Solutions to Rotman’s algebraic topology

Jessica Zhang
Last updated: May 22, 2021
Contents

0 Introduction 3
Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Brouwer Fixed Point Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Categories and Functors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

1 Some Basic Topological Notions 7

Homotopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Convexity, Contractibility, and Cones . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Paths and Path Connectedness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2 Simplexes 12
Affine Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Affine Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

3 The Fundamental Group 14

The Fundamental Groupoid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
The Functor π1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
π1 (S 1 ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

4 Singular Homology 20
Holes and Green’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Free Abelian Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
The Singular Complex and Homology Functors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Dimension Axiom and Compact Supports . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
The Homotopy Axiom . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
The Hurewicz Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

5 Long Exact Sequences 23

The Category Comp . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Exact Homology Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
Reduced Homology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

6 Excision and Applications 30

Excision and Mayer–Vietoris . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
Homology of Spheres and Some Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
Barycentric Subdivision and Proof of Excision . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
More Applications to Euclidean Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

7 Simplicial Complexes 34
Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
Simplicial Approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
Abstract Simplicial Complexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
Simplicial Homology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
Comparison with Singular Homology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
Calculations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
Fundamental Groups of Polyhedra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
The Seifert–van Kampen Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

1
8 CW Complexes 42
Hausdorff Quotient Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
Attaching Cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
Homology and Attaching Cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
CW Complexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
Cellular Homology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

9 Natural Transformations 49
Definitions and Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
Eilenberg–Steenrod Axioms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
Chain Equivalences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
Acyclic Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
Lefschetz Fixed Point Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
Tensor Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
Universal Coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
Eilenberg–Zilber Theorem and the Künneth Formula . . . . . . . . . . . . . . . . . . . . . . . . . . 56

10 Covering Spaces 59
Basic Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
Covering Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
Existence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
Orbit Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

11 Homotopy Groups 63
Function Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
Group Objects and Cogroup Objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

2
0 Introduction
Notation
No exercises!

Brouwer Fixed Point Theorem

Exercise 0.1. As per the hint, observe that if y ∈ G, then we have y = r(y) + (y − r(y)). Obviously, we
have r(y) ∈ H. Moreover, we know that

r(y − r(y)) = r(y) − r(r(y)) = 0,

and so y − r(y) ∈ ker r. Thus G ⊆ H ⊕ ker r.

The reverse is obviously true, since H and ker r are both subgroups of G.
Exercise 0.2. Suppose instead that f : D1 → D1 has no fixed point. Then consider the continuous map
g : D1 → S 0 given by (
1 if f (x) < x
g(x) = .
−1 if f (x) > x
Notice that because f (x) 6= x for all x, the function g is well-defined.
Moreover, we know that f (−1) 6= −1, since f has no fixed point, and so f (−1) > −1. Thus g(−1) = −1.
Similarly, we have g(1) = 1.
Thus we have g(D1 ) = S 0 , which is disconnected. This is a contradiction, so f must have had a fixed
point.
Exercise 0.3. Suppose that r is such a retract. Then we have the following commutative diagram:

Sn
i r

1
S n−1 S n−1 .

Applying Hn−1 , we get another commutative diagram:

Hn−1 (S n )
Hn−1 (i) Hn−1 (r)

Hn−1 (1)
Hn−1 (S n−1 ) Hn−1 (S n−1 ).

We know that Hn−1 (S n ) = 0, however, implying that Hn−1 (1) = 0. This contradicts the fact that
Hn−1 (S n−1 ) = Z 6= 0. Thus the retraction r could not have existed.

Exercise 0.4. Suppose g : Dn → X is a homeomorphism. Then we know that g −1 ◦ f ◦ g is a continuous

map from Dn to itself, and so it has a fixed point x. Then we know that g −1 (f (g(x))) = x, and so it follows
that f (g(x)) = g(x). Thus g(x) ∈ X is a fixed point of f .
Exercise 0.5. Consider the function h : I × I → I × I given by

h(s, t) = f (s) − g(t) + (s, t).

This is the sum of continuous functions, and so it is itself continuous. Moreover, we know that I × I
is homeomorphic to D1 , and so it follows that there is a fixed point (s, t) of h. But this means that
f (s) − g(t) = 0, and so we are done.

3
Exercise 0.6. Observe that x ∈ ∆n−1 must contain some positive coordinate, because
P
xi = 1 and xi ≥ 0
for all i. Since aij > 0 for every i, j, it follows that Ax contains only nonnegative coordinates and, moreover,
contains at least one positive coordinate. Thus σ(Ax) > 0, and so g(x) is well-defined.
Moreover, it is continuous because the linear map A, the map σ, and the division function are all
continuous.
Because ∆n−1 ≈ Dn−1 , it follows that there exists some x with
Ax
x= .
σ(Ax)

Then λ = σ(Ax) > 0 is a positive eigenvalue for A and x ∈ ∆n−1 is a corresponding eigenvector.
We know that x contains only nonnegative coordinates. Suppose then that some coordinate, say x1 , is
zero. Then obviously the first coordinate of λx is zero. However, the first coordinate of Ax is

a11 x1 + a12 x2 + · · · + a1n xn = a12 x2 + · · · + a1n xn .

P
Since xi = 1 and x1 = 0, there exists some k 6= 1 such that xk > 0. Then a1k xk > 0, and since each
i already has a1i xi ≥ 0, it follows that the first coordinate of Ax is strictly positive, contradicting that
Ax = λx.
Thus the eigenvector x has all positive coordinates.

Categories and Functors

Exercise 0.7. We know that
g ◦ (f ◦ h) = g ◦ 1b = g
and
(g ◦ f ) ◦ h = 1A ◦ h = h,
and so associativity implies g = h.

Exercise 0.8.
(i) Notice that if 1A and 10A are both identities, then we must have

1A = 1A ◦ 10A = 10A ,

which proves the desired result.

(ii) If 10A is the new identity in C 0 , then we know that 10A ∈ HomC 0 (A, A) ⊆ HomC (A, A), and so 1A ◦ 10A is
defined. But we know that
10A ◦ 1A = 10A = 10A ◦ 10A ,
and so Exercise 0.7 implies the result.
Exercise 0.9. Clearly, the Hom-sets are pairwise disjoint, since each ixy appears at most once.
It is also obviously associative. In particular, if a ≤ b ≤ c ≤ d, then we know that

icd ◦ ibc ◦ iab = icd ◦ iac = iad ,

and similarly for icd ◦ ibc ◦ iab .
Finally, the map ixx is the identity on x ∈ X. To see that it is a left-identity, note that if y ≤ x, then

ixx ◦ iyx = iyx .

Similarly, we can show that this map is a right-identity as well, and so we are done.
Exercise 0.10. Disjointness is clear, since there is only one object. Because G is a monoid, it is associative
and has an identity, proving that C is a category.

4
Exercise 0.11. It is pretty clear that obj(Top) ⊂ obj(Top2 ). Moreover, a continuous map f : X → Y
between two topological spaces corresponds to the map (f, ∅) in Top2 from (X, ∅) to (Y, ∅), which then
means that Top can be thought of as a subcategory of Top2 .
Exercise 0.12. It is worth noting that Rotman’s definition here is incorrect. The morphisms in M should
be the commutative squares, not merely the ordered pairs (h, k).
Indeed, consider the following counterexample to Rotman’s definition. Let C be the category of sets.
Furthermore, let A be a set with more than one element. Then the following diagrams are both commutative:
1A 0
A A A A
1A 0 1A 0
0 0
A {0} A {0}.

This implies that the ordered pair (1A , 0), where 0 is considered to be the map that sends everything in A
to the zero element, is both in Hom(1A , 0) and in Hom(0, 0), contradicting disjointness.
If we instead consider morphisms of M to be the commutative squares, where composition is defined by
“stacking” the squares on top of one another, disjointness is clear. After all, the squares contain f and g,
and so Hom-sets of different objects must be disjoint.
Associativity is clear, as the morphisms of C are associative.
Finally, there is an identity 1f for every f ∈ HomC (A, B), namely the one where h = 1A and k = 1B .
Exercise 0.13. With the hint, this is clear. In particular, we consider Top2 to be the subcategory of the
arrow category of Top in which the objects are inclusions, and HomTop2 (i, j) = HomTop (i, j).
Exercise 0.14. To see that it is a congruence at all, observe that Property (i) is satisfied because there is
only one Hom-set. Moreover, if x ∼ x0 and y ∼ y 0 , then we know that x(x0 )−1 = hx and y(y 0 )−1 = hy for
some hx , hy ∈ H. But then we know that

(yx)(y 0 x0 )−1 = yx(x0 )−1 (y 0 )−1 = yhx (y 0 )−1 .

However, since (y 0 )−1 = y −1 hy , we know that this is simply

(yx)(y 0 x0 )−1 = yhx y −1 hy .

Because H is normal, we know that yhx y −1 ∈ H. Thus the product of this and hy is in H as well, and so
xy ∼ x0 y 0 , as desired.
To see that [∗, ∗] = G/H simply requires the observation that x ∼ y if and only if x and y are in the
same coset of H.
Exercise 0.15. This follows from the fact that functors preserve (or, in the case of contravariant functors,
reverse) the directions of the arrows. Thus the resulting diagram still commutes.
Exercise 0.16. Note that for (i)–(iv), we can simply use inverses. For instance, for Set, it suffices to note
that if f is a bijection, then f −1 is a bijection, which is clearly true. Similarly, the inverse of a homeomorphism
is a homeomorphism, and the inverse of a group or ring isomorphism is still an isomorphism.
For (v), note that iyx is defined and satisfies the requirements that iyx ◦ ixy = ixx and ixy ◦ iyx = iyy .
For part (vi), notice that f −1 works because f is a homeomorphism. In particular, it is a bijection, and
so f −1 (A0 ) = A. Moreover, it is (bi)continuous since f is.
Finally, for the monoid G, if g has a two-sided inverse h, then hg = gh = 1, which is the identity element
of Hom(G, G).
Exercise 0.17. To prove that T 0 is a functor, first observe that criterion (i) of a functor is satisfied because
T does so. Moreover, if [f ] ∈ HomC 0 (A, B), then f ∈ HomC (A, B), and so T 0 ([f ]) = T f is a morphism in
A . In particular, if [g] ◦ [f ] = [g ◦ f ] is defined in C 0 , then g ◦ f is defined in C . This means, then, that

T 0 ([g] ◦ [f ]) = T (g ◦ f ) = (T g) ◦ (T f ) = T 0 ([g]) ◦ T 0 ([f ]).

Finally, it remains to note that T 0 ([1A ]) = T1A = 1T A = 1T 0 ([A]) for every object A. Thus T 0 is a functor.

5
 Exercise 0.18.
(i) It is clear that tG ∈ obj Ab for every group G. Now suppose that we have a homomorphism f : G → H.
Then we know that t(f ) is a morphism f |tG from tG to tH. To see this, note that it is the restriction of
a homomorphism, and thus is itself a homormophism. Moreover, if x ∈ f (tG), then x = f (y) for some
y ∈ G with finite order. But then there exists some n so that y n = 1. Thus xn = f (y n ) = 1, and so x
has finite order. But x ∈ f (G) ⊆ H implies that x ∈ tH.
Now we must check that t respects composition. Indeed, if g ◦ f is defined, then

t(g ◦ f ) = (g ◦ f )tG = g|f (tG) ◦ f |tG .

But f (tG) ⊆ tH, and so this is simply

t(g ◦ f ) = g|tH ◦ f |tG = t(g) ◦ t(f ),

which proves that composition is respected.

Finally, note simply that t(1G ) = 1|tG , which is the identity on tG.
(ii) Suppose that f is an injective homomorphism from G to H. Then suppose that t(f )(x) = t(f )(y). But
f (x) = f |tG (x) = t(f )(x), and so it follows that f (x) = f (y). Injectivity of f proves the result.
(iii) Let G = Z and H = Z/2Z and let f take even integers to 0 and odd integers to 1. This is evidently
surjective. But tG = {0} while tH = {0, 1}, and so t(f ) : tG → tH cannot be surjective.
Exercise 0.19.

(i) If f is a surjection, then consider an arbitrary coset a + pH of H/pH. We know that there exists some
b ∈ G with f (b) = a, and so it follows that F (f ) takes b + pG to a + pH, proving surjectivity of F (f ).
(ii) Consider the function f : Z → Z taking x to 2x. Then, letting p = 2, we know that F (f ) : Z/2Z → Z/2Z
has F (f )([0]) = F (f )([1]).
Exercise 0.20.

(i) This is evident because R is a ring, and the operations are pointwise.
(ii) By the previous part, we know that if X is a topological space, then C(X) is a ring. Now suppose that
f : X → Y is a continuous map. Then define

C(f ) : C(Y ) → C(X)

g 7→ g ◦ f

and note that this is well-defined. Moreover, we know that C(g ◦ f )(h) = h ◦ g ◦ f , while C(f ) ◦ C(g)
takes h to C(f ) ◦ (h ◦ g) = h ◦ g ◦ f , which proves that C reverses composition. Finally, we know that
C(1x ) takes g to g ◦ 1X = g and is therefore the identity on C(Y ). Thus C (or, rather, the map taking
X to C(X), to be precise) gives rise to a contravariant functor.

6
1 Some Basic Topological Notions
Homotopy
No exercises!

Convexity, Contractibility, and Cones

Exercise 1.1. Suppose H : f0 ' f1 is a homotopy. Then let F (t) = H(x, t) for some fixed x. It is clear
that F (0) = x0 and F (1) = 1. Moreover, since H is continuous, it follows that so too is F . For the converse,
simply let the homotopy H : f0 ' f1 take (x, t) ∈ X × I to F (t).
Exercise 1.2.
(i) There exist functions f : X → Y and g : Y → X such that g ◦ f ' 1X and f ◦ g ' 1Y . Moreover,
there is a homotopy F : 1X ' c, where c denotes the constant map at some x0 ∈ X. Then consider
the map G : Y × I → Y which takes (y, t) to f (F (g(y), t)). In particular, we know that G is continuous
and that it is thus a homotopy from f ◦ g to the constant map c0 at y0 = f (x0 ). But then we find that
1Y ' f ◦ g ' c0 , and so Y is contractible.
(ii) Consider, for example, the subsets X, Y ⊂ R2 where

X = {(x, 0) : x ∈ [0, 1]},

     
1 1
Y = (x, x) : x ∈ 0, ∪ (x, 1 − x) : x ∈ ,1 .
2 2
It is obvious that X is convex, but Y is not, even though there is an obvious homotopy equivalence from
X to Y .
Exercise 1.3. We know that R(x) = eiα x, and so the continuous map F : S 1 × I → S 1 given by F (x, t) =
eiαt x is a homotopy F : 1S ' R. Thus, if g : S 1 → S 1 is continuous, then let θ be such that g(1) = g(ei·0 ) =
eiθ . Then we know that, letting R now be the rotation of −θ degrees, we must have R ◦ g ' 1S ' g = g and
(R ◦ g)(1) = 1, as desired.
Exercise 1.4.
(i) Pick (x1 , y1 ), (x2 , y2 ) ∈ X × Y . Then we know that, for any t ∈ I, we have

t(x1 , y1 ) + (1 − t)(x2 , y2 ) = (tx1 + (1 − t)x2 , ty1 + (1 − t)y2 ).

The result follows from convexity of X and Y .

(ii) If FX : 1X ' cX and FY : 1Y ' cY , where cX and cY are constant maps at cX and cY , respectively, then
the map

F : (X × Y ) × I → X × Y
(x, y, t) 7→ (FX (x, t), FY (y, t))

is clearly a homotopy from 1X×Y to (cX , cY ).

Exercise 1.5. It is clear that X is compact. After all, any open cover of X must contain some set U
containing 0, and thus containing cofinitely many elements of X.
If we have a map h : X → Y , then because Y is discrete, we know that {h−1 (y) : y ∈ Y } is an open
covering of X and thus by compactness admits a finite subcovering. Thus there are only finitely many
elements of y in the image of h.
Now suppose that f : X → Y is a homotopy equivalence. Then there exists some g : Y → X with a
homotopy H : f ◦ g ' 1Y . But H({y} × I) is the continuous image of a connected map and is therefore itself
connected. Because Y is discrete, this means that H(y, 0) = H(y, 1) for all y. But we know that f has finite
image, and Y is infinite, so there exists some y such that y 6∈ im f . In particular, we have y 6= f (g(y)), and
so H(y, 0) = f (g(y)) 6= y = 1Y (y), a contradiction. Thus X and Y are not of the same homotopy type.

7
Exercise 1.6. Suppose X is contractible, with F : c ' 1X , where c is the constant map at p. Note that, for
every x ∈ X, there is a path F (x, t) : {x} × I → X taking x to p ∈ X. In particular, this means that every
x is in the same component as p, proving connectedness.
1
Exercise 1.7. The map H : X → I → X taking (x, t) to x and (y, t) to x if and only if t > 2 works. Indeed,
note that H −1 ({x} × I) is simply {x} × I ∪ {y} × ( 21 , 1], which is open in X × I.
Exercise 1.8.
(i) Consider the map taking the unit interval to S 1 given by t 7→ e2πit .
(ii) If r : Y → X is a retraction, then we know from 1Y ' c that r ◦ 1Y ◦ i ' r ◦ c ◦ i, where i is the injection
X ,→ Y . But the left side is simply r ◦ i = 1X , while the left side is a constant map, proving the result.
Exercise 1.9. We know that there exists some constant map c with f ' c. But then g ◦ f ' g ◦ c, and the
right side is a constant map. Thus g ◦ f is also nullhomotopic.
Exercise 1.10. First, suppose that g is an identification. Note that (gf )−1 (U ) open in X implies that
g −1 (U ) is open in Y because f is an identification. But the hypothesis on g implies that U is open in Z.
Since gf is clearly a continuous surjection, the result follows.
Now, suppose that gf is an identification. It suffices to prove that g −1 (U ) ⊆ Y open implies that U ⊆ Z
is open. But we know by continuity of f that f −1 (g −1 (U )) is open, and so gf being an identification implies
the result.
Exercise 1.11. First, note that this is a well-defined function in the sense that [x] = [y] in X/∼ implies
that f ([x]) = f ([y]).
This is evidently continuous. After all, suppose that U ⊆ Y / is open. Then we know that
−1
f (U ) = {[x] ∈ X/∼ : [f (x)] ∈ U } = U 0 .

If we let v : X → X/∼ and u : Y → Y / be the natural maps, then we know that U 0 is open in X/∼ because

v −1 (U 0 ) = {x ∈ X : f (x) ∈ u−1 (U )} = f −1 (u−1 (U ))

is open.
−1
Finally, we will show that f is an identification. It is obviously surjective. Moreover, if U 0 = f (U ) is
−1 0 −1 −1
open in X/∼, then we simply note that a similar argument as above gives us that v (U ) = f (u (U ))
is open. Since f and u are identifications, it follows that U was an open set in the first place, proving the
result.
Exercise 1.12. Note that if K ⊆ Z is closed, then it is compact and so h(K) is compact in Z, hence itself
closed. Thus h is a closed map, and hence an identification.
Now because v : X → X/ ker h is an identification, Corollary 1.9 applies. Indeed, Corollary 1.9 implies
that hv −1 = ϕ is a closed map. Thus it is an identification, i.e., a continuous surjection.
But the same corollary also implies that ϕ−1 = vh−1 is continuous. This, combined with Example 1.3,
in which it was shown that ϕ is injective, proves the result, as ϕ is now a bicontinuous bijection, i.e., a
homeomorphism.
Exercise 1.13. First observe that f (x) = f (y) implies that [x, t] = [y, t] and so t = 1. Thus f is injective
and hence bijective onto its image CXt = {[x, t] ∈ CX : x ∈ X}. Then open sets in CXt are precisely of the
form U ∩ CXt for an open set U ⊆ CX. But clearly we can assume that [x, 1] 6∈ U because [x, 1] 6∈ CXt ,
and thus we wind up with X × [0, 1), where CXt = X × {t}. This is obviously homeomorphic to X.
Exercise 1.14. The functor takes a map f : X → Y to Cf : CX → CY given by C([x, t]) = [f (x), t]. Note
that this is well-defined. Moreover, it is obvious that this is satisfies the properties of a functor. Indeed, if
g : Y → Z, then
C(g ◦ f )([x, t]) = [g(f (x)), t] = ((Cg) ◦ (Cf ))([x, t])
and clearly C(1X ) is the identity on CX.

8
Paths and Path Connectedness
Exercise 1.15. Using the hint, suppose that g : I → X is a path with g(0) = (0, a) ∈ A and with g(t) ∈ G
for all t > 0. Then note that πi ◦ g is continuous for i = 1, 2, where πi are the projections to the x- and
y-axes. This implies the existence of an ε > 0 such that t ∈ (0, ε) implies that g(t) = (x(t), sin(1/x(t))) has
x(t), | sin(1/x(t)) − a| < δ. But this is obviously impossible, as sin(1/x(t)) will oscillate wildly between −1
and 1.
Exercise 1.16. Let (ai ) and (bi ) be points in S n . We will construct n paths which, when joined together in
the customary fashion (i.e., by traversing each of the n − 1 subpaths in 1/(n − 1) time), will give us a path
from (ai ) to (bi ).
The first path f1 is defined as

f1 (t) = ((1 − t)a1 + tb1 , c2 , a3 , a4 , . . . , an ),

where c2 is chosen to be of the same sign as a2 and in such a way that f (t) ∈ S n . Note that such a c2 always
exists.
In general, for 1 ≤ i ≤ n − 1, the path fi will fix every coordinate except for the i-th, which it will take
to bi , and the (i + 1)-th, which we use as a “free” coordinate to allow for such adjusting. Moreover, observe
that if the first n − 1 coordinates of two points on S 1 are the same, then the n-th coordinates either will be
the same or will be negatives.
If joining the paths f1 , f2 , . . . , fn−1 together gives a path from (ai ) to (bi ), then we are done. Note that
this occurs if an and bn have the same sign.
Otherwise, construct a path g which adjusts the n-th coordinate and uses the (n − 1)-th coordinate as a
“free” one, preserving the sign. This effectively allows us to switch the sign of the n-th coordinate so that
the n-th coordinate is just bn . Moreover, because we preserved the sign of the (n − 1)-th coordinate, it is
still equal to bn−1 .
Exercise 1.17. It suffices to show the forward direction, so suppose that U is not path connected. Then
there are at least two path components.
We will show that each path component is open, which will prove that U is not connected. But because
U is open, we know that open sets in U (as a subspace) or also open in Rn . Thus, for every x ∈ U , there is
a ball Bx centered at x and contained in U . This ball is obviously path-connected. As such, if x is in the
path component A, it must follow that Bx ⊆ A, proving that A is open.
Exercise 1.18. We know that if X is contractible then there exists a point c ∈ X such that 1X is homotopic
to the constant map at c from X to itself. Now consider the map c : I → X satisfying c(t) = c for all t.
In the proof of Theorem 1.13, we saw that any path is homotopic to c. In particular, the constant maps
x : I → X and y : I → X at x and y, respectively, are both homotopic to c. Note that these give rise to
paths from x to c and from c to y, respectively, which in turn give rise to a path from x to y. This proves
path connectedness.

Exercise 1.19.
(i) If X is path connected, then let c and c0 be constant maps. Let f be a path from (the point) c to (the
point) c0 and define H : X × I → X as H(x, t) = f (t). Then H is a homotopy from c to c0 .
For the reverse direction, let H be a homotopy from c to c0 and define the path f : I → X as
f (t) = H(c, t).
(ii) Let f : X → Y be a continuous function. Fix some y0 ∈ Y and consider the map

H :X ×I→Y
(x, t) 7→ px (t),

where px is a path from f (x) to y0 . This is a homotopy from f to the constant map mapping X to y0 .
But if g : X → Y is another continuous function, then the same argument shows that g ' y0 , and so
f ' g, as desired.

