Sketches of solutions to selected exercises

Note: these are intended as sample solutions. There will often be alternative solutions
to problems. Furthermore, solutions presented here are not intended to be 100% complete
but rather to demonstrate the idea of the problem. If the solution is not clear to you, please
come ask me about it!
Due April 24
Let M be the closed Mobius strip (i.e. including its boundary). Let M be
the boundary of M . Compute H (M, M ) using arguments with the long exact
sequence and other basic properties of homology. You can use that you already
know homology computations for circles, but you shouldnt compute H (M, M )
directly from the Delta complex. If necessary, you can also use that H (X) =
H (X) if X can be realized as a Delta complex in order to compute H (M ) and
H (M ).
To get at H (M, M ), we first compute H (M ) and H (M ). The boundary
A is homeomorphic to a circle, so, using what we know about circles (or spheres in general),
H1 (M ) = Z, and H0 (M ) = Z. Well also have Hi (M ) = 0 for i > 1. Similarly, M is
homotopy equivalent to a circle, so we have H0 (M )
= H1 (M )
= Z and all others are 0. So
far we have the long exact sequence

g
H2 (M, M ) - H1 (M ) - H1 (M )
=

=
?

H2 (M, M )

?
-

f
H1 (M, M ) - H0 (M ) - H0 (M )
=

=
?

?
-

H1 (M, M )

?
-

H0 (M, M )

Note that as all Hi (M ) and Hi (M ) are 0 for i > 1, it is immediate that Hi (M, M ) = 0
for i > 2.
So what are the maps? H0 (M ) = Z is generated by any singular 0-simplex, and this
maps to a generator of H0 (M ) = Z, so the map H0 (M ) H0 (M ) is an isomorphism and
H0 (M, M ) must be 0. We can also see this directly as any 0 simplex in M is homologous
via a path to a 0-simplex in M .
Similarly, H1 (M ) = Z is generated by a cycle that goes once around the circle M .
This can most easily be seen using simplicial homology. Furthermore, the inclusion of M
to M followed by the homotopy equivalence of M to the circle provides a map that wraps
around the circle twice. So the map H1 (M )
= Z H1 (M )
= Z is multiplication by 2. It

now follows that H1 (M, M ) = Z2 , using also that the map H1 (M ) H1 (M, M ) must be
surjective because the map H0 (M ) H0 (M ) is injective from above.
Finally, as H1 (M ) H1 (M ) is injective, this forces H2 (M, M ) to be 0.
Recall that RP 2 can be obtained from the Mobius strip by filling in the boundary
circle with a disk. Using this and excision, compute H (RP 2 ). By excision, we
1

?
-

H0 (M, M )

 (RP 2 ) using
have H (M, M )
= H (RP 2 , disk), which is isomorphic to H (RP 2 , pt)
= H
homotopy equivalence and our previous result about reduced homology. Putting this all
(RP 2 )
together, H
= H (M, M ) is 0 in all dimensions except when = 1, in which case it
2
is Z2 . So H1 (RP ) = Z2 , H0 (RP 2 )
= Z, and all other homology groups are 0.
Hatcher 2.1.29 Let T = S 1 S 1 . Let X = S 1 S 1 S 2 and x0 be the union point. By
Example 2.3, H2 (T ) = Z, H1 (T ) = Z Z, H0 (T ) = Z and all other homology groups of
(X)
(S 1 ) H
(S 1 ) H
(S 2 ). Since we know
T are 0. By Corollary 2.25, we have H
=H
i (S n ) = Z if i = n and 0 otherwise, we see that H (X)
H
= H (T ) (using also basic facts
about the relation between ordinary and reduced homology).
The universal cover of T is R2 , which is contractible, so H2 (T) = 0. But the universal
cover of X is like the antenna space over S 1 S 1 but with an S 2 attached at every vertex.
This is homotopy equivalent to a wedge of a (countably) infinite number of S 2 s (one for each
2 (Sa2 )
2 (X)

word in a, b, a1 , b1 ). So H
= a Z 6= 0.
= a H
Hatcher 2.2.2 Suppose for all x, f (x) 6= x and f (x) 6= x, then gt (x) =

(1t)x+tf (x)
|(1t)x+tf (x)|

is

(1t)(x)+tf (x)
|(1t)(x)+tf (x)|

a homotopy from f to the identity and gt (x) =

is a homotopy from f to the
antipodal map. So the identity and antipodal map are homotopic. But the identity has
degree 1 and the antipodal map has here degree 1 because S 2n is an even sphere. So this
is a contradiction, so there must be an x that gets taken to x or x by f .
Now RP 2n is covered by S 2n . Condsider the following diagram
S 2n

f - 2n
S

p
?

RP 2n

f - ?
RP 2n

The map f exists by applying the lifting theorem to f p. By the first part of the theorem,
some x S 2n gets taken to x or x, so f p(x) = p(x).

0
1
Finally consider the map f : R2n R2n given by
in blocks along the diagonal
1 0
and 0 elsewhere. This takes lines to lines, and so induces a map RP 2n RP 2n . But since
there are no real eigenvectors, there are no lines fixed by this.

Due April 17
Let A be the closed annulus, i.e. A = {z C | 1 |z| 2}. Let A be the boundary
of A. Compute H (A, A) using arguments with the long exact sequence and other
basic properties of homology. You can use that you already know homology
computations for circles, but you shouldnt compute H (A, A) directly from the
Delta complex. If necessary, you can also use that H (X) = H (X) if X can be
realized as a Delta complex in order to compute H (A) and H (A) .
To get at H (A, A), we first compute H (A) and H (A). The boundary A is the
disjoint union of two circles, so, using what we know about circles (or spheres in general),
H1 (A) = Z Z, and H0 (A) = Z Z. Well also have Hi (A) = 0 for i > 1. Similarly, A
is homotopy equivalent to a circle, so we have H0 (A)
= H1 (A)
= Z and all others are 0. So
far we have the long exact sequence

H2 (A, A)

H1 (A)

=
-

H2 (A, A)

H1 (A)

=
?

g-

ZZ

H1 (A, A)

H0 (A)

f-

H1 (A, A)

H0 (A)

=
?

