TOPOLOGY

FROM THE
DIFFERENTIABLE
VIEWPOINT

By John W. Jtlilnor
Princeton University

Based on notes by

David W. Weaver

The University Press of Virginia

Charlottesville
 Q

Copyright © 1965
by the Rector and Visitors
of the l�niversity of Virginia

The U ni versity Press of Virginia

First published 1965

Library of Congress
Catalog Card Number : 65-26874
Printed in the
United States of America
To Heinz Hop!
 PREFACE

THESE lectures were delivered at the University of Virginia in December

1963 under the sponsorship of the Page-Barbour Lecture Foundation.
They present some topics from the beginnings of topology, centering
about L. E. J. Brouwer's definition, in 1912, of the degree of a mapping.
The methods used, however, are those of differential topology, rather
than the combinatorial methods of Brouwer. The concept of regular
value and the theorem of Sard and Brown, which asserts that every
smooth mapping has regular values, play a central role.
To simplify the presentation, all manifolds are taken to be infinitely
differentiable and to be explicitly embedded in euclidean space. A small
amount of point-set topology and of real variable theory is taken for
granted.
I would like here to express my gratitude to David Weaver, whose
untimely death has saddened us all. His excellent set of notes made this
manuscript possible.
J.W.M.
Princeton, New Jersey
March 1965

, .

vii
 CONTENTS

Preface VB

l. Smooth manifolds and smooth maps I

Tangent spaces and derivatives 2
Regular values 7
The fundamental theorem of algebra 8
2. The theorem o f Sard and Brown 10
Manifolds with boundary 12
The Brouwer fixed point theorem 13
3. Proof of Sard's theorem 16
4. The degree modulo 2 of a mapping 20
Smooth homotopy and smooth isotopy 20
�
D. Oriented manifolds 26
The Brouwer degree 27
6. Vector fields and the Euler number 32
7. Framed cobordism; the Pontryagin construction 42
The Hopf theorem .50
8. Exercises 52
Appendix : Classifying I-manifolds 55
Bibliography 59
Index 63

IX
TOPOLOGY FROM

THE DIFFERENTIABLE

VIEWPOINT
 §1. SMOOTH MANIFOLDS

AND S MOOTH MAPS

FIRST let us explain some of our terms. Rk denotes the k-dimensional

euclidean space; thus a point x r Rk is an k-tuple x ( Xl) . . . ) Xk ) of
=

real numbers.
Let U C Rk and VeRI be open sets. A mapping f from U to V
(written t : U -7 V) is called smooth if all of the partial derivatives
anf/axi, ... ax,,, exist and are continuous.
More generally let X C Rk and Y C Rl be arbitrary subsets of
euclidean spaces. A map f : X -7 Y is called smooth jf for each x r X
there exist an open set U C Rk containing x and a smooth mapping
F : U -7 Rl that coincides with f throughout un X.
If t : X -7 Y and g : Y -7 Z are smooth, note that the composition
got : X -7 Z is also smooth . The identity map of any set X is auto­
matically smooth.
D EFI�ITIOX. A map f : X -7 Y is called a diffeomorphism if f carries X
homeomorphically onto Y and if both t and rl are smooth.
We can now indicate roughly what d�fferential topology is about by
saying that it studies those properties of a set XC Rk which are invariant
under diffeomorphism.
vVe do not, however, want to look at completely arbitrary sets X.
The following definition singles out a particularly attractive and useful
class.
DEFINITIOX. A subset Me Rk is called a smooth manifold of dimension
m if each x r ilJ has a neighborhood W n ilJ that is diffeomorphic to
an open subset U of the euclidean space R"'.
Any particular diffeomorphism g : U -7 W n il1 is called a para­
metrization of the region W n M. (The inverse diffeomorphism
W n M -7 U is called a system of coordinates on W n M.)
 2 §1. Smooth manifolds

Figure 1. Parametrization of a region in M

Sometimes we will need to look at manifolds of dimension zero . By

definition, M is a manifold of dimension zero if each x E M has a neigh­
borhood W n M consisting of x alone.
EXAMPLES . The unit sphere S2, consisting of all (x, y, z) E R3 with
x2 + y2 + l = 1 is a smooth manifold of dimension 2. In fact the
diffeomorphism
(x, y) -'> (x , y, VI - x2 _ y2),
for x2 + y2 < I, parametrizes the region z > 0 of S2. By interchanging
the roles of x, y, z, and changing the signs of the variables, we obtain
similar parametrizations of the regions x > 0, y > 0, x < 0, y < 0,
and z < O. Since these cover S2, it follows that S2 is a smooth manifold.
More generally the sphere sn-l C Rn consisting of all ( XI' . . , xn) .

with L x: = 1 is a smooth manifold of dimension n 1. For example

-

SO C RI is a manifold consisting of j ust two points.

A somewhat wilder example of a smooth manifold is given by the
set of all (x, y) E R2 with x T" 0 and y = sin (l/x) .

TANGENT SPACES AND DERIVATIVES

To define the notion of derivative dfx for a smooth map f : M -'> N

of smooth manifolds, we first associate with each x E M C Rk a linear
subspace TMx C Rk of dimension rn called the ta ngen t space of jlJ at x.
Then dfx will be a linear mapping from TMx to TNy, where y = f (x) .
Elements of the vector space TMx are called ta ngent vectors to jlJ at x.
Intuitively one thinks of the rn-dimensional hyperplane in Rk which
best approximates M near x; then TMx is the hyperplane through the
Tangent spaces 3

origin that is parallel to this. (Compare Figures 1 and 2.) Similarly

one thinks of the nonhomogeneous linear mapping from the tangent
hyperplane at x to the tangent hyperplane at y which best approxi­
mates f. Translating both hyperplanes to the origin, one obtains dfx .
Before giving the actual definition, we must study the special case
of mappings between open sets. For any open set U C Rk the tangent
space TUx is defined to be the entire vector space Rk. For any smooth
map f: U ---7 V the derivative
dfx : Rk ---7 Rl

is defined by the formula

dfx(h) = lim (f(x + th) - f(x))/ t
t�O

for x E U, h E Rk. Clearly dfx (h) is a linear function of h. (In fact dfx
is just that linear mapping which corresponds to the l X k matrix
(afj ax;). of first partial derivatives, evaluated at x.)
Here are two fundamental properties of the derivative operation:
1 (Chain rule) . If f : U ---7 V and g : V ---7 W are smooth maps, with
f (x) = y, then
d(g 0 f) x = dgy 0 dfx .
In other words, to every commutative triangle

of smooth maps between open subsets of R\ Rl, Rm there corresponds

a commutative triangle of linear maps

2. If I is the identity map of U, then dIx is the identity map of Rk.

M ore genera lly, if U C U' a re open sets and
i : U ---7 U'
4 §1. Smooth manifolds

is the inclusion map, then again di, is the identity map of Rk .

Kote also :
3. If L : Rk --7 Rl is a linea r ma pping, then dLx = L.
As a simple application of the two properties one has the following :

ASSEHTIO�. If f is a diiJeomorphism between open sets U C Rk and

VCR I, then k must equa l l, and the linea r mapping
dlx : Rk --7Rl
must be nonsingula r.

PIWOF . The composition rl 0 I is the identity map of U; hence

d W I ) y 0 dt is the identity map of Rk. Similarly dt 0 d W I ) y is the identity
� x

map of Rl . Thus dl x has a two-sided inverse, and it follows that k = l.

A partial converse to this assertion is valid. Let t : U --7 Rk be a
smooth map, with U open in Rk .

Inverse Function Theorem. If the deriva tive dl x : Rk --7 Rk is non­

singula r, then I maps a ny su.ffi ciently sma ll open set U' a bout x diiJe 01n o r­
phica lly on to an open set t(U') .
(See Apostol [2, p. 144] or Dieudonne [7, p. 2G8]. )

Note that I may not be one-one i n the large, even if every d/r is
nonsingular. (An instructive example is provided by the exponential
mapping of the complex plane into itself. )

Now let u s define the tangent space TMx for a n arbitrary smooth
manifold M C Rk . Choose a parametrization
g:U--7M C Rk
of a neighborhood g (U) of x in M, with g (u ) = x. Here U is an open
subset of Rm. Think of g as a mapping from U to Rk , so that the derivative
dgu : R'"--7Rk
is defined . Set TJl!fx equa l to the image dgu (R'") ot dgu• (Compare Figure 1.)
We must prove that this construction does not depend on the par­
ticular choice of parametrization g. Let h : V --7 M C Rk be another
parametrization of a neighborhood h (V) of x in M, and let v = h- I (x).
Then h- I 0 g maps some neighborhood UI of u diffeomorphically onto
a neighborhood V of v. The commutative diagram of smooth maps
I
ngent spaces 5

,ween open sets

es rise to a commutative diagram of linear maps

��\a;
Rk

- .
d

Rm )R'"',
d(h-1 0 g)"
l it follows immediately that

Image (dgu) = Image (dh,) .