9
 Exercise 1.20. It suffices to show that if a ∈ A and b ∈ B, then there is a path from a to b. But fix some
point x ∈ A ∩ B. Then there is a path from a to x, and a path from x to b. Joining the two paths gives a
path from a to b.
Exercise 1.21. This is simply done by noting that for any (x, y), (x0 , y 0 ) ∈ X × Y , we can join the paths
f (t) = ((1 − t)x + tx0 , y) and g(t) = (x0 , (1 − t)y + ty 0 ).
Exercise 1.22. Suppose f (a), f (b) ∈ Y . Then let p be a path from a to b in X. Now simply note that
q(t) = f (p(t)) is a path from f (a) to f (b), proving the result.
Exercise 1.23.
(i) We already know that there are at least two path components because the entire space is not path
connected. Moreover, both A and G are path connected, and so it follows that they must themselves be
the path components.
 1

(ii) Simply note that the sequence nπ , sin(nπ) ⊂ G approaches (0, 0) ∈ A.
(iii) As per the hint, consider U to be the open disk with center (0, 12 ) and radius 14 . Then X ∩ U is open in
1
X. But note that v(X ∩ U ) is not open in X/A ≈ [0, 2π ]. After all, note that any ball Bε around the
1
point 0 (which is the image of A under the natural map in this case) must contain some point nπ < ε.
1 1


But nπ , which corresponds to the point nπ , 0 ∈ X \ U , is not contained in v(X ∩ U ).
Exercise 1.24. By definition, path components are path connected. Moreover, if C is a path component
and there exists some point x ∈ X and c ∈ C so that there is a path between x and c, then the definition of
path components implies that x ∈ C. Thus path components are maximally path connected.
Finally, suppose that A is path connected and pick a ∈ A. There exists a unique path component C such
that a ∈ C. Then for all b ∈ A, we know that there is a path between a and b, and so b ∈ C. Thus A ⊆ C,
as desired.
Exercise 1.25. Simply use Exercise 1.22 and observe that I is path connected.
Exercise 1.26. Note that, if X is locally path connected, then for all x ∈ X, there exists some open path
connected, hence connected, neighborhood V of x. Alternatively, note that if U ⊆ X is open, then its
components are unions of its path components and thus open.
Exercise 1.27. Given any open subset U of X × Y containing a given point (x, y) ∈ X × Y , there must exist
a basic open neighborhood Ux × Uy ⊆ U of (x, y). Then we know that there exists some path connected Vx
with x ∈ Vx ⊆ Ux , and similarly for y. Then Vx × Vy is path connected by Exercise 1.21. The result follows.
Exercise 1.28. Note that open subsets of open subsets are open in the main space. In particular, let
A ⊆ X be open. Given any x ∈ A, let U be an open neighborhood of x in A. Note that this is also an open
neighborhood in X, and so there exists an open path connected V in X (and hence open in A as well) such
that x ∈ V ⊆ U .
Exercise 1.29. Consider the map F : (Rn+1 \ {0}) × I → Rn+1 \ {0} given by
" #
t
F ((xi ), t) = (1 − t) + pP 2 (xi ).
xi

This is evidently a homotopy which makes S n a deformation retract.

Exercise 1.30. The exact same map as in Exercise 1.29 works for this case.
Exercise 1.31. It is easy to see that the deformation retract of a deformation retract is a deformation
retract, either by a direct argument or by applying Theorem 1.22. Thus the previous exercise implies that it
suffices to show that Dn \ {0} is a deformation retract of S n \ {a, b}. But the map (xi ) 7→ (x1 , . . . , xn−1 , 0)
is exactly the map needed, and so we are done.
Exercise 1.32. If H : f0 ' f1 , then the map H 0 : (y, t) 7→ H(r(y), t) is a homotopy from f˜0 to f˜1 .

10
 Exercise 1.33. Let Y = {y} and observe that (x, 1) ∼ y for all x ∈ X. Thus (x, 1) ∼ (x0 , 1) for alL x, x0 ∈ X.
Moreover, this is the only equivalence. Thus Mf is precisely the quotient space (X × I)/(X × {1}) = CX.
Exercise 1.34.
(i) We first tackle i. It is obvious that i is injective, and thus a bijection onto its image i(X) = {[x, 0] : x ∈
X}. Moreover, the open sets in i(X) are precisely of the U ∩ i(X) for open sets U in Mf .
Note that we can suppose without loss of generality that U is contained in v(X × [0, 1)), where v is
the natural map. Thus U simply looks like the Cartesian product of an open interval with an open set of
X. This proves that i is a homeomorphism, for the open sets of i(X) map exactly to the open sets of X.
We can show that j is a homeomorphism onto j(Y ) in a similar manner. The main idea is simply
that y 6∼ y 0 for any y, y 0 ∈ j(Y ).
(ii) It is obvious that (rj)(y) = r[y] = y = 1Y (y) for any y ∈ Y . It is also clearly continuous by the gluing
lemma. Thus r is indeed a retraction.
(iii) Define F : Mf × I → Mf as suggested in the hint. It is evident that F is continuous. Moreover, for any
[x, t] ∈ Mf , we know that

F ([x, t], 0) = [x, t]

F ([x, t], 1) = [x, 1] = [f (x)] ∈ Y.

Similarly, if [y] ∈ Y , then the definition implies that the remaining criteria for this homotopy to induce
a deformation retraction r(x) = F (x, 1) are satisfied.
(iv) Note that Rotman writes that f is homotopic to r ◦ i; in fact, we can and do prove the stronger statement
that f coincides with r ◦ i.
Let f : X → Y be continuous. Then it is clear that the map f = r ◦ i, where i : X → Mf is an
injection and r : Mf → Y is the retraction taking [x, t] to [f (x)] and taking [y] to itself, proving the
result.

11
2 Simplexes1
Affine Spaces
Exercise 2.1. Note that there is a maximal affine independent subset S of A. This is directly implied by
the fact that any set of greater than n + 1 elements is not affine independent. Hence we can take an affine
independent subset of A with maximum size (because the empty set is affine independent).
Write S = {p0 , . . . , pm }. Then let pm+1 ∈ A \ S. By maximality of S, we know that S ∪ {p} is not affine
independent. Hence there exist si not all 0 such that
m+1
X m+1
X
si pi = 0, si = 0.
i=0 i=0
Pm
Note that the second equation implies i=0 si 6= 0 for some i < m + 1. It follows then that
m  
X si
Pm pi = pm+1 .
i=0 i=0 si

But we know that

m
s
Pmi
X
= 1,
i=0 i=0 si

and so it follows that pm+1 is in fact in the affine span of S.

Exercise 2.2. Let ϕ be the isomorphism from Rn to a subset of Rk . Suppose A ⊆ Rn is an affine set
containing X. Then ϕ(X) ⊆ ϕ(A) ⊆ Rk .
Moreover, we claim that ϕ(A) is affine. After all, for any ϕ(x), ϕ(x0 ) ∈ ϕ(A) and any t ∈ R, the point
tϕ(x) + (1 − t)ϕ(x0 ) = ϕ(tx + (1 − t)x0 ) ∈ ϕ(A) because A is affine.
This implies that the intersection of all affine sets in Rn containing X must contain the intersection of all
affine sets in ϕ(Rn ) containing ϕ(X). Because ϕ is an isomorphism, using ϕ−1 gives the reverse inclusion.
Thus the affine set spanned by X in Rn is precisely the same as that spanned by X in Rk .
Exercise 2.3. This is evident in the case n = 0.
Suppose it is true for n − 1 and consider the canonical injection ι : S n−1 ,→ S n which takes (x0 , . . . , xn−1 )
to (x1 , . . . , xn−1 , 0). It is obvious that we can pick n+1 affine independent points p0 , . . . , pn in this embedding.
Now consider the point pn+1 = (0, . P . . , 0, 1) ∈ S n . Notice
P that the last coordinate of each pi for i 6= n + 1
is zero. Thus suppose we have si with si pi = 0 and si = 0. Then sn+1 = 0, and so this reduces to the
n − 1 case. Affine independence of {p0 , . . . , pn } proves the result.

Affine Maps
Exercise 2.4. Consider the map T 0 (x) = T (x) − T (0). We claim that T 0 is a linear map.
Observe that S = {ei } ∪ {0} spans Rn . Thus we can write any point as the affine sum of elements of S.
Note that the coefficient of the zero vector is flexible, and so we have effectively no restrictions on the sum
of the coefficients.
si ei + s · 0 in Rn , where r = 1 − ri and similarly for
P P P
Consider arbitrary elements ri ei + r · 0 and
s. Let R, S ∈ R. Then note that
 X X  X 
T0 R ri ei + S si ei = T 0 (Rri + Ssi )ei
X  X  
=T (Rri + Ssi )ei + 1 − (Rri + Ssi ) · 0 − T (0)
X X X X
=R ri T (ei ) + S si T (ei ) − R ri T (0) − S si T (0).
1I usually use simplices as the plural of simplex, but Rotman doesn’t; no matter.

12
Considering the R-terms first, simply observe that we can add and subtract RT (0) to give us that
X X  X  
R ri T (ei ) − R ri T (0) = R T ri T (ei ) + r · 0 − T (0) .

This is simply RT 0 (
P
ri ei ). A similar result holds for the S-terms, from which we conclude that
 X X  X  X 
T0 R ri ei + S si ei = RT 0 ri ei + ST 0 si ei ,

proving linearity.
Exercise 2.5. This is obvious from the previous exercise and continuity of linear maps.
Exercise 2.6. Given two m-simplexes [p0 , . . . , pm ] and [q0 , . . . , qm ], the map f taking pi to qi for every i
is a homeomorphism. Bijectivity is obvious by the definition. Continuity is clear by how we extend f from
{pi } to [pi ]. Finally, the inverse is of the same form as f , only with the qi ’s taking the place of the pi ’s and
vice versa; thus f −1 is also continuous.
Exercise 2.7. The following map works:
t2 − t1
f : x 7→ (x − s1 ) + t1 .
s2 − s1
Exercise 2.8. Pick arbitrary T (x), T (x0 ) ∈ T (X) and observe that

tT (x) + (1 − t)T (x0 ) = T (tx + (1 − t)x0 ) ∈ T (X).

Thus T (X) is affine if X is affine, and convex if X is convex. The second statement of the exercise follows
by noting that ` is convex.
Exercise 2.9. Without loss of generality, we delete p0 . Now suppose that
m
X m
X
si pi + sb = 0, si + s = 0.
i=1 i=1

Then we know by definition of the barycenter b that

m m
X s X
si pi + pi = 0.
i=1
m + 1 i=0
Pm Pm
Moreover, letting s0i be the coefficient of pi in the above equation, it is obvious that i=0 s0i = s+ i=1 si = 0.
Thus s0i = 0 for all i because {p0 , . . . , pm } was affine independent. But then we conclude that 0 = s00 = m+1 s
,
0 s
and so s = 0. For every i ∈ {1, . . . , m}, we have 0 = si = m+1 + si . Thus s = 0 implies si = 0 for every i,a
nd so it follows that {b, p1 , . . . , pm } is affine independent, as desired.
P
Exercise 2.10. Once Pmagain, suppose without loss of generality that i = 0. Then the map taking ti pi ∈
[p0 , p1 , . . . , pm ] to ( i=1 ti pi , t0 ) works. Note that this actually requires the affine independence of the pi ’s,
as well as the fact that the coefficients ti are all between 0 and 1.
Exercise 2.11. Notice that [0, e1 , . . . , en ], where ei are the standard basis vectors in Rn , is an n-simplex.
Thus there is a homeomorphism [p0 , . . . , pn ] → [0, e1 , . . . , en ]. If we translate the image by v = (− 41 , . . . , − 14 ),
then we can map the result to Dn by taking a radial mapping. In particular, this map will take
v
p0 7→
kvk
ei + v
pi 7→ for i 6= 0.
kei + vk
Note that this extends to a homeomorphism.

13
3 The Fundamental Group
The Fundamental Groupoid
Exercise 3.1. The homotopy H : X × I → Z given by
(
g0 (F (x, 2t)) if t ≤ 21 ,
H : (x, t) 7→
G(f1 (x), 2t − 1) if t ≥ 21

works. Continuity follows because g0 (F (x, 1)) = G(f1 (x), 0).

Moreover, this homotopy is indeed rel A. For a detailed argument why this is so, simply suppose that
a ∈ A and t ∈ I. If t ≤ 21 , then F (a, 2t) = f0 (a) by definition of F . Hence H(a, t) = g0 (f0 (a)).
Similarly, we can show that if t ≥ 21 , then H(a, t) = g1 (f1 (a)). This follows because f1 (a) ∈ B and G is
a homotopy rel B.
It thus suffices to show that g0 (f0 (a)) = g1 (f1 (a)). But this is obvious because f0 and f1 agree on A,
and g0 and g1 agree on B ⊇ f0 (A).
Exercise 3.2.
(i) First, note that f 0 is well-defined because f (0) = f (1). It is obvious by continuity of f and ln that f 0 is
continuous.
Moreover, consider the map
H 0 : e2πiθ , t 7→ H(θ, t).

This is clearly continuous, for the same reasons that f 0 was continuous. If t = 0, clearly H 0 (e2πiθ , t) =
H(θ, 0) = f (θ) = f 0 (e2πiθ ), and similarly for t = 1. Thus H is indeed a homotopy from f 0 to g 0 .
To see that it is a homotopy rel{1}, simply note that e2πiθ = 1 corresponds to θ = 0, 1. Thus it
follows that
H 0 (1, t) = H(1, t) = f (1)
for all t, proving the result.
(ii) Theorem 3.1 implies that f ∗g ' f1 ∗g1 rel İ. Using the previous part, we find that (f ∗g)0 ' (f1 ∗g1 )0 rel{1}.
Now, using the observation that (f ∗ g)0 = f 0 ∗ g 0 , we find that f 0 ∗ g 0 ' f10 ∗ g10 rel{1}, as desired.

Exercise 3.3. The forward direction is trivial.

For the converse, note that g 0 is a constant map, and so f 0 is nullhomotopic. Then Theorem 1.6 implies
that f 0 ' g 0 rel{1}. In particular, note that g 0 : S 1 → X takes every element of S 1 to g 0 (1) = g(0) = x0 .
Observe that f 0 (1) = x0 as well, and so it follows that f 0 ' g rel{1}, as desired.
Exercise 3.4.

(i) Instead of applying Theorem 1.6, I constructed an explicit homotopy. (If you are interested in a proof
using Theorem 1.6, my guess would be that it relies on the fact that ∆2 ≈ D2 . However, I have not gone
through the details.)
The effective idea of the homotopy I constructed is to, at time t ∈ [0, 1], return the function which
traverses the first t units of the face opposite e0 , then goes along a segment to the point t units away
from e1 on the fact opposite e2 , before returning back to e1 , as shown in the red path below.

e2

e0 e1

14
 The specific homotopy H : I × I → X from (σ0 ∗ σ1−1 ) ∗ σ2 to the constant map at e1 is as follows:

σ0 (4(1 − t)x)
 if x ≤ 41 ,
H(x, t) = σ((1 − x)ε0 (1 − t) + xε2 (t)) if 41 ≤ x ≤ 21 ,
if x ≥ 12 .

σ(2tx − (2t − 1))


We leave it to the reader to check that this works.

(ii) One can generate a similar homotopy, which we do not do here.
(iii) This time, we use the homotopy which goes up along γ for t units, before going parallel to β and coming
back down along δ −1 . The particular formula is as follows:

F (0, 4tx)
 if x ≤ 41 ,
H(x, t) = F (4x − 1, t) if 14 ≤ x ≤ 21 ,
F (1, 2t(1 − x)) if 12 ≤ x.



Once again, we leave the details to the reader to check.

Exercise 3.5. Simply use the homotopy H : I × I → X × Y which takes (s, t) to (F (s, t), G(s, t)). This
is clearly a homotopy from (f0 , g0 ) to (f1 , g1 ). To see that it is still rel İ, simply observe that H(0, t) =
(F (0, t), G(0, t)). Because F and G are both rel İ, it follows that H(0, t) never changes. A similar argument
shows that H(1, t) is always the same, and so H is indeed a homotopy rel İ.

Exercise 3.6.
(i) It is obvious that the homotopy H 0 : (x, t) 7→ H(x, 1 − t) works.
(ii) This is just some slightly annoying manipulation. In particular, note that
(
f (2x) if x ≤ 21 ,
(f ∗ g)(x) =
g(2x − 1) if x ≥ 12 .

By replacing x with 1 − x to get the inverse, we find that

(
−1 f (2 − 2x) if x ≥ 21 ,
(f ∗ g) (x) =
g(1 − 2x) if x ≤ 12 .

However, note that

(
−1 −1 g −1 (2x) if x ≥ 12 ,
(g ∗f )(x) =
f −1 (2x − 1) if x ≥ 21
(
g(1 − 2x) if x ≤ 21 ,
=
f (2 − 2x) if x ≥ 12 .

Thus the two are indeed the same.

(iii) Take the closed path f (t) = e2πit on S 1 . Then note that (f ∗ f −1 )( 18 ) = f ( 14 ) = i, while (f −1 ∗ f )( 18 ) =
f −1 ( 41 ) = −i.
(iv) Suppose ip ∗ f = f and f is not constant. Note that continuity implies that there must exist some
0 < t < 1 so that f (t) 6= p. Thus there exists some k ∈ N so that t < 1 − 2−k .
We claim, however, that f must be constant on [0, 1−2−n ] for every n ∈ N. We prove this inductively.
Clearly, it is true on n = 0. If it is true on n − 1, then we know that ip ∗ f must be equal to p on [0, 21 ],
as well as on [ 21 , 1 − 2−n ] (note that 1 − 2−n comes from 2(1 − 2−n ) − 1, which itself comes from the
equation for the star operator). Thus f is constant on [0, 1 − 2−n ], as desired.
Thus it follows that f (t) = p, a contradiction. Thus f must have been constant in the first place.

15
The Functor π1
Exercise 3.7. Recall that we defined the sin(1/x) space as the union of A = {(0, y) : −1 ≤ y ≤ 1} and
G = {(x, sin(1/x)) : 0 < x ≤ 1/2π}. We also know that A and G are the path components of the sin(1/x)
space. Moreover, both A and G are contractible, and so every path in either A or G is nullhomotopic. In
particular, we conclude that the fundamental group at any basepoint is trivial.
Exercise 3.8. Let X be the sin(1/x) space. We know that CX is contractible. But consider an open
ball around the point x = ((0, 0), 0), that is, the point (0, 0) on the “zeroth” level of the cone. Consider a
small neighborhood (not including the points (t, 1), in particular) around this point and pick some element
y = ((ε, sin(1/ε)), 0) in the neighborhood. Now observe that any path between x and y can be projected
down to a path between (0, 0) and (ε, sin(1/ε)) in X, which we know does not exist. Hence CX is contractible
but not locally path connected.
Exercise 3.9. Note that composition is associative because ◦ is. Moreover, the path class of the trivial loop
based at p is the identity on p. Thus this is a category.
To see that each morphism in C , simply note that the inverse path, i.e., the path f −1 taking t to f (1 − t),
gives a path class [f −1 ] which works as an inverse to [f ] ∈ Hom(p, q).

Exercise 3.10. We simply let π0 take (X, x0 ) ∈ Sets∗ to the set of all path components of X, with basepoint
equal to the path component containing x0 . It takes a morphism f ∈ Hom((X, x0 ), (Y, y0 )) to the map π0 (f )
which takes each path component A of X to the path component B of Y which contains f (A).
Note that this is possible because continuous images of path connected spaces are path connected and
hence contained within a single path component of Y . Moreover, this is indeed a pointed map because the
path component containing x0 must be contained in the path component containing f (x0 ) = y0 , which is
the basepoint of π0 ((Y, y0 )).
It is easy to check functoriality, completing the proof.
Exercise 3.11. Evidently the only possible path is the constant path at x0 . Hence π1 (X, x0 ) is the trivial
group, i.e., {1}.

π1 (S 1 )
Exercise 3.12. Note that 1S is a loop based at 1, i.e., an element of π1 (S 1 , 1). Thus if π1 (S 1 , 1) were trivial,
then 1S would be nullhomotopic. The hint gives the rest of the solution.

Exercise 3.13. We know that deg u = 1. Since 1 is a generator for Z, it follows that [u] generates π1 (S 1 , 1).
Exercise 3.14. Let γ̃(t) = mf˜(t), where f˜ is the lifting of f satisfying f˜(0) = 0. Now simply observe that
 m
exp γ̃(t) = exp f˜(t) = f (t)m

and γ̃(0) = 0. Thus γ̃ is indeed the lifting of f m taking 0 to 0, and so we conclude that

deg(f m ) = γ̃(1) = mf˜(1) = m deg f.

Exercise 3.15. Note that Exercise 1.3 implies that there is a homotopy F : Rf ◦ f ' f , where Rf is the
rotation associated with f . Moreover, from the proof of that same exercise, it follows that F gives a closed
path at every time t. Similarly, we have G : g ' Rg ◦ g. Thus if H : f ' g where H gives a closed path at
every time t, then the homotopy which follows F , then H, and finally G is a homotopy between Rf ◦ f and
Rg ◦ g. Thus Corollary 3.18 implies that f and g have the same degree.
For the converse, simply use Corollary 3.18 to show that deg f = deg g implies that there is a homotopy
rel İ taking Rf ◦ f to Rg ◦ g. Then using F and G defined above, it is clear that g ' Rg ◦ g ' Rf ◦ f ' f .
Exercise 3.16. Theorem 3.7 implies that π1 (T, t0 ) = Z × Z = Z2 .

16
Exercise 3.17. Because D2 is contractible, its fundamental group is trivial. Thus if there were to exist a
retraction r : D2 → S 1 , then r∗ : π1 (D) → π1 (S 1 ) would be a constant. But then, letting i : S 1 → D2 be the
canonical injection, we would have that (r ◦ i)∗ = r∗ ◦ i∗ is a constant. However, we also know that r ◦ i is
the identity on S 1 , and so (r ◦ i)∗ is the identity on π1 (S 1 ), which is not a constant. This is a contradiction,
from which we conclude that S 1 is not a retract of D2 , as desired.
Exercise 3.18. This was proved in Theorem 0.3, which required only the fact proved in the above problem,
namely that S 1 is not a retract of D2 .
Exercise 3.19.
(i) Let f˜ be the unique lifting of f with f˜(0) = 0. Then if f˜(1) ≥ 1, the intermediate value theorem implies
that every point in the interval [0, 1] ⊂ R is in the image of f˜. But this implies that f = exp ◦f˜ must be
surjective, a contradiction.
(ii) Consider the map which traverses the circle once counterclockwise, reaching the point 1 at time t = 21 ,
before looping back and making a clockwise rotation. Clearly it is surjective. However, it is composed of
two loops, one of which has degree 1 and one of which has degree −1. Because deg(f ∗ g) = deg f + deg g,
it follows that this map has degree 0.
Exercise 3.20. As per the hint, consider an arbitrary closed path f in X and let λ be a Lebesgue number of
the open cover {f −1 (Uj ) : j ∈ J} of I. Note that λ exists by the Lebesgue number lemma and compactness
of the unit interval. Picking N ∈ N with N > 1/λ, it follows that if we subdivide I into N equal intervals
k k+1
Ik = [ N , N ], then f (Ik ) ⊆ Ujk for some jk ∈ J.
Define fk as the path in Ujk obtained by restricting
f to Ik and then stretching suitably so that the
domain is all of I. With notation, define fk (t) = f k+t N ∈ Ujk . Because fk is a path in Ujk , it follows that
[fk0 ] = [ijk ◦ fk ] ∈ im(ij )∗ . But now simply observe that [f00 ∗ · · · ∗ fN
0
−1 ] = [f ], implying that [f ] is contained
in the group generated by the subsets im(ij )∗ . This proves the result.
Exercise 3.21. Let U1 and U2 be defined as in the hint, and let ik be the injection from Uk to S n for
k = 1, 2. Observe that, by the previous exercise, it suffices to show that im(ik )∗ is trivial for k = 1, 2.
Without loss of generality, let k = 1. But we know that (i1 )∗ takes a closed path f : I → U1 to the
path class [i1 ◦ f ]. (Note that the basepoint doesn’t really matter for us as long as it is neither the north
nor the south pole.Thus we omit it.) Because U1 ≈ Dn and is therefore contractible, it follows that f is
nullhomotopic. In particular, we know that i1 ◦ f is nullhomotopic, and so [i1 ◦ f ] = [1] for every f . Thus
im(i1 )∗ is trivial, and similarly for k = 2, proving the result.
Exercise 3.22. Corollary 3.11 implies that path connected spaces of the same homotopy type must have
isomorphic fundamental groups. But obviously Z 6∼
= {1}, and so S 1 and S n do not have the same homotopy
type for n > 1.
Exercise 3.23. The multiplication map µ on G/H is continuous. After all, if we let v be the natural map,
then for any open set U ⊆ G/H, we have
µ−1 (U ) = {([x], [y]) : xy ∈ v −1 (U )}.
But this set is open in G/H × G/H because the set consisting of elements (v −1 ([x]), v −1 ([y])) for each
([x], [y]) ∈ µ−1 (U ) is just µ−1 (v −1 (U )), which is clearly open.
For the inversion map i : G/H → G/H, a very similar argument holds. In particular, for any open set
U ⊆ G/H, we have
v −1 (i−1 (U )) = {x−1 : x ∈ v −1 (U )}.
Thus v −1 (i−1 (U )) is open, and so i−1 (U ) is open, proving continuity.
Exercise 3.24. First, we will show that we can lift a loop f : (I, 0) → (G/H, 1) to a unique continuous map
f˜ : (I, 0) → (G, h) for any h0 ∈ H, as shown below.

