1 August 22, 2012: M275 Notes

M275 Notes
Evan Chen
December 10, 2012
Lecture notes from M275 at SJSU, a graduate-level course at SJSU in

Algebraic Topology taught by Professor Richard Kulbelka. These notes were
taken using Vim and/or GVim equipped with LATEX-Suite, which sped up
typesetting significantly; latexmk with the flag -pvc was useful for compiling
these notes in real-time.
1 August 22, 2012

1.1 The Question
Question. Given two topological spaces, can we determine if they are homeomorphic?
In short, this is very hard.
Exercise. Show that R 6≈ [0, 1] with the usual topology.
Solution. Let I = [0, 1]. If we remove 0 from I, then I is still connected. On the
other hand, no point in R has this property, so we lose, since homeomorphism preserves
connectedness.
Definition. Let S 1 = {z ∈ C : |z| = 1} be the unit circle.
1.2 Category Theory

Definition. A category C will consist of a “class” of objects ob(C), as well as a “class”
of morphisms mor(C), such that for every pair of objects there is associated a (possibly
empty) set of morphisms mor(X, Y ).
Example. The category G of groups, with ob(G) being “the set of all groups”, except
this leads to a contradiction, so we call it a class instead. In fact, if the objects of a
category are a set, it is called a “small category” or “kitty-gory”.
Example. A is the category of abelian groups, category V of vector spaces, category R

of rings.
Definition. Give X, Y ∈ ob(C), then mor(X, Y ) is the set of morphisms “from X to

Y ”.
Given two groups G1 , G2 ∈ ob(G), then mor(G1 , G2 ) is the set of all homomorphisms
from G1 to G2 . For vector spaces V1 , V2 ∈ ob(V), then mor(V1 , V2 ) is the set of all linear
maps from V1 to V2 .
Finally, T op is the category of topological spaces. ob(T op) is the set of topologi-
cal spaces, and M or(T op) is the set of morphisms M or(X, Y ), which are continuous
functions from X to Y .
1
Evan Chen M275 Fall 2012: Notes 1 August 22, 2012
Note that morphism sets M or(X, Y ) can be small or even empty. For example, there
are very few homomorphisms from Z/3Z to Z.
Suppose we have a morphism ϕ : X → Y and a morphism ψ : Y → Z. As one might
expect, we define composition to yield another morphism. Also mor(X, X) must contain
the identity.
We define morphisms ϕ : X → Y and ζ : Y → X such that ϕ ◦ ζ = ζ ◦ ϕ = id, then
we say they are inverses, and both of them are called “isomorphs.”
1.3 Outsourcing
In algebraic topology, we say topology is too hard and start with the category T op and
the category of groups G.
Definition. A functor F between categories sends an object X ∈ ob(T op) to a group

F (X) ∈ ob(G).
Given ϕ ∈ M or(X, Y ), we want F to take ϕ to F (ϕ) ∈ M or(F (X), F (Y )). So a

functor is a function on two levels: it sends topological spaces are groups, and morphisms
between spaces to homomorphisms between these groups.
We will be studying the fundamental group functor π1 . It will associate a space X to
a group π1 (X), and we want to send f : X → Y to f∗ : π1 (X) → π1 (Y ). We also want it
to preserve compositions: if f : X → Y and g : Y → Z, then we want π1 (g ◦f ) = (g ◦f )∗ .
Given three functions f : X → Y , g : Y → Z and h : X → Z. The diagram is said to
be commutative when h ≡ g ◦ f . We want the functor to preserve this (aren’t we just
restating?): we need
(g ◦ f )∗ = g∗ ◦ f∗
Fact. If f is an isomorph, then f∗ must be an isomorphism.
This is useful if, say, π1 (X) ∼

6 π1 (Y ), this implies X 6≈ Y . A nice success is that
=
manifolds are completely characterized by the number of “holes” (in a loose sense).
2
2 August 27, 2012

2.1 Announcements and Such
Category theory handout has been updated.
Homework may or may not be due on Wednesday.
2.2 Setting Out

Given a space X, we can either begin algebraic topology by first
(i) Define homology groups H1 (X), H2 (X), · · · .
(ii) Define homotopy groups π1 (x), π2 (X), · · · .
The issue is as follows: homology groups are harder to define, but are relatively easy to
compute. Homotopy groups are easy to define and hard to compute.
Definition. S n denotes an n-dimensional sphere; that is, S n ⊆ Rn+1 can be defined as
x2i = 1}
X
S n = {(x0 , · · · , xn ) :
Example. S 1 ⊂ R2 is a unit circle. S 2 ⊂ R3 is the surface of a sphere.
One can also think of S n as the equator of S n+1 .

Anyways, understanding πi (S n ) is analogous to understanding the symmetric groups
Sn ; by Cayley’s Theorem, this will suffice to understand all of finite group theory.
Definition. π1 (X) is called the fundamental group of X.
Theorem 1 (Brouwer Fixed Point Theorem). If f : D2 → D2 is continuous, then

∃x0 ∈ D2 which is fixed by f .
Corresponding geometric idea: crumple a piece of paper, and lay it on a copy of itself.
Then there are two corresponding points which are directly above each other. (We can
use paper (rectangles) instead of discs, since D2 ≈ I × I.)
Other spaces which satisfy Brouwer Fixed Point Theorem are said to find the fixed
point property. This is strong; for example S1 does not satisfy this property (just rotate
it 90◦ ; now no points are fixed. Remember that S1 is a circle, not a disk.)
We will prove later that ∃ϕ : D2 → S 1 such that ϕ |S 1 = idS 1 (that is, no continuous
function from D2 → S 1 which fixes the circumference.)
2.3 Homotopy Groups

A spider lives in a topological space X. He lives at a base point x0 ∈ X. Any web from
x0 to x0 can be reeled back in; that is, any path is nulhomotopic.
On the other hand, a punctured plane (a plane minus a point) does not have the
property, since if the web encloses the hole, then the web gets stuck.
Consider a bagel/torus. Once again, it does not have this property.
OK, formal nonsense. The terms map, function, etc. are used interchangeably, and
I = [0, 1].
Definition. A path in Y is a continuous map ϕ : I → Y , where I has the standard

topology it inherits as a subspace of R. The initial point is y0 = ϕ(0), and the ending
point is y1 = ϕ(1).
3
Pictorially, we can draw the image of the path ϕ(I). This does not contain all the
information (e.g. speed).
Since I is compact, ϕ(I) cannot go off to infinity or oscillate around a point, or do
any other silly things.
Definition. A path-connected space is one where any two points are connected by a
path.
Definition. A loop in X at x0 is path γ : I → X such that x0 = γ(0) = γ(1). It is

simple if it is not self-intersecting (other than at the endpoint).
Definition. Given two continuous functions f, f 0 : X → Y , we say that f is homotopic

to f 0 if there exists a continuous function F : X × I → Y such that F (x, 0) = f (x) and
F (x, 1) = f 0 (x).
Definition. If A is a subspace of X and f = f 0 on A, (i.e. f (a) = f 0 (a)∀a ∈ A), then we

say that f is homotopic to f 0 relative to A if ∃f : X × I → Y such that F (x, 0) = f (x),
F (x, 1) = f 0 (x) AND F (a, t) = f (a).
A special case we will use a lot is if we have two paths in Y that both start and end
at the same point.
Definition. Paths f, f 0 : I → Y in Y satisfy y0 = f (0) = f 0 (0) and y1 = f (1) = f 0 (1).

Then f and f 0 are path homotopic if they are homotopic relative to {0, 1}.
In other words, f, f 0 : I → Y are path homotopic if ∃F : I 2 → I continuous with
F (0, t) = f (0) = f 0 (0)

F (1, t) = f (1) = f 0 (1)
F (s, 0) = f (s)
F (0, s) = f (s)
These three definitions give equivalence relations.

Consider the homotopy classes of loops beginning at a specified point. We define the
product ∗ to be the “composition” of the paths, in the sense that we glue one path to
the end of the other. Under this operation, the classes form a group at that point.
Theorem 2. All such groups (for a fixed space X) are isomorphic.
It is not hard to check that these operations are all equivalence relations. We write
them as f ' f 0 , f 'A f 0 and f 'p f 0 . This is left as an exercise, since it’s actually not
hard.
Since apparently this is confusing people:
Exercise. Let f : X → Y be continuous, and define F : X × I → Y by F (x, t) = f (x).

Then F is continuous.
Proof. Let V be an open subset in Y . F −1 (Y ) = f −1 (X) × I, however; this is the

product of open sets, and hence itself open.
4
3 August 29, 2012

3.1 Questions
Definition. Given spaces X and Y and a continuous map f : X → Y , we say f is
nulhomotopic if f is homotopic to a constant map; i.e. if there exists y0 ∈ Y such that
when we define the constant map, cy0 : X → Y by x → y0 , then f ' cy0 .
In other words, f is nulhomotopic if ∃y0 ∈ Y and F : X × I → Y continuous for which
F (x, 0) = f (x), F (x, 1) = y0 .
3.2 Straight-Line Homotopy

~
In Rn , one useful way to construct a homotopy is to consider R(t) = (1 − t)~x + t~y .
Example. Given 2 paths in the plane γ1 , γ2 : [0, 1] → R2 by
γ1 (s) = eiπs and γ2 (s) = e−iπs
Show that γ1 and γ2 are path-homotopic.

Solution. If we want to show γ1 'p γ2 , we need F : I × I → R2 with F (s, 0) = γ1 (s),
F (s, 1) = γ2 (s). So we can construct the convenient homotopy
F (s, t) = (1 − t)γ1 (s) + tγ2 (s)
Musing: for any two paths p1 and p2 in X, can we find a larger space Y for which
they are path-homotopic? Probably yes.
This, of course, generalizes. For example, given a loop at the North Pole of S 2 not
passing through the South Pole, we can show that it is homotopic to the constant
function of the North Pole. Consider γ : I → S 2 . Of course we can’t simply use the map
H(s, t) = (1 − t)γ(s) + t · h0, 0, 1i, since this will in general not be on the sphere, but we
can recover this by dividing by the magnitude, forcing it to lie on the sphere; hence, the
1
desired homotopy is simply F (s, t) = H(s, t) · |H(s,t)| .
3.3 Product of Paths
1 3
0 2 4 1
γ1 γ2 γ3
γ1 γ2 γ3
0 1 1 1
4 2
Figure 1: Not Associative, but homotopic
Fix a space X. Given a path γ1 from x1 to x2 , and a path γ2 from x2 to x3 , we can

get a path γ1 ∗ γ2 from x1 to x3 . Specifically, the function is defined as follows.
5
Definition. Given γ1 , γ2 : I → X, define γ1 ∗ γ2 : I → X by

 h i
γ1 (2s) s ∈ 0, 12
(γ1 ∗ γ2 )(s) = h
1
i
γ2 (2s − 1) s∈ 2, 1
It is not hard to check that this is well-defined. It is easy to verify that γ1 (2s) and
γ2 (2s − 1) are continuous, and they agree at the common point s = 21 , so this implies
that the product is continuous by the pasting lemma.
The operation ∗ is not associative; an easy way to see that (γ1 ∗γ2 )∗γ3 6= γ1 ∗(γ2 ∗γ3 ) by
evaluating at s = 1/4. However, it is associative with respect to homotopy classes. The
easy special case is to show that the above quantities are homotopic. . . by distorting time
linearly, we can establish the homotopy by a rather irritating computation. Explicitly,
the homotopy F : I 2 → X is

4s t+1



γ 1 t+1 0≤s≤ 4
t+1
F (s, t) = γ (4s − t − 1)
2 4 ≤s ≤ t+2
4

γ 4s−t−2
 t+2
≤s≤1
3 2−t 4
and this is continuous by the Pasting Lemma and composition. (Since t ∈ I, no denom-
inators are zero.)
6
Evan Chen M275 Fall 2012: Notes 4 September 5, 2012
4 September 5, 2012
4.1 Homeomorphism
Definition. X is homeomorphic to Y , denoted X ≈ Y , if ∃f : X → Y a homeomor-
phism.
Definition. A homeomorphism is a continuous functions f : X → Y such that ∃g :

Y → X is a continuous inverse; that is, g ◦ f = f ◦ g = id.
Definition. We say X is of the same homotopy type as Y , written X ' Y , if ∃f, g with
g ◦ f ' idX and f ◦ g ' idY .
This is an equivalence relation. Clearly, this is weaker than homeomorphism.
Example. R2 \ {~0} ' S1
Proof. Identifying R2 with the corresponding complex number in C, define f : C → S1 by

z
f (z) = |z| , and let g : S1 → C be an “identity” (e.g. inclusion). Evidently, f ◦g : S1 → S1
is actually equal to the identity in S1 . On the other hand, g ◦ f : C → C is basically
equal to f .
4.2 The Fundamental Group

Given X and x0 ∈ X, define the fundamental group of X based at x0 , denoted π1 (X, x0 ),
as the set of path homotopy classes of loops based at x0 . (A loop is a path with the
same start/end.) The operation of the group is essentially ∗, the concatenation of two
paths. Formally, if f, g : I → X are paths, then [f ] · [g] = [f ∗ g].
Question. Is this operation well-defined?
Answer. Yes. This isn’t hard to verify, just follow through.
Question. Is this a group operation?
Answer. Yes. Again, not hard to verify. The constant loop is the identity, and as-
sociativity is evident: more generally, if f, g, h are paths, we showed last time that
f ∗ (g ∗ h) 'p (f ∗ g) ∗ h. The inverse of [f ] is [f¯], where f¯(t) = f (1 − t) for each t.
These groups are, in general, hard to compute. Unlike homology classes, which
are rather easy to compute.
Question. Can we calculate some simple homotopy groups?
Question. Does this group depend on x0 ?
Anyways, here is an actual example.
Example. π1 (R, 0) is the trivial group.
Proof. Well, everything is nulhomotopic, so. . . heh.

In general, any contractible space has trivial fundamental groups. The converse is not
true.
Example. π1 (S 1 , 1) ∼
= Z.
7
Note that these loops are restricted to path components. If X is not path-connected,
then the path components are completely unrelated in terms of their fundamental groups.
So, it really only makes sense to discuss fundamental groups of path-connected spaces.
Now the triumph:
Theorem 3. If X is path-connected, then π1 (X, x1 ) is isomorphic to π( X, x2 ) for any

points x1 , x2 ∈ X.
Proof. In the book.
Define an equivalence relation on I as follows. Then x y iff x = y or {x, y} = {0, 1}.

