Introduction to Combinatorial Homotopy Theory

Francis Sergeraert

Ictp Map Summer School

August 2008

1 Introduction.
Homotopy theory is a subdomain of topology where, instead of considering the
category of topological spaces and continuous maps, you prefer to consider as
morphisms only the continuous maps up to homotopy, a notion precisely defined
in these notes in Section 4. Roughly speaking, you decide not to distinguish two
maps which can be continuously deformed into each other; such a weakening of
the notion of map is quickly identified as necessary when you intend to apply to
Topology the methods of Algebraic Topology. Otherwise the main classification
problems of topology are, except in low dimensions, out of scope.
If you want to “algebraize” the topological world, you will meet another diffi-
culty. The traditional topological spaces, defined for example through collections of
open subsets, cannot be directly processed by a computer; a computer can handle
only discrete objects and in a sense topology is the opposite subject. A combina-
torial intermediary notion between Topology and Algebra is required. Poincaré
started Algebraic Topology about a century ago by using the polyhedra as interme-
diary objects, but since the fifties, the simplicial notions have been recognized as
more appropriate. In this framework of combinatorial topology, the sensible topo-
logical spaces can be combinatorially defined, and also installed and processed on
a computer. This is valid even for very complicated or abstract spaces such as
classifying spaces, functional spaces; the various important functors of algebraic
topology can also be implemented as functional objects.
In a sense there is a conflict between both previous observations. The homotopy
relation is concerned by continuous deformations of maps, while combinatorial
models for topological spaces and maps do not seem to allow enough maps to
model homotopies. But we will see this apparent obstacle is easily overcome, and
the so-called Combinatorial Homotopy Theory is now one of the standard ground
theories for Algebraic Topology.
In particular, if you claim you are mainly interested by constructive results
in Algebraic Topology, it is quickly obvious combinatorial topology is required.

1
Constructive Algebraic Topology is a difficult but fascinating subject, and three
main solutions are now available:

1. Rolf Schön’s solution [21], quite elegant, unfortunately never (?) considered
since his remarkable memoir, in particular from a concrete programming
point of view.

2. The solution studied for years by this author and several collaborators, see
the lecture notes of the previous Map Summer School at Genova [20]. The key
point is that locally effective models for combinatorial spaces are sufficient
to use standard simple Algebraic Topology and make it constructive.

3. The operadic solution where the algebraic world is enriched enough to make
it equivalent to the topological world; more precisely the algebraic structure
of chain complexes is sufficiently enriched, thanks to appropriate operads, to
code in this way the topological spaces up to homotopy.

But whatever solution you decide to study, anyway you will have to use ingredi-
ents coming from combinatorial homotopy. For the solutions 1 and 2 above, it will
even be necessary to implement on your computer the corresponding necessary
simplicial objects and operators; for the solution 3, the objects of the resulting
category, the E∞ -operadic chain complexes, do not seem to use combinatorial ho-
motopy, but the theoretical justifications requires by some means or other this
theory. This Ictp-Map Summer School proposes an introduction to the solution 3
and the present lecture is intended to prepare the audience to the most elementary
facts of combinatorial homotopy.
Section 2 describes the most elementary simplicial techniques, around the no-
tion of simplicial complex. It is already possible in this simple framework to speak
of combinatorial homotopy, for example it is possible to construct simplicial mod-
els for functional spaces, in particular for loop spaces. An important progress at
the end of the forties was the invention (discovery?), mainly by Samuel Eilenberg,
of the notion of simplicial set, to which the rest of these notes is devoted. An
amusing paradox of this terminology must be signaled: the notion of simplicial set
is much more complex than the notion of simplical. . . complex! These simplicial
sets were initially called CSS-sets, an acronym for “complete-semi-simplicial”; but
it was identified a little later the general notion of simplicial object in an arbi-
trary category makes sense and a CSS-set is nothing but a simplicial object in the
category of sets, which explains the modern and natural terminology of simplicial
set.
This notion of simplicial set is one of the most fascinating elementary notions in
mathematics. In a sense it contains the whole richness of topology. Yet an essential
drawback must immediately be pointed out: modelling a topological object as a
simplicial set leads to coherent but arbitrary choices of orders (resp. orientations)
for the vertices (resp. simplices). It happens these choices hide very sophisticated
actions of the symmetric groups Sn ; in a sense, elementary Algebraic Topology
forgets this action and operadic Algebraic Topology on the contrary takes account

2
of this action, in a totally algebraic framework, and in this way, the initial goal
of Algebraic Topology, representing homotopy types as algebraic objects, is finally
reached.
Once the notion of simplicial set is available, most ingredients of algebraic
topology, classifying spaces, loop spaces, functional spaces, homology or cohomol-
ogy groups, any sort of operators between these groups can be more or less easily
described in the framework of simplicial sets. The initial essential step in this
direction was the discovery by Daniel Kan [12] of a purely combinatorial definition
of homotopy groups. The end of these notes shows a few typical examples of sim-
plicial descriptions, mainly to prepare the readers to the lecture about Operadic
Algebraic Topology.

2 Simplicial complexes.
2.1 Definitions.
Definition 1 — A simplicial complex is a pair (V, S) satisfying the properties:

• V , the set of vertices, and S, the set of simplices, are. . . sets, possibly infinite.

• Every simplex σ ∈ S is a non-empty finite set of vertices: σ = {v0 , . . . , vn };

such a simplex is called an n-simplex, the integer n ≥ 0 is the dimension of
the simplex σ. This simplex spans the vertices v0 , . . . , vn .

• For every vertex v ∈ V , the 0-simplex {v} is an element of S.

• For every simplex σ = {v0 , . . . , vn } ∈ S, every m-sub-simplex {vi0 , . . . , vim }

is also an element of S.

For example, let us consider the simplicial complex (V, S) with:

• V = {0 . . . 5}, the integers from 0 to 5.

• S = {0, 1, 2, 3, 4, 5, 01, 02, 12, 23, 34, 35, 45, 345} where 35 for example is a
shorthand for {3, 5}.

Such a simplicial complex is an “abstract” version of the geometrical object:

0• •4
2 3
• •

1• •5

3
 The triangle 012 is hollow, because {0, 1, 2} is not a simplex; on the contrary,
{3, 4, 5} is a simplex and the triangle 345 is filled. In the simplicial complex game,
you have a box with an arbitrary number of available vertices (0-simplices), edges
(1-simplices), triangles, (2-simplices), tetrahedons (3-simplices) and more generally
of n-simplices. Every vertex is labeled by the corresponding element of V and the
simplices of S describe what collections of vertices are spanned by a simplex.
No geometry in this definition; in particular, at this level, a simplex is just an
“abstract” set of vertices, which, when we will geometrically realize in a moment
a simplicial complex, will finally produce an ordinary geometrical simplex.

Definition 2 — Let K = (V, S) be a simplicial complex. The geometrical re-

alization |K| of K is defined as follows: |K| is the set of the indexed families
x = (xσ )σ∈S ∈ [0, 1](S) satisfying the conditions:

• {σ st xσ > 0} ∈ S and in particular is finite;

P
• xσ = 1.

Any topology over [0, 1](S) defines a topology over |K|, but combinatorial topol-
ogy most often is not concerned by such a topology: the combinatorial game is
enough to model, up to homotopy, in this way most “sensible” topological spaces.

2.2 Simple examples.

Let V be an arbitrary set of vertices, possibly infinite. Then the simplex generated
by V , denoted by ∆V , is the simplicial complex (V, S) where S = P∗,f (V ) is the
set of finite non-empty subsets of V . If V = n := {0, 1, . . . , n}, then ∆n is usually
simply denoted by ∆n = (n, P∗ (n)), it is the standard (abstract) n-simplex, and
its realization |∆n | is the common geometrical n-simplex. If V is infinite, then
the simplicial complex ∆V has simplices of arbitrary high dimension, but every
simplex of ∆V has a finite dimension.
The standard model for the n-sphere S n as a simplicial complex is:

S n = (n + 1, P∗ (n + 1) − {n + 1}). (1)

It is the standard n + 1-simplex ∆n+1 from which the maximal simplex n + 1 =

{0, . . . , n + 1} has been removed: think the standard simplex ∆n+1 is solid and
you may so imagine our model for the n-sphere is on the contrary a hollow (n + 1)-
simplex, in other words the boundary of an (n + 1)-simplex. Its realization is
homeomorphic to the boundary of an (n + 1)-disk (or cell, or ball), that is, a
topological n-sphere.
Many topological constructions can be simulated in the framework of simplicial
complexes. For example, if K = (V, S) and K 0 = (V 0 , S 0 ) are two simplicial com-
plexes with base point, that is, two vertices v0 ∈ V and v00 ∈ V 0 are
`distinguished,
then the wedge K ∨ K is defined by K = (V , S ) with V = (V V 0 )/(v0 ∼ v00 )
0 00 00 00 00

4
and S 00 = (S S 0 )/ ∼ where the last relation ∼ identifies any occurence of v0
`
in an element of S with any occurence of v00 in an element of S 0 . Both simplicial
complexes are “attached” at their respective base vertices.
A common construction is however surprisingly difficult to be translated in
the framework of simplicial complexes, namely the product construction. The
difficulty is the following: the elementary piece in the world of simplicial complexes
is a simplex, a point in dimension 0, an edge in dimension 1, a solid triangle in
dimension 2, a tetrahedron in dimension 3, an n-simplex in dimension n. But
the product of two edges is a square, which can be presented as the union of two
triangles, if you cut this square along a diagonal; but two diagonals in a square
and how to choose the right one? A little more difficult, the product of an edge
by a (solid) triangle is a triangular prism which can be presented as the union of
three tetrahedrons, a process neither easy nor deterministic. We will
see later the
product of an m-simplex by an n-simplex can be divided in m+n m simplices of
dimension (m + n), by a process not so obvious, made “automatic” if you work in
the framework of simplicial sets.

2.3 Simplicial maps and homotopy.

Definition 3 — Let K = (V,S) and K 0 = (V 0 , S 0 ) be two simplicial complexes.
A simplicial map f : K → K 0 is a set map f : V → V 0 satisfying the property: for
every simplex σ ∈ S, the image f (σ) is a simplex f (σ) ∈ S 0 .

It is not required f : V → V 0 is injective, and the image of an n-simplex σ ∈ S

can be a simplex f (σ) ∈ S 0 of dimension < n.
Is it possible to define homotopies between simplicial maps? First, let us con-
sider the traditional notion of homotopy between continuous maps.

Definition 4 — Two continuous maps f0 , f1 : X → Y between the topological

spaces X and Y are homotopic if there exists a (continuous) map F : [0, 1]×X → Y
satisfying:
F (0, x) = f0 (x)
(2)
F (1, x) = f1 (x)
for every x ∈ X.

f1
X

f0
Y
[0, 1] X × [0, 1]
F
X

5
Definition 5 — Let f0 , f1 : K = (V, s) → K 0 = (V 0 , S 0 ) be two simplicial maps
between two simplicial complexes. The maps f0 and f1 are elementarily homotopic
if the following property is satisfied: for every simplex σ ∈ V , the union f0 (σ) ∪
f1 (σ) is a simplex of S 0 .

If the required property is satisfied, you can then, at the level of the geometrical
realizations, trivially interpolate the maps f0 and f1 by a continuous family of ft ’s,
for t running the interval [0, 1]. Note ft cannot be simplicially implemented except
for t = 0 or 1.
Definition 5 is natural but not at all satisfactory. Let us consider the situation
with K the interval K = ∆1 = (1, P∗ (1)) and K 0 = (2, S 0 ) with S 0 made of the
three vertices {0}, {1} and {2}, and only two edges {0, 1} and {0, 2}. Let us
consider also the maps f0 , f1 : K → K 0 defined by f0 (0) = f1 (0) = 0, f0 (1) = 1
and f1 (1) = 2. Then these maps are not elementarily homotopic though, in the
topological framework, they are homotopic.

•1
f0
1• •0 •0
f1
•2

This difficulty can be overcome as follows: you decide two simplicial maps f, g :
K → K 0 are homotopic if you can construct a chain f = f0 , f1 , . . . , fk−1 , fk = g
where two successive elements are elementarily homotopic. We let you construct
the simple chain of length 2 describing how the maps of the previous example are
homotopic. But for infinite simplicial complexes, such a solution is not satisfactory.
The technique of Kan simplicial sets allows to overcome this important obstacle,
at the cost of complex technicalities, complex but unavoidable.
It is possible also to define functional spaces in a combinatorial style. Because
the framework of simplicial complexes will be soon given up, we show only a typical
example: how to define the loop space of a pointed simplicial complex (K, ∗), the
base point ∗ being a distinguished vertex of K? Usually a loop γ : [0, 1] → (X, ∗)
in a pointed topological space is a continuous map γ : [0, 1] → X satisfying
γ(0) = γ(1) = ∗. How to copy this notion for simplicial complexes?
The interval [0, 1] is (the realization of) a simplicial complex and the notion
of simplicial map γ : [0, 1] → K makes sense; but combined with the condition
γ(0) = γ(1) = ∗, only one such loop, the trivial constant loop at the base point,
not very satisfactory!
To overcome this obstacle, instead of the simple interval [0, 1], let us consider
the (infinite) simplicial complex I = (N, S) with S = {{n}}n∈N ∪ {{n, n + 1}}n∈N .

I
• • • •
0 1 2 3

6
 First it is natural to decide a loop γ : I → K is a simplicial map satisfying:

• γ(0) = ∗ ;

• For every n ≥ some n0 , γ(n) = ∗.

Our loop starts from the base point, runs various edges of K, and after the time n0 ,
remains fixed at the base point.
Then the loop space ΩK can be naturally defined as a simplicial complex as
follows: ΩK = (Λ, SΛ ) with:

• Λ is the set of loops as just defined;

• A finite set of loops {λ0 , . . . , λn } is an element of SΛ , that is, a simplex of ΩK,

if and only if, for every integer t > 0, the set {λ0 (t − 1), . . . , λn (t − 1)} ∪
{λ0 (t), . . . , λn (t)} is a simplex of K.

The last condition claims that it is possible to interpolate in a barycentric style the
loops λ0 , . . . , λn for every point of the “geometrical” simplex intuitively spanned by
these loops; if the condition is satisfied, we therefore decide to install an “abstract”
simplex between these vertices. It is an interesting exercise of topology to prove
the realization |ΩK| actually has the same homotopy type as the (topological)
loop space Ω(|K|), but this will not be necessary in these notes.
For example if K = S 2 modelled as the boundary of the standard 3-simplex:
K = (0..3, {0, 1, 2, 3, 01, 02, 03, 12, 13, 23, 012, 013, 023, 123}), let us consider the
loops γ0 = 0 → 1 → 2 → 0, γ1 = 0 → 1 → 3 → 0 and γ2 = 0 → 2 → 3 → 0, with
notations made obvious by the figure:

3 3 3
• • •

γ2
γ1
0• •2 0• •2 0• •2
γ0
• • •
1 1 1

Then {γ0 , γ1 }, {γ0 , γ2 } and {γ1 , γ2 } are edges of ΩK, and {γ0 , γ1 , γ2 } is a tri-
angle of ΩK between these edges; drawing the loop which is the “center” of this
triangle is useful.

7
 3
•

0• •2

•
1

Note this “loop” is not an actual loop of our simplicial complex: such a loop is
allowed to run only the edges of our sphere, while our “loop” goes inside some
triangles. This claimed “loop” is only a geometric interpretation of the center of
the triangle of ΩK spanning the actual loops γ0 , γ1 and γ2 .
On the contrary, if γ1−1 is the same loop as γ1 but run in the reverse direction
(meaning?), then {γ0 , γ1−1 , γ2 } is not a triangle of ΩK, why?
Let us decide the base point ∗ ∈ ΩK is the trivial loop constant in 0. Then
∗ → γ0 → γ1 → ∗ is a loop of ΩK, in other words a vertex of ΩΩK =: Ω2 K.
Proving the last loop is not homotopic to the trivial loop is another story.

2.4 Simplicial complexes vs simplicial sets.

A ∆-morphism α : m → n can in particular be a face operator ∂im : m − 1 → m.
The corresponding X-operator X∂im : Xm → Xm−1 is also called the i-th face
operator in dimension m and is most often simply denoted by ∂im or ∂i when
the underlying simplicial set X is implicit. The same for a degeneracy operator
Xηim : Xm → Xm+1 , most often denoted by ηim or ηi . Because of Corollary 11, it is
enough to define the face and degeneracy operators X∂im and Xηim satisfying the
required coherence properties, to define the whole collection of morphisms {Xα }α .
Let us consider the simplest simplicial complex X, the realization of which is
(homeomorphic to) a circle S 1 = {(x, y) ∈ R2 st x2 + y 2 = 1}. Three vertices
and three edges are necessary: X = (V, S) with V = 2 = {0, 1, 2} and S =
{0, 1, 2, 01, 02, 12} where as usual 01 is a shorthand for {0, 1}.

2
•

0• •1

It is not possible to use only two vertices following the figure:

0• •1

for the only possibility to produce an edge consists in choosing a set of vertices;
so that it is possible to install only one edge between two given vertices and the

8
above figure cannot correspond to a simplicial complex. We will see that we do
not meet any problem when associating a simplicial set to the same figure, this
will be explained soon.
We could even consider the following figure:

0•

and observe that is is not possible in the framework of simplicial complexes to

install a “loop” edge from a vertex to itself. This is also possible for simplicial
sets.
We will see it is also possible to give a simplicial set with only two (non-
degenerate) simplices, a vertex 0 and a “triangle” 012, the three edges of which
being collapsed over the unix vertex.

0
× • ×

The realization of this simplicial set will be a triangle where the whole boundary
is identified to a point, that is, a 2-sphere.
More generally any n-sphere can be realized as a simplicial set with only one
vertex and one n-simplex; more precisely only these non-degenerate simplices, for
we will soon learn that any non-degenerate simplex generates an infinite collec-
tion of. . . degenerate simplices, non-visible on the figures, that is, “hidden” in the
geometric realization. For example the minimal simplicial complex correspond-
ing to a 4-sphere requires 6 vertices, 15 edges, 20 triangles, 15 tetrahedron and
6 4-simplices, while as a simplicial set, only one vertex and one 4-simplex are
enough as non-degenerate simplices.
These elementary examples show in general less (non-degenerate) simplices are
necessary to construct an object as a simplicial set than as a simplicial complex.
You can object an infinite number of degenerate simplices is also required, but
precisely these degenerate simplices will give much more flexibility in the con-
struction process. It is true the underlying technology is not obvious, but thanks
to this nice technology, the main parts of topology have a good translation into
the combinatorial world, allowing a constructivist to easily handle topology with
his computer.

3 Simplical sets.
Possible references for this fascinating subjects are:

9
 • [13]: Maybe the most useful reference for the serious user; only one drawback:
hardly any example, no didactic explanation! But many invaluable formulas
and detailed proofs can be found only in this book.
• [15]: See in particular Section VIII.5 of this book for a short introduction to
this subject, which is not the main goal of this book, but unavoidable.
• [14]: See Section §4.2.
• More modern, but also harder, references are [10], a book entirely devoted
to this subject, and also [9, I.2].

3.1 The category ∆.

Some strongly structured sets of indices are necessary to define the notion of
simplicial object; they are conveniently organized as the category ∆. An object
of ∆ is a set m, namely the set of integers m := {0, 1, . . . , m − 1, m}; this set is
canonically ordered with the usual order between integers.
A ∆-morphism α : m → n is an increasing map. Equal values are permitted;
for example a ∆-morphism α : 2 → 3 could be defined by α(0) = α(1) = 1
and α(2) = 3. The set of ∆-morphisms from m to n is denoted by ∆(m, n); the
subset of injective (resp. surjective) morphisms is denoted by ∆inj (m, n) (resp.
∆srj (m, n)).
Some elementary morphisms are important, namely the simplest non-surjective
and non-injective morphisms. For geometric reasons explained later, the first ones
are the face morphisms, the second ones are the degeneracy morphisms.

0 • • 0 0 • • 0
1 • • 1 1 • • 1

∂im = i−1 • • i−1 ηim = i−1 • • i−1

i • • i i • • i
• i+1 i+1 •

m−1 • • m
• m m+1 •

Definition 6 — The face morphism ∂im : m − 1 → m is defined for m ≥ 1 and

0 ≤ i ≤ m by:
∂im (j) = j if j < i,
m (3)
∂i (j) = j + 1 if j ≥ i.

The face morphism ∂im is the unique injective morphism from m-1 to m such
that the integer i is not in the image. The face morphisms generate the injective
morphisms, in fact in a unique way if a growth condition is required.

Proposition 7 — Any injective ∆-morphism α ∈ ∆inj (m, n) has a unique expres-

sion:
α = ∂inn ◦ . . . ◦ ∂im+1
m+1
(4)

10
satisfying the relation in > in−1 > . . . > im+1 .