1S TMx is well defined .
'ROOF THAT TMx IS AN m-DIMENSIONAL VECTOH SPACE. Since
g-1 : g(U) �U
, smooth mapping, we can choose an open set W containing x and
mooth map F : W � Rm that coincides with g-l on W n g(U).
ting Uo = g-I(W n g(U», we have the commutative diagram

g /W� F
(
U inclusion )Rm
therefore

s diagram clearly implies that dg" has rank 117, and hence that its
ge TMx has dimension 117.
row consider two smooth manifolds, M C Rk and N C Rl, and a
6 § 1 . Smooth manifolds

smooth map

with f(x) y. The derivative

dix : TMx --7 TN.
is defined as follows . Since f is smooth there exist an open set W con­
taining x and a smooth map
F : W--7 Rl
that coincides with f on W n M. D efine dix(v) to be equal to dFx (v)
for all v r TMx.
To j ustify this definition we must prove that dFx(v) belongs to TN.
and that it does not depend on the particular choice of F.
Choose parametrizations
g : U --7 M C Rk and h: V --7 N C Rl

for neighborhoods g(U) of x and hey) of y. Replacing U by a smaller

set if necessary, we may assume that g(U) C Wand that i maps g(U)
into h(V). It follows that
h- 1 0 f og : U --7 V
is a well-defined smooth mapping.
Consider the commutative diagram
P
W. )R1

9 r
U
h-' 0 log r
)V
h

of smooth mappings between open sets. Taking derivatives, we obtain

a commutative diagram of linear mappings

where u = g-1(X), v = h- 1 (y) .

It follows immediately that dPx carries TMx = Image (dgu) into
TNy = Image (dh ) . Furthermore the resulting map dfx does not
.

depend on the particular choice of P, for we can obtain the same linear
R egular values 7

transformation by going around the bottom of the diagram. That is:

dtx = dh, 0 d(h lo t 0 g)U 0 (dgut 1 .
-

This completes the proof that

dtx : TMx ----> TNy
is a well-defined linear mapping.
As before, the derivative operation has two fundamental properties :
1 . (Chain rule) . It t : M ----> N and g : N ----> P are smooth, with t (x) = y,
t hen
d(g 0 t)x = dgy 0 dtx'
2. It I is the identity map ot M , then dIx is the identity map of TMx.
iVI ore genera lly, if MeN with inclusion map i, then TMx C TNx with
inclusion map dix. (Compare Figure 2.)

M
™x

TNx

Figure 2. The tangent space of a 8ubrnanifold

The proofs are straightforward .

As before, these two properties lead to the following :
AssEn'rIO�. It t : M --'t N is a diff eomorphism, then dfx : TMx ----> TN"
i s an isomorphism of vector spaces. In particu la r t he dirnension of AI
must be equa l t o the dimension of N.

REGULAR VALUES

Let f : M ----> N be a smooth map between manifolds of the same

dimension.* We say that x r M is a regular point of f if the derivative

* This restriction will be removed in §2.

8 §1. Smooth manifolds
dfx is nonsingular. In this case it follows from the inverse function
theorem that f maps a neighborhood of x in M diffeomorphically onto
an open set in N. The point Y E N is called a r egular val ue if r I (y)
contains only regular points .
If dtx is singular, then x is called a critical point of t, and the image
f(x) is called a critica l value. Thus each Y EN is either a critical value or a
regular value according as r I(y) does or does not contain a critical point .
O bserve that it ]J;f is compact and Y E N is a regula r va lue, then rI (y)
is a finite set (possibly empty) . For r ' (y) is in any case compact, being
a closed subset of the compact space M; and r ' (y) is discrete, since t
is one-one in a neighborhood of each x E r ' (y) .
For a smooth t : M ----7 N, with M compact, and a regular value y r N,
we define #r' (y) t o be the number of point s in r ' (y) . The first observation
to be made about #r' (y) is that it is locally constant as a function of y
(where y ranges only through regular values!) . I . e . , there is a neighbor­
hood V C N ot y such tha t #r' (y') = #r'(y) tor any y' r V. [Let x" . . . ,Xk
be the points of r ' (y) , and choose pairwise disj oint neighborhoods
U" . . . , Uk of these which are mapped diffeomorphic ally onto neighbor­
hoods V ' . . . , Vk in N. We may then take
I
V = V, n V2 n .. . n Vk - t (M - U, - ... - Uk).]

THE FUNDAMENTAL THEOREM OF ALGEBRA

As an application of these notions, we prove the fundamental theorem

of algebra : every nonconstant complex polynomial P (z) must have a zero.
For the proof it is first necessary to pass from the plane of complex
numbers to a compact manifold . Consider the unit sphere S 2 C R3 and
the stereographic proj ection
h+ : S2 - {CO, 0, I)} ----7R 2 X ° C R3
from the "north pole" (0, 0, 1) of S 2 . (See Figure 3.) We will identify
R 2 X ° with the plane of complex numbers . The polynomial map P from
R 2 X ° itself corresponds to a map t from S 2 to itself; where
f(x) = h� 'Ph+ (x) for x � (0, 0, 1)
teO, 0 , 1) = (0, 0, 1).
It is well known that this resulting map t is smooth, even in a neighbor-
Fundamental theor em of algebra 9

(0,0,1)

Figure 3. Stereographic projection

hood of the north pole. To see this we introduce the stereographic

proj ection h_ from the south pole (0, 0, - 1 ) and set

Note, by elementary geometry, that

h+h=\ z) = zl lz l 2 = liz.
Now if P (z ) = aozn + a l zn- 1 + . . . + an, with ao � 0, then a short
computation shows that
Q (z) = znl (a o + a lz + . . . + an zn) .
Thus Q is smooth in a neighborhood of 0, and it follows that f = h=lQ h_
is smooth in a neighborhood of (0, 0, 1 ) .
Next observe that f has only a finite number of critical points; for P
fails to be a local diffeomorphism only at the zeros of the derivative
polynomial P' (z) = L an-i jzi - " and there are only finitely many
zeros since P' is not identically zero . The set of regular values of t,
being a sphere with finitely many points removed, is therefore connected .
Hence the locally constant function #r\ y) must actually be constant
on this set . Since #rl (y) can't be zero everywhere, we conclude that
it is zero nowhere. Thus f is an onto mapping, and the polynomial P
must have a zero .
 §2. THE THEORE M

OF SARD AND B ROWN

IN GENERAL, it is too much to hope that the set of critical values of a

smooth map be finite. But this set will be "small, " in the sense indicated
by the next theorem, which was proved by A. Sard in 1942 following
earlier work by A. P. Morse. (References [30], [24] .)

Theorem. Let f : U � Rn be a smooth map, defined on an open set

U C Rm, and let
C =
(x£ U I rank dfx < n } .
Then t he image f(C) C Rn has Lebesgue measure zero. *
Since a set of measure zero cannot contain any nonvacuous open set,
it follows that the complement Rn f(C) must be everywhere denset
-

in R".
The proof will be given in §3. It is essential for the proof that f should
have many derivatives. (Compare Whitney [38] . )
W e will b e mainly interested i n the case m � n. I f m < n, then
clearly C = U; hence the theorem says simply that f ( U) has measure
zero .