?
-

ZZ

H0 (A, A)

Note that as all Hi (A) and Hi (A) are 0 for i > 1, it is immediate that Hi (A, A) = 0
for i > 2.
So what are the maps? H0 (A) = ZZ is generated by one 0-simplex in each component.
These each map to a generator of H0 (A) = Z, so the map f : Z Z Z is (a, b) a + b.
Therefore, it is onto, so H0 (A, A) = 0. We can also see this directly as any 0 simplex in A
is homologous via a path to a 0-simplex in A.
Similarly, H1 (A) = Z Z is generated by a cycle that goes once around each circle.
This can most easily be seen using simplicial homology. Furthermore, as each inclusion of
a boundary circle into A is a homotopy equivalence, each of these inclusions induces an
isomorphism. This implies that g : Z Z Z is also given by (a, b) a + b. So this is
also surjective, and its kernel must be H2 (A, A). This is easily seen to be isomorphic to Z,
generated by (1, 1).
That leaves H1 (A, A). As g is surjective, H1 (A, A) must inject into H0 (A), and, in
fact, it is the kernel of f . Again, this is Z generated by (1, 1).
So, H2 (A, A)
= H1 (A, A)
= Z and H0 (A, A) = 0.
Find generators for the homology groups Hi (A, A) you computed above. For
each of H1 (A, A) and H2 (A, A), we need generators that map to (1, 1) under , where
here the 1 and 1 are with respect to the generators of H0 (A) or H1 (A) observed above.
So for H1 (A, A), we can use a 1-simplex that runs from one boundary component to the
other. For H2 (A, A) we can use any 2-chain that triangulates A as a Delta-complex; in
3

?
-

H0 (A, A)

other words, form such a complex and then take the sum over all the 2-simplices, oriented
compatibly.

Due April 10
f

Let A
B
C
D
E be an exact sequence. Suppose f is surjective and g is
injective. Show that C = 0. Provide all the details. As g is injective, ker(g) = 0, so
im(k) = 0, so ker(k) = C.
On the other hand, as f is surjective, im(f ) = B, so ker(h) = im(f ) = B. This implies
that im(h) = 0.
But im(h) = ker(k), so C = 0.
Recall that if f : G H is a map of abelian groups, then the cokernel, cok(f ),
is defined to be H/im(f ). Suppose you have the following commutative diagram
of abelian groups in which the horizontal sequences are exact:
0

g
-

?
-

The serpent lemma says that there is an exact sequence

0 ker() ker() ker() cok() cok() cok() 0.
Prove the serpent lemma. Hint: write the kernels and cokernels as homology
groups and use some big results you learned in class. Notice that the map A D

can be included in a chain complex 0 A

D 0 . You can check this
is indeed a chain complex. If we think of D as being in degree 0, then H0 = cok() and
H1 = ker(). Similarly for the other groups. The exact sequence of the serpent lemma is then
just the long exact homology sequence from the short exact sequence of chain complexes.
Hatcher 2.1.16 a) This could be done directly but lets use the exact sequence. First,
notice that if X has one component and A is not empty, then a 0-chain generating H0 (A)
also generates H0 (X). So H0 (A) H0 (X) is onto and H0 (X, A) is 0 from the long exact
sequence. More generally, suppose X has multiple connected components and that A intersects each component. If Xi is a component of X, then H0 (A Xi ) H0 (Xi ) is surjective
by the preceding argument. But then H0 (A) = i H0 (A Xi ) i H0 (Xi ) = H0 (X) is
surjective. So H0 (X, A) = 0.
b) If H1 (X, A) = 0, then H1 (A) H1 (X) is onto immediately by the long exact sequence
and H0 (A) H0 (X) is injective. This last statement cant be true if some path component
Xi of X contains multiple components of A because then H0 (A Xi )
= Zn for some n 2
while H0 (Xi ) = Z. So then H0 (A Xi ) H0 (Xi ) cant be 1-1, and the same follows for
H0 (A) H0 (X).
5

Conversely. If H1 (A) H1 (X) is onto, then ker(p1 : H1 (X) H1 (X, A)) = H1 (X),
so the map p1 : H1 (X) H1 (X, A) is the 0 map. Similarly, if each component of X
contains at most one component of A, then H0 (A) H0 (X) is injective. So its kernel is
0, so the image of H1 (X, A) H0 (A) is 0. But then by the first isomorphism theorem,
0 = H0 (X, A)/ ker( ) = H0 (X, A)/im(p1 ) = H0 (X, A).
Hatcher 2.1.17.a (just the sphere for now) We know from previous computations
that H2 (S 2 )
= H0 (S 2 )
= Z, and otherwise Hi (S 2 ) = 0. Also if A consists of n points,
H0 (A)
= Zn and Hi (A) = 0 otherwise. So the long exact reduced homology sequence looks
like
0

2 (S 2 )
H
= Z - H2 (S 2 , A)
0

0 (A)
0 (X) = 0 - H0 (S 2 , A)
H1 (S 2 , A) - H
= Zn1 - H

2 (S 2 )
So immediately, we have H2 (S 2 , A)
= H
= Z, and H1 (S 2 , A) is isomorphic to
0 (A)
H
= Zn1 . We also must have H0 (S 2 , A) = 0.

Due April 3
i

Hatcher 2.1.11 Since A is a retract, there are maps A

X
A such that ri = idA .
Thus by functoriality, r i is the identity, so i : H (A) H (X) is injective.
Suppose H < G are abelian groups and G/H
= F is a free abelian group. Prove
that G
= H F . Hint: Let q : G G/H be the natural quotient map. First
construct a homomorphism : F G such that q = idF ; note: this implies that
: F G is injective, so (F )
= F . Then show that G
= H (F ) by showing that
H (F ) = {0} and that every element of G can be written as a sum h + f with
h H and f (F ). Double hint: it might help to notice that H is the kernel
of q. Triple hint: to construct f , use the maps! First we construct . As F is free,
it has generators, say {xa }. As q is surjective, there is an element (not necessarily unique)
ya G such that q(ya ) = xa . Define (xa ) = ya . Again, is not unique, but thats okay.
Then clearly q(xa ) = xa . As F is free abelian, defining on the generators determines
P
P
P
on all of F , and as q and are homomorphisms, q( a ca xa ) = a ca q(xa ) = a ca xa for
any ca Z and finite index set for the sum. So is as desired.
Now, let us first show that H (F ) = {0}. Suppose y H (F ). Then q(y) = 0
because H = ker q. But also, if y (F ), the y = (x) for some x and so 0 = q(y) =
q(x) = x. So x = 0 and y = (x) = 0.
Next, lets show that every element of G can be written as h + f with h H and
f (F ). Let y G. Notice y = y q(y) + q(y). Obviously q(y) (F ). But then
q(y q(y)) = q(y) q(q(y)) = q(y) q(y) = 0. so y q(y) H.
(X) H
(Y )
Show that if f : X Y is homotopic to a constant map, then f : H
is the 0 map . Since homotopic maps induce the same map on homology (and reduce
homology), we might as well assume f is the constant map. The we can write f = gh, where
h maps X to a point and g maps a point to the image point f (X). So then the image of f
g
h
(Y ). But H
(pt) = 0, so the result
(X)
H
is the image of the composition H
H (pt)
follows.
The topologists sine curve is the subspace of R2 consisting of all points (x, sin(1/x))
for 0 < x 1 and all points (0, y) for 1 y 1. Compute the singular homology
groups of this space. The space X has two path components, lets say X1 is the set of
points on the (x, sin(1/x)) piece and X2 the other piece. Each of these pieces is contractible.
So H (X) = H (X1 ) H (X2 ) and for each i, H0 (Xi ) = Z and Hj (Xi ) = 0 for j > 0. So
H0 (X) = Z Z and Hi (X) =) otherwise.
Suppose X is a path connected space such that 1 (X) is a non-abelian simple
group. Show that H1 (X) = 0. Since H1 (X) is the abelianization of 1 (X), we must show
that 1 (X) abelianizes to 0. The abelianization of a group is its quotient by its commutator
subgroup C. This is the normal subgroup generated by the commutators, which are elements
7

