MATH 3491 GEOMETRIC TOPOLOGY III

RAPHAEL ZENTNER

RAPHAEL ZENTNER, PARTIALLY BUILT ON EARLIER NOTES BY JOHN HUNTON

1. Introduction: Some notations and conventions, and motivational

problems
1.1. Notations and some reminders from Topology II. To begin with we
have the real line, R, and the plane R2 . We’ll write Rn for n-dimensional euclidean
space, which can also be considered as the topological space of n-vectors (a1 , . . . , an ),
ai ∈ R. In all these cases the standard topology we’ll assume will be the euclidean
one – the topology that comes from the euclidean metric: in R the distance from x
to y is 󳈌x − y󳈌, and in Rn the distance is given by
󳇀
󳇂n
d(x, y) = 󳇂
󳇁󱮦(xi − yi )2 .
i=1

Subspaces. We have some standard subspaces of these (with the subspace topol-
ogy) such as
The closed unit interval I = {x ∈ R 󳈌 0 ⩽ x ⩽ 1}

The closed n-disc Dn = {x ∈ Rn 󳈌 󳈌 x󳈌󳈌 ⩽ 1}

The open n-disc E n = {x ∈ Rn 󳈌 󳈌 x󳈌󳈌 < 1}

The n-sphere S n = 󳆟x ∈ Rn+1 󳈌 󳈌 x󳈌󳈌 = 1󳆣 .

Note, for example, with this notation, the circle S 1 is the set of points in the
plane of unit distance from the origin, and S 2 , the sphere, is the surface consisting
of all points in R3 of unit distance from the origin.
Product spaces. The product construction gives us further basic examples such
as
The torus T = S 1 × S 1 = 󳆟(x, y) 󳈌 x ∈ S 1 , y ∈ S 1 󳆣

The cylinder = S 1 × I, etc.

Quotient spaces. We may also consider the torus T as a space constructed out of
the square I × I by identifying the points on the top edge with the corresponding
points on the bottom edges, and likewise with the left and right sides; we indicate
Date: Michaelmas 2025.
2 RAPHAEL ZENTNER, PARTIALLY BUILT ON EARLIER NOTES BY JOHN HUNTON

this as in the diagram with arrows showing the identifications, including the direction
of identification:

Figure 1. Construction of the torus by identifying sides of a

square. The blue dotted lines on the right show the image of the
glued-up edges.

We can formally record this as the quotient space I × I󳆋 ∼ where the equivalence
relation ∼ is given by
(x, 0) ∼ (x, 1) and (0, y) ∼ (1, y) for all x ∈ I and y ∈ I .
Note that the four corners all get identified to a single point.
This space can also be thought of as a quotient of the plane R2 using the equiv-
alence relation
(x, y) ∼ (x, y + 1) and (x, y) ∼ (x + 1, y) for all x ∈ R and y ∈ R .

Here is a variation on this idea, the Klein bottle K, also a quotient of the square
I × I, but now with one of the identifications running the opposite way.

Figure 2. Construction of the Klein bottle by identifying sides of

a square.

This is given by I × I󳆋∼ using the equivalence relation

(x, 0) ∼ (x, 1) and (0, y) ∼ (1, 1 − y) for all x ∈ I and y ∈ I .

The sphere S 2 , the torus T and the Klein bottle K are examples of closed sur-
faces: compact, connected spaces where every point has an open neighbourhood
homeomorphic to an open disc in R2 . Here are some more: if we consider the torus
T as having ‘one hole’ in it, we can imagine an ‘n-holed torus Mn ’.
So, in our notation M1 is another name for T. Question: can we write Mn as a
simpler space ‘glued up,’ as we described T above?
 MATH 3491 GEOMETRIC TOPOLOGY III RAPHAEL ZENTNER 3

Figure 3. The 4-holed torus M4 .

We shall also meet some 1 dimensional topological spaces. These will be better
known as graphs, which perhaps you met in the first year module Discrete Maths.
In this context, a graph can be thought of as a set of points (the vertices) and a set
of curves (the edges), running between some of the vertices. We can think of this
as a quotient space: we have taken a bunch of intervals and glued them together at
various of their endpoints. At this stage we’ll make no assumption that the number
of vertices or edges are finite or infinite, whether there are any loops, or multiple
edges between two vertices, and so on. Here are some examples.

Figure 4. Two examples of graphs.

Maps. What role does the topology in a topological space play? During Topology II
you will have thought a lot about aspects of the topology on a space, but in short it
is there so as to be able to make sense of a function between two topological spaces
being continuous. Recall that having a topology, for example one given by a metric,
you have a notion of open set. A function f ∶ X → Y between topological spaces is
continuous if
f −1 (any open set in Y ) is an open set in X .
4 RAPHAEL ZENTNER, PARTIALLY BUILT ON EARLIER NOTES BY JOHN HUNTON

During this term, almost all the functions between topological spaces we will ever
see will be continuous, and for shortness of terminology I’ll use the word map as
shorthand for continuous function. This is important terminology; let me repeat it:
Definition 1.1. Throughout this term, by a map we always mean a continuous
function.

Homeomorphisms. The next important concept after continuity is that of home-

omorphism.
Definition 1.2. Two topological spaces X and Y are said to be homeomorphic if
there are maps f ∶ X → Y and g∶ Y → X which are inverses of each other, that is, the
composites satisfy f ○ g = idY and g ○ f = idX where idX , etc, means the identity
function on X. If X and Y are homeomorphic, we write X 󰑔 Y .
At the start of this section we saw three very useful ways to ‘make’ a new topolog-
ical space out of old ones: subspaces, product spaces and quotient spaces. In each
case the new space received its topology from the old ones: though they will not play
a major role this term, check that you know what the respective subspace, product
and quotient topologies are so as to make sense of these as topological spaces.
Path connectedness. In topology II you met the notion of a space being con-
nected. While this is an interesting and somewhat subtle concept, we shall really
only be interested in this module with the perhaps more intuitive notion of path
connectedness. Recall that
Definition 1.3. The space X is path connected if given any two points x0 , x1 ∈ X
there is a map p∶ I → X (i.e., from the unit interval I = [0, 1]) to X with p(0) = x0
and p(1) = x1 .

Compactness. The notion of a space being compact is an important one, though

at first sight can be obscured by the apparent abstraction of the definition. In fact,
almost all the spaces we shall meet this term (excepting, obviously, cases such as Rn )
will be compact. Recall that a closed, bounded subspace of Rn is always compact,
that any closed subspace of a compact spaces is compact as is any finite product of
compact spaces, or a quotient of a compact space. Together they just about cover
everything we will meet.
What will be more important to us will be the consequences of a space being
compact. I shall recall these as and when explicitly needed. As an intuitive rule of
thumb, one of the crucial properties of a space being compact is that the functions
on it are not ‘too crazy’, which is perhaps the main reason for us for making sure
we stay in the world of compact spaces.
Hausdorffness. All the spaces we shall meet this term will be Hausdorff, and this
is not a concept we need to think about here. (Note though that carelessly taken
quotients can produce non-Hausdorff spaces, but we shall avoid those.)
 MATH 3491 GEOMETRIC TOPOLOGY III RAPHAEL ZENTNER 5

Some symbols. In the context of topological spaces we will write X 󰑔 Y to indicate

that X and Y are homeomorphic, that is, there exists a continuous bijective map
f ∶ X → Y whose inverse map is also continuous.
In the context of groups (abelian or non-abelian) we will write G 󰑔 H to indicate
that G and H are isomorphic, that is, there exists a bijective group homomorphism
f ∶ G → H.

2. Some motivational problems

We will discuss some problems informally to motivate the further exposition in
the course, and in particular the introduction of the fundamental group and what
it can do for us for improving our understanding of topology1.

2.1. Proving spaces to be homeomorphic. As above, we consider a quotient

space obtained as follows. We consider the equivalence relation on the square X =
[0, 1] × [0, 1] given by
(x, 0) ∼ (x, 1) for all x ∈ [0, 1] and
(0, y) ∼ (1, y) for all y ∈ [0, 1].
For all other points the equivalence class consists of a point only. We consider the
quotient space T = X󳆋 ∼. This is illustrated by the left picture in Figure 1 above
which shows the identifications we perform on the edges of a square.
We claim that this quotient space T is homeomorphic to S 1 × S 1 . But how do we
prove this? The obvious choice here is construct a map from one space to the other
and prove it is a homeomorphism, either by proving the existence of a continuous
inverse, or by some knowledge from Topology II. For instance, the following Lemma
will often prove useful:
Lemma 2.1. Let X be a compact topological space, and let Y be a Hausdorff-space.
Suppose f ∶ X → Y is a bijective continuous map. Then f is a homeomorphism.
Without the assumptions made the theorem is clearly wrong. For instance f ∶ [0, 1) →
S 1 given by f (t) = exp(i2πt) is clearly a continuous bijective map, but it is not a
homeomorphism.

2.2. Proving spaces not to be homeomorphic. A way of proving that no home-

omorphism f ∶ X → Y between topological spaces X and Y exists consists in showing
that one space does have a certain property that the other one doesn’t, but a hy-
pothetical homeomorphism would show that both spaces to have said property. We
provide some simple examples.

1No theorem stated in this introduction will be examinable material unless we will state and
prove it later on in the course. Also, you should not use these results in assignments as long as we
don’t prove the results later on in the course. In particular, we will probably not prove the Jordan
curve separation theorem.
6 RAPHAEL ZENTNER, PARTIALLY BUILT ON EARLIER NOTES BY JOHN HUNTON

Example 2.2. The closed interval [0, 1] and the open interval (0, 1) are not homeo-
morphic. In fact, one is compact, the other one isn’t. Another argument is this: We
can remove a point from [0, 1] (namely one of the two boundary points), and the
resulting space still is path-connected. The open interval (0, 1) does not have this
property for any of its points, but path-connectedness or compactness are clearly
properties preserved by a homeomorphism.
Example 2.3. The closed interval [0, 1] and the circle S 1 are not homeomorphic.
In fact, removing any point from S 1 results in a space that still is path-connected,
but [0, 1] does not have this property, by the intermediate value theorem.
Properties such as connectedness, compactness, and Hausdorffness are relatively
coarse topological properties that aren’t too refined for showing that two spaces are
not homeomorphic.
For instance, both S 1 ×S 1 and S 2 are compact topological Hausdorff spaces which
stay compact when removing a point, so similar arguments as above do not work.
In this course we will associate the fundamental group2 π1 (X) of a (path-
connected) topological space X. It will have two key properties:
(1) A continuous map f ∶ X → Y induces a homomorphism of groups f∗ ∶ π1 (X) →
π1 (Y ). If g∶ Y → Z is a further continuous maps, then we have three conin-
uous maps, namely f, g, and g ○ f , and consequently three group homomor-
phisms f∗ , g∗ , and (g ○ f )∗ . Then the following property holds:
g∗ ○ f∗ = (g ○ f )∗ (2.1)
(2) For any space X the identity homeomorphism of a space X induces the
identity homomorphism of the group π1 (X),
(idX )∗ = idπ1 (X) . (2.2)
From these properties it easily follows that a homeomorphism f ∶ X → Y induces an
isomorphism of groups f∗ ∶ π1 (X) → π1 (Y ). In fact, let g∶ Y → X be a continuous
inverse. Then by the above properties we have
(2.1) (2.2)
g∗ ○ f∗ = (g ○ f )∗ = (idX )∗ = idπ1 (X) ,
And therefore g∗ ○ f∗ = idπ1 (X) . Similarly one proves f∗ ○ g∗ = idπ1 (Y ) , and therefore
f∗ is a group isomorphism.
We will make the computations π1 (S 1 ×S 1 ) 󰑔 Z2 later and show that π1 (S 2 ) = {e},
the trivial group. Since the first is a non-trivial group, these two spaces therefore
cannot be homeomorphic.