♣ The index set {im+1 , . . . , in } is exactly the difference set n − α(m), that is, the
set of the integers where surjectivity fails. ♣
Frequently the upper index m of ∂im is omitted because clearly deduced from
the context. For example the unique injective morphism α : 2 → 5 the image of
which is {0, 2, 4} can be written α = ∂5 ∂3 ∂1 .
If two face morphisms are composed in the wrong order, they can be exchanged:
∂i ◦ ∂j = ∂j+1 ◦ ∂i if j ≥ i. Iterating this process allows you to quickly compute for
example ∂0 ∂2 ∂4 ∂6 = ∂9 ∂6 ∂3 ∂0 .

Definition 8 — The degeneracy morphism ηim : m + 1 → m is defined for m ≥ 0

and 0 ≤ i ≤ m by:
ηim (j) = j if j ≤ i,
m (5)
ηi (j) = j − 1 if j > i.

The degeneracy morphism ηim is the unique surjective morphism from m+1 to
m such that the integer i has two pre-images. The degeneracy morphisms generate
the surjective morphisms, in fact in a unique way if a growth condition is required.

Proposition 9 — Any surjective ∆-morphism α ∈ ∆srj (m, n) has a unique ex-

pression:
α = ηinn ◦ . . . ◦ ηim−1
m−1
(6)
satisfying the relation in < in+1 < . . . < im−1 .

♣ The index set {in , . . . , im−1 } is exactly the set of integers j such that α(j) =
α(j + 1), that is, the integers where injectivity fails. ♣
Frequently the upper index m of ηim is omitted because clearly deduced from
the context. For example the unique surjective morphism α : 5 → 2 such that
α(0) = α(1) and α(2) = α(3) = α(4) can be expressed α = η0 η2 η3 .
If two degeneracy morphisms are composed in the wrong order, they can be
exchanged: ηi ◦ ηj = ηj ◦ ηi+1 if i ≥ j. Iterating this process allows you to quickly
compute for example η3 η3 η2 η2 = η2 η3 η5 η6 .

Proposition 10 — Any ∆-morphism α can be ∆-decomposed in a unique way:

α=β◦γ (7)

with β injective and γ surjective.

♣ The intermediate ∆-object k necessarily satisfies k + 1 = Card(im(α)). The

growth condition then gives a unique choice for β and γ. ♣

11
Corollary 11 — Any ∆-morphism α : m → n has a unique expression:

α = ∂in ◦ . . . ◦ ∂ik+1 ◦ ηjk ◦ . . . ◦ ηjm−1 (8)

satisfying the conditions in > . . . > ik+1 and jk < . . . < jm−1 . ♣

Finally if face and degeneracy morphisms are composed in the wrong order,
they can be exchanged:

ηi ◦ ∂j = id if j = i or j = i + 1;
= ∂j−1 ◦ ηi if j ≥ i + 2; (9)
= ∂j ◦ ηi−1 if j < i.
All these commuting relations can be used to convert an arbitrary composition
of faces and degeneracies into the canonical expression:

α = η9 ∂6 η3 ∂7 η9 ∂8 η6 ∂2 η4 ∂9 = ∂7 ∂6 ∂2 η2 η4 η6 . (10)
This relation means the image of α does not contain the integers 2, 6 and 7, and
the relations α(2) = α(3), α(4) = α(5) and α(6) = α(7) are satisfied.
The previous propositions show any functor F from ∆ to another category is
entirely known when the image objects F (m) and the image morphisms F (∂im )
and F (ηim ) are given.

Corollary 12 — A contravariant functor X : ∆ → CAT is nothing but a

collection {Xm }m∈N of objects of the target category CAT, and collections of
CAT-morphisms {X(∂im ) : Xm → Xm−1 }m≥1 , 0≤i≤m and {X(ηim ) : Xm →
Xm+1 }m≥0 , 0≤i≤m satisfying the commuting relations:

X(∂i ) ◦ X(∂j ) = X(∂j ) ◦ X(∂i+1 ) if i ≥ j,

X(ηi ) ◦ X(ηj ) = X(ηj+1 ) ◦ X(ηi ) if j ≥ i,
X(∂i ) ◦ X(ηj ) = id if i = j, j + 1, (11)
X(∂i ) ◦ X(ηj ) = X(ηj−1 ) ◦ X(∂i ) if j > i,
X(∂i ) ◦ X(ηj ) = X(ηj ) ◦ X(∂i−1 ) if i > j + 1.

In the five last relations, the upper indices have been omitted. Such a con-
travariant functor is a simplicial object in the category CAT. If α is an arbitrary
∆-morphism, it is then sufficient to express α as a composition of face and degen-
eracy morphisms; the image X(α) is necessarily the composition of the images of
the corresponding X(∂i )’s and X(ηi )’s; the above relations ensure the definition is
coherent.

3.2 Simplicial sets: first definitions.

Definition 13 — Let Set be the category of sets. A simplicial set X is a sim-
plicial object in the Set category; that is, according to the previous section, a
contravariant functor X : ∆ → Set.

12
 This definition is short, but, because of the rich structure of the category ∆,
it is quite complex! You see defining a simplicial set X requires for every non-
negative integer n some object Xn of the Set category, in other words an ordinary
set, and for every ∆-morphism α : m → n some Set-morphism, that is an ordinary
map Xα : Xn → Xm ; furthermore, the set {Xα }α∈∆−morphisms must satisfy the
coherence relations Xα Xβ = Xβα when the composition βα makes sense.
The geometric interpretation of this definition is not obvious, but once under-
stood, this notion is terribly powerful. A power which deserves a little significant
work to reach its marvelous possibilities. Before seriously studying this notion, let
us give a few comments about the comparison between simplicial complexes and
simplicial sets.
A simplicial set X is a simplicial object in the category of sets, and therefore is
given by a collection of sets {X(m)}m∈N and collections of maps {Xα }, the index α
running the ∆-morphisms; the usual coherence properties must be satisfied. As
explained at the end of Section 4, it is sufficient to define the X(∂im )’s and the
X(ηim )’s with the corresponding commuting relations.
The set X(m) is usually denoted by Xm and is called the set of m-simplices
of X; such a simplex has the dimension m. To be a little more precise, these
simplices are sometimes called abstract simplices, to avoid possible confusions with
the geometric simplices defined a little later. An (abstract) m-simplex is only one
element of Xm .
If α ∈ ∆(n, m), the corresponding morphism X(α) : Xm → Xn is most often
simply denoted by α∗ : Xm → Xn or still more simply α : Xm → Xn . In particular
the faces and degeneracy operators are maps ∂i : Xm → Xm−1 and ηi : Xm →
Xm+1 . If σ is an m-simplex, the (abstract) simplex ∂i σ is its i-th face, and the
simplex ηi σ is its i-th degeneracy; we will see the last one is “particularly” abstract.

3.3 The structure of simplex sets.

Definition 14 — An m-simplex σ of the simplicial set X is degenerate if there
exist an integer n < m, an n-simplex τ ∈ Xn and a ∆-morphism α ∈ ∆(m, n)
such that σ = α(τ ). The set of non-degenerate simplices of dimension m in X is
ND
denoted by Xm .

Decomposing the morphism α = β ◦ γ with γ surjective, we see that σ =

γ(β(τ )), with the dimension of β(τ ) less or equal to n; so that in the definition of
degenaracy, the connecting ∆-morphism α can be required to be surjective. The
relation σ = α(τ ) with α surjective is shortly expressed by saying the m-simplex
σ comes from the n-simplex τ .
Eilenberg’s lemma explains each degenerate simplex comes from a canonical
non-degenerate one, and in a unique way.

Lemma 15 — (Eilenberg’s lemma) If X is a simplicial set and σ is an m-

simplex of X, there exists a unique triple Tσ = (n, τ, α) satisfying the following
conditions:

13
 1. The first component n is a natural number n ≤ m;
2. The second component τ is a non-degenerate n-simplex τ ∈ XnN D ;
3. The third component α is a ∆-morphism τ ∈ ∆srj (m, n);
4. The relation σ = α(τ ) is satisfied.

Definition 16 — This triple Tσ is called the Eilenberg triple of σ.

♣ Let T be the set of triples T = (n, τ, α) such that n ≤ m, τ ∈ Xn and

α ∈ ∆(m, n) satisfy σ = α(τ ). The set T certainly contains the triple (m, σ, id)
and therefore is non empty. Let (n0 , τ0 , α0 ) be an element of T where the first com-
ponent, the integer n0 , is minimal. We claim (n0 , τ0 , α0 ) is the Eilenberg triple.
Certainly n0 ≤ m. The n0 -simplex τ0 is non-degenerate; otherwise τ0 = β(τ1 )
with the dimension n1 of τ1 less than n0 , but then (n1 , τ1 , βα0 ) would be a triple
with n1 < n0 . Finally α0 is surjective, otherwise α0 = βγ with γ ∈ ∆srj (m, n1 ) and
n1 < n0 ; but again the triple (n1 , β(τ0 ), γ) would be a triple denying the required
property of n0 . The existence of an Eilenberg triple is proved and uniqueness
remains to be proved.
Let (n1 , τ1 , α1 ) be another Eilenberg triple. The morphisms α0 and α1 are
surjective and respective sections β0 ∈ ∆inj (n0 , m) and β1 ∈ ∆inj (n1 , m) can be
constructed: α0 β0 = id and α1 β1 = id. Then τ0 = (α0 β0 )(τ0 ) = β0 (α0 (τ0 )) =
β0 (σ) = β0 (α1 (τ1 )) = (α1 β0 )(τ1 ); but τ0 is non-degenerate, so that n1 = dim(τ1 ) ≥
n0 = dim(τ0 ); the analogous relation holds when τ0 and τ1 are exchanged, so that
n1 ≤ n0 and the equality n0 = n1 is proved.
The relation τ0 = β0 (α1 (τ1 )) with τ0 non-degenerate implies α1 β0 = id, other-
wise α1 β0 = γδ with δ ∈ ∆srj (n1 , n2 ) and n2 < n1 = n0 , but this implies τ0 comes
from γ(τ1 ) of dimension n2 again contradicting the non-degeneracy property of τ0 ;
therefore α1 β0 = id but this equality implies τ0 = τ1 .
If α0 6= α1 , let i be an integer such that α0 (i) = j 6= α1 (i); then the section β0
can be chosen with β0 (j) = i; but this implies (α1 β0 )(j) 6= j, so that the relation
α1 β0 = id would not hold. The last required equality α0 = α1 is also proved. ♣
Each simplex comes from a unique non-degenerate simplex, and conversely, for
ND
any non-degenerate m-simplex σ ∈ Xm , the collection {α(σ) ; α ∈ ∆srj (n, m) ;
n ≥ m} is a perfect description of all simplices coming from σ, that is, of all
degenerate simplices above σ. This is also expressed in the following formula,
describing the structure of the simplex set of any simplicial set X:
a a a a
Xm = ∆srj (n, m)(σ). (12)
m∈N m∈N ND
σ∈Xm n≥m

In particular a 0-simplex v ∈ X0 is always non-degenerate, it is called a vertex,

and such a vertex generates for every positive dimension n exactly one degenerate
simplex vn = η ∗ v where η is the unique element of ∆srj (n, 0).
The following figures try to explain a little the nature of the collection of
degenerate simplices associated to a non-degenerate simplex σ. If σ ∈ X0N D is a 0-
simplex, only one degeneracy in every positive dimension d, namely ηd−1 · · · η1 η0 σ.

14
The deneneracy operator can be represented by the sequence 0 · · · 0, meaning the
∆-morphism ηd−1 · · · η1 η0 ∈ ∆(d, 0) sends every element of d over 0. This can be
represented as follows:

0 η0 00 η1 000
• • •
σ

Note the expression of an arrow as one degeneracy operator in general is not

unique. For example, the η1 above between 00 and 000 could be replaced by η0 .
But our choice directly gives the canonical expression 000 = η1 η0 (0).
If σ ∈ X1N D has dimension 1, then two degeneracies in dimension 2, three in
dimension 3, and so on.

01111
•
η3
0111 00111
• •
η2 η3
011 0011 00011
• • •
η1 η2 η3
01 η0 001 η1 0001 η2 00001
• • • •
σ

On this diagram, some arrows are labeled, others not. Those which are labeled
constitute a tree rooted at σ giving the canonical expression of a degenerate simplex
from the initial one σ.
Continuing in the same way for an initial non-degenerate simplex σ of dimen-
sion 2 produces the following diagram.

01222
•
η3
0122 01122
• •
η2 η3
01112
•
012 η1 0112 η2
• •
σ 00122
η0 •
0012 η3 00112
• η2 •
η1
00012
•

with the same kind of analysis.

15
4 First examples.
4.1 Discrete simplicial sets.
Definition 17 — A simplicial set X is discrete if Xm = X0 for every m ≥ 1, and
if for every α ∈ ∆(m, n), the induced map α∗ : Xn → Xm is the identity.

The reason of this definition is that the realization (see Section 5) of such
a simplicial set is the discrete point set X0 ; the Eilenberg triple of any simplex
σ ∈ Xm = X0 is (0, σ, η) where the map η is the unique element of ∆(m, 0); the
only non-degenerate simplices are the vertices, the elements of X0 .

4.2 The simplicial complexes.

A simplicial complex K = (V, S) is a pair where the first component V , the vertex
set is an arbitrary “set”; the second component S, the simplex set, is made of finite
subsets of V satisfying a few coherence properties, as explained in Definition 2.
The simplicial complex K = (V, S) is ordered if the vertex set V is provided
with a total order1 . Then a simplicial set, abusively again denoted by K, is
canonically associated; the simplex set of m-dimensional simplices Km in this
new framework is the set of increasing maps σ : m → K such that the image of
m is an element of S; note that such a map σ is not necessarily injective. If α
is a ∆-morphism α ∈ ∆(n, m) and σ is an m-simplex σ ∈ Km , then α(σ) is
naturally defined as α(σ) = σ ◦ α. A simplex σ ∈ Km is non-degenerate if and
only if σ ∈ ∆inj (m, V ); if σ ∈ Km = ∆(m, V ), the Eilenberg triple (n, τ, α) satisfies
σ = τ ◦ α with α surjective and τ injective.
ND
The non-degenerate m-simplices Km is the set of injective increasing maps
m → V where the image is an m-simplex of the initial simplicial complex. There is
so a natural 1-1 correspondance between the m-simplices of the initial simplicial
complex and the non-degenerate m-simplices of the associated simplicial set. The
role of the degenerate simplices will be explained later.
This in particular works for K = (d, P(d)) the simplex freely generated by d
provided with the canonical vertex order. We obtain in this way the canonical
structure of simplicial set for the standard d-simplex ∆d . Its set of m-simplices
∆dm is the set of increasing maps ∆dm = ∆(m, d); the non-degenerate simplices cor-
respond to the injective maps in ∆inj (m, d); in particular, only one non-degenerate
simplex in dimension d, namely idd ∈ ∆(d, d), the fundamental simplex of ∆d .
This section implies the category of simplicial complexes is essentially embed-
ded inside the category of simplicial sets, at least if you forget this matter of order
over the vertices, necessary to obtain a simplicial set. The Zermelo theorem en-
sures such an order over the vertex set V of the initial complex is always possible,
but this matter of order plays a major role in the continuation of the story: such
1
Other situations where the order is not total are also interesting but will be considered later.

16
an order is most often non-natural and the consequent punishment is not far:
these non-natural orders are at the origin of the role of the symmetric groups in
the operadic theories. In a sense, the simplicial set theory succeeds in hiding the
essential role of the symmetric groups in our geometrical space. But the revenge
of the symmetric groups will be terrible: you rejected the symmetric groups at
the geometrical level? Yes, but they will appear again in the algebraic framework
later: under the notion of E∞ -operad.
The category of simplicial sets is designed to allow more flexible combinato-
rial construction processes than those that are possible in the the framework of
simplicial complexes, as roughly explained in Section 2.4.

4.3 The spheres.

Let d be a natural number. The simplest version S = S d of the d-sphere`as srj
a simpli-
cial set is defined as follows: the set of m-simplices Sm is Sm = {∗m } ∆ (m, d);
if α ∈ ∆(n, m) and σ is an m-simplex σ ∈ Sm , then α(σ) depends on the nature
of σ:

1. If σ = ∗m , then α(σ) = ∗n ;

2. Otherwise σ ∈ ∆srj (m, d) and if σ ◦ α is surjective, then α(σ) = σ ◦ α, else

α(σ) = ∗n (the emergency solution when the natural solution does not work).

This is nothing but the canonical quotient, in the simplicial set framework, of
two simplicial complexes S d = ∆d /∂∆d , at least if d > 0; see the figure p. 9 which
illustrates how the 2-sphere can be understood as the quotient S 2 = ∆2 /∂∆2 .
More generally the notion of simplicial subset is naturally defined and a quotient
then appears. In the case of the construction of S d = ∆d /∂∆d , the subcomplex
∂∆d is made of the simplices α ∈ ∆(m, d) that are not surjective.
The Eilenberg triple of ∗m is (0, ∗0 , α) where α is the unique element of ∆(m, 0).
The Eilenberg triple of σ ∈ ∆srj (m, d) ⊂ Sm is (d, id, σ). There are only two non-
degenerate simplices, namely ∗0 ∈ S0 and id(d) ∈ Sd , even if d = 0.

5 Realization.
5.1 Definition and first results.
Before giving other examples of simplicial sets, it is time now to examine the notion
of realization in the framework of the category of simplicial sets.
Let X = ({Xm }m , {Xα }α ) be a simplicial set; the index m runs the non-
negative integers N; the index α runs the ∆-morphisms: a possible α is an increas-
ing map α : m → n.

17
Definition 18 — The (“expensive”) realization |X| of X is:
a
|X| = Xm × |∆m | / ≈ . (13)
m∈N

Each component of the coproduct is the product of the discrete set of m-

simplices Xm by the standard geometric m-simplex |∆m |, that is, the usual topo-
logical m-simplex; in other words, each “abstract” simplex σ in Xm gives birth to
a geometric simplex {σ} × |∆m |, and they are attached to each other following the
instructions of the equivalence relation ≈, to be defined. Let α ∈ ∆(m, n) be some
∆-morphism, and let σ be an n-simplex σ ∈ Xn and t ∈ |∆m | ⊂ Rm . Then the
pairs (α∗ σ, t) and (σ, α∗ t) are declared equivalent. Here α∗ : |∆m | → |∆n | is the
(affine) geometrical map covariantly induced between geometrical simplices by the
“abstract” map α : m → n between the vertices of these simplices, according to
the usual numbering. The map α∗ : Xn → Xm is induced by the simplicial struc-
ture: α∗ = X(α); as usual, the sup-∗ intends to recall the contravariant nature
of the association process. Frequently we omit the sub-∗ or the sup-∗ when the
context clearly implies it.
It is not obvious to understand what is the topological space so obtained. A
description a little more explicit but also a little more complicated explains more
satisfactorily what should be understood.
The cheap realization kXk of the simplicial set X is:
a
ND
kXk = Xm × |∆m | / ≈ (14)
m∈N

where the equivalence relation ≈ is defined as follows. Let σ be a non-degenerate

m-simplex and i an integer 0 ≤ i ≤ m; let also t ∈ |∆m−1 |; the abstract (m − 1)-
simplex ∂i∗ σ has a well defined Eilenberg triple (n, τ, α); then we decide to declare
equivalent the pairs (σ, ∂i∗ (t)) ≈ (τ, α∗ (t)).
Fewer simplices are invoked in the cheap realization: only the non-degenerate
simplices are used, but the equivalence relation assembling them to each other is
more sophisticated.
For example let S = S d be the claimed simplicial version of the d-sphere de-
scribed in Section 4.3. This simplical set S has only two non-degenerate simplices,
one in dimension 0, the other one in dimension d. The cheap realization ||S|| needs
a point |∆0 | = {∗} and a geometric d-simplex |∆d | corresponding to the abstract
simplex id ∈ ∆(d, d); then if t ∈ |∆d−1 | and 0 ≤ i ≤ d, the equivalence relation
asks for the Eilenberg triple of ∂i (id) = ∗d−1 which is (0, ∗0 , η), the map η being the
unique element of ∆(d − 1, 0). Finally the initial pair (id, ∂i∗ t) in the realization
process must be identified with the pair (∗0 , ∆0 ); in other words kSk = |∆d |/∂|∆d |,
homeomorphic to the unit d-ball with the boundary collapsed to one point: the
result is clearly a (d − 1)-sphere.
You observe in this simple example about spheres the role of the degenerate
simplices. Let us now consider the expensive realization |S| of the simplicial set

18
S, simplicial model of the 2-sphere S 2 . In the rough description page 9 of a 2-
sphere as a simplicial set, we explained we would like to attach the whole boundary
∂|∆2 | of the triangle |∆2 |, for example its 0-face ∂0 |∆2 | to the base point ‘∗’. Two
∆-morphisms are invoked in the necessary attachment process:

• The unique map η : 1 → 0 is surjective non-injective. The contravariant

functor which defines the simplicial set S has in particular a map η ∗ : S0 → S1
which associates to the base point ∗ ∈ S0 , in fact the unique element of S0 ,
the degenerate 1-simplex ∗1 ∈ S1 .

• The face map ∂02 : 1 → 2, that is, the unique injective map which avoids 0
(see page 10), applied to the 2-simplex id2 ∈ S2 , gives again ∂02∗ (id2 ) = ∗1 ,
see in Section 4.3 the Rule 2 for the simplicial description of S n .

