THE FUNDAMENTAL GROUP AND COVERING SPACES

JESPER M. MLLER
Abstract. These notes, from a rst course in algebraic topology, introduce the fundamental group and the
fundamental groupoid of a topological space and use them to classify covering spaces.
Contents
1. Homotopy theory of paths and loops 1
1.10. Change of base point and unbased homotopies 4
2. Covering spaces 5
3. The fundamental group of the circle, spheres, and lense spaces 8
3.4. Applications of
1
(S
1
) 9
4. The van Kampen theorem 10
4.1. Fundamental groups of knot and link complements 14
5. Categories 15
6. Categories of right G-sets 17
6.5. Transitive right actions 17
7. The classication theorem 18
7.14. Cayley tables, Cayley graphs, and Cayley complexes 23
7.22. Normal covering maps 27
7.25. Sections in covering maps 27
8. Universal covering spaces of topological groups 27
References 31
1. Homotopy theory of paths and loops
Denition 1.1. A path in a topological space X from x
0
X to x
1
X is a map u: I X of the unit
interval into X with u(0) = x
0
and u(1) = x
1
.
Two paths, u
0
and u
1
, from x
0
to x
1
are path homotopic, and we write simply u
0
u
1
, if u
0
u
1
rel I,
ie if u
0
and u
1
are are homotopic relative to the end-points 0, 1 of the unit interval I.
The constant path at x
0
is the path x
0
(s) = x
0
for all s I
The inverse path to u is the path from x
1
to x
0
given by u(s) = u(1 s)
If v is a path from v(0) = u(1) then the product path path u v given by
(u v)(s) =
_
u(2s) 0 s
1
2
v(2s 1)
1
2
s 1
where we rst run along u with double speed and then along v with double speed is a path from u(0) to v(1).
In greater detail, u
0
u
1
if there exists a homotopy h: I I X such that h(s, 0) = u
0
(s), h(s, 1) = u
1
(s)
and h(0, t) = x
0
, h(1, t) = x
1
for all s, t I. All paths in a homotopy class have the same start point and
the same end point. Note the following rules for products of paths
x
0
u u u x
1
Date: December 10, 2011.
I would like to thank Morten Poulsen for supplying the graphics and Yaokun Wu for spotting some errors in an earlier
version. The author is partially supported by the DNRF through the Centre for Symmetry and Deformation in Copenhagen.
/home/moller/underv/algtop/notes/chp1/[Link].
1
2 J.M. MLLER

u
0
(s)

u
1
(s)

x
0

x
1 h(s, t)
Figure 1. A path homototopy between two paths
u u x
0
, u u x
1
(u v) w u (v w)
u
0
u
1
, v
0
v
1
=u
0
v
0
u
1
v
1
These drawings are meant to suggest proofs for the rst three statements:

u(2s)

x
1

u(s)

x
0

x
1

u(2s)

u(1 2s)

x
0

x
0

x
0
u v w
u v w

x
0

x
3
x
1
x
2
In the rst case, we rst run along u with double speed and then stand still at the end point x
1
for half
the time. The homotopy consists in slowing down on the path u and spending less time just standing still at
x
1
.
In the second case, we rst run along u all the way to x
1
with double speed and then back again along u
also with double speed. The homotopy consists in running out along u, standing still for an increasing length
of time (namely for
1
2
(1 t) s
1
2
(1 +t)), and running back along u.
In the third case we rst run along u with speed 4, then along v with speed 4, then along w with speed 2,
and we must show that can deformed into the case where we run along u with speed 2, along v with speed
4, along w with speed 4. This can be achieved by slowing down on u, keeping the same speed along v, and
speeding up on w.
The fourth of the above rules is proved by this picture,

u
0
(2s)

v
0
(2s 1)

u
1
(2s)

v
1
(2s 1)

x
0

x
1

x
2
We write (X)(x
0
, x
1
) for the set of al homotopy classes of paths in X from x
0
to x
1
and we write
[u] (X)(x
0
, x
1
) for the homotopy class containing the path u. The fourth rule implies that the product
operation on paths induces a product operation on homotopy classes of paths
(1.2) (X)(x
0
, x
1
) (X)(x
1
, x
2
)

(X)(x
0
, x
2
): ([u], [v]) [u] [v] = [u v]
and the other three rules translate to similar rules
[x
0
] [u] = [u] = [u] [x
1
] (neutral elements)
[u] [u] = [x
0
], [u] [u] = [x
1
] (inverse elements)
([u] [v]) [w] = [u] ([v] [w]) (associativity)
THE FUNDAMENTAL GROUP AND COVERING SPACES 3
for this product operation.
We next look at the functorial properties of this construction. Suppose that f : X Y is a map of spaces.
If u is a path in X from x
0
to x
1
, then the image fu is a path in Y from f(x
0
) to f(x
1
). Since homotopic
paths have homotopic images there is an induced map
(X)(x
0
, x
1
)
(f)
(Y )(f(x
0
), f(x
1
)): [u] [fu]
on the set of homotopy classes of paths. Observe that this map
(f) does not change if we change f by a homotopy relative to x
0
, x
1
,
(f) respects the path product operation in the sense that (f)([u] [v]) = (f)([u]) (f)([v]) when
u(1) = v(0),
(id
X
) = id
(X)(x1,x2)
, (g f) = (g) (f) for maps g : Y Z
We now summarize our ndings.
Proposition 1.3. For any space X, (X) is a groupoid, and for any map f : X Y between spaces the
induced map (f): (X) (Y ) is a groupoid homomorphism. In fact, is a functor from the category of
topological spaces to the category of groupoids.
Denition 1.4. (X) is called the fundamental groupid of X. The fundamental group based at x
0
X is
the group
1
(X, x
0
) = (X)(x
0
, x
0
) of homotopy classes of loops in X based at x
0
.
The path product (1.2) specializes to a product operation

1
(X, x
0
)
1
(X, x
0
)
1
(X, x
0
)
and to transitive free group actions
(1.5)
1
(X, x
0
) (X)(x
0
, x
1
) (X)(x
0
, x
1
) (X)(x
0
, x
1
)
1
(X, x
1
)
so that
1
(X, x
0
) is indeed a group and (X)(x
0
, x
1
) is an ane group from the left and from the right.
For fundamental groups, in particular, any based map f : (X, x
0
) (Y, y
0
) induces a group homomorphism
(f) = f

:
1
(X, x
0
)
1
(Y, y
0
), given by
1
(f) = f

([u]) = [fu], that only depends on the based homotopy

class of the based map f.
Proposition 1.6. The fundamental group is a functor
1
: hoTop

Grp from the homotopy category of

based topological spaces into the category of groups.
This means that
1
(id
(X,x0)
) = id
1(X,x0)
and
1
(g f) =
1
(g)
1
(f). It follows immediately that if
f : X Y is a homotopy equivalence of based spaces then the induced map
1
(f) = f

:
1
(X, x
0
)
1
(Y, y
0
)
is an isomorphism of groups. (See Section 5 for more information about categories and functors.)
Corollary 1.7. Let X be a space, A a subspace, and i

:
1
(A, a
0
)
1
(X, a
0
) the group homomorphism
induced by the inclusion map i : A X.
(1) If A is a retract of X then i

has a left inverse (so it is a monomorphism).

(2) If A is a deformation retract of X then i

has an inverse (so it is an isomorphism).

Proof. (1) Let r : X A be a map such that ri = 1
A
. Then r

is the identity isomorphism of

1
(A, a
0
).
(2) Let r : X A be a map such that ri = 1
A
and ir 1
X
rel A. Then r

is the identity isomorphism of

1
(A, a
0
) and i

is the identity isomorphism of

1
(X, a
0
) so i

and r

are each others inverses.

Corollary 1.8. Let X and Y be spaces. There is an isomorphism
(p
X
)

(p
Y
)

:
1
(X Y, x
0
y
0
)
1
(X, x
0
)
1
(Y, y
0
)
induced by the projection maps p
X
: X Y X and p
Y
: X Y Y .
Proof. The loops in X Y have the form u v where u and v are loops in X and Y , respectively (General
Topology, 2.63). The above homomorphism has the form [u v] [u] [v]. The inverse homomorphism is
[u] [v] [u v]. Note that this is well-dened.
We can now compute our rst fundamental group.
Example 1.9.
1
(R
n
, 0) is the trivial group with just one element because R
n
contains the subspace 0
consisting of one point as a deformation retract. Any space that deformation retract onto on of its points
has trivial fundamental group. Is it true that any contractible space has trivial fundamental group?
4 J.M. MLLER
Our tools to compute
1
in more interesting cases are covering space theory and van Kampens theorem.
1.10. Change of base point and unbased homotopies. What happens if we change the base point?
In case, the new base point lies in another path-component of X, there is no relation at all between the
fundamental groups. But if the two base points lie in the same path-component, the fundamental groups are
isomorphic.
Lemma 1.11. If u is a path from x
0
to x
1
then conjugation with [u]

1
(X, x
1
)
1
(X, x
0
): [v] [u] [v] [u]
is a group isomorphism.
Proof. This is immediate from the rules for products of paths and a special case of (1.5).
We already noted that if two maps are homotopic relative to the base point then they induce the same
group homomorphism between the fundamental groups. We shall now investigate how the fundamental group
behaves with respect to free maps and free homotopies, ie maps and homotopies that do not preserve the
base point.
Lemma 1.12. Suppose that f
0
f
1
: X Y are homotopic maps and h: X I Y a homotopy. For
any point x X, let h(x) (Y )(f
0
(x), f
1
(x)) be the path homotopy class of t h(x, t). For any u
(X)(x
0
, x
1
) there is a commutative diagram
f
0
(x
0
)
f0(u)

h(x0)

f
1
(x
0
)
f1(u)

f
0
(x
1
)
h(x1)

f
1
(x
1
)
in (Y ).
Proof. Let u be any path from x
0
to x
1
in X. If we push the left and upper edge of the homotopy I I
Y : (s, t) h(u(s), t) into the lower and right edge

f
0
(u)

f
1
(u)

h(x
0
)

h(x
1
)

we obtain a path homotopy h(x

0
) f
1
(u) f
0
(u) h(x
1
).
Corollary 1.13. In the situation of Lemma 1.12, the diagram

1
(Y, f
1
(x
0
))

= [h(x0)][h(x0)]

1
(X, x
0
)
(f1)

l
l
l
l
l
l
l
l
l
(f0)

R
R
R
R
R
R
R
R
R

1
(Y, f
0
(x
0
))
commutes.
Proof. For any loop u based at x
0
, f
0
(u)h(x
0
) = h(x
0
)f
1
(u) or f
0
(u) = h(x
0
)f
1
(u)h(x
0
).
Corollary 1.14. (1) If f : X Y is a homotopy equivalence (possibly unbased) then the induced homo-
morphism f

:
1
(X, x
0
)
1
(Y, f(x
0
)) is a group isomorphism.
(2) If f : X Y is nullhomotopic (possibly unbased) then f

:
1
(X, x
0
)
1
(Y, f(x
0
)) is the trivial
homomorphism.
THE FUNDAMENTAL GROUP AND COVERING SPACES 5
Proof. (1) Let g be a homotopy inverse to f so that gf 1
X
and fg 1
Y
. By Lemma 1.12 there is a
commutative diagram

1
(Y, f(x
0
))

1
(X, x
0
)
W
W
W
W
W
W
W
W
W
W
W
W
W
W
W
W
W
W
W
W
W
W
W
W
W
W
W
W
W
W
W
W
W
W
W
W
W
W
W
W
W
W
W
W
W
W
W
W
f


1
(Y, f(x
0
))
f
f
f
f
f
f
f
f
f
f
f
f
f
f
f
f
f
f
f
f
f
f
f
f
f
f
f
f
f
f
f
f
f
f
f
f
f
f
f
f
f
f
f
f
f
f
f
f
f
f
g

1
(X, gf(x
0
))

f


1
(Y, fgf(x
0
))

1
(X, x
0
)
which shows that g

is both injective and surjective, ie g

is bijective. Then also f

is bijective.
(2) If f homotopic to a constant map c then f

followed by an isomorphism equals c

which is trivial. Thuse

also f

is trivial.
We can now answer a question from Example 1.9 and say that any contractible space has trivial funda-
mental group.
Denition 1.15. A space is simply connected if there is a unique path homotopy class between any two of
its points.
The space X is simply connected if (X)(x
1
, x
2
) = for all x
1
, x
2
X, or, equivalently, X is path
connected and
1
(X, x) = at all points or at one point of X.
2. Covering spaces
A covering map over X is a map that locally looks like the projection map X F X for some discrete
space F.
Denition 2.1. A covering map is a continuous surjective map p: Y X with the property that for any
point x X there is a neighborhood U (an evenly covered neighborhood), a discrete set F, and a homeo-
morphism U F p
1
(U) such that the diagram
U F
pr
1

