cumulative in the sense that there are canonical natural isomorphisms X[n][k]
= X[n + k].
1
n
The bonding maps : S X X n+1 of a fixed spectrum X determine a
map of spectra : X X[1]. This map is natural in X, and has a natural
adjoint : X X[1], which is defined levelwise by the (twisted) adjoints
: X n hom (S 1 , X n+1 ) of the maps : S 1 X n X n+1 . This construction
can be repeated, to form an inductive system of maps
2 [2]

[1]

X X[1] 2 X[2] . . . .
Write X = lim n X[n]
n
The functor Q : Spt Spt is defined for spectra X by QX = Ex X,
and there is a natural map X : X QX given by the composite

X
Ex X
Ex X.
where : Y Y denotes the canonical map to the colimit.
Say that a map f : X Y of spectra is a stable equivalence if it induces a
strict equivalence Qf : QX QY . The map p : Z W is said to be a stable
fibration if it has the right lifting property with respect to all maps which are
cofibrations and weak equivalences.

532

IX. Localization

Theorem 6.15. With these definitions, the category Spt of spectra, together
with cofibrations, stable equivalences and stable fibrations satisfies the axioms
for a proper simplicial model category.
Proof: We need only verify that the functor Q satisfies axioms A1A3. Of
these, A1 is clear, and A2 is a consequence of the observation that
Y = Y : Y Y
is an isomorphism by a cofinality argument. The functor Q preserves strict
fibrations and pullbacks, giving the pullback part of A3. Recall that this is
enough to give the desired simplicial model structure.
Suppose given maps of spectra
j

B
A
C
where j is a cofibration and g is a stable equivalence. To finish verifying A3, we
have to show that the induced map g : B B A C is a stable equivalence.
Construct a diagram

Bu

C
]
[
g
C
[
u
[
A g wC

where g is a cofibration, and the map C is a strict fibrations and a strict weak
equivalence. The glueing lemma for simplicial sets implies that the induced map
: B A C B A C is a strict weak equivalence. It is therefore enough to
show that the induced map g : B B A C is a stable equivalence. But g
is a stable equivalence as well as a cofibration, and such maps are closed under
pushout.

Remarks 6.16.
1) In view of the coincidence of the homotopy groups of a space with the simplicial homotopy groups of its associated singular complex (Proposition I.11.1),
the definition of QX implies that the homotopy groups k QX n coincide up to
isomorphism with the stable homotopy groups kn |X| of the (pre)spectrum
|X|, so that Q-equivalence is stable equivalence: a map f : X Y of spectra is
a Q-equivalence if and only if it induces an isomorphism in all stable homotopy
groups. For this reason, as is standard, we say that Q-equivalences are stable
equivalences and that Q-fibrations are stable fibrations. The stable homotopy

6. A model for the stable homotopy category

533

groups can, of course, be defined completely in terms of the simplicial homotopy

groups arising from the strictly fibrant model Ex X.
2) Corollary 6.12 implies that a spectrum X is stably fibrant if and only if X
is strictly fibrant and all associated adjoints : X n hom (S 1 , X n+1 ) of the
bonding maps : S 1 X n X n+1 are weak equivalences. In other words, a
stably fibrant spectrum X is a type of -spectrum , and one commonly lapses
into describing it that way.
Write SptQ for the closed model structure on the category of spectra of Theorem 6.15. The associated homotopy category Ho(SptQ ) is a model for the stable
category, in view of the above remarks. This is far from being the only (or even
best) construction of the stable category, but it has the advantage of giving descriptions of stable homotopy theory in a variety of non-standard situations [52].
There is even a collection of ways to describe the closed model structure SptQ :
one could, for example, localize Spt at the family of stably trivial cofibrations,
as is done in [38].
Alternatively, the results of the last section can be used to show that SptQ
arises from the strict category by an f -localization, for a suitable choice of map
(or set of maps) f . The family of maps f is easy to describe: let S be the
(simplicial) sphere spectrum
S 0 = 1 , S 1 , S 1 S 1 , . . . ,
and consider the collection of canonical maps of the form f : S[n] S[n + 1],
n Z. These maps are not cofibrations in the strict closed model category, but
they can be replaced by such. Further, the objects S[n] and S[n] are cofibrant,
so a spectrum X is f -local for this family of maps f if and only if it is strictly
fibrant and all induced maps
f

Hom(S[n], X) Hom(S[n 1], X)

are weak equivalences of simplicial sets. An adjointness argument shows that
this last requirement is equivalent to the assertion that all of the maps

X n hom (S 1 , X n+1 )
are weak equivalences. In particular, a spectrum X is f -local if and only if it is
stably fibrant. Further, a map A B of cofibrant spectra is a stable equivalence
if and only if the induced map Hom(B, X) Hom(A, X) is a weak equivalence
for all stably fibrant spectra X, on account of Lemma II.4.2. It therefore follows
from the definition of f -local equivalence (Definition 4.13) that a map of spectra
is a stable equivalence if and only if it is an f -local equivalence. The weak
equivalences for the stable and f -local theories coincide, as do the cofibrations,
so we recover the stable closed model structure SptQ from Proposition 4.22.