More generally consider a smooth map f : M. � N, from a manifold

of dimension m to a manifold of dimension n. Let C be the set of all
x £ M such that
dfx : TM x � TNf( x )

* In other words, given any E > 0, it is possible to cover fCC) by a sequence of

cubes in Rn having total n-dimensional volume less than E.
t Proved by Arthur B. Brown in 1 935. This result was rediscovered by Dubovickil
in 1953 and by Thorn in 1954. (References [5], [8], [36].)
Regular va lues 11

has rank less than n (i.e. is not onto) . Then C will be called the set
of critica l points, f CC) the set of critica l va lues, and the complement
N - fCC) the set of regula r va lu es of f. CThis agrees with our previous
definitions in the casem = n. ) Since M can be covered by a countable
collection of neighborhoods each diffeomorphic to an open subset of
Rm, we have :

Corollary (A. B . Brown) . The set of regula r va lues of a smooth map

t : M. -7 N i s everywhere dense in N.
In order to exploit this corollary we will need the following :

Lemma 1. If f : M -7 N is a smooth map between manif olds of dimen­

sion m � n, and if y £ N is a regula r va lue, then the set r 1 (y) C M is a
smooth manif old of dimension m - n.
PROOl<'. Let x £ r 1 (y ) . Since y is a regular value, the derivative dfx
must map TMx onto TN•. The null space iJ( C TM.x of dfx will therefore
he an (m - n)-dimensional vector space .
If M C R\ choose a linear map L : Rk -7 Rm-n that is nonsingular
on this subspace iJ( C TMx C Rk . Now define
F : M -7 N X Rm -n
by F (� ) (f (� ) , LCo)· The derivative dFx is clearly given by the formula
dFx(v) = (dfx (v) , L (v» .
Thus dF x is nonsingular. Hence F maps some neighborhood U of x
diffeomorphically onto a neighborhood V of (y, L (x» . No te that r 1 (y)
corresponds, under F, to the hyperplane y X Rm -n• In fact F maps
r 1 (y) n U diffeomorphically onto (y X Rm -n) n V. This proves that
r I (y) is a smooth manifold of dimension m - n.
-1
As an example we can give an easy proof that the unit sphere 8m
is a smooth manifold. Consider the function f : Rm -7 R defined by
f(x) = xi + x� + . . . + x ; .
Any y r£ 0 is a regular value, and the smooth manifold rICl) is the
unit sphere.
If M' is a manifold which is contained in M, it has already been
noted that TJJ![� is a subspace of TMx for x £ M'. The orthogonal com­
plement of TM: in TM.x is then a vector space of dimension m - m'
called the spa ce of normal vect ors to M' in M. a t x.
In particular let M' = r 1 (y) for a regular value y of f : M -7 N.
 y
12 §2. Sard-Brown theorem

Lemma 2. The null spa ce of dfx : TlVI x � TN is precis ely equa l t o

the tangent spa ce TM.� C TM.x of t he submanifold M.' = r 1 (y) . H ence
dfx maps the orthogona l complement of TlVI� isomorphica ll y onto TNv•
PROOF. From the diagram
i
lVI' �M

1
y
1
f
�N l
we see that dfx maps the subspace TlVI; C TlVI x to zero . Counting
dimensions we see that dfx maps the spa ce of norma l vectors t o lVI ' isomor­
phica lly onto TNv•

MANIFOLD S WITH BOUNDARY

The lemmas above can be sharpened so as to apply to a map defined

on a smooth "manifold with boundary . " Consider first the closed
half-space
=

m
Hm { (X l ' . . . , X ) r Rm I X � 0) .
m m
The bounda ry aH is defined to be the hyperplane Rm-1 X 0 C Rm.
DEFINITION. A subset X C Rk is called a smooth m- manifold wit h
bounda ry i f each X r X has a neighborhood U n X diffeomorphic to
an open subset V n Hm of Hm . The bounda ry ax is the set of all points
in X which correspond to points of aHm under such a diffeomorphism.
It is no t hard to show that ax is a well-defined smooth manifold
of dimension m - 1 . The interior X ax is a smooth manifold of
-

dimension m.
The tangent space TXx is defined j ust as in §1, so that TXx is a full
m-dimensional vector space, even if X is a boundary point.
Here is one method for generating examples. Let lVI be a manifold
without boundary and let g : lVI � R have 0 as regular value.

Lemma 3. The set of X in lVI with g(x) � 0 is a smooth manifold, wi th

bounda ry equa l to g- l (O) .
The proof is j ust like the proof of Lemma 1.
Brouwer fixed point theorem 13

EXAMPLE. The uni t dis k Dm, consisting of all x t Rm with

1 - L: x� � 0,

i s a. smooth manifold, with boundary equal to 8m- I •

Now consider a smooth map f : X � N from an m-manifold with
boundary to an n-manifold, where m > n .

Lemma 4. If y £ N is a regula r va lue, both for f a nd for the res triction

f I ax, then rl(y) C X is a smooth (m - n)-ma nifold wi th bounda ry.
Furthermore the bounda ry a(f-l(y) ) is precisely equal to the intersection
of rl(y) wi th ax .
PROOF. Since we have to prove a local property, it suffices to consider
the special case of a map f : Hm � Rn, with regular value y t Rn. Let
x t r\y) . If x is an interior point, then as before rl(y) is a smooth
manifold in the neighborhood of x.
Suppose that x is a boundary point. Choose a smooth map g : U � Rn
that is defined throughout a neighborhood of x in Rm and coincides with
f on u n Hm. Replacing U by a smaller neighborhood if necessary, we
may assume that g has no critical points. Hence g-l(y) is a smooth
manifold of dimension m - n.
Let rr : g-l(y) � R denote the coordinate proj ection,

We claim that rr has ° as a regular value. For the tangent space of

g-l(y) at a point x l
t rr- (O) is equal to the null space of

dgx = dfx : Rm � Rn;

but the hypothesis that f I aHm is regular at x guarantees that this
null space cannot be completely contained in Rm- 1 X 0 .
Therefore the set g-l(y) n Hm = rl(y) n U, consisting o f all x £ g-l(y)
with rr(x ) � 0, is a smooth manifold, by Lemma 3; with boundary
equal to rr-l (O) . This completes the proof .

THE BROUWER FIXED POINT THEOREM

We now apply this result to prove the key lemma leading to the
classical Brouwer fixed point theorem. Let X be a compact manifold
with boundary.
14 §2. Sard-Brown theorem

Lemma 5. There is no smooth map f : X � a X that leaves a X point­

wise fi xed.
PROOF (following M. Hirsch) . Suppose there were such a map f.
Let y r ax be a regular value for f. Since ?J is certainly a regular value
for the identity map f I ax also, it follows that rl(y) is a smooth 1-
manifold, with boundary consisting of the single point

But rl(y) is also compact, and the only compact I-manifolds are finite
disj oint unions of circles and segments, * so that arl(y) must consist of
an even number of points. This contradiction establishes the lemma .
In particular the unit disk
Dn = I x r W I x� + ... + x� � 11
is a compact manifold bounded by the unit sphere sn-I. Hence as a
special case we have proved that the identity ma p of sn-I canno t be ex­
t ended to a smoo th ma p Dn � S,,- I .

Lemma 6. A ny sm ooth ma p g : Dn � Dn ha s a fixed point (i .e. a point

x r Dn with g (x) = x) .

PRO O F . Suppose g has no fixed point. For x r D", let f (x) r sn - I be

the point nearer x on the line through x and g(x) . (See Figure 4.) Then
f : Dn � sn- I is a smooth map with f (x) = x for x r sn-\ which is
impossible by Lemma 5. (To see that f is smooth we make the following
explicit computation : f (x) = x + tu, where
x - g(x)
= t = -x · u + VI - X · X + (X ' U) 2 ,
u Il x - g(x)11 '
the expression under the square root sign being strictly positive . Here
and subsequently I lx l l denotes the euclidean length V x� + ... + x�. )