E
E
E
E
E
E
E
E
E
:

p
1
(U)
p]p
1
(U)
.x
x
x
x
x
x
x
x
x
U
commutes.
Some covering spaces, but not all (7.22), arise from left group actions. Consider a left action GY Y
of a group G on a space Y . Let p
G
: Y GY be the quotient map of Y onto the orbit space GY . The
quotient map p
G
is open because open subsets U Y have open saturations GU =

gG
gU = p
1
G
p
G
(U)
(General Topology 2.82). The open sets in GY correspond bijectively to saturated open sets in Y .
We now single out the left actions G Y Y for which the quotient map p
G
: Y GY of Y onto its
orbit space is a covering map.
Denition 2.2. [5, (*) p. 72] A covering space action is a group action GY Y where any point y Y
has a neighborhood U such that the translated neighborhoods gU, g G, are disjoint. (In other words, the
action map GU GU is a homeomorphism.)
Example 2.3. The actions
Z R R: (n, t) n +t
Z/2 S
n
S
n
: (1, x) x
Z/mS
2n+1
S
2n+1
: (, x) x, where C is an mth root of unity,
m
= 1,
1, i, j, k S
3
S
3
, quaternion multiplication [5, Example 1.43],
are covering space actions and the orbit spaces are ZR = S
1
(the circle), Z/2S
n
= RP
n
(real projective
space), and Z/mS
2n+1
= L
2n+1
(m) (lense space). The action Z S
1
S
1
: (n, z) e
i

2n
z is not a
covering space action for the orbits are dense.
6 J.M. MLLER
Example 2.4. The maps p
n
: S
1
S
1
, n Z, and p

: R S
1
given by p
n
(z) = z
n
, and p

(s) = e
2s
=
(cos(2s), sin(2s)) are covering maps of the circle with bre p
1
n
(1) = Z/nZ and p
1

(1) = Z. There are

many covering maps of S
1
S
1
. The map S
n
C
2
S
n
= RP
n
, n 1, is a covering map of real projective
n-space. The map S
2n+1
C
m
S
2n+1
= L
2n+1
(m) is a covering map of the lens space. M
g
N
g+1
a
double covering map of the unorientable surface of genus g +1 with F = Z/2Z. Can you nd a covering map
of M
g
? Can you nd a covering map of R?
Theorem 2.5 (Unique HLP for covering maps). [5, 1.30] Let p: Y X be a covering map, B be any space,
and h: B I X a homotopy into the base space. If one end of the homotopy lifts to a map B 0 Y
then the whole homotopy admits a unique lift B I Y such that the diagram
B 0

_

e
h0

Y
p

B I
h

e
h

w
w
w
w
w
X
commutes.
Proof. We consider rst the case where B is a point. The statement is then that in the situation
0
y0


_

Y
p

I
u

e u

|
|
|
|
X
there is a unique map u: I Y such that p u = u and u(0) = y
0
. For uniqueness of lifts from I see
Theorem 2.12.(1). We need to prove existence. The Lebesgue lemma (General Topology, 2.158) applied to
the compact space I says that there is a subdivision 0 = t
0
< t
1
< < t
n
= 1 of I such that u maps each
of the closed subintervals [t
i1
, t
i
] into an evenly covered neighborhood in X. Suppose that we have lifted u
to u dened on [0, t
i1
]. Let U be an evenly covered neighborhood of u(t
i1
). Suppose that the lift u(t
i1
)
belongs to U for some F. Continue the given u with (p[(U ))
1
u[[t
i1
, t
i
]. After nitely
many steps we have the unique lift on I.
We now turn to the general situation. Uniqueness is clear for we have just seen that lifts are uniquely
determined on the vertical slices b I B I for any point b of B. Existence is also clear except that
continuity of the lift is not clear.
We now prove that the lift is continuous. Let b be any point of X. By compactness, there is a neighborhood
N
b
of b and a subdivision 0 = t
0
< t
1
< < t
n
= 1 of I such that h maps each of the sets N
b
[t
i1
, t
i
]
into an evenly covered neighborhood of X. Suppose that h(N
b
[0, t
1
]) is contained in the evenly covered
neighborhood U X and let

U p
1
(U) Y be a neighborhood such that p[

U :

U U is a homeomorphism
and

h
0
(b, 0)

U. We can not be sure that

h
0
(N
b
0)

U; only if N
b
is connected. Replace N
b
by
N
b

h
1
0
(

U). Then

h
0
(N
b
0)

U. Then (p[

U)
1
h[N
b
[0, t
1
] is a lift of h[N
b
[0, t
1
] extending

h
0
.
After nitely many steps we have a lift dened on N
b
I (where N
b
is possibly smaller than the N
b
we started
with). Do this for every point b of B. These maps must agree on their overlap by uniqueness. So they dene
a lift B I Y . This lift is continuous since it is continuous on each of the open tubes N
b
I.
We emphasize the special case where B is a point. Let y
0
Y be a point in Y and x
0
= p(y
0
) X its
image in X.
Corollary 2.6 (Unique path lifting). Let x
0
and x
1
be two points in X and let y
0
be a point in the bre
p
1
(x
0
) Y over x
0
. For any path u: I X from x
0
to x
1
, the exists a unique path u: I Y in Y starting
at u(0) = y
0
. Moreover, homotopic paths have homotopic lifts: If v : I X is a path in X that is path
homotopic to u then the lifts u and v are also path homotopic.
Proof. First, in Theorem 2.5, take B to be point. Next, take B to be I and use the HLP to see that homotopic
paths have homotopic lifts.
Corollary 2.7. Let p: Y X be a covering map and let y
0
, y
1
, y
2
Y , x
0
= py
0
, x
1
= py
1
, x
2
= py
2
.
THE FUNDAMENTAL GROUP AND COVERING SPACES 7
(1) By recording end points of lifts we obtain maps
p
1
(x
1
) (X)(x
1
, x
2
)

p
1
(x
2
), p
1
(x
0
)
1
(X, x
0
)

p
1
(x
0
)
given by y [u] = u
y
(1) where u
y
is the lift of u starting at y. Multiplication by a path u from x
1
to
x
2
slides the bre over x
1
bijectively into the bre over x
2
.
(2) The covering map p: X Y induces injective maps
(Y )(y
1
, y
2
)
p
(X)(x
1
, x
2
),
1
(Y, y
0
)
p

1
(X, x
0
)
The subset p

(Y )(y
1
, y
2
) (X)(x
1
, x
2
) consists of all paths from x
1
to x
2
that lift to paths from
y
1
to y
2
. The subbgroup p

1
(Y, y
0
)
1
(X, x
0
) consists of all loops at x
0
that lloft to loops at y
0
.
Denition 2.8. The monodromy functor of the covering map p: X Y is a functor
F(p): (X) Set
of the fundamental groupoid of the base space into the category Set of sets. This functor takes a point in
x X to the bre F(p)(x) = p
1
(x) over that point and it takes a path homotopy class u (X)(x
0
, x
1
)
to F(p)(x
0
) = p
1
(x
0
) p
1
(x
1
) = F(p)(x
1
): y y u. (The notation here is such that F(p)(uv) =
F(p)(v) F(p)(u) for paths u (X)(x
0
, x
1
), v (X)(x
1
, x
2
).)
In particular, the bre F(p)(x) = p
1
(x) over any point x X is a right
1
(X, x)-set.
Corollary 2.9 (The fundamental groupoid of a covering space). The fundamental groupoid of Y ,
(Y ) = (X) F(p)
is the Grothendieck construction of the ber functor (2.8). In other words, the map (p): (Y )(y
0
, y
1
)
(X)(x
0
, x
1
) is injective and the image is the set of path homotopy classes from x
0
to x
1
that take y
0
to y
1
.
In particular, the homomorphism p

:
1
(Y, y
0
)
1
(X, x
0
) is injective and its image is the set of loops at
x
0
that lift to loops at y
0
.
Proof. We consider the functor F(p) as taking values in discrete categories. The objects of (X) F(p)
are pairs (x, y) where x X and y F(p)(x) Y . A morphism (x
1
, y
1
) (x
2
, y
2
) is a pair (u, v) where
u is a morphism in (X) from x
1
to x
2
and v is a morphism in F(p)(x
2
) from F(p)(u)(x
1
) = x
1
u to
y
2
. As F(p)(x
2
) have no morphisms but identities, the set of morphisms (x
1
, y
1
) (x
2
, y
2
) is the set of
u (X)(x
1
, x
2
) such that y
1
u = y
2
. This is precisely (Y )(y
1
, y
2
).
Denition 2.10. For a space X, let
0
(X) be the set of path components of X.
Lemma 2.11. Let p: X Y be a covering map.
(1) Suppose that X is path connected. The inclusion p
1
(x
0
) Y induces a bijection p
1
(x
0
)/
1
(X, x
0
)

0
(Y ). In particular,
Y is path connected
1
(X, x
0
) acts transitively on the bre p
1
(x
0
)
(2) Suppose that X and Y are path connected. The maps

1
(Y, y
1
)(X)(x
1
, x
2
) p
1
(x
2
)
1
(Y, y
0
)
1
(X, x
0
) p
1
(x
0
)

1
(Y, y
1
)u y
1
u
1
(Y, y
0
)u y
0
u
[pu
y
] y [pu
y
] y
are bijections. Here, u
y
is any path in Y from y
1
or y
0
to y. In particular, [
1
(X, x
0
):
1
(Y, y
0
)[ =
[p
1
(x
0
)[.
Proof. The map p
1
(x
0
)
0
(Y ), induced by the inclusion of the bre into the total space, is onto because
X is path connected so that any point in the total space is connected by a path to a point in the bre. Two
points in the bre are in the same path component of Y if and only if are in the same
1
(X, x
0
)-orbit.
If Y is path connected, then
1
(X, x
0
) acts transitively on the bre p
1
(x
0
) with isotropy subgroup

1
(Y, y
0
) at y
0
.
8 J.M. MLLER
Theorem 2.12 (Lifting Theorem). Let p: Y X be a covering map and f : B X a map into the base
space. Choose base points such that f(b
0
) = x
0
= p(y
0
) and consider the lifting problem
(Y, y
0
)
p

(B, b
0
)
f

(X, x
0
)
(1) If B is connected, then there exists at most one lift

f : (B, b
0
) (Y, y
0
) of f over p.
(2) If B is path connected and locally path connected then
There is a map

f : (B, b
0
) (Y, y
0
) such that f = p

f f

1
(B, b
0
) p

1
(Y, y
0
)
Proof. (1) Suppose that

f
1
and

f
2
are lifts of the same map f : B X. We claim that the sets b B [

f
1
(b) =

f
2
(b) and b B [

f
1
(b) ,=

f
2
(b) are open.
Let b be any point of B where the two lifts agree. Let U X be an evenly neighborhood of f(b). Choose

U p
1
(U) = U F so that the restriction of p to

U is a homoemorphism and

f
1
(b) =

f
2
(b) belongs to

U.
Then

f
1
and

f
2
agree on the neighborhood

f
1
1
(

U)

f
1
2
(

U) of b.
Let b be any point of B where the two lifts do not agree. Let U X be an evenly neighborhood of
f(b). Choose disjoint open sets

U
1
,

U
2
p
1
(U) = U F so that the restrictions of p to

U
1
and

U
2
are
homoemorphisms and

f
1
(b) belongs to

U
1
and

f
2
(b) to

U
2
. Then

f
1
and

f
2
do not agree on the neighborhood

f
1
1
(

U
1
)

f
1
2
(

U
2
) of b.
(2) It is clear that if the lift exists, then the condition is satised. Conversely, suppose that the condition
holds. For any point b in B, dene a lift

f by

f(b) = y
0
[fu
b
]
where u
b
is any path from b
0
to b. (Here we use that B is path connected.) If v
b
is any other path from b
0
to b then y
0
[fu
b
] = y
0
[fv
b
] because y
0
[fu
b
fv
b
] = y
0
as the loop [fu
b
fv
b
]
1
(Y, y
0
) xes the point
y
0
by Lemma 2.11.
We need to see that

f is continuous. Note that any point b B has a path connected neighborhood that
is mapped into an evenly covered neighborhood of f(b) in X. It is evident what

f does on this neighborhood
of b.
A map f : B S
1
C0 into the circle has an nth root if and only if the induced homomorphism
f