REFERENCES
[1]. J.F. Adams, Stable homotopy and generalised cohomology, University of Chicago Press,
Chicago, 1974.
[2]. J.F. Adams, Infinite Loop Spaces, Princeton University Press, Princeton, 1978.

[3]. M. Artin and B. Mazur, Etale

homotopy, Lecture Notes in Math., Vol. 100, Springer,
Berlin-Heidelberg-New York, 1972.
[4]. M. Barr and C. Wells, Toposes, Triples and Theories, Springer-Verlag, New York, 1985.
[5]. M. Barratt, V.K.A.M. Guggenheim and J.C. Moore, On semi-simplicial fibre bundles,
Amer. J. Math. 81 (1959), 639657.
[6]. M.G. Barratt and S. Priddy, On the homology of non-connected monoids and their associated groups, Comment. Math. Helv. 47 (1972), 114.
[7]. A.K. Bousfield, A second quadrant homotopy spectral sequence, Trans. AMS 177 (1973),
305318.
[8]. A.K. Bousfield, The localization of spaces with respect to homology, Topology 14 (1975),
133150.
[9]. A.K. Bousfield, The localization of spectra with respect to homology, Topology 18 (1979),
257281.
[10]. A.K. Bousfield, Homotopy spectral sequences and obstructions, Isr. J. Math. 66 (1989),
54-104.
[11]. A.K. Bousfield, Localization and periodicity in unstable homotopy theory, J. AMS 7
(1994), 831873.
[12]. A.K. Bousfield and E.M. Friedlander, Homotopy theory of -spaces, spectra, and bisimplicial sets, Springer Lecture Notes in Math. 658 (1978), 80150.
[13]. A.K. Bousfield and V.K.A.M. Guggenheim, On PL de Rham theory and rational homotopy type, Mem. AMS 179 (1976).
[14]. A.K. Bousfield and D.M. Kan, Homotopy limits completions and localizations, Springer
Lecture Notes in Math., Vol. 304 (2nd corrected printing), Springer-Verlag, Berlin-Heidelberg-New York, 1987.
[15]. K.S. Brown, Abstract homotopy theory and generalized sheaf cohomology, Trans. AMS
186 (1973), 419458.
[16]. K.S. Brown, Buildings, Springer-Verlag, Berlin-Heidelberg-New York, 1989.
[17]. C. Cascuberta, Recent advances in unstable localization, The Hilton Symposium (1993),
CRM Proc. Lecture Notes 6, AMS, Providence, 1994, pp. 122.
[18]. E.B. Curtis, Simplicial homotopy theory, Advances in Math. 6 (1971), 107209.
[19]. J. Dieudonn
e, Les d
ebuts de la topologie alg
ebrique, Expo. Math. 3 (1985), 347357.
[20]. A. Dold and D. Puppe, Homologie nicht-additiver Funktoren; Anwendungen, Ann. Inst.
Fourier (Grenoble) 11 (1961), 201312.
[21]. E. Dror, Pro-nilpotent representations of homotopy types, Proc. AMS 38 (1973), 657660.
[22]. E. Dror-Farjoun, Homotopy localization and v1 -periodic spaces, Algebraic Topology (San
Feliu de Guixols, 1990), Lecture Notes in Math. 1622, Springer-Verlag, Berlin, 1992,
pp. 104113.
[23]. E. Dror-Farjoun, Cellular spaces, null spaces and homotopy localization, Lecture Notes
in Math. 1622, Springer-Verlag, Berlin, 1996.
[24]. W.G. Dwyer and D.M. Kan, Singular functors and realization functors, Ind. Math. 46
(1984), 147153.