In other words, this just glues the endpoints together. Now look at the quotient topology
I/ ≈ S1 . We can now rephrase loops, which are functions f : I → X with f (0) = f (1),
to functions f : S 1 → X. So we can view homotopy groups, then, as the homotopy
classes of maps from S 1 into X, denoted [S 1 , X].
Why π1 ? It turns out that πn represents the homotopy classes of maps from S n into
X.
Comment: π1 is a group, but not abelian. On the other hand, πn is abelian whenever
n≥2
8
5 September 10, 2012

5.1 Homomorphisms Induced by Functions
So far we’ve seen π1 (X, x0 ), the homotopy classes of loops in X based at x0 . If f :
(X, x0 ) → (Y, y0 ),1 , this yields a homomorphism f∗ : π1 (X, x0 ) → π1 (Y, y0 ) by [γ] 7→
[f ◦ γ]. Indeed, one can check that f∗ ([γ][ζ]) = f∗ ([γ])f∗ ([ζ]).
If f : X → Y and g : Y → Z, then (g ◦f )∗ : π1 (X) → π1 (Z) is g∗ ◦f∗ . This is powerful.
That is, given a commutative diagram of spaces, we get for free a commutative diagram
of groups. On another note, this is also completely trivial to prove; it follows immediately
by using definitions.
This is one of the functorial properties of ∗. The other functorial property is that
if idX : X → X is the identity, then (idX )∗ is the identity at the group level.
5.2 Some More Properties of Fundamental Groups

Claim. If X and Y are homeomorphic, then π1 (X, x0 ) ∼
= π1 (Y, y0 ).
Proof. Since X and Y are homeomorphic, ∃f : X → Y, g : Y → X such that g ◦ f = idX

and f ◦ g = idY . Now, we have homomorphisms f∗ and g∗ which are inverses between
the fundamental groups. This implies the conclusion.
Fact. If X is contractible, then π1 (X, x0 ) = 0 is the trivial group.2
Proof. All loops at x0 can be compressed, lol. More formally, since X is contractible,
we know that idX ' cx0 . Now consider a loop γ. Then,
idX ◦ γ ' cx0 ◦ γ
But the left-hand side is γ and the right-hand side is cx0 , implying the loop is itself
nulhomotopic and yielding the conclusion.
5.3 Homework Problem

Problem. Let A be a subspace of Rn . Let h : (A, a0 ) → (Y, y0 ). Show that if h is
extendable to a continuous map of Rn into Y , then h∗ is the trivial homomorphism.
Proof. We are given that ∃g : Rn → Y with g|A = H. Let j be the inclusion. Then
g ◦ j = h.
Now we use functorial properties. We have h∗ = g∗ ◦ j∗ . But this is a map
j∗ g∗
π1 (A, a0 ) −→ π1 (Rn , a0 ) −→ π1 (Y, y0 ).
But π1 (Rn , a0 ) = 0. So j∗ (A) = e, the identity, is trivial. So g∗ ◦ j∗ is trivial, implying

the conclusion.
h
What does this imply? For example, for a non-trivial function π1 (S 1 , 1) −→
∗
π1 (S 1 , 1),
we can’t extend it to R2 .
1
That is, f : X → Y and f (x0 ) = y0 . This generalizes to subsets in place of x0 and y0
2
This is a terrible abuse of notation. But “everyone does it”, so whatever.
9
5.4 Hat Homomorphisms

We saw last time that if α is a path from x0 to x1 , then this gives a homomorphism
α̂ : π1 (X, x0 ) → π1 (X, x1 ) by [γ] 7→ [ᾱ ∗ γ ∗ α]. It turns out this is an isomorphism!
Hence, in a path-connected space, the fundamental groups are all isomorphic. (When
not, they are completely unrelated.)
However, we still pay some attention to the base point, because the isomorphisms are
not necessarily natural.
Now notice that we can do the same thing if α is a loop! This gives an automorphism
from π1 (X, x0 ) to itself. But α̂ : π1 (X, x0 ) → π1 (X, x0 ), and now α ∈ π1 (X, x0 ). Now
the map is
α̂ : [γ] 7→ [ᾱ ∗ γ ∗ α] = [α]−1 [γ][α].
Suddenly we find that for any loop, α̂ is an inner automorphism!
5.5 Even Covering

So far, the fundamental groups are all trivial. Let’s find some more interesting ones.
Definition. Given a function P : E → B, select a base point b0 and U an open neigh-

borhood of b0 . We call U evenly covered (by P ) if P −1 (U ) is a disjoint union of open
sets (possibly infinite) such that P restricted to any of these sets is a homeomorphism.
P is then called a covering projection if this is possible for any b0 . Informally, we will
say “E is a covering projection of B” if every point has an evenly covered neighborhood.
Note that P must be surjective.
Example. Let [n] denote {1, 2, · · · , n} with the discrete topology. Take P : B ×[n] → B
by (b, k) → b. Then this is a simple example of a covering projection.
This is not interesting because B × [n] is path-connected.
Example. Take P : R → S 1 by θ 7→ e2πiθ . This is essentially wrapping the real line

into a single helix and projecting it down.
π π
This is a covering projection. For example,
if U is the arc joining cis 2012 to cis − 2012 ,
1 1
then the pre-image are intervals of the form − 4024 , 4024 + Z. For each interval we have
a homeomorphism, but this is not true for the entire function.
Note that if we take the neighborhood too big, then we may destroy our homeomor-
phism structure.
The idea is that locally, the pre-image of U looks like a product of U and what is
called a fiber ; in the example above, the fiber looks like Z.
10

6.1 Homework Stuff
Hey cool no points lost.
Comment: remember that if you want to show S has one element, it is not sufficient
to show that all elements of S are the same. You do need to additionally check that S
is not empty.
6.2 Covering Spaces and Projections

Consider a function p : E → B. Suppose that ∀b ∈ B, ∃U with b ∈ U evenly covered by
p. Then p is called a covering projection. The key example is the map p : R → S 1 given
by θ 7→ e2πiθ .
Example. Take p : S 1 → S 1 by z 7→ z 3 . This is also a covering projection (check

this). Indeed, the inverse image of a little open interval around 1 is three little intervals
2πi 4πi
centered at e 3 , e 3 , and 1, each three times shorter. We can replace the 3 with any
integer.
Definition. The fiber of a point b0 is the inverse image p−1 ({b0 }).
Sidenote: the covering of a space Y by Y × A, where A has the discrete topology,

is called trivial. (See last class’s example.) It turns out that coverings of contractible
spaces are essentially trivial.
p1 p1 p1 ×p2
Fact. If E1 −→ B1 and E2 −→ B2 are covering projections, then so is E1 × E2 −→
B1 × B2 .
In figure 53.4, we can view S 1 × S 1 as a torus. Take the covering projection p × p,

where p : θ 7→ e2πiθ . The fiber of (1, 1) ∈ S 1 × S 1 is precisely Z2 , and the inverse of a
small patch on the torus is small patches at each lattice point.
If B is connected, then the cardinality of the fiber at any point in B is constant over
all the points in B.
The map (z, w) 7→ (z m , wn ) taking S 1 × S 1 to S 1 × S 1 produces a wrap for a bagel!
Covering projections must be onto since neighborhoods around a point with an empty
fiber are not going to be nonempty by continuity and such. h/handwavei.
Fact. If p : E → B is a covering projection, then p is an open mapping, i.e. p(U ) is

open in B if U ⊆ E is open.
Since p is surjective, then p is a quotient map. That is, if we put an equivalence

relation ∼ on E, then B = E/ ∼= {equiv classes}. This is corollary 22.3 in the book.
p is also a local homeomorphism, i.e. given e ∈ E, ∃V a neighborhood of e such that
p|V : V → p(V ) is a homeomorphism. The proof is not hard; just take an evenly covered
neighborhood of p(e), and use the definition of a projection map. However, not all local
homeomorphisms are covering projections.
Example. Take (0, 2) → S 1 by θ 7→ e2πiθ . This is a local homeomorphism. But now take
a small neighborhood around 1 in S 1 , say the arc joining e−2πi to e2πi . We get small
intervals (0, ), (1−, 1+) and (2−, 2). But this map is not a homeomorphism between
(0, ) to the arc. So local homeomorphisms are not necessarily covering projections.
11
6.3 Group Actions

Definition. Let G be a group and let X be a set. A left action of G on X is a
µ
function (almost a “binary operation”) G × X −→ X by (g, x) 7→ µ(g, x), which will be
abbreviated gx. This action must obey (i) g1 (g2 x) = (g1 · g2 )x for all g1 , g2 ∈ G and
x ∈ X, and (ii) 1x = x for all x ∈ X.
We want to put a topology on G × X, with the discrete topology on G, with µ a

continuous function. For our purposes, G will in general be finite.
Note that if we fix g = g0 , then x 7→ g0 x is a bijection (its inverse is x 7→ g0−1 x).
Example. Let G = Z (additively), X = R, and an action n · x = x + n. It’s not hard

to check this is a group action. Take the equivalence relation whose equivalence classes
are Gx, the orbits of x. If we mod out by this equivalence relation, then this becomes
the covering projection R → S 1 discussed before.
12

7.1 Group Actions
If G is a group and X is a set (space), a right action of G on X is a function X × G → G
that send (x, g) 7→ xg, such that (xg1 )g2 = x(g1 g2 ) for all g1 , g2 ∈ G, x ∈ X and x1 = x
for every X.
For example, X = R and G = (Z, +), then take the action as (x, n) 7→ x + n. We can
look at the orbit of a given x ∈ X.
Example. Let Z/nZ = ht | tn = 1i act on the circle S 1 by S 1 × G 7→ S : (z, ti ) 7→ z · ζ i ,

where ζ is a primitive nth root of unity. Each orbit has size n.
Example. Since (S 1 , ·) is itself a group, it acts on itself by S 1 × S 1 → S 1 : (z, w) 7→ zw.

Now the orbit is the entire circle.
Definition. An action is transitive if there is only one orbit.
Given an action, we can define an equivalence relation ∼ on X by x1 ∼ x2 if they are

in the same orbit. Obviously ∼ is terribly uninteresting when the action is transitive.
This partitions X into equivalence classes.
Given an action and the corresponding equivalence relation ∼, define
X 0 = X/ ∼
= {[x] : x ∈ X}
= {orbits}
We have a function that x takes to its equivalence class, as usual.
7.2 Covering Projections

If X has a topological structure, and we require that the map X × G → X is continuous
where G has the discrete topology, then for any fixed g ∈ G, we get a homeomorphism
from X to Xg (since it has an inverse).
This often provides a covering projection from X to X/ ∼.
Let us consider the example G = (Z, +) again. It is not hard to show that [0, 1) is a
complete set of representatives. This is called the fundamental domain for the action.
Now X/ ∼≈ S1 . We verify this is a covering projection for this particular instance. Each
little interval on S 1 has a pre-image composed of several little intervals.
Let’s take n = 4 in G = Z/4Z, X = S 1 as above. A fundamental domain is half-open
arc joining 1 to i. The resulting covering projection is the map S 1 → S 1 : z 7→ z 4 .
Fact. Product of group actions are group actions. If Gi acts on Xi , then G1 × · · · × Gn

acts on X1 × · · · × Xn by (x1 , · · · , xn )(g1 , · · · , gn ) = (x1 g1 , x2 g2 , · · · , xn gn ).
Example. Take X = R2 and G = Z × Z and consider the right action (x, y)(n, m) =
(x + n, y + m). The fundamental domain is [0, 1)2 , which ends up as a box. This is a
torus; we now have a covering projection from R2 to the torus, which can the be viewed
as S 1 × S 1 .
This hold in general, as follows.
Fact. (X1 × X2 · · · × Xn )0 = X10 × X20 × · · · Xn0 .
13
It is not necessarily the case that we always get a covering projection. Consider the
action S 1 × S 1 → S 1 by (z, w) 7→ zw. This action is transitive, so S 1 / ∼ is a group with
one element. Its pre-image is the entire S 1 . However, we need to write this as a disjoint
union of homeomorphic “spaces” (which are in this case points). But these points are
not open in S 1 .
Coverings which arise from group actions are called regular. In regular coverings
E → B, the coverings correspond to the normal subgroups of π1 (B, b0 ).
Definition. A topological group (G, ·) is a group G with a topology τ for which · is a

continuous function.
In group actions, if G is a topological group we need to impose the discrete topology

on the action X × G → X.
7.3 Liftings
f p
Suppose X −→ B is continuous and E −→ B is a covering projection, then a mapping
f˜
X −→ E is called a lifting if p ◦ f˜ = f .3
Liftings don’t always exist. However, suppose we pick X to be the interval I = [0, 1],
and suppose f (0) = b0 ∈ B. Select a e0 in the fiber p−1 ({b0 }). If B is path-connected,
then there exists a unique lifting f˜ with f˜(0) = e0 .
To prove this, we use the following lemma.
Definition. If (X, d) is a metric space and {Oi } is an open covering for X, then q is a
Lebesgue number for the covering if any set of diameter at most q is contained entirely
in one of the Oi .
Definition. In a metric set X, the diameter of a set A ⊆ X is defined by sup{d(x, y) |

x, y ∈ A}.
Lemma 4 (Lebesgue Number Lemma). A Lebesgue number exists for every covering
whenever X is a compact metric space.
Let {Ui } be a collection of evenly-covered open subsets of B. Look at the collection

of open subsets {f −1 (Ui )}, which is an open covering of I. Also, this covers I. Applying
δ
the Lebesgue Number Lemma. Let n be a large integer with 1/n < 2012 , where δ is the
Lebesgue number. This chops I into n pieces, each of which is contained in some evenly
covered neighborhood.
Now we can just construct our lifting greedily. For each interval Ik = [k/n, (k + 1)/n]
just define f (Ik ) by using the inverse of the homeomorphism determined by the definition
of even covering. Then apply the pasting lemma on f .
We can even lift homotopies. By replacing I with I × I, we can do the same thing
since I × I is metric and compact.
3
The lifting problem, in category theory, is the dual of the inclusion problem.
14

8.1 Path homotopies lifted
γ
If I −→ B is a path is a path γ(0) = b0 , there exists a unique lifting of γ starting at e0 .
F
Furthermore, if I × I −→ B is a homotopy with F (0, 0) = b0 , then there exists a
unique lifting of F to F̃ with F̃ (0, 0) = e0 .
If F : I × I → B is a path homotopy, then F̃ is also a path homotopy. By uniqueness
of lifting, it is not hard to show that F̃ (0, t) and F̃ (1, t) are constant functions in t. This
implies that path homotopies also lifts to path homotopies.
8.2 Fundamental Lifting Correspondence

P
Suppose E −→ B is a covering projection, and fixed e0 ∈ E is in the fiber of some fixed
b0 ∈ B.
Claim (Fundamental Lifting Correspondence). We have a function4 Φ : π1 (B, b0 ) →

p−1 ({b0 }) with [γ] 7→ γ̃(1), where p ◦ γ̃ = γ and γ(1) = γ(1) = b0 and γ̃(0) = e0 .
Proving this claim invokes the fact
Fact. [γ1 ] = [γ2 ] ⇒ γ1 'p γ2 .
The fact that this function is well defined now follows from γ̃1 'p γ2 ⇒ γ̃1 (1) = γ̃2 (2).
Exercise. If E is path connected, then Φ is surjective.
Proof. Suppose e1 ∈ p−1 ({b0 }). Suffices to prove ∃[η] with Φ([η]) = e1 .
We know there is a path η̂ joining e0 to e1 . The function p ◦ η̂ is now a loop in B. We
claim that η = p ◦ η̂ is the desired loop. Indeed, by uniqueness of lifting, η̃ = η̂! (All we
did was project down and lift back up). Now we’re done.
Exercise. If E is simply connected, then Φ is one-to-one.
Proof. Suffices to prove that if Φ([γ1 ]) = Φ([γ2 ]), then [γ1 ] = [γ2 ].
Lift to γ̃1 and γ̃2 . In E we get γ̃1 'p γ̃2 since their endpoints coincide. Project down!
If F̃ is a path-homotopy between γ̃1 and γ̃2 , then F = p ◦ F̃ is now a path-homotopy
between γ1 and γ2 .
In particular, we have a one-to-one correspondence between the cardinality of the fiber

as well as the cardinality of the fundamental group of B.
8.3 Making Headway on Fundamental Groups

p
Recall our standard example R −→ S 1 . The fiber above 1 ∈ S 1 is the integers. Now
since R is contractible and thus simply connected, we find that π1 (S 1 , 1) is in a one-to-
one correspondence between the integers Z. (This doesn’t imply the group is actually
the integers, just that it’s countable and
not finite.)
Now consider 1 in S 2 . Let G = Z2 = T | T 2 = 1 . Now G acts on S 2 by T · ~x = −~x.
This essentially takes a point to the point diametrically opposite it. Modding out by
that action, we get S 2 / ∼ as a hemisphere. The result is S 2 / ∼= RP 2 , called real
projective 2-space.
4
This is actually just a function, since it takes a group. There is no sense of “continuity” here, etc.
15
p
This gives a covering projection S 2 −→ RP 2 (the inverse image of a patch is two
patches diametrically opposite it). But we know that π1 (RP 2 , [1]) is in a one-to-one
correspondence with Z2 . Since there are not that many groups of order two, we find
that π(RP 2 , [1]) ∼
= Z2 . Finally a nontrivial fundamental group!
Let’s try and understand RP 2 . It seems to be the northern hemisphere, with half the
equator included. If we are Santa and look down, we see this is a disk with half the
boundary included. We call it a manifold since locally it looks like Euclidean space.
In this image of a disk, the nontrivial loop is just a path from 1 to −1.
Fact. The fundamental group of X × Y is isomorphic to the fundamental groups of X

and Y .
Since there exist spaces Ln with π( Ln ) ∼

= Zn for each n, via the Fundamental Theorem
of Abelian Groups, we can construct spaces having a fundamental group of any arbitrary
finite abelian group, or even a finitely generated abelian group once we know π(S1 ) below.
8.4 Fundamental Group of the Circle