:
1
(B) Z is divisible by n.
3. The fundamental group of the circle, spheres, and lense spaces
For each n Z, let
n
be the loop
n
(s) = (cos(2ns), sin(2ns), s I, on the circle.
Theorem 3.1. The map : Z
1
(S
1
, 1): n [
n
] is a group isomorphism.
Proof. Let p: R S
1
be the covering map p(t) = (cos(2t), sin(2t)), t R. Remember that the total
space R is simply connected as we saw in Example 1.9. The bre over 1 is p
1
(1) = Z. Let u
n
(t) = nt be
the obvious path from 0 to n Z. By Lemma 2.11 the map
Z
1
(S
1
, 1): n [pu
n
] = [
n
]
is bijective.
We need to verify that is a group homomorphism. Let m and n be integers. Then u
m
(m + u
n
) is a
path from 0 to m+n so it can be used instead of u
m+n
when computing (m+n). We nd that
(m+n) = [p(u
m
(m+u
n
))] = [p(u
m
) p(m+u
n
)] = [p(u
m
)][p(m+u
n
)] = [p(u
m
)][p(u
n
)] = (m)(n)
because p(m+u
n
) = pu
n
as p has period 1.
Theorem 3.2. The n-sphere S
n
is simply connected when n > 1.
THE FUNDAMENTAL GROUP AND COVERING SPACES 9
Proof. Let N be the North and S the South Pole (or any other two distinct points on S
n
). The problem is
that there are paths in S
n
that visit every point of S
n
. But, in fact, any loop based at N is homotopic to
a loop that avoids S (Problem and Solution). This means that
1
(S
n
S, N)
1
(S
n
, N) is surjective.
The result follows as S
n
S is homeomorphic to the simply connected space R
n
.
Corollary 3.3. The fundamental group of real projective n-space RP
n
is
1
(RP
n
) = C
2
for n > 1. The
fundamental group of the lense space L
2n+1
(m) is
1
(L
2n+1
(m)) = C
m
for n > 0.
Proof. We proceed as in Theorem 3.1. Consider the case of the the covering map p: S
2n+1
L
2n+1
(m) over
the lense space L
2n+1
(m). Let N = (1, 0, . . . , 0) S
2n+1
C
n+1
. The cyclic group C
m
= ) of mth roots
of unity is generated by = e
2i/m
. The map
j

j
N, j Z, is a bijection C
m
p
1
pN between the set
C
m
and the bre over pN. As S
2n+1
is simply connected there is a bijection
: p
1
pN = C
m

1
(L
2n+1
(m), pN):
j
[p
j
]
where
j
is the path in S
2n+1
from N to
j
N given by
j
(s) = (e
2isj/m
, 0, . . . , 0). Since
i+j

i
(
i

j
),
it follows just as in Theorem 3.1 that is a group homomorphism.
For the projective spaces, use the paths
j
(s) = (cos(2js), sin(2js), 0, . . . , 0) from N to (1)
j
N, to see
that
: p
1
pN = C
2

1
(RP
n
, pN): (1)
j
[p
j
]
is a bijection.
3.4. Applications of
1
(S
1
). Here are some standard applications of Theorem 3.1.
Corollary 3.5. The nth power homomorphism p
n
: (S
1
, 1) (S
1
, 1): z z
n
induces the nth power homo-
morphism
1
(S
1
, 1)
1
(S
1
, 1): [] []
n
.
Proof. (p
n
)

(1) = (p
n
)

[
1
] = [p
n

1
] = [
n
1
] = [
n
] = (n) = (1)
n
.
Theorem 3.6 (Brouwers xed point theorem). (1) The circle S
1
is not a retract of the disc D
2
.
(2) Any map self-map of the disc D
2
has a xed point.
Proof. (1) Let i : S
1
D
2
be the inclusion map. The induced map i

: Z =
1
(S
1
)
1
(D
2
) = 0 is not
injective so S
1
can not be a retract by 1.7.
(2) With the help of a xed-point free self map of D
2
one can construct a retraction of D
2
onto S
1
. But they
dont exist.
Theorem 3.7 (The fundamental theorem of algebra). Let p(z) = z
n
+a
n1
z
n1
+ +a
1
z +a
0
be a normed
complex polynomial of degree n. If n > 0, then p has a root.
Proof. Any normed polynomial p(z) = z
n
+ a
n1
z
n1
+ + a
1
z + a
0
is nonzero when [z[ is large: When
[z[ > 1 +[a
n1
[ + +[a
0
[, then p(z) ,= 0 because
[a
n1
z
n1
+ +a
0
[ [a
n1
[[z[
n1
+ +[a
0
[ < [a
n1
[[z[
n1
+ +[a
0
[[z[
n1
= ([a
n1
[ + +[a
0
[)[z[
n1
< [z[
n
Therefore any normed polynomial p(z) denes a map S
1
(R) C0 where S
1
(R) is the circle of radius R
and R > 1+[a
n1
[ + +[a
0
[. In fact, all the normed polynomials p
t
(z) = z
n
+t(a
n1
z
n1
+ +a
1
z +a
0
),
t I, take S
1
(R) into C0 so that we have a homotopy
S
1
(R) I C0: (z, t) z
n
+t(a
n1
z
n1
+ +a
1
z +a
0
)
between p
1
(z) = p(z)[S
1
(R) and p
0
(z) = z
n
.
If p(z) has no roots at all, the map p[S
1
(R) factors through the complex plane C and is therefore nullho-
motopic (as C is contractible) and so is the homotopic map S
1
(R) C 0: z z
n
and the composite
map
S
1
zRz
S
1
(R)
zz
n
C0
zz/]z]
S
1
But this is simply the map S
1
S
1
: z z
n
which we know induces multiplication by n (3.5). However, a
nullhomotopic map induces multiplication by 0 (1.14). So n = 0.
A map f : S
1
S
1
is odd if f(x) = f(x) for all x S
1
. Any rotation (or reection) of the circle is
odd (because it is linear).
10 J.M. MLLER
Lemma 3.8. Let f : S
1
S
1
be an odd map. Compose f with a rotation R so that Rf(1) = 1. The induced
map (Rf)

:
1
(S
1
, 1)
1
(S
1
, 1) is multiplication by an odd integer. In particular, f is not nullhomotopic.
Proof. We must compute (Rf)

[
1
]. The HLP gives a lift
0

_

I
1

n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
S
1
Rf

S
1
and we have (Rf)

[
1
] = [p ]. When 0 s 1/2,
1
(s+1/2) =
1
(s) and also Rf
1
(s+1/2) = Rf
1
(s)
as Rf is odd. The lift, of Rf
1
, then satises the equation
(s + 1/2) = (s) +q/2
for some odd integer q. By continuity and connectedness of the interval [0, 1/2], q does not depend on s.
Now (1) = (1/2) + q/2 = (0) + q/2 + q/2 = q and therefore (Rf)

[
1
] = [p ] = [
q
] = [
1
]
q
. We
conclude that (Rf)

is multiplication by the odd integer q. Since a nullhomopotic map induces the trivial
group homomorphism (1.14), f is not nullhomotopic.
Theorem 3.9 (BorsukUlam theorem for n = 2). Let f : S
2
R
2
be any continuous map. Then there
exists a point x S
2
such that f(x) = f(x).
Proof. Suppose that f : S
2
R
2
is a map such that f(x) ,= f(x) for all x S
2
. The composite map
S
1


incl

S
2
x
f(x)f(x)
|f(x)f(x)|

S
1
is odd so it is not nullhomotopic. But the rst map S
1
S
2
is nullhomotopic because it factors through the
contractible space D
2
+
= (x
1
, x
2
, x
3
) S
2
[ x
3
0. This is a contradiction.
This implies that ther are no injective maps of S
2
R
2
; in particular S
2
does not embed in R
2
.
Proposition 3.10 (BorsukUlam theorem for n = 1). Let f : S
1
R be any continuous map. Then there
exists a point x S
1
such that f(x) = f(x).
Proof. Look at the map g(x) = f(x) f(x). If g is identically 0, f(x) = f(x) for all x S
1
. Otherwise, g
is an odd function, g(x) = g(x), and g has both positive and negative values. By connectedness, g must
assume the value 0 at some point.
This implies that there are no injective maps S
1
R; in particular S
1
does not embed in R.
4. The van Kampen theorem
Let G
j
, j J, be a set of groups indexed by the set J. The coproduct (or free product) of these groups is
a group

iJ
G
j
with group homomorphisms
j
: G
j

jJ
G
j
such that
(4.1) Hom(

jJ
G
j
, H) =

jJ
Hom(G
j
, H): (
j
)
jJ
is a bijection for any group H. The group

jJ
G
j
contains each group G
j
as a subgroup and these subgroups
do not commute with each other. If the groups have presentations G
j
= L
j
[ R
j
) then

jJ
L
j
[ R
j
) =

jJ
L
j
[
jJ
R
j
) as this group has the universal property. See [9, 6.2] for the construction of the free
product.
The characteristic property (4.1) applied to H =

jJ
G
j
shows that there is a group homomorphism

G
j

G
j
from the free product to the direct product whose restriction to each G
j
is the inclusion into
the product.
Example 4.2. [9, Example IIIII p 171] Z/2 H Z/2 = Z Z/2 and Z/2 H Z/3 = PSL(2, Z). We can prove
the rst assertion:
Z/2 HZ/2 = a, b [ a
2
, b
2
) = a, b, c [ a
2
, b
2
, c = ab) = a, b [ a
2
, acac, c) = a, b [ a
2
, c, aca = c
1
)
but the second one is more dicult.
THE FUNDAMENTAL GROUP AND COVERING SPACES 11
Suppose that the space X =

jJ
X
j
is the union of open and path connected subspaces X
j
and
that x
0
is a point in

jJ
X
j
. The inclusion of the subspace X
j
into X induces a group homomorphism

j
:
1
(X
j
, x
0
)
1
(X, x
0
). The coproduct

jJ

1
(X
j
, x
0
) is a group equipped with group homomorphisms

j
:
1
(X
j
, x
0
)

jJ

1
(X
j
, x
0
). Let
: H
jJ

1
(X
j
, x
0
)
1
(

jJ
X
j
, x
0
) =
1
(X, x
0
)
be the group homomorphism determined by
j
=
j
.
Is surjective? In general, no. The circle, for instanec, is the union of two contractible open subspaces, so
is not onto in that case. But, if any loop in X is homotopic to a product of loops in one of the subspaces
X
j
, then is surjective.
Is injective? It will, in general, not be injective, because the individual groups
1
(X
i
) in the free product
do not intersect but the subspaces do intersect. Any loop in X that is a loop in X
i
X
j
will in the free
product count as a loop both in
1
(X
i
) and in
1
(X
j
). We always have commutative diagrams of the form

1
(X
i
, x
0
)
i

M
M
M
M
M
M
M
M
M
M
M

1
(X
i
X
j
, x
0
)
ij

n
n
n
n
n
n
n
n
n
n
n
n
ji

P
P
P
P
P
P
P
P
P
P
P
P

1
(X, x
0
)

1
(X
j
, x
0
)
j

q
q
q
q
q
q
q
q
q
q
q
where
ij
are inclusion maps. This means that (
ij
g) = (
ji
g) for any g
1
(X
i
X
j
, x
0
) so that
(4.3) i, j Jg
1
(X
i
X
j
):
ij
(g)
ji
(g)
1
ker
Let N

jJ

1
(X
j
, x
0
) be the smallest normal subgroup containing all the elements of (4.3). The kernel
of must contain N but, of course, the kernel could be bigger. The surprising fact is that often it isnt.
Theorem 4.4 (Van Kampens theorem). Suppose that X =

jJ
X
j
is the union of open and path connected
subspaces X
j
and that x
0
is a point in

jJ
X
j
.
(1) If the intersection of any two of the open subspaces is path connected then is surjective.
(2) If the intersection of any three of the open subspaces is path connected then the kernel of is N.
Corollary 4.5. If the intersection of any three of the open subspaces is path connected then determines
an isomorphism
:

jJ

1
(X
j
, x
0
)/N

=
1
(X, x
0
)
Proof of Theorem 4.4. (1) We need to show that any loop u
1
(X) in X is a product u
1
u
m
of loops
u
i

1
(X
ji
) in one of the subspaces. Let u: I X be a loop in X.
Thanks to the Lebesgue lemma (General Topology, 2.158) we can nd a subdivision 0 = t
0
< t
1
< t
m
=
1 of the unit interval so that u
i
= u[[t
i1
, t
i
] is a path in (say) X
i
. As u(t
i
) X
i
X
i+1
, and also the base
point x
0
X
i
X
i+1
, and X
i
X
i+1
is path connected, there is path g
i
in X
i
X
i+1
from the basepoint x
0
to u(t
i1
). The situation looks like this:
X
1
X
2
X
3

g
1

u[[0, t
1
]

u[[t
1
, t
2
]
[
[
[

g
2

u[[t
2
, 1]
Now u u[[0, t
1
] u[[t
1
, t
2
] u[[t
m1
, 1] (u[[0, t
1
] g
1
) (g
1
u[[t
1
, t
2
] g
2
) (g
m
u[[t
m1
, 1]) is a product
of loops where each factor is inside one of the subspaces.
12 J.M. MLLER
(2) Let N H
1
(X
i
) be the smallest normal subgroup containing all the elements (4.3). Let u
i

1
(X
ji
).
For simplicity, lets call X
ji
for X
i
. Consider the product
u
1
..
1(X1)
u
2
..
1(X2)
u
m
..
1(Xm)
H
jJ

1
(X
j
)
and suppose that (u
1
u
m
) is the unit element of
1
(X). We want to show that u
1
u
m
lies in the
normal subgroup N or that u
1
u
m
is the identity in the quotient group H
1
(X
j
)/N.
Since u
1
u
m
is homotopic to the constant loop in X there is homotopy I I X =

X
j
from the
loop u
1
u
m
in X to the constant loop. Divide the unit square I I into smaller rectangles such that each
rectangle is mapped into one of the subspaces X
j
. We may assume that the subdivision of I 0 is a further
subdivision of the subdivision at i/m coming from the product u
1
u
m
. It could be that one new vertex is
(or more new vertices are) inserted between (i 1)/m and i/m.

