Top 1

Topology I
Sylvain Cappell,
Transcribed by Ian Tobasco
Abstract. This is part one of a two semester course on algebraic topology.
The course was offered in Fall 2011 at the Courant Institute for Mathematical
Sciences, a division of New York University. Though the course is not taken
from any particular text, some references are Introduction to Algebraic Topol-
ogy by Munkres, Hatcher’s series of texts on algebraic topology, Greenberg
& Hauper, Spanier, etc. Milnor’s Topology from the Differentiable Viewpoint
may be useful in the manifolds portion of the course; Kelly’s text may be use-
ful for the review of point-set. There will be a final and suggested homework
problems.
Contents
Chapter 1. Introduction 5
Chapter 2. Point-Set Topology 7
1. Metric Space 7
2. Topological Space 9
3. The Separation Axioms 10
4. Compactness 11
5. The Product Topology 13
6. Connectedness 14
7. Quotient Spaces and Gluing 15
Chapter 3. The Fundamental Group 17
1. Homotopy of Paths 17
2. Independence of Basepoint 18
3. Comparing Spaces Algebraically 20
4. Homotopy of Maps 23
5. Homotopy of Spaces 25
6. Combinatorial Group Theory and Van Kampen’s Theorem 26
7. Application to Knot Theory 34
8. Application to Topological Groups and H-spaces 37
Chapter 4. Covering Space Theory 41
1. Covering Spaces 41
2. Higher Homotopy Groups 42
3. The Lifting Problem 43
4. Comparing Covering Maps 45
3
CHAPTER 1
Introduction
Lecture 1, 9/12/11
We begin with a discussion of some famous topological questions. Historically,
topologists studied the classification of spaces. Soon enough we’ll make the notion
of “space” rigorous; a first example is any X ⊂ Rn with the Euclidean distance
n
!1/2
X 2
dRn (u, v) = ||u − v|| = (ui − vi ) .
i=1
Some “nice” spaces are manifolds, e.g. S 2 . We call this a manifold as it is locally
Eucliean. Another example is the surface of a donut, or T 2 . Notice T 2 ⊂ R3 has
one hole – we say it has genus g = 1. A surface with two holes is genus g = 2,
and so on. There is an infinite family of spaces in this manner, each with different
genus.
One of the basic classification results in topology is that of one-manifolds.
Obviously R1 is a one-manifold. Another is S 1 ⊂ R2 , for it is locally homeomorphic
to R1 . It is a classic result that S 1 is the only compact connected one-manifold.
A second problem in topology concerns embeddings of manifolds into Euclidean
space. Given a manifold M n , can one construct a copy of M n in RN for some N ?
This naturally gets easier as N increases. The interesting question is: What is
the lowest N for which this is possible for all n-dimensional manifolds? Whitney
answered this question with the astounding result that M n ,→ R2n for any n.
Actually, Whitney first proved M n ,→ R2n+1 , then worked very hard to reduce the
dimension of the ambient space by one.
There are more two-manifolds than one-manifolds. One interesting example is
the real projective space RP 2 . Here is a first description. We can view the two-
sphere as the result of gluing two hemispheres together along their boundary (so
along S 1 ). We write S 2 = D+ 2
∪S 1 D− 2
to mean this. Now consider the Mobius
band. It is a two-dimensional manifold with boundary, but it has only one boundary
component, and in fact ∂ (Mobius) = S 1 . To construct RP 2 , we’ll glue D2 to the
Mobius band along S 1 . So RP 2 = D2 ∪S 1 Mobius. We can’t carry this out physically
in R3 ; RP 2 is a closed, compact two-dimensional manifold which does not embed
in R3 . But of course, it will embed in R4 .
Here is a second description of RP 2 . We have RP 1 = S 1 , so perhaps we can
think along the same lines for RP 2 . We proceed by identifying each v ∈ S 2 with
−v, its antipode. This yields RP 2 . An informal proof is as follows: consider S 2
as consisting of north and south polar caps along with an equatorial band. After
identifying the two caps, there is only one cap left. And the space we’re left with
is (cap) ∪ (Mobius). Somehow, we’ve arrived at our previous construction.
A third question from topology stems from the observation that S n = ∂Dn+1
for each n. The question is: Given a manifold without boundary, M n , does there
5
6 1. INTRODUCTION
exists W n+1 such that ∂W = M ? The answer in general is no. For example, RP 2
is not a boundary. Thom answered this question by determining a set of numerical
invariants that determine completely wheter a manifold is a boundary.
A large and powerful part of topology consists of the so-called “fixed point
theorems”. We ask: Given continuous f : X → X, does there exist x ∈ X with
f (x) = x? It is a standard trick to turn an equation of interest into a question
about fixed points. One famous result is the Brouwer fixed point theorem, which
says that every continuous f : Dn → Dn has a fixed point. We can prove this with
calculus in the case n = 1. If f (x) 6= x for all x, define
f (x) − x
g (x) =
|f (x) − x|
which is (pointwise) either +1 or −1. Since g is continuous, we have the two cases
g ≡ 1 and g ≡ −1. In the first case we get f (1) > 1 and in the second f (−1) < 1,
both of which are contradictions.
Here is the idea for arbitrary n. Assuming f (x) 6= x for all x ∈ Dn , construct
g as in the following picture: (D1) Clearly g is continuous, so we’ve exhibited a
continuous map g : Dn → ∂Dn = S n−1 which restricts to the map g|S n−1 = idS n−1 .
Lecture 2, 9/19/11 As it turns out, this is impossible, a fact which we’ll prove later in the course.
Analysis and topology are deeply connected. There are many results in which
analytical quantities are topological quantities as well. A 19th century example is
the Riemann-Rach Theorem, which gives a topological formula for the dimension
of the space of meromorphic functions with prescribed zeros/poles on a Riemann
surface. In the 20th century, the theorem was extended to all dimensions, to sin-
gular varieties, to differential operators (Atiyah-Singer Index Theorem for elliptic
operators). But perhaps this is too grand for now.
Here is a more naive story – Morse Theory. This subject consists of the study
of topological consequences of critical points of functions on a given manifold. A
critical point of a function f is where f looks constant to first order. Consider the
height function h : T 2 ⊂ R3 → R given by h (x) = x3 . h then has four crtical
points. Now observe that we can always increase the number of critical points of h
by deforming the manifold T 2 . A central result of Morse Theory is that no matter
how we deform T 2 , h will always have at least four critical points. The justification
of this fact is in the “Morse inequalities,” which are formulae for the minimum
number of critical points. Another example is S 2 : the height function always will
have two critical points, no matter how S 2 is deformed. Somehow, the topology of
a space (a global result) is determined by the critical points of functions (a local,
analytic result).
In this course we’ll discuss continuous and differentiable functions. In topology,
there is a vast interest in generalizing these notions to manifolds. So one could ask:
how unique is the idea of differentiability on a manifold? More explicitly, suppose
we have a function f defined on a manifold. Can we give that manifold two different
differential structures, one on which f is considered differentiable and the other on
which f is not? Remarkably, the answer is yes! It is a theorem of Milnor that S 7
has 28 differential structures. And R4 has uncountably many differential structures
(though Rn for n 6= 4 only has one). RP 4 also has more than one differential
structure (due to Cappell and Sharean). It is unknown whether S 4 has a unique
differential structure.
CHAPTER 2
Point-Set Topology
1. Metric Space
The first thing to do is make the notion of “space” rigorous. Recall the notion
of a metric space, (X, d), a set X and a distance d : X × X → R≥0 which satisfies
(1) d (u, v) = 0 ⇐⇒ u = v in X
(2) d (u, v) = d (v, u) for all u, v ∈ X
(3) d (u, v) ≤ d (u, w) + d (w, v) for all u, v, w ∈ X.
The third requirement is called the “triangle inequality”.
Examples 1.1.
P 1/2
2
(1) Rn with the Euclidean metric d (u, v) = i (ui − vi ) .
(2) If (X, dX ) is a metric space, then any subset A ⊂ X can be made into a
metric space by setting dA = dX |A .
(3) Function spaces, e.g. C [0, 1] with d (f, g) = max[0,1] |f (x) − g (x)|.
(4) We can give Rn a different metric, dmax (u, v) = maxi=1,...,n |ui − vi |.
Exercise 1.2. Check dmax is a metric.
(5) Any set X can be made into a metric space with d (u, v) = δu=v .
Exercise 1.3. Check d is a metric.
(6) The p-adic metric on Z. The idea is to compare Z with Z/pZ = Zp .
Given a prime number p, we’ll define a metric dp on Z with small ||p||.
Here ||v|| = d (v, 0). Given n ∈ Z, write n = pa b with b prime to n, and
set ||n||p = p−a . (E.g. ||12||2 = 1/4.) Then define dp (u, v) = ||u − v||p for
u, v ∈ Z. (E.g. d2 (4, 12) = 1/8, d2 (11, 12) = 1.)
Exercise 1.4. Prove dp is a metric. In fact, prove the “ultra-metric
inequality” which says
dp (u, w) ≤ max {dp (u, v) , dp (v, w)} .
(Hint: This follows from ||a + b||p ≤ max {||a||p , ||b||p }. Let a = u − v and
b = u − w, then a + b = u − w. The point here is that the power of p
which divides a + b is at least as big as the minimum of the powers of p
dividing a and b.)

We can extend the definition of p-adics to Q. We have Q = u/v v 6= 0 .
So define ||u/v||p = ||u||p /||v||p . In other words, given u/v = pα m where
m has no powers of p (α can be negative ), set ||u/v||p = p−α . (E.g.
||1/12||2 = 4.)
Exercise 1.5. Check this satisfies the metric axioms and the ultra-
metric condition.
7
8 2. POINT-SET TOPOLOGY
We can extend the p-adics even further by completing Q, via limits in

||·||p . This produces the p-adic numbers Qp . These look like power series
in p:

a−i a−(i−1) 2
Qp = + i−1 + · · · + a0 + a1 p + a2 p + · · · .
pi p
Of course, we can also consider objects such as Qnp = Qp × · · · × Qp , and
so on.
Now let f : (X, dX ) → (Y, dY ) be a function.
Definition 1.6. We say f is continuous at a point u if for all > 0 there exists
δ > 0 so that if dX (u, v) < δ then dY (f (u) , f (v)) < . We say f is continuous if
it is continuous at all points x ∈ X.
Exercise 1.7. Consider f : (Rn , dEuclidean ) → (Rn , dmax ) given by f (x) = x.
Is f continuous? Is f −1 continuous?
Definition 1.8. We call f an isometry if it is a 1-1 correspondance which has
dX (u, v) = dY (f (u) , f (v)) for all u, v ∈ X.
Isometry is too strict a condition for our purposes. For example, all circles in
the plane should be topologically equivalent. But only those circles with the same
radii will be isometric. Eventually we’ll see a weaker notion of equivalence.

Definition 1.9. Br (u) = v ∈ X d (u, v) < r is the open ball of radius r > 0
around u.
Proposition 1.10. f is continuous at u ∈ X iff for all > 0 there is δ > 0 so
that f (Bδ (u)) ⊂ B (f (u)).
Definition 1.11. A subset U ⊂ X is open if it contains an open ball around
each of its points.
Proposition 1.12. The collection of open sets satisfy:
(1) X is open, ∅ is open.
(2) The union of open sets is open.
(3) If A, B ⊂ X are open then A ∩ B is open.
Remark. Usually people write (3) in its alternate form, which asserts the finite
intersection of open sets is open. Naturally, they are equivalent. But observe that
the infinite intersection of open sets need not be open, e.g. ∩∞
n=1 (−1/n, 1/n) = {0}.
Proposition 1.13. f : (X, dX ) → (Y, dY ) is continuous iff f −1 (U ) is open in