Φ
We showed earlier that there is a one-to-one correspondence π1 (S 1 , 1) ↔ Z. We now want
to show that Φ is a group homomorphism (somehow); since we have tho correspondence,
we will be done.
That is, keeping in mind that Φ returns integers, we want to show that
Φ([γ1 ] · [γ2 ]) = Φ([γ1 ]) + Φ(γ2 )
Suppose that Φ([γ1 ]) = n = γ̃1 (1), where we view n ∈ R, and Φ([γ2 ]) = m = γ̃2 (1).
But [γ1 ] · [γ2 ] = [γ1 ∗ γ2 ]. Take a lifting of γ1 ∗ γ2 starting at zero defined by
0 ≤ s ≤ 21
(
γ̃1 (2s)
ζ̃ : I → R by ζ̃(s) = 1
.
n + γ̃2 (2s − 1) 2 ≤s≤1
It is trivial to check that this is well-defined and continuous. But one can check that
this is a lifting. Also, we have easily enough that
p(γ̃1 (2s)) = γ1 (2s)

p(n + γ̃2 (2s − 1) = p(γ̃2 (2s − 1))
= γ2 (2s − 1)
Hence, ζ̃(s) is a lifting. Since liftings are unique, we get Φ([γ1 ∗ γ2 ]) = n + m, implying
the conclusion.
16

Poincare’s Conjecture is algebraic topology. Most of algebraic topology is 20th century.
Names: Hopf, Alexandrov, Eilenberg-Steenrod. Euler is sort of considered the father
of topology. As a discipline, topology is quite recent.

p
Same setup as usual. Select b0 ∈ B and a covering projection E −→ B. Up in E we have
p−1 ({b0 }) which we will abbreviate to just p−1 (b0 ). Then we pick some e0 ∈ p−1 (b0 ).
Claim. The group π1 (B, b0 ) acts on p−1 (b0 ) from the right with p−1 (b0 ) × π1 (B, b0 ) →
p−1 (b0 ) via
(e, [γ]) 7→ γ̂(1)
where γ̂ is the unique lifting of γ into E.
We need only show that (e[γ])[η] = (e[γ][η]). This is left to the diligent student.
9.2 Orbits
What is the orbit? eπ1 (B, b0 ). Any time we have an orbit xG, then this is essentially
(i.e. can be modelled by) as the set of right cosets {Hg : g ∈ G} with G acting on the
right by (Hg)g 0 = H(gg 0 ). Here H = Gx is the isotropy subgroup, or the stabilizer, of
x.
We need to find which elements of π1 (B, b0 ) fix e.
We want to characterize e with e[γ] = e. If γ̂(1) = e, then γ̂ is actually a loop. But γ̂
lifts γ; i.e. [p ◦ γ̂] = [γ]. This is p∗ ([γ̂]).
Claim. [γ] fixes e if and only if [γ] ∈ p∗ (π1 (E, e))
Hence, the stabilizer of e is precisely p∗ (π1 (E, e)) and the orbits are right cosets of
this set.
However, we saw last time that if E is path-connected, then the action is transi-
tive. That is, eπ1 (B, b0 ) = p−1 (b0 ). So the entire fiber can be modelled as cosets of
p∗ (π1 (E, e)).
(Based on the restrictions of E, it’s connected iff path-connected.)
Now recall:
Theorem 5 (Lagrange’s Theorem). The number of cosets of G divides |G| for finite G.
So if we know that π1 (B, b0 ) is finite, then we know that the fiber p−1 (b0 ) is pretty
limited (given E is path-connected). Remember, we have a transitive action on a finite
group on a finite set.
p p∗
Claim. If E −→ B is a covering, then π1 (E, e) −→ π1 (B, b0 ), p∗ is an injective homo-
morphism.
In some sense, the smaller the fiber, the closer E is to being B.
17
9.3 Using the Fundamental Group of the Circle

j
Let j be inclusion; for A ⊆ X, the map A −→ X is called inclusion.
Definition. We say that A is a retract of X via the retraction r if ∃r : X → A such

that r |A = idA ⇔ r ◦ j = idA ; that is r(a) = a for all a ∈ A.
Example. Let A = R × {0} be the x-axis in X = R2 . The retraction r will be just

projecting down.
We can generalize this to A = Y × {z0 } ⊆ Y × Z = X. This is just because A is
homeomorphic to Y . In a loose sense, “projecting down into a component is a retraction”.
Example. A Moebius strip can be projected down into its equator.

j∗ ∗ r
Bringing stars back in, note that r∗ ◦ j∗ is the identity π1 (A, a0 ) −→ π1 (X, a0 ) −→
π(A, a0 ). Hence j∗ is injective and r∗ is surjective.
If j ◦ r is homeomorphic to the identity, then something interesting happens; in a
sense, you can collapse in a controlled fashion.
Theorem 6 (No-Retraction Theorem). Consider the two-ball B 2 = {z ∈ C | kzk ≤ 1}

and its boundary S 1 = {z ∈ C | kzk = 1}. (Note that S 1 ⊆ B 2 ). Then S 1 is not a
retract of B 2 .
j∗
Proof. If we had such a retraction, then π1 (S 1 , 1) −→ π1 (B 2 , 1) would have to be injec-
tive. But this is a map from Z to {0}.
18

10.1 Meteorology
Theorem 7. Suppose we have two continuous functions f, g : S 2 → R. Then we can
consider a function Φ : S 2 → R2 as Φ = f × g. Then, there exists z0 ∈ S 2 with
Φ(z0 ) = Φ(−z0 ) by a theorem.
If we let S 2 be the earth, and f , g be temperature and barometric pressure. Then

the theorem implies that there exists two antipodal points on the Earth where the
temperature and pressure coincide. This is a homework problem.
Catastrophe Theory (René Thon)
10.2 Topology
B 2 does not retract onto its boundary S 1 ; i.e.
@r : B 2 → S 1 : r |S 1 = idS 1
We proved this using the fact that B has a trivial fundamental group, but S 1 does not.
The most well-known consequence of this is
Theorem 8 (Brower Fixed Point Theorem). If f : B 2 → B 2 is continuous, then ∃x0 ∈

B 2 such that f (x0 ) = x0 ; i.e. x0 is a fixed point of f .
Munkres uses a proof using vector fields. The proof that most people usually see
doesn’t rely on this.
Proof. Suppose on the contrary that f : B 2 → B 2 has no fixed points. Define r : B 2 →

S 1 by taking the ray starting at f (x), passing through x, and intersecting that with its
boundary. This is r(x). This function is well-defined since f (x) 6= x.
One can check that r is continuous. (Handout on website). But r is clearly a retraction,
and this is a contradiction.
This generalizes to any B n .
Remark. We say that X has the fixed point property if ∀f : X → X continuous, f has
a fixed point.
(i) B n has the fixed point property for every n.
(ii) S 1 doesn’t have the fixed point property (consider rotation). R doesn’t either via
translation. Neither does S 1 × S 1 or Rn .
(iii) If X ≈ Y and Y has the fixed point property, then so does X. In other words, the
property is a topological invariant. If we have h : X → Y a homomorphism, then
for any f : X → X we can construct f˜ = h ◦ f ◦ h−1 . We know that f˜(y0 ) = y0 for
some y0 . Then x0 = h−1 (y0 ) is a fixed point of f .
We only talk about a fixed point when we have a function to itself.
19
10.3 Vector Fields

If O is a non-retarded subset of Rn , then a vector field on O is a function from O to Rn .
A “nowhere vanishing vector field” is a v such that v(~x) 6= ~0 for every ~x ∈ O. In that
case we can have v : O → Rn \ {~0}.
Theorem 9. If v is a nowhere vanishing vector field on B 2 and w is its restriction to

S 1 ⊆ B 2 , then there exists xout and xin in S 1 for which w(xout ) is outward-pointing and
w(xin ) is inward-pointing.
For x, we say x is outward-pointing if w(x) = kx for some scalar k > 0. If k < 0 then
instead we say it is inward-pointing. This result implies the Brower fixed point theorem.
Vector fields can be viewed as e.g. a part of the ocean, where the vector corresponds
to the strength and direction of the current, etc.
A tangent-vector field is trickier; the output of v(x) needs to be “tangent” to the point
(the definition of tangent depends on the space). In other words, the domain depends
on f .
Theorem 10 (Hairy Ball Theorem). S 2 does not have a nowhere vanishing tangent
vector field.
This is of course not true if we replace S 1 by S 2 .
Fact. There exists a non-vanishing tangent vector fields on S n if and only if n is odd.
Since S 2n+1 ⊆ Cn+1 ; we can write it as
S 2n+1 = {(z0 , · · · , zn ) | z02 + · · · + zn2 = 1}
so we can get a tangent vector field by taking (z00 , · · · , zn0 ) where zk0 = −y + xi for
zk = x + yi.
Which S n are parallelizable? Something like {S 1 , S 3 , S 7 }. Prove only in early 1960’s
by Adams.
10.4 Algebra
R1 is a field. R2 can be thought of as C, which form a field.
For R3 we don’t have what we want.
However, R4 has a quaternion structure H. We get a division algebra (almost a field,
except non-commutative multiplication). In R∗ we have the octonians O, but we lose
associativity.
In R16 we have the Cayley numbers, which loses even more. These things are called
algebras. It turns out that we can’t have algebra structures for sufficiently large Rn
which is closely related to the parallelizable spheres above.
Punk hairy ball theorem.
20
Evan Chen M275 Fall 2012: Notes 11 October 1, 2012
11 October 1, 2012
11.1 Something
f
If X −→ Y is a function of sets X, Y , and we have an equivalence relation ∼ which
is preserved under f , then we have an “induced” function fˆ : X/∼ → Y from defined
by fˆ([x]) = f (x). This function is generally continuous when you want it to be (by
composition).
Claim. If h : S 1 → X is nulhomotopic, then there exists an continuous extension

k : B 2 → X of H.
F
Proof. We are given that there is a homotopy S 1 × I −→ X with F (z, 0) = h(z) and
F (z, 1) = cx0 (z) for all z (where cx0 is the constant map sending everything to x0 .)
Define sim on S 1 × I by (z, t) ∼ (z, t) for t < 1 and
(z, 1) ∼ (z 0 , 1)∀z, z 0 ∈ S 1 .
F̂
This is clearly an equivalence relation. Now define (S 1 × I)/∼ −→ X as above. Since
z ∼ z 0 ⇒ F (z, t) = F (z 0 , t) is evident, we see that F̂ is a continuous function. But we
claim that B 2 ≈ (S 1 × I)/∼. Indeed, it’s like a cone, which can be flattened to a disk.
This proves the claim.
h
Any time we have Z −→ X nulhomotopic, we can try the same trick to get an extension
k
CZ −→ X, a “cone”-like version of Z. The cone on Z is actually contractible, since we
can pull everything towards the tip. As a particular example, CS n−1 = B n ; we have
invoked n = 2.
The converse of this turns out to be true.
11.2 Spheres to Spheres

Let [X, Y ] denote the homotopy (equivalence) classes of maps X → Y . It turns out that
π1 (Y ) = [S 1 , Y ], and πn (Y ) = [S n , Y ]. If you understand πn (S m ) for all n and m, then
you completely understand [X, Y ] for reasonable X and Y . This means that it’s sort of
a lost cause, but we do have some “easy” facts:

trivial if n < m


m
πn (S ) = Z if n = m

?

if n > m
For instance, it turns out that π3 (S 2 ) = Z, but πn+1 (S n ) = Z2 when n > 2.
11.3 Antipodes
Definition. If h : S n → S m , then we call h antipode-preserving if h(−x) = −h(x).
It does not make sense to talk about antipodes in most spaces.
Claim. If h : S 1 → S 1 is antipode-preserving (and continuous), then it is not nulhomo-

topic.
21
Proof. Let q : S 1 → S 1 by q(z) = z 2 . Let g = q ◦ h. (See diagram on pg 357).

We want to show that if q(x1 ) = q(x2 ) ⇔ x21 = x22 , then g(x1 ) = g(x2 ). But this is
obvious; h is antipode-preserving! So, we get an induced function k : S 1 → S 1 with
k ◦ q = q ◦ h by k(x2 ) = g(x); this is well defined by the property we have just shown.
Now recall that q is a covering projection. Therefore, q∗ : π1 (S 1 ) → π1 (S 1 ) is a
injective (in general, p∗ : π1 (E) → π1 (B).
k∗
Consider π1 (S 1 , 1) −→ π1 (S 1 , 1). Consider a path γ : I → S 1 with γ(0) = 1 and
γ(1) = −1. First, we claim that [q ◦ γ], the class of a loop at S 1 , is nontrivial. It cannot
be trivial because this would imply that q ◦ γ was nulhomotopic; but the lifting of this
is γ which is not a loop. But now,
k∗ : [q ◦ γ] 7→ k∗ ([q ◦ γ]) = [k ◦ q ◦ γ] = [q ◦ h ◦ γ]
But we claim that [q ◦ h ◦ γ] is nontrivial. Note that q ◦ (h ◦ γ) lifts to h ◦ γ, which is

not a loop; indeed, h(γ(0)) = h(1) 6= −h(1) = h(−1) = h(γ(1)). We conclude that k∗ is
nontrivial.
Finally: the homomorphism k∗ is injective, since it is a nontrivial homomorphism from
Z to Z (check this is sufficient). Also, q∗ is injective as mentioned earlier. But
q∗ ◦ h∗ = k ∗ ◦ q∗
from which we find that h∗ is injective.
Remark. In fact, q∗ : π1 (S 1 ) → π1 (S 1 ) has 1 7→ 2. Indeed, view q as a covering

projection; we have that the fiber (which has cardinality 2) is the right index.
Next week: ham sandwich theorem.
22
12 October 3, 2012
12.1 Something
Want to show the identity i : S n → S n is not nulhomotopic.
If @r : B n+1 → S n a retraction, then f : S n → X is nulhomotopic is equivalent to F
extending to f˜.
f
Sn - X
-
f˜
j
?
n+1
B
If the identity were homotopic, then upon taking f the identity, f˜ ◦ j is the identity,
i.e. a retraction.
12.2 Geography
f
You can’t “map”, i.e. embed S 2 into R2 . That is, there is no S 2 −→ R2 which is
homeomorphic to its image, because the Borsuk-Ulam Theorem implies ∃x0 ∈ S 2 with
f (x0 ) = f (−x0 ); in particular, f cannot be injective.
For all n, if f : S n → Rn , Borsuk-Ulam ∃x0 ∈ S n such that f (−x0 ) = f (x0 ). That is,
Borsuk-Ulam generalizes to arbitrary n instead of just n = 2.
12.3 Ham Sandwich Theorem

Theorem 11 (Ham Sandwich Theorem). Given two slices of bread in R3 with as lice of
ham in between, it’s possible to slice the sandwich in such a way that both silecs of bread
and the ham slice are divided in half (volume-wise).
The proof is analogous to the one given in the book of the bisection theorem.
Proof. Consider R3 × {3} ⊆ R4 which contain the pieces of the sandwich.