We remark en passant that there is a Theorem that allows to argue with connect-
edness after removal of a particular set:

2We sacrifice mathematical precision for more clarity in this introductory section by ignoring
the dependence of a base point
 MATH 3491 GEOMETRIC TOPOLOGY III RAPHAEL ZENTNER 7

Theorem 2.4. (Jordan curve separation theorem) The image f (S 1 ) of any injective
continuous map f ∶ S 1 → S 2 separates S 2 into two different path-connected compo-
nents.
Remark 2.5. It’s unclear whether we will get to the point to proving this theorem
in Geometric Topology III, in fact most likely not.
Now for a point z0 ∈ S 1 the set {z0 }×S 1 is the image of an injective continuous map
f ∶ S 1 → S 1 × S 1 , but the complement S 1 × S 1 ∖ f (S 1 ) is path-connected. Therefore,
by the Jordan curve separation theorem S 1 × S 1 and S 2 are not homeomorphic.
However, the proof of the Jordan curve separation theorem is much more difficult
than introducing the fundamental group and the mentioned computations for S 1 ×S 1
and S 2 .
2.3. Extending continuous maps from the circle to the disc. We recall that
S 1 = ∂D2 , the boundary of the 2-dimensional disc.
Question 2.6. Can the identity map idS 1 ∶ S 1 → S 1 be extended to a continuous
map f ∶ D2 → S 1 ? In other words, does there exist a continuous map f ∶ D2 → S 1
such that the restriction f 󳈌∂D2 satisfies
f 󳈌∂D2 = id∂D2 ? (2.3)
We will see later in the course that no such map exists. The ingredients of the
proof will be the computations π1 (S 1 ) 󰑔 Z, whereas π1 (D2 ) = {e}, together with the
properties (2.1) and (2.2). In fact, suppose that such a map f exists. We denote by
i∶ S 1 → D2 the inclusion map of the boundary. Then the condition (2.3) is equivalent
to the statement
f ○ i = idS 1 .
Passing to the fundamental group and using (2.1) and (2.2), we obtain the equation
(2.1) (2.2)
f∗ ○ i∗ = (f ○ i)∗ = (idS 1 )∗ = idZ ∶ Z → Z (2.4)
But the map f∗ ∶ π1 (D2 ) → π1 (S 1 ) 󰑔 Z is defined on π1 (D2 ) = {e}, and hence the
image of f∗ can consists of one point only. This contradicts f∗ ○ i∗ = idZ . Therefore
no such map f can exist.
We will see that a corollary of this result will be the Brouwer fixed point theorem:
Theorem 2.7. (Brouwer fixed point theorem) Let f ∶ D2 → D2 be a continuous map.
Then f has a fixed point. This means there exists a point z ∈ D2 such that f (z) = z.
A similar result is for instance not true for the circle S 1 or the real line R. (Why?)
2.4. The earth, temperature, and barometric pressure. The Borsuk-Ulam
theorem. We regard the surface of the earth as the 2-dimensional sphere S 2 . This is
not quite correct, but a good model since tunnels etc. are extremely small compared
to the order of magnitude of the earth. We consider a continuous function f ∶ S 2 → R
which we think of, for instance, the temperature on the earth at ground level. To
each point x ∈ S 2 = ∂(B 3 ) ⊆ R3 there is precisely one point opposite of it on the
8 RAPHAEL ZENTNER, PARTIALLY BUILT ON EARLIER NOTES BY JOHN HUNTON

earth, the point −x. For instance, if x denotes Durham, then −x is somewhere near
the Antipodes Islands in the Pacific Ocean, southeast of New Zealand.
We claim that there exists a pair of antipodal points x, −x ⊆ S 2 such that f (x) =
f (−x), that is at both points there is exactly the same temperature. To prove this,
consider the north pole N = (0, 0, 1) and the south pole S = (0, 0, −1) of S 2 . If
f (N ) = f (S) we are done. If not, let γ∶ [0, 1] → S 2 be a continuous map such that
γ(0) = N and γ(1) = S. In other words, γ is a path that starts at the north pole
and ends at the south pole. We consider the function

h∶ [0, 1] → R
t ↦ f (γ(t)) − f (−γ(t)).
This function measures the difference of the temperature at γ(t) and its antipodal
point −γ(t). Suppose that f (N ) > f (S). Then
h(0) = f (N ) − f (−N ) = f (N ) − f (S) > 0.
On the other hand we have
h(1) = f (S) − f (−S) = f (S) − f (N ) < 0.
Since h is a continuous function (as a composition of continuous maps), the inter-
mediate value theorem implies that there exists some t0 ∈ [0, 1] such that h(t0 ) = 0.
But then f (γ(t0 )) = f (−γ(t0 )), and the existence of antipodal points with the de-
sired property is proved.

Now supose that g∶ S 2 → R is an additional continuous function, and we may

think of it as barometric pressure at ground level on the earth. It is natural to ask
the following question:
Question 2.8. Does there exist a pair of antipodal points {x, −x} ⊆ S 2 such that
both equations
f (x) = f (−x)
g(x) = g(−x)
hold? In other words, does there exist a pair of antipodal points on the earth with
identical temperature and barometric pressure at ground level?
It is surprising that the answer still is yes. But a proof as simple as above for
the case of one function is not known, and probably doesn’t exist. It is however a
consequence of the following theorem, known as the Borsuk-Ulam theorem.
Theorem 2.9. (Borsuk-Ulam theorem) Let F ∶ S 2 → R2 be a continuous map. Then
there exists a pair of antipodal points {x, −x} ⊆ S 2 such that F (x) = F (−x).
We will be able to prove this theorem later on in Geometric Topology III. There
exists a higher dimensional analogue of this theorem, but it requires the techniques
of Algebraic Topology IV to prove it.
 MATH 3491 GEOMETRIC TOPOLOGY III RAPHAEL ZENTNER 9

2.5. Telling knots apart: A short outlook to the material in Epiphany

term. A knot is (the image of) an injective continuous map k∶ S 1 → R3 . The
following picture shows three different knots.

Figure 5. In what sense is the knot K2 essentially different from

K0 and K1 ?

In what sense are K0 and K1 “equal”, and in what sense does K2 appear to be
“different” from K0 and K1 ? It turns out that one can deform K1 into K0 through
a one-parameter family of maps all of which are injective. In particular, one can
pass from K0 to K1 without passing through double points. This appears intuitively
clear.
However, how can one show that one cannot deform K0 into K2 without cre-
ating self-intersections during the deformation? This will also be done using the
fundamental group. In fact, we will prove that π1 (R3 ∖ K2 ) is a non-abelian group,
whereas we will see that π1 (R3 ∖ K0 ) and π1 (R3 ∖ K1 ) are both isomorphic to Z,
which is abelian.

3. Elements of homotopy Theory

The idea of homotopy theory is to make rigorous the traditional notion of topology
as ‘rubber sheet geometry’: that we count two topological spaces as equivalent if one
can be deformed into the other in a controlled manner (without cutting, or joining).
To do this properly we have to start by defining a related notion of equivalence
between maps and spaces.
Definition 3.1. Two maps f, g∶ X → Y are homotopic, written f ≃ g, if there is a
continuous map
H∶ X × [0, 1] → Y
with
H(x, 0) = f (x) for all x ∈ X
H(x, 1) = g(x) for all x ∈ X .

We call such a map a homotopy from f to g, and we write f ≃ g to specify that

H

f and g are homotopic via the homotopy H. By defining ht (x) as H(x, t), we can
think of a homotopy as a set of maps ht ∶ X → Y for all t ∈ [0, 1] that continuously
varies from f at t = 0 to g at t = 1.
10 RAPHAEL ZENTNER, PARTIALLY BUILT ON EARLIER NOTES BY JOHN HUNTON

Example 3.2. Consider the circle S 1 as the unit circle in C, i.e., the points of the
form e2iπs for s ∈ [0, 1). Let a∶ S 1 → S 1 be given by a(e2iπs ) = −e2iπs . (This is
sometimes called the antipodal map.) Let id∶ S 1 → S 1 be the identity, id(e2iπs ) =
e2iπs . Then a ≃ id by the homotopy H∶ S 1 × I → S 1 given by
H(e2iπs , t) = eiπ(2s+t) .
(Check this works, i.e., that it takes the right values at t = 0 and t = 1. Remember
that eiπ = −1!)
Lemma 3.3. Homotopy is an equivalence relation between maps.
Proof. We must check ≃ is reflexive, symmetric and transitive. First note that
f ≃ f for any map f ∶ X → Y by the homotopy R∶ X × I → Y given by R(x, t) = f (x).
To see symmetry, if f ≃ g by a homotopy S∶ X × I → Y , then g ≃ f by the homotopy
T given by T (x, t) = S(x, 1 − t).
Transitivity is a little more involved. Suppose f ≃ g by a homotopy A∶ X × I → Y ,
and g ≃ h by some homotopy B, then f ≃ h by a homotopy C given by
for 0 ⩽ t ⩽ 12
C(x, t) = 󳆢
A(x, 2t)
B(x, 2t − 1) for 12 ⩽ t ⩽ 1 .
C is well defined and continuous because A1 = B0 , and it clearly verifies the required
properties. □
Lemma 3.4. If f, g∶ X → Y are homotopic by a homotopy H, and we have further
maps h∶ W → X and k∶ Y → Z, then the composites k○f ○h and k○g○h are homotopic
as maps W → Z.

Proof. If f ≃ g, then k ○ f ○ h ≃ k ○ g ○ h, where we used the notation

H k○H○(id,h)

k ○ H ○ (id, h) for the homotopy

H∶ [0, 1] × W → Z
(t, w) ↦ k(H(t, h(w))).
Definition 3.5. We say that a map f ∶ X → Y is null -homotopic if it is homotopic
to a constant map, i.e., to a map c∶ X → Y with c(x) = y0 for all x ∈ X and some
fixed y0 ∈ Y .
Example 3.6. The identity map id∶ D2 → D2 is null-homotopic. We show this by
giving a homotopy between id and the constant map c which sends all points to the
centre of the disc, that is, c(x) = 0. Here’s the homotopy: check it has the properties
needed, that it is id at one end and c at the other.
H(x, t) = tx (i.e., scalar multiplicaion of the vector x by t.)
Now we can turn to the induced relation on spaces.
Definition 3.7. We call two spaces X and Y are homotopy equivalent, written
X ≃ Y , if there exist maps λ∶ X → Y and µ∶ Y → X whose composites satisfy
µ ○ λ ≃ idX and λ ○ µ ≃ idY .
 MATH 3491 GEOMETRIC TOPOLOGY III RAPHAEL ZENTNER 11

Example 3.8. Let us write p for the one point space. Then D2 ≃ p. To prove this
let us define λ∶ D2 → p the only way we can (that is, λ(x) = p), and µ∶ p → D2 by
µ(p) = 0. Then clearly λ ○ µ∶ p → p is the identity map on p (and so is homotopic to
the identity as every map is homotopic to itself, as we saw in the lemma above). The
other composite, µ ○ λ∶ D2 → D2 sends each point x to 0. We must show µ ○ λ ≃ idD2 ;
actually, we did exactly this in the last example, using the homotopy
H(x, t) = tx .
Lemma 3.9. Homotopy equivalence is an equivalence relation between spaces.
Proof. Exercise.
Definition 3.10. We say a space X is contractible if it is homotopy equivalent to
the one point space p.
Example 3.11. Let X be a space. Define CX, the cone on X, as
CX = X × [0, 1]󳆋∼
where ∼ is a relation that identifies all points of the form (x, 0) with each other, i.e.,
it collapses the 0-end of the cylinder X × [0, 1] to a single point.

Exercise show that D2 󰑔 CS 1 , and in general Dn 󰑔 CS n−1 . Show further that, for
any space X, the cone CX is contractible.
Lemma 3.12. A contractible space is path connected.
Proof. Let X be contractible and suppose a, b are any two points in X. We will
construct a path from a to b. As X is contractible, there are maps X 󲿋→ p 󲿋→ X
f g

with gf ≃ idX by some homotopy H∶ X × I → X, i.e., a map with H(x, 0) = x and

H(x, 1) = gf (x) = g(p). Denote g(p) by c. Then H(a, t), as t runs over the interval
I, is a path from a to c, and H(b, 1 − t), for t ∈ I, is a path from c to b. Running
along the first of these followed by the second is a path from a to b. □
Lemma 3.13. If two spaces are homeomorphic then they are homotopy equivalent,
but the converse does not hold in general.
Proof. Suppose X 󰑔 Y by maps f ∶ X → Y and g∶ Y → X with gf = idX and
f g = idY . Then as any map is homotopic to itself (as seen in the proof of Lemma
3.3), we get that gf ≃ idX and f g ≃ idY .
Example 3.8 showed that D2 ≃ p, but clearly D2 is not homeomorphic to the one
point space, so this shows the converse cannot hold. □
12 RAPHAEL ZENTNER, PARTIALLY BUILT ON EARLIER NOTES BY JOHN HUNTON

Basepoints. In quite a lot of what we do later in this chapter it will be useful for
all spaces to have a particular distinguished base point. This means that each space
X has some particular point, say x0 , and we demand that all maps and homotopies
between spaces take base points to base points. Let us write such a space (call it a
based space) as the pair (X, x0 ), then a based map f ∶ (X, x0 ) → (Y, y0 ) must not only
be a map X → Y in the usual sense, but also satisfy f (x0 ) = y0 , and similarly a based
homotopy H∶ (X, x0 ) × [0, 1] → (Y, y0 ) between the based maps f ∶ (X, x0 ) → (Y, y0 )
and g∶ (X, x0 ) → (Y, y0 ) must satisfy H(x0 , t) = y0 for all t ∈ I as well as the usual
conditions that H(x, 0) = f (x) and H(x, 1) = g(x).
It may be shown that all the results on ≃ being an equivalence relation are also
true in this based version.
Retractions and deformation retractions.
Definition 3.14. Let X be a topological space and A ⊆ X be a subspace, and let
i∶ A → X denote the inclusion map. A map r∶ X → A is called a retraction map, or
retraction if
r ○ i = idA .
We then also call the subset A a retract of X.
Example 3.15. We consider the closed unit ball Dn for n ≥ 1 of which S n−1 ⊆ Dn
denotes the (n − 1)-dimensional sphere. Then the map
r∶ Dn ∖ {0} → S n−1
x↦
x (3.1)
󳈒x󳈒
is a retraction map.
Example 3.16. Keeping well track of the identification, the following sequence of
pictures explains why this is referred to as a bottle:

Figure 6. A sequence of deformations of the Klein bottle

The drawing suggests a self-intersection in this picture in R3 , but in the definition

of the Klein bottle there is no such identification. The reason for the picture with this
 MATH 3491 GEOMETRIC TOPOLOGY III RAPHAEL ZENTNER 13

self-intersection is that it is impossible to find a continuous injective map K → R3

from the Klein bottle into R3 . This is in fact a theorem, but we will not develop the
tools in this course to prove it. (Clearly, there exists an injective continuous map
K → R4 by just using the extra dimension near the intersection in R3 .)
Example 3.17. Consider the subset A = [0, 1] × { 12 } ⊆ X of the square, and denote
by A = π(A) its image in the Klein bottle K = X󳆋 ∼, with π∶ X → K denoting the
quotient map. Then π(A) is homeomorphic to a circle S 1 .