X whenever U ⊂ Y is open in Y .
An immediate consequence is
Proposition 1.14. If f : X → Y , g : Y → Z are continuous, then g ◦ f : X →
Z is continuous.
The ease with which this follows from proposition 1.13 suggests reformulating
the subject in terms of open sets, and not a metric. This brings us to the idea of
an abstract topological space.
2. TOPOLOGICAL SPACE 9
2. Topological Space
Definition 2.1. Let X be a set. T is called a topology on X if it is a collection
of subsets of X, called the open sets, so that
(1) ∅, X ∈ T
(2) If Aα ∈ T for all α then ∪α Aα ∈ T .
(3) If A, B ∈ T then A ∩ B ∈ T .
Examples 2.2.
(1) Any metric space with T consisting of the open sets as determined by the
metric.
(2) Given a topological space (X, TX ) and any A ⊂ X we can define the
relative topology on A by taking TA = B ∩ A B ∈ TX .
Exercise 2.3. (A, TA ) is a topological space.
Remark. The open sets in TA are not necessarily the open sets in
TX . Consider the open sets on R1 and the open sets in A = [0, 1] in the
relative topology.
Exercise 2.4. TA ⊂ TX ⇐⇒ A ∈ TX
Exercise 2.5. Let B ⊂ A ⊂ X and let TX be a topology on X. We
can give A the relative topology from X, then B the relative topology
from A. Show this is the same as going directly from X to B.
(3) Given any set X, we gave it the discrete metric d (u, v) = δu=v . Thus
any set X gets a topology from d, and its not hard to show TX = P (X).
We’ll call this the discrete topology. It is the largest topology. The smallest
topology is TX = {∅, X}, which we’ll call the indiscrete topology. Note
that with any space X and any topology TX we have Tindiscrete ⊂ TX ⊂
Tdiscrete , so the terms “smallest” and “largest” are meaningful.
Exercise 2.6. Consider id : X → Xdiscrete , id : Xdiscrete → X, id :
X → Xindiscrete , and id : Xindiscrete → X. Which of these is continuous,
no matter what topology X has?
(4) The Zariski topology. The idea (from algebraic geometry) is to only study
open sets defined by algebraic equations. For example, we could take
the open sets to be CN \ {solutions of systems of polynomial equations}
along with the empty set. These are “fat” open sets, in that they are very
big. To be precise, if A, B are non-trivial open sets, then A ∩ B will also
be a non-trivial open set.
Theorem 2.7. The Zariski topology is a topology on CN .
This is a highly non-trivial theorem. But we can prove it in C1 . By
definition, the open sets are of the form A = C\ {roots of some poly}, but
as we can write p (z) = Πi (z − ai ) for any polynomial p the result follows.
Lecture 3, 9/26/11
(5) On any set X we can produce the finite-complement topology by set-
ting T = A ⊂ X #Ac < ∞ and adding the emptyset. Note that this
topology does not derive from a metric if X is infinite, because any two
non-trivial sets have non-trivial intersection (one cannot separate open
sets). The Zariski topology with X = C is a specific example of this

topology; in fact, the finite-complement topology is sometimes called the
“Zariski topology” on X. But note that the two Zariski topologies only
coincide on C.
One can also put topologies on finite sets. Recall on any set X there are two
“extreme” topologies: the discrete topology T = P (X) (which is a metric topology)
and the indiscrete topology T = {∅, X} (which is not). On a finite set, there are
a number of topologies; the number of possible topologies grows rapidly with the
number of points. Suppose we have two points a, b. The possible topologies are
T1 = discrete
T2 = indiscrete
T3 = {∅, {a} , X}
T4 = {∅, {b} , X} .
In some sense, T3 and T4 should be the same. We’ll discuss a way to realize this
now.
Definition 2.8. A function f : X → Y is said to be continuous if f −1 (V ) ∈ TX
whenever V ∈ TY .
As we saw in the previous section, this is equivalent to the usual − δ definition
of continuity if X is a metric space.
Definition 2.9. A map f : X → Y is called a homeomorphism if it is a 1-1
correspondance and if both it and its inverse are continuous. If there exists such a
map then we say X and Y are homeomorphic.
Example 2.10. A small circle and a large circle are homeomorphic. So are a
small circle and a large square.
Proposition 2.11. If f : X → Y and g : Y → Z are continuous maps then
g ◦ f : X → Z is continuous.
Proposition 2.12. Homeomorphism on spaces is an equivalence relation.
Let’s go back to the finite set with two points. The map f : ({a, b} , T3 ) →
({a, b} , T4 ) with f (a) = b and f (b) = a is a homeomorphism. So although there
are four topologies on two points, there are only three up to homeomorphism. Of
course if we go to three points, there are many more topologies.
Exercise 2.13. How many topologies does a three point set have? How many
up to homeomorphism? (There is a trick.)
Note that the only topology on the two point set which is a metric topology is
the discrete topology. In general, a metric topology on a finite set is the discrete
topology. (Is it the only metric topology?) There is no formula in general for the
number of topologies on a finite set with n elements.
3. The Separation Axioms

One fundamental problem in topology is to classify spaces (e.g. manifolds) up
to homeomorphism. Depending on what classification means, this problem may be
algorithmically unsolvable. We’ll clarify what we mean later. One way to classify
spaces is with the so-called “separation axioms”. Here are two of the possibilites:
4. COMPACTNESS 11
• For each pair of distinct points p, q ∈ X, there is an open set with one
and not the other.
• For each pair of distinct points p, q ∈ X, there is an open set containing
p but not q (and one containing q but not p).
The first condition is very weak. The second is still weak – it is even satisfied by
the Zariski topology on a set X (take U = X\ {p} and V = X\ {q}). A stronger
condition is the so-called Hausdorff condition (T2 condition):
• Given p, q ∈ X there exist open sets U, V containing p, q with U ∩ V = ∅.
This is the condition one would use to do geometry.
Proposition 3.1. Metric spaces are Hausdorff.
But the Zariski topology is not Hausdorff, which really shows we have found a
stronger condition. Throughout we’ll assume our spaces are T2 .
Exercise 3.2. If a finite topological space is Hausdorff, then it is the discrete
topology.
Consequently, the only metric topology on a finite set is the discrete topology.
4. Compactness
Let’s go back to metric spaces to build some intuition. Let (X, d) be a metric
space. As usual, a sequence of points {xj } in X is said to be convergent if there
exists a point x∞ ∈ X so that d (xj , x∞ ) → 0 as j → ∞. In this case, we write
x∞ = limj→∞ xj or xj → x∞ .
Exercise 4.1. x∞ is unique.
Also recall
Definition 4.2. A sequence of points {xj } in X is said to be Cauchy if for
every > 0 there exists N so that if k, l > N then d (xk , xl ) < .
Exercise 4.3. If xj → x∞ then {xj } is Cauchy.
Remark. The converse does not hold in general. Consider X = (0, 1) and the
sequence xj = j −1 .
Definition 4.4. A space is called complete if a sequence is Cauchy iff it is
convergent.
Example 4.5. R is complete, but (0, 1) is not complete.
Observe, however, that (0, 1) is homeomorphic to R. (Send 1 7→ ∞ and 0 7→
−∞.)
Definition 4.6. A property of a space is called topological if it is equivalent
for homeomorphic spaces.
Thus completeness is not topological.
Definition 4.7. A metric space (X, d) is said to be pre-compact if for every
> 0 there are finitely many points x1 , . . . , xN such that each u ∈ X satisfies
d (u, xj ) < for some j ∈ {1, . . . , N }.
In the previous definition, we have X = ∪N
j=1 B (xj ). So a pre-compact X is
covered by finitely many balls of radius .
Example 4.8. R is not pre-compact, but (0, 1) is pre-compact.

So precompactness is not topological. But consider the following proposition.
Proposition 4.9. The following are equivalent for a metric space X:
(1) X is pre-compact and complete.
(2) Every sequence has a convergent subsequence.
(3) Every cover of X by open sets has a finite subcover.
What does three mean? Three says that if X = ∪α Aα for open Aα , then
actually X = ∪N j=1 Aj . In some sense, this says that the space does not go on to
infinity. A proof of this proposition can be found in Dieudonne’s Foundation of
Modern Analysis, or many other standard texts.
Definition 4.10. A space satisfying any of the properties in the previous
proposition is said to be compact.
Example 4.11. R is not compact (it is not pre-compact). (0, 1) is not compact
(it is not complete). [0, 1] is compact.
Note that condition three is the only topological condition, as it only refers to
the open sets. So for an abstract topological space we’ll use three to characterize
the compact sets.
Definition 4.12. A subset A ⊂ X of a topological space is called compact if
N
whenever A ⊂ ∪α Uα with Uα open in X then A ⊂ Uj=1 Uj .
Exercise 4.13. Let X be a topological space. A ⊂ X is compact as a subset
of X iff A is compact in the relative topology.
Proposition 4.14. If X is Hausdorff, then compact sets are closed.
Remark. The converse is false in general. Just take X = A = R.
Proof. Fix y ∈ X\A. For all x ∈ A we can find open and disjoint Ux , Vx ⊂ X
with x ∈ Ux and y ∈ Vx . Now A ⊂ ∪x Ux so by assumption A ⊂ ∪N j=1 Uj . But then
N
∩j=1 Vj is an open set which contains y, and is disjoint from A by construction,
hence X\A is open.
Proposition 4.15. If X is compact, then A ⊂ X is closed iff it is compact.
Proof. We have compact implies closed. If A is closed, let {Uα } be an open
cover of A. Then {Uα , X\A} is a cover of X, so {U1 , . . . , UN , X\A} covers X.
Hence {U1 , . . . , UN } covers A, so we are done.
This has some useful implications for topology.
Proposition 4.16. On Rn , compact means bounded and closed.
Proof. It’s easy to see compact implies bounded and closed. Conversely, if A
is bounded, then it sits inside a closed ball of some radius. This ball is compact, so
by the previous proposition, A is compact as well.
Proposition 4.17. Suppose f : X → Y is continuous. If X is compact, then
Y is compact as well.
Proposition 4.18. Suppose X, Y are compact and Hausdorff. If f : X → Y
is a continuous 1-1 correspondence, then f is a homeomorphism.
Exercise 4.19. Prove the previous two propositions.
5. THE PRODUCT TOPOLOGY 13
5. The Product Topology

If (A, TA ) and
(B, TB
) are (disjoint) topological
spaces, we give A∪B a topology
by setting T = U ∪ V U ∈ TA , V ∈ TB .
Exercise 5.1. If we have f : A ∪ B → C, then f is continuous iff f |A and f |B
are.
Products arise in the dual situation. Suppose A, B are topological spaces. How
do we topologize A × B? Given maps f : C → A and g : C → B, we want to
combine these to form the product map (f, g) : C → A × B. Then we want to say
that (f, g) is continuous iff f, g are continuous. This tells us the topology we want
on the product space. Is there such a topology? A first guess might be to take
T = {U × V } with U open in A and V open in B. But this would only give us
rectangular open sets. (Consider the picture for R × R and U, V = (0, 1). Should
2
sets like (0, 1) be the only open sets?)
Definition 5.2. Given a topological space X with a topology T , a basis for
T is a subset B ⊂ T such that every non-∅, non-X set in T is a union of sets in B.
Example 5.3. The most natural example is when T is a metric topology. Then
the open balls form a basis; the open balls of radius 1/n for n = 1, 2, . . . form a
basis as well.
Proposition 5.4. A collection B is a basis for a topology on X iff given U, V ∈
B, U ∩V is a union of sets in B. Equivalently, B is a basis if for any point p ∈ U ∩V ,
there exists W ∈ B with p ∈ W ⊂ U ∩ V .

Example 5.5. If X is the discrete topology, we could take B = {u} u ∈ X .
5.6. The
Definition on A × B is the topology generated by
product topology
the basis B = U × V U ∈ TA , V ∈ TB .
Proposition 5.7. If f : C → A and g : C → B, then f, g are continuous iff
(f, g) : C → A × B is continuous.
Proof. Notice that A × B comes with the coorindate projections p1 , p2 onto
A, B. First we claim p1 , p2 are continuous. To see this, let U be open in A, then
p−1
1 (U ) = U × B which is open. It’s the same for p2 . Now if (f, g) are continuous,
then f = p1 ◦ (f, g) and g = p2 ◦ (f, g) are the composition of continuous functions
and hence continuous.
−1
Conversely, suppose f, g are continuous. We only need check (f, g) (U × V )
−1
is open, but (f, g) (U × V ) = f −1 (U ) ∩ g −1 (V ) and so we’re done.
Note. We have developed a satisfactory theory of continuity for maps into
product spaces. There is no characterization in general for maps from product
spaces. This extends beyond topology: one can take a whole course on multivariable
calculus, where the results do not follow simply from single-variable calculus!
For the infinite product Πα∈I Xα , the first guess would be to say that the open
sets are of the form Πα∈I Uα with Uα open in X. This fails for the same reason as
in the finite product.
Definition 5.8. The product topology on Πα∈I Xα has as basis sets of the form
U1 × U2 × · · · × Uk × Πα6=1,...,k Xα .
Theorem 5.9. The product of compact Hausdorff spaces is also compact Haus-
dorff.
The proof of this in the case of two sets is not hard, and neither is the proof in
the finite case. But the infinite product requires some non-trivial logic.
Example 5.10. The solid cubes are compact. Since solid cubes and solid
spheres are homeomorphic, solid spheres are compact as well.
Many spaces (not most) are product spaces.
Examples 5.11.
(1) I × I ' D2
(2) S 1 × I, the cylinder.
(3) S 1 × S 1 ' T 2 ⊂ R3 . This is the only closed two-manifold which has a
product decomposition.
(4) X × I ' X × S 1 by a gluing construction. For S 2 × I we would have to
glue the inner surface to the outer surface. This is possible in R4 .
Lecture 4, 10/3/11
6. Connectedness
What do we mean when we say a space is disconnected? Think of two disjoint
blobs A, B which together make up a space X. This is the intuitive idea. How do
we formalize this? Note that both A and B are open in the topology of X (each
point in A has a ball around it, and the same with B), but so they are both closed.
Definition 6.1. A space X is said to be connected if we cannot decompose X
as a union of disjoint, non-trivial open subsets A, B ⊂ X. Equivalently, there does
not exist A ⊂ X with A 6= φ, X and A is both open and closed.
There is a stronger notion of connectedness. Intuitatively, a space should be
path-connected if any two points have a path between them. What is a path?
Definition 6.2. A (parameterized) path in a space X is a continuous map
ω : I = [0, 1] → X. If u = ω (0),v = ω (1) then ω is said to be the path from u to v,
which themselves are the initial and terminal points of ω (the endpoints of ω).
Given a path ω from u to v in X and a path γ from v to w in X, we can splice
them together to form a new path ω·γ : I → X. But note we need to reparametrize:
(
ω (2t) 0 ≤ t ≤ 12
ωγ (t) = .
γ (2t − 1) 21 ≤ t ≤ 1
We have what begins to look like an algebra among paths – we can combine two
paths to form a new path. But there is a difficulty, as the product is only defined
if ω (1) = γ (0).
Definition 6.3. A space X is said to be path-connected if for any points
u, v ∈ X, there exists a path from u to v.
Given a space X, define a relation ∼ among points of X by setting u ∼ v if
there exists a path from u to v in X. Observe that ∼ is an equivalence relation
among points. We have u ∼ u by the trivial path eu : I → {u} with eu (t) = u for
all 0 ≤ t ≤ 1, and u ∼ v =⇒ v ∼ u for if we have ω from u to v then we have
ω −1 (t) = ω (1 − t) from v to u. Finally, we have u ∼ v + v ∼ w =⇒ u ∼ w by the
7. QUOTIENT SPACES AND GLUING 15
splicing technique above. So X decomposes into equivalence classes, which we call

the path-components of X.
Proposition 6.4. Path-connected implies connected.
Remark. The converse is false. Consider the space consisting of the points
x, sin x1 ∩ {y-axis} ⊂ R2 .