∂02
(∗, ηt)
η × ×
(∗1 , t) (id2 , ∂02 t)
•

∗ × ∆0 ∗1 × ∆1 id2 × ∆2

Now let t ∈ |∆1 |. In the expensive realization process, we must identify:

S0 × |∆0 | 3 (∗, 0) = (∗, ηt) ∼ (η ∗ ∗, t) = (∗1 , t) ∈ S1 × |∆1 |, (15)

and we must also identify:

S2 × |∆2 | 3 (id2 , ∂02 t) ∼ (∂02∗ id2 , t) = (∗1 , t) ∈ S1 × |∆1 |. (16)

Finally we may forget the point (∗1 , t) and directly identify (∗, 0) ∼ (id2 , ∂12 t).
This being valid for every t ∈ |∆1 |, and for ∂12 and ∂22 as well, finally the whole
boundary ∂|∆2 | is identified to the base point ∗.
In this way, any point (σ, t) of the expensive realization, where the (abstract)
simplex component σ is degenerate, can be canonically replaced by the point (τ, αt)
in the same realization if (n, τ, α) is the Eilenberg triple of σ, where τ is a non-
degenerate simplex. The non-degenerate simplices finally do not contribute in
the realization, but they are the necessary intermediary objects to describe the
possibly sophisticated attachments.

Proposition 19 — Both realizations, the expensive one and the cheap one, of a
simplicial set X are canonically homeomorphic.

♣ The homeomorphism f : |X| → kXk to be constructed maps the equivalence

class of the pair (σ, t) ∈ Xm × ∆m to the (equivalence class of the) pair (τ, α∗ (t)) ∈
Xn × ∆n if the Eilenberg triple of σ ` is (n, τ, α). The inverse homeomorphism g
ND
× ∆m ,→ Xm × ∆m . These maps
`
is induced by the canonical inclusion Xm

19
must be proved coherent with the defining equivalence relations and inverse to
each other.
If α = βγ is a ∆-morphism expressed as the composition of two other ∆-
morphisms, then an equivalence (σ, β∗ γ∗ t) ≈ (γ ∗ β ∗ σ, t) can be considered as a
consequence of the relations (σ, β∗ γ∗ t) ≈ (β ∗ σ, γ∗ t) and (β ∗ σ, γ∗ t) ≈ (γ ∗ β ∗ σ, t), so
that it is sufficient to prove the coherence of the definition of f with respect to the
elementary ∆-operators, that is, the face and degeneracy operators.
Let us assume the Eilenberg triple of σ ∈ Xm is (n, τ, α), so that f (σ, t) =
(τ, α∗ t). We must in particular prove that f (ηi∗ σ, t) and f (σ, ηi∗ t) are coherently de-
fined. The second image is the equivalence class of (τ, α∗ ηi∗ t); the Eilenberg triple
of ηi∗ σ is (n, τ, αηi ) so that the first image is the equivalence class of (τ, (αηi )∗ t)
and both image representants are even equal.
Let us do now the analogous work with the face operator ∂i instead of the
degeneracy operator ηi . Two cases must be considered. If ever the composition
α∂i ∈ ∆(m − 1, n) is surjective, the proof is the same. The interesting case happens
if α∂i is not surjective; but its image then forgets exactly one element j (0 ≤ j ≤ n)
and there exists a unique surjection β ∈ ∆(m − 1, n − 1) such that α∂i = ∂j β. The
abstract simplex ∂j∗ τ gives an Eilenberg triple (n0 , τ 0 , α0 ) and the unique possible
Eilenberg triple for ∂i∗ σ is (n0 , τ 0 , βα0 ). Then, on one hand, the f -image of (σ, ∂i∗ t)
is (τ, α∗ ∂i∗ t) = (τ, ∂j∗ β∗ t); on the other hand the f -image of (∂i∗ σ, t) is (τ 0 , α∗ β∗ t);
but according to the definition of the equivalence relation ≈ for kXk, both f -
images are equivalent. The coherence of f is proved.
ND
Let σ ∈ Xm , 0 ≤ i ≤ m, t ∈ ∆m−1 and (n, τ, α) (the Eilenberg triple
∗
of ∂i σ) be the ingredients in the definition of the equivalence relation for kXk;
the pairs (σ, ∂i∗ t) and (τ, α∗ t) are declared equivalent in kXk; the map g is
induced by the canonical inclusion of coproducts, so that we must prove the
same pairs are also equivalent in |X|. But this is a transitive consequence of
(σ, ∂i∗ t) ≈ (∂i∗ σ, t) = (α∗ τ, t) ≈ (τ, α∗ t). We see here we had only described the
binary relations generating the equivalence relation ≈; the defining relation is not
necessarily stable under transitivity. The coherence of g is proved.
The relation f g = id is obvious. The other relation gf = id is a consequence of
the equivalence in |X| of (σ, t) ≈ (τ, α∗ t) if the Eilenberg triple of σ is (n, τ, α). ♣

5.2 Simplicial model for classifying spaces.

5.2.1 The general case of a discrete group.

Illustrating the notion of realization with the classifying spaces of discrete groups
is interesting. This construction can be extended to any arbitrary simplicial group,
see [13, §21].

Definition 20 — Let G be a discrete group, possibly non commutative; the unit

of G is denoted 1. The classifying space BG of G is the simplicial set defined as
follows:

20
 • The simplex set BGm of m-dimensional simplices is BGm = Gm , the
elements of which are called “m-bars” and are traditionally denoted by
σ = [g1 | · · · |gm ]: the separator ‘|’, a bar, is here preferred to the common
comma for clarity.

• Face and degenerator operators are defined by:

∂0 [g1 | · · · |gm ] := [g2 | · · · |gm ];

∂m [g1 | · · · |gm ] := [g1 | · · · gm−1 ];
(17)
∂i [g1 | · · · |gm ] := [g1 | · · · |gi−1 |gi gi+1 |gi+2 | · · · |gm ];
ηi [g1 | · · · |gm ] := [g1 | · · · |gi |1|gi+1 | · · · |gm ].

In particular BG0 = {[ ]} has only one element.

The m-simplex [g1 | . . . |gm ] is degenerate if and only if one of the G-components
is the unit element.
The various commuting relations must be verified; this works but does not
give obvious indications on the very nature of this construction; in fact there is
a more conceptual description. Let us define the simplicial set EG by EGm =
Set(m, G) = Gm ∼ = Gm+1 , that is, the maps of m to G without taking account of
the ordered structure of m (the group G is not ordered); if α ∈ ∆(n, m) there is
a canonical way to define α : EGm → EGn ; it would be fairly coherent to write
EG = G∆ .
There is a canonical left action of the group G on EG, and BG is the natural
quotient of EG by this action. A simplex σ ∈ EGm is nothing but a (m + 1)-
tuple (g0 , . . . , gm ) and the action of g gives the simplex (gg0 , . . . , ggm ). If two sim-
−1
plices are G-equivalent, the products gi−1 gi are the same; the quotient BG-simplex
[g1 , . . . , gm ] denotes the equivalence class of all the EG-simplices (g, gg1 , gg1 g2 , . . .),
which can be imagined as a simplex where the edge between the vertices i−1 and i
(i > 0) is labeled by gi to be considered as a (right) operator between the adjacent
vertices. Then the boundary and degeneracy operators are clearly explained and
it is even not necessary to prove the commuting relations, they can be deduced of
the coherent structure of EG.

5.2.2 BZ2 is a real projective space.

Let us examine what happens for the smallest non-trivial particular case, that
is, the group G with two elements G = Z/2Z =: Z2 ; it is a commutative group
and we then prefer to denote 0 the “unit”. For a bar [g1 | · · · |gm ], two choices
only for a component gi , and the choice gi = 0 implies the simplex is degenerate.
So that finally exactly one non-degenerate simplex for every dimension: GN m
D
=
{[1| · · · |1]}.
Let us carefully examine the beginning of the construction of the realization
BZ2 . The key point is in the next figure.

21
 [ ]1 2
• [1] •

[ ]• [1] 1• [1|1] [ ] = • • = P 2R

BG0 • BG1 [1] •

[ ]0
BG2 0

Only one non-degenerate 1-simplex [1], an interval, both ends beeing identified
to the unique 0-simplex [ ]: the 1-skeleton is a circle, to be understood in fact
as the projective line P 1 R = S 1 /Z2 , that is, the circle where opposite points are
identified, which is again a circle!
Only one non-degenerate 2-simplex [1|1], a triangle, with the faces ∂0 = [1],
∂1 = [0] and ∂2 = [1]. The faces 0 and 2 are the non-degenerate 1-simplex, and the
face 1 is degenerate, therefore collapsed over the base point, the unique vertex [ ];
after this collapsing, there “remains” in fact only two faces for our 2-simplex, both
being identified with our 1-simplex [1]. Examining carefully the orientations of
the faces of our triangle shows finally the 2-skeleton of our realization |BZ2 | is the
2-dimensional real projective space.
More generally the m-skeleton is the m-dimensional real projective space P m R
and the total realization |BG| is the inductive limit, the infinite real projective
space P ∞ R.
In the same way, |EG| is the infinite real sphere S ∞ and |BG| is nothing but
the quotient of this sphere by the antipodal action of Z2 .

6 Simplicial homology.
In Section 5, the strange geometric role of the degenerate simplices in a simplicial
set has been described. It is therefore a good opportunity to introduce now the
subject of simplicial homology, where the role, or rather the absence (!) of role of
the degenerate simplices is also crucial.
For the most elementary notions of homological algebra, many textbooks are
available. The lecture notes [20, Section 2] of another Summer School gives a
careful self-contained exposition of the most elementary parts of this subject. A
useful reference, for a more extended knowledge in this rich area, is [15].

6.1 Basic definitions.

Definition 21 — Let R be a unitary commutative ring, called the coefficient
ring. Let X = (Xn , ∂i , ηi ) be a simplicial set. The R-chain complex associated to
X is the object C∗ (X, R) = (Cm (X, R), dm )m∈Z defined as follows:

22
 • The chain group Cm (X, R) is the null group for i < 0, and the free R-module
generated by the m-simplices Xm of X if i ≥ 0.

• The differential dm : Cm (X, R) → Cm−1 (X, R) is the R-linear map defined

by:
m
X
dm (σ) = (−1)i ∂i σ. (18)
i=0

when σ ∈ Xm .

In this definition, and in general in Homological Algebra, many indices, many

index sets, are omitted, and the reader is assumed to be able to deduce them
from context. The beginners do not like these conscious omissions, but experience
shows it is necessary if you want to avoid terribly cumbersome notations, quickly
making awkward formulas and diagrams. It is even an art in this activity to
select in every situation the right indices to be displayed, and the others to keep
hidden and underlying. It is also a fruitful activity for the reader to systematically
elucidate what are the missing indices, to be sure of one’s understanding.
For example (Xm , ∂i , ηi ) should in principle be displayed as:

({Xm }m∈N , {∂im }m≥1,0≤i≤m , {ηim }i∈N,0≤i≤m ). (19)

Taking account of the very definition of a simplicial set, the reader should admit
there is a unique way to complete the first formula, a little elliptic, to obtain the
second one, where everything is described. And the first formula is in fact so
explicit, at least if you know the underlying definitions, that it is widely preferred.
In the most elementary situations, the coefficient ring R is usually the integer
ring R = Z. Otherwise, some essentially constant coefficient ring is most often
given, which allows to frequently omit the coefficient ring and to simply write
Cm (X, R) = Cm (X).
For example, let us consider the 1-circle S 1 = S defined as in Section 4.3. Then
` srj
Sm = {∗m } ∆ (m, 1). Let us detail the chain groups Cm (S, Z) = Cm (S) for
m = 0, . . . , 3. We must firstly describe the simplices of S0 , S1 , S2 and S2 and their
faces.

• S0 = {∗0 }, only the base point, the unique vertex of this simplicial set, no
faces.

• S1 = {∗1 , id1 }, every face is ∗0 , no choice.

• S2 = {∗2 , η0 , η1 }, see the notations defined Section 3.1 and Proposition 3.1.
For example ∂0 (η0 ) = ∂1 (η0 ) = id1 , but ∂2 (η0 ) = ∗1 ; this is consequence of
Rule 2 in Section 4.3 and of the commuting relations page 11. We encourage
the reader to compute in the same way the faces of η1 .

• S3 = {∗3 , η0 η1 , η0 η2 , η1 η2 }. For example, ∂0 (η0 η2 ) = ∂1 (η0 η2 ) = η1 and

∂2 (η0 η2 ) = ∂3 (η0 η2 ) = η0 .

23
 Knowing all these faces allows the user to compute the first terms of the chain
complex canonically associated to this simplicial set S = S 1 :
d d d
1
(C0 = Z) ←− (C1 = Z2 ) ←−
1
(C2 = Z3 ) ←−
2
(C3 = Z4 ) (20)
where the differentials are the matrices:
 
  0 −1 0 1
  1 1 1
d1 = 0 0 , d 2 = , d3 =  0 1 0 0  (21)
0 0 0
0 0 0 −1
We observe the composition of two successive differentials are null, this is always
the case.

Proposition 22 — If (Cn , d) is the chain complex associated to a simplicial com-

plex X, the composition of two successive differentials dq dq+1 is null. This allows
to define:

• Zq (X, R) := ker dq is the group of q-cycles of X.

• Bq (X, R) := im dq+1 is the group of q-boundaries of X.
• The relation dq dq+1 = 0, always satisfied, is equivalent to Bq ⊂ Zq .
• The quotient group Hq (X, R) := Zq /Bq is the q-dimensional homology group
with coefficients in R. ♣

In the example of C∗ (S), we can determine:

• Z0 = Z, B0 = 0 and H0 = Z.
• Z1 = Z2 , B1 = Z direct summand of Z1 and H1 = Z.
• Z2 = Z2 generated by ∗2 − η0 and ∗2 − η1 , so that B2 = Z2 and H2 = 0.

Writing for example Z0 ‘=’ Z is not correct, in fact the cycle group Z0 is
isomorphic to Z and equal only to Z∗0 , the free Z-module generated by the unique
0-simplex ∗0 . These shorthands are common, often convenient, but can also be
the source of serious drawbacks when Algebraic Topology is examined from a
constructive point of view, see [20].
Many degenerate simplices in a simplicial set! Except for examples as simple
as our circle S 1 , it is not easy in general to compute these homology groups. In
fact, from the homological point of view, the role of these degenerate simplices is
void! The key point is the following: in general a face of a degenerate simplex
can be non-degenerate, for example above ∂02 η0 = id1 , but the differential, that
is, the alternate sum of faces, of a degenerate simplex is always a combination of
degenerate simplices, for example dη0 = ∗1 . So that we can denote by C∗D (X) the
sub-chain complex generated by the degenerate simplices. It happens this chain
complex is “without” homology, which, by a difference process, produces the next
proposition.

24
Proposition 23 — Let X be a simplicial set, C∗ (X) the associated chain com-
plex, C∗D (X) the degenerate sub-complex and C∗N D (X) := C∗ (X)/C∗N D (X) the
quotient chain complex. Then the canonical projection C∗ (X) → C∗N D (X) induces
an isomorphism between the homology groups.

♣ [15, VIII.6]. ♣

Definition 24 — A chain complex morphism f : C∗ → C∗0 is a collection of linear

maps f = {fn : Cn → Cn0 } compatible with the differentials: df = f d, that is, for
every n, the relation dn fn = fn−1 dn holds.

One then says f is of degree 0, for f respects the degree. It is also possible
to consider also maps of arbitrary degrees, but be careful in this case with sign
coherences! Most often, the index n for a component fn of a chain complex mor-
phism is omitted, and except particular cases, elliptic formulas such as df = f d
are preferred.
Because of the compatibility with differentials, a chain complex morphism f :
C∗ → C∗0 induces many natural maps, most often denoted by the same symbol f :

f : Z∗ (C∗ ) → Z∗ (C∗0 )
f : B∗ (C∗ ) → B∗ (C∗0 ) (22)
f : H∗ (C∗ ) → H∗ (C∗0 )

In fact, because of the relation df = f d, the image of a cycle is a cycle, the image
of a boundary is a boundary, so that f naturally induces a map between homology
classes.

6.2 Homotopy and homology for simplicial complexes.

We will examine later in details the notion of combinatorial homotopy in the frame-
work of simplicial sets, not so easy. Considering the particular case of simplicial
complexes is a good introduction. Firstly, a purely algebraic notion of homotopy.

Definition 25 — Let C∗ and C∗0 be two chain complexes, and f0 , f1 : C∗ → C∗0

two chain complex morphisms. These morphisms are (algebraically) homotopic if
0
there exists an operator h = {hn : Cn → Cn+1 }n∈Z satisfying f1 − f0 = dh + hd.

Cn−1 d Cn d Cn+1

f0 f1 h f0 f1 h f0 f1

0
Cn−1 d
Cn0 d
0
Cn+1

25
 Our homotopy operator h has degree +1. If compatible with the differentials,
the relation dh = −hd would be satisfied2 ; our homotopy operator is in fact not
at all compatible with differentials, the “error” being just the difference dh + hd =
f1 − f0 .
Most topologists say such maps f0 and f1 are chain equivalent; we prefer the
more coherent terminology of our definition: as illustrated later, this definition
is nothing but the algebraic translation in the chain complex framework of the
topological notion of homotopy. With a warning: we will see two maps between
various sorts of topological spaces which are (topologically) homotopic induce maps
algebraically homotopic between chain complexes, but the converse in general is
false.

Proposition 26 — Let f0 , f1 : C∗ → C∗0 be two homotopic chain complex mor-

phisms. Then the induced maps f0 , f1 : H∗ (C∗ ) → H∗ (C∗0 ) are equal.

♣ Let h ∈ H∗ (C∗ ) be a homology class represented by some cycle z ∈ Z∗ (C∗ ).

Then (f1 − f0 )(h) is represented by (f1 − f0 )(z) = (dh + hd)(z) = (dh)(z), for z
cycle means dz = 0. The diffference cycle f1 (z) − f0 (z) therefore is the boundary
dh(z) and these cycles are homologous; in other words the homology classes f0 (h)
and f1 (h) are equal. ♣

Proposition 27 — Let K and K 0 be two simplicial complexes, and f0 , f1 : K →

K 0 two simplicial morphisms which are (topologically) homotopic: see Definition 5
and the following discussion. Then the induced maps f0 , f1 : C∗ (K) → C∗ (K 0 ) are
(algebraically) homotopic and therefore the induced maps between homology groups
f0 , f1 : H∗ (K) → H∗ (K 0 ) are equal.

This is a powerful tool for negative results: conversely, if the induced maps
between homology groups are different, then the original continuous maps are not
homotopic. This proposition proved here only in the simplicial complex framework
in fact has a very general scope, and is at the very definition of Algebraic Topology:
a purely algebraic observation implies topological properties.
♣ Given the hypotheses about K, K 0 , f0 and f1 , we have to construct an algebraic
homotopy operator h : C∗ (K) → C∗+1 (K 0 ) between both chain complex morphisms
f0 and f1 . The answer is the following:
n
X
h((v0 , . . . , vn )) := (−1)i (f0 v0 , . . . , f0 vi , f1 vi , . . . , f1 vn )). (23)
i=0

2
Not dh = hd, for the “right” sign is given by the famous Koszul “rule”: dh =
(−1)deg(d)·deg(h) hd.

26
To save some paper and produce less CO2 , we verify the homotopy property only
in the case n = 1:
h((v0 , v1 )) = (f0 v0 , f1 v0 , f1 v1 ) − (f0 v0 , f0 v1 , f1 v1 );
dh((v0 , v1 )) = (f1 v0 , f1 v1 )1 − (f0 v0 , f1 v1 )2 + (f0 v0 , f1 v0 )3
−(f0 v1 , f1 v1 )4 + (f0 v0 , f1 v1 )2 − (f0 v0 , f0 v1 )5 ;
(24)
d((v0 , v1 )) = (v1 ) − (v0 );
hd((v0 , v1 )) = (f0 v1 , f1 v1 )4 − (f0 v0 , f1 v0 )3 ;
(f1 − f0 )((v0 , v1 )) = (f1 v0 , f1 v1 )1 − (f0 v0 , f0 v1 )5 .

where the indices after the simplex expressions show the correspondances which
prove the relation f1 − f0 = dh + hd in this particular case. The general proof is
analogous, a good exercise about index handling. ♣
Is the reader really satisfied with this “proof”? He should not! We have
accumulated here a terrible number of “imprecisions”, let us be simple, a terrible
number of faults. Each one is interesting and illustrates the role of the Algebraic
Topology’s Devil, namely the symmetric group.
The chain complex C∗ (K) associated to the simplicial complex K is given
in Definition 21, which needs in turn the explanations of Section 4.2, where the
simplicial set associated to a simplicial complex is defined. But in the initial
Definition 1 of a simplicial complex, no order over the vertices; on the contrary,
when defining “the” associated simplicial set, a total order over the vertices is
required. If we change this order, what happens for example for the resulting
homology groups? Yes, they are the same up to isomorphism, but the proof is not
so easy.
Second difficulty, Definition 3 for a simplicial map between simplicial complexes
does not require any compatibility conditions with vertex orders, in fact not yet
considered before this definition. But after defining some orders over the vertices
of K and K 0 , if f : K → K 0 is a simplicial map, it can happen v0 < v1 in K
and f v0 > f v1 in K 0 , and the “induced” maps between chain groups, where the
generators are made of “ordered” simplices, is then erroneously defined. Another
difficulty occurs when f v0 = f v1 : the image simplex is then degenerate, but we
did not even mention if we preferred the total version C∗ (K) or the normalized
one C∗N D (K) for “the” chain complex associated to K.
We could require the simplicial maps compatible with orders, which is very
restrictive; but even with such a restriction, the mixed term:

(f0 v0 , . . . , f0 vi , f1 vi , . . . , f1 vn )

of the formula (23) defining the homotopy operator h can produce a simplex with
vertices in a wrong order. You see this question of orders over the simplex vertices
is rather tough.
A really complete solution about this problem of simplex vertices can be found
in [7, Chapter VI], where two chain complexes are associated to a simplicial com-
plex, a big one called the ordered chain complex and a smaller one called the
alternating chain complex; the last one is isomorphic to the normalized chain

27
complex C∗N D (K) defined here through the intermediary notion of simplicial set;
the first one accepts as generators simplices described as sequences of vertices in
any order, taking account of this order; it accepts as well degenerate simplices
with repetitions in the vertices. Both chain complexes have advantages and draw-
backs, but their homology groups are canonically isomorphic, a frequent situation
in Algebraic Topology.