We want to construct a continuous map F : S 3 → R3 because Borsuk-Ulam will say
∃x0 ∈ S 3 such that F (x0 ) = F (−x0 ).
Given ~u ∈ S 3 , there is a unique hyperplane (≈ R3 ) which passes through the origin
and is perpendicular to ~u. Denote this hyperplane by P~u .
Let i = 1, 2, 3 correspond to the first slice, ham, and second slice respectively. Let
fi (~u) be the portion of the volume of the i-th slice of the food that lies on the same side
as ~u.
If ~u = h0, 0, 0, 1i, then fi (~u) = Vol(Ai ). On the other hand, fi (h0, 0, 0, −1i) = 0, since
we see no sandwich looking down. On the other hand, for any other vector, P~u intersects
R3 × {3} in some two-dimensional plane. On the other hand, fi (~u) + fi (−~u) = Vol(Ai ),
since this adds up the sandwich on both sides.
Waving our hands, we get that fi is continuous for the slices. Now, define F : S 3 → R3
by
F (~u) = (f1 (~u), f2 (~u), f3 (~u))
Now Borsuk-Ulam says that ∃~u0 : F (~u0 ) = F (−~u0 ). This says that
∀i : fi (~u0 ) = fi (−~u0 )
1
But we know that their sum is Vol(Ai ), so for each i we have fi (~u0 ) = 2 Vol(ai ).
23
12.4 Retracts
We say A ⊆ X is a retract if there exists a retraction; i.e. a function r : X → A such
that r ◦ j = idA , where j : A → X is the inclusion.
In other words, r is a “left-inverse” function for j. If r were a right-inverse for j; i.e.
j ◦ r = idX , then A = X and this is quite stupid. Let’s weaken the condition that r is a
right-inverse, as follows
Definition. If r is a retraction onto A ⊆ X such that j ◦r ' idX we call r a deformation
retraction and say that A is a deformation retract of X.
Munkres actually requires something stronger; not all books agree on a definition here.
Example. A = {~0} ⊆ Rn = X is a deformation retract of X. Let r be a constant map
to zero; we can check that j ◦ r ' idX simply because Rn is contractible.
Example. In Rn , the map r : hx1 , x2 , · · · , xn i = hx1 , 0, 0, · · · , 0i is a defromation re-
traction.
Now for a less trivial example:
Example. The topologist’s comb (also called the harmonic comb) is the set X ⊆ R2
defined by
[ 1
X= { } × I × ({0} × I) ∪ (I × {0})
n≥1
n
If A = {(0, 1)}, then A is a deformation retraction of f . We can deform it continuously
by smashing the teeth onto the base, pushing that to (0, 0) and the brining the whole
thing up to (0, 1).
Example. The topologist’s fan (also called the harmonic fan) is the set X ⊆ R2 defined
by joining the segments (0, 0) to (1, n1 ) for each integer n, as well as (1, 0). Again
A = {(1, 0)} is a deformation retraction
These aren’t terribly useful deformation retractions, so the alternate definition called
by some authors as a strong deformation retraction is given by
Definition. A strong deformation retraction r : X → A is a deformation retraction
such that the homotopy F between j ◦ r to idX fixes A; i.e. the homotopy doesn’t fix A.
Both the examples above fail under this stronger definition; on the other hand, the
earlier two examples (contractible spaces and projecting onto the x-axis) still are okay.
Munkres just calls this a deformation retraction, so we will do so as well.
12.5 Homotopy Equivalence

f g
The punch line is as follows: spaces are homeomorphic if X −→ Y and Y −→ X satisfy
f ◦ g = idY and g ◦ f = idX . Retractions weaken one of these to g ◦ f ' idX .
It turns out that if we weaken both conditions, we get an equivalence relation called
homotopy equivalence. That is, spaces X and Y are homotopy equivalent if ∃f : X →
Y, g : Y → X is f ◦ g ' idY , g ◦ f ' idX . For example, a point is homotopy equivalent
to Rn .
It turns out that algebraic topology is not able to distinguish between homotopy
equivalent spaces; homology groups, co-homology groups, and the fundamental groups
π1 (X) and even π2 (X), all coincide.
Thus, algebraic topology cannot distinguish between a point and Rn . Also, it cannot
distinguish between the male and female symbols, since both are homotopy equivalent
to S 1 .
24
13 October 8, 2012
13.1 Homotopy Equivalence
Last time we had an equivalence relation, the homotopy type.
Definition. A map f : X → Y is called a homotopy equivalence if ∃g : Y → X such

that
g ◦ f ' idX
f ◦ g ' idY
We say X is homotopy equivalent to Y , denoted X ' Y , if there exists a homotopy

equivalence between them. We say that X and Y have the same homotopy type
For example, {~0} ' Rn , since Rn is contractible.

Last time we said that if X ' Y , then πn (X) ∼= πn (Y ) for each n, and the homology
groups and co-homology groups are all isomorphic as well. That is, most of the constructs
of algebraic topology cannot distinguish between homotopic groups.
This is not too crippling, though:
Fact. If n 6= m, then Rn 6≈ Rm .
Proof. We may assume WLOG that n ≥ m + 1. Recall S m ⊆ Rn . If we had a homeo-

morphism
h - m
Rn R
-
h˜
Sm
So h̃, the restriction of h to S m , must be injective. But the Borsuk-Ulam Theorem

implies that this is not possible.
So there are ways to show spaces X and Y are different even if they are homotopy
equivalent.
13.2 Retraction
Recall that A ⊆ X is a (strong) deformation retract if ∃r : X → A such that r ◦ j is the
identity on A and j ◦ r 'A idX . (See the first lecture).
It turns out that if you take the category of topological spaces, you can make a new
category hTop which are equivalence classes of homotopy classes. In Top, the objects are
topological spaces and the morphisms are continuous functions. In hTop, the objects are
homotopy classes of maps, and the morphisms are equivalence classes of maps. Hence,
the arrows are not actually functions between objects. Hence, the space hTop is not
concrete.
If A is a deformation retract of X, then A is the same homotopy type as X.
Example. The space C \ {−1, 1} is homotopy equivalent to pair of tangent circles; i.e.
the figure eight. These spaces are both deformation retracts of C \ {−1, 1}, but neither
is a deformation retract of the other; however they are of the same homotopy type.
25
It turns out that f : X → Y is a homotopy equivalence, we can build a space Mf such

that X, Y ⊆ Mf are deformation retracts of Mf .
Consider a “cylinder”-like X × I and Y . Take X × I + Y the direct sum topology
(that is, just put the spaces next to each other).
Take Mf = [(X × I) + Y ]/ ∼, where (x, 1) ∼ f (x). Think a top hat.
Now X ≈ X × {0}. So we can collapse everything to Y by compressing the hat down;
so Y is a deformation retraction of this new space. (This does not use the fact that f is
a homotopy.)
This idea is called a mapping cylinder. (Mapping cones are like witch’s hats.)
13.3 More About Categories

If f : X → Y and g : Y → X are inverses, then we can get f∗ : π1 (x) → π1 (Y ) and g∗
similarly.
f,g
Lemma 12. Consider two maps (X, x0 ) −→ (Y, y0 ). If f ' g via a homotopy H and
H(x0 , t) = y0 for all t, then
f∗ ,g∗
π1 (X, x0 ) −→ π1 (Y, y0 )
are homomorphisms and f∗ = g∗ .

Proof. If [γ] ∈ π1 (X, x0 ), then we need f∗ ([γ]) = g∗ ([γ]). But the left-hand side is [f ◦ γ]
and the right-hand side is [g ◦ γ]. We want these to be path-homotopic; that is,
f ◦ γ 'p g ◦ γ
Now define K : I × I → Y by
K(s, t) = H(γ(s), t)
and this does exactly what we want (check this).

f,g
Now, what if X −→ Y but f (x0 ) = y0 and g(x0 ) = y1 ? If f ' g by a homotopy
H : X × I → Y ...
Let’s look at α(t) = H(x0 , t). Now α(0) = f (x0 ) = y0 , and α(1) = g(x0 ) = y1 . So, we
get for free a path α from y0 to y1 .
So we get a commutative diagram
f∗
π1 (X, x0 ) - π1 (Y, y0 )
g∗
α̂
-
π1 (Y, y1 )
Then f∗ and g∗ are most certainly not the same (their domains are different), but if we
consider a loop [ζ] ∈ π1 (Y, y0 ), then we have α̂([ζ]) = [ᾱ ∗ ζ ∗ α]. Remember that α is an
isomorphism.
It turns out that g∗ = α̂ ◦ f∗ . In particular, f∗ is surjective/injective/trivial iff g∗ is
surjective/injective/trivial, etc.
h
In particular, if X −→ Y is nulhomotopic with h(x0 ) = y0 , then h∗ : π1 (X, x0 ) →
π1 (Y, y0 ) is trivial.
Anyways,
26
Theorem 13. If f : (X, x0 ) → (Y, y0 ) is a homotopy equivalence, then f∗ : π1 (X, x0 ) →

π1 (Y, y0 ) is an isomorphism.
Why isn’t this obvious? We could potentially get

f g f
(X, x0 ) −→ (Y, y0 ) −→ (X, x1 ) −→ (Y, y1 )
that is,
(fx0 )-
∗
π1 (X, x0 ) π1 (Y, y0 )
g∗
(fx1 )-
∗
π1 (X, x1 ) π1 (Y, y1 )
But because of our alpha,
g∗ ◦ f∗ = (g ◦ f )∗ = α̂ ◦ (idX )∗ = α̂
So we still have an isomorphism, but g∗ is not the inverse; it goes to the wrong place!
Theorem 14 (Seifert-van Kampen Theorem). If X = U ∪ V where U and V are open

sets, then π1 (X, x0 ) is related to π1 (U, x0 ), π1 (V, x0 ). More details later.
27
14 October 10, 2012

How to get a fundamental group of a space from the fundamental groups of some pieces
from the space?
If U, V ⊆ X, are open, U ∩ V is path-connected (and nonempty), and U, V are simply
connected, then U ∪ V = X, then U ∪ V is simply connected.
This was a homework problem. It was quite good. More generally, consider open
U, V ⊆ X with x0 ∈ U ∩ V path-connected and X = U ∩ V . In essence, π1 (X, x0 ) can
be written as a “product” of π1 (U, x0 ) and π1 (V, y0 ). What product might this be?
14.1 Free Product

Consider groups H, K ≤ G and H ∩ K = {1}. Define a product
H ∗ K := hH, Ki ≤ G
as the smallest subgroup of G containing both H and K, also known as the subgroup of
G generated by H and K. This consists of words of elements of H and K.
This is called the (internal) free product of H and K.
Here is an example of an external free product:
Example. Let H = hai ∼

= Z and K = hbi ∼
= Z. Then
H ∗ K = hwords of a and bi
that is, the elements all look like an1 bm1 · · · ank bmk , where ni , mi ∈ Z.
This is the free group on two generators.
Example. Let H = a | a2 = 1 and K = b | b2 = 1 . Then H ∗ K is the set of words

with letters a and b such that neither the pattern aa nor bb appears.
In general, external free products are huge.
14.2 Von Kampen

If U ∩ V is simply connected, then we almost have π1 (X, x0 ) ∼
= π1 (U, x0 ) ∗ π1 (V, x0 ).
Consider the inclusions:
U
-
jU
iU
x0 ∈ U ∩ V X
-
iV
jV
-
V
Where i and j are inclusions. We have homomorphisms
(jU )∗ : π1 (U, x0 ) → π1 (X, x0 )

(jV )∗ : π( 1V, x0 ) → π1 (X, x0 )
28
which preserve the base-point x0 .

So the precise statement is that if U ∩ V is simply connected, then
π1 (X, x0 ) ∼
= (jU )∗ (π1 (U, x0 )) ∗ (jV )∗ ((π1 (V, x0 ))).
and the main idea is basically the same as the specific problem we did.
Example. Consider the figure eight. Consider two copies of S1 called U and V with
X8 = U ∩ V , U ∩ V = {x0 }. Hence the intersection (which is quite small!) is simply
connected. But U and V are not open in the union! So we need a slight modification:
add a little bit of U and V , open, to the set (so U looks like a fish with the vertical
segment erased).
Now, π1 (U, x0 ) = S 1 . With some handwaving we can get (jU )∗ (π1 (U, x0 )) = Z as well.
Therefore, by Von Kampen we obtain that π1 (X, x0 ) is the free group on two genera-
tors.
So now, we have our first non-abelian fundamental group.
Exercise. Show that the space X8 is a deformation retraction of the punctured torus.
The fundamental group of the torus is Z×Z, and the space X8 is now the abelianization
of the free group F2 .
29
15 October 15, 2012

15.1 von Kampen Again
Open U and V in a space X satisfy X = U ∪V and U ∩V path connected and containing
some element x0 .
Using the Lebesgue number lemma, we decompose any loop [γ] ∈ π1 (X, x0 ) as γ1 ∗
γ2 ∗ · · · ∗ γn , where each γi is a path from xi−1 to xi lying entirely in either U and V ,
and xi ∈ U ∩ V for all i.
Now construct a path in U ∩ V called αi joining xi to xi−1 for each i. Now we see
that we have a composition of loops
γ = (γ1 ∗ ᾱ1 ) ∗ (α1 ∗ ᾱ2 ) ∗ (α2 ∗ γ3 ∗ ᾱ3 ) ∗ · · · ∗ (αn−2 ∗ γn−1 ∗ ᾱn−1 ) ∗ (αn−1 ∗ γn ) (1)
Now the loops alternate between lying in U and V .

However, we are really considering a loop contained in U as a loop in X = U ∪ V ; as
in the inclusion diagram:
U
-
jU
iU
-
U ∩V U ∪V
-
iV
jV
-
V
For example, consider a space B 2 a disk of radius 3. Let V denote the disk with radius
2 and U denote the annulus with outer radius 3 and inner radius 1.
If we take a [η] ∈ π1 (U, x0 ) ≈ Z, we get something that’s not trivial if we consider it
in the larger space B 2 = U ∩ V . In other words, (jU )∗ ([η]) ∈ π1 (B 2 , x0 ) is trivial even
though [η] ∈ π1 (U, x0 ) is not. In still other words, (ju )∗ need not be one-to-one.
15.2 Seifert-Van Kampen Special Cases

The first special case was U, V simply connected implying U ∪V being simply connected.
If U ∩ V is simply connected, then
π1 (U ∪ V, x0 ) ∼
= π1 (U, x0 ) ∗ π1 (V, x0 )
The difference here is that if the intersection is simply connected, then “what you see
what you get” in 1.
For example, if π1 (X8 , x0 ), where X8 is the figure eight, gives the free group F2 = ha, bi.
Generally, let’s look at (jU )∗ ([γ1 ]) ∗ (jV )∗ ([γ2 ]) where [γ1 ] ∈ π1 (U, x0 ) and [γ2 ] ∈
π1 (V, x0 ).
But if [η] ∈ π1 (U ∩ V, x0 ) such that [γ1 ] = (iU )∗ ([η]) and [γ2 ] = (iV )∗ ([η]−1 ), then we
have trouble. It could turn out that we have a ∈ π1 (U, x0 ) and b ∈ π1 (V, x0 ) but ab = 1
in the larger group π1 (X, x0 ).
The full description is
π1 (U ∪ V, x0 ) = ([(jU )∗ π1 (U, x0 )] ∗ [(jV )∗ π1 (V, x0 )])/N
30
where N is the normal closure of (i.e. the smallest normal subgroup which contains)
((iU )∗ [η])−1 · (iV )∗ [η] for every [η] ∈ π1 (U ∩ V, x0 ).
This is called the amalgamated product, written
π1 (U, x0 ) ∗π1 (U ∩V,x0 ) π1 (V, x0 ).
In general, if K is contained in groups G1 and G2 we write the amalgamated product

as G1 ∗K G2 .
15.3 Killing Homotopy

Suppose we have a continuous function f : S 1 → Y , we define the mapping cone Cf as
follows.
We consider the disjoint union Y ∪ (S 1 × I) (literally a cylinder floating above I), and
take
(Y ∪ (S 1 × I))/ ∼
where ∼ identifies S 1 × {0} to a single point, and (z, 1) with f (z) for each z.
The result is what appears to be a witch’s hat.
We now split this into two pieces: let U be a S 1 × [0, 1006
2011 ]; that is, a little more than
1004
the upper half of the hat. The second piece V is the union of Y and S 1 × [ 2011 , 1]; the
base of the hat plus a little above half the hat.
Claim. The fundamental group of Cf = U ∩ V is isomorphic to π1 (V )/ ∼.
Proof. By our theorem,
π1 (Cf , x0 ) = π1 (U, x0 ) ∗π1 (U ∩V,x0 ) π1 (V, x0 ).
But U ∼ B 2 so π1 (U, x0 ) is trivial.