Figure 7. A retract of the Klein bottle

We consider the map

r̃∶ [0, 1] × [0, 1] → [0, 1] × 󳆡 󳆦

1
2 (3.2)
(x, y) ↦ (x, 1󳆋2).
This maps equivalent points to identical points in the image, and therefore defines
a unique continuous map r∶ K → A such that r ○ π = r̃. It is easy to check that the
map r defines a retraction from the Klein bottle K onto the circle A.
That the definition of a retraction map can be a slightly delicate issue is illustrated
by the following example.
Example 3.18. Consider the 2-dimensional real projective space RP2 defined by
the quotient space RP2 = X󳆋 ∼ of the square X = [0, 1] × [0, 1] by the equivalence
relation:
(x, 0) ∼ (1 − x, 1) for all x ∈ [0, 1] and
(0, y) ∼ (1, 1 − y) for all y ∈ [0, 1].
This contains a circle A by the same expression as in the previous example.
However, the map r̃ defined by (3.2) above does not descent to a quotient map
r∶ RP2 → A. More so, we will see later that RP2 does not admit a circle as a retract.
Definition 3.19. Let X be a topological space and A ⊆ X be a subspace, and
let i∶ A → X denote the inclusion map. A map r∶ X → A is called a deformation
retraction map, or deformation retraction if
r ○ i = idA (that is r is a retraction map), and additionally
i ○ r ≃ idX .
We then also call the subset A a deformation retract of X.
14 RAPHAEL ZENTNER, PARTIALLY BUILT ON EARLIER NOTES BY JOHN HUNTON

Example 3.20. The map r defined in Example 3.15 above is a deformation retrac-
tion. To see this, we have to see that the composite i ○ r is homotopic to the identity
on the punctured ball Dn ∖ {0}, that is i ○ r ≃ id. Consider
H∶ [0, 1] × Dn ∖ {0} → Dn ∖ {0}

(t, x) ↦ x + t 󳆚 − x󳆞
x
󳈒x󳈒
Then this is a well-defined homotopy (check that the expression stays within Dn ∖
{0}!) from H0 = idDn ∖{0} to H1 = r.

4. The Fundamental Group

The idea of this section is to show how to every path connected space X with
a base point x0 ∈ X we can associate a group, written π1 (X, x0 ), the fundamental
group3 of X, that has lots of nice properties. These will include the fact that if if two
spaces are homotopy equivalent then their fundamental groups are isomorphic, that
a map f ∶ X → Y will give us a a group homomorphism f∗ ∶ π1 (X, x0 ) → π1 (Y, y0 ),
and that for based maps f, g∶ (X, x0 ) → (Y, y0 ) if f ≃ g through a base-point pre-
serving homotopy then f∗ and g∗ are the same homomorphism. Philosophically,
we are translating topological information about spaces into algebraic information,
which is usually much easier to manipulate.

We begin with all our spaces having base points, as at the end of the last section.
The circle S 1 , thought of as the unit circle in C, will be given the base point 1 ∈ C.
Definition 4.1. A path in the topological space X is a continuous map λ∶ [0, 1] → X.
A loop is a path λ such that λ(0) = λ(1). A loop in the based space (X, x0 ) (or a
loop based in x0 ) is a path λ such that λ(0) = λ(1) = x0 .
We will tend to use the second way of thinking about loops, that of a path
parameterised by the unit interval.
Definition 4.2. Two paths λ and µ are paths with λ(0) = µ(0) and λ(1) = µ(1)
(that is their starting point and end points are equal) are called path-homotopic if
there exists a continuous map
H∶ [0, 1] × [0, 1] → X
satisfying
H(0, t) = λ(t) for all t ∈ [0, 1] ,
H(1, t) = µ(t) for all t ∈ [0, 1] ,
H(s, 0) = λ(0) = µ(0) for all s ∈ [0, 1] and
H(s, 1) = λ(1) = µ(1)) for all s ∈ [0, 1].

3As the notation hints, there are other groups π (X) for n > 1, but we won’t look at them in
n
this module.
 MATH 3491 GEOMETRIC TOPOLOGY III RAPHAEL ZENTNER 15

We write λ ≃ µ if that is the case. This corresponds to the idea that λ can
H

be continuously deformed into µ while keeping endpoints fixed, see the following
figures:

Figure 8. The case of two paths α and β with the same endpoints.
The homotopy Hs = H(s, .) is a family of paths deforming continu-
ously between α and β while the endpoints are kept fixed.

Figure 9. The case of loops: The blue loop λ is deformed to the

red loop µ by moving the points along the green arrows.

It is very useful to think of H diagrammatically as a map defined on a square (see

Figure ??).
Definition 4.3. (Fundamental group) Let (X, x0 ) be a based topological space.
For a loop λ we write
[λ] = {µ 󳈍 λ ≃ µ}
for the set of paths in X based at x0 up to path-homotopy. The fundamental group
of (X, x0 ), denoted π1 (X, x0 ), is the set of path-homotopy classes of loops in (X, x0 ).
In other words,
π1 (X, x0 ) = {[λ] 󳈌 λ is a loop based at x0 }.
So far this is only a definition of a set, not a group. We must now define the
group operations and check they really do satisfy the axioms of a group.
16 RAPHAEL ZENTNER, PARTIALLY BUILT ON EARLIER NOTES BY JOHN HUNTON

Figure 10. The homotopy is a map defined on whole of the unit

square with the properties that one side maps to the loop λ, the
opposite side to µ and the remaining two sides to the basepoint x0 .
If we consider instead paths α and β with the same startpoint and
endpoint which aren’t necessarily equal, then the left vertical side is
mapped to x0 = α(0) = β(0) and the right vertical side is mapped to
x1 = α(1) = β(1).

Definition 4.4. If λ1 and λ2 are two paths, we define their composite λ1 ∗ λ2 by

λ1 (2t) 0 ⩽ t ⩽ 12
λ1 ∗ λ2 ∶ [0, 1] → X t↦󳆢
λ2 (2t − 1) 12 ⩽ t ⩽ 1
where

This is the path which does λ1 at double speed for the first half of the interval, then
λ2 at double speed for the second half.
Definition 4.5. If λ∶ [0, 1] → X is a path, then we define by λ− the path given by
λ− (t) = λ(1 − t).
In other words, the inverse path λ− has the same image as λ, but it is run in the
opposite direction, starting at the endpoint of λ and ending at it starting point.
Definition 4.6. Let x0 ∈ X be a point. Then we denote by x0 the constant path
equal to x0 . In other words, x0 (t) = x0 for all t ∈ [0, 1].
We are now ready to state the main result of this subsection.
Theorem 4.7. Let (X, x0 ) be a pointed topological space. The expression
[λ1 ] ⋅ [λ2 ] = [λ1 ∗ λ2 ]
defines a well-defined binary operation turning the set π1 (X, x0 ) into a group with
neutral element x0 and inverse [λ]−1 = [λ− ].
Proof. To show this is well defined, we must prove that if λ1 ≃ µ1 and λ2 ≃ µ2 then
λ1 ∗ λ2 ≃ µ1 ∗ µ2 . So, suppose λ1 ≃ µ1 by a homotopy H1 ∶ I 2 → X, and λ2 ≃ µ2 by a
homotopy H2 ∶ I 2 → X, shown diagramatically by
then λ1 ∗ λ2 ≃ µ1 ∗ µ2 by specifically, λ1 ∗ λ2 ≃ µ1 ∗ µ2 by a homotopy H∶ I 2 → X
defined as
H (2s, t) 0 ⩽ s ⩽ 12
H(s, t) = 󳆢 1 ◻
H2 (2s − 1, t) 12 ⩽ s ⩽ 1 .
 MATH 3491 GEOMETRIC TOPOLOGY III RAPHAEL ZENTNER 17

So we now have π1 (X, x0 ) is a set with a binary operation. We must show that
this operation is associative, has an identity element and has inverses.
Associativity. We need a homotopy between (λ ∗ µ) ∗ κ and λ ∗ (µ ∗ κ). It is
quite possible to write down the homotopy explicitly, but Section 4 explains best
what happens: We choose a 1-parameter family of re-parametrisations of the path
(λ ∗ µ) ∗ κ to λ ∗ (µ ∗ κ). The images are equal in both cases, but the paths λ and
κ are run through in different speed in the two cases.

Figure 11. Homotopy between (λ ∗ µ) ∗ κ at bottom and

λ ∗ (µ ∗ κ) at top. At any intermediate time the path described
does λ at speed somewhere between twice and four times usual, and
correspondingly adjusts the speed of κ to accomplish the composite
path in unit time. At all times µ is traversed at four times usual
speed.

Identity element. The identity is the class of the constant loop c∶ [0, 1] → X given
by c(t) = x0 . Figure 15 presents the diagrams showing it has the right properly, that
[c] ⋅ [λ] = [λ] = [λ] ⋅ [c].
18 RAPHAEL ZENTNER, PARTIALLY BUILT ON EARLIER NOTES BY JOHN HUNTON

Figure 12. Homotopies between λ ∗ c and λ (left), and between

c ∗ λ and λ (right). The points in the shaded black triangles all map
to the basepoint. At any intermediate time the path described by a
horizontal line though either of the squares stretches λ accordingly.

Inverses. The inverse λ−1 to the loop λ is the loop sending t ↦ λ(1 − t). Check
that λ−1 ∗ λ ≃ c ≃ λ ∗ λ−1 to establish it really does give the inverse for an element
[λ] in π1 (X, x0 ). (What do the corresponding square diagrams look like?) □

For the rest of this section we will establish the following.

󲽨 For a based map f ∶ (X, x0 ) → (Y, y0 ) we get a group homomorphism
f∗ ∶ π1 (X, x0 ) → π1 (Y, y0 )
and this homomorphism depends on only on the based homotopy equivalence
class of f (the homotopies are required to be base point preserving);

󲽨 If (X, x0 ) ≃ (Y, y0 ) then π1 (X, x0 ) 󰑔 π1 (Y, y0 );

󲽨 If X is path connected then π1 (X, x0 ) is independent of the choice of base

point x0 , and so we can speak of the fundamental group of X, just π1 (X).
Definition 4.8. Suppose f ∶ (X, x0 ) → (Y, y0 ) is a based map. Then define the
function f∗ ∶ π1 (X, x0 ) → π1 (Y, y0 ) by
f∗ ([λ]) = [f ○ λ] .
Lemma 4.9. The function f∗ is well defined and is a group homomorphism.
Proof. To show that it is well defined, we must show that it did not depend on
the choice of representative of the equivalence class [λ], that is, we must show that
if λ1 ≃ λ2 , then the composites f ○ λ1 and f ○ λ2 are also homotopic. However, if
H∶ I × I → X is a homotopy from λ1 to λ2 , then the composite f ○ H is a homotopy
from f λ1 to f λ2 .
To show that f∗ is a group homomorphism, we compute
f∗ ([λ] ⋅ [µ]) = f∗ ([λ ∗ µ])
= [f ○ (λ ∗ µ)]
(3)
= [(f ○ λ) ∗ (f ○ µ)]
= (f∗ ([λ]) ⋅ (f∗ ([µ])) .
 MATH 3491 GEOMETRIC TOPOLOGY III RAPHAEL ZENTNER 19

Here the third equality (3) is easily checked from the definition. Thus we have shown
that f∗ is indeed a group homomorphism. □
We will now define a version of homotopy between maps of pointed spaces.
Definition 4.10. Let A ⊆ X and B ⊆ Y be subspaces of topological spaces X
and Y , respectively. Then a map f ∶ X → Y is said to be a map of pairs, written
f ∶ (X, A) → (Y, B) if f (A) ⊆ B.
With this notion, a map f ∶ (X, x0 ) → (Y, y0 ) is a map of pairs if f (x0 ) = y0 . This
is the same as a based map above.
Definition 4.11. We say that two maps f, g∶ (X, x0 ) → (Y, y0 ) are based-point pre-
serving homotopic or homotopic as based map, if there exists a homotopy H∶ [0, 1] ×
X → Y which is a map of pairs H∶ ([0, 1] × X, [0, 1] × {x0 }) → (Y, y0 ). In other words
H(s, x0 ) = y0 for all s ∈ [0, 1].
Lemma 4.12. If f ≃ g∶ (X, x0 ) → (Y, y0 ) are homotopic as based maps, then f∗ = g∗
as group homomorphisms.
Proof. If f ≃ g by a homotopy H∶ X × I → Y , and λ is a based loop in X, then
f ○ λ and g ○ λ are path homotopic as loops in Y by the homotopy K∶ I × I → Y given
by
K(s, t) = H(λ(s), t)
(check this homotopy does what it should). Then
f∗ ([λ]) = [f ○ λ] = [g ○ λ] = g∗ ([λ])
and so f∗ and g∗ are the same functions. □
The following result is rather trivial to prove, but extremely useful, and we will
use it again many times in the course in applications.
Lemma 4.13. The operation of passing from a map f to a homomorphism f∗
preserves composites and identities. This means, if we have based maps