7 Homology groups: a quick survey.

In this section, we give a rough “cultural” presentation of the simplest tools al-
lowing one to compute homology groups. It is just a presentation, the results are
most often stated without any demonstration and references, at least when they
are reachable through any common textbook of Algebraic Topology.
You are interested in some space X, some coefficient group R is given and for
some reason, you would like to determine the groups H∗ (X; R). The space X can
be described as a simplicial complex, or more generally as a simplicial set, and you
could consider the simplicial homology groups. Still more generally, the singular
homology could be used, which is defined for arbitrary topological spaces; if the
space can be triangulated, that is, if it is homeomorphic to some simplicial set,
then the simplicial and singular homology groups are canonically isomorphic; in
particular the simplicial homology groups do not depend on the chosen triangula-
tion.

7.1 Homotopy Types.

Definition 28 — A map f : X → Y between two simplicial sets (or more gen-
erally between two topological spaces) is a homotopy equivalence if there exists a
homotopical inverse g : Y → X, that is, a map satisfying: gf is homotopic to
idX and f g is homotopic to idY . If so, it is said both spaces X and Y have the
same homotopy type. A homotopy type is an equivalence class for this equivalence
relation.

Proposition 27 implies the induced maps between homology groups are then
isomorphisms. So that if you observe Hn X and Hn Y are not isomorphic for some
integer n, you have proved the spaces X and Y do not have the same homotopy
type.
It is well known the homology groups in general do not suffice to distinguish
homotopy types. The standard example is X = S 2 ∨ S 4 and Y = P 2 C; their
homology groups are H0 = H2 = H4 = Z and the others are null ; however
their homotopy types are different, which in this case is proved by considering the
multiplicative structure in cohomology. This extra information is not enough in
general: the next example is X = S 3 ∨ S 5 and Y = ΣP 2 C, this time distinguished
byt a Steenrod operation. The problem of giving a complete invariant set for the
homotopy types is today open [19].

28
Definition 29 — A space X is contractible if it has the homotopy type of a point.

If a space X is contractible, it has the same homology groups as a point, that

is, H0 (X, R) = R and Hn (X, R) = 0 for n > 0. For example a simplex ∆n is
contractible. The converse is false: some non trivial discrete groups G are acyclic,
the same homology as a point, so that the corresponding classifying space BG,
see Section 5.2.1, is not contractible but with trivial homology. In the particular
case of a simply connected space, then the equivalence between contractibility and
trivial homology is true.

Definition 30 — Let K = (V, S) and K 0 = (V 0 , S 0 ) be two disjoint simplicial

complexes, that is, V ∩ V 0 = ∅. Then the join K 00 = K  K 0 is defined as
K 00 = (V 00 , S 00 ) with V 00 = V ∪ V 0 and σ 00 ∈ S 00 if and only if σ 00 ∩ V ∈ S ∪ {∅} and
σ 00 ∩ V 0 ∈ S 0 ∪ {∅}, but σ 00 = ∅ of course remains excluded3 . In particular the cone
CK of a simplicial complex K = (V, S) is the join CK = ∗  K where ∗ is a
simplicial complex reduced to one point, the unique vertex, this vertex not being
a vertex of K.

This definition means the simplices of K  K 0 are made of the simplices of K,

the simplices of K 0 , and any pair (σ, σ 0 ) of S × S 0 generates a simplex of K 00 of
dimension dim σ+dim σ 0 +1. For example the join of two intervals is a tetrahedron:
think you have joined any point of the first interval to any point of the second
one, which explains the terminology.

•
•

•
•

The topological definition of the join is:

X  Y := (X × [0, 1] × Y )/ ∼ (25)

where the equivalence relation ∼ identifies (x, 0, y) ∼ (x, 0, y 0 ) and (x, 1, y) ∼

(x0 , 1, y) for every x, x0 ∈ X and y, y 0 ∈ Y . In particular, if X has only one point,
we find only ∗  Y = (I × Y )/({0} × Y ) = CY .
To construct a cone CK for a simplicial complex K, you just have to add to
every simplex σ of K the simplex {∗} ∪ σ. For example, the cone of an n-simplex
is an (n + 1)-simplex.
A non-trivial exercise consists in proving the join of two spheres is a sphere :

Sp  Sq ∼
= S p+q+1 . (26)
3
It would be convenient to decide there is always, in any simplicial complex, a unique simplex
of dimension -1 corresponding to the void set of vertices; it would be a sort of augmented simplicial
complex.

29
Here you should take the simplicial definition S p = ∂∆p+1 and the same for S q .
The resulting simplicial complex is not isomorphic to ∂S p+q+1 , but its realization is
homeomorphic to. Hint: In a Euclidian S p+q+1 sphere, you have two “orthogonal”
disjoint “large” spheres S p and S q ; for example in the ordinary 2-sphere, you have
two remarkable spheres S 0 and S 1 which are orthogonal: S 0 could be made of both
North and South poles, S 1 being the equator; many other solutions with the same
geometry.

S0 •

•
• •

S0  S1 = S2 •
S1

A cone CK is always contractible, so that the homology groups of a cone are

canonically isomorphic to the homology groups of a point.

7.2 Exact sequences.

Definition 31 — An exact sequence is a linear diagram of groups and group
morphisms:
f g
· · · ←− F ←− G ←− H ←− · · · (27)
where the kernel of every arrow is the image of the previous one. For example, in
the displayed case, ker(f ) = im (g). The inclusion im (g) ⊂ ker(f ) is equivalent
to f g = 0, that is, the sequence is a chain complex. Asking for the equality
ker(f ) = im (g) is claiming the corresponding homology group is null: an exact
sequence is a chain complex the homology groups of which are null.
In particular a short exact sequence is a diagram:
f g
0 ←− F ←− G ←− H ←− 0 (28)

where f is surjective, ker(f ) = im (g) and g is injective.

Frequently, results about homology groups are presented as exact sequences.

An important particular case of this sort is the Mayer-Vietoris exact sequence.

Theorem 32 (Mayer-Vietoris exact sequence) — Let X be a simplicial set,

A and B two simplicial subsets such as X = A ∪ B. Then there exists a canonical
long exact sequence:
∂ A B j ⊕j iA ⊕(−iB )
· · · ←− Hn−1 (A ∩ B) ←− Hn (X) ←− Hn (A) ⊕ Hn B ←− ···
iA ⊕(−iB ) ∂
··· ←− Hn (A ∩ B) ←− Hn+1 (X) ←− · · ·

30
 The maps iA , iB , jA and jB are induced by the canonical inclusions iA :
A ∩ B ,→ A, iB : A ∩ B ,→ B, jA : A ,→ X and jB : B ,→ X. Note the mi-
nus sign given to iB , necessary to obtain (jA ⊕ jB ) ◦ (iA ⊕ (−iB )) = 0. The maps
∂ : Hn (X) → Hn−1 (A ∩ B) are the connection morphisms, more esoteric, not
defined here.
The Mayer-Vietoris exact sequence is important: it allows you to have infor-
mations about the groups H∗ (X) when you know the homology groups H∗ (A),
H∗ (B) and H∗ (A ∩ B): in many cases you can so deduce the homology groups of
the total space X when you know the homology groups of three of its constituents,
A, B and A ∩ B.
As a typical example, let us assume n is an integer n ≥ 2.

Proposition 33 — There exists a canonical isomorphism Hp (S n ) ∼

= Hp−1 (S n−1 )
for p ≥ 2.

3
• • • •
• •0 • • • •
2

• • • •
1 2
S ∆1 S1 CS 1
X =A∪B A A∩B B
In this figure illustrating the particular case n = 2, the 2-sphere S 2 is the boundary
of the 3-simplex; two components in the decomposition, the “lid” A = ∆1 =
{1, 2, 3} and the “cornet” cone CS 1 of the circle S 1 , the boundary of the lid, with
respect to the vertex 0.
♣ Let us consider the n-sphere S n as the boundary of the (n + 1)-simplex, that
is the simplex spanned by n + 1 = {0 . . . n + 1}. In particular the (n − 1)-sphere
S n−1 is the boundary of the n-simplex A spanned by {1 . . . n + 1}. We can also
consider the simplicial subcomplex B defined as the cone of S n−1 of summit 0.
Then A ∩ B = S n−1 and A and B are contractible. Let us consider the following
segment of the Mayer-Vietoris exact sequence:
Hp−1 (A) ⊕ Hp−1 (B) ←− Hp−1 (A ∩ B) ←− Hp (X) ←− Hp (A) ⊕ Hp (B) (29)
which becomes:
0 ←− Hp−1 (S n−1 ) ←− Hp (S n ) ←− 0 (30)
for A and B are contractible. This exact sequence implies the central map is an
isomorphism. ♣
Examining carefully in the same way the beginning of the Mayer-Vietoris exact
sequence, and taking account of the easy calculation of H∗ (S 1 ), we obtain for n ≥ 1:
H0 (S n ; R) = R,
Hn (S n ; R) = R, (31)
Hp (S n ; R) = 0 otherwise.

31
7.3 About the coefficient group.
According to the problem, some or other coefficient group R is preferred. Fre-
quently it is Z the integer ring; this coefficient group can also be a field, in par-
ticular a field Z/pZ for a prime p, or also the field Q of the rational numbers, the
field R of the real numbers. The case where the coefficient group is a field F is
often easier, for when a short exact sequence 0 → Fa → H → Fb → 0 is used to
determine the unknown group H, then the group H is isomorphic to the direct sum
Fa ⊕ Fb = Fa+b : no extension problem in this case to determine the isomorphism
class of H. On the contrary, if the coefficient group is Z, then an exact sequence
such as 0 → Z/6Z → H → Z/2Z → 0 leaves the user of this exact sequence with
a doubt, for the group H could be either Z/2Z ⊕ Z/6Z or Z/12Z: determining the
right choice needs further informations about which most books about Algebraic
Topology are not very loquacious. Obtaining automatic algorithms (pleonasm)
efficiently solving this problem is relatively recent [18].
The coefficient group Z is universal, which needs the notion of torsion group
to be described.

Definition 34 — If G and H are commutative groups, let 0 → G1 → G0 → G →

0 be a free presentation of the group G as Z-module. Then the tensor product
morphism (G1 → G0 ) ⊗ H is not necessarily injective and its kernel is called the
torsion group TorZ (G, H).

You must take for G0 an arbitrary free Z-module allowing you to define a
surjection G0 → G; you could take for example the free Z-module generated by G
itself: G0 = Z(G) and the map sending the generator g of G0 over the element g
of G. The kernel of the surjection G0 → G is necessarily free, and it is the group G1 .
Frequently, more simple choices for G0 are possible, and the final result will not
depend on the choice of the resolution. There is also a canonical isomorphism
Tor(G, H) ∼= Tor(H, G): the procedure can be applied to H if you prefer.
The simplest toy example is G = Z/2Z and H = Z/4Z. The cheapest resolution
of Z/2Z is:
×2
0 → Z → Z → Z/2Z → 0 (32)
The tensor product by Z/4Z produces:
×2
0 → Z/4Z → Z/4Z → Z/2Z → 0 (33)
×2
where the map Z/4Z → Z/4Z is not injective, producing the group:
×2
Tor(Z/2Z, Z/4Z) := ker(Z/4Z → Z/4Z) = Z/2Z. (34)
The reader is advised to verify Tor(Z/4Z, Z/2Z) gives the same result.

Theorem 35 — If X is a space, and R some coefficient group, a short exact

sequence is canonically defined:
0 ← Tor(Hn−1 X, R) ← Hn (X; R) ← Hn (X, Z) ⊗Z R ← 0 (35)

32
Furthermore this exact sequence is split, meaning the solution of the extension
problem is necessarily Hn (X; R) ∼
= (Hn (X, Z) ⊗Z R) ⊕ Tor(Hn−1 X, R), but the
splitting is not canonical.

This theorem implies the knowledge of the groups H∗ (X; Z) is enough to deter-
mine the other homology groups H∗ (X; R). These homology groups with integer
coefficients are therefore so important that most often they are simply denoted by
H∗ X.
Conversely, the cost of determining these groups is high, not amazing: the
Universal Coefficients Theorem shows the Z-groups contain all the others. In the
case, frequent, where these homology groups are known as Z-modules of finite
type, it can be on the contrary more efficient to subdivide the Z-problem into
subproblems as follows:

Hn (X; Z) ∼
M
= Zd0 ⊕ Hn (X; Z(p) ). (36)
p prime

where d0 is the Q-dimension of Hn (X; Q) and Z(p) is the ring integer Z localized
at the prime p. And specific methods with respect to Q and Z(p) often allow to
compute these homology groups.

7.4 Elementary computations.

If the space X is described as a finite simplicial complex, or a finite simpli-
cial set, then the computation of its homology groups is in principle elemen-
tary. You have to compute the boundary matrices dn : Cn (X) → Cn−1 (X)
and dn+1 : Cn+1 (X) → Cn (X) and an elementary computation of Z-linear al-
gebra, mainly the Smith reduction of integer matrices, produces the quotient
ker dn /im dn+1 . But the practical computations become quickly painful or even
impossible, even with the most powerful computers.
For example a triangulation of the torus T2 = S 1 × S 1 is given p.58 with
9 vertices, 27 edges and 18 triangles. Computing naively H1 (T2 ) needs the matrix
describing d1 , 27 columns and 9 rows, and the matrix describing d2 , 18 columns
and 27 rows. With pencil and paper, it is necessary you are a little bit lucid and
careful: you should determine the matrix d1 has rank 8, producing an image of
rank 8 and a kernel of rank 19; the matrix d2 has rank 17, image of rank 17 and
kernel of rank 1. A little more work will produce H1 = Z2 .
The benefit of the technology of simplicial sets becomes obvious if you consider
the following triangulation of the torus:

• •
σ21
σ12 σ13
σ22
σ01• •
σ11

33
As before you must identify on the one hand the left and right edges σ12 of the
square, and on the other hand the top and bottom edges σ11 , and the four corners
σ01 are so identified, producing the unique vertex of this simplicial set. Finally,
1 vertex σ01 , 3 edges σ11 , σ12 and σ13 , and two triangles σ21 and σ22 . The boundary
matrices then are:  
1 1
d1 = [0 0 0] , d2 =  −1 −1  (37)
1 1
It is then significantly easier to deduce the homology groups, the same as before, for
the isomorphism classes of these groups do not depend on the chosen triangulation,
as a simplicial complex or as a simplicial set as well.
But even with this technology of simplicial sets, you could meet severe difficul-
ties with these so called “elementary” computations. For example the standard
triangulation of the Eilenberg-MacLane K(Z/2Z, 4) (see Section 9.1) requires nd
simplices of dimension d with in particular:
n7 = 34359509614
n8 = 1180591620442534312297 (38)
n9 = 85070591730234605240519066638188154620
and no computer in this planet can store into its memory the boundary matrices
d8 and d9 if you intend to compute “elementarily” the group H8 K(Z/2Z, 4). For-
tunately, more powerful methods, mainly the Eilenberg-Moore spectral sequence,
allow the interested algebraic topologist to reduce this computation to the homol-
ogy group of another chain complex, where this time, n07 = 4, n08 = 8 and n09 = 15,
more reasonable; the nature of the corresponding chain complex is much more so-
phisticated, the differentials of the generators are not easy to determine, but once
it is done, the matrices can be easily implemented and the homology group quickly
computed. To be complete about this subject of the homology groups of the terri-
ble Eilenberg-MacLane spaces, we must signal the nice work of. . . Eilenberg and
MacLane (see in particular [5, 6]) finally led Henri Cartan [3] to find a very elegant
method directly giving the homology groups of these Eilenberg-MacLane spaces.

7.5 Find simpler spaces!

Most often, computing H∗ (X) amounts to expressing X as the result of a con-
struction, as elementary as possible, from simpler spaces, the homology of which
being known.

7.5.1 Amalgamated sums.

The Mayer-Vietoris Theorem 32 is a typical example, where X is expressed as the

union of two components A and B, the homology of which being known, and also
the homology of the intersection A∩B. This is to be considered as an amalgamated
sum: X is nothing but the sum of A and B “amalgamated” along their intersection
A ∩ B.

34
7.5.2 Products.

Speaking of sums of spaces naturally leads to think of products. The next section
explains how to define the product of simplicial sets, a simplicial version of the
ordinary product of topological spaces. Let us assume the homology groups of X
and Y are known, how to determine the homology groups of X × Y ? The answer
is the Künneth theorem.

Theorem 36 (Künneth Theorem) — Let X and Y be two spaces and Z =

X × Y their product. Then a canonical short exact sequence is defined for every
integer n:

0 ←− n−1
L
i=0 Tor(Hi (X), Hn−i−1 (Y )) ←− Hn (X × YL
) ←− · · ·
· · · ←− Hn (X × Y ) ←− ni=0 (Hi (X) ⊗ Hn−i (Y )) ←− 0

Some coefficient group R is underlying, all the homology groups are computed
with respect to this coefficient group, the same for the tensor products and torsion
products. This exact sequence is split, but the splitting is not canonical.

This Künneth formula is to be compared with the product of polynomials or

power series:
X X XX n
n n
( an X ) × ( bn X ) = ( ai bn−i )X n (39)
n n n i=0

which suggests the last term of the Künneth exact sequence. This comparison
between product properties leads to the interesting and powerful notion of Poincaré
series. A “slight” error is generated by the naive product, expressed through
torsion products in a similar way, with one dimension less, like in the Universal
Coefficients Theorem. In fact the Universal Coefficients Theorem is a consequence
of a more general statement of the Künneth theorem in the framework of arbitrary
chain complexes.
In the particular case where the coefficient group R is a field, or also when one
of the factors has its homology groups free, which implies the torsion groups are
null, then the Künneth formula quickly gives the homology groups of the product.
For example, the formulas (31) give the homology groups of the n-sphere S n . The
Künneth formula then implies for the product X = S m × S n when 1 ≤ m < n:

H0 = Hm = Hn = Hm+n = R;
(40)
Hk = 0 if k 6= 0, m, n, m + n.

and the case m = n can be processed in the same way.

The simplest case where torsion products are involved is the case of X = P 2 R×
P 2 R. The integer homology groups of P 2 R are H0 = Z, H1 = Z/2Z and the others
are null. The Künneth formula then produces H0 (X) = Z, H1 (X) = Z/2Z⊕Z/2Z,
H2 (X) = Z/2Z and finally H3 (X) = Tor(Z/2Z, Z/2Z) = Z/2Z.

35
7.5.3 Twisted products ⇒ Serre spectral sequence.

