Lecture Notes On Elementary Topology and Geometry-I. M. Singer J. A. Thorpe-Springer

Undergraduate Texts in Mathematics
I. M. Singer
J. A. Thorpe
Lecture Notes on
Elementary Topology
and Geometry
SPACE RIEMANNIAN GROUP THEORY COMPLEX ELEMENTARY
GEOMETRY (group of isometries) VARIABLES GEOMETRY
R2 K=0 Euclidean motions Complex plane Euclidean

in R2 (rotations geometry
+ translations)
X,
S2 " K>0 Rotation group in Riemann sphere Spherical
• R3 geometry
(elliptic
geometry
on P2)
D K <0 Conformal group Disc (hyperbolic Hyperbolic

plane, or upper geometry
half-plane)
Springer-Verlag New York • Heidelberg • Berlin

NUNC COGNOSCO EX PARTE
THOMAS J. BATA LIBRARY

TRENT UNIVERSITY
I
Undergraduate Texts in Mathematics
Editors
F. W. Gehring
P. R. Halmos
Advisory Board
C. DePrima
I. Herstein
J. Kiefer
W. LeVeque
Digitized by the Internet Archive
in 2019 with funding from
Kahle/Austin Foundation
https://archive.org/details/trent_0116301276808
I. M. Singer
J. A. Thorpe
Lecture Notes on
Elementary Topology
and Geometry
Springer-Verlag
New York Heidelberg Berlin
QR Coj
I. M. Singer J. A. Thorpe
Department of Mathematics Department of Mathematics
Massachusetts Institute of Technology SUNY at Stony Brook
Cambridge, Massachusetts 02139 Stony Brook, New York 11790
Editorial Board
F. W. Gehring P. R. Halmos
Department of Mathematics Department of Mathematics
University of Michigan University of California
Ann Arbor, Michigan 48104 Santa Barbara, California 93106
AMS Subject Classifications: 50-01, 53-01, 54-01
Library of Congress Cataloging in Publication Data
Singer, Isadore Manuel, 1924-

Lecture notes on elementary topology and geometry.
(Undergraduate texts in mathematics)
Reprint of the ed. published by Scott, Foresman,
Glenview, Ill.
Bibliography: p. 230
Includes index.
1. Topology. 2. Algebraic topology.
3. Geometry, Differential. I. Thorpe, John A., joint author. II. Title.
[QA611.S498 1976] 514 76-26137
All rights reserved.
No part of this book may be translated or reproduced in any form without written
permission from Springer-Verlag.
© 1967 by I. M. Singer and John A. Thorpe.
Printed in the United States of America.
ISBN 0-387-90202-3 Springer-Verlag New York
ISBN 3-540-90202-3 Springer-Verlag Berlin Heidelberg

Preface
At the present time, the average undergraduate mathematics major finds

mathematics heavily compartmentalized. After the calculus, he takes a course
in analysis and a course in algebra. Depending upon his interests (or those of
his department), he takes courses in special topics. If he is exposed to topology,
it is usually straightforward point set topology; if he is exposed to geom¬
etry, it is usually classical differential geometry. The exciting revelations that
there is some unity in mathematics, that fields overlap, that techniques of one
field have applications in another, are denied the undergraduate. He must
wait until he is well into graduate work to see interconnections, presumably
because earlier he doesn’t know enough.
These notes are an attempt to break up this compartmentalization, at least
in topology-geometry. What the student has learned in algebra and advanced
calculus are used to prove some fairly deep results relating geometry, topol¬
ogy, and group theory. (De Rham’s theorem, the Gauss-Bonnet theorem for
surfaces, the functorial relation of fundamental group to covering space, and
surfaces of constant curvature as homogeneous spaces are the most note¬
worthy examples.)
In the first two chapters the bare essentials of elementary point set topology
are set forth with some hint of the subject’s application to functional analysis.
Chapters 3 and 4 treat fundamental groups, covering spaces, and simplicial
complexes. For this approach the authors are indebted to E. Spanier. After
some preliminaries in Chapter 5 concerning the theory of manifolds, the De
Rham theorem (Chapter 6) is proven as in H. Whitney’s Geometric Integration
Theory. In the two final chapters on Riemannian geometry, the authors
follow E. Cartan and S. S. Chern. (In order to avoid Lie group theory in the
last two chapters, only oriented 2-dimensional manifolds are treated.)
v
Preface
These notes have been used at M.I.T. for a one-year course in topology and
geometry, with prerequisites of at least one semester of modern algebra and
one semester of advanced calculus “done right.” The class consisted of about
seventy students, mostly seniors. The ideas for such a course originated in one
of the author’s tour of duty for the Committee on the Undergraduate Pro¬
gram in Mathematics of the Mathematical Association of America. A
program along these lines, but more ambitious, can be found in the CUPM
pamphlet “Pregraduate Preparation of Research Mathematicians” (1963).
(See Outline III on surface theory, pp. 68-70.) The authors believe, however,
that in lecturing to a large class without a textbook, the material in these notes
was about as much as could be covered in a year.
vi
Contents
Chapter 1
Some point set topology 1

1.1 Naive set theory 1
1.2 Topological spaces 5
1.3 Connected and compact spaces 11
1.4 Continuous functions 13
1.5 Product spaces 16
1.6 The Tychonoff theorem 20
Chapter 2
More point set topology 26

2.1 Separation axioms 26
2.2 Separation by continuous functions 31
2.3 More separability 34
2.4 Complete metric spaces 40
2.5 Applications 43
Chapter 3
Fundamental group and covering spaces 49

3.1 Homotopy 49
3.2 Fundamental group 52
3.3 Covering spaces 62
vii
Contents
Chapter 4
Simplicial complexes 78
4.1 Geometry of simplicial complexes 79
4.2 Barycentric subdivisions 83
4.3 Simplicial approximation theorem 90
4.4 Fundamental group of a simplicial complex 94
Chapter 5
Manifolds 109
5.1 Differentiable manifolds 109
5.2 Differential forms 118
5.3 Miscellaneous facts 132
Chapter 6
Homology theory and the De Rham theory 153

6.1 Simplicial homology 153
6.2 De Rham’s theorem 161
Chapter 7
Intrinsic Riemannian geometry of surfaces 175

7.1 Parallel translation and connections 175
7.2 Structural equations and curvature 184
7.3 Interpretation of curvature 190
7.4 Geodesic coordinate systems 198
7.5 Isometries and spaces of constant curvature 207
Chapter 8
Imbedded manifolds in R3 216
Bibliography 230
Index 231
viii
Some point set topology
1.1 Naive set theory

We shall accept as primitive (undefined) the concepts of a set (collection,
family) of objects and the concept of an object belonging to a set.
We merely remark that, given a set S and an object x, one can determine if
the object belongs to (is an element of) the set, written x e S, or if it does not
belong to the set, written x <£ S.
Definition. Let A and B be sets. A is a subset of B, written A B, if x e A

implies x e B. A is equal to B, written A — B, if A c B and B <= A.
Notation. The empty set, that is, the set with no objects in it, is denoted
by 0.
Remark
(1) 0 <= A for all sets A.
(2) The empty set 0 is unique; that is, any two empty sets are equal. For if
0X and 02 are two empty sets, 0X <= 02 and 02c 0X.
(3) A <= A for all sets A.
Definition. Let A and B be sets. The union A u B of A and B is the set of all x
such that x e A or x e B, written
A u B = [x; x e A or x e B],
The intersection A n B of A and B is defined by
A n B = [x; x e A and x e B],
1
1: Some point set topology
Similarly, if LA is a set (collection) of sets, the union and intersection of all

the sets in Zf are defined respectively by
S — [x; x e S for some S e SA],

SeS?
[^| S = [x; x e S for every S e

SeS?
If A ^ B, the complement of A in B, denoted A' or B — A, is defined by
A' = [x e B\ x $ A],
Theorem 1. Let A, B, C, and S be sets. Then
(1) A u B = B u A.
(2) A n B = B n A.
(3) (A u B) u C = A u (B u C).
(4) (A n B) n C = A n (B n C).
(5) Au(BnC) = (A uB)n (A u C).
(6) A n (Bu C) = (A n B)u (A n C).
(7) IfA^SandB^S then, (A u B)' = A' n B’.
(8) If A ^ S and B ez S, then (A n 5)' = A' U F'.
(9) If ffx and TA2 are two sets (collections) of sets, then
(U s)u (U s) = U s
'SeSC j ’ \se£C 2 SeSC1>jSe2
and
(n s) n (n s) - n
\SeSC1 / \SeSC2 / SeSf’1uSe2
(10) For <5^ as in (9),
(u ^W u $») = u osin so.

\ \ S2eS?2 '
S2eS^ 2
Proof. The proof of this theorem is left to the student.
Definition. Let A and B be sets. The Cartesian product A x B of A and B is the

set of ordered pairs
A x B = [(a, b); a e A, b e B],
A relation between A and B is a subset R of A x B. a and b are said to be

F-related if (a, b) e R.
2
1.1: Naive set topology
Example. Let A = B = the set of real numbers. Then A x B is the plane.

The order relation x < y is a relation between A and B. This relation is the
shaded set of points in Figure 1.1.
Definition. A relation R <= A x A is a partial ordering if
(1) (si, s2) 6 R and (.y2, s3) e R ^ (su s3) e R and

(2) Oi, s2) e R and (s2, sj e R =>■ s± = s2.
A relation R is a simple ordering if it is a partial ordering, and, in addition:
(3) either (5l9 s2) e R or (sa, î) e R for every pair s1} s2 e S.
The order relation for S = real numbers is an example of a simple ordering.

In general, we say that S is partially ordered (simply ordered) by R.
Definition. Let A and B be sets. A function f mapping A to B, denoted

/: A —> B, is a relation (/ A x B) between A and B satisfying the follow¬
ing properties.
(1) If ae A, then there exists b e B such that (a, b) e/.

(2) If (a, b)ef and (a, bj e f then b = bx.
Property (1) says that the function/is defined everywhere on A. Property

(2) says that /is a “single-valued” function.
Notation. hQtf:A->B. By /(a) = b we mean (a, b) e f
3
Definition. Let f\ AB. f is surjective (onto) if for each b e B there exists

a e A such that /(a) = b. If/is surjective, we write f{A) — B./is injective
(one-to-one) if /(a) = f(aj) => a = ax. If/is both surjective and injective,
we say / is a one-to-one correspondence between A and B.
Definition. A set A is countable if there exists a one-to-one correspondence

between the set of all integers and A. A set A is finite if for some positive
integer n there exists a one-to-one correspondence between the set
and A, in which case we say A has n elements.
Theorem 2. If A = {ax,..., an} is a finite set of n elements, then the set of all
subsets of A has 2n elements.
Proof. Consider the set F of all functions mapping A to the set {YES, NO}
consisting of the two elements YES and NO. F has 2n elements. The set FF of
all subsets of A is in one-to-one correspondence with F. For let/: FF -> Fbe
defined as follows.
For BeFF, that is, for B <= A,f(B) is that element of F (that is.
f(B): A -> [YES, NO])
given by
JYES if xeB,
(NO ifx$B.
f is injective because if f(B) = /(C), then f(B)(x) = f(C)(x) for all x e A.

Thus f(B)(x) = YES if and only iff(C)(x) = YES; that is, x e B if and only
if x e C. Thus B = C.f is surjective because every function g: A —> [YES, NO]
determines a B A by
B - [x; g(x) = YES]
and f(B) = g. □
Notation. Motivated by this proposition, we denote by 2A the set of all

subsets of A. Given two sets A and B, BA denotes the set of all functions
A-> B.
Definition. Let/e BA. The inverse f~x of/is the function 2s -> 2A defined by
= [a e A ;f(a) e Bx] (Bx <= B).
f~1{Bj) is called the inverse image of Bx. Note that
Z"1 e(2")2B.
Notation. Let W be a set, and let FF be a collection of sets. We say FF is
4
1.2: Topological spaces
indexed by W if there is given a surjective function <j>\ IVST. For w e IV,

we denote <p(w’) by Sw and denote the indexing of ST by W as {Sw}weW.
Definition. Let be indexed by W. The product of the sets {Sw}weW is

the set
ns.- Sw;f(w) g Sw for all we w\.

weW L weW
If the set W is not finite, this product is called an infinite product. Note that
this notion of the product of sets extends the notion of the product of two
sets Sx x St. For let W = {1, 2}, let = {Slt S2}, and let <p: W S? by
<p(J) = Sj, j = 1,2. Then Sx x S2 = [(j1} s2), s, e S,], and HLew Sw =
[/: {1, 21^-SjU S2',f(j) g Sj], which can be identified with [(/(l),/(2));
/O') g Sj], which can be identified with Sx x S2.
Remark, riioew Sw is a set of functions. One might ask whether there exist
any such functions; that is, is HOew Sw # 0 ? In other words, given infinitely
many nonempty sets, is it possible to make a choice of one element from each
set? It can be shown in axiomatic set theory that this question cannot be
answered by appealing to the usual axioms of set theory. We accept the
affirmative answer here as an axiom.
Axiom of choice. Let {S'U)}lt,eW be sets indexed by W. Assume Sw =£ 0 for all

we W. Then
J Sw 7^ 0.
weW
The axiom of choice is equivalent to several other axioms, one of which is

the following.
Maximum principle. If S is partially ordered by R, and T is a simply ordered

subset, then there exists a set M such that the following statements are
valid.
(1) M c S.
(2) M is simply ordered by R.
(3) If M c N <= S, and N is simply ordered by R, then M = N; that is,
M is a maximal simply ordered subset containing T.
1.2 Topological spaces

Definition. A metric space is a set S together with a function p: S x S —> the
nonnegative real numbers, such that for each slt s2, s3e S:
(1) p(s±, s2) = 0 if and only if sx = s2.
(2) P(sx, s2) = p(s2, sx).
(3) p(slt s3) < p(slt s2) + p(s2, s3).
The function p is called a metric on S.
5
Given a point 50 in a metric space S and a real number a, the ball of radius
a about s0 is defined to be the set
BSo(a) = [seS; P(s, s0) < a\.
Example. Let S be the plane, that is, the product of the set of the real num¬
bers with itself. We define three metrics on S as follows.
For P1 = (*!, yx) and P2 = (x2, y2) two points in S,
Pi(Pi,P2) = V(x2 - xx)2 + (y2 - y-d2,

P2i.P1, = max{|x2 - Xi|, |y2 - Fi|},
P3iPi,Pi) = \Xi ~ w| + \y2 ~ yi\-
The ball of radius a about the point 0 = (0, 0) relative to each of these metrics
is indicated by the shaded areas in Figure 1.2. Note that a ball does not
necessarily have a circular, or even a smooth, boundary.
| §§|!
.
Pi(s,0) < a p2(s,0) < a P3(s,0) < a
Figure 1.2
Remark. The three metrics defined above provide the plane with three
distinct structures as a metric space. Yet for studying certain properties of
these spaces, these metrics are equivalent. Thus, if we want to know, for
example, whether 0 is a limit point of a set T <= S, we ask whether there is a
sequence of points in T which converges to 0; that is, whether a sequence {sn}
of points in T exists such that given any e > 0, there exists an N such that
p(sn, 0) < £
for all n > N. It is not difficult to see that the answer to this question is
independent of whichever of the above metrics we use for p; that is, given
£ > 0, there exists such an N using px if and only if there exists such an N
using p2, etc. The answer does not depend on the shape of the ball of radius e,
but only on its “fatness” or “openness.” For this reason among others, it is
6
convenient to gather together those properties of a metric space that are

essential for describing “openness” and to use such properties to define a
more abstract structure, a topological structure, in which we can still talk
about limit points and in which the three metric structures on the plane
described above will give the same “open sets.”
Definition. A topological space is a set S together with a collection of sub¬

sets of S (that is, ^ is a subset of 2s) satisfying the following conditions:
(la) 0 e
(2a) If JJX,..., f/n e <?/ then P)"=1 Ut e °U.
(3a) Arbitrary unions of elements in lie in that is, if # <= °ll, then
Uce# Ue .
The elements of are called open sets in S. The collection °ll is called a
topology on S.
Remark. We shall often suppress the and simply refer to S’ as a topo¬

logical space.
Definition. Let (S’, be a topological space. A set A <= S is closed if it is the

complement of an open set, that is, if A' e °U.
Remark. By taking complements in conditions (la), (2a), and (3a) above,

one sees that the collection of closed sets satisfies the following conditions.
(lb) 0 e*f, Se%.

(2b) If Alt...,AneV, then (Jf= iAteV.
(3b) Arbitrary intersections of elements in ^ lie in ^.
Remark. A topology can be described by specifying the collection of closed

sets equally as well as specifying the collection of open sets.
Definition. Let (S, be a topological space. Let A <= S. A point s e S is a

limit point of A if for each U e °ll such that s e U,
(U - M) n A ^ 0.
Definition. The closure of a set A S, denoted by A, is the set
A = A u [s e S; s is a limit point of A].
Theorem 1. The closure A of a set A is closed.

Proof. We must show that A' is open. For this it suffices to show that for
each s e A' there exists an open set Us with s e Us ^ A’. Then s e Us for each s
7
implies A' <= (Jsejr Us and Us c= A' for each s implies Usei' Us^ A’. Thus
A' = UseZ' Us is a union of open sets and hence is open.
Now let s e A'. Then s is not a limit point of A, so there exists an open set
Us such that s e Us and (Us - {j}) n A = 0. Furthermore, s $ A because
s$A and hence, in fact, Us n A = 0. Since each element of Us is contained
in an open set, namely Us itself, whose intersection with A is 0, it follows that
Us contains no limit points of A and Us n A = 0 ; that is, Us <= A'. □
Theorem 2. A set A is closed if and only if A — A.
Proof. Assume A is closed. Then A' is open. If s <£ A, then A' is an open set
containing s such that (A' - {5}) n A — 0. Thus s is not a limit point of A.
Hence all limit points of A lie in A; that is, A = A.
Conversely, if A = A, then A is closed by the previous theorem. □
Definition. A set 28 <= 2s is a basis for a topology on S if the following con¬

ditions are satisfied:
(lc) 0ei
(2c) fJBegg B = S.
(3c) If B± and B2 e 28, then Bx n B2 = (JB€$ B for some subset 28 <= 28.
Theorem 3. Let S be a set and & be a basis for a topology on S. Let
= [ U £ 2s; U is a union of elements of 28}.
Then is a topology on S, the topology generated by 28.
Proof. We must verify that °Uss satisfies the three open-set axioms for a
topology on S.
By (lc) and (2c) in the definition of a basis, both 0 and S e so that
condition (la) in the definition of a topological space is satisfied.
Suppose 'V' c <%a. Then Uve-r V = Uve-r (Ube®v B) = {JBe$ B where
28v <= 28 for each Kef and J? = (J ve-r&v c Hence condition (3a) holds.
We prove condition (2a) by induction. We assume that the intersection of k
sets in lies in °llgg. (For k = 1, the statement is automatically true.) Sup¬
pose then Uu ..., Uk + 1e By the inductive hypothesis, Uxr\- • -n Uke
; that is, there exists a subset 28 28 such that U1 n ■ ■ • n Uk = {JBl£3g1 B±.
Since Uk+1 e there exists a subset 282<^ 28 such that Uk + 1 = {JB2e3g2 &2•
Hence
I/i n- • -n Uk+1 = ( U n ( U = IJ (B, n B2).

\BxeSS1 / \Bae#a '
B2e332
But by condition (3c) in the definition of a basis, Bx n B2e

Hence Ux rv • -n Uk + 1 e □
8
Theorem 4. Let (S, p) be a metric space. Let
38 = [5s(o); s e S and a is a nonnegative real number].
Then £8 is a basis for a topology on S.
Proof
(1) Bs(0) = 0 for any s e S, so 0 e 38.

(2) For any a > 0, S = Uses Bfa), so S = UBe® B.
(3) Let sx, s2 e S, let ax, a2 > 0, and let T = Bsfax) n BS2(a2). We may
assume T ^ 0.
To show that T is a union of elements of £8, it suffices to show that for each
seT there exists as > 0 such that Bs(as) c: T. For then T c User Bs(as) <= T.
The first inclusion follows because 5 e Bs(as) for each s e T, and the second
because Bs(as) <= T for each s. Thus T = User Bs(as) is a union of elements
of 38.
Now for s e T, let b, = p(s, s,) for j = 1,2. Then bx < a3- since s e BSj(ay).
Let as = min {ax — bu a2 — b2}. Then as > 0, and we claim that Bs(as) c T.
For suppose t e Bs(as). Then
pit, s,) < p{t, s) + P(s, Sj) < as + bj < a} - bx + bx = ap,
so t e BSj(a,)J =1,2. □
Corollary. A metric space has the structure of a topological space in which the
open sets are unions of balls.
Definition. Let S be a set, and let 38x and 382 be bases for topologies on 5. 38x
and 382 are equivalent if they generate the same topology; that is, if
=°^ss2-
Theorem 5. Let S be a set, and let 38 x and 382 be bases for topologies on S. Then
38! and 382 are equivalent if and only if
(1) for each s e S and B1 e 38x with s e Bx, there exists B2 e 382 such that
s e B2 «= Bx, and
(2) for each s e S and B2 e 382 with s e B2, there exists Bx g 38x such that
se Bxcz B2.
Proof. Suppose 38x and 382 are equivalent, and let seBxe38x. Then
Bx g = ^2, so Bx = UB2eJ2 B2 for some subset 382 <= 382. Hence
s g B2 <= Bx for some B2 e 382 <= 382. Thus (1) is proved, and (2) is proved
similarly.
Conversely, suppose (1) and (2) are satisfied. We first show that
Let B e 38x. By (1) for each s e B there exists Bs e 382 such that s e Bs «= B.
Now B <= UseB Bs c: B, so B = UseB Bs G Thus C Similarly,
using (2), c and so = ^2. □
9
Corollary. The three metrics pu p2, p3 on the plane S described earlier all
determine the same topology on S.
Proof. Conditions (1) and (2) of Theorem 5 are clearly satisfied. □
Remark. It is not true, however, that all metrics on a given space give the
same topology. For example, consider the space R of real numbers with the
following two metrics.
PiOi, r2) U ~ r2 |
Pz(ri, r2)
fO Oi = r2)
U (A ^ r2)
The open sets in the topology defined by p1 are the usual open sets (generated
by open intervals) in R, whereas the collection of open sets defined by p2 is the
set 2R of all subsets of R. For, in fact, relative to p2, Br(f) — {r} for each
r e R, so each “point” is an open set; and, hence, so is each union of “points,”
that is, each subset of R.
Definition. If (S, tft) is a topological space and = 2s, then S is said to have
the discrete topology.
Theorem 6. Let (S, °l() be a topological space and let A <= S. Let
= [A n U; U e
Then A is a topology on A, called the relative topology on A.

Proof. This is a consequence of the following facts.
(1) 0 n A = 0, S n A = A.
(2) = (fl?= 1 Ut)nA.
(3) Uce# (UnA) — U) n A. CD
Remark. When dealing with subspaces, one must be careful to specify which
topology is being used at any given time. Thus, for example, if U is the open
interval (1, 3) <= R and A is the closed interval [2, 4] c= R; then UnA is open
in A, but not open in R. However, if A is either open or closed, we do have
the following theorem.
Theorem 7. Let S be a topological space and let A <= S. If A is open in S, then

every open set in A is open in S. If A is closed in S, then every closed set in A
is closed in S.
Proof. If A is open in 5 and B is open in A, then B = UnA for some open
set C in so 5 is open in S since both U and A are open in S.
If A is closed in S, and B is closed in A, then A — B is open in A; so
A — B = A n U for some open set U in S. Thus
B = A-(A-B) = A- AnU=AnU'.
Since A and U' are both closed in S, so is A n U' = B. □
10
1.3: Connected and compact spaces
Theorem 8. Let S be a topological space and let Q <= S have the relative
topology. Let P be a subset of Q. Then the relative topology °UX on P regarded
as a subset of Q is the same as the relative topology '$/2 on P regarded as a
subset of S.
Proof. Let A <= p. Then
A A = P c\ C/for some open set U <= Q

o A = P n (Q n 0) for some open set U ^ S
o A = P n U for some open set U <= S (since P c O)
o A e <&2. □
1.3 Connected and compact spaces

Definition. A topological space S is connected if the only sets which are both
open and closed are 0 and S.
Theorem 1. A topological space S is connected if and only if it is not the union

of two disjoint nonempty open sets.
Proof. Assume S is connected. Suppose S = Fi u V2 for open sets V1 and

V2 with fj o f2 = 0. Then V1 = V2, so V1 is closed as well as open. Since
S is connected, either V± = 0 or ^ = S. If Vx = S', then V2 — 0; so in
both cases either V1 or V2 must be empty.
Conversely, suppose S' is not the union of two disjoint non-empty sets.
Let V <= S be both open and closed. Then V is also both open and closed,
and S is the union of the disjoint open sets V and V. Thus either V — 0 or
V = 0; that is, either V = 0 or V = S, so S' is connected. □
Examples. It is shown in real analysis that the following spaces are connected:
(1) The space R of real numbers,

(2) Any interval in R,
(3) Real n-space Rn, and
(4) Any ball or cube in Rn.
Remark. A subset of a topological space is said to be connected if it is

connected in the relative topology.
Theorem 2. Let S be a topological space, and let T0 and {Tw}weW be connected

subsets of S. Assume T0 n Tw A 0 for each w e W. Then T0 u (Uu'ew Tw)
is connected.
Remark. This theorem can be used to prove that Rn is connected, given the
fact that lines in Rn are connected. For, in fact, let T0 = {0} and {Tw}weW
denote the set of all lines through 0. (The indexing set W can be taken to be
the unit sphere in Rn.) Then Rn = T0VJ (Uêw Tw).
11
Proof. Let T = T0 u (Uu,ew Tw). Suppose T = V, u V2 for some disjoint

sets Vx and V2 open in T. Then for each w, V1 n Tw and V2 n Tw are disjoint
open sets in Tw, and their union is Tw. Since Tw is connected, either Vx n Tw =
0 or V2 n Tw = 0. Similarly, either V1 n T0 = 0 or V2 n T0 = 0 ; say
F2 n T0 = 0. Then V1 n T0 = T0; that is, T0 c Therefore, since
T0 r\ Tw 0, V-! r\ Tw ^ 0 for each w e W. Thus, from above, V2 n Tw =
0 for all w e Wand V1 n Tw = Tw; that is, Tw cz V± for all weW. Therefore,
^ = T0 u Ty^), and V2 = 0 ■ C!
Example. Rn — {0} is connected for n > 1. Prove that it is.
Definitions. Let S be a set. A collection 'V' <= 2s is a covering of S if

{Jve-r V = S. If S is a topological space and each V e y is an open set,
"V is called an open covering of S.
A topological space S is compact if every open covering has a finite sub¬
covering; that is, if for every open covering Y", there exist a finite number of
sets, say Vlt..., Vkeir for some k, such that S = U?=i V}.
Example. It is shown in real analysis that
(1) the compact subsets of Rn are the closed bounded subsets of Rn (Heine-
Borel theorem),
(2) Rn is not compact, and
(3) an open interval in R1 is not compact.
Definition. A topological space is said to have the finite intersection property

(abbreviated f.i.p.) if every collection ^ of closed sets with property (a)
also has property (/3):
(a) ^ 0 for each finite subcollection {F1,..., Fk} <= J5!
ifi) O F A 0•
Theorem 3. Let S be a topological space. Then S is compact if and only if S

has the f.i.p.
Proof. There is a one-to-one correspondence between collections LF of closed
sets in S and collections "T of open sets in S given by complementation; that
is, y LF if and only if F = [F’; F e iF\ or, equivalently, SF = [Vr; VeF'].
Now if y corresponds to ^ then
PjF=0<>UF=5.
That is, F is an open covering if and only if IF does not satisfy property (y3);
and F has a finite subcovering if and only if FF does not satisfy property (a).
□
Theorem 4. Every closed subset of a compact space is compact in its relative
topology.
12
1.4: Continuous functions
Proof. Let A be a closed subset of a compact space S. We show that A has

the f.i.p. Let JF be any collection of closed subsets of A satisfying property (a).
Since A is closed and each FedF is closed in A, each FedF is closed in S.
Thus SF is a collection of closed sets in S' satisfying (a). Since S is compact,
Of^ F ^ 0. But C)Fe$ir F c A, so A has the f.i.p. □
Theorem 5. Let S be a compact topological space. Then every infinite subset of

S has a limit point.
Proof. We show that if A <= S has no limit point, then A is finite. The proof
is in three steps.
Step (7). A is discrete; that is, its relative topology is the discrete topology.
For suppose a e A. Since a is not a limit point of A, there exists an open set
Ua <= S containing a such that (Ua - {a}) n A = 0; that is, Ua n A = {a}.
Thus each {a} is an open set in the relative topology of A and hence A is
discrete.
Step (2). A is compact. For in fact, since A has no limit points, A — A,
and hence A is closed. But, by Theorem 4, this implies A is compact.
Step (3). A is finite. For let Ua = {a} for each a e A. Then, by Step 1,
{UfacA is an open covering of A. By Step (2), A is compact, so there exists a
finite subcovering {Uai,..., UaJ.
Thus A = (J£=1 Uaj = {«!,..., ak} is finite. □
1.4 Continuous functions

Definition. Let S and T be topological spaces. A function f: S-+T is con¬
tinuous if the inverse images of open sets are open; that is, for each open
set U <= F,/_1(i7) is open in S.
Theorem 1. Let S and T be topological spaces. Let 3Ss and 3ftT be bases for the
topologies on S and T respectively. Then f: S -> T is continuous if and only
if for each s e S and each V edST with f(s) e V, there exists a U e dSs such
that s e U and f(U) <= V.
Remark. For metric spaces, Theorem 1 shows that our definition of con¬
tinuity is equivalent to the usual e, S definition of continuity.
Proof. Assume /is continuous. If 5 e S and f(s) e VeddT, then /_1(F) is
open in S and so is a union of elements of 3&s. Since 5 e/_1(F), s must belong
to one of these basis elements. Call it U. Then U <= /_1(F), so f(U) <= V.
Conversely, suppose for each such s and V there exists such a U. Let V
be open in T. We must show that/_1(F) is open in S. Let s e f~1(V). Then
f(s) e V, so f(s) lies in some basis element Vs eddT with Vs <= V. Therefore,
there exists a basis element Us e&s such that s e Us and f(Us) <= Vs c V;
that is, Us <= f~1(V). Then/_1(F) = Usez-fiv) Us; so/-1(C) is open in S.
□
13
Theorem 2. Let R, S, T be topological spaces. If g: R S and f: S —> T are

both continuous, then f ° g: R-> T is continuous.
Proof. Let Lbe open in T. Since/is continuous,/_1(L) is open in S. Hence,

since g is continuous, g'1(/_1(L)) is open in R. But (f ° g)~1 = g_1 °f~1, so
(/0 g)~1(V) is °Pen in and f ° 8 is continuous. □
Theorem 3. Let S and T be topological spaces. Let f'.S^Tbe continuous and

surjective. If S is connected, then so is T.
Proof. Suppose Vx and V2 are disjoint open sets in T with Vx u V2 = T.

Then/_1(F1) and f~\V2) are disjoint open sets in S with f~1{Vj) u/_1(L2)
= S. Since Sis connected, eitherf~1{Vj) orf~1(V2) must be empty. But since
/is surjective, f~\Vj) — 0 for some j implies Vj = 0. Thus T is connected.
□
Corollary. If f: S ^ T is continuous, and if S is connected, then f(S) is con¬
nected.
Example 1. Let /: [a, b] -> R1 be a continuous real-valued function defined

on the closed interval from a to b. If f{xj) > 0 for some xx e [a, 6], and
f(x2) < 0 for some x2 e [a, b], then f{xj) = 0 for some x0 6 [a, b\.
Proof. Let T — f([a, 6]). Then T is connected by the corollary since [a, b] is
connected. But if 0 j T, then T — (i?_ n T) u (R+ n T) is the union of
nonempty disjoint open sets. Here R+ is the set of positive real numbers,
and the set of negative real numbers. Thus 0 e T; that is, 0 = f(x0) for
some x0. □
Example 2. Let Sn be the unit sphere in Rn + 1; that is,
71+ 1
Sn = (x1,...,xn + 1)eRn + 1; J x,
i= l
Then Sn is connected for n > 0.
Proof. From Section 1.3 we know that Rn + 1 — {0} is connected. Let
/: Rn + 1 - {0}-^Sn
be defined by
*2 *n+l \
Vj'Xf’'"’ VYx?)'
Then /is continuous and surjective; so Sn is connected. □

14
1.4: Continuous functions
Example 3. Let GL{n, R) and GL(n, C) denote respectively the sets of non¬
singular n x n matrices with real and complex entries. Then by stringing out
the rows of each matrix in a line, GL(n, R) may be regarded as a subset of
Rn2, and hence is a topological space as a subset of Rn2 with the relative
topology. Similarly GL(n, C) c C"2 is a topological space. (Note that com¬
plex n2-space C"2 has the topology of the product of R2 with itself n2 times.)
Now GL{n, R) is not connected. For let
A: R^-^R1 - {0}
be the determinant function. A is continuous and surjective. Since R1 - {0}

is not connected, GL(n, R) is not connected.
Note that this argument fails for GL{n, C) since A: GL{n, C)-> C - {0},
and C - {0} = R2 — {0} is connected. In fact, as we shall see later, GL(n, C)
is connected.
Theorem 4. Let S and T be topological spaces, and let f: S -> T be continuous

and surjective. If S is compact, then so is T.
Proof. Let 'V be an open covering of T. Then [/_1(L); Ve'f'] is an open
covering of S. Since S is compact, there exists a finite subcovering
Since / is surjective, /(/_1(K,)) = V,- for j = 1,..., k, and U;=i V, = T

because (J*=i f~\Vj) = S. Thus T is compact. □
Corollary. If f: S -» T is continuous and S is compact, then f(S) is compact.
Corollary 2. Let f'.S-^R1 be continuous. If S is compact, then f assumes its

maximum and its minimum.
Proof. By Theorem 4, f(S) is compact, and therefore a closed bounded set
in R1. The maximum of/is the l.u.b. of f(S). It is a limit point of /(S'), hence
is in /(S) because /(S) is closed. Similarly the minimum of/ is in /(S). □
Definition. Let S and T be topological spaces and f:S^-T.f is a homeo-

morphism if / is a one-to-one correspondence and both / and f~1 are
continuous.
Remark 1. Note that although /-1 has been defined as a map 2T —>2S,
it may, in the case where / is a one-to-one correspondence, be regarded as a
map T-> S by identifying the one-point set {/} with t and /-1({t}) with .y,
where f(s) — t.
Remark 2. The fact that, for a homeomorphism / both / and /-1 are
continuous means that/not only maps points of S to points of T in a one-to-
one manner, but/ also maps open sets of S to open sets of T in a one-to-one
15
manner. This means that S and T are topologically the same; that is, any
topological property enjoyed by S is also enjoyed by T and conversely. Thus
if f: S —> T is a homeomorphism, then S is connected if and only if T is
connected; and S is compact if and only if T is compact.
1.5 Product spaces

Definition. Let T be a set and let °U and y be two topologies on T. The
topology °U is said to be weaker than the topology y if <= 'V. Or
equivalently, if the identity map
I: (T, V) -> (T, <%)
is continuous.
Theorem 1. Let S be a set, and let iV~ <= 2s. Then there exists a weakest topology
°l/ on S such that iV <=- tyi.
Proof. Let
sd = [y-,W c f c 2s, y a topology on S].
stf ^ because 2s e s&. Let

0 = C\-reS# ^ Then iV' ^ °ll. Moreover, °ll is a
topology on S; that is, satisfies the open set axioms. For, in fact,
( ) , S e °ll because 0, S e'f' for each y e stf.
1 0
(2) If Ui,..., Uke °U, then Uu ..., Ukeir for each y e stf so fj> = i U,ei'r
for each y e sd and hence fjy= i U;- e °l/.
(3) If Uwe°l/ for each w e W, W some indexing set, then Uwey for each
w e W and each yestf so Uu>ew Uwey for each f and hence
LJioeW Uw £ .
Thus is a topology on S. °li is the weakest topology containing W

because if $ is any other such topology, then °ii e. sd and hence =
rw r □
Remark. A basis for the weakest topology on S containing W is
& = [U1n---n Uk; UjeWforj= 1

where k is any positive integer].
(Throw in 0 and S.)
Exercise. Prove the above remark.

Remark. We can regard °ll in two distinct ways: as the intersection of all
topologies on 5 containing iy and as the topology “generated” by W.
This is analogous to the situation in linear algebra where, given a subset W
of a vector space V, the smallest subspace of V containing W may be regarded
either as the intersection of all subspaces containing W or as the linear space
spanned by W.
16
1.5: Product spaces
Theorem 2. Let W be an indexing set and let {Tw}weW be topological spaces.

Let S be a set, and let {fw}weW be a collection of functions, fw: S -> Twfor
each w eW. Then there exists a weakest topology on S such that fw: S -> Tw
is continuous for each w e W.
Proof. Let = [fw~\Vw)\ Vw open in Tw, w e W], Let °U be the weakest

topology such that iV <= fy. Clearly, any topology on S such that each /„, is
continuous must contain iV. °l/ is the weakest such topology. □
Remark. The set
@ = fw 1(krw); Wx is a finite subset of W,

LuieWi
Vw open in Tw for each w e W1
is a basis for the topology °l/ of Theorem 2.
Definition. Let {5„}„,6W be a collection of topological spaces. Let P = Sw

(Cartesian product of sets). Let ttw : P Sw be defined for each we W by
^(/)=/(w), (feP).
(Recall that a point /e HLew Sw is a function
/: S-
weW
such that /(w) e Sw for each w e W.) The product topology on P is the
weakest topology in which each ttw is continuous. This topology exists by
Theorem 2.
Remark. The function ttw(w e W) is projection onto the factor Sw. A basis
for the product topology is
SS -
n W1 a finite subset of W and Uw
.HIEWi
an open set in Sw for each w e W1
Let us take a closer look at these basic open sets. First, for each w0 e W and
each open set UWo <= SWo, -rrWo~\UWQ) is the “cylinder” Jflwew Tm where
T _ fSw (w A w0)
w \UW0 (w = w0).
For, in fact, fe ttWq-\UWo) o itWo(J) e UWo o f(w0) e UWo of(w) e Tw for

each we W of e nêw Tw. Taking finite intersections, it follows easily that
n = ne»
weW i weW
17
where
jSw (wtWJ
\ uw (weWJ.
Thus the basis & for the topology on P consists of products of open sets, one
in each factor Sw, with all but finitely many of these being the whole space Sw.
Example 1. Let / denote the open interval (0, 1). Then I x I is the open unit
square in R2. The projection maps are tt}: I x /-> I(j = 1,2), defined by
77^0, b) = a, tr2(a, b) = b.
For Ult an open interval in I, the cylinder t71-1(C/1) is the shaded vertical
strip in Figure 1.3. For U2, an open interval in /, v2~1(U2) is the shaded
Figure 1.4
horizontal strip in Figure 1.4. Thus 771-1(£/1) n tt2~\U2) is the open square
in Figure 1.5. Since the open intervals form a basis for the topology on /,
it follows that the open squares form a basis for the product topology on
I x I. That is, the product topology on I x I is the same as the topology
induced on I x I by the metric space topology on R2.
18
1.5: Product spaces
u*
Figure 1.5
Example 2. The circle S1 is a subset of R2, so it is a topological space in the

induced topology. The product S1 x S1 of the circle with itself is a torus
(the surface of a tire or doughnut). In the product topology on the torus, a
basis for the open sets is given by “rectangular patches,” as in Figure 1.6.
The student should find it instructive to convince himself that the product
topology on S1 x S1 is the same as the induced topology on the torus
considered as a subset of R3.
Example 3. Let R+ denote the positive real numbers, and let Sn denote the
unit sphere in Rn + 1. Then there exists a one-to-one correspondence
tp: Rn + 1 - {0}->£+ x
given by
where ||x|| = (2"=i x2)112. cp is in fact a homeomorphism between Rn + 1 — {0}

in the induced topology and R+ x S’1 in the product topology. Here R+ and
Sn are topologized as subsets of R1 and Rn + 1 respectively.
19
Example 4. Let J+ = [«; n is a positive integer] and let /„ = [0, 1 /»]. Then
P = rine/+ 4 is a topological space in the product topology. We introduce a
metric on the point set of P as follows. If x = (xlt..., xn,...) and
y = (yi, ■ •yn,.. •) are in P, then
Exercise. Show that the series converges so that p is well-defined. Show that p
is a metric on P and the metric topology on P is the same as the product topology.
1.6 The Tychonoff theorem

One of the most important properties of product spaces is given by the
following theorem. In fact, this theorem is one strong reason why the product
topology we have described in Section 1.5 is a good topology for infinite
products (better, for example, than one where products of infinitely many
proper open subsets of the factors are allowed as basic open sets).
Theorem 1 (Tychonoff). The product of compact spaces is compact.
Proof. The proof is deferred to the end of Section 1.6.
Remark. Let us first consider an application of Theorem 1. Recall from

analysis that there are many nice properties enjoyed by continuous real-valued
functions on compact spaces which do not hold when the domain is not
compact. For example, every continuous real-valued function on a compact
set assumes its maximum. Thus it is reasonable to inquire when a continuous
function, defined on some noncompact set S, can be extended to a continuous
function defined on some compact set C => S.
Definition. If/: 5 T and if S <= C, then a continuous extension g of /is

a continuous map g: C-> T, such that g(s) = f(s), s e S; that is, g|s = /
Example 1. Let S = (0, 1], the half-open interval. Let /: R1 be defined

by f(x) = sin x for x g (0, 1]. Although 5 is not compact, it is clear that/can
be extended to a continuous function on the compact set [0, 1], the closed
interval, by setting /(0) = 0.
Example 2. Let S = (0, 1] as before. Now let/: S -> R1 be defined by
f(x) — sin \/x for x e S.
Once again,/is a continuous function on the noncompact set S, but now it is

clear that / cannot be extended to a continuous function on the compact set
20
1.6: The Tychonoff theorem
[0, 1]; that is, we cannot, by adding one point, obtain a compact space on
which / can be continuously extended. But it turns out that if we “add”
sufficiently many points, we can obtain a compact space P; a continuous
function cp: S^P (cp will in fact turn out to be a homeomorphism onto
<p(S) c P); and a continuous function f:P-> R1, such that
(1) cp(S) is dense in P; that is, cp(S) = P,

(2) /« cp(S).
We construct P as follows. Let / = [—1, 1], and let Px = [0, 1] x /, the
closed square.
Let cp: S -» Px be defined by
<p(jc) = (x, sin 1/jc) (x e (0, 1]).
(Note that if/: [0, 1] x /-^ R1 is defined by
Ax, y) = y,
then f ° P. Now P, cp, and the restriction of / to P have the required
properties.
Figure 1.7
21
In summary, we have replaced (0, 1] by the graph <p(S) of sin l/x (a homeo-
morphic copy of (0, 1] (see Figure 1.7). We have carried the function/over to
cp(S) through the homeomorphism 92_1. Then we have extended our function
to the compact set
({0} x [-1, 1]) u ([(*, sin l/x); x e (0, 1]).
Another way to look at the result of this construction is the following.

Let us take the set <p(S) and identify it with S through the homeomorphism <p.
Then the set P is identified with the set (Figure 1.8)
p=m x [-i, i])u«o, i] x {o}).
-1
Figure 1.8
The function / is now a continuous function on P that agrees with / on

(0, 1 ] x {0}. However, note that the topology on P is not the induced topology
from R2; it is the topology of P carried over through the identification
S -> <p(S). Thus, for example, each point of {0} x [ — 1, 1] is a limit point of
the set (0, 1] x {0}.
Example 2 illustrates the main points in the statement and proof of the
following theorem, which is a consequence of the Tychonoff theorem.
Theorem 2. Let Wbe a collection of continuous real-valued (or complex-valued)

functions on a topological space S with the property that each w e W is a
bounded function; that is, vr(S') c Iw, where Iw is, for each w e W, some
closed bounded interval (or ball). Then there exists a compact space P; a
continuous function <p: S -» P; and a family dF of continuous real-valued
(complex-valued) functions on P such that
(1 )W) =p
(2) [fo<p;fe&] = W.
If, in addition, W separates points of S (that is, if, for each s± F s2e S, there
exists a w e W such that w^) F vv(.s-2)), then cp is injective.
Proof. Let P1 — nêw Iw Since each Iw is compact, // is compact by the

Tychonoff theorem. Let tp: S -> Px be defined by
<p(s)(w) = vv(^).
22
Then <p is continuous. To prove this it suffices to show that inverse images of
basic open sets are open, because each open set is a union of such sets.
Suppose U is a basic open set in Px. Then
U = Pi 7Tw~1(Vw)
weWi
for some finite subset Wx c= w, where Vw is open in Iw for each w e Wx. Thus
cp-\U) = f5eS;95(5)G pi
= [5 g S', 99(5 )(w) g Vw for each w e Wx]

= w(s) g Vw for each w g Wx]
= p w-\vwy
we ffi
Thus, since each w e W is continuous, cp~1(U) is a finite intersection of open

sets. Hence cp~1{U) is open, and 99 is continuous.
Now let P = <p(S). P is compact, for it is a closed subset of a compact
space. Then 99: 5 —> P is continuous, where P has the relative topology in Pu
because 93: S^PX is continuous. Moreover, from the definition of <p, it is
clear that
TTw 0 <P(S) = W(^)
for each s e S and each w g W. Thus, if we define
& = WwIp', we W],
then
[/cpj/eJF] = W.
It remains only to prove that if W separates points, then 99 is injective. But

if, for each sx # s2 g S, there exists w e W such that / w(s2), then
9(^1 )(w) = wfo) ^ w(s2) = <p(s2)(w)
for that w, so 9>(^,1) ^ ( ).

99 52 □
Remark. In Example 2 above, W = {/i,/2} where fx{x) = sin \/x and

f2(x) = x. The function f2 was added to obtain separation of points.
Proof of Theorem 1. Let 3? be a collection of closed sets in P — Sw,

where Sw is compact for each w g W. Suppose has the following property:
(a) Intersections of finitely many elements of & are always nonempty.
To prove that P is compact it suffices, by Theorem 3, Section 1.3, to show that

Ope^r F 7^ 0 for each such
23
Now even though each collection
^ = MF);Fe&]
is a collection of closed sets in the compact space Sw with property (a) (and
hence each (~]Fe^ ttw(F) F 0)—one cannot hope to find a point in (~]Fe^ F
merely by selecting a point x such that vw(x) e f)Fs& ™W(F) for each w g IV.
For example, if P is a closed square in R2, let & be the following collection of
pairs of closed balls about two given points P1 = (x1? yx) andP2 = (*2, y^^P'-
& =
Then the point P3 = (xl5 y2) has the property that
^ (p3) e n
and
, ^(Pa) e P) ^2(F)
Fe3?
but yet P3 $ PiFeST F.

To remedy this situation we shall enlarge tF by adding enough sets to
eliminate “poor choices.” Let A = ^ is a collection of subsets of P with
property (a) and ZF c ^]. Note that F e A. Furthermore, A is partially
ordered by inclusion; that is, a partial ordering relation R in A x A is given
by
(#!, ^GRO^CZ 02.

Now {F} c A is a simply ordered subset of A consisting of one point, since
F F. By the maximum principle (see Section 1.1), there exists a maximal
simply ordered Ax c A such that {F} c Ax. Let JF = *&. Then
^ c We claim F? e Ax. Note first that F* e A; that is, F has property
(a). For, if H1,..Hk eF, then Hf e for some ,e Ax (j = 1,..., k).
Since Ax is simply ordered, there exists a j0 such that c <g.Q for all j =
1,..., k. Therefore, all //y e and, since ^jo has property (a), Hy=1^/0.
Thus Jf eA.
But, in fact, F e Ax. For if F Ax, consider the set A2 = Ax u {F}.
Then A2 is a simply ordered subset of A containing {«:F} and Ax c A2. This
contradicts the maximality of Ax.
Note that F has the following property:
08) If Ac p and K n H F 0 for all HeF, then KeF.
For otherwise F = F u {K} satisfies property (a); hence Of e A. Thus

Ax u {Fj is a simply ordered subset of A (F c ^ for each ^ e Ax) con¬
taining Ax, which contradicts the maximality of Ax.
24
Now, for each w e W, let
JT. = [7Tw{H);HejP].
Then is, for each w e W, a collection of closed sets in Sw, satisfying

property (a) because Ji? satisfies property (a). By compactness of Sw,
Tw = TTW(H) 7^ 0
for all w e W. By the axiom of choice (see Section 1.1), HLew Tw + 0 ; that is,
there exists f e P such that f{w) e ^W(H) for each w e W.
It remains only to show that/6 (~]Fe^ F. To prove this it suffices to show
that if V is an open set in P containing/, then VeJF. For if V n H ^ 0 for
all H eJf, then in particular Fn F # 0 for all Fe That this is true for
each open V containing/implies that/e F; that is,/e F since Fis closed for
each F e FF. Thus/e C)Fe^ F, proving the theorem.
So suppose V is an open set containing /. Since V is a union of basis
elements, / e V <= V for some basis element V. Since Fisa basis element,
-i,
(K)
for some finite subset Wx «= W, where Vw is, for each w e Wx, an open set in
Sw. Since /e V, f(w) = vw(f) e Vw for each w e Wx. Furthermore, f(w) e
ttw(H) for each w e Wx and each H eJF, by our choice of /; that is, for each
w e Wx and H e 2FP, either f(w) e or f(w) is a limit point of In
either case, it follows that Vw n nw(H) ^ 0 for each w e Wx, where H e JF.
Therefore, ttw~1(Vw) n H / 0 for all w e Wu HeJF. By property (/3),
7rw~1( Vw) e JF for each w e Wx.
Now by property (a) for any H e JP,
Hence, again by property (/?), V e JF. □
25
More point set topology
2.1 Separation axioms

In Chapter 1 we dealt with topological concepts where it was not particularly
important whether open sets were plentiful or scarce. However, the topological
spaces usually encountered in applications have plenty of open sets, almost
always enough to “separate” points, and often enough to “separate” closed
sets.
Definitions. Let S be a topological space.

S is a T0 space if, given any pair of distinct points, slt s2 e S, there exists
an open set U in S' containing one of these but not the other.
5 is a T1 space if, whenever sx A s2, there exists an open set U1 such that
Si e Ui but s2 $ U1, and there exists an open set U2 such that s2 e U2 but
■A ^ U2.
S is a T2 space, or Hausdorff space, if, whenever ^ s2, there exist open
sets U} with s, e Uj (j = 1, 2) such that U1 n U2 = 0 ; that is, U1 and U2
are disjoint.
Remark. Note that every T2 space is a 7^ space and every T1 space is a T0

space. An example of a T0 space which is not a T1 space is given by
S = {^, s2} — a set consisting of two points.
The topology on S is °U = {0,5,
Exercise. Find a T± space which is not T2.
26
2.1: Separation axioms
Remark. Any metric space is a Hausdorff space. For if sx A .?2, let
a = P(sx, s2) > 0,
and take Uj = BS)(a/2), (j = 1, 2).
Theorem 1. S is a Tx space if and only if each point of S is closed as a subset of S.
Proof. Suppose S is Tx. Let F — {sq} and consider F'. Since S is Tx, there
exists about each point 5 e F' an open set Us such that s0 f Us, that is, such
that Us <= F'. Thus F' = UseF' Us is open in S, and hence Fis closed.
Conversely, suppose points of S are closed. Let Ux = {.y2}' and let U2 =
{■sy}'. Then if sx A s2, sx e Uj and Uj are open. □
Theorem 2. Every compact subset of a Hausdorff space is closed.
Proof. Let C be a compact subset of S’. To show that C' is open, and hence C
is closed, it suffices to show that for each s f C there exists an open set Us
containing s with Us n C = 0. Since 5 is Hausdorff, there exists for each
c e C disjoint open sets Uc and Vc such that s e Uc, c e Vc. The collection
{Vc n CjceC is then an open covering of C. (These are open sets in the relative
topology.) By compactness of C there exists a finite subcovering; that is,
there exists a finite subset Cx C such that Uceej. (Vc n C) = C. Thus
c <= UceCi ve. Let U„ = HceC! Uc. Then Us is open, s e U„ and
[/snCc Us n (UceCi Vc) = 0. This last equality holds because if
t e Usn (UceCi Pc), then t e Uc for all c e Cx and t e VCo for some c0e Cx;
hence t e UCo n Poo = 0, which is impossible. □
Remark. We have actually shown, in the course of proving Theorem 2,

that if C is a compact subset of a Hausdorff space S and if s f C, then there
exist disjoint open sets Ux and U2 such that s e Ux and C <= U2. We can, in
fact, prove more.
Theorem 3. Let Cx and C2 be disjoint compact subsets of a Hausdorff space S.

Then there exist disjoint open sets Uj such that C; <= Uj (j = 1, 2).
Proof. For each s e Cx there exist, according to the remark above, disjoint
open sets Us and Vs such that s e Us and C2 <= Vs. The collection {Us n C!}seCl
is an open covering of Cx. Since Cx is compact, there exists a finite subset
Dx <= Cx such that {Us n Cx}ssDl actually cover Cx; that is, Cx <= UseUi Us.
Let Ux — UseDi Us and U2 = (UseDi Vs. Then C2 c U2, and Ux n U2 = 0
since each Us n Vs = 0. □
Theorem 4. Let S be a compact space, and let T be Hausdorff. Then any

continuous one-to-one correspondence f:S->Tisa homeomorphism.
Proof. We must show that/-1: S is continuous; that is, if U is open in

S, then f(U) = (f~1)~1(U) must be open in T. By considering complements,
27
2: More point set topology
this is the same as proving that if F is closed in S, then f(F) is closed in T.