of the form aba1 b1 . Since C is normal by definition and since 1 (X) is simple, C is either
trivial or all of 1 (X). In the latter case, H1 (X)
= 1 (X)/1 (X) = 0. So we just have to
show that C is not trivial. But if C is trivial, then every commutator aba1 b1 is equal to
the identity e. So aba1 b1 = e, whence ab = ba. So C will be nontrivial as long as there
are two elements of 1 (X) that dont commute.
Hatcher 2.1.12 Equivalence relation:
1. Reflexivity . Suppose f : A B is a chain map. We need to show that f is chain
equivalent to f , i.e. that there is a P : A B+1 such that P + P = f f = 0. Just
take P = 0, and that will work.
2. Suppose f, g : A B are chain homotopic. So there is a P : A B+1 such that
P + P = f g. But now consider the map P . Then using linearity of all the maps,
(P ) + (P ) = g f . So g is chain homotopic to f .
3. Suppose f, g : A B are chain homotopic via P : A B+1 and that g, h : A
B are chain homotopic via Q : A B+1 . Then P + P = f g and Q + Q = g h.
So P + P + Q + Q = (P + Q) + (P + Q) = f h. So the relation is transitive.
Do Hatcher 2.1.13. From the proof of Theorem 2.10, there is a chain homotopy P :
C (X) C+1 (Y ) between f# and g# , ie. P + P = f# g# . If we want a chain
homotopy P : C (X) C+1 (Y ), we can try to let P = P on Ci (X) for i > 0. We only
need to be careful about what happens at the bottom, so lets have a closer look at what
P0 : C0 (X) C1 (Y ) and P1 : Z C0 (Y ) must be.
Note: Since we need f# = f# , we must define f# : C1 (X) = Z C1 (Y ) = Z to be
the identity on Z, and similarly for any chain map.
Lets first look at P1 . We need 0 P1 + P2 1 = f# g# . Notice that 1 = 0, 0 = ,
by definition, and f# g# = 0 at the degree 1. So we can just take P1 = 0.
For P0 , we need 1 P0 + P1 0 = f# g# . We have just defined P1 = 0, and from our
chain homotopy P , we already have a P0 such that 1 P0 + P1 0 = 1 P0 = f# g# (because
in the unreduced chain complex, 0 = 0). So we just use P0 = P0 , and from here we can just
let Pi = Pi for all i 0, and everything will be consistent by the properties of P .
The homology result follows from the chain homotopy result as in the unreduced case.

Due March 27
Hatcher 2.1.1 Its a Mobius strip. (The easiest way to see this is that the instructions
given an RP 2 with a hole cut out of it. Thats always a Mobius strip.)
Hatcher 2.10.a - note that he means that every edge is glued to precisely one
other edge. Let X be such a space. Every point of X lies in the interior of a vertex, an
edge, or a face. If it lies in the interior of a face, it clearly has a euclidean neighborhood. If
it lies on the interior of an edge, then we can find a euclidean neighborhood that looks like
a half disk in each of the two 2-simplices joined together along that edge. At a vertex, the
space looks locally like a pie. Since there are no free edges in the quotient, it must be a full
pie and we get a euclidean neighborhood by adjoining wedges from each of the 2-simplices
Find a way to realize the two-holed torus (what Hatcher calls M2 ) as a complex.
Find a Delta-complex X such that 1 (X) = Z3 .

See me for a picture.

Hatcher 2.1.4 We have 0 (X) = v , 1 (X) = e1 , e2 , e3 , 2 (X) = f .

2 f = e1 + e2 e3 . So ker 2 = 0, so H2 (X) = 0.
Weve seen im2 = e1 + e2 e3 . Also, 1 e1 = 1 e2 = 1 e3 = 0. So ker 1 = 1 (X).
So H1 (X) = e1 , e2 , e3 | e1 + e2 e3 = e1 , e2 
= Z Z.
We know im1 = 0 and clearly ker 0 = 0 (X) = v . So H0 (X)
= Z.
Hatcher 2.1.5 Using the notation from Hatchers picture, 0 (K) = v , 1 (K) =
a, b , and 2 (K) = U, L .
2 U = a + b c and 2 L = c + a b. Its easy then to check that ker 2 = 0, so
H2 (K) = 0.
Weve seen im2 = a + b c, a b + c . Also, 1 a = 1 b = 1 c = 0. So
ker 1 = 1 (K). So H1 (K) = a, b, c | a + b c, a b + c = a, b | a b + (a +
b) = a, b | 2a 
= Z2 Z.
We know im1 = 0 and clearly ker 0 = 0 (X) = v . So H0 (K)
= Z.

Hatcher 2.1.9 Here clearly each j (X) has one generator for each 0 i n. What
are the maps j . If d is the lone j-face (after identification), then lets represent d as
[v0 , . . . , vj ] (of course all the vi have also been identified, but we dont use that yet. Then
P
i [v0 , . . . , vj ] = (1)i [v0 , . . . , vi , . . . , vj ]. But now each of the [v0 , . . . , vi , . . . , vj ] have been
Pj
i
identified to each other. If we call the identified face f , this is
i=0 (1) f , which is 0
if j is odd and f if j is even. So each map 2k+1 is 0 and each map 2k is the identity
 d  f . We just have to be careful around j = 0 and j = n because there are no
1 or n + 1 simplices. Thinking about all these cases shows that H0 (X)
= Z, Hn (X)
=Z

if n is odd, and Hi (X) = 0 otherwise.

