Professional Documents
Culture Documents
Christian Bar Topology
Christian Bar Topology
GEOMETRIE IN POTSDAM
Algebraic Topology
Lecture Notes, Winter Term 2016/17
2 Homotopy Theory 17
2.1 Homotopic maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2 The fundamental group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.3 The fundamental group of the circle . . . . . . . . . . . . . . . . . . . . . . . 28
2.4 The Seifert-van Kampen theorem . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.5 The fundamental group of surfaces . . . . . . . . . . . . . . . . . . . . . . . . 52
2.6 Higher homotopy groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
2.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
3 Homology Theory 79
3.1 Singular homology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
3.2 Relative homology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
3.3 The Eilenberg-Steenrod axioms and applications . . . . . . . . . . . . . . . . 87
3.4 The degree of a continuous map . . . . . . . . . . . . . . . . . . . . . . . . . 96
3.5 Homological algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
3.6 Proof of the homotopy axiom . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
3.7 Proof of the excision axiom . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
3.8 The Mayer-Vietoris sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
3.9 Generalized Jordan curve theorem . . . . . . . . . . . . . . . . . . . . . . . . 126
3.10 CW-complexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
3.11 Homology of CW-complexes . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
3.12 Betti numbers and the Euler number . . . . . . . . . . . . . . . . . . . . . . . 140
3.13 Incidence numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
3.14 Homotopy versus homology . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
3.15 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
i
Contents
Bibliography 159
Index 161
ii
Preface
These are the lecture notes of an introductory course on algebraic topology which I taught
at Potsdam University during the winter term 2016/17. The aim was to introduce the basic
tools from homotopy and homology theory. Choices concerning the material had to be made.
Since time was too short for a reasonable discussion of cohomology theory after homology
had been treated, I decided to skip cohomology altogether and instead included more material
from homotopy theory than is often done. In particular, there is a detailed discussion of higher
homotopy groups and the long exact sequence for Serre fibrations.
The necessary prerequisites of the students were rather modest. The course contains a quick
introduction to set theoretic topology but a certain acquaintance with these concepts was
certainly helpful. Familiarity with basic algebraic notions like rings, modules, linear maps etc.
was assumed.
These notes are based on the lecture notes of a course I taught in 2010. I am grateful to Volker
Branding who wrote a first draft of those lecture notes and produced most of the pstricks-figures
and to Ramona Ziese who improved those notes considerably.
Christian Bär
1
1 Set Theoretic Topology
Here are four versions of a typical question that one asks in mathematics. Fix the integers
1 ≤ n < m.
1.) Does there exist a linear isomorphism ϕ : Rm → Rn ?
No, since linear algebra tells us that isomorphic vector spaces have equal dimension.
2.) Does there exist a diffeomorphism ϕ : Rm → Rn ?
No, otherwise the differential dϕ(0) : Rm → Rn would be a linear isomorphism.
3.) Does there exist a homeomorphism ϕ : Rm → Rn ?
We cannot answer that question yet, but we will develop the necessary tools to find the
answer.
4.) Does there exist a bijective map ϕ : Rm → Rn ?
Yes. We will now explicitly construct an example for such a map in the case n = 1 and
m = 2.
Example 1.1. Since the exponential function maps R bijectively onto R+ = (0, ∞) it suffices to
construct a bijective map ϕ : R+ → R+ × R+ .
3
1 Set Theoretic Topology
x = . . . a3 a2 a1, b1 b2 b3 . . .
Here a j ∈ {0, 1, . . . , 9} and almost all a j are equal to 0. The b j are blocks of the form 0 . . . 0c j
with c j ∈ {1, . . . , 9}. This representation of x is unique. Now define ϕ(x) = (ϕ1 (x), ϕ2 (x))
where
ϕ1 (x) = . . . a5 a3 a1, b1 b3 b5 . . . ϕ2 (x) = . . . a6 a4 a2, b2 b4 b6 . . .
One easily checks that ϕ maps R+ bijectively onto R+ × R+ .
Let us evaluate the construction for an example. Let x = 1987, 30500735 . . .. Then we can read
off the a j and the b j as
a1 = 7, a2 = 8, a3 = 9, a4 = 1, a j = 0 for j ≥ 5
b1 = 3, b2 = 05, b3 = 007, b4 = 3, b5 = 5, . . .
Remark 1.2. In the continuous world counter-intuitive things can happen which are not possible
in the differentiable world. For example, there exist continuous maps
Example 1.3. The first example goes back to Peano [6]. The following example was shortly
after given by Hilbert [3] and is known as the Hilbert curve. It is defined by
4
1.1 Typical problems in topology
ϕ1 ϕ2 ϕ3
ϕ4 ϕ5 ϕ6
Remark 1.4. This cannot happen for smooth curves due to a Theorem by Sard which states that
for smooth ϕ : [0, 1] → R2 the image ϕ([0, 1]) ⊂ R2 is a zero set.
1.) Given two spaces, are they homeomorphic? To show that they are, construct a homeomor-
phism. To show that they are not, find topological invariants, which are different for given
spaces.
Example 1.5 (Classification theorem for surfaces). The classification theorem for surfaces
states that each orientable compact connected surface is homeomorphic to exactly one in the
following infinite list:
5
1 Set Theoretic Topology
S2 (sphere), genus 0
T 2 (torus), genus 1
b
b
b
The genus is the number of “holes” in the surface. We will give a more precise definition later. So
the classification theorem says that to each g ∈ N0 there exists an orientable compact connected
surface Fg with genus g and each orientable compact connected surface is homeomorphic to
exactly one Fg .
Example 1.6 (Brouwer fixed point theorem). Any continuous map f : D n → D n with
D n = {x ∈ Rn |k xk ≤ 1} has a fixed point, i.e., there exists an x ∈ D n such that
f (x) = x.
By the intermediate value theorem we can find an x with g(x) = 0 which is equivalent to
f (x) = x.
6
1.2 Some basic definitions
(i) ∅, Rn ∈ TRn
S
(ii) U j ∈ TRn , j ∈ J ⇒ j ∈J U j ∈ TRn
The importance of the concept of open subsets comes from the fact that one can characterize
continuous maps entirely in terms of open subsets. Namely, a map f : Rn → Rm is continuous
iff for each U ∈ TRm the preimage f −1 (U ) is open, f −1 (U ) ∈ TRn . This motivates taking open
subsets axiomatically as the starting point in topology.
Definition 1.9. A topological space is a pair (X, TX ) with TX ⊂ P(X ) such that
(i) ∅, X ∈ TX
S
(ii) U j ∈ TX , j ∈ J ⇒ j ∈J U j ∈ TX
(iii) U1, U2 ∈ TX ⇒ U1 ∩ U2 ∈ TX
7
1 Set Theoretic Topology
Definition 1.10. Let (X, TX ) be a topological space. Elements of TX are called open subsets
of X . Subsets of the form X \ U with U ∈ TX are called closed.
Examples 1.11. 1.) X arbitrary set, TX = P(X ) (discrete topology). All subsets of X are open
and they are also all closed.
2.) X arbitrary set, TX = {∅, X } (coarse topology). Only X and ∅ are open subsets of X . They
are also the only closed subsets.
3.) Let (X, TX ) be a topological space, let Y ⊂ X be an arbitrary subset. The induced topology
or subspace topology of Y is defined by TY := {U ∩ Y | U ∈ TX }.
4.) Let (X, d) be a metric space. Then we can imitate the construction of the standard topology
on Rn and define the induced metric as the set of all U ⊂ X such that for each x ∈ U there
exists an r > 0 so that B(x, r) ⊂ U. Here B(x, r) = {y ∈ X | d(x, y) < r} is the metric ball
of radius r centered at x.
Remark 1.12. Let two metrics d1 and d2 be given on a set X . If d1 and d2 are equivalent, i.e.,
∃ C ≥ 0 with
C −1 d2 (x, y) ≤ d1 (x, y) ≤ Cd2 (x, y) ∀x, y ∈ X,
then d1 and d2 induce the same topology.
Definition 1.14. Let (X, TX ) and (Y, TY ) be topological spaces. Then a map f : X → Y is
called continuous iff ∀ U ∈ TY : f −1 (U ) ∈ TX .
Example 1.15. If X carries the discrete topology then every map f : X → Y is continuous.
Example 1.16. If Y carries the coarse topology then every map f : X → Y is continuous.
Remark 1.17. In general, a continuous bijective map need not have a continuous inverse. For
example, let #X > 1 and consider f = id X : (X, Tdiscrete ) → (X, Tcoarse ).
Remark 1.18. f : X → Y is continuous iff f −1 (A) ⊂ X is closed for all closed subsets A ⊂ Y .
8
1.3 Compactness
1.3 Compactness
Definition 1.21. Let X be a topological space. A subset Y ⊂ X is called compact iff for
any collection {Ui }i ∈I , Ui ⊂ X open, with Y ⊂ ∪i ∈I Ui there exist i 1, . . . , i n ∈ I such that
Y ⊂ Ui 1 ∪ · · · ∪ Ui n .
Examples 1.22. 1.) Finite sets are always compact. Namely, et Y = {y1, . . . , yn } and let {Ui }i ∈I
be an open cover of Y . Then choose i j such that yi ∈ Ui j . Then {Ui 1 , . . . , Ui n } still covers Y .
2.) If X carries the discrete topology then a subset Y ⊂ X is compact if and only if it is finite.
We have already seen that finite sets are always compact. Conversely, let Y be a compact
subset of X . We cover Y by {{y} | y ∈ Y }. These one-point sets are open because X carries
the discrete topology. Since Y is compact this cover must be finite and therefore #Y < ∞.
Example 1.23. In particular, R is not compact. We can see this directly by looking at the open
cover {(x − 1, x + 1)|x ∈ R} of R. Since all intervals in this cover have length 2, any finite
subcover can cover only a bounded subset of R.
Proposition 1.24. Let X be a compact topological space. Let Y ⊂ X be a closed subset. Then
Y is compact.
9
1 Set Theoretic Topology
K ⊂ f −1 ( f (K )) ⊂ f −1 (∪i ∈I Ui ) = ∪i ∈I f −1 (Ui ) .
| {z }
open
f (K ) ⊂ f ( f −1 (Ui 1 ∪ · · · ∪ Ui n )) ⊂ Ui 1 ∪ · · · ∪ Ui n .
U1 U2
Examples 1.27. 1.) Spaces with discrete topology are Hausdorff spaces, simply put Ui = {x i }.
2.) Let #X ≥ 2. Then X with the coarse topology is not a Hausdorff space.
3.) If the topology of X is induced by a metric d, then X is Hausdorff. Namely, for x 1 , x 2 put
r := d(x 1, x 2 ) > 0. Then the open balls Ui = B(x i , r/2) separate x 1 and x 2 .
4.) Let X be Hausdorff and let Y ⊂ X any subset. Then Y with its induced topology is again
Hausdorff.
10
1.5 Quotient spaces
Proof. Let p ∈ X \ Y . Then for every y ∈ Y the Hausdorff property tells us that there exist open
subsets Uy, p, Vy, p ⊂ X such that y ∈ Uy, p , p ∈ Vy, p and Uy, p ∩ Vy, p = ∅. Since Y is compact
there exist y1, . . . , yn ∈ Y such that Y ⊂ Uy1, p ∪ · · · ∪ Uy n, p . Now put Vp := ∩nj=1 Vy j , p . Since
Vp is a finite intersection of open subsets it is open itself. Moreover, p ∈ Vp . Now
hence Vp ⊂ X \ Y . Therefore
X \ Y = ∪ p ∈X \Y Vp ⊂ X is open.
Thus Y ⊂ X is closed.
Proof. We have to show that ∀A ⊂ X closed f (A) ⊂ Y is closed. Now let A ⊂ X be closed.
Since X is compact, A is compact and also the image f (A) is compact. Since Y is a Hausdorff
space we conclude that f (A) ⊂ Y is closed.
Let π : X → X/∼ , x 7→ [x]. Define U ⊂ X/∼ to be open iff π −1 (U ) ⊂ X is open. One can
easily check that this defines a topology on X/∼ . Observe that π : X → X/∼ is continuous by
definition.
We now state the universal property of the quotient topology: For any topological space Y and
for any maps f : X → Y and f¯ : X/∼ → Y such that the diagram
f
X /Y
⑤⑤=
⑤
π ⑤⑤⑤
⑤⑤ f¯
X/∼
11
1 Set Theoretic Topology
Example 1.30
Let X = [0, 1] and let the equivalence relation be given by x ∼ x ∀x ∈ X and 0 ∼ 1.
x 1 ∼ x 2 ⇔ x 1 = x 2 or x 1, x 2 ∈ Y .
This equivalence relation identifies all points in Y to one point and performs no further identifi-
cations. Example 1.30 is of this form.
Example 1.32. R/(0,∞) is not a Hausdorff space because the points [0] and [1] cannot be sepa-
rated.
X := X1 × · · · × X n
12
1.7 Exercises
Definition 1.33. A subset U ⊂ X is called open (for the product topology) iff for all
p = (p1, . . . , pn ) ∈ U there exist open subsets Ui ⊂ X i with pi ∈ Ui and U1 × · · · × Un ⊂ U.
Examples 1.34. 1.) If all X i carry the discrete topology then X carries the discrete topology.
Namely, the sets {(p1, . . . , pn )} are open, hence all subsets of the product space are open.
2.) If all X i carry the coarse topology then X carries the coarse topology.
3.) Universal property: for all topological spaces Y and for all maps
f = ( f 1, . . . , f n ) : Y → X = X1 × · · · × X n
1.7 Exercises
1.1. Determine all topologies on the set {1, 2, 3} and investigate which ones are homeomorphic.
1.3. Let T be a topology on R which contains all half-open intervals (x, y] and [x, y), x < y.
Show that T is the discrete topology.
13
1 Set Theoretic Topology
a) are not homeomorphic when equipped with the standard topologies induced by R;
1.5. Let D n be as in Exercise 1.2. Let x, y ∈ D̊ n , i.e., k xk, k yk < 1. Construct a homeomor-
phism ϕ : D n → D n with ϕ(x) = y and ϕ(z) = z for all z ∈ ∂D n , i.e., kzk = 1.
b) Show that the universal property on page 11 determines the topology on X/ ∼ uniquely, i.e.,
if T is a topology on X/ ∼ such that π : (X, TX ) → (X/ ∼, T ) is continuous and if the
universal property holds for (X/ ∼, T ) then T is the topology defined above.
b) W n /∂W n ≈ S n .
1.8. Let X = R/Q, i.e., the quotient space of R under the equivalence relation x ∼ y iff x − y ∈ Q.
Show that X has uncountably many elements but carries the coarse topology.
1.9. Let X = R2 /Z2 , i.e., the quotient space of R2 under the equivalence relation x ∼ y iff
x − y ∈ Z2 . Let Y = S1 × S1 with the product topology. Show that X and Y are homeomorphic.
1.10. Let X be a topological space. One obtains the cone C X over X by considering the cylinder
X × [0, 1] and then passing to the quotient C X = (X × [0, 1])/ ∼ where the equivalence relation
∼ is given by (x, t) ∼ (x ′, t ′ ) if and only if (x, t) = (x ′, t ′ ) or t = t ′ = 1. Show:
a) CS n ≈ D n+1 .
b) If X is compact so is C X .
c) If X is Hausdorff so is C X .
1.11. Let X be a topological space. One obtains the suspension ΣX of X as the quotient
ΣX = (X × [0, 1])/ ∼ where the equivalence relation ∼ is given by (x, t) ∼ (x ′, t ′ ) if and only if
(x, t) = (x ′, t ′ ) or t = t ′ = 1 or t = t ′ = 0.
If furthermore f : X → Y is a map then f × id : X × [0, 1] → Y × [0, 1] induces a map
Σ f : ΣX → ΣY . Show:
a) If f is continuous so is Σ f .
14
1.7 Exercises
b) ΣD n ≈ D n+1 .
c) ΣS n ≈ S n+1 .
15
2 Homotopy Theory
Definition 2.1. Let X and Y be topological spaces. Let A ⊂ X be a subset. Two continuous
maps f 0, f 1 : X → Y are called homotopic relative to A iff there exists a continuous map
F : X × [0, 1] → Y such that
Remark 2.2. f 0 ≃ A f 1 ⇒ f 0 A = f 1 A
Remark 2.3. If A = ∅, then we say “ f 0 and f 1 are homotopic” instead of “ f 0 and f 1 are
homotopic relative to ∅”. Similarly, we write “ f 0 ≃ f 1 ” instead of “ f 0 ≃∅ f 1 ”.
2.) Let f 0 : Rn → Y be continuous. Put F (x, t) := f 0 ((1 − t)x), then f 0 ≃ const map.
3.) Lef f = Exp : R → S1 ∈ C, Exp(x) = e2πi x . From the previous example we know
f ≃ const map, but we will shortly see that f ;Z const map.
Given two topological spaces X, Y we set
C (X, Y ) := { f : X → Y | f is continuous}
Lemma 2.5. Let X, Y be topological spaces, let A ⊂ X and let ϕ ∈ C (A, Y ). Then “≃ A ” is
an equivalence relation on { f ∈ C (X, Y ) | f | A = ϕ}.
17
2 Homotopy Theory
Proof. a) “≃ A ” is reflexive:
f ≃ A f , because we can put F (x, t) := f (x).
b) “≃ A ” is symmetric:
Let f ≃ A g. We have to show g ≃ A f . Let F : X × [0, 1] → X be a homotopy relative to
A from f to g. Put G(x, t) := F (x, 1 − t). This is a homotopy relative to A from g to f ,
therefore g ≃ A f .
c) “≃ A ” is transitive:
Let f ≃ A g and g ≃ A h. We have to show f ≃ A h. Let F : X × [0, 1] → Y be homotopy
relative to A from f to g and let G : Y × [0, 1] → X be homotopy relative to A from g to h.
Then put H : X × [0, 1] → Y ,
F (x, 2t),
H (x, t) :=
0 ≤ t ≤ 1/2
G(x, 2t − 1),
1/2 ≤ t ≤ 1
Definition 2.7. A map f ∈ C (X, Y ) is called a homotopy equivalence iff there exists a g ∈
C (Y, X ) such that g◦ f ≃ id X and f ◦g ≃ idY If there exists a homotopy equivalence f : X → Y
then X and Y are homotopy equivalent. In symbols, X ≃ Y .
Remark 2.8. This defines an equivalence relation on the class of topological spaces.
X ≈ Y ⇒ X ≃ Y.
Example 2.10. Euclidean space is homotopy equivalent to a point, Rn ≃ {0}. Namely, put
f : {0} → Rn, f (0) = 0, and g : Rn → {0}, g(x) = 0. Then g ◦ f = id {0} and f ◦ g ≃ idRn by
Example 2.4.1. Remark 2.8 implies Rn ≃ Rm for all n, m ∈ N.
18
2.1 Homotopic maps
Definition 2.12. Let A ⊂ X and let ι : A → X be the inclusion map. Then A is called
1.) a retract of X iff there exists r ∈ C (X, A) such that r A = id A , i.e. r ◦ ι = id A . Then r is
called a retraction from X to A.
3.) a strong deformation retract of X iff there exists a retraction r : X → A such that
ι ◦ r ≃ A id X .
Examples 2.13. 1.) Let X any topological space, let A = {x 0 } ⊂ X consist of just one point.
Then r : X → A, r (x) = x 0 , is a retraction, hence A is a retract of X . The one-pointed set A
is a deformation retract of X iff X is contractible.
2.) Let X = Rn+1 \ {0} and A = S n . Consider the map r : X → A with r (x) = k xx k .
The composition ι ◦ r : Rn+1 \ {0} → Rn+1 \ {0} then satisfies ι ◦ r = k xx k . The map
F ∈ C (Rn+1 \ {0} × [0, 1], Rn+1 \ {0}) given by
x
F (x, t) = t x + (1 − t)
k xk
is continuous and satisfies
F (x, 0) = (ι ◦ r)(x), F (x, 1) = id(x)
for all x ∈ Rn+1 \ {0} and
F (a, t) = a, a ∈ Sn
We thus conclude that ι ◦ r ≃S n idRn+1\{0} , hence S n is a strong deformation retract of
Rn+1 \ {0}.
The difference between a deformation retract and a strong deformation retract is rather subtle.
Example 2.14
We consider the comb space given by
( 1 )
X = (x, y) ∈ R2 | 0 ≤ y ≤ 1 and (x = 0 or x = for some n ∈ N)
( ) n
2
∪ (x, y) ∈ R | y = 0 and 0 ≤ x ≤ 1
19
2 Homotopy Theory
A
1
......
Z
1 1 1
0 4 3 2 1
First we show that A is a deformation retract of X . The map F : X × [0, 1] → X given by
F (x, y, t) := (x, (1 − t)y) is continuous and satisfies
Since X × [0, 1] is compact the map G would be uniformly continuous. Therefore for ε = 1/2
we can find δ > 0 such that
G(x, y, t) − G(x ′, y ′, t ′ )
< 1
2
whenever |x − x ′ | < δ, |y − y ′ | < δ and |t − t ′ | < δ.
Then
1
G , 1, t − G(0, 1, t)
< ,
!