But by Theorem 4, Section 1.3, F is compact. Thus, by Theorem 4, Section
1.4,/(F) is compact. Hence /(F) is closed by Theorem 2 above. □
Examples. That both assumptions are necessary in Theorem 4 is illustrated

by the following two examples.
(1) S — R1 with the discrete topology (S is not compact). T — R1 with the

usual topology.
(2) S = {$i, .y2} with the discrete topology (S is compact). T = .y2} with
topology °ll — {0 ,S, {si}} (T is not Hausdorff).
f: S->T = the identity map in both examples.
Remark. Theorem 3 above shows that in a Hausdorff space there are

enough open sets to separate compact sets. Sometimes we need to be able to
separate closed sets. This requires a stronger axiom.
Definitions. A topological space S is regular (or T3) if
(1) S is 7\, and

(2) for every closed set Fin S' and each s <£ F, there exist disjoint open sets
U1 and U2 such that s e U1 and F <= U2.
A topological space S is normal (or F4) if
(1) S is Jj, and

(2) for every pair of disjoint closed sets F, in S, there exist disjoint open
sets Uj such that F, <= U} (j = 1, 2).
Remark. Notice that every Tk space {k = 0, 1,..., 4) is a F, space for each

j < k.
Example. We shall construct a space which is Hausdorff, but not regular. Let
s = [x = (x1} x2) e R2; x2 > 0]; that is, S is the “closed” upper half-plane.
We define a topology on 5 by giving it a basis F8 as follows. Let R1 denote the
xx axis in R2. Let S+ = S — R1, so S+ is the “open” upper half-plane. Let
= [Bx(r)l x e S+ and r < x2, where x = (x1? x2)]
and
^2 = [(^/) n S+) u {x}; x e R1],
where the balls Bx(r) are defined relative to the usual metric on R2. Now let
& = u @2. Verification that 3S is a basis for a topology on S is left to the
student. The basis elements in ^ are illustrated in Figure 2.1.
28
2.1: Separation axioms
Figure 2.1
Note that the topology on S is not the topology induced from the usual
topology on R2. Now it is easy to check that S is T2. Hence points are closed.
Furthermore, the set R1 — {0} is closed in S since its complement S+ u {0} is
clearly open. However, the closed set R1 — {0} and the point 0 cannot be
separated by open sets: if Ux is open and 0 e Uly then some basis element
(B0(r) n S+) u {0} is contained in Ui (for some real number r). Thus any
open set containing (r/2, 0) must intersect Ux. Hence S is not regular. (We
shall later give an example of a regular space which is not normal.)
Remark. Every compact Hausdorff space is normal. This is a consequence
of Theorem 3 by the fact that closed subsets of a compact space are compact.
Remark. The open sets Ut and U2 whose existence is asserted in the defini¬
tion of a regular space (or of a normal space) might conceivably have the
property that their boundaries intersect; that is, Ux n U2 might be nonempty.
According to the following two theorems (Theorems 5 and 6), it is possible
to choose our sets Ut more judiciously.
Theorem 5. Let S be a regular topological space. Suppose F is closed in S and

s $ F. Then there exist open sets Ux and U2 in S such that s e Ux, F <= U2,
and Ux n U2 = 0.
Proof. By the definition of regularity, there exist disjoint open sets Vx and
V2 with s e Vx and F <= V2. Consider the set V[. It is closed in S, and s V[.
Therefore, since S is regular, there exist disjoint open sets and W2 such
that s e Wx and V[ <= W2. Let Ux = Wx and U2 = V2. Then 5 e U1} and
F <=■ U2. Moreover, Ux n U2 = 0 because
Ux = Wx c Wl = W2 C Vx
and
u2 - v2 c vl = Vi
so that
Ux n U2 <=■ Vx n Vi = 0. □
29
Theorem 6. Let S be a normal topological space. Suppose Fl and F2 are closed

subsets of S. Then there exist open sets Ux and U2 in S such that Fx c U1,
F2 c U2, and Ux n U2 = 0.
Proof. Similar to the proof of Theorem 5.
Theorem 7. Let (S, p) be a metric space. Then S, with the associated metric
topology, is normal.
Proof. We have already seen that S is Hausdorff, hence Tx. For C any subset
of S, we consider the function dc: S -> R1 defined by
dc(s) = g-l.b. {P(s, c)},

ceC
where g.l.b. denotes greatest lower bound. Now, for Fx and F2 disjoint closed
sets in S, let dx = dFl and d2 — dF2, and let
Ux = [5 e S; dx(s) — d2(s) < 0],

Ua = [se S; dx(s) - d2(s) > 0].
Then Ux n U2 = 0. We shall show that Fx <= Ux, F2 <= U2, and that Ux and
U2 are open, thereby proving that S is normal.
First, suppose 5 e Fx. Then dx(s) — 0. So to prove that s e Ux, we must
show that d2(s) > 0. But if d2(s) = 0, then for every e > 0 there exists
f2 e F2 such that f2 e Bff). Hence s e F2 = F2, contradicting the fact that
Fx n F2 = 0.
Thus Fx <= Ux. Similarly, F2 «= U2. It now remains only to show that Ux
and U2 are open.
To show that Ux is open, it suffices to show that each sx e Ux is contained
in a ball which is contained in Ux. (The proof for U2 is identical.) Let
a = d2(sx) - dx(sx). Then a > 0. We prove that BSl(a/3) <= Ux. In fact, let
5 e BSl(a/3). Then p(s, sx) < a/3, and we must show that dx(s) — d2(s) < 0.
But
dx(s) - d2(s) = [rfx(5) - dt.isx)] + - d2(sx)] + [J2(^) - d2(s)]

= [î(-y) - ^1(^1)] - a + - ^2(5)].
Now, for each fxe Fx,
p{s,fi) < p(s,sx) + p(sx,fx) < | + P(sx,fx).
Taking the g.l.b. over all fx g Fx,
dx(s) < ^ + dx(sx)
or
î(^) - dx(sx) <
30
2.2: Separation by continuous functions
Similarly, for each f2 e F2,
p(sx,f2) < p(sx, 5) + p(s,f2) < ^ + P(s,f2)
so
d2(sx) < 3 + d2(s)
or
<40i) - d2(s) < |.
Thus
dx(s) - d2{s) <f-a + f<0,
so s eUx\ that is, BSl(a/3) <= £/, and £/ is open, as claimed. □
2.2 Separation by continuous functions

If 5 is a topological space in which the only open sets are 0 and S, then
one sees easily that the only continuous real-valued functions on S are the
constant functions. On the other hand, if the space is such that, for every
pair sx, s2 of distinct points in S, there exists a continuous real-valued
function on S with f(sx) ^ f(s2), then S is Hausdorff. (Take Ux and U2
disjoint open sets in R1 with f(sx)eUx and f(s2)eU2; then sjef~1{Uj)
(j = 1, 2), and f~1(U1) n f~1(U2) = 0.) Thus the existence of continuous
real-valued functions on a space is related to the separation axiom satisfied
by the space. Theorem 1 and the following remark characterize normality in
these terms.
Theorem 1 (Urysohn’s lemma). Let A0 and Ax be disjoint closed subsets of a

normal space S. Then there exists a continuous function f: S —> [0, 1] such
that f(A0) = 0 (that is, f(s) — 0 for all s e A0) and f(Ax) = 1.
Remark. Conversely, if a 7\ space S has the property that for each pair
A0, Ax of disjoint closed sets there exists such a function, then S is normal.
For let U0 =/-1([0,i)) and Ux =/"1((i, !])• Then A0 c= U0, Ax c= Ux,
and U0 O Ux = 0.
Proof. Given A0 and Au we construct/by first constructing approximations

Ur to the sets/-1([0, r)) for all r = k/2n (k = 1,..., 2n; n = 1, 2,...).
First, by normality, there exist disjoint open sets Ull2 and Vll2 such that
A0 <= Ull2 and Ax <= V1/2. Then we have
A0 c UX/2 c VX/2 c Ax.
31
Second, consider the disjoint closed sets A0 and U[l2. By normality, there
exist disjoint open sets UVi and Vvi such that A0 <= Ulti and Ulj2 c Vvi.
The sets V'v2 and Ax are disjoint closed sets, so, by normality again, there exist
disjoint open sets U3li and V3li such that Vl/2 <=■ U314 and Ai c F3/4. Now we
have
Ao c C/1/4 <— b^l/4 Uii2 c Fi/2 c C3/4 c F3/4 c
Continuing by induction, we obtain open sets Ur and Vr, defined for each
dyadic r = A:/2n, with this property: for each n,
A0 C Cl/2n C Vll2n C= C2/2n C C2/2n C b^2n-i)/2n C ^(2n-l)/2n C ^1-
In particular, for each pair r4 and /-2 of dyadics with r4 < r2,
U c f' u c F'
c L'r2 K r2*
Now we define /: S -*> [0, 1] by
'g.l.b. [r;je [/r] if 5 e Ur

m 1 ifîjl/r.
Since A0 c Ur for all r, /(^0) = 0- Similarly, since t/r c: A[ for all r,

Ur n A1 = 0 for all r and f(Ax) = 1. Thus to complete the proof we need
only verify that /is continuous. To do so, it suffices to show that for some
basis 88 of the open sets in [0, 1],/-1([/) is open for each U e 88. Let us take
as basis:
88 = [[0, a), (c, d), (b, 1]; a, b, c, d are irrational].
Now, since (c, d) — [0, d) n (c, 1], and hence /_1(c, d) = /_1([0, d)) n
1]), it suffices to show that/_1([0, d)) and/_1((c, 1]) are open for all
irrationals c and d.
But /_1([0, d)) is open because/-1([0> d)) = Ur<d Ur. For, if s e UTq for
some/-0 < /then/(s) < r0 < /so.ye/_1([0, d)). Conversely, if s £/-1([0, d)),
then f(s) < d, so there exists a dyadic r0 < d such that s e Uro Ur<d UT.
Similarly, 1]) is open because/_1((c, 1]) = Ur>c K- For if s e VTo
for some r0 > c, then 5 £ V'To, so s$ Uro, and f(s) > r0 > c. Conversely, if
.se/_1((c, 1])> then s$ Uro for some r0 > c. Since the dyadics are dense in
[0, 1], there exists a dyadic rx with c < rx < r0. Now s <£ Uro implies
■s' £ V'Tl <= Uro; that is, se Vn <= \Jr>c Vr. □
Corollary. Let A0 and A1 be disjoint closed subsets of a normal space S. Then

there exists a continuous function g: S [a, b] with g(A0) = a and
g(,Aj) = b.
32
2.2: Separation by continuous functions
Proof. Let /:S'-^[0, 1] be the function given by Theorem 1. Define

g = h of where h: [0, 1] -> [a, b] is defined by
h{x) = a + (b - a)x (x e [0, 1]). □
Remark. Theorem 1 may be rephrased in terms of continuous extensions as

follows.
Let S be a normal space and let A0 and Ax be closed subsets of S. Let
Si = A0 u Ax, and let /: Sx [0, 1] be defined by f(A0) = 0, f{Afi = 1.
Then there exists a continuous extension g of / to all of S.
Note that A0 and Ax are each both open and closed in Sx: they are open
because S is normal. Hence the function/: Sx -> [0, 1] is continuous.
Remark. Theorem 2 of Section 1.6 was also an extension theorem. Essen¬

tially it says that any family of continuous bounded real-valued or complex¬
valued functions on a topological space S can be extended to a family of
continuous functions on a compact space containing S.
Example. One should not expect that, given any Sx c= S and/: Sx -> T, there
exists a continuous extension of/ to S. For example, let S be the disc
s = [(•*> y) G R2', x2 + y2 < 1],
and let Sx be the circle
Si = [(*, y) e R2; x2 + y2 = 1],
Let T = Si, and let f: SX-^T be the identity map. Then there exists no
extension g to/to all of S. (This fact will later be proved rigorously when we
have more machinery at our disposal.)
Intuitively, we can see why no extension exists: If g were a continuous
extension, consider the behaviour of g on the concentric circles Sr of radius
r (0 < r < 1). Since g is continuous, nearby points are mapped into nearby
points. Thus, any point s of Sr, for r close to 1, must be mapped close to the
point s/r on Sx. Letting s move around Sr, we see that g(j') must move around
Sx, and g must map Sr onto Sx. Now, as r gets smaller, the same argument
still shows that g must map Sr onto Sx for all r > 0. But, for small enough r,
the center 0 is close to every point on Sr. Hence g(0) must be close to every
point in the image of Sr; that is, g(0) must be close to every point of Sx. Of
course, this is impossible.
Remark. The question When does a function defined on a subset of topo¬

logical space admit an extension to the whole space ? is one of the fundamental
questions of topology. It is this question which motivates much of the
machinery developed in algebraic topology. However, for real-valued func¬
tions on normal spaces, we can already prove the following.
33
Theorem 2 (Tietze extension theorem). Let Sx be a closed subset of a normal

space S, and let f: Sx-+ [—1, 1] be continuous. Then f has a continuous
extension g to all of S.
Proof. Let Ax = /_1([£, 1]) and A_x — /-1([—1, — £]). Then Ax and A_x
are disjoint closed subsets of Si. Since Sx is closed, Ax and A_x are closed in
S. By Urysohn’s lemma, there exists a continuous function /j: S[—j,
such that ffA ±1) = ± -J-. (We are actually using here the corollary to Theorem
1 above.) Then, for all s e Sx,
1/(0 -A(0| < f-

Next, consider the function/-/!: Si [-f, f]. Letzf2 = (/-/i)-1(f, f])
and A_2 = if ~ /i)_1([ — f, — f]). Then A2 and A_2 are disjoint closed
subsets of S. By Urysohn’s lemma, there exists a continuous function
/2: S-> [ —|, f] such that ffAf) = f and /204_2) = —f. Now, for all s e Sl5
1/(0 -/i(0 ~/a(Ol < t-

Continuing by induction, we construct functions/n for all positive integers
n, such that
— 2n-1 2n~1
fn:S
3" ’ 3n
and furthermore, for all s e Sx,
As) - | fls) < (f)n-

i=i
We may regard the/„ as functions fn\ [-1, 1] such that |/n(0| <
2n_1/3n. By the Weierstrass M-test, the series 2n=i/n converges uniformly to
a continuous function g: S [-1, 1], (We leave to the student the verifica¬
tion that these convergence properties, familiar from analysis on R1, are also
valid in the more general setting considered here.) Since
1/(0 - g(OI = lim 2m

m - i=i < lim (f)n = 0
for all ^ e Sx, we see that g(s) = f(s) on Sx. □
2.3 More separability

Definition. A topological space S is locally compact if, for each s e S, there
exists an open set Us with s e Us such that Us is compact.
34
2.3: More separability
Remark. A finite product of locally compact spaces is locally compact.

For, in fact, if
^ = (slt ...,sk)eS1 x • • • x Sk
'-V-'
(/r-fold product)
then there exist open sets U, <= (j = 1 with sj e U, and U, compact.

Thus s e U1 x • • • x Uk and U1 x • • • x Uk — U1 x • ■ ■ x Uk is compact.
Note that this argument fails in the case of infinite products since sets of the
form HLew Uw (Uw open in Sw) are in general not open in ELew Sw.
Definition. Let S be a 7} space. Let oo denote a point not in S. The 1-point

compactification of 5 is the topological space S obtained as follows. As a
point set, S = S u {co}. Let °lls denote the topology on S. Let
^ = [V c S; V is a compact closed subset of 5]

= [5 — F; F is compact and closed in S’],
A basis for the topology on S is then
38$ = ous \j V'
Exercise. We leave to the student the verification that &§ is in fact a basis for a
topology on S. Note that the relative topology on S as a subset of Sis the same as
the original topology on S.
Example 1. Let S = R1. Then S = R1 u {oo}, where an open set about oo is

a complement of a compact subset of R1. In particular, given an open set U
containing oo, there exists a real number Mv > 0 such that |x| > Mv
implies x e U. S is homeomorphic to the circle S1.
Exercise. Why is the last statement true?
Example 2. Let S = R2. Then S = R2 U {oo}, where the open sets about oo
are complements of compact sets in R2. In particular, each open set contain¬
ing oo must contain all points in the exterior of some sufficiently large ball.
S is homeomorphic to the 2-sphere S2. In fact, a homeomorphism <p: S2 —> S
is given by stereographic projection (Figure 2.2). We regard the sphere as
sitting on the xy plane with its south pole at the origin. Given a point s e S2,
<p(s) is the intersection with the xy plane (R2) of the line through the north
pole of S2 and s. It is clear that <p is a one-to-one correspondence mapping
S2 — {north pole} onto R2 and mapping the north pole to oo. That <p is, in
fact, a homeomorphism is easily checked.
Example3. Let S = Rn. Then S is homeomorphic to the ^-sphere Sn. A

homeomorphism is given, as in Example 2, by stereographic projection.
35
Remark. If S is compact, then {00} is open in S. Thus the 1-point compacti-

fication is, in this case, uninteresting.
Theorem 1. Let S be a T1 space. Then the l-point compactification S of S is

compact and Tx. Furthermore, S is Hausdorjf if and only if S is Hausdorjf
and locally compact.
Proof. S is Tx. For if and s2 e S, then appropriate separating open sets Ux

and U2 exist in S (since S is Tx), and Ux and U2 are also open in S. If sx e S
and s2 = °°> Ux — S and U2 = {.$1}'. Then Ux is open in S since it is open
in S, and U2 is open in S since it is the complement of the compact closed set
{sx}. ({si} is closed because S is Tx.) Clearly sx e Ux, sx $ U2, s2 e U2, and
s2 Ux. Thus S is Tx.
S is compact. For let {Uw}weW be an open covering of S. Then 00 e UWo for
some vv0 e W. Claim: Uw0 is compact in S. For UWo = K where 0&x is
some subset of the basis 3#$ for the topology on S. Since 00 e UWQ, at least one
element in 3SX is of the form F', where Fis a closed compact subset of 5. Thus
= Pi = F;
Ve£x
that is, Uw0 is a closed subset of the compact set F, hence Uw0 is compact in S,
as claimed.
Since {Uw}weW covers S, {Uw n U„0}weW_{Wo) must be a covering of U^0
by open sets in £/^0. Since U^0 is compact, there exists a finite subcollection
36
{UWl n Uw0,..UWk n Uw0} which covers Uf0. Hence the finite collection
UW0, UWl,..., UWk must cover S. Thus S is compact.
Now assume S is Hausdorff. We must verify that S is Hausdorff and
locally compact. S is certainly Hausdorff, because every subset of a Hausdorff
space is Hausdorff in the relative topology. S is locally compact because, for
each s e S, there exist disjoint open sets Ux and U2 in S such that seUx and
oo e U2 (S is Hausdorff). Now U2 is compact (by the above argument that
shows Uf0 compact), and Ux <= U2 — U2, so Ux is compact by Theorem 4,
Section 1.3.
Conversely, suppose 5 is Hausdorff and locally compact. Let sx, s2 e S. If
both sx and s2 e S, then there exist disjoint open sets Ux and U2 in S such that
Si e Ux and 52 e U2. But Ux and U2 are also open in S, so sx and s2 can be
separated in 5. If sx e S and s2 = oo, then, since S is locally compact, there
exists an open set U in S such that sxe U and U is compact (and closed).
Then V — U' is open in S; and sx e U, s2e V, and U n V = 0. Thus S is
Hausdorff. □
Definition. A topological space S is completely regular if
(1) it is 7\, and

(2) given any s e S and any closed set C with s <£ C, there exists a con¬
tinuous function /: S -> [0, 1] such that f(s) = 0 and /(C) = 1.
Remark. Every normal space is completely regular (by Urysohn’s lemma)

and every completely regular space is regular. The latter is true because if C
is closed in a completely regular space S and s$C, then
Ux =/-mi)) and U2=f~\(b 1])
separate 5 and C, where / is a function such that/(j) = 0 and/(C) = 1.
Theorem 2. Every subset of a completely regular space is completely regular

(in the relative topology).
Proof. Trivially, every subset of a Tx space is Tx. Suppose T <= S. Let C be

closed in T, and t e T, t$C. Then C = T n F for some closed set F in S.
Furthermore, t £ F, for otherwise teTn F = C. Since S is completely
regular, there exists a continuous function /: S [0, 1] such that//) = 0
and/(F) = 1. The restriction of / to T is continuous and has the required
properties. D
Theorem 3. Every locally compact Hausdorff space is completely regular.
Proof. Let S be locally compact Hausdorff. Then the 1-point compactifica-

tion S of S is compact and Hausdorff by Theorem 1, hence normal. In
particular, S is completely regular; hence so is S by Theorem 2. □
37
Remark. We have seen that certain separation properties of a topological

space imply certain others. These implications are gathered together in the
following diagram:
Compact Hausdorff => Locally compact Hausdorff

\ ^
Normal => Completely regular => Regular => Hausdorff => Tx => T0.
/
Metric
By Theorem 2, every subset of a completely regular space is completely

regular. Since an arbitrary product of compact spaces is compact (Tychonoff
theorem) and an arbitrary product of Hausdorff spaces is Hausdorff, we see
that an arbitrary product of closed intervals is compact Hausdorff, hence
completely regular. Thus every subset of an arbitrary product of closed
intervals is completely regular. Theorem 4 below shows that all completely
regular spaces may be regarded as such subsets.
Theorem 4. Every completely regular space is homeomorphic to a subset of a

product of closed intervals.
Proof. We proceed as in the proof of Theorem 2, Section 1.6. Let S be

completely regular. Let W be the set of all continuous functions mapping S
into the closed interval [0, 1]. For each w e W, let Iw = [0, 1], so that {Iw}weW
is a collection of closed intervals. Let P = ELew and let y. SP be
defined by
y(s)(w) = w'(s) (w e W).
We shall show that y is a homeomorphism onto y(S).

It has already been shown in the proof of Theorem 2, Section 1.6, that y is
continuous and injective. (Note that since S is completely regular, W sepa¬
rates points and closed sets. In particular, since points are closed, W separates
points.) Thus we need only verify that y~'L is continuous; that is, that y maps
open sets of S onto open sets of y(S).
Let U be open in S. It suffices to show that for each s e U, there exists an
open set V in y(S) with y(s) e V c y(U). For then y(U) is the union of these
open sets. So suppose s e U. By complete regularity of S, there exists aw e W
such that iv(s) = 0 and w(U') = 1. Then 1)) is open in P, where
ttw:PÎw is projection. Let V = 1)) n y(S). Then V is open in
y(S). y(s) e V because 7Tw(y(s)) — y(s)(w) = w(^) = 0. Moreover, V <= y(U)
because if fe V, then /= y(t) for some t e S, and irJJ)e[0, 1); that is,
1 > ir-wif) = nJMt)) = <p(0(w) = w(t); that is, t e w-1([0, 1)) <= U, and
f=y{t)ey{U). □
38
Remark. Recall that when we first defined a topological space, we were

motivated by properties of more familiar spaces, namely metric spaces.
Theorem 4 shows that under the assumption of complete regularity, we are
back on familiar ground. We understand closed intervals quite thoroughly,
and we have a pretty good feeling for the operation of taking products of
spaces. Now we find that all topological spaces satisfying the axiom of
complete regularity are subsets of products of closed intervals. However, we
are not yet back into the realm of metric spaces, as Theorem 5 will
show.
Definition. A topological space 5 is first countable if, for each point s e S, there
exists a countable collection {t/s(rt)}nG{positlVeintegers> of open sets containing
s such that if U is any open set containing s, then there exists an n such
that Ufn) <= U.
Remark. Let (S, p) be a metric space. Then S is first countable because the
balls {Bs(\/n)} have the required property.
Theorem 5. For each real number r, let Ir = [0, 1]. Let P = Ores1 4- Then
P is not first countable. In particular, P is not metrizable; that is, there
exists no metric p on P such that the metric topology is the product
topology on P.
Proof. Let p e P be such that p(r) = 0 for all r e R1. Suppose there exists a
countable collection {Vn} of open sets containing p such that for each open
set U containing p, Vn c U for some n. We shall show that this assumption
leads to a contradiction, thereby proving that P is not first countable.
Now each Vn is open in P and hence contains a basis element Vn with
p e Vn. In particular, for each open set U containing p, there exists an n such
that Vn <=■ U. But, for each n, there exists a finite subset Fn of R1 such that
(r*FJ
(r e Fn)
and UnJ is, for each r e Fn, an open set in /r. Let C = U“= i Tn. Then C is a
countable union of finite sets, and hence is countable. In particular, C / R1,
so there exists an r0e R1 — C, and TnJo — Iro for each /?; that is, the r0-
coordinate is unrestricted in each Vn. Thus, if U0 g ITo, then U — rireKi Sr,
where
Ir 0 A r0)
Ur o (r = r0)
is open in P, and U contains no V, □

39
2.4 Complete metric spaces
Definitions. Let (S, p) be a metric space. Recall that a sequence {.sn} of points
in S converges if there exists an s e S such that lim*-,* p(sn, s) = 0.
{sn} is a Cauchy sequence if lirn^^ p(sn, sm) = 0. S is called a complete
metric space if every Cauchy sequence converges.
Let (Si, Px) and (S2, p2) be two metric spaces. An isometric embedding of Sx
into S2 is a map f: Sx-+ S2 such that p2(f(s),f(t)) = px(s, t) for all s, t e Sx.
In particular, iff: Sx -> S2 is an isometric embedding, then/maps Sx homeo-
morphically onto f(Sx), where Sx and S2 are provided with the metric topology.
For any metric space (S', p), there exists an isometric embedding/of (S, p) into
a complete metric space, which is called a completion of (S, p) iff(S) is dense.
Remark. Unlike most other concepts we have thus far considered, the
notion of completeness is not a topological concept; that is, it is not invariant
under homeomorphisms. For example, if Sx = (—77-/2, 7t/2) and S2 — R1,
then Sx and S2 are both metric spaces in the usual way. S2 is complete, but Sx
is not. However, there is a homeomorphism/: Sx -> S2 given by f(s) — tan s.
We have discussed compactness for topological spaces in general. When
we restrict our attention to metric spaces, there are several equivalent ways of
saying that a space is compact.
Theorem 1. Let S be a metric space. The following four conditions on S are

equivalent.
(1) S is compact. (Heine-Borel property)
(2) S is countably compact; that is, every infinite subset of S has a limit
point. (Bolzano-Weierstrass property)
(3) S is sequentially compact', that is, every sequence in S has a convergent
subsequence.
(4) S is complete and totally bounded', that is, S is complete, and for every
real number e > 0, there exists a finite number of balls of radius e, say
BSl(e),..., BSm(e), which cover S; that is, S = {J?=1 BSj(e).
Proof
(1) => (2) by Theorem 5, Section 1.3.

(2) => (3). Let {5n} be a sequence in S. If an infinite number of the sn are
equal, then they form a convergent subsequence. On the other hand, suppose
there exist infinitely many distinct elements in {^n}. By (2), this infinite set has
a limit point 50 G S. Consider the ball BSo( 1). Since s0 is a limit point of {^n},
sni e £So(l) for some nx. Next consider Bsf£). Then since s0 is a limit point of
{sn}, sn2 e £So(£) for some n2 > nx. Continuing by induction, we obtain a
subsequence $ni, sn2,... such that sn, e RSol/(2/-1), (j - 1,2,...). This sub¬
sequence converges to s0 because
lim p(snp s0) < lim ^-x = 0.

j —* <x> y-400 Z
40
2.4: Complete metric spaces
(3) => (4). s is complete because if {sn} is a Cauchy sequence, then by (3),
there exists a subsequence {snk} converging to some point s0. In fact, sn
converges to s0 because
kn — •*'o| ^ |jn — Jnj + |sntc — s0| -> 0 as n and nk —> oo.
Moreover, S is totally bounded. For otherwise there exists e > 0 such that no
finite collection of balls of radius e cover S. Let sx e S. Then BSl(e) # S so
there exists s2£BSl(e); that is, P(sx, s2) > e. Now BSl(e) u BS2(e) ^ S so
there exists 53 £ BSl(e) U BS20). Thus P(sm, sn) > e for m # n, {m, n}
c {1,2,3}. Continuing by induction, we obtain a sequence {sn} such that
p(sm, $n) ^ e for all m, n (m ^ n). By (3), there exists a convergent sub¬
sequence {s„}}. Let s0 = limy.,*, snj. Then there exists a j0 such that, for
> jo, sn, e BSo(e/2). For j\,j2 > j0 and .A + j2, this implies that
P(snfi, ^n,2) — p(snjl, Sq) + p(s0, S„j2) < ^ T ^ = £-
But nh ^ /7y2 implies /5(^nn> ^2) > This contradiction establishes (4).
(4) => (1). We shall prove this remaining implication in two steps:
(a) (4) => (3),

(b) (4) + (3) => (1).
(a) Let {5n} be a sequence. By (4), there exists a finite number of balls
Bx i,..., B1<kl of radius 1 that cover S. Thus for some jx, there exists an
infinite set Jx of positive integers such that n e Jx implies sn e B1Jv Let nx be
the first integer such that sni e BX Jl.
Now by (4) again, there exists a finite number of balls B21,..., B2k2 of
radius \ which cover S'. Hence for some j2, there exists an infinite subset J2
of Jx such that nej2 implies sn e B2j2. Let n2 be the first such integer with
n2 > nx.
Continuing by induction, we obtain a sequence {.ynJ such that for all
l > k, sni G BKh (a ball of radius 1 jk). Thus, for /, m > k,
P(.Sni, Snm) ^ 2/At,
and is a Cauchy sequence. Since S is complete, {^nJ converges.

(b) Let 'V be an open covering of S. To prove compactness, it suffices to
show that for some positive integer n, each ball of radius \\n lies in some
V e V. If so, then by total boundedness, there exists a finite number of balls,
say Bx,..., Bk, of radius \/n which cover S. But for each j there exists a
Vj e Y’ with B, <= Vit so S = (J?=i bj c U/=i Vh and VU...,VK is a
finite subcovering of S.
Suppose, for each n, there exists a ball 5Sn(l/«) of radius l/n which is not
contained in any V e "T. We shall show that this assumption leads to a
contradiction, thereby establishing the theorem. The sequence {.?„} of centers
41
of these balls has a convergent subsequence {snfc} by (3). Let Sq = linbc-oo snic.
Then s0 e V0 for some V0eir. Since V0 is open, there exists r > 0 such that
BSg(r) c Vo- Since snic s0, there exists a k0 such that snk e Bsfr/2) for all
k > k0. Let l > k0 be such that 1 /nt < r/2. Then Bs (1/ni) c BSQ(r) c V0.
But this is a contradiction. □
Remark. Condition (4) of Theorem 1 shows that in a complete metric

space, compactness is equivalent to the existence, given any e > 0, of a
finite set of points in S such that each point of S is within e of one of these.
Theorem 2. Let (S, p) be a complete metric space. Suppose { Un} is a countable

collection of open sets each of which is dense in S, that is,
Un = S for n = 1,2,....
Then p|”=i Un / 0.
Proof. We shall construct a sequence which will converge to a point in

Pl”=i Un-Choose sx e Ux. Since Ux is open, there exists a ball BSl(rx) Ux
for some rx. Since sxe S = U2, BSl(rx) n U2 ¥= 0. Choose s2 e BSl(rx) n U2.
Let r2 be such that BS2{r2) c: BSl(rx) n U2 and r2 < min {rx/2, rx - p(sx, s2)}.
BS2(r2) <= Bsfrx) because 5 e Bsfr2) implies
p(s, sx) < p(s, s2) + p(s2, sx) < r2 + p(sx, s2) < rx.
We continue by induction to obtain a sequence {sn} of points in S, and about

each sn, a ball BSn(rn) such that
rn < jSri and Btn^(rn + 1) <= Bsfrn) <= Un.
{.?„} is a Cauchy sequence. For, given e > 0, we can choose n0 such that
rno < e/2. Since sn e BSn(rn) <= BSnQ(rno) for all n > n0.
p(.^mf $n) — ^nQ) T Pfng> $n) ^nQ T ^ £
for all m,n> n0. Since S' is complete, {sn} converges to some point s0.
We must verify that s0 e Un for all n. But, for any n, sn+k e Bs (rn + 1) for
all k > 1. Since s0 — limô, sn+k.
s0eBs (rB + 1) c Bs (rn) <= Un,

n +1 n
as was to be shown. □
Theorem 2, stated somewhat differently, has many applications in analysis.

We now proceed to recast this theorem into its more standard form.
42
2.5: Applications
Definition. Let S' be a topological space. A subset T of S is nowhere dense if T

contains no nonempty open set.
Remark. T <=■ S is nowhere dense if and only if (T)' is dense in S.
Definition. A subset T of a topological space S is of the first category if it is a

countable union of nowhere dense sets. Otherwise, T is said to be of the
second category.
Corollary to Theorem 2 (Baire Category Theorem). A complete metric space

is of the second category; that is, it is not the union of a countable number of
nowhere dense sets.
Proof. Let (S, p) be a complete metric space. Suppose S = \J ®= 1 Tn, where

each Tn is nowhere dense. Then S = Un = i Tn, so, taking complements,
0 = n»-i (Tn)'. But each (Tn)' is open and dense by the above remark. This
contradicts Theorem 2. □
Exercise. Theorem 2 and its corollary are also true if the term “complete
metric” is replaced by “compact Hausdorff.” Can you prove it?
2.5 Applications
Definition. A normed linear space is a linear space (vector space) L, over the
reals or complexes, together with a real-valued function (denoted by || ||)
on L satisfying the following conditions for all vectors a and b and all
scalars A:
(1) ||a|| > 0 and ||a|| = 0oa — 0
(2) ||Ao|| = |A| ||a||
(3) jja + 61 < ||a|| + ||£||
Remark. Let (L, || ||) be a normed linear space. Define a metric on L by
P(a, b) = ||o - b||.
Then (L, p) is a metric space because
(1) p(a, 6) = 0 o ||a — 6|| = 0oa — b = 0oa = b;

(2) p(a, b) = ||a - b\\ = | — 111|« — b\\ = ||6 - a\\ = P(b, a); and
(3) p(a, c) = ||a — c\ — \a — b + b — c|| < ||a - b\ + \\b - c||
= p(a, b) + p(b, c).
Definition. A normed linear space L is a Banach space if (L, p) is a complete

metric space.
Remark. A sequence an in L is Cauchy if
lim p(an, am) = 0, that is, if lim \\an — am|| = 0.
43
Example. Let L = C([0, 1]) be the space of all continuous real-valued

functions on [0, 1], For /e L, define ||/|| = max*e(0>:L1 |/(x)|. Then L is a
Banach space. That || || is a norm is easily verified. That (L, p) is a complete
metric space follows from the theorem that a uniformly convergent sequence
of continuous functions converges to a continuous function.
Remark. More generally, if S is any compact Hausdorff space, the set of all
continuous real-valued functions on S is a Banach space, where ||/|| =
maxseS |/($)|.
Theorem 1. There exists a continuous real-valued function f e C{[0, 1]) such

that f has a derivative at no point of [0, 1].
Proof. For n any positive integer, let
fit + h)~ f[t)

Cn /eC([0, 1]); < n
h
for some t and all h with t -f h e [0, 1 ]
We shall show that Cn is nowhere dense for each n. Since C([0, 1]) is a
complete metric space and hence is of the second category, it will then follow
that
U c" * c([°> *]);

n=1
that is, there exists a function/e C([0, 1]) such that/Cn for any n. This/is
the required function because f $ Cn means that
fit + h) -f(t)
> n
h
for all t e [0, 1] and some h (depending on t and «). Note that for each
fixed t, h 0 as n -> oo because given e > 0 the difference quotient
|(fit + h) -f(t))/h\ is bounded as a function of h for |/z| > e. Thus
fit + h)-fjt)
lim sup = oo.
ft-* o h
and the derivative of/at t fails to exist for each t e [0, 1],
To prove that Cn is nowhere dense, we must show that Cn contains no
nonempty open set. First we show that Cn is closed; that is, that Cn = Cn.
Note that since C([0, 1]) is a metric space, C([0, 1]) is first countable, and
hence any limit point of a set f c C([0, 1]) is in fact a limit of a countable
44
2.5: Applications
subset of T. Thus to show C„ is closed, it suffices to show that if {fk} <= c„ is

a sequence which converges in C([0, 1]), then the limit /e Cn. But fk e Cn
implies that there exists tk e [0, 1] such that
fk{tk + h) - fk(tk) < n

h
for all h. Since [0, 1] is compact, the sequence {4} has a convergent subse¬
quence, which we also denote by {4}. Let 4 = limfc_oo 4. Then
f(t0 + h) — m f(t 0 + h) — f(tk + h) + f(tk + h) — fk(tk + h)
] Mh + /Q - fkih) j fk(tk) - /(4) [ /(4)-/Oo)
/(to + h) — /(4 + A) /(tfc + A)

< +
h h
(D (2)
A(tfc) — /(4)
+ /c(4 + A) - /c(4) +
h h
+ /(4) -/(*«,)
Now fix h. For any e > 0, if k is large enough, (D and (D are smaller than e
because / is continuous and 4-^4; while Q) and @ are smaller than e
because fk converges uniformly to /. Because of the previous paragraph,
<3) < n. Hence
/(to + h) — /(4)
< n + 4e (any e > 0);
h
that is,
/(to + h) -7(4)
< n and /e Cn.
h
Thus each Cn is closed. Now we show that Cn is nowhere dense; that is,
given any g e Cn — Cn and any e > 0, there exists /eC[0, 1] such that
45
II/— g|| < e and f $ Cn. Now a typical example of a function in C[0, 1]
which is not in Cn is the “sawtooth” function (Figure 2.3). For any n we
can find such a function, whose norm is less than or equal to any prescribed
Figure 2.3
e > 0, and where the slope of each line segment is greater than n in absolute
value. To find a function f <£ Cn within £ of g, we need only construct a
sawtooth function close to g, as in Figure 2.4.
We leave this construction to the reader, but include the following hint.
Using the uniform continuity of the function g, we can find a continuous
function gu so that ||g — gi|| < e/2 and gx is piecewise linear; that is, gx
looks like the line segments in Figure 2.5. It suffices to find the appropriate
sawtooth function for each linear piece of g1 and then to patch. □
Figure 2.5
Remark. Theorem 1 above is of considerable interest, but more important

than the statement is the method of proof. This method is used frequently in
analysis and topology to prove the existence of functions with specified
properties.
We exhibit a regular topological space which is not normal as another
application of the Baire category theorem.
46
2.5: Applications
Example. Let S be the upper half-plane in R2 together with the x axis R1.
Let 38x be the usual open sets in the upper half-plane. Let
^2 = [£<r,6)(e) u {r};r e R1, e > 0],
so that an element of 382 is an open disc tangent to R1 together with r the

point of tangency, as in Ligure 2.6. Take u 382 as basis for the topology
on S.
- -
r
Figure 2.6
Note that in the relative topology, Rl is discrete. In fact, for each p e R1,
R1 — {p} is closed in S because its complement is clearly a union of open sets.
Since F = Pjpe^-F (R1 — {/?}) for any subset F c: R\ it follows that every
subset of R1 is closed in S.
S is regular. Suppose Lis closed in S andp $ F. Ifp $ R1, there exists a ball
Bp(e) contained in the open set F'; and the open sets Bp(e/2) and (Bp(e/2)'
separate p and F. If p e R1, there exists a basic open set V containing p and
contained in the open set F'; that is, V n F = 0. Let Vx be a basic open set
containing p such that Vx <= V (Figure 2.7). Clearly each point in F (in fact,
each point in V') is contained in a basic open set disjoint from Vx. Let U be
the union of these. Then Vx and U separate p and F.
S is not normal. Let Fx <= R1 be the rationals and F2 c R1 be the irra¬

tionals. Then Fx and F2 are closed sets in S. Moreover, it is not possible to
find disjoint open sets Ux and U2 in S such that Fx <= Ux and F2 <= U2. For
suppose such sets Ux and U2 did exist. Then for each p e Fx (j = 1, 2), there
exists a basic open set Dv <= U) with p e Dp (Dv is the union of {p} with a disc
tangent at p). UpeFi DP c Ux and (JP6f2 Dp c U2 are also disjoint open sets
containing Fx and F2 respectively. Let/: R1 -> R1 be defined by f(p) = radius
of Dp. (/ is not necessarily continuous.) For each positive integer n, let
T =
1 n pcF2;f(p) > -
41
Then U”=i Fn = F2 becausep e F2 implies f(p) > 1 /n for some n. Moreover,
^ = (u4u(u{4
\
\n= 1 / qeF1 J
so we have expressed the set R1 as a countable union of sets Tn and {q}. Now
consider the usual topology on R1. Then R1 is a complete metric space, and
hence is of the second category. Since we have expressed R1 as a union of
countably many sets, not all of these sets can be nowhere dense in R1 (in the
usual topology). Each {q} is nowhere dense, so Tn must contain an open set
in R1 for some n. In particular, Tn must contain some interval /. Let q e 1 be
rational. Since q e Tn, there exists a sequence {pk} c Tn which converges to q.
But (Ufc=i Dp,c) n Dq <= Ux n U2 = 0. This is impossible since each DPk
has radius > \/n; that is, f(pk) does not tend to zero as k 00 and pk q.
This contradiction completes the proof. □
Theorem 2 (Uniform boundedness principle). Let S be a complete metric

space. Let SF be a family of continuous real-valued functions on S with the
property that for each s e S, there exists a constant Ms such that |/(s)| < Ms
for allfe Then there exists a nonempty open set U ^ S and a constant M
such that |/(s)| < M for all f e IF and all s e U.
Proof. For each f e ZF and each positive integer n, let
Tnj = [seS; |/(j)| < n].
TnJ is closed because TnJ = f~x{[-n, n]). Let
Fn — Pi TnJ = [5 g S; |/(j)1 < n for all f edF\
Tn is closed because it is the intersection of closed sets. U”=i Fn = S because

if s e S, then s e Tn for all n > Ms. By the Baire category theorem, not all Tn
are nowhere dense; that is, some Tn = Tn must contain a nonempty open set
U. Then, for all s e U and f edF, |/($)| < n. Take M = n. □
48
Fundamental group and
covering spaces
3.1 Homotopy
Notation. The letter / will henceforth denote the closed interval [0, 1]. A map
is a function; the two words will be used interchangeably.
Definition. Let X and Y be topological spaces, and let/0 and f be continuous

maps X -> Y.f0 is homotopic to fx (written f0 ~ f) if there exists a con¬
tinuous map F: X x I -> Y such that F(x, 0) = /0(x) and F(x, 1) = /i(x)
for all x e X. The map F is called a homotopy from f0 to f1.
Remark. For (x, t) e X x I, we may regard t as measuring time. Then

f(x) = F(x, t) is a 1-parameter family of maps X -> Y. At time 0 we have the
map/0. At time 1 we have the map f. As time increases from 0 to 1, the map
/0 is deformed continuously into the map f1.
Example. Let i: Rn -» Rn be the identity map. Let c be defined by

c(v) = 0 for all v e Rn. Then c ~ i. For let F: Rn x / -> Rn be defined by
F(v, t) = tv. Then Fis continuous, and F{v, 0) = 0 = c(v) and F{v, 1) = v =
i(v) for all v e Rn. Geometrically, the map F shrinks the image Rn of i to the
point {0} as t varies from 1 to 0.
Theorem 1. Homotopy is an equivalence relation; that is, for f g, h continuous

maps X -> Y,
0 )/-/,
(2) / ~ g implies g — f, and
(3) f ~ g and g ~ h implies f ~ h.
49
3: Fundamental group and covering spaces
Proof
(1) Let F: X x I-> Y be defined by F(x, t) = /(x). F is continuous

because it is the composition of the continuous maps / and projection onto
the first factor.
(2) Given a homotopy F: X x / -> Y such that F(x, 0) = /(x) and
F(x, 1) = g(x), let G: X x IY be defined by G(x, t) = F(x, 1 - t). G is
continuous because t -> 1 — t is continuous. G is a homotopy from g to /.
(3) Given homotopies F, G: X x / -> Y, with F(x, 0) = /(x), F(x, 1) =
g(x) = G(x, 0), and G{x, 1) = h{x), let H: X x /-> Y be defined by
(F{x, 2t) (0 < t < i)

H(x, t)
\G(x,2t- 1) (i< t< 1).
Then H(x, 0) = /(x), and H(x, 1) = h{x). H is continuous by the following

lemma. □
Glueing (or pasting) lemma. Let X and Y be topological spaces. Assume

X = A U B, where A and B are closed {open) sets in X. Suppose/j: A —x Y
and f2: B—> Y are continuous functions such that /i(x) = /2(x) for all
x e A n B. Let g: X -> Y be defined by
(x e A)
(x g B).
Then g is continuous.
Remark. The function g of the lemma is thus obtained by “glueing”/!

and /2 together along their common domain.
Proof of the Glueing Lemma. Assume A and B are both closed. We show
that inverse images of closed sets are closed. Let F be a closed subset of Y.
Then
g~\F) = g~\F)r\ {A v B)
= {g~\F) n A) u {g~\F) n B)
= f,-\F)yjf2~\F).
Since fi is continuous, /i-1(F) is closed in A, and hence in X because A is

closed in X. Similarly, f2~1{F) is closed in X. Moreover, the union of two
closed sets is closed.
If both A and B are open, the same argument shows that inverse images of
open sets are open. □
Theorem 2. Let X, Y, Z be topological spaces. Suppose that f0 and f are

homotopic maps X -> Y and that g0 and gx are homotopic maps Y^-Z.
Then g0 ° f0 and gx ° fi are homotopic maps X —>- Z.
50
3.1: Homotopy
Proof. We break the proof into two steps:
(a) go o/o ~ g0 °A
(b) go °/i - gi °/i-
Then g0 ° /o — gi °/i by (3) of Theorem 1.

(a) Let F: X x /-> 7 be a homotopy from /0 to /i. Let G = g0° F:
X x /-» Z. Then G is a homotopy from g0 ° f0 to g0 o fx.
(b) Let 77: Y x I^Z be a homotopy from g0 to gx. Let f1: X x /->
7 x / be defined by/^x, t) = (ffx), t). f± is easily seen to be continuous.
Let
K = Hofi: X x 7—>Z.
Then K is a homotopy from g0 ° f\ to gx ° fx. □
Definition. Two spaces X and Y are of the same homotopy type if there exist
continuous maps f:X^Y and g: Y—> X such that g°/~ ix and
/ ° g ~ iY, where ix and iY are the identity maps on X and Y respectively.
Remark. It is easy to verify that “same homotopy type” is an equivalence

relation. Thus, the collection of all topological spaces is partitioned into
equivalence classes. Two spaces are in the same class if and only if they are of
the same homotopy type. Clearly homeomorphic spaces are of the same homo¬
topy type. Much of algebraic topology is concerned with the study of those
properties of topological spaces which are invariants of homotopy type, that
is, those properties which, when possessed by one topological space X, are
possessed by every topological space of the same homotopy type as X.
Definition. A topological space X is contractible if the identity map ix: X X

is homotopic to a constant map; that is, if ix ~ c, where c: {x0} for
some x0 e X.
Theorem 3. A space X is contractible if and only if X is of the same homotopy

type as a single point.
Proof. Suppose X is contractible. Then ix ~ c for some c: X -> {x0} (x0 e X).
Let / = c: Z->{x0} and let g: {x0} -> X be defined by g(x0) = x0. Then
clearly g °f = c ~ ix, and /° g = ilXo) ~ i{Xo), so X and {x0} have the same
homotopy type.
Conversely, suppose Y — {y} consists of a single point, and X has the
same homotopy type as Y. Then there exist continuous maps f:X^Y and
g:Y->X such that g°/~/x and /° g ~ iY. Let x0 = g(y), and let
c: X ^ {jt0}. Then c = g °/~ ix, so X is contractible. □
Remark. Both Theorem 3 and the example at the beginning of this section
show that Rn has the same homotopy type as a single point. Thus from the
viewpoint of homotopy theory, Rn is a trivial space.
51
3.2 Fundamental group

Definition. Let A" be a topological space. A path in X from x0 to xx (with
origin x0 and end Xi) is a continuous map a: I X such that a(0) = x0
and a(l) = xx.
Remark. Note that a path is a function and not a set of points. A path is a
“parameterized curve.”
Definition. A space X is arcwise connected if, given any two points x0 and xx in
X, there exists a path with origin x0 and end xx.
Theorem 1. If a topological space is arcwise connected, then it is connected.
Proof. Let Abe arcwise connected. Suppose U0 and U1 are nonempty disjoint
open sets with t/0 u Ux = X. Let x0 e U0 and xx e U1. Since X is arcwise
connected, there exists a path a from x0 to xx. Clearly a^1(U0) and a_1(C/1)
are disjoint open sets in /with a~1(U0) u a-1(t/1) = I. Moreover, 0 e a~1(U0)
and 1 so these sets are nonempty. But / is connected, so this is
impossible. □
Remark. The converse of Theorem 1 is false, as is shown by the following

example.
Example. Let X = A u B <=■ R2 with the relative topology, where
A = [(0, y) e R2; \y\ < 1],

B = [(x, sin 1/x) e R2; 0 < x < 1],
X is connected. For let U1 and U2 be disjoint open sets in X with

U1U U2 = X. The point (0, 1) is in one of these two sets, say (0, 1) g Ux.
We shall show that U2 = 0, and hence X is connected.
First, (0, 1) g Ux n A, so U1 n A ^ 0. Since A is connected, and
A = (t/i n A) u (U2 n A),
it follows that U2 n A = 0. Next, consider Ux n B. Since any ball in R2

about (0, 1) must contain points of the form (x, sin 1/x), Uxr\ B ^ 0. But
B is connected—it is the continuous image of the connected set (0, 1]—and
B = (Uxn B) u (U2 n B). Thus U2 n B = 0. Hence U2 = (U2 n A) u
(U2nB) = 0.
X is not arcwise connected. For let a be a path in X with origin (0, 1). We
shall show that a(I) c A, and hence no point in B can be joined to (0, 1) by a
path in X. Consider a~\A). Since A is closed in X, a~X(A) is closed in /
a~\A) # 0 because 0 g a X(A). Hence it suffices to show that a~1(A) is open
52
3.2: Fundamental group
in /, for then a~1(A) = I because I is connected. Suppose t0 e a~1(A). Then

a(t0) e A. Let U = X n Ba(to){\). Then U is open in X. Since a is continuous,
there exists an e > 0 such that a(t) e U whenever |/ - r0| < e. Claim:
a((t0 - e, t0 + e)) c: A. For suppose |c - t0| < e and a(tx) e B. Now
U n B is a union of disjoint arcs (homeomorphic images of open intervals),
and the arc containing a(C) is both open and closed in U. This contradicts the
connectedness of «((t0 - e, t0 + e)). Thus (t0 - e, t0 + e) c a-Â) and
a_1(y4) is open in I (see Figure 3.1).
Figure 3.1
Definition. Let a be a path from x0 to xx and let p be a path from x1 to jc2. The
product of a and p is the path ap from x0 to x2 defined by
(0 < / < i)
(i<t< 1).
The inverse of a is the path a 1 from xx to x0 defined by a 1(t) = a(l — t).
Remark. That the product of two paths is continuous is a consequence of

the “Glueing lemma” (Section 3.1).
Definition. Two paths a and p from x0 to xx are homotopic (written a ~ P) if

there exists a continuous map F: I x /-> X such that
F(0, t2) = x0 and F(l, t2) = xx for all t2 e /;

F(tu 0) = a(tx) and F(tx, 1) = P{tx) for all tx e I.
(See Figure 3.2).
Remark. Thus a homotopy of paths is a homotopy in the usual sense

together with the additional requirement that the end points remain fixed
53
throughout the homotopy. Note that without this additional requirement

every path is homotopic to a constant path.
Remark. The relation ~ is an equivalence relation. The proof is similar to
the corresponding proof for ~ (see Theorem 1).
Theorem 2. Suppose a0 ~ and /S0 ~ /3X are paths such that a0l80 is defined.
Then is defined and ~ a0/?0.
Proof. is defined because
end of ax = end of a0 = origin of yS0 = origin of j8x.
Let F and G be homotopies from a0 to ax and from /30 to j8x respectively. Then
a homotopy H from a0/30 to is given by
tf) (0 <h< i)
H(t\, tf)
\G(2h - 1, t2) (i < h < 1).
H is continuous by the “glueing lemma.” □
Theorem 3. Suppose a0 and ax are homotopic paths. Then a0“1 ~ ai_1.
Proof. Let F be a homotopy from a0 to ax. A homotopy H from a0 ~1 to ax -1

is given by
H{tlt t2) = F(1 - h, t2). □
Notation. Let <a> denote the ~ equivalence class of a; that is, <a> is the
set of all paths homotopic to a. Since homotopic paths have the same end
points, the origin and end of <a> are defined.
54
Definition. The product and inverse of ~ equivalence classes are defined by
<a><i8> = <«/?> (if oc/3 is defined),

<«>_1 = <a_1>.
These are well defined by Theorems 2 and 3 above.
Theorem 4. For each x e X, let ex: I X be the path defined by ex(t) = x for
all tel. Then:
(1) //<«> has origin x0, then <eXo><a> = <a>.