10

Due March 20
Note: the following exercise is essentially a special case of Hatchers 1.3.24, which
actually follows from the same sort of ideas as this by putting more spaces in
X is a covering space with X
path connected,
the middle. 1. Suppose p : X
is a normal cover
locally path connected, and simply connected. Show that X
over X is isomorphic to 1 (X).
and that the group of deck transformations for X
is trivial, so is its image in 1 (X). The trivial subgroup is normal, so the
Since 1 (X)
= 1 (X)/{e} = 1 (X).
covering is normal, and G(X)
over x0 X. Let H be a subgroup of 1 (X, x0 ) and
2. Let x0 be a basepoint of X
over X. Let
hence also a subgroup of the group of deck transformations of X

0 in XH , which we let be the basepoint of

XH = X/H,
and let xH
0 be the image of x
pH
q

xH . Notice that we can factor p as X XH

X = X/G. Show that q : XH X
is a covering space of X. For each x X, let Ux be a neighborhood such that p1 (Ux )
consists of homeomorphic copies of Ux . In fact, if Ux is one such copy, then all the other
copies will have the form g Ux , where g 1 (X, x0 ) and g Ux is the image of Ux under the
by identifying equivalence classes of the form {hy | h H}
action of g. XH is formed from X
So going from X
to XH identifies all of the hUx for h H. Thus q 1 (Ux )
for each y X.
is the quotient p1 (Ux )/H, which consists of a copy of Ux for each left coset gH.
3. Show that q (1 (XH , xH
Recall that q (1 (XH , xH
0 ) = H.
0 ) will be consist of those

loops in X that are images of loops in XH . Since X is simply connected, the loops in XH are
of the paths in X
starting at x0 and ending at points of the form h
the image from X
x0 for
h H. Thus there is a correspondence between elements of H and elements of 1 (XH , xH
0 ).
One then checks that this is a homomorphism.
4. Suppose H is normal. Show that G/H acts properly discontinuously on XH
(in particular, show that g1 x = g2 x if and only if g1 H = g2 H) and observe that
X = X/G = XH /(G/H).
(Clarification: recall that each point of XH corresponds to an orbit of a point x
under the action of H. In particular, every point in XH is an image pH (x)
in the X
Define an action of G on XH by letting g(pH (x)) = pH (gx). Part
for some x in X.
with
of your job is to show that this is well-define, i.e. if y is another point in X
pH (y) = pH (x), show that g(pH (x)) = g(pH (y)) so that this definition is consistent.
Then to show that we really have an action of G/H, show that every element
of the coset gH acts on pH (x) the same way. Lastly, you should argue that the
XH identifies points
action of G/H is properly discontinuous.) Recall that pH : X
that are in the same orbit of H. So let x X,
and let pH (
of X
x) be its image in xH . Define
gpH (
x) to be pH (g
x). To see that this is well-defined, suppose pH (y) = pH (x). Then y = hx
for some h H. But then gy = ghx, so gpH (y) = pH (gy) = pH (ghx) = gpH (hx) = gpH (x)
11

because x and hx are in the same orbit. So the map is well-defined. Now, to show that G/H
acts on XH , observe ghpH (x) = pH (ghx) = gpH (hx) = gpH (x). Therefore, each element of
the coset gH acts on XH in the same way, and we can consider this an action of GH on
XH . This action is properly discontinuous because if Ux is as above, then g takes pH (Ux ) to
pH (g Ux ), which is disjoint from Ux if g
/ H. For freeness, if g(pH (x)) = pH (x), that means
that pH (gx) = pH (x), which happens only if g H for some h H. But then gH represents
the identity element of G/H.
Recall from last weeks homework the spaces Xn , n {1, 2, 3, . . .} obtained from
the two-dimensional disk D2 by identifying points on the boundary that differ
by an angle 2/n; let q : D2 Xn be the quotient map. In that problem, you
n is
should have shown 1 (Xn )
= Zn and determined that the universal cover X
2
homeomorphic to n copies of D with their boundaries identified to each other.
n . The group of transformations
1. Describe how the deck transformations act on X
is Zn . The generator 1 acts by taking the ith copy of D2 to the (i + 1)st copy (the nth
copy goes to the first) and rotating it by an angle 2/n.
4 you would identify to get the quotient space of
2. Describe what points of X
4 corresponding to the unique subgroup of index 2 of 1 (X4 ). Each point on
X
disk i would be identified with its antipodal point on the disk i + 2 mod 4
Describe all connected covering spaces of RP 2 , up to equivalence. The covering
spaces are RP 2 and S 2 .
Describe all connected covering spaces of RP 2 RP 2 (Note: some of these will be
familiar, others you made need to describe as quotients of universal cover under
certain group actions). 1 (RP 2 RP 2 ) = Z2 Z2 , so the covering spaces correspond
to 0, Z2 0, 0 Z2 , and {0, (1, 1)}. The covering spaces corresponding to the first three of
these are S 2 S 2 , RP 2 S 2 and S 2 RP 2 . The last space is S 2 S 2 /(p, q) (p, q).
Let X = RP 2 RP 2 with basepoint the union point x0 .
of X?
1. Whats the universal cover X
Its a string of S 2 s with the north pole of one identified to the south pole of the next.
is simply-connected, its group of deck transformations is 1 (X, x0 )
2. Since X
=
2
2
Z2 Z2 . Let a, b be the respective generators of 1 (RP x0 ) and 1 (x0 RP )
as subgroups of 1 (X, x0 ). Describe the covering actions of a and b. Then
describe what a covering action by a general element of 1 (X, x0 ) would
look like. Can you see why all such transformations are generated by those
corresponding to a and b? (Note, while the group of deck transformations
is isomorphic to 1 (X, x0 ), the exact isomorphism might depend upon the

choice of basepoint in X.)

12

Take two consecutive spheres and label them A and B. a acts as the antipodal map
on sphere A and consequently flips the entire string of spheres. b acts similarly but
inverts B instead. A general element would flip the string of spheres, inverting one
them.
Due 3/6/2014
Describe all connected covering spaces of RP 2 , up to isomorphism. 1 (RP 2 ) = Z2
which has only one subgroup - the trivial subgroup. So the only cover is the universal cover,
which is S 2 .
Describe all connected covering spaces of RP 2 S 2 , up to isomorphism. This
should look like three spheres in a row.
Describe all connected covering spaces of S 1 S 2 , up to isomorphism. This should
look like R with a sphere attached at every integer.
Describe all connected covering spaces of S 1 S 2 , up to isomorphism. Since S 2
is simply connected, we need only take products of S 2 with the covering spaces of S 1 . The
trivial subgroup of 1 (S 1 ) = Z corresponds to the covering ei : R S 1 , while the subgroup
nZ corresponds to eni : S 1 S 1 . So the desired covers are ei id : R S 2 S 1 S 2 and
eni id : S 1 S 2 S 1 S 2 .
Let X = S 1 S 1 with 1 (X, x0 ) = ha, bi in the standard way. For each of the
following subgroups of 1 (X, x0 ), find (draw) a covering space (with basepoint)
x0 ) of X such that p (1 (X,
x0 )) is the given subgroup:
(X,
1. {(ab2 )an (ab2 )1 | n Z}
2. {(ab)n | n Z}
3. {b3 (ab)n b3 | n Z}
4. the subgroup freely generated by a2 and b
5. the subgroup freely generated by a3 , b, aba1 and a1 ba.
I cant draw these here. Come see me if you want to talk about them.