(X, x0 ) 󲿋→ (Y, y0 ) 󲿋→ (Z, z0 )

f g

and id∶ (X, x0 ) → (X, x0 ) denotes the identity map, then

(g ○ f )∗ = g∗ ○ f∗ ∶ π1 (X, x0 ) → π1 (Z, z0 )
and (idX )∗ ∶ π1 (X, x0 ) → π1 (X, x0 ) is the identity homomorphism,
(idX )∗ = idπ1 (X,x0 ) .
Proof. These follow straight from Definition 4.8. Suppose λ is a based loop in X.
Then
(g ○ f )∗ ([λ]) = [(g ○ f ) ○ λ] = [g ○ (f ○ λ)] = g∗ ([f ○ λ] = (g∗ ○ f∗ )([λ]),
and
(id)∗ ([λ]) = [id ○ λ] = [λ]
hold for all [λ] ∈ π1 (X, x0 ). □
20 RAPHAEL ZENTNER, PARTIALLY BUILT ON EARLIER NOTES BY JOHN HUNTON

Corollary 4.14. Suppose the based spaces (X, x0 ) and (Y, y0 ) are homotopy equiv-
alent through base point preserving homotopies. Then their fundamental groups are
isomorphic. More precisely, if f ∶ (X, x0 ) → (Y, y0 ) and g∶ (Y, y0 ) → (X, x0 ) are
pointed maps which are base-point preserving homotopic, then we have
g∗ ○ f∗ = idπ1 (X,x0 ) and
f∗ ○ g∗ = idπ1 (Y,y0 ).
Proof. By assumption g ○ f ≃ idX and f ○ g ≃ idY through based point preserving
homotopies. Then Lemma 4.12 implies that (g ○ f )∗ = (idX )∗ . By the functoriality
property Lemma 4.13 we conclude that
(g ○ f )∗ = g∗ ○ f∗ = idπ1 (X,x0 ) ,
and the other case is proved in the same way. □
The fundamental group of the circle, and covering spaces I. We will give
here only a statement with no proof which will be given and completed in the section
of covering spaces I. In the lecture we sketched a proof relying on future results.
Theorem 4.15. The fundamental group π1 (S 1 , 1) of the circle S 1 ⊆ C is isomorphic
to Z, generated by the loop α∶ [0, 1] → S 1 given by t ↦ exp(2πit) ⊆ C,
π1 (S 1 , 1) = ⟨[α]⟩ 󰑔 Z.
Retractions and the Brouwer fixed point theorem.
Base point dependence. Let us suppose now that X is a path-connected topolog-
ical space, and that x0 , x1 ∈ X are two points. We ask whether or how the fundamen-
tal groups π1 (X, x0 ) and π1 (X, x1 ) are related. By assumption, there exists a path
λ∶ [0, 1] → X such that λ(0) = x0 and λ(1) = x1 . Now suppose that α∶ [0, 1] → X is a
loop based at x1 , that is x1 = α(0) = α(1). Then the composed/concatenated path
λ ∗ (α ∗ λ− ) (4.1)
is a loop based at x0 , see Section 4. It follows first λ from x0 to x1 , then follows
the loop α, and then travels back along the inverse path λ− from x1 to x0 . We will

Figure 13. The loop λ ∗ (α ∗ λ− )

now show that this defines in fact an isomorphism between the fundamental groups
based at x0 and x1 .
 MATH 3491 GEOMETRIC TOPOLOGY III RAPHAEL ZENTNER 21

Proposition 4.16. In the situation above, the map I λ defined by

I λ ∶ π1 (X, x1 ) → π1 (X, x0 )
[α] ↦ [λ ∗ (α ∗ λ− )]
defines a well-defined group isomorphism.
Proof. The fact that I λ is well-defined, i.e. only depends on the path-homotopy
class [α] of α is proved similarly to the proof of the statement that multiplication
in the fundamental group given by composites of paths is well-defined.
To show that it is a homomorphism, we compute
I λ ([α] ⋅ [β]) = [λ ∗ ((α ∗ β) ∗ λ− )]
= [λ ∗ (((α ∗ λ− ) ∗ (λ ∗ beta))λ− )]
= [(λ ∗ (α ∗ λ− )) ∗ (λ ∗ (β ∗ λ− ))]
= I λ ([α]) ⋅ I λ ([β]).
Here the first equation used the definition of I λ and the definition of the group
law in π1 (X, x0 ). The second equation uses a homotopy similarly to that in the
argument that for an arbitrary path µ starting at a point x1 the composite path
µ ∗ µ− (a loop based at x1 ) is path-homotopic to the constant path x1 that we
used in the proof about inverse elements in Theorem 4.7, and the third equation
is a homotopy through re-parametrisation similar to the proof of associativity in
Theorem 4.7. This proves that I λ is a homomorphism.
We leave it as an exercise to check that I λ is in fact an isomorphism with inverse
−
homomorphism I λ ,
(I λ )−1 = I λ .
−

□
Remark 4.17. In light of the last Proposition 4.16 we may simply speak of “the
fundamental group of the path-connected space X” and write π1 (X) instead of
π1 (X, x0 ). This is justified because we then know that for any two points x0 , x1 ∈ X
we have
π1 (X, x0 ) 󰑔 π1 (X, x1 ).
However, one has to be careful because the isomorphism depends on the chosen path
from x0 to x1 in general, as we will now further exhibit. The issue becomes a subtle
and delicate one in particular once we deal with homotopies that do not necessarily
keep base points fixed.
We now stress the fact that this isomorphism depends essentially on the choice
of path λ connecting the points x0 and x1 .
Proposition 4.18. Suppose X is a path-connected topological space, and suppose
λ, µ∶ [0, 1] → X are two paths with λ(0) = µ(0) = x0 and λ(1) = µ(1) = x1 . Then the
composite isomorphism
I λ ○ (I µ )−1 ∶ π1 (X, x0 ) → π1 (X, x0 )
22 RAPHAEL ZENTNER, PARTIALLY BUILT ON EARLIER NOTES BY JOHN HUNTON

is given by conjugation with the path-homotopy class [λ ∗ µ− ] of the loop λ ∗ µ− based

at x0 , that is we have
(I λ ○ (I µ )−1 )([γ]) = [λ ∗ µ− ][γ][λ ∗ µ− ]−1
for all elements [γ] ∈ π1 (X, x0 ).
Proof. We compute this, using the definition of I λ and using (I µ )−1 = I µ :
−

(I λ ○ (I µ )−1 )([γ]) = [λ ∗ ((µ− ∗ (γ ∗ µ)) ∗ λ− )]

= [(λ ∗ µ− ) ∗ (γ ∗ (µ ∗ λ− ))]
= [λ ∗ µ− ] ⋅ [γ] ⋅ [µ ∗ λ− ]
󳆵󳆺󳆺 󳆺 󳆹󳆺󳆺 󳆺 󳆷 󳆵󳆺󳆺 󳆺 󳆹󳆺󳆺 󳆺 󳆷
a loop based at x0 a loop based at x0
− − −1
= [λ ∗ µ ] ⋅ [γ] ⋅ [λ ∗ µ ] .
Here we made use of the same kind of arguments as in the proof of Proposition
4.16. □
We recall Corollary 4.14. If we have homotopy equivalences through base-point
preserving homotopies, then the maps they induce on the fundamental groups are
inverses of one another. What if we lose control of the base point during the homo-
topy? There will be practical situations as in the proof of the fundamental group
of Algebra below where we lose control of the base point. We will investigate this
next.
We will consider the cases of homotopies between loops now that do not keep
the base point fixed. To stress the difference to path-homotopies, we will call such
homotopies “free homotopies”. We do not expect that this preserves the path-
homotopy class, as illustrated in the next figure. We consider the square from
which we remove two points z0 and z1 , and loops α and β encircling both the point
z1 , but moving differently around the other. They are visibly freely homotopic, but
we will show later that they are not path-homotopic by showing that they represent
different elements in the fundamental group.

Figure 14. The loops α and β are freely homotopic, with interme-
diate stage of the free homotopy sketched on the right. The loop γ
traces the movement of the base point under the homotopy.
 MATH 3491 GEOMETRIC TOPOLOGY III RAPHAEL ZENTNER 23

Lemma 4.19. Let X be a topological space and let x0 , x1 be points. Suppose

α∶ [0, 1] → X and β∶ [0, 1] → X are loops based at α(0) = α(1) = x0 , respectively
β(0) = β(1) = x1 . Suppose that H∶ [0, 1] × [0, 1] → X is a continuous map such that
H(0, t) = α(t) and H(1, t) = β(t) for all t ∈ [0, 1], and such that
t ↦ H(s, t)
is a loop for all s ∈ [0, 1]. Let γ(t) = H(t, 0) = H(t, 1) be the path “traced out by the
homotopy H”, starting at H(0, 0) = α(0) = α(1) and moving to H(1, 0) = β(0) =
β(1). Then we have
[α] = I γ ([β]),
with the isomorphism I γ from Proposition 4.16 above. In particular, if γ is a closed
loop (and hence both α and β are loops based at x0 ), we have
[α] = [γ] ⋅ [β] ⋅ [γ]−1 .
Proof. We need to provide a path-homotopy G between α and γ ∗ (β ∗ γ − ). The
following picture explains how this can be achieved, keeping endpoints fixed despite
the fact that the homotopy H does not keep the endpoints fixed. Some explanations

Figure 15

are in place: We define the path γs ∶ [0, 1] → X by the formula

γs (t) = γ(st).
This is a path from γs (0) = H(0, 0) = x0 to γ(s) = H(s, 0). For a fixed value of s,
the path-homotopy defined by the picture in Figure ?? is the path
Gs = G(s, −) = γs ∗ (Hs ∗ γs− ).
It is easy to check that this gives a continuous map G∶ [0, 1] × [0, 1] → X which is a
path-homotopy between α and γ ∗ (β ∗ γ − ) which is a representative of I γ ([β]). If
γ is a loop, we argue as in the proof of Proposition 4.18 that we have
I γ ([β]) = [γ] ⋅ [β] ⋅ [γ]−1
in this case. □
24 RAPHAEL ZENTNER, PARTIALLY BUILT ON EARLIER NOTES BY JOHN HUNTON

We will next consider the effect of a homotopy between maps not necessarily
keeping based points fixed.
Corollary 4.20. Suppose f, g∶ X → Y are freely homotopic through a homotopy H,
from H0 = H(0, .) = f to H1 = H(1, .) = g, that is f ≃ g. Let x0 ∈ X be a point. Let
H

γ(s) = H(s, x0 ). Then for the path γ(t) = H(t, x0 ) from f (x0 ) to g(x0 ) we have the
identity
I γ ○ f∗ = g∗ ∶ π1 (X, x0 ) → π1 (Y, g(x0 )).
Proof. This is an easy consequence of the preceeding Lemma 4.19 applied to freely
homotopic paths f ○ α and g ○ α, for a loop α based at x0 . □
Lemma 4.21. Let X be a topological space, and x0 ∈ X. Let f ∶ X → X be a map
which is not necessarily base-point preserving. Suppose that f is freely homotopic
to the identity, f ≃ id (not necessarily base point preserving even if f (x0 ) = x0 ) by
a homotopy F with F0 = f and F1 = id. Then the map f∗ ∶ π1 (X, x0 ) → π1 (X, f (x0 ))
is given by
f∗ = I γ ∶ π1 (X, x0 ) → π1 (X, f (x0 )),
where γ∶ [0, 1] is the path given by γ(t) = F (t, x0 ). In particular, f∗ is an isomor-
phism.
Proof. Let Fs be a free homotopy from F0 = f to F1 = id, where Fs = F (s, −),
and F ∶ [0, 1] × X → X is continuous. Let γ(s) = F (s, x0 ). This is a path from
F (0, x0 ) = f (x0 ) to F (1, x0 ) = x0 . Let [α] ∈ π1 (X, x0 ) be represented by a loop α
based at x0 . By definition
f∗ ([α]) = [f ○ α].
Putting H(s, t) = F (s, α(t)), the map H∶ [0, 1] × [0, 1] → X defines a free homotopy
from H0 = f ○ α to H1 = id ○ α = α. By Lemma 4.19 we have
[f ○ α] = I γ ([α]).
□
Theorem 4.22. Suppose f ∶ X → Y is a homotopy equivalence with homotopy in-
verse g∶ Y → X, that is g ○ f ≃ idX and f ○ g ≃ idY . Let x0 ∈ X be a point. Then
f∗ ∶ π1 (X, x0 ) → π1 (Y, f (x0 )) and
g∗ ∶ π1 (X, f (x0 )) → π1 (Y, g(f (x0 )))
are isomorphisms. In particular, (freely) homotopy equivalent spaces have isomor-
phic fundamental groups.
Proof. Lemma 4.21 implies that
(g ○ f )∗ = g∗ ○ f∗ ∶ π1 (X, x0 ) → π1 (X, g(f (x0 ))) and
(f ○ g)∗ = f∗ ○ g∗ ∶ π1 (Y, f (x0 )) → π1 (X, f (g(f (x0 ))))
are isomorphisms. (Notice that despite our notation, the maps f∗ are homomor-
phisms between different fundamental groups, in the first one
f∗ ∶ π1 (X, x0 ) → π1 (Y, f (x0 )),
 MATH 3491 GEOMETRIC TOPOLOGY III RAPHAEL ZENTNER 25

but in the second one it is

f∗ ∶ π1 (X, g(f (x0 ))) → π1 (X, f (g(f (x0 )))).
The first equation implies that g∗ ∶ π1 (Y, f (x0 )) → π1 (X, g(f (x0 ))) is surjective (it
has a right inverse), and the second that it is injective (it has a left inverse). There-
fore g∗ is an isomorphism. But now if g∗ ○ f∗ is an isomorphism and g∗ is an
isomorphism, then f∗ π1 (X, x0 ) → π1 (Y, f (x0 )) also is. □
Corollary 4.23. Suppose the topological space X is contractible, and suppose x0 ∈
X. Then π1 (X, x0 ) is the trivial group {x0 }.
Proof. That X is contractible means that it is homotopy equivalent to the one-point
space {p}. Clearly the fundamental group of the latter is trivial. Theorem 4.22 now
implies the claim. □
Application: The fundamental theorem of algebra.