(2) //<a) has end xu then = <«>.
(3) //<a) has origin x0 and end xx, then
<a)<a~ 1> = (eXQ) and <a-1><«> = <eXl>.
(4) «a><iS»<y> = <a>«/3><y» (i/(aj8)y is defined).
Corollary. Le/ X be a topological space and let x0 e X. The set of ~ equivalence

classes of paths with origin = end = x0 forms a group under the operations
of multiplication and inverse as defined above. This group is denoted by
TrfiX, jc0) and is called the fundamental group, or first homotopy group,
of the pair (X, x0).
Proof of Theorem 4
(1) We must show that eXoa ~ a. Thus we want to construct a continuous
mapping F: I x / -> X such that
-^l/xfO} = Ccoa>
F |/x{i> = a,
^1(0,x/ — Xo,
-f’lajx/ = = “(!)•
55
We do this by defining Fon the triangle in Figure 3.3 to be the constant map
into jc0, and by requiring that F, restricted to each horizontal line segment
in the trapezoid of the figure, be equal to a (after suitably parameterizing the
line segment). Explicitly,
x0 (2 ti < 1 — t2\
F(ti, t2) 2tx — 1 + t2\
(l — t2 < 2tx).
1 + t2 /
(2) We must show that cceXl ~ a. The proof is similar to the proof of (1)
(see Figure 3.4).
a
Figure 3.4
(3) It suffices to prove that (aXa'1) = <e*0>, f°r then we may interchange
the roles of a and a-1. Thus we must show that aa_1 ~ eXo. We do this by
“pulling the end point in to x0 along the path a” (see Figure 3.5).
Exercise. Find the analytic expression for this homotopy.
(4) We must show that (af3)y ~ «(/8y). Now
f«(40 (0 < t < i),

((«f3)(y))(0 = \ft4t - 1) (i < t < i),
(y(2t - 1) (i < t < 1);
a(21) (0 < t < i),

W y)X0 = y8(4/ - 2) (i < t < |),
y(4t - 3) (i < t < l).
56
a 0 y
a 13 y
Figure 3.6
The homotopy is (see Figure 3.6)
(4/i - 1 < fa),
12) — \ £(4fx — t2 — 1) (4/i - 2 < /2 < 4/,. - 1),
(fa < 4fx - 2). □
Remark. Given two base points x0 and xx in X, we cannot in general expect

any relationship between 771(T, x0) and tt^X, xx). For example, if x0 and xx
do not lie in a common connected subset of X, there can be no relationship.
Consider the disjoint union X of a circle and a point xx (Figure 3.7). For x0
in the circle, ■n1{X, x0) has nontrivial elements, whereas 7r1(Z, jq) is clearly
trivial.
Similarly, if X is not arcwise connected, we can expect no relationship

between these groups. Consider the space X = Xx u X2 u X3 <= R2, where
X1 is the graph of sin \/x (0 < x < 1), X2 = {0} x I, and X3 is a circle
tangent to the y axis at (0, 1) with center to the left of the y axis (Figure 3.8).
For x0 e Xx, tt^X, x0) is trivial. But for xx = (0, 1), n^X, xx) has nontrivial
elements.
For arcwise connected spaces, the situation is better.
57
Figure 3.8
Theorem 5. Let X be an arcwise connected topological space. Let x0, x1 e X.

Then there exists a group isomorphism of rrfX, x0) onto -rrfX, x^.
Proof. Let y be a path from x0 to xx. Let yf. ttx(X, x0) —> ttx(X, Xi) be defined
by
y#(<«» = <y>_1<«)<y> = <y_1«y> for <«> e ttx(X, x0).
y# is a homomorphism because
y#«“»y#«£» = <y>~1<«><y><y>_1<^><y>
= <y>-1<«><ô><^><y>
= <y> -1<a)<^)<y>
= y#««)<j8»-
y# is an isomorphism because it has an inverse, namely, y#_1 = (y-1)#- □
Remark. Recall that for a group G and a e G, the inner automorphism of G

due to a is the isomorphism ia of G onto itself given by ifb) = aba~1 for
b e G. Now, given two paths in X from x0 to xl9 we obtain two isomorphisms
?r1( X, x0) -> tti(X, Xj). The above proof actually shows that these isomorphisms
differ by an inner automorphism of ttx(X, x0).
Corollary. Let yx and y2 be two paths in X from x0 to xx. Then
(ya)# = (yi)#0 ia,
where ia is the inner automorphism of vx(X, x0) due to a — <yiy2_1) e

TTfX, x0).
58
Proof. For <«> e tt^X, x0),
(yi)#-1 ° (y2)#(<«» = <yi><y2>_1<a><y2><yi>_1

= (y^^XaXy^-1)-1
= 4<«>;
that is, (yO#-1 ° (y2)# = ia, or (y2)# = (y^ ° ia. □
Remark. Theorem 5 shows that all fundamental groups of an arcwise

connected space X are isomorphic; that is, associated with the space X is a
certain abstract group, the fundamental group of X. However, the above
corollary shows that unless this group is commutative, no natural way exists
of identifying the groups arising from different base points. (The isomorphism
depends on the homotopy class of the path joining the base points.) Hence we
shall regard the fundamental group as being a concrete group, computed with
respect to a given base point. The importance of the base point will become
clearer as we study the behavior of the fundamental group relative to con¬
tinuous maps.
Definition. Let X and Y be arcwise connected spaces. Let /: X -> Y be con¬

tinuous. For x0 g X, let /*: -n-fX, x0) -> 7t-1(7,/(x0)) be defined by
/*(<«>) - </° «>

for <ce> g tti(X, x0). Note that this definition makes sense because if a and
are two paths in X—both beginning and ending at x0 with a ~ j8—then
f o a ~ fo p. (A homotopy from f ° a to f ° /3 is given by f ° F where F is a
homotopy from a to /3.)
Remark./*: 7rfX, x0) -> tt1( Y, f(x0)) is a homomorphism, because for <«>
and <£> e 7t^X, x0),
/*««><!s» = /*(<«£>) = </° («/*)>

= <(/°«)(/°£)> = </°«X/°0>
=/*(<«>)/*(</*>).
/* is called the homomorphism induced by f.
Theorem 6. Let A", 7, Z arcwise connected, and let x0 e Z. TTzefl:

(1) Iff: X^ Y andg: F-^Z are continuous, then (g°/)* = g* °/*-
(2) Iff0 and_/j: X—> Y are homotopic maps and F: X x I —> Y is a homo¬
topy from f0 to f, then if)* = <j# o (/0)*, wAere o- « the path in Y from
fo(Xo) toffxo) given by
a(t) = F(x 0, 0-
59
Remark. Part (2) of Theorem 6 says that homotopic maps induce the same
homomorphism on fundamental groups, up to an inner automorphism that
compensates for the fact that the two maps may send the base point in X into
different points in Y.
Proof of Theorem 6
(1) Let <a> e x0). Then
(g °/)*(<«» = (g °/° “> = g*K /0 «>) = g* °/*(<«»•
(2) Let <a) e -n^X, ;c0). We want to show that
(/i)*<«> = <¥((/o)*<a»; that is> -
Thus we must show that
, fi ° « - o 1(fo ° «)CT-
For this, consider the map G : I x /-> Y defined by
G(*i, t2) = F(a(G), t2) (tu t2 e I).
G is continuous because F and a are. Moreover,
G(h, 0) = ^(«(L), 0) = /o(a(ti)) =fo° a(A),

G(h, 1) = 1) = fMh)) = fi ° <ti),
G(0, t2) = F(a(0), f2) = F{xo, t2) = cr(/2),
G(l, /a) = F(a(l), t2) = F(x0, t2) = o(t2).
Thus the boundary of / x / is mapped by G as indicated in Figure 3.9.
fi ° a fi ° a
The required homotopy H from fx ° a to <r _1(/0 o a)a is then obtained by

deforming G as indicated in Figure 3.10.
60
fi 0 a
Analytically, H is given by
F 2?o — 2 + 3\
A) — ‘ 5 t: < h <
312 F 1 4 j’
a(4h - 3) > □
4 /•
Corollary 1. If X and Y are arcwise connected spaces of the same homotopy

type, then their fundamental groups are isomorphic.
Proof. Since X and Y are of the same homotopy type, there exist maps
/: X-» Y and g: Y-> X such that g ° f ~ ix and f ° g — iY. Choose a base
point x0e X in the image of g, say x0 = g(y0)- We shall show that
f*:"i(X,X0 )~>rri(Y,f(x0))
is an isomorphism. Now,
g* : rrf Y,f(x0)) -> 77,(A, g o/(x0))
and, by Parts (1) and (2) of Theorem 6,
g* °f* = (g °/)* = ° (ix)* =
where of. -nfX, xr0) -» -ofX, g ° f(x0)) is an isomorphism. Thus /* is injective.

On the other hand, we may consider the homomorphism, also denoted by
g*, mapping 77,( Y, y0) -> rrfX, x0). Then
/* ° g* = (f° g)* = (°i)# ° Or)* = (CTi)#
where (<?,)#: ^(T, y0) -> 77,( Y,f° gO0)) is an isomorphism. Thus/* is
surjective also, hence an isomorphism. □
61
Corollary 2. If X is contractible, then -n-fX, x0) = (e); that is, the fundamental
group of X consists of the identity element only.
Proof. By definition, X contractible means that X is of the same homotopy

type as a one-point space. Thus Corollary 1 implies the result. □
Corollary 3. irfR71, 0) = (<?).
3.3 Covering spaces

All spaces throughout this section are Hausdorff topological spaces.
Definition. A space X is locally connected if for each point x e X and each

open set V containing x, there exists a connected open set U such that
x e U <= V. A space X is locally arcwise connected if for each point x e X
and each open set V containing x, there exists an open set U, with xei/c
V, such that whenever xl5 x2 e U, there exists a path a from xx to x2 with
a(7) c V.
Remark. Not every arcwise connected space is locally arcwise connected.

For let Abe the union of the graph of sin 1/x, x e (0, 1], with an arc connect¬
ing (1,0) and (0, 1) (Figure 3.11). Then X is arcwise connected, but if B is any
ball in R2 about (0, 1) of radius < 1, then V = B n X is open in X, yet V
contains no open set U with the property required for local arcwise
connectedness.
Figure 3.11
Definition. Let X and X be arcwise connected, locally arcwise connected

spaces, and let p: X X be continuous. The pair (X, p) is called a covering
space of X if
62
3.3: Covering spaces
(1) p is surjective, and

(2) for each x e X, there exists an open set U in X containing x such that
p~l(U) is a disjoint union of open sets, each of which is mapped
homeomorphically onto U by p. Such an open set U will be called
admissible.
Example 1. Let
X = S1 — [z; z a complex number with \z\ = 1].
Let X = R1, and let p: X—X be given by p(r) = e2nir. Then (X, p) is a
covering space of X. For x e X, let U be a small open arc containing x. Then
p~\U) is a disjoint union of open intervals, each a translate of every other
by some integer (Figure 3.12). X may be looked at as an infinite spiral over
S1, with p being ordinary projection (Figure 3.13).
Figure 3.12
o
Figure 3.13
Example 2. Let
X — P2 — the set of all lines through the origin in R3.
The topology on P2 is generated by “open cones” of lines through the origin

(Figure 3.14). P2 is called the real projective plane. Let X = S2 R3, and
let p: X-> X be defined by
p(x) = the line through the origin which passes through x (x e X).
Then (X, p) is covering space of X, in fact, a “double covering” of X; that is

p-1(x) consists of two points for each x e X.
63
Figure 3.14
Example 3. Let X = X = S1 = [z; \z\ = 1]. Let p: X~^ X be given by

p(z) = z2. Then (X, p)-is a double covering of X.
Example 4. Let I=S1xS1- the torus. Let X = R2 and p: X-> X be

defined by
p(ri, r2) = (e2niri, e2nir2).
Then (X, p) is a covering space of X. For a “patch” U on X, p~1(U) is a

union of “patches” in R2, each a translate of every other by a vector with
integer coordinates (Figure 3.15). (Note that unit square gets mapped onto
S1 x S1 by p, and, in fact, S1 x S1 may be regarded as a closed unit square
with one pair of opposite edges identified to form a cylinder and the other
pair of opposite edges (now circles) identified to form a tire.)
Figure 3.15
64
Example 5 (A “nonexample”). Let A = S1, and let Abe a finite open spiral
over S1 with projection p: X —> X (Figure 3.16). (A, p) is not a covering space
of A because if x lies under an “end” of A, then there exists no open set U
about x satisfying (2) of the definition.
iO
Figure 3.16
Example 6 (Another “nonexample”). Let A = S2 and A = an infinite

cylinder circumscribed about S2; and let p be radial projection (Figure 3.17).
Then (A, p) is not a covering space because p is not surjective.
.
Theorem 1 Let (A, p) be a covering space of A. Then p: X -> A is an open
mapping; that is, p{U) is open in X for each open set U <= A.
Proof. Let U be open in A. For xep(U), let V be an admissible open set

containing x, so that p_1(F) is a disjoint union of open sets, each mapped
homeomorphically onto V by p. Let x e U be such that p(x) = x. Then x lies
in one of these open sets mapped homeomorphically onto V. Call it W. Then
W n t/isopenin W, so p(Wn U) is open in Fsinceplwisahomeomorphism.
But V is open in A, so p(W n U) is an open set in A with x e p( W n U) «=
p{U). Since xep(U) was arbitrary, p(U) is a union of open sets, and hence is
open. □
Remark. Given two spaces Y and A, a continuous map /: Y -> A, and a

covering space (A, p) of A, we find it often of interest to know whether we
can “lift” the map/to a map/: T-» A; that is, whether we can construct a
65
map/: X such that p°f = f The map/is also said to coverf. The next
theorem (Theorem 2) shows that under certain mild conditions, any two such
maps (if they exist) either agree everywhere or agree nowhere. The following
theorem (Theorem 3) and its corollaries tell us that certain maps can, in fact,
be “lifted.”
Theorem 2. Let (X, p) be a covering space of a space X, and let Y be a connected

and locally connected space. Suppose a, ft: Y -> X are continuous maps such
that
(1) p ° a = p o j3, and

(2) a(y0) = /3(Fo) for some y0 e Y.
Then a = /3.
Proof. Let
Z = [ye Y; «(y) = p(y)].
We must show that ’Z = Y. Since Y is connected, and Z ^ 0 (y0 eZ),

it suffices to show that Z is both open and closed.
Z is closed: For consider the map a x ft: Y -> X x X defined by
(« * P)(y) = («0)> P(y))-
Then a x ft is continuous. Let D be the diagonal in X x X; that is,
D = [(x, i);xe X],
D is closed because X is Hausdorff. Hence Z = (a x /5)-1(Z>) is closed in Y.

(Note that this argument shows that for any two continuous maps from any
topological space S into any Hausdorff space T, the set of points where the
two maps agree will be closed in S.)
Z is open: Suppose z e Z. Let x = p ° a(z) = p ° |8(z). Let U be an ad¬
missible open set about x. Then p~l{U) is a union of disjoint open sets one of
which, call it W, contains a(z) = /3(z). Now IF is both open and closed in
p_1(£/). Hence any connected subset of p_1(t/) which has points in Wmust
lie entirely in W. Now z e(p ° a)_1(C/), an open set in Y. Since Y is locally
connected, there exists a connected open set V± in Twithze Vx c (p ° a)_1((7).
Since a is continuous, ce(F1) is connected in /?~ 1(C7) with a(z)eJF; so
aiVi) c W. Similarly, there exists a connected open set V2 in Y containing z,
such that |8(F2) c W. Since p\w is injective, and since p ° a = p ° /?, it follows
that a — j8 on Vx n V2. Thus V1 n V2 is contained in Z and is an open set
containing z. Thus Z is open. □
Theorem 3 (Covering homotopy theorem). Let (X, p) be a covering space of a

space X, and let Y be a compact, connected, and locally connected space. Let
66
f: Y ^ X, and let F: Y x / —>• X be a homotopy with F(y, 0) = p °f{y)

for y e Y. Then there exists a homotopy G: Y x I —> X such that
(1) G(y, 0) = /O), and

(2) p o G = F.
Moreover, G may be chosen to be stationary; that is, whenever y e Y is such

that F(y, t) is constant for t in some interval, then G(y, t) is also constant for
t in that interval.
Remark. Before proving Theorem 3, we derive several consequences.
Corollary 1. Let (X, p) be a covering space of X, and let x e X. Then

p*'. TrfX, x) -> rrfX, p(x)) is injective.
Proof. Suppose <a> e vfX, x) is in the kernel of p*. Then
(P ° «> = />*(<«» = e = <e*>; that is, p ° & ~ ex,
where x = p(x). We want to show that <«> is the identity element in nfX, x),
that is, that a ~ e .
Let F: I x I X be a homotopy from p o « to ex. By the covering homo¬
topy theorem, with Y = I, there exists a homotopy G: I x I~^X with
pa G = F such that G(tu 0) = a^) for all t1 e I. Furthermore, since
F(0, t2) = x and F(l, t2) = x are constant for t2e I, G may be chosen so that
(7(0, t2) = «(0) = x and (7(1, t2) = fi(l) = x
for all t2 e /. Sincep ° G = Fand since F(tx, 1) = ex(tx), the paths 4-^ (7(^,1)
and tx -> e^fx) are both mapped by p into ex. Moreover, G(0, 1) = x = e*(0),
so these paths agree at a point. Thus, by Theorem 2, G(c, 1) = c5(g) for all
tx e I, and G is a homotopy from a to ex. □
Corollary 2. Let (X, p) be a covering space of X. Let a be a path in X. Let

x0 = a(0), and choose x0 e X with p(x0) = x0. Then there exists a unique
path a in X, with origin x0, covering a, that is, with p ° a = a.
Proof. The uniqueness is a consequence of Theorem 2 above. To prove
existence, we apply the covering homotopy theorem to the case where
Y = {To} is a 1-point space and /: T-> X is defined by f{y0) = x0. Since
projection on / is a homeomorphism of Y x / onto /, we may identify these
two spaces through this homeomorphism and regard a as mapping Y x I-+X.
Then a(y0, 0) = x0 — p ° f(y0), so by the covering homotopy theorem, there
exists a homotopy G : Y x /-* X such that p ° G — a and
G(y0, 0) = f(y0) = x0.
Let a: I^ X be defined by a(t) = G(y0, t). Then a has the required proper¬
ties. □
67
Corollary 3. Let (X, p) be a covering space of X. Let x e X, and x e X with

p(x) = x. Then there exists a “natural” one-to-one correspondence between
p_1({x}) and the coset space rrfX, x)/p^fX, x).
Proof. We define c: 7tx(X, x) —> /?_1({x}) as follows. Letg e nfX, x). Suppose
a and p are two representations of g; that is, g = <ct> = <)8>. By Corollary 2,
a and p have unique lifts a and p to X, with origin x. Now a ± p, so by the
covering homotopy theorem, & ± p. In particular, a and P have the same end
point; that is, d(l) = /3(1). Thus we can define
c(g) = d(l),
where a is any representative of g.

Since X is arcwise connected, x can be joined to each element of tt~ Ktt)
by a path in X, which is in turn the lift of a closed path (its projection) in X.
Thus c: -nfX, x) -> p_1(W) is surjective.
We now show that c is constant on cosets of p^fX, x) in nfX, x).
Suppose <a> and </?> lie in the same coset. Then <)8> = <y><a> = <ya> for
some <y> e p^fX, x). Thus
c«£» = C«ya» = ya(l).
But ya = yd by Corollary 2, so that
= y«(l) = a(l) = c«a»,
and the restriction of c to each coset is constant, as claimed.

Thus c defines a map c: irfX, x^p^nfjt, x) ->p_1({x}) by
c(H <«» = C«a», <a> 6 irfx, x),
where //<a> is the coset of H = p^-n-fX, x) containing <a>. c is surjective

because c is. c is injective because if c(//<«» = c(H(py), then c«a» = c«j8»;
that is, d(l) = P(l), so that <(d/J_1> evfX, x). Letting h = p*«dj8_1», we
have
kp> = ,

so that <a> and <j8> are in the same coset of H; that is, f/<a> = □
Example 1. Consider the line R1 as a covering space for the circle S1, with
p(r) = e2nir. Take as base point in S1 the point z = 1. Thenp_1({l}) is the set
of integers. Since nfR1, 0) is trivial, p^fR1, 0) = (e), so n^S1, 1) is in
one-to-one correspondence with the integers. (We shall see later than this
correspondence is, in fact, a group isomorphism.)
68
Example 2. Consider the plane R2 as a covering space for the torus S1 x S1,
with p(ru r2) = (e2niTi, e2niT2). Then ttx(R2, 0) is trivial so x S'1, (1, 1))
is in one-to-one correspondence withp~\( 1, 1)}, the Cartesian product of the
set of integers with itself. (This correspondence also turns out to be a group
isomorphism, of •n-1(5'1 x S1, (1, 1)) with the direct product of the integers
with itself.)
Example 3. The sphere S2 is a covering space of the projective plane P2 (see

Example 2 at the beginning of this section). Let neP2 be the z axis (n is a line
through the origin in R2 and hence is a point in P2). Then
p~\{n}) = (north pole, south pole} <= S2.
Later we shall prove that ttx(S2, north pole) = {e}. Assuming this fact for the
moment, we get from Corollary 3 that rrx{P2, n) has two elements; that is,
rrx{P2, n) = Z2, the cyclic group of order 2. In particular, there is a path a
in P2 which is not homotopic to a constant, but its square a2 is homotopic to
a constant. The curve in P2 defined by a thus has the property that the path
obtained by traveling around once cannot be shrunk to a point, and yet the
path obtained by traveling around twice can be shrunk to a point. Geo¬
metrically, given a great circle on S2 through the north and south poles, let
x(t) e S2 move along this great circle from the north to the south pole as t
varies from 0 to 1. Let a(t) be the line through the origin which passes through
x(t). Then a is a path in P2 with the above property. Note that the lift t -> x{t)
of a to S2 is not a closed path in S2, whereas the lift of a2 to S2 is closed. This
amounts to traveling all the way around the great circle.
Proof of Theorem 3. Since Yand /are compact, Y x /is compact, hence so

is F(Y x I). Thus F(Y x I) can be covered by finitely many admissible open
sets Uu ..., Ur (cover F{Y x I) by admissible open sets and take a finite
subcovering). Since {F~1(Ul)}rl = 1 covers Y x /, and since a basis of open sets
in 7 x 7 is given by (open sets of Y) x (open intervals in I), we can find a
finite covering {Va} of Y of connected, open sets and a decomposition of the
unit interval 0 = t0 < tx < • • • < tk = 1 such that each F(Va x [tu h + 1]) <=
some i/j.
We construct G by constructing Gt: Y x [fi-n, fi] -»■ X{i = 1,..., k) with
the properties that (1) p ° Gt = F|yxBi_lft(]; (2) Gt is continuous; and (3)
Gi — Gj+i on the closed set
Y x [#*_!, U] n Y x [tu tl+1\ - [(y, tt);y e Y].
By the glueing lemma for closed sets, G will be continuous, and, of course,
p o G = F.
By induction, assume the G{ have been defined for / < j so that they satisfy
(1), (2) (for i = 1 and (3) (for 7 = 11). We construct Gj+1.
69
In order for (3) to be satisfied, Gj+1(y, tj) must equal Gj(y, tj). Let
G“+1: Va x [t,; tj+1] -> X be defined as follows. F(Va x [tu tj+1\) <= Uh an
admissible open set in X. Since Va is connected, the set G;(Fa x {t,}) is
connected. But
p o Gj(Va x {tj}) — F(Va x {tj}) c Uh
Hence, Gj{Va x {ty}) lies in one of the open sets W <= p~1(Ui) on which
p:W->Ul is a homeomorphism. Define G“+1 = (p|w)-10 F\v0Lx[tj,t, + 1-i-
Note that Gf+1 = Gj on Va x {tj}. To construct the continuous map Gj+1
on the topological space
Y x \tj, tj+1],
we paste the maps G“+1 defined on the open sets Va x \th tj+1\ of the space
Y x [th tj+x\. By the glueing lemma for open sets, we need only verify that
G“+i and Gj+1 agree on (Va n VB) x [t}, tj+1], which we can assume is not
empty.
Suppose Gj(Vy x {tj}) lies in the open set Wy (y = a, ft), with p: Wa —> Ux
and p: WB-> Um homeomorphisms. Since G]+1 = Gj on Vy x {tj} (y = a, j8),
then G“+ = GBj + x on (Fa n F5) x {fy}. Since any point of (Fa n F^) x [th tj+1\
j.
can be connected to (Fa n F^) x {tby an arc in (Va n F^) x [tj, tj+1\, we
must have
GUidK n F5) x [t;, L+1]) c lVan We (y = «, 0).
But p o GJ+1 = L,|VyX[S.iij + l] for y = a,p, and p|W(r = p|W/J on IFa n UV

Hence G?+1 = G?+1 on (F„ n F„) x [q-, /y+1].
Note that this induction argument also works to start the induction, that is,
for the construction of G1; because we are given Gx on Y x {t0} equal to /.
We leave to the reader the verification that the above construction auto¬
matically makes G stationary. □
Definitions. Let X be an arcwise connected, locally arcwise connected space.

X is simply connected if its fundamental group is trivial, or equivalently, if
every closed path in X is homotopic to a constant.
A covering space (X, p) of a space X is called a universal covering space
if X is simply connected.
Remark. The line, the plane, and the sphere are universal covering spaces,
respectively, of the circle, the torus, and the projective plane. That the sphere
S2 is simply connected, however, remains to be shown.
Definition. A space X is locally simply connected if for each x e X, there exists

an open set F containing x such that any path a in F, with a(0) = a(l) = x,
is homotopic in X to the constant path ex.
70
Theorem 4. Let X be arcwise connected, locally arcwise connected, and locally

simply connected. Let H be a subgroup of rrfX, x). Then there exists a
covering space (X, p) such that p^rrfX, x) = H, where x e X with p(x) = x.
In particular, if H = (e), then X is simply connected, so each such X has a
universal covering space.
Proof. From the covering homotopy theorem and its corollaries, we know
that each a in X is the unique lift starting at 5(0) of the path p ° 5 in X; more¬
over, a is a closed path if and only if (p ° a) e p^rrfX, x) (Corollary 3). Hence
the point 5(1) in X is determined by <p ° 5). So it makes sense to try and
construct the points of X from paths in X.
Let Q denote the set of all paths in X beginning at x. We define an equiva¬
lence relation = on D by a = p if and only if a(l) — /3(1) and e H.
This is an equivalence relation:
(1) a = a because <aa_1> = e H.

(2) If a I P, then = <a/3-1>-1 6 H, sop I a.
(3) If a = p and /3 = y, then a|(l) = /?(1) = y(l), and
(ay-1) = (taP~1 2Py~1'} = (a^"1)^-1) e H,
H
SO a = y.
Let X be the set of all = equivalence classes. For aeQ, let [a] denote the
= equivalence class of a. Define p\ X —> X by p([a]) = a(l). p is surjective
because X is arcwise connected.
We define a topology on X as follows. For [a] e X, let U be an open set in
X containing a(l). Let (see Figure 3.18)
([a], U) = [[aP]; p a path in U beginning at a(l)].
The collection of all such ([a], U), together with 0, forms a basis for the
topology on X. It is a basis because
(1) X = ([ex\, X) and

(2) for (K], UP) and ([«2], U2) two such sets, if [y] e (K], Ux) n ([a2], U2),
then ([y], Ux n U2) c ([«i]> Up) n ([a2]> U2).
71
The topology on X is Hausdorff. For suppose [aj and [a2\ e X with [cti] ^
[a2], If <*2(1) + a2(l), then there exist disjoint open sets U1 and U2 in X with
0^(1) g Ui and a2(0 e U2 {X is Hausdorff), and clearly ([0^], £/) n [a2], U2)
= 0. So suppose ai(l) = a2(l). Since X is locally simply connected, there
exists an open set U containing ax(l) = a2(l) such that any closed path in U
is homotopic in X to the constant path. Then ([ai], U) n ([a2], U) — 0, for
otherwise there would exist paths and y in JJ starting at «i(l) = a2(l) with
[«iyS] = [a2y]; that is, with c^/S = a2y. Thus (a^y-1^-1) e H. But /3y_1 is
a closed path in U, so fiy~l ~ eai(1) and
<aias-1) = <a1eaifX)a2~1y = <ai£y_1a2_1> e H\
that is, <*! = a2 and [aj = [a2\, which is a contradiction.

p: Xis continuous. For if U is open in X, then for each [a] e p_1(i7),
([a], U) is open in X and is contained in p_1(t/).
X is arcwise connected. For let [a] and |)S] e X. To construct a path in X
from [a] to [j8], we shrink a in X by pulling its end in to its origin, and then
snake out along /3 toward its end, and then take = equivalence classes.
Explicitly, an/: /-> X, with/(0) = [a] and/(l) = [/?], is defined as follows:
< t2 <
m =
(0
a < t2 < i),
where
«t20i) = a((l - 2/2)/!) (/2 < i)
and
A2(î) — PM* ~ 1)0 (t2 2: i).
/is continuous, because if ([y], U) is a basic open set in X, then
/_1(M U) = [t2 g I;f(t2) e ([y], U)] = A u B
where
A — [t2e [0, ^]; [a(2] g ([y], U)]
and
[/aeft, l];[AJe([y], £/)].
We show first that A is open. Suppose t2 g A. Then aj2(l) e U. Since t2 ->«(2(1)

is continuous, there exists an open interval J about t2 such that at (1) e U for
72
all t2eJ. Moreover, for t2eJ, <a(2> = <a^>, where t? is the path along a
from oj2(l) to ««2(1). Furthermore, since t2 e A, [cqj = [yp\ for some path p
in U from y(l) to aj2(l); that is, <«%,£“ V-1> e H. (Figure 3.19). But
<“t2(y^)_1> = <at27?-1^_V_1> = <ar2^_1y_1> e //.
Thus
««2 = y(^)>
where prj is a path in C/ from y(l) to at2(l); that is,
K1 = [y(^)]e(M, U)
for all t2 e J. Hence J is an open set about t2, contained in A; and A is open.
Similarly B is open, and hence so is /-1([y], U). This proves that / is con¬
tinuous, thus completing the proof that X is arcwise connected.
(J?, p) is a covering space. For given xx e X, let U be an arcwise connected
open set containing x±, with the property that each path in U is homotopic in
X to a constant. As in the proof above that X is Hausdorff, we see that if
[«i] A [“2] is such that ^([aj) — p([a.2]), that is, «i(l) = a2(l), then
([«i], U) n ([«2], U) = 0. Also, since U is arcwise connected, />|([a]tC0:
([a], U)-^ U is surjective for any [a] with /?([a]) = xx. Moreover, is
injective. For suppose p([af3]) = p([uy]) for some paths p and y in U starting
at x1. Then p(l) = y(l), so Py~1 is a closed path in U and hence is homotopic
to the constant path eXl; that is,
<(«£)(«y)_1> = <«(£y-1)«-1> = (ae*!05-1) = <eXl> e H
and [ap] = [ay].

Now p~\U) = ([a], U) where the union is over all a with/?([«]) = x1.
For if y is a path, with p([y]) = y(l) e U, let p be a path in U from y(l) to xx.
Then p([yp]) = xlt and [y] e ([yp], U).
To complete the proof that (X, p) is a covering space, it remains only to
show that p is an open map; that is, that it maps open sets onto open sets.
73
For this, suppose V is an open set in X. Since V is a union of basic open sets
U = ([a], U), it suffices to show thatp(U) is open for such U. So let xx ep(U);
say Xi — p([y3]) = /3(1) for [yS] e U. Let U1 be the set of all points in U which
can be joined to x± by paths in U. Since X is locally arcwise connected, it is
easy to verify that Ux is an open set. Clearly xx g Ux. We claim that Ux <= p(U),
implying that p(U) is open. In fact, Ux = p(W], Ux). Moreover, [y3] = [ay],
where y is a path contained in U. Thus any element [17] e ([yS], Ux) is of the
form
M = m = [«yS]6([«], U) = U,
where 8 is a path in Ux from yS(l) to 17(1), and ([y3], Ux) c U.

Finally, to complete the proof of the theorem, we must show that
P*(ti(^ x)) = H, where x = [ex]. For this, let a be any path in X starting
at x. Let a be the path in X starting at Jc defined by a(t2) = [<xt2], where
at2(tx) = a(CC)- « is continuous: see the argument used above for/in the
proof that X is arcwise connected, a covers a because p(a(t2)) = p{[at2]) =
af2(l) = a(t2). Now suppose a is closed; that is, suppose <a> e ttx{X, x). Then
<a> e p*(ttx(X, x)) o a is a closed path in X

o a(l) = [ex]
[«] = iex]
H
<>a = ex
O <a> e H.
Thus p*(jTx(X, x)) — H, as required. □
Remark. Recall that, for a group G, two subgroups F/j and H2 are said to
be conjugate if
= gHxg_1 = [gh1g~1; hx g Hx] for some g e G.
Theorem 5. Let (X, p) be a covering space of a space X, let x e X, and let

Xx, x2£p 1({x}). Thenp*TTx(X, andp*ttx(X, x2) are conjugate in vx(X, x).
Proof. Let y be a path X from Xx to x2. Then, by Theorem 5, Section 3.2,
TTx(X, Xx) = [<y><6t><y_1>; <a> e TTx(X, x2)].
Projecting,
p*TTx(X, Xx) = [<p o y}h(p of1); he p*TTx(X, Jc2)],
and (p o y) e x). □
Theorem 6. Let X be locally simply connected and let (X, p) be a covering space
of X. Suppose (X, p) is a covering space of X. Then (X, p ° p) is a covering
space of X.
74
Proof. We leave the proof as an exercise.
Theorem 7. Let X be locally simply connected and let (Xx, px) and (X2, p2) be
covering spaces of X. Suppose x] e Xj (j = 1, 2), with Pl(xx) = p2(x2), and
Al) C= (P2)*7Tl(X2, X2).
Then there exists a unique map p: Xx -> X2 with p(xx) = x2 such that (Xx, p)
is a covering space of X2. Furthermore, p2 ° p — px.
Proof. For y e Xx, we define p(y) as follows. Let yx be a path in fx from xx

to y. Then px ° yx is a path in X with origin px(icx) = p2(x2). Let y2 be the lift
of px »yitoa path in X2 starting at x2. Set p(y) = y2( 1). Then p is well-
defined because if is another path in Xx from xx to y, then <7iî x> e
ttx(Xx, xx) so that
(Pi)*«yi& x)) e (/’i)*77'i(^Ti> ^1) <~ (P2)*7Ti(X2, x2),
and hence px ° y^x-1 lifts to a closed path in X2; that is, y2(l) = yS2(l).
The proofs that p is continuous and unique and (Xx, p) is in fact a covering
space of X2 involve no new techniques and are left as exercises for the
student. □
Definition. Two covering spaces (Xx,px) and (X2,p2) of a space X are iso¬
morphic if there exists a homeomorphism h: Xx -> X2 such that p2° h = px.
Remark. According to Theorem 4, given a subgroup H of the fundamental

group of a space, there exists a covering space whose fundamental group is
mapped isomorphically onto H by the projection map. The following result
asserts that this covering space is unique up to isomorphism.
Theorem 8. Let X be locally simply connected and let (Xx,px) and (X2,p2) be
covering spaces of X. Suppose xx e Xx (j — 1, 2), with px(xx) = p2(x2) and
(/?l)*77l(^ri> -*l) = (/,2)*77l(^2> -^2)-
Then (Xx,px) and (X2, p2) are isomorphic.

Proof. By Theorem 7, there exist maps/5: Xx -> X2 and q: X2 -> Xx such that
(Zl5 p) is a covering space of X2, (X2, q) is a covering space of Xx, px = p2° P,
p2 = Pi0 q, 1) = x2, and q(x2) = xx. Thus
Pi = P2 °P = Pi 0 (q°p)
and (Xx, q ° p) is a covering space of Xx by Theorem 6. From the uniqueness

part of Theorem 7, it follows that q° p = 7>r Similarly, p °q = ix2. There¬
fore q = p^1 and p is a homeomorphism with p2° p = px; that is, (Xx,px)
and (%2, p2) are isomorphic. □
75
Definition. Let (X, p) be a covering space of a space X. A covering transforma¬

tion, or deck transformation, of {X, p) is a homeomorphism h: X -> X such
that p o h = p.
Remark. The set of covering transformations is a group under composition.

It is called the group of deck transformations and is denoted by X, p).
Note that a deck transformation h permutes the “decks” of X; that is, if U
is an admissible open set in X, then h permutes the copies of U in p-1(t/).
Definition. A covering space {X, p) is called a regular covering space of X if

P*tti{X, x) is a normal subgroup of n1(X,p(x)) for some x e X. Note that
since a normal subgroup equals all its conjugate subgroups, the condition
of regularity of a covering space is independent of the base point x.
Theorem 9. Let (X, p) be a regular covering space of a locally simply connected

space X. Then the group id(X, p) of deck transformations is isomorphic to
the quotient group nfX, p(x))/p* -n-fX, x).
Corollary 1. If (X, p) is a universal covering space, then vfX, x) = C${X, p);

that is, rrfX, x) is isomorphic to r${X, p).
Corollary 2. ( '1, 1) ~ «/, where ./ is the group of integers.

771 5
Proof of Corollary 2. The universal covering space of S1 is (R1, p) where

p{r) = e2nir. A deck transformation of (Rx,p) is a homeomorphismh:R1-^R1
such that p o h = p, that is, such that e2nih(r) = e2nir. Thus h(r) — r is an
integer for all r e R1. But r -> h(r) — r is a continuous map R1 -> J. Since R1
is connected, so is its image under this map. Hence h(r) - r = k for some
fixed k\ that is, h(r) = r + k for some k, and h is translation by the integer
k. □
Corollary 3.771(S’1 x S'1, (1, 1)) ^ J @ J.
Proof of Theorem 9. Let H = p^fX, x). By Theorem 8, we may replace

(X, p) by any covering space whose fundamental group projects onto H. In
particular, we may assume that (X, p) is the covering space constructed in the
proof of Theorem 4 and that x = [ex], x e X. We shall construct a homo¬
morphism <p: 77±{X, x) —> &(X,p) which is surjective and has kernel H. The
theorem then follows from group theory.
For <a> e TrfX, x), let <p«a» be the deck transformation defined by
?(<«»([£]) = [op] for [p] e X.
This map y is well-defined because if « ~ ax and I p1 then <«> = <ttl> and

e H so that
(ap(aiP1) ~ 1> = <a><^1_1><a1>-1 = G if
since H is normal; thus aP I a1pi and [a/3] = [a^].
76
Note that 9>(<«» is injective and surjective because its inverse is

<P««» is continuous because clearly 9>«a»_1(([j8], U)) = «p<«-1>(([yS], U)) =
([a-1^], U) for any ([/?], U). Similarly gX**))-1 is continuous so <p«a» is a
homeomorphism. Since
/>((?<«»([£])) = «|8( 1) = j8(l) - Pmi

gXa» is a deck transformation for each <a).
cp is a homomorphism because
<P««lXa2»([^]) = 93(<«l«2»([^])
= K«20]
= <p«al»([a2^])
= <P««1» ° <P««2»([/S])
for all [0] e X; that is, <?X«iX«2» = <K<«i» 0 <P««2» for all <«!>, <«2> e
1Ti(X, x).
The kernel of cp is H because
<K«xm mow] = m
=
o a/3 = fi
o e H.
 X is a deck transformation. Let

[a] = h([ex]). Then a(l) = />([«]) = x, so a is a closed path in X; that is,
<a> e vi(X, x). Now <p(<a» — h. For, in fact, <p«a» anci ^ are both maps
Z-> 2, with
P ° (<Xa» = P° h = p
and with 95«a»([e*]) = [aex] = [a] = h{[ex]), so, by Theorem 2, go«a» =

h. □
(Note that X is locally connected because it is locally arcwise connected:

for y e X and V open in X containing y, the set of all points in V which can
be joined to x by a path is a connected open set in V.)
77
Simplicial complexes
The goal of this chapter is to develop some machinery which will enable us to
compute the fundamental group of a large class of spaces. These spaces are
the ones which can be obtained by piecing together in a nice way basic topo¬
logical building blocks called simplices. A O-dimensional simplex is a point,
a 1-dimensional simplex a line segment, a 2-dimensional simplex a triangle,
a 3-dimensional simplex a tetrahedron, and so on. All the spaces which will
occupy our attention in the coming chapters will be homeomorphic to spaces
built up from simplices.
Given a decomposition (triangulation) of two spaces into small enough
simplices, we shall show that any continuous map from one space to the
other can be approximated by a map which is linear on each simplex. More¬
over, this approximating map will be in the same homotopy class as the
original one. Thus we will have reduced difficult topological problems of
mappings to more accessible algebraic problems of “piecewise linear” maps
and spaces.
Figure 4.1 illustrates these ideas. It shows two triangulations of the unit
78
4.1: Geometry of simplicial complexes
interval I and a piecewise linear approximation to the function /:/->/

defined by
f(x) = 4x — 4x2.
4.1 Geometry of simplicial complexes

Definition. Let V be a vector space over R1 and let C be a subset of V. C is
convex if
Ci, c2 e C => tcx + (1 - t)c2 e C
for all tel.
Definition. A set {v0, vx, . . ., cfe} of vectors in a vector space V is convex-

independent, or c-independent, if the set {z^ — v0, v2 — v0,vk — c0} is
linearly independent. Note that this definition does not depend on which
vector is called Co-
Example. In R2, {v0, vx, v2} is c-independent if and only if v0, vx, and v2 are
not collinear.
Theorem 1. Suppose {c0, vu ..., vk} is a c-independent set. Let C be the convex
set generated by {v0, vu ..., vk}; that is, C is the smallest convex set con¬
taining {v0, Pi,..., vk}. Then C consists of all vectors of the form ]> f=0 aivi,
where a, > 0 for all i and 2f= o ai — 1 • Furthermore, each v e C is uniquely
expressible in this form.
Proof. First note that the intersection of convex sets is a convex set, so C
exists; C is the intersection of all convex sets containing {v0, vly..., vk}.
Now let
n. n,
Ci = v; v = ^ aivu ai - 0, 2 = 1
{= 0 i=0
Ci is convex because if v — 2?= o <*ivi and w = 0 biVu then

k
tv + (1 — 0^ = 2 \-tQi + (1 —
i=0
and
y [tat + (1 - t)bt] = t 2 + (1 ~ 0 2 bi = t + a “ 0 = !•
i=0 i=0 1=0
Thus Ci is a convex set containing {c0, vlf..., vk}; hence Cx C.

Conversely, Cx c C. For certainly 2f=o aivi e C whenever all but one at
are zero.
We proceed by induction. Assume 2?=o aivi e C whenever, at most, n of
the at are nonzero (n < k + 1). Let 2.^=o Wt have n + 1 nonzero au which
79
4: Simplicial complexes
we may assume (by relabelling if necessary) are a0, an. an A 1 for

otherwise all other flj = 0. Thus
Since
Sr-o1 [a(/(l — a„)]Pj e C by the induction assumption. Hence
n— 1
for tel since C is convex. Let t = 1 — an. It follows that 2?=o e C; that
is, Ci <= C, and Cx =< C.
We proceed to show uniqueness. Suppose
k k
i=0 i=0
where 2?=o = 2f=o = 1. Then

k
o = 2 («»-
i=0
k
2 (ai _ t)i){Vi - U0).
i=0
Since {px — p0, v2 — v0,vk — p0} is linearly independent, at — bt = 0 for

all i > 0. Then clearly a0 — b0 also. □
Definition. Let V be a vector space over R1. A convex set generated by c-
independent vectors {v0, vx,..., vk} is called a (closed) k-simplex and is
denoted by [p0, vx,..vk]. k is called the dimension of the simplex. If
v e [p0, vx,..., vk], then the coefficients au with at > 0 and 2JL0 <k = 1
such that v — 2f=o aivu are called the barycentric coordinates of v.
Examples. For {u0, vx} vectors in R1, the simplex [p0, pj is the closed interval
[p0, Pi]- For {v0, vx, y2} c R2, [Po> vlt v2] is the triangle with vertices v0, vx, and
v2. The centroid of this triangle is the point with barycentric coordinates
(t> i). For V = Rn, the simplex [p0, vu ..., vk] is a compact metric space
(it is closed and bounded) in the relative topology. In fact, using barycentric
80
4.1: Geometry of simplicial complexes
coordinates, it is not difficult to see that [e0, vlf..yfc] is homeomorphic to a

product of k unit intervals. However, this homeomorphism is not an isometry.
Definitions. Let {i>0, vlt..., vk} be a c-independent set. The set
[v e [i>0, vu ..., vk]; aiv) > 0, i = 0, 1,..., k]
is called an open simplex and is denoted by (v0, vu ..., vk). We shall also
denote an open simplex by (5) and the corresponding closed simplex by [5].
Let [5] = [y0, vk] be a closed simplex. The vertices of [5] are the
points v0, Vi,..., vk. The closed faces of [s] are the closed simplices
[vJo,v}l,...,vjh] where .. .,jh} is a nonempty subset of
{0, 1 ,...,k}. The open faces of the simplex [5] are the open simplices
(vjo, vh,..., vj.
Remarks
(1) A vertex is a 0-dimensional closed face. It is also an open face.
(2) An open simplex (5) is an open set in the closed simplex [5]. Its closure is
[si
(3) The closed simplex [s] is the union of its open faces.
(4) Distinct open faces of a simplex are disjoint.
(5) The open simplex (s) is the interior of the closed simplex [5]; that is, it is
the closed simplex minus its proper open faces (faces ^ (5)).
Definition. A simplicial complex K (Euclidean) is a finite set of open simplices

in some Rn such that
(1) if (s) e K, then all open faces of [s] e K;

(2) if (ii), (.y2) e K and (5X) n (.y2) ^ 0, then fo) = (s2).
The dimension of K is the maximum dimension of the simplices of K.
Remarks. If K is a simplicial complex, let [X] denote the point set union of
the open simplices of K. Then [Z] is compact, and [K] = (Jwe*: CO =
U(s)e£T CO-
If [.?] is a closed simplex, the collection of its open faces is a simplicial
complex which we denote by s.
Examples. Figure 4.2 shows examples of simplicial complexes. Those shown
Figure 4.2
81
Figure 4.3
Figure 4.4
in Figure 4.3 are not simplicial complexes. By adding simplices, however, the
point sets in Figure 4.3 can be made into complexes (Figure 4.4). Note that a
complex is more than just a point set. It is a set with additional structure. It is
possible to have two different complexes with the same point set, as in
Figure 4.5.
Figure 4.5
Definition. A subcomplex of a simplicial complex A is a simplicial complex

L such that (s)eL implies (s)e K.
Remark. For each (s) e K, the simplicial complex s is a subcomplex of K.
Definition. Let K be a complex. Let r be an integer less than or equal to

dim K. The r-skeleton Kr of K is the collection KT = [(5) e K; dim s < r].
Remark. The r-skeleton Kr is a subcomplex of K.
82
4.2: Barycentric subdivisions
4.2 Barycentric subdivisions

Definition. Let v e Rn and let A Rn. The pair (v, A) is in general position
if v $ A and, for each au a2e A with a1 + a2, [v, ad n [v, a2] = {y}.
Examples The points and sets in the plane shown in Figure 4.6 are in general
position. Those shown in Figure 4.7 are not in general position. Note that if A
Figure 4.7
is a triangle with its interior, then there is no v e R2 such that (v, A) is in

general position.
Definition. Let (v, A) be in general position. The cone with vertex v and base A
(or the join of v with A), denoted by v * A (Figure 4.8), is the set
v * A — \^J [v, a].

aeA
Theorem 1. Let [s] = [y0, vu ..., vk] be a k-simplex. Let ve(s). Then
(v, [Ate-1 ]) is in general position, and v * [sfc-1] = [j].
Figure 4.8
83
Proof. Let au a2 e [sk ~1]. Suppose there exists w e [v, oj n [v, a2] with
w / v. We must then show that a1 — a2. Consider the expressions for au a2,
and v in terms of barycentric coordinates in [s]:
k k k
a-L = 2 aiiVi> fl2 = 2 aziVi’ v = 2 P'Vi-

i=0 i=0 i=0
Since (s) and a2 $ (s), aUl = 0 and a2iz = 0 for some ft and i2 < k. More¬
over, since v e (5), ft ^ 0 for all i. Since w e [u, aj, w = ftp + (1 — ft)#! for
some ft el. Thus w = 2?=o [ftft + (1 — ft)«u]ft-
Similarly, since we [v, a2], we have w = 0 [ftft + (1 — ftOîlft for
some t2 e I. By the uniqueness of barycentric coordinates,
ftft + (1 — ft)aii = ftft + (1 — ft)a2i (/ = 0, 1,..., ft).
Hence ft t2 — (1 /ft)[(1 — ft)«2i — (1 — ft)a1{].

Taking z = ft yields
ft _ ft = 7T- (1 _ ft)a2i1 — 0.
ftl
Taking z = ft yields
ft — ft 0 ~ ft)ali2 < 0.
Hence ft — ft = 0, ft = ft, and
(1 ft)o:ij — (1 — ft)a2j
for all z. Now ft ^ 1 since vv ^ p, so this implies that alf = a2j for all z; hence
«i = «2, completing the proof that (v, [s*-1]) is in general position.
Now v * [5fc_1] c [5] because [5] is convex. Conversely, |>] c v * [s*-1].
For if w e [sk~1], then certainly wev * [s'*-1]. So suppose w e (s). We may
assume w ^ v. In barycentric coordinates,
k
w = 2 “‘ft’ V — (all «i5 ft > 0).
i=0
Since
-ft) i «< - i ft =1 -1 = 0,
i=0 i—0
and since at — ft ^ 0 for some ft ay — ft < 0 for some y. For each such y,
let///) = ft + /(ay - ft). Since //!) > 0 and ///) < 0 for large /, there
84
4.2: Barycentrie subdivisions
Figure 4.9
exists a t} > 1 such that fa + tj{af — fa) = 0. Choose i0 from among the
numbers j so that tlo < t„ for all such j. Then fa0 + tio(aio - fa0) = 0, and
A + ho(ai — fa) > 0 for all /'. Hence v + tio(w — v) = re[il_1] (see Figure
4.9). Also,
w = — x + ——- v = Hjc + (1 — tl)v

ho ho
with t1 = \/tio < 1. Hence w e v * [5|C_1]. □
Exercise. Let [5] = [t>0, vlt..., vk]. Prove that (v, [5]) is in general position if
and only if vlt.. ., vk, t>} is c-independent, in which case
v * H = [t>0, vu ..vk, v].
Definitions. Let s be a A>simplex. The barycenter of s, denoted b(s), is the

point in (j) with barycentric coordinates (1 /(k + 1),..., \/(/c + 1)); that
is, if
(5) = (0O, »i,, 0fc),

then
b(s) = irn Xv"

Let K be a simplicial complex. A subdivision of K is a simplicial complex
K+ such that
(1) IK<] = [K]

(2) if s e K+, then (s) some open simplex of K.
Examples. Each of the complexes in Figure 4.10, second column, is a sub¬

division of the corresponding complex in the first column. Note that although
the second and third complexes in the second column have the same point
set, neither is a subdivision of the other.
85
Figure 4.10
A
Theorem 2. Let s be a k-simplex. Let Kf be a subdivision of sk~1. Let v e (5).
Then (v, [A^]) is in general position. Furthermore, v * [A^+] is the point set of a
complex K defined by K — K* U (Us+eK+ (5+> v)) u (v) (see Figure 4.11).
Here, for
0+) = (»o. vx,...,vr)eK\
(s\ v) — (v0, Pi,, vr, v). The complex K is a subdivision of s.
Figure 4.11
86
Proof. By Theorem 1, (v, [s*-1]) is in general position and v * [s*-1] = [s].

But [Kf] = [5fc_1], so (v, [Kf]) is in general position, and v * [AT+] = [5].
We must show that K is a simplicial complex. It is a set of open simplices.
Moreover, each simplex v) in K is either in Kf or is of the form (s1, v). If it
is in K\ then all its open faces are in K\ hence in K. If it is of the form (s+, v),
then its open faces are s+, (v), and [(5,1+, v); an open face of s+]. Thus, in
each case, all open faces are in K, and Condition (1) for a complex is satisfied.
To verify Condition (2) for a complex, we must show that distinct open
simplices have void intersection. Since Kf is a complex, this is certainly
satisfied by pairs of simplices each of which is in K+. Moreover, if s+ e A"1
then (,s+) n (j1+, v) = 0 for all .y1+ e Kf because, in fact, (jy1, v) (5).
Clearly, (v) meets no other open simplices of K. So now suppose s1+, sf e A+
have (5x+, v) n (sf, y) A 0. Let w e (s^, v) n (s2\ v). Since these are open
simplices, w + y. Now there exists a unique xe[5fc_1] = [/C] such that
w e [y, x] (see Theorem 1). Since [s^, v] = v * [5/], it follows that x e (51+).
Similarly x e (.y2+), so (5i+) n (s2+) A 0. Therefore i,1t = sf since Kf is a
complex, and (5X+, v) = (s2+, v). Thus K is a complex.
The point set of K is
[£] = U [S] = u v * = v * m= [ji-

seK sfeKf
Since each open simplex of K is contained in an open simplex of s—those in

Kf are, because K+ is a subdivision of sk~1; the rest are contained in (s)—
K is a subdivision of s. □
Definition. Let K be a simplicial complex. A partial ordering is defined on K

by Sx < s2 if and only if Sx is a face of s2. The notation sx < s2 shall mean
5X < s2 and Sx ^ s2.
Theorem 3. Let K be a simplicial complex. Let
Ka) = [(i(j0), b(sx),. ..,b(sk)); s0 < s± < • • • < sk; s0, slt..sk e K].
Then Ka) is a subdivision of K. Furthermore, for each s0, slf...,sreK with

s0 < Sx <■■■< sr, (b(s0),..., b(sr)) c (sr).
Remark. The subdivision Ka) is called the first barycentric subdivision of K.