X
k
X

X
i
Connect the image of the new vertex with a path g inside X
i
X
k
X

to the base point. Now u

i
is
homotopic in X
i
to the product (u
i
[[(i 1)/m, ] g) (g u
i
[[, i/m]) of two loops in X
i
. This means that we
may as well assume that no new subdivision points have been introduced at the bottom line I 0. Now
perturb slightly the small rectangles, but not the ones in the bottom and top row, so that also the corner of
each rectangle lies in at most three rectangles. The lower left corner may look like this:

X
5
X
6
X
1
u
15
u
16
X
2
u
12
u
1
The loop u
1
in X
1
is homotopic to the product of paths u
15
u
16
u
12
by a homotopy as in the proof of 1.12.
Connect the image of the point to the base point by a path g
156
inside X
1
X
5
X
6
and connect the image
of the point to the base point by a path g
126
inside X
1
X
2
X
6
. Then u
1
is homotopic in X
1
to the
product of loops (u
15
g
156
) (g
156
u
16
g
126
) (g
126
u
12
) in X
1
. The rst of these loops is a loop in X
1
X
5
, the
second is a loop in X
1
X
6
, and the third is a loop in X
1
X
2
. In H
1
(X
j
) and modulo the normal subgroup
N we have that
u
1
..
X1
u
2
..
X2
= u
15
g
156
. .
X1
g
156
u
16
g
126
. .
X1
g
126
u
12
. .
X1
u
2
..
X2
= u
15
g
156
. .
X5
g
156
u
16
g
126
. .
X6
g
126
u
12
u
2
. .
X2

After nitely many steps we conclude that modulo N the product u
1
u
m
equals a product of constant
loops, the identity element.
Corollary 4.6. Let X
j
be a set of path connected spaces. Then

jJ

1
(X
j
)

=
1
(

jJ
X
j
)
provided that each base point x
j
X
j
is the deformation retract of an open neighborhood U
j
X
j
.
Proof. Van Kampens theorem does not apply directly to the subspaces X
j
of
_
X
j
because they are not
open. Instead, let X
t
j
= X
j

_
iJ
U
i
. The subspaces X
t
j
are open and path connected and the intersection
of at least two of them is the contractible space
_
iJ
U
i
. Moreover, X
j
is a deformation retract of X
j
.
For instance, punctured compact surfaces have free fundamental groups.
THE FUNDAMENTAL GROUP AND COVERING SPACES 13
Corollary 4.7 (van Kampen with two subspaces). Suppose that X = X
1
X
2
where X
1
, X
2
, and X
1
X
2
,=
are open and path connected. Then

1
(X
1
X
2
, x
0
)

=
1
(X
1
, x
0
) H
1(X1X2,x0)

1
(X
2
, x
0
)

=
for any basepoint x
0
X
1
X
2
.
This means that when X
1
X
2
is path connected the fundamental group functor takes a push out of spaces
to a push out, amalgamated product, of groups
X
1
X
2
i1

i2

X
2

1
(X
1
X
2
)
(i1)

(i2)

1
(X
2
)

X
2

X
1
(X
2
)


1
(X)
As a very special case, we see that a space, that is the union of two open simply connected subspaces with path
connected intersection, is simply connected. This proves, again (Theorem 3.2), that S
n
is simply connected
when n > 1.
We can use this simple variant of van Kampen to analyze the eect on the fundamental group of attacing
cells.
Corollary 4.8 (The fundamental group of a cellular extension). Let X be a path connected space. Then

1
(X

D
n

) =
_

1
(X)/

) n = 2

1
(X) n > 2
where

is a path from the base point of X to the image of the base point of S
1

D
2

.
Proof. Let Y be X with the n-cells attached. Attach strips, fences connecting the base point of X with the
base points of the attached cells, to Y and call the results Z. This does not change the fundamental group
as Y is a deformation retract of Z (Corollary 1.7). Let A be Z with the top half of each cell removed and let
B = Z X. Then Z = A B and A B are path connected (the fences are there to make A and B path
connected) so that

1
(Z) =
1
(A) H
1(AB)

1
(B)
by the van Kampen theorem in the simple form of Corollary 4.7. Now B is contractible, hence simply
connected (Corollary 1.14), so
1
(Y ) =
1
(Z) is the quotient of
1
(A) by the smallest normal subgroup
containing the image of
1
(A B)
1
(A). But A B is homotopy equivalent to a wedge
_

S
n1

of
(n 1)-spheres. In particular, A B is simply connected when n > 2 (Corollary 4.6, Theorem 3.2) so
that
1
(Y ) =
1
(Z) =
1
(A) =
1
(X). When n = 2,
1
(A B) is a free group and the image of it in

1
(A) =
1
(X) is generated by the path homotopy classes of the loops

.
Corollary 4.9. Let X be a CW-complex with skeleta X
k
, k 0. Then

0
(X
1
) =
0
(X),
1
(X
2
) =
1
(X)
Corollary 4.10. The fundamental groups of the compact surfaces of positive genus g are

1
(M
g
) = a
1
, b
1
, . . . , a
g
, b
g
[

[a
i
, b
i
]),
1
(N
g
) = a
1
, . . . , a
g
[

a
2
i
),
The compact orientable surfaces M
g
, g 0, are distinct,
1
(M
g
)
ab
= Z
2g
, and the compact nonorientable
surfaces N
h
, h 1, are distinct,
1
(N
g
)
ab
= Z
g
Z/2.
Corollary 4.11. Let M be a connected manifold of dimension 3. Then
1
(M x) =
1
(M) for any
point x M.
Proof. Apply van Kampen to M = MxD
n
, MxD
n
S
n1
and remember that S
n1
is simply
connected when n 3.
Which groups can be realized as fundamental groups of spaces? For instance, C

= S
1
and C
m
= S
1

m
D
2
so that any nitely generated abelian group can be realized as the fundamental group of a product of these
spaces.
14 J.M. MLLER
Corollary 4.12. For any group G there is a 2-dimensional CW-complex X
G
such that
1
(X
G
)

= G.
Proof. Choose a presentation G = g

[ r

) and let
X
G
= D
0

g]
D
1

]
D
2
be the 2-dimensional CW-complex whose 1-skeleton is a wedge of circles, one for each generator, with 2-discs
attached along the relations.
Observe that X
HIG
= X
H
X
G
. Also, X
1(Mg)
= M
g
, X
1(Ng)
= N
g
, g 1.
4.1. Fundamental groups of knot and link complements. The complement of a pair of unlinked circles
in R
3
deformation retracts to S
1
S
1
S
2
S
2
and a pair of linked circles to (S
1
S
1
)S
2
. The fundamental
groups are Z Z and Z Z, respectively. Thus the two complements are not homeomorphic.
Let m and n be relatively prime natural numbers and K = K
mn
the (m, n)-torus knot. We want to
compute the knot group
1
(R
3
K).
According to (4.11),
1
(R
3
K) =
1
(S
3
K). Now
S
3
= D
4
= (D
2
D
2
) = D
2
D
2
D
2
D
2
is the union of two solid tori intersecting in a torus S
1
S
1
. Let K be embedded in this middle torus. Then
S
3
K = (D
2
D
2
K) (D
2
D
2
K), (D
2
D
2
K) (D
2
D
2
K) = S
1
S
1
K
and van Kampen says (if we ignore
1
the condition that the subsets should be open)

1
(S
3
K) =

1
(D
2
D
2
K) H
1
(D
2
D
2
K)

1
(S
1
S
1
K)
Here, D
2
D
2
K deformation retracts onto the core circle D
2
0, and S
1
S
1
K (the torus minus
the knot) is an annulus S
1
(0, 1). (Take an open strip [0, 1] (O, 1) and wrap it around the torus so that
the end 0 (0, 1) meets the end 1 (0, 1)). The image of the generator of this innite cyclic group is the m
power of a generator, respectively the nth power. Hence

1
(S
3
K) = a, b [ a
m
= b
n
) = G
mn
It is now a matter of group theory to tell us that if G
m1n1
and G
m2n2
are isomorphic then m
1
, n
1
=
m
2
, n
2
. In order to analyze this group, note that a
m
= b
n
is in the center. Let C be the central group
generated by this element. The quotient group
G
mn
/C = a, b [ a
m
, b
n
) = Z/mHZ/n
has no center. (In general the free product GHH of two nontrivial groups has no center because the elements
are words in elements from G alternating with elements from H.) Therefore C is precisely the center of G
mn
.
Thus we can recover mn as the order of the abelianization of G/Z(G). Also, any element of nite order
in Z/mH Z/n is conjugate to an element of Z/m or Z/n. Thus we can recover the largest of m, n as the
maximal order of a torsion element in G/Z(G). Thus we can recover the set m, n.
Corollary 4.13. There are innitely many knots. (Here are some of them.)
Another way of saying this is that D
2
D
2
K deformation retracts onto the mapping cylinder of the
degree m, respectively n, map S
1
S
1
. Thus the union of these two spaces, S
3
K, deformation retracts
onto the union of the two mapping cylinders, which is the double mapping cylinder X
mn
for the two maps.
Thus X
mn
embeds in S
3
and R
3
when (m, n) = 1. On the other hand X
22
is the union of two Mobius
bands. A M obius band is RP
2
minus an open 2-disc, so X
22
= RP
2
#RP
2
, the Klein bottle, which does not
embed in R
3
.
1
To x this, thicken the knot and enlarge the two solid tori a little so that they overlap.
THE FUNDAMENTAL GROUP AND COVERING SPACES 15
5. Categories
A category ( consists of [7]
Objects a, b, . . .
For each pair of objects a and b a set of morphisms ((a, b) with domain a and codomain b
A composition function ((b, c) ((a, b) ((a, c) that to each pair of morphisms g and f with
dom(g) = cod(f) associates a morphism g f with dom(g f) = dom(f) and cod(g f) = cod(g)
We require
Identity: For each object a the morphism set ((a, a) contains a morphism id
a
such that g id
a
= g
and id
a
f = f whenever these compositions are dened
Associativity : h (g f) = (h g) f whenever these compositions are dened
A morphism f ((a, b) with domain a and codomain b is sometimes written f : a b. A morphism
f : a b is an isomorphism if there exists a morphism g : b a such that the two possible compositions are
the respective identities.
Denition 5.1. A group is a category with one object where all morphisms are isomorphisms. A groupoid
is a category where all morphisms are isomorphisms.
Example 5.2. In the category Top of topological spaces, the objects are topological spaces, the morphisms
are continuous maps, and composition is the usual composition of maps. In the category hoTop, the objects
are topological spaces, the morphisms are homotopy classes of continuous maps, and composition is induced
by the usual composition of maps. In the category Grp of groups, the objects are groups, the morphisms are
groups homomorphisms, and composition is the usual composition of group homomorphisms. In the category
Mat
R
the objects are the natural numbers Z
+
, the set of morphisms m n consists of all n by m matrices
with entries in the commutative ring R, and composition is matrix multiplication. The fundamental groupoid
(X) of a topological space X is a groupoid where the objects are the points of X and the morphisms x y
are the homotopy classes (X)(x, y) of paths from x to y, and composition is composition of path homotopy
classes.
A functor F : ( T associates to each object a of ( an object F(a) of T and to each morphism f : a b
in ( a morphism F(f): F(a) F(b) in T such that F(id
a
) = id
F(a)
and F(g f) = F(g) F(f).
A natural transformation : F = G: ( T between two functors F, G: ( T is a T-morphism
(a) T(Fa, Ga) for each object a of ( such that the diagrams
a
f

Fa
Ff

(a)

Ga
Gf

b Fb
(b)

Gb
commute for all morphisms f ((a, b) in (. A natural transformation is a natural isomorphism if all the
components (a), a Ob((), are T-isomorphisms.
Example 5.3. The fundamental group is a functor from the category of based topological spaces and based
homotopy classes of maps to the category of groups.
The fundamental groupoid is a functor from the category of topological spaces to the category of groupoids.
Any homotopy h: f
0
f
1
induces a natural isomorphism h: (f
0
) =(f
1
): (X) (Y ) between functors
between fundamental groupoids (Lemma 1.12).
Denition 5.4. Let ( and T be categories. The functor category Func((, T) is the category whose objects
are the functors from ( to T and whose morphisms are the natural transformations.
Denition 5.5. Two categories, ( and T, are isomorphic (equivalent) when there are functors (
F

T
G

such that the composite functors are (naturally isomorphic to) the respective identity functors.
Lemma 5.6. A functor F : ( T is an equivalence of categories if and only if
any object of T is isomorphic to an object of the form F(a) for some object a of (
16 J.M. MLLER
F is bijective on morphism sets: The maps ((a, b)
fF(f)
T(F(a), F(b)) are bijections for all objects
a and b of (
Proof. Suppose that F : ( T is an equivalence of categories. Then there is a functor G in the other direction
and natural isomorphisms : GF = 1
c
and : FG = 1
1
. Let d be any object of T. The isomorphism

d
: FG(d)

=
d shows that d is isomorphic to Fa for a = Gd. Let a, b be objects of (. We note rst that
((a, b) T(Fa, Fb) ((GFa, GFb) is injective for the commutative diagram
a
f