As for V , we just push down part of the hat, so π1 (V ) = π1 (Y ).
Finally, what have we identified when we amalgamated π( U ∩ V, x0 )? We have U ∩ V
homotopic to S 1 , whose fundamental group is Z, generated by some loop η.
But η is trivial in U . So ∼ kills loops that can be deformed from η.
31
16 October 17, 2012

16.1 Something
Let Σ3 denote the permutations of the set {1, 2, 3}. We know Σ3 is generated by the
transpositions
τ1 = (2 3) τ2 = (3 1) τ3 = (1 2).
This does not imply that Σ3 is isomorphic to Z2 ∗ Z2 ∗ Z2 . In fact, τ2 τ3 τ2 = τ1 so not all
the generators are necessary.
16.2 Torus
Consider the frame of a torus T (the two loops with generate π1 (S 1 × S 1 )). It is not
hard to see that this is a figure-eight.
Denote by U a square patch open in T . Let V be all of T minus some smaller co-centric
square patch contained entirely in U . Then U ∪ V = T , and U ∩ V is isomorphic to S 1 .
Then π1 (T, x0 ) = π1 (U ∪ V, x0 ) ∼
= [π1 (U, x0 ) ∗ π1 (V, x0 )]/N for some N .
Now once again we look at the torus as the orbits of Z2 acting on R2 by addition. We
can take as representatives the square given by [− 21 , 12 )2 .
Claim. V is homotopy equivalent to the frame X8 .
Proof. Push out the hole.
So, we get π1 (V ) ∼
= ha, bi = F2 . But although it’s not trivial in the figure eight, we
find that aba−1 b−1 becomes trivial in T .
In other words, in the torus, ab = ba.
That is; we find now that π1 (U ∪ V, x0 ) = ha, b | ab = bai ∼
= Z2 .
Actually, consider a torus with a rectangle cut out. We can consider patching it with
a cone pointing outwards; so it looks like a torus with a spike on it. This is a mapping
cone!
16.3 Group Presentations

Hey cool I already know all this!
D E
D8 = r, s | r4 = s2 = 1, rs = s−1 r .
Now we can consider a free group hr̃, s̃i and the canonical homomorphism ϕ to D8 .
Denote by K the kernel of ϕ.5
Zorn’s Lemma states that any poset such that every totally ordered subset has an
upper bound, then there is a maximal element in the whole poset (that is, no element
which exceeds it). Once again, this is equivalent to the axiom of choice.
Consider the following subgroup of ha, bi:
hbab, babab, bababab, · · · i
This is an infinitely generated subgroup of a finitely generated group.
Definition. If G and K are called finitely generated we call G finitely presented.
5
Completely random sidenote: every vector has a basis if and only if the Axiom of Choice holds.
32
In our particular example, G = hr, si and K = r4 , s2 , rsr−1 s N (where the subscript

N includes the rest of the elements that makes the group normal). These are called
relators; any relation can be converted to a relator by moving everything to one side.
4 2 −1

We get presentations at this point as well, namely D8 = r, s | r , s , rsr s .
Go back to the figure eight with π1 (X8 ) = hr, si. We will mold this into the dihedral
group D8 as follows:
• Take a disk and map its boundary onto the loop [r] by z 7→ z 4 . This forces r4 = 1.
• Take another disk and map its boundary onto the loop [s] by z 7→ z 2 . This forces
s2 = 1.
• Take a third disk and map it in this super troll way that causes rsr−1 s = 1.
16.4 For Next Time

Universal covers, and unwrapping spaces (reverse of covering projections) such that each
space above a given space X corresponds to as subgroup of π1 (X), with the universal
subgroup corresponding to the trivial subgroup.
33
17 October 22, 2012

17.1 Something
ϕ
E1 - E2
p1
p2
-

B
ϕ is a morphism of covering spaces if the diagram commutes; that is, p2 ◦ ϕ = p1 .
In particular, φ must take fibers to fibers.
If ϕ is one-to-one and onto, and ϕ and ϕ−1 are both continuous, then ϕ is an “iso-
morphism”.
17.2 van Kampen again

Last time, we did stuff with a torus. If you consider a square with the sides identified
(i.e. a torus), we have aba−1 b−1 = 1, but not so with the patch.
Now, let’s consider identifying the edges of a square in opposite directions to get a
space X.
a
4 -
6
b b
?
4
a
When we identify the edges a, we get a Möbius strip. At this point, ba is a loop. The
Möbius strip deformation retracts onto the “core” circle and thus is homotopy equivalent
to S 1 . Hence, the fundamental group of the Möbius strip is π1 (M ) = Z, generated by
ba.
What happens when we make the second attachment? First, delete a patch in the
center. Now baba is a nontrivial loop which corresponds to 2 in π1 (M ). But when we
fill in the patch, suddenly baba = 1.
So let’s call U be a large band-aid, and V be X minus a patch. Then,
π1 (X) = π1 (U ) ∗π1 (U ∩V ) π1 (V ) = {1} ∗Z Z = Z2 .
17.3 n-Cells
We’ve actually seen this space X before! All we are doing is identifying antipodal points
on the boundary, which is our much earlier space S 2 /A, where antipodal points are
mapped. Now if we look downwards, we get our space (except circular).
This is also equivalent to applying a mapping cone on S 1 , but with the loop we are
applying it to being 2. The whole process is called attaching a 2-cell to Y ; we are
attaching the northern hemisphere of B 2 onto the space Y .
We can generalize this to n-cells by replacing “half of B 2 ” without “half ofB n ”.
In this way, we can build up spaces form the bottom; starting with a bunch of points,
we attach 1-cells. Now ∂B 1 = [−1, 1] and S 0 is {−1, 1}. Hence, adding 1-cells let us add
“loops” and “edges”. Then we go on to n-cells, etc.
34
These lead to things called CW complexes.

One last example: we can build S 1 by considering two points, adding two 1-cells to
create S 1 , then two 2-cells to build S 2 . Or we can be lazy, so that we start with a point,
then just throw on S 2 . In other words, give a space, there are multiple ways to produce
CW complexes.
17.4 Manifolds
A manifold is a space which locally looks like a flat Euclidean sphere; that is, every
point has a neighborhood homeomorphic to Euclidean space. For example, S 2 is a
two-dimensional manifold.
These manifolds are called closed, meaning it is compact and without boundary.
Definition. Given a closed manifold M a M , a morse function is a “nice” function

f : M → R, which is an “altitude function”.
We assume that M is smooth (a C ∞ manifold). For example, if M ⊆ R3 , then an

altitude function might just be (x, y, z) 7→ z. D E
∂g ∂g
Suppose g(x, y) is a function. The gradient ∇g = ∂x , ∂y . A critical point is a point
where the gradient is zero. We want our critical points to be nondegenerate as well. (e.g.
0 of x3 is considered degenerate.)
We view our manifold M as a bunch of functions in coordinates, such that at any given
altitude, there is at most one critical point. What to do these critical points represent?
Starting at the bottom, each critical point is equivalent to attaching an n-cell.
17.5 Poincaré Conjecture

Theorem 15 (Poincaré Conjecture). If an n-manifold Σn has the same homotopy type
of S n , then Σn is homeomorphic to S n
n = 2 is easy, but n ≥ 5 was proven in the 60’s. The n = 4 case is hard and involves
“elegant procrastination”, but the n = 3 case is the hardest.
Gauss-Bonnet Theorem
35
18 October 24, 2012

18.1 Klein Bottle
One last application of van Kampen:
a
4 4
6 6
b b
4 - 4
a
Etymology: flasche and fiasco are related, but fläche means surface.
Once again, we take punch a hole in the middle of the square. Let V be K minus the
hole, and U be a slightly large band-aid. Again, by Seifert van-Kampen, we can get
D E
π1 (K) = π1 U ∗π1 (U ∩V ) πV = 1 ∗Z Z = a, b | bab−1 a .
Note: in general, there is no easy way to prove that we’ve gotten all the relationships,
but in our case our U ∩ V are all generated by one generator, which makes it easy.
The “word problem” basically states that given a presentation of a group, it is not
algorithmically possible to determine whether a word is equal to the identity. It is also
not algorithmically possible to determine whether two groups are isomorphic, given their
presentations.

Now we have diagrams!
E
p
?
B
Given two covering projections (E1 , p1 ) and (E2 , p2 ) onto the same space B, then am
orphism between the two is a map ϕ such that p2 ◦ϕ = p1 (see last week). Geometrically,
ϕ maps p−1 −1
1 (b0 ) to p2 (b0 ).
In other words, the morphisms between covering projections are continuous functions
which preserve fibers. Isomorphisms are those which are invertible.
Suppose a “finite” group G acts freely6 on a space E.
Example. Let G = (Z, +) and E = R. Let G act on E by n · x = x + 2πn. This

corresponds to the classical covering projection of R onto S 1 , which occurs since E/G ≈
S1.
Later we will see that the nicest covering projections are those which can be repre-
sented by this group modding.
6
That is, the stabilizer of each x ∈ E is trivial. In still other words, @g ∈ G \ {1}, x ∈ E with g · x = x.
36
Again, we have our fundamental lifting correspondence:

path γ E
-
p
?
I - [b0 ] ∈ π1 (b) B
We will see that with “nice” projections, then either the lifts of a loop [γ]inπ1 (B) are
all loops or all non-loops.
18.3 More General Liftings

e0 ∈ p−1 (b0 ) ⊆ E
-
f̃ p
f ?
- b0 ∈ B
(Y, y0 )
Let f : Y → B with b0 = f (y0 ). We seek a lifting f˜ of f that takes y0 to e0 ∈ p−1 (b0 ).
We restrict our attention to connected and locally path connected7 spaces Y , E, B. (It
turns out that spaces which are connected and locally path connected are automatically
path-connected).
Theorem 16. Let f : (Y, y0 ) → (B, b0 ) be continuous and consider a covering projection
p : (E, e0 ) → (B, b0 ). Then a lifting f˜ with f˜(y0 ) = e0 if and only if
f∗ (π1 (Y, y0 )) ⊆ p∗ (π1 (E, e0 )).
If this lifting exists, it is unique.
Conveniently, p∗ (π1 (E, e0 )) ∼

= π1 (E, e0 ) since we saw earlier that p∗ is an injection. In
particular, if Y = I, then f∗ (π1 (Y, y0 )) is trivial and the inclusion is evident.
Question. If f turns out to be a covering projection, is it the case that f˜ is also a

covering projection?
Proof. For the “only if” part, suppose such a lifting f˜ exists. Then p ◦ f˜ = f . Thus,
f∗ = p∗ ◦ f˜∗ . Hence,
f∗ (π1 (Y, y0 )) = p∗ (f∗ (π1 (Y, y0 ))) ⊆ p∗ (π1 (E, e0 ))
where the last step follows from the obvious f˜∗ (π1 (Y, y0 )) ⊆ π1 (E, e0 ).
For the other direction, assume f∗ (π1 (Y, y0 )) ⊆ p∗ (π1 (E, e0 )). First, we will show that
if f˜ exists, it is unique! (This turns out to be instructive).
Suppose f˜1 and f˜2 are both liftings of f that start at e0 . We wish to show f˜1 (y1 ) =
˜
f2 (y1 )∀y1 ∈ Y . If y1 = y0 this is obvious. Hence assume y1 6= y0 and consider a path α
from y0 to y1 . In that case, f ◦ α is a path from f (y0 ) = b0 to f (y1 ). Lift f ◦ α to a path
γ starting at e0 ; this path is unique. But p˜◦(f1 ◦ α) = f ◦ α, so f˜1 ◦ α is a lift of f ◦ α
starting at f˜1 (α(0)) = f˜1 (y0 ) = e0 . Hence, the ending points are certainly the same;
7
X is locally path-connected if ∀x ∈ X and U an open neighborhood of x0 , ∃V a path connected open
set with x0 ∈ V ⊆ U . This lets us exclude things like that harmonic fan...
37
that is, γ(1) = f˜1 (α(1)) = f˜1 (y1 ) (that is, their ending points are the same). Similarly
f˜2 (y1 ) = γ(1) as well, so we’re done.
Anyways, to define f˜ we just use the same thing. That is, if α is a path from y0 to
y1 , then let γ be the lifting of f ◦ α and define f˜(y1 ) = γ(1). (In practice, it is very hard
to clean up this definition of f˜ to something reasonable.)
We need to check that this map is both well-defined (since α was chosen arbitrarily)
and that it is continuous. USE THE CONDITION NOW (oops I did not notice that).
You need the first condition to get the well-defined part, and the second part is just sort
of blah.
To be finished next time. . .
The “only if” direction intuitively should be easy: going from topology to algebra
tends to lose information, so it is natural to guess that the existence of a lifting is
strong, while the algebraic condition is weak.
38
19 October 29, 2012

19.1 More on Covering Projections
Let B be connected and locally path connected; let E, Y be path connected and Y
path-connected.
e0 ∈ P −1 (b0 ) ⊆ E
f ?
y0 ∈ Y - b0 ∈ B
Want to show f lifts to a unique map f˜ with f˜(y0 ) = e0 if and only if f∗ (π1 (Y, y0 )) ⊆
p∗ (π1 (E, e0 )).
Idea: connect y0 to y1 via α; then f ◦ α is a path in B with f (α(0)) = f (y0 ) = b0 and
f (α(1)) = f( y1 ) = b1 . Lift this to a path γ starting at e0 and define f˜(y1 ) = γ(1).
We got up to here last time, and now we need to show this is well-defined and contin-
uous. This is probably a good time to use the condition.
Example. To show what goes wrong without said hypothesis, consider the following:
E=R θ
f = id
? ?
1- B = S 1 e2πiθ
Y =S
Here, this hypothesis doesn’t hold since f∗ (π1 (Y, y0 )) ≈ π1 (Y, y0 ) ≈ Z and π1 (E, e0 ) =
{1} ⇒ p∗ (π1 (E, e0 )) = {1}. Indeed, this causes the function to not be well-defined: the
two paths α1 , α2 : [0, 1] → S 1 from 1 to −1 given by t 7→ eiπt and t 7→ eiπ3t yield different
values when lifted.
OK, let’s try to prove well-definedness. Consider two paths α and β in Y with the
same endpoints, and consider paths f ◦ α and f ◦ β in B from b0 to b1 . Lift f ◦ α and
f ◦ β̄ to γ and δ, where γ(1) = δ(0). Also, γ ∗ δ lifts (f ◦ α) ∗ (f ◦ β̄), since
p(γ ◦ δ) = (p ◦ γ) ∗ (p ◦ δ) = (f ◦ α) ∗ (f ◦ β̄).
Now
[f ◦ (α ∗ β)] = f∗ ([α ∗ β̄]) ∈ f∗ (π1 (Y, y0 )) ⊆ p∗ (π1 (E, e0 ))
so that f ◦ (α ∗ β̄) 'p p ◦ (η) where η is some loop at e0 .
Hence, η is a lifting of (f ◦ α) ∗ (f ◦ β̄) = f ◦ (α ∗ β̄) which starts at e0 . Now by
uniqueness of lifting, then, this is a loop. So δ̄ = ¯β = β ends at the same place as γ.
Hence, the function is well-defined.
Continuity can be shown from the fact that f is continuous and we have evenly covered
neighborhoods (i.e. homeomorphisms on a local scale.)
39
19.2 Equivalence of Projections

h
π1 (E, e0 ) e0 ∈ p−1 (b0 ) ⊆ E - e0 ∈ (p0 )−1 (b0 ) ⊆ E 0
0 π1 (E 0 , e00 )
p 0
p
p∗ )
0 ∗
- (p
-
B
π1 (B, b0 )
Theorem 17. There’s a unique equivalence h of coverings (E, p) and (E 0 , p0 ) such that
h(e0 ) = e00 if and only if
p∗ (π1 (E, e0 )) = p0∗ (π1 (E 0 , e00 )).
Recall that an equivalence if a homeomorphism h such that p0 ◦ h = p.
Proof. Suppose h exists. Then p0 ◦ h = p ⇒ (p0 ◦ h)∗ = p∗ ; i.e. p0∗ ◦ h∗ = p∗ . In that case,
h is a lifting of p; by our previous theorem we have p∗ (π1 (E, e0 )) ⊆ p0∗ (π1 (E 0 , e0 )).
E0 E
-
-
1
h−
p0
h
p ? p0 - ?
E - B E0 B
But now by doing the same thing with h−1 , we find p0∗ (π1 (E 0 , e00 )) ⊆ p∗ (π1 (E, e0 ));
this establishes the reverse inclusion and hence the groups in question are equal.
For the reverse direction, we can perform the same trick with double inclusion to find
that p lifts to h : E → E 0 with h(e0 ) = e00 , and p0 lifts to g : E 0 → E with g(e00 ) = e0 .
We want to show that g and h are inverses, but don’t know anything about them yet.
E E
-
-
h
E
◦
p p
id
g
p ? p ?
E - B E - B
So we look at the map g ◦ h. Since (p ◦ g ◦ h) = (p ◦ g) ◦ h = p0 ◦ h = p, so g ◦ h lifts p.