1
∀t ∈ [0, 1] .
n
2
Hence G( n1 , 1, t) ∈ B((0, 1), 12 ) for all t ∈ [0, 1]. Z
1
0 n 1
20
2.2 The fundamental group
On the other hand the mapping t 7→ G( n1 , 1, t) is a continuous path in X from ( n1 , 1) to (0, 1) and
must for some t take values in Z.
⇐
A is a deformation retract of X
⇐
⇐
:
A is a retract of X ; A≃X
That both possible implications in the bottom row do not hold in general can be seen by
counterexamples. Let A = {x 0 } be a point in X and let X be not contractible. Then A is a retract
of X but A and X are not homotopically equivalent. This is a counterexample for the implication
“ ⇒ ”.
A counterexample for the other direction “ ⇐ ” is given by X = [0, 1] × [0, 1] ⊂ R2 and
A = comb space. Then X and A are contractible, hence X ≃ A, but one can show that there is
no retraction X → A.
ω(2t),
(ω ⋆ η)(t) =
0 ≤ t ≤ 1/2,
η(2t − 1),
1/2 ≤ t ≤ 1.
Moreover, we consider ω−1 ∈ Ω(X ; x 1, x 0 ) with ω−1 (t) = ω(1 − t) and ε x0 ∈ Ω(X ; x 0 ) where
ε x0 (t) = x 0 .
21
2 Homotopy Theory
x1
X
η x0 x2
ω
0 1 0 1
t → 2t t → 2t − 1
0 1
Definition 2.17. For ω ∈ Ω(X ; x 0 ) denote by [ω] the homotopy class of ω relative to {0, 1}.
Then
π1 (X ; x 0 ) := {[ω] | ω ∈ Ω(X ; x 0 )}
is called the fundamental group of (X, x 0 ).
(iv) (ω ⋆ η) ⋆ ζ ≃ {0,1} ω ⋆ (η ⋆ ζ ).
Proof. The proof will be given graphically. In the following diagrams we draw the domain of
the required homotopies. The horizontal axis denotes the loop parameter whereas the vertical
axis represents the deformation parameter. The interpolations are piecewise linear. The red area
22
2.2 The fundamental group
gets mapped to x 0 .
ω′ η′
ω η
(ii) The first diagram shows ε x0 ≃ {0,1} ω, the second proves that ω ≃ {0,1} ε x0 .
ω ω
ε x0 ω ω ε x0
ω(1 − s)
s
ω ω−1
23
2 Homotopy Theory
For any s ∈ [0, 1] the corresponding ”blue“ line segment gets mapped to ω(1 − s). To prove the
statement ε x0 ≃ {0,1} ω−1 ⋆ ω interchange the roles of ω and ω−1 .
ω η ζ
Hence π1 (X ; x 0 ) together with “ · ” is a group with neutral element 1 := [ε x0 ] and inverse element
[ω]−1 = [ω−1 ].
To any topological space with a preferred point we have associated a group, the fundamental
group of the space with that point. Now we consider continuous maps. Let f ∈ C (X, Y ) and
x 0 ∈ X . We put f (x 0 ) =: y0 ∈ Y . If ω ≃ {0,1} ω ′ and H is a homotopy between them relative to
{0, 1} then f ◦H is a homotopy between f ◦ω and f ◦ω ′ relative to {0, 1}. Hence f ◦ω ≃ {0,1} f ◦ω ′ .
Therefore we have a well-defined map f # : π1 (X ; x 0 ) → π1 (Y ; y0 ), f # ([ω]) = [ f ◦ ω].
24
2.2 The fundamental group
f # : π1 (X, x 0 ) → π1 (Y, y0 )
is a group isomorphism.
Now we want to deal with the question to what extent π1 (X ; x 0 ) depends on the choice of x 0 .
X
To study this question let x 0, x 1 ∈ X and assume ω
there exists a path γ ∈ Ω(X ; x 0, x 1 ). If such a
path does not exist π1 (X ; x 0 ) and π1 (X ; x 1 ) are x1
not related. γ
x0
25
2 Homotopy Theory
Proof. a) Assertions 2., 3., and 4. follow directly from Lemma 2.18 and the definitions.
c) We compute
f g f g f g
κ · Φ̂γ ([ω]) · κ −1 = γ ′ ⋆ γ −1 · γ ⋆ ω ⋆ γ −1 · γ ⋆ (γ ′ ) −1
f g
= γ ′ ⋆ γ −1 ⋆ γ ⋆ ω ⋆ γ −1 ⋆ γ ⋆ (γ ′ ) −1
f g
= γ ′ ⋆ ω ⋆ (γ ′ ) −1
= Φ̂γ′ ([ω]).
26
2.2 The fundamental group
Proposition 2.22. Let X, Y be topological spaces and x 0 ∈ X . Let f , g ∈ C (X, Y ) and let
H : X × [0, 1] → Y be a homotopy from f to g. Define η ∈ Ω(Y ; f (x 0 ), g(x 0 )) by
η(s) := H (x 0, s).
π1 (X ; x 0P)
PPP
g#
PPPf#
PPP
PP(
π1 (Y ; g(x 0 )) / π1 (Y ; f (x 0 ))
Φ̂η
Proof. Let [ω] ∈ π1 (X ; x 0 ) and put F : [0, 1] × [0, 1] → Y with F (t, s) := H (ω(t), s).
Y
g◦ω
η
s F
f ◦ω
t
| |
1 2
3 3
Now we can improve Corollary 2.20 and show that homotopy equivalent spaces have isomorphic
fundamental groups.
27
2 Homotopy Theory
f # : π1 (X ; x 0 ) → π1 (Y ; f (x 0 ))
Lemma 2.26. Let t 0 ∈ R and z0 = Exp(t 0 ) ∈ S1 . Then for all f ∈ C (S1, S1 ) with f (1) = z0
there exists a unique fˆ ∈ C (R, R) with fˆ(0) = t 0 such that the following diagram commutes:
fˆ
R /R
Exp Exp
S1 / S1
f
28
2.3 The fundamental group of the circle
Assume that besides fˆ there is a second map f˜ ∈ C (R, R) with the same proper-
ties as fˆ. The equality of Exp(x) = Exp(x ′ ) is equivalent to x − x ′ ∈ Z. Since
Exp( fˆ(t)) = f (Exp(t)) = Exp( f˜(t)) we deduce that fˆ(t) − f˜(t) ∈ Z for all t ∈ R. Since both
f˜, fˆ are continuous it follows that fˆ − f˜ is constant. Finally, we know that fˆ(0) = t 0 = f˜(0),
hence f˜ = fˆ.
Moreover, the restriction of Exp to each of these intervals is a homeomorphism onto its image,
" #
Exp[t 1 +k, t 1 +k+ 1 ]
1 ≈
: t 1 + k, t 1 + k + −→ C.
2 2
Its inverse can written down explicitly in terms of logarithms but we will not need this.
Now we construct fˆ step by step.
fˆ := (Exp | I0 ) −1 ◦ f ◦ Exp,
1
where I0 ⊂ R is the compact interval of length 2 with Exp(I0 ) = C0 and t 0 ∈ I0 . This insures
that
Put t 1 := fˆ(ε).
Next, f (Exp([ε, 2ε])) is contained in a closed semi-circle C1 and we define fˆ on [ε, 2ε] by
fˆ := (Exp | I1 ) −1 ◦ f ◦ Exp
where I1 ⊂ R is the compact interval of length 21 with Exp(I1 ) = C1 and t 1 ∈ I1 . This insures
that the two definitions of fˆ at ε agree so that we obtain a continuous function fˆ : [0, 2ε] → R.
Repeating this procedure infinitely many times we can extend fˆ continuously to [0, ∞) → R.
The extension to the left is done similarly so that we obtain a continuous function fˆ : R → R.
Commutativity of the diagram holds by construction.
29
2 Homotopy Theory
Definition 2.27. For a map f ∈ C (S1, S1 ) a map fˆ ∈ C (R, R) for which the diagram in
Lemma 2.26 commutes is called a lift of f .
Remark 2.30
1.) We have seen that different lifts of f differ by an additive constant in Z. Hence the degree
deg( f ) is well defined, independently of the choice of lift fˆ.
3.) The map t 7→ fˆ(t + 1) − fˆ(t) is continuous and takes values in Z, by the same argument as
above. We conclude that fˆ(t + 1) − fˆ(t) = deg( f ) for all t ∈ R.
Hence fˆ + ĝ is a lift of f g and we get the following formula for the degree
deg( f g) = fˆ(1) + ĝ(1) − fˆ(0) + ĝ(0) = deg( f ) + deg(g) .
30
2.3 The fundamental group of the circle
Thus f is onto.
Example 2.31. For the map f n : S1 → S1 with f n (z) = z n a lift is given by fˆ(t) = nt so that its
degree is deg( f n ) = fˆn (1) − fˆn (0) = n.
(·, s) and not surjective because −1 is not contained in the image. Hence, by Remark 2.30,
is continuous
deg FF(·, s ′ ) = 0. We now compute using 2.30.5:
! !
F (·, s) ′ F (·, s)
deg(F (·, s)) = deg · F (·, s ) = deg + deg F (·, s ′ ) = deg F (·, s ′ ) .
F (·, s ′ ) F (·, s ′ )
We see inductively that
where 0 = s0 < s1 < · · · < sr = 1 is a partition of the unit interval [0, 1] satisfying
|si+1 − si | < δ.
31
2 Homotopy Theory
Proof. The map F ∈ C (S1 × [0, 1], S1 ), F (z, s) := g(sz), defines a homotopy from a constant
map to f . By Lemma 2.32 we conclude that deg( f ) = deg(const) = 0.
Since dividing by an does not change the roots, we assume without loss of generality that an = 1.
Now assume that p has no roots. Then the map f : C → S1 with f (z) = | p(z)p(z) | is a well-defined
continuous map. To compute deg( f S 1 ) consider
The map F ∈ C (S1 × [0, 1], S1 ) is a homotopy from f n (z) = z n to f S 1 . Computing its degree
we find deg( f S 1 ) = deg( f n ) = n ≥ 1.
On the other hand, f is a continuous map defined on all of C, hence Corollary 2.33 implies
deg( f S 1 ) = 0. This is a contradiction, thus p must have a root.
Lemma 2.35. Suppose the map f ∈ C (S1, S1 ) satisfies f (−z) = − f (z) for all z ∈ S1 . Then
the degree deg( f ) is odd.
Moreover,
− f Exp(t) = − Exp fˆ(t) = Exp 21 Exp fˆ(t) = Exp fˆ(t) + 21 .
32
2.3 The fundamental group of the circle
From f (− Exp(t)) = − f (Exp(t)) we conclude Exp fˆ t + 21 = Exp fˆ(t) + 21 and hence
fˆ t + 12 − fˆ(t) + 21 =: k (t)
is an integer for every t. Due to the continuity of the map, k (t) it is constant, k (t) = k. Hence
fˆ(t + 12 ) − fˆ(t) = k + 12 for all t ∈ R. Now we can compute
deg( f ) = fˆ(1) − fˆ(0) = fˆ(1) − fˆ 21 + fˆ 21 − fˆ(0)
= (k + 12 ) + (k + 21 ) = 2k + 1
Theorem 2.36 (Borsuk-Ulam). Let f ∈ C (S2, R2 ) satisfy f (−x) = − f (x) for all x ∈ S2 .
Then f has a zero.
Proof. Assume that the map f has no zero. Then the map g : S2 → S1 with g(x) := k ff (x) (x)
k
is defined and continuous. Moreover, g satisfies g(−x) = −g(x) for all x ∈ S 2 . Now define a
p
map G : D 2 → S1 by G(y) = g(y, 1 − k yk 2 ). The map G ∈ C (D 2, S1 ) has the property that
GS 1 = g|S 1 . By Corollary 2.33 we know that deg(gS 1 ) = 0. On the other hand we know by
Lemma 2.35 that deg(gS 1 ) is odd, which gives a contradiction.
Corollary 2.37. Let f ∈ C (S2, R2 ). Then there exists a point x 0 ∈ S2 with f (−x 0 ) = f (x 0 ).
Proof. Put g(x) := f (x) − f (−x). Then the map g ∈ C (S2, R2 ) satisfies g(−x) = −g(x)
for all x ∈ S2 . Hence by the Borsuk-Ulam Theorem 2.36 there exists an x 0 ∈ S2 with
0 = g(x 0 ) = f (x 0 ) − f (−x 0 ), which proves the theorem.
Remark 2.38. In particular, the map f ∈ C (S2, R2 ) cannot be injective. Thus the sphere S2
cannot be homeomorphic to a subset of R2 . This also shows that R3 cannot be homeomorphic to
R2 .
33
2 Homotopy Theory
H x, t := {y ∈ R3 | hy, xi = t} . tx • +
H x, t
S2
•x
It is clear that H−x,−t = H x, t . We define the
half space H x,+ by
t
0•
+ 3
H x, t := {y ∈ R | hy, xi ≥ t}.
H x, t ′
H x, t
b) Now fix x ∈ S2 . Look at the function
a x : R → R defined by
+
a x (t) := vol(A ∩ H x, t ). B(0, R A )
vol(A)
αx
Moreover, a x (t) = vol(A) for t ≪ 0
and a x (t) = 0 for t ≫ 0.
t −x t +x
34
2.3 The fundamental group of the circle
c) By continuity and monotonicity the pre-image of any value under α x is a closed interval. In
particular, a−1 − +
x (vol(A)/2) = [t x , t x ].
t − +t +
Now put α(x) := x 2 x . Hence H x,α(x) divides A into two pieces of equal volume. Moreover,
α(−x) = −α(x) for all x ∈ S2 and it is not hard to check that α is continuous. Similarly,
define the functions β for B and γ for C.
d) Consider the map f ∈ C (S2, R2 ) with f (x) = (α(x) − β(x), α(x) − γ(x)). We have
Thus the Borsuk-Ulam Theorem 2.36 applies and there exists a point x 0 ∈ S2 with
Hence α(x 0 ) = β(x 0 ) = γ(x 0 ). Thus the hyperplance H x0,α(x0 ) does the job.
Remark 2.40. The ham-sandwich theorem is optimal in the sense that it fails for more than three
sets in R3 .
Example 2.41. A ball B(x, r) ⊂ R3 with center x is cut into two halves of equal volume exactly
by those planes that contain x. If you choose four balls in R3 in such a way that their centers are
not contained in one plane, then no plane will cut them all into halves of equal volume.
In the following we want to use the concept of degree to determine π1 (S1 ; 1).
fω
Exp !ω
a) For ω ∈ Ω(S1 ; 1) and I = [0, 1] consider the fol- S1 a❈o I / S1
❈❈ ④=
lowing diagram: ❈❈≈ ④④④
❈❈ ④④
Exp ❈ ④④ ω
I/∂I
Here ω and Exp are the continuous maps induced on the quotient space. They exist because
ω(0) = ω(1) and Exp(0) = Exp(1), compare Section 1.5. By Example 1.30 we know that
Exp : I/∂I → S1 is a homeomorphism. Now put
f ω := ω ◦ (Exp) −1 ∈ C S1, S1
deg : Ω(S1 ; 1) → Z.
35
2 Homotopy Theory
c) This map is surjective because for n ∈ Z we can consider the map ω(t) = Exp(nt). The
commutative diagram
z7→z n = f n (z)
S1 f◆◆ / S1
O
◆◆◆
◆◆◆ t 7→ω(t )=Exp(nt )=Exp(t ) n
◆
t 7→Exp(t ) ◆◆◆
◆
I
shows f ω (z) = z n = f n (z) and we get deg(ω) = deg( f n ) = n.
d) Now let ω, ω ′ ∈ Ω(S1 ; 1) and consider the map
f ω⋆ω′ (Exp(t)) = ω ⋆ ω ′ (t)
ω(2t),
=
0 ≤ t ≤ 1/2
ω ′ (2t − 1), 1/2 ≤ t ≤ 1
f ω (Exp(2t)),
=
0 ≤ t ≤ 1/2
f ω′ (Exp(2t − 1)), 1/2 ≤ t ≤ 1
Let fˆω , fˆω′ be lifts of f ω , f ω′ with fˆω′ (0) = fˆω (1). Then we have
Exp( fˆω (2t)),
f ω⋆ω′ (Exp(t)) =
0 ≤ t ≤ 1/2
Exp( fˆω′ (2t − 1)), 1/2 ≤ t ≤ 1
!
= Exp fˆω (2t), 0 ≤ t ≤ 1/2
f ω′ (2t − 1),{z 1/2 ≤ t ≤ }
ˆ 1
|
=:g (t )
Note that the map g(t) is continuous because of fˆω′ (0) = fˆω (1). Thus g(t) is a lift of f ω⋆ω′ .
Now we compute the degree of ω ⋆ ω ′.
deg(ω ⋆ ω ′ ) = g(1) − g(0)
= fˆω′ (1) − fˆω (0)
= fˆω′ (1) − fˆω′ (0) + fˆω (1) − fˆω (0)
= deg(ω ′ ) + deg(ω)
Hence the map deg : π1 (S1 ; 1) → (Z, +) is a group homomorphism.
36
2.3 The fundamental group of the circle
e) Finally we compute its kernel. Let ω ∈ Ω(S1 ; 1) with deg(ω) = 0. Let fˆω be the lift of f ω
with fˆω (0) = 0. Since 0 = deg(ω) = fˆω (1) − fˆω (0) we have fˆω (1) = fˆω (0) = 0. Next
consider the continuous map F : I × I → S1 with F (t, s) := Exp(s fˆω (t)). It satisfies:
F (t, 0) = 1 = ε 1 (t) ,
F (t, 1) = f ω (Exp(t)) = ω(t) ,
F (0, s) = Exp(s · 0) = 1 ,
F (1, s) = Exp(s · fˆω (1)) = Exp(s 0) = 1 .
We conclude that ω ≃ {0,1} ε 1 , hence [ω] = [ε 1 ] = 1 ∈ π1 (S1 ; 1). Therefore the kernel is
trivial and the map deg : π1 (S1 ; 1) → Z is injective.
Example 2.43. We already know that S1 is a strong deformation retract of C \ {0}. Hence the
inclusion ι : S1 → C \ {0} induces an isomorphism
ι# : π1 (S1 ; 1) → π1 (C \ {0}; 1)
S1 = ∂D 2
Example 2.44
We show that S1 = ∂D 2 is not a retract of D 2 . Suppose the map D2
r : D 2 → S1 is a retraction and denote the inclusion by ι : S1 → D 2 .
Then we would have r ◦ ι = idS 1 , hence r# ◦ ι# = (r ◦ ι)# = (idS 1 )# = idπ 1 (S 1 ;1) . We then get a
contradiction because of the following diagram
ι# r#
Z π(S1 ; 1) / π(D 2 ; 1) {1} / π(S 1 ; 1) Z
:
idπ 1
1 (S ;1)
As a corollary we get a proof of Brouwer’s fixed point theorem in dimension two. See page 92
for the theorem in general dimensions.
37
2 Homotopy Theory
Proof. Assume that f ∈ C (D 2, D 2 ) has no fixed point. Then f (x) , x for all x ∈ D 2 so that we
can consider the half line emanating from f (x) through x. We let r (x) be its intersection point
with ∂D 2 = S1 as indicated in the picture.
x
f (x) • r (x)
Fixed points of f (x, y) are then the same as solutions to the above system of equations (2.1). To
apply Brouwer’s fixed point theorem we have to show that f (D 2 ) ⊂ D 2 . Let x, y ∈ D 2 . Then
!2 !2
2 1 sin(x)y cos(x)y x 2
k f (x, y)k = + + +
2 4 4 4
1 sin(x)y sin(x) y 2 2 cos(x) 2 y2 cos(x)yx 2 x 4
= + + + + +
4 4 16 16 8 16
1 sin(x)y y2 cos(x)yx 2 x 4
= + + + +
4 4 16 8 16
1 |y| y2 |y|x 2 x 4
≤ + + + +
4 4 16 8 16
1 1 1 1 1
≤ + + + +
4 4 16 8 16
38
2.3 The fundamental group of the circle
3
=
4
≤ 1.
Example 2.47. Consider f n ∈ C (S1, S1 ) with f n (z) = z n . We check that the diagram
deg
π1 (S1 ; 1)
/Z
( f n )# n·
deg
π1 (S1 ; 1)
/Z
Remark 2.48. From Exercise 2.5 we know that for X1, X2 and x 1 ∈ X1, x 2 ∈ X2 we have
1.) We call X connected iff X and ∅ are the only subsets of X which are both open and closed.
2.) We call X path-connected iff for all x 1, x 2 ∈ X there exists a path ω ∈ Ω(X ; x 1, x 2 ).
Example 2.50. The interval [0, 1] is connected. To see this let I ⊂ [0, 1] be open and closed
and assume that I is neither empty nor all of [0, 1]. Then there exists t 0 ∈ I and t 1 ∈ [0, 1] \ I.
W.l.o.g. let t 0 < t 1 , the other case being analogous. We put T := sup(I ∩ [0, t 1 )). Then
0 ≤ t 0 ≤ T ≤ t 1 ≤ 1. Since I is closed T ∈ I.
If T = 1 then T = t 1 which contradicts t 1 < I. If T < 1 then there exists ε > 0 such that
[T, T + ε) ⊂ I because I is open. This contradicts the maximality of T.