Proof. If not, then we can write X = A ∪ B for A, B 6= ∅, A ∩ B = ∅, and

A, B open. Let u ∈ A and v ∈ B and let ω be a path in X from u to v. Consider
ω −1 (A) and ω −1 (B). These are open disjoint subsets which seperate I = [0, 1],
which is not possible.
The situation can be remedied as follows. Recall a neighborhood about a point
is an open set containing that point.
Definition 6.5. Say X is locally path-connected if each point u ∈ X has a
path-connected neighborhood.
Proposition 6.6. If X is connected and locally path-connected, then it is path-
connected.
Remark. This result applies to manifolds, for instance.
Proof. Decompose X into path-components. The components are open sub-
sets of X, by local path-connectedness. So X decomposes into disjoint open subsets.
As X is connected, all but one component is null, and the remaining one is all of
X. So X is path-connected.
7. Quotient Spaces and Gluing

We want to be able to take an interval from 0 to 1 and glue the points 0, 1
together with an equivalence relationship, so that we get a circle. That is, we want
to say S 1 = I/0 ∼ 1. We’ll formalize this notion now.
Given a space X and an equivalence relation ∼ among points of X, we can
form a new space, called the quotient space X/ ∼, whose points are the equivalence
classes of ∼. Given a ∈ X, we’ll write [a] to mean its equivalence class.
Note. When writing an equivalence relationship ∼, we often omit that u ∼ u.
Given an equivalence relation ∼, we always have a quotient map q : X → X/ ∼
which maps m 7→ q (u) = [u]. So we give X/ ∼ a the quotient topology by setting
U ⊂ X/ ∼ to be open if q −1 (U ) is open in X. This makes q is continuous.
Example 7.1. S 1 = I/0 ∼ 1. This is easy to see for the quotient map is a
1-1 correspondence between compact, Hausdorff spaces and is clearly continuous in
one direction.
Now we can describe a nice way to construct new spaces from old, by “gluing”.
Let A, B be disjoint spaces and let f : C → A and g : C → B be maps from a third
space C into A and B. We might insist that there is a copy of C in A, B, i.e. that
f and g are injective. But this is too restrictive. Similarly, it would be too much
to allow both to be not injective. So let’s only insist, say, that f is injective. Given
all this, we can form a new space X by gluing:
X = A ∪ B/f (u) ∼ g (u) for u ∈ C.
If X is a space formed in such a way, we write X = A ∪C B to mean A glued to B

along C.
Example 7.2.
(1) S n = D+
n n
∪S n−1 D− .
(2) RP = S /u ∼ (−u). The quotient map q : S 2 → RP 2 is two-to-one.
2 2
(3) RP 2 = Mobius ∪S 1 disc.

(4) RP 2 = D 2 1
/u ∼ (−u) for u ∈3 S .
2
(5) RP = lines
through 0in R ←→ pairs of antipodal points on S 2 .
(6) RP n = lines through 0in Rn+1 ←→ S n /u ∼ (−u). Then, RP 0 =
{1 point} and RP 1 = S 1 /u ∼ (−u) = D+ 1
/endpoints identified. So
RP w S . This looks familiar! Consider the map f (z) = z 2 in the
1 1
complexplane, which maps S 1 2-to-1

to S .
1
n n+1
(7) CP = C-lines through 0in C . It turns out that CP 1 w S 2 .
CHAPTER 3
The Fundamental Group
1. Homotopy of Paths
Consider a two dimensional space X with a hole cut out. Given points u, v ∈ X
on opposite sides of the hole, there is a family of paths which can be deformed into
each other. But there is a second such family, and any path in that second family
cannot be deformed into a path in the first family. Let’s formalize this.
Definition 1.1. Given two paths α, β from u to v in X, a homotopy of α to
β is a continuous map H : I × I → X with
H (0, s) = u, 0 ≤ s ≤ 1
H (1, s) = v, 0 ≤ s ≤ 1
H (t, 0) = β (t) , 0 ≤ t ≤ 1
H (t, 1) = α (t) , 0 ≤ t ≤ 1.
There is another way to think of what homotopy means. Think of Hs as the
the path from u to v given by Hs (t) = H (t, s). So Hs is a parameterized family of
paths (from u to v) from H0 = β to H1 = α.
Exercise 1.2. The relation α ∼ β (α ∼ β if α is homotopic to β) is an
h
equivalence relation among paths from u to v. (Hint: The tricky one is transitivity.)
One of our goals in this course is to decide when paths are homotopic to each
other. We’re not ready for that now, though.
n
Exercise
1.3. Recall
a subset A ⊂ R is said to be convex if for all p, q ∈ A
we have tp + (1 − t) q 0 ≤ t ≤ 1 ⊂ A. Show that if A is convex, then any two
paths from u to v in A are homotopic.
Recall the product operation between paths. A nice fact is that the product
operation is compactible with homotopy.
Proposition 1.4. Suppose α ∼ β with
h
α (0) = β (0) = u
α (1) = β (1) = v
and γ ∼ δ with
h
γ (0) = δ (0) = v
γ (1) = δ (1) = w,
Then α · γ ∼ β · δ.
h
17
18 3. THE FUNDAMENTAL GROUP
Exercise 1.5. Prove the proposition above.

Remark. But associativity fails. Even if (α · β) · γ makes sense (the endpoints
have to match), it is usually the case that (α · β) · γ 6= α · (β · γ). The images will
be the same, but the parametrizations are different.
Proposition 1.6. Under the endpoint matching conditions above, (α · β) · γ ∼
h
α · (β · γ).
Exercise 1.7. Prove the proposition above.
So we have associativity up to homotopy of paths. Now if we are given a path
α from u to v in X, then its inverse α−1 will be given by α−1 (t) = α (1 − t), the
reverse path from v to u. Note that α · α−1 is a path from u to itself.
Definition 1.8. A loop is a path from a point u to itself. It is said to be fixed
at the point u.
Proposition 1.9. Let α be a path from u to v in X. Then α · ev ∼ α ∼ eu · α.
h h
Exercise 1.10. Prove the proposition.

Here is a slightly less obvious proposition, where homotopy becomes interesting.
Proposition 1.11. α · α−1 ∼ eu .
h
This result really cannot be acheived by reparameterization.

Proposition 1.12. Let π1 (X, u) = {loops in Xbased at u} / ∼. This is a
h
group.
Definition 1.13. π1 (X, u) is called the (first) fundamental group for X at u.
As the notation suggests, there are π2 , π3 , . . . – we’ll get to these later. We’ll
see that π1 is a measure of the one-dimensional holes in a space.
Remark. We proved homotopy is compatible with the product. So [αβ] =
[α] · [β]. But [α] · [β] 6= [β] · [α] in general. We’ll see why later on by considering a
figure eight.
Here is the easiest example of a fundamental group:
Example 1.14. If X is convex, then π1 (X, u) = {[eu ]} for all u ∈ X.
Definition 1.15. A space is called simply connected if it is connected and if
π1 = {0}.

Example 1.16. π1 S 1 = Z. Proof is by winding number.
Lecture 5, 10/17/11
2. Independence of Basepoint
Having defined the (first) fundamental group π1 (X, x) of a space X with base-
point x, the first natural question to ask is: how important is the choice of base-
point? The answer is not at all, so long as we stay within a single path-connected
component of X.
Theorem 2.1. If x, x0 are points in the same path-component of X, then
π1 (X, x) ' π1 (X, x0 ).
2. INDEPENDENCE OF BASEPOINT 19
As a result,
Corollary 2.2. If X is path-connected, π1 (X, x) is independent of the choice
of x ∈ X.
Remark. As we’ll see, there is not a natural isomorphism. We’ll have to make
a choice. But there is not always a natural isomorphism: Z ' Z in (at least) two
ways (via x 7→ x + 1 or x 7→ x − 1).
To prove the theorm, we’ll first need some ideas from algebra.
Definition 2.3. If G is a group, then φ is an automorphism of G if φ : G → G
is an isomorphism. We’ll write Auto (G) to mean the set of automorphisms of G to
itself.
Example 2.4. x 7→ x + 1 and x 7→ x − 1 on Z are automorphisms.
Proposition 2.5. Auto (G) forms a group under composition.
Example 2.6. Auto (Z) ' {±1} ' Z2 . Also, Auto (Zp ) ' Z∗p (' Zp−1 ). The
last result is a theorem from number theory.
As in the previous example, given a ring R we write R∗ to mean the multi-
plicative invertible elements of R. So R∗ forms a group under multiplication.

Definition 2.7. G is cyclic if it is generated by one element, G = g k k ∈ Z .
Proposition 2.8. If G is infinite cyclic, then G ' Z±1 . If G is finite cyclic,
then G ' Zm+ (generated by 1).
Definition 2.9. Given a group G and an element g ∈ G, define conjugation
of u by g to be the map φg : G → G, u 7→ gug −1 . We call φg (u) the conjugate of
u by g.
Proposition 2.10. φg = idG if G is abelian.
We call conjugation by an element an inner automorphism of G. The set of all
inner automorphisms of G is denoted Inner (G).
−1
Exercise 2.11. φg (uv) = φg (u) φg (v). Also φg is 1-1, as φg−1 = (φg ) .
Exercise 2.12. φhg = φh ◦ φg .
As a consequence, we get
Proposition 2.13. Inner (G) ⊂ Auto (G).
Exercise 2.14. G is abelian iff Inner (G) = {e}.
We can relate the group of inner automorphisms to G very easily, by the map
Φ : G → Inner (G), g 7→ φg .
Exercise
2.15. Show
Φ is a homomorphism. Also show Ker (Φ) = Z (G) =
g ∈ G gh = hg ∀h ∈ G .

Definition 2.16. Z (G) = g ∈ G gh = hg ∀h ∈ G is called the center of G.
G is abelian iff Z (G) = G. And if A is abelian then Z (G × A) ⊃ A. So we can
easily produce centers out of any group in this way.
Now let’s go back to topology. Our goal is to see why the group of homotopic
loops is independent of the basepoint. The proof is almost geometric.
0
Proof of theorem.
0
−1
Draw a path γ from x to x. Define a map χγ : π1 (X, x) →
π1 (X, x ), [α] 7→ γαγ . (This is not a conjugation, since γ is not in the group.
But it is close.) Now it’s easy to check
(1) This depends only, up to homotopy of loops, on the homotopy class of α.
So it is well-defined.
(2) It
it a−1homomorphism. Check χγ(αβ) = χγ (α) χγ (β). The LHS is
and the RHS is γαγ −1 γβγ −1 .

γαβγ
(3) It has an inverse, hence χγ is an isomorphism. This proves the proposition.

−1
Exercise 2.17. Check step three. Specifically, show that (χγ ) = χγ −1 .
How canonical was the choice of isomorphism in the proof? Notice χγ depends
only on the choice of the homotopy class of paths from x0 to x. One can check that
this isomorphism of π1 (X, x) and π1 (X, x0 ) is “almost” canonical, in that any two
isomorphisms differ by an inner automorphism. If χγ1 , χγ2 : π1 (X, x) → π1 (X, x0 )
are two choices of isomorphism via paths γ1 , γ2 from x0 to x, then the composite
χ−1
γ2 χγ1 is an automorphism of π1 (X, x).
Claim. This is an inner automorphism, in fact given by χγ2 γ −1 . So χγ1 =

1
(inner) χγ2 .
Exercise 2.18. Prove the claim. This is a harder exercise than the others.
We’ll never use this fact in this course. But it does come up sometimes.
Recall we called X simply connected if X is path-connected and if π1 (X, x) =
{e} = {0}. (The first notation indicates multiplication, and the second addition.)
And we saw that if X is convex, then π1 (X, x) = {0}. More generally, we have
Definition 2.19. X ⊂ Rn is called star-like if it has a star-point x ∈ X such
that for any u ∈ X, the ray tu + (1 − t) x, 0 ≤ t ≤ 1 lies in X.
Example 2.20. If X is star-like, then π1 (X, x) = {0} for all x ∈ X. First start
at the starpoint then use the fact that there is one path-component.
3. Comparing Spaces Algebraically

Now given spaces and basepoints we can produce groups. The next question
is: given a function between such spaces, can we produce a homomorphism? We’ll
write X,x to show the basepoint x explicitly.
Definition 3.1. A continuous map f : X,x → Y,y is called basepoint-preserving
if f (x) = y.
We’ll use the notation space+ to mean a space with a basepoint. We’ll write
map+ to mean a basepoint preserving map.
Let f be a map+ from X,x to Y,y . Then f induces a homomorphism f∗ :
π1 (X, x) → π1 (Y, y) via the following
Definition 3.2. Given a loop α in X based at x, define the induced map
f∗ : π1 (X, x) → π1 (Y, y) by f∗ [α] = [f ◦ α].
This is well-defined, for α ∼ β in X implies f ◦ α ∼ f ◦ β in Y . To see this,
h h
take the homotopy and compose it with f .
3. COMPARING SPACES ALGEBRAICALLY 21
Claim. f∗ is a homomorphism, i.e. f∗ ([α] · [β]) = f∗ ([α]) · f∗ ([β]).