But so does idE . By uniqueness of lifting, we have g ◦ h = idE . Doing the same thing in
the other direction gives the conclusion.
Note that in this theorem, we have specified base points!

Recall:
Fact. If H1 , H2 ≤ G, then H2 is called conjugate to H1 if ∃g ∈ G such that H2 =

gH1 g −1 . Since conjugation is an automorphism (the inner automorphism), then H1 ∼
= H2
must take place.
40
Now, consider a single covering projection with two points e0 , e1 .
e0 , e1 ∈ p−1 (b0 ) ⊆E
p
?
b0 ∈ B
Let Hi = p∗ (π1 (E, ei )), for i = 0, 1. If γ is a path from e0 to e1 then α = p ◦ γ is the
loop from b0 to itself.
Now, [α]−1 ∗ p∗ (π1 (E, e0 )) ∗ [α] = p∗ (π1 (E, e1 )); indeed,
[α]−1 ∗ p∗ ([δ]) ∗ [α] = [ᾱ ∗ (p ◦ δ) ◦ α]

= [(p ◦ γ̄) ∗ (p ◦ δ) ◦ (p ◦ γ)]
= p∗ ([γ̄ ◦ δ ∗ γ])
In fact, the converse is true: if some subgroup H is conjugate to p∗ (π1 (E, e0 )) in

π1 (B, b0 ), then ∃e1 ∈ E and a path α from e0 to e1 such that H = p∗ (π1 (E, e1 )).
If (E, p, e0 ) and (E 0 , p0 , e00 ) are coverings
h - e0 ∈ E 0
e0 ∈ E 0
p0
p
b0 ∈ B
these coverings are equivalent if and only if p∗ (π1 (E, e0 )) is conjugate to p0∗ (π1 (E 0 , e00 )).
If they are conjugate, then, ∃e01 ∈ E 0 with p0 (e01 ) = b0 for which p∗ (π1 (E, e0 )) =
p0∗ (π1 (E 0 , e01 )); actually equal.
Next time: universal covers!
41
Evan Chen M275 Fall 2012: Notes 20 Halloween, 2012
20 Halloween, 2012
Assume all spaces are connected and locally path connected (which implies path-connected
and locally path-connected).
Last time, we saw that a lifting of a map f : Y → b exits if and only if f∗ (π1 (Y, y0 )) ⊆
p∗ (π1 (Y, y0 )). Hence, given two coverings of the same space B, there exists an equivalence
h with h(e0 ) = e00 iff p∗ (π1 (E, e0 )) = p0∗ (π1 (E 0 , e00 )).
E0
-
p0
h
?
E - B
p
Then, there exists a equivalence h, period, if and only if p∗ (π1 (E, e0 )) is conjugate to
p0∗ (π1 (E10 , e00 )) in π1 (B, b0 ). Here, e0 ∈ E and e00 ∈ E are both in the fiber of b0 .
20.1 Universal Covers

p
Definition. Given a projection E −→ B, E is called a universal cover8 if E is simply
connected (1-connected).
In particular, if E1 and E2 are both universal covers, then they are equivalent by the
above, since p∗ (π1 (E1 , e1 )) = p∗ (0) = 0 and similarly for E2 . In particular, E1 and E2
are homeomorphic. Hence, up to homeomorphism there is a unique universal covering.
E2
-
p2
p1 - ?
E1 B
Hence, R is the universal covering of S 1 since R is contractible (via θ 7→ e2πiθ ). Similarly,
R2 = R × R covers S 1 × S 1 = T by (θ, φ) = (e2πiθ , e2πiφ ), so it is also the universal
covering. This implies, for example, that S 2 cannot cover the torus because S 2 is also
simply connected, but we already have R2 as the universal covering and R2 6≈ S 2 .
Why do we call these universal? First, recall the facts:
q r
Fact. Let r ◦ q = p be continuous maps by E1 −→ E2 −→ B.
E2
-
r
q
?
E1 - B
p
If p, r are coverings then so is q. If p, q are coverings then so is r.
Claim. If (Ẽ, p̃) is a universal covering of B and (E, p) is any covering of B,
8
Strictly, (E, p).
42
Proof. Notice that p̃∗ π1 (Ẽ, ẽ0 ) = p̃∗ 0 = 0 ⊆ p∗ (π1 (E, e0 )); this inclusion is strict if E is
not a second universal covering (since p∗ is trivial). Hence, we must have a lifting φ; by
the fact above we see that we φ is actually a covering projection!
E
-
φ
p
?
Ẽ - B
p̃
20.2 Category Theory

Given a category C, an object Ur ∈ ob(C) is called unusually repelling if ∀A ∈ ob(C), ∃ϕA ∈
Hom(U, A) : Ur → ϕA A. By uniqueness and such, it is clear that a universally repelling
object is unique. Also called an initial object.
Here, our category is the covers (E, p) of B and the morphisms are functions f12 such
that p2 ◦ f12 = p1 .
f12
E1 - E2
p1
p2
-
B
Now are categories are starting to look more complicated! Here’s another example:
Example. Fix groups {Gi }. The objects of the category are a group and families of
homomorphisms from some external group H, not fixed. That is, we can view our objects
in the form
ϕi
(H, {H −→ Gi }).
Our morphisms will be homomorphisms Φ12 such that ϕ2i ◦ Φ!2 = ϕ1i for all i.
Φ12
H1 - H2
ϕ1
2i
ϕ
i
Gi
The object
pr
Y nY o
i
Gj , Gj −→ Gi ,
which is the product of all the groups where the homomorphisms are just protections,
are universally attractive. Indeed, just take
Φ12 : h → (ϕi (h))i ;
i.e. just project each thing individually.
43
20.3 Do Universal Covers Always Exist

Answer: no. But usually.
IF E is a universal cover, then ∀b ∈ B∃U a neighborhood such that the inclusion
j
U ,→ B for which j∗ : π1 (U, b) → π1 (B, b) is trivial. This property is called semi-locally
simply connected.
It turns out this is sufficient!
Example (Hawaiian Earring). In R2 take the subspace topology on the set

( 2 )
1 1

2
[
(x, y) | x − +y = 2 .
n≥1
n n
All these points are tangent at (0, 0). You can show this is continuous and such. Here,
Figure 2: Hawaiian Earing
this is NOT semi-locally simply connected, so there is no universal covering of this space.
You can also look at the higher-dimensional version, where spheres are replaced by
circles, then we get some other interesting stuff.
20.4 Subgroups vs Coverings

Suppose a base B is semi-locally simply connected.
0 = π1 (Ẽ) Ẽ
q
-
p̃ E
p

?
B
44
0 = p̃∗ (π1 (Ẽ)) < p∗ (π1 (E)) ≤ π1 (B)

Given (E, p) let C(E, p, B) denote the group of covering transformations of (E, p); that
is self-equivalences of the covering (E, p).
h
E - E
p
-

B
This is a group under composition; the identity is idE , et cetera.
Next time: how does C(E, p, B) act on E?
45
Evan Chen M275 Fall 2012: Notes 21 November 5, 2012
21 November 5, 2012
21.1 Examples of Group Actions
Definition. Let G be a group. It acts freely from the left on X if ∀x ∈ X : Gx = {1}.
That is, the stabilizer of any x ∈ X is the trivial subgroup.
Example. G = (R, +) acts on X = R by addition.
Definition. Let X/G denote the set of the orbits of elements in X, endowed with the
quotient topology.
Our goal is, given a group G and a space X, to get a covering projection p : X → X/G
by x 7→ [x]; that is, we project each element of the space to its image in the group X/G
with the quotient topology.
Example. Z acts on X = R. We get X/G ≈ S 1 . This is really just our standard

projection θ 7→ e2πiθ .
2πi
Example. G = Z/nZ acts on X = S 1 by m · z = ζ m · z, where ζ = e m . This is a
well-defined action but is also a covering projection. Now X/G is just an “interval” of
S 1 . This corresponds to the covering map S 1 → S 1 : z 7→ z n .
Why do we need the group to act freely? Consider the following example with X = B 2 ,
where G does not act freely be the stabilizer of 0 is all of G.
Example. Let X = B 2 = {z | |z| ≤ 1}. Let G = Z/nZ act on B 2 by m · w = ζ m · w.

Now B 2 /G cannot be identified with some covering projection, because the cardinality
of the fiber must be constant; this is not possible here.
This isn’t enough to guarantee a covering projection: for example, if X = R and

G = R acts on X by addition, then X/G is trivial. It is not possible to cover using R
the singleton space.
21.2 Stronger Conditions

Definition. G acts on a space X properly discontinuously if ∀x ∈ X, there exists a
neighborhood U such that U ∩ gU = ∅∀g 6= 1.
Intuitively, this means that G acts in such a way that if 1 6= g ∈ G, then elements of
X are moved “sufficiently far away”. The example X = G = R does not work, but. . .
Fact. If G acts on X properly discontinuously, then X → X/G is a covering projection.
Note. Some book use freely and properly discontinuously. However, if an action is
properly discontinuous, then it is automatically free (for obvious reasons).
We can start viewing X as topological groups; a group endowed with a topology. For
example,
hζn i = Zn ≤ R ∼
= S1
Other important groups: SO(n), GLn (R). Anyways, on a side note
O(n) ⊆ GLn (R) = O(n) × R? .
46
21.3 Covering Transformations

Let G = C(E, p, B) be the group of covering transformations of (E, p, B); that is, it is
the set of all homeomorphisms
{h : E → E | p ◦ h = p}.
This is a group with identity is idE .
Claim. G acts on E from the left by h · e := h(e).
Proof. Check that this is an action. To see that it is a free action, consider an h(e) = e,
so that h is the stabilizer of e ∈ E.
E
-
h
p
?
E - B
p
h is a lifting of p, where h(e) = e. But by uniqueness of lifting, we immediately obtain
h = idE .
It turns out that this is properly discontinuous as well. Consider some e ∈ E, let
b0 = p(e), take some neighborhood Ub0 of b0 . Now p−1 (Ub0 ) is a collection of disjoint
neighborhoods around p−1 (b0 ), per even covering. Let Ue be the one around V .
Then, p(Ue ) = Ub0 . If h is some homeomorphism which preserves projection, then
h(Ue ) must be Ub0 . Unless h is the identity, then h(Ue ) ∈ p−1 (Ub0 ), and this is a bunch
of disjoint sets.
21.4 Coverings and Such

Again, G denotes C(E, p, B).
E
p
π
-
?
E/G B
Let π : E → E/G be a group quotient map, and p : E → B be the standard covering
projection. Can E/G = B?
Let ∼π be the relation that equates elements of E in the same orbit of π. Let ∼p be
the relation that equates elements in the same fiber of p.
Fact. e ∼π e0 ⇒ e ∼p e0 . The converse is not necessarily true.
Proof. Compute
e ∼π e0 ⇒ e0 = ge = g(e)
⇒ pp(e0 ) = p(g(e))
= p(e)
0
⇒ e ∼p e
⇒ e ∼p e0
47
22 November 7, 2012
22.1 Groups
We are given a covering projection p : E → B, with b0 ∈ B having fiber F = p−1 (b0 ).
Let G = π1 (B, b0 ) from now on.
1. G acts from the right on F . The action F × π1 (B, b0 ) → F sends (e, [γ]) to γ̃(1)
where γ̃ is a path starting at e.9
2. This G-action is transitive. (This is obvious.) Therefore, eG = F ; i.e. the orbit of

E is all of F . Furthermore, F is equivalent to the cosets of Ge , the stabilizer of E.
3. p−1 (b0 ) is essentially equivalent to p∗ (π1 (E, e))\π1 (B, b0 ).
Consider C(E, p, B) as defined in the previous section. Then, define Aut(F ) to be the
set of G-automorphisms of F ; that is
Aut(F ) = {Ψ : F → F bijective | Ψ(eg) = Ψ(e)g ∀g ∈ G}.
These are both groups under composition. Now we claim these groups are isomorphic!
(This implies that C(E, p, B) is completely determined by what p does to a fiber.)
Claim. C(E, p, B) ∼
= Aut(F ).
Proof. Consider φ ∈ C(E, p, B). Now consider some e ∈ F = p−1 (b0 ).
p(ϕ(e)) = (p ◦ ϕ)(e) = p(e) = b0 ⇒ ϕ(e) ∈ p−1 (b0 ).
Hence, ϕ|F is a function from F to F .