(G, h0 )
f˜
v

(I, 0) f
(G/H, 1)

17
In the above diagram, the map v is the natural map taking g to the coset gH ∈ G/H.
First, we will find a suitable neighborhood U of 1 such that the family {hU : h ∈ H} is pairwise disjoint.
Discreteness of H implies that there exists an open neighborhood V of 1 with V ∩ H = {1}. It is clear that
the map ϕ : (x, y) 7→ xy −1 is the composition µ ◦ (id ×i) and is therefore continuous. Thus ϕ−1 (V ) ⊆ G × G
is an open neighborhood of (1, 1). This implies that we can find an open neighborhood U of 1 such that
U × U ⊆ f −1 (V ).
Now suppose that there are h1 , h2 ∈ H and x, y ∈ U with h1 x = h2 y. But this would require that
xy −1 = h−11 h2 . It is clear that xy
−1
∈ ϕ(U ) ⊆ V . Moreover, because H is a subgroup, we know that
−1 −1
h1 h2 ∈ H, and so xy ∈ V ∩ H. Thus x = y and h1 = h2 , proving that the sets hU are disjoint, as desired.
Note that any translate Ug = gU of U is a neighborhood of g ∈ G and has {hUg : h ∈ H} disjoint.
Note that v is an open map, and so the set W = v(U ) ⊆ G/H is open. Moreover, because v|U is the
restriction of a continuous open map to an open set, it follows that v|U is itself continuous and open. It is
also a bijection onto W , and so v|U : U → V is a homeomorphism.
Note that the collection of sets V [g] for [g] ∈ G/H forms an open cover of G/H. Thus, if we are given
some f : (I, 0) → (G/H, 1), then we can consider the open cover

{f −1 (V [g]) : [g] ∈ G/H}

of I. Note that we can find a finite subcover of this open cover. This means that we can take subsets of the
sets in this open cover, given us a finite collection open overlapping subintervals which are, in order of their
smaller coordinate, labeled I1 , . . . , Ik . Let the group elements g1 , . . . , gk be such that Ij ⊆ f −1 (V [gj ]). This
is simply because I = [0, 1] is connected compact.
Now we can lift f to each interval f −1 (V [g]) in this finite subcover. Note that 0 = t1 ∈ I1 ⊆ f −1 (V [g1 ]).
Moreover, we know that v −1 (V [g1 ]) consists of disjoint unions of U , and so we can pick the one containing
h0 . Now, for each t ∈ I1 , we let f˜(t) to be the unique element in this copy of U such that v(f˜(t)) = f (t).
Because the intervals overlap, we know that there is some t2 ∈ I2 ∩ I1 , and so we can do the same thing, all
the way to tk . This lets us define f˜(t) for all t ∈ I, and it is easy to show that our construction is indeed a
lifting satisfying the commutative diagram above.
Now consider the map d : π1 (G/H, 1) → H taking a loop [f ] to d([f ]) = f˜(1), where f˜ is the unique lifting
of f with f˜(0) = 1. It is obvious that im d ⊆ H because v(f˜(1)) = f (1) = [1] implies that d([f ]) = f˜(1) ∈ H.
Moreover, the reverse inclusion holds, showing surjectivity. In particular, if h ∈ H, then path connectedness
of G implies that there is a path f˜ from 1 to h. Taking its projection f = v ◦ f˜, note that f is a loop
because v(f˜(1)) = v(h) = [1]. Thus d([f ]) is defined and equal to h. To show injectivity, simply note that
f˜(1) = 1 implies that f˜ is a loop in G. Because G is simply connected, however, it follows that f˜, and
hence f , is nullhomotopic. Thus f ∈ ker d implies that [f ] = [1]. Finally, we must show that d is indeed a
homomorphism, i.e., that d(f ∗ g) = d(f )d(g). But this is clear if we lift f to f˜ with f˜(0) = 1, and if we lift
g to g̃ with g̃(0) = d(f ). This follows the same proof layout as Theorem 3.16, and proves the result.
Exercise 3.25. If S ⊆ GL(n, R) is a subgroup of GL(n, R), then note that µ : S × S → S is continuous.
After all, the product, entrywise, is simply a polynomial, and so µ is a polynomial in each of its n2 entries.
Since polynomials are continuous in R2 , it follows that each of the n2 components of µ is continuous. Hence
µ is continuous.
To see that the inversion i is continuous, observe that the determinant det A is a continuous function,
since it too is a polynomial (and is never zero, by definition of GL). It thus suffices to show that the function
A 7→ adj A is continuous. But it is easy to see that the adjugate matrix, which is the transpose of the cofactor
matrix, is also a polynomial in the entries of A, and so adj A is continuous too. Thus i is continuous, and so
S is a topological group, as desired.

Exercise 3.26. As hinted in the exercise, fix h0 ∈ H and let ϕ : G → H be the map taking x to xh0 x−1 h−1
0 .
Note that xh0 x−1 ∈ H because H is normal, and so ϕ(x) is indeed an element of H. Moreover, we know
that ϕ is continuous and {h} ⊆ H is open for each h ∈ H. Thus {ϕ−1 (h) : h ∈ H} is an open cover of G
consisting of disjoint open sets.
−1
S are two elements h1 , h2 ∈ H such that ϕ (hi ) 6= ∅ for i = 1, 2, then setting
In particular, if there
A = ϕ−1 (h1 ) and B = h6=h1 ϕ−1 (h) will give us two disjoint open sets A and B that cover G. This implies
that G is disconnected, a contradiction. Thus for all but one element of H, we must have ϕ−1 (h) = ∅,

18
proving that ϕ is constant. But obviously, setting x = h0 , we find that ϕ(x) = 1. Thus xh0 x−1 h−1
0 = 1 for
all x ∈ G, and so xh0 = h0 x for each h0 ∈ H. This proves the result.

19
 4 Singular Homology
Holes and Green’s Theorem
No exercises!

Free Abelian Groups

P
Exercise 4.1. If γ ∈ F , then we can write γ = b∈B mb b, where mb ∈PZ is zero for almost all b. Now,
P B = ∪Bλ for disjoint Bλ , we can define for each λ the value γλ = b∈Bλ mb b ∈ Fλ . Then obviously
writing
γ = γλ . P 0
To see that this expression is unique, simply observe that if γ = γλ , then because the sums are formal
sums only, it follows that γλ = γλ0 for every λ. But then it follows that the coefficient for each b ∈ Bλ must
be the same in γλ and in γλ0 , and so the two expressions are the same. Moreover, it is clear that almost every
γλ is zero. After all, only finitely many mb ’s are nonzero, and so only finitely many γλ contain a nonzero
coefficient. P P P
Finally, the converse is clear. In particular, if γ = γλ and γλ = b∈Bλ mb b, then γ = b∈B mb b.
Exercise 4.2. To see the forward direction (isomorphic implies same rank), simply restrict to the basis. In
particular, if ϕ : F → F 0 is an isomorphism between two free abelian groups, and if B is a basis for F , then
ϕ(B) is a basis for F 0 . But clearly B and ϕ(B) have the same cardinality because ϕ is injective. Thus F
and F 0 have the same rank.
To see the converse, consider bases B and B 0 for F and F 0 , respectively. Because B and B 0 have the same
cardinality, there is a bijection ϕ|B between them. Pick such a bijection and extend it to all of F linearly.
Theorem 4.1 tells us that this is a homomorphism; indeed, it is an isomorphism because ϕ|B was a bijection.
Exercise 4.3.
P
(i) An arbitrary
P element
P of S1 (X) looks like mσ σ, where σ ranges over paths in X. Then we know that
∂1 takes mσ σ + nσ σ to
X X X X
mσ σ(1) + nσ σ(1) − mσ σ(0) − nσ σ(0) = ∂1 (m) + ∂1 (n),
σ σ σ σ
P
where m = mσ σ and similarly for n. Thus this is a homomorphism.
(ii) If x0 and x1 lie in the same path component of X, then there is a path σ between them. This path is an
element of X (indeed, it is a basis element of X), and satisfies ∂1 (σ) = x1 − x0 .
The converse is slightly trickier, however. Suppose that x0 and x1 belong to different path components,
say X0 and X1 , respectively. Then consider the map ϕ : S0 (X) → Z which takes x ∈ X to 1 if x ∈ X0
and to 0 otherwise. This defines ϕ on the basis of S0 (X), so we can linearly extend it to a group
homomorphism (Theorem 4.1). P P
Any element in the image of ∂1 can be written as ( mσ σ)(1) − ( mσ σ)(0). Then we know that
X X  X
ϕ ( mσ σ)(1) − ( mσ σ)(0) = mσ ϕ(σ(1) − σ(0)).

But because σ is a path, obviously σ(1) and σ(0) are in the same path component. In particular, we have
ϕ(σ(1) − σ(0)) = 0, and so im ∂1 ⊂ ker ϕ. Now observe that ϕ(x1 − x0 ) = −1. Thus x1 − x0 6∈ im ∂1 ,
proving the converse.
(iii) By definition, we have that σ ∈ ker ∂1 if and only if σ(1) − σ(0) = 0. Because σ is a path, however, this
condition is equivalent to saying that σ is a closed path.
To see that the path condition on σ is necessary, note that the sum of two closed paths is in ker ∂1
but is not itself a closed path.

The Singular Complex and Homology Functors

No exercises!

20
Dimension Axiom and Compact Supports
Exercise 4.4. Note that Sn (X) = ∅ for all n, because there is no function ∆n → X = ∅. Thus ker ∂ =
im ∂ = ∅, and so Hn (X) is trivial.
Exercise 4.5. We know that ∂0 is the zero map, and so ker ∂0 = S0 (X). Moreover, the proof of the
dimension axiom shows that ∂1 is the zero map as well. In particular, we find that Z0 (X)/B0 (X) ∼
= S0 (X).
But we know, once again from the proof of the dimension axiom, that S0 (X) is infinite cyclic and hence
H0 (X) ∼
= Z.
Exercise 4.6. We already know how Sn acts on objects of Top. Defining Sn (f ) = f# on morphisms, it is
easy to see that Sn satisfies the functorial properties Sn (1X ) = 1Sn (X) and Sn (g ◦ f ) = Sn (g) ◦ Sn (f ).
Exercise 4.7. We know that S 0 is the disjoint union of two points, and so Hn (S 0 ) = Hn ({0}) ⊕ Hn ({1}).
But the dimension axiom and Exercise 4.5 imply that
(
Z2 if n = 0
Hn (S 0 ) = .
0 otherwise.

Exercise 4.8. Because the Cantor set is the disjoint union of countably many points, it follows that H0 (X) =
Zω and Hn (X) = 0 for all n > 0.

The Homotopy Axiom

Exercise 4.9.
(i) For n = 0, note that β1 = [a0 , b0 ], and so ∂1 β1 is the constant map taking e0 ∈ ∆0 to b0 − a0 =
∆
(e0 , 1) − (e1 , 0). On the other hand, we know that P−1 is the zero map, and λ∆ ∆
i # (δ) = λi . Thus the
right-hand side of the equation is simply
λ∆ ∆
1 − λ0 ,

which is the map taking e0 ∈ ∆0 to (e0 , 1) − (e1 , 0). The two sides are therefore the same.
For n = 1, we first consider the left-hand side. Note that

∂2 β2 = [b0 , b1 ] − [a0 , b1 ] + [a0 , b0 ] − [a1 , b1 ] + [a0 , b1 ] − [a0 , a1 ]

= [b0 , b1 ] + [a0 , b0 ] − [a1 , b1 ] − [a0 , a1 ],

and so it is simply the constant map ∆1 → ∆1 × I taking everything to b0 − a1 = (e0 , 1) − (e1 , 0). For
the right-hand side, on the other hand, we already know that

λ∆ ∆ ∆ ∆
1 # (δ) − λ0 # (δ) = λ1 − λ0 : t 7→ (t, 1) − (t, 0).

Moreover, because ∂1 ∆1 = e1 − e0 , we know that

P0∆ ∂δ : t 7→ ((e1 − e0 )(e0 ), t) = (e1 , t) − (e0 , t).

Thus the right-hand side takes e0 to

(e0 , 1) − (e0 , 0) − (e1 , 0) + (e0 , 0) = (e0 , 1) − (e1 , 0)

and takes e1 to
(e1 , 1) − (e1 , 0) − (e1 , 1) + (e0 , 1) = (e0 , 1) − (e1 , 0).
hus the two sides agree on e0 and e1 , from which we conclude the result.
(ii) We know that

P1X (σ) = (σ × 1)# (β2 )

= (σ × 1) ◦ [a0 , b0 , b1 ] − (σ × 1) ◦ [a0 , a1 , b1 ].

21
 The first term takes an arbitrary element (t0 , t1 , t2 ) ∈ ∆2 , where we use barycentric coordinates, to the
point (σ((t0 + t1 )e0 + t2 e1 ), t1 + t2 ). By corresponding a point (1 − t)e0 + te1 ∈ ∆1 to t, we find that the
first term takes (ti ) to (σ(t2 ), t1 + t2 ). Similarly, the second term takes (ti ) to (σ(t1 + t2 ), t2 ). Thus we
find the following explicit formula:

P1X (σ) : (t0 , t1 , t2 ) 7→ (σ(t2 ), t1 + t2 ) + (σ(t1 + t2 ), t2 ).

Exercise 4.10. Let σ : ∆n → X be a simplex. Then note that PnX (σ) = (σ × 1)# (βn+1 ). Thus

(f × 1)# PnX (σ) = (f σ × 1)# (βn+1 ).

On the other hand, we know that

PnY f# (σ) = (f# σ × 1)#(βn+1 ),

which is the same as the previous expression because σ is a simplex and so f# σ = f σ.

Exercise 4.11. The inclusion i is a homotopy equivalence, and so Corollary 4.24 implies that i∗ is an
isomorphism.

Exercise 4.12. Note that the sin(1/x) space has two path components, both of which are contractible.
Thus H0 (X) = Z2 and Hn (X) = 0 for n > 0.

The Hurewicz Theorem

Exercise 4.13. We know that ϕ ◦ h# takes the path class [f ] to ϕ[h ◦ f ] = cls hf η. On the flip side, we
know that h∗ ◦ ϕ takes ϕ to h∗ cls f η. But because f η is a simplex, this is simply cls hf η as well.
Exercise 4.14. We know that
f ∗ f −1 ∗ (f ∗ f −1 )−1 ' c
for some constant map c. But note that (f ∗ f −1 )−1 = f ∗ f −1 . Thus we can apply the Hurewicz map to
find that
2 cls((f + f −1 )η) = [0].
It follows that f + f −1 ∈ B1 (X), where f and f −1 are considered as 1-chains. Thus f and −f −1 are
homologous, as desired.

Exercise 4.15. Note that the boundary of the second triangle is α ∗ β + γ − (α ∗ β) ∗ γ. Thus cls(α ∗ β ∗ γ) =
cls(α ∗ β + γ). Repeating this procedure on the first triangle, we find that cls(α ∗ β ∗ γ) = cls(α + β + γ).
Note that, in the text, there is a second equality, namely that these expressions equal cls α + cls β + cls γ.
However, homology classes are not actually defined for paths which are not closed, so this seems to be an
error.

Exercise 4.16. This is proved in Theorem 6.20.

22
 5 Long Exact Sequences
The Category Comp
Exercise 5.1. These results all follow directly from the definition of exactness.

(i) Note that ker f = im 0 = 0, and so f is injective.

(ii) In this case, we have im g = ker 0 = C.
(iii) By the previous two parts, we know that f is bijective. Because f is a homomorphism as well, it follows
that f is an isomorphism.
(iv) Either observe that 0 → A → 0 → 0 is exact and apply the previous part, or note that A → 0 is injective
while 0 → A is surjective, implying that A ∼
= 0, i.e., that A = 0.
Exercise 5.2. Note that f is surjective if and only if ker g = im f = B. But ker g = B if and only if g is
the zero map, which is itself true exactly when ker h = im g = 0. Since ker h = 0 if and only if h is injective,
we are done.
i
Exercise 5.3. We know that 0 → A → − B implies that i is an injection. But because i is a surjection onto its
image, this implies that iA ∼
= A. Moreover, because ker p = im i = iA, we know that B/iA = B/ ker p ∼ = im p.
Because p is a surjection (see Exercise 5.1), the result follows.
Exercise 5.4. This amounts, effectively, to following the arrows and the equations given by exactness. In
more detail, let fn : Bn → Cn and gn : Cn → An−1 . Now observe that Bn = im hn = ker fn . Thus fn is the
zero map. Moreover, because ker gn = im fn , we know that gn is injective. Finally, we have im gn = ker hn−1 .
But hn−1 is an isomorphism, and so its kernel is trivial. Thus im gn = 0. Because gn was injective, it follows
that Cn = 0.
Exercise 5.5.

(i) Let f be the map from A to B and g be the map from B to C. Let {aα } and {cγ } be maximal independent
−1
sets of A and C, respectively. ForP P bα = f (aα ). For every γ, pick some bγ ∈ g (cγ ), which
every α, let
is possible by surjectivity of g. If nα bα + n0γ b0γ = 0, then we know that
X X 
g n α bα + n0γ b0γ = 0.

But we also know that im f = ker g, andPso g(bα ) = 0. Thus this simply implies that n0γ cP
P
γ = 0, implying
that n0γ = 0. But now we know that nα bα = 0, and so injectivity of f implies that nα aα = 0 as
well. Thus nα = 0 for all α as well, and so {bα } ∪ {bγ } is independent. Thus rank B ≥ rank A + rank C.
To show the opposite inequality, it suffices to show that {bα } ∪ {bγ } is maximally independent. Note
that b 6∈ f (A). Otherwise, we could take f −1 on {bα } ∪ {b}, which is not independent. Now consider
{bγ } ∪ {b}. If g(b) = g(bγ ) for any γ, then we know by Exercise 5.3 that b − bγ ∈ f (A). Obviously, we
cannot have b−bγ = bα for any α, otherwise that would give us our linear dependence. Thus {b−bγ }∪{bα }
is a subset of f (A) with rank A + 1 elements. This is not independent, a contradiction.
(ii) We prove this by induction. The previous part takes care of the base case. Consider the following
commutative diagrams.
fn v
0 An An−1 An−1 / im fn 0

f¯n−1 fn−2 f2 f1
0 An−1 / ker fn−1 An−2 ... A1 A0 0.

Here v is the natural map and f¯n−1 is the well-defined map taking x + ker fn−1 to fn−1 (x).
Let ri be the rank of Ai . Then the first diagram implies that rn − rn−1 + r = 0, where r is the rank of
An−1 / im fn . Because im fn = ker fn−1 , the second diagram implies by induction that r − rn−2 + rn−3 +
· · · = 0. Thus we subtract the first from the second to find that rn − rn−1 + rn−2 − · · · = 0, as desired.

23
Exercise 5.6. If ∂n = 0 for all n, then we know that Hn (S∗ ) = ker ∂n / im ∂n+1 = Sn /{0} = Sn .
Exercise 5.7. If f : S∗ → S∗0 is an equivalence, then it has an inverse g : S∗0 → S∗ . Thus at every n, there is
a gn : Sn0 → Sn so that gn ◦ fn = idSn and fn ◦ gn = idSn0 . It follows that fn is an isomorphism for every n.
Conversely, suppose fn is an isomorphism for every n. Then let g = {gn }, where gn = fn−1 . It is clear
that f and g are inverses, and so f is indeed an equivalence in Comp.
Exercise 5.8. If the former sequence is exact in Comp, then we know that im f = ker g. Then the terms
of degree n of both im f and ker g must be the same. In other words, we must have im fn = ker gn , and so
the latter sequence is exact in Ab.
On the other hand, suppose that the latter sequence is exact for every integer n. Then we know that
the degree n terms of ker f and im g are the same. Moreover, we know that the differentiation operators are
the same because they are defined, in both cases, simply as restrictions of the differentiation operator in S∗ .
Thus the two complexes are the same, as desired.
Exercise 5.9.
(i) We have the following diagram, where ∂¯n represents the map taking sn + Sn0 7→ ∂n (sn ) + Sn−1
0
.

∂n+1 ∂n
... Sn+1 Sn Sn−1 ...
vn+1 vn vn−1

0 ∂¯n+1 ∂¯n
... Sn+1 /Sn+1 Sn /Sn0 0
Sn−1 /Sn−1 ...

To show that v is a chain map, we must show that vn−1 ∂n = ∂¯n vn for every n. Pick a simplex σ : ∆n → X.
We know that vn−1 (∂nσ) = ∂n σ + Sn−1 0
. However, we also have ∂¯n vn σ = ∂¯n (σ + Sn0 ) = ∂n σ + Sn−1
0
.
Thus this is indeed a chain map.
Moreover, it is obvious that ker vn = Sn0 for every n. The definition of a subcomplex implies that the
∂n | ker vn is the operation in S∗0 . Thus ker v = S∗0 , as desired.
(ii) At each n, we know from the previous part that we have the following commutative diagram in Ab.
∂n
Sn Sn−1
vn vn−1
∂¯n
Sn / ker fn Sn−1 / ker fn−1

By the first isomorphism theorem for groups, however, we know that there is an isomorphism θn from
Sn / ker fn → im fn such that θn vn = fn .
We claim that θ = {θn } is the desired chain map. To see this, observe that

θn−1 ∂¯n (σ + ker fn ) = θn−1 (∂n σ + ker fn−1 ) = fn−1 ∂n σ.

On the other hand, because σ + ker fn = vn (σ), we know that

∂n0 θn (σ + ker fn ) = ∂n0 (fn (σ)) = ∂n fn σ.

The two final expressions in the above equations are equal, moreover, because {fn } is itself a chain map.
Exercise 5.10. First, note that both S∗0 /(S∗0 ∩ S∗00 ) and (S∗0 + S∗00 )/S∗00 are well-defined because everything
is abelian. Now consider the map
S∗0 + S∗00
ϕ : S∗0 →
S∗00
Sn 7→ Sn + S∗00 .
0 0

Note that we have boundary maps

∂n0 : Sn0 → Sn−1
0

24
and 0 00
Sn0 + Sn00 Sn−1 + Sn−1
∂n : → 00 ,
Sn00 Sn−1
where ∂ n takes (s0n + s00n ) + Sn00 to (∂n0 s0n + ∂n00 s00n ) + Sn−1
00
.
We claim that ϕ is a chain map. To see this, it suffices to show that ϕn−1 ∂n0 = ∂ n ϕn . But for any
σ ∈ Sn0 , we know that
0

ϕn−1 ∂n0 (σ) = ∂n0 σ + Sn−1 00

= ∂ n (σ + Sn00 ) = ∂ n ϕn (σ),
as desired. Moreover, the second isomorphism theorem for groups implies that ϕn is a homomorphism with
kernel Sn0 ∩ Sn00 , from which it follows that ϕ is a chain map with ker ϕ = S∗0 ∩ S∗00 . The first isomorphism
theorem (Exercise 5.9) implies the result.

Exercise 5.11. Consider the sequence in the problem, namely

i p
0 T∗ /U∗ S∗ /U∗ S∗ /T∗ 0 .

Clearly, we have
im in = {tn + Un : tn ∈ Tn }.
Moreover, we know that pn (sn + Un ) = sn + Tn , so

ker pn = {sn + Un : sn ∈ Tn }.

Clearly these are equal.