Iterating, K<-n) = (((A(1))a)) • • -)a> is the nth barycentric subdivision of K.
v—-V-
n times
Proof of Theorem 3. (By induction on dim K.) For dim K = 0, Ka) = K

so there is nothing to prove. Assume the theorem is true for all simplicial
complexes of dimension <n — 1. Let dim K = n. Then the (n — l)-skeleton
Kn~x is a complex of dimension <n - 1; hence the theorem is true for A:71-1.
In particular, if
So, • • •, sr s A
87
s0 < Si <■■■< sr, and dim sr < n — 1, then {60o)> £>0i),..., 60r)} is c-
independent and gives an open simplex (b(s0), 60i),..60r)) in (^'l_1)(1).
Furthermore,
Wo), 60i),b(sr)) <= Or)-
Now suppose s0, sx,..., sr e K (s0 <••• < 5>), and dim sr = n. Since
sr_ 1 < sr, dim iSr_] < n — 1, so that (b(s0), b(s1),..., 60r-i)) is a simplex
<= Or _ x), which is a face of sr. Since b(sr) e Or), Theorem 1 implies that
(b(sr), (b(s0),..., 60r-i))) is in general position. Hence (b(s0), 6O1),..., b(sr))
is an open simplex, the interior of the closed simplex
Wo), &0i), • • ■, &0r)] = &0r) * Wo), • • - , b(sT_ 1)] <= [sr].
Thus Wo),b(sr)) <= Or)-

Thus far we know that Ka) is a collection of open simplices. It is in fact a
simplicial complex. Condition (1) is clearly satisfied: any face of
(b(s0), b(si),..., 60r)) is of the form (b(sio), b(sh),..., b(sjh)) and is hence in
Ka\ We now verify Condition (2). Suppose s0 < ■ ■ ■ < sr, s0 < ■ • • < sq, and
Wo),..., b(sr)) n Wo),60,)) ^ 0.
Let w belong to the intersection. Then vv e (sr) n (5,). Since K is a complex,

sr = sq and b(sr) = b(sq). Moreover,
Wo),b(sr _ 0) c (Sr _ 0 and (b(s0),..., 60, _ 0) c 0, _ x)
where Or-i) and 0,_ 1) are both faces of Or); hence (b(s0),..., 60r-i)) and
Wo),..., 60,- 0) g Since
w g Wo),..., b(sr_ 1), b(sr)) n (b(s0),..., 60,- 0, 600)

c b(sr) * (b(s0),.. .,b(sr_1)) n 600 * (b(s0),..., 60,-0),
we conclude by Theorem 2 and the induction assumption that
Wo),.60,-0) = Wo), • • - , b(Sq~ 1)).
This shows that X(1) is a simplicial complex. To complete the proof of the
induction step and of the theorem, we must show that [A^(1)] = [A^]. Clearly,
[Ka)] <= [AT]. Also, [*(1)] =3 [(/f"-0(1)] = [AT--1], using the induction
assumption. Hence we must show that
[A(1>] ^ [tf] _ [K*-1].
So suppose w g [AT] — [A^n_1], Then w must lie in some open simplex (j) of
dimension n. Thus
w g 0) c 0] = 60) * 0n_1].
Now O'1-1] <= [X71-1] = [(A^n_1)(1)] so w g 60) * 0i) for some
W = Wo), • • •, b(sk)) g (/fn-1)(1).
88
If w = b(s), then w is a vertex in Ka). If w ^ b(s), then
w e (fe(j0), • • •, b(sk), b(s)) c= [*<»]. □
Definitions. Let (S, p) be a metric space, and let T be a compact subset of S.

The diameter of T is
diam T = sup p{tx, t2).

*i**a«r
Since T is compact and p is continuous, the maximum is assumed, and we

may write:
diam T = max p(tx, t2).

ti.f2er
Let K be a simplicial complex in Rn, where Rn is provided with the usual

metric. The mesh of K is the maximum diameter of simplices of K:
mesh K = max diam [5].

seK
Lemma. If s is a simplex in Rn, then diam [5] = p(vx, v2) for some pair vx, v2 of
vertices of s. If K is a simplicial complex, then mesh K = p(v1, v2), where vx
and v2 are vertices of some simplex of K.
Proof. Let s be a simplex and let vx, v2 e [5] be such that diam [5] = p(vx, v2).
Suppose, say, that v2 is not a vertex. Then v2 is in some open simplex of
dimension > 1. In particular, there exist wx, w2 e [s] with wx 7^ w2, such that
v2 = twx + (1 - t)w2
for some t with 0 < t < 1. But the convex function / defined on / by
fit) = pOi, twx + (1 - t)w2),
has no maximum for 0 < t < 1, contradicting the maximality of p at (y1; v2).
The second statement of the lemma follows immediately from the defini¬
tion of mesh K. □
Theorem 4. Let K be a simplicial complex of dimension m. Then mesh Ka> <

(m/(m + 1)) mesh K. In particular, limn-.^ mesh K(n) = 0.
Proof. By the lemma, there exists a simplex (b(s0), b(sx),..., b(sr)) e Ka) such
that mesh Ka) = p(b(sk), b(sh)) with sk < sh. By renumbering vertices if
necessary, sk = (v0,..., vp), sh = (v0,..., vp, vp + 1,..., vq), and
mesh Ka'’ = ||6(^) - fc(jh)ll

1
P + 12
1 i=0 - jti 2
1/ T 1 j=0 v>
89
1 q + 1
q + 1 P + 11 i=o i=o
1 * /
q + 1 j=0 l/5
1 1
p + \ q -\- 1 life- •>,)
j = 0 i=0
<
1 1 â v
P + I1 vTTT 2 2 Ife - «’/!-

^ 1 i=o i = o
But \\vt - Vj\\ < diam OJ < mesh K. Moreover, whenever i - j, the (i,j)th
term in this summation is zero. There are p + 1 such terms. The number of
nonzero terms is therefore
(p + 1)(<7 + 1) — (P + 1) = (P + ^)<i-
Since each term in the summation is < mesh K, and since q < m,
mesh Ka) < —mesh K < —- mesh K. □

q + 1 m + 1
4.3 Simplicial approximation theorem

Definition. Let K and L be simplicial complexes. A map y: [A] -> [L] is a
simplicial map if
(1) for each vertex v of K, y(v) is a vertex of L,

(2) for each simplex (v0, vu ..., vk) e K, the vertices <p(y0X <p(Ti), • • •, <p(vk)
all lie in some closed simplex of L, and
(3) for each (s) = (i>0, vu ..vk) e K and p = 2?=o aivi e (■y)> the image
of p is given by
k
9>(/0 = 2 awivi).
i=0
Remark. Condition (1) says that <p must map the 0-skeleton of K into the
0-skeleton of L. Condition (2) says that for each simplex (v0, vu ..., vk) e K,
the set cp(v0), <p(Vi),..., y(vk), with redundancies removed, is the set of vertices
of some simplex in L. Condition (3) says that the mapping cp is linear on each
simplex.
Since a simplicial map depends on K and L, not just [A] and [Lj, we will
denote it by <p: K -> L.
90
4.3: Simplicial approximation theorem
Examples. In Figure 4.12, projection is a simplicial map. However, in Figure

4.13, projection is not a simplicial map, even though conditions (1) and (3)
are satisfied.
Remarks. With the glueing lemma, it is easy to check that a simplicial map
is continuous.
A simplicial map is, by Condition (3), determined by its effect on vertices.
Conversely, a vertex map K° -> L° from the vertices of K into the vertices of
L can be extended to a simplicial map K —> L if and only if Condition (2)
above is satisfied.
Definition. Let K be a simplicial complex and let v be a vertex of K. The star of

v is the point set
St(u) = (J (s)
vet s]
(s)eK
Theorem 1. Let K be a simplicial complex. For v a vertex of K, St (v) is an

open set in [K] containing v, and v is the only vertex of K which lies in St (v).
The collection {St (f)}ue^0 is an open covering of [K].
Proof. We shall show that the complement of St (v) in [AT] is closed.
st (vy = U CO.
vi[s]
91
Since v $ [5] implies v $ face of s, we have (5) c St (y)' implies [5] £ St (v)'.
Since [s] is compact, [s] is closed. Hence St(y)' = U(s)=sta>y is closed.
Next, v is the only vertex in St (v) because the only open simplex containing
a vertex is the 0-simplex consisting of that vertex alone.
Finally, (Jpe£:o St (v) = [K] because if p e [K], then p e (5) for some
(5) £ K, and hence p e St (y) for any vertex v of (s). □
Definition. Let K and L be simplicial complexes. Let /: [K\ -> [L] be con¬
tinuous. A simplicial map <p: K->L is a simplicial approximation to /if
/(St (y)) <=■ St (rp(v)) for each vertex v of K.
Theorem 2. Suppose <p: K—> L is a simplicial approximation to f: [A] -> [L],

Then, for any p e [K\j\p) and (p(p) lie in a common closed simplex of [L\.
Proof. Let p e [K], Then p e (s) for some simplex (s) = (v0, vx,..vT) e K,
and
f(p) ef((s)) <= /(St (y,)) c St
for all j £ {0, 1,..., r}. Now f(p) e (t) for some simplex (t) e L, so for (t),
(t) n St (c-p(Vj)) ^ 0 for all j. But, since L is a complex and St (<p(vx)) is a
union of open simplices, (t) <= St (<p(vy)) for all j; that is, 9o{vj) is a vertex of (t)
for all j. In terms of barycentric coordinates in s, p = 25 = 0 aivj> and
r
<p(p) = 2 a^vi)6 [?]-

)=0
This completes the proof. □
Corollary. Suppose 9r. K-> L is a simplicial approximation to f: [A] -> [L],

Then
d(f,cp) < mesh L,
where d{f <p) - suppem P(f(p), <p(p)).
Theorem 3. Let cp be a simplicial approximation to f: [A] -> [L\. Let Kx be a

subcomplex of K, and suppose that the restriction of f to [A\] is a simplicial
map. Then there exists a homotopy between f and cp which is stationary on
[Kx].
Proof. Define F: [K] x /—> [L] by
F(p, t) = tcp(p) + (1 - t)f(p).
F does map into [L] because, by Theorem 2, f(p) and cp(p) lie in a common
simplex that, being convex, also contains the line joining them. It is easily
verified that F is continuous, and clearly F(p, 0) = f(p) and F(p, 1) = cp(p)
for all p £ [A].
92
4.3: Simplicial approximation theorem
F is stationary on [A^] because <p|[Kl] is a simplicial approximation to

/ItKiJ (since /(StKl (f)) c /(StK (v)) c St (y(v)) for each vertex v e K0, and
hence / = <p on [A^] by the following lemma. □
Lemma. Suppose f: K—^L is a simplicial map and that <p is a simplicial

approximation to f. Then <p = f
Proof. For each vertex v e K,
f(v) 6/(St (v)) <= St (93(f)).
But, since/is a simplicial map,/(f) is a vertex and, by Theorem 1 ,f{v) = 93(f).

Thus/and [L] be continuous and cp\ K° -> L° be a vertex map.
<p can be extended to a simplicial approximation to f if and only if /(St (f)) <=
St (cp(v)) for all v e K°.
Proof. The implication in one direction is obvious. For the other direction
we need only verify that L; that is,
we need only show that if (s) = (f0, fl5..., vr) is a simplex in K, then
cp(v0), yivx),..., cp{vr) are vertices of a common simplex of L. But, in fact,
/(CO) c /(St (v,)) c: St (<p(i/)) for ally e{0, 1,..., r} so p|5=o St =£ 0.
This implies there exists an open simplex (t) c St (<p(n,)) for all j. <p(vj) must
then be a vertex of (t) for all j. □
Theorem 5. Let f: [/f] —> [L] be continuous. Let {A'n} be a sequence of sub¬
divisions of K such that limn_oo mesh Kn — 0. Then, for n sufficiently large,
there exists a simplicial map <p: Kn-> L such that <y is a simplicial approxi¬
mation to f.
Proof. By Theorem 1, (St (w)}weL« is an open covering of [L\. Since / is

continuous, {/_1(St (w))}weL° is an open covering of [K], Since [/f] is a com¬
pact metric space, there exists a S > 0 such that any ball of radius 8 lies in
some open set of this covering. Choose n large enough so that mesh Kn < 8/2.
Then diam [5] < 8/2 for each s e Kn. Hence, for each vertex v in Kn,
St (f) <= Bv(8). But 2/(8) <= /-1(St (w)) for some w e L°, so for each v e (Kn)°,
St (y) c /- !(St (w)) for some w e L°. For each v e (Kn)°, define cp(v) to be any
such vertex w eL°. (There are only finitely many such w; pick any one.) Then
<p:(Kn)°->L° has the property that St (v) <= /_1(St (<p(v))); that is,
/(St (t>)) c St (93(f)) for each v e (Kn)°. By Theorem 4, 93 can be extended to a
simplicial approximation to /. □
Corollary. Let /: [A"] -> [L] be continuous. Then, for any e > 0, there exist
subdivisions Kn of K and Lm of L, and a simplicial approximation 93: A"n -> Lm
to f such that d(f 9?) < e.
93
Proof. By Theorem 4 of Section 4.2, there exist subdivisions with arbitrarily

small mesh. Given e > 0, Let Lm be a subdivision of L such that mesh Lm < e.
Then/: [K] -> [Lm\. By Theorem 5, there exists a subdivision Kn of K and a
simplicial approximation cp: Kn-+Lm to/. By the corollary to Theorem 2,
d(f, cp) < mesh Lm < e. □
4.4 Fundamental group of a simplicial

complex
Definition. Let K and L be simplicial complexes. Let 9?! and <p2: K-+L be
simplicial maps. <p1 and <p2 are contiguous if, for each simplex (v0, vly..vk)
e K, there exists a simplex t e L such that
9>iOo), ¥i(»i\¥i(vk) and ¥2(1)0), ¥2(1)1), • • •,
are vertices of t.
Example. Let K be the complex of a 3-dimensional simplex, with vertices

{v0,Vi,v2,v3}. Let L be a one dimensional complex with three vertices
(w0, w±, w2} and two 1-simplices (w0, u/) and (wl5 w2), as in Figure 4.14.
w0 Wi W-1
K
Figure 4.14
Define simplicial maps <pu cp2, <p3: K -> L by prescribing them on vertices as
follows:
<Pi(Po) = ¥i(Vi) = Wo, ¥1(1)2) = <Pi(p3) = wx;

¥2(1) o) — <P2(g) — ¥2(1)2) — ¥2(1)3) — Wi ;
¥3(vo) = W2, <P3(Vl) — ¥3(1)2) = ¥3(1)3) = Wi.
It is easily checked that cpx and <p2 are contiguous and that cp2 and <p3 are
contiguous. Note, however, that <p± and <p3 are not contiguous. Hence the
property of being contiguous is not an equivalence relation.
Definition. Two simplicial maps cp,: K -* L are contiguous equivalent,

denoted ~, if there exists a finite sequence <p0, ¥1, ■ • •» ¥k of simplicial
maps K^L such that cp0 = <p, ¥k — </•, and <pt is contiguous with cpi_1 for
each / e{l, 2,..., k).
94
4.4: Fundamental group of a simplicial complex
Remark. It is easily checked that ~ is an equivalence relation.
Theorem 1. Let K and L be simplicial complexes and let f: [/v ] —> [L\. Suppose
<Pi, <p2: K L are both simplicial approximations to f. Then <px and <p2 are
contiguous.
Proof. Let (5) = (v0,..., vk) be a simplex of K. Then
/(CO)c/(St(y,))^ Stfa(i;y))
for all je {0,..., k} and ie{l, 2}. Thus
/((j))c nst n n st (<p2(Vj)).

1=0 y=o
Let (r) be an open simplex in L such that f((s)) n (t) ^ 0. Then
(t)cz p|St (9l(yy)) n P) St (<p2(v,)),

1=0 1=0
and hence 9>i(u0), • • •, <Pi(vk) and 9o2(v0),..., cp2(vk) are vertices of (t). □
Theorem 2. Suppose 9^ and <p2: K^-L are contiguous simplicial maps. Then
9>i and cp2 are homotopic.
Proof. Note first that for each p e [A"], cpfip) and <p2(p) lie in a common
simplex of L. For p e (.s) with (s) = (v0,..., vk) e K. In barycentric coor¬
dinates, p = 2f=o Since 9and ?2 are contiguous, <px(v0),..., yfv^ and
<p2(v0),..., <p2(vk) are vertices of some simplex (t) e L. Then yfp) =
2i = o ai<Pj(Vi) e (/) fory e {1, 2}.
Now define F: [K] x /-> [L] by
F(P, 0 = (1 - t)<Pi(p) + t<p2(p), (pe[K];te I).

F makes sense: since cp^p) and <y2{p) lie in a common simplex for each
p e K, so does the line segment joining them. F is a homotopy from 9^ to
<P2■ □
Corollary. Contiguous equivalent simplicial maps are homotopic.
Theorem 3. Let K be a simplicial complex. Let oc0 and a/ -> [A'] be paths in
[A']. Suppose a0 ± ax. Then there exists a subdivision F of I and simplicial
maps 9>0 and 9^: F —> K such that
(1) 9>y is a simplicial approximation to a( j e {0, 1}) and
(2) 9>0 ~ 951.
Moreover, the subdivision F can be chosen to be finer than any given simplicial
subdivision of I.
Remark. More generally, if K and L are two simplicial complexes and

fo,fc[K]->[L]
95
are homotopic maps, then there exists a subdivision K' of K and simplicial
maps <p0, 9ox \ K' ->L satisfying Conditions (1) and (2) above. The proof of
this is a generalization of the following proof.
Proof of Theorem 3. Let F: I x /-^ [K] be a homotopy from a0 to ax.

Since {St (vv)}w6Ko is an open covering of [K], {F_1(St (w))}^^ is an open
covering of / x I. Since / x / is a compact metric space, there exists a 8 > 0
such that each ball of radius 8 is contained in F-1(St (w)) for some w e K°.
Choose a subdivision /' of /, with vertices v0 = 0, vu ..., vs — 1, and
another subdivision /" of / with vertices //2fc, (/ = 0,..., 2k). Then /' x I"
can be made into a simplicial complex M with vertices vrl — (vr, l/2k) and
2-simplices of the form (vrl, vlr+1, vlrX\) or (vrl, vlT + 1, (see Figure 4.15).
The subdivisions can be chosen fine enough so that
St (vT) x [(/ - l)/2fc, (/ + l)/2k]
is contained in a ball of radius 8 and therefore contained in F~1(St (w)) for

some w e K°. Since St (vTl) <= St (vr) x [(/ - l)/2\ (/ + l)/2fc] c /^(St (w)),
there exists, by Theorem 4 of Section 4.3, a simplicial map O: M -> K, which
is a simplicial approximation to F and for which
St (vr) x [(/ - l)/2fc, (/ + l)/2fc] c F-\St d)(pr')).
Let <pi = (i = 0, 1), so that (pf is a simplicial approximation to

F|/*{i} — aj.
We now show that (p0 ^ 9?i. Leti/-, = <h|/x;/2'csothat(/«o = To and^2* =

It suffices to show that i/j, and i/<J + 1 are contiguous for / = 0,..., 2k - 1;
that is, for each simplex (vr, vr + 1) £ /', the vertices
•AiOr) = MVr+l) = +
+ i(Pr) = and iAi + i(wr + 1) =
96
lie on a simplex of K. But
/ l + 1 / / + 1
Q St => F^St (vr) x ) n F(st(vr + 1) x
u=o 2k’ 2k 2k’ 2k
11 + 1
= FI vr+1) x
2k’ 2k )■
which is not empty and hence contains a simplex (t) of K. Hence the four
vertices in question are vertices of (/). □
Definitions. Let K be a simplicial complex. An edge in K is an ordered pair

e = | of vertices of K, such that vx and v2 lie in some simplex of K.
vx is the origin of e, and v2 is the end of e. If e = |pji;2|, the edge \v2vx\ is
denoted by e~1. A route in K is a finite sequence tu = e1e2 - ■ -ek of edges in
K such that, for each / e {1,..., k — 1}, the end of et equals the origin of
ei+1. The origin of oj is the origin of eu and the end of a> is the end of ek.
Given two routes tu = ex • • • ek and r = e[ ■ ■ ■ e'm with end of cu equal to
origin of r, their product tor is defined by
a>T = ex- ■ • -e'm.
The inverse of a route a> = ex ■ ■ ■ ek is the route
tu-1 = ek 1- • -ei1.
An equivalence relation on the set of all routes in K is defined as follows.

If e — \vxv2\ and / = |y2y3| are such that vx, v2, v3 are vertices of a simplex,
then the product ef is edge equivalent to the edge . Two routes tu and r
E
are edge equivalent, denoted tu ~ r, if t can be obtained from tu by a
sequence of such elementary edge equivalences.
E
Example. Suppose K is the complex in Figure 4.16. Then lôî 11 fit?2| —
I ^0^311 v3v21 because
E E
\v0Vi\\vxv2\ ~ \v0v2\ ~ \v0v3\\v3v2\.
Figure 4.16
Remark. Edge equivalence is an equivalence relation. Moreover, if tu is a

E
route with origin v, then tutu-1 — \vv\. Also, if vx, v2,..., vk are vertices of a
E
simplex, then \vxv2\\v2v3\ - ■ ■\vk-1vk\ ~
97
Theorem 4. Let K be a simplicial complex, and let v0 be a vertex of K. Let

E(K, v0) be the set of edge equivalence classes of routes in K with origin v0
and end v0. Then E(K, v0) is a group, with identity |i>0tf0|, under the opera¬
tions of multiplication and inverse defined above for routes. E(K, v0) is called
the edge path group of (K, v0).
Proof. Routine. □
Remark. The edge path group of a complex K is a purely “combinatorial”

object; that is, it depends on only the vertices of K and those subsets which
are vertices of a simplex. Its definition does not use the topological
properties of the space \K\.
Indeed, we can define an “ abstract ” simplicial complex as follows. Let V be
a finite set, elements of which we shall call vertices. Let A be a collection of
subsets of V, called (abstract) simplices, such that
(1) if v e V, then {v} e A;

(2) if S e A, then each nonempty subset of S is also in A.
Such a collection A is called an abstract simplicial complex.

Note that every simplicial complex determines an abstract simplicial
complex. Conversely, it can be shown that every abstract simplicial complex
A has a “realization” as a simplicial complex; that is, there exists a
simplicial complex whose abstract complex is A. (Note, however, that each
abstract complex corresponds to many (nonisometric) simplicial complexes.)
The edge path group can then be defined for abstract complex A. It is
the same as the edge path group of any realization of A. It is in this sense
that we mean E(K, v0) is a purely combinatorial object. In contrast, the
mesh of a complex is not a combinatorial concept.
Theorem 5. Let K be a simplicial complex, and let v0 be a vertex of K. Then

E(K, v0) is isomorphic with rr1 ([AT], Vo)-
Proof. We construct an isomorphism h: E(K, v0) -»• ^([AT], v0) as follows.

Let w be a route in K beginning and ending at v0. Then co = IfoîHî^l • • •
\vk_1vk\ for some set^, v2,..., vk} of vertices in K, with vk — v0. Now regard
the interval / as the space of a simplicial complex with vertices
{0, 1 /k, 2Ik,.. .,(k — 1 )/k, 1}. Consider the vertex map <pm: 7° K° defined
by <Pco(jlk) = Vj, (J £ {0, 1,. ..,k}). Since |o0»i| •' • K-i®»| is a route, <pra
extends to a simplicial map <pa: I^ K. Set h(a>) =
E
Note that if a> ~ r, then (pa ~ <pT, so h{w) = h(j). Thus h is well defined.
h is a homomorphism because if a> = e1 ■ • ■ ek and r = e[ ■ ■ • e'm are routes
with origin and end v0, then a homotopy between <paz and cpacpz is obtained
from Figure 4.17.
98
h is surjective. For suppose <a> e 7r1([^], v0). Then, by the simplicial

approximation theorem, there exists a subdivision /' of / and a simplicial
approximation 93: 7' -> K to a. Moreover 93 ± a, so <9?> = <a>. Let t0 <
?!<■••< 4 be the vertices of /'. Let a> be the route
• • • |<p(4-iMhc)l
in if. Then /z(to) - <93) = <a>.
h is injective. For this we must show that if to is a route such that cpw ~ eVo,
E
then cj ~ |y0y0|. But, by Theorem 3, there exists a subdivision /' of I and
simplicial maps cp0, cp1:/' —> K such that y0 and cp1 are simplicial approxima-
Q
tions to <pa and eVo respectively, and such that 9>0 — <px. (This subdivision /'
can be chosen finer than the subdivision of / used to define Now, since
eVQ is a simplicial map and 9ox is a simplicial approximation to it, cp1 = eVo.
Then to show that
E
OJ ~ |p0t>o|,
it suffices to prove the following.
(1) If 93 and jj are contiguous equivalent simplicial maps I' ^ K, then

~ awhere and oj^ are the routes associated to 93 and 4>.
(2) If if>: /->- AT is a simplicial map and 93: /' -> K is a simplicial approxima-
. E
tion to ^ on a finer subdivision of I, then u>^ ~ a>0.
For, by (2), w = u>9a ~ w0o, and, by (1),

E
^ OJ01 = “>evo = 14)4) |-
E
Proof of (1). Since ~ is an equivalence relation, it suffices to prove that
contiguous simplicial maps have this property. So suppose 93,«/»: /' —>■ K are
contiguous. Now
“>(4)95(4)||95(4)93(4)| • • * |9>(4-iM4)|,

and
— I ^(4)^(4) 11 ^(4)0(4) I • •4 l*A(4-i)i/'(4)l-

99
Hence
aâ^-1 = |<p(/0)<p(/1)l'' ■ |95(4-i)93(4)I li/'(4)iA(4-i)l ■ ■ ■ l'/,(4)lA(4)l-
But since 93 and </< are contiguous, ( -1), 93(4), </<4-i)> and ip(tk) are vertices
93 4
of a common simplex. Moreover 93(4) = «A(4) = 4- Hence
19>(4-i)9>(4)110(4)^(4-1)I ^ k(4-i)<A(4-i)|
and
~ |<p(4)<p(4)l • • • 19^(4-2)9?(4-1)11<p(4-i),A(4-1)110(4-i)*/'(^fc-2) 1
• • •|iA(4>/,(4)|-
Similarly, since 93 and </> are contiguous,
19^(4 - 2)9>(4 -1) 119^(4 -1)’/’(4 -1) I — 1<p(4 - a)0(4 -1) I
and
193(4 - 2>/<4 - x) 11 ^(4 -i)0(4- 2) I ^ 19^(4 - 2)0(4 - 2) U

so that
~ |9>(4)<p(4)| • • • |93(4—3)9,(4—2)119^(4-2)^(4-2)11*A(4-2)^(4-3)I
• • • |</'(4)i/'(4)l-
Continuing in this way we find, by induction, that
E
OW-1 - |<K4)!/'(4)| = K»o|.
Proof of (2). Since the restriction of ifj to the subcomplex of /' consisting of
the vertices {4, 4,..., 4} is a simplicial map, 9>(q) = <p(tt) for i e (0, 1,..., k}.
Moreover, since </<: /-» K is a simplicial map, (0(4, ti +1)) is a simplex (5) in K
of dimension 0 or 1.
Claim: For each vertex u of I' with tt < u < ti+1, 9?(«) is a vertex of (s).
For, in fact, since 93 is a simplicial approximation to 0,
m 6 0(Str (“)) c StK (9°(m))-

Since 1Jj(u)e(s), (s) n St 93(1/) ^0, so (s) c St 93(h), and 93(h) must be a
vertex of (s); that is, cp(u) is equal either to 0(4) or to 0(4+1) as claimed.
Thus, if u0 = ti <«!<•••< «r = tt + l are the vertices of 7' between
and fi + 1, then {99(14), 93(4),..., 93(4)} are vertices of a common simplex of K.
Note that 93(14) is a vertex of (s) because 93(14) = 93(h) = <A(h); similarly
93(4) is a vertex of (s).
Now consider the parts of 04 and 04 arising from the restrictions of 0 and
93 to [4 ti +1]. This part of 04 is just |</>(h)^(h + i)|- The corresponding part of
04 is
|93(w0)93(Wi)| [93(4)93(401 • * • \<P(Ur-l)<p(Ur)\-
100
Since {<p(«0), • • •, <p(«r)} are vertices of a common simplex of K, this is edge

equivalent to |<p(w0)<p(«r)| — |<p(fi)<p(fi + i)| = |i/<(/‘j)i/<(ti + 1)|. Thus these parts of
and are edge equivalent. Since this is true for each i, ~ □
Corollary. Let K be a simplicial complex, let v0 e K°, and let i: K2 -> K be the
injection of the 2-skeleton of K into K. Then i induces an isomorphism
i*- E(K2, v0)Ê(K,v0).
Consequently, the induced map
U- ”i([K2], v0) -> v0)

is an isomorphism.
Proof. The definition of edge equivalence depends only on K2. □
Theorem 6. The n-sphere Sn is simply connected for n > 1; that is, ■n1(Sn,p)
— (e) for each p e Sn.
Proof. First note that Sn is homeomorphic to the /z-skeleton of an {n + 1)-

simplex. In fact, if s is an (n + l)-simplex in Rn + 1, the following map
<P: [s'1] —> Rn +1 maps [^n] homeomorphically onto Sn c Rn + 1. Let b =
(bi,..bn + j) e Rn + 1 be the barycenter of s. For x = (xl5..xn + 1) e [jn],
define <p(x) e Rn + 1 by
rn+1 "11/2 C^l *2 • • •> %n +1 ^n + l)

[ (Z - bd2\
Geometrically, [^] may be regarded as inscribed in Sn, and 9? is projection
outward from the barycenter of [s].
It suffices then to show that ui([jn], v0) = (e). By Theorem 5, every element
of 7Ti([5,n], y0) has a representative a which is a route, and, in particular, its
image lies in [s1]. If n > 1, then there exists a point p e [5"] with p $ [s1]. But
[■*"] — {p} is homeomorphic with Rn, which is contractible. Hence a ~ eVo. □
Definitions. A graph is a simplicial complex of dimension less than 2. A tree is

an arcwise connected graph T such that, for each 1-simplex s e T, [T] — (5)
is not connected (see Figure 4.18).
An end of a graph is a vertex which is the vertex of at most one 1-simplex.
Remark. Every tree has an end. For otherwise we could build up a route by
starting at one vertex, moving to another vertex along a 1-simplex, moving
to a third vertex along a different 1-simplex, etc. The route never touches a
vertex twice because otherwise the 1-simplex which brings the route back to
that vertex could be removed without disconnecting the tree. If the route
never reaches an end, we will touch infinitely many vertices in this way
(induction). But a complex has only finitely many vertices.
101
Figure 4.18
Theorem 7. Every tree is contractible.
Proof. By induction on the number of vertices. If T has 1 vertex, the theorem

is trivial. Assume the theorem for trees with n vertices. Let T have n + 1
vertices, and let v0 be an end of T. Then there exists a unique 1-simplex s sT
with vertex v0.
Let L = T — {(5), v0}. Then L is a simplicial complex, and [L] = [T] —
(s) u {y0}. L is a tree because if t is a 1-simplex in L such that [L] — (t) is
connected, then [T] — (t) would also be connected. Now L has only n
vertices, so L is contractible. Moreover, [L] and [T] are of the same homotopy
type. (Let /: [T] -> [L] map (5) u {t>0} into the other vertex of s and map L
onto itself. Let g: [L] —> [T] be inclusion. Then f0 g — ilL) and g °f ~ im.)
Hence [T] is contractible. □
Corollary. Let T be a tree and let v0 be a vertex of T. Then
*i([n »o) = E(T, v0) = (e).
Definition. Let K be a graph. Let a0 be the number of vertices of K, and let cq

be the number of 1-simplices. Let x(K) = a0 ~ «i- The integer x(A) is
called the Euler characteristic of K.
Remark. Note that the integer x(^) is invariant under subdivisions,
because inserting an extra vertex into K splits some 1-simplex into two
1-simplices, so that a0 and both increase by one.
Theorem 8. Let T be a tree. Then y(T) = 1.
Proof. By induction on the number n of vertices of T. For n = 1, the theorem

is clear. Assume the theorem for trees with n vertices, and let T have n + 1
102
vertices. Let L be the tree obtained in the proof of Theorem 7. Then L has n
vertices so X(L) = 1. But a0(T) = «0(L) + 1, and Ul(T) = afL) + 1, so
x(T) = X(^) = L □
Theorem 9. Let K be an arcwise connected graph. Let n be the maximum

number of open 1 -simplices which can be removed from K without disconnect¬
ing the space, (n is the number of “basic” circuits in K.) Then n = 1 - X(K).
Proof. If K is a tree, then n = 0, and Theorem 8 applies. If K is not a tree, let

Oi) be an open 1-simplex such that [W] - (sf is connected. If K - fo) is a
tree, stop. Otherwise, let (s2) be an open 1-simplex such that [/f] — (s1) u (s2)
is connected. Continue. Since there are only finitely many 1-simplices in K,
the process must stop; that is, for some n, K - {(^), (j2),..., (jn)} is a tree T.
Then
x(K) = X(T) - n = 1 - n;
that is,
n = 1 - x(K). □
(Note that the above formula implies that although the particular 1-
simplices which we delete are by no means unique, the number which must be
deleted to obtain a tree is independent of the particular method of deletion
employed.)
Remark. Recall the definition of the free group Fn on n generators. Consider
an alphabet consisting of n letters ol5 a2,..., an. Consider the symbols
a1~1,a2-1,...,an-1 and e.
Let S be the set of all “words” obtained by arranging these symbols in any
order in a row of finite length—repetitions are allowed. The “product” a/3
of two words a and /3 is defined by juxtaposition: /3 is attached to the end of a.
The “inverse” of a word is obtained by reversing the order of the arrange¬
ment and at the same time replacing ay by af1, af1 by ah and e by e. An
equivalence relation ~ is defined on S as follows. We decree that ee ~ e and
that for each j,
OjOj 1 ~ e. lai ~ e;
OjC ~ Qj, af1e ~ ai
eaj ~ fly, eafr ~ aj
Furthermore, any two words are ~ equivalent if one may be obtained from
the other through a sequence of such “elementary” equivalences. The set of
equivalence classes forms a group with multiplication and inverse as above,
and with identity the equivalence class of e. This group is Fn.
Example. The free group on one generator is isomorphic with the integers.
Remark. For n > 1, Fn is not commutative. For, in fact, a1a2a1~1a2~1 ^ e.
103
Remark. If Fn is a free group with generators au a2,..an, and G is any

group, then any map h: {ax, a2,.. ■, an} G can be extended to a homo¬
morphism h: Fn-+ G. The homomorphism h is defined by
■ Qj.*1) = Ka,yK
Moreover, this property characterizes the group Fn; that is, if H is a group
generated by n elements such that any map from these generators into an
arbitrary group extends to a group homomorphism, then H is isomorphic
with Fn.
Theorem 10. Let K be an arcwise connected graph, and let v0 be a vertex of K.

Then ttx([K], v0) is isomorphic with the free group on n = 1 — y(A")
generators.
Example. Let px and p2 be distinct points in R2. Then for p e R2 — [px, p2}>
■nx(R2 — {px, P2}, P) is the free group on two generators. For consider a graph
K as in Figure 4.19. (Note that px and p2 are not part of the graph.) Then
R2 — {px,p2} and [/f] are of the same homotopy type. In fact, the map
104
— {Pu P2} -> [Ai] defined by projection, as in Figure 4.20, together with
the inclusion map [Ai] -> R2 — {px, p2} give a homotopy equivalence. Hence
tti(A2 - {/?!, p2}, p) = -n^K], v0).
But x(K) = 5 — 6 = — 1, so n — 1 — (— 1) = 2 and, by Theorem 10,

"1 ([AT], r0) £ F2.
Proof of Theorem 10. We shall construct homomorphisms
h: E(K, v0) -+ Fn,
hi - Fn-> E(K, v0)
such that hx ° h and h ° hx are identity maps. This will show that E(K, v0) is
isomorphic with Fn. The theorem is then a consequence of Theorem 5.
Construction of h. Let (sx), (s2),..., (j„) (n = 1 — x(X)) be open 1-
simplices of K such that T = K — {(5X), (s2),..., (^n)} is a tree (cf. Theorem 9).
Let Fn be the free group with generators (^x), (j2), • • •, (sn). For each
j e {1, 2,let 5;+ be the edge \vjv'j\ in K, where Vj and v) are the vertices
of Sj. (We are here implicitly choosing some ordering for the vertices of each
Sj.) Let Sj~ be the edge \v'jVj\. Then each route a> in K is of the form
w = PiSh*PaPta* ■ ■ ■pkSjk±pk+1,
where each pt is a route (possibly the trivial route |r>;,y;(|) in the tree T. Now set
Koi) = (Sj1yl(Sj2yl--fSjky\
We must check that h is well defined; that is, that h(o>) depends only on the
edge equivalence class of <0. For this, it suffices to show that if ojx and o>2 are
routes in K which differ by an elementary edge equivalence, then =
h(a>2). So suppose
COl = o\v1V2\\v2V3\t,
w2 = o\vxV2\t,
where o and r are routes in K, and vx, v2, v3 are vertices of a common simplex
in K. Since A" is a graph, and hence has no 2-simplices, either vx = v2 — v3,
Vl = v2, v2 = v3, or vx = v3. In each of the first three cases, at least one of the
simplices (vx, v2) and (v2, v3) is a 0-simplex, hence is not an (sf), and the other
is equal to (vu v3). Thus in each of the first three cases, /2(wx) = h(oj2). In the
fourth case,
<t>! — o\ViV2\\v2Vi\t
<jj2 -- cr\ViVi | r.
If (vx, v2) is not an (s}), then clearly h(o>x) = h(oj2). If (vx, v2) = (sf) for somey,
then
U)1 = OSj^Sj^T,
h( tux) = h(o)(Sj)±1(Sj)Tlh(T) = h(cr)eh(r) = h(o)h(T) = h(a>2).
105
Thus in all cases, h(a>*) = h(co2) as required, and h is well defined. Clearly, h
is a homomorphism.
Construction of hx. Since Fn is a free group, it suffices to define h1 on the
generators (sy) = (vy, vy). For this, let oy be a route in the tree T from v0 to vy,
and Tj be a route in T from v0 to v). Define hf{sf) to be the edge equivalence
class of the route ajsj + rj~1. This definition is independent of oy because any
other route in T from v0 to vy is edge equivalent to oy. (T is simply connected
by Theorem 7.) (Proceed similarly for ry.) Now hx extends uniquely to a
homomorphism Fn -> E(K, v0).
h o hx is identity because, for each generator (s,) of Fn,
h o hx((s})) = h(ojSj+r;) = (sf.
hx ° h is identity because if
w = PlSh±P2S)2± ■ ‘ ■pkSile±Pk +1
is a route in K, then
^ ° h(a>) = /i1((^1)±1(â)±1- • •(5*fc)±1)
= (ahSh + Th ~1)± V/A + 'ri2~1)±1--‘ + rik~1)±1
= yhish±v'hVi2Sj2±'n'i2- ■ 'V/kM
where
(ojl (if sj{ appears as sy +)

Vu = J\ re
KTJt (it sjt appears as sj{ ~)
and
, f Tu ~1 (if su appears as sJ{+)
^ = 1lcr„ _! re
(it sh appears as syf).
But pi and Vh are both routes in T from v0 to the origin of hence they
are edge equivalent. (T is simply connected.) Similarly, p2 and r}'ji7]j2 are both
routes in T from the end of syi± to the origin hence they are edge
equivalent. Continuing by induction, we conclude that hx °h(u>) ~ cu; that
is, hi o h is identity. j-j
Corollary. Let K be a simplicial complex. Then the fundamental group

vo) (Vo a vertex of K) is in a natural” way a quotient group of a free
group.
Proof. We may assume that K (and hence K1 also) is arcwise connected. Let
i: K1 K be injection. Then
W-E{K\ v0)->E(K,v0)
is surjective from the definition of the edge path group. Hence, by Theorem 5,
**: 77([/C], v0) TT^], v0)
106
is surjective. Let n — 1 — xC^1)- Then 771([A'1], v0) = Fn, the free group on n
generators, so
i*' Fn *l([XL t-’o)
is surjective. Let H be the kernel of /*. Then
ttx([K], v0) ~ FJH. □
Remark. Regarding E(KX, v0) as Fn, the subgroup H is the subgroup

generated by routes of the form pi\v1v2\\v2v3\\v3v1\p2~1, where and p2 are
routes in the tree T from v0 to vu and (t?l5 v2, v3) is a 2-simplex in K.
Corollary. Let D2 = {(x, y) e R2; x2 + y2 < 1} be the unit disc in R2. There
exists no continuous map f: D2 -» S1 such that/|si is the identity.
Proof. Suppose such an/exists. Let g: S1 -> 1,

D2 be inclusion. Then f ° g = is
so that (f° g)*: iTiiS1, 1) -> ^(S1, 1) is the identity map. But (/°g)* =
/* ° g*, and 7r1(D2, 1) = (e) since D2 is homotopic to a point. Then
Im (/o g)* = Im (/* o g*) c Im/* = (e).
Since (f ° g)* is surjective, we conclude ^(S1, e) — (e), which contradicts

Theorem 10. □
Corollary. Special case of Brouwer fixed point theorem.) Let D2 be the unit disc
in R2. Suppose f: D2 -> D2 is continuous. Then f has a fixed point; that is,
there exists an x e D2 such that f(x) — x.
Proof. Suppose there exists no fixed point. Then, for each x e D2,f{x) ^ x,
so that x — /(x) ^ 0 (vector addition in R2). Let g: D2 —> S1 be defined as
follows. For each xe D2, g(x) is the projection of f(x) onto S1 along the
vector x — f(x) (see Figure 4.21). Then g is continuous, and g|si is the
identity. This contradicts the previous corollary. □
gf)
107
Remark. These two corollaries admit the following generalizations to

higher dimensions. (1) There exists no continuous map from the «-disc Dn
(closed ball in Rn) onto its boundary (an (n — l)-sphere S'"-1) whose restric¬
tion to 5'n_1 is the identity. (2) Every continuous map from the closed n-disc
Dn into itself has a fixed point. However, 5'n_1 is simply connected for n > 2,
so the above proof breaks down. The fundamental group in the proof must
be replaced by another topological invariant, the (« — l)-th homology group.
In the case n = 1, the analogues of these corollaries are consequences of the
connectedness of D1 — I.
108
Manifolds
5.1 Differentiable manifolds

Definition. A locally Euclidean space X of dimension n is a Hausdorff topo¬
logical space such that, for each x e X, there exists a homeomorphism cpx
mapping some open set containing x onto an open set in Rn.
Remark. We may, if we wish, choose each cpx so that <px(x) = 0 and so that
the image of cpx is a ball i?0(Y). Given any (px homeomorphically mapping an
open set U about x onto an open set in Rn, let e > 0 be such that B^fe) <=
<Px(U). Let
B9lx)(e) -> B0(e)

be translation by — <p(x). Then
*Px = ° x | <Px ~ (*)(£))
maps cpx~1(Btl)(x)(e)) homeomorphically onto B0(e).
Example 1. Rn is locally Euclidean. For each x e Rn, take cpx to be the identity
map.
Example 2. Sn is locally Euclidean. Given x e Sn, let y e Sn, y ^ x. Then

cpx = stereographic projection from y maps Sn — {j} homeomorphically
onto Rn.
Example 3. Projective space Pn; that is, the space of all lines through 0 in
Rn + 1, is locally Euclidean. For since Pn is covered by Sn, each xePn is
contained in an open set homeomorphic to an open set in Sn that itself
contains, about each of its points, an open set homeomo phic to an open set
in Rn.
109
5: Manifolds
Example 4. Each open subset U of a locally Euclidean space X is locally

Euclidean. For if jc e U, let tpx be a homeomorphism mapping an open set
about x in X onto an open set in Rn. Take cpx = ipx\undomam
Example 5. The set of all nonsingular k x k matrices forms a locally Euclid¬

ean space of dimension k2. Each k x k matrix may be identified with a k2-
tuple by stringing out the rows in a line. The nonsingular matrices then form
an open set of Rk2, namely A-1^1 — {0}) where A: Rk2 -» R1 is the deter¬
minant function.
Definition. A Ck-differentiable manifold of dimension n is a pair (X, ) where

X is a Hausdorff topological space, and O is a collection of maps such that
the following conditions hold (see Figure 5.1).
(1) {domain cp}^* is an open covering of X,

(2) each maps its domain homeomorphically onto an open set in Rn,
(3) for each <p,ipe<$> with (domain <p) n (domain f) # 0, the map
i{j ° cp~1 is a C^-map from <p(domain <p n domain f) <= Rn into Rn,
(4) O is maximal relative to (2) and (3); that is, if </r is any homeomorphism
mapping an open set in X onto an open set in Rn such that, for each
cp s O with domain <p n domain ^ # 0, ip ° <p-x and <p ° 0-1 are Ca¬
rnaps from
<p(domain cp n domain f) and (/((domain cp n domain f)
into Rn—then ip e <h.
Here k may be 0, 1, 2,..oo, to. C° means continuous. Ck for k finite

means all partial derivatives of order less than or equal to k exist and are
continuous. C°° means all partial derivatives of all orders exist and are
continuous. C“ means real analytic; that is, the function may be expressed
as a convergent Taylor series in a neighborhood of each point.
110
5.1: Differentiable manifolds
Note that a Cfc-manifold is a locally Euclidean space and a locally

Euclidean space gives rise to a C°-manifold.
If n = 2 and, in Condition (3), “C*” is replaced by “complex analytic”
(where R2 is identified with the complex numbers C1), (X, O) is called a
complex analytic manifold of complex dimension 1 or a Riemann surface.
O is then called a complex structure or conformal structure on X.
The maps is called a coordinate system on the open set (domain <f) <= X. For
x e X, a coordinate system about x is a coordinate system such that
x e domain cp.
Remark. Each of the above Examples 1, 2, 3, and 5 of locally Euclidean

spaces form the underlying space of a C“-manifold. You need only check
that the maps <px satisfy Condition (3) for a manifold, and then take d> to be a
maximal set containing {<px}xeX- Example 4 above also carries over to mani¬
folds. Namely, if (X, O) is a C^-manifold and U is an open set in X, then
(U, <E|c;) is a Cfe-manifold, where = {<p|[7}4)6,]>.
Definitions. Let (X, O) be a Ck-manifold. A real-valued function /: X —> R1

is a Cs-function (5 < k), denoted/e C\X, R1), if, for each cp e d>, /° <p_1
is a Cs-function mapping the image of cp <= Rn into R1.
Let (X, O) be a Ck-manifold, and let x e X. A real-valued function/is
said to be of class Cs (5 < k) in a neighborhood of x, denoted
fe C%X, x, R1), if
U = (domain /)
is an open set in X containing x, and /e C\U, R1), where U has the Ck-
manifold structure as an open set in X.
Remarks. Note that we are able to define Cs-functions on X because (1) X

looks locally (via the coordinate systems <p e O) like Rn, and we know what it
means for a function on Rn to be Cs; and (2) if U = domain cp and V =
domain ift for <p, e O, with U n V f 0, the concept of a CMunction in a
neighborhood of x in U n V is the same relative to the coordinate system cp
as to the coordinate system */>, because ^^-Msa Cfc-homeomorphism and
k > s.
Note also that if / and g are Cs-functions in a neighborhood of x, then
f + g and fg (product) are CMunctions in a neighborhood of x, where
domain (/ + g) = domain (fg) = (domain/) n (domain g).
Definition. Let (X, <£) be a C^-manifold, and let cp e $ be a coordinate system

on U = domain cp. Let r, : Rn R1 be the y'th coordinate function on Rn;
that is, rfau a2,..., an) = aj for (ax, ...,an)e Rn. The yth coordinate
function of the coordinate system cp is the function xp. U R1 defined by
xj = n o cp.
111
5: Manifolds
Remark, xy. [/-> R1 is a Ck-function. The n-tuple of functions (xl5. .xn)

is sometimes also referred to as a coordinate system.
Definition. Let (X1, ®x) and (X2, ®2) be Cfc-manifolds (not necessarily of the
same dimension). A mapping Y: X1 -> X2 is of class Cs (s < k), denoted
Y e Cs(Xx, X2), if, whenever/6 CS(X2, R1), then/«Te CS(X1, R1).
Exercise 1. Show that, if Y: X1 X2 is of class Cs (s > 0), then Y is

continuous.
Remarks. We shall confine our attention to C“-manifolds. This will

include, in particular, (^“-manifolds and complex analytic manifolds of
dimension 1. We shall use the word “smooth” to denote C“.
We now proceed to define the concept of tangent vector on a manifold.
Recall that, in Euclidean space, a vector at a point defines a map which sends
each smooth function into a real number, namely, the directional derivative
with respect to the given vector. Moreover, the vector is determined by its
values on all smooth functions. We shall use this property to define tangent
vectors on a manifold.
Definition. Let (X, O) be a smooth manifold and let x e X. A tangent vector at

jc is a map v: Cx(X,x, R1) -> R1 such that, if cp is a (fixed) coordinate
system with x e U = domain cp, then there exists an rc-tuple (al5 a2,..., an)
of real numbers with the following property. For each /e C”(Z, x, R1),
v(f) = 2 a* Jr(/0<P_1)Uw
i=1
(Note that if W = domain/, then cp and/are both defined on the open set
U n W containing x, so that /° 99_1 is a smooth function with domain
cp(U n W) c: Rn containing <p(x).)
Remark. If v: Cro(A, x, R1) -> R1 has the property required above of a

tangent vector with respect to one coordinate system cp about x, then it also
has this property with respect to any other coordinate system about x. For,
if i/> is another such coordinate system, then, using the chain rule,
<0= u,
i=x vr'
= 2 Qi Jk f° ( iA_10 ^0 <p_1)Uu)
i = 1 Fi
n 71 d
= 2ai 2 gr(/o^"i)i*<*)î(ô9>~i)u*),
i=1 3=1 't
where Ju(ip 0 95”1) is the Jacobian matrix of the function if> ° cp"1. Lienee
VW) =2(2 aiJ^ 0 95~1)U*>) -ir (/° 'A_1)Uu)-

3=1 \i = i / vri
112
Setting
n
bi = 2
i=1
aiJiM ° <P"1)U(x),
we obtain
K/>= 2 4,
i= 1 'i
Thus, to check if v is a tangent vector at x, it suffices to check the required

property in any one coordinate system at x.
Notation. Given a coordinate system cp about x, let xy = rf ° <p denote the

yth coordinate function of <p. By 8/8xf (J = 1,...,«) is meant the tangent
vector at x defined by
~(f) =
for fe CCC(X, x, R1). Thus 8/8x,- corresponds, relative to the coordinate
system <p, to the «-tuple (0, 0,..1,..0), where the 1 is in the yth spot.
Remark 1. If xu ...,xn are the coordinate functions of a coordinate

system <p about x, and yu ..., yn are those of a coordinate system </< about x,
then the above computation shows that
Remark 2. A tangent vector v at x e X has the following properties. For

any f,ge Cco(Tr, x, R1) and for A e R1,
(1) v(f+ g) = v(f) + v{g)

(2) v(Xf) = Xv(f)
(3) v(fg) - v(f)g(x) + f(x)v(g).
These three properties say that the map v: C™{X, x, R1) —> R1 is a derivation.
Moreover, these properties characterize tangent vectors; that is, we could
have defined a tangent vector to be a map v: Cco(A, x, R1) -» R1 satisfying
(l)-(3) above, and then proved that, relative to any coordinate system 95 about
x, v — 2i"=i ^(S/SXj) for some «-tuple (alf..an) of real numbers, where x(
is the ith coordinate function of <j>.
Remark 3. The set Xx of tangent vectors at x forms a vector space under the
following rules of addition and scalar multiplication:
Oh + v2\f) = + v2(f) (»i, v2 e Xx),

(XVl)(f) = X(Vl(f)) (VleXx,XeRi).
113
5: Manifolds
To see that v1 + v2 and Xv± are tangent vectors at x, let <p be a coordinate
system about x, with coordinate functions (xi,..., xn). Then
n Jk
v± = V a^d/dXi) and v2 = bi(8f8Xi)
i= 1 i=1
for some (alt ...,an) and (bltbn). It is then easy to check that
n
_a_
t?i + V2 = 2 (°i + bi)
0X,’
i=1
2 (M> Sx/
i= 1
The map (alt..., an) -> Oi(d/dxt) gives a vector space isomorphism
Rn -> Xx, so has dimension n. Moreover, it is clear that {d/dxi}ie{1.n} is a
basis for Xx. The space Xx is called the tangent space to X at x. It is also
denoted by T(X)x or by T(T, x).
For (p and </« two coordinate systems at x, with coordinate functions
(x1}..., xn) and (^l5 ...,yn) respectively, the formula
_a_
8y(
merely expresses the vector d/dxj in terms of the basis ie{1.n}. Thus the
change of basis matrix from the basis {8/dy^ of Xx to the basis {8/8xJ is
precisely the Jacobian matrix ((0/9xy)(.yi))-
Remark 4. The tangent space T(Rn, a) to Rn at a point a e Rn is naturally
isomorphic with Rn itself. The isomorphism Rn T(Rn, a) is given by
(A,.An) —, j Ai S-.
j= 1 V’i
Notation. We shall henceforth omit the O from our notation for a differen¬
tiable manifold {X, <h). To be sure, a locally Euclidean space X may have two
or more distinct differentiable structures on it (or it may have none), but we
shall denote a manifold (X, ®) merely by X and shall assume that a definite
differentiable structure is given on it.
Definition. Let X and Y be smooth manifolds. Let T: X —>• Y be a smooth

map. The differential of Y at x e X is the map JY: Xx -> YTU) defined as
follows. For v e Xx and g e CX(Y, Y(x), R1), (JY(y))(g) = v(g ° Y).
Remark. It is easily checked that e?Y(u) is indeed a tangent vector at Y(x).