Remark 2.51. If X is 1-connected then it is path-connected by definition but the converse is not
true. For example, S1 is path-connected but not 1-connected.
39
2 Homotopy Theory
Again, the converse implication does not hold in general. Consider for example the space
Remark 2.53. Let X be path-connected. Then the following are equivalent (see Exercise 2.2):
(i) X is simply connected;
g · H = H · g for all g ∈ G .
40
2.4 The Seifert-van Kampen theorem
The condition on a subgroup of being normal can be reformulated in various ways. It is equivalent
to any of the following:
Example 2.55. Consider the cartesian product of two groups G = G1 ×G2 = {(g1, g2 ) | g j ∈ G j }
with componentwise multiplication. Then {(g1, 1) | g1 ∈ G1 } G1 is a normal subgroup of G
because
(g1, g2 ) · (g̃1, 1) · (g1, g2 ) −1 = g1 g̃1 g1−1, g2 1g2−1 = g1 g̃1 g1−1, 1 .
(g · H ) · (g̃ · H ) = (g g̃) · H.
Normality of H ensures that this multiplication is well defined. The group H is then the kernel of
G → G/H with g 7→ g · H. Thus the normal subgroups are exactly those which arise as kernels
of group homomorphisms.
is the smallest normal subgroup containing S. We call N (S) the normal subgroup generated by
S.
41
2 Homotopy Theory
Definition 2.59
Let G1 and G2 be groups. A group G is called free product G1 ❆
of G1 and G2 iff there exist homomorphisms i j : G j → G such i1 ⑥⑥⑥ ❆❆ ϕ1
❆❆
⑥
that for all groups H and for all homomorphisms ϕ j : G j → H ~ ⑥⑥ ⑥
ϕ 1 ⋆ϕ 2
❆❆
there exists a unique homomorphism G `❆ /H
❆❆ ⑥>
❆❆ ⑥⑥⑥
❆ ⑥⑥ϕ
ϕ1 ⋆ ϕ2 : G → H i 2 ❆❆ ⑥⑥ 2
G2
such that the diagram commutes.
G1
Namely, let G ′ be another free product of G1 and G2 with i1 ⑥ ❇❇❇ i ′
⑥⑥ ❇❇1
i ′j : G j → G ′ the corresponding homomorphisms. By the ⑥ ⑥ ❇❇
~⑥⑥ ′ ′
i 1 ⋆i 2
universal property of G with H = G ′ and ϕ j = i ′j we get the G `❆ / G′
❆❆ ⑤>
following commutative diagram: ❆❆ ⑤⑤
⑤
❆ ⑤⑤i ′
i 2 ❆❆ ⑤
⑤ 2
G2
G1
i ′1 ⑤ ❆❆❆ i
⑤⑤⑤ ❆❆1
⑤ ⑤ ❆❆
~
⑤ ⋆i
Interchanging the roles of G and G ′ we get another commutative G ′ `❇
i 1 2 /G
diagram: ❇❇ ⑥ ⑥>
❇❇ ⑥⑥
❇ ⑥⑥ i 2
i ′2 ❇❇ ⑥⑥
G2
42
2.4 The Seifert-van Kampen theorem
G1 ❆
i1 ⑥⑥⑥ ′ ❆❆❆i 1
i1
⑥⑥ ❆❆
Combining both diagrams we obtain ~⑥⑥i ′1 ⋆i ′2 i 1 ⋆i 2 ❆
G `❆ / G′ /G
❆❆ O >
❆❆ ′ ⑥⑥⑥
❆i ⑥⑥
i 2 ❆❆2 ⑥⑥⑥ i 2
G2
G1 ❆
i1 ⑥ ❆❆ i
On the other hand, this diagram commute as well. ⑥⑥⑥ ❆❆1
⑥ ❆❆
~⑥
⑥ id
G `❆ /G
❆❆ ⑥ ⑥>
❆❆ ⑥⑥
❆ ⑥⑥ i 2
i 2 ❆❆ ⑥⑥
G1
By the uniqueness of the induced homomorphisms we have
(i1 ⋆ i 2 ) ◦ i1′ ⋆ i 2′ = idG
and similarly
i1′ ⋆ i 2′ ◦ (i1 ⋆ i 2 ) = idG′ .
Hence the map i 1 ⋆ i 2 : G → G ′ is a group isomorphism with inverse i 1′ ⋆ i2′ .
Next we show the existence of the free product of two groups by a direct construction. For this
purpose let G1 and G2 be groups. For formal reasons we assume without loss of generality that
G1 ∩ G2 = ∅.1 We define
G1 ⋆ G2 := (x 1, . . . x n ) | n ∈ N0, x j ∈ G1 \ {1G1 )} ∪ (G2 \ {1G2 } such that
if x i ∈ G1 then x i+1 ∈ G2 or conversely .
(x 1, . . . , x n ) · (x n+1, . . . , x n+m )
(x 1, . . . , x n−1, x n · x n+1, x n+2, . . . , x n+m ) if (x n , x n+1 ∈ G1 or x n, x n+1 ∈ G2 )
and x n · x n+1 , 1,
:=
(x 1, . . . , x n−1 ) · (x n+2, . . . , x n+m ) if (x n , x n+1 ∈ G1 or x n, x n+1 ∈ G2 )
and x n · x n+1 = 1,
(x , . . . , x , x , . . . , x
1 n n+1 n+m ) otherwise.
This turns G1 ⋆G2 into a group with neutral element the empty sequence (). The inverse element
for (x 1, . . . x n ) is given by (x −1 −1
n , . . . , x 1 ) because of
−1 −1
(x 1, . . . , x n ) · x −1
n , . . . , x1 = (x 1, . . . , x n−1 ) · x −1
n−1, . . . , x 1 = · · · = (x 1 )(x −1
1 ) = ().
1 Otherwise replace G2 by an isomorphic group which is disjoint to G1 .
43
2 Homotopy Theory
Remark 2.62
1.) The subset i j (G j ) ⊂ G1 ⋆ G2 is a subgroup isomorphic to G j . The intersection i1 (G1 ) ∩
i2 (G2 ) = {1} is trivial. The union i 1 (G1 ) ∪ i2 (G2 ) generates i 1 (G1 ) ⋆ i 2 (G2 ) as a group.
2.) If G1 = {1} then G1 ⋆ G2 = i 2 (G2 ) G2 . Similarly, if G2 = {1} then G1 ⋆ G2 G1 .
3.) If G1 , {1} and G2 , {1} then we may choose x ∈ G1 \ {1} and y ∈ G2 \ {1}. Then
(x), (x, y), (x, y, x), (x, y, x, y), . . .
yields infinitely many pairwise different elements in G1 ⋆ G2 , hence |G1 ⋆ G2 | = ∞ (even if
|G j | < ∞). In addition, we have
(x) · (y) = (x, y) , (y, x) = (y) · (x) ,
hence the group G1 ⋆ G2 is not abelian, even if G1 and G2 are.
Remark 2.63. Usually one identifies G1 with i1 (G1 ) and G2 with i2 (G2 ) and writes
x 1 · x 2 · . . . · x n instead of (x 1, x 2, . . . , x n ).
Example 2.64. Consider G1 = Z/2Z = {1, −1} and G2 = Z/2Z = {1′, −1′ }. Elements of
Z/2Z ⋆ Z/2Z are for example
a = (−1) · (−1′ ) · (−1) · (−1′ )
b = (−1′ ) · (−1) · (−1′ ) · (−1) · (−1′ ) .
Now we calculate
a · b = (−1) · (−1′ ) · (−1) ·(−1′ ) · (−1′ ) ·(−1) · (−1′ ) · (−1) · (−1′ )
| {z }
1′
′
= (−1) · (−1 ) ·(−1) · (−1) ·(−1′ ) · (−1) · (−1′ )
| {z }
1
= (−1) ·(−1′ ) · (−1′ ) ·(−1) · (−1′ )
| {z }
1′
= (−1) · (−1) ·(−1′ )
| {z }
1
= (−1′ ) .
44
2.4 The Seifert-van Kampen theorem
Proposition 2.65. Let X be a topological space and let U, V ⊂ X be open such that U ∪V = X .
Let x 0 ∈ U ∩ V . Furthermore, assume that U, V and U ∩ V are path-connected. Then X is
path-connected and the map
i # ⋆ j# : π1 (U; x 0 ) ⋆ π1 (V ; x 0 ) → π1 (X ; x 0 )
Proof. First of all we note that the space X is path-connected because each point in X lies in U
or in V and can therefore be connected to x 0 by a path.
The statement of the proposition is equivalent to saying that
i# π1 (U; x 0 ) ∪ j# π1 (V ; x 0 )
generates π1 (X ; x 0 ) as a group. Now let [ω] ∈ π1 (X ; x 0 ) and subdivide the unit interval I = [0, 1]
by 0 = t 0 ≤ t 1 · · · < t n = 1 such that ω([t i , t i+1 ]) ⊂ U or ⊂ V .
X
ω0 ω2
U
x0
U ∩V η1 η2 η3 U ∩V
ω(t 1 ) ω(t 2 ) ω(t 3 )
ω1
By removing subdivision points if necessary we can assume that if ω([t i−1, t i ]) ⊂ U then
ω([t i , t i+1 ]) ⊂ V or conversely. Reparametrize ωi (t) = ω((1 − t)t i + tt i+1 ). Then ωi ∈
Ω(U or V ; ω(t i ), ω(t i+1 )) and in particular ω(t i ) ∈ U ∩ V . Since by assumption U ∩ V is
path-connected there exist η j ∈ Ω(U ∩ V ; x 0, ω(t j )). Then
ω0 ⋆ η 1−1, η 1 ⋆ ω1 ⋆ η 2−1, η 2 ⋆ ω2 ⋆ η 3−1, . . . , η n−1 ⋆ ω n−1
45
2 Homotopy Theory
Corollary 2.66. Let X is a topological space and let U, V ⊂ X be open such that U ∪ V = X
and U ∩ V , ∅. If U and V are 1-connected and U ∩ V is path-connected, then the space X is
1-connected.
Proof. Since
{1} = {1} ⋆ {1} = π1 (U; x 0 ) ⋆ π1 (V ; x 0 ) → π1 (X ; x 0 )
is onto we get that π1 (X ; x 0 ) = {1}.
Example 2.67. Consider X = S n and put U = S n \ {e1 }. The stereographic projection yields
a homeomorphism U → Rn . Hence U is 1-connected. Similarly, V = S n \ {−e1 } is also
1-connected. Now
U ∩ V = S n \ {e1, −e1 } ≈ Rn \ {0}.
For n ≥ 2 the space U ∩ V is path-connected. Corollary 2.66 shows that S n is simply connected
for n ≥ 2.
Recall that we know already that this is not true for n = 1 because π1 (S1 ; 1) Z.
46
2.4 The Seifert-van Kampen theorem
For α ∈ π1 (U ∩ V ; x 0 ) we calculate
−1
1 = i # i#′ (α) · j# j#′ (α) = i # i #′ (α) · j# j#′ (α) −1 = (i # ⋆ j# ) i #′ (α) · j#′ (α) −1 .
| {z }
∈π 1 (U;x 0 )⋆π 1 (V ;x 0 )
Let ω1, ω3, · · · ∈ Ω(U; x 0 ) and ω2, ω4, · · · ∈ Ω(V ; x 0 ) be such that
1 = i # ⋆ j# [ω1 ]U · [ω2 ]V · [ω3 ]U · . . .
= i # ([ω1 ]U ) · j# ([ω2 ]V ) · i # ([ω3 ]U ) · . . .
= [ω1 ] X · [ω2 ] X · [ω3 ] X · . . .
= [ω1 ⋆ ω2 ⋆ ω3 ⋆ . . . ] X
s
1 −
Then there exists a homotopy H : [0, 1] ×
[0, 1] → X such that
47
2 Homotopy Theory
| | |
ω1 ω2 ωN
u
>
>
d
η v0 ⋆ (H ◦ c) ⋆ η −1
v 1 ∈ Ω(U or V ; x 0 ) .
48
2.4 The Seifert-van Kampen theorem
u1 u2 uN
> > >
ε x0 = l 1 ∧ ∧ ∧ ∧ ∧ r N = ε x0
> > >
d1 d2 dN
We find that
Di := η d i (0) ⋆ (H ◦ d i ) ⋆ η −1
d i (1) ∈ Ω(U or V ; x 0 ).
Similarly we define L i , Ri , Ui ∈ Ω(U or V ; x 0 ) and we conclude that by (2.2)
[Di ]Wi · [Ri ]Wi = [L i ]Wi · [Ui ]Wi ∈ π1 (Wi ; x 0 ) (2.3)
where Wi = Uor V . We now compute in π1 (U; x 0 ) ⋆ π1 (V ; x 0 ) modulo N ({i#′ (α) · j#′ (α) −1 | α ∈
π1 (U ∩ V ; x 0 )}):
[D1 ]W1 · [D2 ]W2 · . . . · [D N ]W N = [D1 ]W1 · [D2 ]W2 · . . . · [D N ]W N · [RN ]W N
= [D1 ]W1 · . . . · [D N −1 ]W N −1 · [L N ]W N · [UN ]W N
= [D1 ]W1 · . . . · [D N −1 ]W N −1 · [RN −1 ]W N · [UN ]W N .
If now W N −1 = W N then we can apply (2.3) once more and get
[D1 ]W1 · [D2 ]W2 · . . . · [D N ]W N = [D1 ]W1 · . . . · [D N −2 ]W N −2 · [L N −1 ]W N −1 · [UN −1 ]W N −1 · [UN ]W N
(2.4)
In case W N −1 , W N , say W N −1 = U and W N = V , then L N = RN −1 ∈ Ω(U ∩ V ; x 0 ). Hence
mod N (... ) ′
[RN −1 ]V = j#′ ([RN −1 ]U∩V ) = i# ([RN −1 ]U∩V ) = [RN −1 ]U .
In
π1 (U; x 0 ) ⋆ π1 (V ; x 0 )
G :=
N ({i #′ (α) · j#′ (α) −1 | α ∈ π1 (U ∩ V ; x 0 )})
the computation (2.4) is still possible. By induction on all squares in the row from right to the
left we find that
=1
z }| {
[D1 ]W1 · . . . · [D N ]W N = [L 1 ]W1 ·[D1 ]W1 · . . . · [D N ]W N = [U1 ]W1 · . . . · [UN ]W N .
49
2 Homotopy Theory
and hence
ker(i # ⋆ j# ) ⊂ N i #′ (α) · j#′ (α) −1 | α ∈ π1 (U ∩ V ; x 0 ) .
π1 (X ; x 0 ) π1 (U; x 0 ) ⋆ π1 (V ; x 0 )
Example 2.70. Consider the figure 8 space and the two subsets U and V as indicated in the
picture:
X
V U
b b
x0 x0
U ∩ V ≃ point
π1 (X ; x 0 ) π1 (S1 ; x 0 ) ⋆ π1 (S1 ; x 0 ) Z ⋆ Z .
50
2.4 The Seifert-van Kampen theorem
M
b
V
p
U ∩ V ≈ Rn \ {0} ≃ S n−1
π1 (X ; x 0 ) π1 (U; x 0 ) ⋆ π1 (V ; x 0 ) = π1 (X \ {p}; x 0 ) .
Removing a point from a manifold of dimension at least 3 does not change its fundamental group.
M N
Let X = M#N.
U ∩V
| {z }
=U | {z }
=V
51
2 Homotopy Theory
Now choose U and V as in the picture. Then we have that U ≈ M \ {p} and V ≈ N \ {q}. Since
U ∩ V ≈ S n−1 × (0, 1) ≃ S n−1
is 1-connected we find by Corollary 2.69 once again that
π1 (M#N ) π1 (U ) ⋆ π1 (V ) π1 (M ) ⋆ π1 (N ) .
For example, if M = N = T 3 then π1 (T 3 #T 3 ) = (Z3 ) ⋆ (Z3 ).
Remark 2.73. We now see easily that the torus T n with n ≥ 3 cannot be homotopy equivalent
to the connected sum T n ≃ M#N of two non-1-connected manifolds M and N.2 If it were
possible then π1 (T n ) π1 (M ) ⋆ π1 (N ) would not be abelian but we know that π1 (T n ) Zn , a
contradiction.
2
Definition 2.74. We call Fg := T · · · #T}2 a surface of genus g ≥ 1.
| # {z
g−times
Fg = ···
| {z }
g−times
We now want to compute π1 (Fg ) and show that the fundamental groups for surfaces of different
genus are not isomorphic. This then shows in particular that they are not homotopy equivalent.
52
2.5 The fundamental group of surfaces
b−1
g a1
a−1 b1
g
a1−1
b−1
1
··· ≈ D2
a2
b2
a2−1
Proof. We do an induction on g.
Induction basis for g = 1: We see that the labelled disk D 2 / ∼ is homeomorphic to the labelled
W 2 / ∼ which is homeomorphic to the two-dimensional torus, i.e. to F1 .
a1
b−1
1 a1 a1
≈ b−1
1 b1 ≈ b−1
1 b1 ≈
a1−1 b1
a1−1
53
2 Homotopy Theory
b−1
g a1 a1
b1 b1
a−1
g b−1
g
a1−1 a−1 a1−1
g
c
b−1
1 b−1
1
≈ c
a2
a2
b2
b2
a2−1
a2−1
The endpoints of c in the two pieces are identified. This yields a homeomorphism where the
interior of the now closed loop c is cut out of the two remaining disks.
a1
b1
b−1
g a1−1
a−1
g b−1 a2
g
a−1 b2
c b−1 g
1 a2−1 b−1
1 a1
c
c b−1 c
≈ 2
a2
b2 a1−1 b1
a2−1
c c ≈
| {z } | {z }
(g−1)−times g−times
54
2.5 The fundamental group of surfaces
Remark 2.78. Let G be a group. Assume G is generated by g1, ..., gn ∈ G as a group. We have
group homomorphisms ϕi : Z → G, ϕi (k) = gik . Repeated application of the universal property
of free products of groups yields the group homomorphism
(...(ϕ1 ⋆ ϕ2 ) ⋆ ϕ3 ) ⋆ · · · ⋆ ϕ n =: ϕ1 ⋆ · · · ⋆ ϕ n : (...(Z ⋆ Z) ⋆ Z) ⋆ · · · ⋆ Z =: Z ⋆ · · · ⋆ Z → G
with
(ϕ1 ⋆ · · · ⋆ ϕ n )((k1, i 1 ), ..., (kr , i r )) = gik11 · · · gikrr ,
where i j ∈ {1, ..., n} is an index used to make the Z-factors formally disjoint.
The fact that g1, ..., gn generate G is equivalent to the fact that ϕ1 ⋆ · · · ⋆ ϕ n : Z ⋆ · · · ⋆ Z → G
is onto. It follows that
Z⋆···⋆Z
G
ker(ϕ1 ⋆ · · · ⋆ ϕ n )
If the normal subgroup ker(ϕ1 ⋆ · · · ⋆ ϕ n ) is also finitely generated as a group, with generators
x 1, ..., x m , then we call G finitely presentable and
a presentation of G.
Example 2.79. For the cartesian product Z2 of two copies of Z we have the presentation
Z/2Z ha | a2 i.
Remark 2.80. A word of caution: isomorphic groups may have several, quite different presen-
tations. For example
because the generators xyxw −1 and zy−1 x −1 on the right-hand side can be used to eleminate w
and z. It is therefore often not obvious whether two presentations give rise to isomorphic groups.
55
2 Homotopy Theory
b−1
g a1
b1
a−1
g
a1−1
b−1
1
V
a2
b2
a2−1
The subset U is contractible and therefore π1 (U ) = {1}. The subset V is homotopy equivalent
to the boundary ∂D subject to the identifications of the equivalence relation, V ≃ ∂D/ ∼. The
identifications induced by ∼ generate a bouquet of 2g circles, one for each relation ai and bi .
The bouquet is denoted by S1 ∨ · · · ∨ S1 , where the wedge sum “∨” of two topological spaces X
and Y is defined to be the disjoint union of X and Y with identification of two base points x 0 ∈ X ,
y0 ∈ Y such that X ∨ Y := X ∪ Y /{x 0 ∼ y0 }. For the bouquet of circles all S1 ’s are joined at the
same base point. Graphically we can depict the bouquet of circles as follows
S1 ∨ · · · ∨ S1 =
So we have
V ≃ ∂D 2 / ∼ ≈ |
S1 ∨ {z
· · · ∨ S}1 .
2g times
56
2.5 The fundamental group of surfaces
with generators again denoted by a1, b1, a2, b2, ..., ag , bg . For the intersection of U and V we
have U ∩ V = D̊ 2 \ D̄ 2 ( 21 ) ≃ S1 and thus by Theorem 2.42 we have π1 (U ∩ V ) Z. A generator
of π1 (U ∩ V ) is given by a loop c of degree 1.
b−1
g a1
b1
a−1
g
a1−1
b−1
1
c
a2
b2
a2−1
Example 2.82. For the two-dimensional torus T 2 we find with Example 2.79
π1 (T 2 ) = π1 (F1 ) = ha, b | aba−1 b−1 i Z2,
in agreement with Remark 2.48.