Brouwer Fixed Point Theorem. A ny continuous f unction G : Dn � Dn

ha s a fixed point .
PROO F. We reduce this theorem to the lemma by approximating G
by a smooth mapping. Given f > 0, according to the Weierstrass
approximation theorem, t there is a polynomial function PI : R" � Rn
with I IP (x) - G(x)11 < f for x r D". However, PI may send points
I
* A proof is given in the Appendix.
t See for example Dieudonne [7, p. 133].
Brouwer fixed point theorem 15

g(x)
x

Figure 4

of Dn into points outside of Dn . To correct this we set

P(x) = P , (x) / ( l + €) .
Then clearlyP maps Dn into Dn and I IP (x) - G(x)11 < 2€for x r Dn.
Suppose that G(x) ,t: x for all x r Dn. Then the continuous function
I IG(x) - xii must take on a minimum j.L > 0 on Dn. ChoosingP : Dn ---'> Dn
as above, with I I P (x) - G(x)11 < j.L for all x, we clearly have P (x) ,t: x.
Thus P is a smooth map from Dn to itself without a fixed point . This
contradicts Lemma 6, and completes the proof.
The procedure employed here can frequently be applied in more
general situations : to prove a proposition about continuous mappings,
we first establish the result for smooth mappings and then try to use an
approximation theorem to pass to the continuous case . (Compare §8,
Problem 4 . )
 §3. PROOF OF

SARD'S THEORE M*

FIRST let us recall the statement :

Theorem of Sardo Let f: U -t RP be a smooth ma p, with U open in Rn,
a nd l et C be the set of cri tica l points; tha t is the set of a ll x E U wi th
rank dfx < p .
Then f(C) C R P has mea sure zero.
REMARK. The cases where n ::::; p are comparatively easy. (Compare
de Rham [29, p. 10] .) We will, however, give a unified proof which
makes these cases look j ust as bad as the others.
The proof will be by induction on n. Note that the statement makes
sense for n 2: 0, p 2: 1 . (By definition R O consists of a single point .)
To start the induction, the theorem is certainly true for n = 0.
Let C1 C C denote the set of all x E U such that the first derivative
dfx is zero . More generally let Ci denote the set of x such that all partial
derivatives of f of order ::::; i vanish at x. Thus we have a descending
sequence of closed sets
C � C1 � C2 � C3 � • • • •

The proof will be divided into three steps as follows :

STEP 1. The image fCC - C1) has measure zero .
STEP 2 . The image f(Ci - Ci+1) has measure zero, for i 2: 1 .
STEP 3. The image f(Ck) has measure zero for k sufficiently large.
(REMARK. If f happens to be real analytic, then the intersection of

* Our proof is based on that given by Pontryagin [28]. The details are somewhat
easier since we assume that f is infinitely differentiable.
Step 1 17

the C i is vacuous unless t is constant on an entire component of U .

Hence in this case it is sufficient to carry out Steps 1 and 2 . )
PRO O F O F STEP 1 . This first step i s perhaps the hardest . We may
assume that p � 2, since C = CI when p = 1 . We will need the well
known theorem of Fubini* which asserts that a mea sura ble set
A C RP = Rl X RP-I
must have mea sure zero if it intersects ea ch hyperplane (constant) X RP-1
in a set of (p - I )-dimensiona l mea sure zero.
For each x E C - CI we will find an open neighborhood Ve Rn
so that t(V n C) has measure zero . Since C - C1 is covered by countably
many of these neighborhoods, this will prove that fCC - CI) has measure
zero .
Since x ¢ CI, there is some partial derivative, say afl /ax1, which
is not zero at x. Consider the map h : U � Rn defined by
h ex) = C/J ex) , X2 , • • • , xn).
Since dhi; is nonsingular, h maps some neighborhood V of x diffeomor­
phically onto an open set V'. The composition g = f 0 h- l will then
map V' into RP• Note that the set C' of critical points of g is precisely
h ( V n C); hence the set g (C') of critical values of g is equal to t ( V n C).

Figure 5. Construction of the map g

* For an easy proof (as well as an alternative proof of Sard's theorem) see Stern­
berg [35, pp. 51-52]. Sternberg assumes that A is compact, but the general case
follows easily from this special case.
18 §3. Proof of Sard's theorem

For each (t, X2, . . . , xn ) {: V' note that g et , X2, • • • , Xn) belongs to
the hyperplane t X R P-1 C R P: thus g carries hyperplanes into hyper­
planes. Let
I
g' : (t X R n- 1 ) n V' � t X W-
denote the restriction of g. Note that a point of t X Rn-I is critical for
g' if and only if it is critical for g; for the matrix of first derivatives of g
has the form

According to the induction hypothesis, the set of critical values of g'

has measure zero in t X R P -I• Therefore the set of critical values of g
intersects each hyperplane t X R P -1 in a se t of measure zero . This
set g(C') is measurable, since it can be expressed as a countable union
of compact subsets. Hence, by Fubini's theorem, the set
g(C') = f( V n C)
has measure zero, and Step 1 is complete.
PROOF O F STEP 2. For each x {: Ck - Ck+I there is some (k + 1)-8'
k
derivative a +1fr/ax" ...ax'kH which is not zero . Thus the function
• . •
k
w(x) = a tr/axs, axm•
vanishes at x but aw/ax,. does not . Suppose for definiteness that s, = l.
Then the map h : U � Rn defined by
hex) = (w(x), X2, ... ,x)
n
carries some neighborhood V of x diffeomorphically onto an open set V'.
Note that h carnes Ck n V into the hyperplane 0 X W-1• Again
we consider

Let

denote the restriction of g. By induction, the set of critical values of g

has measure zero in RP• But each point in h(Ck n V) is certainly a
critical point of g (since all derivatives of order :;'k vanish) . Therefore
gh(Ck n V) = t(Ck n V) has measure zero.
Since Ck - CHI is covered by countably many such sets V, it follows
that teCk -- Ck+I) has measure zero.
Step 3 19

PRO O F OF STEP 3. Let r CUbe a cube with edge o. If k is sufficiently

large (k > nip - 1 to be precise) we will prove that t(Ck n r) has
measure zero . Since Ck can be covered by countably many such cubes,
this will prove that t(Ck) has measure zero .
From Taylor's theorem, the compactness of r, and the definition
of Ck, we see that
f(x + h) = t(x) + R(x, h)
where
k
1) IIR(x, h)II :::;: c Ilh Il +1
for x £ Ck n r, x + h £ r. Here c is a constant which depends only
on t and I" . Kow subdivide r into rn cubes of edge ol r. Let II be a cube
of the subdivision which contains a point x of Ck• Then any point of II
can be written as x + h, with
2) Ilhll :::;: y:;;(olr).
From 1) it follows that t(11) lies in a cube of edge all+1 centered
about t(x), where a = 2c (y:;; 0/+1 is constant . Hence t(Ck n r) is
n
contained in a union of at most r cubes having total volume

If k + 1 > nip, then evidently V tends to 0 as r � OJ; so f(Ck n J")

must have measure zero . This completes the proof of Sard's theorem.
 ,..

§4. THE DEGREE MODU LO 2

OF A MAPPING

CONSIDER a smooth map f : S' -+ sn. If y is a regular value, recall that

#r 1 (y) denotes the number of solutions x to the equation f (x) = y.
We will prove that the residue cla ss modulo 2 of #rl (y) does not depend on
the choice of the regula r va lue y. This residue class is called the mod 2
degree of f. More generally this same definition works for any smooth
map
f :M-+N
where M is compact without boundary, N is connected, and both
manifolds have the same dimension. (We may as well assume also that N
is compact without boundary, since otherwise the mod 2 degree would
necessarily be zero .) For the proof we introduce two new concepts.

SMOOTH HOMOTOPY AND SMOOTH ISOTOPY

Given X C Rk, let X X [0, 1] denote the subset* of Rk+l consisting

of all (x, t) with x E X and 0 ::; t ::; 1 . Two mappings

f,g :X-+ Y
are called smoothly homotopi c (abbreviated f rv g) if there exists a

* If M is a smooth manifold without boundary, then M X [0, 1] is a smooth

manifold bounded by two "copies" of M. Boundary points of M will give rise to
"corner" points of M X [0, 1].
H omotopy a nd isotopy 21

smooth map F : X X [0, 1] � Y with

F (x , O ) = f(x) , F(x , 1) = g(x)

for all x £ X. This map F is called a smooth homotopy between f and g.
Note that the relation of smooth homotopy is an equivalence relation.
To see that it is transitive we use the existence of a smooth function
<p : [0, 1] � [0, 1] with

<p (t) = ° for 0::::;; t ::::;; t

<p (t) = 1 for i::::;; t ::::;; 1.
(For example, let <p (t) = A (t - t)/(A(t - t) + A(i - t)), where AH = °
for 7 ::::;; ° and A(7) = exp ( _7-1 ) for 7 > 0.) Given a smooth homotopy F
between f and g, the formula G(x, t) = F (x, <p ( t ) ) defines a smooth
homotopy G with
G(x, t) = f(x) for 0::::;; t ::::;; t
G(x, t) = g(x) for i::::;; t ::::;; 1 .
Now i f f "" g and g "" h, then, with the aid o f this construction, it i s
easy t o prove that f "" h.
If f and g happen to be diffeomorphisms from X to Y, we can also
define the concept of a "smooth isotopy" between f and g. This also
will be an equivalence relation.