5. Covering spaces I
We will next study the notion of covering maps, which will have an intimate
relationship with fundamental groups, as we will see later on.
5.1. Definition and examples.
Definition 5.1. Let X, Y be topological spaces. A map π∶ X → Y is called a
covering map if it is locally trivial in the following sense: For any point y ∈ Y there
exists a neighbourhood V 󰑡 y, such that its preimage under π is a union of open sets
π −1 (V ) = ∪j∈J Uj ,
where J is some index set (which may depend on y), where the open sets Uj are
mutually disjoint, that is Uj ∩ Uk = ∅ if j ≠ k, and where the restriction map
󰑔
π󳈌Uj ∶ Uj → V
is a homeomorphism for any j ∈ J. The space Y is often called the base of the
covering, the space X is often called the total space of the covering, and often one
says that “X is a covering space of Y” without specifying the map π, which is of
course some abuse of language. For a point y ∈ Y the set π −1 (y) ⊆ X is also called
the fibre over y. A covering map is called globally trivial or just trivial if π −1 (Y ) is
a disjoint union of open sets as above, each being homeomorphic to Y . If the index
set J is finite, then its cardinality is also called the number of sheets of the covering
map over y.
Convention 5.2. We will also write ∐j∈J Uj for a disjoint union of open sets.

Example 5.3. (Covering of the circle by the real line) We consider the circle S 1 as
the set of unit complex numbers. The map
π∶ R → S 1
t ↦ exp(2πit)
26 RAPHAEL ZENTNER, PARTIALLY BUILT ON EARLIER NOTES BY JOHN HUNTON

Figure 16. The local triviality of a covering map

is a covering map with fibre π −1 (1) = Z, the integers. To prove this, let z0 =
exp(2πit0 ) ∈ S 1 be a point. Then the open sets

Uj = (t0 − + j, t0 + + j) ⊆ R
1 1
3 3
indexed by j ∈ Z are disjoint open sets which are homeomorphic, under π, to a
neighbourhood of z0 . In fact, if we consider the respective closed sets Uj = [t0 − 13 +
j, t0 + 13 + j] then the restriction of π to Uj is an injective continuous map, and hence
is a homeomorphism onto its image by an earlier lemma from Topology II. Hence
π󳈌Uj is also a homeomorphism.

Figure 17. The covering of the circle by the real line

Example 5.4. (Covering of a circle by itself) For any non-zero integer n ∈ Z the
map
ηn ∶ S 1 → S 1
z ↦ zn
 MATH 3491 GEOMETRIC TOPOLOGY III RAPHAEL ZENTNER 27

is a covering map. The proof is left as exercise. In the case n = 2, this is illustrated
by the following figure.

Figure 18. A 2-sheeted cover of the circle by itself

Observation 5.5. We observe the following immediate consequences of the defini-

tion of a covering map:
󲽨 A covering map is always surjective by definition.
󲽨 The fibre π −1 (y) over a point y inherits (as a subspace of X) the discrete
topology.
󲽨 The map y ↦ 󳈌π −1 (y)󳈌 ∈ N ∪ {∞} is locally constant on Y . In particular, if
Y is connected, then the number of sheets of the cover is constant.
Definition 5.6. (Wedge of two pointed topological spaces) Given two pointed topo-
logical spaces (Y, y0 ) and (Z, z0 ) , the wedge Y ∨ Z of these two topological spaces
is defined to be the quotient space
Y ∨ Z = (Y 󱮣 Z)󳆋 ∼
under the equivalence relation y0 ∼ z0 , and all other points or only equivalent to
themselves. Thus the unique equivalence class of more than two points is given by
the set {y0 , z0 }.
As an example, the wedge of two circles is homeomorphic to the symbol ∞, see
the lower part of Figure 19.
Example 5.7. (Covering of a wedge of two circles by an infinite tree) Figure 19
indicates a covering map π∶ X → Y , where Y = S 1 ∨ S 1 , and where X is an infinite
graph. The way to understand this covering maps is that all the vertical edges
between the vertices (the dots in the graph) are mapped to the left circle, and all
horizontal edges between vertices are mapped to the right circle, with the orientation
as indicated by the lifts of the paths α and β shown.

5.2. Lifting properties. Let π∶ X → Y be a covering space. Let ϕ∶ Z → Y be a

map. We can ask about the existence of a map ϕ̃∶ Z → X such that the following
diagram commutes,
28 RAPHAEL ZENTNER, PARTIALLY BUILT ON EARLIER NOTES BY JOHN HUNTON

Figure 19. A covering map of a wedge of two circles by an infinite

󳇊
tree. Indicated are lifts α ∗ β and β󳇊
∗ α, both starting at the same
point, but with different endpoints. We will see below that this
implies that the wedge of two circles has non-abelian fundamental
group.

X
ϕ̃
π

ϕ
Z Y
meaning that π ○ ϕ̃ = ϕ.
Convention 5.8. In a diagram of maps, we indicate by dashed arrows the existence
question of a map making the diagram commutative
Lifts may not always exist. For instance, the map ϕ∶ S 1 → S 1 given by z ↦ z
cannot be lifted to the covering π∶ R → S 1 , as we shall see later. In fact, ϕ can be
seen as a loop in S 1 , and we will see that its lift to R is given by a path which is not
 MATH 3491 GEOMETRIC TOPOLOGY III RAPHAEL ZENTNER 29

a closed loop anymore. Uniqueness of liftings, which we will prove now, will imply
that this ϕ cannot be lifted.
The following definition should be familiar from Topology II.
Definition 5.9. A topological space Z is connected, if the following holds: When-
ever we have a union Z = U ∪V by open sets U and V which are disjoint, then either
U = Z (and V = ∅) or V = Z (and U = ∅).
We recall also that path-connected topological spaces are always connected with
this definition, but the converse statement does not always hold.
Proposition 5.10. (Uniqueness of liftings) Let π∶ X → Y be a covering space, and
suppose that X is a Hausdorff space. Suppose that ϕ∶ Z → Y is a map, and suppose
that Z is a connected space. Suppose we have two lifts ϕ˜1 and ϕ˜2 ,
X
ϕ˜1
π
ϕ˜2
ϕ
Z Y
such that there exists a point z0 ∈ Z for which ϕ˜1 (z0 ) = ϕ˜2 (z0 ). Then ϕ˜1 = ϕ˜2 .
Proof. We define the subset W ⊆ Z to be the points of coincidence of ϕ˜1 and ϕ˜2 ,
W = {z ∈ Z 󳈌 ϕ˜1 (z) = ϕ˜2 (z)}.
We will show that W is open, and that its complement W c = Z ∖ W is also open.
By connectedness of Z, it then has tollow that either W c or W is all of Z. Since
by assumption z0 ∈ W , the set W is non-empty, and therefore we can conclude that
W = Z which is what we want to prove.
We first show that W c is open. So suppose there is some z ∈ Z such that ϕ˜1 (z) ≠
ϕ˜2 (z). Since X is Hausdorff, there exist disjoint open sets U1 and U2 in X, such
that ϕ˜1 (z) ∈ U1 and ϕ˜2 (z) ∈ U2 . Now all maps are continuous, and therefore there
exists a subset O1 ⊆ Z and O2 ⊆ Z such that ϕ˜1 (O1 ) ⊆ U1 and ϕ˜1 (O2 ) ⊆ U2 . Then
clearly O = O1 ∩ O2 is an open set containing z, and such that O ⊆ W c . Hence W c
is open.
Now suppose that z ∈ W , that is ϕ˜1 (z) = ϕ˜2 (z). Let V be a neighbourhood of
ϕ(z) ∈ Y which is trivialising for π, that is
π −1 (V ) = 󱮣 Uj
󰑔
such that π󳈌Uj → V is a homeomorphism for all j. We suppose that ϕ˜1 (z) = ϕ˜2 (z) ∈
Uj . By continuity of ϕ˜1 and ϕ˜2 there is an open neighbourhood O of z such that
ϕ˜1 (O) ⊆ Uj and ϕ˜2 (O) ⊆ Uj . But this means that we have the equation
ϕ󳈌O = π󳈌Uj ○ ϕ˜1 󳈌O = π󳈌Uj ○ ϕ˜2 󳈌O .
But since πUj is a homeomorphism, this implies that
(π󳈌Uj )−1 ○ ϕ󳈌O = ϕ˜1 󳈌O = ϕ˜2 󳈌O ,
and hence O ⊆ W . □
30 RAPHAEL ZENTNER, PARTIALLY BUILT ON EARLIER NOTES BY JOHN HUNTON

Proposition 5.11. (Existence of path lifting) Let π∶ X → Y be a covering map. Let

α∶ [0, 1] → Y be a path, and let x0 ∈ π −1 (α(0)) a point in the fibre over the starting
point of α. Then there exists a unique lift α̃∶ [0, 1] of α (i.e. π ○ α̃ = α?, starting at
α̃(0) = x0 .
Proof. The uniqueness part of the statement follows from proposition 5.10, so we
only need to prove the existence part of the statement.
The proof works inductively. Since α([0, 1]) ⊆ Y is compact (being the image of a
compact set under a continuous map), there is a finite number of trivilialising open
neighbourhoods U1 , . . . , Un covering α([0, 1]). We would like to conclude that as a
consequence, there is a partition of the interval [0, 1] by
0 = t0 < t1 < ⋅ ⋅ ⋅ < tn = 1
such that for each i = 0, . . . , n − 1 we have that
α([ti , ti+1 ]) ⊆ Ui ,
where Ui is one of the sets U1 , . . . , Un .
This would be an easy conclusion were the α−1 (Ui ) all open intervals in [0, 1],
but this will in general not be the case. However, the existence of such a partition
is guaranteed by the
Lemma 5.12. (Lebesgue Lemma) Let (X, d) be a compact metric space, and let
U = (Ui ) be a cover by open sets of X. Then there exists some positive number
δ > 0, called the Lebesgue number, such that the following holds: Whenever A ⊆ X
is a set of diameter smaller than δ, then there exists some Ui ∈ U such that A ⊆ Ui .
Here the diameter diam(A) of some subset A is given by
diam(A) = sup{d(x, y) 󳈌 x, y ∈ A}.

Proof. (Proof of the Lebesgue Lemma) Because X is compact, there is a finite

subcover U1 , . . . , Uk of X. We define a function h∶ X → R≥0 as follows. For i = 1, . . . , k
we define hi (x) as the supremum of the radii ri > 0 such that B(x, ri ) ⊆ Ui , where
B(x, ri ) is the closed ball of radius ri centered at x. If there is no radius r such
that B(x, r) ⊆ Ui , we set hi (x) = ∞. Then we set h(x) = min{h1 (x), . . . , hk (x)}.
Clearly h(x) < ∞ because the sets U1 , . . . , Uk cover X. Using the triangle inequality,
it is not hard to check that h is a continuous function. Since X is compact and
h is continuous, there is some point x0 at which h admits a minimum. Were the
statement only about diameters of open sets, the number δ = h(x0 ) would be a
Lebesgue number. To allow for any sets, we set δ = 10 9
h(x0 ). (The supremum of
values r such that [−r, r] is contained in the open set (−1, 1) is equal to 1, but
we do not have [−1, 1] ⊆ (−1, 1).) Indeed, let now A ⊆ X be some set of diameter
smaller than δ. Let x ∈ A. Then A ⊆ B(x, δ). Indeed, for any point y ∈ A we have
d(x, y) ≤ diam(A) < δ.
Now h(x) ≥ 10 9 δ. Since hi (x) is the supremum of the radii ri such that B(x, ri ) ⊆
Ui , we clearly have B(x, 10 9
hi (x)) ⊆ Ui . So in particular B(x, 10
9
h(x)) is contained
 MATH 3491 GEOMETRIC TOPOLOGY III RAPHAEL ZENTNER 31

in some set Ui . As δ = 10 h(x0 ),

9
we see that δ ≤ 9
10 h(x), and hence B(x, δ) ⊆ Ui for
some i. □

It is now clear that the existence of a partition as above exists. We only need to
find a subdivision so that for all i we have 󳈌ti+1 − ti 󳈌 < δ.
To start the induction, notice that α([0, t1 ]) ⊆ Ui0 for some i0 . By assumption
that Ui0 is a trivialising open set of the covering map, we have
π −1 (Ui0 ) = 󱮣 Oj,i0 such that π󳈌Oj,i0 ∶ Oj,i0 → Ui0
j∈J

is a homeomorphism. Clearly, there is some j such that x0 ∈ Oj,i0 . We can now

define a lift α̃∶ [0, t1 ] → X of α󳈌[0,t1 ] by defining

α̃(t) = (π󳈌Oj,i0 )−1 (α(t)).