For, if • • •> Tm), then v = 2?=i Oi(d/dXi) for some real numbers a{\ and if
g e C”(y, Y(x), R1), then
idX¥(v)](g) = v(g o Y) = âi~(go Y)

i =i
âi^r(SOT 1 ° t o Y o cp !)| «>(JC)

i=i i
ill ^ ^
2 Qi 2 nr(S° ’■-1)|T.,(X)=-(JJ0 roT o<p-i)| <P(x)

i=i j = l Udj °'i
[(■Si,..sm) coordinates on i?m]
W,(s) (y‘ °
n 171 r\ r\
= 2 2 a‘ a* T)
2 <y> - *> S7 (*)•

.1 = 1 dY)_
Since this holds for all g e C™(Y, Y(.x), R1),
and, in particular, dx¥(v) is a tangent vector. Furthermore, it is clear that dY

is a linear transformation Xx —> FTU). Since
this linear transformation dY has matrix
relative to the bases {8/8xi}ie{1.n) and {dldy,}j€{1.m).

Remark. Let X, Y, and Z be smooth manifolds. Let Y: X —>■ Y and
O: Y —>Z be smooth maps. Then J(0 ° Y) = dd> ° d'Y.
Proof. Suppose v e Xx and h e C™(Z, )
= [dd>(dY(v))](h)
= [(d<& o (PV)(v)](h). □
Remark. Let X be a smooth manifold, and let U be open in X. Then U is

itself a smooth manifold. Moreover, the inclusion map i: U -> X is a smooth
115
5: Manifolds
map. Indeed, fe C°°(A, R1) implies f\v e C°°(t7, R1). Furthermore, the
differential
di: T(U, u0) -> T(X, M0) («0 e 17)
is an isomorphism; we shall identify these two linear spaces.
Exercise 2. If u0 e U an open set in X, construct a function h e CX(X, R1) such

that
(xeW an open set containing u0),

(x $ U).
[Hint: Make use of the smooth function g: R1 —v R1 defined by
(t > 0)
g(t) =
(t < 0).]
If fx e CX(U, w0, R1), use Exercise 1 to show that there exists a smaller open
set W and fe C°°(Z, R1) such thas f\w = fx |w.
Remark. Let X be a smooth manifold, and let /e C°°(Z, i?1). Let us com¬
pute df. For v e T(X, x), <77(y) e T(R1,f(x)). Since T(R1,f(x)) is 1-dimen-
sional, df(v) = A (d/dr) for some A e R1. To determine A, it suffices to evaluate
df(v) on the coordinate function r : R1 -> R1 as follows.
A = (r) = [df(v)](r) - v(r of) = v(f).
Thus df(v) = v(f) (d/dr). Now T(R1,f(x)) is naturally isomorphic with R1

via the isomorphism A(d/dr)-> A. Let us identify these two spaces through
this isomorphism. Then df: T(X, x) -> R1 is a linear functional on T(X, x);
that is, df is a member of the dual space T*(X, x) and is, as such, given by
df(v) = v(f) (v 6 T(X, x)).
T*(X, x) is called the cotangent space at x.
Definition. Let Xbe a smooth manifold. A smooth curve in A is a smooth map

« from some (open or closed) interval c= R1 into X. If the domain of a is a
closed interval [a, b], smoothness of a means that a admits a smooth
extension
a: (a — e, b + e) X.
(Note that open intervals are open sets in jR1 and hence are smooth
manifolds.)
A broken Cx-curve in A is a continuous map a: [a, b] -> X together
with a subdivision of [a, b] on whose closed subintervals a is a C00 curve.
116
Example.
(/, / sin \jt) (t g (0, 1 ])

(0, 0) (t = 0)
is not a smooth curve in R2 because it admits no smooth extension past 0.
Definition. Let a: / —»- X (/an interval cij1) be a smooth curve in X. The

tangent vector to a at time t (t e /), denoted by d(f), is defined by
Note that d(f) is well defined, even at the endpoints of /.
Remark. Given a tangent vector v e Xx, let a: / —> X be a smooth curve

whose tangent vector at time t = 0 is v. (Such a curve may be obtained by
taking a coordinate system <p about x, finding a curve (for example, the
straight line) in Rn whose tangent vector at time 0 is d<p(v), and pulling this
curve back to X by <p_1.) Then, for /e CX(X, x, R1),
Thus v(f) is the derivative of the “ restriction ” of / to the curve a. Moreover,

two curves ax and a2 have the same tangent vector v at time 0 if and only if
«i(0) = a2(0) and
for all /e Cm(X, x, R1) (see Figure 5.2). We may use this equation to define
an equivalence relation on the set of all curves a with a(0) = x. Then we get a
one-to-one correspondence between equivalence classes of curves through x
and tangent vectors at x. Thus, we could have defined a tangent vector at x
to be such an equivalence class of curves through x.
Figure 5.2
117
5: Manifolds
5.2 Differential forms

Definitions. Let X be a smooth manifold. Define (see Figure 5.3)
T(X) = IJ T(X, x) and T*(X) = (J T*(X, x).

xeX xeX
T(X) is called the tangent bundle of X. T*(X) is called the cotangent bundle
of X.
T(X,x 0 T(X,x2)
A projection map v. T(X) -> X is defined as follows. If v e T{X), then

v e T(X, x) for some (unique) xe X; set tt(v) = x. Similarly, there is a
projection map from T*(X) onto X that we shall also denote by w.
A vector field on A is a map V: X -> T(X) such that v o V = ix. A
vector field V is smooth if for each/e C°°(X, R1), Vf e C”(Z, R1). Here Vf
is defined by
(f//)(x) = V (x)/.
A differential \-form on A is a map a>: X T*(X) such that tt o o> = ix.

A differential 1-form a> is smooth if for each smooth vector field V on X,
a>(F)eCco(JF, i?1).
Here o>(K) is defined by (co(K))(x) = w(x)(F(x)). We shall denote the set
of all smooth vector fields on X by C00(Ar, T(X)) and the set of all smooth
1-forms by C°°(X, T*(X)).
Exercise. Define a manifold structure on T(X) so that n is a smooth map and

so that a vector field V is smooth if and only if it is a smooth map from X —»■ T(X).
[Hint: For <p: [/—> i?n a local coordinate system on X, with coordinate func¬
tions (xi,..., xn), define <p: 7r-1((/) -> R2n by
V(v) = (cp o bx,..., bn),
where bx,..bne R1 are such that v = 2"=i bt S/Sx,.]
Remark 1. Let fe C°°(A, R1). Then dfeCa(XtT*(X)). For if

Ve C°°(A, T(X)), then df(V) = Vfe C°°(A, R1).
118
5.2: Differential forms
Remark 2. CX(X, T(X)) and C®(A", T*(X)) are both vector spaces over
the reals under the operations of pointwise addition and scalar multiplication.
For example, if Vx and V2 e CX(X, T(X)\ then V1 + V2 is defined by
(yi + F2)(x) = Vx(x) + V2(x); and if A e R1, then AJ0 is defined by
(AFi)O) =
Remark 3. Let y be a coordinate system on X with domain U and coordi¬
nate functions (x1; x2,..., xn). Then the following hold.
(1) (d/dxi) 6 C°°(t/, T(U)) for ie{l,...,n}. d/dXi is smooth because if
fe C™(U, R1), then fo e C“(<p(t0, R1),

and, for each x e U,
(*) = (/° 9 X) (cp(x))
o cp
(*);
that is,
° <P e CX(U, R1).
(2) If Ve C°(U, T(U)), then there exist functions at e CW(U, R1) for
i e{l,..., «}, such that V = 2"=i a^d/dxf). These functions ax exist because
{(8/dXl)(x)}ie{1.n}
is a basis for T(X, x). They are smooth because (d/dx^x,) = 8Xj, so that
Uj = 2 a(Siy = 2 ai ^ (*/) = y(xJ e C"(U, R1).

i=1 i=1
(3) If Ve C"(Z, 7X20), then V\ve C°°(f7, T(U)) by a previous exercise,

and V\v = 2?=i Oi(d/dXi) as in (2) with € C°°(U, R1).
(4) dxj e CX(U, T*(U)) for/' e{l,..., «} because xf e CX(U, R1). Further¬
more, {dxj} is at each point the dual basis to {d/dx,} because
dx{t) = ty* - 8»-

(5) If co e CîU, then there exist at e C°°(i7, R1) such that
w = 2?=i These functions ax exist because {dx{} is at each point a basis
for the cotangent space. They are smooth because
(6) If fe C°(U, R1), then
119
5: Manifolds
because df — 2"= 1 Qj dx} for some af, and
ai = ^2 dXj{jf) = df{dx)
ai = dXi
We have just seen that if f e C°°(A, R1), then df is a smooth differential

1-form. We now introduce differential A;-forms.
Review of exterior algebras

Let V be an ^-dimensional vector space over the reals. Then the following
hold.
(1) The vector space Afc(F*) is the space of all skew-symmetric ^-linear
functions on V; that is, each r e A.k(V*) is a map r: F © • • ■© V-> R1 such
'-V-'
that for all vu ..., vk, v) e V, A e R1, fc_tlmes
(1) t(vx,. . Oy_i, Vj + v), vj + 1, ...,vk)
= r(vlt . . Cy-i, Vj, »y + 1, . . Vk) + t(vu . . ., I7y_!, uj, ®y + i, . . tffc) 5
(2) r(yn(1),..., yn(fe)) = (-1 y-r{vx, ...,vk); and
(3) r(pi, . . Oy-i, Ayy, »y + 1,. . yfc) = Ar(y1? . . ., V„ . . l>fc).
where tt is any element of the permutation group Sk on k letters, and (— l)n
is +1 if 77- is an even permutation or — 1 if -n is an odd permutation. This
second condition is equivalent to requiring that if two vectors in the argument
of r are interchanged, then the value of r on these vectors changes sign. The
dimension of Afc(F*) is equal to the binomial coefficient ^ j for k < n; it is
zero for k > n.

(2) If we set &(V*) = 2"=o © Afc(F*), where A°(F*) = R1, a product is
defined on &(V*) as follows. If reAfc(F*) and fie A'(V*), their product
r a fi is the element of Ak+l(V*) defined by
T A fliVi, ..vk + l)
1 v
(k- 4- r\\ ( 0 T(Fi(l)> • • - j V]t(.k))lx(Pn(.k +1)» • • •» Fi(fc + I))-
\K X l). „esk + t
Since &(V*) is generated by such fi and r, this multiplication extends to

&(V*) by linearity, that is, by requiring that exterior multiplication A be
distributive with respect to vector addition. This multiplication is associative
and &(V*) is an algebra, with unit 1. However, multiplication is not com¬
mutative: if fie Afc(F*) and r e A*(F*), then
fi A T = ( — 1 )klT A fl.
(3) If cpi,..., <p„ is a basis for F*, then

[<Ph A • ■ ■ A <pik; l < f < i2 <■•■< ik < n]
is a basis for Afc(F*). Hence the union of these sets over k e{l,..., n}, to¬
gether with 1 e A°(F*), is a basis for &(V*). It follows that the dimension of
&(V*) is 2n.
120
If vi> —> vk 6 K the value of 9?^ A • • • A <pik on these vectors is given by
(<p(l A • • • A Vk) = T^y 2 (_1)XW' •
(4) &(V*) has the following properties:

(1) le&(V*), V* c &(V*);
(2) ^(F*) is generated by 1 and F*;
(3) 9 A <p = 0 whenever <p e V*; and
(4) dimension ^(F*) = 2".
These properties in fact characterize ^(F*); that is, if #(F*) is any algebra
over the reals satisfying properties (1)—(4), then #(F*) and ^(F*) are iso¬
morphic (as algebras).
(Note that Condition (3) is equivalent to the condition that cpx A <p2 =
—<p2 A for all <px, cp2 e F*.)
(5) If L: V* —> F* is a linear transformation, then L induces a unique
algebra homomorphism L\ ^(F*)->^(F*) which extends the map L. L
preserves degrees; that is, L: Afc(F*) —> Afc(F*). In particular, L: An(F*) ->
An(F*). Hence, since dim An(F*) = 1, there exists a scalar A such that
-£|a\K*) = ^*An(V)- This scalar A is precisely A(L), the determinant of L.
(6) The algebra ^(F*) is called the Grassmann algebra, or exterior algebra,
of V*. Elements of ^(F*) are called forms on F. Forms in Afc(F*) are said
to be of degree k.
Now let A be a smooth manifold. Let
Afc(A) = (J A\T*(X, x)),

xeX
and let
nX) = (J 9(T*(X, x)).

xeX
As usual, we shall denote the projection maps from these spaces onto A by v.
These spaces can each be given the structure of a smooth manifold such that
7r is a smooth map.
Definition. A k-form on A is a mapping /*: A-> Ak(A) such that v ° ^ = ix.

A &-form [j. on A is smooth if whenever V1}..., Vk are smooth vector fields
on A, then
f(Fi, . •Vk) e C“(A, R1),
where
KVi, ■■■, Vk)(x) = KxWfx),vk(x)).

A differential form on A is a mapping w. A-> ^(A) such that ttoW = ix;
it is smooth if its component in Ak(A) is smooth for each k. The set of
smooth A>forms on A is denoted by C°°(A, Afc(A)). The set of all smooth
121
5: Manifolds
differential forms is denoted by Cco(X, @(X)). Note that C°°(A, Ak(X)) is

a vector space under pointwise addition and scalar multiplication, and that
C°°(A, &(X)) is an algebra under the additional operation of pointwise
exterior multiplication.
Remark 1. A 0-form on X is just a real-valued function on X; it is a smooth

0-form if and only if it is a smooth function.
Remark 2. Let <p be a local coordinate system on X, with domain U and

coordinate functions (xj,..., xn). Then {dxl5..., dxnj is a basis for T*(X, x)
for each x e U. Hence
[dxh A • • • A dxik \ :x < ■ < ik\
is a basis for Ak(T*(X, x)) for each xe U. Thus, the restriction to U of each
A>form p on A can be expressed as
h< — <ilc
where each aiv..ilc is a real-valued function on U. Furthermore, p is smooth if

and only if, for each (<p, U), au...ifc e CX(U, R1). This is because
Theorem 1. Let X be a smooth manifold. There exists a unique linear map

d: CCC(X, @(X)) -» C°°(A, @(X)), called the exterior differential, such that
the following properties hold.
(1) d: CCC(X, Ak(X)) -> Cco(X, Ak + 1(X));

(2) d(f) = df (ordinary differential) for f e C°°(A, A°(AY);
(3) if p e Cco(X, Ak(X)) and t e CX(X, &(X)), then
d(p A t) = (dp) A r + (— l)kp a dr; and
(4) d2 = 0.
Remark. For the proof we need the following lemma, which asserts that
for any exterior differentiation operator d, (dco)(x) depends only on the
behavior of w in a small neighborhood of x.
Lemma. Let d: Ca>(X, &(X))—> C'X>(X, &(X)) be linear and satisfy the con¬
ditions of the theorem. Suppose to e C”(A, @(X)) is such that w\w = 0 for
some open set W <=- X. Then (da>)\w = 0. Hence, if u>, t e CX(X, &(X)) are
such that to|w = t\w for some open set W, then (da>)|w = (c/r)|w.
Proof. Suppose w\w = 0. Let x0 e W. Let /e C°°(A, R1) be such that/(x0)

= 1 and/(x) = 0 for all xfW. Then fw is identically zero on X, so that
0 = d(fcu) = (df) A ai + fdoj.
122
Evaluating at x0 gives (rfo>)(x0) = 0. Since this holds for all x0 e W, da>\w = 0.

If w|w = r|w, then (a> — r)|w = 0, so that
0 = [d(u) - t)]|w = [dw - dr]|w and da>|w = dr\w. □
Proof of Theorem 1
Uniqueness. Suppose d: C°°(T, &(X)) -> C™(X, &(X)) satisfies the con¬
ditions of the theorem. Let x e X, and let 95 be a local coordinate system about
x with domain U and coordinate functions (x1?..., xn). Let oj e C °°(X, Ak(X)).
Then the restriction of a> to U can be expressed as
“>\u = 2 •••«* dxh A • • • A dxik
for some aiv..i]c e CC0(C/, R1). Now the right-hand side of this equation is not a
differential form on X, so we cannot apply d to it. However, let U± be an
open ball containing x with U1 compact and <= jj, and let g e C°°(T, R1) be
such that g(x) = lforxe £4 andg(x) = Oforx^ U. Then <5 e C°°(X, Ak(X)),
where
& = 2 (^1-0 d(gxh) A • • • A d(gxik).
Here, by gh, for h e C “(£/, R1), is meant the smooth function on Xdefined by
, ,v v /g(x)h(x) (xeU)
te,I>W = to (xiU).
Furthermore, <l>\Ul = a>\Uv By the lemma, {d<X)\Ul = (d<l>)\Ul. Now
d“> = 2 dâhd(gxtl) A ■ • • A d(gxtl)] (by linearity)
= 2 d(gaii -iJ A d(gxh) A • • • A d(gxik)

+ 2 d(.d(gxt 1) A • • • A c7(gx,J) (by Property (3))
= 2 d(gail -ii) A ^(g*ii) A • ■ • A ^(gxijc),
since each term of the second sum is zero by Properties (3) and (4). In par¬
ticular, since g is identically 1 on Uu and since (da>)\Ul = (Jcu)]^,
u 3
(^)k = 22^ K-i*) dxi A dxh A--- A dxik.
ii<-<lk i = l 0Xi
Thus if d exists, its value at x on &-forms must be given by this formula. Since
x was arbitrary in X, and since every differential form is a sum of &>forms,
k e{0, 1uniqueness is established.
Existence. We first define d locally. Let 95 be a local coordinate system on
X with domain U and coordinate functions (xl5..., xn). (Note that U is itself
a smooth manifold.) Define dv: C°°(£/, ^(£/))—> C°°(£/, @(U)) as follows.
For
co = 2 aii-ifc A • • • A Jxifc e C°°(£/, Ak(U)),
123
5: Manifolds
define
Extend dv to C°°(t/, @(U)) by linearity. Then Properties (1) and (2) are
clearly satisfied. To verify (3) and (4), note first that each form in C °°(f/, U))
is a sum of forms of the type ah...ik dxh A • • • A dxik. By linearity of dv,
together with distributivity of exterior multiplication with respect to addition,
it suffices to check (3) and (4) on forms of this type.
Property (3). Suppose
P = ah- ik dxh A ■ • • A dxik and r = bh...u dxh A • • • A dxh.

Then
dv{p A t) = du[ail...ikbh...il dxh A • • • A dxik A dxyi A • • • A dx;i]
n
=z r=1
3
Qx , (ahd" aH - ik
3
('fiii—ii)
dxr A Jx;i A • • • A Jxy,
(n 3 \
2 A dxil A A dxA
A (bh...h dxh A • • • A fibq,)

+ (~ l)fc(a«1...it ^x4l A ■ • • A Jxjfc)
(n 3 \
2 •••;,) dxr A ^ A • • • A rfXy, J
= (^tr/Lt) A T + (—1)V A c/yT.
Property (4). For ,x = air..j)c dxh A ■ ■ • A dxik.
V 8
du2p — dL / . av ' A ^Xjj A • • • A
_r = i
z _8_
. Q = dxs
dxs A dxT A dxfl A • • • A dxjfc.
But certainly the terms in this expression with r = 5 are zero, since dxr A dxr
— 0. Moreover, for r ^ s, the equality of mixed partial derivatives on Rn
implies that
8xs 8xr ~ 8xr 8xs (a^-A
so that
_a_ 3 3 3
8xs ~Sxr (ah -0 dxs A dxr = — yyy -yy (aiv..ik) dxT A dxs\
8xr 8xs
thus the remaining terms match up in pairs which cancel each other.
124
Thus the operator dv has Properties (l)-(4). By uniqueness, every linear

operator on C®(£/, ^(t/)) having these properties must be given by the above
boxed formula. In particular, if U1 is any open subset of U, then <p| Vl is a
coordinate system, and dUx:C°{Ult <3(UX)) -► C“(C/l5 ^(C/,)) is given in
the coordinate system <p|Pl by the same formula. Thus, if u> e C°°(A, &(X)),
then
— (du(a)\ t/))| Uv
This relation enables us to define d globally by {dw)\v = du(w|u) for all
o> e C"(JT, &(X))
and any coordinate neighborhood U. This d is well defined because if U and

V are overlapping coordinate neighborhoods, then
{dU{u>\u')')\Uny = dUnV(a>\ UnV) = (^((olpr))! UnV.
Clearly, d has the required properties, since dv has them for each U. □
Digression on vector analysis

The multilinear algebra developed above is particularly simple in the case
n — 3. We want to show how the classical approach of vector analysis fits
into the scheme of differential forms.
In order to develop the connection, we consider first the general situation
in an n-dimensional vector space T.
Definition. A volume element of T is a choice of basis in An(T*); since An(T*)

is 1-dimensional, a volume element is a choice of a nonzero element in
An(T*).
Example. If T is the tangent space to a manifold and {dxu ..., dxn} is a

basis for T*, then dx± A • • • A dxn is a volume element of T. (Note that a
volume element o> determines an isomorphism An(T*) ~ R\ where ru
corresponds to r. Conversely, such an isomorphism defines a volume element
co corresponding to 1.)
Remark. Given a volume element <o of T, there exists a natural isomorphism

m: An~\T*) -» Tdefined as follows. Recall that Tis naturally isomorphic to
its double dual T**. Identifying T** with T through this isomorphism, m will
have values in T**. For e T*, A is the real number such that <p a ip = Aco. To show that
m is an isomorphism, let {<pu .. ,,<pn} be a basis for T* such that co =
(Pi A • • • A cpn. Then the set {<}?! A • • • A y, _i A <Pj+i A • • ■ A <pn} is a basis
for An-1(r*). The value of m on these basis vectors is then given by
m(y>i A • • • A A cpj + 1 A • • • A <pn) = (-\)n + 1ej,
where {eu ..., en} is the basis for T dual to {<?!,..., <pn}.
125
5: Manifolds
Remark. Given an inner product < , > on a finite dimensional vector space
T, there exists a natural isomorphism g:T-^T* defined by
[g(u)](w) = (v, w> {v, w e T).
If {eu ..., en} is a basis for T, let gi} = (e{, cy>, (/,;e{l,..n}). Then in
terms of the dual basis {<pl5.. .,<pn} for T*,
n
g(ed = 2 SaVi
1=1
0‘6 {!»•••» «})•
In particular, if {el9 ..., en} is orthonormal, then gi} = Sij} and
g(e d = 9>i-
Applications. Take T = Rn. Then T has an inner product and a natural

volume element a> = <px A • • • A cpn, where {cpi} is the dual basis to the natural
basis {e4} for Rn. Thus the isomorphisms m and g are defined. Also, we have
natural identifications T(Rn, x) -«-> Rn for each x e Rn.
(1) Let /e C°°(Rn, R1). Then the gradient of / is the vector field on Rn
given by
grad / = g_1 o (df).
Relative to the usual coordinates (xu ..., xn) = (rly..., rn) on Rn,
df_ d_ (df_ df\

grad / = g 1 W) = r‘(i dXj dXj \ex/' ‘ ■’ dxnJ'
(2) Let V be a vector field on R3. Then g ° V is a 1-form and d(g ° V) is a

2-form. Now for dimension T = 3, A2(T*) — An_1(r*), so the isomorphism
m maps A2(T*) -> T. Thus m(d(g ° V)) is a vector field on R3. It is called the
curl of V.
curl V — (m o d° g)(F).
Exercise. Compute the coordinate expression for curl V.
(3) Let vx and v2 be vectors in R3. Then g(vx) and g(v2) are 1-forms. Their
exterior product is a 2-form; its image under m is a vector, called the cross
product of Vx and v2.
Vx X v2 = m(g(px) A g(v2)).
(4) Let V be a vector field on Rn. Then m_1(F) is an (n — l)-form on Rn.

Its differential is an n-form; that is, a multiple of the volume element a>. This
multiple is (up to sign) the divergence of V:
(—\)n~1dom~1(V) — (div V)a>.
126
Remark. Using these formulas, certain important formulas of vector

analysis become trivial consequences of d2 = 0.
A. curl grad/ = 0 because
curl grad f = m° d° g(g~1 ° d(J))
= m(d2f)
= 0.
B. div curl V = 0 because
d ° m~ J(curl V) = d ° m_ 1(m ° d ° g( V))

= d\g{V))
= 0.
Definition. Let X and 7 be smooth manifolds, and let Y: X -> 7 be a smooth

map. Then an induced map Y*: C°°(7, ^(7))-> CX(X, &(X)) is defined
as follows. For f e C°°(Y, A°(7)), Y*(/) = /»T; for oj e C°°(7, Afc(7))
[k > 0],
= w(T(*)X^0h), • • •, dY(vk)) [olf ...,vke T(X, x), x <= X];

Y* is extended to CX(Y, &(Y)) by linearity.
Remarks. It is easy to check that, if w is a smooth differential form, then

so is Y*w. It is clear that Y* maps /c-forms into fc-forms. In fact, it is easily
checked that Y* is an algebra homomorphism; i.e., Y* is linear and
Y*(w a r) = (Y*a>) A (Y*t)
for all o), r.
Theorem 2. Let X and Y be smooth and let Y: X Y be a smooth map. Then
d o Y* = Y* o d.
Proof
(1) If/e C"(7, A°( 7)), then for v e T(X, x),
[doY*(f)M=[d(foY))(v)
= [df o c/Y](y) (since d on functions is ordinary differential)
= [Y *(df)](v)
= [0^* ° d)(f)](v).
(2) For w a 1-form on 7 of the type oj — df
(d ° Y*)(w) = dQYfdf))
= dQY* o d(f))
= d(d o Y*(/)) (by (1))
= 0,
and
(Y* o d)(co) = Y*(dco) = Y*(ddf) = Y*(0) = 0.
127
5: Manifolds
(3) Using (1) and (2), together with the fact that Y* is an algebra homo¬
morphism, the result is established in general by checking it locally on k-
forms a> restricted to local coordinate neighborhoods:
o>|, = 2 ah—ik ^xh A • • • A dxik.
(Details are left to the reader.) □
Definitions. Let X be a smooth manifold. A smooth differential form a> on X

is closed if dco = 0. A form a> is exact if it is the differential of another form
on X; that is, o> is exact if to = dr for some smooth form r. (Note that
every exact form is closed, since d2 = 0. The converse question is funda¬
mental to our subject.)
Let Zk(X, d) denote the vector space of closed &-forms on X. Let
Bk(X, d) denote the space of exact £-forms on X. Then B\X, d) <= Zk(X, d)
because d2 = 0. Let Hk(X, d) = Zk(X, d)/Bk(X, d). Hk(X, d) is called the
A:th De Rham cohomology group of X. Its dimension, which we shall see is
finite for compact X, is called the kth Betti number of X.
Remark. Although these cohomology groups are defined in terms of the

manifold structure of X, they are topological invariants; that is, if two mani¬
folds are homeomorphic (by a not necessarily smooth homeomorphism),
then they have isomorphic cohomology groups. In fact, these groups can be
defined directly using only the topological structure of X.
Example 1. H°(X, d) ^ R1 if X is connected. For since there are no forms of

degree less than 0, B°(X, d) = 0. Thus
H\X, d) = Z°(X, d) = [fe C°°(X, R1); df = 0],
If U is any connected coordinate neighborhood of X, with coordinate func¬

tions (xlt..., xn), then df = 0 on U means
0 = rf/= 2 fwdx,-,
i=i
that is, (d/dxi)(f) = 0 for all i. But this implies that/ is constant on U. Since
Xis connected, and since/is constant on each connected coordinate neighbor¬
hood in X, then / must be constant on X; that is, Z°(X, d) = [constant
functions on X] ^ R1.
Example 2. H1(S1, d) ^ R1, where S1 is the circle. For since there are no
nonzero £-forms on S1 for k > 1, Z\S\ d) = C00^1, AH-S1)). Moreover,
B\S\d)= [df;f e CîS1, A1)].

Now, if 9 denotes the polar coordinate on S1, then d/89 is a nonzero vector
128
field on S1 and its dual 1-form dd is a nonzero 1-form on S1 (see Figure 5.4).
Furthermore, dd is not exact (in spite of the notation!)—but, given any 1-form
w — g(9) dd on S1, w — (c dd) is exact for some c e R1. Thus
Z\S\ d)IB\S\ d) ^ [c dd- c e R1] ~ R1.
Exercise. Verify the above facts. Take c = (l/2ir) jl" g(6) dd.
Remarks. Let </<: X 7 be smooth. Then
<A* ■ Zk(Y, d) -> Z\X, d) and <p*: B\ Y, d) Bk(X, d).

For if oj is a closed £>form on Y, then d(tp*oj) — i/j*(du>) = */>*(()) = 0. If
cv = dr is an exact /:-form on Y, then ifi*(a>)= <p*(dr) = d(</>*(t)). Thus </>*
induces a linear map $ on cohomology, such that
• 4>: Zk( Y, d)/Bk( Y, d) -> Z\X, d)/B\X, d);
that is,
4>:Hk(Y,d)^Hk(X,d).
If S: W—> X and T: XY are smooth, it is easy to check that (T° S)* =
S*°T*, and hence (T^S) = S°T:

S T
W-> X-> Y,
Hk(W, d) Hk(X, d) Hk( Y, d).
Thus we have attached to each smooth manifold X new algebraic invariants

Hk(X, d) such that given smooth maps between manifolds, there are induced
algebraic maps between these algebraic objects. As in the case of the funda¬
mental group, we are thus able to solve certain difficult topological problems
by studying their algebraic counterparts.
Now let us show that Hk(Rn, d) = 0 for all k > 0. Since Rn is diffeo-
morphic (isomorphic as a smooth manifold) with the unit ball 50(1) about 0
in Rn, we may as well show that Hk(B0( 1), d) = 0 for all k > 0. For this we
need the following technical lemma.
129
5: Manifolds
Lemma. Let X be a smooth manifold. Then, for each k, consider the maps
C“(A, Ak~1(X)) Ca(X, Ak(X)) Cm(X, Afc + 1(Z)).

hk -1 hk
Suppose there exist linear maps
h,\ C°°(A, A*+ x( A)) -> C°°(A, A’(X)) O' = k - 1 or k)
such that hk°d + d°hk_ 1 is the identity map on CX(X, Ak(X)). Then
Hk(X, d) = 0; that is, every closed k-form is exact.
Proof. Suppose o» e Cm(X, Afc(A)) is closed. Then
o» = (hk o d + do /?te_!)(oj) — hk{dcd) + d(hk_1w) - d{hk_1 w). □
Remark. If a sequence of such linear maps hj is defined for all j > 0, the
sequence h, is called a homotopy operator.
Theorem 3 (Poincare’s lemma). Let U = 50(1) <= Rn, Then Hk(U, d) = 0 for
all k > 0.
Proof. We shall construct maps satisfying the conditions of the

lemma. This is done through an integration process. Since these maps are to
be linear, it suffices to define hk^1 on forms a> — g dxh A • • • A dxilc; similarly
for hk. For such a>, set
hk-i(oj)(x) = ^ tk~lg(tx)dt^p,
where
p = xtl dxh A • • • A dxik - xi2 dxh A dxi3 A • • • A dxik

+-h (-1 )k~1xtkdxh A • • • A dxik_1.
(Note that dp = k dxh A • • ■ A dxik.)

The map hk is defined similarly by replacing k everywhere by k + 1.
Now, for m = g dxh A • • • A dxik e C°°(t/, Ak(U)) and x e U,
(d°hk_ i)(<«)(x)
(Jo tk Xg(tx) dt^jp
A p + (J0 tk 1g(tx) dt^j dp
= 2 (J0 (S(tx)) dt^ dx, A p + ^ £ tk~1g(tx) dt^j dp
= 2 (|0 {k dt^j dxj A p + tk~1g{tx) dt\ dxh A • • • A dx
130
5.2: Differentia] forms
and
(hk d)(w)(x) = 2 dXj A dxh A • • • A dxik
[Xj dxh A • • • A dxik — dx, A /x].
Thus,
(d o hk_x + hk° d)(w)(x)
= k(jd’-1g(‘x)dt) + | (Jo‘ '* J|(r*)xy*) J.Xj, A • • • A dx{*Jc
-{/: ktk *g(tx) + tk jt(g(tx)) dt >• dx< A ■*k

dXi
= {Jo ^ ['*#(>*)] dXifl A • • • A
= ffcg(/x)|£ </xh A • • • A dxi)c

= £(*) A • • • A Jxj)c
= oi(jc) (for all x e C/).
Since d ° hk^1 + hk° d acts as identity on such a>, it acts by linearity as identity
on all A:-forms. □
Remark 1. The maps hk^1 and hk used in this proof were not just picked
out of the air. They were constructed as follows. Given a vector space T and
v e T, v defines a map i(v): Ak(T*) —> Ak~1(T*) by
P(®)(")](®1» • • Ofc-i) = w(o, t>fc_i).
Note that z is a bilinear map T 0 Ak(T*) -> Afc_1(r*). This map z* is called
interior multiplication. The map hk_x was obtained by applying i(x) to o> and
averaging over the line through the origin in the direction x.
Remark 2. Theorem 3 is a special case of a more general result. Let U be a
smooth manifold. Suppose there exists a smooth map 'F: U x Ie-> U, where
Ie = [re/?1; —e < r < 1 + e], such that T(u, 1) — u for all ueU, and
0) — u0 for all ue U; some u0e U (see Figure 5.5). Then Hk(U, d) = 0
for all k > 0. The map Y is a smooth homotopy. This theorem says that if U
is smoothly homotopic to a point, then the cohomology of U is that of a point.
In the case covered by Theorem 3, a smooth homotopy is given by
T(x, t) = tx (t € IE; x e 50(1)).
131
5: Manifolds
Note that the above proof of Poincare’s lemma works equally well for a
star-shaped region, that is, an open set U such that for some x0 e U, the line
segment joining x0 to any other point in U lies completely in U.
5.3 Miscellaneous facts

Theorem 1. Let X and Y be smooth manifolds, with X connected, and let
ip\ XY be smooth. Assume dift = 0. Then ifj is a constant map; that is,
i/»(x) = y0 for some y0 e Y and for all x e X.
Proof. Let y0 e ifi(X). Then »/»_ 1(>’o) is a closed set in X. We shall show this
set is also open, hence = X since X is connected.
Suppose x0 e ifj~1(y0)- It is sufficient to find an open set U in X such that
x0e U and U ifi~1(y0). Let V be a coordinate neighborhood of y0> with
coordinate functions (j>l5..., ym). Take U to be any connected coordinate
neighborhood of x0 such that [/<= V). Let (xl5 ...,xn) denote the
coordinate functions in U. Then, for each x e U, the matrix for difi(x) relative
to the bases {3/<?x(} for T(X, x) and {d/dyi} for T(Y, ijj(x)) is
Now, difj = 0 implies (b/dxîyi ° </>) = 0 on U for all i, j. But this implies that
y{ o ifj is constant on U for all i. Hence yt i/j(x) = yt° ^(x0) for all i and
°
all xe U; that is, </»(x) = t/<(xo) = To for all x e U, and U <=■ i/>-1(y0) as
required. □
Definition. Let X be a smooth manifold, and let V and W be smooth vector

fields on X. The bracket [V, W] of V and W is the smooth vector field on
X defined by
[V, W](J) = V(Wf) - W{Vf) ife C”(X, i?1)).
132
5.3: Miscellaneous facts
Remark. [V, W] is a vector field, because if 93 is a local coordinate system

with domain U and coordinate functions (x^, ..., xn), then
for some au e CX(U, R1). Since [V, W] is clearly bilinear, it suffices to

check that [V, W] is a vector field when V = a(8/8xi) and W = b(8/8x,).
Then, since mixed partials are equal,
4-,
8 8 8 8
Q 8x( 8xj 8Xj ^ 8xi (/)
Since a(8/8xi)(b) and b^d/dxâ) <= C”(Z, R1), [V, W] is indeed a smooth
vector field.
Remark. The bracket of vector fields has the following properties, each of
which is easily verified.
(1) [V, W] = -[W, V]

(2) [V, + V2, W] = [V±, W] + [V2, W]
(3) [cV, W] = c[V, W] for c e R1
(4) [[V, W],Z] + [[W,Z], V] + [[Z, V], W] = 0.
Property (4) is called the Jacobi identity. These four properties say that
Ca(X, T(X)) is a Lie algebra under bracket multiplication. Note that such an
algebra is non-associative.
Theorem 2. Let a> be a smooth \-form, and let V and W be smooth vector fields
on X. Then
dofiV, W) = V(oj(W)) - W(oj(V)) - <o([V, W])}.
Remark. In some texts, the fraction \ is missing from this formula. This is
due to a slightly different definition of exterior multiplication.
Proof of Theorem 2. It suffices to verify this formula in a local coordinate

neighborhood. Furthermore, since both sides are linear in u>, we need only
133
5: Manifolds
check it on forms of the type u> = f dg (since every 1-form is locally a sum
2 at dxi). For cu = f dg,
doj(V, W) = (df A dg)(V, W)

= WAV) dg(W) - df{W) dg(V)}
= \m){Wg) - {Wf){Vg)).
On the other hand, we also have
i{V(co(W)) - W(co(V)) - w([V, W])}

= ${V(fdg(W)) - W(fdg(V)) — f dg([V, W])}
= \{V{f(Wg)) - W(f{Vg)) — f([V, W]g)}
= \m\Wg) +fV(Wg) - (W/XVg) -fW(Vg) - fV(Wg) +fW(Vg)}
= \m\wg) - (wfxvg)}. □
Theorem 3 (Inverse function theorem). Let X and Y be smooth manifolds of
dimension n. Let ji: X Y be a smooth map. Suppose x0e X is such that
d<Kx0):T(X,x0)->T(Y,,p(x0))
is an isomorphism. Then there exists a neighborhood U0 of x0 such that

(1) Uo is injective,
(2) ip(U0) is open in Y, and
(3) «/»“1: ip(U0) -> U0 is smooth.
Proof. Let <p2 be a coordinate system about jj{x0) with domain V and co¬
ordinate functions {yx,..., yn). Let <px be a coordinate system about x0 such
that
U = domain <p1 <= <p~x(V).

Let (xl5..., xn) denote the coordinate functions of <px. Then, relative to the
bases {d/dxt} for T(X, x0) and {djdyjf for T(Y, <p(x0)), dip(x0) has the matrix
which is nonsingular since dif>(x0) is an isomorphism.
c > C
<PI
NK
d d 7 —>d d
Figure 5.6
134
Now transfer everything to Rn via <px and <p2. Let U = <pfU), F = <p2(F),
and iJ>: U —> Lbe defined by f = <p2 o jj o ^ -1 (see Figure 5.6). Then </<(x) =
OAiCrh • ■ • > W-’f)) for .x e 0, where i/y = o The Jacobian of >p at x0 =
<Pi(x0) is
which is nonsingular. Flence, by the classical inverse function theorem, there

exists an open set U0 <= U containing x0 such that V0 = f(U0) is open, and
such that the equations
■ ■ ■, rn) = Si (/ e {1,. . ., «})
have a unique solution in U0 for each (j1}..., jn) e V0. Moreover, this solution
depends smoothly on (sy,..., sn). In other words, there exist smooth functions
hr- (jg{i,...,«})
such that for each s = (su ..., sn) e V0,
$i(hi(s),..., hn(s)) = st.
Setting h(s) = (hx(s),..., hn(s)) for se V0, this says that h = Trans¬
ferring back to X and Y, we find the conditions of the theorem are satisfied,
with
U0 = <Pi-1(tf0). □
Theorem 4 (Implicit function theorem). Let X and Y be smooth manifolds,

with dim X > dim Y. Let </>: X -> Y be a smooth map. Let y0 e ifi(X) and
let
Xo = r\Yo) = [xe X; f(x) = y0].
Assume that for each x s X0, djj(x): T(X, x)T(Y, ifi(xj) is surjective.
Then X0 has a manifold structure, whose underlying topology is the relative
topology of X0 in X, and in which the inclusion map X0 -> X is smooth.
Furthermore, dim X0 = dim X — dim Y.
We give some applications before proving Theorem 4.
Applications
(1) The n-sphere Sn is a smooth manifold whose topology is the induced
topology in Rn + 1. For let ifr. Rn+1 -> R1 be defined by
71+ 1
4>(ri,...,rn + 1) = ^ ri2.
i = 1
Then Sn = *A_1(1)- Since dim R1 — 1, we need only check that djj ^ 0 at

each point of </>_1(l). But dj> = 2 jjifj rt drt. Since {drx,..., drn + 1} is linearly
independent, djj 0 unless rt = 0 for all i. In particular, df ± 0 on _ 1(1).
Note that dim Sn = dim Rn + 1 — dim R1 — n, as expected.
(2) Let X = Rn2, viewed as the space of all real n x n matrices. Let
Y = R1, and let «/»: A" —> Y be the determinant function. Then */'_1(l) is the
135
5: Manifolds
group of all n x n matrices of determinant 1. It is called the unimodular

group. To verify that this group has a manifold structure, we need only show
that dip ^ 0 at each point of </'“10)- Now, for (rj;) coordinate functions on
Rn\
1P°ru — 2
jieS„
Hence
n
^=22 (— 1 V'HjKI) ■ ' 'ri-lJiU-lri + nnU+l)' ' ’rnn<.n) ^rinU)-

1 = 1 neSn
For each (i, j), the coefficient of dru in this sum is, up to sign, the determinant
of the cofactor of rtj in (rtj). These cannot all be zero at any point of «/»_1(l)
since det (riy) # 0 at such points. Since {drtj} is a linearly independent set, we
are done.
Note that this unimodular group has dimension n2 — 1.
(3) Let X = Rn2 as in (2). Let Y be the set of all symmetric n x n real
matrices. Y is a manifold, for it can be naturally identified with Rn<-n + 1'>'2:
merely string out in a row the entries on and below the main diagonal. Let
ip: X-> Y be defined by <p(x) — xxl where, for each x e X, xl denotes the
transpose of x. Note that ip is smooth, since each entry of >p(x) is a polynomial
in the entries of x. Let X0 = 0_1( 1). Thus X0 is the group of orthogonal
n x n matrices; that is, X0 is the orthogonal group.
To verify that X0 is a manifold, we must show that dip(x) is surjective for
each x e X0. For this, it suffices to show that dip(e) is surjective, where
e = (8iy) is the identity matrix. For assuming that dip(e) is surjective, let
x e X0. Then the map Rx: XX, defined by Rx(y) = yx (matrix multiplica¬
tion), is a smooth map with a smooth inverse, namely Rx-1, and hence dRx
is everywhere an isomorphism. Moreover, ip ° Rx = P(y).

Hence,
dip\x = d(>p o ^-1)1* = dip\Rx-iw o dRx-i\x = dip\e ° dRx-i\x,
so dip(x) is a composition of surjective maps, hence is surjective.
We still must check that dip(e) is surjective. But
(ru ° = 2 rih(x)rjh(x) (1 < i <j < n);

h=l
hence the entries in the matrix for dip(x), where 1 < k, l < n, and 1 < / <
j < n, are
rd*) (if k = i + j),
ru(x) (if k =j ^ /),

eT
cr kl (r«
0)U
2r tl(x) (if k = i = j),
0 (otherwise).
136
In particular, the entries in the matrix for dip(e), where 1 < k, l < n, 1 < i <
j < n, are
1 (if (k, /) = (ij); i ^ j)
(if (Jc, /) = (J, 0; i / j)

(if (k, /) = (ij); i=j)
.0 (otherwise).
Thus the square submatrix, consisting of those entries with k < /, is a diagonal
matrix with diagonal entries 1 and 2, and so dp{e) has rank n{n + l)/2; that
is, dp{e) is surjective.
Note that dim X0 = dim X - dim Y = n{n - l)/2.
(4) Let X = the set of all complex n x n matrices — R2n2. Let Y =
[x e X; x1 = x]. Let *p: Ar—> Y be defined by ip(x) = xxK Then, as in (3),
the set i/(_ 1(e) is a manifold. */<-1(e) is the unitary group. Its dimension is
2 n2, — n2 = n2.
Remark. Examples (2), (3), and (4) are examples of Lie groups; namely,
they are groups whose underlying spaces are C“-manifolds and are such that
the group operations are analytic.
Proof of Theorem 4. Let V be a coordinate neighborhood of y0 in Y, with

coordinate functions (yu .. .,ym). For x0 e X0, let U be a coordinate neigh¬
borhood of x0 in X such that U <= i/i-1(F). Let (x1} ...,xn) denote the
coordinate functions on U. We may assume that this coordinate system is
chosen so that Xi(x0) = 0 (1 < i < n). Now dp surjective at x0 means that
the m x n matrix ((d/8xj)(yi ° p)\Xo) has rank m. By renumbering the co¬
ordinate functions on U if necessary, we may assume that the last m columns
of this matrix are independent, that is, that this matrix has the form
(* = JX
where / is a nonsingular m x m matrix. Let <p: URn~m x Fbe defined by
P(x) = (Xi(x),..., xn_Jx), <P(x)) (x e U).
Then dp(x0) has matrix
where I is the identity (n — m) x (n — m) matrix. Hence dp(x0) is an iso¬

morphism. By the inverse function theorem, there exists a neighborhood U0
of x0 such that <p\ Uo is injective, <p(U0) is open in Rn~m x V, and p~1: <p{U0) ->
U0 is smooth. We may assume that <p(U0) is of the form W0 x V0, where
0 e W0 and y0 e V0, since open sets of this type form a basis for the topology
on r_m x V (see Figure 5.7). Now note that i/?_1(B/0 x {y0}) = X0 n U0.
Since p\Uo is a homeomorphism, ^xqûo maps X0 n U0 homeomorphically
onto W0 x {j0} ^ W0 <= Rn~m. Thus p\Xonu0 is a coordinate system about
*o in X0.
137
5: Manifolds
To see that such coordinate systems actually define a smooth manifold

structure on X0, we must check that they behave properly on overlaps. So
suppose
<A: U0^W0 x V0 and <p: U1->W1 x Vx
are such that (X0 n U0) n (X0 o Uf) # 0 (see Figure 5.8). Since is
smooth, so is
V ° 'P 1|«t70nUi)-
Restricting to $(X0 n U0 n Ux) = i]>{U0r\ Uf) n (Rn~m x {y0}), it follows

that
y0 4> 1|i?(x0ni;oû1); to U0 Ux) —> <p(ô n t/0 n t/i)
is smooth. Hence Z0 is a smooth manifold, of dimension n — m. □
Figure 5.8
Definition. A submanifold of a smooth manifold Y is a pair (X, if), where Xis

a smooth manifold and i/j : X -» Y is an injective smooth map such that difi
is injective at each point of X.
Examples. The manifold X0 of the previous theorem, together with the

inclusion map X0 X, is a submanifold of X. In particular, Sn is a sub-
138
manifold of Rn + 1, and each of the Lie groups discussed above are sub¬
manifolds of the space of all n x n real (complex in the case of the unitary
group) matrices.
Remark. Note that ip: XY being injective does not imply that dip is
injective at each point. For example, the smooth map <p: R1 R1 defined by
<p(x) = x3 is injective, and yet dip(0) = 0. Note also that (X, ip) being a sub¬
manifold of Y does not imply that tp is a homeomorphism of X onto *p(X)
with the relative topology.
Example. Consider the torus
S1 x S1 = [(zl5 z2); zx, z2 complex, with |zx\ = \z2\ = 1].
Define ip: R1 -> S1 x S1 by >p(t) = (e2nit, e2niat), where a is an irrational

number. Then (R1, ip) is a submanifold of S1 x S1. However, is dense
in S1 x S1, so ip is not a homeomorphism. This submanifold is called the
skew line on the torus. Representing the torus as a square with opposite sides
identified, ip maps R1 as in Figure 5.9.
a c
Figure 5.9
Theorem 5. Let (X, >p) be a submanifold of Y, with X compact, Suppose X has

dimension m and Y has dimension n, where m < n. Then, for each x0 e X,
there exists a coordinate system <pY: V -> Rn about 4>(x0) with coordinate
functions (>>i,..., jn), such that
<P(X) n V = [ye V; ym + 1(y) = ■■■ = yn(y) = 0],
Furthermore, a coordinate system <px: U Rm can be chosen about x0, with

coordinate functions {xY, ...,xm), such that U c </,-1(F) and such that
Xj = y, ° >p for all j < m. Thus, on U,
Xj (j < m)
.0 (J > m).
Proof. The proof of Theorem 5 is left as an exercise. □

139
5: Manifolds
Remark. When a coordinate system <pY is chosen as in Theorem 5, ^(X) n

V is said to be a slice in <pY. Note that the coordinate systems obtained in the
proof of Theorem 4 are of this type.
Corollary. If (X, f) is a submanifold of Y, and X is compact, then
t-.X-^KX)
is a homeomorphism. Moreover, for each submanifold obtained by applying
the implicit function theorem, the inclusion map is a homeomorphism.
Proof. Since ip(X) is Hausdorff in the relative topology, the first statement is
proved. □
Definition. Let X be a smooth manifold, and let V be a smooth vector field

on X. An integral curve of V is a smooth curve a: (a, b) X (Figure 5.10),
such that the tangent vector to a at each point is equal to the value of V
at that point; that is,'
d(t) = V(a(t)) (for all t e (a, b)).
Remark. Let <p: U -> Rn be a local coordinate system on X, with coordinate

functions (x1;..., xn). Let a: (a, b) —> U be a smooth curve in U. Then, by
definition, & = da(d/dt). Hence, the zth component of a relative to the basis
{8/dxj} is
dxj(d) = dXi(da(d/dt)) = d(xt ° a)(d/dt) = d/dt(xt ° a),

so that
Thus « is an integral curve of a vector field V = 2 aib/dx^) if and only if
(*) jt (*i ° «) = at (/ e {1,..., nj).
140
Thus, to find integral curves of a given vector field V on a coordinate neigh¬

borhood U, we need solve the system (*) of differential equations. Solutions
are guaranteed by the following classical theorem.
Theorem 6. Let W be an open set in Rn, let w0 e W, and let a{ e C°°(W, R1),
(1 : ( —e, e) x W0-> W such that, for each
w e WQ, 0|( - e,e) x<u,> is a solution of the equations
f = amt),...,fn(t)) (1 < i<n)
subject to the initial conditions f(0) = wt; that is, if W is

defined by
«»(0 = <A(b w),

then for 1 W satisfying (A)
and (B).
Reinterpreting Theorem 6 in terms of vector fields, we obtain Theorem 7.
Theorem 7. Let X be a smooth manifold and let V be a smooth vector field on

X. Let x0 e X. Then there exist an open set U about x0, an interval ( — e, e) c
R1, and a smooth map jr. ( — e, e) x U -> X, such that for each u e U, the
curve
au ’■ ( ~ e, e) -> X
defined by a fit) = j(t, u) is the unique integral curve: ( — e, e) -> X of V,
with au(0) = u.
Furthermore, the smooth maps if>t: U X, defined for each t e ( — e, e) by
4>t(u) = ijj(t, u), have the properties
(1) iptl + t2 = o O on whenever tx, t2 and / + t2e (-e, e)
(2) i/r_t = on <pt(U) n Ufor each t e ( — s, e).
Proof. Let W be a coordinate system about x0, with coordinate functions

(*i,. . ., xn). Then, on W, V = 2"=i ajd/dxi) for some smooth functions
at e CX(W, R1).
By Theorem 6, there exist U <= W, (-e, e) c R1, and </>:( — e, e) x U
W c X with the required properties. The last statement is a consequence of
141
5: Manifolds
the uniqueness of the solution; namely, it is easy to check that tx —>

+ t2, u) and ti -» </>(*!, ipt2(u)) are both integral curves of V which send
0 into </>f2(w), and hence they are equal; that is, <ptl+t2 = >At1 ° Similarly,
«A-t = -Ar1. □
Remark. Properties (1) and (2) of Theorem 7 express the fact that ipt is a
local one-parameter group of transformations.
Remark. The previous theorem guarantees the existence locally of integral
curves for vector fields. However, it is not always possible to obtain integral
curves globally; that is, it is not possible in general to find a curve a: R1 -> X
through x0 such that a is an integral curve of a given vector field V. For
example, let X = R2 — {0} and let V = 8/dr^ Then the integral curve of V
through (—1,0) cannot be extended to values of t > 1 (see Figure 5.11).
->
-> ->
4 ->
—> ->
Figure 5.11
However, if X is compact, then every vector field admits through each point
integral curves defined on all of R1.
Remark. In studying the motion of a particle in R3 under the influence of
a force field F, Newton’s law tells us that the path of motion is a curve (xf(0)
such that
d2Xi(t)
m = Ft (1 < i < 3),
~dF~
where m is the mass of the particle. Setting pt = m(dXi/dt), we have
dxi = A, dPi =F (1 < i < 3).

dt m dt (
But (x1; x2, x3. Pi, p2, pf) may be regarded as coordinate functions on the
cotangent bundle of R3. Hence the orbit of the particle is just the projection
onto R3 of the integral curve of a vector field on the cotangent bundle. In
fact, the cotangent bundle is the natural domain for the study of mechanics
on a manifold.
142
Remark. The use of integral curves provides a geometric interpretation of

the bracket of two vector fields. Let V and W be smooth vector fields on X,
and let x0 e X. Suppose we move along the integral curve of V through x0
until the parameter has moved from 0 to af~s\ then move along an integral
curve of W from 0 to V^; then move back along an integral curve of V, the
parameter now varying from 0 to —Vs; and finally move back along an
integral curve of W from 0 to — Vs, as in Figure 5.12. We will not in general
through x;
through x)
Figure 5.12
return to our starting point. As s -> 0, our end point will trace out a curve
through x0. The bracket [V, W](x0) is precisely the tangent vector to this
curve.
Definition. Let V be an ^-dimensional real vector space. Then A"(F*) has

dimension 1, so it is isomorphic to R1. Thus An(F*) — {0} is disconnected;
it is the union of two connected components. An orientation of V is a
choice of one of these components. An oriented vector space is a pair
(V, s/) where s# is an orientation of V.
Remarks. Thus each vector space V has two possible orientations. An

ordered basis {<pl5 ..., <pn} of V* determines an orientation of V; namely, the
component of An(V*) in which <px A • • • A <pn lies. Given two ordered bases
{<Pi, • • •, <pn} and {<p'n, ■ ■ ■, Vn) of V*, with <p’{ = 2 then <Pi A ■ • • A <p'n =
det (Ciy)^! A • • • A (pn. Hence two ordered bases determine the same orienta¬
tion if and only if the determinant of the change of basis matrix is positive.
In particular, if {<px,..., <pn} is an ordered basis for V*, then the orientation
determined by the basis
W2, <Pl> 93> • • • > SPn}
is different from the one determined by {<?!, y2, ■ ■., <pn}-

In R2, an orientation amounts to a sense of rotation. The orientation
determined by {dru dr2} gives the usual sense of positive rotation on R2;
namely, so that the rotation sending d/8r1 into d/dr2 is one of +7r/2. The
143
5: Manifolds
orientation determined by {dr2, drx} defines the opposite sense of rotation, so

that 8/dr2 -> d/8r1 is a rotation of +tt/2 (see Figure 5.13). Similarly, an
orientation of R3 amounts to choosing either the right-handed rule or the
left-handed rule for cross products.
Orientation of dri A dr2 Orientation of dr2 A dri
Figure 5.13
Definition. A smooth manifold (X, O) is orientable if there exists a subset

O' <= <1> such that
(1) {domain <p}0ev is a covering of X, and
(2) If 9o1 and <p2 are coordinate systems in O', with domains U and V and
coordinate functions (xu ..., xn) and (yu .. .,yn) respectively, then
the function A: U n V-> R1 determined by
dxx A • • • A dxn = A dyx A • • • A dyn
is everywhere positive.
An orientation of an orientable manifold (X, d>) is a choice of subset
O' c O satisfying (1) and (2) and maximal with respect to (2). An oriented
manifold is a triple (X, O, O'), where (X, O) is an orientable manifold and
O' is an orientation of {X, O).
Remark. The function A such that

dxx A • • • A dxn = A dy± A • • • A dyn
is just the Jacobian determinant of <px o q>2 ~1; that is,
A = det = det dlcp-x <P2 X)-
In view of this, it is easy to check that a connected orientable manifold (X, O)

has exactly two orientations O' and O", and that O is the disjoint union
O' u O".
Remark. A more sophisticated approach to orientation of manifolds is to
consider the set An(2f). This set can be given the structure of an (n + 1)-
dimensional manifold as follows. Let y: URn be a local coordinate
144
system on X, with coordinate functions (xlt Then a coordinate

system <p: 7r-1(C/) -» Rn + 1 is defined on tt--1([/) by
<p(w) = A(oj)) (to e
where A: 77-_1(t/) R1 is the function such that
A(w) dx1 A • • • A dxn = co (a> G 7r_1(t/)).
In terms of An( A), we have the following characterization of orientability.
Theorem 8. Let X be a connected smooth manifold (see Figure 5.14). Let
O = {0 element in An(T*(X, x))} <= An(A).

xeX
Then either An(X) — O is connected, in which case X is not orientable, or

A"(A) — O breaks up into exactly two connected components, in which case
X is orientable. An orientation of an orientable manifold X amounts to a
choice of one of these two components.
Proof. We omit the proof. □
Theorem 9. Let (X, O) be a smooth manifold of dimension n. Suppose there

exists a smooth n-form co on X which is nowhere zero. Then X is orientable.
Proof. Let R1. Since co is never zero, neither is fv.
Thus either f0> 0 everywhere, or/,, < 0 everywhere. Let
4)' = [cpe^-Jv > 0],
145
5: Manifolds
Then O' is an orientation of X. O' covers X because if x e X and 9 is a co¬

ordinate system about x with f0 < 0, then the new coordinate system 0. Furthermore, if 99, tp e O' have domains U and V and coordinate
functions
(*1, •••,*„) and Oi, ...,yn)

respectively, then on U n V
1 f
dyx A • • • A dyn = to = -f dxx A • • • A dxn
Ji! J\i!
and fjf# > 0. Maximality is clear. □
Theorem 10. Let (X, </>) be an n-dimensional submanifold of Rn + 1. Suppose

(X, 1/>) admits a nonzero “ normal vector field’’’’; that is, suppose there exists
a smooth map V: X ^ T(Rn + 1) such that for each x e X, V(x) is a nonzero
vector in T(Rn + 1, ipix)) perpendicular to dijj(T(X, x)) (see Figure 5.15). Then
X is orientable.
ip(x)
Figure 5.15
Remark. Perpendicularity in T(Rn + 1,1fj(x)) means with respect to the inner

product < , ) given by
Proof of Theorem 10. Given a normal vector field V, consider the n-form p.
defined at points of >p(X) by
p = i(V) drx A • • • A drn + 1.
Let a) = 4>*p. Then o> is a smooth «-form on X. By Theorem 9, it suffices to
show oj is never zero on X. Suppose it were; that is, suppose w(x) = 0 for
some x e X. Then
0 = wOOOi, ...,vn)
= K^(*))(<#(g), • • •, difi(vn))
146
for all »n e T(X, x). Now each vector w e T(Rn + 1, 0(*)) is of the form
w = difi(v) + cV(x) for some veT(X, x), (c e R1). Thus, for arbitrary
vectors
wt = + CiV(x) e T(Rn + 1, </.(*)) (1 </<«),
we have
K'AWXh’i, ..wn) = ^W)(#(Bl) + Cl V(x),..., diP(vn) + cnV(x))

= /*(0W)(#(t> 0, ■ • •, #(yn))
n
+ 2 ciKiK*)X#(»i)» • •#(yy-i), F(x),

y=i
<#0fi+i), ■ d*l>(vn)).
All other terms are zero since F(x) appears twice as an argument, and /u, is
skew symmetric. Moreover, the first term vanishes by the above discussion,
and each term of the sum is zero because
fOAMX- •V,...) = /(F) drx A • • • A drn + 1(..., F,...)