It is explained Section 12 how a product of simplicial sets can be twisted to obtain a

much larger variety of spaces. It is the simplicial version of the notion of fibration.
The general framework is as follows, you have to consider a diagram or spaces
which is similar to a short exact sequence:

∗ → F ,→ E → B → ∗ (41)

Most often, the initial and terminal point spaces are omitted:

F ,→ E → B (42)

but they should not, for the analogy with the short exact sequences is really good.
The first component F is called the fiber space, the last one B is the base space,
and the central space E is the total space. The simplest case happens when this
“exact sequence” is split; you must consider B as a pointed space, a point ∗ ∈ B is
given. Then a particular case of twisted product, in fact in this case not twisted,
is the following:
F ,→ F × B → B (43)
where the first arrow is the map x 7→ (x, ∗), this is why a base point in B is
required, and the second arrow is the canonical projection. The general situation
could be written:
F ,→ F ×τ B → B (44)
where the index τ means some method is used to change the trivial product F × B
into a deeply modified one F ×τ B. The simplicial case is described in detail in
Section 12.
The simplest non-trivial case maybe is the trivial product Z × S 1 producing a
space with a countable number of circles, which can be modified to Z ×τ S 1 = R
giving this “exact sequence”:

{0} ,→ Z ,→ R → S 1 → ∗ (45)

with in particular the canonical inclusion Z ,→ R and the exponential map exp :
R → S 1 : t 7→ e2iπt , considering S 1 as the unit circle of the complex plane C.
A detailed description of this particular case is given Section 12.1; note also this
topological twisted product is simultaneously an actual exact sequence of groups,
where the central group is a non-trivial extension of the last one S 1 by the first
one Z. In more general situations, this is nomore true.
What about the homology groups of a twisted product? The homology groups
H∗ (F ) of the fiber space F and the homology groups H∗ (B) of the base space B
are assumed known and you intend to compute the homology groups of the total
space E = F ×τ B. How to process?
We have explained when commenting the Künneth Theorem 36 that the ho-
mology groups of a product can be more or less interpreted as the product of
the homology groups of the factors, if you consider the “total” homology H∗ as

36
 Hn tn . For the twisted product, there remains in a sense to
P
a Poincaré series
twist again the Künneth result, to be considered in this situation as an interme-
diary step. This complex twisting process for the homology groups is formalized
through the important notion of spectral sequence. But this process works only if
the base space is simply connected, and therefore cannot be applied to our exam-
ple of the exponential map Z ,→ R → S 1 : the first homotopy group π1 S 1 = Z
is non-trivial and the circle S 1 is certainly the simplest example of a non-simply
connected space. So that for our illustration, we need another example of twisted
product where the base is simply connected.
A popular example of this sort is the Hopf fibration: S 1 ,→ S 3 → S 2 . Think
of S 3 as the unit sphere of C2 = R4 , that is:

S 3 = {(z1 , z2 ) ∈ C2 st |z1 |2 + |z2 |2 = 1}. (46)

The group S 1 = {z ∈ C st |z| = 1} acts over S 3 by the diagonal action z ·

(z1 , z2 ) := (zz1 , zz2 ), and the quotient of this action, in other words the orbit space,
is homeomorphic to the sphere S 2 : it is the classical result that the projective space
P 1 C is nothing but the Riemann sphere. Combined with the canonical inclusion
S 1 ,→ S 3 : z 7→ (z, 0), we obtain the twisted product:

S 1 ,→ S 3 → S 2 (47)

which is the Hopf fibration. It can be proved this fibration actually can be orga-
nized as a twisted product as described Section 12, not so easy. Note this time the
base space S 2 is not a group, for S 1 is not a normal subgroup of S 3 ; the 2-sphere S 2
is only a homogeneous space; but the sphere S 2 is simply connected and the so
called Serre spectral sequence can be used.
2
This works as follows. First an array of groups Ep,q is to be constructed, where
2
every Ep,q is made of two groups appearing in the Künneth theorem:
2
Ep,q = (Hp (B) ⊗ Hq (F )) ⊕ Tor(Hp−1 (X), Hq (F )). (48)

2 2
In our case of the Hopf fibration, only four Ep,q are non-trivial, namely E0,0 = Z,
2 2 2 2
E0,1 = Z, E2,0 = Z, E2,1 = Z. The Ep,q ’s can be organize as an array with two
axes p and q, giving in our case:

q
2 0 0 0 0

1 Z 0 Z 0

0 Z 0 Z 0
0 1 2 3 p

This diagram is traditionally called the E 2 -page of the spectral sequence, the
starting page for the Serre spectral sequence. If we add all the displayed groups

37
along the diagonals p + q = n, we obtain the homology groups of the (non-twisted)
product S 1 ×S 2 , namely Hn = Z for 0 ≤ n ≤ 3. But in the twisted case, the process
is not finished, we must run all the next pages of the spectral sequence, starting
from the displayed E 2 -page. This consists in installing mysterious differentials
2
d2p,q : Ep,q 2
→ Ep−2,q+1 , mysterious, but of course well defined by the theory of
spectral sequences, not detailed here4 . In our case, please admit there is only one
2 2
non-trivial differential, between E2,0 and E0,1 , isomorphic to idZ . No choice for the
2
others dp,q . We obtain the diagram:

q
2 0 0 0 0

1 Z 0 Z 0
idZ

0 Z 0 Z 0
0 1 2 3 p
E 2 -page

These d2p,q ’s really are differentials: the composition of two successive ones is null.
This strange table of d2p,q ’s is nothing but a collection of chain complexes, and the
E 3 -page is made of all the homology groups of these:
q
2 0 0 0 0

1 0 0 Z 0

0 Z 0 0 0
0 1 2 3 p
E 3 -page

Again you must now install new differentials d3p,q : Ep,q

3 3
→ Ep−3,q+2 , but because
3
of the geometry of our E -page, all these differentials are null, which implies the
E 4 -page is the same as the E 3 -page. And so on up to the E ∞ -page, the same
as the E 3 -page. There remains to sum along the diagonals p + q = n, giving
H0 (S 3 ) = H3 (S 3 ) = Z, the other homology groups being null and to compare with
the formulas (31) to verify our computation is coherent with a previous one for
the same groups.
Here, we were lucky, for in the general case, it is even not enough to sum the
∞ ∞
Ep,q along the diagonals p + q = n, for these Ep,q ’s are only the components of the
graduated module associated to some filtration of the unknown group Hn (F ×τ B),
leading sometimes again to terrible extension problems.
4
In this case, the non trivial d22,0 is essentially the Chern class c1 of the fibration.

38
7.5.4 Classifying spaces and loop spaces.

Did you think of this problem: Given a space X, is it possible to find an “inverse”
space Y , that is, a space satisfying X ×Y = 1. The righthand term 1 should be the
unit for the product, that is the point space ∗ = ∆0 and the equation to be solved
becomes X × Y = ∗. Because of the definition of the product, if X is not also a
point, it is clearly impossible. But we can ask the same question up to homotopy:
is it possible to find a space Y such that the product X × Y ∼ ∗? If X is simply
connected and non-contractible, it is impossible: certainly some homology groups
Hn (X) are non-null and the Künneth Theorem implies it is not possible to find a
solution for Y , because it is clearly impossible to annihilate the non-null Hn (X)
when computing the Hn (X × Y ). In the non-simply connected case, an analogous
argument based over homotopy groups, in fact quite simpler, gives the same result.
It seems there are no possible solutions for inversion in topology, except in trivial
situations.
Yes, there are, but you must consider some twisted products.

Theorem 37 — Let X be a connected space. Then there exists a contractible

twisted product ΩX ×τ X with ΩX the loop space of X. Every space satisfying this
condition has the homotopy type of ΩX.

The product is contractible, so that up to homotopy the relation ΩX ×τ X ∼ ∗

is satisfied. This is true in the topological framework with an appropriate notion
of fibration. It is true also in the simplicial framework and such a solution, due to
Daniel Kan, is detailed in Section 9.2.
The loop space ΩX being defined in this way as a left inverse of the original
space X, another natural question is now to design a method computing the ho-
mology groups H∗ ΩX from the given groups H∗ X. In fact the last groups are
not enough, some further topological informations not contained in the homology
groups are necessary, but the setting of these notes does not allow the author to go
further along this problem. See [20, Section 9] for a detailed study of this problem,
and a complete algorithmic solution.
Note also these twisted products are not symmetric: the respective roles of
the fiber space and the base space are quite different, so that it is natural to ask
also the symmetric question: given a space X, is it possible to find another space
Y and a twisted product X ×τ Y ? Examining this question leads to restrict the
nature of the space X: it must be a topological group G or something analogous
up to homotopy, and the solution Y is then called the classifying space BG.

Theorem 38 — Let G be a topological group. Then there exists a contractible

twisted product G ×τ BG with a space BG called the classifying space of the group
G. Every space satisfying this condition has the homotopy type of BG.

A particular case is G the Eilenberg-MacLane space G = K(π, n) for some

commutative group π and some positive integer n; this Eilenberg-MacLane space

39
is itself naturally provided with a structure of commutative group, in particular
π = K(π, 0), and a recursive definition of K(π, n) is K(π, n) = BK(π, n − 1)).
This space is called the classifying space, because it is essential when classifying
the fibrations using the structural group G. In the discrete case, Section 5.2.1, a
simplicial description of BG is given and also the twisting function τ . The case of
K(π, n) for π a discrete group is explained in Section 9.1.
Because G is a group, the chain complex C∗ G is provided with a natural product,
producing a structure of differential algebra. Eilenberg and MacLane proved this
structure is the key point to determine the homology groups H∗ (BG).

7.5.5 Eilenberg-Moore spectral sequences.

The loop space ΩX of a connected space X and the classifying space BG of a

topological group G are “inverse” spaces in the world of twisted products:

ΩX ×τ X ∼ ∗ G ×τ BG ∼ ∗ (49)

If a number x is close to the unit 1, an inverse x−1 of x can be computed as a

sum of geometric series :
∞
X
x−1 = 1 + (1 − x) + (1 − x)2 + · · · = (1 − x)n (50)
n=0

If X is simply connected and G connected, the homology groups of ΩX and

BG can be computed essentially in the same way: the Eilenberg-Moore spectral
sequences give sense to the formulas:

H∗ ΩX = ∞ n
P
n=0 (1 − H∗ X)
P∞ (51)
H∗ BG = n=0 (1 − H∗ G)n

In particular the expression (1 − H∗ X), for example, means you must cancel the
H0 X = R in H∗ X, that is, only the “tail” H1 t + H2 t2 + . . . must be considered in
the total homology. The “negative” sign in this total homology is a consequence of
a desuspension (loop space) or suspension (classifying space) process to be applied
to the total homology. Furthermore, the infinite sum ∞
P
n=0 must be twisted too,
using the Alexander-Whitney coproduct in C∗ X, sketched in the next Section, or
the product structure in C∗ G coming from the group structure of G. Algorithmic
versions of these spectral sequences are explained in [20], using the methods of
constructive algebraic topology [18].

7.6 Alexander-Whitney coproduct and operads.

As explained in the previous section, a coalgebra structure, mainly defined by the
Alexander-Whitney coproduct, can be installed on the chain complex associated
to a simplicial set. The process is remarkably simple, so simple that it is a little

40
strange this structure finally has so much importance, but this is the starting point
of a long process leading to the modern and outstanding point of view of operadic
structure. A good opportunity to present the general status of modern Algebraic
Topology.

Definition 39 — Let X be a simplicial set. The Alexander-Whitney coproduct

is a chain-complex morphism:

∆ : C∗ X → C∗ X ⊗ C∗ X (52)

defined as follows for a generator σ ∈ Xn :

n
X
∆(σ) := ∂i+1 . . . ∂n a ⊗ ∂0 . . . ∂i−1 a. (53)
i=0

Typically, if X = ∆3 and σ is the maximal simplex 0123 of ∆3 , then, with an

obvious interpretation of the integer sequences:

∆(0123) = 0 ⊗ 0123 + 01 ⊗ 123 + 012 ⊗ 23 + 0123 ⊗ 3. (54)

It happens this is nothing but an algebraic version of the diagonal map X →

X ×X : x 7→ (x, x), interpretation valid thanks to the Eilenberg-Zilber equivalence
C∗ (X × X) ∼ (C∗ X ⊗ C∗ X). This coproduct is associative, of if you prefer co-
associative, with an obvious definition, but not commutative. You could consider
it is a drawback, but it is not. The induced map in cohomology is commutative,
you could think it is an advantage, but it depends on the point of view.
It is a good opportunity to point out a terrible drawback of the very definition of
the notion of simplicial set. Essentially, all the vertices of a simplex are numbered,
from 0 to n for an n-simplex. This defines an order over the vertices, orientation of
the edges, ortientation of triangles, and so on. But all these extra data are arbitrary
and do not correspond at all with the pure notion of an “abstract” simplex where
there is no reason to choose some order of the vertices: the very definition of
a simplicial set leads to arbitrary choices in the description, which will terribly
resctrict the scope of simple algebraic topology. You could object on the contrary
the study by Eilenberg and Steenrod in [7, Chapter VI], already mentioned, allows
you to close this matter? In a sense, simple algebraic topology makes commutative
the non-commutative world of topology, but it is cheating, the non-commutativity
of the topological world cannot be indefinitely hidden and this non-commutativity
must finally be considered and processed.
It happens the first step in this direction is the Alexander-Whitney coproduct,
which is not commutative, and in a sense which captures a (very small) part of
the unavoidable non-commutativity of the topological world.
If you continue to study this matter about commutativity in Algebraic Topol-
ogy, you will certainly go to the notion of operadic structure over a chain complex.
The notion of E∞ operad can be defined, an E∞ -structure makes sense for a chain
complex and the final essential result along this line is the following.

41
Theorem 40 (Michael Mandell [16]) — Let X be a simply connected homotopy
type, all the Z-homology groups of which have finite type. Let C∗ be a free Z-chain
complex of finite type whose homology groups are those of X. Then it is possible
to define an E∞ -structure over C∗ , this structure being equivalent to the homotopy
type.

It was explained after Definition 28 the homology groups are not enough to
identify a homotopy type, in fact are far from being enough. Mandell’s theorem
explains that completing the homology groups, more precisely a free chain complex
having the right homology groups, with an E∞ -operadic structure allows you to
provide the missing information to complete the definition of this homotopy type.
In other words, any “reasonable” homotopy type can be defined by a chain com-
plex of finite type provided with an E∞ -structure. The given homotopy type has
been entirely “algebraized”. It happens the inductive system of symmetric groups
S := (Sn )n∈N plays a major role in the definition of the E ∞ -operad: the “non-
commutative” component of a homotopy type is captured in an E∞ -structure.
But the next natural problem: “Is it possible to constructively classify in this
way the collection of the reasonable homotopy types?” is today entirely open.

8 Products of simplicial sets.

The general work style in Algebraic Topology consists in firstly proving results for
simple spaces, next deducing analogous results for more complicated spaces con-
structed from these spaces. The product constructor is important, as in most parts
of mathematics, a more sophisticated one in topology being the twisted product
constructor, invoking fibrations.
A simple but terrible observation is to be made about products, if one works
in the simplicial framework: the product of two simplices is not a simplex. For
example a 1-simplex is an interval, the product of two intervals is a square, which
cannot be naturally identified to a 2-simplex. But this square can be divided
into two triangles, that is, two 2-simplices, and we must carefully organize this
remark, not so easy. We will see the simplicial set structure magically gives the
right solution, rather amazing!
• • • • •
?
• • × = =
?
• • • • •

Definition 41 — If X and Y are two simplicial sets, the simplicial product Z =

X ×Y is defined by Zm = Xm ×Ym for every natural number m, and αZ∗ = αX ∗
×αY∗
if α is a ∆-morphism.

The definition of the product of two simplicial sets is perfectly trivial and
is however at the origin of several landmark problems in algebraic topology, for

42
example the deep structure of the twisted Eilenberg-Zilber theorem, still quite
mysterious, and also the enormous field around the Steenrod algebras.
Every simplex of the product Z = X × Y is a pair (σ, τ ) made of one simplex
in X and one simplex in Y ; both simplices must have the same dimension. It is
tempting at this point, because of the “product” ambience, to denote by σ × τ
such a simplex in the product but this would be a terrible error! This is not
at all the right point of view; the pair (σ, τ ) ∈ Zm is the unique m-simplex in Z
whose respective projections in X and Y are σ and τ , again some m-simplices, and
this is the reason why the pair notation (σ, τ ) is the only one which is possible.
For example the diagonal of a square is a 1-simplex, the unique 1-simplex the
projections of which are both factors of the square; on the contrary, the “product”
of the factors is simply the square, which does not have the dimension 1 and which
is even not a simplex.

Theorem 42 — If X and Y are two simplicial sets and Z = X × Y is their

simplicial product, then there exists a canonical homeomorphism between |Z| and
|X| × |Y |, the last product being the product of k-spaces.

If you consider the product |X| × |Y | as the ordinary product of topological

spaces, the same accident as for CW-complexes (see [8, p.59]) can happen. The
framework of k-spaces avoids this obstacle, reducing the problem to finite simpli-
cial subsets. Furthermore this esoteric problem does not exist when both factors
are countable simplicial sets (countable sets of simplices), most often the case in
concrete constructive topology.
♣ There are natural simplicial projections X × Y → X and Y which define a
canonical continuous map φ : |X × Y | → |X| × |Y |. The interesting question is to
define its inverse ψ : |X| × |Y | → |X × Y |.
First of all, let us detail the case of X = ∆2 and Y = ∆1 where the essential
points are visible. The first factor X has dimension 2, and the second one Y
has dimension 1 so that the product Z shoud have dimension 3. What about
the 3-simplices of Z? There are 3 such non-degenerate 3-simplices, namely ρ0 =
(η0 σ, η2 η1 τ ), ρ1 = (η1 σ, η2 η0 τ ) and ρ2 = (η2 σ, η1 η0 τ ), if σ (resp. τ ) is the unique
non-degenerate 2-simplex (resp. 1-simplex) of ∆2 (resp. ∆1 ). This is nothing but
the decomposition of a prism ∆2 × ∆1 in three tetrahedrons.
Note no non-degenerate 3-simplex is present in X and Y and however some
3-simplices must be produced for Z; this is possible thanks to the degenerate
simplices of X and Y where they are again playing a quite tricky role in our
workspace; in particular a pair of degenerate simplices in the factors can produce
a non-degenerate simplex in the product! This happens when there is no common
degeneracy in the factors.

43
 σ×τ = (η0 σ, η2 η1 τ ) ∪ (η1 σ, η2 η0 τ ) ∪ (η2 σ, η1 η0 τ )
11 11 11

1
01 21 01 21 21 21

τ = ∪ ∪
10 10 10

0
00 20 00 00 00 20
1
1 1 1
σ
0 2 0 2 0 2 0 2

For example the tetrahedron ρ0 = (η0 σ, η2 η1 τ ) inside Z is the unique 3-simplex

the first projection of which is η0 σ, and the second projection is η2 η1 τ ; the first
projection is a tetrahedron collapsed on the triangle σ, identifying two points when
the sum of barycentric coordinates of index 0 and 1 (the indices where injectivity
fails in η0 ) are equal; the second projection is a tetrahedron collapsed on an interval,
identifying two points when the sum of barycentric coordinates of index 1, 2 and
3 are equal.
Let us take a point of coordinates r = (r0 , r1 , r2 , r3 ) in the simplex ρ0 . Its
first projection is the point of X = ∆2 of barycentric coordinates s = (s0 =
r0 + r1 , s1 = r2 , s2 = r3 ); in the same way its second projection is the point of
Y = ∆1 of barycentric coordinates t = (t0 = r0 , t1 = r1 + r2 + r3 ). So that:

φ(ρ0 , (r0 , r1 , r2 , r3 )) = ((σ, (r0 + r1 , r2 , r3 )), (τ, (r0 , r1 + r2 + r3 ))) (55)

In the same way:

φ(ρ1 , (r0 , r1 , r2 , r3 )) = ((σ, (r0 , r1 + r2 , r3 )), (τ, (r0 + r1 , r2 + r3 )))

(56)
φ(ρ2 , (r0 , r1 , r2 , r3 )) = ((σ, (r0 , r1 , r2 + r3 )), (τ, (r0 + r1 + r2 , r3 )))

The challenge then consists in deciding for an arbitrary point:

(σ, (s0 , s1 , s2 )), (τ, (t0 , t1 )) ∈ |X| × |Y |

what simplex ρi it comes from and what a good φ-preimage (ρi , r) could be. You
obtain the solution in comparing the sums u0 = s0 , u1 = s0 + s1 , u2 = t0 ; the
sums s0 + s1 + s2 and t0 + t1 are necesarily equal to 1 and do not play any role.
You see in the three cases, the values of ui ’s are:

((η0 σ, η2 η1 τ ), r) ⇒ u0 = r0 + r1 , u1 = r0 + r1 + r2 , u2 = r0 ,
((η1 σ, η2 η0 τ ), r) ⇒ u0 = r0 , u1 = r0 + r1 + r2 , u2 = r0 + r1 , (57)
((η2 σ, η1 η0 τ ), r) ⇒ u0 = r0 , u1 = r0 + r1 , u2 = r0 + r1 + r2 ,

so that you can guess the degeneracy operators to be applied to the factors σ and
τ from the order of the ui ’s; more precisely, sorting the ui ’s puts the u2 value in

44
position 0, 1 or 2, and this gives the index for the degeneracy to be applied to σ;
in the same way the u0 and u1 values must be installed in positions “1 and 2”, or
“0 and 2”, or “0 and 1” and this gives the two indices (in reverse order) for the
degeneracies to be applied to τ . It’s a question of shuffle. Furthermore you can
find the components ri from the differences between successive ui ’s. Now we can
construct the map ψ:

φ((σ, s)(τ, t)) = (ρ0 , (u2 , u0 − u2 , u1 − u0 , 1 − u1 )) if u2 ≤ u0 ≤ u1 ,

= (ρ1 , (u0 , u2 − u0 , u1 − u2 , 1 − u1 )) if u0 ≤ u2 ≤ u1 , (58)
= (ρ2 , (u0 , u1 − u0 , u2 − u1 , 1 − u2 )) if u0 ≤ u1 ≤ u2 .