534

References

535

[25]. W.G. Dwyer and D.M. Kan, Homotopy theory and simplicial groupoids, Indag. Math. 46
(1984), 379385.
[26]. W.G. Dwyer and D.M. Kan, Equivalences between homotopy theories of diagrams, Algebraic Topology and Algebraic K-theory, W. Browder, ed., Annals of Math. Studies, Vol.
113, pp. 180205, Princeton Univ. Press, Princeton, 1987.
[27]. W.G. Dwyer, D.M. Kan and J.H. Smith, Homotopy commutative diagrams and their
realizations, J. Pure Applied Albegra 57 (1989), 524.
[28]. W.G. Dwyer, D.M. Kan and C.R. Stover, An E 2 model category structure for pointed
simplicial spaces, Preprint.
[29]. W.G. Dwyer, D.M. Kan and C.R. Stover, The bigraded homotopy groups i,j X of a
pointed simplicial space X, Preprint.
[30]. W.G. Dwyer and J. Spalinski, Homotopy theories and model categories, Handbook of
Algebraic Topology, Elsevier, Amsterdam, 1995, pp. 73126.
[31]. S. Eilenberg, Singular homology theory, Ann. of Math. 45 (1944), 407477.
[32]. S. Eilenberg and J.A. Zilber, On semi-simplicial complexes and singular homology, Ann.
of Math. 51 (1950), 409513.
[33]. P. Gabriel and M. Zisman, Calculus of Fractions and Homotopy Theory, Springer-Verlag,
New York, 1967.
[34]. H. Gillet and D. Grayson, The loop space of the Q-construction, Ill. J. Math. 31(4)
(1987), 574597.
[35]. P. Goerss, Andr
e-Quillen cohomology and the homotopy groups of mapping spaces: understanding the E2 term of the Bousfield-Kan spectral sequence, J. of Pure and Applied
Algebra 63 (1990), 113153.
[36]. P.G. Goerss, Homotopy fixed points for Galois groups, Preprint.
[37]. P.G. Goerss, Comparing completions of a space at a prime, Preprint (1997).
[38]. P.G. Goerss and J.F. Jardine, Localization theories for simplicial presheaves, Preprint
(1996).
[39]. J.P.C. Greenlees and J.P. May, Derived functors of I-adic completion and local homology,
J. Algebra 149 (1992), 438453.
[40]. T. Gunnarsson, Abstract Homotopy Theory and Related Topics, Thesis, Chalmers Tekniska H
ogskola, G
oteborg, 1978.
[41]. J-C. Haussmann and D. Husemoller, Acyclic maps, Enseign. Math. 25 (1979), 5375.
[42]. P.S. Hirschhorn, Localization of Model Categories, http://www-math.mit.edu/psh.
[43]. L. Illusie, Complexe cotangent et d
eformations I, Springer Lecture Notes in Math., Vol.
239, Springer-Verlag, Berlin-Heidelberg-New York, 1971.
[44]. U. Jannsen, Continuous
etale cohomology, Math. Ann. 280 (1988), 207245.
[45]. J.F. Jardine, Simplicial objects in a Grothendieck topos, Contemporary Math. 55 (1986),
193239.
[46]. J.F. Jardine, Simplicial presheaves, J. Pure Applied Algebra 47 (1987), 3587.
[47]. J.F. Jardine, Stable homotopy theory of simplicial presheaves, Can. J. Math. 39(3)
(1987), 733747.
[48]. J.F. Jardine, The homotopical foundations of algebraic K-theory, Contemporary Math.
83 (1989), 5782.
[49]. J.F. Jardine, Supercoherence, J. Pure Applied Algebra 75 (1991), 103194.
[50]. J.F. Jardine, Homotopy and homotopical algebra, Handbook of Algebra, Vol. 1, Elsevier,
Amsterdam, 1996, pp. 639670.
[51]. J.F. Jardine, Boolean localization, in practice, Doc. Math. 1 (1996), 245275.