= dr1 A • • • A <frn + 1(F,..., F,...)
= 0.
Since wl5..., wn e T(Rn + 1, ifj(x)) were arbitrary, this shows that K’KX)) = 0.
But
P = z(F) dr1 A • • • A Jrn + 1

n
= 2 iy_1(Fr/) A • • • A A dry + 1 A • • • A drn + 1.
Since F(x)r,- ^ 0 for some y, p(<p(x)) + 0. This contradiction proves the

theorem. □
Corollary. The unit sphere Sn is orientable.
Proof. Sn admits a nonzero normal vector field, namely, the restriction to Sn

of the unit vector field on i?n + 1 — {0} pointing radially outward. □
Remark. It can be shown that every compact connected ^-dimensional

submanifold of Rn + 1 separates Rn + 1 into two connected pieces, one bounded
and one unbounded. Thus every such submanifold admits a unit normal
vector field (for example, the one pointing into the unbounded component),
hence is orientable.
Remark. A nonorientable 2-dimensional manifold is called a one-sided
surface.
Example 1. The Mobius strip S, obtained from an open rectangular strip by

giving the strip a half twist and glueing the ends, is nonorientable. Note that a
nonzero normal vector field cannot exist on S, for if such a field varies
147
5: Manifolds
continuously along the center line, it would have to point in the opposite
direction after a full circuit.
->-
->-
Figure 5.16
Example 2. The Klein bottle K, obtained from 7x7 by identifying opposite

sides (to get a cylinder) and then identifying the other pair of sides with a
twist (Figure 5.16), is nonorientable. This surface cannot be represented as a
submanifold of R3. However, there does exist a map ip: K-^> R3 with dip
injective at each point, and such that tp is one-to-one except along a circle in
R3 (see Figure 5.17).
Definition. Fet X be a topological space, and let be an open covering of X.

The covering is locally finite if, for each x e X, there exists an open set
Wx containing x such that
[UeW; 17 n Wx ^ 0]
is a finite set.
Definition. A topological space X is paracompact if every open covering of X

has a locally finite refinement; that is, if for every open covering °ll, there
exists a locally finite open covering V such that for each Kef there exists
at/ef with V <= U.
Remark. It can be shown that all metric spaces are paracompact. Also,
every regular topological space whose topology has a countable basis is
paracompact.
148
Definition. Let X be a smooth manifold. A smooth partition of unity on A is a

pair (if ST), where tF is a locally finite covering of X and ^ = {fv)vey is
a collection of smooth real-valued functions on X such that
(1) each fv > 0,

(2) for each the support offv — the closure of the set
[x e X;fv(x) A 0]
is contained in F, and
(3) 'fve'T fv = L
(Note that this sum makes sense since for each x e X,fv(x) = 0 for all but
finitely many V e'Y'.
Theorem 11. Let X be a paracompact manifold. Then, given any open covering
of X, there exists a smooth partition of unity (A’ ST) on X such that Y~ is
a refinement of
Proof. Since A is a manifold, there is a refinement if of such that each

open set W e iV is a coordinate neighborhood, and W is compact. Since X
is paracompact, there is a locally finite refinement Y~ of the open covering iV.
Note that Y” is a refinement of °Y, and if V e if then V is compact, and V is
a coordinate neighborhood.
Suppose we can “shrink the covering Y~ slightly” and still get a covering.
That is, suppose for each Veif we can choose an open set a(V) such that
a(F) c= v and {<x(V)}Ve-r is a covering. We then proceed as follows. Since
Ve'Y' is a coordinate neighborhood, and a(V) is a compact set in V, we can
find a smooth nonnegative function gv: X -» R1 such that gv(x) = 1 for
x e a(V) and gv(x) = 0 for x ^ V. Let g = fVe-r gv- Then g is well defined and
in CX(X, R1) because 'Y' is locally finite. Furthermore, g never vanishes on
X because {cc(V)}Ve-r is a covering; hence fv = gv/g e Cco(A, R1). Let ST —
{fv}vs-r\ then (i/',dY) is a smooth partition of unity.
To “shrink the covering Y" slightly,” proceed as follows. Consider the
family (3 of all functions j8 such that
(1) domain of /3 is a subset of 'Y'; _

(2) if V eS>B, then jS(F) is an open set in V such that /3(F) <= V; and
(3) the collection of open sets |jS(F); V e Q)f] u [V; V <£ S>B) is an open cover¬
ing of X.
The family (3 is partially ordered: /3 < y if <=■ and Ve 2>B => /3(F) =
y(F). We leave the following point set argument to the reader: since ir is
locally finite, the maximum principle implies that (3 has a maximal element a
and S>a = 'Y', so that a is the required shrinkage. □
Theorem 12. Let X be a paracompact manifold that is orientable. Then there

exists a smooth n-form a> on X such that a> never vanishes.
149
5: Manifolds
Proof. Let d>' be an orientation of X. Let = {domain cp}^^. Then °U is an

open covering of X. Let (X', SX) be a smooth partition of unity such that X~
is a refinement of °U. For each V e X, let <j>v e O' be such that V c domain <pv-
Then the restriction of cpv to V is also an element of O'. Let (x^,...,
denote the coordinate functions on V. Then the //-form u>v = dxf A • • • A
dxj € C°°(F, An(F)) is nowhere zero on V. Let w = 'Zve-rfv^, where fvu>v
is by definition zero outside V. Then o> g Cx(X, An(X)).
We must show that at is nowhere zero. For x g X, let cp e O' be a coordinate
system about x, with domain U and coordinate functions (yx,..., yn). Then,
for each Kef with U n V A 0,
ojv = dx± A • • • A dxnv = gv dyx A • • • A dyn on U O V,
and gv > 0 on U n V since both cpv and |u — 2 C/v-£t>V)lrr = ( 2 fySv\ dyi A • • • A dyn.
Since 2v€^ fv = 1, there exists V^g'V such thatfVo(x) > 0. SincegVo(x) ^ 0

and each fvgv > 0, (^ve-r fvgv)(x) / 0 and ot(x) + 0. □
Remark. Theorems 9 and 12 completely characterize orientability of

paracompact manifolds by the existence or nonexistence of a nonzero /7-form.
This characterization can be applied to show that the projective space Pn is
orientable if and only if n is odd. This is done by considering the sphere Sn
as a covering space of Pn with covering map p. Let a> be the nonzero n-form
on Sn constructed in the proof of Theorem 10 and its corollary. Then one
can show that for n odd, o> defines an //-form d> on Pn such that a> = p*d>. If
Pn were orientable for n even, then there would exist a nonzero //-form on
Pn and then p*w = ga> for some g ^ 0. On the other hand, one can check
that if Xx ^ x2g Sn are such that //(x^ = p(x2), then g(xx) > 0 o g(x2) < 0,
contradicting the fact that g is never zero.
Remark. A nonzero smooth //-form on a smooth //-manifold is called a
volume element. Thus every orientable paracompact manifold admits a
volume element. The form /(F) dr1 A • • • A drn + 1 on Sn discussed in Theorem
10 and its corollary is the usual volume element on the //-sphere.
Definition. A Riemannian manifold is a smooth manifold X, together with a

map
<,): X—> {inner products on T(X, x)}

xeX
such that for each x e X, <,)(x) (usually denoted <,>*) is an inner product
on T(X, x), and such that <,) is smooth; that is, for each pair Vx, V2 of
smooth vector fields on X, (Vx, F2> is a smooth function, where
<Vi, F2>(x) = <F1(x), F2(x)>*.
The map <(,) is called a Riemannian structure on X.
150
Theorem 13. Let X be a paracompact smooth manifold. Then there exists a

Riemannian structure on X.
Proof. Let (ffS?) be a smooth partition of unity on X such that each

V e i is a coordinate neighborhood. Define a Riemannian structure <,)v on
each V e V by
where (xl5..xn) are the coordinate functions on V. Then define <,) on Xby
O— 2 -/rOy- □
Remark. The converse of Theorem 13 also holds; namely, every Rieman¬

nian manifold is paracompact.
Example 1. Rn is a Riemannian manifold: take {d/dr^ as an orthonormal

basis for the tangent space at each point.
Example 2. Let Xbe a Riemannian manifold, and let (Y, i) be a submanifold

of X. Then a Riemannian structure is given on Y by
<»i, v2>y = <diOi), di(v2)}m (t>i, v2 e T( Y, y)).
Example 3. In view of Example 2, every submanifold of Rn has a Riemannian

structure.
Example 4. Let X and Y be Riemannian manifolds. Then the manifold

X x Y has a Riemannian structure given as follows. For (x, y) e X x Y, the
tangent space T(X x Y, (x, y)) is naturally isomorphic to the direct sum of
the vector spaces T(X, x) and T(Y, y). An inner product on 7(1 x Y, (x, y))
is then given by requiring that this isomorphism be an isometry with the
orthogonal direct sum T(X, x) © T(Y, y).
Definition. Let X and Y be Riemannian manifolds. A map <p: Y is an

isometry if it is smooth, injective, surjective, has a smooth inverse, and is
such that d*
for all 0l5 02 e T(X, x) and x e X.
Remark. Thus an isometry preserves all the structure of a Riemannian

manifold. Two manifolds are equivalent from the viewpoint of Riemannian
geometry if there exists an isometry between them. Such manifolds are said
to be isometric. Note that two Riemannian manifolds as smooth manifolds
can be the same; yet as Riemannian manifolds, be distinct.
151
5: Manifolds
Figure 5.18
Example 5. Consider the torus S'1 x S'1. It has a Riemannian structure as a

submanifold of R3 (see Figure 5.18). On the other hand, it has a Riemannian
structure as a product S1 x S1, where S1 is given a Riemannian structure by
way of its usual imbedding into R2. These two structures are distinct. In fact,
the product structure on S1 x S1 cannot be obtained by representing
S1 x S1 as a submanifold of R3 (see Chapter 8). However, it can be obtained
as a submanifold of R4 since
S1 x S1 c i?2 x R2 = R\
152
Homology theory and the
De Rham theory
6.1 Simplicial homology

We have defined the De Rham cohomology groups H\X, d) for a smooth
manifold X. These groups came from a sequence of maps
C"(X, Al-\X)) C”(jr, Al(X)) -^+C°(X, Al + 1(X))
and H\X, d) = Ker d/Im d. We saw that the dimension of H°(X, d) meas¬
ured the number of connected components of X, and we saw, at least for the
circle X = S1, that the dimension of H\X, d) measured the number of
“holes” in X. We shall now develop similar groups for simplicial complexes.
We shall study a sequence of maps
8 8
Ci-1 Ci cl+1,
where each Ck is an abelian group and where d2 = 0. Then homology groups
Hi will be defined by Hx — ZxIBh where Zx = Ker d: Ct -» C;_x and Bx =
Im 8: C; + i —> Ci. An element of Z; will geometrically be a “chain” of
/-simplices without boundary. An element of Bx will geometrically be a
boundary of a chain of (/ + l)-simplices. The boundary of a 1-simplex
(v0, yx) will be the sum of the 0-simplices v0 and vl with appropriate signs
attached. Similarly, the boundary of a 2-simplex (v0, vlf v2) will be an
appropriate linear combination of its edges (v0, vx), (vx, v2), and (v2, v0).
Definition. Let 5 be an /-simplex, with vertices v0, vlf..., v:. Two orderings
(vH, vj2,..., vu) and (vkl, vk2,..., vkl) of the vertices of j are equivalent if
(k1,...,kl) is an even permutation of (j\,.. This is clearly an
equivalence relation, and for / > 1, it partitions the orderings of v0,..., vx
into two equivalence classes. An oriented simplex is a simplex s together
153
6: Homology theory and the De Rham theory
with a choice of one of these equivalence classes. If v0, vx, ..., vt are the
vertices of 5, the oriented simplex determined by the ordering (v0,..., v,)
will be denoted by <v0, vu ..., v{).
Remark. Note that an oriented 1-simplex has a sense of direction attached

to it, an oriented 2-simplex has a sense of rotation attached to it, and so on
(see Figure 6.1). In fact, each /-simplex s lies in an /-dimensional plane in
Vq» -)-»Vl Vo •-<-
;
0
<V ,Vi> /
Vi
Figure 6.1
some Rm. Orienting s by <t>0, vu ..., is the same as orienting the /-plane
containing s by means of the ordered basis {yx — v0, v2 — v0,..., vt — v0}.
Definition. Let K be a simplicial complex, and let J denote the group of

integers. Let Ct(K, J) denote the factor group of the free abelian group
generated by all oriented simplices of K, modulo the subgroup generated
by all elements of the form <v0, vu v2,..., v{) + <1^, v0, v2,..v{y. Thus
Ci(K, J) is an abelian group called the group of l-chains of K with integer
coefficients. A typical element of this group is of the form
2
s an 1-simplex
»«<J> (”.e S),
where, for each /-simplex s, <s) is some fixed orientation of 5, and where 5
with the opposite orientation is identified with — <5>.
Remark. Given an arbitrary abelian group the group C,(K, &) of

/-chains of K with coefficients in ^ can be defined as the set of all formal
linear combinations
2 Ss<s> (gs 6 <&)

8
subject to the identifications -gs<>o, vlf..t?,> = gs<pu v0,..., t?,>. (We are
writing the group operation in ^ additively.) In particular, Ct(K, is defined
154
6.1: Simplicial homology
for any field in which case Ci(K, #") is a vector space over whose dimen¬
sion equals the number of /-simplices of K. We shall only be interested in the
cases where ^ constitutes the integers the reals R, the complexes C, or the
integers J2 modulo 2; that is, the group of order 2.
Definition. Let <j> = <u0, vu..vt + 1} be an oriented (/ + l)-simplex. The

boundary of <5) is the /-chain defined by
;+1
8<s> = 2 (_1)y<ro5 vl + 1},
1=0
where ~ over a symbol means that symbol is deleted.
Remark. Note that d<-?> is well defined and that Uyio iTo* v1}.. .,v},..
yJ+1], the union of the faces occurring in S<5>, is the topological boundary of
[si
Examples
(1) S<v0, Pi> = <t;i> - <t?0>.

(2) a<r0, vu v2y = <»i, v2y - <v0, v2y + <»„, »i> = <v0, v^y + <»lf v2y + <y2, v0y
(see Figure 6.2).
Vi Vi
Definition. Let K be a simplicial complex, and let & be an abelian group.

The boundary map
Ct(K, 9) Cl+1(K, &)
is the group homomorphism defined by
0(2 &<J>) = 2 gs S<5>.

Lemma. The maps
C; _ x(K, &) Q(K, <&) Cl+1(K, 9)
satisfy d2 = 8 ° 8 — 0.
Proof Since 8 ° d is linear, it suffices to check this on generators
<v0, v±,..., vl + 1y
155
as follows:
i +1
d(8(v0, ...,vl + 1}) = d 2 (-iyoo,..V),..ui+i>
-i = 0
l+1
= 2 (_1); a^°’ • • •» 0/» • • •» y/+i>
y=o
= 2 (—!y 2 (-Ôo,..4..Vj, ■.ui+i>

y=o Li = o
i +1
+ 2 (—l)1_1<t’o» •••> fy. Vl + 1)
i^T+i
= 2 (-i)1+/<ô, vl+1y
i<]
+ 2 ( — 1)J + ^ 1<»0. • • • 5 Ey, . . . , Vt, . . ., V{ + l)>

i>i
= 2 K-!)1^ + (-i)i+5_1Kô,vl+1y
i<J
= 0. □
Definition. Given K and let
Zt(K, &) = [c e Q(K, $); 8c = 0],

BIK,&) = [0c;ceCl+1(tf,SO],
SO = WC*, SO-
Elements of &) are called cycles, and of Bt(K, are called boundaries.
The group Ht(K, @) is called the /th homology group of K with coefficients
in
Remark. It turns out that the groups Ht(K, @) depend only on the topology
of [AT]. If f: [A][L] is a homeomorphism, then there is induced an
isomorphism
/*: HX{K, <$) -> Ht(L, (S).
In particular, if Kx and K2 are simplicial complexes with |Xx] = [i<f2], then

they have the same homology groups.
Exercise. Show that the vector space H0(K, R) has dimension equal to the
number of connected components in [A'].
Example 1. Let K be the 1-skeleton of a 2-simplex; so K consists of three

vertices v0, vx, v2 and three 1-simplices (y0, vx), (vu v2), and {v2, v0) (see
Figure 6.3). Then both C0(K, J) and CX(K, «/) are isomorphic to J © J @ J.
Ci{K, J) = 0 for / > 1. A typical element cx of CX(K, «/) is of the form
Ci = mx(y>0, vxy + m2(vx, v2y + m3(v2, voy (m1; m2, m3 e J).
156
»2
Figure 6.3
Its boundary 8cx is given by
8Cl = m1«u1> - <f0» + m2{(v2y - (vxy) + w3«y0> - <c2»
= («3 - miKvo) + {mx — wiaXî) + (m2 - m3)<y2>.

Thus cx e ZX(K, J) if and only if
m3 — mx — 0, mx — m2 = 0, m2 — m3 = 0,
that is, if and only if mx = m2 — m3, so
ZX(K, J) = [«(Oo, vxy + <»!, v2y + (v2, voy);
Furthermore, BX{K, J) = 0 because C2(K, Jf) = 0. Hence
HX(K, J) = ZX(K, J)!BX{K, J) ~ J.
To compute H0(K, J), note that a typical cycle c0 e Z0(K, J) is of the form
c0 = «i<c0> + «2<>i> + «3<^2> («i, n2, n3 e J).
Then c0 = 8cx for some
Ci = mx(v0, »!> + m2(vi, v2y + m3(v2, c0> e CX(K, J)
if and only if there exist (integer) solutions to the equations
m3 - mx = nx,
mx - m2 — n2,
m2 — m3 — n3.
It is easy to check that such a solution exists if and only if nx + n2 + n3 = 0.
Thus
B0(K, <f) = [«i<u0> + «2<î> + «3<^2>; nx + n2 + n3 = 0].
Let <p: Z0(K,«/) -> J be the homomorphism defined by
<p(nx(v0y + n2(vxy + n3(v2y) = nx + n2 + n3.
Then the kernel of <p is just B0(K, «/); thus
H0(K, J) = Z0(K, S)/B0(K, S)%J.
157
Example 2. Let K be the complex consisting of all the faces of a 2-simplex

(v0, vx, v2). Then, as in Example 1,
H0(K, J) ~ J.
Moreover, as before,
ZX(K, J) = [n«n0, »i> + <*>!, v2y + <v2, v0y); n e J].

Now, however,
C2(K, J) = [n(v0, vx, v2y;ne J],

so that
BX(K, J) = [d(n(p0, vx, v2y); neJ]

= [n«v 1, v2y - <t>0, v2y + <y0, vxy); ns J]
= [n«v0, vxy + <»!, u2> + <u2, ô»; ns J]
= ZX(K, J).
Hence
HX(K, J) = ZX(K, S)/BX(K, JO = 0.
Finally, since d(n(y0, vx, t>2» = 0 if and only if n = 0, Z2(K, Jf) — 0,

and hence
tf2(tf, JO = 0.
Definitions. Let K be a simplicial complex. The /th Betti number j8, of K is

the integer
j8, = dim #,(£, i?).
The Euler characteristic x(K) of K is the integer
dim K
X(K) = ^ (-O'ft-
1 = 0
Theorem. Let K be a simplicial complex. For each l with 0 < / < dim K, let
a, denote the number of l-simplices in K. Then
dim K
x(K)= 2 (-!)■“<;
1=0
that is, x(K) is equal to the number of vertices — the number of edges + the
number of 2-faces — • • • .
Proof. For each /, 0 < / < dim K, consider the linear map
C>-X(K, R) Q(K, R),
158
where C_x is by definition the zero space. Then, by the rank and nullity
theorem of linear algebra,
ai = dim Ct(K, R) = dim Ker 8 + dim Im 8

= dim ZX{K, R) + dim Bx -X(K, R) (/ = 0, 1,..., dim K).
Moreover,
ft = dim Ht(K, R) = dim [Z,(K, K)/Bt(Kt R)]

— dim ZX(K, R) — dim BX(K, R).
Thus
dim K
X(K) = 2 (-l)'ft
1=0
dim K
= J (—^‘[dim ZX(K, R) — dim BX(K, i?)]

1 = 0
dim K dim K
= J (—l)1 dim ZX(K, R) + 2 (-1),+Idim5^,i?)

1=0 1 = 0
dim K dim K
= 2 (-1)' dim ZX(K, i?) + 2 (-1)' dim B‘-1(^> R)

1=0 1 = 1
(since dim Bx = 0 for / = dim K)
dim K
= 2 (—l)z[dim Zx(K, R) + dim Bi_±(K, i?)]
1=0
(since dim = 0)
dim K
= 2 (-i)' «!• □
1 = 0
Remark. If [X] is homeomorphic to a connected compact orientable

2-dimensional manifold, then it turns out that ft = 1 and ft = 1, so that
X(K) = ft - ft + ft = 2 - ft
or
ft = 2 - X(K).
Furthermore, ft is always even for such K. It can be shown that any such
surface is homeomorphic to a sphere with a certain number of “handles”
attached; ^ft is just the number of handles (see Figure 6.4).
Figure 6.4
159
Thus the homology groups completely determine the homeomorphis'm

class of connected compact orientable surfaces. However, for higher di¬
mensional manifolds, the homology groups contain comparatively little
information.
Remark. We have been discussing a homology theory for simplicial
complexes, that is, a theory arising from a sequence of groups and homo-
morphisms
• • • cueK, R) Q(K, R) Cl + 1(K, *)«—•••,

where the map 8 lowers the dimension of chains. On the other hand, in
studying De Rham cohomology, we used a sequence
•• Cm(X, Al~ *(X)) C °°(X, A\X)) C "(X, A! + \X)) —>
where the map d raised dimension (degree). In order to compare these two
theories, it is convenient to define a simplicial cohomology theory. This is
done by passing to dual spaces.
Definition. Let K be a simplicial complex. For 0 < / < dim K, let
C\K) = [CIK, A)]*.
Let 8*: C\K) -» Cl + 1(K) be the adjoint of the map 8: Cl + 1(K, R) ->
Ci(K, R). Thus 8* is defined by
[g*(<p)](c) - <p(8c) (cpeC \K) ;ceC1 + ,(K, R)).
Then we get a sequence
-* C'~\K) C\K) Cl + 1(K) —> • • •.
Moreover, g* o 8* = 0 since 8 o g = 0. Let
ZW - [<pGCTO;g*<p = 0],
B\K) = [8*r,cpeCl-\K)],
H\K) = Z\K)/B\K).
Elements of C!(A) are called cochains; elements of Z\K) are cocycles',

elements of R!(/f) are coboundaries. The map 8* is the coboundary operator.
H\K) is the /th cohomology group of K.
Exercise. Verify that Hl(K) is isomorphic to [Hi(K, i?)]*.
We shall need an explicit formula exhibiting the effect of the coboundary

operator 8*. For each oriented /-simplex > of K, let <p<s> e C l(K) be defined by
f 1 (if <*> = <*»
<P<s><0 = < -1 (if <0 = -<J»

I 0 (if t ± j).
160
6.2: De Rham’s theorem
Thus, if RiiX • • • > is a basis for C((X, R) (so that {5l5..., sm} is the set
of all /-simplices of K), then {<p<Sl>,..T<Sm>} is the dual basis for C\K).
Since d* is linear, we need only compute the effect of d* on these generators
<P<s>-
Lemma
d*<P< VO.*,> = 2 <P<t),Vo.t)j>5
where 2' denotes the sum over all vertices v e K such that (y, v0, yl5..., y,)
is an (/ + l)-simplex of K.
Proof. We need only check this formula on oriented (/ + l)-simplices
<0 = <>o, wl3..., w, + 1>
of K. If we set (s) = <y0, yl5..., y;>, the left side yields
(8*<P<s>)«0) = <P<s>(8<t»
= (-!)'< w0, ..., w, + i>^

i+i
= 2 (—1)V<s>«M;o, • • ., wt, . . W, + 1».
i= 0
But each term of this sum is zero unless, for some j, (vv0,..., w>y,..., wi+1) = (s);
that is, unless (s’) is a face of (t). If (^) is a face of (t), then (t) = (y, v0,..y,)
for some vertex v e K, in which case either
(1) <0 = <v, v0,..., vt) and (3*<p<s>)«0) = 1; or

(2) <0 = -<y, v0,..., y,> and (SV<s>)«0) = -1.
Thus
{ 1 (if = <y, v0,..v^ for some y)

-1 (if <t> = — <y, y0,..., y,> for some v)
0 (in all other cases)
= ^2 <p<».»o.
Since this holds for arbitrary <t>, the formula is established. □
6.2 De Rham’s theorem

Definition. A smoothly triangulated manifold is a triple (X, A', A), where X is a
C" manifold, K is a simplicial complex, and h: [K] -> X is a homeo-
morphism such that for each simplex s of K, the map A|[s]: [s] -*■ X has
an extension hs to a neighborhood U of [s] in the plane of [s] such that
hs: U —> X is a smooth submanifold.
161
Remark. If dim X = n, we need only require that this last condition be

satisfied for each n-simplex of K, since every simplex of K is a face of an
^-simplex and since restrictions of smooth maps to submanifolds are smooth.
Example. Let X = Sn. Let K be the n-skeleton of an (n + l)-simplex

circumscribed about Sn. Let h: [.£] -> Sn be radial projection. Then (X, K, h)
is a smoothly triangulated manifold (Ligure 6.5).
u
Figure 6.5
Remark. It can be shown that every compact smooth manifold can be

smoothly triangulated. The proof is difficult and will not be presented here.
Note that smoothly triangulated manifolds are compact because [W] is
compact for each (finite) simplicial complex K.
The goal of this section is to show that for smoothly triangulated manifolds
(X, K, h), the De Rham cohomology of X is isomorphic to the simplicial
cohomology of K. Lor this, we shall need the following facts about barycentric
coordinates. Recall that we have previously discussed the barycentric coordi¬
nates of a point relative to the vertices of a simplex containing it. We now
extend this concept.
Definition. Let K be a simplicial complex and let vlf...,vm denote the vertices
of K. Suppose x e [!£]. For j e {1,..., m}, the y'th barycentric coordinate
b,(x) of x is defined as follows. If x ^ St(y;), then Z>y(x) = 0; if x e St(yy),
then xe(i) for some simplex s having as a vertex, and bj(x) is equal to
the barycentric coordinate of x in s relative to the vertex Vj.
162
Remark. The following facts are easily verified.
(1) bj\ [.K] -> R is a continuous function.

(2) bj(x) > 0 and 27=1 b,(x) = 1 for each x e [AT].
(3) x = 27=i blx)v}.
(4) bJo(x) ^ 0, bh(x) A 0,..., bu(x) A 0 for some x e [K] if and only if
vjo,..., vu are the vertices of an /-simplex of K.
Definition. Let K be a simplicial complex, and let s be a simplex of K. The

star of s is the union of all the open simplices (t) of K such that (5) is a
face of (t).
Remarks
(1) For s = v a 0-simplex (i.e., a vertex) of K, St(» = St(y), as defined above.
(2) St(s) is an open set in [AT], (This is an elementary consequence of (3).)
(3) If (s') = (vi0,..., vu) and x e [K], then x e St(s) if and only if bu(x) A 0
for all 1 e {0,, /}.
(4) If (s) = (vi0,vh), then
[A/] — St(s) = [x e [A/]; bu(x) = 0 for some i e {0,..., /}].
(5) If sx and s are /-simplices of K with s± / s, then [sx] <= [A/] — St(s).
Given a smoothly triangulated manifold (X, K, h), we want to define, for

each /, an isomorphism from H\X, d) onto H\K). To do this, note that
homomorphisms f: Hl(X, d)-> Hl(K) are defined whenever there is given
a sequence of linear maps f: C°°(X, Al(X)) -> Cl(K) such that 8* of =
f + 1 o d for all /.
->C °°(X, Al(X)) C ™(X, Al + 1(X)) -> ■ ■ ■

A A+i
v 8*
->C\K) -> Cl + 1(K) —> • • •
For then f(Z\X, d)) c Z\K), because rfu = 0(ue C‘(X, d)) implies that
=f+iW =f+1(0) = 0.
Also f(Bl(X, d)) <= B\K), because to = dr (t e Cl~1(X, d)) implies that
f(oj) =f(dr) = e*(/,_1r)elma*.
Thus/; induces
f: H\X, d) = Z\X, d)/B\X, d) -> Z‘(K)/B‘(K) = H\K).
We now proceed to define such a sequence of linear maps
J\ C”(X, A‘(X))-^Cl(K).
For to e C °°(X, A!(3Q), (to) will be a linear functional on Ct(K). Thus it

suffices to specify the values of J; (o>) on basis elements of Ct(K), that is, on
163
oriented /-simplices <s>. To do this, consider the smooth map hs: UX.
Then h*{u>) is a smooth /-form on U, an open set in the plane of [5]; that is,
in an /-dimensional Euclidean space. We define (a»)«5>) to be the integral
of this /-form over <s>:
f («x<j» - f
Jl J <s>
hfn-
In other words, let (rl5..., rt) denote coordinates in the plane of [5] consistent
with the orientation of <.s>; so if <5) = <v0,. let (rl5 ...,/*,) be
coordinates relative to the ordered basis {v1 — v0,..., vt — y0}- Then
hf(aj) — gdrx A • • • A dr,
for some continuous function g on U, and
(w)«j» — gdr^-'drx (Riemann integral).

Jl J[s]
Note that this integral is independent of the homeomorphism h\ that is, it

depends only on the point set h([s]) and its orientation by the change of
variables theorem for integrals.
Claim:
This is just Stokes’s theorem. For given any smooth /-form a> and oriented
(/ + l)-simplex <5),
f ° d(co)]«sy) = f (hs)*(da>)
Jl +1 J<S>
=f J <S>
d(hf(w))
= (by Stokes’s theorem)

J8<s>
= <«>)(0<s»
= 8*°J, MW-
Thus Jl induces a homomorphism J(: Hl(X, d) H\K).
Theorem (De Rham’s Theorem). Let (X, K, h) be a smoothly triangulated

manifold. Then
ji:H\X,d)^H\K)
is an isomorphism for each l (0 < / < dim X).
This theorem is a consequence of the following two lemmas.
164
Lemma 1. There exists a sequence of linear maps

«i: C \K) -> C"(X, A'(X)) (0 < / < dim X)
with the following properties.
(1) d ° a; = ai +1 o d*.
(2) j; o a, = identity.
(3) //" c° denotes the 0-cochain such that c°(v) = 1 for each vertex v in K,
then a0(c°) = 1; that is, a0(c°) is the 0-form equal to the constant
function 1.
(4) If (s ') is an oriented l-simplex of K, then the l-form a,(<p<s>) is identically
zero in a neighborhood of X — St (s).
Lemma 2. Let oj be a closed l-form on X. Suppose J( (cu) = 8*c for some

c e C!_1(-^)- Then there exists an (/ — 1 )-form r on Xsuch that l (r) = c
and dr = oj.
Remark. Lemma 1 shows that J; is surjective. For given zgZ\K), let

oj = a;(z). Then a> eZ\X, d) because
dco = d o cc,(z) = al + 1 o d*(z) = a; + i(0) = 0.
Furthermore, (a>) = ° a,(z) = z. Thus J(: Z\X, d) -> Z\K) is surjective;

hence so is J;. (Note that Property (1) says that the map a, induces a homo¬
morphism a,: H\K) -> H‘(X, d). Property (2) says that this map is a right
inverse to J;.)
Lemma 2 shows that J( is injective. For if o> eZ\X, d) and J\(a>)eBl(K),
then oj e B\X, d) by Lemma 2.
Thus Lemmas 1 and 2 together do, as claimed, imply De Rham’s theorem.
Proof of Lemma 1. For notational convenience, we shall identity [X] and X

through the homeomorphism h\ that is, we shall assume that [X] = X and
that h = identity.
Step 1. We first construct a special partition of unity, subordinate to the
open covering
{St (v); v is a vertex of X}
of X. Let vlt...,vm denote the vertices of X. For each je{ 1,..., m), let
bj denote the yth barycentric coordinate function on [X] = X and let
1
x e X; bj(x) > (n = dim X),
n + 1
x s A'; Hx) - VT~1

165
Then Fy and Gy are closed sets in X with the following properties (see Figure
6.6).
(1) F, c St (Vj).
(2) X - St (»,) c Gy.
(3) Fj r\ G) = 0 that is, Fy <=■ Gy.
(4) There exists a smooth function fi> 0 such that /y > 0 on Fy, and
/y — 0 on G;. (Fy is a closed set in the compact space X, hence is compact;
thus an /y > 0 can be found which is greater than 0 on Fy and equal to 0
outside the open set Gy => Fy.)
(5) The closed sets Fy cover X. (Given xe X, then x e (5) for some simplex
(5) = (vjo,..., vu) of dimension / < n. Now 6y(x) = 0 for j {j0,.. .,j\} and
2i=obu(x) = 1. Since/ + 1 < n + 1, bfcx) > 1 /(n + 1)for some;eO'0,.
Thus x e Fy for this j.) In particular, for each x e X, /y(x) + 0 for some j.
Furthermore, Gy is an open covering of X.
(6) From (5), 27=i/y > 0, so that gy = /y/2fe=1 A is defined and smooth
on X. Furthermore, {gy} is a smooth partition of unity on X subordinate to
{Gy}; that is, 27= 1 gy = 1, and gy vanishes outside G{. Since G) c St (v,), the
partition of unity {gy} is also subordinate to the open covering {St (uy)}.
Step 2. We shall now define a, in terms of the smooth functions {g,}
defined above. Since a, is to be linear, it suffices to specify the values of a, on
the generators <p<s> of Cl(K). For <s> = <Uy0,..., an oriented /-simplex,
we define a,(9J<s>) to be the /-form
«i(<p<s>) = /! 2 dSio A • • • A dgh A • • • A dgu.

i= 0
Verification of properties (l)-{4)
Property (1). Clearly,
d° «i(«p<s>) = (/ + 1)! dgjo A • • • A dgu.
166
On the other hand,
= (/ + 1)! X f Sk dgj0 A • • • A dgh

k L
2 dSk A dgi0 A • • • A dgh A • • • A dgu
Claim. If the vertices vk, v,0,..., vu are distinct and yet are not the vertices
of an (/ + l)-simplex of K, then
gk dgio A • • • A dgh = 0.
For, if x $ St (vk), then gk(x) = 0. If x e St (vk), then bk(x) # 0. But now

bh(x) = 0 for some / e {0,..., /}, for otherwise bk(x) # 0, bjo(x) ^ 0,...,
bit(x) ^ 0, hence (vk, vjo,..., vu) is an (/ + l)-simplex. But this is a contra¬
diction. Using this i, let
V=\yeXih,iy) < ~
Then U is an open set in X containing x, and gu is identically zero on U

because U <= Gh. Hence dgh = 0 on U, and, in particular, dgh{x) = 0. This
completes the proof of the claim.
Applying this result to the terms of the above expression for al + 1° d*(gs<s>)
yields
(A) 2' gk dg>o A • • • A dgu = 2 gk dgio A • •' A dgu,
since those terms on the right-hand side which do not appear on the left are
identically zero; and
(B) 2' 2 dg* A dgj0 A • • • A dgh A • • • A dgh

k i=0
l
= 2 t-1)1 2' gu dgk A • • • A dgu A • • • A dgh
i=0 k
2 C-iy
i= 0 kiVo
2 gu dgk A dgj° A • • • A dg]{ A • • • A dgh
l
2 (-!)' 2 Zh dgk A dgio A • • • A dgh A • • • A dgh
i=0 k*Jt
i-v s~rJ,
I
2 (.-tygiti-dgii) A dgio A • • • A dgh A • • • A
1=0
167
Hence, subtracting (B) from (A),
«J + 1 ° S*(<p<s>) = (/ + 1)! dgj o A • • • A dgh
= (/ + 1)! 4>/o A • • • A rfgy,

= d ° «,(<p<s>).
Property (3). Since cc0(<p<Vj>) =
(m \ m
2 ^<«i>) = 2 & =
x=i / r=x
Property (4). Suppose <5> = (vio,..vu>. Then
«i(<P<s>) = /! 2
;=l
dgjo A • • ■ A 4?;t A • • • A dgu.
Note that if x e X is such that &;;c(x) < l/(n + 2) for some k e {0,../},
then x e GJk, so that gjk and dgjk, and hence aj(<p<s>), are zero at Thus
ai(9,<s>) is identically zero on
x e X; bfk(x) < ^ for some k e {0,..., /} ,
which is an open set containing X — St (s).
Property (2). Proof by induction. For / = 0, J0 ° «0(<p<Uy>) (J e {1,..., m})

is the 0-cochain given by
Jo ° «o(9»<0,>) «»*>) = JQ (gi) <Vk> = gj(vk).
But note that gj(vk) — 0 for k =£ j since vk $ St {vj) and gj = 0 outside

St (Vj). Furthermore,
1 = (for each k).

2 g&k) = gk(vk)
y=i
Hence
(if k = j)
Jo ° «o(9<*,>) «»*>) =
(if k + j)
= >«»*».
Since this holds for all j and k, |0 ° “o = identity as required.
Now assume Property (2) for dimension / - 1. For <s> and <t> oriented
/-simplices of K,
I ° «i(9><») <0 = «i(«P<s>)-

IJ‘ J J<t>
168
We must show that this equals 1 if <s> = </>, and 0 if j # t. That this is zero
for 5 # tis a consequence of Property (4) since a,(<p<s>) is identically zero in a
neighborhood of X - St($)=> [f]. So we need only check that J<s) «;(<p<s>) = 1.
For this, let <r> = (vh,. and 5 - fvio, vh,vu). Then
J
f <S>
«i(0*9<r>) =fJ <S>
d[a, _1((p<r>)]
= JL
8(.S>
“«-!(?<'>)•
Butd<» = <r> plus an alternating sum of other oriented (/- l)-simplices, so
L v “l-lfoo) =
J8<s> J<r>
«I-l(?<r>) = 1
by induction. Hence
1 = ai(a*<p<r>) = al(<p<s> + terms of type «p<t> (f ^ j))

J<s> J<s>
= “j(<□
J <S>
In order to prove Lemma 2, we shall need the following lemma.
Lemma 3. Let s be a k-simplex in Rn.

(ar) Suppose r > 0 and k > 1. Let a> be a smooth closed r-form defined
“near” [jk-1]; that is, defined in a neighborhood of[sk~1]. If k = r + 1,
assume further that J.<s) w = 0. Then there exists a smooth closed r-form t
defined near [s] such that t = a> near [s'0-1].
(br) Suppose r > 1 and k > 1. Let w be a smooth closed r-form defined
near [s]. Suppose t is a smooth (r — \)-form defined near [.s*-1] such that
dr = a> near [sfc_1]. If k = r, assume further that [0 t = | w. Then
there exists a smooth (r — \)-form T defined near [v] such that T = r
near [s'1-1], and dr' = w near [j].
Remark. That the integral conditions are necessary in (ar) and (br) is a
consequence of Stokes’s theorem. For in (ar), if r exists, then
f --f .
Jd<s> J8<s>
and in (br), if r exists, then
f --f
J<s> J<s>
Proof of Lemma 3. Proof by induction. We shall first verify (a0) and then
establish that
(ao) ^ (î) ^ (ai) ^ (^2) ^ •
169
(a0): to is a smooth 0-form; that is, a smooth function defined near [5-fc “1 ];
and dcj — 0. Hence co is constant on the components of its domain. If k > 1,
[s*-1] is connected, so w is a constant function c in a neighborhood of
[jfc-1]. Set t = c in a neighborhood of [5]. If k = 1, then <s> = <v0, vx)
for some pair of vertices v0, v1} and
0 = I co = — co(v0).
Je<s>
Thus the constant value of co near vx equals the constant value of co near u0;
that is, co is constant near [s*-1] as before. Once again, set r equal to this
constant function on a neighborhood of [5].
(ar-i) => (br) : co is a closed r-form (r > 1) defined on an open set con¬
taining [5]. By Poincare’s lemma (Section 5.2), co is exact near [5]; that is,
there exists a smooth (r — l)-form tx defined near [5] such that drx = co
near [5]. (To see that Poincare’s lemma applies here, we need only note that
any open set containing [s] must contain another open set about [5] which is
diffeomorphic (smoothly homeomorphic with a smooth inverse) to an open
ball. In fact, we can choose a star-shaped region containing [5].) Now in
general, rx will not be equal to t near [s'0-1]. Consider the difference — t
near [s*-1]. It is closed since, near [s*-1],
d(Ti — r) — co — co = 0.
Furthermore, if k = {r — 1) + 1 = r, then
= 0 (by hypothesis).
Thus we can apply (ar_x) to the form tx — There exists a smooth closed
r.
(r — l)-form n defined near [5] such that /u. = tx — r near [s*-1]. Let
T' = Ti — v- Then r' is a smooth (r — l)-form defined near [5] such that
t' = t\ — m = T near [ste-1], and dr' = drx - dp = co - 0 = co near [s].
(br) => (ar): <s> = <y0,..., vk} for some choice of vertices v0,..vk; let
<t) = <vx,..., vk}. LetF = [^'c“1] — (t). Since co is closed, Poincare’s lemma
asserts the existence of a smooth (r — l)-form ^ defined near F such that
dfj. = co near F. (F is star-shaped; hence any open set containing F contains
a star-shaped neighborhood U of F) (see Figure 6.7). In particular, d^ = co
near [t*-2].
170
If k > 1, we would like to apply (br) to the forms w and /x and the
(k — l)-simplex t. In order to do this we must check if k — 1 = r, then
J<t> w - Jg<{> n = 0. But, letting c = d(s} - <t> so that dc = -0<t>,
f
Jew
F = |
J<t>
« + 1
J u
,
+
3
II
\/d
A
,
~L"+J [“ (since each simplex of c is contained

in F and d/x — a> near F)
= f
Jew
= 0 (by hypothesis).
Applying (br), there exists a form /a' defined near [/] such that n' — /x near
[,tk~2] and d\x = a> near [/]. Let /x2 be the form defined near [s^1] by glueing
together /x and /a' along their common domain, an open set where they agree
(Figure 6.8). Then dfx2 = near [sfc_1], since both fx and /z' have this property
in their domains of definition.
If k = 1, then [s*-1] consists of two vertices v0, vx. Since o> is closed,
Poincare’s lemma guarantees the existence of smooth (r — l)-forms ^ near vt
domain n'
171
(/ = o,1), with dî = co. By shrinking domains, we can assume (domain /x0)
and (domain î) are disjoint (Figure 6.9). This defines /x2 near [sk x], with
d/c2 — co near [s'0-1] as before.
domain /u i
Finally, let / be a smooth function which is identically 1 in a small neigh¬

borhood of [s*-1], and identically zero outside the domain of /x2. Then ffi2
is a smooth (r — l)-form defined near [s]. Let r = J(//x2). Then r is a closed
/■-form defined near [s], and we have near [s*-1]
r = d(f/x2) = df A fi2 + f dfi2 = dfi2 = co,
since /= 1 and df = 0 near □
Proof of Lemma 2. We shall construct inductively a sequence
To, t15 ..., Tn = T {n = dim X)
of (/ — l)-forms such that
(1) rk is defined in a neighborhood of the k-skeleton [Kk\ of K,

(2) drk — co near [Afc],
(3) rk = rk_1 near and
(4) J,_,(t,_1) = c.
Note that this will prove the lemma because (4) implies that for each oriented
(/ — l)-simplex <$> of [/£] and each k > 1 — 1,
f (Tfc)«J» = f rk=( T,.! = f (t1-1)«‘S’» = C«5».

