Fundamental group and covering spaces: the facts.

Intro.
I assume all spaces nice.
Details of all of the facts asserted here can be found in any
of the standard references. They are collected here for your
convenience.
Suppose Z is a space, and * a point of Z. We define !1(Z,*)
as homotopy classes of maps f:[0,1] -> Z, such that f(0) =
f(1) = *.
(Homotopy here means that one views as equivalent two
functions which lie on a 1-parameter family of functions
ft:[0,1] ->Z, all of which satisfy the boundary condition ft(0) =
ft(1) = *.)
These boundary conditions are absolutely critical for getting
a nontrivial theory. !1(Z,*) is a group using concatenation of
paths; the constant path is the identity and going backwards
is the inverse. !1(Z,*) is referred to as the fundamental group
of Z. (If Z is path connected, the choice of * is irrelevant, in

the sense that for a different choice of points, one gets an

isomorphic group.)
Example: If Z is the circle S1 = {u in C | |u| = 1}, we can
define a map !1(S1,1) -> Z (the integers) by sending a map f
to (log(f)(1) log(f)(0))/2!i. This is referred to as the winding
number.
Youll want to check that this makes sense, i.e. that one can
define log(f) continuously on the interval, and that the above
definition is independent of all choices, once one decides
that log(f)(0) = 0 and that the winding number doesnt
change during a homotopy.
If f and g have the same winding number then
ft(x) = exp(tlog(f)(x) + (1-t)log(g)(x))
is a homotopy between g and f. (What goes wrong if they
have different winding number?).
The maps x -> exp(2!inx) have winding numbers n, so this
homomoprhism is an isomorphism.

By the way,we say that a space is simply connected if

its fundamental group is trivial.
Categorical Nonsense
If f:(X,x) -> (Y,y) is a map, (this notation means that f(x) = y)
then there is an induced homomorphism, defined in the most
obvious way, f*: !1(X,x) -> !1(Y,y).
If g: (Y,y) -> (Z,z) is another map, then (gf)* = g* f*. This is
both obvious and useful. The first is clear and the second
will be shown countless times in class.
Pointedly homotopic maps of pointed spaces, i.e. f,g :(X,x) ->
(Y,y) homotopic through such maps, induce the same
homorphisms on fundamental group, i.e. f* = g*. This implies
a homotopy invariance property of !1.

Covering Spaces
Fundamental group is important for at least two reasons.
The first is the connection to covering spaces. We assume

here that all spaces are connected (unless they arise in the
middle of a proof, or something).
Definition. A map p: A -> B is a covering space, if:around
each point b in B, there is a neighborhood N of b, so that
p-1(N) is a disjoint union of sets Ai each of which is mapped
homeomorphically onto N by p.
The map exp: R -> S1 considered before is a good example.
Let us pick a point b in B and fix a point a in A which maps to
B. In other words, let us refine our perspective and consider
covering spaces as maps p:(A,a) -> (B,b). This way, to
every cover one can unambiguously assign the subgroup p*
(!1(A,a)) inside of !1(B,b).
If we did not pin down a basepoint in A, a covering space
would only give a well defined conjugacy class of subgroup
of !1(B,b).
Theorem (Classification of Covers): To every subgroup of
!1(B,b) there is a covering space of B so that the induced
subgroup is the given one.

Moreover if (A,a, p) and (A,a, p) are two different

covers corresponding to the same subgroup, then there is a
(unique) homeomorphism h:(A,a) -> (A,a) so that p = ph.
Note that a cover of a cover is a cover, so that smaller
subgroups correspond to higher covers.
The trivial subgroup corresponds to the universal
cover of B. Theorems stated below imply that all of the
maps p* on fundamental groups are 1-1 (for covering maps!!!
not for maps in general!!). So the universal cover is the
only simply connected cover that a space has.
Note too, that if p:(A,a) -> (B,b) is a universal cover,
then for any a in p-1(a), there is a unique homeomorphism
ha:A -> A so that
ha(a) = a
p ha = p.
The set of homeomorphisms which satisfy the second
condition form a group (sometimes called the group of
covering transformations or the group of deck
transformations), and we are essentially saying that one can
determine the group element by seeing where a goes. It
turns out that this group is isomorphic to the fundamental
group of B.

Remark: There is a nice analogy to field theory. Let F be a

field; covering spaces are like algerbaic extension fields.
The universal cover is like the algebraic closure. The
fundamental group is like the Galois group of the algebraic
closure. The relation between covers and subgroups of the
fundamental group is just like ordinary Galois theory. There
are situations where this analogy is actually implemented
geometrically, but lets not digress.....
Examples: The 2-sphere S2 is simply connected. The
projective plane RP2 has fundamental group Z/2Z since it is
the quotient of S2 by making the identifications x = -x. The
projection map is a covering map, and the group of covering
transformations is just Z/2Z = {id, x -> -x}. The nontrivial
element in the fundamental group of RP2 can be thought of
as the quotient of a great chord on S2 that connects the north
pole to the south pole.
The torus R2/Z2 has fundamental group Z2. Again, elements
of the fundamental group can be thought of as the
projections of chords connecting the origin to (m,n) in the
lattice Z2.

An example that takes more work to think through is the

universal cover of the figure 8. It is the unique tree where
every vertex has 4 edges emanating from it. It is easy to see
that this space is simply connected (indeed it is contractible,
i.e. homotopy equivalent to a point). If one colors the edges
blue and green, so that at each vertex two edges are blue
and two green, then one can use this to define a covering
map to 8, where the top loop is colored blue and the bottom
green. What are the covering translations in this example?
In general, you should have little trouble using the the
method of the calculation of !1(S1) to show that if one has a
simply connected space A and a group of homeomorphisms
! of A so that A -> A/! is a covering map (this condition is
called proper discontinuity of the action. It can be phrased
as the condition that for each a in A there is a neighborhood
N of a, so that for all g not the identity in !, gN and N are
disjoint), then the quotient A/! has fundamental group !.
What requires more thought is why this is the general
case, i.e. how does one construct universal covers and how
does one get the group of deck transformations.
More on covering spaces.