Assume now that a lift (which we continue to denote α̃) of α󳈌[0,tk ] has been
constructed. Then α([tk , tk+1 ]) ⊆ Uik , where we have
π −1 (Uik ) = 󱮣 Oj,ik such that π󳈌Oj,ik ∶ Oj,ik → Uik
j∈J

is a homeomorphism. For some j we will have α̃(tk ) ∈ Oj,ik (notice that α̃(tk ) is
constructed in the previous step).
We can now extend our lift α̃ from the previous step to one that is a lift of α󳈌[0,tk+1 ]
by defining
α̃(t) = (π󳈌Oj,i0 )−1 (α(t)) for tk ≤ t ≤ tk+1 .
Since this extension from [0, tk ] to [0, tk+1 ] started at α̃(tk ), the lift α̃ that we
obtained of α󳈌[0,tk+1 ] is clearly unique. By induction, the general case follows. □

Example 5.13. Consider the path αn ∶ [0, 1] → S 1 given by t ↦ exp(2πint). For

every integer n ∈ Z this is a closed loop based at 1 ∈ S 1 . Consider the covering map
π∶ R → S 1 given by t ↦ exp(2πit) from above. Then the unique lift α 󳇈n of αn starting
at 1 ∈ S is given by the
1

󳇈n (t) = nt.
α
More examples of liftings of paths appear on Problem sheet 3.
Theorem 5.14. (Path-homotopy lifting property of covering maps) Let π∶ X → Y
be a covering map. Let
F ∶ [0, 1] × [0, 1] → Y
be a continuous map. Let y0 = F (0, 0), and let x0 ∈ π −1 (y0 ). Then there exists a
unique lifting
F̃∶ [0, 1] × [0, 1] → X such that F̃(0, 0) = x0 .
If F is a path-homotopy from the path F0 = F (0, −) to the path F1 = F (1, −), then
F̃ is also a path-homotopy, and in particular keeps start and end point fixed.
32 RAPHAEL ZENTNER, PARTIALLY BUILT ON EARLIER NOTES BY JOHN HUNTON

Proof. Uniqueness follows from Proposition 5.10 above, so we will only prove the
existence of a lifting.
By the Lebesgue Lemma, there exist subdivisions

0 = s0 < s1 < ⋅ ⋅ ⋅ < sm = 1

0 = t0 < t1 < ⋅ ⋅ ⋅ < tn = 1

such that F ([si , si+1 ] × [tj , tj+1 ]) ⊆ Uij , where Uij ⊆ Y is a trivialising open set for
the covering map π∶ X → Y . Let y0 = F (0, 0), and let x0 ∈ π −1 (y0 ).
As a first step in our construction, we define F̃ on the set A = {0}×[0, 1]∪[0, 1]×{0}
by lifting the paths s ↦ F (s, 0) and t ↦ F (0, t) with the condition that the lifts of
the paths start at x0 . We can do this by Proposition 5.11.
In the second step of our construction, we will work inductively on rectangles in
constructing the lift F̃. We will start with the rectangle [0, s1 ] × [0, t1 ] in the lower
left corner, then continue with [0, s1 ]×[t1 , t2 ], and continue with the lower row until
we are done with it, and then continue in the second row, work through it from the
left to the right, go up one row etc. This is illustrated in Figure 20.

Figure 20. Lifting by induction on rectangles, starting in the lower

left corner, finishing the row, continue on the left of the second row
etc. In each step, F̃ was already constructed in the previous step on
the red “L-shape” of the new rectangle, and this needs to be extended
over the new rectangle.
 MATH 3491 GEOMETRIC TOPOLOGY III RAPHAEL ZENTNER 33

Suppose the new rectangle is Rij = [si , si+1 ] × [tj , tj+1 ], so that F (Rij ) ⊆ Uij . Now
π −1 (Uij ) is a disjoint union of open sets,
π −1 (Uij ) = 󱮣 Ok such that π󳈌Ok ∶ Ok → Uij is a homeomorphism.
k

We denote
Lij = {si } × [tj , tj+1 ] ∪ [si , si+1 ] × {tj }.
the “L-shape” as sketched in red in Figure 20. The lift F̃󳈌Lij was already constructed
in previous steps of the induction. We claim that only one of the open sets Ok
contains the image F̃(Lij ). Indeed, the preimage
F̃󳈌−1
Lij (󱮣 Ok )

is going to be a disjoint union of open sets (with the subspace topology) of Lij .
Since Lij is connected, we conclude that we must have Lij = F̃−1 (Ok ) for some k,
and ∅ = F̃−1 (Ol ) for l ≠ k. Now we can extend F̃ to the rectangle Rij as follows:
F̃󳈌Rij = (π󳈌Ok )−1 ○ F 󳈌Rij .

This was the induction step, and so we conclude inductively that we obtain a lift F̃
of F on the whole square [0, 1] × [0, 1].
Now suppose that F is a path-homotopy. In particular, s ↦ F (s, 0) = y0 for all
s ∈ [0, 1], and s ↦ F (s, 1) = y1 for all s ∈ [0, 1], for some y1 ∈ Y . But that means
that for the lift F̃ the image of the continuous paths
s ↦ F̃(s, 0) and s ↦ F̃(s, 1) (5.1)
takes value in the fibre π −1 (y0 ) respectively π −1 (y1 ). A fibre has the discrete topol-
ogy (meaning that every point in the image π −1 (y) is open in the subspace topol-
ogy of the fibre). In fact, for a trivialising open set U for π, the disjoint union
π −1 (U ) = ∐ Ol provides open sets such that each Ol contains exactly one point of a
fibre π −1 (y) for y ∈ U , and the Ol ’s are pairwise disjoint. Therefore, by connected-
ness of the interval [0, 1], the images of the two paths in Equation (5.1) each take
value in exactly one point. Therefore, F̃ is a path-homotopy as well. □
Corollary 5.15. Let π∶ X → Y be a covering map, and let α, β∶ [0, 1] → Y be two
paths with α(0) = β(0) and α(1) = β(1). Suppose α and β are path-homotopic. Let
̃ and β̃ be lifts of α and β starting at some point x0 ∈ π −1 (α(0)) = π −1 (β(0)). Then
α
̃ and β̃ are path-homotopic, and, in particular, have the same endpoint:
α
̃
̃(1) = β(1).
α
Corollary 5.16. Let π∶ X → Y be a covering map, and let π(x0 ) = y0 . Then the
induced map
π∗ ∶ π1 (X, x0 ) → π1 (Y, y0 )
is injective.
34 RAPHAEL ZENTNER, PARTIALLY BUILT ON EARLIER NOTES BY JOHN HUNTON

Proof. Suppose that π∗ ([α]) = π([β]), where α, β are loops based at x0 . We have
π∗ ([α]) = π([β]) ⇐⇒ [π ○ α] = [π ○ β]
⇐⇒ π ○ α ≃ π ○ β.
Now obviously α and β are lifts of π ○ α and π ○ β, and hence theorem 5.14 implies
that α ≃ β are path-homotopic. Hence [α] = [β]. □
5.3. The lifting correspondence.
Definition 5.17. Let π∶ X → Y be a covering map, and let x0 ∈ π −1 (y0 ). Then the
map
φx0 ∶ π1 (Y, y0 ) → π −1 (y0 )
̃(1)
[α] ↦ α
̃ denotes the (unique) lift of α starting
is called the lifting correspondence, where α
at x0 .
By Corollary 5.15, this is indeed well-defined. We will next show that there are
situations where this is a bijection. In some cases, it allows to completely determine
the fundamental group of Y .
Remark 5.18. The lifting correspondence may in fact depend essentially on x0 , as
the following figure indicates.

Theorem 5.19. Let π∶ X → Y be a covering map, and let x0 ∈ π −1 (y0 ) for some
point y0 ∈ Y .
(a) If X is path-connected, then the lifting correspondence φx0 ∶ π1 (Y, y0 ) → π −1 (y0 )
is a surjection.
(b) If X is simply-connected (path-connected and with trivial fundamental group),
then the lifting correspondence is a bijection.
Remark 5.20. Notice that even though φx0 may be a bijection, it is not necessarily
a group homomorphism. In fact, there are many situations where the fibre doesn’t
even have the structure of the group. In the case of the covering π∶ R → S 1 above,
it does in fact define a group isomorphism.
Proof. (Proof of lifting correspondence) For part (a), let x ∈ π −1 (y0 ) be a point. By
assumption there is a path γ∶ [0, 1] → X with γ(0) = x0 and γ(1) = x. Clearly γ is
the lift of the loop π ○ γ∶ [0, 1] → Y based at y0 . Hence φx0 ([π ○ γ]) = x.
For injectivity, we rely on the following Lemma
Lemma 5.21. (Path homotopy in simply connected spaces) Let X be a simply con-
nected space, and let α and β be two loops with the same starting point α(0) = β(0) =
x0 and same endpoint α(1) = β(1) = x1 . Then α and β are path-homotopic.
Details of the proof of the Lemma are left as exercise, and a sketch of proof was
given in class. The proof of part (b) follows easily from this Lemma, and the details
are left as exercise. Again, more details were provided in the lecture. □
 MATH 3491 GEOMETRIC TOPOLOGY III RAPHAEL ZENTNER 35

Figure 21. The lifting correspondence may depend essentially on

the starting point x0 . In the example shown, some lifts are loops,
some have different starting and end point.

5.4. Fundamental groups of real projective spaces and the Lie groups
SO(n). We will show next that the lifting correspondence is useful enough to de-
termine the fundamental group of some spaces completely. We will only borrow
one result in this subsection from the next section, namely that the spheres S n for
n ≥ 2 have trivial fundamental group, which we will only prove when approaching
the Seifert-van-Kampen theorem.
Definition 5.22. For n ≥ 0 let ∼ denote the following equivalence relation on the
n-dimensional sphere S n . We declare x ∼ y if y = ±x, that is, the point y is either
x itself or its antipodal point −x. Here we consider the n-sphere as the boundary
of the closed ball around the origin in Rn+1 , making sense of −x in the usual sense.
The relation ∼ is clearly an equivalence relation. Then the quotient space
RPn = S n 󳆋 ∼
is called the n-dimensional real projective space.
We should think of RPn as the set of lines through the origin in Rn+1 , as a set.
Indeed, every line through the origin in Rn+1 intersects the sphere S n in precisely
two antipodal points x ∼ (−x). The definition as a quotient of S n gives the real
projective space a topology, and hence gives a suitable notion of lines that are close
to one another. Also, since RPn is a quotient of S n , it is clearly compact.
36 RAPHAEL ZENTNER, PARTIALLY BUILT ON EARLIER NOTES BY JOHN HUNTON

Example 5.23. Clearly RP0 is a space consisting of one point only. The map
f ∶ RP1 → S 1
[z] ↦ z 2
is easily checked to be well-defined, where [z] denotes the equivalence class of the
point z ∈ C 󰑔 R2 . In fact, this map f yields a homeomorphism, as is easily checked: It
is clearly surjective and injective, and it is therefore a bijective map from a compact
space space to a Hausdorff space, and therefore a homeomorphism.
We did define the 2-dimensional real projective space RP2 previously as the quo-
tient of the disc where two antipodal points on the boundary of the disc are consid-
ered equivalent. The next Lemma shows that this is in fact the same description,
up to homeomorphism.
We will denote by H+n and H−n the upper respectively lower hemispheres of the
n-dimensional sphere S n :
H+n = {x = (x1 , . . . , xn+1 ) ∈ S n ⊆ Rn+1 󳈌 xn+1 ≥ 0},
H−n = {x = (x1 , . . . , xn+1 ) ∈ S n ⊆ Rn+1 󳈌 xn+1 ≤ 0}.

Figure 22. In the case n = 2, the map h is consisting of mapping

from the unit disc tangent to the north pole “down to the upper
hemisphere”.