Ji-l J<S> J<s> Jl-1
To construct r0, cover K° by a collection of mutually disjoint balls. Since
cu is closed, co is exact in each of these balls by Poincare’s lemma. Hence
there exists a smooth (/ — l)-form t'0, defined on the union of these balls,
such that dr'0 = co there. If / — 1 # 0, take r0 = t'0. If / — 1 = 0, we want
J0 (To) = c. But for Vj a vertex of [AT],
f (To)«^» = f To = T^).
JO J <v,>
Let cij — c(vj) — To(yy), and define t0 on the ball about v5 by r0 = t'0 +
Then dr0 = dr'0 = <o near [A0], and JQ (r0) — c as required.
172
Now assume that rk_1 has been constructed with Properties (l)-(4). To
construct rfc, note that if we can find, for each ^-simplex s, a smooth (/ — 1)-
form rk(s) defined in a neighborhood of [5] such that d(rk(s)) = w near [>],
and Tfc(5) = rk_1 near then glueing will yield a smooth (/ - l)-form
r'k satisfying (l)-(3).
To construct rk(s), we shall apply (b,) of Lemma 3. Note that w is a smooth
closed /-form defined near [5] and that rk_1 is a smooth (/ - l)-form defined
near [s'1'1] such that drk_1 = u> near [s*-1]. Furthermore, if k = /, then
f U) = f (to)«J» «5> = 5 together with either orientation)

J<s> Jl
= 8*c«s» (by hypothesis)

= c(d(s»
- jfc-i
f Ofc-iXaO)) (by (4) since k — /)
-f J8<s>
Hence we can apply (/?,). There exists a smooth (/ - l)-form rfc(s) near [5]
such that t^) = rk_1 near 0fc-1] and d(rk(s)) = a> near [5].
This constructs r'k satisfying (l)-(3). If k ^ l — 1, set rfc = rk. If k = l — 1,
we have satisfying (1)—(3), and we want rl_1 such that J,_1(Ti-i) = c.
Let cx = c — J,_1 and define in a neighborhood of [A'-1] by
Ti-i = T;-i + «/-i(Ci),
where a;_! is the linear map C'_1(^) -> Ca(X, A!_1(30) defined in Lemma 1.
For each r and each oriented /--simplex <$>, note that ur(cp<s>) is identically
zero on a neighborhood of X — St (,y). In particular, «r(q9<s>) is identically zero
near [Kr_1]. Since each ce Cr(K) is a linear combination of such <p<s>, we
have ar(c) identically zero near [A'r_1] for each r-cochain c.
Applying this first with r — /, then with r = l — 1, we find
drt_x = dr'i _ x + d o a:_x (cx) = dj\-x + «; 0 3*(cx) = <7rJ_x = w
near [Kl x] and
ti- 1 = T(-i + K;-i(ci) = Ti-i = Ti-2
near [Kl 2]. Thus r,_x satisfies (l)-(3) with k — l — 1. Property (4) is also
satisfied:
= (c - Cx) + Cx
= c. □
Remark 1. De Rham’s theorem shows that the simplicial cohomology

groups (with coefficients in R) of a smoothly triangulated manifold (X, K, h)
173
6: Homology Theory and the De Rham Theory
are isomorphic to the De Rham cohomology groups of X. In particular, these

groups are independent of the triangulation (K, h) of X. Since the cohomology
groups are dual to the homology groups, the groups H^K, R), for |W] a
smooth manifold, also depend on [K] only, not on the particular simplicial
subdivision K.
Remark 2. The direct sum 2fimox © H\X, d) can be given the structure of
an associative algebra as follows. Recall that 2f=nox © CX(X, Al(X)) is an
associative algebra under exterior multiplication A. 2©Z\X, d) is a
subalgebra, for if daj = 0 and dr = 0, then
d(to A t) = (da>) A T + to A (dr) - 0.
2 © B\X, d) is an ideal in 2 © Z l(X, d), for if to — dp and dr — 0, then

to A r = d(p A t). Hence
y@H'(X,d) =
l
y®{Z\X,d)IB\X,d))^^®(z\X,d) N®B\X,d)\
l 1 \ I l /
is also an associative algebra. In particular, 2; © H\X, d) is a ring, called

the De Rham cohomology ring of X.
It is also possible to define a product, called the cup product, of simplicial
cochains in such a way that 2i © Hl(K) becomes an algebra. It can be shown
that the isomorphism J: 2; © H\X, d) -* 2i © Hl{K) is then an algebra
isomorphism.
Remark 3. Lemma 3 contains in disguise a proof that
(if 0 < / < ri),

(if / = 0, ri).
For if to is a closed /-form (0 < l < ri) defined on a neighborhood of the

n-skeleton [sn] of an (n + l)-simplex s, then it was shown that to extends to
a closed (and hence exact) /-form on [j]. This implies that H\Sn, d) — 0 for
0 < / < n. It was shown for / = n that any closed n-form to, defined near
[sn] such that Jg<s> to = 0, is also exact. The map Zn(Sn, d) -> R defined
by to->Ja<s>to is then a homomorphism with kernel B\Sn, d). Hence
Hn(Sn, d) = R. Also H°(Sn, d) = R because Sn is connected.
We have tacitly assumed here that any closed /-form to on Sn can be
extended to a closed /-form defined in a neighborhood of Sn. Y*to is such an
extension, where Y: Rn + 1 - {0}-^ Sn is radial projection.
174
Intrinsic Riemannian
geometry of surfaces
7.1 Parallel translation and connections

Definition. Let M be an oriented Riemannian manifold of dimension 2. Let
T{M) denote the tangent bundle of M. Let
S(M) = [(m,v)eT(M);<v,v> = 1],
S(M) is called the sphere bundle, or circle bundle, of M.
The notation (m, v) for a point of T(M) {or S'(M)} is redundant since
v e T(M, m). Nevertheless, we use it to emphasize that v is a tangent vector
at m.
Remarks
(1) S(M) is a smooth manifold of dimension 3. The function/: T(M)—> R1,
given by f(m, v) = <[v, v), is smooth, and df ^ 0 whenever f — 1, so the
implicit function theorem applies.
(2) Note that the circle S1 = [z e C; |z| = 1] is a group under (complex)
multiplication. Since eiei-ewz = e*(0i+ea), the group S1 is just the group of
rotations of the oriented plane R2. This group acts on S(M): there exists a
smooth map
A: S1 x S(M)->S(M)
given by
A(g, (m, v)) = (m, gv) (g e S1; (m, v) e S(M)),
where gv is the image of the vector v under rotation by g in the oriented plane
T(M, m) (Figure 7.1). So, if g = ew, and {vl9 v2} is any oriented orthonormal
basis for T(M, m), then v = c1v1 + c2t>2 f°r some cl5 c2 e R1, and
gv = (cx cos 0 — c2 sin 9)v1 + (cx sin 9 + c2 cos 9)v2.
175
7: Intrinsic Riemannian geometry of surfaces
We shall often denote A(g, (m, v)) by g(ni, v). Then g: S(M) -> S(M) is a
smooth map for each g e S1.
(3) If it: S(M) M denotes projection, then -tt_1(m) is just the unit circle
in T(M, m). Moreover, if (m, vx) and (m, v2) are any two elements of 7r_1(m),
then there exists a unique g e S1 such that (m, v2) = g(m, vx). (Take g = ew,
where 6 is the positive angle of rotation from v1 to v2.)
(4) S(M) is locally a product space. For let U be a coordinate neighbor¬
hood in M, with coordinate functions (xl5 x2). Let ex be the vector field
(8/dx1)/\\8/dx1\\, where ||0/cbc1|| = <(S/dxi), (d/dxi)>1/2. Then e1 is a smooth
vector field on U, which is everywhere of length 1. Thus ex defines a smooth
map
c: [/-> 77--1([/) by c(m) — (m, e^m)).
Clearly v o c = iv. Now define B: U x S1 -> 7t_1([/) by
B(m, g) = gc(m) = (m, ge^m)) = A(g, (m, ex(m)).
Then it is easy to verify that B is smooth, injective, and surjective; and that
dB is everywhere nonsingular so that 2?-1 is also smooth.
(5) It is not true that S(M) is globally a product of S1 with M. If there
exists a smooth nonzero vector field on M, then the above argument shows
that S(M) is diffeomorphic with M x S1. However, there do not exist such
nonzero vector fields in general. (For example, M = S2.)
For M = R2, the notion of translating a tangent vector parallel to itself is
clear. We now propose to generalize it and introduce the concept of parallel
translation of tangent vectors on arbitrary 2-dimensional oriented Riemannian
manifolds. It will turn out that we will be able to parallel translate vectors
along curves from one point to another, but that the result will depend on the
curve. In particular, if we parallel translate around a closed curve, we may not
get back to our original vector. The new vector will differ from the original
vector by a rotation; i.e., by an element of S1. For M = R2, a “flat” space,
this rotation is zero. For arbitrary M, this rotation (or, more precisely, the
limit of it as the curve shrinks to a point m) will measure the “curvature” of
M at m.
We develop this notion of parallel translation in order to obtain an intrinsic
meaning for curvature of the Riemannian manifold M; that is, a meaning
independent of any ambient space in which M may lie. In Chapter 8 we will
interpret this curvature differently when M is a submanifold of R3.
176
7.1: Parallel translation and connections
We shall require that parallel translation be an isometry. Thus, parallel

translation of a unit vector along a curve a: [a, £>] —> M will determine a unit
tangent vector &(t) e T(M, «(/)) for each te[a,b\. If ve-n~\a{d)\ then
parallel translation of v will determine a curve «: [a, b] -> S(M) such that
tt o a = a. Moreover, if
vx e and vx = gv (geS1),
then the curve ax: [a, b] -> S(M), determined by parallel translating vlf will
be given by
«i(0 = ga(0 (t e [a, 6]).

Conversely, if, corresponding to each curve <x:[a,b]->M and each unit
tangent vector v at a (a), there existed a unique “lift” 5: [a, b] -> S(M), with
the above properties, then a notion of parallel translation is defined (see
Figure 7.2).
<-7r '[Image a]
Recall that in the theory of covering spaces, each curve had a unique lift
because the fibers p_1(x) were discrete. However, here the fibers 77-_1(m) are
not discrete; they are circles. Hence lifts are not unique. In fact, we do not
even know in which direction to start moving. (There is a whole line of
vectors v e T(S(M), (m, v)) such that cItt(v) = a (a); each of these is a candi¬
date for a(a).) So given m e M and v e T(M, m), we need a way of determining,
for each curve a through m, an initial direction for a; that is, we need a way
of choosing, for each a(a) e T(M, m), a vector a (a) e T(S(M), (m, t>)) such
that dv(a(a)) = d(a). Choosing the vector &(a) is more primitive than finding
the lift a, but it will turn out that when the choice is made at every point of
77-_1(a([fl, 6])), the lift—hence the parallel translate—is determined.
A natural way of uniquely determining such a vector a(a) would be to
require that it lie in a given two-dimensional subspace of T(S(M), (m, v))
that is mapped isomorphically onto T(M, m) by cItt. Such a subspace will be
complementary to the vertical space
d7r~\0) =[te T(S(M), (m, vj); dn(t) = 0],
177
Definition. A connection on S(M) is a choice of a two-dimensional subspace

(m, v) of T(S(M), (m, v)) at each point (m, v) e S(M) such that the
following hold.
(1) T(S(M), (m, v)) = v) © that is, the subspace ^(m, v)

is complementary to the vertical space at (m, v).
(2) dg{3^{m, n)) = gv) for each g e S'1.
(3) The choice of Stf is smooth; that is, for each point (m, v) e S(M), there
exists an open set [/ about (m, n) and smooth vector fields X and Y
defined on U such that {X, 7} spans XY at each point of U.
Remark. There exists a smooth vector field V on S(M) such that V spans
the vertical space <7tt-1(0) at each point of S(M). It is constructed as follows.
Let d/86 denote the usual unit tangent vector field on S1. Then 8/89 is
invariant under the action of g e S1 on S1; that is,
for each h e S1.
For (m, v) e S(M), consider the smooth map A1: S1 ->■ S(M) defined by
g) = g(m, v) = (m, gv).
Define (see Figure 7.3)
V(m, v) = dAx
Figure 7.3
178
In terms of a local coordinate neighborhood U of m in M and of the

corresponding direct sum representation
T(S(M), (m, v)) = T(M, m) 0 T(S\ g) ((m, v) e n~\U)),
where g is such that v = gex, the vector field V is given by
In particular, note that V is smooth and never zero, that drr( V) = 0, and that
dh{V) = V for each h e S1.
Definition. Let 3#* be a connection on S(M). The 1 -form of or the connec¬

tion \-form, is the 1-form g> on S(M) defined as follows. Let
q: T(S(M), (m, i?)) = 3^{m, v) 0 dn~1(0) —> d'n-~1(0)
be the projection map. For t e T(S(M), (m, v)), set cp(t) = A, where A is
the real number such that q(t) = A V(m, v).
Local description of <p. Let X and Y be smooth vector fields defined in an

open set U of S(M) such that {X(m, v), Y(m, v)} spans <50%m, v) for each
(m, v) e U. Then {V(m, v), X(m, v), Y{m, i;)} is a basis for (m, v))
at each (m, v) e U. Let {<Pi(m, y), go2(m, v), go3(m, v)} be the dual basis for
T*(S(M), (m, v)). Then <pu <p2, g?3 are smooth 1-forms on U, and g> = cpx.
In particular,
(1) g? is smooth, since <px is smooth.

(2) g?( V) ^ 1.
(3) g*)(0 = ° dg(t)
= g>(A dg(V) + dg(tf))
— g?(AK) (since dg(3Y) c:
= A
= <p(0-
Lemma. Suppose <p is any smooth \-form on S(M) such that ifj(V) = 1 and
g*f = ip. Then 3^ = <A_1(0) is a connection on S(M) with the property that
its connection l-form is tp.
Proof. For each (m,v)eS(M), <p(m, v): T(S(M), (m, v)) -> R is a linear
functional. Since dim T(S(M), (m, v)) = 3, 0_1(0) has dimension 2. V^ <A_1(0),
so <A_1(0) is a complement to the vertical space. dg(tp~1(0)) = j/» _ 1(0) because
g* <A = D
179
Remark. Let U be a coordinate neighborhood in M. We now exhibit a

connection on v— S(U) = U x S1. Recall that, given coordinates
(xi, x2) in U, a smooth map c: U-» is defined by
c{m) = (m, (e/gx1)/||0/Sx1||).
For me M, let
^(c(m)) = <fc(7X*7, m)).

Then «3^(c(m)) is complementary to the vertical. For
dir^cim))) = dir ° m)) = o c)(T(C7, m)) = T(i7, m)
so that «#j(c(m)) is two-dimensional and dv|jpl(c(m)) is an isomorphism.
Furthermore, F ^ since dîV) = 0.
Now set ^j(gc(m)) = <:/g(^(c(m)).
■u
Figure 7.4
In terms of the product representation 7r_1(f/) = U x S'1 given by c,
m, u) is just the tangent space at (m, t>) to the submanifold U x {r>} (Figure
7.4). More precisely, letting B: U x S'1 7r-1(t/) be the isomorphism de¬
fined by B(m, g) = gc(m) = (m, ge^m)),
v) = dB(T(U x {g}, (m, g))),
where g e S1 is such that gex(m) = v. The 1-form 9^ of this connection is
<Pi = (B-')*(d9),
wherep: U x S'1 -* S'1 is projection, dd is the 1-form on S1 dual to d/86, and
de = P*(dd).
Note that dcpr — 0 for this special connection, for
dcPl = </[(£-x)* o p*(d0)] = d[(p o I?_1)*(c/0)]
= (p o B~1)*(d(d9)) = 0.
180
Warning. d{d9) = 0, not because dd is the differential of a 0-form (it is not),

but because there are no nonzero 2-forms on S1.
Our definition of a connection was motivated by a desire to construct a
notion of parallel translation. We now prove that given a connection on
S(M), parallel translation is indeed defined.
Theorem. Let be a connection on S(M) with \-form <p. Let a: [a,b]-> M

be a broken C00 curve in M. Let v e T{M, a (a)) with ||y|| = 1. Then there
exists a unique broken C“ curve a: [a, b] -> S(M), called the horizontal lift
of a, through (a(a), v), such that
(1) 77 o oc = a.
(2) a(t) e J^(a(t)); that is, <p(a(t)) = 0 for all t e [a, b\.
(3) a(a) = (a(a), v).
The vector a(b) e T{M, a{b)) is the parallel translate of v along a to a(b).
The proof of this theorem requires two preliminary lemmas.
Lemma 1. Let ^ and 2T2 be two connections on S(M) with connection \-forms
9>! and <p2. Then
(1) (<P2 - <Pi)(V) = 0.

(2) g*(cp2 - cpf = cp2 - 9! (for all g e S1).
(3) cp2 — <px = 77*(r) {for some smooth \-form r on M).
Proof. (1) and (2) are clear. We shall show that (1) and (2) imply (3). If is
any smooth 1-form on S(M) with = 0 and g*for all g e S1, then
ifj = 77*(t) for some r. To define ronpe T{M, m), let {m, vx) e 77_1(m), and
let w e T{S{M), (m, vf) be such that dn{w) = v. Set t(v) = ip(w). t(v) is
independent of the w chosen in d-n-1(h) since dTr{wft) = v implies that
^(vtq — w) = 0, so that nq — w = AFfor some A. Thus
i/.(wi) = 4>{w + XV) = ip(w) + Aip(V) = >p(w).
Also, t(v) is independent of the point {m, vf) chosen in 77_1(m) because if
(m, v2) e 77_1(m),
then v2 — gvx for some g e S1. Moreover, if w e T(S(M), (m, vf) satisfies
dn(w) = v, then dg{w) e T{S{M), (m, v2)) satisfies dn(dg(w)) = v, and
0|(.n,v2)(dg{w)) = i/j\g(m,vo(dg(w))
= = *P\
t is smooth because in a coordinate neighborhood U, r{v) = i/> ° dc{v), where
c: U->tt-\U)
is defined by c(m) = (m, ^(m)). □
Lemma 2. Let a: [a, b] -> M be a smooth curve in M. Let a: [a, b] -» S'(M)

and ft: [a, b] —> S'(M) smooth curves such that tt o a = a am/ rr ° ft = a
181
(see Figure 7.5). Suppose a is horizontal relative to some connection JF on

S(M) with connection \-form <p; that is, suppose <p(a(t)) = 0. Then there
exists a smooth function 9: [a, 6] —> R such that
(1) jS(t) = eima(t) (t e [a, 6]) and

(2) ,*0(0) = (dd/dt)(t) (t e [a, b]).
Furthermore, if &(a) = 0(a), then 9 can be chosen such that 9(a) = 0.
Proof. Let g: [a, b] S1 be defined by
0(0 = g(0«(0 (f e [a, 6]).
It is easy to verify that g is a smooth curve. Since R is a covering space of S1,

and [a, b] is simply connected, there exists a lift 9: [a, b\ -> 7? of g (see Figure
7.6). Furthermore, if a(a) = fi(a), then g(a) — 1, and there exists a unique
such lift with 9(a) — 0.
Figure 7.6
182
Since p is smooth and has a smooth inverse locally, 9 is smooth. Further¬

more,
Rt) = g(t)«(0 = P ° 6(t)a(t) = eima(t)

so (1) is satisfied.
To verify (2), first note that the tangent vector to the curve
g: [a, 6]->5* [g(0 = <?«««]
is given by
iiO = ip ° 0)(t)
Restricting attention to a coordinate neighborhood U <= M and the corre¬

sponding product representation ^ U x S1,
5(0 = h(t)c(cc(t)) = (cc(t), h(t))
for some h{t) = emt) e S1, and
kt) = («(/), g(t)h(t)) = («(*), ei(0(O+*(t»).

The tangent vector to 5 at 5(t) is then (5(7), (dip/dt)(d/8d)), whereas the tangent
vector to ft at kO is («(/), [(dd/dt) + (d>p/dt)](d/d9)); that is,
kO = d(gmkO) + (o. § 4) = (0X«(0) +

where d(g(t)) is the differential of the map git): S(M) —> S{M). Since a(t)
is horizontal, and dig{t))i3^) c ^
.(to) = = f• a
Proof of the Theorem. Note that it suffices to prove the theorem for a
smooth curve a. For then we can uniquely lift each smooth portion of any
broken curve.
Local existence. Let U be a coordinate neighborhood in M. We shall
show the existence of unique horizontal lifts in 7r_1(t/) — S(U). Let c: £/ ->
S(U) be as usual: c(m) — (m, ex{m)) for me U. We shall first show that if 3^
is the special connection ^ on 7t-1((7), constructed via the product structure,
then a has a unique horizontal lift 5X such that aâ) = c(a(a)). Indeed, let
5X: [a, b] —> 77-_1(t/) be defined by 5X = c ° a. Then 77°51 = 77-°c°a = a
and ai(t) = dc{a{t)) e <?f[(c(t)), so 5X is a horizontal lift. Moreover, a± is the
unique J^-horizontal lift such that
«i id) = c(a{a)).
For if a2 were another such lift, then, by Lemma 2,
52(t) = e'°«\it)
for some smooth function 9 with 0(a) = 0; and <Pi(a2(0) = d9/dt, where <px
is the connection 1-form of Now 52 is ^-horizontal if and only if
183
<Pi(«2(0) = 0; that is, dd/dt = 0. Hence 0(0 must be constant. Since 6(a) = 0,
6(t) = 0; that is, d2(t) = ax(t) for all t; that is, «2 = dx.
Thus a admits a unique ^-horizontal lift ax with dx(a) = c(a(a)). Now
consider our original connection with connection 1-form <p. Then, by
Lemma 1,
<px — (p = 77*T
for some smooth 1-form t on [/. Let & be any curve in n~1(U) such that
■n o a — a. Then, by Lemma 2, 5(0 = eJ0(t)al(t) and 93i(d(0) = dd/dt. Thus a
is an Jf-horizontal lift of a if and only if cp(d(t)) = 0; that is, if and only if
(<Pi - 9>)(«(0) = 9>i(«(0) =
But on the other hand,

(<Pi - <?>)(«(0) = (77*T)(«(0) = T(dira(t)) = r(a(t)).
Thus a is ^-horizontal if and only if dd/dt = r(d(r)); that is, d = t(«(0) dt
+ for some constant d0. Hence each ^-horizontal lift a of a is of the form
«(0 = g(t)(c ° «(0)>
where
g(t) — eieoel
For each unit vector v in T(U, a(a)), there is precisely one d0 with 0 < d0 < 2-u
and (a(a), v) = eie°(a.(a), ex). The above formula, with this value of d0, then
gives the unique ^-horizontal lift a with d(a) = (a(a), v).
Global existence. To establish global existence, let a: [a, b] -> M and let
t0 = sup [t e [a, b]; a|[aj6} has a (unique) lift a].
We shall show that t0 = b. Suppose t0 + b. Then consider the restriction of
a to the interval [/0 — e, t0 + e]. By local existence, this has a unique lift 5
for some sufficiently small e > 0, say with a(t0) = (a(L>X w) e S(M). Then
a(t0 — e) = ga(t0 — e) for some g e S1, and ga is a horizontal lift with
ga(t0 — e) = a(t0 — e). By uniqueness, ga = d on the interval [t0 — e, t0).
Hence ga extends a beyond t0, contradicting the definition of t0. □
Remark. Note that, relative to the special connection on tt~ \U),

parallel translation is independent of the curve. In fact, the vector field
ex = (d/dx^/Wd/dxiW is parallel along every curve in U.
7.2 Structural equations and curvature

Definition. Consider the circle bundle S(M) of a smooth oriented Riemannian
2-manifold M. Two smooth 1-forms and w2 are defined on S(M) as
follows. For t e T(S(M), (m, v)),
"i(0 = <drr(t), o>,

OJ2(0 = (d-n(t), ivy.
184
7.2: Structural equations and curvature
where iv — e'nl2v is the image of v under rotation through an angle of rr/2

in T(M, m). (We shall show below that these 1-forms are indeed smooth.)
Remark 1. Thus o>x(t) and uj2(t) are the components of d-n(t) relative to
the orthonormal basis {v, iv} for T(M, m); that is,
d-n{t) = wx(t)v + io2(t)(iv).
Remark 2. Suppose is a connection on S(M) with connection 1-form <p.

Then {cp, to1, co2} is a basis for T*(S(M), (m, v)) for each (m, v) e S(M).
Proof. Since dim T*(S(M), (nr, v)) = 3, it suffices to show that there exists
no nonzero t e T(S(M), (m, v)) that is simultaneously annihilated by these
three forms—for then these forms are linearly independent. But if w1(t) =
o»a(0 = 0, then = 0, so tis vertical; that is, / = XVfor some A. Further¬
more, if <p(t) = 0, then A = <p(AF) = <p(t) = 0, so t = 0. □
Remark 3. Let g = eie e S1. Then
g*a>1 = (COS + (sin 9)a>2,

g*a>2 = —(sin 6)co1 + (cos 6)a>2.
Proof, gv = (cos 9)v + (sin 6)(iv). Hence, for t e T(S(M), (m, v)),
£*a,l|(m,e)(0 = W11 (m,gu)W§'(0)
= <c/t7 o dg(t), gV>

= <M0, gV>
= (dir(t), (cos 9)v + (sin 9)iv}
= (cos 9)cu1(t) + (sin d)a>2(t).
Similarly,
g*cD2(t) = —(sin 9)qji(t) + (cos 9)oj2(t). □
Remark 4. g*(<nx A a>2) = wj a oj2 for all g = ew e S1. For,
g*(co1 A co2) = g*CD1 A g*co2 = (cos2 9 + sin2 9)u)x A C02 = u)x A C02.
Furthermore, wx A co2(t1; t2) = 0 if either tx or t2 is vertical. Hence, as in the

proof that <p — cpx = it*t for some r (Section 7.1), the 2-form A co2 is
the image under 77* of a (unique) form on M.
Definition. The volume element of a smooth oriented Riemannian 2-manifold

M is the smooth 2-form, vol, on M such that
7T*(VOl) = 0)! A OI2;
that is, for vx, v2 e T(M, m), vol (vx, v2) - A <^>2|(m,u)(î, *4) f°r anY
(m, y) e v~1(m) <= 5(M) and i?i, i4 g T(S(M), (m, v))
such that dTr(v'i) — vt (i = 1, 2).
185
Remark 5. Suppose U is a coordinate neighborhood in M with coordinate

functions (jcl5 jc2). Let ex = (d/dxx) || 8/dxx || and let a)'x, a>2 be the smooth
1-forms on U that at each me U form the basis for T*(M, m) dual to
{ex{m), iex(m)}. Let
c: U->tt-\U) c S(M)
be given by c(m) = (m, ex{mj). Then, for v e T(M, m),
(c*wx)(v) = u>x(dc(v)) -- <dir o dc(v), ex} = <v, ex) - co'x(v),

so u)i — c*wx. Similarly, a>2 — c*a>2. In particular,
a)'x A a>2 = c*0)1 A C*oj2 = C*(oj1 A a>2) = C* ° 77*(vol) = (n ° c)*(vol);
so, since n ° c = iu,
Vol|y = oj'x A o/2.
Now let a>i = TT*a)'t (i = 1,2). Then wj and d>2 are smooth 1-forms on
tt~\U) ^ S(M) and
<jjx A d>2 = TT*0>\ A 77*CU2 = TT*(a>'x A Cl>2) = 7T*(V0l) = A 0>2.
Moreover, at each point (m, ex(m)) of c(t/), a>4 = dij. For, if
teT(S(M),(m, ej),
then
wi(Oei + w2(O0’î) =
= w’x(d'TT(t))e1 + oj’2(dTr{t ))(iex)
= 6>i(t)ex + a>2(t)(iex).
Note further that, for g — ew e S1,
g*&i — g* o TT*(X)\ = (jt o g)*COj = = d>i.
Thus, from Remark 3 above,
(S*wi)|<m,ei) = (cos 0)cux + (sin 0)co2|(m>ei)

= (cos e)<*i + (sin 0)w2|(m>8l)
= (cos 6)g*d>x + (sin d)g*d>2|(m>ei).
Applying (g-1)*, the forms wx and dij at (m, gex) are related by
cdx = (cos 6)d>x + (sin 0)d>2.
Similarly,
w2 = —(sin d)d>x + (cos &)d)2.
In particular, the above formulae show that a>x and co2 are smooth.
Remark 6. For higher dimensional Riemannian manifolds, the volume
element is obtained similarly. If U is a coordinate neighborhood in the
oriented Riemannian manifold M, with coordinate functions (jcl5..., x„)
186
such that dxx A • • • A dxn gives the orientation of U, consider the vector
fields 8/8xx, ..., 8/8xn. Using the Gram-Schmidt orthogonalization process,
we obtain smooth vector fields ex,..., en on U which form an orthonormal
basis for the tangent space at each point. Let a/1;..., u'n be the dual 1-forms.
Then the n-form vol| v = o>i A • • • A wj, is independent of the (oriented)
coordinate system on U and thus defines a global nonzero «-form vol.
Given an oriented Riemannian 2-manifold M and a connection on S{M)
with connection 1-form <p, the 1-forms cp, wx, o»2 form a basis for the cotangent
space at each point of S(M). Hence the 2-forms oq a o>2, a>x a cp, o>2 a <p
form a basis for the 2-forms at each point of S(M). Hence dcp, dcox, dw2 can
be expressed in terms of this basis. The resulting formulae are called the
Cartan structural equations. We now derive them, beginning with the second
structural equation.
Second structural equation. On for a coordinate neighborhood U,
let <px denote the connection 1-form of the special connection JFX. Then
dcpx — 0 so that
dcp = d<p — dcpx = d(cp — cpx) = d(jr*T) = 7T*(dr)
for some smooth 1-form r on U. Now dr is a 2-form on U, hence is a multiple

of the volume element; that is, dr = - AT vol for some smooth function K on
U. Thus
dcp = 77*(— K VOl) = 77*(-/077*(VOl)
or
dcp = — (K ° 77)aq A a>2.
The smooth function K is independent of the coordinates used, since it is

determined by this last formula. Thus K is a smooth function on M, called
the curvature of the connection cp.
First structural equation. On 7r_1([7), for a coordinate neighborhood U,
we have seen that at ei8c(m),
oq = (cos 9)d>1 + (sin 0)cv2,

a>2 = —(sin 9)d)1 + (cos 9)a>2.
Now
dcl)i - J(77*w;) = TT*(du)'^ - 77*(fli VOl) = (flj ° Tt)^ A W2
for some smooth function at on U. Thus setting dx = at° n,
da>1 = —(sin 9) d9 a + (cos 9)alcx>1 A w2
+ cos 9 d9 A 2 + (sin 9)a2cu1 A co2

= d9 A a>2 + (ax cos 9 + a2 sin 9)a>1 A a>2.
If is the special connection ^ on tr_1(C7), then <px = d9, thus for this
special connection,
(*) da>i = cp± A co2 + bxcx>x A cx)2
187
for some smooth function b± on tt~1(U). Similarly,

doj2 = — <pi A coi + b2u>1 A o>2.
For an arbitrary connection form cp, cpx — 95 = v*t for some smooth 1-form
r = cxoji + c2oj'2 on U. Hence
cpi — cp = 7r*(C!wi + C2a>2)
— (C1 0 77)<^l + (C2 0 7r)u»2
= flwl + f2u>2
for some smooth functions fx,f2 on 7t_1({7), since oh, o>2 span the same
space at each point as a>l5 a>2. Thus
<Pi = i = (p A oj2 + fico1 A oj2 + b1a>1 A <u2
== i A o>2.
This, together with the corresponding equation for du)2, gives the first
structural equations as follows:
da>1 — cp A co2 + h1a>1 A o)2,

da>2 — —(p A + h2a>1 A ta2,
where /j1} h2 are smooth functions on S(M). Note that although these
equations were derived over a coordinate neighborhood, they are independent
of coordinates. Thus they are valid globally.
Although one might expect that by choosing an appropriate connection { relative to the basis {93 A o^, 2, a*! a co2}
could be prescribed fairly arbitrarily, this is not the case. In fact, dw1 never
has a component in the 2 in the
<p a w2 and <p A co1 directions, respectively, must always be +1 and —1.
It is natural to ask whether the first structural equations can be made
simpler by an appropriate choice of connection on S(M). In particular, can
cp be chosen such that hx = 0 and h2 = 0? The answer is yes, and the choice
is unique.
Theorem. Let M be an oriented Riemannian 2-manifold. Then there exists a

unique connection 1p on S(M) such that
da>1 = ifl A o>2,

du>2 = — ip A oq.
This connection is called the Riemannian connection.
188
Proof. Let is any other connection on
S(M), then, as above,
2
for some x1 and x2. Solving for cp and substituting in the first structural
equations for cp, we obtain
du>i = i/j A oj2 + (hx + Xi)aq A u>2,

d(o2 — —if) A (l>i + (h2 + X2)cox A co2.
Thus
da>1 = if) A to2,

da>2 = —if) A oq
if and only if x1 = —hx, x2 = —h2. This gives both existence and unique¬
ness. □
The Cartan structural equations have a dual formulation in terms of

vector fields. Let V, Eu E2 be the smooth vector fields on S(M) that form
the dual basis to 9, oq, cu2. Then E± and E2 are horizontal at each point
since <p(£'i) = <p(E2) = 0. Moreover,
diriE^m, v)) = oq(Ex)v + o)2(E1)(iv) = v,
so E^m, v) is the unique horizontal vector at (m, v) whose image under d-n is
v. Similarly, dTr(E2(m, v)) = iv. The structural equations then become
[K Ex\ = e2,
[V,E2] =-Elt
[Eu E2\ = (Ko tt)V - hxEx - h2E2.
If cp = if), the 1-form of the Riemannian connection, the last boxed

equation reduces to
[Ei,E2] = (K on)V.
To verify these equations, apply the formula
dr(V1, V2) - \{Vxr(V2) - V2t(Vx) - r([Vx, V2])}

nine times, as r runs through the set {cp, a>x, o2}, and Vx, V2 runs through
the set {V, Ex, E2}.
Remark. If K is constant, these formulae show that {V, Ex, E2j spans a
finite-dimensional Lie algebra.
From now on, for an oriented Riemannian 2-manifold M, let the connec¬
tion chosen be the Riemannian connection, and let K be the curvature
function for that connection.
189
7.3 Interpretation of curvature

We now show that the curvature K of M measures the amount of rotation
obtained in parallel translating vectors around small closed curves in M. The
intuitive reason is this. On S(M) we have the vector fields E±, E2, and V, and
we know that for the Riemannian connection,
[Ex,E2\ = (Kott)V.
But [Eu E2](m, v) is just the tangent vector to the curve through (m, v)
obtained by following the integral curves of E1 and E2 forward and then
backward through parameter distances of Vs (Figure 7.7; see Section 5.3).
Projecting this figure down to M, we obtain a rectangular-shaped figure

which is “nearly” closed; that is, the curve obtained through m has zero
tangent vector at m because it is the projection of [Eu E2](m, v), which is
vertical (see Figure 7.8).
Now the integral curves in S(M) are the horizontal lifts of the curves in
M; that is, these curves are obtained by parallel translating v around the
curves in M. The endpoints of the curve through (m, v)—dotted in Figure
7.9—essentially differ by an element of S1, namely the rotation g = eie, which
sends v into its parallel translate around the rectangle in M.
Since the area of the rectangle in M is approximately Vs-Vs = s, the
190
7.3: Interpretation of curvature
(m,gv) -^
(m,v)
tt 1(m)
Figure 7.9
limit as s 0 of the angle of rotation 0 divided by the area of the rectangle

is equal to the coefficient of V, namely K{m).
Stated precisely and in somewhat greater generality, the result we have
been discussing is as follows.
Theorem 1. Let M be an oriented Riemannian 2-manifold. Let <s> be an

oriented 2-simplex in R2, and let h: [s] M be a map which has a smooth
extension mapping a neighborhood of [s] into M. Let a be the closed broken
C “ curve in M obtained by restricting h to b(s'). Then the rotation obtained
by parallel translation around the closed curve a is
ei ;<a> h'(K vol)
so that the angle of rotation is J<s> h*(K vol).
Remark. Note that this result contains the result discussed above. To
obtain K{m), take the limit of J<s> h*(K vol)/J<s> h*(vol) as <s> shrinks to
zero and A«j)) shrinks to m. However, the theorem says more. For example,
it is possible to have K > 0 on /?([s]) and still get a trivial rotation upon
parallel translating around a, namely when the total angle of rotation
J<s> h*(K vol) is an integer multiple of 2tt.
Proof of Theorem 1. Let <s> = fv0, vu vf) for some vertices v0, vu v2, and
let vv0 e T(M, h(v0)) be a unit vector. The lines in [s] through vx cover [5];
their images under h are curves in M which cover /j([s]). Let h\ [5] —> S(M)
be obtained by mapping each of these curves into its horizontal lift in S(M)
through
(/*0h), wx) e S(M),
191
0 Oi),wi)
where w1 is the parallel translate of w0 along the curve a|<1)0jVl> to h{v^)

(Figure 7.10).
By construction, tt o h = h. Moreover, h has a smooth extension mapping
a neighborhood of [5] into S(M). This may be checked via local coordinates;
we omit the computation.
Now
I /z*(/fvol)=f (770 £)*(/: vol)

J <s> J <s>
= f £*[^*(^vol)]
J <s>
— j h*[(K ° v)o)1 A o»2]

J <s>
= — h*(d<p) (second structural equation)
J <s>
= -fJ <s>
d(h\)
= — h*w (Stokes’s theorem)

J 8 <s>
- -LAS{*))*
192
where = h\e<s>. Let a denote the horizontal lift of a through h(v0) =

(°c(v0), w0). Since ft\<VOfVl> = and ft\<Vl,v2> = ^|<»lf»a> are horizontal
(h(vi),wt)
(h(v0),w o)
Figure 7.11
by construction of h, we have /?|<„0tt)1> - a|<t)0>t)1> and fi\<VltV2> = a|<„1>V2>.

By Lemma 2, Section 7.1, there exists a function f:(v2,v0}-+R with
f(v2) = 0 such that
j?(0 = e</(t)a(t).
But a is the horizontal lift of a, so that a(v0) is the parallel translate of w0
around a. On the other hand, /?(v0) = vv0. Hence eif(vo) is just the rotation
mapping the parallel translate of w0 around a into vv0; that is, e~if(vo) rotates
vv0 into its parallel translate around a. By the second statement of Lemma 2,
Section 7.1,
for t e <v2, v0}. Moreover, <p{d$(djdt)) = 0 on <v0, vx) and (vx, v2) since £ is
horizontal there. Thus since d<s> = (v0, vx)> + <pl5 v2) + <v2, t>0>,
= ~f(v 0)
= the angle of rotation from w0 to its parallel
translate around a. □
193
Definitions. Let a: [a, b]-+ M be a smooth curve. The length 1(a) of a is the
real number
/(«) = KOIdL
The arc length along a is the function s: [a, b] -> R given by
s(t) = || «(■>■) 1 dr.
Remark. I and s are defined because t ||a(t)|| is continuous. Note that

the function s is of class C1, but it is not necessarily smooth because
||d(t) || is not necessarily differentiable where a(t) = 0. If, however, ||«(r) || # 0
for all t, then s is smooth and monotonically increasing.
Definition. A curve «: [a, b] -> M is said to be parameterized by arc length if

||a(t)|| = 1 for all t e [a, b]. In this case, s(t) = t — a for all t e [a, b].
Remark. Given any curve a: [a, b] M with ||a(0|| ^ 0 for all t e [a, b],
a new curve a1: [0, /(a)] -> M, parameterized by arc length, is obtained by
setting
ax = a ° S _1.
Then Im a! = Im a, and l(a±) = 1(a).
Remark. The concept of arc length extends to broken C00 curves a since
||«(/) || is defined at all but a finite number of points.
Definition. Given a smooth curve a: [a, b] —M parameterized by arc length,

a smooth curve a : [a, b] —> S(M) is defined by
a'(t) = (a(t),d(t)) (te[M]).
a is said to be a geodesic in M if a is horizontal; that is, if a is the hori¬

zontal lift of a through (a(a), a(a)) e S(M). Note that if a is a geodesic,
the parallel translate of a(0) along a to a(t) is just a(r); that is, the tangent
to a translates into itself, and a is a “straight line” of the surface.
To measure how far a curve a is from being “straight,” we measure how

far a is from being horizontal. Suppose, then, a is parameterized by arc
length so that a : [a, b] -> S(M) is a curve in S(M).
Definition. The geodesic curvature ka(t) of a at t e [a, b] is ifj(da'(dldt)), where

i/> is the 1-form of the Riemannian connection.
Notation. If a: [a, b] M is a broken C00 curve with a(t) 0 for all

t e [a, b], let r(«) = j'0(a) kai(t) dt, where a! is the new curve obtained from a
by parameterizing by arc length.
If M is a smoothly triangulated manifold, then t can be considered as a
1-cochain (relative to the triangulation).
194
Lemma (the Gauss-Bonnet theorem for 2-simplices). Let M be an oriented

Riemannian 2-manifold. Let <s> be an oriented 2-simplex in R2, and let
h: [s\ M be a map which has a smooth nonsingular extension mapping a
neighborhood of [5] into M. Let a be the closed broken C°° curve in M
obtained by restricting h to d(s}. Then
J< /z*AT(vol) = — r(a) + ^ (interior angles ofh[s]) — v.
Proof. From Theorem 1 above, el^> ^vod is the rotation obtained by

parallel translation around the closed curve a. Suppose a is broken up into
its three smooth curves a0, «l5 and «2 so that r(«) = 2 r(ai) and at: [au ai + 1]
—> M with a0 = a and a3 = b. By Lemma 2, Section 7.1, is the rotation
from the parallel translate of d,(a,) to dj(ai+1). Hence, from the picture in M
(Figure 7.12), we get that parallel translation around the closed curve a is
given by e,( ~T(a)~ 2 exterior angles). Hence, by taking logarithms, we get
h*(K vol) = — t(oc) — y (exterior angles) + 2tt7,

J <s>
where / is an integer.
We use a continuity argument to show / = 1. Suppose p0 is a flat Rieman¬

nian metric in a neighborhood of A[s] (say transferred from R2 via h). Then
K = 0, t(cc) = 0, and f (exterior angles) is 2n. Hence for the flat Riemannian
metric, / = 1. Suppose p is our given Riemannian metric, and let pt = tp0 +
(1 — t)pbe a family of metrics, t e [0, 1]. Let Kt, rt(a), exterior angles( be the
usual entities for pt. These are continuous functions of t. Hence / is a con¬
tinuous function of t. Since it is an integer for all t and equal to 1 for t = 0,
we obtain l — 1. (You can also obtain this result by checking that / = 1 for
small triangles and taking barycentric subdivisions.)
Since interior angle + exterior angle = v, the lemma is proved. □
Definition. Let M be an oriented, connected, smoothly triangulated 2-mani¬

fold. For each 2-simplex 5 in M, let <5) denote this simplex oriented
consistently with M. That is, if h: K-+ M is the triangulation and a> is a
195
2-form on M giving its orientation, let the orientation of 5 be given by the

2-form h*u>. Let
c = 2<‘>-
Then c is a cycle called the fundamental cycle of M. Given any 2-form [x
on M, the integral of [x over M is defined by
Exercise. Prove that c is a cycle.
Remark. The integral can be defined without use of a triangulation. Let

M be a compact oriented n-manifold, and let /x be an «-form on M. Let
{Uj,fj} be a smooth partition of unity on M, where {£/,} is a finite covering
of M by coordinate neighborhoods. Then integration of rc-forms is defined
on each U, by pulling the forms back to Rn through the coordinate systems.
The integral of /x over M is then given by
This is independent of the partition of unity used, for if {Vk, gk} is another
such partition, then
Theorem 2 (Gauss-Bonnet theorem). Let M be an oriented, connected,

smoothly triangulated, Riemannian 2-manifold. Then
(x(M) is the Euler characteristic of M.)
Proof. Note that each 1-simplex t of M is an edge of precisely two 2-simplices

of M. For given any point m e (t), there exists, by the implicit function
theorem, a coordinate ball U about m such that (t) n U is mapped into a
straight line in R2. By choosing U small enough, t must divide U into precisely
two pieces. These pieces must lie in distinct 2-simplices, and, since open
simplices are disjoint, there can be no other 2-simplex with t as an edge.
Thus, since each 2-simplex has three 1-simplices as edges, the total
number n1 of 1-simplices of M is given by nx — 3n2/2, where n2 is the number
of 2-simplices of M. Letting n0 denote the number of vertices in M, the Euler
characteristic (Section 6.1) is given by
X = »0 - «i + n2
= n0 - (3m2/2) + n2
-no - (n2/2).
196
Now we apply the previous lemma, and
K vol
277 Jm
“sj?L**(jcvo1)
= 2“ 2 + 2 0nteri°r angles of /t[s]) — 77 j
= y- ^ —r(Sc) + ^ ^2 interi°r angles of /z|T] j — n2rrj.
But Sc = 0, and 2s (2 interior angles of h[s]) equals the sum over all vertices
v in M of the sum of the interior angles at v of all 2-simplices with v as a
vertex. Taking a coordinate neighborhood of V contained in St (c), we see
that for each v, the sum of these interior angles at v is exactly 2tt (Figure 7.13).
Hence
= ^ (277«0 — n2v)
2tt
= »o - J = x(M). □
Remark. Note that this theorem holds for any connection on S(M), since
only the second structural equation was used in the proof.
Corollary 1. Let M be any Riemannian 2-manifold homeomorphic with the

sphere S2. Then
K vol = 477.
Corollary 2. Let M be any Riemannian manifold homeomorphic with the torus

S1 x S1. Then
K\ ol = 0.
197
Corollary 3. Let M be as in the theorem. Suppose there exists on M a smooth

vector field which is never zero. Then y(A/) = 0. In particular, there exists
no nonzero vector field on S2.
Proof. Let X be such a vector field and let Y = Ar/||A'’||. Then Y is smooth,
and || Y(m) || = 1 for all me M, so Y is a smooth map M -> S(M). On S(M),
df = — A w2 = — 7r*(AT vol).
Hence
d(Y**f>) = Y*(dip) = -7*o 77*(/fvol)
= -(770 7)*(/:voi)
= — K vol,
since -n ° 7 = iM. Thus K vol is exact and
2t7X = J Kvol = —J* d(Y*cp) = -Ja Y*cp = 0. □
7.4 Geodesic coordinate systems

Let M be an oriented Riemannian 2-manifold, and consider the vector
field E1 on S(M). Let Ut denote the local 1-parameter group of transforma¬
tions on S(M) associated to the vector field E1. Then, for each m0e M and
z e ■tT~1(m0), there exists an e > 0 and an open set Wz about z in S(M) such
that the map [—e, e] x Wz->- S(M) given by (t, w) -> Ufw) is smooth. Since
rr~\m0) = S'1 is compact, 7r_1(m0) can be covered by a finite number of
such sets Wz. Taking e to be the minimum of the corresponding numbers e,
the map /x: [0, e] x S1 -> S(M), given by
p{t, g) = Ut(m0, gv0) (t?0 fixed e T(M, m0); ||f0|| = 1),
is well defined and is smooth. Since for each g e S1, the curve t -> p,(t, g) in
S(M) is horizontal with tangent vector Elf the curve t —> 7T o p(t,g) is a
geodesic starting at m0 with tangent vector dfiEfmo, gv0)) = gv0 at m0 (see
Figure 7.14).
Let De denote the open disc of radius e about the origin in R2, and let
P: [0, e) x S1 De (P for polar map) be given by
P(t, g) = tg.
Then P is a smooth map. Moreover, the restriction of P to (0, e) x S1 is

injective and maps (0, e) x S1 onto De - {0}. Since, morever, n ° /lx(0, g) — m0
for all g e S1, a map G: De M is defined byG = v o ^ o p-1,
[0, e) x S1 S(M)
Db -> M
198
7.4: Geodesic coordinate systems
Exercise. Verify that this map G is smooth and that dG(0) is an isomorphism,
so that G maps De diffeomorphically onto an open set of M for e sufficiently
small. (Note that G(0) = m0 and that G maps radial straight lines in De into
geodesics through m0; see Figure 7.15.)
Definition. The map G: De-^M, where e > 0 is sufficiently small that G

maps De diffeomorphically onto its image, is called a polar, or geodesic,
coordinate system about m0.
Remark. Let d/dr and d/dd denote the natural vector fields on (0, e) x S1.
Then dP{d/dr) and dP(d/dd) are the tangent vectors to the polar coordinate
curves in De — {0}. Note that dP(d/dd) can be smoothly extended to De by
setting dP(d/dd)\0 = 0. The following lemma asserts that these orthogonal
vector fields dP(d/dr) and dP(d/dd) are mapped by geodesic coordinate systems
into orthogonal vector fields; that is, that the orthogonal curves r — constant
199
and 0 = constant are mapped into orthogonal curves in M by geodesic

coordinate systems.
Gauss’s Lemma. <dG ° dP(d/dr), dG ° dP(d/86)y = 0.
Proof. dG ° dP = d(G ° P) = d(n ° p) = dn ° dp. Moreover, d^d/dr) = Elt

hence for (m, v) e Image p,
d-n dJl) — drrÊ^m, v)) = V.

L \8r! (m,v)
Thus
<{da°dp{^)’da°dp{re)y=° ^
To see that this is zero, consider

= p*{du>^) = A W2) (by first structural equation)
- A |U,*o>2.
Then
But dfj.(d/dr) = Eu so that <^(d^d/dr)) = 0 since is horizontal, and

<jj2{dii{djdr)) = 0. Hence
0 = ^<v,(|4)
_0 _0
= 1/1 '
2 \0r 0? 00 }•
But
«,i(£i) = 1,
so the second term is zero. Moreover, the bracket [(0/0r), (0/00)] = 0 by

equality of mixed partial derivatives, so the third term is zero also. Thus
0_
= 0,
dr
200
and (n*oj1)(d/89) is independent of r; that is,
"•‘"’(I>)>
is independent of r. But as r —^0, dP(8/89)-> 0, so that this inner product->0.

Since it is independent of r, it must therefore be identically zero. □
Remark. By Gauss’s lemma, dG(dP(8/89)) is always orthogonal to the

radial geodesics of our geodesic coordinate system.
Next we study the behavior of the length of dG(dP(8/89)) as we move
along a radial geodesic. Since w1(d/j.(8/89)) and a>2(d/u(8/89)) are the com¬
ponents of dir(diJ.(d/89)) relative to an orthonormal basis, and since
to i(d 11.(8/89)) = (M*o,1)(8/89) = 0
from the proof of Gauss’s lemma, (fj.*a)2)(8/89) = a>2(dii(8/89)) is, at least

up to sign, equal to the length of d-n ° d^(8/89) — dG ° dP(8/89).
Now from the first structural equation,
d(jj.*a)2) = H-*(.da>2) = —A Wj) — — A
Thus
d{dr, b() |
But (ix*a>1)(8/89) = 0 by Gauss’s lemma, and
so that
On the other hand,
d(n*oj2)
= i/l
2 \dr
0x*W2) G**<*>a) - -Ill
dr' dd\I)
18_
(/x*cu2)
~ 2 8r
since
(M*w2)^lj = w2 = "2(^1) = 0
201
and [(d/dr), (d/dff)] = 0. Thus
a r / a \i „ * d\
dr = ^(sej-
Now applying the second structural equation,
G?(ju*i/i) = H*(dif) = fj.*(—(K 077)0)! A tu2) = A ^*o>2.
Thus,
\0r’ 00/
1
= (K ° IT ° fl)
1
= -^(KoGoP)(h*W2)
(»«)’
as before. Finally,
d_
= i/i
2 \dr ^(A) 86 y»(A)] - A])
- 1JL
2 dr ^(A)
since
_0 _0'
0**«(|:) = = «£i) = 0 and = 0.
dr’ 86
Thus
a r / a \l / a \.
= -(KoGoP)(^2)If
dr L
<^y
\ / J yC/(7/
Differentiating the first boxed equation above and substituting into the
second, we obtain
0“*w2) + (K0G0P) (P*"2) = 0.

8r2 (*) (A):
Moreover,
o>*(V) = 0,
= “’HA)) r=0
=
202
and, from the first boxed equation above,
r=0
0 (*(£)) r=0
KV) = i.
In particular, (/a*o>2)(5/d0) > 0 (at least for small values of r), and so
(n*w2)(d/dd) is indeed equal to the length of dG ° dP(d/d9).
Thus we have shown the following theorem.
Theorem 1. Let G: De —> M be a geodesic coordinate system about m0. For

g e S1, consider the geodesic r -> G ° P(r, g) through m0. Let f(r) denote
the length of the vector field dG ° dP(d/S9) along this geodesic. Then f
satisfies the differential equation
d2f
■jp; + (K o Go P)f = 0 (Jacobi's equation)
as well as the initial conditions /(0) = 0 and/'(0) = 1.
Remark. Note that if K is constant, this differential equation can be solved

explicitly. Namely, if K > 0 then f(r) = (1 /VK) sin (VKr). If K = 0, then
f(r) = r; and if K < 0, then/(r) = (l/V~K) sinh (V — Kr).
Theorem 2. Geodesics minimize arc length locally; that is, if a is a geodesic in
M starting at m0, then there exists an e > 0 such that /(a|[0>e]) < l(J3) for
all broken C ” curves ft in M from m0 to a(e).
Proof. Let G : DE M be a geodesic coordinate system about m0 with

e < 1(a). Then since a is a geodesic,
a(t) = G O P(t, go) = G(tg0) (0 < t < e)
for some g0 e S1. Let p’:DE — {0} -> De be the map which rotates each point
of De onto the radial line through g0; that is, p' is defined by p'(rg) = rg0.
Figure 7.16
Let U = G(De), and let p: U - {m0} -> U be the corresponding map in U;

that is, p = G°p'oG~1 (Figure 7.16). Then dp: T(U, m) -> T(U, p(m))
decreases lengths; that is,
||dp(L)|| ^ ||t?|| for all v e T(U, m), m e U — {m0}.
203
For by Gauss’s lemma, dG ° dP(8/8r) and dG ° dP(8/8d) are orthogonal

vectors in T(U, m). Moreover,
= =
"•"(I)
= = ||<«?(0)I = 0.
It follows that the magnitude of dp(v) equals the magnitude of the orthogonal
projection of v onto dG ° dP(8j8r), which is less than or equal to ||u||.
Figure 7.17
Now let ft [a, ft —s- M be any broken C ” curve from m0 to a(e). Let
c = inf [/ g [a, ft; p(r) <£ U],
Then p(c) $ U, but p(t) e U for all t < c. Let ft = j8|[a>cj (Figure 7.17). Then
l(Pi) < l(P). Now consider the curve p ° ft: [a, ft M. Then
\\p°Pi(t)\\ = \\dp(pim < 1^(011

since dp decreases lengths, hence
kp°Pi) < KPi) < m.

But Image p ° fix = Image a, so, since a is one-to-one, we have p ° Pi(t) =
a of(t) for t g [a, c) where/ = a'1 op o ft. Thus
/(/3) > l(p o ft) - £ ||p o ft(0|| dt
= Ja II““/(Oil dt
=£ K/tO)ll/'(OI *
a J' IK/M)II/'W * = £ IKt)|| dr
= ^(a|[0.s])- □
204
Remark. The proof above also shows that if G: De-+M is geodesic

coordinate system about m0 and j8 is any broken C00 curve starting at m0
and ending at some point outside the geodesic neighborhood G(De), then
/(/?) > e.
Theorem 3. Let p: M x M —» R be defined by

p(mx, m2) = inf [/(a); a a broken C00 curve in M starting at
mx and ending at m2].
Then p is a metric on M, and the metric topology on M is the same as the
manifold topology on M.
Proof. Clearly p{mx, m2) = p(m2, mx). Also,

p(mi, m3) < p{mx, m2) + p(m2, m3)
because, given any e > 0, let a and /3 be curves from mx to m2 and from
m2 to m3, respectively, such that
Ka) < p(nh, m2) + ^, Kfi) < p{m2, m3) +
Then
p(pii, mz) < l(a) + l(fi) < p(tnu m2) + p{m2, m3) + e.
The remaining requirement p(mx, m2) = 0 o mx = m2 for a metric is also
satisfied because if mx ^ m2, let G: De —>■ M be a geodesic coordinate
system about mx not containing m2. Then every curve from mx to m2 has
length at least e, so p(mx, m2) > e.
The two topologies on M agree because for each m0 e M, there exists
e0(m0) > 0 and a geodesic coordinate system G : Deo(mo)-+M about m0. Then
G(De) = Bmo(e) for all e < e0(m0). □
Theorem 4. Let Mx and M2 be oriented Riemannian 2-manifolds. Let mx 6 Mx

and m2eM2. Suppose G1:De->M1 and G2:D£-^-M2 are geodesic
coordinate systems about mx and m2, respectively {see Figure 7.18). If
K2 ° G2 ° Gi_ 1 = Kx, where Kx is the curvature of Mx (i = 1,2), then
G2°Gr1 is an isometry.
Proof. We must show that d{G2» Gf1) = dG2 ° dG1~1 is an isometry at

each point. Since G2°GX~1 maps radial geodesics into radial geodesics,
dG2° dGx~x preserves lengths in the radial direction. By Gauss’s lemma,
dG2 o dGx~x preserves orthogonality. Thus we need only verify that lengths
in the ^-direction are preserved; that is, we need only show that
But, if we fix g e S1 and let

/ 8 \
m = dG‘°dr\se (j = 1, 2),
0,9)/
205
then Jacobi’s equation says that
p + (K,. G, . P)f, - 0,
where//0) = 0 and//(0) = 1 O' = 1, 2). By uniqueness of solutions,/x(r) =

/2(r) for all r\ that is,
Figure 7.18
Corollary. If M has constant curvature K, then given any pair ml5 m2 of points
in M and ul5 v2 of unit tangent vectors (vt e T(M, mf), there exists an
isometry <J> mapping a neighborhood of m1 onto a neighborhood of m2 such
that <!>(/«!) = m2 and JO0x) = v2.
Proof.Let Gp. Ds-> be a geodesic coordinate system about m{ (i = 1, 2).