It now suffices to prove that ker in = 0 and im pn = Sn /Tn . But note that in (tn + Un ) = 0 implies
that tn ∈ Un . Hence tn + Un = 0 as an element of Tn /Un as well. Moreover, consider an arbitrary element
sn + Tn ∈ Sn /Tn . It is equal to pn (sn + Un ), which proves that p is surjective. Hence the sequence of
complexes is exact.
Exercise 5.12. We claim that X  X
ker ∂nλ = ker ∂nλ .
P λ P λ

If sn ∈ ker ∂n , then by definition we must have ∂nλ (sλn ) = 0 for each λ. The converse is also clearly
true. Similarly, we find that
X  nX o X
λ λ
im ∂n+1 = ∂n+1 (sλn+1 ) = λ
im ∂n+1 .

Thus we conclude that

ker ∂nλ X ker ∂ λ
P
X  ker ∂n n
X
Hn S∗λ = =P λ
= λ
= Hn (S∗λ ).
im ∂n+1 im ∂n+1 im ∂n+1

Exact Homology Sequences

P 0
mσ σ + Sn (A). This implies that mσ − m0σ = 0 for every σ
P
Exercise 5.13. Suppose mσ σ + Sn (A) =
with im σ 6⊆ A. Hence we can pick the unique representative
X X
mσ σ = m0σ σ,
im σ6⊆A im σ6⊆A

thus showing that this is indeed the free abelian group generated by σ with im σ 6⊆ A.
Exercise 5.14.

25
 (i) It suffices to prove that pn is surjective and that im in = ker pn . To see that pn is surjective, consider the
following segment of the long exact sequence:
pn in−1
Bn Cn An−1 Bn−1

Note that the map fn : Cn → An−1 has im fn = ker in−1 = 0, and so ker fn = Cn . Thus im fn = Cn ,
proving surjectivity.
To see that im in = ker pn , simply consider the following segment:
in pn
An Bn Cn

The result immediately follows.

(ii) There exists a map r with r ◦ i = idA . Thus r∗ ◦ i∗ = idHn (A) , from which it follows that i∗ is injective.
Theorem 5.8 gives an exact sequence
in pn d
... Hn (A) Hn (X) Hn (X, A) ...

where p∗ is induced by the quotient map S∗ (X) → S∗ (X)/S∗ (A). Since i∗ injective implies that in is
injective, we can apply the previous part to find an exact sequence
in pn
0 Hn (A) Hn (X) Hn (X, A) 0

Then Exercise 5.3 implies that Hn ⊕ Hn (X, A) = Hn (X), as desired.

(iii) We now have i ◦ r ' idX as well. In particular, since A and X have the same homotopy type, we must
have Hn (A) ∼= Hn (X) by Corollary 4.24. Thus in in the exact sequence given in the previous part must
be the identity, and so ker pn = im in = Hn (X). But since pn is surjective, it follows that Hn (X, A) = 0,
as desired.
Exercise 5.15. We prove this in cases. We will use the follow commutative diagram, where the columns
are exact:
0 0 0

0 0
∂n+1 ∂n
0
... Sn+1 Sn0 0
Sn−1 ...

in+1 in in−1

∂n+1 ∂n
... Sn+1 Sn Sn−1 ...

pn+1 pn pn
00 00
∂n+1 ∂n
00
... Sn+1 Sn00 00
Sn−1 ...

0 0 0
Case 1. S∗ and S∗0 are acyclic.
We would like to show than Zn00 = ker ∂n00 is equal to Bn00 = im ∂n+1
00
. We already know that Bn00 ⊆ Zn00 .
00
Surjectivity of p implies that we can rewrite Zn as

Zn00 = pn (ker(∂n00 pn )) = pn (ker(pn−1 ∂n )).

This, in turn, can be written as

Zn00 = {pn sn : pn−1 ∂n sn = 0}.

26
On the other hand, we can rewrite Bn00 as
Bn00 = im(∂n+1
00
pn+1 = im(pn ∂n+1 )) = pn im ∂n+1 .
Since S∗ is acyclic, we know that im ∂n+1 = ker ∂n , and so we find that
Bn00 = {pn zn : ∂n zn = 0}.
Now consider an arbitrary element pn sn ∈ Zn00 . Since pn−1 ∂n sn = 0, we know that ∂n sn ∈ ker pn−1 =
im in−1 , where again we use the fact that S∗ is acyclic. Injectivity of in−1 implies the existence of a unique
s0n−1 ∈ Sn−1
0
with in−1 s0n−1 = ∂n sn . We know, however, that ∂n−1 ∂n = 0, and so
0 = ∂n−1 ∂n s = ∂n−1 in−1 s0n−1 = in−2 ∂n−1
0
s0n−1 .
0
Since in−2 is injective, it follows that ∂n−1 sn−1 = 0, and so acyclicity of S∗0 implies that sn−1 ∈ ker ∂n−1
0
=
0 0 0 0
im ∂n . In particular, we can write sn−1 = ∂n sn .
Now notice that
∂n in s0n = in−1 ∂n0 s0n = in−1 s0n−1 = ∂n sn ,
where the last equality follows from the definition of s0n−1 . We know that zn = sn − in s0n ∈ Zn since ∂n0 is a
homomorphism. But we also know that
pn zn = pn (sn − in s0n ) = pn sn − pn in s0n = pn sn ,
where we use exactness of the columns. In other words, we have a zn with ∂n zn = 0, such that pn zn = pn sn .
Thus pn sn ∈ Bn00 , proving that Zn00 = Bn00 . To be even more explicit, this implies that Hn00 = Zn00 /Bn00 = 0 for
all n, proving that S∗00 is an acyclic complex as well.
Case 2. S∗0 and S∗00 are acyclic.
00
Suppose sn ∈ Zn , i.e., that ∂n sn = 0. Then ∂n00 pn = pn−1 ∂n implies that pn sn ∈ ker ∂n00 = im ∂n+1 . Hence
00 00 00
write pn sn = ∂n+1 sn+1 . Since pn+1 is surjective, we can find sn+1 with pn+1 sn+1 = sn+1 , and so
00
pn ∂n+1 sn+1 = ∂n+1 pn+1 sn+1 = pn sn .
But then we know that ∂n+1 sn+1 − sn ∈ ker pn = im in . Thus there exists a unique s0n with in s0n =
∂n+1 sn+1 − sn . We can take ∂n of both sides to find that
0 = ∂n ∂n+1 sn+1 − ∂n sn = ∂n in s0n = in−1 ∂n0 s0n ,
0
and so it follows that ∂n0 s0n = 0. In particular, we know that s0n ∈ im ∂n+1 , so we can find s0n+1 whose
0
boundary is sn . Recall that we had
sn = ∂n+1 sn+1 − in s0n .
0
But the last term is equal to in ∂n+1 s0n+1 = ∂n+1 in+1 s0n+1 , and so this is in turn equal to
sn = ∂n+1 (sn+1 − in+1 s0n+1 ).
This proves that sn ∈ Bn , and so Zn = Bn .
Case 3. S∗ and S∗00 are acyclic.
This final case is handled similarly to the first two, but we lay out the details below. Let s0n ∈ ker ∂n0 = Zn0
be arbitrary. Then in s0n ∈ ker ∂n = im ∂n+1 , and so
in s0n = ∂n+1 sn+1
00 00
for some sn+1 . We know that pn+1 sn+1 ∈ ker ∂n+1 = im ∂n+2 because pn in s0n = 0. Hence there exists s00n+2
00
with ∂n+2 s00n+2 = pn+1 sn+1 . But then it follows that
00
pn+1 ∂n+2 sn+2 = ∂n+2 pn+2 sn+2 = pn+1 sn+1 ,
from which it follows that sn+1 − ∂n+2 sn+2 ∈ ker pn+1 = im in+1 . Thus there exists s0n+1 with in+1 s0n+1 =
sn+1 − ∂n+2 sn+2 . We then find that
0
in ∂n+1 s0n+1 = ∂n+1 in+1 s0n+1 = ∂n+1 sn+1 = in s0n .
Injectivity implies s0n = ∂n+1
0
s0n+1 ∈ Bn , thus proving the final case.

27
Exercise 5.16. To show that f# (Zn (X, A)) ⊆ Zn (X 0 , A0 ), consider γ ∈ Zn (X, A). Note that γ ∈ Sn (X),
and so Lemma 4.8 implies that
∂n0 f# γ = f# ∂n γ.
P
We know, moreover, that ∂n γ = mσ σ, where the sum ranges over all σ with im σ ⊆ A. Thus it follows
that X  X
∂n0 f# γ = f# mσ σ = mσ f σ.
im(f σ)⊆f (A)

Since f σ is a simplex into X with image contained in A0 , it follows that this is an element of Sn−1 (A0 ).
0

Hence we conclude that f# γ ∈ Zn (X 0 , A0 ), as desired.

The proof for boundaries is similar.
Exercise 5.17. As defined, we have that f# : Hn (X, A) → Hn (X 0 , A0 ) is given by
f# : γ + im ∂ n+1 7→ f# (γ) + im ∂ 0 n+1 ,
where ∂ and ∂ 0 denote the boundary maps of the quotient complexes S∗ (X)/S∗ (A) and S∗ (X 0 )/S∗ (A0 ),
respectively, and where
γ ∈ ker ∂ n = Zn (X, A)/Sn (A).
The third isomorphism theorem gives an isomorphism
Zn (X, A)/Sn (A) Zn (X, A)
Hn (X, A) = →
Bn (X, A)/Sn (A) Bn (X, A)
which takes
γ + Bn (X, A)/Sn (A) 7→ γ + Bn (X, A).
Since f# (γ) = f# (γ), we find that, thinking of f# as a map from Zn (X, A)/Bn (X, A) to a map from
Zn (X 0 , A0 )/Bn (X 0 , A0 ), it takes
γ + Bn (X, A) 7→ f# (γ) + Bn (X 0 , A0 ),
as desired.
Exercise 5.18. Recall the definition of εi : ∆n−1 → ∆n as taking {e0 , . . . , en−1 } to {e0 , . . . , êi , . . . , en−1 }.
Thus we have
n
X
∂n σ = (−1)i σεi .
i=0

Since σεi : ∆n−1 → X has image in A by hypothesis, this is in Sn−1 (A), as desired.

Reduced Homology
Exercise 5.19. By Theorem 5.6, it is sufficient to show that we have a short exact sequence

0 Se∗ (A) Se∗ (X) S∗ (X, A) 0.

When n ≥ 1, this is clear by Theorem 5.8. When n = 0, we have the sequence

0 → Z → Z → 0 → 0,
which is easily verified to be exact.
Exercise 5.20. Consider the exact sequence

H1 (CX) H1 (CX, X) H
e 0 (X) H
e 0 (CX).

Note that H1 (CX) = 0 because CX is contractible. On the other hand, Corollary 5.18 implies that H ∼
e 0 (X) =
4 ∼
Z , while H0 (CX) = 0. Thus the map H1 (CX, X) → H0 (X) is surjective. Moreover, its kernel is equal to
e e
the image of the map H1 (CX) → H1 (CX, X), which is simply 0 since H1 (CX) = 0. Thus the map is also
injective, from which it immediately follows that H1 (CX, X) ∼
= Z4 .

28
Exercise 5.21. Consider the exact sequence

e 1 (S 0 )
H e 1 (S 1 )
H H1 (S 1 , S 0 ) e 0 (S 0 )
H e 0 (S 1 ),
H

which is simply equal to

0 Z Z 0.

Now note that the first map has im = 0, so the second map has ker = 0. Thus the second map has im ∼
= Z,
and so the third map has ker ∼
= Z. Yet we also know that the last map is the zero map, and so the third
map has im = Z, from which it follows that the H1 (S 1 , S 0 ) = Z × Z.
Exercise 5.22. When n = 0, this follows from the exact sequence

H
e 0 (X) H
e 0 (X) H0 (X, X) 0 .

After all, the first map is the identity, and so the second map is the zero map. But the second map is
surjective, and so H0 (X, X) = 0.
For n > 0, we have the exact sequence

H
e n (X) H
e n (X) Hn (X, X) H
e n−1 (X) H
e n−1 (X).

The first map is the identity, and so the second map is everywhere zero. Thus the image of the second map,
which is the kernel of the third map too, is equal to 0. Since the kernel of the last map, which is 0 (the map
is the identity), is equal to the image of the third map, it follows that the third map is everywhere zero.
Hence the third map is injective, but also everywhere zero, and so Hn (X, X) must have been 0 in the first
place.

29
6 Excision and Applications
Excision and Mayer–Vietoris
Exercise 6.1. Since A and B are both open, we know that A◦ = A and B ◦ = B. Thus A◦ ∪ B ◦ = X, and
so we can use the Mayer–Vietoris sequence, along with the fact that A ∩ B = ∅, to find an exact sequence
(i1∗ ,i2∗ ) g∗ −j∗ D
0 = Hn (∅) Hn (A) ⊕ Hn (B) Hn (X) Hn−1 (∅) = 0 .
Thus the middle map g∗ − j∗ is an isomorphism, which proves the result.
Exercise 6.2. We use excision directly. In particular, Excision II gives us an isomorphism i∗ : Hn (B, ∅) →
Hn (X, A). But Hn (B, ∅) = Hn (B) for all n ≥ 0, and so the conclusion follows.
Exercise 6.3. This is simply a diagram chase. Suppose Dn ([x]) = [x0 ]. Then by definition of D, we know
that there exists some [z] ∈ Hn (X1 , X1 ∩ X2 ) such that dn ([z]) = [x0 ] and hn ([z]) = qn ([x]). It thus follows
that
gn−1 (Dn ([x])) = [fn−1 (x0 )] = fn−1 (dn ([z])).
Now set y = f (x), so that [y] = fn ([x]). We would like to show that Dn0 ([y]) = fn−1 (dn ([z])). To do this,
set z 0 = fn (z). Then since d commutes with f by definition (cf. Theorem 5.7), we know that
d0n ([z 0 ]) = d0n (fn (z)) = fn−1 (dn ([z])) = gn−1 (Dn ([x])).
Moreover, because h and q are just inclusions, we know that they commute with f . In particular, from the
fact that hn ([z]) = qn ([x]), and so f (hn ([z])) = f (qn ([x])), we find that
h0n (f ([z])) = qn0 (fn ([x])),
from which it follows by definition that h0n ([z 0 ]) = qn0 ([y]). Thus it follows that
D0 (f ([x])) = D0 ([y]) = gn−1 (Dn ([x])),
which proves that the desired diagram commutes.
Exercise 6.4. First note that the condition implies that, for all n ≥ 1, we have
Hn (Xi ) = Hn (Xi ∩ Xj ) = Hn (X1 ∩ X2 ∩ X3 ) = 0.
Since each Xi is open, we can apply the Mayer–Vietoris sequence. Applying it on X1 and X2 gives an exact
sequence
0 Hn (X1 ∪ X2 ) 0 ,
and so we conclude that Hn (X1 ∪ X2 ) = 0.
Now we can apply Mayer–Vietoris to X1 ∪ X2 and X3 to find an exact sequence
0 Hn (X) Hn−1 ((X1 ∪ X2 ) ∩ X3 ). .

But the last term is exactly Hn−1 ((X1 ∩ X3 ) ∪ (X2 ∩ X3 )).

To see that this is 0, apply Mayer–Vietoris to X1 ∩ X3 and X2 ∩ X3 . In particular, setting H =
Hn−1 ((X1 ∩ X3 ) ∪ (X2 ∩ X3 )) as the desired homology group, we know that

Hn−1 (X1 ∩ X3 ) ⊕ Hn−1 (X2 ∩ X3 ) H Hn−2 ((X1 ∩ X3 ) ∩ (X2 ∩ X3 ))

is exact. If n > 2 or if X1 ∩ X2 ∩ X3 = ∅, then the first and last terms are clearly 0, which proves that the
middle homology group is indeed 0. If, on the other hand, we have n = 2 and X1 ∩ X2 ∩ X3 is contractible,
then the last term is Z. However, the next map in the Mayer–Vietoris sequence is an injective map, since it
is induced by inclusions. Thus we have an exact sequence
0 H Z A ,
where A is some homology group (in fact, it is Z2 ) and the map Z → A is injective. Note that the image of
the first map is 0, and so the map H → Z is injective. But we also know that im(H → Z) = ker(Z → A) = 0.
Thus H = 0, as desired.

30
Homology of Spheres and Some Applications
No exercises!

Barycentric Subdivision and Proof of Excision

Exercise 6.5. We use induction. In particularly, for Σ0 we know we have (0 + 1)! = 1 0-simplexes. Now
suppose the statement is true for n − 1. Using the notation in the definition, note that each n-simplex in Σn
is spanned by b and an (n − 1)-simplex in Sd ϕi . Since there are n + 1 total possible ϕi ’s, and since there
are n! total (n − 1)-simplexes in Sd ϕi , it follows that Sd Σn has (n + 1)! total n-simplexes, as desired,
Exercise 6.6.
(i) By construction, every point is the barycenter of at least one face. Moreover, writing Σn = [p0 , . . . , pn+1 ],
suppose that b is the barycenter of [pi1 , . . . , pik ], as well as of [pj1 , . . . , pj` ]. Then
1 1
(pi + · · · + pik ) = (pj + · · · + pj` ).
k+1 1 `+1 1
This implies linear dependence, unless the two subsimplexes of Σn are actually the same simplex.
(ii) This is clear for n = 0. Now suppose that the statement is true for n − 1 Note that, by definition, every n-
simplex of Σn is spanned by the barycenter b of Σn and an (n−1)-simplex of Sd ϕi . But every face of Σn is
a subset of Σn . Thus we can write an n-simplex of Σn as [bσ0 , . . . , bσn ] with σ0 < · · · < σn−1 < σn = Σn .
Exercise 6.7.
(i) Note that Sd1 (δ 1 ) is exactly b1 . Sd0 (∂δ 1 ). Since Sd0 is the identity, we know that this is b1 .(∂δ 1 ). But
∂δ 1 = e1 − e0 , while b1 = 21 (e0 + e1 ), and so it follows that

Sd1 (δ 1 )(t) = b1 .e1 (t) − b1 .e0 (t)

   
t t
= (e0 + e1 ) + (1 − t)e1 − (e0 + e1 ) + (1 − t)e0 .
2 2
Note that both terms within the large parentheses are 1-simplexes, and so we cannot “cancel” the
t
2 (e0 + e1 ) terms.
For n = 2, we would like to evaluate b2 . Sd1 (∂δ 2 ). Note that ∂δ 2 = [e1 , e2 ] − [e0 , e2 ] + [e0 , e1 ]. Thus we
know, either by using the same argument as before or by appealing to the case n = 1 in part (ii) below,
that
   
t t
Sd1 (∂δ 2 )(t) = (e1 + e2 ) + (1 − t)e2 − (e1 + e2 ) + (1 − t)e1
2 2
   
t t
− (e0 + e2 ) + (1 − t)e2 + (e0 + e2 ) + (1 − t)e0
2 2
   
t t
+ (e0 + e1 ) + (1 − t)e1 − (e0 + e1 ) + (1 − t)e0 .
2 2
Thus we may evaluate Sd2 (δ 2 ) on a term-by-term basis. For example, the first term of Sd2 (δ 2 )(t1 , t2 ) is
t2 /(1 − t1 )
   
t2
t1 b2 + (1 − t1 ) (e1 + e2 ) + (1 − t2 /(1 − t1 ))e2 = t1 b2 + (e1 + e2 ) + (1 − t1 − t2 )e2 .
2 2
We can do this with each term to find Sd2 (δ 2 ).
(ii) We can simply evaluate this using the previous part. In particular, we have
Sd1 (σ) = σ# Sd1 (δ 1 )
   
t t
= (σ(e0 ) + σ(e1 )) + (1 − t)σ(e1 ) − (σ(e0 ) + σ(e1 )) + (1 − t)σ(e0 ) .
2 2
Similarly, by replacing each ei in Sd2 (δ 2 ) with σ(ei ), we have Sd2 (σ).

31
Exercise 6.8. It is sufficient to show commutativity for generators σ : ∆n → X. But note that f# Sdn (σ) =
f# σ# Sdn (δ n ). However, since f# σ# = (f ◦ σ)# , it follows that this is in turn equal to (f ◦ σ)# Sdn (δ n ) =
Sdn (f ◦ σ) = Sdn f# σ. This proves commutativity, as desired.
Exercise 6.9. Recall that the j-th face of σ is σεj : [e0 , . . . , en−1 ] → [e0 , . . . , êi , . . . , en ]. Now observe that
εj is clearly affine. After all, we know that
!
X X X X
εj ti ei = ti ei + ti ei+1 = ti εj (ei ).
i i<j i≥j i

Thus we know that X  X  X

σεj ti ei = σ ti εj (ei ) = ti σ(εj (ei )).

Thus σεj is affine. Since σεj (ei ) is either ei (if i < j) or ei+1 (if i ≥ j), it follows that the vertex set of σεj
is a (proper) subset of the vertex set of σ. Since ∂σ is just an alternating sum of faces of σ, it follows that
∂σ is affine whenever σ is.
Exercise 6.10. Recall the definition of a cone:
(
b if t0 = 1,
b.σ(t0 , . . . , tn+1 ) = 
t1 tn+1

t0 b + (1 − t0 )σ 1−t 0
, . . . , 1−t0 if t0 < 1.

It is clear that b is affine. Since the case t0 < 1 results in the sum of affine maps, it follows that this is also
affine.
Now note that b.σ(e0 ) = b and b.σ(ei ) = σ(ei ) for i 6= 0. Thus the vertex set of b.σ is the union of
{b} and the vertex set of σ. Note that Sd0 σ is affine whenever σ : ∆0 → E is affine. If Sdn−1 preserves
affineness, then note that Sdn must as well. After all, we know that Sdn σ = bn . Sdn−1 (∂σ) is the cone of
some point bn ∈ E and the affine function Sdn−1 (∂σ). (Note that this last function is affine because ∂σ is,
by Exercise 6.9, affine.) The result follows.

More Applications to Euclidean Space

Exercise 6.11. Writing (1 + an# )γ as γ + an# γ, note that

∂(γ + an# γ) = ∂γ + an−1

# ∂γ
= γ(e1 ) − γ(e0 ) + (−γ(e1 )) − (−γ(e0 )).

But recall that −γ(e1 ) = γ(e0 ) and −γ(e0 ) = γ(e1 ), and so we know that the terms cancel out to 0. Thus
(1 + an# )γ is a 1-cycle.
Exercise 6.12. We can simply compute evaluate (1 + an# )(1 − an# ) on a simplex σ. In particular, we find
that

(1 + an# )(1 − an# )σ = (1 + an# )(σ − an σ)

= σ + an# σ − an σ − an# (an σ)
= σ + an σ − an σ − an an σ.

But of course we have σ = an an σ, and so the terms all cancel out. We can simply extend to any 1-chain γ.
Exercise 6.13. Again, we evaluate the expression on a simplex σ and find that

(1 + an# )(1 + an# )σ = (1 + an# )(σ + an σ)

= σ + an σ + an σ + an an σ
= 2σ + 2an σ = 2(1 + an# )σ.

As before, we can extend.

32
Exercise 6.14. Letting τ be, as in Theorem 6.22, the southerly path in S 1 from a1 (y) to y, recall that the
homology class of the cycle σ + τ generates all of H1 (S 1 ). But notice that [(1 + a1# )σ] = [σ + a1 σ] = [σ + τ ],
which proves the result.
Exercise 6.15. Suppose f : S 1 → R is continuous. Note that such a function is effectively a continuous
function f : [0, 1] → R with f (0) = f (1), and so the intermediate value theorem implies the result.

Exercise 6.16. Suppose S ⊆ R2 is homeomorphic to S 2 . Then there is a function ϕ : S 2 → S ,→ R2 , where

the S 2 → S part is a homeomorphism. The Borsuk-Ulam theorem, however, implies that there exists some
point x ∈ S 2 with ϕ(x) = ϕ(−x). But ϕ is an injection, a contradiction.
Exercise 6.17. This is obvious from Borsuk-Ulam. Since there exists an x with f (x) = f (−x), but
f (−x) = −f (x) for all x, it follows that there exists an x with f (x) = −f (x), i.e., with f (x) = 0.
Exercise 6.18. This follows the same proof as that of Borsuk-Ulam. In particular, we use the function

g(x) : S n → S n−1
f (x) − f (−x)
x 7→ .
kf (x) − f (−x)k

This would be an antipodal map, a contradiction.