57
2 Homotopy Theory
Proof. The statement follows once we see that π1 (Fg ) π1 (Fg ′ ). Attention here: as noted in
2.80 different presentations can yield isomorphic groups.
For any group G let [G, G] be the normal subgroup generated by all commutators aba−1 b−1 ,
a, b ∈ G. The abelian factor group G/[G, G] is called the abelianization of G. We now calculate
the abelianization of π1 (Fg ).
π1 (Fg )
= ha1, b1, ..., ag , bg | a1 b1 a1−1 b−1 −1 −1 −1 −1
1 · · · a g bg a g bg , a1 b1 a1 b1 , ...
[π1 (Fg ), π1 (Fg )]
..., a1 a2 a1−1 a2−1, ...i
= ha1, b1, ..., ag , bg | a1 b1 a1−1 b−1 −1 −1
1 , ..., a1 a2 a1 a2 , ...i
Z2g ,
where the second equality follows because the simple commutators ai bi ai−1 b−1
i , i = 1, ..., g
−1 −1 −1 −1
generate the relation a1 b1 a1 b1 · · · ag bg ag bg .
Hence if Fg ≃ Fg ′ then π1 (Fg ) π1 (Fg ′ ) and thus
π1 (Fg )/[π1 (Fg ), π1 (Fg )] π1 (Fg′ )/[π1 (Fg′ ), π1 (Fg′ )].
′
Thus Z2g Z2g and therefore g = g ′.
Remark 2.84. This proves the uniqueness part of the classification for surfaces, see Example 1.5.
Definition 2.85. Let W n = [0, 1] × . . . × [0, 1] be the n-cube. Let X be a topological space
| {z }
n times
and x 0 ∈ X . Then
πn (X ; x 0 ) := {[σ]∂W n | σ ∈ C (W n, X ), σ(∂W n ) = {x 0 }}
is called the n-th homotopy group of X with base point x 0 . Here [σ]∂W n denotes the homotopy
class of σ relative to ∂W n .
58
2.6 Higher homotopy groups
σ σ⋆τ
X t 2, . . . , t n
b
t1
x0
Now put [σ]∂W n · [τ]∂W n := [σ ⋆ τ]∂W n . The proof that this yields a well-defined group
multiplication on πn (X ; x 0 ) for n ≥ 2 is literally the same as in the case for n = 1. The neutral
element is represented by the constant map ε nx0 : W n → X , ε nx0 (t 1, . . . , t n ) = x 0 , and [σ]−1
∂W n is
represented by σ(1 − t 1, t 2, . . . , t n ).
Unlike for the case n = 1 the higher homotopy groups are abelian:
Proposition 2.86. Let X be a topological space and x 0 ∈ X . Then for n ≥ 2 the group
πn (X ; x 0 ) is abelian.
59
2 Homotopy Theory
σ τ σ σ σ
deform deform deform
−→ −→ −→
τ τ τ
τ σ
deform
−→
Remark 2.87. This proof also shows that replacing t 1 by any other variable in the definition of
σ ⋆ τ gives the same group multiplication on πn (X ; x 0 ).
f # : πn (X ; x 0 ) → πn (Y ; y0 )
Remark 2.88. Lemma 2.18, 2.19, Corollary 2.20, Propositions 2.21, 2.22, Theorem 2.23 and
Corollary 2.24 also hold for πn (X ; x 0 ). In particular, if f : X → Y is a homotopy equivalence
then the map f # : πn (X ; x 0 ) → πn (Y ; f (x 0 )) is an isomorphism. For γ ∈ Ω(X ; x 0, x 1 ) there is
an isomorphism
Φγ : πn (X ; x 1 ) → πn (X ; x 0 )
given by [σ]∂W n 7→ [σ ′]∂W n where
σ(2t),
σ ′ (t) =
kt kmax ≤ 21 ,
γ(2 − 2kt kmax ),
1
2 ≤ kt kmax ≤ 1.
γ
x0 x1 σ
60
2.6 Higher homotopy groups
π ϕ
Wn / W n /∂W n
≈
/ Sn
Remark 2.90. The Seifert-van-Kampen theorem for πn works only under very restrictive as-
sumptions. For this reason the computation of πn for explicit examples can be very difficult. We
will be able to compute πn (S m ) for n ≤ m but many of the πn (S m ) for n > m are actually still
unknown.
π0 (X ; x 0 ) = {path components of X }.
Definition 2.92. Let W, E and B be topological spaces and let p ∈ C (E, B). We say that p has
the homotopy lifting property, abbreviated by HLP, for W iff for every f ∈ C (W, E) and every
h ∈ C (W × [0, 1], B) with h(w, 0) = p( f (w)) for all w ∈ W there exists H ∈ C (W × [0, 1], E)
such that
H (w, 0) = f (w) ∀w ∈ W and h = p◦H.
In other words, there exists an H ∈ C (W × [0, 1], E) such that the diagram
f
∈ W /
w
❴ ✉: E
H✉ ✉
p
✉
✉ h
(w, 0) ∈ W × [0, 1] /B
commutes.
61
2 Homotopy Theory
Example 2.94. Now consider E = [0, ∞) × {0} ∪ (−∞, 0] × {1} ⊂ R2 , B = R and let p be the
projection onto the first factor, p(t, s) = t.
f
b
E
W = {w0 }
| B
h
The map p is surjective but still does not have the HLP for W = W 0 . For example, choose
h(w0, t) = t, f (w0 ) = (0, 1).
Definition 2.95. A map p ∈ C (E, B) is called a Serre fibration or weak fibration iff it has the
HLP for all W n , n ≥ 0. The space E is called the total space and B is called the base space of
the fibration. For b0 ∈ B we call p−1 (b0 ) the fiber over b0 .
Example 2.96. For topological spaces F and B put E := B × F and p = pr1 , the projection on
the B-factor. Then p has the HLP for all W . In particular, p is a Serre fibration. Namely, let
f ∈ C (W, B × F) and h ∈ C (W × [0, 1], B) be given such that p( f (w)) = h(w, 0) for all w ∈ W .
Now write f (w) = ( β(w), ϕ(w)) with β ∈ C (W, B) and ϕ ∈ C (W, F). Hence
h(w, 0) = p( f (w)) = β(w).
Now put H (w, t) := (h(w, t), ϕ(w)). Then H ∈ C (W × [0, 1], B × F) and
p(H (w, t)) = pr1 (h(w, t), ϕ(w)) = h(w, t) ,
H (w, 0) = (h(w, 0), ϕ(w)) = ( β(w), ϕ(w)) = f (w) ,
as required. Hence p has the HLP for any W .
62
2.6 Higher homotopy groups
Definition 2.97. A map p ∈ C (E, B) is called fiber bundle with fiber F iff for each b ∈ B there
exists an open subset U ⊂ B with b ∈ U and a homeomorphism Φ : p−1 (U ) → U × F such
that the diagram
Φ
U×F o ≈ p−1 (U )
tt
pr 1 ttt
tt
zttt p | p −1 (U )
U
commutes.
Proof. Let f ∈ C (W n, E) and h ∈ C (W n × [0, 1], B) such that h(w, 0) = p( f (w)) for all w ∈ W .
Now subdivide W n × [0, 1] such that h maps each subcube entirely into an open subset U as in
the definition of the fiber bundle.
W n × [0, 1]
[0, 1]
Q11
Wn
By Example 2.96 products have the HLP, hence we can extend the map f to a continuous map
H11 : (W n × {0}) ∪ Q11 → E such that p ◦ H11 = h.
63
2 Homotopy Theory
W n × [0, 1]
[0, 1]
Q11 Q12
Wn
Next we want to extend the lift to Q12 . Now there seems to be a problem because the required
lift H12 need not only coincide with f along the edge Q12 ∩ (W n × {0}) but also with H11 along
Q11 ∩ Q12 .
But there are homeomorphisms of a cube onto itself mapping two edges onto one as indicated in
the picture.
Thus we can apply the HLP and extend the lift to (W n × {0}) ∪ Q11 ∪ Q12 . Iteration of this
procedure proves the assertion.
Example 2.99. Let G be a Lie group, e.g. a closed subgroup of GL(n; K ) with K = R or K = C.
Let H ⊂ G be a closed subgroup. We equip the space B = G/H = {g · H | g ∈ G} with the
quotient topology. Such a space is called a homogeneous space. Then G → G/H with g 7→ g · H
is a fiber bundle with fiber H. The proof of this fact requires some technical work, see [8, p. 120
ff].
64
2.6 Higher homotopy groups
B 0
( ! )
0 1
H= B ∈ SO(n) .
Then H is a closed subgroup of G isomorphic to SO(n). We now show that G/H ≈ S n . Consider
the map f : G → S n with A 7→ A · en+1 where en+1 the (n + 1)-st unit vector of the canonical
basis. The map f is continuous and surjective. We observe
where the map π : G → G/H is the canonical projection. We conclude that the map f descends
to a bijective map f¯ : G/H → S n . By the universal property of the quotient topology the map
f¯ : G/H → S n is continuous. Since G is compact the space G/H is also compact. Moreover,
the sphere S n is a Hausdorff space, hence the map f¯ is a homeomorphism. Thus we obtain a
fiber bundle SO(n + 1) → S n with fiber SO(n).
Let p : E → B be any Serre fibration and fix e0 ∈ E. Put b0 := p(e0 ) ∈ B and let F := p−1 (b0 ).
Then e0 ∈ F. Let ι : F → E be the inclusion map. We obtain the following two homomorphisms:
ι# : πn (F; e0 ) → πn (E; e0 ),
p# : πn (E; e0 ) → πn (B; b0 ).
Then we have
65
2 Homotopy Theory
b
x
b
η k (x)
b
( 12 , . . . , 21 , −1)
This homeomorphism maps the faces out of which Boxk is built onto the regions depicted
in the right figure. In order to define ∂ : πn (B; b0 ) → πn−1 (F; e0 ) let σ ∈ C (W n, B) with
σ(∂W n ) = {b0 }.
F E
tn
b e0
σ p
b
b0
t 1, . . . , t n−1
Since (t 1, . . . , t n−1 ) 7→ σ(t 1, . . . , t n−1, 0) = b0 is constant we can lift it to the constant map
(t 1, . . . , t n−1 ) 7→ e0 . Now the HLP of p for W n−1 yields a continuous map Σ : W n → E with
66
2.6 Higher homotopy groups
a) We have to show that [σ̃]∂W n−1 does not depend on the particular choice of the lift Σ.
Let Σ ′ ∈ C (W n, E) be another lift of σ with Σ ′ (t 1, . . . , t n−1, 0) = e0 . Then Σ−1 • Σ ′ is a lift of
σ −1 • σ. Here • denotes the concatenation with respect to the variable t n , Σ−1 (t 1, . . . , t n ) =
Σ(t 1, . . . , t n−1, 1−t n ) and similarly for σ −1 . Since σ −1 • σ ≃∂W n ε bn0 we can find a homotopy
h : W n+1 → B relative to ∂W n with
Then h(Boxn ) = {b0 }. We apply the HLP of p for W n+1 to get a lift H ∈ C (W n+1, E) of h
with
H (t 1, . . . , t n, 0) = Σ−1 • Σ ′ (t 1, . . . , t n ).
Σ′
Σ−1
t n+1 tn
t 1, . . . , t n−1
67
2 Homotopy Theory
σ′
B
h
b
b0
t n+1 tn
t 1, . . . , t n−1
σ
The HLP for W n yields a lift H : W n+1 → E of h with H (t 1, . . . , t n−1, 0, t n+1 ) = e0 . We thus
obtain a homotopy
Ĥ (t 1, . . . , t n−1, s) = H (η −1
n−1 (t 1, . . . , t n−1 ), s)
We have shown that the map ∂ : πn (B; b0 ) → πn−1 (F, e0 ) is well defined.
Lemma 2.101. For n ≥ 2 the map ∂ : πn (B; b0 ) → πn−1 (F, e0 ) is a group homomorphism.
(Σ ◦ η −1 −1 −1
n−1 ) ⋆ (T ◦ η n−1 ) ≃ ∂W n−1 H ◦ η n−1
68
2.6 Higher homotopy groups
b b b
We conclude that
∂([σ]∂W n ) · ∂([τ]∂W n ) = [Σ ◦ η −1 −1
n−1 ]∂W n−1 · [T ◦ η n−1 ]∂W n−1
= [(Σ ◦ η −1 −1
n−1 ) ⋆ (T ◦ η n−1 )]∂W n−1
= [H ◦ η −1
n−1 ]∂W n−1
= ∂([σ ⋆ τ]∂W n )
= ∂([σ]∂W n · [τ]∂W n ).
∂ ι# p# ∂ ι#
... / πn (F; eo ) / πn (E; e0 ) / πn (B; b0 ) / πn−1 (F; e0 ) / ...
∂ ι# p#
... / π0 (F; e0 ) / π0 (E; e0 ) / π0 (B; b0 )
69
2 Homotopy Theory
Remark 2.103. Exactness means that the image of the incoming map equals the kernel of the
outgoing map. The question arises what this means on the π0 -level where we do not have
homomorphisms. The image is defined for an arbitrary map. For the kernel we recall that π0 is
a set together with a distinguished element which corresponds to the neutral element of a group.
It is therefore natural to define the kernel of a map to be the set of all elements in the domain of
the map which are mapped to the distinguished element. Having clearified this, exactness of the
above sequence also makes sense on the π0 -level.
[0, 1]
Wn
[τ] E, ∂W n = [H ◦ η −1 −1
n−1 ] E, ∂W = ι# ([H ◦ η n−1 ] F, ∂W ) ∈ im ι# .
n n
70
2.6 Higher homotopy groups
ε en0
ii) ker ι# ⊂ im ∂:
Let [τ]∂W n ∈ πn (F; e0 ) with 0 = ι# ([τ]∂W n ). Hence τ ≃∂W n ε en0 in E. Let H be a
homotopy in E relative to ∂W n from ε en0 to τ. Then H is a lift of h := p ◦ H. Since
H maps ∂W n+1 to F, the map h maps ∂W n+1 to b0 . Thus h represents an element in
πn+1 (B; b0 ). By definition of ∂ we have [H ◦ η −1
n−1 ]∂W = ∂([h]∂W n+1 ).
n
[0, 1]
ε en0
Wn
71
2 Homotopy Theory
e0
e0
e0 τ e0 deform
−→ e0 τ e0 deform
−→ τ
e0
e0
Therefore [τ]∂W n = [H ◦ η −1
n−1 ]∂W = ∂[h]∂W n+1 ∈ im(∂).
n
ii) ker ∂ ⊂ im p# :
Let [σ]∂W n ∈ πn (B; b0 ) with ∂([σ]∂W n ) = 0. Let Σ be a lift of σ with initial condition
ε en−1
0
. Then Σ ◦ η −1
n−1 represents ∂([σ]∂W ) = 0. Hence we have in F
n
Σ ◦ η −1 n−1
n−1 ≃ ∂W n−1 ε e 0
e0
maps to F
e0 F F e0
Σ
e0
72
2.6 Higher homotopy groups
b0
deform deform
−→ −→
b0 σ b0 σ σ
p# : πn (E, e0 ) → πn (B; b0 )
Example 2.106. The map Exp : R → S1 with t 7→ e2πit is a covering with fiber Z. Hence
πk (S1 ; 1) πk (R, 0) = {0}
for all k ≥ 2. More generally, Exp : Rn → T n = S1 × . . . × S1 with
(t 1, . . . , t n ) 7→ (e2πit 1 , . . . , e2πit n )
is a covering with fiber Zn . Hence πk (T n ) πk (Rn ) = {0} for all k ≥ 2.
Example 2.107. Consider the real projective space, defined by RPn := S n / ∼, where x ∼ y
⇐⇒ x = y or x = −y. Then the map p : S n → RPn with x 7→ [x]∼ is a covering with fiber
Z/2Z. Hence πk (RPn ) πk (S n ) for all k ≥ 2. For n ≥ 2 we investigate the sequence:
{0} = π1 (S n ) / π1 (RPn ) / π0 (Z/2Z) / π0 (S n ) = {0}
73
2 Homotopy Theory
We deduce that π1 (RPn ) π0 (Z/2Z) Z/2Z as sets. But then π1 (RPn ) Z/2Z also as groups
because there is only one group of order 2 (up to isomorphism).
Example 2.108. We know that SO(n + 1)/SO(n) ≈ S n from Example 2.99. In the case of n = 2
this means that SO(3)/SO(2) ≈ S2 . We also know that SO(2) ≈ S1 . For k ≥ 3 we consider the
long exact sequence
We know that πk (S1 ) = {0} and πk−1 (S1 ) = {0}. Hence we find
* +
x 2 + y 2 − u2 − v 2 2(yu − xv) 2(yv − xu)
f (x, y, u, v) = .. 2(yu + xv) x2 − y 2 + u2 − v 2 2(uv − xy) //
, 2(yv − xu) 2(uv + xy) x 2 − y 2 − u2 + v 2 -
satisfies f (−x, −y, −z, −v) = f (x, y, z, v) and therefore induces a continuous map
f¯ : RP3 → SO(3). This map is bijective and thus a homeomorphism. Hence we have
SO(3) ≈ RP3 . It follows that
for all k ≥ 3. Later we will see (Example 3.121 on page 151) that π3 (S3 ) Z and consequently
π3 (S2 ) Z.
Example 2.109. By the same proof as for SO(n + 1)/SO(n) ≈ S n we get that
Therefore there is a fiber bundle SU(n + 1) → S2n+1 with fiber SU(n). For n ≥ 1 we consider
the following part of the long exact homotopy sequence:
ι# p#
π0 (SU(n)) / π0 (SU(n + 1)) / π0 (S 2n+1 ) = {0}.
Hence the map ι# : π0 (SU(n)) → π0 (SU(n+1)) is onto. Since SU(1) = {1} we have π0 (SU(1)) =
{0} and thus π0 (SU(n)) = {0} by induction on n. Thus SU(n) is path-connected for all n ≥ 1.
Now let us analyze π1 (SU(n)). Consider
ι# p#
π1 (SU(n)) / π1 (SU(n + 1)) / π1 (S 2n+1 ) = {0}.
Again, we conclude that the map ι# : π1 (SU(n)) → π1 (SU(n+1)) is onto. By the same induction
as before we find π1 (SU(n)) = {0} for all n ≥ 1. Thus SU(n) is simply connected for all n ≥ 1.
74
2.7 Exercises
2.7 Exercises
2.1. Let X be a set and let x 0 ∈ X . Determine π1 (X ; x 0 ) where
2.2. Let X be a topological space and let ω : S1 → X be continuous. Show that the following
are equivalent:
2.3. Let X be a topological space and let f , g : X → S n be continuous. Assume f (x) , −g(x)
for all x ∈ X . Show that f and g are homotopic.
2.4. Let X and Y be topological spaces. Show that X × Y is contractible if and only if X and Y
are contractible.
is a group isomorphism.
2.6. Let X = [0, 1] × [0, 1] ⊂ R2 and let A ⊂ X be the comb space. Show that there is no
retraction X → A.
f (z) − p
f p (z) = .
| f (z) − p|
Then
U ( f , p) := deg( f p )
is called the winding number of f around p.
a) Show that for p, q ∈ R2 \ f (S1 ) which can be joined by a continuous path in R2 \ f (S1 ) we
have U ( f , p) = U ( f , q).
75
2 Homotopy Theory
2.10. On [0, 1] × [−1, 1] consider the equivalence relation ∼ given by (t, s) ∼ (t ′, s ′ ) iff (t, s) =
(t ′, s ′ ) or |t − t ′ | = 1 and s ′ = −s. The quotient space M := [0, 1] × [−1, 1]/ ∼ is called the
Möbius strip. The image S of [0, 1] × {0} is called the chord of M, that of [0, 1] × {−1, 1} is the
boundary ∂ M of M.
a) Show that ∂ M is homeomorphic to S1 .
b) Show that S is a strong deformation retract of M.
c) Determine π1 (M; x 0 ) for some x 0 ∈ ∂ M and the subgroup ι# (π1 (∂ M; x 0 )) ⊂ π1 (M; x 0 )
where ι : ∂ M ֒→ M is the inclusion map.
d) Show that ∂ M is not a retract of M.
2.14. Let X be a path-connected topological space. Show that the suspension ΣX (see Exer-
cise 1.11) is also path-connected.
2.15. Show that the map C → C, z 7→ z 2 , has the homotopy lifting property for W 0 but not for
W 1.
76
2.7 Exercises
π1 (X, {0}) Z
π1 (RPn ) Z/2Z.
2.18. Decide by proof or counter-example whether or not the following assertion holds true:
The Seifert-van Kampen theorem also holds if one only assumes that U ∩ V is connected rather
than path-connected.
2.19. Let E = B × F and p = pr1 : E → B be the product fibration. Show that the boundary
map ∂ : πn+1 (B; b0 ) → πn (F; e0 ) is trivial in this case.
2.20. Let p : E → B be a Serre fibration. Show that for [σ]∂W n+1 ∈ πn+1 (B; b0 ) the class
[Σ ◦ η −1
n ]∂W n ∈ π n (F; e0 ) does not depend on the choice of lift Σ of σ with the “initial condition”
ε en0 .
πk (SU(n)) πk (U(n))
for all k ≥ 2.