DE FINITION . The diffeomorphism f is smo othly isoto pi c to g if there

exists a smooth homotopy F : X X [0, 1] � Y from f to g so that,
for each t £ [0, 1 ] , the correspondence
x � F(x , t)
maps X diffeomorphically onto Y.

It will turn out that the mod 2 degree of a map depends only on its
smooth homotopy class :

Homotopy Lemma. Le t j, g : M � N be sm oothl y hom otopic maps

between manifolds of the same dimension, where M is compa ct and without
bounda ry. If y £ N is a regula r va lue for both f and g, then

PROO F . Let F : M X [0, 11 � N be a smooth homotopy between

f and g. First suppose that y is also a regular value for F. Then F- 1 (y)
22 §4 . Degr ee mo dulo 2

is a compact I-manifold, with boundary equal to

rl(y) n (NI X 0 U M X 1) = rl(y) X 0 U g-I(y) X 1.
Thus the total number of boundary points of rl(y) is equal to

#rl(y) + #g-I(y).
But we recall from §2 that a compact I-manifold always has an even
number of boundary points . Thus #r\y) + #g-I(y) is even, and
therefore

(�
MxO Mxl

Figure 6. The number of boundary points on the left is congruent to the number on the
right modulo 2

Now suppose that y is not a regular value of F. Recall (from §I)

that #rl(y') and #g-l(y') are locally constant functions of y' (as long
as we stay away from critical values) . Thus there is a neighborhood
VI C N of y, consisting of regular values of f, so that

for all y' {: VI ; and there is an analogous neighborhood V2 C N so that

'
for all y {: V2• Choose a regular value z of F within VI n V2• Then

which completes the proof.

We will also need the following :

Homogeneity Lemma. Let y and z be a rbit ra ry int erior points of the

smooth, connected manifold N . Then t here exist s a diffeomo rphism h: N ----> N
t hat is smoo th ly isot opi c to the identity and ca rries y into z.
H omotopy and i sotopy 23

( For the special case N sn the proof is easy : simply choose h to

be the rotation which carries y into z and leaves fixed all vectors orthog­
onal to the plane through y and z. )
The proof in general proceeds as follows : We will first construct a
smooth isotopy from Rn to itself which

1) leaves all points outside of the unit ball fixed, and

2) slides the origin to any desired point of the open unit ball .

/ ""

/
'" ...-V
Figure 7. Deforming the unit ball

Let 'P : Rn ----+ R be a smooth function which satisfies

'P (x) > 0 for I Ixll < 1
'P(x) = 0 for Ilx I I � l.
(For example let 'P(x) = A ( 1 - IlxW) where A(t) = 0 for t < 0 and
A (t) = exp ( _C1) for t > 0.) Given any fixed unit vector c £ sn-\
consider the differential equations
dx ·
df = Ci'P(X1, • • • , x ); � = 1, . . . , n.
n
For any X £ Rn these equations have a unique solution x = x(t), defined
for all* real numbers which satisfies the initial conditi on
x(O) = x.
W e will use the notation x(t) = F,(x) for this solution. Then clearly
1) F t (x) is defined for all t and x and depends smoothly on t and x,
2) Fo ( x) = x,
3) F,+t (x) = F, 0 Fl (x).

* Compare [22, §2.4].

 §4 . D egree modul o 2

Therefore each F, is a diffeomorphism from Rn onto itself. Letting t

vary, we see that each F, is smoothly isotopic to the identity under an
isotopy which leaves all points outside of the unit ball fixed . But clearly,
with suitable choice of c and t, the diffeomorphism F, will carry the
origin to any desired point in the open unit ball.
Now consider a connected manifold N. Call two points of N "isotopic"
if there exists a smooth isotopy carrying one to the other. This is
clearly an equivalence relation. If y is an interior point, then it has a
neighborhood diffeomorphic to R\ hence the above argument shows
that every point sufficiently close to y is "isotopic" to y. In other words,
each "isotopy class" of points in the interior of N is an open set, and
the interior of N is partitioned into disjoint open isotopy classes.
But the interior of N is connected; hence there can be only one such
isotopy class. This completes the proof.
We can now prove the main result of this section. Assume that M
is compact and boundaryless, that N is connected, and that f : JJ![ ----> N
is smooth .

Theorem. If y and z a re regula r va lues of f t hen

#r 1 (y) == #r 1 (z) (modulo 2).
This common residue cla ss, which is ca lled t he mod 2 degree of f, depends
only on the smooth homotopy cla ss of f.
PROOF. Given regular values y and z, let h be a diffeomorphism
from N to N which is isotopic to the identity and which carries y to z.
Then z is a regular value of the composition h 0 f. Since h 0 f is homotopic
to f, the Homotopy Lemma asserts that
# (h 0 f)-l(Z) == #r\ z) (mod 2) .
But

so that

Therefore

as required .
Call this common residue class deg2(f) . Now suppose that f is smoothly
homotopic to g. By Sard's theorem, there exists an element y I: N
H omotopy a nd isotopy 25

which is a regular value for both f and g. The congruence

deg2 f == #r\y) == #g-l(y) == d eg2 g (mod 2 )
now shows that deg2 f i s a smooth homotopy invariant, and completes
the proof.
EXAM PLES . A constant map c : M � M has even mod 2 degree.
The identity map I of M has odd degree. H ence the identity ma p of a
compa ct bounda ry less manifold is not homotopic to a constant .
In the case M. = sn, this result implies the assertion that no smooth
map f : Dn+! � sn leaves the sphere pointwise fixed. (I .e., the sphere
is not a smooth " retract" of the disk. Compare §2, Lemma 5 . ) For such
a map f would give rise to a smooth homotopy
F(x, t) = f(tx) ,
between a constant map and the identity.
 §5. ORIENTED MANIFOLDS

IN ORDER to define the degree as an integer (rather than an integer

modulo 2) we must introduce orientations .

DE FINITIONS. An orientation for a finite dimensional real vector

space is an equivalence class of ordered bases as follows : the ordered
basis (b] , . . . , bn ) determines the same orienta ti on as the basis (bj , . . . , b� )
if b: = L a i ibi with det (aii) > 0 . It determines the opposite orienta tion
if det (a i i) < 0 . Thus each positive dimensional vector space has precisely
two orientations. The vector space Rn has a sta nda rd orientation corre­
sponding to the basis ( 1 , 0, . . . , 0) , (0, 1 , 0, . . . , 0) , . . . , (0, . . . , 0, 1 ) .
I n the case o f the zero dimensional vector space it i s convenient to
define an "orientation" as the symbol + 1 or - 1.
An oriented smooth manifold consists of a manifold M together
with a choice of orientation for each tangent space TMx. If m � 1,
these are required to fit together as follows : For each point of M there
should exist a neighborhood U C M. and a diffeomorphism h mapping U
onto an open subset of Rm or Hm which is orienta tion preserving, in the
sense that for each x £ U the isomorphism dhx carries the specified
orientation for TJJifx into the standard orientation for Rm.
If M is connected and orientable, then it has precisely two
orienta tions.
If M has a boundary, we can distinguish three kinds of vectors in
the tangent space TM", at a boundary point :

1) there are the vectors tangent to the boundary, forming an (m - 1 ) ­

dimensional subspace T(aM)", C TMx;
2) there are the "outward" vectors, forming an open half space
bounded by T(aM)x;
3) there are the "inward" vectors forming a complementary half
space.
The Bro uwer degr ee 27

Each orientation for M determines an orientation for aM� as follows :

For x £ aNI choose a positively oriented basis (VI' V2, , vm) for TMx
• • •

in such a way that V2, • • • , Vm are tangent to the boundary (assuming

that m :2: 2) and that VI is an "outward" vector. Then (V2' , vm) . . •

determines the required orientation for a M at x.