GFa
GFf

a
f

b GFb

b
shows that f =
b
GFf
1
a
can be recovered from GFf. Thus ((a, b) T(Fa, Fb) is injective.
Symmetrically, also the functor G is injective on morphism sets. To show that F is surjective on morphism
sets let g be any T-morphism Fa Fb. Put f =
b
Gg
1
b
. The commutative diagram
Fa
g

GFa
Gg

a
f

GFa
a

GFf

a
f

Fb GFb

b GFb

b
shows that GFf = Gg and so Ff = g since G is injective on morphism sets.
Conversely, suppose that F : ( T is a functor satisfying the two conditions. We must construct a
functor G in the other direction and natural isomorphisms : FG = 1
1
and : GF = 1
c
. By the rst
condition, for every object d T, we can nd an object Gd ( and an isomorphism
d
: FGd d. By the
second condition, ((Gc, Gd)

= T(FGc, FGd) for any two objects c and d of T. Here, T(c, d)

= T(FGc, FGd)
because FGc

= c and FGd

= d. Thus we have T(c, d)

= T(FGc, FGd)

= ((Gc, Gd). This means that for
every T-morphism g : c d there is exactly one (-morphism Gg : Gc Gd such that
FGc
FGg

c
g

FGd

d
commutes. Now G is a functor and a natural isomorphism FG = 1
1
. What about GF? Well, for any
object a of (, ((GFa, a)

= T(FGFa, Fa)
Fa
so there is a unique isomorphism
a
: GFa a such that
F
a
=
Fa
. This gives the natural isomorphism : GF =1
c
.
It follows that when (
F

T
G
is an equivalence of categories then there are bijections
((c, Gd) = T(Fc, d) ((Gd, c) = ((d, Fc)
of morphism sets.
Lemma 5.7. If (, (
t
and T, T
t
are equivalent, then the functor categories Func((, T) and Func((
t
, T
t
) are
equivalent.
The full subcategory generated by some of the objects of ( is the category whose objects are these objects
and whose morphisms are all morphisms in (.
Example 5.8. The category of nite sets is equivalent to the full subcategory generated by all sections
S
<n
= x Z
+
[ x < n, n Z
+
, of Z
+
. The category of nite dimensional real vector spaces is equivalent
to the category Mat
R
. If f : X Y is a homeomorphism (homotopy equivalence) then the induced morphism
(f): (X) (Y ) is an isomorphism (equivalence) of categories. The fundamental groupoid of a space is
equivalent to the full subcategory generated by a point in each path component.
THE FUNDAMENTAL GROUP AND COVERING SPACES 17
6. Categories of right G-sets
Let G be a topological group and F and Y topological spaces.
Denition 6.1. A right action of G on F is a continuous map F G F : (x, g) x g, such that
x e = x and x (gh) = (x g) h for all g, h G and all x F. A topological space equipped with a right
G-action is called a right G-space. A continuous map f : F
1
F
2
between two right G-spaces is a G-map if
f(xg) = f(x)g for all g G and x F
1
.
Denition 6.2. A left action of G on Y is a continuous map GY Y : (g, y) g y, such that e y = y
and (gh) y = g (h y) for all g, h G and all y Y . A topological space equipped with a left G-action is
called a left G-space. A continuous map f : Y
1
Y
2
between two left G-spaces is a G-map if f(gy) = gf(y)
for all g G and x Y
1
.
The orbit spaces (with the quotient topologies) are denoted F/G = xG [ x F for a right action
F G F and GY = Gy [ y Y for a left action GY Y .
The orbit through the point x F for the right action F G F is the sub-right G-space xG = xg [ g
G obtained by hitting x with all elements of G; the stabilizer at x is the subgroup
x
G = g G [ xg = x
of G. The universal property of quotient spaces gives a commutative diagram
G
gxg

gxGg

E
E
E
E
E
E
E
E
xG
x
GG
xGgxg

x
x
x
x
x
x
x
x
of right G-spaces and G-maps (General Topology, 2.81). Note that G-map
x
GG xG:
x
Gg xg is
bijective. (In particular, the index of the stabilizer subgroup at x equals the cardinality of the orbit through
x.) In many cases it is even a homeomorphism so that the orbit xG through x and the coset space
x
GG of
the isotropy subgroup at x are homeomorphic.
Proposition 6.3 (G-orbits as coset spaces). Suppose that F is a right G-space and x a point of F. Then
x
GG xG is a homeomorphism G
gxg
xG is a quotient map
Proof. Use that the a bijective quotient map is a homeomorphism, the composition of two quotient maps is
quotient, and if the composition of two maps is quotient than the last map is quotient (General Topology,
2.77). By denition, G
x
GG is quotient.
By a right (or left) G-set we just mean a right (or left) G-space with the discrete topology. In the following
we deal with G-sets rather than G-spaces.
Denition 6.4. GSet is the category of right G-sets and G-maps. The objects are right G-sets F and the
morphisms : F
1
F
2
are G-maps (meaning that (xg) = (x)g for all x F
1
and g G).
6.5. Transitive right actions. The right G-set F is transitive if F consists of a single orbit. If F is transitive
then F = xG for some (hence any) point x F so that F and HG are isomorphic G-sets where H is the
stabilizer subgroup at the point x (Proposition 6.3). Thus any transitive right G-set is isomorphic to the
G-set HG of right H-cosets for some subgroup H of G.
Denition 6.6. The orbit category of G is the full subcategory O
G
of GSet generated by all transitive right
G-sets.
The orbit category O
G
of G is equivalent to the full subcategory of GSet generated by all G-sets of the
form HG for subgroups H of G. What are the morphisms in the orbit category O
G
?
Denition 6.7. Let H
1
and H
2
be subgroups of G. The transporter is the set
N
G
(H
1
, H
2
) = n G [ nH
1
n
1
H
2

of group elements conjugating H

1
into H
2
.
The transporter set N
G
(H
1
, H
2
) is a left H
2
-set. Let Let H
2
N
G
(H
1
, H
2
) be the set of H
2
-orbits.
18 J.M. MLLER
Proposition 6.8. There is a bijection
: H
2
N
G
(H
1
, H
2
) O
G
(H
1
G, H
2
G), (H
2
n)(H
1
g) = H
2
ng
This map takes H
2
n to left multiplication H
1
G
H1gH2ng
H
2
G by H
2
n. In case H
1
= H = H
2
, the map
: HN
G
(H) O
G
(HG, HG), (Hn)(Hg) = Hng
is a group isomorphism.
Proof. The inverse to is the map that takes a G-map H
1
G

H
2
G to its value (H
1
) = H
2
n at
H
1
H
1
G. Since H
2
n = (H
1
) = (H
1
H
1
) = H
2
nH
1
, the group element n conjugates H
1
into H
2
. In case
H
1
= H = H
2
and n
1
, n
2
N
G
(H), we have
(Hn
1
)(Hn
2
)(H) = (Hn
1
)(Hn
2
) = Hn
1
n
2
= (Hn
1
n
2
)(H)
so is group homomorphism in this case.
In particular we see that
all morphisms in O
G
are epimorphisms
all endomorphisms in O
G
are automorphisms
every object HG of O
G
is equipped with left and right actions
(6.9) HN
G
(H) HGG = O
G
(HG, HG) HGG HG: Hn Hg m = Hngm
where the left action are the G-automorphisms of HG in O
G
.
the maximal G-orbit is G = eG and O
G
(eG, HG) = HG, the minimal G-orbit is = GG
and O
G
(HG, GG) = (GG = is the nal object of O
G
)
Remark 6.10 (Isomorphism classes of objects of O
G
). The set of objects of O
G
corresponds to the set
of subgroups of G. The set of isomorphism classes of objects of O
G
corresponds to the set of conjugacy
classes of subgroups of G: Two objects H
1
G and H
2
G of the orbit category O
G
are isomorphic if and
only if H
1
and H
2
are conjugate: If there exist an inner automorphism that takes H
1
into H
2
and an inner
automorphism that takes H
2
into H
1
such that the composite maps are the respective identity maps of
H
1
G and H
2
G, then these inner automorphisms must in fact give bijections between H
1
and H
2
as the
factorizations H
1
Inn(n1)
H
2
Inn(n2)
H
1
Inn(n1)
H
2
of the respective identity maps imply that the inner
automorphism Inn(n
1
) is a bijection.
7. The classification theorem
In this section we shall see that covering maps are determined by their monodromy functor.
Denition 7.1. Cov(X) is the category of covering spaces over the space X. The objects are covering maps
Y X and the morphisms Cov(X)(p
1
: Y
1
X, p
2
: Y
2
X) are continuous maps f : Y
1
Y
2
over X
(meaning that f preserves bres or p
1
= p
2
f).
How can we describe the category Cov(X)? We are going to assume from now on that X is path connected
and locally path connected.
Let Func((X), Set) be the category of functors from the fundamental groupoid (X) to the category Set
of sets. There is a functor
Cov(X) Func((X), Set)
which takes a covering map p: Y X to its monodromy functor F(p): (X) Set (2.8) and a covering
map morphism to the induced natural transformation of functors. Conversely, does any such functor come
from a covering space of X?
Suppose that F : (X) Set is any functor. Let Y (F) =

xX
F(x) be the union of the bres and let
p(F): Y (F) X be the obvious map taking F(x) to x for any point x X.
Denition 7.2. A space X is semi-locally simply connected at the point x X if any neighborhood of x
contains a neighborhood U of x such that any loop at x in U is contractible in X. The space X is semi-locally
simply connected if it is semi-locally simply connected at all its points.
All locally simply connected spaces are semi-locally simply connected.
THE FUNDAMENTAL GROUP AND COVERING SPACES 19
Lemma 7.3. Suppose that X is locally path connected and semi-locally simply connected. Then there is a
topology on Y (F) such that p(F): Y (F) X is a covering map. The monodromy functor of p(F): Y (F) X
is F.
Proof. Suppose that x is a point in X and U X an open path connected neighborhood of x such that any
loop in U based at x is nullhomotopic in X. Observe that this implies that there is a unique path homotopy
class u
z
from x to any other point z in U so that
U F(x) p
1
(U): (y, z) F(u
z
)(y)
is a bijection.
For each y F(x), let (U, y) Y be the image of U y under the above bijection. By assumption, the
topological space X has a basis of sets U as above. The sets (U, y) then form a basis for a topology on Y .
The covering map Y (F) X determines a bre functor (2.8) from the fundamental groupoid of X to the
category of sets. By construction, this bre functor is F.
Denition 7.4. A covering map p: Y X is universal if Y is simply connected.
According to the Lifting Theorem 2.12, any two universal covering spaces over X are isomorphic in the
category Cov(X) of covering spaces over X. We may therefore speak about the universal covering space of
X. Is there always a universal covering space of X?
By Corollary 2.9 the fundamental groupoid of Y (F) has the set Y (F) as object set and the morphisms are
(7.5) (Y (F))(y
1
, y
2
) = u (X)(x
1
, x
2
) [ F(u)y
1
= y
2

for all points x

1
, x
2
X and y
1
F(x
1
), y
2
F(x
2
). In particular, let x
0
be a base point in X. There is a
right action F(x
0
)
1
(X, x
0
) F(x
0
) and
Y (F) is path connected The right action of
1
(X, x
0
) on F(x
0
) is transitive
Y (F) is simply connected The right action of
1
(X, x
0
) on F(x
0
) is simply transitive
We can always nd a functor that satises the last condition in that
F = (X)(x
0
, ): (X) Set
is a functor and the action of
1
(X, x
0
) on F(x
0
) =
1
(X, x
0
) is simply transitive.
Corollary 7.6. X admits a simply connected covering space if and only if X is semi-locally simply connected.
Proof. The covering space Y (F) of the functor F = (X)(x
0
, ) is simply connected.
Conversely, suppose that p: Y X is a covering map and U X and evenly covered open subspace then
U X factors through Y X. If
1
(Y ) is trivial then
1
(U)
1
(X) is the trivial homomorphism.
Example 7.7. The Hawaiian Earring

nZ+
C
1/n
and the innite product

S
1
of circles are connected and
locally path connected but not semi-locally simply connected. Thus they have no simply connected covering
spaces. The innite join
_
S
1
does have a simply connected covering space since it is a CW-complex. Indeed
any CW-complex or manifold is locally contractible [5, Appendix], in particular locally simply connected.
Theorem 7.8 (Classication of Covering Maps). Suppose that X is semi-locally simply connected. The
monodoromy functor and the functor F Y (F)
Cov(X)

Func((X), Set)
are category isomorphisms.
Proof. Let p
1
: Y
1
X and p
2
: E
Y
X be covering maps over X with associated functors F
1
and F
2
. A
covering map
Y
1
p1