13

Due 2/28/2014
S 1 S 1 be the covering space shown in box 2 of Hatchers table of
Let p : X
examples of coverings of S 1 S 1 . Let f : S 1 S 1 S 1 S 1 be the map given
by the composition of the projection S 1 S 1 S 1 y0 followed by the map

S 1 y0 S 1 S 1 described by the loop a3 b3 . Does f lift to X?

Yes! The image
1
1
3 3
2
2
f in 1 (S S ) is the cyclic subgroup generated by a b = a (ab)b , which is a subgroup
of the image of p because a2 , b2 , ab are all in this image (by direct inspection or using the
information given in Hatchers table).
Use covering space theory to show that every map : RP 2 S 1 is homotopic to
a constant map. Since 1 (RP 2 ) = Z2 , and 1 (S 1 ) = Z, the homomorphism must be
the trivial homomorphism with image 0. So must lift to a map : RP 2 R. But every
map to R is nullhomotopic, so we have a homotopy, say t from 0 = to 1 = 0. Then if
p : R S 1 is the covering map, p t is a nullhomotopy of .
X is a covering space with X

Suppose Y is simply connected and p : X

contractible. Show that every map f : Y X is homotopic to a constant map.
Since Y is simply connected, the criterion for the lifting theorem is automatically satisfied,
of f . Since X
is contractible, say by a homotopy Ht , the
so there is a lift f : Y X
composition Ht f gives a homotopy from f to a constant map and pHt f gives a homotopy
from pf = f to a constant map.
Describe the universal cover of T 2 S 1 , where T 2 is the torus S 1 S 1 . This is like
an antenna space, except it alternates lines and planes. Start with a line. At every integer
points attach a plane. At ever Z Z lattice point in each plane, attach a line (disjoint from
the rest of the space otherwise). At every integer point of each new line, attach a plane.
And so on.
Let Xn be the space obtained from the two-dimensional disk D2 by identifying
points on the boundary that differ by an angle 2/n; let q : D2 Xn be the
quotient map.
Whats 1 (Xn ) (you dont need to write out the argument)? Zn
Suppose x is in the interior of D2 ; describe what small neighborhoods of q(x)
look like. Now suppose x is in the boundary of D2 ; describe what small neighborhoods of q(x) look like.
For points in the interior of D2 , the corresponding points
in Xn have neighborhoods homeomorphic to Euclidean 2-space R2 . For points on D2 , their
neighborhoods in Xn look like n copies of the half-plane {(x, y) R2 | y 0}, all joined
along the x-axis.

14

 n of Xn using what you figured out in

Find and describe the universal cover X
the previous sections; hint: think about the familiar case X2 = RP 2 .
Recall

that each point in Xn has to have n points in its preimage in Xn . Thinking about RP 2 ,
the space can be assembled as follows: start with n copies of D2 , then glue together their
boundaries such that the point at angle on the ith copy ( mod n) gets glued to the point
at angle + 2/n on the i + 1st copy ( mod n). It is not difficult to see that this space is
simply connected, using the van Kampen Theorem, and then we can check that each point
n has neighborhoods that get taken homeomorphically to Xn in the appropriate way:
in X
the preimage of each point in the interior of D2 has a neighborhood that looks like n copies
of D2 (displaced in the copies of D2 by angles of 2/n), while the preimage of each point
corresponding to a boundary point of the original D2 consists of n neighborhoods around
n , each of the form described above (a union of half planes).
the edge of X
Due 2/20/2014
Show that a covering map p : X Y is an open map (and hence a quotient
map). Let {Ua } be a covering of Y by open sets such that each p1 (Ua ) is a disjoint union
of open sets each homeomorphic to Ua . We can write p1 (Ua ) = b Uab , where each Uab is
homeomorphic to Ua . Then a,b Uab covers X, so there is a basis {Bi } for X for which each
elements of the basis lies inside some Uab . But then if Bi Uab is an open set in the basis,
then p(Bi ) is also open because p|Uab is an isomorphism. Any open set V of X is a union of
some collection of the Bi , V = iJ Bi . Then p(V ) = p(iJ Bi ) = iJ p(Bi ) is open. So p is
an open map.
Alternative: Let U be an open set of X. Let x U . Then, as p is a covering map, p(x)
Y has a neighborhood V such that p1 (V ) is a disjoint union of open sets homeomorphic
to U . Let V0 be the subset of p1 (V ) that is homeomorphic to V and contains x. Then
V0 U is an open neighborhood of x that maps to an open neighborhood of p(x) in p(U ).
Since x U was arbitrary and p : U p(U ) is surjective, every point of p(U ) has an open
neighborhood in p(U ). So p(U ) is open.
X
Hatcher 1.3.1. If Ua is the open covering of X in the definition of what makes p : X
a covering map, then {Ua A} is an open covering of A. Furthermore, it has the desired
property because if p1 (Ua ) = qb Ua,b , then the homeomorphism Ua,b Ua induced by p
restricts to a homeomorphism Ua,b p1 (A) Ua A.
Hatcher 1.3.1. Let {U } and {V } be appropriate coverings of X1 and X2 . Consider
1
(p1 p2 )1 (U V ) = p1
1 (U ) p2 (V ) (this is an easy set theory exercise). Suppose
1
p1
1 (U ) = qU,i and p2 (V ) = qV,j . Then the sets U,i V,j are disjoint from each other
(again by easy set theory). And p1 p2 restricts to a homeomorphism from U,i V,j to
U V because the product of homeomorphisms is a homeomorphism.
Use a covering space of S 1 S 1 to show that the free group on 3 generators is
isomorphic to a subgroup of the free group on 2 generators. Generalize 3 to n.
15

Use a covering space that is a union of 3 (or n) circles and so has 1 = Fn .