2.22. The complex projective space is defined as CP n := S2n+1 / ∼ where z ∼ w iff there exists
u ∈ S1 ⊂ C such that z = u · w. Here we have regarded S2n+1 = {z ∈ Cn+1 | |z| = 1} as a subset
of Cn+1 . The map p : S2n+1 → CP n , z 7→ [z]∼ , is called the Hopf fibration.
Compute
πk (CP n )
for all k ≤ 2n.
Hint: You can use πk (S m ) = {0} for all 1 ≤ k < m and m ≥ 2.
77
2 Homotopy Theory
i p
2.25. Let 0 → A → − A′ − → A′′ → 0 be an exact sequence of abelian groups. Show that the
following three conditions are equivalent:
(i) There exists an isomorphism Ψ : A′ → A × A′′ such that the following diagram commutes:
i p
0 /A / A′ / A′′ /0
= Ψ =
0 /A / A × A′′ / A′′ /0
where the arrows A → A× A′′ and A× A′′ → A′′ are given by the canonical maps a 7→ (a, 0)
and (a, a ′′ ) 7→ a ′′, respectively.
If these conditions hold then we say that the exact sequence is split.
i p
2.26. a) Show that every exact sequence 0 → A → − A′ − → A′′ → 0 of vector spaces (i.e. all
spaces are vector spaces over some fixed field and all homomorphisms are linear maps) is
split.
2
b) Show that the exact sequence 0 → Z →
− Z → Z/2Z → 0 is not split.
78
3 Homology Theory
Homotopy groups are in general hard to compute. For instance, for spheres not all homotopy
groups are known even now. In this chapter we introduce rougher invariants which are much
easier to determine, the homology groups.
e0 = (0, . . . , 0) ∈ Rn
e1 = (1, . . . , 0) ∈ Rn
..
.
en = (0, . . . , 1) ∈ Rn
Now we define
e2 b
n=2 ∆2 = triangle b b
e0 e1
79
3 Homology Theory
∆3
n=3 ∆3 = tetrahedron b b
Now we fix a commutative ring R with 1. The most important examples will be
R = R, Q, C, Z, Z/nZ.
i
For n ≥ 0 we consider the affine linear map Fn+1 : ∆n → ∆n+1 given by
e j,
(e j ) =
j <i
e j+1,
i
Fn+1 (3.1)
j ≥i
80
3.1 Singular homology
X
e2 σ
b
F21
b
b
e0 e1 e0 e1
σ (i)
Lemma 3.6. ∂ ◦ ∂ = 0.
Proof. It suffices to prove ∂∂σ = 0 for all n-simplices σ. For j < i we have
j
Fni ◦ Fn−1 = Fnj ◦ Fn−1
i−1
. (3.2)
We compute
n
∂∂σ = ∂ * (−1) i σ ◦ Fni +
X
, i=0 -
Xn
= (−1) i ∂ σ ◦ Fni
i=0
81
3 Homology Theory
n
X n−1
X
i j
= (−1) (−1) j σ ◦ Fni ◦ Fn−1
i=0 j=0
X X
j j
= (−1) i+ j σ ◦ Fni ◦ Fn−1 + (−1) i+ j σ ◦ Fni ◦ Fn−1
0≤ j <i ≤n 0≤i ≤ j ≤n−1
X X
j ′ ′ j′ ′
= (−1) i+ j σ ◦ Fn i−1
◦ Fn−1 + i −1
(−1) j +i −1 σ ◦ Fn ◦ Fn−1
0≤ j <i ≤n 0≤ j ′ <i ′ ≤n
= 0.
In the last step we used (3.2) for the first sum and changed the summation indices from i → j ′
and j → i ′ − 1 in the second sum.
Example 3.9. Assume that X = {point}. Then there is only one singular n-simplex, namely
the constant map σ n : ∆n → X . In other words, Sn (X ; R) = R · σ n . Consequently,
σ n(i) = σ n ◦ Fni = σ n−1 and
n
0, n odd,
X
i
= σ n−1,
∂σ n = (−1) σ n−1 n even, n , 0,
0,
i=0 n = 0.
This implies
R · σ n,
Z n (X ; R) =
n odd or n = 0,
0,
n even and n , 0,
82
3.1 Singular homology
and
R · σ n,
Bn (X ; R) =
n odd,
0,
n even.
We conclude that
R, if n = 0,
Hn (X ; R)
0, otherwise.
As in homotopy theory we not only associate groups to spaces but also homomorphisms to
maps. Let f : X → Y be a continuous map. Then we obtain an R-module homomorphism
Sn ( f ) : Sn (X ; R) → Sn (Y ; R) by setting
m
X m
X
Sn ( f ) α i · σi := α i · ( f ◦ σi ).
i=1 i=1
Sn ( f )
Sn (X ; R) / Sn (Y ; R)
∂ ∂
S n−1 ( f )
Sn−1 (X ; R) / Sn−1 (Y ; R)
Proof. We compute
∂(Sn ( f )(σ)) = ∂( f ◦ σ)
Xn
= (−1) i ( f ◦ σ) ◦ Fni
i=0
n
X
= (−1) i f ◦ (σ ◦ Fni )
i=0
n
X
= Sn−1 ( f ) (−1) i σ ◦ Fni
i=0
= Sn−1 ( f )(∂σ).
This lemma implies Sn ( f )(Z n (X ; R)) ⊂ Z n (Y ; R) and Sn ( f )(Bn (X ; R)) ⊂ Bn (Y ; R). Hence
we obtain a well-defined R-module homomorphism Hn ( f ) : Hn (X ; R) → Hn (Y ; R) where
Hn ( f )([x]) := [Sn ( f )(X )]. Here the square brackets denote the homology classes of the
n-cycles. One sees directly from the definition that Hn (·) has the functorial properties
(i) Hn (id X ) = idH n (X ;R) ,
83
3 Homology Theory
(ii) Hn ( f ◦ g) = Hn ( f ) ◦ Hn (g).
Exactly as for homotopy groups these functorial properties imply that a homeomorphism f :
X → Y induces an isomorphism Hn ( f ) : Hn (X ; R) → Hn (Y ; R). Homeomorphic spaces have
isomorphic homology groups.
Sn (X ) / Sn (X, A)
∂ ∂¯
Sn−1 (X ) / Sn−1 (X, A)
commutes. Since ∂ ◦ ∂ = 0 and since Sn (X ) → Sn (X, A) is onto we also have that ∂¯ ◦ ∂¯ = 0. Set
Z n (X, A) = Z n (X, A; R) := ker ∂¯ : Sn (X, A) → Sn−1 (X, A) ,
Bn (X, A) = Bn (X, A; R) := im ∂¯ : Sn+1 (X, A) → Sn (X, A) .
Since Z n (X, A) = Z n′ (X, A)/Sn (A) and Bn (X, A) = Bn′ (X, A)/Sn (A) we obtain
Z n′ (X, ∅) = Z n (X ),
Bn′ (X, ∅) = Bn (X ),
Hn (X, ∅) = Hn (X ).
84
3.2 Relative homology
σ : ∆1 → X,
(te1 + (1 − t)e0 ) 7→ (cos (2πt), sin (2πt), 1).
|
>
σ X
∆1
A
∂σ = σ(e1 ) − σ(e0 ) = 0.
85
3 Homology Theory
Therefore σ ∈ Z1 (X ) ⊂ Z1′ (X, A) and, as we will see later, represents a nontrivial element in
H1 (X ). For the 2-simplices σ1 and σ2 defined as indicated in the following picture
σ
>
>
σ1 X
∆2
σ2
we find for the 2-chain σ1 + σ2 the boundary ∂(σ1 + σ2 ) = σ + a where a ∈ S1 (A). Hence,
modulo a ∈ S1 (A), we have σ ∈ B1 (X, A) and therefore 0 = [σ] ∈ H1 (X, A).
where c ∈ Z n′ (X, A). Note that ∂c ∈ Z n−1 (A) since ∂ 2 = 0. The connecting homomorphism is
well defined because replacing c by another representative c + ∂b + a where b ∈ Sn+1 (X ) and
a ∈ Sn (A) yields
c + ∂b + a 7→ ∂(c + ∂b + a) = ∂c + ∂a.
Since ∂a ∈ Bn−1 (A) we get [∂c + ∂a] = [∂c] ∈ Hn−1 (A). Since ∂ : Z n′ (X, A) → Sn−1 (A) is a
homomorphism, the connecting homomorphism is also a homomorphism.
86
3.3 The Eilenberg-Steenrod axioms and applications
Hn ( f )
Hn (X, A) / Hn (Y, B)
∂ ∂
H n−1 ( f | A )
Hn−1 (A) / Hn−1 (B)
Proof. We compute
X X
Hn−1 ( f | A ) ∂ α i σi = Hn−1 ( f | A ) ∂ α i σi
i i
X X
j
= Hn−1 ( f | A ) αi (−1) j σ j ◦ Fn−1
i j
X X
j
= αi (−1) j ( f | A ) ◦ (σi ◦ Fn−1 )
i j
X X
j
= αi (−1) j ( f ◦ σi ) ◦ Fn−1
i j
X
= ∂ α i ( f ◦ σi )
i
X
=∂ α i ( f ◦ σi )
i
X
= ∂ Hn ( f ) α i σi .
i
Dimension Axiom. (
R, n = 0
Hn ({point}; R)
0, otherwise
The next axiom deals with homotopy invariance. Two continuous maps f 0, f 1 : (X, A) → (Y, B)
of pairs of spaces are called homotopic (in symbols f 0 ≃ f 1 ) iff there exists an H ∈ C (X ×[0, 1], Y )
87
3 Homology Theory
H (x, 0) = f 0 (x),
H (x, 1) = f 1 (x),
H (A × [0, 1]) ⊂ B.
Homotopy Axiom.
Let f 0, f 1 ∈ C ((X, A), (Y, B)) be homotopic, f 0 ≃ f 1 . Then the induced maps on homology
coincide, i.e.,
Hn ( f 0 ) = Hn ( f 1 ) : Hn (X, A) → Hn (Y, B)
holds for all n.
Remark 3.14. Let (X, A) ≃ (Y, B), i.e., there exist f ∈ C ((X, A), (Y, B)) and
g ∈ C ((Y, B), (X, A)) such that f ◦ g ≃ id(Y, B) and g ◦ f ≃ id (X, A) . It then follows as be-
fore that Hn (X, A; R) Hn (Y, B; R).
Exactness Axiom. For any pair of spaces (X, A) and inclusion maps i : A ֒→ X ,
j : (X, ∅) ֒→ (X, A), the sequence
H n+1 ( j)
Hn+1 (X, A) o ···
∂
H n (i) H n ( j)
Hn (A) / Hn (X ) / Hn (X, A)
∂
H n−1 (i)
··· o Hn−1 (A)
Excision Axiom.
For every pair of spaces (X, A) and every U ⊂ A with Ū ⊂ Å the homomorphism
Hn ( j) : Hn (X \ U, A \ U ) → Hn (X, A)
88
3.3 The Eilenberg-Steenrod axioms and applications
j : (X \ U, A \ U ) ֒→ (X, A)
is an isomorphism.
The proofs of axioms A2, A3, and A4 will be given later. Before that we wil show their usefulness
by studying some basic examples.
R2 R
∈
Since the map (a, b) 7→ a + b is onto, the map H0 (D 1 ) → H0 (D 1, S0 ) must be zero. Hence
H0 (D 1, S0 ) = 0.
For n ≥ 2 we have the following exact sequence
H0 (S n−1 ) −→ H0 (D n ) −→ H0 (D n, S n−1 ) −→ 0
id
R −→ R
Again the second arrow has to be trivial and therefore H0 (D n, S n−1 ) = 0. This settles the
case m = 0.
89
3 Homology Theory
H1 (D n ) −→ H1 (D n, S n−1 ) −→ H0 (S n−1 ) −→ H0 (D n )
For n = 1 we have
H1 (D 1 ) −→ H1 (D 1, S0 ) −→ H0 (S0 ) −→ H0 (D 1 )
H1 ({point}) R2 R
∈
0 (a, b) 7−→ a+b
0
H1 (D n ) −→ H1 (D n, S n−1 ) −→ H0 (S n−1 ) −→ H0 (D n )
=
Sn
D−n
For n ≥ 1 we have
H1 (D−n ) −→ H1 (S n ) −→ H1 (S n, D−n ) −→ H0 (D n ) −→ H0 (S n )
=
90
3.3 The Eilenberg-Steenrod axioms and applications
Sn
D−n
U−n
0 0
and
Hm (D n ) −→ Hm (D n, S n−1 ) −→ Hm−1 (S n−1 ) −→ Hm−1 (D n )
=
0 0
This yields
Hm (S n ) Hm (S n , D−n ) Hm (D n, S n−1 ) Hm−1 (S n−1 ) (3.5)
Induction over m concludes the proof.
91
3 Homology Theory
Remark 3.17. Let us describe a generator of H1 (S1 ) Z geometrically. We use the isomor-
phisms in (3.4) and start with H1 (D 1, S0 ). The map c : ∆1 → D 1 , t 7→ cos(π(1 − t)), is a singular
1-simplex with ∂c = const1 − const−1 . Under the isomorphism H0 (S0 ) R2 it maps to (1, −1)
which generates the kernel of the map R2 → R given by (a, b) 7→ a + b. Hence c represents a
generator of H1 (D 1, S0 ).
The isomorphism H1 (D+1 , S0 ) → H1 (D 1, S0 ) induced by vertical projection gives us a generator
of H1 (D+1 , S0 ), namely the homology class represented by c ′ : ∆1 → D+1 , t 7→ eiπ (1−t ) .
The isomorphism H1 (D+1 , S0 ) → H1 (S1, D−1 ) is induced by the inclusion. Hence [c ′] ∈
H1 (S1, D−1 ) is a generator. Now
F (t, 0) = c ′ (t),
F (t, 1) = eiπ (1−2t ) =: c ′′ (t),
F (0, s) = −1 ∈ D−1 ,
F (1, s) = e−iπ s ∈ D−1 .
Thus c ′ ≃ c ′′ as maps (∆1, {0, 1}) → (S1, D−1 ). Using the homotopy axiom we compute
Therefore [c ′′] ∈ H1 (S1, D−1 ) is a generator. Since ∂c ′′ = const c ′′ (1) − constc ′′ (0) = const−1 −
const−1 = 0, the 1-simplex c ′′ represents a homology class in H1 (S1 ). Moreover, since inclusion
induces an isomorphism H1 (S1 ) → H1 (S1, D−1 ) we find that [c ′′] ∈ H 1 (S1 ) is a generator.
Using the homotopy (t, s) 7→ eiπ (s−2t ) we see that t 7→ e−2iπt also represents a generator of
H1 (S1 ). Finally, playing the same game with lower and upper hemispheres interchanged shows
that t 7→ e2iπt represents a generator of H1 (S1 ) as well.
As a first application of Theorem 3.16 we now prove Brouwer’s fixed point theorem in all
dimensions.
Proof. We assume that the map f does not have a fixed point and then derive a contradiction.
Let n ≥ 2 (the case n = 1 having been treated in Remark 1.6).
92
3.3 The Eilenberg-Steenrod axioms and applications
Dn
Now consider the continuous map g : D n → S n−1 as in
x
the picture. Denote the inclusion map by ι : S n−1 → D n f (x)
• g(x)
and note that g ◦ ι = idS n−1 .
∂D n = S n−1
s(x 0, x 1, . . . , x n ) = (−x 0, x 1, . . . , x n ).
93
3 Homology Theory
Proof. The proof is given by induction. First consider the case n = 1. Let c : ∆1 → S1 be a
1-simplex generating H1 (S1 ) and let S1 (s)(c) be its image under the induced homomorphism on
1-chains.
S1 (s)
b
∧ b
∨
We need to show that H1 (s)[c] = [S1 (s)c] = −[c]. We construct two 2-simplices. The first one
is given by
e2
b σ1
b
∧
b b
e0 e1
94
3.3 The Eilenberg-Steenrod axioms and applications
e2
b σ2
b b
e0 e1
∂(σ1 + σ2 ) = S1 (s)c + c
and hence
0 = [∂(σ1 + σ2 )] = [S1 (s)c + c] = [S1 (s)c] + [c] .
This yields the desired result in the case of n = 1.
The induction step n − 1 ⇒ n follows from the following commutative diagram:
Hn (S n ) / Hn (S n , D n ) o
− Hn (D+n, S n−1 ) / Hn−1 (S n−1 )
where the horizontal isomorphisms are the ones in (3.5). Commutativity of the last square
follows from Lemma 3.13 and that of the first and the second one from the fact that the horizontal
isomorphisms are induced by inclusion maps (which commute with s). Note here that we have to
choose the lower hemisphere with respect to a different coordinate than the one which is reflected
by s so that s(D−n ) = D−n .
Remark 3.21. Let a : S n → S n with a(x) = −x for all x ∈ S n be the antipodal map then
Hn (a) = (−1) n+1 . This follows from the fact that a is the composition of n + 1 reflections.
95
3 Homology Theory
Definition 3.22. A vector field on S n is a map v : S n → Rn+1 such that v(x) ⊥ x for all
x ∈ Sn .
x
v(x)
b
Loosely speaking, this means that every continuously combed hedgehog has a “bald” spot.
Proof. If n is odd we simply set
Hn ( f ) : Hn (S n ; Z) Z → Hn (S n ; Z) Z
96
3.4 The degree of a continuous map
Definition 3.24. The number deg( f ) is called the degree of the map f . The degree can be
defined in the same way for continuous maps f : (D n, S n−1 ) → (D n, S n−1 ).
2.) For the antipodal map we have seen that deg(a) = (−1) n+1 .
(i) deg(id) = 1;
(ii) deg(const) = 0;
Proof. The first and third assertion follow directly from the functorial property of Hn ( f ). The
fourth and fifth statement follow from the homotopy axiom. The second assertion follows from
the fact that the homomorphism induced by a constant map factors through Hn (pt) = 0. The last
statement of the lemma follows fromm the commutativity of the following diagram:
H n+1 ( f )
Hn+1 (D n+1, S n ) / Hn+1 (D n+1, S n )
∂ ∂
H n+1 ( f | S n )
Hn (S n ) / Hn (S n )
Theorem 3.27. (i) Every f ∈ C (S n, S n ) without fixed points satisfies deg( f ) = (−1) n+1 .
(ii) Every f ∈ C (S n, S n ) without an antipodal point, i.e., f (x) , −x for all x ∈ S n , satisfies
deg( f ) = 1.
97
3 Homology Theory
Proof. (i) Let f ∈ C (S n, S n ) be without a fixed point. Then the line segment joining f (x)
and −x does not contain the origin.
x b
b
0
f (x) b b
−x
x b
b
0
b
−x
b
f (x)
Since G(x, 0) = f (x) and G(x, 1) = x we have that f ≃ id and deg( f ) = deg(id) = 1
follows.
(iii) Finally, assume that f ∈ C (S n, S n ) has neither fixed points nor antipodal points. Then
deg( f ) = (−1) n+1 by (i) and by deg( f ) = 1 by (ii). Thus n must be odd.
98
3.4 The degree of a continuous map
j1 : S n → S n × S n , x 7→ (x, p),
n n n
j2 : S → S × S , x 7→ (p, x) .
(H n ( j1 ), H n ( j2 )) H n (µ)
Z2 Hn (S n ) ⊕ Hn (S n ) / Hn (S n × S n ) / Hn (S n ) Z
6
(d1, d2 )
with d µ ∈ Z.
Definition 3.28. The pair of numbers (d1, d2 ) ∈ Z2 is called the bidegree of the map µ.
Remark 3.29. The bidegree does not depend on the choice of p. If one chooses another p′ , then
a path from p to p′ will yield a homotopy between the corresponding embedding maps jν and jν′ .
Hence they induce the same homomorphisms on homology and therefore the same bidegrees.
Examples 3.30. 1.) Consider the case n = 1, S1 ⊂ C and let the map µ be given by µ(z1, z2 ) =
z1 z2 . Choose p = 1 ∈ S1 . Then µ ◦ j1 = id and thus
d1 = deg(µ ◦ j1 ) = deg(id) = 1.
Remark 3.31. The quaternions H form a division algebra isomorphic to R4 as a vector space.
The algebra H is associate and noncommutative. The standard vector space basis we be denoted
by 1, i, j, k. Therefore any h ∈ H can be uniquely written as
h = h0 + h1 i + h2 j + h3 k .
99
3 Homology Theory
Proof. The proof is by induction on k. For k = 0 and k = 1 the statement is trivial because
constant maps have degree 0 while the identity has degree 1. The case of k = −1 has already
been shown for n = 1, since here z 7→ z −1 = z̄ is a reflection and hence deg( f −1 ) = −1. On the
other hand for k = −1, n = 3 we note that S1 ⊂ S2 ⊂ S3 regarding C ⊂ H. Now the following
commutative diagram
implies that deg( f −1 ) = −1 in the quaternionic case too. The horizontal isomorphisms are
obtained as the composition of the isomorphisms
We used that the quaternionic multiplication restricted to the complex numbers C is just complex
multiplication.
Now we are ready to carry out the induction over k. We consider k > 0 and we show that the
statement for k − 1 implies that for k. Let µ be complex (resp. quaternionic) multiplication. Then
Now we can show that the fundamental theorem of algebra 2.34 also holds for quaternionic
polynomials.