A
A
A
A
A
A
A
f

Y
2
p2
.}
}
}
}
}
}
}
X
induces a natural transformation
f
: F
1
= F
2
of functors given by
f
(x) = f[p
1
1
x: p
1
x
1
p
1
x
2
.
Conversely, any natural transformation : F
1
= F
2
induces a covering map Y (f): Y (F
1
) Y (F
2
) of the
associated covering spaces.
20 J.M. MLLER
For example, let F : (X) Set be any functor, let x
0
X and y
0
F(x
0
). There is a natural transfor-
mation (X)(x
0
, ) = F whose x-component is (X)(x
0
, x) F(x): u F(u)y
y
for any point x of X.
This conrms that the universal covering space lies above them all.
Corollary 7.9. The functor
Cov(X)
1
(X, x
0
)Set: (p: Y X) p
1
(x
0
)
is an equivalence of categories.
Proof. The inclusion
1
(X, x
0
) (X) of the the full subcategory of (X) generated by x
0
into (X) is
an equivalence of categories. The induced functor Func((X), Set) Func(
1
(X, x
0
), Set) is then also an
equivalence. But Func(
1
(X, x
0
), Set) is simply the category of right
1
(X, x
0
)-sets.
In particular, the full subcategory Cov
0
(X) of connected covering spaces over X is equivalent to the
category of transitive right
1
(X, x
0
)-sets which again is equivalent to the orbit category O
1(X,x0)
(6.6). The
set of covering space morphisms from the connected covering space p
1
: Y
1
X to the connected covering
space p
2
: Y
2
X is
Cov(X)(p
1
: Y
1
X, p
2
: Y
2
X) = Func((X), Set)(F(p
1
), F(p
2
))
=
1
(X)Set(p
1
1
(x
0
), p
1
2
(x
0
))
= O
1(X)
(
1
(Y
1
)
1
(X),
1
(Y
2
)
1
(X))
=
1
(Y
2
)N
1(X)
(
1
(Y
1
),
2
(Y
2
))
and, in particular,
Cov(X)(p: Y X, p: Y X) =
1
(Y )N
1(X)
(
1
(Y ))
for any connected covering space p: Y X over X. If we map out of the universal covering space X1) X
this gives
Cov(X)(X1) X, Y X) =
1
(Y )
1
(X) Cov(X)(X1) X, X1) X) =
1
(X)
which means that the universal covering space admits a left covering space
1
(X)-action with orbit space

1
(X)X1) X = X.
Corollary 7.10. Let G =
1
(X) for short. The functor
O
G
Cov
0
(X): H (HX1) GX1))
is an equivalence of categories.
Is this
C
2

F
F
F
F
F
F
F
F
id

e
3
3

x
x
x
x
x
x
x
x

x
x
x
x
x
x
x
x

x
x
x
x
x
x
x
x

F
F
F
F
F
F
F
F

F
F
F
F
F
F
F
F
3

3
id
.
C
3

x
x
x
x
x
x
x
x
C2

a picture of the orbit category of symmetric group

3
or is it a picture of the path connected covering spaces
over a path connected, locally path connected, and semi-locally simply connected space with fundamental
group
3
? Both! The space could be X
3
from Corollary 4.12; see Example 7.20 for more information.
Here are some examples to illustrate the Classication of Covering Spaces.
Covering spaces of the circle: The category Cov
0
(S
1
) = O
C
of path connected covering spaces of
the circle S
1
= ZR consists of the covering spaces nZR ZR where n = 0, 1, 2, . . .. There is a
THE FUNDAMENTAL GROUP AND COVERING SPACES 21
covering map nZR mZR if and only if m[n and in that case there are m such covering maps,
namely the maps
S
1
z
n

A
A
A
A
A
A
A
A
z
m/n

S
1
z
m
.}
}
}
}
}
}
}
}
S
1
where is any mth root of unity.
Covering spaces of projective spaces: The category Cov
0
(RP
n
) = O
C2
of connected covering spaces
of real projective n-space RP
n
, n 2, has 2 objects, namely the trivial covering map RP
n
RP
n
and the universal covering map S
n
RP
n
.
Covering spaces of lense spaces: The universal covering space of the lense space L
2n+1
(m) = C
m
S
2n+1
,
n 1, is S
2n+1
. The other covering spaces are the lense spaces L
2n+1
(r) = C
r
S
2n+1
for eah divisor
r of m. The category of connected covering spaces of L
2n+1
(m) is equivalent to the orbit category
O
Cm
.
Covering spaces of surfaces: The category Cov
0
(M
g
) = O
1(Mg)
is harder to describe explicitly.
Any nite sheeted covering space of a compact surface is again a compact surface. The paper [8]
contains information about covering spaces of closed surfaces.
Example 7.11 (Covering spaces of the Mobius band). The cylinder S
1
[1, 1] = Z(R [1, 1]) where
the action is given by n (x, t) (x +n, t)
-
-
6 6 6 6 6 6 6 6

The Mobius band MB = Z(R[1, 1]) where the action is given by n (x, t) (x +n, (1)
n
t)
-
-
? ? ? ?
6 6 6 6


Every even-sheeted covering space of the Mobuis band is a cylinder, every odd-sheeted covering space is a
Mobuis band.

_
_
Example 7.12 (Covering spaces of S
1

m
(S
1
I)). Let X
m
= S
1

m
(S
1
I) be the mapping cylinder
of the degree m map of the circle. We can construct X
m
in the following way: Take a (codomain) circle
of circumference 1/m and a square [0, 1] [0, 1]. Wrap the bottom edge [0, 1] 0 of the square m times
around the circle in a screw motion so that each time the square travels once around the circle it is also being
rotated an angle of 2/m. Finally, glue the two ends, 0 [0, 1] and 1 [0, 1], of the square together.
There is a picture of X
m
in [5, Example 1.29]. The codomain circle is the core circle and the domain circle is
the boundary circle. The fundamental group
1
(X
m
) is Z since X
m
deformation retracts onto the codomain
(core) circle so that the inclusion S
1
i1
X
m
S
1
is a homotopy equivalence. The inclusion S
1
i0
X
m
S
1
of the domain (boundary) circle induces multiplication by m on the fundamental groups; this is simply
because of the general mapping cylinder diagram which becomes
S
1
m

P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
i0

(S
1
I)
m
S
1
= X
m
DR

S
1
i1
:

in this special case. It may help to envision the boundary circle in X

m
sliding towards the core circle.
The universal covering space of X
m
is X
m
1) = CZ/m R where CZ/m = (Z/m I)/(Z/m 1) is
the cone on the set Z/m with m points. (CZ/m is a starsh with m arms). We may realize CZ/m R in
22 J.M. MLLER
Figure 2. The universal covering space of X
m
R
3
with CZ/m placed horizontally in the XY -plane and R as the vertical Z-axis. The covering space action
of the unit 1 Z on CZ/mR is then the screw motion ([[a]
m
, t], x) ([[a +1]
m
, t], x +1/m) with matrix
_
_
cos(2/m) sin(2/m) 0
sin(2/m) cos(2/m) 0
0 0 1/m
_
_
that rotates CZ/m counterclockwise 1/mth of a full rotation and moves up along the Z-axis 1/mth of a unit.
(In Figure 7.12 the R-axis isnt exactly vertical since that would take up too much space. The covering space
action takes the indicated lines, situated at distance 1/m, to each other.) What is the lift of the domain and
the codomain circles of X
m
to the universal covering space X
m
1)? (One of them will lift to a loop.)
Since m Z acts trivially on CZ/m there is an m-sheeted covering map
CZ/mS
1
= CZ/mmZR = mZ(CZ/mR) Z(CZ/mR) = X
m
with mZZ as deck transformation group. What is the lift of the domain and the codomain circles to this
m-fold covering space?
Let X = X
1
X
2
be a CW-complex that is the union of two connected subcomplexes X
1
and X
2
with
connected intersection X
1
X
2
. According to van Kampen, the fundamental group G =
1
(X) = G
1
H
A
G
2
is the free product of G
1
=
1
(X
1
) and G
2
=
1
(X
2
) with A =
1
(X
1
X
2
) amalgamated. We will assume
that the homomorphisms G
1
A G
2
are injective. Then also the homomorphisms G
1
G G
2
of the
push-out diagram
A

G
2

G
1

G
are injective according to the Normal Form Theorem for Free Products with Amalgamation [6, Thm 2.6].
Let X1) be the universal covering space of X = GX1) and let p: X1) X be the covering projection
map. The spaces p
1
(X
1
) and p
1
(X
2
) are left G-spaces with intersection p
1
(X
1
)p
1
(X
1
) = p
1
(X
1
X
2
).
Let y
0
p
1
(X
1
X
2
) be a base point. The commutative diagram [3, II.7.5]

1
(p
1
X
1
, y
0
)

_

1
(p
1
X, y
0
) = 1

_

1
(X
1
, p(y
0
))

1
(X, p(y
0
))
tells us that the component of p
1
(X
1
) containing y
0
is simply connected so it is the universal covering space
X
1
1) of X
1
= G
1
X
1
1). We see from this that there is a homeomorphism of left G-spaces
G
G1
X
1
1)
:
p
1
(X
1
)
induced by the map G X
1
1) p
1
(X
1
) sending (g, y) to gy. Similar arguments apply to p
1
(X
2
) and
p
1
(X
1
X
2
), of course, and hence
X1) = G
G1
X
1
1)
G
A
(X1X2)1)
G
G2
X
2
1)
is the union of the two G-spaces G
Gi
X
i
1), i = 1, 2. This means that the universal covering space of X is
the union of the G-translates of the universal covering spaces of X
1
and X
2
joined along G-translates of the
universal covering space of X
1
X
2
. The next example demonstrates this principle.
THE FUNDAMENTAL GROUP AND COVERING SPACES 23
Example 7.13. [5, 1.24, 1.29, 1.35, 1.44, 3.45] Let X
mn
= X
m

S
1 X
n
be the double mapping cylinder for
the degree m map and the degree n map on the circle. X
mn
is the union of the two mapping cylinders with
their domain (boundary) circles identied, X
m
X
n
= S
1
. By van Kampen, the fundamental group has a
presentation

1
(X
mn
) =
1
(X
m
) H
1(S
1
)

1
(X
n
) = a, b [ a
m
= b
n
) = G
mn
with two generators and one relation. We shall now try to build its universal covering space.
We may equip X
mn
with the structure of a 2-dimensional CW-complex. The 1-skeleton of X
mn
consists
of two circles, a and b, joined by an interval, c, and X
mn
= X
1
mn

a
m
cb
n
c
D
2
is obtained by attaching a 2-cell
_
`

_
`

a b
c

Figure 3. 1-skeleton X
1
mn
of X
mn
along the loop a
m
cb
n
c. (If we use the corollary to van Kampen [5, 1.26] instead of the van Kampen theorem
itself we get that
1
(X
mn
) = a, cbc [ a
m
(cbc)
n
).)
The universal covering space X
mn
1) is also a 2-dimensional CW-complex. The inverse image in X
m
1)
of the left half of the 1-skeleton is the vertical line R with spiraling rungs attached 1/mth of a unit apart.
Rungs with vertical distance 1 point in the same direction so they can be joined up with the inverse image
in X
n
1) of the right half of the 1-skeleton. Now ll in 2-cells in each of the rectangles with sides a
m
, c,
a
m
b
n

Figure 4. Part of 1-skeleton of X

mn
1)
b
n
and c. Continue this process. There will be similar rectangles shifted up 1/mth unit along the left axis
and rotated 2/m or up 1/nth unit along the right axis and rotated 2/n. The 2-dimensional CW-complex
X
mn
1) built in this way is the universal covering space; it is the product T
mn
R of a tree T
mn
and the real
line, hence contractible [10, Chp 3, Sec 7, Lemma 1]. The element a G
mn
acts by skew motion around one
of the vertical lines in X
m
1) and b G
mn
acts by skew motion around one of the vertical lines in X
n
1).
Note that a
m
= b
n
acts by translating one unit up. What is the lift of X
m
X
n
(the circle parallel to circle
a but passing through the point of the 1-skeleton) to the universal covering space?
What is the universal abelian covering space G
t
mn
X
mn
1) of X
mn
? Its deck transformation group is
G
t
mn
G
mn
= (G
mn
)
ab
= a, b[a
m
= b
n
, ab = ba) = Z Z/d
where d = (m, n) is the greatest common divisor. What is the mn fold covering space with funda-
mental group equal to the normal closure N of

a
m
, aba
1
b
1
_
and deck transformation group NG =
a, b [ a
m
= b
n
, a
m
, ab = ba) = a, b [ a
m
, b
n
, ab = ba) = Z/m Z/n? What is the lift of X
m
X
n
to this
covering space?
7.14. Cayley tables, Cayley graphs, and Cayley complexes. [6, III.4] [3] For any group presentation
G = g

[ r

) there exists (Corollary 4.12) a 2-dimensional CW-complex

X
G\G
= D
0

g]
D
1

]
D
2
= (GGD
0
)

g]
(GGD
1
)

]
(GGD
2
)
with fundamental group
1
(X
G\G
) = g

[ r

) = G. This is the most simple space with fundamental group

G so it is natural to apply Theorem 7.8 to X
G\G
. So what are the connected covering spaces of X
G
? There
is an equivalence of categories
X
?
: O
G
Cov
0
(X
G\G
): HG (X
H\G
X
G\G
)
24 J.M. MLLER
and the Cayley complex of HG is the 2-dimensional CW-complex X
H\G
while the Cayley graph is its 1-
skeleton. We now dene these CW-complexes more explicitly for any object of O
G
(or for any right G-space
for that matter) relative to the given presentation of G.
The 0-skeleton of X
H\G
is the right G-set X
0
H\G
= HG; this is the bre of the covering map X
H\G