Find a simply connected cover of the space . This is the free graph of valence 3.
Due 2/13/2014
Think of the real projective plane RP 2 as the space obtained from the unit disk
by identifying opposite points on the boundary. 1. Show that 1 (RP 2 )
= Z2 . 2.
Draw a picture showing a loop representing the nontrivial element [f ] 1 (RP 2 ).
3. Draw a series of pictures showing the nullhomotopy from 2[f ] to the constant
path. Using the picture, we see that RP 2 can be formed as a CW complex with one cell in
each dimension 0, 1, 2. If a represents the one-cell, then the attaching map takes the boundary
of the 2-cell to a2 . So by our theorems about 1 of cell complexes, 1 (RP 2 ) = Z/ha2 i
= Z2 .
Let X be the quotient space of the disk D2 obtained by identifying points on
the boundary that are 120 degree apart. Compute 1 (X) Just like for RP 2 , this
1 = Z/3.
Let X be the complement of n points in R2 . Compute 1 (X). Do the same for the
complement of n points in R3 . [Hint: use induction on n]. If n = 1, we can assume
without loss of generality that the point is the origin. Then X deformation retracts to S 1
so 1 (X) = Z. Now, suppose weve shown that for n 1 points 1 (X) = n1 Z, the free
product on n 1 copies of Z. Let X be the complement of n points. Its possible to find a
plane that separates the points into two groups of size m and m0 , each less than n. Now write
X = A B where A and B are each halfspaces determined by the plane, but thickened up
slightly so that A B
= Rn1 (, ). Then by induction 1 (A)
= m Z and 1 (B)
= m0 Z.

1 (A B) = 0, so by the van Kampen theorem, 1 (X) = (m Z) (m0 Z) = n Z.

For R3 , the proof is the same except all groups are trivial.
Hatcher 1.2.7 Start with the cell structure on S 2 with two cells in each dimension. So
there are two 0 cells, v0 , v1 . We can assume the two one cells e1 , e2 are attached so that
they both run from v0 to v1 . Then to attach the two 2-cells, we can attach each one by
sending its boundary to the loop e1 e1
2 . Now form the quotient X by gluing v0 to v1 . Then
e1 and e2 are loops, and the 1-skeleton X 1 of X is the figure eight. So 1 (X 1 ) = Z Z.
But now using Proposition 1.26, 1 (X) = 1 (X 1 )/N , where N is the subgroup generated
by the loops along which we glue the 2-cells. But the only such loop is still e1 e1
2 . So
1
1 (X) = Z Z/he1 e2 i = Z Z/he1 = e2 i. So this group is generated by {e1 , e2 } but e1 = e2 .
So this is just Z.
Consider the annulus. Identify antipodal (opposite) points on the outer circle
with each other (as if youre forming RP 2 ). Also identify antipodal points on
the inner circle with each other. Call the resulting space X. Compute 1 (X).
16

Let the space be X. Suppose the annulus consists of the points in the plane with radius
1 r 2. Write the annulus as AB where A consists of the points of radius 1 < r 2 and
B consists of the points with radius 1 r < 2. Then each of A and B deformation retracts
to the boundary, which is just a circle (note: if you identify the antipodal points of a circle,
you get a circle). So 1 (A)
= 1 (B)
= Z. So 1 (X)
= Z 1 (AB) Z. Now A B also retracts
to a circle, so 1 (A B)
Z.
The
inclusions

(A
B) 1 (A) and 1 (A B) 1 (B)
=
1
can each be identified as multiplication by 2 Z Z. So 1 (X) = Z Z/hx2 = y 2 i
= Z Z2 .
Let X be the union of a sphere S 2 with one of its diameters. Compute 1 (X).
1 (X) = Z using the van Kampen theorem. Alternatively, note that this space is homotopy
equivalent to the one in Hatcher 1.2.7.
Due 2/6/2014
1. Suppose that X is a compact metric space and that U is an open covering of X. Show
that there is a number > 0 (called a Lebesgue number ) such that for every set Z X
with diameter < , there is an element U U such that Z U .
Suppose there is no such . Then for each n N, there is a subset Zn X with
diameter < 1/n and such that Zn is not contained in any single element of U. Let
zn Zn . Then zn has a convergent subsequence, as X is compact. Let z be the limit
of the subsequence. But then there is some U U such that x U , and so there is a
ball of some diameter d centered at x and contained in U . But then an easy inequality
argument shows that there must be some Zm contained in U , a contradiction.
2. Use the results of the previous exercise to show that if f : I Y is a path in
the arbitrary space Y and if V is an open covering of Y then there is a partition
0 = t0 t1 tn = 1 of I such that for each interval [ti , ti+1 ], the image
f ([ti , ti+1 ]) is contained in some single element V V.
Let be the Lebesgue number for the covering {f 1 (V ) : V V} of I. Then 0
/2 3/2 . . . N /2 1, for an appropriate N , gives a partition of the
desired form.
Hatcher 1.1.13 Let : A X be the inclusion. Suppose every path in X with endpoints
in A is homotopic to a path in A. Let f be a loop in X based at x0 . By the assumption, f
is homotopic to a loop f 0 in A. But then f 0 represents an element [f 0 ] 1 (A, x0 ) and also
([f 0 ]) = [f ] 1 (X, x0 ) because f is homotopic to f 0 . So is surjective.
Conversely, let f be a path with endpoints in A. Let h1 be a path from f (0) to x0 in
A, and let h2 be a path from x0 to f (1) in A. Since is onto, there is a homotopy from
1
1
1
h1
1 f h2 to a loop g contained in A, i.e. h1 f h2 g. But then f h1 gh2 using previous
problems.