Proof. The claim holds on the level of loops, so it holds on the homotopy
classes.
Now we have a way to linearize the problems of topology. We have a way to
turn the problems of topology into problems of groups, and the problems of maps
between spaces to homomorphisms between groups. This subject can get quite
close to linear algebra.
Example 3.3. Consider idX : X → X. Clearly (idX )∗ [α] = [α], as it holds on
the level of paths.
As we saw in the previous example, our construction has the fundamental
property that
(3.1) idπ1 (X,x) = (idX )∗ : π1 (X, x) → π1 (X, x) .
A second fundamental property is compatibility with composition: If f : X,x → Y,y
and g : Y,y → Z,z are map+ s, then
(3.2) (g ◦ f )∗ = g∗ ◦ f∗ .
This is a consequence of associativity of composition. Relations (3.1) and (3.2)
are essential to our algebraic view of topology. Whenever we look at topology in
an algebraic way, we’ll look for these properties. Some people call this naturality;
others call it functoriality.
Example 3.4. Consider the map idS 1 : S 1 → S 1 . This is a loop in the circle,
1
so [idS 1 ] ∈ π1 S,1 . (We’re using complex numbers in the plane here to denote the

basepoint.) There is an isomorphism by winding number between π1 S 1 , 1 and
Z. The isomorphism sends the generator [idS 1 ] to 1.
There is a more geometrical way to see this. Given z ∈ S 1 write z = e2πiθ and
take the map gk (z) = z k from S 1 to itself. Then gk e2πiθ = e2πikθ is the function
which winds around the circle k times. If we regard [gk ] as an element of π1 S 1 , 1 ,
then
k
[gk ] = [g1 ] · [g1 ] · · · [g1 ] = [g1 ] .
| {z }
k times

And here, g1 = idS 1 . If we write it additively, then we have winding # : π1 S 1 , 1 →
Z, [gk ] 7→ k · 1 = k.
Now if we regard gk as a map on S 1 , then (gk )∗ : π1 S 1 → π1 S 1 has
h i
k
(gk )∗ [α (t)] = (α (t)) , 0 ≤ t ≤ 1.
Additively, we have (gk ) : Z → Z, 1 7→ k. Note (gk )∗ is only a homomorphism (it
is not invertible, think of k = 3).
Theorem 3.5. π1 (X × Y ) = π1 (X) × π1 (Y ) (i.e. they are isomorphic).
Proof. Let p1 , p2 be the coordinate projections from X × Y onto X, Y . Then
we have induced projections (p1 )∗ : π1 (X × Y ) → π1 (X) and (p2 )∗ : π1 (X × Y ) →
π1 (Y ). Then we have a natural map (p1 )∗ × (p2 )∗ : π1 (X × Y ) → π1 (X) × π1 (Y ).
It’s automatically a homomorphism, since it is a product of homomorphisms. Now
we claim this is an isomorphism. Given paths α, β : I → X, Y we can get α × β :
I → X × Y , t 7→ (α (t) , β (t)) easily. Just check p1 × p2 (α × β) = α × β then
apply this to homotopy classes to get (p1 )∗ × (p2 )∗ ([α × β]) = ([α] , [β]). This is
surjectivity.
For injectivity, we apply this principle to homotopy classes (go one dimension
higher, it’s a general principle in topology when we want to check injectivity). We
only need to check that ker (p1 )∗ × (p2 )∗ = {0} = {e}. Suppose for some loop in
the product α × β : I → X × Y we have (p1 )∗ × (p2 )∗ ([α × β]) = e and so
(p1 )∗ ([α × β]) = e
(p2 )∗ ([α × β]) = e.
Hence, [α] = e in π1 (X) and [β] = e in π1 (Y ), so we have a homotopy Hx in X of
α to the constant loop at x and similarly a homotopy Hy in Y of β to the constant
loop at y. Thus Hx × Hy is a homotopy of α × β in X × Y to the constant loop at
(x, y).
Note. The proofs of surjectivity and injectivity followed the same pattern.
Surjectivity was done at the level of paths; injectivity was done at the level of
homotopies (one dimension higher).
Now we can compute some fundamental groups.

Example
3.6. Consider that T 2 ' S 1 × S 1 . So π1 S 1 × S 1 = π1 S 1 ×
π1 S 1 = Z × Z =Z2 . This is the group of lattice points in the plane. Similarly,
π1 S 1 × · · · × S 1  = Zn . In fact, the same proof carries through for infinite prod-

| {z }
n times
ucts. (Think about how the product topology was defined for infinite products.)
Now suppose we want to describe maps from the n-torus T n = S 1 × · · · × S 1 to
| {z }
n times
the m-torus T m = S 1 × · · · × S 1 . For the circle to the circle, the maps were given
| {z }
m times
by the gk ’s. On the torus, we can map each of the circles to each of the circles.
Given an n×m matrix with integer entries A, we can produce a map gA : T n → T m
so that (gA )∗ : π1 (T n ) = Zn → π1 (T m ) = Zm is just matrix multiplication by A.
We’ll see an explicit version of this later on. Here is an example.
Example 3.7. Consider the map f : S 1 × S 1 → S1 × S1
which has f (x, y) =
0 1
(y, x). Then f∗ : Z ⊕ Z → Z ⊕ Z is multiplication by .
1 0
Lecture 6, 10/24/11
Now suppose X,x is a set+ and let [α] ∈ π1 (X, x). What does it mean for [α]
to be trivial? Let’s give a geometric answer.
Definition 3.8. Let α : S 1 → X be a (continuous) map and let i : S 1 ,→ D2
be inclusion. If there exists (continuous) a : D2 → X so that a ◦ i = α, then we say
α can be extended to a.
Proposition 3.9. [α] = e iff α : S 1 → X can be extended to a map a : D2 →
X.
Proof. There is a homotopy H : I × I → X between α and ex . Think of I × I
as a box with α on the top and x on the left, right, and bottom sides. Identify the
4. HOMOTOPY OF MAPS 23
left, right, and bottom sides and let q : I 2 → I 2 / ∼ be the quotient map. Let H
be so the diagram commutes.
I2
H /X
z<
zz
z
q
zz
zz H
I 2/ ∼

Since I 2 / ∼= D2 in fact we can think of H as a map from D2 to X. Then H S 1 = α
so this is the extension we want.
The converse is proved similarly.
Exercise 3.10. Work out the details above.
Given an inclusion i : A,x ,→ X,x , note that i∗ : π1 (A, x) → π1 (X, x) need not
be injective, and may even be zero.

Example
3.11. If i : S 1 ,→ D2 is the inclusion, then i∗ : π1 S 1 = Z →
π1 D2 = 0 must be the zero map.
But inclusions can still useful.
Definition 3.12. Given i : A ,→ X, a retraction is a map r : X → A with
r ◦ i = idA . If there is a retraction r : X → A, we say A is a retract of X.
Example 3.13. Let X,x and Y,y be disjoint and define X∨ Y = X ∪ Y /x ∼ y.
Let i : X ,→ X∨ Y be inclusion, q : X∨ Y → X∨ Y /Y the quotient map, and
g : X∨ Y /Y → X the natural identification.
i /
XG X∨ Y
GG
GG
G
g GGG
q
#
X∨ Y /Y
Given the above, the map r = g ◦ q −1 is a retraction. Check that r ◦ i = idX .
As usual, a retract+ map will be a retract that preserves basepoint.
Proposition 3.14. Suppose A is a retract+ of X and let i : A ,→ X be the
inclusion. Then the induced map i∗ : π1 (A) ,→ π1 (X) is injective.
Example 3.15. i∗ : π1 (X) ,→ X∨ Y is injective.
Proof. (D2) r ◦ i : idA so (r ◦ i)∗ = (idA )∗ and hence r∗ ◦ i∗ = idπ1 (A) by
naturality. So i∗ is injective.
Example 3.16. As a result, S 1 is not a retract of D2 . We’ve already seen the
importance of this fact.
4. Homotopy of Maps
(D3) Suppose we have a space X,x and consider Y = X × I/x × I. There are
(at least) two inclusions, j : X ,→ Y which maps to X × {1} and i : X ,→ Y which
maps to X × {0}. The picture suggests that for [α] ∈ π1 (X, x), i∗ ([α]) = j∗ ([α])
in π1 (Y ). So we expect i∗ = j∗ . We’ll capture this in general with a new notion of
deformation for maps.
Definition 4.1. Given two maps f, g : X → Y , a homotopy of f to g is a map

H : I × X → Y which satisfies H (u, 0) = g, (u), and H (u, 1) = f (u) for all u ∈ X.
We write f ∼ g if there is a homotopy between them.
h
Remark. As with paths, another way of thinking about homotopy is to con-

sider the family of maps Ht (u) = H (u, t). Since H is a homotopy, Ht has H0 = g
and H1 = f .
Proposition 4.2. ∼ is an equivalence relation on maps from X → Y .
h
Topologists use the following notation:

[X, Y ] = {continuous maps from Xto Y } / ∼ .
h
As we’ll see, [X, Y ] will turn out to be countable for reasonable spaces. In many
cases of interest, [X, Y ] will also turn out to be a group. Let’s also introduce the
notation f ∼ g to mean f (x) = y, g (x) = y, and there exists a homotopy+ H
h+
with H (x, t) = y for all 0 ≤ t ≤ 1. And then
[X, Y ]+ = [X, Y ] = {continuous maps+ from Xto Y } / ∼ .
h+
These considerations enormously shrink the subject of maps from X to Y .

Example 4.3. If Y ⊂ Rn is convex, then [X, Y ] (or [X, Y ]+ ) has just one
element, represented by the constant map to the basepoint of Y . Proof is by linear
interpolation: generic f, g are homotopic by tf + (1 − t) g.
But we still retain useful information, topologically speaking.
Proposition 4.4. If f ∼ g, then f∗ = g∗ as homomorphisms from π1 (X, x)
h+
to π1 (Y, y).
Proof. The idea is that if we can deform the whole space, then we can deform
loops. We want to show that f∗ ([α]) = g∗ ([α]) for [α] ∈ π1 (X, x). So let H be a
homotopy+ from f to g, and let L : X × I → X × I be given by L (u, t) = (α (u) , t).
Then H ◦ L is a homotopy (in the sense of loops) between f ◦ α to g ◦ α. So
[f ◦ α] = [g ◦ α], and so f∗ ([α]) = g∗ ([α]).
Definition 4.5. A map+ f : X → Y is said to be null-homotopic if it is

homtopic+ to the constant map (to the basepoint of Y ).
These maps play the role of zero maps.
Proposition 4.6. If f is null-homotopic, then f∗ is the zero map.
Proof. Let c be the constant map. If f ∼ c, then f∗ = c∗ = 0.
h+
Example 4.7. If Y (or X) is convex, then all maps of X to Y are null-

homotopic.

Notice that in our notation, π1 (X, x) = S 1 , X + . What makes π1 so special
is that it is a group.

Definition 4.8. The higher homotopy groups are the sets πk (X, x) = S k , X + .
5. HOMOTOPY OF SPACES 25
The name here is due to Hurewicz, who discovered them.1 A priori it’s not
obvious these are groups, but we’ll prove it, and in fact we’ll also see they are
abelian for k > 1. As k increases, πk provides a way to study higher dimensional
(spherical) holes.
Definition 4.9. A is called a deformation retract(+) of X if there is an inclusion(+)
i : A ,→ X and a map(+) r : X → A so that r ◦ i ∼ idA .
h (+)
Proposition 4.10. Suppose A is a deformation retract+ of X and let i : A ,→

X be the inclusion. Then the induced map i∗ : π1 (A) → π1 (X) is an inclusion.
Example 4.11. Suppose A is convex and i : A ,→ X is inclusion.
Claim. A is a deformation retract of X.
Proof. Take r : X → A to be a constant map. As A is convex, any two maps
are homotopic. So r ◦ i ∼ idA .
h
5. Homotopy of Spaces
For the purposes of algebraic calculation in topology, we need to loosen the no-
tion of homeomorphism. Recall a continuous map f : X → Y is a homeomorphism
if there is a continuous map g : Y → X so that g ◦ f = idX and f ◦ g = idY .
Definition 5.1. A map f : X → Y is called a homotopy equivalence(+) (h.e.)
if there is a map(+) g : Y → X so that g ◦ f ∼ idX and f ◦ g ∼ idY . If there
h (+) h (+)
is such a map, then we say X and Y are homotopy equivalent(+) spaces.
Proposition 5.2. Homotopy equivalence is an equivalence relation among
spaces.
Exercise 5.3. Check this in detail. The main point (as usual) is transitivity.
Proposition 5.4. If f : X → Y is a homotopy equivalence, then the induced
map f∗ : π1 (X) → π1 (Y ) is an isomorphism.
Proof. Suppose g ◦ f ∼ idX and f ◦ g ∼ idY as in the definition. Then
h+ h+
g ◦ f ∼ idX and (g ◦ f )∗ = (idX )∗ so by naturality g∗ ◦ f∗ = idπ1 (X) . It’s the same
h+
for f ◦ g.
In light of this,
Definition 5.5. Spaces h.e. to X are said to have the same homotopy type as
X.
Exercise 5.6. [X, Y ] depends only on the homotopy types (respectively) of X
and Y .
Remark. The converse does not hold. There are counterexamples showing
πk (X) and πk (Y ) can be the same for all k even though X and Y are not homotopy
equivalent.
1 is an interesting non-trivial map from S 3 to S 2 , discovered by Hurewicz, which shows
` There
π3 S 2 6= 0.
´
Theorem 5.7 (J.H.C. Whitehead). Suppose f : X → Y is a map between