X
H\G
as a right G-space. The 1-skeleton of X
H\G
is the Cayley graph for HG, the 1-dimensional HN
G
(H)-
CW-complex
X
1
H\G
= (HGD
0
)

HgHgg]
(HGD
1
)
obtained from the 0-skeleton HG by attaching to each right coset Hg HG an arrow from Hg to Hgg

for each generator g

; note that we have no other choice since the loop g

in the base space lifts to a path

in the total space that goes from Hg in the bre HG to Hgg

in the bre. (The Cayley graph is simply a

graphical presentation of the Cayles table for Cayley table for group multiplication HG G HG.) In
this way, the Cayley table for HG is a [G: H[-fold covering space of the 1-skeleton
_
g]
S
1
of X
G
. The
Cayley graph is connected since each group element g is a product of the generators which means that there
is a sequence of arrows connecting the 0-cells He and Hg.
Next attach 2-cells at each Hg HG along the loop r

for each relation r

. Since the relation r

is a
factorization of the neutral element e in terms of the g

, it denes loops Hg Hgr

= Hg based at each
0-cell Hg in the Cayley graph X
1
H\G
. The resulting left HN
G
(H)-CW-complex
X
H\G
= (HGD
0
)

HgHgg]
(HGD
1
)

Hg
r

Hg]
(HGD
2
)
is the Cayley complex of HG. The Cayley complex is still connected for attaching 2-cells does not alter the
set of path components (Corollary 4.9). Clearly, every G-map H
1
G H
2
G extends to a covering map
X
H1\G
toX
H2\G
.
In particular, taking H = e to be the trivial group, the Cayley complex for the right G-set eG = G,
X
e]\G
= (GD
0
)

g]
(GD
1
)

]
(GD
2
)
is a 2-dimensional left G-CW-complex, the universal covering space of X
G\G
. The 0-skeleton is G, at each
g G there is an arrow from g to gg

for each generator g

and a 2-cell attached by the loop g

r

gr

= g.
In other words, there is one 0-G-cell G D
0
, one G-1-cell G D
1
for each generator g

, attached by the
left G-map that takes e D
1
= e 0, 1 to e and g

, and one G-2-cell G D

2
for each relation r

attached by the left G-map that on e D

2
is the loop r

at e. The orbit space under the left action of

H < G on X
G
(eG) is the Cayley complex for the orbit space HG: HX
e]\G
= X
H\G
. In particular,
GX
e]\G
= X
G\G
= D
0

g]
D
1

]
D
2
= X
G
is a point Ge with an arrow Ge
g
Ge for each generator g

and with one 2-cell attached along the loop

Ge
r

Ge for each relation r

.
It is very instructive to do a few examples. See [4] for information about graph theory.
Example 7.15 (Cayley complexes for cyclic groups). For the innite cyclic group G = C

= a), X
e]\G
=
Z (Z D
1
) = R and X
G\G
= GR = S
1
. For the cyclic group G = C
2
= a [ a
2
) of order 2, X
e]\G
=
(C
2
D
0
) (C
2
D
1
) (C
2
D
2
) = S
2
and X
G\G
= GS
2
= RP
2
. For the cyclic group G = C
m
= a [ a
m
)
of order m, X
e]\G
is a circle with C
m
D
2
attached and and X
G\G
= GX
e]\G
is the mapping cone for
S
1
m
S
1
.
Example 7.16 (Cayley graphs for F
2
-sets). Let G = a, b) = Z H Z be a free group F
2
on two generators.
Then X
G\G
= S
1
S
1
and O
G
= Cov
0
(S
1
S
1
). Since there are no relations, Cayley complexes for right
THE FUNDAMENTAL GROUP AND COVERING SPACES 25
G-sets are Cayley graphs. In particular, X
e]\G
= (GD
0
) (GD
1

GD
1
) is the G-graph
gb
ga
1
g ga
gb
with vertex set G and two edges from g to ga and gb for every vertex g G, and X
G\G
= GX
e]\G
is the
graph, S
1
S
1
, with one vertex GG and two edges. In general, for any subgroup H of G, the Cayley graph,
X
H\G
, for HG is the covering space of X
G\G
= S
1
S
1
characterized by any of these three properties:
X
H\G
is the Cayley table for HGG HG relative to the generators a and b
the bre of X
H\G
X
G\G
is the right G-set HG
the image of the monomorphism
1
(X
H\G
)
1
(X
G\G
) = G is conjugate to H
Here are some examples:
If H =

a
2
, ab, b
2
_
then the Cayley table and the Cayley graph of the G-set HG = He, Ha are
a b
He Ha Ha
Ha He He
He
a
b

Ha
b

because He
a
Ha, He
b
Hb = Habb
2
b = Ha, Ha
a
Haa = He, and Ha
b
Hab = He. The
subgroup H is normal since it has index two. Note that H is free of rank 3.
If H = [G, G] is the commutator subgroup of G then the Cayley graph gives a tiling of the the plane
by squares with edges labeled aba
1
b
1
.
If H = G
2
is the smallest subgroup containing all squares in G, the right cosets are HG =
He, Ha, Hb, Hab and the Cayley graph is the graph of the Cayley table.
If H is the smallest normal subgroup containing a
3
, b
3
, and (ab)
3
, then the Cayley graph gives a
tiling of the plane by hexagons, with edges ababab, and triangles with edges aaa or bbb. Observe that
Hxa
3
y = Hxa
3
x
1
xa
3
y = Hxy.
It is, in general, a dicult problem to enumerate the cosets of H in G.
Exercise 7.17. [5, (3) p 58] Let G =< a, b >be the free group on two generators and H =

a
2
, b
2
, aba
1
, bab
1
_
.
Draw the Cayley graph for HG with the help of the information provided by this magma session:
> G<a,b>:=FreeGroup(2);
> H:=sub<G|a^2, b^2, a*b*a^-1, b*a*b^-1>;
> Index(G,H);
3
> T,f:=RightTransversal(G,H);
> T;
{@ Id(G), a, b @} //The vertices of the Cayley graph
> E:={@ <v,(v*a)@f,(v*b)@f> : v in T @};
> E;
{@ <Id(G), a, b>, <a, Id(G), a>, <b, b, Id(G)> @} //The edges
>
Exercise 7.18. Let G be a free group of nite rank and H a subgroup of G. Show that H is free and that
[G: H[(rk(G) 1) = rk(H) 1. (This exercise is most easily solved by using the Euler characteristic.)
Example 7.19. When G = C
m
= g [ g
m
) is the cyclic group of order m > 0, the Cayley complex
X
G
(eG) = (GD
0
) (GD
1
) (GD
2
)
is the universal covering space of the mapping cone for the degree m map on the circle. It is the left G-
CW-complex consisting of a circle with m 2-discs attached. (When m = 2, this is the 2-sphere which is the
26 J.M. MLLER
Figure 5. HX
e]\G
for H =

(ab)
3
, a
_

a, b [ a
2
, b
2
_
= G
universal covering space of the mapping cone RP
2
for the degree 2-map of the circle.) What is the covering
space action of G on X
G
(eG)?
Example 7.20. [5, Example 1.48, Exercise 1.3.14] Let G =

a, b [ a
2
, b
2
_
= Z/2HZ/2
4.2
= ZZ/2 be the free
product of Z/2 with itself. Then X
G\G
= RP
2
RP
2
and Cov
0
(RP
2
RP
2
) = O
C2IC2
. The total space
X
e]\G
of its universal covering space X
e]\G
X
G\G
is an innite string of S
2
s. Indeed, the 0-skeleton is
G, the 1-skeleton obtained by attaching two 1-discs to each 0-cell, is
ba
a
a
b
b
b
e
a
a
a
b
b
ab
and the 2-skeleton is obtained by attaching two 2-discs at each 0-cell along the maps a
2
and b
2
. The left
action of a G which swaps e a, b ab, etc is the antipodal map on the sphere containing e and a.
The subgroup H =

(ab)
3
_
= 3Z Z is normal in G so that the orbit set
HG = He, Ha, Hb, Hab, Hba, Haba = 3ZZ 2ZZ
is actually a group; it is the dihedral group of order 6, isomorphic to
3
. The quotient space HX
e]\G
=
X
H\G
is a necklace of six S
2
s formed from the 1-skeleton
He
a

H
H
H
H
H
Hb
b

T
T
T
T
T
T
T
T
T Ha
a

Hba
a

H
H
H
H
H
Hab
b

Haba
b

by attaching 2-discs at each vertex along the loops a

2
and b
2
. The fundamental group of H

X
G
is H and the
deck transformation group is HN
G
(H) = HG since H is normal. The dashed arrows show the covering
space left action by Ha HG; the orbit space for this action is the Cayley complex of the next example.
The element Hab HG acts by rotating the graph two places in clockwise direction.
For another example, take H =

(ab)
3
, a
_
= 3Z Z/2; H is not normal for N
G
(H) = H, HG =
He, Hb, Hba has 3 elements, and
He
b

Hb
b

Hba
a

is the Cayley graph for HG. The Cayley complex, obtained by attaching six 2-discs along the maps a
2
and b
2
at each vertex, is RP
2
, S
2
, S
2
, RP
2
on a string as shown above. The deck transformation group
HN
G
(H) = HH is trivial.
THE FUNDAMENTAL GROUP AND COVERING SPACES 27
Example 7.21. Let G = Z/2 H Z/3
4.2
= PSL(2, Z) be the free product of a cyclic group of order two and a
cyclic group of order three. This graph
b
b
b e
a
a
a
b
2
b
is the beginning of the Cayley complex for G. Describe the left G-CW-complex X
G
(G)!
7.22. Normal covering maps. Let p: Y X be a covering map between path connected spaces.
Denition 7.23. The covering map p: Y X is normal if the group Cov(X)(Y, Y ) of deck transformations
acts transitively on the bre p
1
(x) over some point of X.
If the action is transitive at some point, then it is transitive at all points. Why are these covering maps
called normal covering maps?
Corollary 7.24. Let X be a path connected, locally path connected and semi-locally simply connected space
and p: Y X a covering map with Y path connected. Then
The covering map p: Y X is normal The subgroup
1
(Y ) is normal in
1
(X)
Proof. The action HN
G
(H) HG HG of the group of covering maps on the bre is transitive i and
only if H is normal in G.
All double covering maps are normal since all index two subgroups are normal.
7.25. Sections in covering maps. A section of a covering p: E X is a (continuous) map s: X E such
that s(x) lies above x, ps(x) = x, for all x X. In other words, a section is a lift of the identity map of the
base space. Each section traces out a copy of the base space in the total space (and that is why it is called a
section).
Lemma 7.26. Let p: E X be a covering space over a connected, locally path connected and semi-locally
simply connected base space X. Then the evaluation map s s(x)
sections of p: E X

p
1
(x)
1(X,x)
is a bijection.
Proof. Since X is connected, sections are determined by their value at a single point (2.12), so the map is
injective. It is also surjective because any
1
(X, x)-invariant point corresponds (under the classication of
covering spaces over X) to the trivial covering map X X which obviously has a section.
In fact, E contains the trivial covering p
1
(x)
1(X,x)
X as a subcovering.
If either of Y
1
X or Y
2
X is normal, then
Cov(X)(Y
1
, Y
2
) =
_

1
(Y
2
)
1
(X)
1
(Y
2
)
1
(Y
1
)
otherwise
for the transporter N
1(X)
(
1
(Y
1
),
1
(Y
2
)) equals
1
(X) if
1
(Y
1
)
1
(Y
2
) and otherwise.
8. Universal covering spaces of topological groups
Suppose that G is a connected, locally path connected, and semi-locally simply connected topological
group (for instance, a connected Lie group) and let G1) be the universal covering space (7.4) of G. We can
use the group multiplication in G to dene a multiplication in G1) simply by letting the product [] [] of
two homotopy classes of paths [], [] G1) equal the homotopy class [ ] G1) of the product path
( )(t) = (t) (t) whose value at any time t is the product of the values (t) G and (t) G.
28 J.M. MLLER
Lemma 8.1. G1) is a topological group and G1) G is a morphism of topological groups whose kernel
is the subgroup [] [ (0) = (1) =
1
(G, e) of homotopy classes of loops based at the unit e G.
The set
1
(G, e) is here equipped with the group structure it inherits from G1) where multiplication of
paths is induced from group multiplication in G. However, we have also dened a group structure on
1
(G, e)
using composition of loops. It turns out that these two structures are identical.
Lemma 8.2. Let
1
and
2
be two loops in G based at the unit element e. Then the loops
1

2
(group
multiplication) and
1

2
(loop composition) are homotopic loops.
Proof. There is a homotopy commutative diagram
S
1

N
N
N
N
N
N
N
N
N
N
N
N

S
1
S
1

_

12

G G 
_

M
M
M
M
M
M
M
M
M
M
M
M
S
1
S
1
12

GG

G
where is the diagonal and the folding map. The loop dened by the top edge from S
1
to G is the
composite loop
1