Suppose that p:(A,a) -> (B,b) is a covering map. Then for

any f: [0,1] -> (B,b) such that f(0) = b, there is a unique lifting
f^: [0,1] -> (A,a) such that f^(0) = a and pf^ = f.
Now suppose that f: (X,x) -> (B,b) is an arbitrary map with
f(x) = b, X path connected. We can try to use the previous
remark to try to lift f to a map f^: (X,x) -> (A,a) simply by
defining f^(x) to be the result of lifting the composite of f with
any path connecting x to x. Of course there will be many
different paths connecting these two points, and this
indeterminacy causes an obstruction. However, the analysis
of this gives rise to the following critically important theorem:
Lifting condition for Covering Spaces: Suppose p:(A,a) ->
(B,b) is a covering map, and f:(X,x) -> (B,b) is given. Then
there is an f^:(X,x) ->(A,a) so that pf^ = f (and, of course,
f^(x) = a, as is included in this notation) if and only if
f* !1(Z,z) is a subgroup of p* !1(A,a).
The proof is actually straightforward and it has many
corollaries, which I leave to you.

Definition/Corollary: Say that a cover p:(A,a) -> (B,b) is

normal if there are deck transformation sending any element
of p-1(b) to any other. The cover p is normal iff p* !1(A,a) is a
normal subgroup of !1(B,b).
Corollary: The uniqueness statement in the classification
theorem holds.
Corollary: If Z is simply connected, then any map f:(Z,z) ->
(B,b) can be lifted to any cover.
Corollary: For any cover, p:(A,a) -> (B,b), the induced map,
p* is 1-1.
Now for existence of covers.... we will only explain the
universal cover, since other covers can be obtained as
quotients of it by subgroups of !1(B,b).
Part of B being a nice space includes the fact that two
nearby (in the C0 topology) functions are homotopic. I leave
it to you to decide if the spaces you know and love are nice
this way. (I bet they all are.) We will assume that nearby
paths conncting b to b are homotopic through paths

connecting these points. (Using an embedding in Euclidean

space and the tubular neighborhood theorem, you should be
able to check this for manifolds. For cell complexes, you
might want to do an induction...)
Given (B,b) we will build a new space, called A. The
points of A are equivalence classes of maps a:[0,1] -> B, so
that a(0) = b. Two paths a,a define the same point in A if
a(1) = a(1)
a and a are homotopic relative to their endpoints.
Note that without the second condition, the space obtained
would just be B. With both of them, one can check that the
map a -> a(1) is a covering map, and that a map from a
circle to B lifts to A iff it is homotopic to a constant. By the
lifting criterion, this implies that A is simply connected. So it
is a universal cover.
Remark: This construction is not as bad as it looks. After a
while you can really do it. As an exercise, do it to 8 to get
the universal 4-valent tree as described above.
Of course, after a while, with experience you can see
universal covers of lots of things. Practice on your friends
and on household items.

Computation of fundamental group.

I had said that fundamental group was important for at
least 2 reasons, the first being its connection to covering
space theory. The second is that it is quite computable (in
one sense anyway).
Example: If X is contractible then the fundamental group is
trivial.
Example: If one sees the universal cover and group of deck
transformations, then one also knows the fundamental
group.
But actually, the key practical tool is Van Kampens
theorem. It describes the fundamental group of a union in
terms of the fundamental groups of the pieces. I will
describe a pretty useful, but not the most general version of
it.
Van Kampens theorem. Let Z denote the union of A and B,
and X denote their intersection. We assume A, B, and X are
all connected (and nonempty), and that the inclusions of X in
A and B are nice. Then !1(Z,x) is generated by !1(A,x) and

!1(B,x). The only relations among the elements of !1(A,x)

and !1(B,x) are the ones forced by the fact that the elements
of !1(C,x) can be thought of as elements of both of these
groups.
Examples.
1. If A and B are simply connected, and their intesection
is connected, then their union is simply connected.
(Give an example where this fails when the intersection
is disconnected.)
2. If X is simply connected, then !1(Z,x) is the free product
!1(A,x)* !1(B,x). The elements of the free products are
just finite strings of elements of !1(A,x) and !1(B,x), and
one multiplies strings by concatenting them, ignoring
the identity, and combining contiguous elements of the
same group.
As a special case, consider the free group on two
generators F2 = Z*Z. A typical element of that group looks
like xyx-3y2x. Its inverse is x-1y-2x3y-1x-1. Do you see it?

3. These things can be tricky if !1(X,x) is nontrivial. The

group described is called a free product with
amalgamation and is denoted by !1(A,x)*

!1(X,x)

!1(B,x).

There is a nice interpretation of what the elements of this

look like when the induced maps of !1(X,x) into the other
two pieces are injective, but without this it can get
complicated. As a simple example suppose that X is a
circle and that !1(A,x) = Z/2Z and !1(B,x) = Z/3Z, so that
the induced homomorphisms are the obvious surjections.
(By the way, can you build such spaces?) Then, Van
Kampens theorem tells us that Z is simply connected. Do
you see why?
The fundamental groups of both A and B are generated by
that of the circle, i.e. there is one generator, say g. From
A we learn that g2 = e and from B we learn that g3 = e. So
in the amalgamated free product (i.e. !1(Z,x)) g = e, so the
who group vanishes.
In principle any space that can be broken up into pieces
can have its fundamental group described by generators
and relations via Van Kampens theorem.

For more, see the usual references.