536

References

[52]. J.F. Jardine, Generalized Etale Cohomology Theories, Progress in Math., Vol. 146, Birkh
auser, Boston-Basel-Berlin, 1997.
[53]. A. Joyal, letter to A. Grothendieck, (1984).
[54]. D.M. Kan, Semisimplicial spectra, Ill. J. Math. 7 (1963), 463478.
[55]. D.M. Kan, Abstract homotopy I, Proc. of Nat. Acad. of Sci. 41 (1955), 10921096.
[56]. D.M. Kan, Abstract homotopy II, Proc. of Nat. Acad. of Sci. 42 (1956), 225228.
[57]. D.M. Kan, Abstract homotopy III, Proc. of Nat. Acad. of Sci. 42 (1956), 419421.
[58]. D.M. Kan, On c.s.s. complexes, Amer. J. Math. 79 (1957), 449476.
[59]. D.M. Kan, Adjoint functors, Trans. AMS 87 (1958), 294329.
[60]. D.M. Kan, A combinatorial definition of homotopy groups, Ann. of Math. 67 (1958),
288312.
[61]. K. Lamotke, Semisimpliziale algebraische Topologie, Springer-Verlag, Berlin-HeidelbergNew York, 1968.
[62]. J. Lannes, Sur les espaces fonctionnels dont la source est le classifiant dun p-groupe
ab
elien
el
ementaire, I.H.E.S. Publications Math
ematiques 75 (1992), 135244.
[63]. L.G. Lewis, J.P. May and M. Steinberger, Equivariant Stable Homotopy Theory, Springer
Lecture Notes in Math., Vol. 1213, Springer-Verlag, Berlin-Heidelberg-New York, 1986.
[64]. S. Mac Lane, Homology, Third Corrected Printing, Springer-Verlag, New York Heidelberg
Berlin, 1975.
[65]. S. Mac Lane, Natural associativity and commutativity, Rice University Studies 49 (1963),
2846.
[66]. S. Mac Lane, Categories for the Working Mathematician, Graduate Texts in Math., Vol.
5, Springer, Berlin-Heidelberg-New York, 1971.
[67]. J.P. May, Simplicial Objects in Algebraic Topology, Van Nostrand, Princeton, 1967.
[68]. J.P. May (with contributions by F. Quinn, N. Ray, and J. Tornehave), E ring spaces
and E ring spectra, Springer Lecture Notes in Math., Vol. 577, Springer-Verlag, BerlinHeidelberg-New York, 1977.
[69]. J.P. May, Pairings of categories and spectra, J. Pure Applied Algebra 19 (1980), 299346.
[70]. D. McDuff and G. Segal, Homology fibrations and the group-completion theorem, Invent.
Math 31 (1976), 279287.
[71]. H. Miller, The Sullivan conjecture on maps from classifying spaces, Ann. of Math. (2)
120 (1984), 3987.

[72]. J.S. Milne, Etale

Cohomology, Princeton University Press, Princeton, 1980.
[73]. I. Moerdijk, Bisimplicial sets and the group-completion theorem, Algebraic K-Theory:
Connections with Geometry and Topology, NATO ASI Series C, Vol. 279, pp. 225240,
Kluwer, Dordrecht, 1989.
[74]. J.C. Moore, Semi-simplicial complexes and Postnikov systems, Symposium Internationale
de Topologica Algebraica, 1958, pp. 232247.
[75]. D. Quillen, Homotopical Algebra, Lecture Notes in Math., Vol. 43, Springer, BerlinHeidelberg- New York, 1967.
[76]. D. Quillen, The geometric realization of a Kan fibration is a Serre fibration, Proc. AMS
19 (1968), 14991500.
[77]. D. Quillen, Rational homotopy theory, Ann. of Math. 90 (1969), 205295.
[78]. D. Quillen, Higher algebraic K-theory I, Springer Lecture Notes in Math. 341 (1973),
85147.
[79]. D. Quillen, Homotopy properties of the poset of non-trivial p-subgroups of a group, Advances in Math. 28 (1978), 101128.

References

537

[80]. C.L. Reedy, Homotopy theory of model categories, Preprint (1973).

[81]. M. Ronan, Lectures on Buildings, Perspectives in Math., Vol. 7, Academic Press, Boston,
1989.
[82]. H. Schubert, Categories, Springer, New York-Heidelberg-Berlin, 1972.
[83]. G. Segal, Categories and cohomology theories, Topology 13 (1974), 293312.
[84]. J-P. Serre, Trees, Springer-Verlag, Berlin-Heidelberg-New York, 1980.
[85]. D. Sullivan, Genetics of homotopy theory and the Adams conjecture, Ann. of Math. 100
(1974), 179.
[86]. E.H. Spanier, Algebraic Topology, First corrected Springer edition, Springer-Verlag, New
York, 1981.
[87]. M. Sweedler, Hopf Algebras, W.A. Benjamin, New York, 1969.
[88]. R.M. Switzer, Algebraic Topology Homotopy and Homology, Springer-Verlag, BerlinHeidelberg-New York, 1975.
[89]. D.H. Van Osdol, Simplicial homotopy in an exact category, Amer. J. Math. 99 (1977),
11931204.
[90]. R.M. Vogt, Homotopy limits and colimits, Math. Z. 134 (1973), 1152.
[91]. F. Waldhausen, Algebraic K-theory of spaces, Springer Lecture Notes in Math. 1126
(1985), 318419.