Therefore we are motivated to construct the isomorphism
Φ : C(E, p, B) → Aut(F ) : ϕ 7→ ϕ|F .
Lots of things to check:
• Check that ϕ|F ∈ Aut(F ). Since ϕ is a homeomorphism, it is globally bijective,

so immediately ϕ|F , and it is surjective since ϕ−1 |F is also injective (as ϕ−1 is also
bijective). Hence ϕ |F is bijective.
To show that ϕ|F preserves the right action by G = π1 (B, b0 ), take some e ∈ F .
Then we want to show that ϕ|F (e[γ]) = (ϕ|F (e))[γ]. But
ϕ|F (e[γ]) = ϕ(e[γ]) = ϕ(γ̃(1)).
But then
p ◦ (ϕ̃ ◦ γ̃) = p ◦ γ̃ = γ.
So ϕ ◦ γ̃ is a lifting of γ starting at ϕ(γ̃(0)) = ϕ(e). So, ϕ(γ̃(1)) = ϕ(e)[γ] by
definition. This establishes the desired equality. Hence, we finally get ϕ|F ∈
Aut(F ).
• Check that ϕ|F is a homomorphism. This is trivial since the operations are
composition in both groups.
9
Since p(γ̃(1)) = γ(1) = b0 , we see that this is well-defined in the sense that γ̃(1) ∈ F .
48
• Check that Φ is injective. We can just check that the kernel is trivial. Suppose
ϕ|F is the identity; i.e. in the kernel. Then ∀e ∈ F : ϕ|F (e) = e ⇒ ϕ(e) = e. Then
ϕ is a lifting of p that takes e to e, so it must be the identity.
• Check that Φ is surjective. We prove an algebraic lemma: Suppose F is a

transitive right G-set. We wish to show for all e1 , e2 ∈ F , we have the equality
Ge1 = Ge2 if and only if ∃φ ∈ Aut(F ) such that φ(e1 ) = e2 . (This is purely
algebraic).
First suppose such a φ exists. Take some g ∈ Ge1 . Then e2 = φ(e1 ) = φ(e1 g) =
φ(e1 )g = e2 g. Therefore, g ∈ Ge2 . Hence, Ge1 ⊆ Ge2 . Doing the same thing with
φ−1 yields the reverse inclusion.
For the other direction, suppose that Ge1 = Ge2 . We wish to construct φ ∈ Aut(F ).
Select some e ∈ F ; by transitivity, e = e1 g̃ for some g̃ in G by transitivity. We
select the map
φ : e1 g̃ = e 7→ e2 g̃.
This satisfies the condition φ(e1 ) = φ(e1 · 1) = e2 · 1 = e2 . To see that φ is well-
defined suppose that e = e1 ĝ = e1 g̃. Then ĝg̃ −1 ∈ Ge1 = Ge2 . Thusly e2 ĝg̃ −1 = e2
which implies e2 ĝ = e2 g̃ and hence φ is well defined.
Let us now check injectivity; suppose φ(e) = φ(e0 ) and e = e1 g̃, e0 = e1 g̃ 0 . Evidently
e2 g̃ = e2 g̃ 0 . Then e2 g̃˜(g 0 )−1 = e2 , so g̃˜(g 0 )−1 ∈ Ge2 = Ge1 which forces e = e0 .
glosses over remaining details
We can now use the lemma to show Φ is onto. Let φ ∈ Aut(F ), take e ∈ F , and
let e0 = φ(e). Now Ge = Ge0 . Recalling that G = π1 (B, b0 ) we see that
Ge = p∗ π1 (E, e) Ge0 = p ∗ π1 (E, e0 )
So p∗ π1 (E, e) = p∗ π1 (E, e0 ).
E
-
p
ϕ
p ?
E - B
By our general lifting criterion, we see that ∃ϕ : (E, e) → (E, e0 ) an equivalence.
It turns out later that Aut(F ) = NGe (G)/Ge , where NGe (G) is the normalizer of Ge in
G. In the case of covering projections, we know that Ge = p∗ π1 (E, ee ) and G = π1 (B, b0 )
and we can compute stuff. In particular, if π1 (E) = 1 then Aut(F ) = π1 (B, b0 ).
49
23 November 14, 2012

23.1 Groups of Covering Transformations
We discussed a group of covering transformations C(E, p, B) ∼ = Aut(F ), where F is the
p
fiber of a b0 ∈ B with a covering projection E −→ B. Here Aut(F ) consists of the
permutations of F that preserve the right action of π1 (B, b0 ).
If F is a transitive right G-set (i.e. E is connected) and e0 ∈ F , then F is equivalent
(as a G-set) to the set of right cosets Ge0 \G, where Gx denotes the stabilizer of x.
Now, the new claim is:
Claim. If G = π1 (B, b0 ) we have
Aut(F ) ∼
= NG (Ge0 )/Ge0 .
Here Ge0 = p∗ π1 (E, e0 ) is the stabilizer of the element e0 and NG (Ge0 ) = {g ∈ G |

gGe0 g −1 = Ge0 } is the normalizer of Ge0 in G.
Note that in general
NG (Ge0 ) 6= {g ∈ G | gGe0 g −1 ⊆ Ge0 }
although it is certainly true for finite groups. Also, recall the facts (i) H ≤ NG (H) ≤ G
for any H ≤ G, (ii) H E G ⇔ NG (H) = G, and (iii) H E NG (H)
Proof. We will do this completely in the context of group theory; let F = Ge \G where
Ge0 is the stabilizer of some e0 ∈ F and suppose G acts transitively on F . Consider some
φ ∈ Aut(F ) = Aut(Ge \G) mapping Hg 7→ φ(Hg) = φ(H)g. Notice that Φ is completely
determined by any H. So, for any automorphism φ, we let gφ denote a representative of
the coset φ(H1G ) = φ(H).
Hence, we are motivated to construct the automorphism Γ : Aut(F ) → NG (Ge0 )/Ge0
by
φ 7→ Hgφ .
Okay, details. . .
• First we claim gφ ∈ NG (Ge0 ). We have
Hgφ = φ(H) = φ(Hh) = φ(H)h ∀h ∈ H.
So, H = Hgφ hgφ−1 which implies gφ hgφ−1 ∈ H ∀h ∈ H. Thus gφ Hgφ−1 ⊆ H. . . which

is not enough. To establish the reverse inclusion observe that because H =
φ−1 (φ(H)) = φ−1 (H)gφ = (Hgφ−1 )gφ ⇒ gφ−1 gφ ∈ H. Do some blah.
• Next, we claim that Γ(φ1 φ2 ) = φ1 φ2 (H), so that Γ is a homomorphism. Compute
φ1 φ2 (H) = φ1 (Hgφ2 ) = φ1 (H)gφ2 = (Hgφ1 )gφ2 = H(gφ1 gφ2 ) = Hgφ1 Hgφ2 = Γ(φ1 )Γ(φ2 ).
• To get that Γ is injective, notice that Γ(φ) = φ(H) = H. Then φ(Hg) = φ(H)g =
Hg for every g, so φ is the identity, and hence ker Γ = {1} and we win.
• To demonstrate that Γ is onto, consider H ĝ ∈ NG (Ge0 )/Ge0 and define φ by

φ(Hg) = φ(H)g = H ĝg. Then trivially Γ(φ) = H ĝ. Do blah.
50
Returning to the context of topology we have

C(E, p, B) ∼
= Nπ1 (B,b0 ) (p∗ π1 (E, e0 ))/p∗ π1 (E, e0 ).
In particular, p∗ π1 (E, e0 ) E π1 (B, b0 ) if and only if
C(E, p, B) ∼
= π1 (B, b0 )/p∗ π1 (B, b0 ).
In this case, we call the covering regular.
23.2 A Diagram
We have that C(E, p, B) acts properly discontinuously on E.
E
p
π
? -
E/C(E, p, B) ............
- B
Last time we showed that ∼1 and ∼2 defined by e ∼1 e0 ⇔ ∃φ ∈ C(E, p, B) : φ(e) = e0

and e ∼2 e0 ⇔ p(e) = p(e0 0) satisfy e ∼1 e0 ⇒ e ∼2 e0 . Hence we get an induced map
E/C(E, p, B) → B.
When are ∼1 and ∼2 ? This occurs iff e ∼2 e0 ⇒ e ∼1 e0 ; that is p(e) = p(e0 ) ⇒ ∃φ :
φ(e) = e0 . But the former condition is equivalent to e, e0 being in the same fiber. In
English (?), that means that for every e, e0 in the same fiber one can find a covering
transformation taking one to the other.
Such a φ is a lifting of p.
E
-
φ
E p

p - ?
E B
Hence, we can find a φ if and only if p∗ π1 (E, e) = p∗ π1 (E, e0 ); yet p∗ π1 (E, e0 ) = [γ] ∗
p∗ π1 (E, e)[γ̄] for every [γ] ∈ π1 (B, b).
Yet this is normality.
In conclusion, p : E → B is effectively an orbit projection if and only if it is a regular
covering.
23.3 Universal Coverings

Suppose E is simply connected; that is E is the universal cover of B. Then
C(E, p, B) ∼
= π1 (B, b0 )
since p∗ π1 (E, e) is trivial and hence normal in everything. Thus, universal coverings are
regular.
-
?
E/H .........
- B = E/G
51
Furthermore, suppose that H ≤ π1 (B, b0 ) = C(E, p, B). Now π1 (E/H) = H!

We will also get an induced map from E/H to B that turns out to be a covering
projection, although not necessarily regular.
23.4 Example
Let E = R2 and G = Z×Z, where G acts on E in the usual manner giving E/G = S 1 ×S 1 .
Now π1 (S 1 × S 1 ) = π1 (S 1 ) × ı1 (S 1 ) = Z × Z.
Consider Z × 2Z ≤ π1 (S 1 × S 1 ).
Then E/(Z × 2Z) we get another torus which is “not as tightly wrapped”. Now our
induced map is a covering projection from a torus to a torus.
52

24.1 Manifolds
Manifolds are spaces which locally look like Euclidean space. For example, S 1 is a
manifold since sufficiently small neighborhoods are homeomorphic to R1 .
A n-manifold with boundary is a space such that every point has either a neighborhood
homeomorphic to Rn or has a neighborhood homeomorphic to
Rn+ := {(x1 , x2 , · · · , xn ) | xn ≥ 0}.
It turns out that the boundary of a manifold with boundary is an n − 1 manifold with
boundary.
24.2 Groups acting dis-continuously

Here, we consider left actions. Recall
Definition. A group G acts properly discontinuously on a space X if for every x ∈ X

there exists an open neighborhood U of x such that U ∩ gU = ∅ for every g 6= 1G in G.
If G acts properly discontinuously on E, then the projection E onto its orbit space is
a covering projection.10 The proof was discussed in an earlier section which I am too
lazy to look up; the basic idea is that one takes a small neighborhood around each of
the elements in the fiber, which is possible because the action is properly discontinuous.
Claim. The group of covering transformations of this projection is precisely G; that is,
C(E, p, E/G) = G.
Proof. Define the map λg : E → E by e 7→ ge for every g ∈ G, and then construct

the map τ : G → C(E, p, E/G) by g 7→ λg . We claim this is an isomorphism. Clearly
λg ∈ C(E, p, E/G) by construction.
24.3 An Example of a Semidirect product

Let G = Z oφ Z, where φ : Z → Aut(Z) is given by 1 7→ (n 7→ −n).
Then, our group operation is simply

(h1 , k1 ) · (h2 , k2 ) = h1 + (−1)k1 h2 , k1 + k2 .
We observe that (m, n) = (m, 0)(0, n) = (1, 0)m (0, 1)n . Now define a = (1, 0) and
b = (0, 1); these generate the group. Now, check that bab−1 = a−1 .
Hence, we may write D E
G = a, b | bab−1 = a−1 .
24.4 The Action

Now let G act on E = R2 by
(
(x + n, y + m) m even
(n, m)(x, y) =
(1 − x + n, y + m) m odd
53
y
3
b2 x
2
bx
1
a−1 x x ax a2 x
0
−1 0 1 2 3
x
−1
Figure 3: Diagram of group action
It is easy to see that

a(x, y) = (x + 1, y)
b(x, y) = (1 − x, y + 1)
Now this action is properly discontinuous. Hence, we get a regular covering p : E →
E/G. Then G = Z o Z = C(E, p, E/G) = π1 (E/G)/p∗ π1 (E). But π1 (E) is trivial and
we obtain π1 (E/G) = G.
Anyways, everything is equivalent to something in the green square, so that:
Definition. A fundamental domain is a set containing exactly one representative of
each equivalence class.
Now a fundamental domain of our group is our unit square with the left and right
edges identified nicely, and the top/bottom twisted. This is a Klein Bottle.

6 6
-
Therefore, the fundamental group of a Klein bottle is Z o Z. We’ve seen this earlier
with the van Kampen Theorem, when the condition βαβ −1 α = 1 was added by the
patching.
10
This is also a regular covering projection; it’s equivalent to p∗ π1 (E) E E/G.
54

25.1 Subgroups and Snowmen
Check that groups are nonempty. That is all.
Also, snowmen are topologically equivalent to B 3 , but it does not have an optimal
volume-per-surface-area ratio, and hence melts.
25.2 Unoriented Cobordism and Manifolds

Let M1 and M2 be closed n-manifolds. We write M1 ∼ M2 if the disjoint union M1 tM2 =
∂W for some space W ; in English, the two manifolds can be joined in such a way that
they are the boundaries of an n + 1 manifold.
This equivalence relation is called bordism.
If we make the convention that ∅ is an n-manifold for every n, then we can define Nk
to be the equivalence classes of k-manifolds.
We now have
N0 = {∅, {1}}.
because a 0-manifold with k points is cobordant to a 0-manifold with k + 2 points, but
∅ and {1} are not cobordant.
Figure 4: The three-element set is cobordant to the one-element set.
Now the only closed manifold of dimension one is S 1 ; hence N1 = {∅}.

Now we can define an addition by [M1 ] + [M2 ] = [M1 t M2 ] which will give us a group
action with identity [∅]. The inverses are themselves: check that [M ] + [M ] = [∅].
Fun fact: since the Cartesian product of an m-manifold and n-manifold is an m + n-
manifold, with some work we can get our structure to become a graded ring.
25.3 Bottles Again

Last time, we had a group
D E
G = Z o Z = a, b | bab−1 = a−1
defined by
(m, n)(k, `) = (m + (−1)n k, n + `).
Our group action gave a map R2 → R2 /G from R2 to its orbit space, which gave a
regular covering projection from R2 to the Klein Bottle K.
Again, as we saw last time, since this is a regular covering we get π1 (K) = G = Z o Z.
Now, suppose H ≤ G. Then H acts on R2 as well, and we can get mod out again:
R2
-
?
R2 /H ........
- R2 /G = K
55
Obviously, two things equivalent under the H action are equivalent under the G action,
but not vice versa. Hence, the equivalence classes of H are “finer” than those of G. In
still other words, there is less collapsing.
Now we get a covering projection of R2 /H onto R2 /G. Note, however, that if H is
nontrivial then this cannot be regular.
25.4 Example
Let
H12 = Z o 2Z = {(n, 2m) | n, m ∈ Z} ≤ G.
What does our action look like?
y
3
(0, 2)x
2
(−1, 0)x x (1, 0)x (1, 0)2 x

0
−1 0 1 2 3
x
−1
Figure 5: H12 acts on R2 .
This time, there’s no “twist”: so the result is that R2 /H12 ≈ T , the torus. Now we
can rewrite our diagram earlier:
Torus
-
?
R2 - Klein Bottle
Actually, H12 ∼ = Z × 2Z if we just check that the group multiplication is “normal”.
This is consistent with the fact that the fundamental group of the torus is Z × Z.
Now, consider the covering we have obtained, q : T → K. Then
C(T, q, K) = Nπ1 (K q∗ π1 (T )/q∗ π1 (T )
56
But q∗ π1 (T ) is precisely Z o 2Z with index 2 in Z × Z. Since index 2 subgroups are

always normal, and with some easy computations we get that the group of covering
transformations in Z/2Z.
25.5 The other way around

Let H21 = 2Z o Z ≤ G now. What is the new picture?
y
3
(0, 2)x
2
(0, 1)x
1
x (2, 0)x
0
−1 0 1 2 3
x
−1
Figure 6: H21 this time
Note that the [0, 2] × [0, 1] doesn’t work here, since (2, 0) ∼ (−1, 1) which doesn’t get
mapped to anything on the line y = 1 within that rectangle.
And. . . we get a Klein bottle. Hence the Klein bottle covers itself. In fact, if we use
H33 = 3Z o 3Z we get the Klein bottle once again. Unfortunately, H33 is not normal in
Z o Z.
25.6 GAP
Just something to Google.
57

Once again we begin with G = Z o Z as before acting on X = R2 . If H ≤ G, then H
acts on R2 . Again, we get a covering from R2 /H to R2 /G.
Question. When is such a covering regular?
Example. If H = 3Z o 3Z, do we have H E G?
Solution. This is equivalent to ghg −1 ∈ H∀g ∈ G. It suffices to check that this is true
when g is a generator of G. Compute
(1, 0)(3s, 3t)(1, 0)−1 = (3s + 1 + (−1)3t , 3t) = (3s + 1 + (−1)t , 3t).
Unfortunately, this is false for t odd. Hence it is not the case that H E G.
Returning to our example, suppose H E N ≤ G. Now
[G : H] = [G : N ][N : H].
We have in fact seen that [G : N ] 6= 1. Hence, we have two possibilities:
• [G : N ] = [N : H] = 3, in which case G/N and N/H are both Z3 .
• [G : N ] = 9 and [N : H] = 1 in which case N/H is trivial.
What we care about is the group of covering transformations, which is N/H is the stuff
p̂
above. We also know that the covering R2 /H −→ R2 /G is not regular because it is not
the case that H is normal in G. (Recall that E → B is regular iff p∗ π1 (E) E π1 (B), but
π1 (X/G) = G.)
OK, so let us try to compute NG (H). If (m, n) ∈ N then g(3s, 3t)g −1 . Computing,

(m, n)(3s, 3t)(m, n)−1 = (−1)n · 3s + m(1 − (−1)2n+3t ), 3t .
This lies in 3Z o 3Z if and only if m(1 − (−1)t ) is divisible by 3. Since this must hold
for all t, this is true when m ∈ 3Z.
We then conclude that NG (H) is 3Z o Z, by our index argument.
26.1 Pictures
Let H = 3Z × 3Z.
The fundamental domain is [−1, 2) × [0, 3). This is obvious. Both yield Klein bottles.
26.2 A Characterization of Regular Coverings

p
Suppose E −→ B. We are not assuming that E is simply connected. This covering p
is regular if and only if for each individually [γ] ∈ π1 (B), either all lifts of γ to E are
open11 , or they are all closed.
For example, in our example above, we can take a “loop” α from (0, 0) to (0, 1) in
R /G with fundamental domain [0, 1)2 .
2
Then, a lifting of α3 can be viewed as a path from (−1, 0) to (−1, 3) and also one from
(0, 0) = (1, 3) to (0, 3). The latter is closed but the former is open.
11
i.e. γ̃(0) 6= γ̃(1).
58
We can also show generally that
Z o `Z E Z o Z
by showing that the covering of G/(Z o `Z) onto the Klein bottle is regular.
How de prove this claim? One direction is straightforward. Suppose p∗ π1 (E) E π1 (B)
so that C(E, p, B) = π1 (B)/p∗ π1 (E).
ϕ
E - E
p
-

B
Take [ϕ] ∈ π1 (B). Lift it to γ̃ starting at e0 . Then every ϕ◦γ̃ is a lift of γ if ϕ ∈ C(E, p, B).
Now every ϕ ◦ γ̃ is also a lift of γ if ϕ ∈ C(E, p, B). This turns out to be all the possible
liftings; some technical details omitted here.
Conversely, consider the case where ∀[γ] all lifts are open or all are closed. We wish
to show
p∗ π1 (E, e0 ) E π1 (B, b0 ). ⇔ [α] ∗ p∗ π1 (E, e0 ) ∗ [α]−1 = p∗ π1 (E, e0 ).
But this is equivalent to p∗ π1 (E, e1 ) = p∗ π1 (E, e0 ).
59

Definition. The order of a covering is the number of elements above the fiber of any
given element. This is well-defined.
If G is a group then the covering E → E/G = B is the order of G.