17

Hatcher 1.1.16b If there were a retraction S 1 D2 S 1 S 1 , then there would be

homomorphisms 1 (S 1 S 1 ) 1 (S 1 D2 ) 1 (S 1 S 1 ) that compose to the identity.
But 1 (S 1 D2 )
= 1 (S 1 ) 1 (D2 )
= Z, while 1 (S 1 S 1 )
= Z Z. This leads to a
contradiction as there is no surjection Z Z Z (pick your favorite reason from abstract
algebra; mine is that the image would have to consist of multiples of some (a, b) Z Z
because the image of a cyclic group is cyclic, but then its not hard to argue that this cannot
include all possible (x, y)).
Hatcher 1.1.17 For n Z define n : S 1 S 1 S 1 so that n is the identity on the first
S 1 of S 1 S 1 and n restricts to the angle map n for S 1 . So n is the identity
on the first circle and wraps the second circle n times around the image circle. n cannot
be homotopic to m for m 6= n because if [f ] 1 (S 1 S 1 ) is represented by the loop that
generates 1 of the second circle in S 1 S 1 , then n ([f ]) and m ([f ]) represent different
elements of 1 (S 1 ) by what we know about the fundamental group of the circle. Thus n is
not homotopic to m using Lemma 1.19 of the book.
Hatcher 0.3a Suppose f : X Y and g : Y Z are homotopy equivalences. Let f 0 and
g 0 be the homotopy inverses. We claim f 0 g 0 is a homotopy inverse to gf . To see this, we
recall that, by definition, there are homotopies, say Ft from f 0 f to idX , Gt from g 0 g to idY ,
Ht from f f 0 to idY , and It from gg 0 to idZ . Then we can obtain a homotopy from f 0 g 0 gf
to idX as follows. First do the homotopy f 0 Gt f from f 0 g 0 gf to f 0 idY f = f 0 f . Then do the
homotopy Ft from f 0 f to idX . The other homotopy from gf f 0 g 0 to idZ is similar.
This shows that the relation of being homotopy equivalent is transitive. But clearly for
any space X, X is homotopy equivalent to X by the identity. Symmetry is evident from the
definition. So homotopy equivalence is an equivalence relation.
Hatcher 0.3c (You may assume 0.3b) Suppose f 0 : X Y is homotopic via Ft to
the homotopy equivalence f : X Y , with homotopy inverse g : Y X. Then f 0 g is
homotopic to f g via Ft g, which is homotopic to idY and gf 0 is homotopic via gFt to gf ,
which is homotopic to idX . Thus g is also a homotopy inverse to f 0 .
Hatcher 0.4 Let F be the weak deformation retraction. Let f1 (x) = F (x, 1) : X A,
and let g : A , X be the inclusion. Then for gf1 : X X, we have gf1 = f1 , which is
homotopic to f0 , which is the identity. On the other hand, f1 g : A A is the same as the
restriction of f1 to A. But then the restriction of F to A I is a homotopy from f1 |A to
f0 |A = id|A .
Hatcher 0.6a [Hint to students. Heres a point set topology lemma that might be useful:
Suppose X Y is a product space, Y is compact, and N is an open set in X Y . Suppose
also that x0 Y N for some x0 X. Then there is a neighborhood W of x0 in X such
that W Y N . This is a very important lemma, sometimes called the tube lemma.]

18

First we can perform a deformation retraction ft (x, y) = (x, ty). At time t = 1 this is
the identity and at time t = 0 every point is on the horizontal segment. Now perform a
deformation retract from the horizontal segment to your favorite point x0 . Explicitly, this
can be done as gt (x) = tx0 + (1 t)x. The desired deformation retraction comes by first
doing ft and then gt .
It is impossible to have a deformation retract to any other point (x0 , y0 ) X for y0 > 0
because this would contradict the continuity of the retraction. Suppose it were possible.
Then the homotopy F : X I X giving the deformation retract would have to be
continuous. Pick a ball neighborhood U = B ((x0 , y0 )) of (x0 , y0 ) where the radius  satisfies
0 <  < y0 . Now F 1 (U ) would have to be open in X I and contain [X 1] [(x0 , y0 ) I].
In particular, it follows from the tube lemma that F 1 (U ) would contain some neighborhood
B I around (x0 , y0 ) I in X I. But then every point in B would have to stay within
 of (x0 , y0 ) through the homotopy, which is impossible.
Give a presentation in terms of generators and relations of Z4 involving one
generator; involving two generators; involving three generators. There are many
answers to this. Here are some: hx | x4 = ei, hx, y | x4 = 1, y = x2 i, hx, y, z | x4 = 1, y =
x, z = x.
Recall the fundamental theorem of finitely generated abelian groups. If G is a
finitely generated abelian group, explain how you would write down a presentation for it. Every finitely generated abelian group has the form Zr Zp1 Zpn }, where
Zr is the product of r copies of z. So a presentation would be hx1 , . . . , xr , y1 , . . . , yn | yipi =
e, xi xj = xj xi , ya yb = yb ya , xk y` = y` xk i.
Consider the group G with presentation hx, y | x4 = e, y 2 = e, (xy)2 = ei. Show
that G is a finite group and determine how many elements it has. Hint: start
by rewriting the relation (xy)2 = e in a more useful form. Extra credit: can you
identify G as a familiar group? Using (xy)2 = xyxy = e, we see xy = y 1 x1 = yx1
(since y 2 = e implies y = y 1 ). So given any word, we can use this relation to pull any ys
from right to left across powers of x (for negative powers, note that x1 = x3 , so we can
always write xs in terms of positive powers). Thus every word can be rewritten as y n xm , and
clearly n = 0, 1 and m = 0, 1, 2, 3, so there are at most 8 elements of the group. Furthermore,
this group is isomorphic to the dihedral D4 ; its not hard to find an explicit isomorphism.
Due 1/30/2014
Hatcher 1.1.2. Suppose h and g are homtopic paths from x0 to x1 and that [f ] 1 (X, x0 ).
Then hf h1 is homotopic to gf g 1 (see the paragraph in which Hatcher defined the product
of paths), so h ([f ]) = g ([f ]).

19

Hatcher 1.1.3. First suppose h depends only on the endpoints of h and not the actual
path. Let x0 X be a basepoint. Let [h] 1 (X, x0 ). Because h has the same endpoints
as the constant path, h : 1 (X, x0 ) 1 (X, x0 ) is the same as c : 1 (X, x0 ) 1 (X, x0 ),
where c is the constant path. But it is clear that c = id. Therefore, using the definition of
h , [g] = c ([g]) = h ([g]) = [hgh1 ] = [h][g][h]1 . But then [g][h] = [h][g]. Since [h] and [g]
were arbitrary, 1 (X, x0 ) is abelian.
Conversely, assume 1 (X, x0 ) is abelian for any x0 . Let h and g be two paths from x0 to
x1 , and let [f ] 1 (X, x1 ). We must show that h ([f ]) = g ([f ]), i.e. that [hf h1 ] = [gf g 1 ],
which is the same as showing that hf h1 gf 1 g 1 is homotopic to the constant map to x0 . But
now notice that h1 g and f are both loops based at x1 , so [h1 g][f ] = [f ][h1 g] as elements
of 1 (X, x1 ). But now using this homotopy, we see that hf h1 gf 1 g 1 is homotopic to
hh1 gf f 1 g 1 , which is homotopic to the constant map.
Hatcher 1.1.6. Since I cant draw pictures here, Ill just outline the ideas. Let [g]f denote
the equivalence class of the loop g : S 1 X in [S 1 , X] (the g stands for free - these are
sometimes called free homotopy classes). This is an equivalence relation just like homotopy
of paths is (the proof is essentially identical).
One of the keys to this problem is to notice the following: suppose f : S 1 X can be
written as the composition of paths g1 g2 . . . gk , where the gi are paths but not necessarily
loops. Then [g1 g2 . . . gk ]f = [g2 . . . gk g1 ]f ; in other words, cyclically permuting the order of the
paths gives the same equivalence class. This can be seen by just rotating the parametrization,
i.e. if the parametrization of g2 starts at angle from s0 , then we have a homotopy ft () =
f ( + t) from g1 g2 . . . gk to g2 . . . gk g1 .
Now, let f [S 1 , X], and let x1 = f (s0 ), where s0 is a basepoint of S 1 . Let h be any path
from x0 to x1 (this exists by the assumption of path-connectedness). We can consider the
map f as representing an element of 1 (X, x1 ) and form h ([f ]). This gives a loop, hf h1
based at x0 . By our observation above, this is homotopic to f h1 h, and this is homotopic to
f by reeling in h1 h to x1 analogously to the proof that h1 h is homotopic to the constant
path. This shows that is surjective.
Now suppose [f ], [g] 1 (X, x0 ) are conjugate. Then by definition, [f ] = [h][g][h]1 for
some [h] 1 (X, x0 ). So [f ] = [hgh1 ]. But then [f ]f = ([f ]) = ([hgh1 ]) = [hgh1 ]f =
[gh1 h]f = [g]f .
Finally, suppose ([f ]) = ([g]), so there is a free homotopy F : S 1 I X from f to
g. Let h be the path followed by s0 during this homotopy. Then f is homotopic, preserving
basepoints, to hgh1 . (This would be easier in pictures (come by my office if you want to
S 1