2
and the loop dened by the bottom edge is the product loop
1

2
.
One can also show that in this situation
1
(G, e) must be abelian.
Let H = R1 Ri Rj Rk be the quaternion algebra where the rules i
2
= j
2
= k
2
= 1, ij = k = ji,
jk = i = kj, ki = j = ki dene the multiplication. Let Sp(1) denote the topological group of quaternions
of norm 1.
Sp(1) acts in a norm preserving way on the real vector space H = R
4
by the rule v = v
1
for
all Sp(1) and v R
4
= H. This give a homomorphism : Sp(1) SO(4). Since R1 is invariant
under this action, it takes R

= Ri Rj Rk = R
3
to itself, so there is also a group homomorphism
: Sp(1) SO(3) [2, I.6.18, p 88]. The kernel is R Sp(1) = 1. Convince yourself that is surjective
(see the computation below and recall that an element of SO(3) is a rotation around a xed line), so that
: Sp(1) SO(3) = 1Sp(1) is a double covering space.
Let
R() =
_
cos sin
sin cos
_
be the matrix for rotation through angle .
Lemma 8.3. The map : Sp(1) SO(3) is the universal covering map of SO(3). The fundamental group

1
(SO(3), E) = 1 is generated by the loop
(t) = (t) =
_
R(2t) 0
0 1
_
, 0 t 1,
Proof. The topological space Sp(1) = S
3
is simply connected, so Sp(1) SO(3) is the universal covering
space of SO(3). (We have seen this double covering before: It is the double covering S
3
RP
3
.)
The fundamental group
1
(SO(3), E) = C
2
is generated by the image loop (t) = (t) of a path (t) in
Sp(1) from +1 to 1. If we take
(t) = cos(t) + sin(t)k, 0 t 1,
then the image in SO(3) is the loop
(t) = (t) =
_
R(2t) 0
0 1
_
, 0 t 1,
This follows from the computation
(t)i(t)
1
= (cos(t) +k sin(t))i((cos(t) k sin(t))
= cos
2
(t)i + cos(t) sin(t)j + cos(t) sin(t)j sin
2
(t)i = cos(2t)i + sin(2t)j
and similarly for (t)j(t)
1
= sin(2t)i + cos(2t)j and (t)k(t)
1
= k.
THE FUNDAMENTAL GROUP AND COVERING SPACES 29
It is also known that the inclusion SO(3) SO(n) induces an isomorphism on
1
for n 3. We conclude
that the fundamental group
1
(SO(n), E) has order two for all n 3 and that it is generated by the loop
(t) in SO(n). Thus the topological groups SO(n), n 3, have universal double covering spaces that are
topological groups.
Denition 8.4. For n 3, Spin(n) = SO(n) 1) is the universal covering space of SO(n) and : Spin(n) SO(n)
is the universal covering map.
The elements of Spin(n) are homotopy classes of paths in SO(n) starting at E and, in particular, Spin(3) =
Sp(1). The kernel of the homomorphism : Spin(n) SO(n) is e, z where e is the unit element and z = []
is the homotopy class of the loop .
Proposition 8.5. The center of Spin(n) is
Z(Spin(n)) =
_

_
C
2
= e, z n odd
C
2
C
2
= e, x e, z n 0 mod 4
C
4
= e, x, x
2
, x
3
n 2 mod 4
for n 3.
Proof. From Lie group theory we know that the center of Spin(n) is the inverse image of the center of SO(n).
Thus the center of Spin(n) has order 2 when n is odd and order 4 when n is even.
Suppose that n = 2m is even. Then Z(Spin(2m)) = e, z, x, zx where x = [] is the homotopy class of
the path
(t) = diag(R(t), . . . , R(t))
from E to E. Note that x
2
is (8.2) represented by the loop
(t)
2
= diag(R(2t), . . . , R(2t))
Conjugation with a permutation matrix from SO(2n) takes
(t) = diag(R(2t), E, E, . . . , E) to diag(E, R(2t), E, . . . , E)
and since inner automorphisms are based homotopic to identity maps, both the above loops represent the
generating loop . It follows that
x
2
= [(t)
2
] = [(t)
m
] = z
m
=
_
e m even
z m odd
Thus Z(Spin(2m)) = z x = C
2
C
2
if m is even and Z(Spin(2m)) = x = C
4
if m is odd.
What is the fundamental group
1
(PSO(2n)) of the topological group PSO(2n) = SO(2n)/ E)?
When will two diagonal matrices in SO(n) commute in Spin(n)? Let D = diag(1, . . . , 1 be the
abelian subgroup of diagonal matrices in SO(n). By computing commutators and squares in Spin(n) we
obtain functions
[ , ] : D D e, z, q : D e, z
given by q(d) = (

d)
2
and [d
1
, d
2
] = [

d
1
,

d
2
] where (

d) = d, (

d
1
) = d
1
, (

d
2
) = d
2
. They are related by
formula
q(d
1
+d
2
) = q(d
1
) +q(d
2
) + [d
1
, d
2
]
which says that [ , ] records the deviation from q being a group homomorphism (using additive notation
here). It suces to compute q in order to answer the question about commutativity relations.
Proposition 8.6. q(d) = e i the number of negative entries in the diagonal matrix d D is divisible by 4.
[d
1
, d
2
] = e i the number of entries that are negative in both d
1
and d
2
is even.
Proof. Note that two elements of D are conjugate i they have the same number of negative entries. Use
permutation matrices and, if necessary, the matrix diag(1, 1, . . . , 1). Consider for instance
d
1
= diag(1, 1, 1, . . . , 1), d
2
= diag(1, 1, 1, 1, 1, . . . , 1)
with two, respectively four, negative entries. The paths

d
1
(t) = diag(R(t), 1, . . . , 1),

d
2
(t) = diag(R(t), R(t), 1, . . . , 1)
30 J.M. MLLER
represent lifts of d
1
and d
2
to Spin(n). Then (

d
1
)
2
= z = q(d
1
) and (

d
2
)
2
= e = q(d
2
). Computations like
these prove the formula for q and the formula for [ , ] follows. The number of negative entries in d
1
+ d
2
is the number of negative entries in d
1
plus the number of negative entries in d
2
minus twice the number of
entries that are negative in both d
1
and d
2
.
Exercise 8.7. Let D
n
Spin(n) be the inverse image of D SO(n). How many elements of order 4 are
there in D
n
? Can you identify the group D
n
? Show that there is a homomorphism SU(m) Spin(2m).
When m is even, what is the image of E SU(m)? What is the image of the center of SU(m)? Describe
the covering spaces of U(n).
The inclusions SO(n) SO(n + 1), n > 2, and SO(m) SO(n) SO(m + n), m, n > 2, of special
orthogonal groups lift to inclusions
Spin(n)

Spin(n + 1)

Spin(m)
(z1,z2))
Spin(n)

Spin(m+n)

SO(n)

SO(n + 1) SO(m) SO(n)

SO(m+n)
of double coverings. (Here, Spin(m)
(z1,z2))
Spin(n) stands for (z
1
, z
2
)) (Spin(m) Spin(n))).
The inclusion U(n) SO(2n), that comes from the identication C
n
= R
2n
, lifts to an inclusion of double
covering spaces as shown in the following diagrams.
SU(n)
C
k
U(1)
(A,z)(A,z)

Spin(2n)

SU(n)
C

n
U(1)
(A,z)(A,z
2
)

Spin(2n)

SU(n)
Cn
U(1) = U(n)

SO(2n) SU(n)
Cn
U(1) = U(n)

SO(2n)
To the left, n = 2k is even, and to the right, n = 2k + 1 is odd; C
n
= (E,
1
) [
n
= 1 and C
t
n
=
(E,
k
) [
n
= 1 are cyclic groups of order n and C
k
= (E,
1
) [
k
= 1 C
2k
= C
n
is cyclic of order
k. The isomorphism SU(n)
Cn
U(1) U(n) takes (A, z) to zA. When n is divisible by 4, z = (E, 1)
and x = (E, 1) have order two; when n is even and not divisible by 4, x = (E, i) has order four and
x
2
= (E, 1) = z. This explains the computation of the center of Spin(2n). (Is the group in the upper left
corner of the right diagram isomorphic to U(n)? See [1] for more information.)
There is a double covering map pin(n) O(n) obtained as the pullback of Spin(2n) SO(2n) along the
inclusion homomorphism O(n) SO(2n).
Example 8.8. The inclusion U(2) SO(4) lifts to an inclusion SU(2) U(1) Spin(4). Let G
16

SU(2) U(1) Spin(4) be the group
G
16
=
___
i 0
0 i
_
, i
_
,
__
0 i
i 0
_
, i
__
G
16
has order 16, center Z(G
16
) = x, z = C
2
C
2
= Z(Spin(4)), and derived group [G
16
, G
16
] = xz = C
2
.
Its image under the covering maps
Spin(4)/ z) = SO(4)

V
V
V
V
V
V
V
V
V
V
V
V
V
V
V
V
V
Spin(4)

l
l
l
l
l
l
l
l
l
l
l
l
l

R
R
R
R
R
R
R
R
R
R
R
R
R
Spin(4)/ xz)

Spin(4)/ x, z) = PSO(4)
Spin(4)/ x) = SSpin(4)

h
h
h
h
h
h
h
h
h
h
h
h
h
h
h
h
h
is dihedral D
8
in SO(4), abelian C
4
C
2
in Spin(4)/ xz), quaternion Q
8
in the semi-spin group SSpin(4) =
Spin(4)/ x), and elementary abelian C
2
C
2
in PSO(4). (All proper subgroups of G
16
are abelian but itself
and some of its quotient groups are nonabelian.)
THE FUNDAMENTAL GROUP AND COVERING SPACES 31
Example 8.9. There exists a covering space homomorphisms of topological groups
U(1) SU(n) U(n): (z, A) zA
with kernel C
n
= (z, z
1
E) [ z
n
= 1 =

(,
1
E)
_
where
n
= e
2i/n
. The universal covering space
homomorphism is R SU(n) U(n): (t, A)
t
n
A with kernel C

(1,
1
n
E)
_
. Any covering space of
U(n) is of the form

(k,
k
n
)
_
(RSU(n)) for some integer k 0.
Similarly, let S(U(m) U(n)) denote the closed topological subgroup (U(m) U(n)) SU(m + n) of
U(m+n). There exists a covering space homomorphisms of topological groups
U(1) U(m) U(n) S(U(m) U(n)): (z, A, B) diag(z
n1
A, z
m1
B)
with kernel C
lcm(m,n)
= (z, z
n1
E, z
m1
E) [ z
lcm(m,n)
= 1 =

(
lcm(m,n)
,
1
m
,
n
)
_
where m
1
= m/ gcd(m, n) =
lcm(m, n)/n and n
1
= n/ gcd(m, n) = lcm(m, n)/m. The universal covering space homomorphism of
S(U(m) U(n)) is R SU(m) SU(n) S(U(m) U(n)): (t, A, B) (
t
m
A,
t
n
B) with kernel C

(1,
1
m
,
n
)
_
. Any covering space of S(U(m) U(n) is of the form

(k,
k
m
,
k
n
)
_
(RSU(m) SU(n)) for
some integer k 0.
All nite covering spaces of U(n) are covered by U(1) SU(n). To see this, let n and k be integers and
put k
1
= k/ gcd(n, k). Then there is a commutative diagram
U(1) SU(n)

(z,A)(z
k
1
,A)

R
(k,
k
))
SU(n)

U(1) SU(n)
(
t
n
,A)(t,A)

R
(1,
1
))
SU(n) = U(n)
of covering space homomorphisms.
References
[1] P.F. Baum, Local isomorphism of compact connected Lie groups, Pacic J. Math. 22 (1967), 197204.
[2] Theodor Brocker and Tammo tom Dieck, Representations of compact Lie groups, Graduate Texts in Mathematics, vol. 98,
Springer-Verlag, New York, 1985. MR 86i:22023
[3] Kenneth S. Brown, Cohomology of groups, Graduate Texts in Mathematics, vol. 87, Springer-Verlag, New York, 1982.
MR 83k:20002
[4] Reinhard Diestel, Graph theory, third ed., Graduate Texts in Mathematics, vol. 173, Springer-Verlag, Berlin, 2005.
MR MR2159259 (2006e:05001)
[5] Allen Hatcher, Algebraic topology, Cambridge University Press, Cambridge, 2002. MR 2002k:55001
[6] Roger C. Lyndon and Paul E. Schupp, Combinatorial group theory, Springer-Verlag, Berlin, 1977, Ergebnisse der Mathe-
matik und ihrer Grenzgebiete, Band 89. MR 58 #28182
[7] Saunders Mac Lane, Categories for the working mathematician, second ed., Graduate Texts in Mathematics, vol. 5, Springer-
Verlag, New York, 1998. MR 2001j:18001
[8] A. D. Mednykh, On the number of subgroups in the fundamental group of a closed surface, Comm. Algebra 16 (1988),
no. 10, 21372148. MR 90a:20076
[9] Derek J. S. Robinson, A course in the theory of groups, second ed., Springer-Verlag, New York, 1996. MR 96f:20001
[10] Edwin H. Spanier, Algebraic topology, Springer-Verlag, New York, 1981, Corrected reprint. MR 83i:55001
Matematisk Institut, Universitetsparken 5, DK2100 Kbenhavn
E-mail address: moller@[Link]
URL: [Link]
~
moller