Exercise 6.19. Suppose an (Fi ) ∩ Fi = ∅ for i = 1, . . . , n. There exist functions gi : S n → I with gi (Fi ) = 0
and gi (an Fi ) = 1. Then define f : S n → Rn by f (x) = (g1 (x), . . . , gn (x)). Note that Exercise 6.18 implies
that there exists some x0 ∈ S n with f (x0 ) = f (−x0 ). Thus it follows that

gi (x0 ) = gi (−x0 ) = gi (an x0 )

for all i. Hence if x0 ∈ Fi then the left side of the equation is 0 while the right side is 1, and if x0 ∈ an Fi
then the left side is 1 while the right side is 0. Either way, this is a contradiction, and so it follows that
x0 , an x0 6∈ Fi for all i. Thus x0 an x0 ∈ Fn+1 .
I have not come up with a counterexample in the n + 2 case, unfortunately.
Exercise 6.20. Suppose A ⊆ S n is a subspace, and suppose that h : S n → A is a homeomorphism.
Invariance of domain implies that A is open in S n . But we also know by compactness of S n that A must be
compact, and hence closed in its ambient space. Thus A is clopen. Since A is obviously nonempty, it follows
by connectedness that A = S n .
Exercise 6.21. This follows because we can just write S n = Rn ∪ {∞}. Hence any open set in Rn , including
Rn itself, is just an open set in S n .
Walking through this in more detail, suppose U, V ⊆ Rn with a homeomorphism h : U → V and with U
open. Then U is an open subset of S n because Rn is open in S n . Hence invariance of domain on S n implies
that V is open in S n , and so V ∩ Rn = V is open in Rn as well.
Exercise 6.22. Suppose ϕ : X → Y is a homeomorphism. Then we can simply pass to a homeomorphism
between U and V in X to a homeomorphism between ϕ(U ) and ϕ(V ) in Y . Since every open set in Y is of
the form ϕ(U ) for some open U in X, the result follows immediately.
x
Exercise 6.23. Consider the map h : Dn → D 12 (0) defined by h(x) = 2. It effectively shrinks Dn n down
to the closed ball with radius 12 . Obviously the two disks are homeomorphic. But Dn is open while D 12 (0)
is not.

33
7 Simplicial Complexes
Definitions
Exercise 7.1.

Exercise 7.2. Consider some (nondegenerate) triangle with vertices P, x0 , y0 in R2 . Then define xi to be
the midpoint of P and xi−1 , and similarly define yi . Then the union X of the triangle with all the line
segments xi yi is compact and connected.
We claim that it is not a polyhedron. Otherwise, there exists some simplicial complex K admitting a
homeomorphism h : |K| → X. But observe that K must have an infinite vertex set.
To see this, for each i, define si to be \
si = s,
h−1 (xi )∈s
S
where s ranges over all simplices of K. Note that this intersection is over a nonempty set because s = |K|,
−1
so there must exist some s containing h (xi ). Moreover, there are only finitely many simplices, so the
intersection exists. Condition (ii) implies that si is a common face of s, and thus is a simplex. It must
be 0-dimensional since the segment P xi , xi yi , and x0 xi cannot all be part of the same 1-simplex. In other
words, xi must be a common face of two 1-simplices, and so it must be a point.
Hence there are infinitely many vertices of K, a contradiction.
Exercise 7.3. Note that the upper right and lower right triangles are the same.
Exercise 7.4.
(i) The forwards direction is just the definition of the subspace topology. To see the backwards direction,
suppose F ∩ s is closed in s for
S every s ∈ K. Each s is closed in |K|, so F ∩ s is closed in |K|. Since
there are finitely many s and s = |K|, it follows that we can take the union of all F ∩ s. In particular,
we have [
F = (F ∩ s)
s∈K
is the finite union of closed sets, hence is itself closed in |K|.
(ii) This is obviously true if K has dimension 0.
If K (and hence s) has dimension > 1, then consider the complement of s◦ :
(s◦ )c = (|K| − s) ∪ ṡ.
Then notice that  
(|K| − s) ∪ ṡ ∩ s = ṡ,
which is closed in s. Suppose t ∈ K is not equal to s. Then consider
 
At = (|K| − s) ∪ ṡ ∩ t.
If s ∩ t = ∅, then At = ∅ is closed in t. Otherwise, we know that s ∩ t is a face of t. Since s is of highest
dimension, we know that either s = t, which we already took care of above, or s ∩ t is part of ṡ, in which
case we know that
ṡ ∩ t = s ∩ t, (|K| − s) ∩ t = t − s ∩ t.
Hence At = t, which is still closed in t.
The previous part proves the result.

34
Exercise 7.5. We begin by showing s◦ ∩ t◦ = ∅ when s 6= t. Note that
s◦ ∩ t◦ = (s − ṡ) ∩ (t − ṫ) = s ∩ t − ṡ ∩ t − s ∩ ṫ.
But s ∩ t is a face of both s and t. It can’t be equal to both s and t since s 6= t. Thus s ∩ t is a proper face of
at least one of sSand t, say s. This means that s ∩ t is part of ṡ, and thus is in ṡ ∩ t. This proves disjointness.
To see that s◦ = |K|, simply do this in the case of K as a simplex, and take unions. (To do this when
K is a single simplex, use induction.)
Exercise 7.6. The backwards direction is obvious by the definition of st. For the forwards direction, suppose
[
x ∈ st(p) = t◦ .
p∈Vert(t)

◦
Then we know that x ∈ t for some t having p as a vertex. Uniqueness implies that s = t, so p ∈ Vert(s).
Exercise 7.7.
(i) Obviously the union is |K| because every s ∈ K has at least one vertex, hence is contained in at least
one star. To see that st(p) ⊆ |K| is open, notice that
[
(st(p))c = s◦ .
p6∈Vert

Intersect this with t ∈ K. If p 6∈ Vert(t), then this intersection is equal to t since no simplex of ṫ can
have p as a vertex. If p ∈ Vert(t), then write t = [p, p1 , . . . , pk ]. The intersection can be seen to simply
be {p1 , . . . , pk }, which is obviously closed. Thus Exercise 7.4 implies the result.
(ii) If x ∈ st(p), then x ∈ s◦ for some s with p ∈ Vert(s). Since x, p ∈ s and s is convex, it follows that the
line segment is also contained in st(p).
Exercise 7.8. The forwards direction is because [p0 , . . . , pn ] is in the intersection. The backwards direction
is because there must exist some simplex [p0 , . . . , pn , q0 , . . . , qm ] ∈ K. Since any face of a simplex in K is
also in K, it follows that [p0 , . . . , pn ] is a simplex in K.

Simplicial Approximation
T
Exercise 7.9. In the forwards direction, suppose ϕ is a simplicial map. If st(pi ) 6= ∅, then there exists a
simplex in K with vertices [pi ]. The definition implies that there must exist a simplex with vertices [ϕ(pi )],
proving this direction. The backwards direction follows directly from Exercise 7.8.
Exercise 7.10. Suppose ϕ is a simplicial approximation to f , and suppose x ∈ |K| with f (x) ∈ s◦ . Write
x ∈ t◦ for t ∈ K, and write t = [p1 , . . . , pn ]. Then we know that x ∈ st(pi ) implies that f (x) ∈ st(ϕ(pi )), so
that s◦ ⊆ st(ϕ(pi )). Thus s has ϕ(pi ) as a vertex for each i = 1, . . . , n.
Hence |ϕ|(x), which is determined by ϕ(pi ), is in s by affineness.
Now suppose that f (x) ∈ s◦ implies |ϕ|(x) ∈ s. Let p be some vertex of K so that x ∈ st(p). Then
f (x) ∈ s◦ , so |ϕ|(x) ∈ s. Hence ϕ(p) is a vertex of s by affineness and the definition of |ϕ|, from which it
follows that
f (x) ∈ s◦ ⊆ st(ϕ(p)).
We can take the union over all x ∈ st(p):
[
f (x) ⊆ st(ϕ(p)).
x∈st(p)

Of course, this left side is exactly f (st(p)), and so we’re done.

Exercise 7.11. Suppose ϕ : K → L is a simplicial approximation. Consider the obvious homotopy:
H(t, x) = (1 − t)|ϕ|(x) + tf (x).
We can do this because |ϕ|(x) and f (x) are, by Exercise 7.10, in the same simplex.

35
 Exercise 7.12.
(i) This is true because it’s true for simplices.
(ii) Order the vertices of K, and define ϕ(bs ) to be the smallest vertex of s under this order. We claim that
this gives a simplicial approximation to the identity. Consider a vertex bs of Sd(K). Then we know that

f (st(bs )) = s◦ ⊆ st(ϕ(bs ))

by the definition of ϕ(bs ), where f is the identity.

(iii) There exists a homeomorphism g : |L| → X. If g(v), then we are done. Otherwise, we know that
x ∈ g(s◦ ) for some unique s ∈ L. Consider the subdivision K of L obtained by drawing lines from s to
every vertex of s. This gives a function h : |K| → X which is equal to g, and thus is a homeomorphism,
as desired.
λi bsi = 0. Since s0 < · · · < sq , we knwo that there exists some vertex
P
Exercise 7.13. Suppose that
pq which only appears in b , so λq = 0. But then there is a vertex pq−1 which only appears in bsq−1 , so
sq

λq−1 = 0, and so on. Thus λi = 0 for all i, proving independence.

Exercise 7.14. Every point of Sd K is contained in a unique open simplex of K, so it follows that an open
simplex of Sd K can be contained in at most one open simplex of K. To see that there is at least one such
simplex, note that [bs0 , . . . , bsq ]◦ is contained in s◦q .
Exercise 7.15. This follows from the triangle inequality:

|x − y| ≤ |x − p| + |p − y| ≤ 2µ,

because x and p are in one simplex, and y and p are in another.

Exercise 7.16. Write s = [bs0 , . . . , bsq ], where s0 < · · · < sq . Then diam s = sup ||bsi − bsj ||. If i < j, then
we know that
nj
||bsi − bsj || ≤ diam sj ,
nj + 1
n n
where nj = dim sj . But diam sj ≤ mesh K since sj ∈ K, and nj +1
j
≤ n+1 , since nj ≤ n. Hence it follows
that
n
diam s ≤ mesh K,
n+1
and so mesh Sd K ≤ (n/n + 1) mesh K. Induction implies the general result.
Exercise 7.17. If s ∈ K (q) , then s = [p0 , . . . , pr ] for some r ≤ q. Thus ϕ(s) = [ϕ(p0 ), . . . , ϕ(pr )] ∈ L(q) , as
desired.
Exercise 7.18. Let b be the barycenter of the (n + 1)-simplex, and consider

x−b
f (x) = + b.
||x − b||

This is the desired homeomorphism.

Abstract Simplicial Complexes

No exercises!

36
Simplicial Homology
Exercise 7.19. In general, there are n+2

n
q+1 total q-simplices in an (n+1)-simplex. Since S is the n-skeleton
of such a simplex, it follows that we must simply evaluate
n   n+2
X n + 2    
n
X n+2 q q+1 n+2 n+2
χ(S ) = (−1) = (−1) + + =2
q=0
q+1 q=0
q 0 n+2

when q is even. When q is odd, the last term is negative, and we find that χ(S n ) = 0.
Exercise 7.20. Here, we have α2 = 18, α1 = 27, and α0 = 9. Thus χ(T ) = 18 − 27 + 9 = 0.
P P
Exercise 7.21. Note that i is obviously an injection. Moreover, since the element b ∈ B1 mb b+ c∈B2 mc c ∈
F (b) is equal to X 
X X X
b ∈ B1 mb b + mc c ∈ F (b) = p mb b, −mc c ,
c∈B2

we see that p is surjective. Finally, note that

nX X  X X o
ker p = mb b, mc c : mb b = mc c
= {(x, x) : x ∈ F (B1 ) ∩ F (B2 )}
= {(x, x) : x ∈ F (B1 ∩ B2 )} = im i,

which completes the proof of exactness.

Comparison with Singular Homology

Exercise 7.22. For q ≥ 1, the complexes are the same. If q = 0, we use the same argument as in Theorem
5.17, in particular, by restricting our attention to the ending:

∂˜0
0 ker ∂˜0 C0 (K) C−1 (K) 0 .

Exercise 7.23. This is simply because ker ∂˜−1 = C−1 (K).

Exercise 7.24.
(i) We can simply use the straight line homotopy between ϕ(p) and ψ(p) for all vertices p of K; the rest of
the point follow by affineness. The reason this works is simply because ϕ(p) and ψ(p) belong to the same
simplex, which is convex.
(ii) Since |ϕ| ' |ψ, we know that |ϕ|∗ = |ψ|∗ , which in turn implies that ϕ∗ = ψ∗ by Theorem 7.22.
Exercise 7.25. Let L be a line segment, along with its endpoints and its midpoints. Thus it is composed
of two 1-simplices, and three 0-simplices. Then let ϕ1 map a 1-simplex to the left side of L, and ϕ2 map it
to the right side of L. Finally, if ϕ3 maps the 1-simplex to the midpoint, it follows that ϕ1 ∼ ϕ3 ∼ ϕ2 , but
obviously ϕ1 6∼ ϕ2 .
Exercise 7.26.
(i) This is clear by mapping the base points together, and mapping a given equivalence class to the corre-
sponding equivalence class. For example, we have some point x ∈ X, then the homeomorphism would
take [[x]] ∈ (X ∨ Y ) ∨ Z to [x] ∈ X ∨ (Y ∨ Z). Similarly, it would take [[y]] 7→ [[y]] and [z] 7→ [[z]].
(ii) For i = 1, 2, there exists a simplicial complex Li and a homeomorphism hi : |Li | → Ki . Fix some vertex
xi ∈ Vert(Li ). Then let L = L1 ∨ L2 . Then, identifying each Li with the natural corresponding set in L,
we can apply Theorem 7.17 to find the exact senuence

... Hn (L1 ∩ L2 ) Hn (L1 ) ⊕ Hn (L2 ) Hn (L) Hn−1 (L1 ∩ L2 ) ....

37
 Of course, we have L1 ∩ L2 is a singleton, so the homology groups are 0. Thus, if n ≥ 2, then we know
that Hn (L1 ∩ L2 ) = Hn−1 (L1 ∩ L2 ) = 0, and so Hn (L) ∼
= Hn (L1 ) ⊕ Hn (L2 ), as desired. Otherwise, we
can simply use the tail:

... H1 (L) H0 (L1 ∩ L2 ) H0 (L1 ) ⊕ H0 (L2 ) H0 (L) 0.

If Li has ci components, then notice that L has c1 +c2 −1 components. Since the map H0 (L1 )⊕H0 (L2 ) →
H0 (L) is surjective, it follows that its kernel is Z (or, more accurately, a free abelian group of rank 1).
Hence the image of H0 (L1 ∩ L2 ) → H0 (L1 ) ⊕ H0 (L2 ) is Z. The fact that H0 (L1 ∩ L2 ) = Z implies
that this map is an isomorphism, thus with empty kernel. Finally, we conclude that the image of
H1 (L) → H0 (L1 ∩ L2 ) is trivial, and so we again have the exact sequence

0H1 (L1 ) ⊕ H1 (L2 ) H1 (L) 0 .

The result follows.

(iii) Use Corollary 7.19. In particular, let Kq consist of all proper faces of an oriented (q + 1)-simplex. Then
the corollary implies that Hq (Kq ) = H̃q (Kq ) = Z and Hr (Kq ) = 0 for any r 6= q. (Note that reduced
homology matches the regular homology since q ≥ 1.) Thus the previous part shows that the space
mq
n _
_
Kq ,
q=1 i=1

where the wedge occurs at some identified vertex, satisfies the desired properties.
Exercise 7.27.
(i) This follows directly from the five lemma and Theorem 7.22, namely by looking at the following commu-
tative diagram with exact rows:

... Hn (L) Hn (K) Hn (K, L) Hn−1 (L) Hn−1 (K) ...

.

... Hn (|L|) Hn (|K|) Hn (|K|, |L|) Hn−1 (|L|) Hn−1 (|K|) ...

(ii) This follows from the previous part, Corollary 7.17, and Theorem 7.22.
Exercise 7.28. We can simply use the straight line homotopy to p. Exercise 7.7 implies that this is well-
defined.
Exercise 7.29. In particular, we must show that
\  \ 
Lαi ∩ Lβi 6= ∅.

But notice that σ0 < σ1 < · · · < σq implies that σ0 ∈ Lβi for each βi . We also know that σ0 ∈ Lα0 ∩· · ·∩Lαq ,
and so it follows that σ0 is in the displayed intersection above. Hence g and f are contiguous.
Exercise 7.30. We have the following exact sequence:

Hq (M ∩ L1 ) Hq (M ) ⊕ Hq (L1 ) Hq (M ∪ L1 ) Hq−1 (M ∩ L1 ).

The conditions imply that Hq (M ) ⊕ Hq (L1 ) = Hq (M ) and the two outermost terms are both trivial. Thus
Hq (M ) ∼
= Hq (M ∪ L1 ).
Now consider the following exact sequence:

Hq ((M ∪ L1 ) ∩ L2 ) Hq (M ∪ L1 ) ⊕ Hq (L2 ) Hq (M ∪ L1 ∪ L2 ) Hq−1 ((M ∪ L1 ) ∩ L2 ).

But notice that

(M ∪ L1 ) ∩ L2 = (M ∩ L2 ) ∪ (L1 ∩ L2 ) = M ∩ L2
since L1 ∩ L2 ⊆ M . Hence the flanking terms of the exact sequence displayed above are again 0. Since L2
is acyclic, it follows that Hq (M ∪ L1 ) ∼
= Hq (M ∪ L1 ∪ L2 ). Repeating this proves the result.

38
Calculations
Exercise 7.31. Consider the Klein bottle, as in Figure 1. Let P be the entire square. Then we can define
v v

v v
Figure 1: The Klein bottle

the adequate subcomplex with chains

E2 = hP i, E1 = hai ⊕ hbi, E2 = hvi.

We have

∂P = a + b + a − b
∂a = ∂b = 0
∂v = 0.

Hence it follows that we have

Z2 = 0, Z1 = hai ⊕ hbi, Z0 = hvi,
B2 = 0, B1 = h2ai, B0 = 0.

The results are obvious.

Exercise 7.32. This time, if we let a denote each edge and v denote each vertex, we have

∂P = ka, ∂a = 0, ∂v = 0.

Thus we now have

Z2 = 0, Z1 = hai ⊕ hbi, Z0 = hvi,
B2 = 0, B1 = hkai, B0 = 0.
This gives the desired homology groups.

Fundamental Groups of Polyhedra

Exercise 7.33. This is true because equality is an equivalence relation.
Exercise 7.34.
(i) It suffices to show that o(()α) cannot be changed in a single move. But this is clear. In particular, using
the definition, note that o(()α) is o(()β) if β 6= ∅, and is p if β is empty. The same holds for o(()α0 ), so
o(()α) is preserved. Similarly, e(()α) = e(()α0 ).
(ii) Again, it suffices to show this for a single elementary move. We can further assume that β = β 0 . Write
α = γ(p, q)(q, r)δ and α0 = γ(p, r)δ. Then

αβ = γ(p, q)(q, r)δβ = γ(p, r)δβ = α0 β 0 .

(Recall β = β 0 .)

39
 Exercise 7.35. An edge path, by definition, only goes along the 1-skeleton. Thus K being connected
automatically implies that K (1) is.
If K (1) is connected, then let x, y ∈ |K|. There are unique open simplices s◦ , t◦ with x ∈ s◦ and y ∈ t◦ .
Pick vertices v and w of s and t, respectively. Then consider the path taken by going straight line from x to
v, then along the edges to w, then along a straight to y. Hence |K| is connected (indeed, path-connected).
If |K| is connected, then |K| is clearly path-connected.
Finally, if |K| is path-connected, then we can find edge paths between any two vertices of K in the
following manner: Each time the path crosses the 1-skeleton, say along the edge between v and w, pick
either v and w and append that vertex (or, rather, the edge between that vertex and the previous one) to
the edge path. That this works is clear.
Exercise 7.36. This is exactly the proof of Theorem 3.6, with γ as the edge path from p0 to p1 .
Exercise 7.37. Since an elementary move only moves across a 2-simplex, it follows that the edge path group
is only dependent on the 2-skeleton.
Exercise 7.38.
(i) This is clear.
(ii) If v and w are in the same component as some point x, then by taking an edge path from v to x, then
from x to w, we have an edge path between v and w. This proves that components are connected.
Obviously the union of the components is K. To see that the unions are disjoint, suppose v ∈ [x] ∩ [y]
and w ∈ [x]. Then the path w → x → v → y implies that w ∈ [y]. Since w was arbitrary, and since
w ∈ [y] would similarly imply w ∈ [x], it follows that [x] = [y]. This proves disjointness.
(iii) Suppose [α] ∈ π(K, x). Then we claim that [α] ∈ π(L, x). But this is simply because any vertex along α
is necessarily connected to p via an edge path, hence belongs to L.
Exercise 7.39.
(i) Write α = e1 . . . em and β = em+1 . . . em+n . Then (αβ)◦ : Im+n → K takes vi to pi , where pi = α◦ (vi )
for 0 ≤ i ≤ m and pi = β ◦ (vi−m−1 ) otherwise. This is exactly γ.
(ii) It suffices to show this if α and β are separated by one step. But, writing α = γ(p, q)(q, r)δ = γ(p, r)δ = β,
simply note that we can use the straight line homotopy from the center of (p, r) to go to q. Resizing
intervals as necessary, as in the previous part, gives the result.
Exercise 7.40. It suffices to show that trees are contractible. This is true for zero or one 1-simplices. For
(n + 1) total 1-simplices, simply pick an edge one of whose endpoints is a leaf. Then we can contract that
edge to the other vertex, which is connected to the rest of the tree. Induction implies the result.
Exercise 7.41. Suppose e1 . . . en were a circuit in T1 ∪ T2 . Suppose without loss of generality that e1 ∈ T1 .
Let i and j be the first and last indices, respectively, such that ei , ej ∈ T1 ∩T2 . There is a path α which starts
with ei and ends with ej contained in T1 ∩ T2 . Now notice that e1 . . . ei−1 αej+1 . . . en is a circuit contained
entirely within T1 , contradicting that T1 is a tree.
Exercise 7.42. Let G be any abelian group, and let ϕ : {xF 0 : x → X} → G. Our goal is to show that
there is a unique homomorphism ψ : F/F 0 → G with ψ(xF 0 ) = ϕ(xF 0 ) for all xF 0 ∈ F/F 0 . (See Theorem
4.1(i).)
As in the definition of a free group, let ϕ̃ be the unique homomorphism from F to G with ϕ̃(x) = ϕ(xF 0 )
for all x ∈ X. Now define

ψ : F/F 0 → G
f F 0 7→ ϕ̃(f ).

To see that this is well-defined, notice that f ∈ F 0 implies that f = g −1 h−1 gh for some g, h ∈ F . Thus

ϕ̃(f ) = ϕ̃(g)−1 ϕ̃(h)−1 ϕ̃(g)ϕ̃(h).

But G is abelian, so this is exactly 1, which proves well-definedness.