Lemma 5.24. The map

h∶B n → H+n
󳆼
(y1 , . . . , yn ) ↦ (y1 , . . . , yn , 1 − 󱮦󳈌yi 󳈌2 )
is a homeomorphism, and it induces a homeomorphism
h∶ B n 󳆋∼ → RPn ,
where ∼ denotes the equivalence relation on B n given by x ∼ y if x = y or if x, y lie
on the boundary ∂B n and satisfy y = ±x.
 MATH 3491 GEOMETRIC TOPOLOGY III RAPHAEL ZENTNER 37

Proof. The proof is an easy exercise that we leave to the interested reader. An
illustration of the map h is given by Figure 22. □
Proposition 5.25. The quotient map q∶ S n → RPn is a 2-sheeted covering map.
Proof. Suppose [x] ∈ RPn is a point where [x] = ±(x1 , . . . , xn+1 ). Then for some
i = 1, . . . , n + 1 we must have xi ≠ 0. We define the two sets
Ui+ = {y = (y1 , . . . , yn+1 ) 󳈌 yi > 0},
Ui− = {y = (y1 , . . . , yn+1 ) 󳈌 yi < 0},
Then V = q(Ui+ ) = q(Ui− ) is an open set containing [x], and q −1 (V ) = Ui+ ∐ Ui− . It
is easy to check that the restrictions
q󳈌Ui± ∶ Ui± → V
is a homeomorphism. □
We are now in the place for computing the fundamental groups of real projective
spaces. In light of Example 5.23 above and our known result that π1 (S 1 , ∗) 󰑔 Z, we
only need to consider the case n ≥ 2. The result we need from a later chapter is
Proposition 5.26. Suppose n ≥ 2. Then the sphere S n is simply connected, that
is, S n is connected and the fundamental group is trivial:
π1 (S n , ∗) 󰑔 {e}.
Since the map q∶ S n → RPn is a 2-sheeted covering map, and since S n is simply
connected for n ≥ 2, the lifting correspondence
φx ∶ π1 (RPn , [x]) → q −1 ([x])
is a bijection by part (b) of Theorem 5.19, and we have q −1 ([x]) = {±x}. Hence
π1 (RPn , [x]) is a group with two elements. There is only one such group, and we
conclude the following Theorem.
Theorem 5.27. We have π1 (RPn , [x]) 󰑔 Z󳆋2.
To conclude this section, we point out that we also obtain the fundamental groups
of some special orthogonal groups SO(n).
Proposition 5.28. The group SO(2) is isomorphic to S 1 . The group SO(3) is
homeomorphic to RP3 .
Proof. Indeed, any element in R ∈ SO(2) is a matrix given by
cos(θ) − sin(θ)
R = Rθ = 󳆚 󳆞
sin(θ) cos(θ)
for some θ ∈ [0, 2π], and the map Rθ ↦ exp(iθ) gives the desired isomorphism.
For the second claim, we consider the map
φ∶ B 3 → SO(3)
0 ↦ id
x ↦ Rx󳆋󳈒x󳈒,π󳈒x󳈒 .
38 RAPHAEL ZENTNER, PARTIALLY BUILT ON EARLIER NOTES BY JOHN HUNTON

Here Rv,θ denotes the rotation by angle θ ∈ [0, 2π] in the mathematically positive
sense around the normed vector v ∈ S n . In an orthonormal basis of R3 given by the
vectors (v, v2 , v3 ), this rotation is given by the matrix
⎛1 0 0 ⎞
Rv,θ = ⎜0 cos(θ) − sin(θ)⎟ .
⎝0 sin(θ) cos(θ) ⎠
For a normed vector v ∈ R3 it is easy to check that Rv,θ = R−v,2π−θ . In particular,
we have Rv,π = R−v,π . It is now easy to check that the map φ above is a homeomor-
phism onto its image on the interiour of the ball B 3 , it is surjective, and is 2-to-1
on its boundary, expressed by this last equation Rv,π = R−v,π .
We therefore conclude that φ descends to a homeomorphism
φ∶ B 3 󳆋 ∼→ SO(3),
where ∼ is the equivalence relation which identifies antipodal points on the boundary
of B 3 . Together with the homeomorphism h∶ B 3 󳆋 ∼→ RP3 from Lemma 5.24 we
therefore obtain the homeomorphism
−1
φ ○ h ∶ RP3 → SO(3).
□
Corollary 5.29. We have the following isomorphisms for the fundamental groups
of SO(2) and SO(3):
π1 (SO(2), ∗) 󰑔 Z and π1 (SO(3), ∗) 󰑔 Z󳆋2.
It can be shown that π1 (SO(n), ∗) 󰑔 Z󳆋2 for all n ≥ 3, but in this course we will
not have the tools available to prove this.
5.5. Distinguishing some surfaces via the fundamental group.

6. The Seifert-van Kampen Theorem

In this section our aim is to develop a very useful tool for computations of the
fundamental group of a space X by the fundamental group of smaller portions of
it. More precisely, we suppose that X = U ∪ V is the union of two open sets, both
containing some chosen base point x0 ∈ X. We will see that it will be possible to
determine the fundamental group of X from the fundamental groups of U and V ,
if U ∩ V , U , and V are all path-connected.
We start with the following proposition. We recall that a group G is said to
be generated by a subset S ⊆ G if any element in G can be written as a product
g = g1 . . . gk for some elements gi , i = 1, . . . , k where gi ∈ S or gi−1 ∈ S.
Proposition 6.1. Let X = U ∪ V be the union of two open sets, and let x0 ∈ U ∩ V .
Let
iU ∶ (U, x0 ) → (X, x0 ) and
iV ∶ (V, x0 ) → (X, x0 )
 MATH 3491 GEOMETRIC TOPOLOGY III RAPHAEL ZENTNER 39

be the inclusion maps of the sets U and V , respectively. Then the fundamental group
π1 (X, x0 ) is generated by elements in the image of
(iU )∗ ∶ π1 (U, x0 ) → π1 (X, x0 ) and
(iV )∗ ∶ π1 (V, x0 ) → π1 (X, x0 ).
Proof. Let α∶ [0, 1] → X be a loop based in x0 . By the Lebesgue lemma, there exists
a subdivision
0 = t0 < t1 < ⋅ ⋅ ⋅ < tn = 1
such that for each i = 0, . . . , n − 1 we have
α([ti , ti+1 ) ⊆ U or α([ti , ti+1 ) ⊆ V.
Without loss of generality, we may furthermore assume that if α([ti , ti+1 ) ⊆ U then
α([ti+1 , ti+2 ) ⊆ V , and if α([ti , ti+1 ) ⊆ V then α([ti+1 , ti+2 ) ⊆ U , for all i. With this
assumption, we have
α(ti ) ∈ U ∩ V for all i.
Now the paths α󳈌[ti ,ti+1 ] need not be even loops. However, for each i, there exists
a path γi ∶ [0, 1] → U ∩ V such that γi (0) = x0 and such that γi (1) = α(ti ). This is
illustrated in the following figure.

Figure 23. Connecting the basepoint to α(ti ) ∈ U ∩ V .

But then we have, up to the abuse of notation that the restrictions α󳈌[ti ,ti+1 are
considered to be paths defined on [0, 1] (this can be achieved by a reparametrisation)
α ≃ (α󳈌[0,t1 ] ∗ γ1− ) ∗ (γ1 ∗ α󳈌[t1 ,t2 ] ∗ γ2− ) ∗ ⋅ ⋅ ⋅ ∗ (γn−1 ∗ α󳈌[tn−1 ,1] ).
󳆵󳆺󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺󳆹 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆷 󳆵󳆺󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆹󳆺󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆷 󳆵󳆺󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆹󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺 󳆺󳆷
a loop based in x0 a loop based in x0 a loop based in x0

Here, every loop of the form γi ∗ α󳈌[ti ,ti+1 ] ∗ γi− )

has image entirely in U or entirely in
V . That this is really a homotopy equivalence relies on the same proof as showing
that α ∗ α− ≃ [x0 ] in Theorem 4.7. We sketch the idea by considering only a single
ti , to keep the notation simple. It will be clear how the general case follows.
For every value of s ∈ [0, 1], we define the path γi,s −
by γi,s−
(t) = γi− (st). For a fixed
value of s this is a path running from γi,s −
(0) = α(ti ) to γi,s−
(1) = γi,s
−
(1) = γi− (s). For
s = 0 this is the constant path α(ti ), and for s = 1 this is the path γi− .
Now the homotopy between α󳈌[ti−1 ,ti+1 ] and α󳈌[ti−1 ,ti ] ∗ γi− ∗ γi ∗ α󳈌[ti ,ti+1 ] is given
by
−
H(s, −) = α󳈌[ti−1 ,ti ] ∗ γi,s ∗ γi,s ∗ α󳈌[ti ,ti+1 ] ,
up to an obvious reparametrisation sketched by the triangle in Figure 24.
We conclude that we obtain
[α] = [α󳈌[0,t1 ] ∗ γ1− ] ⋅ [γ1 ∗ α󳈌[t1 ,t2 ] ∗ γ2− ] ⋅ ⋅ ⋅ ⋅ ⋅ [γn−1 ∗ α󳈌[tn−1 ,1] ],
40 RAPHAEL ZENTNER, PARTIALLY BUILT ON EARLIER NOTES BY JOHN HUNTON

Figure 24. The path-homotopy in the previous argument

and each of the elements on the right hand side lies either in the image of (iU )∗ or
in the image of (iV )∗ . □
Corollary 6.2. Let n ≥ 2 and let S n ⊆ Rn+1 be the n-dimensional sphere. Then its
fundamental group is trivial, that is, for all x0 ∈ S n , we have
π1 (S n , x0 ) = {e}.
Proof. We consider the two open sets
−1
U = {(x1 , . . . , xn+1 ) ∈ S n 󳈌 xn+1 > } and
2
V = {(x1 , . . . , xn+1 ) ∈ S n 󳈌 xn+1 < }.
1
2
We take x0 ∈ S n ∩ Rn × {0}. Here S n ∩ Rn × {0} 󰑔 S n−1 is “the equator” of S n .
Clearly U ∩ V is path-connected. The intersection in fact deformation retracts
onto the equator S n−1 . Now both sets U and V are homeomorphic to the interiour
of the open unit ball ○B n , by an argument similar to the one in Lemma 5.24. One
only has to “stretch the disc placed tangentially to the sphere at the north pole a
little beyond the equator”. It should be clear how this could be formalised more
rigorously.
By Proposition 6.1, we conclude that S n is symply connected, because both U
and V are. □
Remark 6.3. It is worth spending a moment to think about why this argument fails
for n = 1. Which hypothesis is not met for applying Proposition 6.1?
6.1. The free product of groups. Before going on with the formulation of the
Seifert-van Kampenn theorem, we need to digress into group theory.
Suppose G and H are groups. We first define a set G ∗ H from G and H, and in
a next step define a binary operation that we prove is a group law.
 MATH 3491 GEOMETRIC TOPOLOGY III RAPHAEL ZENTNER 41

Definition 6.4. Let G and H be arbitrary groups. Then G ∗ H is defined to be the

set of finite sequences
(x1 , . . . , xn )
(where n is variable) such that the following two conditions hold for all i ∈ N ∪ {0}:
(1) xi ∈ G∖{eG } or xi ∈ H ∖{eH }, where eG and eH denote the identity element
in G respectively H.
(2) xi and xi+1 lie in different groups.
We include as element the empty sequence ( ).
The binary operation we will define is simple in principle, but a little cumbersome
to write down. It will be the concatenation of sequences, except when we can simplify
because the last element of the first sequence lies in the same group as the first of
the second sequence.
(x1 , . . . , xn ) ⋅ (y1 , . . . , ym )
⎧
⎪ (x1 , . . . , xn , y1 , . . . , ym )
⎪
⎪
⎪
if xn and y1 lie in different groups
⎪
⎪
⎪(x1 , . . . , xn−1 , xn y1 , y2 , . . . , ym ) if xn and y1 lie in the same group, but xn y1 ≠ e
=⎨
⎪
⎪
⎪ (x1 , . . . , xn−1 y2 , . . . , ym ) if xn and y1 lie in the same group, xn y1 = e, but xn−1 y2 ≠ e
⎪
⎪
⎪
⎪
⎩ ⋮ ⋮
In other words, we reduce the sequence whenever we can because the product of
consecutive elements in the concatenated sequence makes sense in G or in H. If
we obtain the identity element, we omit from the sequence, and possibly reduce the
sequence further.
We notice that with this definition we have
(x1 , . . . , xn ) ⋅ (x−1 −1
n , . . . , x1 ) = ( ). (6.1)
Proposition 6.5. The set G ∗ H together with the binary operation Section 6.1 is
a group with neutral element the empty sequence ( ) and with inverses
(x1 , . . . , xn )−1 = (x−1 −1
n , . . . , x1 ).

Proof. It is easy to verify that the binary operation is associative, and the claim
about the neutral element follows trivially. The claim about inverses is a direct
consequence of ??. □
Definition 6.6. The group G ∗ H is called the free product of the groups G and H.
The sequences which are elements in G∗H are also called words in the corresponding
elements of G respectively H.
Observation 6.7. If the groups G and H are both non-trivial groups, that is,
G ≠ {eG } and H ≠ {eH }, then the free product G ∗ H is non-abelian and infinite.
Remark 6.8. If G and H are groups, then the maps
jG ∶ G → G ∗ H
g ↦ (g)
42 RAPHAEL ZENTNER, PARTIALLY BUILT ON EARLIER NOTES BY JOHN HUNTON

and
jH ∶ H → G ∗ H
h ↦ (h)
are injective group homomorphisms.
Convention 6.9. We will often drop the notations jG and jH and build words from
elements in G and elements in H, if there is no risk of ambiguity. For instance, for
g ∈ G ∖ {eG } and h ∈ H ∖ {eH } we will write
gh instead of the sequence (g, h).
Also, we will not make the sequence (x1 , . . . , xn ) always explicit as above. For
instance, the notation
x1 . . . xn ∈ G ∗ H
will denote an element in the free product G ∗ H if each element xi is an element of
G or of H (without the requirement that xi and xi+1 are to lie in different groups),
and it is clear how to understand this, given that we can make simplifications like
in the definition of the group law in the free product.
The usefulness of the free product is expressed in the fact that it behaves in a very
simple way with respect to group homomorphisms from G and H into a common
group K, as expressed by the following proposition.
Proposition 6.10. Let G, H, and K be groups, and let φ∶ G → K and ψ∶ H → K
be group homomorphism. Then there exists a unique homomorphism Ξ∶ G ∗ H → K
such that we have
Ξ ○ jG = φ and Ξ ○ jH = ψ.
In other words, the following diagram is commutative:
G
φ
jG

G∗H Ξ
K

jH
ψ

H
Proof. We first show uniqueness. Suppose Ξ󰐞 verifies the same properties as in ??.
If (x1 , . . . , xn ) ∈ G ∗ H is a word, say with x1 ∈ G and xn ∈ H, then we have
Ξ󰐞 ((x1 , . . . , xn )) = Ξ󰐞 ((jG (x1 )) ⋅ (jH )(x2 ) ⋅ ⋅ ⋅ ⋅ ⋅ (jH )(xn ))
= φ(x1 )ψ(x2 ) . . . ψ(xn )
= Ξ((jG (x1 )) ⋅ (jH )(x2 ) ⋅ ⋅ ⋅ ⋅ ⋅ (jH )(xn ))
= Ξ((x1 , . . . , xn )).
 MATH 3491 GEOMETRIC TOPOLOGY III RAPHAEL ZENTNER 43

Similar equations hold if x1 ∈ G and xn ∈ G, if x1 ∈ H and xn ∈ G, and if x1 ∈ H and

xn ∈ H.
To prove existence, we define Ξ in the natural appearing way:
⎧
⎪
⎪φ(x1 )ψ(x2 ) . . . if x1 ∈ G, and
Ξ((x1 , . . . , xn )) = ⎨
⎪
⎪ψ(x1 )φ(x2 ) . . . if x1 ∈ H.
⎩
It is easy to see that this is a group homomorphism. □
Definition 6.11. The unique homomorphism Ξ from the previous proposition is
denoted by φ ∗ ψ.
This should be compared with the following similar statement about the direct
product of two groups:
Proposition 6.12. Let G, H, and K be groups, and let φ∶ G → K and ψ∶ H → K be
group homomorphism such that
φ(g)ψ(h) = ψ(h)φ(g) for all g ∈ G, hinH.