“reasonable” path-connected spaces. Then f∗ : πk (X) → πk (Y ) is an isomorphism
iff f is a homotopy equivalence.
The hard part is the converse. We won’t do this here.
Homotopy equivalence is much weaker (even suprisingly so) than homeomor-
phism.
Examples 5.8.
(1) Rn is homotopy equivalent to a point. In general, suppose X is convex,
then X will be homotopy equivalent to a point. Indeed, given p ∈ X, let
f : X → {p} be the constant map and let g : {p} → X be a map to any
point in X. Then f ◦ g = id{p} , and g ◦ f maps X to (a point in) X so
that g ◦ f ∼ idX .
h
(2) R2 \ {p} is homotopy equivalent to S 1 .
Exercise 5.9. Check this. Let i : S 1 ,→ R2 \ {0} be inclusion and let
f : R2 \ {0} → S 1 be given by f (v) = v/||v||.
(3) Rn \ {p} is homotopy equivalent to S n−1 .
(4) Rn \ {p1 , p2 } with p1 6= p2 is homotopy equivalent to a “figure eight”.
(5) Rn \ {p1 , . . . , pk } with pi 6= pj for i 6= j is homotopy equivalent to ∨k S n−1 .
(6) R2 \ {0} ' S 1 × R+ which is homotopy equivalent to S 1 × {p} by the next
exercise.
Exercise 5.10. If X is h.e. X 0 and Y is h.e. Y 0 , then X × Y is h.e.
X 0 × Y 0.
Similarly, Rn \ {p} ' S d−1 × R+ which is homotopy equivalent to
d−1
S × {p}.
Next time, we’ll see that π1 (S n ) = 0 for n > 1.
Theorem 5.11 (Poincare Conjecture). If M is a closed, simply-connected,
three-dimensional manifold, then M ' S 3 .
It’s not that hard (relatively, at least) to prove
Proposition 5.12. If M is a closed, simply-connected, three-dimensional man-
ifold, then M is homotopy equivalent to S 3 .
6. Combinatorial Group Theory and Van Kampen’s Theorem

Lecture 7, 10/31/11
We already know how to compute the fundamental group of a product.
Proposition 6.1. π1 (X × Y ) ' π1 (X) × π1 (Y ).
Now let’s consider a different type of construction. Suppose we have X = A∪B
with A ∩ B 6= ∅. Can we describe π1 (X) in terms of π1 (A) and π1 (B)? First we
need some group theory.
Definition 6.2. If G is a group and S ⊂ G is a set of elements of G, the
subgroup generated by S (in G) is the smallest subgroup of G containing S.
The definition makes sense by the following easy proposition.
Proposition 6.3. If S1 , S2 ⊂ G are subgroups, then so is S1 ∩ S2 .
6. COMBINATORIAL GROUP THEORY AND VAN KAMPEN’S THEOREM 27
So the smallest subgroup of G containing S is the intersection of all such sub-

groups. And the intersection is non-empty, for at least S ⊂ G.
Can we explicitly describe the subgroup H which is generated by S? Suppose
S = {s1 , s2 , . . . , sk , . . . }, then
H = all finite products s±1 ±1 ±1

j1 sj2 · · · sjm sji ∈ S .
Here we call the product s±1 ±1 ±1

j1 sj2 · · · sjm a word. So H is the subgroup which consists
of all words built from elements in S. And if H = G then we say S generates the
group G. S is the group-theory equivalent to the basis from linear algebra.
Example 6.4. Z is generated by {+1}, {−1}, {2, 7}, etc.
Now suppose A,x , B,x are open in X,x and suppose X = A ∪ B. Then a natural
guess is that π1 (X, x) is generated by π1 (A, x) and π2 (B, x). Note this implies
that if S is a set generated from π1 (A) and T a set generated from π1 (B), then
S ∪ T generates all of π1 (X). The next example shows this guess is incorrect.

Example 6.5. Let A = S 1 / eiθ , θ ∈ [−π/4, π/4] and B = S 1 / eiθ , θ ∈ [3π/4, 5π/4] .
Then there are loops in A ∩ B which are not in A or in B.
The problem here is path-connectedness (In the example above, A ∩ B has two
path-components.)
Theorem 6.6 (1/2 of van Kampen’s Theorem). Let X,x = A,x ∪B,x where A, B
are open in X, and suppose A ∩ B is path-connected. Then π1 (X, x) is generated
by π1 (A) and π1 (B).
Corollary 6.7. Let A, B as in the above theorem. Then if A, B are simply
connected, then π1 (A ∪ B) = 0.
Examples 6.8.
(1) Thus S n is simply connected for n > 1. Write S n = A ∪ B where A is
a bit more than the upper hemisphere and B a bit more than the lower
hemisphere. We can choose A, B to be open sets so that A ∩ B is the
equatorial band. Then A ∩ B ' S n−1 × (−1, +1) is path-connected, and
since A, B ' Dn , A, B are simply connected. Now apply the corollary.
(2) Πki=1 S ni for ni > 1 is simply connected. For example, S 2 × S 2 is simply
connected.
Proof of theorem. The proof goes as follows:
(1) Break a loop based at x ∈ A ∩ B into a finite number of pieces which are
(alternately) contained in A ∩ B.
(2) Modify the resulting paths to make them into loops based at x.
The details are not hard. For step one, say γ : I → X is a loop based at x and
consider γ −1 (A) and γ −1 (B). These are open subsets of I, each of which can be
written as a union of open intervals. By compactness, there is a finite subcover. We
can combine sequential ones belonging all in γ −1 (A) (similarly for those in γ −1 (B)).
So γ is decomposed into a series of sequentially overlapping open intervals, each
contained in γ −1 (A) or in γ −1 (B) (and we can assume these alternate). Finally,
picking points in the sequential overlaps (which lie in γ −1 (A) ∩ γ −1 (B)), we get
γ = Πki=1 γi where γi is a path in A or B, alternately.
Now step two. Suppose the points we chose above correspond to times 0 =
t0 < t1 < t2 < · · · < tk−1 < 1 = tk . Let γi be the path γ|[ti−1 ,ti ] . Finally,
[γ] = γ1 δ1−1 δ1 γ2 δ2−1 δ2 γ3 δ3−1 · · · (δk−1 γk )

in π1 (X) by cancellation up to path homotopy (each parenthesis contains a loop

in A ∩ B). Here δi is a path from x to γi (ti−1 ). (CHECK)
Before we can compute more complicated fundamental groups, we need to
develop some basic combinatorial group theory.
6.1. Group Presentation. First we introduce the notion of a “free group”.
Definitions 6.9. Let A be any set, then the set W = x±1 ±1

±1

1 x2 · · · xn xi ∈ A
consists of all the words generated from the alphabet A. A reduced word is a word
which does not contain sequentially the letters x, x−1 for x ∈ A. The reduction
map r = red. : W = words → RW = reduced words sends a word to its reduced
form.
Note. For example, r : xyy −1 z −1 zzxy −1 7→ xzxy −1 . One has to prove this is
well-defined, i.e. reducing in any order results in the same word. For details, see
any book on combinatorial group theory.
We can define a group whose elements are the reduced words, where product
is given by concatenation followed by reduction.
Definition 6.10. The free group on A, F (A), consists of the reduced words
RW from A with the product w1 · w2 = r (w1 · w2 ).
Note. F (A) is said to be “free” as it is constructed so that there are no
relations between the generating elements.
F (A) depends, up to isomorphism, on |A|. If |A| = k, we show this explicitly
by writing F (A) = Fk .

Example 6.11. F0 = {e} and F1 = xk k ∈ Z ' Z.
 
 e, 
a, b, a−1 , b−1 ,

 

 
F2 = F (a, b) = a 2 −2
, a , ab, ba, a−1
b, a−1 −1 −1
b , b a, ba−1 −1 −1
, b a .
 


 .
..



Already F2 is non-abelian.
By construction, A ⊂ F (A). A behaves almost like a basis in linear algebra
(but now F (A) can be non-abelian).
Proposition 6.12. Let G be a group and i : A ,→ F (A) inclusion. Given a
map φ : A → G, there is a unique homomorphism Φ : F (A) → G with Φ ◦ i = φ.

Proof. Set Φ (x11 x22 · · · xkk ) = φ (x1 ) 1 φ (x2 ) 2 · · · φ (xk ) k .
Exercise 6.13. Check the details.
The construction above is very concrete. It would be better, philosophically,
to say the free group F (A) on a set A is the unique group with the property that
given the inclusion i : A ,→ F (A) any map φ : A → G there exists a unique
homomorphism Φ so that Φ ◦ i = φ. (So that the diagram commutes.) (D1) This
is the true defining characteristic of the free group. Of course, one can check that
everything we’ve done above holds in the more general view.
Proposition 6.14. Every group is a quotient of a free group. Every finitely
generated group is a quotient of a free group on a finite set.
Proof.
Say a group G is generated by a set S ⊂ G. WriteS = {a, b, c, . . . } and
let S = a, b, c, . . . . Now define a homomorphism Φ : F S → G by Φ (u) = u.
(By the previous proposition
it is enough to specify Φ on the generators.) Since Φ
is surjective, G ' F S / ker (Φ).
Exercise 6.15. Let Φ : H → G be a surjective homomorphism. Check that
G ' H/ ker (Φ).
Examples 6.16.
(1) Zn = F1 /an where F1 = F (a).
(2) Consider the symmetric group (permutation group) on n elements, Sn .
Exercise 6.17. Sn is generated by two elements a : (1, 2, . . . ) → 7
(2, 1, . . . ) (permutation of the first two entries) and b : (1, . . . , n) → 7
(n, 1, . . . , n − 1).
Define Φ : F (A, B) → Sn by Φ (A) = a and Φ (B) = b. Then
Sn = F2 / ker (Φ). Note ker (Φ) contains A2 , B n , etc.
Now we know we can always write G = F (A) /something. In fact the “some-
thing” must be a so-called “normal” subgroup.
Definition 6.18.
Let
H be a subgroup
of G. Define the right coset of H in G
to be the set Hg = hg h ∈ H, g ∈ G .
Exercise 6.19. Show that if Hg1 ∩ Hg2 6= ∅, then Hg1 = Hg2 .
So we can write write G = He ∪ Hg1 ∪ Hg2 ∪ · · · which is a disjoint union of
right cosets. (This is the right coset decomposition of G). What about left cosets?
Consider that if Hg = gH for all g ∈ G, then H = gHg −1 for all g. So we have
Definition 6.20. A subgroup H ⊂ G is called normal if H = gHg −1 for all
g ∈ G.
Proposition 6.21. If G is abelian, all subgroups are normal.
Example 6.22. S3 has 3! = 6 elements. It has subgroups Z2 of order two
and Z3 of order three. The first permutes elements (ab) and the second shuffles
(abc) 7→ (cab).
Exercise 6.23. Check that Z2 is not normal and Z3 is normal.
If H . G (read: H is normal in G), then the left and right cosets are the same.
Definition 6.24. Supose
H . G,
then the quotient group G/H is the group
consisting of the cosets Hg g ∈ G with product (Hg1 ) · (Hg2 ) = Hg1 g2 .
Proposition 6.25. |G/H| = |G|/|H|.
If we let [G : H] be the number of cosets of H in G, then the proposition reads
[G : H] = |G|/|H|.

Proposition 6.26. Given a homomorphism χ : K → L, ker (χ) = u ∈ K χ (u) = e
is a normal subgroup.
Proof. Let H = ker (χ) so that χ (H) = e. Then χ gHg −1 = χ (g) χ (H) χ g −1 =

−1
χ (g) eχ (g) = e for all g ∈ G. Hence gHg −1 ⊂ H and H ⊂ gHg −1 , so we’re
done.
Exercise 6.27. Check the proof.
Now suppose we have a surjective homomorphism Φ : F (A) → G, then as above
G ' F (A) / ker (Φ). Of course, we could describe ker (Φ) by listing its generators,
but this is in general too difficult and not necessary. Recall xg = gxg −1 is the
conjugate of x by g. And recall we said a set S generates a subgroup H ⊂ G if
H is the smallest subgroup containing S. Explicitly, H is all the finite products of
elements of S or their inverses. Now we have
Definition 6.28. A set S normally generates the normal subgroup H of G if
H is the smallest normal subgroup of G containing S.
We’ll write hSi to mean the subgroup H normally generated by S. H now
consists of all products of elements of S, their inverses, or conjugates of S.
Example 6.29. Let F (a, b) be the free group on two generators. Then we
have F (a, b) / hbi = F (a). Lots of elements are killed (not just b)! hbi contains
bk , abk a−1 , aj bk a−j , etc. The point here is that we did not need to specify that
all these elements are killed; we only needed to say b (and the normal subgroup it
generates) are killed.
This is a general phenomena: ker (Φ) is often easily describable by its normal
generators. This gives a more efficient way to describe the quotient F (A) / ker (Φ).
Example 6.30. A clock with ten hours (dihedral group of order 2n). There is
an element σ which is rotation by 1/n. Then σ n = e. And there is an element τ
which is a flip. Then τ 2 = e (τ = τ −1 ). And one can show τ στ = τ −1 (flip, rotate,
2
then flip). This is equivalent to writing (τ σ) = e. These three relations describe
the whole multiplication table of the group. There are 2n elements in the group;

and n flips. In the notation above, we have D2n = free/ hrelationsi =
n rotations
F (s, t) / S n , T 2 , T ST S . This is a very convenient way to describe the whole
group. We can write this in the notation
 

 

n 2
D2n = s, t S , T , T ST S .
 |{z}
 | {z }

generators relations
This is the “presentation” of the group D2n .