40
 To see that ψ does indeed satisfy that ψ(xF 0 ) = ϕ(xF 0 ), simply notice that ψ(xF 0 ) = ϕ̃(x), which is
defined to be ϕ(xF 0 ).
Finally, to see that ψ is the unique homomorphism with this property, note that any other function ψ 0
would have to have ψ 0 (f F 0 ) = ϕ̃(f ), and thus be exactly equal to ψ.
Hence F/F 0 is indeed free abelian, with the desired basis.
Exercise 7.43. Exercise 7.42 shows that the rank of the free group F is the rank of the free abelian group
F/F 0 . But this latter rank is invariant with respect to X.
Exercise 7.44.
(i) By picking a maximal tree T , and setting some edge not in the tree to be x, we can see that every other
edge becomes either x, x−1 , or 1. Hence GRP 2 ,T ∼
= Z/2Z, and Corollary 7.37 implies the result.
(ii) Hurewicz’s theorem applies since RP 2 is obviously path-connected. Moreover, since Z/2Z is abelian, its
commutator subgroup is trivial. Thus H1 (RP 2 ) ∼
= Z/2Z, as desired.
Exercise 7.45.
(i) Pick points x, y ∈ X. Then consider vertices p and q of the simplices containing x and y, respectively.
Consider the following path: Take the straight line from x to p, then take the path mapped out by F (p, t)
as t ∈ I, then the path mapped out by F (q, 1 − t), and finally the straight line from q to y.
(ii) Let F : X × I → X have F (v, 0) for all v ∈ X (1) and F ( , 1) a constant function. Then by taking the
homotopy along F , we can go from (p, q) to the constant point, then back to some arbitrary edge of
T , where T is a maximal tree of X. Hence (p, q) = 1, implying a trivial edge path group. Thus the
fundamental group is trivial too.
Exercise 7.46. Since there are n vertices, we know that there are n − 1 edges of a maximal tree. The result
follows from Corollary 7.35.
Exercise 7.47. If X has m edges and n vertices, then χ(X) = −m + n. Thus 1 − χ(X) = m − n + 1. Now
use Hurewicz’s theorem, Exercise 7.35, Exercise 7.42, and Corollary 7.35 to find the result for H1 . Note that
H0 (X) = Z because X is connected, and Hq (X) = 0 for q ≥ 2 because X has dimension 1.
Exercise 7.48. We know that S m is the boundary of an (m + 1)-simplex. Thus there is an edge between
any two vertices, so we can fix one vertex p and let T be the star consisting of all edges (p, q). Now consider
any other edge (q, r). Note that {p, q, r} forms a simplex, so (p, q)(q, r) = (p, r). But in GK,T , we know that
(p, q) = (p, r) = 1, so (q, r) = 1 as well. Thus π(K, p) ∼
= GK,T = 1, and so π1 (S m ) = 1. Hence S m is simply
connected.
Exercise 7.49.
(i) Since every vertex is contained in K (q) , we can pick any simplex of maximal dimension. Its vertices are
contained in Vert(K (q) ), but it does not itself belong in the q-skeleton.
(ii) If a full subcomplex L exists, we know that it would need to include every simplex of K with vertices in
A. Moreover, adding any other simplex would introduce new vertices. Thus such a subcomplex would
be unique. Note that the set thus described is indeed a subcomplex, since any faces of s ∈ L would have
to have vertices in A as well.
The second part of the statement follows from the description of L.
Exercise 7.50. Consider some element [α] =∈ π(K, v0 ). Then there is some path α0 ' α with α0 ∈ π(L, v0 ).
Thus i[α0 ] = [α], proving surjectivity.
If K is the 2-simplex and L is its boundary, then obviously any closed edge path in K is also in L (and,
in particular, is homotopic to a closed edge path in L). But the fact that K is simply connected while L is
not implies that there cannot be an isomorphism.

The Seifert–van Kampen Theorem

No exercises!

41
8 CW Complexes
Hausdorff Quotient Spaces
Exercise 8.1. For any x, y 6= 0, let λ = xy −1 . Then x = λy, so [x] = [y]. Hence F P 0 is just a single point.
Exercise 8.2. In each case, first usepthe fact that each space is a division ring to map [x0 , x1 ] 7→ [1, x] =
[1, x−1
0 x1 ], then divide each term by 1 + ||x||2 so that the result has magnitude 1. This gives the desired
homeomorphisms.
Exercise 8.3. Note that U (R) = {±1} ≈ S 0 . To see the homeomorphism for C, use the map eiθ 7→
cos θ + i sin θ. Finally, to see the homeomorphism for H, write a given quaternion as an ordered quadruple,
and divide by its magnitude.
Exercise 8.4. Note that the real projective plane is just the quotient of S 2 , where antipodal points are
identified. This is in turn equal to the quotient of R3 where points on a line through the origin are identified,
i.e., RP 2 .

Exercise 8.5. Use the map [x] 7→ [x/|x|] to get a homeomorphism RP n 7→ S n /∼.
Exercise 8.6. Consider the map f taking [x1 , . . . , x2n+2 ] ∈ S 2n+1 /∼ to [z1 , . . . , zn+1 ] ∈ CP n , where zj =
x2j−1 + ix2j for each j. This is easily seen to be well-defined. If x, y ∈ S 2n+1 with x ∼ y, then x = λy for
some λ with |λ| = 1. If f (x) = [zi ] and f (y) = [wi ], then notice that (zi ) = λ(wi ) as well, so [zi ] = [wi ].

Exercise 8.7. The same argument as above holds, this time by defining zn = x4n−3 +ix4n−2 +jx4n−1 +kx4n .

Attaching Cells
Exercise 8.8. By Corollary 1.9, it suffices to show that α q β is constant on the fibers of v. Thus suppose
v(s) = v(t). It is sufficient to suppose that t = f (s) and s ∈ A, since the relation ∼ is generated by all
(a, f (a)). But if t = f (s), then we have

(α q β)(s) = α(s) = β(f (s)) = β(t) = (α q β)(t).

This proves the result.

Exercise 8.9.
(i) It is clear that B and B −1 are contained in the equivalence relation generated by B. Note that D is as
well due to reflexivity. Finally, K is attained by (a, f (a))(f (a), a0 ) = (a, a0 ), where f (a0 ) = f (a). Now
note that repeating this with (a0 , f (a0 )) simply takes us back to (a, f (a)), so there are no other elements
in the equivalence relation.
(ii) Simply note that

K = {(a, a0 ) : f (a) = f (a0 )}

= (f × f )−1 {(x, y) ∈ im(f × f ) : x = y}
= (f × f )−1 (∆ ∩ im(f × f )).

This is exactly what we wanted.

Exercise 8.10. It is easy to verify that the diagram commutes. Now suppose we have some Z with functions
α : X → Z and β : Y → Z so that β ◦ f α ◦ i. We would like to find a function ϕ : X qf Y → Z making the
pushout diagram commute.
We will first show that, if such a function exists, then it must be unique. Since the maps X → X qf Y
and Y → X qf Y are induced by v, it follows that ϕ ◦ v|X = α and similarly for Y and β. Hence ϕ ◦ v
would have to be equal to α q β, i.e., ϕ = (α q β) ◦ v −1 . Note that Exercise 8.8 applies, so this is indeed a
well-defined map.

42
 Exercise 8.11. The only case too check is if x ∈ X and y ∈ Y . Note that there exists some a ∈ A, so
consider the path from x to a = f (a) to y. Hence X qf Y is path-connected.
Exercise 8.12.
(i) The equation given in the hint follows from Exercise 8.9. Now, for the forwards direction, observe that
v(C) closed implies v −1 v(C) closed in X q Y . Hence its intersection with Y is closed in Y . But notice
that f (C ∩ A) ∩ Y = f (C ∩ A). Furthermore, we know that f −1 (f (C ∩ A)) and f −1 (C ∩ Y ) are completely
disjoint from Y . Thus
v −1 v(C) ∩ Y = (C ∩ Y ) ∪ f (C ∩ A)
is closed in Y , as desired.
Going backwards, observe that v −1 v(C) ∩ Y is closed in Y , using the hypothesis and the argument
above. Moreover, since f is continuous, the hypothesis implies that
f −1 ((C ∩ Y ) ∪ f (C ∩ A)) = f −1 (C ∩ Y ) ∪ f −1 (f (C ∩ A))
is closed. Since C ∩ X is closed in X and f (C ∩ A) ∩ X = ∅, this implies that v −1 v(C) ∩ X is closed in X.
Hence v −1 v(C) is closed in X q Y . Since v is an identification, it follows that v(C) is closed in X qf Y .
(ii) The function is clearly bijective and continuous. To see that it is a homeomorphism, we will show that
it is a closed map. Thus suppose C ⊆ Y is closed. Obviously, i(C) ⊆ X q Y is closed and has empty
intersection with X. Then to see that v(i(C)) is closed in X qf Y , simply use the previous part. In
particular, observe that
i(C) ∩ Y = i(C),
which is closed in Y , while
f (i(C) ∩ A) = ∅,
which is also closed, so that their union is closed. Thus v(i(C)) is closed in X qf Y , proving that the
given function is a homeomorphism.
(iii) Again, this is clearly bijective and continuous. Since X − A is open in X, if U is open in X − A, then
it is also open in X. Note that i(U ) is open in X q Y . Now note that i(U )c is closed in X q Y , and
its intersection with X is U c , which is closed in X. Moreover, we know that its intersection with Y is
Y itself, while f (i(U )c ∩ A) = f (A). Since Y ∪ f (A) = Y , which is closed in Y , part (i) implies that
v(i(U )c ), which, by surjectivity of v, is exactly v(i(U ))c , is closed. Thus v(i(U )) is open, proving that
this is an open, bijective, continuous map, thus a homeomorphism.
(iv) Note that Φ takes A ⊆ X to A ⊆ X q Y , which is then exactly equal to the attached region of X qf Y .
Exercise 8.13.
(i) Since f is from a compact set to a Hausdorff set, it is closed. Let C be closed. Then A being compact
implies that it is closed, so C ∩ A is closed in X. Thus f (C ∩ A) is closed in Y . Since C ∩ Y is closed in
Y , it follows from part (i) that v(C) is closed in X qf Y .
(ii) First, suppose that z ∈ im Φ|A. Then there is some x ∈ X with v(i(x)) = z, so i(x) is in the fiber. We
know that {z} is closed because Y is Hausdorff, so v −1 (z) is also closed. Since v −1 (z) ⊆ A, and closed
subsets of compact sets are compact, it follows that v −1 (z) is compact.
Otherwise, we know that we can use either the homeomorphism in Exercise 8.12(ii) or the homeo-
morphism Φ|(X − A) to show that v −1 (z) = {z}, which is indeed a nonempty compact subset of X.
Exercise 8.14. This is just invariance of boundary. Alternatively, see the proof of Lemma 8.15.
Exercise 8.15. If n = 0, this is obviously true. Otherwise, let e = s− ṡ ≈ Dn−1 −S n−1 and let Y = |K (n−1) |
be a closed subset of |K| (since it’s the finite union of (closed) simplices). Then e ∩ Y = ∅ and e is an n-cell.
Hence Theorem 8.7 says that we need only exhibit a relative homeomorphism Φ : (Dn , S n−1 ) → (e ∪ Y, Y ).
But letting Φ be the obvious homeomorphism from Dn to s works.
Exercise 8.16. Write Y = {y}. Then define the relative homeomorphism Φ : (Dn , S n−1 ) → (en ∪ Y, Y )
which takes Dn − S n−1 to en in the obvious way, and takes x ∈ S n−1 to y. Theorrem 8.7 tells us that the
attachment of Dn to Y along f = Φ|S n−1 is a homeomorphisms between Dn /∂Dn = S n and en ∪Y ≈ en ∪e0 .

43
 Homology and Attaching Cells
Exercise 8.17. Note that χ(K) = 1 − 2 + 1 = 0, so rank H2 (K) + 1 = rank H1 (K). Furthermore, doing the
same thing as with the torus in Example 8.7, we see that the projections are f α ∗ f α1−1 , which has degree
0, and f β ∗ f β1 , which has degree 2. Thus, since H1 (S 1 ∨ S 1 ) ∼
= H1 (S 1 ) ⊕ H1 (S 1 ), we can consider f∗ to be
the map x 7→ (0, 2x). It has trivial kernel and image isomorphic to Z ⊕ Z/2Z. Working through the exact
sequence in Theorem 8.11 gives the result.
Exercise 8.18. There are a couple typos here: In the first part, the wedge for M is of 2h circles, and in the
second part, we should have χ(M ) = 2 − 2h, not χ(M ) = h.

(i) Note that each (αi , αi−1 ) and (βi , βi−1 ) pair gives a circle. Since all the vertices are identified with each
other, this gives us the desired wedge product. More formally, we can define a function Φ from a polygon
P to W , and let f = Φ|∂P . Then f αi = (f αi−1 )−1 , and similarly for β, which gives us our 2h circles. A
similar argument can be done for M 0 .
(ii) Note that H2 (S 1 ∨ · · · ∨ S 1 ) = 0. Thus we have the following exact sequence:

f∗ i∗
0 H2 (M ) H1 (S 1 ) H1 (S 1 ∨ · · · ∨ S 1 ) H1 (M ) Z Z2 Z 0 ,

where the last few terms are just H0 (S 1 ), Z ⊕ H0 (S 1 ∨ · · · ∨ S 1 ), and H0 (M ), since all three spaces
are path-connected. The fact that this sequence is exact implies that Z2 → Z is a surjection, so the
map Z → Z2 is an injection. Hence H1 (M ) → Z is the zero map. Thus i∗ is surjective and has kernel
(isomorphic to) Z2n /H1 (M ).
Looking at the maps from left to right now, observe that H2 (M ) → H1 (S 1 ) = Z is injective. Thus
ker f∗ = H2 (M ), so im f∗ = H1 (S 1 )/H2 (M ). But ker i∗ = im f∗ , and so it follows that

χ(M ) = rank H2 (M ) − rank H1 (M ) + rank H0 (M ) = 2 − 2h,

where we use the fact that rank H0 (M ) = 1.

Now notice that the same argument as in Example 8.7 implies that f∗ is the zero map. Thus

H2 (M ) = ker f∗ = H1 (S 1 ) = Z.

For H1 (M ), since the flanking terms are torsion-free, it follows that H1 (M ) is also torsion-free. Since it
has rank 2h, the result follows.
(iii) The same argument as before shows that χ(M 0 ) = 2 − n. This time, however, the map f∗ is not the zero
map. In particular, by composing with projections, we find that f∗ : H1 (S 1 ) → H1 (S 1 ∨ · · · ∨ S 1 ) takes
x 7→ (2x, . . . , 2x), where we have identified H1 (S 1 ∨ · · · ∨ S 1 ) with H1 (S 1 ) ⊕ · · · ⊕ H1 (S 1 ).
In particular, we have ker f∗ = 0 and im f∗ = (Z/2Z)n . The argument before shows that ker f∗ =
H2 (M 0 ), and so H2 (M 0 ) = 0. Using the Euler characteristic (i.e., a rank argument), we can conclude
that rank H1 (M 0 ) = n − 1. (Note that, this time, the first homology group isn’t torsion-free, thanks to
the Z/2Z terms.)
(iv) We first consider M . Note that it only has one vertex, say v. Thus, with chains

E2 = hW i, E1 = hα1 i ⊕ hβ1 i ⊕ · · · ⊕ hαn i ⊕ hβn i, E0 = hvi,

we have
∂W = α1 + β1 − α1 − β1 + · · · = 0, ∂αi = ∂βi = v − v = 0, ∂v = 0.
Hence it follows that
Z2 = hP i, Z1 = hα1 i ⊕ hβ1 i ⊕ . . . , Z0 = hvi,
B2 = 0, B1 = 0, B0 = 0.
Thus we have
H2 (M ) = Z, H1 (M ) = Z2h , H0 (M ) = Z.

44
 Now, for M 0 , with the natural chains, we have

∂P = 2α1 + · · · + 2αn , ∂αi = 0, ∂v = 0.

Thus we conclude that

Z2 = 0 Z1 = hα1 i ⊕ · · · ⊕ hαn i, Z0 = hvi,
B2 = 0, B1 = h2(α1 + · · · + αn )i, B0 = 0.

This gives us that

H2 (M 0 ) = 0, H1 (M 0 ) = Zn /2Z = Z/2Z ⊕ Zn−1 , H0 (M 0 ) = Z,

which coincides with the previous parts.

CW Complexes
Exercise 8.19. Note that U ⊆ X is open if and only if U c ⊆ X is closed, which is in turn the case if and
only if U c ∩ Aj is closed in Aj for all j ∈ J. But U ∩ Aj = Aj − U c ∩ Aj , so this last conditiono is true if
and only if U ∩ Aj is open in Aj for j ∈ J.
Exercise 8.20. It is obvious that {Y ∩ Aj } fits the conditions (i)–(iii). Now note that if F ⊆ Y is closed in
the subspace topology, then F = Y ∩ F 0 for some closed F 0 ⊆ X. Hence

F ∩ (Y ∩ Aj ) = Y ∩ F 0 ∩ Y ∩ Fj = (Y ∩ Aj ) ∩ F 0 .

Of course, this is closed in Y ∩ Aj , so F must be closed in the weak topology.

Now suppose F is closed in the weak topology. Then F ∩ (Y ∩ Aj ) is closed in all Y ∩ Aj . Since Y is
closed, we know that F ∩ Y ∩ Aj must also be closed in Aj . This is true for all j, so F ∩ Y is closed in the
weak topology on X, i.e., as a subset of X. Thus F = F ∩ Y is closed as a subspace of Y .
Exercise 8.21. By Lemma 8.20, we know that A closed in X implies that A ∩ X 0 is closed in X 0 for all finite
subcomplexes X 0 . Theorem 8.19 says that this implies that, for all compact K of X, we must have A ∩ K
closed in K. Thus, by definition, it follows that A is closed in the weak topology generated by compact
subsets.
Now suppose A ∩ K is closed in K for all compact K. Since finite subcomplexes are compact, it follows
that A ∩ X 0 is closed in X 0 for all such X 0 . Hence A is closed in X.
Exercise 8.22. To see that X (0) is discrete, simply let A ⊆ X (0) . Then A ∩ ē is the finite union of 0-cells,
and hence is closed.
To see that the 0-skeleton is closed, note that X (0) ∩ ē is a finite union of 0-cells, and thus is closed in ē.
This is true for all e, so X (0) is closed.
Exercise 8.23. If A is closed, then obviously A ∩ X (n) is closed in X (n) . Now if A ∩ X (n) is closed in
X (n) for each n, pick X 0 to be any finite complex. Let n be the highest dimension of any cell in X 0 . Then
A ∩ X 0 = (A ∩ X (n) ) ∩ X 0 must be closed in X 0 . Lemma 8.20 implies the result.
Finally, the corresponding statement for open sets follows from Exercise 8.19.
Exercise 8.24. This is simply because each n-cell is just Dn − S n−1 ; the attachment is given by Φe (S n−1 )
according to Exercise 8.12(iv).
Exercise 8.25. This is visually clear. Alternatively, with α and β as the edges, and v as the vertex, we can
notice that there is a relative homeomorphism

Φ : (D2 , S 1 ) → (T, α ∪ β ∪ {v})

since D2 ∼
= I × I. Since α and β are 1-cells, and v is a 0-cell, it follows that this map gives T as the union of
two 1-cells, one 0-cell, and one 2-cell (namely im Φ|(D2 − S 1 )).
The same argument can be done for the Klein bottle.

45
Exercise 8.26.
(i) They both violate closure finiteness since the closure of the base point intersects infinitely many cells.
(ii) The set of all {1/n} is closed in the weak topology, but not as a subspace.
Exercise 8.27. The same proof as Theorem 7.1 holds, but with Dn in place of ∆n .

Exercise 8.28. First, observe that the 1-skeleton is always nonempty (as long as X is nonempty). In
particular, suppose e is an n-cell in X, where n is the smallest dimension of a cell in e. Then the relative
homeomorphism (Dn , S n−1 ) → (e∪X (n−1) , X (n−1) ) implies that there is a map between S n−1 and X (n−1) =
∅, which is impossible. Thus there must be some 0-cell, and so the 1-skeleton is nonempty.
In fact, there must be some part of the 1-skeleton in each path component. Thus if X is disconnected,
then its 1-skeleton must be as well.
Now suppose that the 1-skeleton is disconnected. We can easily show that X (n) disconnected implies that
(n+1)
X is disconnected. Since X is the union of all its skeletons, and since X (n) ⊆ X (n+1) , it follows that X
being connected would have to imply that there is some n with X (n) connected. Since X (0) is discrete, hence
disconnected (unless it has one element only, in which case the 1-skeleton would be connected), it follows
that n ≥ 1, and so this provides the desired contradiction.

Exercise 8.29. The forward direction is obvious. Now suppose that f Φe is continuous for all e. Let K ⊆ Y
be closed and let e be a k-cell. We want to show that

f −1 ∩ Φe (Dk )

is closed in ē = Φe (Dk ). But Φe is a relative homeomorphism and is, in particular, a closed map on
Dk − S k−1 . Now, because
Φ−1
e (f
−1
(K) ∩ Φe (Dk )) = (f Φe )−1 (K) ∩ Dk
is closed in Dk , we’re done.
Exercise 8.30.
(i) Consider attaching the (closed) top half of the circle to the topologist’s sine curve (which maps 0 to 0
and x to sin(1/x) for x ∈ (0, 2π]). Then attach the (closed) bottom half of the circle to the same curve,
but running backwards. Obviously this is a CW complex. But it is connected and not path-connected,
violating Exercise 7.35. Hence this is not a polyhedron.
(ii) If n = 0, this is obvious. Suppose this is true for n. Say we attach k total (n + 1)-cells. (Note that this
kind of inductive creation of CW complexes is made possible byy Theorem 8.24.) Note that an (n+1)-cell
is homeomorphic to an open (n + 1)-simplex. Furthermore, the attachment map can be approximated by
a simplicial map. Since simplicial approximations are homotopic to the original maps, the result follows.
Exercise 8.31. Here we can use the same cells and attaching maps, only with the basepoints all identified.
For any cell not equal to the basepoint, its closure is contained in whichever Xλ the cell was originally in,
and thus intersects only finitely many cells. The closure of the basepoint is itself, and only intersects itself.
This proves closure finiteness.
W simply note that if A is closed, then A ∩ Xλ is closed, where we
To see that this has the weak topology,
identify Xλ with its natural image in Xλ . Thus, since the closure of any cell is contained within Xλ for
some λ, it follows that A ∩ ē = A ∩ Xλ ∩ ē is closed in ē for every e.
Exercise 8.32. The first and second conditions of a CW complex are clearly satisfied since Di+j is home-
omorphic to [0, 1]i+j .
To see the third condition holds, use the equation in the hint. Notice that all four expressions on the
right side intersect finitely many cells in X or X 0 . In particular, it follows that (ē − e) × ē0 intersects finitely
many cells of E 00 , and similarly for the other term. Thus e × e0 intersects e × e0 , plus these finitely many
other cells. This proves closure finiteness.
Finally, for the fourth condition, note that the weak topology is just the product topology when working
with finitely many factors.

46
Exercise 8.33. With notation as suggested in the hint, suppose the intersection of A and every cell in E 00
is closed in the cell. Now observe that
e × a0 = ē × a0 ,
and similarly for b0 . Moreover, we know that
e × c1 = (ē − e) × I ∪ ē × (a0 ∪ b0 ) ∪ (e × c1 ).
 

But now observe that ē = e ∪ (ē − e), so that the middle term can be rewritten as
ē × (a0 ∪ b0 ) = e × (a0 ∪ b0 ) ∪ (ē − e) × (a0 ∪ b0 ) .

Since a0 ∪ b0 ∪ c1 = I, it then follows that

e × c1 = (ē − e) × I ∪ (e × I) ∪ (ē − e) × (a0 ∪ b0 ) = ē × I.
 

Now let πX and πI be the projections to X and I, respectively. We know that πX (A) ∩ ē is closed in
each e, since πX (A) ∩ ē is closed in each ē, and similarly for πI (A) ∩ a¯0 and πI (A) ∩ b¯0 . Moreover, since
πI (A) ∩ c¯1 = πI (A) ∩ I, and since A ∩ (ē × I) is closed in ē × I, it follows that the intersection πI (A) ∩ c¯1 is
also closed in I.
Thus πX (A) and πI (A) are closed, so A is closed, as desired.
Exercise 8.34. Let i : Z → Y and j : Y → X be the injections. We can now easily check the criteria for a
strong deformation retraction. In particular, note that
r1 r2 ji = r1 1Y i = r1 i = 1Z ,
while
jir1 r2 = j(ir1 )r2 ' jr 2 rel Z ' 1X rel Z,
since Y ⊆ Z.

Cellular Homology
Exercise 8.35. /
Exercise 8.36.
(i) If X is compact, then it is finite. Thus Wk (X, Y ) = Hk (XYk , XYk−1 ) is free abelian of rank equal to the
number of k-cells in E − E 0 . Say this rank is rk . Then we know that
Hk (X, Y ) ∼
= Hk (W∗ (X, Y )) = ker dk / im dk+1 ,
but both ker dk and im dk+1 are subsets of Zrk . Thus Hk (X, Y ) is finitely generated.
(ii) This is the same proof, since rk is at most the number of cells of dimension k.
Exercise 8.37. Using the cellular decomposition for RP ∞ = RP n , we find that
S

Wk (RP ∞ ) = Hk (e0 ∪ · · · ∪ ek , e0 ∪ · · · ∪ ek−1 ),

which is obviously free abelian of rank 1. It follows that the we get a chain · · · → Z → Z → Z → . . . , so the
kernels and images of each map must be 0 or Z. Hence Hk (RP ∞ ) = Hk (W∗ (RP ∞ )) is either 0 or Z.
W `
Exercise 8.38. Let bλ be the basepoint of Xλ . Then we know that Xλ = Xλ /{bλ }. Let v be the
natural map. Then Theorem 8.41 implies that v∗ induces an isomorphism from
a  _ 
Hk Xλ , {bλ } → H̃k Xλ .