If the dimension of M is 1, then each boundary point x is assigned
the orientation -1 or + 1 according as a positively oriented vector
at x points inward or outward . (See Figure S.)

Figure 8 . How to orient a boundary

m
As an example the unit sphere s -I C Rm can be oriented as the
boundary of the disk Dm .

THE BROUWER DEGREE

Now let M and N be oriented n-dimensional manifolds without

boundary and let

be a smooth map . If M is compact and N is connected, then the degree

of f is defined as follows :
Let x £ 111 be a regular point of f, so that df" : TMx -t TNf(x) is a
linear isomorphism between oriented vector spaces. Define the sign
of dfx to be + 1 or -1 according as dfx preserves or reverses orientation.
For any regular value y £ N define
deg(f ; y) = L: sign dfx .
Xf,f-l (y)

As in § 1 , this integer deg (f; y) is a locally constant function of y. It is

defined on a dense open subset of N.
28 §5. Oriented manifolds

Theorem A. The integer deg(f; y) does not depend on the choice of

regula r value y.
I t will be called the degree of f (denoted deg f) .

Theorem B. If f is smoothi y homotopic to g, then deg f = deg g.

The proof will be essentially the same as tha t in §4 . It is only necessary
to keep careful control of orientations.
First consider the following situation : Suppose that M. is the boundary
of a compact oriented manifold X and that M is oriented as the boundary
of X .

Lemma 1. If f : M --* N extends to a smooth map F : X --* N, then

deg(f; y) = ° for every regula r value y.
PROOF. First suppose that y is a regular value for F, as well as for
f = F I M. The compact I-manifold F- I (y) is a finite union of arcs and
circles, with only the boundary points of the arcs lying on M = ax.
Let A C r I (y) be one of these arcs, with aA = {a} U {b}. We will
show that
sign dfa + sign dfb = 0,
and hence (summing over all such arcs) that deg (f ; y) 0.

0,
Figure 9. How to orient F-I(y)

The orientations for X and N determine an orientation for A as

follows : Given x [ A, let (VI' . . . , vn+ ) be a positively oriented basis
I
for TXx with VI tangent to A . Then VI determines the required orientation
for TAx if and only if dFx carries (V2' . . . , vn +l ) into a positively oriented
ba sis j or TNy•
Let VI (x) denote the positively oriented unit vector tangent to A at x.
Clearly VI is a smooth function, and VI (x) points outward at one boundary
point (say b) and inward at the other boundary point a.
The Brouwer degre e 29

It follows immediately that

sign dfa = - 1, sign dfb = + 1;
with sum zero . Adding up over all such arcs A, we have proved that
deg(f; y) = 0.
More generally, suppose that Yo is a regular value for f, but not for F.
The function deg(fi y) is constant within some neighborhood U of Yo·
Hence, as in §4, we can choose a regular value y for F within U and
observe that
deg(f ; Yo) = deg(f; y) = 0.
This proves Lemma 1.

Now consider a smooth homotopy F : [0, 1] X 111 --+ N between

two mappings f(x) = F(O, x), g (x) = F(I, x).

Lemma 2. The degree deg(g; y) is equal to deg(f; y) for any common

regular value y.
PROOF. The manifold [0, 1] X lVr can be oriented as a product,
and will then have boundary consisting of 1 X Ar (with the correct
orientation) and ° X lVr (with the wrong orientation) . Thus the degree
of F I a([O, 1] X lVr) at a regular value y is equal to the difference
deg(g ; y) - d eg(f ; y) .
According to Lemma 1 this difference must be zero .
The remainder of the proof of Theorems A and B is completely
analogous to the argument in §4 . If y and z are both regular values
for f : 111 --+ N, choose a diffeomorphism h : N --+ N that carries y to z
and is isotopic to the identity. Then h will preserve orientation, and
deg(f ; y) = d eg(h 0 f; he y»�
by inspection. But f is homotopic to h 0 fi hence
deg(h 0 f; z) = deg(f ; z)

by Lemma 2. Therefore deg(f; y) = deg(f; z) , which completes the proof.

The complex function z --+ Z , Z � 0, maps the unit circle

k
EXAMPLE S .
onto itself with degree k. (Here k may be positive, negative, or zero .)
The degenerate mapping

f : 111 --+ constant £ N

30 §5 . Oriented manifolds

has degree zero. A diffeomorphism f : M ---7 N has degree + 1 or -1

according as f preserves or reverses orientation. Thus a n orienta tion
reversing d�fJ eom orphism of a compa ct bounda ry less manifold is not
smoothly homotopic to the identity.
One example of an orientation reversing diffeomorphism is provided
by the reflection r i : sn ---7 sn, where

The antipodal map of sn has degree (-If+'' as we can see by noting

that the antipodal map is the composition of n + 1 reflections :

Thus if n is even, the a ntipoda l map of sn is not smoothly homotopic to

the identity, a fact not detected by the degree modulo 2 .
A s an application, following Brouwer, w e show that sn a dmits a
smooth field of nonzero tangent vectors if a nd only if n is odd. (Compare
Figures 10 and 11.)

o
Figure 10 (above). A nonzero vector field on the I-sphere

Figure 11 (below). Attempts for n = 2

DEFIN ITION . A smooth tangent vector field on M C Rk is a smooth

map v : M ---7 Rk such that v (x) 1: TMx for each x 1: M. In the case of
the sphere sn C Rn+1 this is clearly equivalent to the condition
1) v(x) · x = 0 for all x 1: W,
The Brouwer degree 31
using the euclidean inner product.
If vex) is nonzero for all x, then we may as well suppose t h a t
2) ( ) ·v(x)
l) X = 1 for all x (: sn.
For in any case vex) = v(x) / l l v (x) 1 1 would be a vector field which does
satisfy this condition. Thus we can think of v as a smooth function
from sn to itself.
Now define a smooth homotopy
F : sn X [0 , 11"J � sn
hy the formula F(x, 0) = x cos 0 + v ex) sin o. Computation shows that
F(x, O) · F(x , 0) = 1
and that
F(x , O) = x, F(x , 11") = -x.
Thus the antipodal map of sn i s homotopic to the identity. But for n

even we have seen that this is impossible.

On the other hand, if n = 2k - 1 , the explicit formula

defines a nonzero tangent vector field on S" . This completes the proof.
It follows, incidentally, that the antipodal map of sn is homotopic
to the identity for n odd. A famous theorem due to Heinz Hopf asserts
that two mappings from a connected n-manifold to the n-sphere are
smoothly homotopic if and only if they have the same degree . In §7
we will prove a more general result which implies Hopf's theorem.
 §6. VECTOR FIELDS AND

THE EULER NUMBER

As a further application of the concept of degree, we study vector fields

on other manifolds.
Consider first an open set U C R m and a smooth vector field

with an isolated zero at the point z t U . The function

ve x) = v (x)/ \ \v(x) \ \
maps a small sphere centered at z into the unit sphere. * The degree
of this mapping is called the index L of v at the zero z.
Some examples, with indices - 1, 0, 1, 2, are illustrated in Figure 12 .
(Intimately associated with v are the curves "tangent" to v which are
obtained by solving the differential equations dx . j dt = V i (Xl, . . . , xn) .
It is these curves which are actually sketched in Figure 12.)
A zero with arbitrary index can be obtained as follows : In the plane
of complex numbers the polynomial l defines a smooth vector field
with a zero of index k at the origin, and the function l defines a
vector field with a zero of index - k.
We must prove that this concept of index is invariant under diffeomor­
phism of U. To explain what this means, let us consider the more
general situation of a map f : M ----> N, with a vector field on each
manifold.

DEFIN ITION . The vector fields v on M and V i on N correspond under f

if the derivative dfx carries v ex) into v' (f (x) ) for each X t M.

* Each sphere is to be oriented as the boundary of the corresponding disk.

The index 33

{, = - I {' = Q

{, = + I (, = + 2

Figure 12. Examples of plane vector fields

If f is a diffeomorphism, then clearly Vi is uniquely determined by v .

The notation
Vi = df 0 v o rl
will be used .