There seems an ambiguity occurs when there is an equality between u2 and u0

or u1 , but it is easy to see both possible preimages are in fact the same in |Z|.
Now this can be extended to the general case, according to the following recipe.
Let σ ∈ Xm and τ ∈ Yn be two simplices, s ∈ ∆m and t ∈ ∆n two geometric points.
We must define ψ((σ, s), (τ, t)) ∈ |Z| = |X × Y |. We set u0 = s0 , u1 = s0 + s1 , . . . ,
um−1 = s0 + . . . + sm−1 , um = t0 , um+1 = t0 + t1 , . . . , um+n−1 = t0 + . . . + tn−1 .
Then we sort the ui ’s according to the increasing order to obtain a sorted list (v0 ≤
. . . ≤ vm+n−1 ). In particular um = vi0 , . . . , um+n−1 = vin−1 with i0 < . . . < in−1 ,
and u0 = vj0 , . . . , um−1 = vjm−1 with j0 < . . . < jm−1 . Then:

ψ((σ, s), (τ, t)) = ((ηin−1 . . . ηi0 σ, ηjm−1 . . . ηj0 τ ),

(59)
(v0 , v1 − v0 , . . . , vm+n−1 − vm+n−2 , 1 − vm+n−1 )) .

Now it is easy to prove ψ ◦ φ = id|Z| and φ ◦ ψ = id|X|×|Y | , following the proof

structure clearly visible in the case of X = ∆2 and Y = ∆1 .
It is also necessary to prove the maps φ and ψ are continuous. But φ is the
product of the realization of two simplicial maps and is therefore continuous. The
map ψ is defined in a coherent way for each cell σ × τ (this time it is really the
product |σ| × |τ | ⊂ |X| × |Y |) and is clearly continuous on each cell; because of
the definition of the k-topology, the map ψ is continuous. ♣
If three simplicial sets X, Y and Z are given, there is only one natural map
|X × Y × Z| → |X| × |Y | × |Z| so that “both” inverses you construct by applying
twice the previous construction of ψ, the first one going through |X × Y | × |Z|, the
second one through |X| × |Y × Z| are necessarily the same: the ψ-construction is
associative, which is interesting to prove directly; it is essentially the associativity
of the Eilenberg-MacLane formula.

8.1 The case of simplicial groups.

Let G be a simplicial group. The object G is a simplicial object in the group
category; in other words each simplex set Gm is provided with a group structure
and the ∆-operators α∗ : Gm → Gn are group homomorphisms.

45
 This gives in particular a continuous canonical map |G × G| → |G|; then
identifying |G × G| and |G| × |G|, we obtain a “continuous” group structure for
|G|; the word continuous is put between quotes, because this does not work in
general in the topological sense: this works always only in the category of k-
spaces where the group structure is a map |G| × |G| → |G|, the source of which
being evaluated in the k-space category; because of this definition of product, it
is then true that |G| × |G| = |G × G|. The composition rule so defined on |G|
satisfies the group axioms; in particular the associativity property comes from
the considerations about the associativity of the ψ-construction in the previous
section.

9 Other examples of simplicial sets.

9.1 The Eilenberg-MacLane spaces.
The classifying spaces of groups, see Section 5.2.1, are particular cases of Eilenberg-
MacLane spaces. Let G be some discrete group; the realization process applied to
the simplicial set BG produces a model for K(G, 1): all the homotopy groups are
null except π1 canonically isomorphic to G. The construction can be generalized
to construct K(π, d), d > 1, when π is an abelian group. See also [13, Chapter V]
where these questions are carefuly detailed.
Let π be a fixed abelian group, and d a natural number. The simplicial set
E(π, d) is defined as follows. The set of m-simplices E(π, d)m , shortly denoted
by Em , is Em = C d (∆m , π), the group of normalized d-cochains on the standard
m-simplex with values in π. Such a cochain σ is nothing but a map σ : ∆m d → π,
m
defined on the (degenerate or not) d-simplices of ∆ , null for the degenerate
simplices. If α is a ∆-morphism α : n → m, this map defines a simplicial map
α∗ : ∆n → ∆m which in turns defines a pullback map α∗ : C d (∆m , π) → C d (∆n , π)
between m-simplices and n-simplices of Em .
The simplicial set E(π, d) so defined contains the simplicial subset K(π, d),
made only of the cocycles, those cochains the coboundary of which (d :
C d (∆m , π) → C d+1 (∆m , π) is null. In fact E(π, d) is a simplicial group, that
is, a simplicial object in the category of groups, and K(π, d) is a simplicial sub-
group. The quotient simplicial group E(π, d)/K(π, d) is canonically isomorphic to
K(π, d + 1) and this structure defines the Eilenberg-MacLane fibration:

K(π, d) ,→ E(π, d) → K(π, d + 1) (60)

See later the section about simplicial fibrations for some details.

9.2 Simplicial loop spaces.

Let X be a simplicial set. We can construct a new simplicial set DT (X) (the
acronym DT meaning Dold-Thom) from X, where DT (X)m is the free Z-module

46
generated by the m-simplices Xm ; the operators of DT (X) are also “generated”
by the operators of X. This is a simplicial version of the Dold-Thom construction,
producing a new simplicial set DT (X), the homotopy groups of which being the
homology groups of the initial X. The simplicial set DT (X) is also a simplicial
group; its simplex sets are nothing but the chain groups at the origin of the
simplicial homology of X, but in DT (X), each simplicial “chain” of X is one
(abstract) simplex. See [13, Section 22].
The same construction can be undertaken, but instead of using the abelian
group generated by the simplex sets Xm , we could consider the free non-
commutative group generated by Xm . This also works, but then the obtained
space is a simplicial model for the James construction of ΩΣX, the loop space
of the (reduced) suspension of X. See [2] for the James construction in general
and [4] for the simplicial case.
It is even possible to construct the “pure” loop space ΩX, without any suspen-
sion. This is due to Daniel Kan [12] and works as follows. It is necessary to assume
∗
X is reduced, that is with only one vertex: the cardinality of X0 is 1. Let Xm the
∗
set of all m-simplices, except those that are 0-degenerate: Xm = Xm − η0 (Xm−1 );
this makes sense for m ≥ 1. Then let GXm be the free non-commutative group
∗ ∗
generated by Xm+1 ; to avoid possible confusions, if σ ∈ Xm+1 , let us denote by
τ (σ) the corresponding generator of GXm . The simplicial object GX to be defined
is a simplicial group, so that it is sufficient to define face and degeneracy operators
for the generators:

∂i τ (σ) = τ (∂i+1 σ), if 1 ≤ i ≤ m;

−1
∂0 τ (σ) = τ (∂1 σ)τ (∂0 σ) ; (61)
ηi τ (σ) = τ (ηi+1 σ), if 0 ≤ i ≤ m.

These definitions are coherent, and the simplicial set GX so obtained is a

simplicial version of the loop space construction. See [13, Chapter VI] for details
and related questions, mainly the twisted Eilenberg-Zilber theorem, at the origin of
the general solution described in [20] for the computability problem in algebraic
topology.

9.3 The singular simplicial set.

Let X be an arbitrary topological space. Then the singular simplicial set associ-
ated to X is constructed as follows. The set of m-simplices SXm is made of the
continuous maps σ : ∆m → X; one (abstract) simplex is one continuous map but
no topology is installed on SXm ; in particular when SX will be realized in the
following section, the discrete topology must be used. The source of the abstract
m-simplex σ is the geometric m-simplex ∆m ⊂ Rm provided with the traditional
topology. If α ∈ ∆(n, m) is a ∆-morphism, this α defines a natural continuous
map α∗ : ∆n → ∆m between geometric simplices, and this allows us to naturally
define α∗ (σ) = σ ◦ α∗ . An enormous simplicial set is so defined if X is an arbitrary
topological space; it is at the origin of the singular homology theory.

47
10 Kan extension condition.
Let us consider the standard simplicial model S 1 of the circle, with one vertex
and one non-degenerate 1-simplex σ. This unique 1-simplex clearly represents
a generator of π1 (S 1 ), but its double cannot be so represented. This has many
disadvantages and correcting this defect was elegantly solved by Kan.

Definition 43 — A Kan (m, i)-hat (Kan hat in short) in a simplicial set X is a

(m + 1)-tuple (σ0 , . . . , σi−1 , σi+1 , . . . , σm+1 ) satisfying the relations ∂j σk = ∂k−1 σj
if j < k, j 6= i 6= k.

For example the triple (∂0 id, ∂1 id, ∂2 id, ) is a Kan (2, 3)-hat in the standard
3-simplex ∆3 if id is the unique non-degenerate 3-simplex, see Section 4.2. In
the figure below representing this hat, the faces 013, 023 and 123 are all the 2-
dimensional faces except one (012) of the 3-simplex id.

1
∂2 id • ∂0 id

0 • • 2

∂1 id
•
3

In general, a Kan (m, i)−hat is organized as all the faces except one, the i-th
one, of a potentially existing simplex of dimension m + 1, but this hypothetic sim-
plex maybe in fact does not exist. In the simplex example above, the “hypothetic”
simplex does exist, it is id. But let us now consider the following example for the
circle S 1 , this circle being modelled as a simplicial set as described in Section 4.3.
∗• •∗ id
id id
• •
∗ ∗

The righthand figure is the circle drawn as usual, showing the base point ‘∗’ and
the fundamental 1-simplex id. The lefthand figure represents the 1-hat (id, id),
but this definition is incomplete, an important point to be detailed now.
The first example (∂0 id, ∂1 id, ∂2 id) in ∆3 is a correct Kan (2,3)-hat, but it is
not a Kan (2,2)-hat. Why? In this case the definition of this (2,2)-hat would be:
σ0 = ∂0 id
σ1 = ∂1 id (62)
σ3 = ∂2 id

48
where the critical indices are underlined. Because we are considering whether our
triple is a (2-2)-hat, the coherence condition of Definition 43 must in particular be
? ?
satisfied for j = 1 and k = 3: ∂1 σ3 = ∂2 σ1 , which becomes ∂1 ∂2 id = ∂2 ∂1 id but
the righthand member of this relation is ∂1 ∂3 id and this relation is false: it would
contradict Proposition 7.
Let us consider now the possible hat (id, id) of the circle S 1 ; every face of the
fundamental simplex id is the base point ‘∗’, so that the coherence conditions of
Definition 43 are certainly satisfied: our pair (id, id) is as you like a (1, 0)-hat, a
(1, 1)-hat and a (1, 2)-hat as well.

Definition 44 — If (σ0 , . . . , σi−1 , σi+1 , . . . , σm+1 ) is a Kan (m, i)-hat in the sim-
plicial set X, a filling of this hat is a simplex σ ∈ Xm+1 such that ∂j τ = σj for
j 6= i.

The “fundamental” 3-simplex id of ∆3 is a filling of the example Kan (2, 3)-hat:

(∂0 id, ∂1 id, ∂2 id) in ∆3 .
Let us now consider the hat (id, id) of S 1 . Then η1 id fills the (1, 0)-hat (id, id),
for ∂1 η1 id = id and ∂2 η1 id = id as well. In the same way, η0 id fills the (1,2)-hat
(id, id), for ∂0 η0 id = ∂1 η0 id = id. But on the contrary the (1, 1)-hat (id, id)
cannot be filled. Only three 2-simplices in S 1 , all degenerate, namely η1 η0 ∗, η0 id
and η1 id. But:
∂0 η1 η0 id = η0 ∗ =
6 id
∂2 η0 id = η0 ∗ = 6 id (63)
∂0 η1 id = η0 ∗ = 6 id
and no 2-simplex is able to fill our (1, 1)-hat. You see the second index in the
definition of a hat is crucial.

Definition 45 — A simplicial set X satisfies the Kan extension condition if any

Kan hat has a filling.

The standard simplex ∆d does not 5 satisfy the Kan condition: Consider the
(1, 2)-hat of ∆2 (σ0 = ∂1 (id), σ1 = ∂0 (id)) ; the only necessary incidence relation
∂0 σ0 = ∂0 σ1 is satisfied, but the wrong order in the indices prevents from filling
this hat. Exercise: prove the same for ∆1 . Most elementary simplicial sets do not
satisfy the Kan condition.
The simplicial sets satisfying the Kan extension condition have numerous in-
teresting properties; for example their homotopy groups can be combinatorially
defined, see the next section, a canonical minimal version is included, also sat-
isfying the extension condition [13, Section 9], a simple decomposition process
produces a Postnikov tower [13, Section 8].
The simplicial groups are important from this point of view: in fact a simpli-
cial group always satisfies the Kan extension condition [13, Theorem 17.1]. For
5
Claude Quitté and Zhao Gong Yun noticed the erroneous opposite statement in a previous
version, thanks!

49
example the simplicial description of P ∞ R (see Section 5.2) is a simplicial group
and therefore satisfies the Kan condition, which is not so obvious; it is even min-
imal. The singular complex SX of a topological space X also satisfies the Kan
condition but in general is not minimal. These simplicial sets satisfying the Kan
condition are so interesting that it is often useful to know how to complete an
arbitrary given simplicial set X and produce a new simplicial set X 0 with the
same homotopy type satisfying the Kan condition. The Kan-completed X 0 can be
constructed as follows.
Let us define first an elementary completion χ(X) for X. For each Kan (m, i)-
hat of X, we decide to add the hypothetical (m + 1)-simplex (even if a “solution”
preexists), and the “missing” i-th face; such a completion operation does not
change the homotopy type of X. Doing this completion construction for every
Kan hat of X, we obtain the first completion χ(X). Then we can define X0 = X,
Xi+1 = χ(Xi ) and X 0 = lim→ Xi is the desired Kan completion. You can also run
this process in considering only the failing hats.

11 Homotopy groups.
The simplest invariants allowing one to distinguish different homotopy types are
homology groups and homotopy groups. The initial definition of homotopy groups
is due to Hurewicz, and is strangely much simpler than the definition of homology
groups, due many years before to Poincaré. Another strange property is that,
except in special cases, these homotopy groups are hard to be computed.

11.1 The topological case.

11.1.1 The Poincaré group.

The definition of Hurewicz goes as follows. The case of the first homotopy group
π1 X of a pointed topological space X = (X, ∗) is a little simpler and we start with
this particular case. This group was already considered by Poincaré and is often
called the Poincaré group of X. A pointed topological space is a space where some
point ∗ ∈ X is distinguished. This point will be used as the mandatory starting
and arriving points of the loops.
The pointed circle S 1 = (S 1 , ∗) is the usual unit circle of the complex plane C
pointed at the unit ∗ = 1 ∈ C. A (topological) loop ΩX 3 γ : (S 1 , ∗) → (X, ∗) is a
continuous map sending the base point of S 1 to the base point of X. Installing the
compact-open topology over ΩX defines a structure of topological space; don’t
be anxious with the technicalities of such a topology, we will soon switch to a
combinatorial framework where this esoteric subject disappears. It is natural to
decide the trivial constant loop ∗ : S 1 → X : ∗(t) :≡ ∗ is the base point of ΩX, so
that the process can be iterated, defining Ωn X for arbitrary n ∈ N.
Two loops γ0 and γ1 are homotopic if a continuous map h : I × S 1 → X can be

50
installed between them; on one hand h|{i}×S 1 = γi for i = 0, 1; but the “intermedi-
ary” loops must also be. . . loops, that is, h|I×{∗} ≡ ∗ ∈ X. The homotopy relation
‘∼’ is an equivalence relation, allowing us to define the quotient π1 X := ΩX/ ∼.
Poincaré discovered a natural group structure can be installed over π1 X. If
γ, δ ∈ ΩX, the composition ε = γ · δ is defined by:

ε(z) = γ(z 2 ) if =z ≥ 0,
(64)
= δ(z 2 ) if =z ≤ 0.

You may not like this parametrization of the circle by the complex numbers of
modulus 1. If you prefer a parametrization by the interval I = [0, 1], often consid-
ered as a time interval, deciding that z = e2πit , you obtain:

ε(t) = γ(2t) if t ∈ [0, 0.5],

(65)
= δ(2t − 1) if t ∈ [0.5, 1].

The following figure illustrates this composition.

0.5 0.75
0.5 0.25

• • • := •
∗ ∗ ∗

where the “time” parametrization of the circle is used, and the surrounding space X
is not shown.
For later generalization, a dual point of view is convenient. The wedge S 1 ∨ S 1
of two circles is the (categorical) sum of two circles in the category of the pointed
spaces; it is the disjoint union of two circles with the base points identified:

S 1 ∨ S 1 := •

Two canonical loops λ1 and λ2 of this wedge are the lefthand and righthand inclu-
sions S 1 ,→ S 1 ∨ S 1 . The composition of these loops is a loop λ = λ1 · λ2 : S 1 →
S 1 ∨ S 1 running the whole wedge S 1 ∨ S 1 . Also, two loops γ, δ ∈ ΩX define a con-
tinuous map γ ∨ δ : S 1 ∨ S 1 → X, this is the universal property of the categorical
sum. And finally γ · δ := (γ ∨ δ) ◦ λ. In this way the unique composition of λ1 and
λ2 is extended to compositions of arbitrary loops.
The composition of loops is not associative, but is associative up to homo-
topy. It is a matter of continuous change of parametrization, from the time slicing
[0.00, 0.25, 0.50, 1.00] to [0.00, 0.50, 0.75, 1.00].
The unit of π1 X is the homotopy class ‘∗’ of the constant loop ‘∗’. It is not
a unit in ΩX for the composition, for in general, except if γ = ∗ is the constant
loop, ∗ · γ 6= γ; but the loop ∗ · γ is homotopic to γ. The proof is the following
figure:

51
 t
∗
γ γ γ γ
∗ ∗∗
∗ ∗
s

which produces the formulas:

s + 2t − 1
h(s, t) := γ( ) if (1 − s)/2 ≤ t ≤ 1;
s+1 (66)
:= ∗ otherwise.

A non-constant loop has no inverse, but has a canonical inverse up to homotopy,

consisting in running the same path, but in the opposite direction: γ −1 (t) :=
γ(1 − t).
" #−1
:=
• •

The composition γ · γ −1 is homotopic to the constant loop, thanks to the map

h : I × S 1 → X defined by the formulas:
p
h(s, t) := γ(1 − 2 (1/2 − t)2 + s2 )) if (1/2 − t)2 + s2 ≤ 1/4,
(67)
:= ∗ otherwise.

The following figure explains these formulas:

t
1∗ ∗ ∗
γ −1 ∗
∗ ∗ ∗
∗
γ
0∗ ∗ ∗
0 1 s

In this way, the restriction to the lefthand edge of the square, in fact a circle S 1 , is
the composition γ ◦ γ −1 , while the restriction to the righthand edge is the constant
loop at the base point.
The quotient π1 X := ΩX/ ∼ inherits from loop composition a group structure,
this is the Poincaré group of X, more precisely of the pointed space (X, ∗).

11.1.2 Higher homotopy groups.

The construction of π1 X sketched in the previous section can be extended to higher

dimensions, producing a collection of groups πn X for every integer n ≥ 0. In the
very particuliar case n = 0, it is natural to decide π0 X is the set of the path

52
connected components of X, pointed by the component containing the base point.
But except in particular situations, there is no natural group structure over this
set π0 X.
The most direct definition of πn X for n ≥ 2 is simply πn X := π1 (Ωn−1 X). We
give now an equivalent definition closer to the particular definition of the π1 group.
In the previous section, the composition of loops has in particular been defined
through a particular map λ : S 1 → S 1 ∨ S 1 .

Definition 46 — Let (X, ∗) be a pointed space. Then the (pointed) suspension

of X is the space:
X ×I
ΣX := . (68)
(X × 0) ∪ (X × 1) ∪ (∗ × I)

For example the suspension of the pointed circle (S 1 , ∗) is the pointed 2-sphere
(S 2 , ∗).

S1 × 1
•
•
I 7−→ 7−→ •
∗×I
S1 × 0 •

ΣS 1
S1 •∗

If you collapse over one point only the upper and lower circles S 1 × 1 and S 1 × 0,
you already obtain a 2-sphere where the interval ∗ × I maybe is become the Green-
wich meridian. There remains to collapse also this meridian over a unique point,
a sort of unique pole of the suspension ΣS 1 , the base point of this suspension.
The symmetric meridian which crossed the Pacific ocean is become a “meridian”
running a whole big circle of the final sphere.
In general, any n-sphere is the suspension of the previous sphere: (S n , ∗) =
Σ(S n−1 , ∗). More generally, the suspension operator Σ is a functor : if f : (X, ∗) →
(Y, ∗) is a pointed map, that is f (∗) = ∗, then a natural process constructs Σf :
Σ(X, ∗) → Σ(Y, ∗).
The suspenction functor commutes with the wedge operator: there is a canon-
ical homeomorphism Σ(X ∨ Y ) ∼ = ΣX ∨ ΣY . Exercise.
So that if we apply the suspension functor to the map λ : S 1 → S 1 ∨ S 1 ,
the loop which runs both components of S 1 ∨ S 1 , then a new map is produced
λ2 : S 2 → S 2 ∨ S 2 . Both meridians Greenwich and its symmetric are collapsed
over the “central” base point of S 2 ∨ S 2 .