Given a fundamental domain D, we can act on it by the elements of G, which will
construct a “tessellation” of our space X.
27.1 Fractional Linear Transformations

Consider a function F (z) = az+b
cz+d . This is either called a fractional linear transformation
or linear fractional transformation. We assume ad − bc 6= 0.
Then, we can define a group G functions, and this is a group under transformation.
Remarkably, the “product” corresponds to exactly the matrix multiplication; that is,
the canonical bijection between G and M2 (C) respects multiplication.
However, G = 6 GL2 (C) since a function can have multiple representations (indeed,
scale the coefficients). This ends up being:
G∼
= GL2 (C)/{I2 , −I2 }.
where I2 is the 2 × 2 identity.

Now it is not hard to show that GL2 (C) = det−1 (C \ {0}) is a topological group. Now,
GL2 (Z) is a discrete subgroup of GL2 (C).
The modular group is the subgroup of G with the addition stipulation that ad − bc =
1 and a, b, c, d ∈ Z. Now look at its fundamental domain is given at http://en.
wikipedia.org/wiki/Fundamental_domain#Fundamental_domain_for_the_modular_
group.
60
Evan Chen M275 Fall 2012: Notes 28 December 3, 2012
28 December 3, 2012
Recall the definition of a semidirect product.
R2
-
?
K = R2 /G R2 /H
Same setup as before, where H ≤ G.
28.1 An Aside
Actually, as a curiosity
D E D E
Z o Z = a, b | bab−1 = a−1 ∼
= α, β | α2 = β 2 .
The bijection is α = ab and β = b.

It thus turns out that K = RP 2 #RP 2 , where we are taking the connected sum. Now
we can also use the Seifert van-Kampen theorem to get
π1 (RP 2 #RP 2 ) = π1 (U ) ∗π1 (U ∩V ) π1 (V ) = Z ∗Z Z.
Here U = V = RP 2 . This is where we get the relation α2 β −2 = 1.

Sidenote : isomorphism problem in combinatorial group theory. Word problem also.
28.2 Galois and Intermediate Covers

Let B be a “nice” space with a universal cover Ẽ.
Ẽ
r
-
p̃ Ẽ/H
ν

?
Ẽ/G = B
We have π1 (B) ∼
= G = C(Ẽ, p̃, B). Now G acts on Ẽ to yield B, et cetera.
r
Let H ≤ G. Then H acts on Ẽ, and Ẽ −→ Ẽ/H is a covering transformation as well.
So, we find a get a map from the subgroups of G to the quotients of Ẽ by
H 7→ (Ẽ, r, Ẽ/H).
At the extremes {1} 7→ E and G 7→ E/G = B. On the other hand, any covering
Y → B can be recovered12 as Ẽ/H for some H.
12
No pun intended.
61
Now we get
r
Ẽ - Y
q

?
B
We claim that C(Ẽ, r, Y ) ≤ C(Ẽ, p̃, B). Suppose that
ϕ ∈ C(Ẽ, r, Y ) ⇒ ϕ : Ẽ → Ẽ has r ◦ ϕ = r.
We wish to show that p̃ ◦ ϕ = p̃. But this follows by p = q ◦ r.
ϕ ∈ C(Ẽ, r, Y )
Ẽ - Ẽ
r r
-

Y
q
?
B
Hence, with a few more details we obtain
Fact. For “nice” B, the spaces which cover B correspond precisely to the subgroups of
G = C(Ẽ, p̃, B).
How does this correspond to Galois theory? Let k be a subfield of K and suppose
G = Autk (K) = {ϕ : K → K | ϕ ∈ Aut(K), ϕ |k = idk }.
For every subgroup {1} ≤ H ≤ G define K H to be the fixed field of H, defined by
K H = {k ∈ K | σ(x) = x ∀σ ∈ H}.
Clearly, k ⊆ K G and GL = {σ ∈ G : σ |L = idL } ≤ G
28.3 Pullback Diagrams

pr2
f ! E ..................
- E
..
..
..
.
pr1 .... p
..
..
.?
f - ?
B0 B
0
Given a covering p : E → B and a map f : B → B, we define
f ! E = {(x, e) ∈ B 0 × E | f (x) = p(e)}.
62
Then we get a canonical projection
pr1 (x, e) = x and pr2 (x, e) = e.
It turns out that pr1 is a covering. For an x0 ∈ B 0 , the fibre of x in f ! E is
pr−1 (x0 ) = {(x̃0 , e) | pr1 (x̃, e) = x; p(e) = f (x)} = {(x0 , e) | p(e) = f (x0 )} = {x0 }×p−1 (f (x)).
so the fibers of pr correspond canonically to the fibers of p.

Now we can show that if f1 and f2 are homotopic, then the corresponding pr1 are
“equivalent”. And more to come. . .
63
29 December 5, 2012
If n < m then any map from S n → S m if n < m since it is not surjective, and then one
can take the missed point and expand it to something else. However, S n → S m is often
not nulhomotopic if n ≥ m > 1.
29.1 Fiber Bundles

Definition. A map from p : E → B is called a fibre bundle with structure group G if
for every b ∈ B there exits a neighborhood Ub of b such that ∃h : p−1 (Ub ) → U × F such
that (i) h is a homeomorphism (ii) h(e) = (p(e), h2 (E))
Example. Locally the Möbius Strip looks like U × [−1, 1]
q
E0 - E
p0
?
B
q is called a bundle map if p ◦ q = p0 ; that is q “preserves” fibers.
If b0 ∈ B then
−1

q p0 (b0 ) ⊆ p−1 (b0 ).
29.1.1 Vector Bundles

Tangent bundles. Some very small subset of the tangent bundles have trivial tangent
bundles.
Also disk bundles.
29.1.2 Covering Projections

A covering projection is a fiber bundle where the bundle projection is actually a local
homeomorphism. In particular, the fiber is discrete.
Consider b0 ∈ U, V where U, V are some subsets of B. Suppose we have p : E → B.
hV hU
(V ∩ U ) × F p−1 (V ∩ U ) - (V ∩ U ) × F
pr (V (U
)
1 ) r 1
-
? p
V ∩U
hU ◦h−1
V
We have hU and hV are homeomorphisms. Then (b, f ) 7→ (b, L(b, f )) for some
arbitrary function L.
64
29.2 Ending Notes

If E → B is a fiber bundle13 with fiber F then there exists a magical space BG and EG
and a map f : B → BG such that the pullback f ! EG is equivalent to E.
E = f ! EG - EG
? f - ?
B BG
These are called “classifying spaces”.
13
e.g. F = Rn and G = GLn (R) or F = {a1 , · · · , an } and G = Sn the symmetric group
65
30 December 10, 2012

Last class!
30.1 Fiber Bundles

Fix F and a group G which acts on F from the left.
Example. Let F = Rn and G = O(n) the group of orthogonal n × n matrices. Then if
A ∈ G, ~x ∈ F we could make a group action A~x ∈ F .
Example. Let F = {1, 2, · · · , n} and G = Sn the symmetric group. Then the action is
just σ · i = σ(i).
Definition. Suppose p : E → B. This is a fibre bundle with fibre F and structure
group G if ∀b ∈ B there exists a neighborhood Ub ⊆ B such that p−1 (Ub ) is essentially
a product; that is, ∃ϕU : Ub × F → p−1 (Ub ) a homeomorphism such that p ◦ ϕU is
projection. In other words, the following diagram should commute:
ϕU
- p−1 (Ub )
Ub × F
≈
p
pr 1
?
Ub
Furthermore, if b ∈ Ub ∩ Vb , then p−1 (Ub ∩ Vb ) we get a diagram
ϕ-U −1 ϕV
(Ub ∩ Vb ) × F p (Ub ∩ Vb ) (Ub ∩ Vb ) × F
≈ ≈
pr p pr 1
1
- ?

Ub ∩ Vb
Then, we require that there is a “transition” function ΓV U : Ub ∩ Vb → G for which
x 7→ ΓV U x for which
(ϕ−1
V ◦ ϕU )(x, f ) = (x, ΓV U (x)f )
for all f ∈ F .
Finally, if b ∈ Ub ∩ Vb ∩ Wb we require ΓW V ◦ ΓV U = ΓW U .
Finally, we can define morphisms as such:
u
E0 - E
p0 p
? ?
B0 - B
f
A morphism between fiber bundles with the same fiber F and structure group G is a
pair (u, f ) such that p ◦ u = f ◦ p0 .
Thus, we have a category of fiber bundles with fiber F and group G.
66
30.2 Categories
Now, let H be the category of topological spaces and homotopy classes of maps. The
objects consists of category of topological spaces X, and arrows are the homotopy classes
of maps [X, Y ].
Also, Set is the category whose objects are sets and arrows are arbitrary set functions.
Now we consider a map kG : X 7→ kG (X), the set of all fiber bundles with structure
group G and fiber F over X. This is a set, of course. Then kG is a contravariant14
functor. We want to send [f ] ∈ [X, Y ] to some map f ∗ : kG (Y ) → kG (X).
f ! Ey - Ey
? ?
X - Y
f
So, we just use the pullback. The claim is that f ! Ey → X is a fiber bundle with the
same fiber and such.
Now, we say that kG is representable in the sense that there is a unique (up to ho-
motopy) space BG and an element of kG (BG ) (i.e. a fiber bundle EG → BG ) such that
∀X : kG (X) = [X, BG ].
That is, if E → X is a fiber bundle, there is a unique [f ] ∈ [X, BG ] such that this
commutes:
E ≈ f ! EG - EG
? ?
X - BG
f
Hence, for every group G we can get a “master” EG → BG which describes all fiber
bundles with structure group G.
As an example, we have for F = {1, 2} and G = Z2
S1 ⊆ S2 ⊆ S3 ⊆ · · ·
? ? · · · RP 1 ⊆ RP 2 ⊆ RP 3 ⊆ · · ·
?
So we can define an S ∞ = n≥1 S n and RP ∞ = n≥1 RP n .

S S
14
Meaning the direction is flipped.
67

1 August 22, 2012: M275 Notes

Uploaded by

Copyright:

Available Formats

1 August 22, 2012: M275 Notes

Uploaded by

Document Information

Original Description:

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

1 August 22, 2012: M275 Notes

Uploaded by

Copyright:

Available Formats

M275 Notes

December 10, 2012

Lecture notes from M275 at SJSU, a graduate-level course at SJSU in

1 August 22, 2012

In short, this is very hard.

Exercise. Show that R 6≈ [0, 1] with the usual topology.

Definition. Let S 1 = {z ∈ C : |z| = 1} be the unit circle.

1.2 Category Theory

Example. A is the category of abelian groups, category V of vector spaces, category R

Definition. Give X, Y ∈ ob(C), then mor(X, Y ) is the set of morphisms “from X to

Definition. A functor F between categories sends an object X ∈ ob(T op) to a group

Given ϕ ∈ M or(X, Y ), we want F to take ϕ to F (ϕ) ∈ M or(F (X), F (Y )). So a

Fact. If f is an isomorph, then f∗ must be an isomorphism.

This is useful if, say, π1 (X) ∼

2 August 27, 2012

2.2 Setting Out

(i) Define homology groups H1 (X), H2 (X), · · · .

(ii) Define homotopy groups π1 (x), π2 (X), · · · .

Definition. S n denotes an n-dimensional sphere; that is, S n ⊆ Rn+1 can be defined as

Example. S 1 ⊂ R2 is a unit circle. S 2 ⊂ R3 is the surface of a sphere.

One can also think of S n as the equator of S n+1 .

Definition. π1 (X) is called the fundamental group of X.

Theorem 1 (Brouwer Fixed Point Theorem). If f : D2 → D2 is continuous, then

2.3 Homotopy Groups

Definition. A path in Y is a continuous map ϕ : I → Y , where I has the standard

Definition. A loop in X at x0 is path γ : I → X such that x0 = γ(0) = γ(1). It is

Definition. Given two continuous functions f, f 0 : X → Y , we say that f is homotopic

Definition. If A is a subspace of X and f = f 0 on A, (i.e. f (a) = f 0 (a)∀a ∈ A), then we

Definition. Paths f, f 0 : I → Y in Y satisfy y0 = f (0) = f 0 (0) and y1 = f (1) = f 0 (1).

In other words, f, f 0 : I → Y are path homotopic if ∃F : I 2 → I continuous with

F (0, t) = f (0) = f 0 (0)

These three definitions give equivalence relations.

Theorem 2. All such groups (for a fixed space X) are isomorphic.

Exercise. Let f : X → Y be continuous, and define F : X × I → Y by F (x, t) = f (x).

Proof. Let V be an open subset in Y . F −1 (Y ) = f −1 (X) × I, however; this is the

3 August 29, 2012

3.2 Straight-Line Homotopy

γ1 (s) = eiπs and γ2 (s) = e−iπs

Show that γ1 and γ2 are path-homotopic.

F (s, t) = (1 − t)γ1 (s) + tγ2 (s)

3.3 Product of Paths

Figure 1: Not Associative, but homotopic

Fix a space X. Given a path γ1 from x1 to x2 , and a path γ2 from x2 to x3 , we can

Definition. Given γ1 , γ2 : I → X, define γ1 ∗ γ2 : I → X by

Definition. A homeomorphism is a continuous functions f : X → Y such that ∃g :

This is an equivalence relation. Clearly, this is weaker than homeomorphism.

Example. R2 \ {~0} ' S1

Proof. Identifying R2 with the corresponding complex number in C, define f : C → S1 by

4.2 The Fundamental Group

Question. Is this operation well-defined?

Answer. Yes. This isn’t hard to verify, just follow through.

Question. Is this a group operation?

Question. Can we calculate some simple homotopy groups?

Question. Does this group depend on x0 ?

Anyways, here is an actual example.

Example. π1 (R, 0) is the trivial group.

Proof. Well, everything is nulhomotopic, so. . . heh.

Theorem 3. If X is path-connected, then π1 (X, x1 ) is isomorphic to π( X, x2 ) for any