Lemma 1. S1lppose that the vector field v on U corresponds to

Vi = df 0 v o rl
on U' under a diffeomorphism f : U -> U'. Then the index of v at an isolated
zero z is equal to the index of Vi at f Cz) .
34 §6. Vector fields

Assuming Lemma 1, we can define the concept of index for a vector

field w on an arbitrary manifold M as follows : If g : U ----> M is a para­
metrization of a neighborhood of z in M, then the index L of w at z is de­
fined to be equal to the index of the corresponding vector field dg - l O W 0 g
on U at the zero g- I (Z) . It clearly will follow from Lemma 1 that L is
well defined.
The proof of Lemma 1 will be based on the proof of a quite different
result :

Lemma 2. A ny orientation preserving diffeomorphism f of R m is

smoothly isotopic to the identity.
(In contrast, for many values of m there exists an orientation pre­
serving diffeomorphism of the sphere s m which is not smoothly isotopic
to the identity. See [20, p . 404] . )

PROOF . W e may assume that f (O) = 0 . Since the derivative a t °

can be defined by
dfo (x) = l im f(tx)/t ,
t�O

it is natural to define an isotopy

by the formula
F(x , t) = f(tx)/ t for ° < t :::; 1 ,
F(x, 0) = dfo (x) .
To prove that F is smooth, even as t ----> 0, we write f in the form*

where gl , . . . , gm are suitable smooth functions, and note that

for all values of t .

Thus / is isotopic to the linear mapping d/o, which is clearly isotop i ( '
t o the identity. This proves Lemma 2 .

PROOF O F LEMMA 1 . W e may assume that z = f ez ) = 0 and that U

is convex. If / preserves orientation, then, proceeding exactly as above,

* See for example [22, p. 5] .

The index swn 35

we construct a one-parameter family of embeddings

ft : U ----> Rm
with fo = identity, fl = f, and f, (O) = 0 for all t. Let v, denote the
vector field df, 0 v 0 f� 1 on !, ( U) , which corresponds to v on U. These
vector fields are all defined and nonzero on a sufficiently small sphere
centered at o. Hence the index of v = Vo at 0 must be equal to the
index of v ' = VI at o. This proves Lemma 1 for orientation preserving
diffeomorphisms.
To consider diffeomorphisms which reverse orientation it is sufficient
to consider the special case of a reflection p. Then
V ' = p O V O p- l ,
so the associated function v'(x) = v
'
(x ) / l l v ' (x ) 1 1 on the e-sphere satisfies
v' = p 0 v 0 P- I .

Evidently the degree of v' equals the degree of v, which completes the
proof of Lemma 1 .
We will study the following classical result : Let M be a compact
manifold and w a smooth vector field on M with isolated zeros. If M
has a boundary, then w is req1lired to point outward at all boundary points.
Poincare -Hopf Theorem. The sum L� of the indices at the zeros at
such a vector field is equal to the Euler number*
m
i
x(M) = L ( _ I) rank H i (M) .
i=Q
In particular this index S1lm is a topological invariant at M : i t does not
depend on the particular choice of vector field.
(A 2-dimensional version of this theorem was proved by Poincare
in 188.5 . The full theorem was proved by Hopf [ 14] in 1926 after earlier
partial results by Brouwer and Hadamard . )
W e will prove part o f this theorem, and sketch a proof o f the rest .
First consider the special case of a compact domain in Rm.
Let X C R'" be a compact In-manifold with boundary. The Gauss
mapping
g : ax ----> 8m- I
assigns to each x 1: aX the outward unit normal vector at x.

* Here HiCM) denotes the i-th homology group o f M . This will b e our first and
last reference to homology theory.
36 §6. Vector fields

Lemma 3 (Hopf) . If v : X -+ Rm is a smooth vector field with isolated

zeros, and if v points out of X along the boundary, then the index sum L �
is equal to the degree of the Gauss mapping from ax to sm- I . In particular,
L � does not depend on the choice of v .
For example, if a vector field on the disk Dm points outward along
the boundary, then L � = + 1. (Compare Figure 13.)

Figure 13. An example with index sum +1

PROOF. Removing an e-ball around each zero, we obtain a new

manifold with boundary. The function v ex ) = v (x) / l l v (x) 1 1 maps this
manifold into sm- I . Hence the sum of the degrees of v restricted to the
various boundary components is zero . But v I a x is homotopic to g,
and the degrees on the other boundary components add up to - L � .
(The minus sign occurs since each small sphere gets the wrong orien­
tation.) Therefore
deg(g) - L � = 0
as required .
REMARK. The degree of g is also known as the "curvatura integra"
of a x , since it can be expressed as a constant times the integral over
a x of the Gaussian curvature. This integer is of course equal to the
Euler number of X. For m odd it is equal to half the Euler number of a x .

Before extending this result to other manifolds, some more pre­

liminaries are needed .
It is natural to try to compute the index of a vector field v at a zero z
The index sum 37

in terms of the derivatives of v at z. Consider first a vector field v on

an open set U C R m and think of v as a mapping U -; R m , so that
dvz : Rm -; R m is defined .
DEFIN IT ION . The vector field v is nondegenerate at z if the linear
transformation dv, is nonsingular.
It follows that z is an isolated zero .

Lemma 4. The index of v at a nondegenerate zero z is either + 1 or - 1

according as the determinant of dv, is positive or negative.
PROOF . Think of v as a diffeomorphism from some convex neighbor­
hood Uo of z into R m . We may assume that z = o. If v preserves orien­
tation, we have seen that vi Uo can be deformed smoothly into the
identity without introducing any new zeros. (See Lemmas 1, 2. ) Hence
the index is certainly equal to + 1.
If v reverses orientation, then similarly v can be deformed into a
reflection ; hence L = - l .
More generally consider a zero z of a vector field w on a manifold
M e Rlo . Think of w as a map from M to Rk so that the derivative
dw, : TM, -; Rk is defined.
Lemma 5. The derivative dw, actually carries TIYI, into the s1lbspace
TNI, C R\ and hence can be considered as a linear transformation from
TM, to itself. If this linear transformation has detmninant D � 0 then
z is an isolated zero of w with index equal to + 1 or - 1 according as D is
positive or negative.
PROOF. Let h : U -; M be a parametrization of some neighborhood
of z. Let ei denote the i-th basis vector of R m and let

so that the vectors t', . . . , en form a basis for the tangent space TIYIh (u) .
We must compute the image of ti = ti (u) under the linear transforma­
tion dW h(n) . First note that
1)
Let v I: V i e ; be the vector field on U which corresponds to the vector
field w on M. By definition v = dh- I o W 0 h, so that
w(h(u» = dh,,(v) = I: Vi ti .
Therefore
2)
38 §6. Vector fields

Combining 1) and 2), and then evaluating at the zero h-1 ( z) of v, we

obtain the formula

3)
Thus dwz maps TMz into itself, and the determinant D of this linear
transformation TM. -+ TM. is equal to the determinant of the matrix
(av ; / aUi) ' Together with Lemma 4 this completes the proof.

Now consider a compact, boundaryless manifold M C Rk. Let N,

denote the closed t-neighborhood of M (i . e . , the set of all x £ Rk with
[ [x - y [ [ � t for some y £ M) . For t sufficiently small one can show
that N, is a smooth manifold with boundary. (See §8, Problem 1 1 .)

Theorem 1. For any vector field v on M with only nondegenerate zeros,

the index sum L L is equal to the degree of the Gauss mapping*

g : aN, -+ Sk-l .
In particular this sum does not depend on the choice of vector field.
PROOF. For x £ N, let r ex) (: M denote the closest point of M. (Compare
§8, Problem 12.) Note that the vector x - r ex) is perpendicular to the
tangent space of M at r ex) , for otherwise r ex) would not be the closest
point of M. If t is sufficiently small, then the function r ex) is smooth and
well defined.