Let geS1 be the rotation of De which maps ^Gi“1(t>1) onto dG2~l(v2).
Then G2° g: De M is another geodesic coordinate system about m2. Set
f = G2ogoG1"1. □
Remark. Thus, if K = 0, M is locally isometric with R2 (with its usual

Riemannian structure). However, it is not true that M is globally isometric
with R2 (consider, for example, the torus S1 x S1 with its product structure).
However, it can be shown that if M is simply connected and K = 0, then M
is isometrically the same as R2.
Remark. Jacobi’s equation also gives some information in the case of non¬
constant curvature. For example, suppose K > C for some constant C. Then
206
7.5: Isometries and spaces of constant curvature
if / is the solution of Jacobi’s equation f" + Kf = 0, and fx is the solution

of/; + Cfx = 0, then the theory of ordinary differential equations (Sturm-
Liouville systems) tells us that/ < fx; that is, the geodesics in M spread more
slowly than do geodesics in a space of constant curvature C; that is, geodesics
emanating from m0 will come back together faster in M than in a space of
constant curvature C.
7.5 Isometries and spaces of constant

curvature
Let Mx, M2 be smooth oriented Riemannian 2-manifolds. Suppose
/: Mx -> Mz is an isometry that preserves the orientation. Thus / is smooth,
injective, surjective, and it has a smooth inverse. Moreover, df preserves the
inner product on tangent spaces.
<df(vi), df(v2)} = <i?!, v2y (vx, v2 e T(MX, mx); mx e Mx),
and if r is any 2-form giving the orientation of M2, then f*r gives the orienta¬
tion of Mx. Note that since / preserves the Riemannian structure and the
orientation,/preserves everything defined in terms of these; for example, the
curvature, the Riemannian connection, and the parallel translation. Five
explicit statements follow.
(1) /induces a map/: S(MX) -> S(M2), defined by
f(mx, v) = (f(mx), df{v)) {{mx, v) e S(MX)).
This map is smooth and has a smooth inverse, namely/_1 = f~1. Note that
ti-2 °/ = f° ”1-
S(MX) > S(M2)
ni w2
Y
f Y
Mx —> M2
Moreover, since df is an orientation preserving isometry at each point,
v)) = gf(mx, v)
for all (mx, v) e S(MX) and g e S1; that is, f° g = g °f for each g e S1.
(2) / preserves the forms a>x, w2 and </< the 1-form of the Riemannian
connection). For, if t e T(S(MX), (mx, v)), then
<J*o>xM>)(t) = a>xM<dftt))
= <df(v), dn2(df(t))y
= <df(v), df(drrx{t))y
= <», dirx{t)y
= a*-1(0;
207
that is.
f*a>1M2 . cu1Mi.
Similarly,
f*OJ2M2 — a>2Ml.
To check that the connection is preserved, we use the uniqueness of the
Riemannian connection. Note that f*ifj is a connection form on S(MX) since
M2
(/*^2)(F) = >pM<df(V)) = i/»m2(F) = 1,

and
2) = g* o/*(</-M2) = (/o g)*(lfjM2) = (g o /)*</lM2
for each geS1. Moreover,
da>iM2 = ifjM2 A ca2M2,
da)2M2 = A cv1M 2,
so that
JWlMl = d(f*OJ1M2) = f^dco^ 2) = (/*i/«M2) A (/*a,2M 2)
= (/VM2) A C2M1,
^0)2M1 = d(f*CV2M 2) =f*(daJ2M2) = 2) A (7*0^2)
= -(Z*^2) A «XM1.
Thus the connection form f*4>M2 satisfies the simplified form of the structural
equations. By uniqueness,
7*^2 = lfjM 1.
(3) /preserves curvature; that is,f*K M2 = KM1. For
#M< = -(£"* o A <a2M. (i=l, 2),

so that
— (KM1 o 771)oi1Ml a 0)2M1 =

- j(7vM2)
= 7*(#"2)
— f*((— KM2 o tt^LUxM2 A OJ2M2)
= “C^2 0 772 °f)(f*CU1M2) A (f*C02M2)

= ~{KM2 ofo 7T1)cu1M1 A C02M1,
hence
O 77-1 = XM2 °f T,
01 1
that is,
Kmi = KM2 o f = f*KM2.
208
(4) / preserves parallel translation; that is, if a: [a,b]-+Mu then the

horizontal lift/° a of f ° a through (/ ° a (a), df(v)) is given by
/o a =/o a,
where a is the horizontal lift of a through (a(a), v). The reason is that
/*«AMa = and df(jeMi) =

(5) / maps geodesics into geodesics, for a is a geodesic if and only if & is
parallel along a.
Remark. Note that, given M, the set of all orientation-preserving isometries

h: M -» M forms a group under composition. We call this group Stf the
group of isometries of M.
Definition. is transitive if for each mx, m2 e M, there exists h e 3f such that

h{mf = m2.
Theorem 1. Let M be a smooth oriented Riemannian 2-manifold. Suppose the

group of isometries of M is transitive. Then M has constant curvature.
Proof. Let m0 e M. Then for any me M, there exists he 34k such that
h(m0) = m. Thus K(m0) = (h*K)(m0) = K(h(m0)) = K(m). □
Definition. Let 34k denote the group of isometries of M. For m0 e M, let 3fmo
denote the subgroup of 3%* leaving m0 fixed; that is,
•^mo = [heJf; h(m0) = m0].
34kmQ is called the isotropy group of M at m0.
Remark. Note that for k e JTmo, dk: T(M, m0) T(M, m0) is an orienta¬
tion-preserving isometry; that is, dk is a rotation of T(M, m0). Thus for each
k e Jfmo, there exists g e S1 such that dk{v) = gv for all v e T(M, m0). More¬
over, since d{kx ° k2) = dkx ° dk2, the map O: Jf~mo -> 51 defined by k ->
dk(m0) is a homomorphism.
Lemma. If M is connected, the homomorphism O: JTmo —> S1 is injective.
Proof. Let G: De -> M be a geodesic coordinate system about m0. Note

that for k e 3fm k: G(DS) —^ G{De) since k is an isometry. Moreover,
G_1 ° k ° G is orientation-preserving isometry of De leaving 0 fixed; that
is, G~1 o k ° G = g for some rotation g e S1. Since Jg(0)is also rotation byg,
dk(m0) = d(G o g o G_1)(w0) = dG ° g ° dG~1(m0).
Thus
k e kernel O o dk(m0) — identity
og — identity
<► k\G(De) = Go go G'1 = identity.
209
Thus if k e kernel O, the set
N = [m s M\ k{m) = m and dk: T{M, m) —> T{M, m) = identity]
is an open set in M. On the other hand, since k and dk are continuous, N is

closed in M. Since M is connected, N = M; that is, k = identity on M. Thus
kernel O = (identity)
and is injective. □
Theorem 2. Let M be a connected oriented Riemannian 2-manifold. Suppose

Jf1 is a subgroup of the group JF of isometries of M such that
(1) Jf’1 is transitive, and

(2) for some mQ e M, the homomorphism
is surjective, where mf — Jfmo n JF1. Then -FF — £FX.
Proof. Suppose h e JF. By transitivity of JFX, there exists h1 e JF1 such that
h\h{mjj) = m0.
Then hxh e Jfmo. Since is surjective, there exists k1 e CF mf such that
<f> (k1) = <J>(/F/z)-
Since ® is injective (by the lemma), k1 = hxh; that is, h = k1{hx)~x eJFx.
Thus FF c FFX\ that is, FF = #FX. □
Definition. For each me M, the set

[km = k{m); k e JTmJ
is called the orbit of Jfmo through m in M.
Theorem 3. Let M be an oriented Riemannian 2-manifold. Suppose, for some

m0e M, ®: Jfmo -> S'1 is surjective. Then, in a geodesic coordinate neighbor¬
hood of m0, the geodesics through m0 are the orthogonal trajectories
{reparameterized by arc length) of the orbits of Jfmo {see Figure 7.19).
210
Proof. If G: Ds M is a geodesic coordinate neighborhood of m0, then the

orbits of Jf „0 are the images under G of the concentric circles about the origin
in De (see the proof of the lemma above). These orbits are orthogonal to the
radial geodesics by Gauss’s lemma. By the uniqueness of orthogonal trajec¬
tories, these trajectories are the radial geodesics. □
Example 1. Consider the plane R2 with the usual Riemannian structure. Since
\\d/8fi\\ = 1, the map c: R2 -> S(R2) given by c(m) = (m, ex) = (m, (d/drj)
is globally defined, so the special connection is defined on all of S(R2).
Moreover, ex = d/dr1 and iex = d/dr2, so a>'x = drx and w2 = dr2. Thus, by
Remark 5, Section 7.2,
wx = (cos 9)d>x + (sin d)u>2
= (cos 6)v* drx + (sin 6)ir* dr2.
Similarly,
u>2 = — (sin 9)it* drx + (cos 0)t7* dr2.
Since <px = dd, the first structural equations become
da>1 = — (sin 6) dd A (tt* drx) + (cos 9) dd A (tt* dr2)
= <Pi A o)2,
and, similarly,
du>2 = — A
Thus the structural equations are in the simplified form, and, by uniqueness,
the special connection on S(R2) is the Riemannian connection. Moreover, as
we have seen for the special connection, d<px = 0, so the curvature K of R2 is
identically zero.
Clearly, each translation and each rotation of R2 is an isometry. Let be
the group generated by rotations and translations of R2. Then J^1 is transitive
(in fact, the group of translations is transitive), and d>: JTq1 S1 is surjective
(Jfo1 the group of rotations about 0 in R2); so, by Theorem 2, Jt?1 is the full
group of isometries of R2.
For m0 e R2, Jfmo is the group of rotations of R2 about m0. Since the orbits
of Jfmo are circles about m0, the geodesics through m0 are straight lines by
Theorem 3.
Example 2. Let S2 denote the sphere of radius r about the origin in R3, with
the induced Riemannian structure. Note that the group of rotations about
the origin in R3 is a group of isometries of S'2. It is the full group of isometries
of S2 by Theorem 2. Since Jf? is transitive on S2, the curvature K of S2 is
constant by Theorem 1. By the Gauss-Bonnet theorem,
2;J>vol = xOT = 2;
211
that is,
Kvol(S2) = 4„, = ^ = i
The geodesics through min S2 are the orthogonal trajectories of the orbits of
the group of rotations of R3 that leave m fixed. These orbits are circles about
m, so the geodesics through m are the great circles through m (Figure 7.20).
Example 3. Let D denote the open disc of radius 1 in R2. Regarding D as a

subset of the complex plane, consider the Riemannian structure (called the
Poincare, or hyperbolic, metric) on D defined by
<01, 02> = (01, »a e T(D, p);pe D),
where v1-v2 denotes the usual inner product (dot product) on T(D, p) and p
is the complex conjugate of p, so that pp = r2, where r is the Euclidean
distance from p to the origin. Thus,
8 8 1
8rx p 8r2 p " 1 ~ PP
and <(3/3^), (3/3r2)) = 0. Note that the radial lines in D through the origin
have infinite length. For, if a: [0, 1) -a- D is given by a(t) = teiB for some 9,
then
Ka) = JQ KOI dt = Jo | t2 dt = oo.
Let
<€ = ew -^ for some p e D, 0 < 9 < 2v

1 — zp
From elementary complex variable theory, ^ is a group (the conformal group)

of transformations of D onto itself. ^ is, in fact, a group of isometries of D.
212
For let/e If we identify T{D, z) with T(D,f(z)) by identifying both with

R2 in the usual way, then, since conformal maps preserve angles, df(z) =
Kz)g(z) for some real number A(z) > 0 and some rotation g(z) e S1. The
magnification factor A(z) is given by A(z) = \(df/dz)(z)\. But an elementary
computation shows that
dl(z) - zz) = 1 - /(z)/(z).

dzy } (1
Since vectors at z which are unit vectors in the Poincare metric have Euclidean
length 1 - zz, this implies that df{z) is a rotation which maps unit vectors
(in the Poincare metric) into unit vectors; that is, df(z) is an isometry for each
z, so/: D -> D is an isometry.
Thus % is a group of isometries of D. % is transitive because, for p e D,
(z ~ P)
fp-z
(1 - zp)
maps p onto the origin, hence for each pair py, p2 e D, /P2_1 °/Pl maps
Pi onto p2. Moreover, the isotropy subgroup of # at the origin is given by
•ô = [f: D -*■ D; /(z) = ei9z];
that is, is the group of rotations in R2. Hence d>: Jf0 S1 is surjective,
so, by Theorem 2, ^ is the full group of isometries of D. In particular, since
is transitive, the curvature K of D is constant by Theorem 1.
Since the isotropy group JT0 at 0 is the rotation group, whose orbits are
circles about 0, Theorem 3 tells us that the geodesics through 0 are the radial
lines (suitably parameterized). Since isometries map geodesics into geodesics,
and since the image of a radial straight line under a conformal transformation
of D is either a radial straight line or a circle which meets the boundary of D
at right angles, we see (Figure 7.21) that such lines and circles are the geodesics
Figure 7.21
in D. In particular, the sum of the interior angles of a geodesic triangle in D is

less than v, so that, by the Gauss-Bonnet theorem, the constant value of K
is negative (see Figure 7.22).
213
Exercise. Find the constant value of K. [Hint : K can be found either by com¬
puting the area of a geodesic triangle and applying the Gauss-Bonnet theorem or
by use of Jacobi’s equation.]
Figure 7.22
Remark. Although we have exhibited the hyperbolic space D as a sub¬

manifold of R2, note that the inclusion map is not an isometry. In fact, it can
be shown that the space D cannot even be imbedded as a Riemannian
submanifold of R3.
Remark. We have exhibited three Riemannian manifolds R2, S2, and D of
constant curvature K = 0, K > 0, and K < 0, respectively. It turns out that
any two simply connected complete Riemannian 2-manifolds of the same
constant curvature K are isometric. So these three examples are essentially
the only examples of simply connected complete 2-manifolds of constant
curvature. These three examples are important in several different mathe¬
matical disciplines, as illustrated in Table 7.1.
Table 7.1
Group theory
Riemannian (group of Complex Elementary
Space geometry isometries) variables geometry
R2 K = 0 Euclidean motions Complex plane Euclidean

in R2 (rotations geometry
4- translations)
S2 K > 0 Rotation group Riemann sphere Spherical
in R3 geometry
(elliptic
geometry
on P2)
D K < 0 Conformal group Disc (hyperbolic Hyperbolic
plane, or upper geometry
half-plane)
214
Relative to the last column in particular, these three examples are the only
“elementary” geometries. Under the replacements points —>points, straight
lines -> geodesics, length -> arc length, angles —> angles, and congruence ->
isometry, most of the axioms for Euclidean geometry are satisfied by these
examples. The most noteworthy exception is that Euclid’s fifth postulate, the
parallel postulate, fails to hold on D: given a “straight line” / and a pointp
not on /, there are infinitely many “straight lines” through p which never
meet / (see Figure 7.23).
In the case of elliptic geometry, it is customary to use P2 as model rather

than S2, so that any pair of “straight lines” will meet in at most one point.
In this elliptic geometry a “straight line” fails to divide the remaining points
into two disconnected parts.
In the case of complex variable theory, R2, S2, and D are the only simply
connected complex 1-manifolds, a strong form of Riemann’s mapping
theorem.
215
8 Imbedded manifolds in R3
In the previous chapter we studied the intrinsic geometry of a surface. Now

we look at the properties of a surface as it lies in R3.
Let (M,f) be a submanifold of R3, where M is an oriented 2-manifold. Let
M be given the Riemannian structure induced from /: for v1,v2e T(M, m),
<»i> >m = df(v2y>nmy
Definition. Given (M,f), the spherical map on M is the map 5: M^-S2 defined
as follows. For me M, let s(m) e T(R3, f(m)) be the unit vector perpen¬
dicular to df(T(M, m)) that is consistent with the orientations of T(M, m)
and T(R3, f(m)); that is, if v is a unit vector in T(M, m), then s(m) is the
(unique) unit vector in T(R3, firrij) such that {df(v), df(iv), s(m)} is an
oriented orthonormal basis for T(R3, f(m)). Identifying T(R3,f(m)) with
R3 in the usual way, we may regard s(m) as a point in S2. (Figure 8.1).
s(m) s(m)
Figure 8.1
We shall find it convenient to identify T(M, m) with df(T(M, m)). Further¬

more, since df(T(M, m)) and T(S2, s{m)) are the same subspace of R3 under
216
8: Imbedded manifolds in R3
the usual identification of the tangent spaces to R3 with R3, we shall identify
T(M, m) and T(S2, s(m)). Then
ds(m): T(M, m) -> T(5'2, .s(m)) = T(M, m),
so Gfc(m) is a linear transformation from T(M, m) into itself. The main result
of this chapter is the theorem of Gauss: The curvature K(m) of M at m is
equal to the determinant of ds(m). In fact. Gauss originally defined K (for
imbedded manifolds) to be this determinant, and he was surprised to discover
that K was in fact intrinsic; that is, K depends only on the Riemannian
structure on M and not on the particular imbedding /.
Let us first note what manifolds in R3 look like in local coordinates in
terms of the spherical map. For/: M R3, letf = r} °/(y = 1,2, 3), so that
Am) = ffm), ffm))
for all m e M. Here rf are the coordinate functions on R3. Let p.\ UM be
a smooth injective map from an open set U <= R2 onto an open set in M such
that /x_1 is smooth, and such that if r is a 2-form on M which gives the
orientation of M, then /x*r is a positive multiple of drx A dr2. Thus /x_1 is
an oriented coordinate system on /x(£/). Let xt = r, ° /x_1 (i = 1,2) denote
the coordinate functions on /j-(U). Then d/cbq = d^{8j8rb) and
8_ g(A ° /-0 g
df = d(fo n) (i = 1, 2),
dx i (D -1 dri 8rk
so that the matrix for df, relative to the bases {d/Sx(} for T{M, m) and {d/drj}
for T(R3, f(ni)), is
/a(/l0 a(/i ° z4)

ar 1 8r2
d(/2 ° Z4) a(/2 ° m)
df■
a/-i 8r2
S(/3 /x) a(/3 ° m)
0
\ a/-! Sr2
/x(t7) is then given t
/ A JL\ = y■ Kfk 0 A4) s(/fc 0 m)

\ dx/ 8Xj / = ii
fc^ ar, dri
Since dfiS/Sxf) x df(8/8x2)—cross product in R3—is perpendicular to
df{T(M, m)) at each point me U and is consistent with the orientations of
T(M, m) and T(R3,f(m)), the spherical map on /x(t/) is given by
_ df(8/dXl) x dfiojdxf
Sm \\df(8J8xi) x df(8/8x2)\\
1 l d(/i ° I1) a(/2 0 f) Z(f3 0 /x) \ v j 8{f1 o /x) 8{f2 o /x) b(f3 o /x)^ ^
“a\ a/-x ’ ’ ar, / \ ar2 ’ ar2 ’ ar2 J
217
where a is the magnitude of this cross product. In particular, we see that s is a

smooth map.
Now consider a fixed point m0 e M. Altering/by a translation and rotation
if necessary (this will leave the induced Riemannian structure on M un¬
changed), we may assume that /(m0) = 0 and that df(T(M, m0)) is tangent
to the (ru r2)-plane R2 and is oriented consistently with the natural orientation
of R2; that isA dr2) gives the orientation of T(M, m0) (Figure 8.2).
Let p : R3 -> R2 be projection; that is,
p(ui, u2, w3) = (l/j, u2).
rz
Then p °f: M R2; p ° f(m0) = 0; and
d(p 0 f)(m0) = dp o df(m0) = df(m0)
maps T(M, m0) isomorphically onto T(R2, 0). By the inverse function theo¬
rem, there exists a neighborhood U of 0 in R2 such that (p °/)_1 is defined
and is smooth on U. Let p = (p of)-1: UM. Then p-1 is a local coordi¬
nate system about m0. Moreover, since p of o p = iUf
fo p(Ul, u2) = («i, u2, g(uu u2)) ((«!, u2) e U),
where g = r3 °/° p: UR1. Thus, if xl5 x2 are the coordinate functions

on p(U), then the matrix for df relative to this coordinate system is
\ 8g_ ?g_ I
\ drx dr 2 )
218
and the spherical map on ij-(U) is given by
“a)) = *^3 ('• °- W,("*’ "a)) x (°> >> W2(">• “»>)

“ a(u„ «2) (“8^ "2>' lu" “2,> ');
that is.
8g_
dr i
*')■
where a = V(8g/8rx)2 + (8gj8r2f + 1. Note that at m0 = ^(0), s(mQ) _L
df(T(M, m0)) = i?2 implies that (dg/d/qXO) = (8g/8r2)(0) = 0 and a(0) = 1.
Now under our identification of T(M, m0) with df(T{M, ra0)), (8/8xi)(m0)
becomes identified with (8/8ri)(0) (z = 1, 2), so that the entries b{j for the
matrix for
ds: 7XM, rn0) -> r(M, m0)
relative to the basis {d/8xt} are given by
bn = < ds ±\°\
8x i 1 8x
=
dr A d(s — d(fj o S o fji)
aF,(r’ °" ^
Thus
8 ( 1 5g\ 1 8a 8g 1 82g d2g
=
8r1 \ a 8rJ 0 a2 8r1 8r1 o a 8rx2 '8ri2
and, similarly,
S2g
*12 — — *21?
*/*i 8r2
*2g
*22 _ er22
that is,
a2g 82g
dzy fiz-! dr2 dzq
ds(m0)
a2g *2g
d/q 8r2 8r2 dr2 / 0
In particular, assuming Gauss’s theorem (which still remains to be proved)
that K is the determinant of ds, we obtain a qualitative description of the
219
behavior of f(M) near the point f(m0) under various assumptions on the
curvature K(m0). For from the critical point theory of functions of two
variables, we know that the function g will have a maximum at the critical
point 0, provided that the eigenvalues of the matrix
are both negative and that g will have a minimum if the eigenvalues are both
positive. Thus if K(m0) > 0, then near 0, f(M) lies either completely above
or completely below the i?2-plane. On the other hand, if K(m0) < 0, then
ds(m0) has one positive and one negative eigenvalue, and g has a saddle point
at 0. If K(m0) = 0, the behavior of /(M) near 0 is undetermined (see Figure
8.3).
In order to prove Gauss’s theorem, we need an explicit description of the
local geometry of imbedded manifolds.
K(m0) > 0 K(m0) > 0
K(mo) < 0
Figure 8.3
220
Theorem 1. Let (A/,/) be an oriented 2-manifold in R3. Let ex, e2: S(M) -> R3
be defined by
efm, v) = df(v),
efm, v) = df(iv).
Let ifi be the l-form on S(M) defined by
= (dex, ef)\
that is, for (m, v) e S(M) and t e T(S(M), (m, v)), deft) e T(R3, efm, v))
= R3, and
4>{t) = <deft), efm, v)>.
Then >p is the l-form of the Riemannian connection on M.
Proof
Invariance. For g = eiB e S1,
g*f = if, o dg = <<&! o Jg, e2 o g> = <0d(ex o g), e2 o g>.
But, for (m, i>) e S(M),
ex ° g(ra, u) = ^(m, cos 6 v + sin 0 (zV))

= cos 9 df(v) + sin 9 df(iv),
e2 o g(m, v) = e2(m, cos 9 v + sin 9 (iv))
= — sin 9 df(v) + cos 9 df(iv),
so
e1 ° g = cos 9 ex 4- sin 9 e2,

e2 ° g — ~ sin 9 e1 + cos 9 e2.
Thus
d(ei og) = cos 9 de1 + sin 9 de2.
Now since (ex, ef) = 1, <e2, e2> = 1, and <el5 c2> = 0,
0 = d(ex, <?!> = 2(delt ef),

0 = d(e2, e2} = 2(de2, e2>,
0 = d(ex, e2} = (dex, e2> + (ex, def)\
so that fidex, ef) = (de2, e2> = 0, and <el5 def> = -{dex, e2y. Hence
g*>p = (d(ex o g), e2ogy = (cos2 9 + sin2 9)(dex, e2> = </>.
Normalization. For (m, v) e S(M), V(m, v) = y(0), where y: i? ->• S'(M) is

given by
y(0) = (m, cos 9 v + sin 9 (iv)).
221
Hence
>P(V(m, v)) = 0)) = <<&i(y(0)), <?2(y(0))>
= <Oi o y)(0), df(iv)).
But
ex o y(9) = cos 9 df(v) + sin 9 df{iv),
so that
{ex o y)(0) = — sin 9 df(v) + cos 9 df(iv)\0 = df(iv),
and
>KV{m, v)) = 1.
Thus i/r is a connection form on M. To verify that it is the form of the

Riemannian connection, it suffices to check that the simplified forms of the
first structural equations are satisfied. To do this, note that when (m, v) e S(M)
and t e T(S(M), (m, v)),
"i(0 = <dn(t), v> = (df° dir(t), df(v)y = (d(f° rr)(t), ex(m, v)\
so
"i = (d{fo tt), ex>.

Similarly,
“2 = (d(f° 7r), C2>.

Also note that the component of d(f ° tt) in each /y-direction is the exterior
differential of a function, namely,
dr} ° d(f o 77) = d(r}- °f° tt),
so that d(d(fo tt)) = 0, where the exterior derivative of a 3-tuple of forms is

computed componentwise. Hence
dori = d(d(f o v), et}

= <dd(fo tt), e{y - <J(/o 77 ), dety.
(*) dwt = —id{fo tt), dety.
But
d(f°7T) = (d(f° v), «!>«! + (d(J° tt), e2ye2,
since ex{m, v) and e2(m, v) form an orthonormal basis for df(T(M, m)) for
each (m, v) e S(M). Thus
d(f°v) = + <o2e2,
222
hence, from (*),
dw1 - —(wiei + oj2e2, del) — — a>2 A (e2, dexy = — co2 a */> = */» A oj2,
dw2 — —(u>1e1 + w2e2, de2y = —wx a (ex, de2y = a)1 a = — </» A a>1. □
Interpretation of parallel translation for imbedded manifolds

Let a be a curve in M, and d some lift to S(M). Then ex ° d is a curve in
R3 lying on the unit sphere, so that d(ex ° a)(d/dt)to = (d/dt)(ex ° d)io is
always perpendicular to ex ° d(r0). Now d is horizontal if and only if = 0.
By Theorem 1, this is the case if and only if d(ex ° d)(d/dt)tQ is perpendicular
to e2 o a(t0). Since ex and e2 span the tangent space, a is horizontal if and only
if (d/dt\ex ° a)to is perpendicular to df(T(M, «(t0))) for all t0.
Thus is X is a unit vector field defined along a and d(r) = (a(t), X(a(t))),
then X is parallel along a o a is horizontal o (d/dt)(ex ° a) = (d/dt)(X o «)
is always perpendicular to the manifold; that is, X is parallel along a if and
only if X° a is “constant along the manifold”: the tangential component of
the derivative of X ° a in R3 is identically zero. For imbedded manifolds, this
characterization can be used in place of the abstract notion of connection.
Theorem 2 (Gauss). Let (M,f) be an oriented 2-manifold in R3. Let s: M -> S2

be the spherical map. Then for each me M, det ds (m) = K{m).
Proof. Let s: S(M) -> S(S2) be defined by
s(m, v) = (s(m), df(v)).
Let ojx, a>2 and if/ be the structural 1-forms on S(M), and let o>l5 a>2, and ip be
the corresponding forms on SOS2). Then
(I)
For
s*ip = ip o ds = (fiei o ds, e2° sy = ° s), e2 ° i).
But
ex o s(m, v) = efsfn), df(v)) = df(v) = efm, v),

e2 o s(m, v) = e2(s(m), df(v)) = i df(v) = s(m) x df(v)
= df(iv) = e2(m, v);
so that eto s — eu and s*ip = (dex, e2y = ^.

s*o)x = OnOJi + a12o>2,
,s*u>i = a2Xiox + a22oj2,
where the aif are the smooth functions on S(M) such that
is the matrix for ds(m) relative to the basis {v, iv} for T(M, m).
223
For if {m, v) e S(M) and t e T(S(M), (m, v)\ then
(5*Wl)(0 = O
= (dir(ds(t)), df(v)') (since s(m, v) = (s{m), #(f)))
= ° s)(t), df(v)>
= <</($ ° w)(0, #(*>)>
= <ds(dn{t)), J/(y)>
= + oj2(t)(iv)), df(v)>
= <ds(v), y)w!(?) + (ds(iv), v}oj2(t),
since d/fy) is identified with v. Similarly,
(i*o>2)(r) = (ds(v), ivyw^t) + <ds(iv), iv}a>2(t),
completing the proof of (II).
(Ill) For each me M, det ds(m) = K(m).
For
-(K°tt)o>1 A W2 = di/j = d(s*f) - s*(df)
= £*(—afi A a>2) (since S2 has constant curvature 1)
= —5*60! A S*CO2
= —(Unî + #12^2) A (021^1 "F ^22^2)
= — (Rlla22 a12a2l)(ul A a>2,
so that
(£?ii ^12\
1 = det ds{m), □
#21 #22' (m,u)
Remark. The linear transformation ds(m) is called the second fundamental

form of M at m. Note that it is a self-adjoint linear transformation, since its
matrix relative to the special coordinates discussed above is the symmetric
matrix
/ g2g \
I 8rf dr2 8r1 \
I 82g 82g I
\ 8rx 8r2 8r2 /
Theorem 3. Let (M,/) be a compact oriented 2-manifold in R3. Then there

exists m0 e M such that K(m0) > 0.
Proof. Consider intuitively any sphere with M inside. Shrink this sphere until
it just touches M at some point m0. Then the curvature of M at m0 is greater
than the curvature of this sphere.
Consider more precisely the function r2 of on M, where r is the distance
from the origin in R3. Let m0 e M be a point where r2 of assumes its maxi¬
mum. By rotating about the origin in R3 if necessary, we may assume that
224
f(m0) is on the r3 axis; that is, that f(m0) = (0,0, c) for some c. Then r3 of also
has an extremum at m0 since |r3 ° /1 (m0) = r» f(m0) > r ° /(m) > |r3 ° f | (/rz)
for all m e M. Hence for v e T(M, m0),
= dr3(df(v)) = </(r3 °/)(tz) = 0;
that is, df(T(M, m0)) is perpendicular to the r3 axis. We may assume that the
orientation of df(T(M, m0)) agrees with the orientation on the ru r2 plane.
(Otherwise, a rotation through an angle of v about the rx axis will accomplish
this.) We may further assume, by rotating about the r3 axis if necessary, that
d/dr1 and d/dr2 are eigenvectors of ds(m0). Then since f(m0) = (0, 0, c), the
special coordinates are valid on the translate if /(A/) by (0, 0, — c); that is,
there exists g: U -> M, U R2 such that
/° +(« 1, «2) = (Ml, «2, c + g(ux, «2))

for some g: U -> R with g(0, 0) — 0 and
dg_
= 0.
drx o dr2
Then, since d/8r1 and d/dr2 are eigenvectors of ds(m0), ds(m0) has matrix
d2g
0
dr2
d2g
0
dr.2
2/0
But r2 o f°g. has a maximum at (0, 0). Hence since
r2 of o fjfm, u2) = uf + «22 + [c + g(i/i, «2)]2,
then
2 + 2
0 > J-s (r2ofof)
(t;) + 3/y2
for j = 1,2. Thus
g2g 1 , 32g
< — and < —>
e/-,2 c dr2 o c
so that
K(m0) — det dsirrio)

S2g S2g
^2^2 ^ 4 > °- □
0 C
Corollary. The torus S1 x S'1 with its Riemannian product structure cannot
be imbedded as a submanifold of R3.
225
Proof. Since the covering map R2 -> S1 x S1 is a local isometry, S1 x S'1

has curvature identically zero. □
Theorem 4. Let (M,f) be a compact connected oriented 2-manifold in R3.

Suppose the curvature K of M is never zero. Then in fact K > 0 everywhere.
Furthermore, the spherical map s: M —> S2 is injective, surjective, and has
a smooth inverse; that is, M and S2 are diffeomorphic (isomorphic as
manifolds).
Proof. The first statement is a consequence of the previous theorem. For the
second statement, consider the map v os mapping M into the projective plane
P2, where v. S2 P2 is projection. We show that tt o s is a covering map.
First, since det ds = K > 0, ds is everywhere nonsingular, hence so is
d-n o ds = d(v o s). By the inverse function theorem, tt o s is locally one-to-one
and is an open map. Moreover, -tt ° s is onto. For if p eP2—say p — tt(P),
p e S2—consider the plane in R3 perpendicular to p. Moving this plane out
in the direction of p toward infinity and then back until it just touches M at
m0, we find that s(m0) = ±p, so that tt ° s(mj) = p. More precisely, let
m0 e M be such that the function m -» <f(m), pj assumes its maximum at
m0. Then p _L df(T(M, m0)) because, for v e T(M, m0),
<df(v),p} = d(( f p})(v) = 0.
Hence s(m0) — ±p, and n°s(m0)=p. Finally, since M is compact,

(tt o s)~1(p) is finite for eachp e P2. (It is closed in M, hence compact; on the
other hand, each point in v■~1(p) is open in the relative topology, since tt is
locally one-to-one.) Hence for U a sufficiently small open set about p,
(tt o 5)_1([/) is a union of disjoint open sets each mapped homeomorphically
onto U by tt ° s. Thus tt ° s is a covering map, and M is a covering space of P2.

But the only covering spaces of P2 are S2 and P2 itself. Since P2 is not
orientable, M must be homeomorphic to S2 and s must be a smooth homeo-
morphism with a smooth inverse. □
Theorem 5. Let (Mu fj) and (M2,f2) be oriented 2-manifolds in R3. Suppose
Mx and M2 are tangent along a curve a in R3; that is, suppose there exist
curves at: [a, b] -> Mt(i = 1, 2) such that
fi ° «i = h ° «2 = «,
and
dfx(T(M^, «,(/))) = df2(T(M2, aft)))
for all t e [a, b]. Then parallel translation along a is the same in both mani¬
folds; if v1 e T(Mi, afa)) and v2 e T(M2, a2(a)) with dffvf = df2(v2), then
dfl (parallel translate of vx along g^)

= df2 (parallel translate of v2 along a2).
226
Proof. Let s^Mi-^S2 denote the spherical map (/= 1,2), and let
Si'. S(Mt) -> S(S2) denote the corresponding map on the circle bundles. Since
dfi(T(Mx, ax(t))) = df2(T(M2, a2(t)))

for all t g [a, b], we have o « = + s2 o «. We may assume that sx a = °
+ s2o a, for otherwise we may reverse the orientation on M2 so that this is

the case. (Note that orientation reversal on M2 has the effect of changing the
sign of the connection form on S(M2) and hence does not affect parallelism.)
From the proof of Theorem 2,
^ *4> = 4>i,
where ifjt is the connection form on Let at: [a, b] -> S(Mt) denote the
horizontal lift of through (aâ), vt). Then
° *°(£)) = ffefaii,)))= R*?)(*‘(i))

*'{da'{i)) =0
=
since a, is horizontal. Hence ° [a, b] S(S2) (i = 1, 2) is the horizontal

lift of iioctj = s2o a2 through
h ° «i(«) = (Ji(«i(a)), dfM)

= (s2(a2(a)), df2(v2))
= 0 “2(0)-
In particular, sx ° ax = s2 ° a2. Hence, if v{(t) denotes the parallel translate of

Vi along ccx to ai(f), then a^t) = («j(t), v^t)) (i = 1, 2), and
0i(«i(0), dfiivyit))) = ii(«i(0)

= sx° «i(0
= S2° a2(t)
= df2(v2(t)));
that is, dfx(vx(t)) = df2(v2(t)). □
Remark. Theorem 5 gives us a way of seeing geometrically how parallel

translation behaves for submanifolds in R3. Let «:[a,i]->-Mbea curve.
Consider the family of tangent planes to f(M) along the curve /° a in R3.
For t0, t g [a, b], the intersection of the tangent plane at/° a(t0) and the tangent
plane at /° a(t) will generally be a line. The limit of this line, as t —»■ t0, will
be a line through /° a(/0) in the plane df(T(M, a(t0)). The collection of all
these lines, for t0 e [a, b], forms a surface D, called a developable surface. It
turns out that such a surface D is flat (K = 0) and is tangent to f(M) along
f o a. Hence parallel translation in M along a is the same as parallel trans¬
lation along/o « in D. But since D is flat, it is locally isometric to a piece of
the plane; that is, in a neighborhood of a, D can be rolled out on the plane
where parallel translation is ordinary translation. In particular, parallel
227
translation along a circle on a sphere in R3 is the same as parallel translation

along that circle in the cone tangent to the sphere along that circle. But the
cone can be rolled out on the plane where parallel translation is ordinary
translation; then the cone can be rolled back around the sphere to find the
parallel translate on the sphere (Figure 8.4).
Theorem 6. Let M be an oriented connected 2-manifold, and let f g: M —x R3

be two isometric imbeddings. Suppose that the second fundamental forms
of (M,f) and (M, g) are the same. Then there exists a rigid motion (isometry)
® of R3 such that f = <D ° g.
Sketch of Proof. Choose a point (m0, v0) e S(M) and a rigid motion d>
carrying g(m0) into f(m0), dg(v0) into df(v0), and dg(T(M, m0)) onto
df(T{M, m0)). Let /i - / and f2 = O ° g; then ffm0) = /2(m0), dffv0) =
dfzipo), and
dfi(T(M, m0)) = df2(T(M, m0)).
Hence sfm0, v0) = s2(m0, v0). Furthermore, because 0 is an isometry, the

second fundamental forms of (M,/x) and (M,/2) are equal.
Note that as a result of formulas (I) and (II) in the proof of Theorem 2, the
fact that the second fundamental forms for (M,f), i = 1,2, are equal implies
st: S(M)^S(SZ)
has the following property: sffp) — s2*(f) and iy*^) = i2*(<;i>i), i = 1,2.
Note also that S(S2) can be identified with the space of 3 x 3 orthogonal
matrices of determinant one, for a point of S(S2) consists of a pair (u, h),
where u is a unit vector in R3, and v is a unit vector perpendicular to u. The
matrix corresponding to (u, v) is the one whose columns are u, v, and u x v.
228
Lemma. Let ax, a2 be two curves [0, 1] -> S(S2) such that aflf) = a2*(f) and
ai*(î) = «2*(<5i) (* = 1, 2). Let p be the curve P(t) = a2(t)-1 -aft), where
inverse is matrix inverse and the dot indicates matrix multiplication. Then
P*$) = P*(&d = 0.
Proof. We leave the proof of this lemma to the student. It involves computing
d^ in terms of dax and da2.
Corollary to the Lemma. If in addition, aÔ) = a2(0), then ax — a2.
Proof. Since ajl5 a>2, f span the cotangent space at each point of S(S2), the
lemma implies that dp is identically 0, so that p is a constant map. Since
«i(0) = a2(0), P(0) is the identity matrix, and, therefore, so is p(t) for all
t e [0, 1], Hence aft) — a2(t) for all t e [0, 1],
We are now in a position to show that fx = /2. Let y be a curve in S(M)

starting at (m0, v0), and let at = st ° y. Then the curves a, satisfy the hypothesis
of the lemma, and ^(0) = a2(0), so we conclude from the corollary that
ai — «2- In particular, sx{y(t)) = i2(y(0) id £ [0, 1]). Since any point of S(M)
is reachable by a smooth curve starting at (m0, v0), we have sx = s2; that is,
for every (m, v) £ S(M), sfm) = s2(m) and dffv) = df2(v).
Consider the map fx — /2: M -> R3, where (fx — /2)(m) = fx(m) — f2(m).
Since dffv) — df2{v) = 0 for any unit vector, we conclude d(fx — f2) = 0;
that is,/i = f2 + constant. Since fx(m0) = /2(m0), we have fx = f2. □
229
Bibliography
Topology
1. Alexandroff, P. and Hopf, H. Topologie. Berlin: Springer-Verlag, 1935.
2. Bourbaki, N. Topologie Generate. Paris: Hermann, 1953.
3. Eilenberg, S. and Steenrod, N. E. Foundations of Algebraic Topology. Prince¬
ton, N.J.: Princeton University Press, 1952.
4. Greenberg, M., Lectures on Algebraic Topology. New York: Benjamin, 1967.
5. Hocking, J. G. and Young, G. S. Topology. Reading, Mass.: Addison-Wesley,
1961.
6. Hu, S. T. Elements of General Topology. San Francisco: Holden-Day, 1964.
7. Kelley, J. L. General Topology. New York: Van Nostrand, 1955; Springer-
Verlag, 1975.
8. Lefschetz, S. Introduction to Topology. Princeton, N.J.: Princeton University
Press, 1949.
9. Pontryagin, L. S. Foundations of Combinatorial Topology. Rochester, N.Y.:
Graylock, 1952.
10. Seifert, H. and Threlfall, W. Lehrbuch der Topologie. Leipzig: Teubner, 1934.
11. Sierpinski, W. Introduction to General Topology. Toronto, Ontario: University
of Toronto Press, 1934.
12. Spanier, E. H. Algebraic Topology. New York: McGraw-Hill, 1966.
13. Wallace, A. H. Introduction to Algebraic Topology. New York: Pergamon,
1957.
Geometry
1. Auslander, L. and Mackenzie, R. E. Introduction to Differentiable Manifolds.
New York: McGraw-Hill, 1963.
2. Bishop, R. L. and Crittenden, R. J. Geometry of Manifolds. New York:
Academic, 1964.
3. De Rham, G. Varietes Dijferentiables. Paris: Hermann, 1960.
4. DoCarmo, M. Differential Geometry of Curves and Surfaces. Englewood
Cliffs, N.J.: Prentice-Hall, 1976.
5. Flanders, H. Differential Forms, with Applications to the Physical Sciences.
New York: Academic, 1963.
6. Guggenheimer, H. W. Differential Geometry. New York: McGraw-Hill, 1963.
7. Hicks, N. J. Notes on Differential Geometry. Princeton, N.J.: Van Nostrand,
1965.
8. Kobayashi, S. and Nomizu, K. Foundations of Differential Geometry, Vol. I.
New York: Interscience (Wiley), 1963.
9. Laugwitz, D. Differential and Riemannian Geometry. New York: Academic,
1965.
10. O’Neill, B. Elementary Differential Geometry. New York: Academic, 1966.
11. Sternberg, S. Lectures on Differential Geometry. Englewood Cliffs, N.J.:
Prentice-Hall, 1964.
12. Warner, F. Foundations of Differentiable Manifolds and Lie Groups. Glenview,
Ill.: Scott, Foresman, 1971.
13. Whitney, H. Geometric Integration Theory. Princeton, N.J.: Princeton Uni¬
versity Press, 1957.
230
Index
Admissible open set, 63 Complete, 40 Derivation, 113

Arc length, 194 Completely regular, 37 Diameter, 89
Arcwise connected, 52 Completion, 40 Diffeomorphic, 129
Axiom of choice, 5 Complex structure, 111 Differential, 114
Cone, 83 exterior, 122
Conformal structure, 11 I Differential form, 118, 121
Baire category theorem, 43
Conjugate subgroups, 74 closed, 128
Ball. 6
Connected, 11 degree of. 121
Banach space, 43
Connection, 178 exact, 128
Barycenter, 85
Riemannian, 188 Divergence, 126
Barycentric coordinates, 80,
Connection form, 179
162
Contiguous, 94 Edge, 97
Barycentric subdivision, 87
Contiguous equivalent, 94 Edge equivalent, 97
Basis, 8
Continuous function, 13 Edge path group, 98
equivalence of, 9
extension, 20 End (of a graph), 102
Betti number, 125, 158
Contractible, 51 Euler characteristic, 102, 158
Boundary, 155, 156
Convex, 79 Exterior multiplication, 120
Boundary map, 155
Convex independent, 79
Bracket, 132
Coordinate function, 11 1 Face, 81
Broken C" curve, 116
system, 111 Finite, 4
Cotangent bundle, 118 Finite intersection property, 12
c-independent, 79 Cotangent space, 1 16 First category, 43
Cartan structural equations, Countable, 4 First countable, 39
187-188 Countably compact, 40 Form, 120, 121
Cartesian product, 2, 5 Covering, 12 Free group, 103
Cauchy sequence, 40 Covering homotopy theorem, Function, 3
Chain, 154 66 Fundamental cycle, 196
Circle bundle, 175 Covering space, 62 Fundamental group, 55
Class Cs, 111, 112 Covering transformation, 76
Closed set, 7 Cross product, 126 Gauss-Bonnet theorem, 196
Closure, 7 Curl, 126 for 2-simplices, 195
Coboundary, 160 Curvature, 187 Gauss’s lemma, 200
Coboundary operator, 160 Cycle, 156 General position, 83
Cochain, 160 Geodesic, 194
Cocycle, 160 Geodesic coordinate system,
Cohomology, De Rham, 128 Deck transformation, 76 199
simplicial, 160 De Rham cohomology, 128, Geodesic curvature, 194
Compact, 12 174 Glueing lemma, 50
Complement, 2 De Rham’s theorem, 164 Gradient, 126
23]
Index
Graph, 101 Manifold, complex analytic, Simple ordering, 3

Grassmann algebra, 121 1 1 1 Simplex, 80
Group of isometries, 209 differentiable, 1 10 open, 81
orientable, 144 Simplicial approximation, 92
oriented, 144 Simplicial complex, abstract,
Hausdorff space, 26 smooth, 112 98
Homeomorphism, 15 Euclidean, 81
smoothly triangulated, 161
Homology group, 156 Simplicial map, 90
Maximum principle, 5
Homotopic maps, 49 Mesh, 89 Simply connected, 70
Homotopic paths, 53 Metric, 5 Skeleton, 73
Homotopy, 49 Metric space, 5 Slice, 140
Homotopy operator, 130 Smooth, 1 12
Metrizable, 39
Homotopy type, 51 Mobius strip, 147 curve, 116
Horizontal lift, 181 vector field, 118
Hyperbolic metric, 212 Normal topological space, 28 Sphere bundle, 175
Normed linear space, 43- Spherical map, 216
Nowhere dense, 43 Star, 91. 163
Implicit function theorem, 135
Star-shaped region, 132
Indexing set, 4-5
One -to-one correspondence, 4 Stationary, 67
Induced map on De Rham
1-form of a connection, 179 Structural equations, 187-188
cohomology, 129
1-point compactification, 35 Subcomplex, 82
on differential forms, 127
Open covering, 12 Subdivision, 85
on fundamental groups, 59
Open set, 7 Submanifold, 138
Infinite product, 5
Orbit, 210 Subset, 1
Injective, 4
Inner automorphism, 58 Orientable, 144 Surjective, 4
Orientation, 143
Integral curve, 140
Oriented, 143, 144, 153
Integra] of a 2-form, 195-196
Orthogonal group, 136 Tangent bundle. 118
Interior multiplication, 131
Tangent space, 114
Interior of a simplex, 81
Paracompact, 148 Tangent vector, 1 12
Intersection, 1
Parallel translate, 181 to a curve, 117
Inverse function theorem, 134
Parametrized by arc length, Tietze extension theorem, 34
Inverse image, 4
194 7Vspace, 26, 28
Inverse of a function, 4
Partial ordering, 3 Topological space, 7
of a homotopy class of paths,
Partition of unity, 149 Topology, 7
55
Pasting lemma, see Glueing discrete, 10
of a path, 53
lemma. generated by, 8. 16
of a route, 97
Path. 52 relative, 10
Isometric embedding, 40
Poincare metric, 212 Torus, 19
Isometric manifolds, 151
Poincare’s lemma, 130 Totally bounded, 40
Isometry, 151
Product Transitive, 209
Isomorphic covering spaces, 75
of homotopy classes of Tree, 101
Isotropy group, 209
paths, 55 Triangulated manifold, 161
of paths, 53
Jacobi identity, 133 of routes, 97
Jacobi's equation, 203 Product topology, 17 Uniform boundedness
Join, 83 Projective plane, 63 principle, 48
Unimodular group, 136
Regular covering space, 76 Union, 1
Length of a curve, 194 Regular topological space, 28 Unitary group, 137
Lie algebra, 133 Relation, 2 Universal covering space, 70
Lie group, 137 Relative topology, 10 Urysohn’s lemma, 31
Limit point, 7 Riemannian connection, 188
Locally arcwise connected, 62 Riemannian manifold, 150
Locally compact, 34 Riemannian structure, 150 Vector field, 118
Locally connected, 62 Riemann surface. 111 Vertex, 81
Locally Euclidean, 109 Route, 97 Volume element, 125, 150,
Locally finite, 148 185
Locally simply connected, 70 Second category, 43
Local one-parameter group of Second fundamental form, 224
transformations, 142 Sequentially compact. 40 Weaker topology, 16
232
Date Due
AUG 3 t 19BG
l» rt \ f e
«
° r -'.fi
Shw 4f / i$| J 'i
C fIAN
r• ■' ' iqqn
31 IjOU
. - n nfl in
A Z “ 20 JU
J6
J QA 611 .S498 1976
Singer, I. M. (Isadore Ma 010101 000
Lecture notes on elementary to
II 1 IIIMlII 1 HI
0 1163 01276 30 8
TRENT UNIVERSITY
<
QA611 .S498 1976

Singer, Isadore Manuel, 1924-
Lecture notes on elementary
topology and geometry.
DATE
ISSUED TO j (.1
From the Preface:
At the present time, the average undergraduate mathematics

major finds mathematics heavily compartmentalized. After the
calculus, he takes a course in analysis and a course in algebra.
Depending upon his interests (or those of his department), he
takes courses in special topics. If he is exposed to topology, it is
usually straightforward point set topology; if he is exposed to
geometry, it is usually classical differential geometry. The
exciting revelations that there is some unity in mathematics, that
fields overlap, that techniques of one field have applications in
another, are denied the undergraduate. He must wait until he is
well into graduate work to see interconnections, presumably
because earlier he doesn’t know enough.
These notes are an attempt to break up this compartmentali-
zation, at least in topology-geometry. What the student has
‘ «
learned in algebra and advanced calculus are used to prove

some fairly deep results relating geometry, topology, and group
theory. (De Rham’s theorem, the Gauss—Bonnet theorem for
surfaces, the functorial relation of fundamental group to
covering space, and surfaces of constant curvature as
homogeneous spaces are the most noteworthy examples.)
ISBN 0-387-90202-3 Springer-Verlag • New York • Heidelberg • Berlin

ISBN 3-540-90202-3 Springer-Verlag • Berlin • Heidelberg • New York

Lecture Notes On Elementary Topology and Geometry-I. M. Singer J. A. Thorpe-Springer

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Lecture Notes On Elementary Topology and Geometry-I. M. Singer J. A. Thorpe-Springer

Uploaded by

Copyright:

Available Formats

Undergraduate Texts in Mathematics

R2 K=0 Euclidean motions Complex plane Euclidean

D K <0 Conformal group Disc (hyperbolic Hyperbolic

Springer-Verlag New York • Heidelberg • Berlin

THOMAS J. BATA LIBRARY

AMS Subject Classifications: 50-01, 53-01, 54-01

Library of Congress Cataloging in Publication Data

Singer, Isadore Manuel, 1924-

All rights reserved.

© 1967 by I. M. Singer and John A. Thorpe.

Printed in the United States of America.

ISBN 0-387-90202-3 Springer-Verlag New York

ISBN 3-540-90202-3 Springer-Verlag Berlin Heidelberg

At the present time, the average undergraduate mathematics major finds

Some point set topology 1

More point set topology 26

Fundamental group and covering spaces 49

Homology theory and the De Rham theory 153

Intrinsic Riemannian geometry of surfaces 175

Imbedded manifolds in R3 216

1.1 Naive set theory

Definition. Let A and B be sets. A is a subset of B, written A B, if x e A

The intersection A n B of A and B is defined by

A n B = [x; x e A and x e B],

Similarly, if LA is a set (collection) of sets, the union and intersection of all

S — [x; x e S for some S e SA],

[^| S = [x; x e S for every S e

If A ^ B, the complement of A in B, denoted A' or B — A, is defined by

Theorem 1. Let A, B, C, and S be sets. Then

(10) For <5^ as in (9),

(u ^W u $») = u osin so.

Proof. The proof of this theorem is left to the student.

Definition. Let A and B be sets. The Cartesian product A x B of A and B is the

A x B = [(a, b); a e A, b e B],

A relation between A and B is a subset R of A x B. a and b are said to be

Example. Let A = B = the set of real numbers. Then A x B is the plane.

Definition. A relation R <= A x A is a partial ordering if

(1) (si, s2) 6 R and (.y2, s3) e R ^ (su s3) e R and

A relation R is a simple ordering if it is a partial ordering, and, in addition:

The order relation for S = real numbers is an example of a simple ordering.

Definition. Let A and B be sets. A function f mapping A to B, denoted

(1) If ae A, then there exists b e B such that (a, b) e/.

Property (1) says that the function/is defined everywhere on A. Property

Notation. hQtf:A->B. By /(a) = b we mean (a, b) e f

Definition. Let f\ AB. f is surjective (onto) if for each b e B there exists

Definition. A set A is countable if there exists a one-to-one correspondence

f(B): A -> [YES, NO])

f is injective because if f(B) = /(C), then f(B)(x) = f(C)(x) for all x e A.

B - [x; g(x) = YES]

Notation. Motivated by this proposition, we denote by 2A the set of all

= [a e A ;f(a) e Bx] (Bx <= B).

f~1{Bj) is called the inverse image of Bx. Note that

Notation. Let W be a set, and let FF be a collection of sets. We say FF is

indexed by W if there is given a surjective function <j>\ IVST. For w e IV,

Definition. Let be indexed by W. The product of the sets {Sw}weW is

ns.- Sw;f(w) g Sw for all we w\.

Axiom of choice. Let {S'U)}lt,eW be sets indexed by W. Assume Sw =£ 0 for all

The axiom of choice is equivalent to several other axioms, one of which is

Maximum principle. If S is partially ordered by R, and T is a simply ordered

1.2 Topological spaces

BSo(a) = [seS; P(s, s0) < a\.

Pi(Pi,P2) = V(x2 - xx)2 + (y2 - y-d2,

Pi(s,0) < a p2(s,0) < a P3(s,0) < a