Then there exists a unique homomorphism Ξ∶ G ∗ H → K such that we have

Ξ ○ jG = φ and Ξ ○ jH = ψ.

Then there exists a unique homomorphism Θ∶ G × H → K such that we have

Θ ○ jG = φ and Θ ○ jH = ψ,

where here jG ∶ G → G × H and jH ∶ H → G × H are defined by jG (g) = (g, eH ) for all

g ∈ G, and jH (h) = (eG , h) for all h ∈ H. In other words, the following diagram is
commutative:
G
φ
jG

G×H Θ
K

jH
ψ

H
Proof. The proof is easy and we leave its proof to the reader. □
The following example illustrates that the condition in Proposition 6.12 is in
general necessary.
Example 6.13. We consider the following homomorphisms φ∶ Z󳆋3 → S3 and ψ∶ Z󳆋2 →
S3 , where S3 is the permutation group of three elements. We specify these on gen-
erators of Z󳆋3 respectively Z󳆋2:
φ(1) = (1 2 3) and ψ(1) = (1 2).
44 RAPHAEL ZENTNER, PARTIALLY BUILT ON EARLIER NOTES BY JOHN HUNTON

If there were a homomorphism Θ as in Proposition 6.12, it would clearly by surjective

because the permutations (1 2 3) and (1 2) generate S3 . But both S3 and Z󳆋3 × Z󳆋2
have six elements, so a surjective homomorphism would be also injective, and hence
an isomorphism. However, the group Z󳆋3 × Z󳆋2 is abelian, whereas the permutation
group S3 is not, a contradiction. Therefore, the condition in Proposition 6.12 cannot
be omitted.
6.2. Seifert-van-Kampen theorem, simplified version. We are now ready to
state a first version of the Seifert-van-Kampen theorem.
Theorem 6.14. Suppose that X = U ∪ V is a union of two open sets U and V such
that the intersection U ∩ V is simply connected. Suppose that x0 ∈ U ∩ V . Then the
homomorphism
(jU )∗ ∗ (jV )∗ ∶ π1 (U, x0 ) ∗ π1 (V, x0 ) → π1 (X, x0 )

is an isomorphism. Here jU ∶ U ↪ X and jV ∶ V ↪ X are the inclusion maps, and

(jU )∗ ∗ (jV )∗ is the homomorphism guaranteed by Proposition 6.10.
Proof. In class we discussed a “proof by example” which we will omit here, because
we will give a full proof of the more general statement later on. □
Example 6.15. Suppose X = S 1 ∨ S 1 . We denote by α∶ [0, 1] → S 1 a loop in the
first S 1 such that π1 (S 1 , x0 ) = ⟨[α]⟩ 󰑔 Z, and by β∶ [0, 1] → S 1 a loop in the second
S 1 such that π1 (S 1 , x0 ) = ⟨[β]⟩ 󰑔 Z. We claim that we have
π1 (S 1 ∨ S 1 , x0 ) 󰑔 ⟨[α]⟩ ∗ ⟨[β]⟩ 󰑔 Z ∗ Z.
Indeed, this is a consequence of the above version of the Seifert-van-Kampen theo-
rem, as we will sketch by pictures, rather than attempting to write down U and V
by formulae. The following figure suggests two open sets, each of which admits a
base-point preserving deformation retraction onto a circle S 1 , as sketched.

Figure 25. The space S 1 ∨ S 1 as a union of two open sets with

simply-connected intersection.

The statement of the above version of the Seifert-van-Kampen theorem naturally

raises the question what happens if π(U ∩V, x0 ) is not the trivial group. We start by
 MATH 3491 GEOMETRIC TOPOLOGY III RAPHAEL ZENTNER 45

observing that we have an obvious commutative diagram of inclusions if X = U ∩ V .

U
iU jU

U ∩V X

iV jV

V
If x0 ∈ U ∩ V is a base point, then this yields the following commutative diagram of
fundamental groups.
π1 (U, x0 )
(iU )∗ (jU )∗

π1 (U ∩ V, x0 ) π1 (X, x0 ) (6.2)

(iV )∗ (jV )∗

π1 (V, x0 )
Now there is still the homomorphism (jU )∗ ∗(jV )∗ ∶ π1 (U, x0 )∗π1 (V, x0 ) → π1 (X, x0 ),
but we expect it to have kernel now. In fact, we have the commutative diagram
π1 (U, x0 )
(iU )∗ (jU )∗
jπ1 (U )

(jU )∗ ∗(jV )∗
π1 (U ∩ V, x0 ) π1 (U, x0 ) ∗ π1 (V, x0 ) π1 (X, x0 )

jπ1 (V )
(iV )∗ (jV )∗

π1 (V, x0 )
With the slight abuse in notation from Convention Convention 6.9, the expression
(iU )∗ (󰂃)((iV )∗ (󰂃))−1 denotes an element of π1 (U, x0 ) ∗ π1 (V, x0 ), for any 󰂃 ∈ π1 (U ∩
V, x0 ). However, such an element lies in the kernel of (jU )∗ ∗ (jV )∗ , as we will now
show:

(jU )∗ ∗ (jV )∗ ((iU )∗ (󰂃)((iV )∗ (󰂃))−1 ) = (jU )∗ ((iU )∗ (󰂃)) ((jV )∗ ((iV )∗ (󰂃)−1 )
= ((jU )∗ ○ (iU )∗ )(󰂃)) (((jV )∗ ○ (iV )∗ )(󰂃))−1
= e.
(6.3)
46 RAPHAEL ZENTNER, PARTIALLY BUILT ON EARLIER NOTES BY JOHN HUNTON

The last equality holds because (jU )∗ ○ (iU )∗ = (jV )∗ ○ (iV )∗ . Thus, all elements
of the normal subgroup generated by the elements (iU )∗ (󰂃)((iV )∗ (󰂃))−1 lies in the
kernel of (jU )∗ ∗ (jV )∗ , and so the normal subgroup

N = ⟪{(iU )∗ (󰂃)((iV )∗ (󰂃))−1 󳈌 󰂃 ∈ π1 (U ∩ V, x0 )}⟫

generated by these elements lies in the kernel of (jU )∗ ∗ (jV )∗ . We recall that for a
subset S ⊆ G of a group G the normal subgroup
langleS⟫ is the smallest normal subgroup containing the set S, or equivalently, it is
the intersection of all normal subgroups of G which contain S:
⟪S⟫ = 󱮉 H.
H⊇S
H◁G

Here we use the notation H ◁G to indicate that H is normal in G, i.e. that xN = N x

for all x ∈ G, or equivalently, that xhx−1 ∈ N for all h ∈ N .

6.3. The free product with amalgamation. The preceding considerations lead
us naturally to the following definition.
Definition 6.16. Suppose L, G, and H aree groups, and that α∶ L → G and β∶ L → H
are group homomorphisms. As before, we denote by α(l)β(l) an element of the free
product G ∗ H. We consider the normal subgroup

N = ⟪{α(l)β(l)−1 󳈌 l ∈ L}⟫ ◁ G ∗ H. (6.4)

Then the quotient group

G ∗L H ∶= G ∗ H󳆋N

is called the free product of G and H, amalgamated over L.

This is of course an abuse of notation, since clearly the group homomorphisms α
and β are part of the definition, but they do not appear in the notation G ∗L H.
The analogue of the property of Proposition 6.10 for the free product of G and
H is now the following.
Proposition 6.17. Let L, G, H and K be groups, and let φ∶ G → K and ψ∶ H → K be
group homomorphism, and likewise α∶ L → G and β∶ L → H, such that the condition
φ○α=ψ○β

holds.
Then there exists a unique homomorphism χ∶ G ∗L H → K such that we have
χ ○ q ○ jG = φ, and χ ○ q ○ jH = ψ,

where q∶ G ∗ H → G ∗L H denotes the quotient homomorphism, and where jG ∶ G →

G∗H and jH ∶ H → G∗H denote the inclusion homomorphisms into the free product.
 MATH 3491 GEOMETRIC TOPOLOGY III RAPHAEL ZENTNER 47

In other words, the following diagram is commutative:

G
α φ
q○jG

G ∗L H
χ
L K.

q○jH
β ψ

H
Proof. Proposition 6.10 provides us with the homomorphism φ ∗ ψ∶ G ∗ H → K such
that (φ ∗ ψ) ○ jG = φ and such that (φ ∗ ψ) ○ jH = ψ. It is easy to see that the normal
subgroup N defined in Equation (6.4) lies in the kernel of φ ∗ ψ. In fact, we did this
in a special but analogous case in Equation (6.3). The first isomorphism theorem,
together with the conclusion of Proposition 6.10 yields the desired conclusion of the
existence of χ with the desired properties.
The proof of uniqueness is left as an easy exercise. □
Example 6.18. (The modular group Sl(2, Z)) This example is not course/exam-
relevant, and neither do we provide a full proof, but it illustrates that free products
with amalgamations are not “exotic” but occur in other areas of mathematics. The
group Sl(2, Z) is by definition the group of 2 × 2-matrices with integer coefficients
and determinant one. It is also known as the modular group and plays a crucial role
in number theory. It acts by Möbius transformations on the upper half plane in
C. Entire books were written only about Sl(2, Z). We refer to Modular group on
Wikipedia.
It is known that Sl(2, Z) is generated by the two elements
0 −1
S=󳆚 󳆞 which is an element of order 4, and
1 0
1 −1
R=󳆚 󳆞 which is an element of order 6.
1 0
Thus we obtain two group homomorphisms φ∶ Z󳆋4 → Sl(2, Z) and ψ∶ Z󳆋6 → Sl(2, Z)
specified by the image of a generator of the cyclic groups Z󳆋4 and Z󳆋6,
φ(1) = S and ψ(1) = R.
We also have the two homomorphisms α∶ Z󳆋2 → Z󳆋4 given by 1 ↦ 2 and β∶ Z󳆋2 → Z󳆋6
given by 1 ↦ 3. These are well-defined. Since
−1 0
S2 = 󳆚 󳆞 = R3
0 −1
we obtain the equality
φ ○ α = ψ ○ β.
48 RAPHAEL ZENTNER, PARTIALLY BUILT ON EARLIER NOTES BY JOHN HUNTON

Now Proposition 6.17 yields a homomorphism

f ∶ Z󳆋4 ∗Z󳆋2 Z󳆋6 → Sl(2, Z). (6.5)
Since S and R generate the modular group, this homomorphism is surjective. It is
a consequence of the Ping-pong Lemma that this homomorphism is injective, and
hence an isomorphism. Thus
Sl(2, Z) 󰑔 Z󳆋4 ∗Z󳆋2 Z󳆋6.
The subgroup
󳆢± 󳆚 󳆞󳆧 󰑔 Z󳆋2
1 0
0 1
is normal (and in fact central) in Sl(2, Z), and the quotient group is denoted
PSl(2, Z) = Sl(2, Z)󳆋{±id}.
The above homomorphism f then induces an isomorphism
f ∶ Z󳆋2 ∗ Z󳆋3 → PSl(2, Z).
Thus the group PSl(2, Z) is isomorphic to the free product of Z󳆋2 and Z󳆋3.
6.4. Seifert-van-Kampen theorem, general case. The commutative diagram in
Equation (6.2) together with the property of the free product with amalgamations
Proposition 6.17 provides us with a unique group homomorphism
χ∶ π1 (U, x0 ) ∗ π1 (V, x0 ) → π1 (X, x0 )
π1 (U ∩V,x0 )

such that the following diagram commutes:

π1 (U, x0 )
(iU )∗ (jU )∗
q○jπ1 (U )

π1 (U ∩ V, x0 ) π1 (U, x0 ) ∗ π1 (V, x0 ) π1 (X, x0 )

χ
(6.6)
π1 (U ∩V,x0 )

(iV )∗ q○jπ1 (V ) (jV )∗

π1 (V, x0 )
commutes.
Theorem 6.19. (Seifert-van-Kampen) Suppose that X is a topological space, and
that U and V are open subsets, and suppose x0 ∈ U ∩ V is a point. Suppose that U ,
V , and U ∩ V are path-connected. Then the homomorphism
χ∶ π1 (U, x0 ) ∗ π1 (V, x0 ) → π1 (X, x0 )
π1 (U ∩V,x0 )

which turns the diagram in Equation (6.6) commutative, is an isomorphism.