100
3.4 The degree of a continuous map
Proof. Suppose the polynomial p has no root. Then we can define the continuous map
p̂ : S3 → S3 with p̂(z) := | p(z) 3 3 p(t z)
p(z) | . Now consider F (z, t) : S × [0, 1] → S with F (z, t) := | p(t z) | .
We observe that F (z, 0) = const and F (z, 1) = p̂(z), hence we have p̂(z) ≃ const and conse-
quently deg( p̂) = 0.
On the other hand, we can put for z ∈ S3 and t > 0
t k p zt
G(z, t) :=
|t k p zt |
and observe that the expression
z
k
t p = z k + tα 1 z k−1 + . . . + t k α
t
extends continuously to t = 0. The map G : S3 × [0, 1] → S3 satisfies G(z, 0) = f k (z) and again
G(z, 1) = p̂(z). Thus p̂ ≃ f k and hence deg( p̂) = k. This contradicts k > 0.
(i) (ii)
Z Hn (S n ; Z) / Hn (S n , S n \ f −1 (p); Z) o Hn (U, U \ f −1 (p); Z) (3.6)
deg p ( f ) Hn ( f )
(iii)
Z Hn (S n ; Z)
/ Hn (S n, S n \ {p}; Z)
We observe:
Concerning (i): This homomorphism is induced by the inclusion
S n = (S n , ∅) ֒→ (S n, S n \ f −1 (p)).
101
3 Homology Theory
Definition 3.34. The number deg p ( f ) is called the local degree of f over p.
2.) If f : U → S n is the inclusion map then deg p ( f ) = 1 for all p ∈ U. Namely, in this case the
homomorphisms (i) and (iii) in (3.6) coincide and Hn ( f ) is the inverse of (ii).
Proposition 3.36. Assume that f −1 (p) ⊂ K ⊂ V ⊂ U where K is compact and V open. Then
the degree deg p ( f ) is given by:
Hn (S n ) / Hn (S n , S n \ K ) o Hn (V, V \ K )
deg p ( f ) H n ( f |V )
Hn (S n )
/ Hn (S n, S n \ {p})
Hence we can replace f −1 (p) by a larger compact set in U and also U by a smaller open
neighborhood of f −1 (p). For this reason we call deg p ( f ) local.
Proof. The assertion follows from the commutativity of the following diagram:
Hn (S n6 , S n O \ f −1 (p)) o Hn (U, U \O f −1❙(p))
♠♠ ❙❙❙❙
♠♠♠♠ ❙❙H❙n❙( f )
♠ ❙❙❙❙
♠♠♠
♠♠♠ ❙❙)
Hn (S n ) ◗ Hn (S
5
n , S n \ {p}) o Hn (S n )
◗◗◗ ❦❦
◗◗◗
◗◗◗ ❦❦❦ ❦❦❦❦
◗◗◗ ❦❦
(
❦❦❦❦ H n ( f |V )
Hn (S n , S n \ K ) o Hn (V, V \ K )
Remark 3.37. For f : S n → S n we have that deg( f ) = deg p ( f ) for all p ∈ S n . Simply choose
K = V = S n in Proposition 3.36.
102
3.4 The degree of a continuous map
Lemma 3.38. Let f : (D+n, S n−1 ) → (D+n, S n−1 ) be continuous and p ∈ D̊+n such that f −1 (p) ⊂
D̊+n is compact. Then
deg( f ) = deg p f | D̊+n .
Proof. We extend f to a continuous map F : (S n , D−n ) → (S n, D−n ) by mapping the circular arc
from a point x on the equator S n−1 to the south pole to the corresponding arc from f (x) to the
south pole. Then by Remark 3.37 and Proposition 3.36 with K = f −1 (p), V = D̊+n and U = S n
we find
deg(F) = deg p (F) = deg p (F | D̊+n ) = deg p ( f | D̊+n ) . (3.7)
On the other hand, the commutative diagram
(D+n, S n−1 ) / (S n, D n ) o
− Sn
f F F
(D+n, S n−1 ) / (S n, D n ) o
− Sn
Hence deg(F) = deg( f ). This together with (3.7) concludes the proof.
Hn (S n ) / Hn (S n , S n \ ( f ◦ g) −1 (p)) Hn (g −1 (U ), g −1 (U ) \ ( f ◦ g) −1 (p))
103
3 Homology Theory
Example 3.42 U
Assume that f −1 (p) is a finite set, i.e.
b
deg p ( f ) = deg p ( f λ ) .
λ=1
Uλ
Example 3.43. Consider the map f k : S1 → S1 with f (z) = z k , k > 0, and set p = 1. We write
f k−1 (1) = {k-th unit roots} = {ξ1, . . . , ξ k }
and find that f k |small neighborhood of ξ λ is a homeomorphism. Hence deg( f k, λ ) = ±1. Since k > 0,
the restriction of f k to a small neighborhood of ξ λ is homotopic to an embedding of a (k times
larger) neighborhood of ξ λ into S1 . Hence deg( f k, λ ) = 1. We conclude deg1 ( f k ) = k.
104
3.5 Homological algebra
Proof of Proposition 3.41. Choose open neighborhoods Vλ such that f λ−1 (p) ⊂ Vλ ⊂ Uλ with
Vλ ∩ Vµ = ∅ for λ , µ. Now put V = ∪rλ=1 Vλ . Proposition 3.36 tells us deg p ( f ) = deg p ( f |V ).
The commutative diagram
deg p ( f |V )
H n ( f |V ) ,
Hn (S n ) / Hn (V, V \ f −1 (p)) / Hn (S n, S n \ {p}) o Hn (S n
O O O )
*.1+/
.. .. //
.. //
. (1, ...,1) (1, ...,1)
,1-
Lr Lr ⊕λ Hn ( f λ ) Lr Lr
λ=1 Hn (S n ) /
λ=1 Hn (Vλ, Vλ \ f λ−1 (p)) /
λ=1 Hn (S n, S n \ {p}) o 2 λ=1 Hn (S n )
*.deg p ( f 1 ) +/
.. .. //
.. . //
. /
, deg p ( f r ) -
yields
* +/ * . +
deg p ( f 1 ) 1
deg p ( f |V ) = (1, . . . , 1) · ... ..
. .
// · . .. // = deg p ( f 1 ) + . . . + deg p ( f r ).
, deg p ( f r ) - ,1-
∂n+1 ∂n
... / K n+1 / Kn / K n−1 / ...
∂n ◦ ∂n+1 = 0 (3.8)
Z n K∗ := {x ∈ K n | ∂n x = 0} = ker(∂n ) ,
105
3 Homology Theory
Sn (X, A; R) = Sn (X ; R)/Sn (A; R),
Kn =
n≥0
0,
n<0
Definition 3.46. Given two complexes K∗ and K∗′ a chain map ϕ∗ : K∗ → K∗′ is a sequence of
homomorphisms ϕ n : K n → K n′ such that
ϕ n+1
K n+1 / K′
n+1
∂n+1 ∂′n+1
ϕn
Kn / K′
n
that ϕ n (Z n K∗ ) ⊂ Z n K∗′ and also that ϕ n (Bn K∗ ) ⊂ Bn K∗′ . Hence ϕ∗ induces homomorphisms
Hn (ϕ∗ ) : Hn K∗ → Hn K∗′ by Hn (ϕ∗ )([z]) = [ϕ n z]. This construction is functorial in the sense
that
Hn (ϕ∗ ◦ ψ∗ ) = Hn (ϕ∗ ) ◦ Hn (ψ∗ ) and Hn (idK∗ ) = idH n K∗ .
ϕ∗ ψ∗
Definition 3.48. A sequence · · · −→ K∗′ −→ K∗ −→ K∗′′ −→ · · · of chain maps is called exact
ϕn ψn
iff the sequence · · · −→ K n′ −→ K n −→ K n′′ −→ · · · is exact for every n ∈ Z.
106
3.5 Homological algebra
i∗ p∗
Proposition 3.49. If 0 → K∗′ −→ K∗ −→ K∗′′ → 0 is an exact sequence of complexes then the
H n (i) H n (p)
sequence Hn K∗′ −→ Hn K∗ −→ Hn K∗′′ is exact for every n ∈ Z.
i∗ p∗
Definition 3.50. Let 0 → K∗′ −→ K∗ −→ K∗′′ → 0 be an exact sequence of complexes. We
construct the connecting homomorphism
as follows: Let z ′′ ∈ Z n K∗′′ represent an element [z ′′] ∈ Hn K∗′′. Since pn is surjective we can
choose x ∈ K n with pn x = z ′′. For ∂ x ∈ K n−1 we observe
pn−1 ∂ x = ∂ ′′ pn x = ∂ ′′ z ′′ = 0.
i n−2 ∂ ′ y ′ = ∂i n−1 y ′ = ∂∂ x = 0.
∂∗ [z ′′] := [y ′].
107
3 Homology Theory
Lemma 3.51. The conncecting homomorphism ∂∗ : Hn K∗′′ → Hn−1 K∗′ is well defined, i.e.,
independent of the choices made in its construction.
Proof. There are two choices in the construction of the connecting homomorphism: that of the
preimage x with pn x = z ′′ and that of the representing cycle z ′′ itself.
As to the choice of x let pn x = pn x̄ = z ′′. For the corresponding elements y ′, ȳ ′ ∈ K n−1
′ we
have and i n−1 y = ∂ x and i n−1 ȳ = ∂ x̄. Since pn (x − x̄) = 0 there exists an ω ∈ K n′ with
′ ′ ′
x − x̄ = i n (ω ′ ). We compute
i n−1 (y ′ − ȳ ′ ) = ∂(x − x̄) = ∂(i n ω ′ ) = i n−1 ∂ω ′ ,
from which we conclude that y − ȳ ′ = ∂ω ′ and therefore [y ′ ] = [ ȳ ′ ].
As to the choice of the representing cycle z ′′ it suffices to show that if z ′′ is a boundary then
[y ′ ] = 0. Let z ′′ = ∂ ′′ ζ ′′ be a boundary. Since pn+1 is onto we can choose ξ ∈ K n+1 with
pn+1 ξ = ζ. Then x = ∂ξ is an admissible choice because
pn x = pn ∂ξ = ∂ ′′ pn+1 ξ = ∂ ′′ ζ = z ′′ .
Thus ∂ x = ∂∂ξ = 0 and hence y ′ = 0.
It is easy to see that the connecting homomorphism is indeed a homomorphism. Now we are
ready to prove the Exactness Axiom.
i∗ p∗
Proposition 3.52. If 0 → K∗′ −→ K∗ −→ K∗′′ → 0 is an exact sequence of complexes then the
long homology sequence
∂∗ ∂∗
··· / Hn+1 K ′′ / Hn K ′ / Hn K∗ H n (p∗ )/ Hn K ′′
H n (i ∗ )
/ Hn−1 K ′ / ···
∗ ∗ ∗ ∗
is also exact.
Proof. In view of Proposition 3.49 it remains to show ker H (i) = im ∂∗ and ker ∂∗ = im H (p).
a) im ∂∗ ⊂ ker H (i):
Using the notation of Definition 3.50 we compute
H (i)∂∗ [z ′′] = H (i)∂∗ [px] = H (i)[i −1 ∂ x] = [∂ x] = 0 .
b) ker H (i) ⊂ im ∂∗ :
Let H (i)[z ′] = 0. Then we have [iz ′ ] = 0 and therefore iz ′ = ∂ x. Put z ′′ := px. Then
∂ ′′ z ′′ = ∂ ′′ px = p∂ x = piz ′ = 0. Hence z ′′ ∈ Z n K∗ represents an element in homology. We
compute
∂∗ [z ′′] = ∂∗ [px] = [i−1 ∂ x] = [z ′] .
Hence [z ′] ∈ im ∂∗ .
108
3.5 Homological algebra
c) im H (p) ⊂ ker ∂∗ :
This follows from
∂∗ H (p)[z] = ∂∗ [pz] = [i−1 |{z}
∂z ] = 0 .
=0
d) ker ∂∗ ⊂ im H (p):
Let ∂∗ [z ′′] = 0. We write z ′′ = px and compute
∂(x − ix ′ ) = ∂ x − i∂ ′ x ′ = 0 .
0 / K′ / Kn / K ′′ /0
n n
is again exact. By Proposition 3.52 we obtain the long exact homology sequence for a triple:
... /H (X, A)
❣ ❣❣❣ n+1
∂∗ ❣❣❣❣ ❣
❣❣
❣❣❣❣❣
s❣❣❣❣❣
Hn (A, B) / Hn (X, B) / Hn (X, A)
❣❣ ❣❣❣
∂∗ ❣❣❣❣❣
❣
❣❣❣ ❣❣❣❣❣
s❣❣
Hn−1 (A, B) / ...
109
3 Homology Theory
Proposition 3.55. The long homology sequence is natural, i.e., if the diagram of chain maps
i∗ p∗
0 / K′ / K∗ / K ′′ /0
∗ ∗
ϕ ∗′ ϕ∗ ϕ ∗′′
j∗ q∗
0 / L′ / L∗ / L ′′ /0
∗ ∗
H n+1 (p∗ ) ∂∗ ∂∗
... / Hn+1 K ′′ / Hn K ′ / Hn K∗ H n (p∗ )/ Hn K ′′
H n (i ∗ )
/ ...
∗ ∗ ∗
is commutative as well.
Hn (ϕ) ◦ Hn (i) = Hn ( j) ◦ Hn (ϕ ′ ),
Hn (ϕ ′′ ) ◦ Hn (p) = Hn (q) ◦ Hn (ϕ).
H n (ϕ ′′ ) H n−1 (ϕ ′ )
∂∗
Hn L ′′ / Hn L ′
commutes. We calculate
Example 3.56. Let f ∈ C ((X, A), (Y, B)). For K n′ = Sn (A), K n = Sn (X ), K n′′ = Sn (X, A)
and ϕ n′ = Sn ( f | A ), ϕ n = Sn ( f ), and ϕ n′′ the induced homomorphism on Sn (X, A) we recover
Lemma 3.13.
110
3.5 Homological algebra
such that
′
ϕ n − ψ n = ∂n+1 hn + hn−1 ∂n ,
for all n ∈ Z. The maps ϕ and ψ are called homotopic if such a homotopy exists. In this case
we write ϕ ≃ ψ.
Proposition 3.58. (i) The relation “≃” is an equivalence relation on the set of all chain
maps K → K ′.
ϕ ′ ϕ − ϕ ′ψ = ϕ ′ (∂ ′ h + h∂) = ∂ϕ ′ h + ϕ ′ h∂ ,
ϕ ′ψ − ψ ′ ψ = (∂ ′ h ′ + h ′ ∂)ψ = ∂ ′ h ′ψ + h ′ψ∂ ,
ϕ ′ ϕ ≃ ϕ ′ψ ≃ ψ ′ ψ .
111
3 Homology Theory
for j = 0, . . . , n by
(ek , 0),
Tn (ek ) =
j k ≤ j,
(ek−1, 1),
k > j.
In the case n = 0 we have
T00 : ∆1 → ∆0 × I = {e0 } × I ,
which can be visualized as
b
(e0, 0)
T00
b b b
(e0, 1)
e0 e1
112
3.6 Proof of the homotopy axiom
e2 (e0, 1) (e1, 1)
b b b
T10
b b b b
e0 e1 (e0, 0) (e1, 0)
e2 (e0, 1) (e1, 1)
b b b
T11
b b b b
e0 e1 (e0, 0) (e1, 0)
Lemma 3.62. The composition of the operators T and the affine linear map F (defined in (3.1)
on page 80) yields:
j+1 i j
Tn ◦ Fn+1 = (Fni × id) ◦ Tn−1 ( j ≥ i),
j i+1 j
Tn ◦ Fn+1 = (Fni × id) ◦ Tn−1 ( j < i),
Tni ◦ Fn+1
i
= Tni−1 ◦ i
Fn+1 (1 ≤ i ≤ n),
Tn0 ◦ Fn+1
0
= i 1,
Tnn ◦ Fn+1
n+1
= i 0,
If i ≤ k ≤ j then
Tnj+1 ◦ Fn+1
i j
(ek ) = Tnj+1 (ek+1 ) = (ek+1, 0) = (Fni × id)(ek , 0) = (Fni × id) ◦ Tn−1 (ek ).
If k > j then
j+1 i j+1 j
Tn ◦ Fn+1 (ek ) = Tn (ek+1 ) = (ek , 1) = (Fni × id)(ek−1, 1) = (Fni × id) ◦ Tn−1 (ek ).
The proofs of the other formulas are similar exercises in index shifting.
113
3 Homology Theory
j
Now let σ : ∆n → X be a singular n-simplex. Note that (σ × id) ◦ Tn ∈ C (∆n+1, X × I). We
define Pσ ∈ Sn+1 (X × I) by
n
X
Pσ := (−1) j (σ × id) ◦ Tnj .
j=0
P : Sn (X ) → Sn+1 (X × I).
Lemma 3.63. Let (X, A) be a pair of spaces. The operator P is a chain homotopy for
0≤ j <i ≤n
X
j+1
− (−1) i+ j+1 (σ × id) ◦ Tn i
◦ Fn+1 .
0≤i ≤ j ≤n−1
114
3.6 Proof of the homotopy axiom
j=0 i=0
X
j
= (−1) i+ j (σ × id) ◦ Tn ◦ Fn+1
i
(i < j)
0≤i< j ≤n
Xn
+ (σ × id) ◦ Tni ◦ Fn+1
i
(i = j)
i=0
n+1
X
− (σ × id) ◦ Tni−1 ◦ Fn+1
i
(i = j + 1)
i=1
X
+ (−1) i+ j (σ × id) ◦ Tnj ◦ Fn+1
i
(i > j + 1)
1≤ j+1<i ≤n+1
X ′ ′
= (−1) i+ j +1 (σ × id) ◦ Tnj +1 ◦ Fn+1
i
0≤i ≤ j ′ ≤n−1
× id) ◦ i1 − (σ × id) ◦ i0
+(σ X
′ j ′
+ (−1) i + j+1 (σ × id) ◦ Tn ◦ Fn+1
i +1
0≤ j <i ′ ≤n
In the last step we changed the summation indices to j ′ = j − 1 and i ′ = i − 1. We see that
P∂σ + ∂Pσ = (σ × id) ◦ i1 − (σ × id) ◦ i0
= i 1X ◦ σ − i 0X ◦ σ
= Sn (i 1X )σ − Sn (i 0X )σ
which proves the lemma.
Proof. Let F be a homotopy for f and g, i.e., F ∈ C ((X × I, A × I), (Y, B)) with
f = F ◦ i 1X , g = F ◦ i0X .
Then by (3.9) we get
S( f ) − S(g) = S(F)S(i 1X ) − S(F)S(i 0X ) = S(F)P∂ + S(F)∂P = S(F)P∂ + ∂S(F)P.
Hence S(F)P is a chain homotopy for S( f ) and S(g).
115
3 Homology Theory
Hn (X \ U, A \ U ) Hn (X, A)
provided Ū ⊂ Å.
X
A U
SnU (X ) := {σ ∈ Sn (X ) | σ is U -small}
= im * Sn (Ui ) −−−−−−−−−→ Sn (X ) +
M ⊕i S n ( j i )
, i ∈I -
with ji : Ui → X the inclusion map. For A ⊂ X we put
SnU (X )
SnU (X, A) := .
SnU (A)
Theorem 3.66 (Small chain theorem). The inclusion SnU (X, A) → Sn (X, A) induces an iso-
morphism in homology.
Before coming to the proof we show that the small chain theorem implies the excision axiom.
To this extent let (X, A) be a pair of spaces and let U ⊂ A be such that Ū ⊂ Å. Now set U1 := Å
116
3.7 Proof of the excision axiom
and U2 := X \ Ū. Then U = {U1, U2 } forms an open cover of the space X . We compute
SnU (X )
SnU (X, A) =
SnU (A)
Sn ( Å) + Sn (X \ Ū)
=
Sn ( Å) + Sn (A \ Ū)
Sn (X \ Ū )
=
(Sn ( Å) + Sn (A \ Ū)) ∩ Sn (X \ Ū)
Sn (X \ Ū)
= .
Sn (A \ Ū)
Similarly we get
Sn ( Å \ U ) + Sn (X \ Ū) Sn (X \ Ū)
SnU (X \ U, A \ U ) = = .
Sn ( Å \ U ) + Sn (A \ Ū) Sn (A \ Ū)
In the following commutative diagram all arrows are induced by inclusions.
U (X \ U, A \ U ) = / S U (X, A)
Sm n
Sm (X \ U, A \ U ) / Sn (X, A)
By the small chain theorem the vertical arrows induce isomorphisms on homology and the
excision theorem is proved.
It remains to prove Theorem 3.66. We define the barycenter of ∆n by
n
1 X
Bn := ej .
n + 1 j=0
B0 = e0,
1
B1 = (e0 + e1 ), B1
b
B2
2
1 e0 e1 e0 e1
B2 = (e0 + e1 + e2 ).
3
For an affine map σ : ∆n → ∆n+1 we define the affine map Cn σ : ∆n+1 → ∆n+1 by
Bn+1,
(Cn σ)(ek ) =
k = 0,
σ(ek−1 ),
k ≥ 1.