Of course, Theorems 5.13 and 5.17 also imply that the left side is equal to
X
H̃k (Xλ ),
λ

which implies the result.

47
 Exercise 8.39. To do this, we simply compute d2 , d1 , d0 . In particular, since W2 (T ) is generated by e2 , we
know that d2 = ∂e2 = 0. Similarly, we find that d1 = d0 = 0. This gives the result.
Exercise 8.40. Use Exercise 7.19. In particular, this implies that
(
m n 0 if m or n odd
χ(S × S ) = .
4otherwise

Exercise 8.41. Use the cellular decomposition of CP n . In particular, we know that CP n = e0 ∪ · · · ∪ e2n ,
and so the only nonzero αi are for even i. Thus
X
χ(CP n ) = (−1)i αi = 1 + 1 + · · · + 1 = n + 1.

The same argument holds for HP n .

Exercise 8.42. This is again obvious:
χ(RP n ) = 1 − 1 + 1 − 1 + . . .
1
is equal to 0 if n is odd and 2 if n is even. This is exactly 2 (1 + (−1)n ).
Exercise 8.43. This is simply the principle of inclusion-exclusion.
Exercise 8.44. It is sufficient to show that ∼ is closed. Notice that
p
[ p
[
{(z0 , z1 , z2 , z3 ) : hm (z0 , z1 ) = (z2 , z3 )} = {(z0 , z1 , z2 , z3 ) : hm (z0 , z1 ) = (z2 , z3 )} = Sm .
m=1 m=1

because hp = h. Thus it suffices to check that each Sm is closed. Suppose that (z0 , z1 , z2 , z3 ) 6∈ Sm . Say that
z2 6= ζ m z0 ; note that a similar argument can be given if z3 6= ζ mq z1 . Then there is an open neighborhood
with coordinates (x0 , x1 , x2 , x3 ) on which ζ m (x0 ) 6= x3 since ζ m x − y is continuous. Thus (Sm )c is open,
which proves that Sm is closed, as desired.
Exercise 8.45.
(i) Note that ζ = 1, so h is just the identity. Thus S 3 /∼ = S 3
(ii) Now we have ζ = −1, so h maps antipodal points to each other. Thus S 3 /∼ = RP 3 .
0
(iii) In this case, we know that ζ q = ζ q , so hq = hq0 . Thus L(p, q) = L(p, q 0 ).
Exercise 8.46.
(i) Since this is a finite decomposition, we only need to verify the first two conditions for a CW complex.
The first is clear by definition. For the second condition, the maps are obvious for e0r and e1r . For e2r , we
use the fact that z1 = z1 (z0 ) is determined by z0 . Thus the map
z0 7→ (z0 , z1 (z0 ))
works. Finally, for e3r , take (z0 , θ) and map θ linearly onto (2πr/p, 2π(r + 1)/p).
(ii) It is easy to check that eir ∼ eir0 for each i.
Exercise 8.47.
(i) This is the cellular boundary formula, or just a generalization of the argument for Lemma 8.46.
(ii) For D(γ1 ), simply notice that
−1
D(γ1 ) = v# d1 v# (γ1 ) = v# d1 e1r = v# (e0r − e0r+1 ) = 0.
A similar argument holds for the other differentiations.
(iii) This is obvious from the chain complex:
0 ×p 0
W4 = 0 Z Z Z Z 0 = W−1

48
9 Natural Transformations
Definitions and Examples
Exercise 9.1. This is obvious from Lemma 4.8.
Exercise 9.2. This is exactly the statement of Exercise 4.10.
Exercise 9.3. The commutative diagram in Exercise 4.13 is exactly the statement that the map is natural.
Exercise 9.4. Commutativity of the diagram
f∗
Hn (X, A) Hn (Y, B)
∂ ∂

Hn−1 (A, ∅) Hn−1 (B, ∅)

(f |A)∗

follows from the exact sequence in Theorem 5.8, since Hn−1 (A, ∅) = Hn−1 (A).
Exercise 9.5. This is again precisely the statement from Exercise 6.8.
Exercise 9.6.
(i) Suppose σ : F → G and τ : G → H are natural. Then we can “stack” the commutative diagrams:
Ff
F (C) F (D)
σC σD
Gf
G(C) G(D)
τC τD
Hf
H(C) H(D)

Hence it follows that τ σ = (τC σC ) gives a natural transformation.

(ii) Reflexivity is due to the commutativity of the following diagram:
Ff
F (C) G(C)
1F (C) 1G(C)
Ff
F (C) G(C)

To see symmetry, simply choose τC−1 for each object C. This can be done because each τC is an equivalence.
Finally, transitivity follows from the previous part and the fact that the composition of equivalences is
an equivalence.
Exercise 9.7.
(i) If ϕ ∈ Nat(Hom( , A), F ), then ϕA is a map from Hom(A, A) to F (A). Since 1A ∈ Hom(A, A), it follows
that ϕA (1A ) ∈ F (A). Thus y is a well-defined function.
(ii) We must check that τ ∈ Nat(Hom( , A), F ) whenever µ ∈ F (A). First, observe that τX is indeed
a morphism from Hom(X, A) to F (X). After all, if f : X → A, then F f : F A → F X. Hence
τX (f ) = (F f )(µ) is an element of F (X).
To see that τ is natural, we must show that the following diagram commutes for all f : X → Y .
Hom(f ,A)
Hom(X, A) Hom(Y, A)
τX τY

F (X) Ff
F (Y )

49
 But for each g ∈ Hom(X, A), we know that

(F f ) ◦ τY (g) = (F f )(F g(µ)) = F (g ◦ f )µ,

while we have Hom(f, A) = f ∗ , so that

τX ◦ Hom(f, A)(g) = τX ◦ f ∗ (g) = τX (g ◦ f ) = F (g ◦ f )µ.

These are equal, so τ is a natural transformation.

(iii) First, we will show that y ◦ y 0 : F (A) → F (A) is the identity. Let µ ∈ F (A). Then we know that

y 0 (µ) = {τX : f 7→ (F f )(µ)},

and so we have that

y(y 0 (µ)) = (y 0 (µ))A (1A ) = F (1A )(µ).
But F is a functor, so F (1A ) = 1F (A) , and so this is exactly equal to 1F (A) (µ) = µ, which proves that
y ◦ y 0 is the identity on F (A).
Now to check y 0 y, suppose ϕ ∈ Nat(Hom( , A), F ). Then we know that

y 0 (y(ϕ)) = {τX : f 7→ (F f )(ϕA (1A )).

We would like to show that

(F f )(ϕA (1A )) = ϕX f,
since this will imply that y 0 (y(ϕ)) = ϕ. But we know that the following diagram commutes:

f∗
Hom(X, A) Hom(A, A)

ϕX ϕA

F (X) Ff
F (A)

Thus we know, in particular, that

F f ◦ ϕA (1A ) = ϕX f ∗ (1A ) = ϕX (1A ◦ f ) = ϕx ◦ f.

This is what we wanted.

(iv) If ϕ : Hom( , A) → Hom( , B) is natural, then we have the following commutative diagram:

Hom(f ,A)
Hom(X, A) Hom(Y, A)

ϕX ϕY

Hom(X, B) Hom(Y, B)
Hom(f ,B)

Let F be the functor Hom( , B). Then we know that

ϕX (f ) = y 0 (y(ϕ))X (f )
= (F f )(ϕA (1A ))
= Hom(f, B)(ϕA (1A ))
= ϕA (1A ) ◦ f,

where f : X → A. Thus ϕX (f ) = µf , as desired.

(v) Same proof.

50
Exercise 9.8. We must verify the properties of a category. To see that the family of Hom(F, G)’s, where F
and G are functors C → A , is disjoint, notice that this means that there exists some τ = (τC : F (C) → G(C))
and σ = (σC : F 0 (C) → G0 (C)) which are equal. Hence F (C) = F 0 (C) and G(C) = G0 (C) for all C, since
τC = σC is in both Hom(F (C), G(C)) and Hom(F 0 (C), G0 (C)). Since this is true for all C ∈ C , it follows
that F = F 0 and G = G0 .
Composition of natural transformations reduces to composition of morphisms, which is associative.
Finally, note that 1A ∈ Hom(F, F ) given by

1A = {(1A )C = 1F (C) }

works as an identity morphism.

Exercise 9.9.
(i) We shall verify the properties of a contravariant functor. The functor gives us a complex

∂n+1 ∂n
... Cn+1 Cn Cn−1 ... C0 C−1 ...

Since Cn is abelian, we know that n ∈ Z implies that C(n) = Cn ∈ Ab.

The only morphisms in Z are ιxy when x ≤ y. Note that C(ιxy ) is the composition ∂x+1 ◦ · · · ◦ ∂y :
Cy → Cx . We must verify that composition is reversed and identities are respected. But this is clear
from the definition:
]iotayz ◦ ιxy = ιxz = ∂x+1 ◦ · · · ◦ ∂z
is exactly C(ιxy ) ◦ C(ιyz ), and C(ιxx ) is the composition of an empty set of differentiation operators, and
thus is the identity.
(ii) The chain map condition is exactly the condition of commutativity.

Eilenberg–Steenrod Axioms
No exercises!

Chain Equivalences
Exercise 9.10. To prove (i) implies (ii), note that ps = 1C implies s is injective. Then the same argument
as in Corollary 9.2 implies that B = ker p ⊕ im s. Of course, we have ker p = im i and C 0 = im s = s(C) ∼
= C.
Since p(C 0 ) = C, this proves the first implication.
The second implication is clear. In particular, consider q : B → A defined by (i(x), c) 7→ x. Then
qi(a) = q(i(a)) = a.
Finally, to show (iii) implies (i), define s(c) as

s(c) = p−1 (c) − iqp−1 (c).

To see that this is well-defined, pick b ∈ ker p = im i, so b = i(a). Thus b − iq(b) = i(a) − iqi(a) = 0. Hence
p(b) = p(b0 ) means that b − iqb = b0 − iqb0 , proving well-definedness. To see that this choice of s gives a split
exact sequence, simply verify that

ps(c) = p(p−1 (c) − iqp−1 (c)) = c − piqp−1 (c).

Since pi = 0, this is equal to c.

Acyclic Models
Exercise 9.11. First we show that the diagram given by Rotman commutes, i.e., that

∂n (tn − t;n −sn−1 dn ) = 0.

51
We know that

∂n tn − ∂n t0n − ∂n sn−1 dn = tn−1 dn − t0n−1 dn − ∂n sn−1 dn .

The inductive hypothesis implies that

∂n sn−1 = tn−1 − t0n−1 − sn−2 dn−1 .

Plugging this value in and canceling gives us

∂n tn − ∂n t0n − ∂n sn−1 dn = sn−2 dn−1 dn = 0,

because dd = 0.
Thus the diagram commutes. In particular, we know that

im(tn − t0n − sn−1 dn ) ⊆ ker ∂n = im ∂n+1 ,

where the final equality comes from the fact that E∗ is an acyclic complex. This means that we can rewrite
the diagram as follows:
Fn

tn −t0n −sn−1 dn

En+1 ∂n+1
im(tn −t0n −sn−1 dn ) ∂n =0
0
= im ∂n+1 = ker ∂n

Thus Theorem 9.1 implies that we can find sn with the desired properties.
Exercise 9.12. We have F (g) = F (0 + g) = F (0) + F (g), so F (0) acts as the 0 element. If A is the zero
group, then its identity is the zero homomorphism. Hence 1F (A) = F (1A ) is the zero homomorphism, so
F (A) = 0.
Exercise 9.13.
(i) We’ll prove the covariant case. By Exercise 9.10, we have a morphism q : A → B with qi = 1A . Note
that (F p) ◦ (F s) = F (p ◦ s) = F (1C ) = 1F (C) , and similarly for q and i, so that we still have a split
sequence, as long as it is exact. Moreover, these imply that F p is surjective and F i is injective.
It now suffices to check that im F i = ker F p. But notice that B ∼= iq(B) ⊕ sp(B) implies that F (B) is
equal to the functored version of the right side, thus making the center of the short functored sequence
exact.
(ii) This simply uses induction on |I| = n + 1 and the following short exact sequence:
Pn i Pn+1 p
0 i=1 Ai i=1 Ai An+1 0.

Note that this is split exact with s : an+1 7→ (0, . . . , 0, an+1 ). Thus the previous part applies, and
Exercise 9.10 implies that
n+1
! n
! n+1
Ai ∼
X X X
F = F A i ⊕ F (A n+1 ) = F (Ai ),
i=1 i=1 i=1

where the last equality follows from the inductive hypothesis.

Exercise 9.14.
(i) If ∂n ∂n+1 = 0, then F (∂n ∂n+1 ) = 0 thanks to additivity. This proves that the functored complex is a
chain complex too.
(ii) Note that fn−1 ∂n = ∂n0 fn implies that F (fn−1 ∂n ) = F (∂n0 fn ). Since functors respect composition, this
proves the result.

52
(iii) Note that additive functors respect homotopy because they rerspect both composition and addition.
Hence if g : B∗ → A∗ makes f an equivalence, i.e., if g ◦ f ' 1A∗ and f ◦ g ' 1B∗ , then it follows that

F g ◦ F f ' F 1A∗ = 1F A∗ ,

and similarly for B. Hence F g is an inverse for F f , so F f is a chain equivalence.

Exercise 9.15.
(i) This simply involves applying Corollary 9.13(ii). In particulars, we know that Fp and Sp arer both
free with basis in M = {∆p }. We want to show that ∆p is totally S- and F -acyclic. But notice that
H̃n (S∗ (∆k )) = 0 because ∆k is contractible, and similarly for Fp , since it coincides with Cp ono ∆. This
proves acyclicity, and so the two are naturally chain equivalent.
(ii) Theorem 9.8 implies that singular and large simplicial homology are the same, while Theorem 7.22 implies
that normal simplicial and singular homology are the same.

Lefschetz Fixed Point Theorem

Exercise 9.16. Notice that 1G induces 1G/tG : x + tG 7→ x + tG. Hence, with any basis {x1 , . . . , xn } of
G/tG, we have 1G/tG equal to the identity matrix whose dimension is rank G/tG.
Exercise 9.17. A basis {x1 , . . . , xk } of G0 /tG0 can be extended to {x1 , . . . , xn } of G/tG. Since G00 is just
G/G0 and f 00 (g + G0 ) = pf (g) = f (g) + G0 . Thus f 00 (xi + G0 ) = f (xi ) + G0 for i = k + 1, . . . , n. Thus the
matrix of f¯ is diagonal, of the form shown on p. 259 of the textbook, which implies the result.
Exercise 9.18. If f : S n → S n , then f0∗ and fn∗ are maps Z → Z. Note that f0∗ is the identity, and thus
has trace 1. If tr fn∗ = 1 as well, then the whole map is homotopic to either the identity or the antipodal
map, implying that f is a homotopy equivalence. Thus tr fn∗ = 0, and so λ(f ) = 1 6= 0. The Lefschetz fixed
point theorem implies the result.

Tensor Products
Exercise 9.19. Note that

a ⊗ 0 + a0 ⊗ b0 = a ⊗ 0 + (a ⊗ b0 + (a0 − a) ⊗ b0 )
= a ⊗ b0 + (a0 − a) ⊗ b0
= a0 ⊗ b0 ,

and similarly for 0 ⊗ b.

Exercise 9.20. We would like to show that m(a, b) ∼ (ma, b) for m ∈ Z. It is true for m > 0 by induction,
true for m = 0 by Exercise 9.19, and true for m < 0 by inverses.
Exercise 9.21. The hint gives the full solution. If a ∈ A then there exists some m > 0 so that ma = 0.
Hence a ⊗ q = ma ⊗ (q/m) = 0. Since this is true fora all generators of A ⊗ Q, the result follows.
Exercise 9.22. Let m be the order of a ∈ A and n the order of b ∈ B. Then we know that gcd(m, n) = 1,
so that there exist integers x, y with mx + ny = 1. Hence we have that

a ⊗ b = a ⊗ (mx + ny)b
= (mx + ny)(a ⊗ b)
= mx(a ⊗ b) + ny(a ⊗ b)
= (mxa ⊗ b) + (a ⊗ nyb) = 0.

Exercise 9.23. This is the exact same argument as Corollary 9.27.

Exercise 9.24.

53
(i) Use Theorem 9.25(ii) with the fact that A × B ∼
= B × A.
(ii) To see this, simply consider the following commutative diagram:

1A ⊗f
A⊗B A⊗C

B⊗A f ⊗1A
C ⊗A

Exercise 9.25. Note that TA (f + g) = 1A ⊗ f + 1A ⊗ g, since both maps complete the diagram

A×B A⊗B
ϕ

A⊗B

where ϕ(a, b) = (a, (f + g)(b)). But note that 1A ⊗ f + 1A ⊗ g is just TA (f ) + TA (g), proving additivity.
Exercise 9.26. This is clear, since we have

1A ⊗ f : A ⊗ B → A × B
a ⊗ b 7→ a ⊗ f b = a ⊗ mb = m(a ⊗ b).

Exercise 9.27.
(i) This is easy to show directly. In particular, we show that a 7→ 1 ⊗ a is an isomorphism. We would like to
show that 1 ⊗ a = n ⊗ b if nb = a. But n ⊗ b = n(1 ⊗ b) = 1 ⊗ (nb) = 1 ⊗ a, as desired. Hence this map
is surjective. It is injective because, otherwise, every 1 ⊗ a would be 0, which would violate the universal
property of tensor products given by Theorem 9.25. Hence this is an isomorphism.
(ii) We must show that the following commutes:

1A ⊗f
Z⊗A Z⊗B
τA τB

A f
B

This commutes because

τB ◦ (1A ⊗ f ) : (n, a) 7→ (n, f (a)) 7→ nf (a),
while
f ◦ τA : (n, a) 7→ na 7→ f (na),
and nf (a) = f (na) since f is a homomorphism.

Universal Coefficients
Exercise 9.28.
Aj where Aj = Zxj , and F 0 = A0k where A0k = Zx0k . Then F ⊗ F 0 is just
P P
(i) We can write F =
X X
F ⊗ F0 = F ⊗ A0k = (F ⊗ A0k )
X X 
= Aj ⊗ A0k
X
= Aj ⊗ A0k .
j,k

But it is easy to verify that Aj ⊗ A0k = Z(xj ⊗ x0k ) ∼

= Z, which proves the result.

54
(ii) This is obvious from the previous part since

rank F ⊗ F 0 = |J × K| = |J||K| = rank F rank F 0

Exercise 9.29. This is simply an application of Theorem 9.28 and Corollary 9.30, along with Exercise 9.27.
We end up with
A ⊗ B = Z/2Z ⊕ Z/3Z ⊕ Z/3Z ⊕ Z/5Z ⊕ Z/5Z ⊕ Z/5Z.
Exercise 9.30.
(i) Using coordinate-wise addition and scalar multiplication of the form
X X
p (qi , gi ) = (pqi , gi )

shows that Q ⊗ G is a Q-vector space. Hence dim Q ⊗ G is defined.

(ii) This follows immediately from the Tor exact sequence, along with the fact that Tor(Q, B) = 0 for all B.
Exercise 9.31. This is simply a calculation using the properties of Tor. We end up with

Z/3Z ⊕ Z/5Z.

Exercise 9.32. Using Exercise 9.30 with the short exact sequence

0 → F → G → G/F → 0

gives us
dim Q ⊗ G = dim Q ⊗ F + dim Q ⊗ G/F.
But dim Q ⊗ G/F = 0 by Exercise 9.21 and the fact that G/F is torsion. Moreover, we know that Q ⊗ F
has basis (1, xi ), where xi is a generator of F , so dim Q ⊗ F = rank F = rank G, which proves the result.

Exercise 9.33. Note that [Tor 1] and [Tor 5] imply that there is an exact sequence

0 → Tor(B 0 , A) → Tor(B, A) → Tor(B 00 , A) → B 0 ⊗ A → B ⊗ A → B 00 ⊗ A → 0,

since B ⊗ A ∼= A ⊗ B by Exercise 9.24. But if A is torsion-free, then Tor(B 00 , A) = 0 by [Tor 2], which gives
us the desired exact sequence.

Exercise 9.34. This is false! Consider, forPexample, when F = Z and H = Z/2Z, and a = 2, h = 1. In
general, we need the condition that if a = mj xj , where {xj } is a basis for F , then mj h 6= 0 for at least
some j. After all, we need that
a ⊗ h = (mj xj ⊗ h)j = (mj h) 6= 0.
Exercise 9.35. Let α be the map (cls z) ⊗ g →
7 cls(z ⊗ g). Then the Universal Coefficients Theorem implies
that
0 Hn (X) ⊗ G α Hn (X; G) Tor(Hn−1 (X), G) 0

is exact. Of course, since G is torsion-free, we know that Tor(Hn−1 (X), G) = 0. Hence α is an isomorphism.

Exercise 9.36. Use the second part of the Universal Coefficients Theorem. In particular, it gives us that

Hn (X; Z/mZ) ∼
= (Hn (X) ⊗ Z/mZ) ⊕ Hn−1 (X)[m],

since
Tor(Hn−1 (X), Z/mZ) = Hn−1 (X)[m]
by [Tor 4]. If Hn−1 (X) is torsion-free, the second term is zero, which gives the conclusion.

55
Eilenberg–Zilber Theorem and the Künneth Formula
Exercise 9.37. This is a straightforward calculation. In particular, we find that
(λ ⊗ µ)n−1 Dn (ci ⊗ ej ) = (λ ⊗ µ)n−1 (dci ⊗ ej + (−1)i ci ⊗ ∂ej )
= (λi−1 ⊗ µj )(dci ⊗ ej ) + (λi ⊗ µj−1 )((−1)i ci ⊗ ∂ej )
= λi−1 dci ⊗ µj ej + (−1)i λi ci ⊗ µj−1 ∂ej .
A similar calculation gives
Dn0 (λ ⊗ µ)n (ci ⊗ ej ) = Dn0 (λi ⊗ µj )(ci ⊗ ej )
= Dn0 (λi ci ⊗ µj ej )
= dλi ci ⊗ µj ej + (−1)i λi ci ⊗ ∂(µj ej ).
Of course, we know that dλ = λd and ∂µ = µ∂, which implies the result.
Exercise 9.38. Note that it suffices to prove the hint, since transitivity will finish the proof. The proof of
the hint is a routine, if long, computation.
Exercise 9.39. Suppose λ : C∗ → C∗0 and λ0 : C∗0 → C∗ with λ ◦ λ0 ' 1C∗0 and λ0 ◦ λ ' 1C∗ . Similarly define
µ and µ0 . Then Exercise 9.38 implies that
λ ⊗ µ : C∗ ⊗ E∗ → C∗0 ⊗ E∗0 ,
and similarly for λ0 ⊗ µ0 ). But
(λ ⊗ µ) ◦ (λ0 ⊗ µ0 ) = (λλ0 ) ⊗ (µµ0 ) ' 1C∗0 ⊗ 1E∗0 = 1C∗0 ⊗E∗0 .
The same calculation holds for the other composition, which proves chain equivalence.
Exercise 9.40. Each n (i.e., each 0 → Sn0 → Sn → Sn00 → 0) works because E∗ is a chain complex, hence
En is free.
Exercise 9.41. For n ≥ 1, we know that Hn (X) = 0 = Hn (Y ). Hence the Künneth formula implies that
Hn (X × Y ) ∼
X X
= Hi (X) ⊗ Hj (Y ) ⊕ Tor(Hp (X), Hq (Y )).
i+j=n p+q=n−1

But the first term is 0 since one of i, j is at least 1, and thus one of Hi (X), Hj (Y ) is 0. The second term
is zero since the only way for Hp (X) and Hq (Y ) to both be nonzero is if p = q = 0, in which case both
homology groups are free. Hence the torsion Tor(H0 (X), H0 (Y )) is zero in that case too.
Exercise 9.42. For path-connected X and Y , we have
H1 (X × Y ) = H0 (X) ⊗ H1 (Y ) ⊕ H1 (X) ⊗ H0 (Y ) ⊕ Tor(H0 (X), H0 (Y )).
But H0 (X) = H0 (Y ) = Z, and so using Exercise 9.27 gives us that the first two terms are H1 (Y ) and H1 (X),
respectively, while [Tor 2] implies that the last term is 0. This gives the first equation.
For H2 , notice that the Tor terms have either H0 (X) or H0 (Y ), so [Tor 2] implies that they are 0. Hence
H2 (X × Y ) = [H0 (X) ⊗ H2 (Y )] ⊕ [H1 (X) ⊗ H1 (Y )] ⊕ [H2 (X) ⊗ H0 (Y )].
Using H0 (X) = H0 (Y ) = Z again gives the result.
Exercise 9.43. This splits into multiple cases and is slightly annoying. We end up with the following:


 0 p≥n+2

⊕ p = n + 1, n odd



 Z Z/2Z

p = n + 1, n even



 Z/2Z

Z ⊕ Z/2Z p = n, n odd
Hp (K × RP n ) =
Z/2Z ⊕ Z/2Z
 p = n, n even

⊕ 1<p<n



 Z/2Z Z/2Z

Z/2Z ⊕ Z/2Z ⊕ Z/2Z p = 1, p 6= n





p=0

Z

56
Exercise 9.44. We once again have many, many cases.



 Z p = 0, m 6= 0
⊕ p=m=0



 Z Z

p odd, p < min(m, n)



 Z/2Z

Z ⊕ Z/2Z m odd, p = m




n m
Z/2Z m odd, p odd between m and n
Hp (RP × S ) =


 Z m odd, p even, p ≤ m + n
m even, p = m



 Z

Z ⊕ Z/2Z m even, p odd between m and n




m even, p odd, p ≤ m + n

Z



0 otherwise

(Something like that, I can’t quite read my work anymore.)