Lecture 8, 11/7/11
We have shown that every group can be described by its presentation,
n o D E
G = (xi )i∈I (Rj )j∈J = F (xi )i∈I / (Rj )j∈J
where the xi are the generators and the Rj = word in xj ’s are relations. The angled
brackets on the right mean that not only are the Rj set to e, but also all products
and conjugates of them and their inverses.

Example 6.31. Zn = x xn .
Note that it is both diffucult in theory and in practice to recognize that a group
is trivial from its presentation.
Definition 6.32. A group is finitely presented if both |I| and |J| are finite.
Exercise 6.33. Q is not finitely generated. (Hint: Think about denominators.)
Exercise 6.34. (Rapaport) Let x, y, z be the generators. Let yxy −1 = x2 ,

−1
zyz = y 2 , xzx−1 = z 2 . (When the number of relations equals the number of
generators, the group is said to have a balanced presentation.) Show this is the
trivial group. (Hint: Once we get, say, x = e then by symmetry y = z = e.)
Example 6.35. There is an analogue due to Serre in four generators x, y, z, w.

The relations are are
yxy −1 = x2
zyz −1 = y 2
wzw−1 = z 2
xwx−1 = w2 ,
but now the conclusion is that the group is infinite.
How could we go about deciding if a group is trivial? Suppose G has the

presentation
n o
G = (xi )i∈I (Rj )j∈J .
The following Tietze moves modify the presentation while keeping G the same:

(1) G = n(xi ) , y (Rj ) , y o
±1 ±1

(2) G = (xi ) (Rj ) , Rα Rβ

(3) G = (xi ) (Rj ) with one replaced by a conjugate
These moves are the non-commutative analogue to Gaussian elimination from linear
algebra. But here, the number of relations can change (in Gaussian elimnation the
number of rows/columns stay the same).
Theorem 6.36. Given two finite presentations of a group G, they are related
by a sequence of such moves.
Proof sketch. Represent the second set of relations in terms of the first and
mimic with a sequence of moves.
Remark. There are moves which preseve balance. But the corresponding
theorem in this case is unknown (see the Andrews-Curtis conjecture, which people
believe is false). This is interesting to topologists since balanced presentations arise
naturally for low-dimensional manifolds.
Now suppose G is trivial, but we are given a presentation with n relations.

One can show the number of Tietze moves needed to conclude G is trivial grows
faster than any computable function. Hence it is not algorithmically decidable if
an arbitrary group G is the trivial group!
6.2. Van Kampen’s Theorem. Van Kampen’s theorem describes the fun-
damental group of X = A ∪ B in terms of the fundamental groups of A and B.
So we need a way to combine fundamental groups. We already have the direct
product, but this was only useful in describing π1 (X × Y ). Here is a second way to
combine groups.

Definition 6.37. Suppose G1 = xi Rj and G2 = yk Sl are groups.
Their free product is the group

G1 ? G2 = (xi ) , (yk ) (Rj ) , (Sl ) .
Note. This is independent of the choice of presentations of G1 , G2 .
Example 6.38. Fk ? Fl = Fk+l . So Fk = Z ? Z ? · · · ? Z where there are k copies
of Z.
Here is a more conceptual way to describe the free product. Given groups
G1 , G2 , a group K is called a free product of G1 and G2 if there exists a diagram
of group homomorphisms (D1) such that given any homomorphsisms φ1 : G1 → L
and φ2 : G2 → L into some group L, there exists a unique Φ : K → L so that
Φ ◦ i = φ1 and Φ ◦ j = φ2 . (D2) Then one can prove:
(1) If K exists it must be unique (use uniqueness of Φ to get, when L, K
satisfy the hypotheses, maps both ways, etc.).
(2) There exists such a K. (Take K = G1 ? G2 .)
Recall the direct product G1 × G2 . First, note that (a, b) · (a0 , b0 ) = (aa0 , bb0 ) so
then (a, e) and (e, b) commute. So G1 ? G2 6= G1 × G2 , and in fact
G1 × G2 = (xi ) , (yk ) (Rj ) , (Sl ) , xi yk x−1 y −1 .

i k
But still the two products are related: (D3). Though the existence of Φ is guaran-
teed by the definition of free product we gave above, we already knew it was possible
to find such a homomorphism. Indeed, it is always possible to get a homomorphism
just by adding more relations (to get a quotient group).
Why does all of this matter for topology? Here is another weak form of Van
Kampen’s theorem.
Theorem 6.39 (Weak van Kampen). Let X = A ∪ B with A, B open and
assume A, B, A ∩ B are path connected. Choose a common base point x ∈ A ∩ B.
Assume A ∩ B is simply connected, then
π1 (X, x) = π1 (A, x) ? π1 (B, x) .
Remark. We usually can apply van Kampen’s theorem even when A, B are
not open. If A is not open, the trick is to replace A by a slightly larger open set
A0 such that i : A ,→ A0 is a homotopy equivalence (similarly replace B with B 0 ),
then apply van Kampen’s theorem to A0 and B 0 .

Example 6.40. We know π1 S 1 = Z. So then π1 ∨2 S 1 = Z ? Z. More
generally, π1 ∨k S 1 = Fk . And π1 ∨k S 2 = {0} since S k = D+ k k
∪S k−1 D− .
Here is the first non-trivial example of a free product.
Example 6.41. The infinite dihedral group, Z2 ? Z2 . What does this look like?
Say the first is generated by a, the second by b, then
Z2 × Z2 = a, b a2 , b2 .

We can list out the words:


ababab · · · a


ababab · · · b
Z2 × Z2 = .
bababa · · · a


bababa · · · b

In fact Z ,→ Z2 × Z2 as a normal subgroup, look for products of (ab). So then

Z2 × Z2 /Z = Z2 .
n
Exercise 6.42. Prove that D2n = a, b a2 , b2 , (ab) .
(D4)
Example 6.43. π1 (X∨ Y ) = π1 (X) ? π2 (Y ).
What if A ∩ B is not simply connected? Then the free product is not enough,
since π1 (A) ? π1 (B) includes possibly two copies of π1 (A ∩ B). We need a way to
“count” the overlap once.
Definition
6.44. Let G1 , G2 be groups and suppose we are given a group
H = (zm ) · · · with homomorphisms i : H → G1 and j : H → G2 . Then the
free product of G1 and G2 with amalgamation by H is given by

G1 ?H G2 = (xi ) , (yk ) (Rj ) , (Sl ) , i (zm ) = j (zm ) ∀m .
Note. At this point, we don’t need to assume that i, j are injective.
Theorem 6.45 (van Kampen). Suppose X = A ∪ B with A, B open and
A, B, A ∩ B path connected. Choose a common basepoint x ∈ A ∩ B. Then
π1 (X) = π1 (A) ?π1 (A∩B) π1 (B)
with the maps i∗ : π1 (A ∩ B) → π1 (A) and j∗ : π1 (A ∩ B) → π1 (B) induced by
the inclusions i : A ∩ B ,→ A and j : A ∩ B ,→ B.
Examples 6.46.
(1) G ?G G = G with the maps i, j = idG .
(2) G1 ?{e} G2 = G1 ? G2 .
As remarked before, i∗ and j∗ may fail to be injective. Here is a interesting
example from geometry.

Example 6.47. The solid torus is D2 × S 1 , which has π1 D2 × S 1 = Z.
Claim. S 3 = D2 × S 1 ∪S 1 ×S 1 S 1 × D2 .

Before giving a proof, let’s see what this says about π1 . We have π1 S 3 = {0}
and
π1 D2 × S 1 ∪S 1 ×S 1 S 1 × D2 = π1 D2 × S 1 ?π1 (S 1 ×S 1 ) π1 S 1 × D2

= Z ×Z×Z Z
with the respective maps i∗ = (0, 1) and j∗ = (1, 0) (written as matrices). So
π1 D2 × S 1 ∪S 1 ×S 1 S 1 × D2 = x, y i∗ (a) = j∗ (a) , i∗ (b) = j∗ (b)

= x, y e = y, x = e
= {e} .
Now here are two proofs of the claim. The first is symbolic. For manifolds
M, N we have
∂ (M × N ) = ∂M × N ∪∂M ×∂N M × ∂N.
(Think about ∂ (I × I).) So
∂ (Dp × Dq ) = ∂Dp × Dq ∪∂Dp ×∂Dq Dp × ∂Dq
and then
S p+q−1 = ∂ Dp+q = S p−1 × Dq ∪S p−1 ×S q−1 Dp × S q−1 .

In particular,
S 3 = ∂ D2 × D2 = S 1 × D2 ∪S 1 ×S 1 D2 × S 1 .

And now here is a picture: (D5)
7. Application to Knot Theory

Lecture 9, 11/14/11
This section discusses an interesting geometric application of the fundamental
group to knots. A typical knot looks like (D1), known as the trefoil knot. (The
knot lives in R3 , but we’ve drawn the “under-over” projection of the knot onto a
plane.) The most naive idea of a knot is an injective map ω : I → R3 . But we
could pull apart this knot completely since the ends are not permanently attached.
So instead we have
Definition 7.1. An injective map ω : S 1 → R3 is called a knot.
It is a basic question of knot theory to classify the knots. The trivial knot
(unknot) is just the inclusion S 1 ,→ R2 ,→ R3 . In this section we hope to prove
that the trefoil knot and the unknot are not the same. We’ll do so by coming up
with invariants and then computing these for the trefoil knot and unknot.
Definition 7.2. The crossing number of a knot is the minimum number of
crossings in a projection.
The unknot has crossing number zero, and the trefoil knot has crossing number
three. Here is another view of the trefoil knot: (D2) The figure eight knot has four
crossings: (D3) All of these so far are examples of alternating knots, or knots where
the crossings alternate under then over then under again. Another alternating knot,
this time with five crossings: (D4) But have we proven yet that the crossing numbers
are what we’ve asserted above? We only guessed what the numbers are, because
the pictures are so simple. In fact, the crossing number is not a useful invariant, as
there is no formula in general.
And what exactly do we mean when we say two knots are the same? All of the
knots are really just circles, being injective maps defined on S 1 . But what makes
them different is that they are sitting in R3 in different ways. Note that R3 is not
essential; a common (classical) way to think of a knot is as a map S 1 ,→ S 3 =
R3 ∪ ∞. So we have
Definition 7.3. A knot is a map K ,→ S 3 where K ' S 1 .
And equivalence is defined via
Definition 7.4. Two knots K1 ,→ S13 , K2 ,→ S23 are equivalent if there is a
homeomorphism φ : S13 → S23 with φ (K1 ) = K2 .
7. APPLICATION TO KNOT THEORY 35
Note. This particular definition of equivalence does not keep track of direction.
For example, it cannot tell a knot from its mirror image.
Usually we restrict further the notion of a “knot.” One could have a knot with
a infinite amount of “knottedness”, (D5) Pictures like this are excluded by making
any of several requirements: one could require the knot to be
• piecewise linear (a finite number of line segments)
• a differential submanifold (with tangents non-zero)
• topologically locally flat (near each point it looks standard).
But we won’t do any of these, since we only care to make a difference between the
trefoil and the unknot.
Step one is to consider everything that’s not the knot.
Definition 7.5. The knot complement of a knot K1 ,→ S 3 is the space S 3 \K1 .
The knot complement of the unknot is quite simple. But we’ll see that the
knot complement of the trefoil knot is quite complicated. The idea now is to show
that S 3 \trefoil and S 3 \unknot are not homeomorphic. Indeed, if there were an
equivalence of knots φ : S 3 → S 3 then restricting it to the knot φ| : K1 → K2 gives
a homeomorphism and so does restricting it to the knot complement φ| : S 3 \K1 →
S 3 \K2 . This raises a philisophical question: Is a knot uniquely determined by its
complement? The answer is yes in dimension three (due to Gordon and Leucke).
The answer is no in higher dimensions, e.g. dimension five.
The second step is to realize that it would suffice to show the knot complements
are not homotopy equivalent. Are we throwing away more information again? Yes,
but we’ll skip over this for the discussion here. We have only one technique for
showing spaces are not homotopy equivalent. So the third step is to show that the
fundamental groups of the complements are not isomorphic.
Definition 7.6. Given a knot K1 ,→ S 3 , the group of the knot G is the group
G = π1 S 3 \K1 .
We’ll show
Proposition 7.7. The group of the trefoil knot is not abelian, whereas the
group of the unknot is Z.
From this it immediately follows
Proposition 7.8. The unknot and the trefoil are not equivalent.
There are generalizations of knot theory to higher dimensions.
Theorem 7.9 (Zeeman unknotting). Any knot S k ,→ S n , k ≤ n, is equivalent
to the standard knot unless k = n − 2.
The first person to discover knots in higher dimensions was Artin.
It’s often annoying to deal with non-compact spaces. So we often work with
what’s called the closed complement of the knot. Suppose we have S 1 ,→ S 3 a knot.
Then one can prove there exists a tube S 1 ×D2 in which the knot is included. Then
0
S 3 \ S 1 × D2 ,→ S 3 \S 1 × {0}. On the left we have the closed complement, on the
right the knot complement. These are not very different. The closed complement is
a three-dimensional manifold with boundary, but the manifold without its boundary
is homotopy equivalent to the knot complement.
Now recall we saw

S 3 = D2 × S 1 ∪S 1 ×S 1 S 1 × D2 .
(D6) From the picture it follows that the closed complement of the unknot is
S 3 − S 1 × D2 = D2 × S 1 .
So
π1 (closed complement of the unknot) = π1 D2 × S 1

=0×Z
= Z.
And the complement of the unknot is
S 3 \unknot ' (open disc) × S 1 ' R2 × S 1
so again π1 = Z. (D7)
Now for the trefoil knot. The trefoil can be drawn on the standard torus in R3 .
In general, torus knots are knots that can be drawn on the torus. (D8) Segments
I, II, III, IV altogether form a loop on the torus. We claim I ∪ II ∪ III ∪ IV =
trefoil. (D9) So the trefoil is the (2, 3)-torus knot. In general we have the (p, q)-torus
knots where p, q are relatively prime.
Now we want to prove (2, 3)-torus knot is not the unknot. Let’s compute
G(p,q) = π1 S 3 \ (p, q) -torus knot . Now
S 3 \K(p,q) = D2 × S 1 \K(p,q) ∪(S 1 ×S 1 \K(p,q) ) S 1 × D2 \K(p,q) .