117
3 Homology Theory
e2
b
G
b
e0 e1
e0 e1 C1 G
e2
b
b b
e0 e1
Now we set
Proof. (i) It suffices to show the assertion for an affine simplex c. We then find
and hence
118
3.7 Proof of the excision axiom
as desired.
Lemma 3.70. To each topological space X and each n ∈ N0 we can associate homomorphisms
Sdn : Sn (X ) → Sn (X ),
Q n : Sn (X ) → Sn+1 (X ),
such that
(iii) Sd∗ and Q∗ are natural, i.e., for every f ∈ C (X, Y ) the following diagrams commute:
Sd n Qn
Sn (X ) / Sn (X ) Sn (X ) / Sn+1 (X )
Sn ( f ) Sn ( f ) Sn ( f ) S n+1 ( f )
Sd n Qn
Sn (Y ) / Sn (Y ) Sn (Y ) / Sn+1 (Y )
(iv) If the map σ : ∆n → ∆n is affine then each simplex σ j occuring in Sdn (σ) or in Q n (σ)
is affine and
n
diam(σ j ) ≤ diam(σ).
n+1
Proof. The construction of Sdn and Q n will be done recursively, the proof is by induction over
n. We start by considering the case n = 0. We put Sd0 := id : S0 (X ) → S0 (X ) and Q0 := 0. It
is obvious that the four assertions hold.
In the case n ≥ 1 we assume that Sdn−1 and Q n−1 are already defined and we set for a singular
simplex σ : ∆n → X :
119
3 Homology Theory
and
Q n (σ) = Sn+1 (σ)(Cn (id∆ n −Sdn (id∆ n ) − Q n−1 ∂ id∆ n ) ).
| {z }
∈S aff n
n (∆ )
| {z }
∈S aff
n+1
(∆ n )
Now we check (ii): Again we first consider the case X = ∆n and σ = id∆ n .
120
3.7 Proof of the excision axiom
p q
2. One vertex is Bn :
Bn
pi
Then we find
1 X
n
d(pi , Bn ) = pi −
n + 1 j=0
pj
1 X n
= (pi − p j )
n + 1 j=0
n
1 X
≤ |pi − p j |
n + 1 j=0 | {z }
≤diam(σ ) and =0 for j=i
n
≤ diam(σ) .
n+1
121
3 Homology Theory
Sd − Sd2 = Sd ◦ ∂ ◦ Q + Sd ◦ Q ◦ ∂
= ∂ ◦ Sd ◦ Q + Sd ◦ Q ◦ ∂.
We conclude that Sd2 ≃ Sd ≃ id. Iterating this procedure we find that Sdr ≃ id for all r ∈ N.
Hence there exist homomorphisms Q n(r ) : Sn (X ) → Sn+1 (X ) with
Lemma 3.72. For σ : ∆n → X continuous there exists a ε > 0 such that all ε-balls ∩∆n are
completely contained in σ −1 (Ui ) for some Ui ∈ U .
Proof. Assume the assertion were false. Then for ε k = 1/k there exists a point pk ∈ ∆n such
that
B 1 (pk ) = {x ∈ ∆n | |x − pk | < ε k }
k
is not contained in any of the σ −1 (Ui ). After passing to a subsequence we have that pk → p ∈ ∆n
by compactness of ∆n . Choose i0 with p ∈ σ −1 (Ui 0 ). Since σ −1 (Ui 0 ) is open there exists a δ > 0
such that Bδ (p) ⊂ σ −1 (Ui 0 ). Now choose k so large, that |pk − p| < δ/2 and ε k = k1 < δ/2. We
then find
B 1 (pk ) ⊂ Bδ (p) ⊂ σ −1 (Ui 0 ),
k
a contradiction.
Corollary 3.73. Assume that σ : ∆n → X is continuous. Then there exists an r (σ) ∈ N such
that every simplex σ j occuring in Sdr (σ) or in Q (r ) (σ) for r ≥ r (σ) is completely contained
in one of the Ui , i.e., Sdr (σ), Q (r ) (σ) ∈ SnU (X ).
122
3.8 The Mayer-Vietoris sequence
∂ Sdr x
|{z} = Sdr ∂ x
∈S U
n+1
(X ) for large r
= Sdr z
= z − ∂Q (r ) z − Q (r ) |{z}
∂z .
=0
Hence
z = ∂( Sdr x + Q (r ) z )
| {z }
∈S U
n (X ) for large r
b) Now we pass to general (X, A). Consider the commutative diagram with exact rows:
By part a) of the proof we know that the outer four arrows are isomorphisms. The Five
Lemma (Exercise 3.11) implies that the map HnU (X, A) → Hn (X, A) is also an isomorphism.
with the inclusion maps iν : U1 ∩ U2 → Uν and jν : Uν → X . We then get the following long
exact homology sequence:
H n (i 1 )
−H n (i 2 ) ∂
· · · → Hn (U1 ∩ U2 ) / Hn (U1 ) ⊕ Hn (U2 ) (H n ( j1 ), H n ( j2 )) / H U (X ) → Hn−1 (U1 ∩ U2 ) → · · ·
n
123
3 Homology Theory
By the small chain theorem HnU (X ) is canonically isomorphic to Hn (X ). Using this isomorphism
we can replace HnU (X ) in the above exact homology sequence by Hn (X ).
(H n ( j1 ), H n ( j2 )) ∂
· · · → Hn (U1 ) ⊕ H❱n (U2 ) / H U (X ) / Hn−1 (U1 ∩ U2 ) → · · ·
❱❱❱❱ n 4
❱❱❱❱ ✐✐✐✐
❱❱❱❱ ✐✐✐✐✐✐✐
(H n ( j1 ), H n ( j2 ))❱❱❱❱❱❱ ✐✐✐✐ ∂ M V
❱* ✐✐✐✐
Hn (X )
Theorem 3.74 (Mayer-Vietoris sequence). Let X be a topological space, let A ⊂ X and let
U1, U2 ⊂ X be open such that U1 ∪ U2 = X . Set Aν := A ∩ Uν and let
be the inclusion maps, ν = 1, 2. Then the following sequence is exact and natural
H n (i 1 )
∂MV / Hn (U1 ∩ U2, A1 ∩ A2 ) −H n (i 2 )
/ Hn (U1, A1 ) ⊕ Hn (U2, A2 )
···
❝ ❝ ❝❝ ❝ ❝ ❝❝❝❝❝
❝
❝❝❝❝❝❝❝❝
❝❝❝❝ ❝❝❝❝❝❝❝❝❝❝❝ (H n ( j1 ), H n ( j2 ))
q ❝❝❝❝❝❝
❝
Hn (X, A) M V / Hn−1 (U1 ∩ U2, A1 ∩ A2 ) / ···
∂
Example 3.75. We give a new computation of the homology of S1 . To this extent we cover the
circle as indicated in the picture.
U1 U2
124
3.8 The Mayer-Vietoris sequence
If n ≥ 2 then all homologies of the point occuring in this diagram vanish. Hence Hn (S1 ) = 0
for all n ≥ 2. Since S1 is path-connected we have that H0 (S1 ) R. In the case n = 1 we find
11
/ H1 (S 1 ) / R2 11 / R2
0
To compute H1 (S1 ) we calculate
−x
! ( ! )
1 1
x
1
H1 (S ) ker = x ∈ R R.
1 1
U1 U2
H0 ({p}) R / H0 (S 1 ) ⊕ H0 (S 1 ) R2
;
1
1
125
3 Homology Theory
Lemma 3.77. Let X be a topological space and let Ui ⊂ X be open with Ui ⊂ Ui+1 and
∪i ∈N Ui = X . Furthermore let ιn : Un ֒→ X and ιn, m : Un ֒→ Um for m ≥ n be the inclusion
maps. Then we have:
(i) For each α ∈ Hk (X ; R) there exists an n0 such that α ∈ im(Hk (ιn )) for all n ≥ n0 .
(ii) For each α n ∈ Hk (Un ; R) with Hk (ιn )(α n ) = 0 there exists an m0 such that
Hk (ιn, m )(α n ) = 0 for all m ≥ m0 .
P
(ii) Again represent α n ∈ Hk (Un ) by lj=1 α j σ j . From Hn (ιn )(α n ) = 0 we know that there
P
exists a β ∈ Ck+1 (X ; R) such that lj=1 α j σ j = ∂ β. As before there exists a compact
subset C ′ ⊂ X such that β ∈ Ck+1 (C ′; R). Since there exists an m0 with C ′ ⊂ Um for all
m ≥ m0 and thus β ∈ Ck+1 (Um ; R) we have that Hk (ιn, m )(α n ) = 0.
Proposition 3.78. Let n ∈ N and let Y be a compact topological space such that for any
embedding f : Y → S n
H∗ (S n \ f (Y ); R) H∗ (point; R) .
126
3.9 Generalized Jordan curve theorem
0 = Hi+1 (S n \ f ({ 12 } × Y )) / Hi (S n \ f ([0, 1] × Y ))
Hi (S n \ f ([0, 21 ] × Y )) ⊕ Hi (S n \ f ([ 21 , 1] × Y ))
Thus the inclusions S n \ f ([0, 1]×Y ) ֒→ S n \ f ([0, 12 ]×Y ) and S n \ f ([0, 1]×Y ) ֒→ S n \ f ([ 12 , 1]×Y )
induce an injective homomorphism
Hi (S n \ f ([0, 1] × Y )) → Hi (S n \ f ([0, 12 ] × Y ) ⊕ Hi (S n \ f ([ 21 , 1] × Y )) .
Hence
0 , Hi (inclusion)(α) ∈ Hi (S n \ f ([0, 21 ] × Y )) or
0 , Hi (inclusion)(α) ∈ Hi (S n \ f ([ 21 , 1] × Y )) .
By iterating this procedure we obtain a sequence of intervals Ik such that
I0 = [0, 1], Ik+1 ⊂ Ik , |Ik | = 2−k
and
0 , Hi (ι0, k )(α) ∈ Hi (S n \ f (Ik × Y ))
for all k. Here Vk := S n \ f (Ik × Y ) is open in S n and ιk,l : Vk ֒→ Vl for l ≥ k denotes the
inclusion map. We find
∩k ∈N Ik = {t}
∪k ∈N Vk = S n \ f ({t} × Y ) =: X
Now we apply the previous Lemma 3.77 for the inclusion ι : V0 ֒→ X . Hence, for i ≥ 1,
0 , Hi (ι)(α) ∈ Hi (X ) = Hi (S n \ f ({t} × Y ) ) = 0
| {z }
≈Y
giving a contradiction. The proof for i = 0 is similar (or in fact the same if one uses augmented
homology).
127
3 Homology Theory
H∗ (S n \ f (D r )) H∗ ({point})
S n \ f (D 0 ) ≈ Rn ≃ {point}
H∗ (S n \ f (Sr )) H∗ (S n−r −1 ).
For the induction step “r − 1 ⇒ r” we write Sr = D+r ∪ D−r . Then D+r ∩ D−r = Sr −1 . We put
U+ := S n \ f (D+r ) and U− := S n \ f (D−r ). Both sets U+, U− are open because f (D±r ) is compact.
In addition
U+ ∩ U− = S n \ f (Sr ) and U+ ∪ U− = S n \ f (Sr −1 ).
Now we look at the Mayer-Vietoris sequence and use Corollary 3.79:
H (U ) ⊕ H (U ) / Hi+1 (S n \ f (S r −1 )) / Hi (S n \ f (S r ))
| i+1 + {z i+1 −}
H i+1 (point) ⊕H i+1 (point)=0
H (U ) ⊕ H (U )
| i + {z i −}
H i (point)⊕ H i (point)=0
if i ≥1
128
3.9 Generalized Jordan curve theorem
Proof. a) We know that H0 (S n \ f (S n−1 )) H0 (S0 ) R2 , hence S n \ f (S n−1 ) has exactly two
path-components U and V .
b) Since f (S n−1 ) is compact the union U ∪ V = S n \ f (S n−1 ) is open and therefore both
connected components U and V are open.
c) Since U and V are open they contain no boundary point of U, thus ∂U ⊂ f (S n−1 ). Assume
that there exists p ∈ S n−1 with f (p) < ∂U. Then there exists W ⊂ S n open with f (p) ∈ W
and W ∩ U = ∅. Since f is continuous we can choose an open ball B ⊂ S n−1 with f (B) ⊂ W .
Since S n−1 \ B ≈ D n−1 we find that the space Y := S n \ f (S n−1 \ B) is path-connected because
of Corollary 3.79:
Moreover, we have
Y = U ∪ V ∪ f (B) ⊂ U ∪ V ∪ W and U ∩ (V ∪ W ) = ∅
and hence
Y = (Y ∩ U ) ⊔ (Y ∩ (V ∪ W ))
| {z } | {z }
U ⊃V
can be written as a disjoint union of open non-empty subsets. This contradicts Y being
path-connected.
Corollary 3.82 (Generalized Jordan curve theorem). For every embedding f : S n−1 → Rn
the complement Rn \ f (S n−1 ) consists of exactly two path components U and V . Both U and
V are open, U is bounded, V is unbounded and ∂U = ∂V = f (S n−1 ).
S n = Rn ∪ {∞}, f : S n−1 → Rn ⊂ S n .
129
3 Homology Theory
Remark 3.83. Consider an embedding f : S n−1 → S n . If i ≥ 1 we find for the homology of the
components of the complement
H (U ⊔ V ) = Hi (U ) ⊕ Hi (V ) Hi (S0 ) = 0
and hence U and V have the same homology as a point. This makes us suspect that U, V ≈ D̊ n .
For n = 2 this is indeed true, but it is false for n ≥ 3. The Alexander horned sphere is an
example of an embedding of S2 into S3 where one component of the complement is not even
simply connected:
The red circle in the picture is a non-contractible loop giving rise to a non-trivial element in
the fundamental group. A very nice video illustrating this embedding of S2 can be found at
http://www.youtube.com/watch?v=d1Vjsm9pQlc.
3.10 CW-complexes
We now describe a type of topological spaces for which there is a particularly efficient way to
compute their homology. These space are obtained by gluing together balls of various dimensions.
130
3.10 CW-complexes
ϕσ : D n → σ̄ ⊂ X
Definition 3.85. An element σ ∈ Xn is called an n-cell. The map ϕσ is called the characteris-
tic map of σ and X n is called the n-skeleton of (X, X). The map ϕσ S n−1=∂D n : S n−1 → X n−1
is called the attaching map of σ.
X2
b b
X0
X1
b b
X0 = {{e1 }},
Xn = {σ n = S n \ {e1 }},
Xm = ∅, otherwise,
131
3 Homology Theory
e1
Dn b
Sn
ϕσ n
The attaching map to the n-cell is the constant map S n−1 → {e1 }. We have
X 0 = X 1 = X 2 = . . . = X n−1 = {e1 },
S n = X n = X n+1 = . . .
Example 3.87. If a space X has the structure of a CW-complex there are in general many
different ways to write X as a CW-complex, i.e., there are many different X for the same X . Let
us look again at X = S n . We start with the case n = 0. Here the CW-structure is unique:
{{e1 }, {−e1 }},
Xm =
m=0
∅,
m > 0.
n−1
XmS ,
n n
m ≤ n−1
:=
n
XmS
{ D̊+ , D̊− }, m=n
∅,
m>n
Sn
◦+
Dn
S n−1
◦−
Dn
132
3.10 CW-complexes
X 0 = S0, X 1 = S1, . . . , X n = S n .
Example 3.88. Real projective space RPn . We define the real projective space as
where
x = (x 0, . . . , x n ) ∼ y = (y0, . . . , yn )
iff there exists a t , 0 such that x = t y. We consider the canonical projection map
Then
σ̄k = [x 0, . . . , x n ] ∈ RPn | x k+1 = . . . = x n = 0 ≈ RPk .
For the characteristic map ϕk : D k → σ̄k we can take
q
ξ 7→ [ξ1, . . . , ξ k , 1 − |ξ | 2, 0, . . . , 0] .
It is clear that the map ϕk is continuous. Now we check that it is also surjective. Let [x ′, x k , 0] ∈
σ̄k where x ′ = (x 0, . . . , x k−1 ). Without loss of generality we assume that x k ≥ 0. Set
′
ξ := √ ′ 2x 2
∈ D k . We compute
| x | + | xk |
" s #
x′ |x ′ | 2
ϕk (ξ) = p , 1− ′ 2 ,0
|x ′ | 2 + |x k | 2 |x | + |x k | 2
" s #
x′ |x k | 2
= p , ,0
|x ′ | 2 + |x k | 2 |x ′ | 2 + |x k | 2
= [x ′, |x k |, 0]
= [x ′, x k , 0] .
Next we show that ϕk | D̊ k is injective. Let ϕk (ξ) = ϕk (η) for ξ, η ∈ D̊ k . This leads to the two
equations: q q
ξ = tη and 1 − |ξ | 2 = t 1 − |η| 2
for some t , 0. Squaring and adding both equations we find
|ξ | 2 + 1 − |ξ | 2 = t 2 |η| 2 + t 2 (1 − |η| 2 )
133
3 Homology Theory
p
leading
p to t 2 = 1 and consequently t = ±1. Since |ξ |, |η| < 1 it follows that 1 − |ξ | 2 > 0 and
1 − |η| 2 > 0 and therefore t > 0. Hence t = 1 and thus ξ = η.
The map ϕk : D k → σ̄k is closed, therefore the restriction
ϕk D̊ k : D̊ → σk
X 0 = {point} ⊂ |{z}
X 1 ⊂ |{z}
X 2 ⊂ · · · ⊂ |{z}
Xn .
≈RP1 ≈RP2 =RP n
Finally let us discuss the gluing map. For ξ ∈ ∂D k = S k−1 , i.e. |ξ | = 1, the gluing map is given
by ϕk (ξ) = [ξ, 0, 0] and hence ϕk = ψ : S k−1 → X k−1 ≈ RPk−1 .
Example 3.89. Complex projective space CPn . For the complex projective space the discussion
is analogous to the real case with complex parameters instead of real parameters,
1,
|Xm | =
m even and 0 ≤ m ≤ 2n
0,
otherwise
and
X 0 = {point} = X 1 ⊂ |{z}
X 2 = X 3 ⊂ |{z}
X 4 = X 5 ⊂ · · · ⊂ X 2n−1 = |{z}
X 2n .
≈CP1 ≈CP2 =CP n
Example 3.90. Quaternionic projective space HPn . Similarly, for the quaternionic projective
space
1,
|Xm | =
m divisible by 4 and 0 ≤ m ≤ 4n,
0,
otherwise,
and
X 0 = {point} = X 1 = X 2 = X 3 ⊂ |{z}
X 4 = . . . ⊂ X 4n−3 = · · · = |{z}
X 4n .
≈HP1 =HP n
Remark 3.91. Every compact differentiable manifold can be triangulated and is consequently a
finite CW-complex.
134
3.11 Homology of CW-complexes
L ⊕σ∈Xn H i (ϕ σ )
Hi (D n, S n−1 ) / Hi (X n, X n−1 )
σ ∈Xn
is an isomorphism.
R | Xn |,
Hi (X n, X n−1 )
for i = n
0,
otherwise
Dn
We define a a a
Y n := D n, Y n−1 := S n−1, Ẏ n := Ḋ n
σ ∈Xn σ ∈Xn σ ∈Xn
135
3 Homology Theory
X2
b
X1
Now consider:
Φ:=∪σ∈Xn ϕ σ
(Y n, Y n−1 ) / (Y n, Ẏ n ) / (X n, Ẋ n ) o ? _ (X n, X n−1 )
= =
Hi (Ẏ n ) / Hi (Y n ) / Hi (Y n , Ẏ n ) / Hi−1 (Ẏ n ) / Hi−1 (Y n )
and the Five Lemma the map Hi (Y n, Y n−1 ) → Hi (Y n, Ẏ n ) is also an isomorphism. Similarly,
we get that the inclusion (X n, X n−1 ) ֒→ (X n, Ẋ n ) induces an isomorphism Hi (X n, X n−1 ) →
Hi (X n, Ẋ n ). The inclusions
(Y n \ Y n−1, Ẏ n \ Y n−1 ) ֒→ (Y n, Ẏ n ),
(X n \ X n−1, Ẋ n \ X n−1 ) ֒→ (X n, Ẋ n ),
Hence we find
M
Hi (D n, S n−1 ) Hi (Y n, Y n−1 )
σ ∈Xn
Hi (Y n, Ẏ n )
Hi (Y n \ Y n−1, Ẏ n \ Y n−1 )
Hi (X n \ X n−1, Ẋ n \ X n−1 )
Hi (X n, Ẋ n )
Hi (X n, X n−1 ) .
136
3.11 Homology of CW-complexes
Lemma 3.94. The sequence of groups K n (X, X) together with ∂n forms a complex.
The pair (K∗ (X, X), ∂∗ ) is called the cellular complex of (X, X).
∂
Hn (X n, X n−1 ) / Hn−1 (X n−1 )
= ∂n
*
∂
Hn−1 (X n−1 ) / Hn−1 (X n−1, X n−2 ) / Hn−1 (X n−2 )
2
=0 = ∂n−1
*
Hn−1 (X n−2 ) / Hn−2 (X n−2, X n−3 )
Proof. The proof is by induction on p−q. The assertion is certainly true for p−q = 0. To analyze
the situation p − q > 0 we look at the exact homology sequence for the triple (X p, X q+1, X q ):
ind. hyp.