see it), but the homotopy can be described by the composition I I

I
, where H
takes I 0 to S 1 0 by the obvious quotient and 0 I and 1 I to s0 0 and I 1 up
s0 I, around S 1 1 and then down s0 I).
Hatcher 1.1.10 Suppose [f ] 1 (X, x0 ) and [g] 1 (Y, y0 ). We will show that as a loop
in X Y , f g is homotopic to the loop s (f (s), g(s)). The same argument in reverse
shows that that gf is also homotopic to s (f (s), g(s)), so f g gf . For the homotopy,
20

we simply note that f g consists of the path s (f (2s), cy0 ) X Y for s [0, 1/2], where
cy0 is the constant path at y0 Y , and similarly, (f g)(s) = (cx0 , g(2s 1)) for s [1/2, 1].
In other words, if f 0 is the path that does f (2s) for s [0, 1/2] and is constant after that
and g 0 is the path that does g(2s 1) for s [1/2, 1] and is constant before that, then
f g = (f 0 , g 0 ). But clearly f 0 f and g 0 g, so (f 0 , g 0 ) (f, g), using the isomorphism
1 (X Y, x0 y0 )
= 1 (X, x0 ) 1 (Y, y0 ).
Topological group problem. i. The operation is well-defined and Ill let you show its
associative (should be easy). The identity is the constant path ce since (f ce )(s) = f (s)e =
f (s). If f is a loop, then its inverse in this group is the loop s (f (s))1 .
ii. Suppose F (s, t) is the homotopy from f to f 0 and G(s, t) is the homotopy from g to
0
g . Let H(s, t) = F (s, t) G(s, t). Then H(s, 0) = F (s, 0) G(s, 0) = f (s) g(s) = (f g)(s)
and H(s, 1) = F (s, 1) G(s, 1) = f 0 (s)g 0 (s) = (f 0 g 0 )(s). So H is a homotopy from f g
to f 0 g 0 .
iii. On the one hand f ce f and ce g g. So by part ii, (f ce ) (ce g) f g. On
the other hand, ((f ce ) (ce g))(s) = (f ce )(s) (ce g)(s). Notice that if s 1/2, then ce g(s)
is just e, and if s 1/2, f ce (s) = e. So (f ce )(s) (ce g)(s) = f g. So f g f g. So
[f g] = [f g] 1 (G, e).
iv. In the last problem we saw that (f ce ) (ce g) f g, but f ce f ce f and
ce g g gce . So (f ce ) (ce g) (ce f ) (gce ) g f (the last relation for essentially the
same reason as in part iii). So [f ] [g] = [g] [f ] 1 (G, e), and since [f ] [g] = [f ][g] by
part iii, also [f ][g] = [g][f ].
Due 1/23/2014
Hatcher 1.1.1. Since g0 g1 , it follows that g0 g1 : Recall that g0 (s) = g0 (1 s)
and g1 (s) = g1 (1 s). So if G(s, t) is the homotopy from g0 to g1 , then G(1 s, t) is the
homotopy from g0 to g1 . Now, as f0 g0 f1 g1 and g0 g1 , we have f0 g0 g0 f1 g1 g1 . But g0 g0
is homotopic to the constant path at f0 (1) = f1 (1), so f0 g0 g0 f0 . Similarly f1 g1 g1 f1 .
Hatcher 1.1.4. UPDATED: Lemma: if Z is a star-shaped subset of Rn , then Z is simply
connected. Proof: Let z Z be the center and consider 1 (Z, z). Let f be any loop based
at z. Let F (t, s) = tz + (1 t)f (s). This is a continuous homotopy from f to the constant
path at z. Since every point in the homotopy lies along the straight path from f (s) Z to
z, F is contained in Z. Therefore, Z is simply connected.
Now, let f : I X be a path. By compactness, we can divide I into subintervals
0 = s0 s1 sn = 1 such that each f ([si , si+1 ]) is contained in a single star-shaped
neighborhood, say Si . Let zi be the center of the star Si . As Si is simply connected, any
two paths with the same endpoints are homotopic (we proved this in class). So f |[si ,si+1 ]
is homotopic to the path that goes straight from f (si ) to zi and then straight from zi to
f (si+1 ).
More general than homotopies of paths are homotopies of maps. Two maps
f, g : X Y are called homotopic (written f g) if there exists a map F :
21

X I Y such that F (x, 0) = f (x) and F (x, 1) = g(x). Show that, for any fixed X
and Y , homotopy is an equivalence relation on the set of maps from X to Y .
Same proof as for path homotopies! (see class notes)
Suppose f, f 0 : X Y with f f 0 and g, g 0 : Y Z with g g 0 . Show that
g f g 0 f 0 , where denotes composition of functions.
Let F : X I Y be the homotopy from f to f 0 and G : Y I Z be the homotopy
from g to g 0 . Now use G(F (x, t), t).
Suppose f, f 0 : X Y are homotopic and g, g 0 : Z W are homotopic. Show that
f g, f 0 g 0 : X Z Y W are homotopic.
Let F : X I Y be the homotopy from f to f 0 and G : Z I W be the homotopy
from g to g 0 . Now use (F (x, t), G(z, t)).

22