Exercise 9.45. This is the exact same idea, but I’ll admit I didn’t work it all out.
Exercise 9.46. It turns out that the machinery we have (i.e., fundamental groups) isn’t sufficient to dis-
tinguish S 1 ∨ S 2 ∨ S 3 from S 1 × S 2 , as they both have fundamental group Z. In fact, this seems to require
cohomology (see ??).
Exercise 9.47. We use the Künneth formula here. Note that the homology groups of S 1 are all cyclic or
zero, so the Tor terms are zero. Hence
X
Hn (S 1 × S 1 ) = Hi (S 1 ) ⊗ Hj (S 1 ).
i+j=n

If n > 2, then one of Hi (S 1 ) and Hj (S 1 ) is zero, so

Hn (S 1 × S 1 ) = 0 n > 2.

When n = 0, then we have i = 0, j = 0, so

H0 (S 1 × S 1 ) = Z ⊗ Z = Z

If n = 1, then we have (i, j) = (0, 1), (1, 0), and so

H1 (S 1 × S 1 ) = Z ⊗ Z ⊕ Z ⊗ Z = Z ⊕ Z.

Finally, if n = 2, then we only have to consider (i, j) = (1, 1), so that

H2 (S 1 × S 1 ) = Z ⊗ Z = Z.

Now recall that

Hn (K1 ∨ K2 ) ∼
= Hn (K1 ) ⊕ Hn (K2 ),
so we know that
Hn (S 2 ∨ S 1 ∨ S 1 ) ∼
= Hn (S 2 ) ⊕ Hn (S 1 ) ⊕ Hn (S 1 ).
We know the homology groups of S 1 and S 2 , and working them out gives the same homology groups as
those of S 1 × S 1 .
(Interestingly, the fundamental groups of these two spaces are different from one another, which one can
show using Seifert–Van Kampen. I wonder if Rotman mixed up this problem with Exercise 9.46.)
Exercise 9.48.
(i) This is straightforward using the fact that the homology of wedges is the direct sum of homology groups.
Hence both homology groups are Z when n = 0, 3, Z/2Z when n = 1, and 0 otherwise.

57
(ii) According to a cursory search online, this requires universal coverings.
(iii) This seems to be another mistake on Rotman’s part, as he seems to have thought that RP 3 and RP 2 ∨ S 3
had different fundamental groups. In fact, they both have Z/2Z as their fundamental group, and so it is
obvious that RP 3 × RP 2 and (RP 2 ∨ S 3 ) × RP 2 have the same homology groups and fundamental group.

Exercise 9.49. Since the homology groups of S 1 are all cyclic or zero, the Tor terms in the Künneth formula
don’t count. Suppose that
r−1
Hn (T r−1 ) = Z( n ) .
Note that this is true for r = 1. Then we have that

Hn (S 1 × T r−1 ) ∼
X
= Hi (S 1 ) ⊗ Hj (T r−1 ) = Hn (T r−1 ) ⊕ Hn−1 (T r−1 ).
i+j=n

But of course we know that

r−1 r−1
     
r
+ = ,
n n−1 n
and so it follows that r
Hn (T r ) = Hn (S 1 × T r−1 ) = Z(n) .

58
10 Covering Spaces
Basic Properties
Exercise 10.1. Obviously R is path-connected and exp is continuous. Furthermore, for each exp(2πit) ∈ S 1 ,
consider the neighborhood S 1 \ {exp(π + 2πit)}. Of course, we know that
[ 1 1

exp−1 (U ) = n + t − ,n + t + ,
2 2
n∈Z

so U is evenly covered.
Exercise 10.2. To see that pk : z 7→ z k is continuous, pick some open U ⊆ S 1 . Pick exp(2πit) ∈ p−1
k (U ),
so that exp(2πikt) ∈ U . Then there is some ε > 0 so that
{exp(2πikx) : t − ε < x < t + ε} ⊆ U.
Then it follows that the open set {exp(2πix) : t − ε < x < t + ε} is contained in p−1 −1
k (U ), so that pk (U ) is
open. Hence pk is continuous.
To see that (S 1 , pk ) is a covering space, let e exp(2πit) ∈ S 1 . Pick the open neighborhood
1
U = {exp(2πix) : t − 2 < x < t + 21 }.
Note then that
1 1
[ t−
 
+n t+ +n
p−1
k (U ) = exp 2πix) : 2
<x< 2
,
k k
n∈Z
proving that U is evenly covered.
Exercise 10.3. Informally: Note that a point of RP n corresponds to a pair of antipodal points in S n . Given
some point in S n , there is always a small open neighborhood which does not intersect its reflection (which
is a neighborhood of the antipodal point). This neighborhood is evenly covered.
Exercise 10.4.
(i) Consider any (basic) open neighborhood of x0 . The preimage of this neighborhood under q looks like
two disjoint intervals on S 1 . The only possibility is if q restricted to a homeomorphism on each of these
intervals. But this isn’t the case (surjectivity fails), so no neighborhood of x0 is evenly covered.
(ii) A non-tangency point obviously has an evenly covered neighborhood, while x0 has an evenly covered
neighborhood whose sheets correspond to small neighborhoods of the infinitely many tangency points of
X.
e

Exercise 10.5. Each element of p−1 (x0 ) belongs to a different sheet, since p is a homeomorphism on each
sheet. Thus p−1 (x0 ) is discrete.
Exercise 10.6. Since a covering projection is a local homeomorphism, it follows that local topological
properties are all inherited by picking a suitably small neighborhood of any given point.
Exercise 10.7. If p−1 (U ) = Si and Si0 ⊆ Si is p−1 (V ) ∩ Si , then note that p−1 (V ) = Si0 . Furthermore,
S S
we know that p : Si0 → V is a homeomorphism since p : Si → U is a homeomorphism and p(Si0 ) = V . Hence
V is evenly covered.
Exercise 10.8.
(i) Pick (x1 , x2 ) ∈ (X1 , X2 ). Suppose neighborhoods Ui of xi are pi -admissible for i = 1, 2. In particular,
write [ [
p−1
1 (U1 ) = Si , p−1
2 (U2 ) = Tj .
Then it is easy to check that [
(p1 × p2 )−1 (U1 × U2 ) = Si × Tj .
Note that R is a covering space of S 1 (Exercise 10.1), and so it follows that R×R covers the torus S 1 ×S 1 .

59
(ii) Either using Exercise 10.2 or by noting that (X, 1X ) covers X for any path-connected space X, it is easy
to see that S 1 is a covering space of S 1 . Hence the conclusion follows from the previous part.
Exercise 10.9. Note that q = α−1 pβ is continuous. Since Ye and X e are homeomorphic, we know that Ye is
path-connected.
Now letSy ∈ Y correspond to x ∈ X, i.e., have α(y) = x. Note that x has a neighborhood Ux such that
p−1 (Ux ) = Si for sheets Si . Write β −1 (Si ) = Ti , where Ti and Si are homeomorphic. Observe that
q(Ti ) = α−1 pβ(Ti ) = α−1 p(Si ) = α−1 (Ux ).
Moreover, we know that q|Ti is a homeomorphism, because p is a homeomorphism on Si , making q = α−1 pβ
a composition of homeomorphisms. Hence α−1 (Ux ) is a q-admissible neighborhood of y.
Exercise 10.10. In this case, we have p∗ π1 (X,
e x̃0 ) trivial, so that

e x̃0 )] = |π1 (X, x0 )|.

m = [π1 (X, x0 ) : p∗ π1 (X,
If m is prime, then the only group of order m is Z/mZ.
Exercise 10.11. This is the same logic as Exercise 10.10.
Exercise 10.12. Pick u ∈ U . Then p−1 (U ) has cardinality m. But no two elements of p−1 (U ) can be in
the same sheet, else p|Si would not be injective, while there is at least one element of p−1 (U ) in each Si , else
p|Si would not be surjective onto U . Since no element can be in multiple sheets, as the sheets are disjoint, it
follows that the elements of p−1 (U ) can be put into 1-1 correspondence with the sheets Si . Hence |I| = m.
Exercise 10.13.
(i) Note that [f ] ∈ ker θ iff x̃[f ] = x̃ for all x̃ ∈ p−1 (x0 ). By definition, we know that x̃[f ] = f˜(1). Hence the
following are all equivalent:
• [f ] ∈ ker θ
• f˜(1) = x̃[f ] = x̃ for all x̃ ∈ p−1 (x0 )
• f˜ is a closed loop at x̃ for all x̃
• p[f˜] ∈ p∗ π1 (X,
e x̃)
•
T
[f ] ∈ x̃∈Y p∗ π1 (X,e x̃)
The equivalence of the first and last conditions is what we originally wanted.
(ii) Note that Xe being simply connected implies that T p∗ π1 (X,
e x̃) is trivial. Hence ker θ is trivial, so θ is an
injection.
Exercise 10.14. We know that G is a covering space for G/H with p : g 7→ gH. Let θ : π(G/H, 1) → SH be
as defined in Exercise 10.13. Since G is simply connected, we know that θ is injective. Hence π1 (G/H, 1) ∼
=
im θ. Note that θ takes x̃ 7→ f˜(1), where f˜(0) = x̃. But these path lifts in im θ correspond precisely to
elements of H, since f˜(1) ∈ ker p = H.
Exercise 10.15. Every subgroup of an abelian group is normal. Hence p∗ π1 (X,
e x̃0 ) is a normal subgroup
of π1 (X, x0 ). This is the definition of a regular covering space.
Exercise 10.16. Pick x ∈ X. Let ỹ ∈ q −1 (x) and x̃ = h(ỹ). We have the following diagram:
h
(Ye , ỹ) (X,
e x̃)

q p

(X, x)

Pick some path f˜X : I → X e with f˜X (0) = x̃. Define f = pf˜X . There is a unique f˜Y lifting f to (Ye , ỹ),
i.e., with fY (0) = ỹ. But now notice that hf˜Y is a path lifting f into X
˜ e such that (hf˜Y )(1) = h(ỹ) = x̃.
˜
Uniqueness implies that hfY = fX .˜
Now im h contains im f˜x , which contains x. Thus x ∈ im h. Since x was arbitrary, this proves that h is
surjective.

60
Exercise 10.17. Let x̃0 ∈ p−1 (x0 ). Now consider the following statements, all of which are equivalent:

• [f ] ∈ p∗ π1 (X,
e x̃0 )
• [f ] stabilizes x̃0
• x̃0 [f ] = x̃0
• f˜(1) = x̃0 where f˜ is the lifting with 0 7→ x̃0
• f˜ is closed at x̃0

Covering Transformations
Exercise 10.18. The isomorphism is the composition of two isomorphisms. The first is Cov(X/X) e →
Aut(p (x0 )) which takes h to h|p (x0 ). The second is Aut(p (x0 )) → π1 (X, x0 ) which takes ϕ to [f −1 ]
−1 −1 −1

where f is the well-defined path so that ϕ(x̃0 ) = x̃0 [f ] for x̃0 in the fiber over x0 .
Hence the isomorphism Cov(X̃/X) takes h to the map [f −1 ] defined by h(x̃0 ) = x̃0 [f ] for x̃0 ∈ p−1 (x0 ).
Exercise 10.19. No. Any neighborhood of p ∈ S n is necessarily going to include some q p̃, thus making an
even covering of any neighborhood impossible. To see why any neighborhood of p intersects {q : q ∼ p} at a
point that isn’t p, simply note that the equivalence class of p is connected.
Exercise 10.20. This is exactly the same argument as Example 10.2.
Exercise 10.21. Suppose that, for each closed path f : I → X, either every lifting f˜ of f is closed, or
no lifting f˜ is closed. Exercise 10.17 implies that p∗ π1 (X, e x̃1 ) for all x̃0 , x̃1 ∈ p−1 (x0 ). Now
e x̃0 ) = p∗ π1 (X,
−1 −1 e x̃1 ) for some x̃1 ∈ p−1 (x0 ).
Corollary 10.12 says that, if x̃0 ∈ p (x0 ), then gp∗ π1 (X, x̃0 )g e = p∗ π1 (X,
Hence the conjugate of p∗ π1 (X, e x̃0 ) is itself, making it a normal subgroup. This is true for every x0 , so
(X, p) is regular.
e
Now suppose (X, e p) is regular. Then p∗ π1 (X, e x̃0 ) is normal for each x̃0 . Corollary 10.12(i) implies that
p∗ π1 (X,
e x̃0 ) and p∗ π1 (X,
e x̃1 ) are conjugate for all x̃0 , x̃1 in the fiber over x0 . Hence they are equal. Now use
Exercise 10.17:
e x̃0 ) = {[f ] : f˜ closed at x̃0 }.
p∗ π1 (X,
Hence if the lifting f˜ to x̃0 is closed, then so too is the lifting to x̃1 .
T
Exercise 10.22. The monodromy group is π1 (X, x0 )/ ker θ where ker θ = x̃∈p−1 (x0 ) p∗ π1 (X, e x̃). But the
−1
p∗ ’s are all equal by Corollary 10.12(i). Hence ker θ = p∗ π1 (X, x̃0 ) for some x̃0 ∈ p (x0 ). Corollary 10.28
e
implies the result.
Exercise 10.23. If X is an H-space, then π1 is abelian. Hence every subgroup is normal, so every covering
space X is regular.

Existence
Exercise 10.24. It suffices to show that [f¯] = [f −1 ]. After all, if this is true, then f¯ ∗ f is nullhomotopic,
hence [f¯ ∗ f ] ∈ G, hence hf iG = hf −1 iG . But note that e ' f˜ ◦ f ' f˜ ∗ f rel İ. Hence [f¯] = [f −1 ] as desired.
Exercise 10.25. A similar argument, on multiplication only, holds.
Exercise 10.26. Let U be an open cover of X.
e For x ∈ X, consider an admissible neighborhood Vx of x:
j
[
p−1 (Vx ) = Si .
i=1

Let Wx be admissible with Wx ⊆ Vx . Then we can write

j
[
−1
p (Wx ) = Ti ,
i=1

61
 where Ti ⊆ Si . Note that Ti ≈ Wx and Si ≈ Vx for each i = 1, . . . , j. Hence Ti is compact, since it is
homeomorphic to a closed subset of the compact set Wx .
Thus there are, for each i = 1, . . . , j, finitely many sets of U which together cover Ti . Take all of them to
obtain a (finite) cover of [ [
Ti ⊇ Ti = p−1 (Wx ).
Call this finite cover to be Wx .
Note that the family of all Wx ’s covers X. Since X is compact, it follows that finitely many Wx ’s cover
X, say Wx1 , . . . , Wxn . Then
[n
Wxi = X,e
i=1

which gives a finite subcover of U.

Exercise 10.27. The j-sheeted covering spaces is exactly the number of X eG where X eG is j-sheeted. By
Theorem 10.9(iii), this is exactly the number rof G with [π1 (X, x0 ) : p∗ π1 (XG , x̃0 )] = j. Of course, this
e
p∗ -group is exactly G, and so this is exactly the number of subgroups having index j.
Exercise 10.28. The result follows as long as any finite CW complex has a finitely generated π1 . But this
can be seen to be true by Van Kampen.

Orbit Spaces
Exercise 10.29. We know by Theorem 10.54 that Cov(X/( e X/H))
e = H, and so we can think of G as a
subgroup of Cov(X/(
e X/H)).
e Now use Theorem 10.52, with X = X/H. e We know, in particular, that G is
a subgroup of Cov(X/X), and thus is exactly a covering space (X/G, v) of X = X/H,
e e e as desired.

Exercise 10.30.
(i) Suppose gx = x and consider a proper neighborhood V of x. Then we know that gV ∩ V = ∅, but
x = gx ∈ gV ∩ V , contradiction.
(ii) If G = {e, g1 , . . . , gn } and x ∈ X, then, since X is Hausdorff and since gi x 6= x, there exists a neighborhood
V of X which does not contain any gi x. Obviously, this V is a proper neighborhood.

Exercise 10.31. This is exactly the argument in the proof of Theorem 10.2, namely in the first full paragraph
on p. 276.
Exercise 10.32.
(i) The group Z/pZ acts on S 3 via m • (z0 , z1 ) = (ζ m z0 , ζ mq z1 ). This action is proper because part (ii) of
Exercise 10.30 obviously applies.
(ii) Note that S 3 /(Z/pZ) is exactly L(p, q). Thanks to the previous part, Theorem 10.54(ii) applies, which
implies that
π1 (L(p, q)) = π1 (S 3 /(Z/pZ)) = Z/pZ.
(iii) We know that L(p, q) inherits the local properties of S 3 , since there is a local homeomorphism between
them. Thus L(p, q) is a 3-manifold.
If U is an open cover of L(p, q), then p−1 (U) is an open cover of S 3 . Hence finitely many elements
of p−1 (U), say p−1 (Ui ) for i = 1, . . . , n, cover S 3 . But then {U1 , . . . , Un } is a finite subcover of U which
covers L(p, q), proving compactness.
Finally, note that A ⊆ L(p, q) clopen implies that p−1 (A) is clopen in S 3 . Hence p−1 (A) = ∅, S 3 , and
so A = ∅, L(p, q). Thus L(p, q) is connected too.
Exercise 10.33. Notice that T → T /G is a universal covering space since T is simply connected. Moreover,
since T /G is a connected 1-complex, we know by Corollary 7.35 that π1 (T /G) is free. But Theorem 10.54(iii)
implies that π1 (T /G) ∼
= G, and so G is free.

62
11 Homotopy Groups
Function Spaces
No exercises!

Group Objects and Cogroup Objects

Exercise 11.1.

(i) By definition of a product, there is a unique morphism θ : (X, q1 , q2 ) → (C1 × C2 , p1 , p2 ) in C making

the diagram commute, namely θ = (q1 , q2 ).
(ii) The objects are ordered triples (X, k1 , k2 ) where X is a set and ki : Ci → X are functions. Morphisms
θ : (X, k1 , k2 ) → (Y, `1 , `2 ) are functions θ : X → Y making the following commute:

X
k1 k2

C1 θ C1

`1 `2
Y

Exercise 11.2. We first tackle Ab.

The map θ : X → G1 ⊕ G2 in the product diagram is given by θ(g) = (q1 g, q2 g). Commutativity follows
from the fact that pi (θ(g)) = qi (g). Uniqueness of θ follows from the fact that any other θ0 must satisfy
θ0 (g) = (g1 , g2 ) where gi = pi (g1 , g2 ) = qi (g). Hence θ0 = θ.
The map η : G1 ⊕ G2 → X in the coproduct diagram is given by (g, h) 7→ k1 (g) + k2 (h), where + denotes
the operation in the abelian group X. We can easily check commutativity and uniqueness using the fact
that η must be a group homomorphism.
Now, for Grp, note that the free product property on p. 173 is exactly the coproduct property. The
same argument as in the abelian case shows that direct product is the product in Grp.
Exercise 11.3.

(i) We will show this for Top∗ . Suppose we have ((X, x), k1 , k2 ). It is obvious that the map θ : (A1 ∨A2 , ∗) →
(X, x), if it exists, must take ∗ to x, and ∗ =
6 ai ∈ ji (Ai ) to ki (ai ). We need only show that this map θ is
continuous. (In contrast, the proof has already been completed for Set∗ ; commutativity of the relevant
diagram is obvious from the definition of θ.)
Suppose U ⊆ X is closed. Note that θ−1 (U ) ∩ Ai = ki−1 (U ). (This statement is clear if ∗ 6∈ U . If
∗ ∈ U , then

θ−1 (U ) ∩ Ai = (θ−1 (U \ {x}) ∩ Ai ) ∪ {∗} = ki−1 (U \ {x}) ∪ {ai } = ki−1 (U ),

which proves the statement anyway.) The definition of the topology of the wedge (see Example 8.9)
implies that θ−1 (U ) is closed. Hence θ is continuous, completing the proof.
(ii) Call this subset S. The map f : A1 ∨ A2 → S which takes a ∈ Ai (or, more accurately, a ∈ ji (Ai )) to
(a, a2 ) if i = 1 and to (a1 , a) if i = 2 is continuous by the previous argument. It is clearly bijective and
closed, since a closed set F in A1 ∨ A2 is still closed in A1 × A2 . Thus it is a homeomorphism.
Exercise 11.4. Commutativity follows from the interchanging of C1 and C2 in the definition. To see

63
associativity, consider the following diagram:

(C1 × C2 ) × C3

C1 × C2

C1 C2 C3

There is a unique map X → (C1 × C2 ) × C3 making this diagram commute.

Now define p1 : (C1 × C2 ) × C3 → C1 to be the composition of the red arrows below. Furthermore, the
product property of C2 × C3 implies the existence of the following blue and green arrows:

(C1 × C2 ) × C3

C1 × C2 C2 × C3

C1 C2 C3

Let p2 be the blue arrow. The fact that there is still the same unique map X → (C1 × C2 ) × C2 making this
commute, then, implies that (C1 × C2 ) × C3 is the product of C1 and C2 × C3 , thus proving associativity.
Exercise 11.5.

(i) We would like to find f1 × f2 making the following commute:

f2
C2 D2

f1 ×f2
C1 × C2 D1 × D2

f1
C1 D1

But the existence of maps C1 × C2 → Ci → Di implies, by the product property of D1 × D2 , a unique

map f1 × f2 into D1 × D2 making the diagram commute.
(ii) Same idea.
Exercise 11.6.
(i) Note that ∆X is the unique map making the red part of the diagram commute, while q1 × q2 is the unique

64
 map making the blue part commute:

D1 × D2

D1 q1 ×q2 D2

X ×X

X ∆X X

But of course, since the maps qi ◦ 1X = X → X → Di are equal to simply the maps qi : X → Di , we
know that the unique map X 7→ D1 × D2 making this entire diagram commute is (q1 , q2 ). Uniqueness
implies that (q1 , q2 ) must be equal to (q1 × q2 )∆X .
(ii) This is the same idea.
(iii) We already showed the first statement. For the second, notice that ∇B (f ×g) = (f, g). But (f, g)∆A (a) =
(f (a), g(a)) = (f + g)(a) because A ⊕ B = A q B.
Exercise 11.7.
(i) Everything follows from the hint, except that we must verify that 1X×Z and θλ complete the given
diagram. Commutativity of the left triangle is obvious in both cases. To see that q1X×Z = t, note that
Z being terminal implies that q = t. To show that qθλ = t, note that qθλ : X × Z → X → X × Z → Z.
Thus Z being terminal again implies the result.
Now θ and λ are inverses, and so X × Z and X are equivalent.
(ii) This is the dualized version of the previous part.

65