Hn (X q+1, X q ) / Hn (X p, X q ) / Hn (X p, X q+1 ) = 0.
Proof. Lemma 3.95 with q = 0 says Hn (X p, X 0 ) = 0. The claim now follows from the exact
sequence
0 = Hn (X 0 ) −→ Hn (X p ) −→ Hn (X p, X 0 ) = 0.
137
3 Homology Theory
Hn (X ) Hn (X r ) .
Proof. The assertion follows from Corollary 3.97 and the exact sequence
Hn (X, X q ) Hn (X r , X q ) .
Proof. Since r ≥ n + 1, Corollary 3.97 gives us Hn+1 (X, X r ) = Hn (X, X r ) = 0. The assertion
now follows from the exact homology sequence of the triple (X, X r , X q ):
Hn K∗ (X, X) Hn (X )
138
3.11 Homology of CW-complexes
Proof. Consider the commutative diagram with exact columns and rows
Hn+1 (X n+1, X n
❘ ) H (X n−2 ) = 0
n−1
❘❘❘
❘❘❘∂n
∂ ❘❘❘
❘❘❘
)
i∗
0 = Hn (X n−1 ) / Hn (X ) n / Hn (X n, X n−1 )
❘❘❘
/ Hn−1 (X n−1 )
❘❘❘ ∂n−1
❘❘❘
❘❘❘
❘)
Hn (X n+1 ) Hn−1 (X n−1, X n−2 )
Hn (X n+1, X n ) = 0
Now we compute
Hn (X ) Hn (X n+1 )
Hn (X n )
∂Hn+1 (X n+1, X n )
i∗ Hn (X n )
i ∗ ∂Hn+1 (X n+1, X n )
ker(Hn (X n, X n−1 ) → Hn−1 (X n−1 ))
∂n Hn+1 (X n+1, X n )
ker(∂n−1 )
∂n Hn+1 (X n+1, X n )
= Hn K∗ (X, X)
R ←− 0 ←− · · · ←− 0 ←− R ←− 0 ←− · · ·
Since all arrows are 0 the homology coincides with the complex, hence
R, j = 0 or n,
H j (S n )
0, otherwise,
which confirms our earlier findings.
139
3 Homology Theory
Example 3.102. Now look at X = CPn . Then we have for the CW-decomposition from Exam-
ple 3.89
α 0 = α 2 = . . . = α 2n = 1, α j = 0 otherwise.
Again all arrows in the complex
R ←− 0 ←− · · · ←− 0 ←− R ←− 0 ←− · · ·
Example 3.103. Consider X = HPn . For the CW-decomposition from Example 3.90 we have
α 0 = α 4 = α 8 = . . . = α 4n = 1, α j = 0 otherwise
0 ←− R ←− 0 ←− 0 ←− 0 ←− R ←− · · · ←− 0 ←− R ←− 0 ←− · · ·
Now let (X, X) be a finite CW-complex. Denote the number of j-cells in (X, X) by α j . Then
we get the following estimate for the Betti numbers:
b j (X ; R) = dim R H j (X ; R)
ker(∂ j : K j (X, X) → K j−1 (X, X))
= dim R
im(∂ j+1 : K j+1 (X, X) → K j (X, X))
≤ dim R ker(∂ j : K j (X, X) → K j−1 (X, X))
≤ dim R K j (X, X)
= αj.
140
3.12 Betti numbers and the Euler number
Note that the sum in this definition is finite. It ends at the highest dimension occuring in the cell
decomposition.
Proposition 3.106. We have the following relation between Euler and Betti numbers:
∞
X
χ(X, X) = (−1) i bi (X ; R).
i=0
In particular, the Euler number does not depend on the cell decomposition because the Betti
numbers don’t. On the other hand, the Euler number does not depend on the coefficient field
because the α i ’s don’t. We will henceforth write χ(X ) instead of χ(X, X).
In order to prove the proposition we use the following
141
3 Homology Theory
X
= (−1) j dim H j V∗ .
j
Proof of Proposition 3.106. The proposition follows from Lemma 3.107 with Vj = K j (X, X)
Rα j .
α0 α1 α2
Tetrahedron 4 6 4
Cube 8 12 6
Octahedron 6 12 8
Dodecahedron 20 30 12
Icosahedron 12 30 20
142
3.13 Incidence numbers
Hence
1,
χ(RPn ) =
n even
0,
n odd
∂
H1 (X 1, X 0 ) / H0 (X 0 )
L L
H1 (ϕτ )(H1 (D 1, S0 )) R
τ ∈X1 σ ∈X0
Hence ∂1 is given by the (α 1 × α 0 )-matrix (∂στ ) where τ ∈ X1 and σ ∈ X0 . The entries ∂στ ∈ R
of this matrix are easily computed. Namely, recall that D 1 = [−1, 1] and that a generator of
H1 (D 1, S0 ) is represented by c : [0, 1] → [−1, 1], t 7→ 2t − 1. Then
Thus if ϕτ (−1) = ϕτ (1) then ∂στ = 0 for all σ. If ϕτ (−1) , ϕτ (1) then
1, for σ = ϕτ (1),
∂σ = −1,
τ
for σ = ϕτ (−1),
0,
otherwise.
Example 3.110. We compute the homology of the following CW-complex consisting of two
0-cells and three 1-cells:
τ2
τ1 σ1 b b σ2
τ3
Clearly, ∂στ11 = ∂στ12 = 0. Depending on how the characteristic maps parametrize the 1-cells τ2
and τ3 we get
∂στ21 = ∂στ31 = 1 and ∂στ22 = ∂στ32 = −1,
143
3 Homology Theory
or possibly different signs which will not affect the homology however. Thus the cellular complex
is
*.0 1 1+
/
o o 2 ,o 0 −1 −1- 3 o
··· 0 R R 0o ···
The image of the matrix is {(x, −x) | x ∈ R} = R · (1, −1) and hence
H0 (X ; R) R2/R · (1, −1) R
and !
0 1 1
H1 (X ; R) ker = {(x, y, −y) | x, y ∈ R} R2 .
0 −1 −1
L L
Hn+1 (ϕτ )(Hn+1 (D n+1, S n )) Hn (ϕσ )(Hn (D n, S n−1 ))
τ ∈Xn+1 σ ∈Xn
τ)
(∂σ
Rα n+1 / Rα n
Hence ∂n+1 is given by the (α n+1 × α n )-matrix (∂στ ) where τ ∈ Xn+1 and σ ∈ Xn . The entries
(∂στ ) ∈ R of this matrix are called the incidence numbers. We want to see how we can compute
them.
Fix σ ∈ Xn , τ ∈ Xn+1 , and p ∈ D̊ n . From the commutative diagram
∂n+1
+ prσ
∂
Hn+1 (X n+1 n
O ,X )
/ Hn (X n )
O
/ Hn (X n, X n−1 ) / Hn (ϕσ )Hn (D n, S n−1 )
O
H n+1 (ϕ τ ) H n (ϕ τ | S n )
−1 ◦ϕ :ϕ −1 (σ )→D n ⊂S n )
deg p (ϕ σ
∂ τ τ
Hn+1 (D n+1, S n )
/ Hn (S n ) / Hn (S n ) Hn (D n, S n−1 )
Hn (S n , S n \ ϕτ−1 (ϕσ (p)))
O
H n (ϕ τ )
Hn (σ, σ \ {ϕσ (p)})
−1 )
H n (ϕ σ
Hn ( D̊ n, D̊ n \ {p}) / Hn (D n, D n \ {p})
144
3.13 Incidence numbers
we conclude that
∂στ = deg p (ϕσ
−1
◦ ϕτ : ϕτ−1 (σ) → D n ⊂ S n ).
We used the canonical isomorphism Hn (S n ) Hn (D n, S n−1 ). We have intepreted the incidence
numbers as certain local mapping degrees. In applications they can often be computed by
counting preimages.
145
3 Homology Theory
Proposition 3.112. Let (X, X) be a finite CW-complex and let x 0 ∈ X 0 . Then the inclusion
map j : X 2 ֒→ X induces an isomorphism
j# : π1 (X 2 ; x 0 ) → π1 (X ; x 0 ) .
Proof. We have to show that attaching a k−cell for k ≥ 3 does not alter π1 in the sense that the
inclusion induces an isomorphism on π1 . We assume that X 2 is path-connected, since otherwise
we may replace X 2 by the path-component that contains x 0 .
Let Y be a path-connected finite CW-complex and let Ỹ be obtained from Y by attaching a k-cell.
More precisely, Ỹ = Y ∪ϕ D k = Y ⊔ D k / ∼ where x ∼ ϕ(x) for all x ∈ S k−1 . Here ϕ : S k−1 → Y
is a continuous map. We have to show that the inclusion map induces an isomorphism
j# : π1 (Y ; x 0 ) → π(Ỹ ; x 0 )
U1 = D̊ k , U2 = Ỹ \ D k ( 21 ) ≃ Y
Y U2
146
3.14 Homotopy versus homology
π1 (U1 ) = {1},
π1 (U2 ) π1 (Y ),
π1 (U1 ∩ U2 ) π1 (S k−1 ) = {1}, (here we use k − 1 ≥ 2)
and we find
π1 (U1 ) ⋆ π1 (U2 )
π1 (Ỹ ) π1 (Y )
im π1 (U1 ∩ U2 )
the isomorphisms being induced by inclusions.
Example 3.113. Consider complex-projective space X = CPn . We use the cell decomposition
from Example 3.89:
X 0 = {point} = X 1 ⊂ |{z}
X 2 = X 3 ⊂ |{z}
X 4 = X 5 ⊂ . . . ⊂ X 2n−1 ⊂ |{z}
X 2n
≈CP1 ≈CP2 =CP n
Example 3.114. Similarly, for X = HPn we use the cell decomposition from Example 3.90:
X 0 = {point} = X 1 = X 2 = X 3 ⊂ |{z}
X 4 = . . . ⊂ X 4n−3 = . . . ⊂ |{z}
X 4n
≈HP1 =HP n
Remark 3.115. Proposition 3.112 can be generalized as follows: For a finite CW-complex the
inclusion map j : X n+1 ֒→ X always induces an isomorphism j# : πn (X n+1 ; x 0 ) → πn (X ; x 0 )
where x 0 ∈ X 0 .
Now we relate homotopy and homology groups. Recall that for the n-dimensional cube
W n = [0, 1]n we have (W n, ∂W n ) ≈ (D n, S n−1 ). Fix a generator α n ∈ Hn (W n, ∂W n ; Z)
Hn (D n, S n−1 ; Z) Z. The elements of πn (X ; x 0 ) are homotopy classes relative to ∂W n of
maps f : W n → X with f (∂W n ) = {x 0 }. Then Hn ( f )(α n ) ∈ Hn (X, {x 0 }; Z). The long exact
homology sequence of the pair (X, {x 0 }) yields for n ≥ 1
147
3 Homology Theory
Namely, if n ≥ 2 then Hn−1 ({x 0 }; Z) = 0. For n = 1 the arrow emanating from H0 ({x 0 }; Z)
is injective so that the incoming arrow must again be zero. In either case the inclusion
j : (X, ∅) ֒→ (X, {x 0 }) induces an isomorphism Hn ( j) : Hn (X ; Z) −→ Hn (X, {x 0 }; Z). Now set
h([ f ]) := Hn ( j) −1 Hn ( f )(α n ) ∈ Hn (X ; Z) .
Due to homotopy invariance the expression h([ f ]) only depends on the homotopy class of the
map f . Hence we have constructed a well-defined map
h : πn (X ; x 0 ) → Hn (X ; Z), n ≥ 1.
(x 1, . . . , x n ) 7→ (2x 1, x 2, . . . , x n )
(x 1, . . . , x n ) 7→ (2x 1 − 1, x 2, . . . , x n ) .
Wn W1n W2n W n
s1
s2
f 1 (s1 (x)),
g(x) =
x 1 ≤ 12 ,
f 2 (s2 (x)),
x 1 ≥ 12 .
148
3.14 Homotopy versus homology
n=1 n=2
σ1 σ3 c1 = σ1 + σ2
c2 = σ3 + σ4
σ2 σ4
c1 c2
h([ f 1 ] · [ f 2 ]) = h([g])
= Hn ( j) −1 Hn (g)(α n )
= Hn ( j) −1 Hn (g)([c1 + c2 ])
= Hn ( j) −1 Hn (g)([c1 ]) + Hn ( j) −1 Hn (g)([c2 ])
= Hn ( j) −1 Hn ( f 1 ◦ s1 )([c1 ]) + Hn ( j) −1 Hn ( f 2 ◦ s2 )([c2 ])
= Hn ( j) −1 Hn ( f 1 )Hn (s1 )([c1 ]) + Hn ( j) −1 Hn ( f 2 )Hn (s2 )([c1 ])
= Hn ( j) −1 Hn ( f 1 )(α n ) + Hn ( j) −1 Hn ( f 2 )(α n )
= h( f 1 ) + h( f 2 ).
Proposition 3.118. The Hurewicz homomorphism h is natural, i.e., for every f ∈ C (X, Y )
with f (x 0 ) = y0 the following diagram commutes:
h / Hn (X ; Z)
πn (X ; x 0 )
f# Hn ( f )
h
πn (Y ; y0 ) / Hn (Y ; Z)
j X : (X, ∅) ֒→ (X, {x 0 })
jY : (Y, ∅) ֒→ (Y, {y0 })
149
3 Homology Theory
satisfy
Hn ( jY ) ◦ Hn ( f ) = Hn ( f ) ◦ Hn ( j X )
and consequently
Hn ( f ) ◦ Hn ( j X ) −1 = Hn ( jY ) −1 ◦ Hn ( f ) .
Thus
as claimed.
π0 (X ; x 0 ) = π1 (X ; x 0 ) = . . . = πn−1 (X ; x 0 ) = {1} .
We then have
H1 (X ; Z) = . . . = Hn−1 (X ; Z) = 0
and the map
h : πn (X ; Z) → Hn (X ; Z)
is an isomorphism.
Remark 3.120. For n = 1 this theorem cannot be true as it stands because H1 (X ; Z) is always
abelian while π1 (X ; x 0 ) is not in general. However, h induces a homomorphism
Already Poincaré showed that if π0 (X, x 0 ) = {1} (i.e., X is path-connected) then the map h̄ is an
isomorphism.
150
3.14 Homotopy versus homology
Example 3.122. We know that X = CPn is 1-connected. With the help of the Hurewicz
isomorphism we calculate
π2 (CPn ) H2 (CPn ; Z) Z .
Example 3.123. We also know that X = HPn is 1-connected. We apply the Hurewicz isomor-
phism three times and we get
π2 (HPn ) H2 (HPn ; Z) = 0,
π3 (HPn ) H3 (HPn ; Z) = 0,
π4 (HPn ) H4 (HPn ; Z) Z.
Example 3.124. Now let us analyze the space X = RPn for n ≥ 2. The map ψ : S n → RPn is a
two-fold covering and by Theorem 2.102
ψ♯
π1 (S n ) −−→ π1 (RPn ) → π0 ({p, q}) → π0 (S n )
Remark 3.125. Under the assumptions of the theorem of Hurewicz 3.119 not much can be
said about h : πk (X, x 0 ) → Hk (X ; Z) for k > n. For example, consider the Hopf fibration
S3 → S2 with fiber S1 . By Theorem 2.102 it induces an isomorphism π3 (S2 ) π3 (S3 ) Z.
But H3 (S2 ; Z) = 0 and hence h : π3 (S2 ) → H3 (S2 ; Z) is not injective.
On the other hand, for the 2-torus T 2 we have again by Corollary 2.105 π2 (T 2 ) π2 (R2 ) = 0
while one can compute H2 (T 2 ; Z) Z. Thus h : π2 (T 2 ) → H2 (T 2 ; Z) is not surjective.
Remark 3.126. We have seen that Hk (S n ; Z) = 0 whenever k > n. But in general this is not
true for the higher homotopy groups of the sphere, e.g. π3 (S2 ) Z. The computation of πk (S n )
for k > n is a difficult problem and many of these groups are not known to date.
151
3 Homology Theory
3.15 Exercises
3.1. Let X be a topological space. Show:
a) If X is path-connected then
H0 (X ; R) R
3.2. Let Yk = {1, ..., k} be equipped with the discrete topology. Compute Hn (Yk ; R) without
using Exercise 3.1 directly with the help of the Eilenberg-Steenrod axioms.
Hint: Consider the pair (Yk , Yk−1 ).
∂ # : S0 (X ; R) → R,
X X
∂# α i σi = αi,
where σi ∈ C (∆0, X ).
a) Verify ∂ # ◦ ∂ = 0.
ker(∂ # : S0 (X ; R) → R)
H0# (X ; R) :=
im(∂ : S1 (X ; R) → S0 (X ; R))
for X = {point}.
3.6. Let p be a complex polynomial without zeros on the unit circle S1 ⊂ C. Show: The degree
of the map
p(z)
p̂ : S1 → S1, p̂(z) = ,
|p(z)|
coincides with the number of zeros of p in the interior of the unit disk (counted with multiplicities).
152
3.15 Exercises
3.7 (Homotopy invariance of the local mapping degree). Let V ⊂ S n be open and F : V ×
S
[0, 1] → S n continuous. We put f t (x) := F (x, t). Let p ∈ S n such that t ∈[0,1] f t−1 (p) is
compact. Show:
deg p ( f 1 ) = deg p ( f 0 ).
3.8. Let (X, A) be a pair such that A is closed and a strong deformation retract of an open
neighborhood U. Show that Hn (X, A) = Hn (X/A) for n , 0.
3.9. Let Z = S1 × [0, 1] be the cylinder. Compute Hn (Z, S1 × {0} ∪ S1 × {1}) for all n. Sketch
generators of the nontrivial homology groups.
Hint: Use the homology sequence of the triple
3.10. Let (X, A) be a pair. Describe the 0th singular relative homology group H0 (X, A).
3.11 (Five lemma). Let the rows in the following commutative diagram of abelian groups be
exact:
f1 f2 f3 f4
A1 / A2 / A3 / A4 / A5
ϕ1 ϕ2 ϕ3 ϕ4 ϕ5
g1 g2 g3 g4
B1 / B2 / B3 / B4 / B5
3.12. Suppose
f n+1 fn
· · · → G n+1 −−−−−−−−−−→ G n −−−−−−−−→ G n−1 → · · ·
is a long exact sequence of abelian groups and
f′ f′
′ n+1 ′ −−−−−n−−−→ ′
··· → G n+1 −−−−−−−−−−→ G n G n−1 → ···
is a subsequence, i.e., G n′ ⊂ G n and f n′ = f n |G′n . Prove that the subsequence is exact if and only
if the quotient sequence
′
· · · → G n+1 /G n+1 → G n /G n′ → G n−1 /G n−1
′
→ ···
is exact.
153
3 Homology Theory
3.14. Let f ∈ C (S n, S n ) with f (D+n ) ⊂ D+n and f (D−n ) ⊂ D−n . We identify S n−1 with D+n ∩ D−n
and therefore have f (S n−1 ) ⊂ S n−1 . Show
3.15. Show that H1 (R, Q; Z) is a free abelian group and find a basis as a Z-module.
3.16. Let M be an n-dimensional manifold, n ≥ 3. Let p ∈ M. Show that the inclusion map
M \ {p} ֒→ M induces an isomorphism
H j (M \ {p}; R) H j (M; R)
b) Let p1, ..., pn ∈ R2 be pairwise distinct. Compute H1 (R2 \ {p1, ..., pn }) and sketch generators.
Sketch generators.
3.21. Show:
3.22. Let n ≥ 2 and k ≥ 1. Let X be the topological space obtained from k copies of S n by
identifying them all at one point. More formally,
k
[
X= { j} × S n ∼
j=1
154
3.15 Exercises
3.23. Find a CW-decomposition of the 2-torus with exactly one 0-cell, two 1-cells, and one
2-cell and use it to compute the homology.
155
Sources list for pictures
titlepage http://www.sxc.hu
p. 4 taken from [3]
p. 5 based on http://commons.wikimedia.org/wiki/File:Hilbert_curve.png
p. 6, 2. and 4. picture http://www.sxc.hu
p. 33 based on http://www.sxc.hu
p. 130 http://www.math.ohio-state.edu/∼fiedorow/
157
Bibliography
[1] Greenberg, M. J.; Harper, J. R.: Algebraic topology. A first course, Benjamin/Cummings
Publishing, 1981
[3] Hilbert, D.: Ueber die stetige Abbildung einer Line auf ein Flächenstück, Math. Ann. 38
(1891), 459–460
[4] Lück, W.: Algebraische Topologie. Homologie und Mannigfaltigkeiten, Vieweg, 2004
[6] Peano, G.: Sur une courbe, qui remplit toute une aire plane, Math. Ann. 36 (1890), 157–160
[8] Warner, F. W.: Foundations of differentiable manifolds and Lie groups, Springer-Verlag,
New York, 1983
159
Index
161
Index
B E
162
Index
163
Index
164