By deleting the knot we have only

taken away from the boundary of D2 × S 1 and
1 2 2 1
S × D . So D × S \K(p,q) is still homotopy equivalent to S 1 , and the same
with S × D \K(p,q) . And S × S \K(p,q) is homotopy equivalent to S 1 ×(0, 1).
1 2 1 1
(D10) So
π1 S 3 \K(p,q) = π1 S 1 ?π1 (S 1 ) π1 S 1

= Z ?Z Z
where the inclusions are multiplication by p on the first Z and q on the second Z.
So
π1 S 3 \K(p,q) = x, y xp = y q .

To finish the proof we’ll check this group is abelian.
Claim. G(2,3) is not abelian. In fact, it has a non-abelian quotient.
To see this, recall that

n o
2
D6 = σ, τ σ 2 = e, τ 3 = e, (στ ) = e

is the non-commutative group of order six. These relations are a quotient of the
relations on G(2,3) . Take the homomorphism φ : x 7→ σ, y 7→ τ since σ 2 = τ 3
anyways in D6 . So if we take G(2,3) and add additional relations then we get D6 .
So the trefoil knot is not the unknot.
8. APPLICATION TO TOPOLOGICAL GROUPS AND H-SPACES 37
8. Application to Topological Groups and H-spaces

Lecture 10
11/21/11. This is another application for which the notion of fundamental group is useful.
We’ll study spaces that are both topological spaces and groups.
Examples 8.1.
(1) Rd
(2) G, any finite group with the discrete topology.
(3) S 1 ⊂ R2 is a topological
space because it is sitting in the plane. But
also S 1 = z ∈ C |z| = 1 . So z gets a group structure from complex
multiplication, z1 · z2 = eiθ1 · eiθ2 = ei(θ1 +θ2 ) .
(4) The n-torus S 1 × S 1 × · · · × S 1 .
(5) Matrix groups. The general linear group is the group
GL (n, R) = {real n × n invertible matrices}
2
under matrix multiplication. Since GL (n, R) ⊂ Rn this is also a topo-
logical space. And as M ∈ GL iff det M 6= 0, GL (n, R) is an open subset
2
of Rn . But GL (n, R) is not compact. There are many familiar subspaces
of the general linear group, e.g. the orthogonal group
On = A ∈ GL (n, R) A−1 = At .

One can show On ⊂ GL (n, R) is compact. Note that if A ∈ On then

det A = ±1. The special orthogonal group is

SOn = A ∈ On det A = +1 .
(6) Complex analogues. Cn is a topological group under addition. C∗ =
C\ {0} is a topological group under multiplication. The complex general
linear group is GL (n, C). The unitary group
Un = A ∈ GL (n, C) A−1 = A∗

is a compact subspace. Note A ∈ Un iff |det A| = 1. The special unitary

group is

SUn = A ∈ Un det A = 1 .
This last group is important in physics.
Definition 8.2. A topological group is a set G so that
(1) G is a group, i.e. G has multiplication m : G × G → G and inversion
i : G → G.
(2) G is a topological space.
(3) m and i are continuous.
Note. The third condition says that the two structures are compatible.
Topological groups often represent the symmetries of geometrical systems.
They can also represent the symmetries of physical systems. Einstein realized that
understanding the symmetries of the universe was important to physics.
The previous examples are all topological groups. In fact, they are are also
Lie groups, in that they are both groups and differentiable manifolds. It is known
that all finite dimensional topological groups are Lie groups. This was proved 60-70
years ago.
A natural question is: Which spaces are topological groups? If a space X has
a toplogical group structure, then X satisfies a certain “uniformity”. Let e ∈ X be
the identity element. Then given any g ∈ X, we can produce a homeomorphism
by (left or right) multiplication by g. That is, the map ×g : X → X, x 7→ xg is
a homeomorphism. Notice that ×g : e 7→ g. So given any point g ∈ X, there is a
homeomorphism carrying e to g, and therefore there is a homeomorphism between
any two points x, y ∈ X. Thus if a space is to be a topological group, then it
must look the same at all points. For example, the figure eight, X, and Y are not
topological groups.
Theorem 8.3. If G is a topological group, then π1 (G, e) is abelian.
This is another way to see that the figure eight is not a topological group, for
its fundamental group is the free group on two elements.
Example 8.4. R2 \ {p} is a topological group. But R2 \ {p1 , p2 } is not a topo-
logical group, for R2 \ {p1 , p2 } ∼ S 1 ∨ S 1 .
h.e.
Before we prove the theorem, consider

Proposition 8.5. Given a group K, the multiplication map m : K × K → K
is a homomorphism iff K is abelian.
Proof. We have m : (h, g) 7→ hg. In the product group we have (h1 , g1 ) ·
(h2 , g2 ) = (h1 h2 , g1 g2 ). Apply the homomorphism m to get (h1 g1 ) · (h2 g2 ) =
h1 h2 g1 g2 and hence g1 h2 = h2 g1 . This happens iff the group is abelian.
Now we’ll see two proofs of the theorem. The first is quick, the second is
pictorial.
First proof. Consider multiplication m : G × G → G, which is continuous.
It induces a map m∗ : π1 (G) × π1 (G) = π1 (G × G) → π1 (G) which is given by
m∗ ([α] · [β]) = [α · β]. To see this, write [α] = [α] · e and [β] = e · [β], so by
considering pointwise multiplication we find m∗ (([α] , [β])) = m∗ (([α] · e, e · [β])) =
[α] · [β]. This says m∗ is a homomorphism, and hence π1 (G) is abelian.
Second proof. In π1 (G) there seem to be two products:
(1) The usual loop product, α · β.
(2) The pointwise product of loops, α × β (t) = α (t) · β (t).
We claim [α · β] = [α × β] and that both multiplicatoins are abelian. The first
claim follows as α · β ∼ α × β in G. The homotopy is depicted below. (D1) To
h
prove the multiplication are abelian, either slide some more, or turn the picture
upside down. (D2)
Remark. We never used that G has an inverse. What’s more, we never used
that multiplication is associative. We only used continuous multiplication and the
identity element.
This last remark leads to the following generalization of topological groups.
Definition 8.6. An H-space is a topological space X equipped with a multi-
plication m : X × X → X satisfying
(1) m is continuous.
8. APPLICATION TO TOPOLOGICAL GROUPS AND H-SPACES 39
(2) There exists an identity element e, with m (e, u) = u = m (u, e) for all
u ∈ X.
Example 8.7. Any topological group is a H-space. This is the main example.
The proofs above give the next theorem.
Theorem 8.8. If X is a H-space, then π1 (X, e) is abelian.
Remark. Lots of topological spaces have non-abelian fundamental groups.
This theorem says that such topological spaces cannot have multiplication.
There are spaces that have more than one H-space structure. There is a large
literature on finite dimensional H-spaces. Before we move on, let us see an inter-
esting example of a Lie group.
Example 8.9. S 3 is a Lie group. A concrete way
to see this is via a system
of
arithemetic called the quaternion numbers, H = a + bi + cj + dk a, b, c, d ∈ R .
The quaternions are associative and satisfy i2 = j 2 = k 2 = −1, ij = k, ji = −k.
1/2
They form a group. One can define a norm via |z| = a2 + b2 + c2 + d2 . Then
1/2
(|z||z|) = |z| where z = a − bi − cj − dk.

Claim. S 3 = z ∈ H |z| = 1 .
So S 3 has a multiplication via the quaternions, so it is a group. The quaternions
were discovered by Hamilton.
What spheres are Lie groups? As it turns out, S 0 = {±1}, S 1 , and S 3 are Lie
groups. S 7 is an H-space, via the Cayley numbers. But the multiplication is not
associative. In fact, S 7 has no associative H-space structure, where multiplication
is associative up to homotopy. So it cannot be a Lie group. In general, we would
hope that “nice” multiplication in a Euclidean space satisfies the property
(8.1) a 6= 0, b 6= 0 =⇒ ab 6= 0.
As it were, R , R = C, R = H, and R8 = Cayley numbers have such a multipli-
1 2 4
cation. The surprise is that these are the only Euclidean spaces on which such a
multiplication can be defined.
Theorem 8.10 (J. Frank Adams). S n is a H-space iff n = 0, 1, 3, 7.
Exercise 8.11. Suppose Rn , n 6= 1, 2, 4, 8, is made into a H-space via the
multiplication m : Rn × Rn → Rn . Show that the above theorem implies that
property 8.1 must fail. Therefore show that the only Euclidean spaces Rn on which
a “nice” multiplication is defined are with n = 1, 2, 4, 8.
There is strong evidence for the following conjecture.
Conjecture 8.12. Every finite dimensional H-space is homotopy equivalent
to a manifold.
CHAPTER 4
Covering Space Theory
1. Covering Spaces
Intuitively, a covering space gives a way to unwrap a space. As we’ll see, there
is a way to translate any question about a fundamental group into a question about
a covering space, and vice-versa.
Examples 1.1.
(1) Consider the real line R and take the map f : R → S 1 , f (x) = e2πx .
Then f −1 (1) = Z ⊂ R. In some sense this map wraps the line around
and around the circle. We want to think about the circle being unwrapped
to the line.
1 1
k −1 2πil/k
So consider the map gk : S → S ,
(2) Wrap a circle k times arounda circle.
gk (z) = z . Then gk (1) = e , and gk is a k-to-1 map.
We’ll see that these are the only covering spaces of the circle, the first is the
infinite cover and the second the finite cover.
Example 1.2. Wrap C around C\ {0}. The map is f : C → C\ {0}, f (z) = ez .
The inverse map on a peice of this is a log.
Definition 1.3. A continuous map f : Y → X is said to be a covering map
if each point u ∈ X has an open neighborhood U so that f −1 (U ) = ∪j∈J Vj where
the Vj are disjoint open sets so that f |Vj : Vj → U is a homeomorphism. Such a
neighborhood U is said to be evenly covered by f .
Note. So a mapping f is a covering map if X is the union of evenly covered
open subsets.
Example 1.4. The map f : (−5, 7) → S 1 which wraps the open interval around
1
S is a local homeomorphism, for we can find a neighborhood on which f restricts
to a homeomorphism. But it does not hit each point the same number of times. So
f is not a covering map.
Proposition 1.5. Let X be a connected space. If f : Y → X is a covering
map, then for all p, q ∈ X we have #f −1 (p) = #f −1 (q).
Proof. We claim #f −1 (p) is locally constant, i.e. #f −1 (p) is constant in a
neighborhood of every p ∈ X. To see this, pick U an evenly covered neighborhood
of p, then #f −1 (q) = |J| for all q ∈ U . Then the set of points q so that #f −1 (q) =
#f −1 (p) is an open subset of X. Therefore, X is a disjoint union of open sets
(whose non-trivial inverse images have given cardinality). Since X is connected,
one of these open subsets is all of X, and the rest are null.
Definition 1.6. Let X be connected and f a covering map. Its degree is the
number deg (f ) = #f −1 (p).
41
42 4. COVERING SPACE THEORY
Note. This is well-defined by the proposition.

Exercise 1.7. If f1 : Y1 → X1 , f2 : Y2 → X2 are covering maps, then f1 × f2 :
Y1 × Y2 → X1 × X2 is a covering map.
Exercise 1.8. If f : Z → Y , g : Y → X are covering maps, then g ◦ f : Z → X
is a covering map. And deg (g ◦ f ) = deg (g) · deg (f ).
Example 1.9. Let f be the quotient map f : S n → RP n given by the descrip-
tion S n /v ∼ (−v) = RP n . Then f is a 2-to-1 covering map. In particular, the
quotient map g : S 1 → RP 1 = S 1 /v ∼ (−v) is given by g (z) = z 2 .
As we’ll see, the only covers of RP n , n > 1, are given by the quotient map in
the example above, and the identity idRP n .
2. Higher Homotopy Groups

Lecture 11
11/28/11 Recall elements of π1 (X, x) can be represented in various ways:
1
• If ω : S,p → X,x is a loop with ω (p) = x, then [ω] ∈ π1 (X, x). How do we
multiply loops? Let q : S 1 → S 1 /S 0 = S 1 ∨ S 1 be the quotient map, and
given two loops α, β, let α∨β : S 1 ∨S 1 → X be a map defined on the figure
eight (α on the first S 1 , β on the second). (D1) Then α · β = (α ∨ β) ◦ q.
And we had [α] · [β] = [αβ]. This description of loop multiplication will
generalize nicely to higher homotopy groups.
• If ω : I = [0, 1] → X,x is a path with ω (0) = ω (1), then [ω] ∈ π1 (X, x).
We already saw how to multiply, by splicing.
• If ω : R → X is a continuous map and ω (R\K) = {x} where K ⊂ R is
compact, then [ω] ∈ π1 (X, x). Once again we splice together to multiply,
but now on the whole
real line. (D2)
• π1 (X, x) = S 1 , X + .