53
 λ2
•

• S2 S2 ∨ S2

In the same way, canonical maps λn : S n → S n ∨ S n are defined for every integer
n > 0.
It is natural to define πn X as the set of homotopy classes of pointed maps:
πn X := C((S n , ∗), (X, ∗))/ ∼. A multiplicative structure is defined as in dimen-
sion 1: γ · δ = (γ ∨ δ) ◦ λn . It is not hard to proof a group structure is so defined;
the group so obtained is in fact canonically isomorphic to π1 Ωn−1 X.

Proposition 47 — Let X be a pointed topological space. Then the homotopy

groups πn X are abelian for n ≥ 2.

On the contrary, the group π1 in general is not commutative. For example,

π1 (S 1 ∨ S 1 ) is the free non-commutative group with two generators.
Proposition 47 is a particular case of a more general result.

Proposition 48 — Let (X, ∗) be a connected space provided with a product

(x, y) 7→ x · y satisfying the following properties:

• ∗ · ∗ = ∗;
• A homotopy map h : I × X → X satisfies h(0, x) = x, h(1, x) = ∗ · x and
h(s, ∗) ≡ ∗: the left-product by ∗ is homotopic to the identity.
• A homotopy map k : I × X → X satisfies h(0, x) = x, h(1, x) = x · ∗ and
k(s, ∗) ≡ ∗: the same for the right-product.

Then the group π1 X is commutative.

In particular a topological group certainly satisfies these hypotheses.

♣ The proof is the figure:
γ
∗ γ·∗ k ∗

h γ·δ h
δ δ
∗·δ ∗·δ
γ·∗
∗ k ∗
γ

54
which defines a map I × I → X. The restriction to the twelve segments of the
figure is clearly shown. How to fill in the central square and the four trapezoids is
also clearly indicated. This figure illustrates why the path γ · δ = is homotopic
to the path δ · γ = . ♣
♣ [Proposition 47] If n ≥ 1, Ωn X := Ω(Ωn−1 X) is provided with a product
satisfying the properties required in Proposition 48: it is the path composition.
Therefore π1 Ωn X =: πn+1 X is a commutative group for n ≥ 1. ♣

11.2 Computability obstacles.

In general, even the computation of a π1 X is not so easy. The two following results,
stated without any proof, are in this respect rather striking.

Theorem 49 — There does not exist a general decision algorithm:

• Input: A finite simplicial complex K.

• Output: A boolean β ∈ {⊥, >}, with β = > if and only if π1 K = 0.

If X is connected and its Poincaré group π1 X is null, the space X is said simply
connected. It happens no general algorithm can decide whether this property is
true or not, even for “simple” spaces such as the realizations of finite simplicial
complexes. More precisely there exists a “semi -algorithm” α: if X is simply
connected, then the algorithm α working over X will terminate and prove π1 X = 0.
If π1 X 6= 0, then the algorithm does not terminate, but the user of this algorithm
will not be “informed”!
The set of simplicial complexes with k given vertices is finite and the iso-
morphism problem between them is easily solved. Considering successively the
isomorphism classes of simplicial complexes with 0 vertex, 1 vertex, 2 vertices,
. . . , k vertices, and so on, produces a countable list without repetition of all the
isomorphism classes of the finite simplicial complexes.

Theorem 50 — The set of isomorphism classes of simplicial complexes is count-

able, and can be presented as a sequence (Kn )n∈N . Then there exists an integer
n0 ∈ N satisfying:

• For every n < n0 , there exists a demonstration proving or disproving the

relation π1 Kn = 0;

• π1 Kn0 6= 0, but there des not exist a proof of this fact.

These strange results are consequences of the Novikov theorem about the un-
decidability of the word problem in Group Theory. Note the sequence (Kn )n∈N can
be explicitly constructed, for example by a computer program, but this integer n0
will remain definitively “hidden”. When the mathematicians will try to study

55
π1 Kn0 , they will not succeed in proving π1 Kn0 = 0, for it is false (!), but they
will not succeed either in proving π1 Kn0 6= 0, because such a proof does not exist!
And our mathematicians will definitively remain in doubt, unable to decide if this
n0 is this terrible number, or if more ordinarily they have not yet been skillful
enough to elucidate this mystery. This a striking avatar of the famous Gödel’s
incompleteness theorem, which has quite concrete consequences!

11.3 A recipe for “computing” the Poincaré group of a

finite simplicial complex.
This title seems to contradict the theorems stated in the previous section! On the
contrary this section will settle more clearly the obstacle, which illustrates also the
frequent confusions about the notion of “result” in Mathematics.
Let us consider the simple simplicial complex K made of 12 edges organized
as four contiguous squares.

• • •

K • • •

∗• • •

No triangle in this simplicial complex, the squares are hollow. The base point
is the down left vertex. The first step to “compute” the Poincaré group consists
in drawing a maximal tree rooted at the base point: this consists in taking the
maximum number of edges, connected to the base point by some path, but without
introducing any cycle. Such a maximal tree, shown by thick dashes overloading
the corresponding edges on this figure, could be for K:

• • •

K • • •

∗• • •

You observe adding any edge to this tree would generate a cycle, which is forbidden
in a tree. There remains to denote the other edges, those which are complementery
to this tree and, important, to orient them.

g4
• • •
g3
K • • •
g1 g2
∗• • •

56
Then the group π1 (K, ∗) is canonically isomorphic to the free non-commutative
group generated by these particular edges: π1 (K, ∗) =< g1 , g2 , g3 , g4 | >. The
recipe to obtain a representant of these generators is simple: start from the base
point, go to the origin of the marked edge along the maximal tree, a unique path,
run the edge, and go back to the base point again along the chosen tree. For
example the loop represented by the generator g4 runs the perimeter of the whole
square, in the clockwise sense.
A non-commutative free group G of finite type admits many bases, but with
the same cardinality n, for every basis of n elements induces an isomorphism
G/[G, G] ∼
= Zn and Zm ∼ = Zn implies m = n. We obtain in this way a proof that
the number of edges of a maximal tree is independent of the choice of this tree.
The same if you change the base point in the same connected component:

Proposition 51 — Let X be a connected topological space and x0 , x1 ∈ X. Then

every path connecting x0 and x1 in X induces a canonical isomorphism π1 (X, x0 ) ∼
=
π1 (X, x1 ).

♣ Exercise. ♣
Our recipe is valid only for a 1-dimensional simplicial complex, without any
triangle. Let us decide to fill the down-left square by two triangles to obtain the
new simplicial complex K 0 :

• • •

K0 • • •

∗• • •

Because of the triangles, one further edge is needed. The maximal tree can be the
same as before, which could give the figure:

g4
• • •
g3
K0 • g1 • •
g2
g5
∗• • •

We have now five generators, because of the added edge. But each triangle is the
source of one relation in the resulting group, consisting in expressing the path
running the perimeter of a triangle is homotopic to the constant loop, that is, the
“null” loop. If we run the triangle perimeters in the counterclockwise sense, we
obtain here the two relations: g5−1 = ∗ and g5 g1−1 = ∗. In the traditional notation
of finitely presented groups, we obtain:

π1 (K 0 , ∗) =< g1 , . . . , g5 |g5−1 , g5 g1−1 > (69)

57
But the first relation simply cancels the generator g5 and the second one in turn
cancels the generator g1 , so that finally:

π1 (K 0 , ∗) =< g2 , g3 , g4 | > (70)

The role of the filled square is simply, with respect to the initial situation with K,
to cancel the generator g1 , for the loop running the boundary of the added square
is now homotopic to the trivial loop.
These explanations should be enough for the general case. You do not have
to take account of the possible simplices in dimensions ≥ 3. To be sure you have
understood, consider the natural triangulation of the 2-torus T2 = S 1 × S 1 by 18
triangles:

∗
•

maybe more easily described by the planar figure:

∗ ∗
• • • •

• • • •

• • • •
∗ ∗
• • • •

Pay attention to the repetitions in this figure, where on one hand the left and
right edges of the total square are to be identified, which gives maybe the torus
meridian in front of us on the 3D figure; in the same way the upper and lower
edges are to be identified, giving some “parallel” of our torus; so that finally the
four corners of this representation represent in fact a unique point of the torus,
maybe the basepoint highlighted on the 3D figure.
The preparatory work to compute π1 (T2 ) could be this figure:

58
 ∗ ∗
• • • h3 •
v7 d7 v8 d8 v9 d9 v7
• h7 • h8 • h9 •
d4 d5 d6
• h4 • h5 • h6 •
d1 d2 d3
∗ ∗
• • • h3 •

The maximal tree is made in fact of only 8 edges; adding any other edge would
generate a cycle, in particular for example if you finish the lefthand edge of the
global square, because the down left and up left corners represent the same point of
the torus, the base point: adding this edge would produce a whole torus meridian,
certainly a cycle.
27 edges in our simplicial complex, so that the 19 remaining edges are the gen-
erators of our group. With obvious conventions in naming the edges and orienting
them, we find:

π1 (T2 ) =< h3 , . . . , h9 , d1 , . . . , d9 , v7 , v8 , v9 |R > (71)

where the relations R are made of 18 elements, one for each triangle:

d−1
1 , d−1
2 , h3 d−1
3 , d1 h−1 −1
4 , d2 h5 , d3 h−1
6 ,
h4 d−1
4 , h d
5 5
−1
, h d
6 6
−1
, d h
4 7
−1
, d h−1
5 8 , d h−1
6 9 , (72)
h7 v8 d−1 −1 −1 −1 −1 −1 −1
7 , h8 v9 d8 , h9 v7 d9 , d7 v7 , d8 v8 , d9 h3 v9 .

But we can use these relations to cancel successively many generators, for example:

d1 , d2 , h4 , h5 , d4 , d5 , h7 , h8 . (73)

Also:
v7 = d7 = v8 = d8 = v9 and h3 = d3 = h6 = d6 = h9 (74)
which leads to the simplified presentation:

< h3 , v7 , d9 |d9 = v9 h3 = h3 v9 >=< h, v|hv = vh > (75)

It is the group with two commuting generators. In other words π1 T2 = Z2 . Note

that S 1 is the group of the complex numbers of modulus 1, and T2 = S 1 × S 1
can be provided with the product group structure. In particular T2 satisfies the
properties of Proposition 48, and the commutativity of the π1 was in this case
mandatory.
How make compatible the relatively simple method described in this section
and the negative results of the previous section? The process described here allows
one to easily give a finite presentation of the Poincaré group of any finite simplicial

59
complex, but a problem remains: such a finite presentation is in a sense a partial
result. For example the isomorphism problem between such groups is in general
undecidable: no general algorithm can consider two finite presentations Pr1 and
Pr2 respectively “describing” some groups G1 and G2 and then decide whether
these groups are isomorphic or not. And claiming that our combinatorial method
“computes” the Poincaré group has in fact a limited scope.
The problem would be solved if, given some finite presentation Pr of some
group, it would exist a canonical process to obtain from this maybe complicated
presentation the simplest one, a process usually called the search of a normal form
among equivalent presentations. It fortunately or unfornately happens, depending
of the point of view, such a normalization process in general does not exist, oth-
erwise the Novikov theorem about the undecidability of the word problem would
be false.

11.4 Analogous methods in higher dimensions?

Methods of this sort do not exist! The combinatorial methods describing the
higher homotopy groups exist only for the Kan simplicial sets, and these simplicial
sets, even for a very simple initial simplicial set, are not in general of finite type.
For example any Kan simplicial set having the homotopy type of the 2-sphere
S 2 has an infinite number of non-degenerate simplices of dimension n for any
dimension n ≥ 2. This is essentially caused by the deep topological nature of these
homotopy groups. These difficulties are from another respect paradoxical : these
higher homotopy groups are necessarily commutative, and of finite type for any
“reasonable” space, a striking result due to Jean-Pierre Serre, thanks to his famous
spectral sequence.
We nevertheless describe here the main components of the combinatorial def-
inition of the homotopy groups, due to Daniel Kan. This definition cannot be
directly used for computations, but is useful in many cases, typically to easily
understand the notion of Postnikov tower.
We work with a pointed Kan simplicial set (K, ∗), the base point ∗ being some
vertex ∗ ∈ K0 , the set of 0-simplices. Definition 45 ensures the existence of some
extension function ε:
ε : (m, i, σ0 , . . . , σi−1 , σi+1 , . . . , σm+1 ) 7→ σ (76)
where the input is a Kan (m, i)-hat (see Definition 43) and the outpout is a filling
of this hat (Definition 44). The component m is redundant in the input but the
component i is not! Look at the discussion p.49 to understand why this component
is important.

‘σ1 ’
• • • •
ε
(m = 1, i = 1) σ
σ2 σ0 σ2 σ0
• •

60
This extension function is a sort of “simplex composition”: when m+1 m-simplices
are coherently contiguous to each other, then the filling produces another m-
simplex to be considered as the composition of the input simplices. In the simplest
case of the figure above, the edge σ1 is the composition of σ0 and σ2 produced by ε;
the justification of this composition is the simplex σ to be considered as a homotopy
between σ0 · σ2 and σ1 .

Definition 52 — An n-sphere of the pointed simplicial complex (K, ∗) is an n-

simplex σ ∈ Kn satisfying ∂i σ = ∗n−1 for 0 ≤ i ≤ n.

More precisely ∂i σ = ηn−2 · · · η0 ∗ =: ∗n−1 : every face of our n-sphere is the base
point degenerated up to dimension n − 1. Such a sphere defines a canonical simpli-
cial map γ : S n → K from the n-sphere, presented as a simplicial set as explained
Section 4.3, toward our simplicial set K: the fundamental simplex of S n is sent
over σ and this is enough to define our simplicial map.

Proposition 53 — Let σ and σ 0 be two n-spheres of the Kan pointed simplicial

set (K, ∗). Then (n, n, ∗n , . . . , ∗n , σ, σ 0 ) is a Kan (n, n)-hat. This allows to define
the composition of these spheres:

σ · σ 0 := ∂n ε(n, n, ∗n , . . . , ∗n , σ, σ 0 ). (77)

It is important to decide the “missing” simplex is in position n, between the

first factor σ in position n − 1 and the second one σ 0 in position n + 1. This comes
from the common rule about alternating signs reflecting the orientations of faces:
we must put σ and σ 0 in such a way they have naturally the same sign, which
needs a gap between them, gap which will be filled by the composition, which is
to be interpreted as if σ − (σ · σ 0 ) + σ 0 = 0, the wished relation.
To finish to convince the reader about this process, let us decide to define the
composition as:
σ · σ 0 := ∂n+1 ε(n, n + 1, ∗n , . . . , ∗n , σ, σ 0 ). (78)
where this time the simplex in position n + 1 is missing. In particular, if σ = σ 0 ,
we obtain:
σ · σ := ∂n+1 ε(n, n + 1, ∗n , . . . , ∗n , σ, σ). (79)
But there is always in this case a trivial solution for the Kan extension, namely
ηn−1 σ, for ∂n−1 ηn−1 σ = ∂n ηn−1 σ = σ and the other faces of ηn−1 σ are the base
point. We would obtain σ · σ = ∗n , this relation is correct with the right signs:
(−1)n−1 σ + (−1)n σ = ∗n ; but nothing can be concluded about the composition of
σ and itself, that is, with the same sign.

Definition 54 — Let (K, ∗) be a pointed simplicial set satisfying the Kan ex-
tension condition. Then two n-spheres σ and σ 0 are homotopic if there exists an
(n + 1)-simplex ρ such that ∂i ρ = ∗n for 0 ≤ i ≤ n − 1, ∂n ρ = σ and ∂n+1 ρ = σ 0 .

61
For example the discussion preceding this Definition was simply a proof any sphere
is homotopic to itself.

Proposition 55 — Let (K, ∗) be a pointed simplicial set satisfyng the Kan exten-
sion condition. Let Pn (K, ∗) be the set of n-spheres of K. Then:

πn (K, ∗) := Pn (K, ∗)/ ∼ (80)

is the quotient of the n-sphere set byt the homotopy relation defined above, which
is an equivalence relation. π0 (K, ∗) is naturally identified to the connected compo-
nents of K, pointed by the one containing the base point ∗. The composition of
spheres define a group structure over πn (K, ∗) for n ≥ 1, commutative for n ≥ 2.

♣ The proof consists in copying which was explained in the topological case in
Section 11.1, replacing the various topological constructions by simplicial ones,
frequently using the Kan condition to guarantee such constructions are possible.
See [13, Section I] for a detailed discussion. ♣

12 Simplicial fibrations.
A fibration is a map p : E → B between a total space E and a base space B
satisfying a few properties describing more or less the total space E as a twisted
product F ×τ B. In the simplicial context, several definitions are possible. The
notion of Kan fibration corresponds to a situation where a simplicial homotopy
lifting property is satisfied; to state this property, the elementary datum is a Kan
hat in the total space and a given filling of its projection in the base space; the
Kan fibration property is satisfied if it is possible to fill the Kan hat in the total
space in a coherent way with respect to the given filling in the base space. This
notion is the simplicial version of the notion of Serre fibration, a projection where
the homotopy lifting property is satisfied for the maps defined on polyhedra. The
reference [13] contains a detailed study of the basic facts around Kan fibrations,
see [13, Chapters I and II].
We will examine with a little more details the notion of twisted cartesian prod-
uct, corresponding to the topological notion of fibre bundle. It is a key notion in
topology, and the simplicial framework is particularly favourable for several rea-
sons. In particular the Serre spectral sequence becomes well structured in this
environment, allowing us to extend it up to a constructive version, one of the main
subjects of another lecture series of this Summer School. We give here the basic
necessary definitions for the notion of twisted cartesian product.
A reasonably general situation consists in considering the case where a sim-
plicial group G acts on the fiber space, a simplicial set F , the fiber space. As
usual this means a map φ : F × G → F is given; source and target are simpli-
cial sets, the first one being the product of F by the simplicial set G, underlying
the simplicial group; the map φ is a simplicial map; furthermore each component
φm : (F × G)m = Fm × Gm → Fm must satisfy the traditional properties of the

62
right actions of a group on a set. We will use the shorter notation f. g instead of
φ(f, g). Let also B be our base space, some simplicial set.

Definition 56 — A twisting operator τ : B → G is a family of maps {τm : Bm →

Gm−1 }m≥1 satisfying the following properties.

1. ∂0 τ (b) = τ (∂1 b)τ (∂0 b)−1 ;

2. ∂i τ (b) = τ (∂i+1 (b)) if i ≤ 1;

3. ηi τ (b) = τ (ηi+1 b);

4. τ (η0 b) = em if b ∈ Gm+1 , the unit element of Gm being em .

In particular it is not required τ is a simplicial map, and in fact, because of

the degree -1 between source and target dimensions, this does not make sense.

Definition 57 — If a twisting operator τ : B → G is given, the corresponding

twisted cartesian product E = F ×τ B is the simplicial set defined as follows. Its
set of m-simplices Em is the same as for the non-twisted product Em = Fm × Bm ;
the face and degeneracy operators are also the same as for the non-twisted product
with only one exception: ∂0 (f, b) = (∂0 f. τ (b), ∂0 b).

The twisting operator τ , the unique ingredient at the origin of a difference

between the non-twisted product and the τ -twisted one, acts in the following way:
the twisted product is constructed in a recursive way with respect to the base
dimension. Let B (k) be the k-skeleton of B and let us suppose F ×τ B (k) is already
constructed. Let σ be a (k + 1)-simplex of B; we must describe how the product
F × σ is to be attached to F × B (k) ; what is above the faces ∂i σ for i ≥ 1
is naturally attached; but what is above the 0-face is shifted by the translation
defined by the operation of τ (b). It is not obvious such an attachment is coherent,
but the various formulas of Definition 57 are exactly the relations which must
be satisfied by τ for consistency. It was not obvious, starting from scratch, to
guess this is a good framework for working simplicially about fibrations; this was
invented (discovered ?) by Daniel Kan [12]; the previous work by Eilenberg and
MacLane [5, 6] in the particular case of the fibrations relating the elements of the
Eilenberg-MacLane spectra was probably determining.

12.1 The simplest example.

Let us describe in this way the exponential fibration exp : R → S 1 : t → e2πit . The
base space B is S 1 , a simplicial model with one vertex ∗0 and one non-degenerate
edge σ = id(1) had been described Section 4.3. The structural group G is Z, a
discrete simplicial group, see Section 4.1: any simplex set Zn is simply Z, and for
every ∆-morphism α : m → n, the induced map α : Zn = Z → Z = Zm is the
identity; every simplex of dimension ≥ 1 is degenerate. The fiber space F = Z is

63
again the structural group, acting on itself according to the group law. When this
is the case, the fibration is called principal.
If we define the twisting operator 0 = τ : S 1 → Z • •
as the trivial one, τ (σ) = 0, we obtain simply the • •
trivial product Z × S 1 . Because of the required com- Z• • Z × S1
patibility with degeneracy operators, this is enough to • •
define our twisting operator, and it it easy to verify • •
all the required properties are satisfied.
S1
Let us now define a non-trivial twisting operator
• • deciding τ (σ) = 1; again a unique coherent way to
• • extend this definition to all the degenerate simplices
Z• • Z ×τ S 1
of S 1 . Let us carefully consider which happens for the
• • 1-simplex (η0 n, σ), the first component being the 1-
dimensional degeneracy of the integer n ∈ Z. Nothing
S1 happens for the face 1: ∂1 (η0 n, σ) := (∂1 η0 n, ∂1 σ) =
(n, ∗0 ).
On the contrary, for the face 0, because of the formula:

∂0 (f, b) := (∂0 f · τ (b), ∂0 b) (81)

we obtain: ∂0 (η0 n, σ) := (∂0 η0 n · τ (σ), ∂0 σ) = (n · 1, ∗0 ), but the right action of Z

over itself, produces ∂0 (η0 n, σ) = (n + 1, ∗0 ). Because of the twisting operator,
every circle in the product is in some sense “broken”: it starts at the floor n,
but arrives at the floor n + 1. The realization of this twisted product therefore is
homeomorphic to the real line, and the projection Z ×τ S 1 → S 1 is isomorphic to
the exponential map t 7→ e2πit .