Definition 2.1. The higher homotopy groups are πk (X, x) = S k , X + .
We can describe πk in ways analogous to the above:
k
• If ω : S,p → X,x then [ω] ∈ πk (X, x). Let q : S k → S k /S k−1 = S k ∨ S k
be the quotient map, and let α ∨ β : S k ∨ S k → X be the wedged map.
Then we define α + β = (α ∨ β) ◦ q, and we define [α] + [β] = [(α ∨ β) ◦ q].
• If ω : Dk → X with ω S k−1 = x then ω ∈ πk (X, x). Multiplication is
similar to above.
• If ω : Rk → X with ω Rk \K = {x} where K ⊂ Rk is compact, then
[ω] ∈ πk (X, x). Multiplication is similar to above.
Exercise 2.2. Prove that πk (X, x) is a group under loop multiplication.
So the picture is all the same so far. But in fact, we get more information when
k > 1.
Theorem 2.3. For k > 1, πk (X, x) is abelian.
Remark. This is why we use additive notation for πk , k > 1.
Exercise 2.4. Prove the theorem. (Hint: Use the third description of πk (X, x)
above. Rotate the space to get a homotopy between [α] + [β] and [β] + [α].)
3. THE LIFTING PROBLEM 43
As mentioned in the previous chapter, the homotopy groups do not completely

determine the space. But the following facts hold.
Proposition 2.5. Let X be path-connected. If πk (X, x) = 0 for all k ≥ 1,
then X is contractible.
Proposition 2.6. If X has only one non-zero homotopy group, say in dimen-
sion k, then that group G determines the homotopy type of X.
Remark. The Eilenberg-MacLane spaces, denoted K (G, k), are those spaces
whose kth (and only non-zero) homotopy group is G. By the proposition above,
for k ≥ 2 it is neccesary that G is abelian to ensure the existence of a space.
Example 2.7. We’ll prove later that
(
1
Z k=1
πk S = .
0 k>1
So S 1 = K (Z, 1).
The following theorems are analogues from before. Their proofs are similar as
before.
Theorem 2.8. πk (X × Y ) ' πk (X) × πk (Y ).
Example 2.9. Combining the previous example with this theorem gives that
(
1 1
Z×Z k =1
πk S × S = .
0 k>1
So S 1 × S 1 = K (Z × Z, 1).
Note that the torus T 2 = S 1 × S 1 has a two-dimensional hole (in the center).
But the example above shows that π2 fails to capture this hole. This is because the
center hole is not spherically shaped (it is shaped like a torus). Since we arrived at
the theory of homotopy groups by investigating spherically shaped holes, we should
not expect π2 to see this hole. Next semester, we’ll study homology theory, which
investigates holes of arbitrary shape. Such a theory would find the center hole in
T 2.
Theorem 2.10. An element ω : S k → X vanishes in πk (X, x) ([ω] = 0 in
πk (X)) iff ω extends to a map ω : Dk+1 → X, i.e. iff there exists a map φ so that
the following diagram commutes. (D3)
Exercise 2.11. Prove the previous two theorems.
3. The Lifting Problem

Suppose X, Y are path connected spaces, and f : Y → X,x is a covering map.
Recall we defined the degree of f as deg (f ) = #f −1 (p) for any p ∈ X. This was
independent of p. Finally, let any y ∈ f −1 (x) be the basepoint for Y .
We now investigate the so-called “lifting problem”. Notice that for a small
enough set A ⊂ X, we get exactly deg (f ) copies of A upstairs in Y . For example,
if f : R → S 1 is the infinite-cover, then there are an infinite number of copies of
a small neighborhood of any point s ∈ S 1 . But there is no copy of the entire S 1
in R1 . More formally, we have the lifting problem: Given a map g : A,a → X,x ,
does there exist a map+ ĝ : A,a → Y,y so that f ◦ ĝ = g? (This is existence and
uniqueness, combined.) Such a map ĝ is called a lift of g (to the cover Y ). The lift
preserving the base-point is unique.
Theorem 3.1 (Lifting Theorem). Let X, Y be path connected, let f : Y,y → X,x
be a covering map+ , and let g : A,a → X,x be a map+ . There exists a lift+ iff
Im g∗ ⊂ Im f∗ . Moreover, if a lift+ exists, then it is unique.
Proof of =⇒ . Formally, if the diagram of spaces (D6) exists, then the dia-
gram of groups (D7) exists. That is, if there exists a map ĝ so that f ◦ ĝ = g, then
the induced maps must satisfy f∗ ◦ ĝ∗ = g∗ . Hence Im g∗ ⊂ Im f∗ .
The direction ⇐= is much harder, and we won’t see it now.

Corollary 3.2. If A is simply connected, then there exists a unique lift+ .
Proof. The condition is that Im g∗ ⊂ Im f∗ where g∗ : π1 (A) → π1 (X). But
π1 (A) = 0 so the condition is vacuous.
Corollary 3.3. If f : X → Y is a covering map, then f∗ : πk (Y ) → πk (X)
is an isomorphism for k > 1.

Example 3.4. πk S 1 ' πk R1 = 0 for k > 1 by the corollary. Also
πk S 1 × · · · × S 1 = 0 either since πk S 1 = 0 or because there is a covering
map wrapping Rn around T n .
Proof sketch. Given ω : S k → X representing [ω] ∈ πk (X), we can lift
this to Y as S k is simply connected for k > 1. Hence f ◦ ω̂ = ω for some ω̂, and
hence f∗ ([ω̂]) = [ω] in πk (X). This is surjectivity. The proof that f∗ is injective is
similar, by lifting maps of Dk+1 .
Exercise 3.5. Fill in the details above.
Example 3.6. As we observed before, there is a map q : S n → RP n = S n /v ∼
(−v) which is the quotient map. This is a covering map of degree two. Thus
(
0 1<k<n
πk (RP n ) ' πk (S n ) = .
?
The cases k ≥ n are unknown in general.
Let us go in another direction. Consider S 0 ⊂ S 1 ⊂ S 2 ⊂ · · · , then define
∞
S = ∪k S k . This space is contractible (for the spheres are nested to begin with).
Similarly, RP 0 ⊂ RP 1 ⊂ RP 2 ⊂ · · · and RP ∞ = ∪k RP k . One can show that
π1 (RP ∞ ) = Z2 (we’ll prove it later). And πk (RP ∞ ) ' πk (S ∞ ) = 0 for S ∞ is
contractible. So RP ∞ = K (Z2 , 1).
Here is a nice fact about covering spaces.
Proposition 3.7. Let X, Y be path connected, and f : Y → X a covering map.
Then f∗ : π1 (Y ) → π1 (X) is injective.
Remark. So we can view π1 (X) as being isomorphic to Im f∗ .
Exercise 3.8. Prove the proposition via the technique used in the proof of
corollary 3.3.
4. COMPARING COVERING MAPS 45
4. Comparing Covering Maps

Suppose f1 : Y1 → X, f2 : Y2 → X are covering maps. Is there a map between
Y1 and Y2 so that the diagram commutes?
Example 4.1. Take X = S 1 , Y1 = R1 and f1 : x 7→ e2πix the infinite cover,
Y2 = S 1 and f2 the k degree cover f2 : z 7→ z k . The comparison map is h : R1 → S 1
is given by h (x) = e2πix/k . Then f2 ◦ h = f1 .
Definition 4.2. Given covers f1 : Y1 → X, f2 : Y2 → X, a map of covering
spaces is a map h : Y1 → Y2 so that f2 ◦ h = f1 .
Proposition 4.3. There exists such a map h iff Im (f1 )∗ ⊂ Im (f2 )∗ .
Proof. Apply the lifting theorem.
Definition 4.4. Two covers Y1 , Y2 are said to be equivalent if there are maps
of covers h, k in both directions so that k ◦ h = idY1 and h ◦ k = idY2 .
Proposition 4.5. Covers Y1 , Y2 are equivalent iff Im (f1 )∗ = Im (f2 )∗ .
Proof. Use the previous proposition to get inclusions both ways.
Now consider the mapping from covers of X to subgroups of π1 (X). We have
proved that this mapping must be injective. Thus covers of X are determined by
their fundamental groups, regarded as subgroups of π1 (X).

Example 4.6. We have π1 S 1 = Z. The subgroups are (kZ) ' Z (k > 0)
and {0}. We’ve shown this must be the whole story when it comes to covering
maps. The covering space corresponding to the subgroup {0} is R1 , and the cover
corresponding to kZ is the map fk : S 1 → S 1 , z 7→ z k . These are all the connected
covering spaces of S 1 .
Corollary 4.7. If X is simply connected, then the only connected cover is
idX : X → X.
Proof. π1 (X) = {0}.
As it turns out, the mapping from covers of X to subgroups of π1 (X) is also
surjective. So it is bijective. We’ll prove this next time via a construction.
Here is a special case of the above.
Proposition 4.8. If X is a connected space, then X has at most one simply
connected covering space. Moreover, this covering space maps to all the other covers.
We’ll
prove a simply connected covering space always exists. And πk (X) =
πk X̃ . (??)
Definition 4.9. Given a connected space X, its simply connected covering

space is called the universal cover of X, and is denoted X̃.
Corollary 4.10. If X is simply connected, then X̃ = X.
Exercise 4.11. X
^ × Y = X̃ × Ỹ .
Examples 4.12.
1 × S n = R × S n with n > 1.
(1) S^
(2) S 1 ×^
· · · × S 1 = Rn
| {z }
n times
]n = S n with n > 1.
(3) RP
] = R2 . (D9)
(4) Klein
1 × X = R × X.
(5) If X is simply connected, then S^
Exercise 4.13. Work out the details in the previous examples.
Exercise 4.14. What is the universal cover of S 1 ∨ S 2 ?

Top 1

Uploaded by

Copyright:

Available Formats

Top 1

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Top 1

Uploaded by

Copyright:

Available Formats

Topology I

We can extend the p-adics even further by completing Q, via limits in

Proposition 1.13. f : (X, dX ) → (Y, dY ) is continuous iff f −1 (U ) is open in

sets). The Zariski topology with X = C is a specific example of this

3. The Separation Axioms

Example 4.8. R is not pre-compact, but (0, 1) is pre-compact.

5. The Product Topology

splicing technique above. So X decomposes into equivalence classes, which we call

Proof. If not, then we can write X = A ∪ B for A, B 6= ∅, A ∩ B = ∅, and

7. Quotient Spaces and Gluing

If X is a space formed in such a way, we write X = A ∪C B to mean A glued to B

(3) RP 2 = Mobius ∪S 1 disc.

complexplane, which maps S 1 2-to-1

The Fundamental Group

Exercise 1.5. Prove the proposition above.

Exercise 1.10. Prove the proposition.

This result really cannot be acheived by reparameterization.

Claim. This is an inner automorphism, in fact given by χγ2 γ −1 . So χγ1 =

3. Comparing Spaces Algebraically

Claim. f∗ is a homomorphism, i.e. f∗ ([α] · [β]) = f∗ ([α]) · f∗ ([β]).

π1 S 1 × · · · × S 1  = Zn . In fact, the same proof carries through for infinite prod-

Definition 4.1. Given two maps f, g : X → Y , a homotopy of f to g is a map

Remark. As with paths, another way of thinking about homotopy is to con-

Topologists use the following notation:

These considerations enormously shrink the subject of maps from X to Y .

Definition 4.5. A map+ f : X → Y is said to be null-homotopic if it is

Example 4.7. If Y (or X) is convex, then all maps of X to Y are null-

Proposition 4.10. Suppose A is a deformation retract+ of X and let i : A ,→

Theorem 5.7 (J.H.C. Whitehead). Suppose f : X → Y is a map between

6. Combinatorial Group Theory and Van Kampen’s Theorem

So the smallest subgroup of G containing S is the intersection of all such sub-

Here we call the product s±1 ±1 ±1

in π1 (X) by cancellation up to path homotopy (each parenthesis contains a loop

This is the “presentation” of the group D2n .

Exercise 6.33. Q is not finitely generated. (Hint: Think about denominators.)

Exercise 6.34. (Rapaport) Let x, y, z be the generators. Let yxy −1 = x2 ,

Example 6.35. There is an analogue due to Serre in four generators x, y, z, w.

but now the conclusion is that the group is infinite.

How could we go about deciding if a group is trivial? Suppose G has the

Now suppose G is trivial, but we are given a presentation with n relations.

We can list out the words:

In fact Z ,→ Z2 × Z2 as a normal subgroup, look for products of (ab). So then

And now here is a picture: (D5)

7. Application to Knot Theory

Now recall we saw

S 3 \K(p,q) = D2 × S 1 \K(p,q) ∪(S 1 ×S 1 \K(p,q) ) S 1 × D2 \K(p,q) .

By deleting the knot we have only

To finish the proof we’ll check this group is abelian.

Claim. G(2,3) is not abelian. In fact, it has a non-abelian quotient.

To see this, recall that

8. Application to Topological Groups and H-spaces

One can show On ⊂ GL (n, R) is compact. Note that if A ∈ On then

is a compact subspace. Note A ∈ Un iff |det A| = 1. The special unitary

Before we prove the theorem, consider

Covering Space Theory

Note. This is well-defined by the proposition.

2. Higher Homotopy Groups

As mentioned in the previous chapter, the homotopy groups do not completely

3. The Lifting Problem

complexplane, which maps S 1 2-to-1