12.2 Fibrations between K(π, n)’s.

Let us recall (see Section 9.1) E(π, d) is the simplicial set defined by E(π, d)m =
C d (∆m , π) (only normalized cochains) and K(π, n) is the simplicial subset made
of the cocycles. The maps between simplex sets to be associated to ∆-morphisms
are naturally defined. A simplicial projection p : E(π, d) → K(π, d+1) associating
to an m-cochain c its coboundary δc, necessarily a cocycle, is also defined. The
simplicial set ∆m is contractible, its cochain complex is acyclic and the kernel of p,
the potential fibre space, is therefore the simplicial set K(π, d). The base space is
clearly the quotient of the total space by the fibre space (principal fibration), and
a systematic examination of such a situation (see [13, Section 18]) shows E(π, d)
is necessarily a twisted cartesian product of the base space K(π, d + 1) by the fiber
space K(π, d).
It is not so easy to guess a corresponding twisting operator. A solution is
described as follows; let z ∈ Z d+1 (∆m , π) a base m-simplex; the result τ (z) ∈
Z d (∆m−1 , π) must be a d-cocyle of ∆m−1 , that is a function defined on every
(d + 1)-tuple (i0 , . . . , id ), with values in π, and satisfying the cocycle condition; the

64
solution τ (z)(i0 , . . . , id ) = z(0, i0 + 1, . . . , id + 1) − z(1, i0 + 1, . . . , id + 1) works,
but seems a little mysterious. The good point of view consists in considering the
notion of pseudo-section for the studied fibration; an actual section cannot exist if
the fibration is not trivial, but a pseudo-section is essentially as good as possible;
see the definition of pseudo-section in [13, Section 18]. When a pseudo-section is
found, a simple process produces a twisting operator; in our example, the twisting
operator comes from the pseudo-section ρ(z)(i0 , . . . , id ) = z(0, i0 + 1, . . . , id + 1),
quite natural.
The fibrations between Eilenberg-MacLane spaces are a particular case of uni-
versal fibrations associated to simplicial groups. See [13, Section 21].

12.3 Simplicial loop spaces.

A simplicial set X is reduced if its 0-simplex set X0 has only one element. We
have given in Section 9.2 the Kan combinatorial version GX of the loop space of
X. This loop space is the fiber space of a co-universal fibration:

GX ,→ GX ×τ X → X. (82)

Only the twisting operator τ remains to be defined. The definition is simply. . .

τ (σ) := τ (σ) for both possible meanings of τ (σ); the first one is the value of the
twisting operator to be defined for some simplex σ ∈ Xm+1 and the second one
is the generator of GXm corresponding to σ ∈ Xm+1 , the unit element of GXm if
ever σ is 0-degenerate (see Section 9.2). The definition of the face operators for
GX are exactly those which are required so that the twisting operator so defined
is coherent.
It is again an example of principal fibration, that is the fiber space is equal
to the structural group and the action GX × GX → GX is equal to the group
multiplication. This fibration is co-universal, with respect to X; in fact, let H ,→
p
H ×τ 0 X → X another principal fibration above X for another twisting operator
τ 0 : X → H. Then the free group structure of GX gives you a unique group
homomorphism GX → H inducing a canonical morphism between both fibrations.
If the simplicial space X is 1-reduced (only one vertex, no non-degenerate 1-
simplex), then an important result by John Adams [1] allows one to compute
the homology groups of GX if the initial simplicial set X is of finite type; an
intermediate ingredient, the Cobar construction, is the key point. One of the main
problems in Algebraic Topology consists in solving the analogous problem for the
iterated loop spaces Gn X when X is n-reduced; it is the problem of iterating the
Cobar construction; one of the lecture series of this Summer School is devoted to
this subject, organized around a constructive version of Algebraic Topology.

65
13 Homotopy groups: a quick survey.
13.1 Sphere homotopy groups.
In the same spirit as in Section 7, this section is a “cultural” presentation of the
most elementary facts around the computation of homotopy groups.
The first natural question is the comparison problem between homology and
homotopy groups. For simple spaces, it is common to observe the homology groups
are relatively easy to be computed, and on the contrary the homotopy groups can
be very hard.
The first amazing result of this sort is for the spheres: the homology groups are
easily computed, it was done in these elementary notes as a consequence of Propo-
sition 33. On the contrary, computing the homotopy groups of spheres remains
a difficult subject far from being totally understood today, some popular “con-
jectures” meaning it is essentially impossible to achieve it. Typically the unique
non-trivial homology group of the 4-sphere is H4 S 4 = Z, while the homotopy
groups of the same sphere are rather chaotic:

n 4 5 6 7 8 9 10 11 12 13 · · ·
πn S Z Z2 Z2 Z12 ⊕ Z Z2 Z2 Z3 ⊕ Z24 Z15 Z2 Z32 · · ·
4 2 2

where Zk means Z/kZ as it is common in algebraic topology. This dissymetry is

amazing, but Hilton’s duality gives an interesting explanation. This duality ex-
presses on the contrary there is a good symmetry between homology and homotopy
groups, at least if you choose the good point of view.
A sphere is simply (!) simple, but another simplicity is precisely the fact that
its homology groups are. . . simple: an n-sphere is a Moore space with only one
non-trivial homology group in dimension n, which is denoted by the relation S n =
MS(Z, n) (MS = Moore space). Such a Moore space is unique up to homotopy
equivalence.
Symmetrically a homotopically simple space should have only one non-trivial
homotopy group; such a space does exist, it is also unique up to homotopy equiv-
alence, it is the Eilenberg-MacLane space K(π, n) studied in Section 9.1. For
example the sphere S 4 is the (up to homotopy equivalence) space with H4 = Z as
unique non-trivial homology group. In front of this sphere, it is natural to con-
sider the space, unique up to homotopy equivalence, having π4 = Z as the unique
non-trivial homotopy group; this space is the Eilenberg-MacLane space K(Z, 4),
its homotopy groups are simple, it is the very definition of this space, but the
homology groups are complicated:

n 4 5 6 7 8 9 10 11 12 13 · · ·
Hn K(Z, 4) Z 0 Z2 0 Z3 ⊕ Z 0 Z2 ⊕ Z2 0 Z60 ⊕ Z Z2 · · ·

However the symmetry is not totally perfect: the structure of the homology
groups of the Eilenberg-MacLane spaces is well understood, for a long time, due

66
to Henri Cartan’s work in the fifties [3]. On the contrary the entire structure of
the homotopy groups of the spheres seems today out of scope for long.

13.2 Hilton’s duality continued.

The right continuation of the story is along the same line. The Mayer-Vietoris
exact sequence is a simple process often allowing one to compute the homology
groups of A∪B when the homology groups of A, B and A∪B are known. However
note the Mayer-Vietoris method is not an algorithm, because of a computability
lack; transforming it and the other analogous exact and spectral sequences into
actual algorithms is relatively recent; the article [18] gives in particular an account
about this matter of computability in Algebraic Topology.

13.2.1 Serre exact sequence.

No Mayer-Vietoris exact sequence for the homotopy groups, but there is such an
exact sequence for the fibrations. The respective roles of fibrations and cofibra-
tions for homology and homotopy are exchanged. The sophisticated tool “spectral
sequence” is required to study the homology of a fibration, but for the difficult
homotopy groups, the simple tool exact sequence is enough.

Proposition 58 (Serre exact sequence) — Let F ,→ E → B be a fibration

(see Section 12) with a connected base space B. Then a long exact sequence connect
the homotopy groups of the components:

· · · ←− πn−1 F ←− πn B ←− πn E ←− πn F ←− πn+1 B ←− · · · (83)

The morphisms πn F → πn E and πn E → πn B are naturally induced by the

maps defining the fibration. The connection morphism πn B → πn−1 F is more
esoteric.

13.2.2 Homotopy groups of S 1 .

Two elementary examples, already considered in these notes, are convenient to

understand how to use this exact sequence. The first one is the fibration:

Z ,→ R → S 1 (84)

which was used to illustrate the notion of twisting operator in Section 12.1. The
situation is particularly simple but already interesting. The total space is con-
tractible, so that all its homotopy groups are null, πn R = 0 for every n ∈ N. The
fiber space Z is a discrete space. In general, because of the role of the base point
when defining the homotopy groups, πn X = πn X0 for n ≥ 1 if X0 is the connected
component of the base point: the other connected components cannot play any
role for n ≥ 1. Here the base point is 0 ∈ Z and the connected component of 0 is

67
{0}, a point, so that πn Z = 0 for every n ≥ 0. On the contrary, see Section 11.1.2,
π0 X is just the set of the connected components, and therefore π0 Z = Z which in
this particular case is a group.
The Serre exact sequence gives in particular for n ≥ 2:

(πn−1 Z = 0) ←− πn S 1 ←− (πn R = 0) (85)

and a group placed in an exact sequence between two null groups is null also, so
that this proves πn S 1 = 0 for n ≥ 2. The circle S 1 is connected and π0 S 1 = 0. To
determine π1 S 1 , again we use the Serre exact sequence:

(π0 R = 0) ←− (π0 Z = Z) ←− π1 S1 ←− (π1 R = 0) (86)

In an exact sequence, two consecutive groups placed between two other groups
which are null are necessarily isomorphic. This proves the relation π1 S 1 = Z.
In this particular case, the connection morphism ∂ : π1 S 1 → π0 Z is easy to
understand. Remembering the circle can be viewed as the unit circle of the complex
plane, a loop γ : [0, 1] → S 1 of the circle is also a loop γ : [0, 1] → (C∗ := C − {0}),
but such a loop has an index :
Z
1 dz
αγ := (87)
2iπ γ z

which counts the number of “rounds” of the path γ around the origin of C; this
index is an integer and ∂(γ) = αγ .
This intrusion of the complex plane and its analysis environment could not be
generalized to other connection morphisms. The key point is not at all complex
analysis, it is a matter of being able to lift appropriate maps. The projection
exp : R → S 1 satisfies an essential property: if γ : [0, 1] → S 1 is a continuous
map and if x0 ∈ R satisfies exp(x0 ) = γ(0), then there exists a unique lifting
b : [0, 1] → R satisfying γ
γ b(0) = x0 and exp ◦ γ
b = γ:

x0
{0} R
γ
b exp

[0, 1] γ S1

A compacity argument allows to construct the lifting γ b by pieces concern-

ing only small “segments” [α, α + ε] of the circle, where the inverse image
exp−1 ([α, α + ε]) is homeomorphic to the trivial product Z × [α, α + ε]. This last
property is typical of a covering. More generally, the homotopy lifting property,
generalizing our situation, with lifting in general non-unique, allows to prove the
Serre exact sequence, see for example [22, Section 7.2].
So that only one homotopy group of S 1 is non-null, and the circle is an example
of Eilenberg-MacLane space: S 1 = K(Z, 1). Note the standard model of K(Z, 1)

68
described in Section 9.1 is a “monster” of infinite dimension, but because of the
unicity of an Eilenberg-MacLane space K(π, n), certainly this monster6 K(Z, 1)
has the same homotopy type as the simple circle S 1 . In fact it is not hard, using
the method of round counts to construct a canonical map |K(Z, 1)| → S 1 . The
inverse map up to homotopy is the canonical inclusion S 1 ,→ |K(Z, 1)| the image
of which is the edge [1] of K(Z, 1).

13.2.3 The Hopf fibration revisited.

The next example is the Hopf fibration already considered p.37:

S 1 ,→ S 3 → S 2 (88)

We know πn S 1 = 0 for n ≥ 2 and the Serre exact sequence in particular produces

the exact sequence segment for n ≥ 3:

(πn−1 S 1 = 0) ←− πn S 2 ←− πn S 3 ←− (πn S 1 = 0) (89)

where the central morphism placed between two null groups is an isomorphism.
We will see a little further a simple reason proving π3 S 3 = Z, so that π3 S 2 = Z,
which was discovered by Hopf in 1935 and at this time was a surprise: it was
rather expected that, as for the homology groups, πn S 2 = 0 for n ≥ 3; in fact we
know now an infinity of groups πn S 2 are not null. And these groups are the same
for S 2 and S 3 , not really intuitive!
It can be proved by relatively simple means that πi S n is null for i < n: any
continuous map S i → S n is homotopic to a simplicial map modulo a subdivision
of the source space [7, II.7], but such a map is not surjective, and S n − {∗} ∼= Rn
is contractible. In particular π1 S 3 = π2 S 3 = 0. So that this segment of the Serre
exact sequence:

(π1 S 3 = 0) ←− π1 S 1 ←− π2 S 2 ←− (π2 S 3 = 0) (90)

proves π2 S 2 ∼
= π1 S 1 = Z.

13.2.4 Hurewicz and Adams.

The isomorphism π2 S 2 = Z is confirmed by the Hurewicz theorem, quite funda-

mental.

Theorem 59 (Hurewicz Theorem) — Let X be a simply connected space. If

Hi X = 0 for 1 ≤ i ≤ n − 1 with some n ≥ 2, then there exists a canonical
isomorphism Hn X ∼
= πn X.
6
An amusing fact: this monster is the minimal (!) model of S 1 in Kan’s theory, see [13,
Section 9]

69
♣ [13, §13] ♣
This is simply understood as follows: the first non trivial homology and ho-
motopy groups are canonically isomorphic. Note in particular H0 X being the free
Z-group generated by the connected components of X, it is isomorphic to Z if X
is simply connected and in particular connected. But this does not matter in the
Hurewicz theorem. In the particular case n = 1, there is an analogous result:

Theorem 60 (Poincaré Theorem) — Let X be a connected space. Then there

is a canonical surjective map π1 X → H1 X the kernel of which is the commutator
subgroup [π1 X, π1 X].

♣ [13, §13] ♣
In other words this produces a canonical exact sequence:

0 ←− H1 X ←− π1 X ←- [π1 X, π1 X] ←− 0 (91)

inducing a canonical isomorphism H1 X ∼= π1 X/[π1 X, π1 X]: The first homology

group is the first homotopy group made commutative by division by the normal
subgroup generated by the commutators [a, b] := aba−1 b−1 , this normal subgroup
is known as the commutator subgroup.
Let us go back to the Hurewicz theorem for the spheres: if n ≥ 2, then π1 S n =
0, for 1 ≤ n and the Hurewicz theorem can be applied. The first non-trivial
homology group Hn S n = Z therefore is isomorphic to the first non-trivial homotopy
group πn S n = Z.
The Hurewicz theorem, combined with the Serre spectral sequence, can be
used to compute the homotopy groups as follows. More generally let X be a
simply connected space the first non-trivial homotopy group being πn X = G,
some commutative group. Then it can be proved it is possible to construct a
canonical fibration:

K(G, n − 1) ,→ K(G, n − 1) ×τ X → X (92)

where the fiber space is an Eilenberg-MacLane space K(G, n − 1): its unique
non-trivial homotopy group is πn−1 K(G, n − 1) = G; furthermore the twisting
operator τ can be chosen in such a way the connection morphism of the Serre
exact sequence πn X → (πn−1 K(G, n − 1) = G) is the isomorphism πn X ∼ = G.
Then working exactly as for the Hopf fibration, it was in fact a particular case of
this situation, we obtain two results:

• πn (K(G, n − 1) ×τ X) = 0;

• For i ≥ n + 1, πi (K(G, n − 1) ×τ X) ∼
= πi X.

Let us denote the total space K(G, n − 1) ×τ X =: Xn+1 . You see this total space
Xn+1 has the same homotopy groups as X except πn Xn+1 = 0: one says Xn+1
is X where the πn = G has been killed. This allows the topologist to apply again

70
the Hurewich theorem to Xn+1 : we know now that πn+1 Xn+1 ∼ = Hn+1 Xn+1 . If the
homology groups of K(G, n − 1) are known and if we can apply the Serre spectral
sequence 7.5.3, then we could compute the homology groups of Xn+1 and decuce
in particular of Hn+1 Xn+1 the group πn+1 Xn+1 ∼
= πn+1 X, and we could continue.
Let us apply for example this process to the 3-sphere S 3 . The first homology
group of S 3 is H3 S 3 = Z, which allows us to construct a fibration:
K(Z, 2) ,→ (X4 = K(Z, 2) ×τ S 3 ) → S 3 (93)
It can be proved, using the homology groups of S 3 and K(Z, 2) that the first
homology group of X4 is H4 X4 = Z/2Z; therefore Z/2Z = H4 X4 = π4 X4 = π4 S 3 ,
which proves π4 S 3 = Z/2Z.
The next step will produce a fibration:
K(Z/2Z, 3) ,→ (X5 = K(Z/2Z, 3) ×τ X4 ) → X4 (94)
It can be proved, using the homology groups of X4 and K(Z/2Z, 3) that the first
homology group of X5 is H5 X5 = Z/2Z; therefore Z/2Z = H5 X5 = π5 X5 =
π5 X4 = π5 S 3 , which proves π5 S 3 = Z/2Z.
And so on, but the computations become more complicated; Jean-Pierre Serre,
using essentially these ingredients, succeeded in 1950 in computing almost all the
groups πi S n for i ≤ 10, which was an enormous progress with respect to which
was known before: very few. He was rewarded for this work by a Fields Medal in
1954.
A sophisticated and powerful extension of the Hurewicz theorem is the Adams
spectral sequence: it is a spectral sequence which essentially starts with the homol-
ogy groups of X and which converges to the homotopy groups of X. But this tool
is totally out of scope in these elementary notes. All the modern research works
around the homotopy groups are now based upon the Adams spectral sequence,
see for example [17].

References
[1] J. Frank Adams. On the Cobar construction. Proceedings of the National
Academy of Science of the U.S.A., 1956, vol. 42, pp 409-412.
[2] Gunnar Carlsson and R. James Milgram. Stable homotopy and iterated loop
spaces. in [11], pp. 505-583.
[3] Henri Cartan. Algèbres d’Eilenberg-MacLane et homotopie. Séminaire Henri
Cartan, 1954-5. Or Henri Cartan, Œuvres, Springer, 1979.
[4] Dominique Dancète. Sur la Cobar construction. Thèse, Grenoble, Institut
Fourier, 1998
[5] Samuel Eilenberg, Saunders MacLane. On the groups H(π, n), I. Annals of
Mathematics, 1953, vol. 58, pp 55-106.

71
 [6] Samuel Eilenberg, Saunders MacLane. On the groups H(π, n), II. Annals of
Mathematics, 1954, vol. 60, pp 49-139.

[7] Samuel Eilenberg and Norman Steenrod. Foundations of Algebraic Topology.

Princeton University Press, 1952.

[8] Rudolf Fritsch and Renzo A. Piccinini. Cellular structures in topology. Cam-
bridge University Press, 1990.

[9] Sergei I. Gelfand and Yuri I. Manin. Methods of Homological Algebra.

Springer-Verlag, 1996.

[10] Paul G. Goerss and John F. Jardine. Simplicial Homotopy Theory. Birkhäuser,
1999.

[11] Handbook of Algebraic Topology (Edited by I.M. James). North-Holland

(1995).

[12] Daniel M. Kan. A combinatorial definition of homotopy groups. Annals of

Mathematics. 1958, vol. 67, pp. 282-312.

[13] J. Peter May. Simplicial objects in algebraic topology. Van Nostrand, 1967.

[14] John McCleary. A user’s guide to spectral sequences. Cambridge Universityt

Press, Second Edition, 2001.

[15] Saunders MacLane. Homology. Springer-Verlag, 1975.

[16] Michael A. Mandell. Cochains and homotopy types. Publications Mathéma-

tiques de l’IHES, vol 103, pp 213-246.

[17] Douglas C. Ravenel. Complex cobordism and stable homotopy groups of

spheres. Academic Press, 1986.

[18] Julio Rubio, Francis Sergeraert. Constructive Algebraic Topology. Bulletin des
Sciences Mathématiques, 2002, vol. 126, pp. 389-412.

[19] Julio Rubio, Francis Sergeraert. Algebraic Models for Homotopy Types. Ho-
mology, Homotopy and Applications, 2005, vol.7, pp.139160.

[20] Julio Rubio, Francis Sergeraert. Genova Lecture Notes.

www-fourier.ujf-grenoble.fr/∼sergerar/Papers/Genova-Lecture-Notes.pdf

[21] Rolf Schön. Effective algebraic topology. Memoirs of the American Mathemat-
ical Society, 1991, vol. 451.

[22] Edwin H. Spanier. Algebraic Topology. McGraw Hill, 1966.

72