Figure 14. The .-neighborhood of M

* A different interpretation of this degree has been given by Allendoerfer and

Fenchel : the degree of g can be expressed as the integral over M of a suitable curva­
ture scalar, thus yielding an m-dimensional version of the classical Gauss-Bonnet
theorem. (References [ 1J, [9]. See also Chern [6J . )
The index sum 39

We will also consider the squared distance function

cp (x) = I l x - r (x) W·
An easy computation shows that the gradient of 'I' is given by
grad 'I' = 2(x - r(x» .
Hence, for each point x of the level surface aN, = cp-\i) , the outward
unit normal vector is given by
g (x) = grad '1'/ 1 I grad 'I' l l = (x - r(x» /E .
Extend v to a vector field w on the neighborhood N, by setting
w(x) = (x - r ex»� + v (r(x» .
Then w points outward along the boundary, since the inner product
w (x) · g (x) is equal to E > O. Note that w can vanish only at the zeros
of v in M; this is clear since the two summands (x - r ex» � and v (r(x»
are mutually orthogonal. Computing the derivative of w at a zero
z tM, we see that
dWz (h) = dUz(h) for all h t TMz
dWz (h) = h for h t TM� .
Thus the determinant of dwz is equal to the determinant of dvz • Hence
the index of w at the zero z is equal to the index L of v at z.
Now according to Lemma 3 the index sum I: L is equal to the degree
of g. This proves Theorem 1.

EXAMPLE S . On the sphere 8m there exists a vector field v which

points "north" at every point. * At the south pole the vectors radiate
outward ; hence the index is + 1 . At the north pole the vectors converge
inward ; hence the index is ( - 1 ) '" . Thus the invariant I: L is equal to 0
or 2 according as rn is odd or even. This gives a new proof that every
vector field on an even sphere has a zero.
For any odd-dimensional , boundaryless manifold the invariant I: L is
zero . For if the vector field v is replaced by - v, then each index is
multiplied by ( _ 1 ) m, and the equality
I: L = ( _ 1) m I: L,

for m odd, implies that I: L = O.

* For example, v can be defined by the formula v(x) = p - (p ' x)x, where p is
the north pole. ( See Figure 1 1 . )
40 §6. Vector fields

REMARK . If L L = ° on a connected manifold AI, then a theorem

of Hopf asserts that there exists a vector field on M with no zeros at all.

In order to obtain the full strength of the Poincare-Hopf theorem,

three further steps are needed .
STEP 1. Identification of the invariant L L with the E1der n1l1nber x(M) .
It is sufficient to construct j ust one example of a nondegenerate vector
field on AI with L L equal to x (M) . The most pleasant way of doing
this is the following : According to Marston Morse, it is always possible
to find a real valued function on M whose " gradient" is a nondegenerate
vector field . Furthermore, Morse showed that the sum of indices
associated with such a gradient field is equal to the Euler number of M .
For details o f this argument the reader i s referred t o Milnor [22, pp . 29,
36] .
STEP 2 . Proving the theorem for a vector field with degenerate zeros.
Consider first a vector field v on an open set U with an isolated zero

at z. If
A : U --7 [0 , 1 ]
takes the value 1 o n a small neighborhood N l of z and the value °
outside a slightly larger neighborhood N, and if y is a sufficiently
small regular value of v, then the vector field
v' (x) = v ex) - A (X) Y
is nondegenerate* within N. The sum of the indices at the zeros within N
can be evaluated as the degree of the map
ii : aN --7 sm - l ,

and hence does not change during this alteration.

More generally consider vector fields on a compact manifold M.
Applying this argument locally we see that any vector field with isolated
zeros can be replaced by a nondegenerate vector field without altering the
7·nteger L L .
STEP 3. Manifolds with boundary. If M C Rk has a boundary, then
any vector field v which points outward along alJ1 can again be extended
over the neighborhood N, so as to point outward along aN, . However,
there is some difficulty with smoothness around the boundary of M.
Thus N, is not a smooth (i .e. differentiable of class COO) manifold,

* Clearly v' is nondegenerate within N1• But if Y is sufficiently small, then v'
will have no zeros at all within N - N1 •
The index sum 41

but only a C l -manifold . The extension w, if defined as before by

w (x) = v (r (x) ) + x - r ex) , will only be a continuous vector field
near aM. The argument can nonetheless be carried out either by showing
that our strong differentiability assumptions are not really necessary
or by other methods.
 §7. FRAMED COBORDIS M

THE PONTRYAGI N CONS TRUCTION

THE degree of a mapping M � M' is defined only when the manifolds

M and M' are oriented and have the same dimension. We will study
a generalization, due to Pontryagin, which is defined for a smooth map
f : M � SP
from an arbitrary compact, boundaryless manifold to a sphere. First
some definitions .
Let N and N' be compact n-dimensional submanifolds of M with
aN = aN' = aM = 0. The difference of dimensions m - n is called
the codimension of the submanifolds .
D EFINITION. N is coboTdant to N' within M if the subset
N X [0 , �) U N' X (1 - �, 1]
of M X [0, 1 ] can be extended to a compact manifold
X C M X [0 , 1 ]
s o that
ax = N X ° U N' X 1 ,
and so that X does not intersect M X ° U M X 1 except at the points
of ax.
Clearly cobordism is an equivalence relation. (See Figure 1 5 . )
DEFINITION . A framing o f the submanifold N C M i s a smooth
function \:) which assigns to each x t N a basis
=
' m-n
\:) (x) (v (x) , . . . , v (x»
for the space TN; C TM� of normal vectors to N in M at x. (See
The Pontryagin construction

(�
MXO M X I M x2

Figure 15. Pasting together two cobordisms within M

Figure 1 6 .) The pair (N, b) is called a framed submanifold of M. Two

framed submanifolds (N, b) and (N', tl:J) are framed cobordant if there
exists a cobordism X C M X [0, 1] between N and N' and a framing
u of X, so that
'
u (x , t) = (V i (x) , 0) for (x , t) (: N X [0 , E)
i
u (x , t) = (W i (x) , 0) for (x , t) (: N ' X ( 1
- E, 1].

Again this is an equivalence relation.

Now consider a smooth map f : M ----> W and a regular value y (: SP.
l
The map f induces a framing of the manifold r (y) as follows : Choose
a positively oriented basis b = (v\ . . . , vP) for the tangent space T ( SP ) •.
For each x (: r l (y) recall from page 12 that

maps the subspace Tr l (y)x to zero and maps its orthogonal complement
Tr l (y) � isomorphically onto T(SP)y. Hence there is a unique vector
i
w (x) (: Tr \ y); C TMx
i
that maps into V under dtx ' It will be convenient to use the notation
tl:J = f*)J for the resulting framing w l (x ) , . . . , wP (x) of r l (y) .
D EFINITION. This framed manifold (f - l (y) , f*b) wi ll be called the
Pontryagin manifold associated with f.
Of course f has many Pontryagin manifolds, corresponding to dif­
ferent choices of y and b, but they all belong to a single framed cobordism
class :

Theorem A. If y' is another regular val1le of f and b' is a positively

oriented basis for T(SP) . " then the framed manifold (f- l (y'), f*b') is
framed cobordant to (f- l (y) , f*b) .
Theorem B. Two mappings from M to SP are smoothly homotopic
if and only if the associated Pontryagin manifolds are framed cobordant.
44 §7. Framed cobordism

N� ___

M x [O, IJ
Figure 1 6. Framed submanifolds and a framed cobordism

Theorem C. A ny compact framed submanifold (N, \t1) of codimension p

in M occurs as Pontryagin manifold for some smouth mapping f : JJ1 -> SP .
Thus the homotopy classes of maps are in one-one correspondence
with the framed cobordi:·nn classes of subl1lanifolds .
The proof of Theorem A will be very similar to the arguments in
§§4 and 5. It will be based 011 three lemmas .

Lemma 1 . If 0 and 0' are two different positively oriented bases at y,

then the Pontryagin manifuld (r ' (y) , f* o ) is framed cobordant to (f- 1 (y) ,
t* o' ) .
PROOF. Choose a smooth path from 0 to 0 ' in the space of all positively
oriented bases for T(SP ) y . This is possible since this space of bases
The Pontryagin construction 45

can be identified with the space GL + (p, R) of matrices with positive

determinant, and hence is connected. Such a path gives rise to the
required framing of the cobordism r 1 (y) X [0, 1 ] .
B y abuse o f notation w e will often delete reference to f*o and speak
simply of " the framed manifold r 1 (y) . "

Lemma 2. If y is a regular value off, and z is sufficiently close t o y , then

r l (z) is framed cobordant to r 1 (y) .
PROOF. Since the set f (C) of critical values is compact, we can choose
E > 0 so that the E-neighborhood of y contains only regular values.
Given z with l i z - y l l < E, choose a smooth one-parameter family
of rotations (i.e. an isotopy) r, : SD --+ SD so that rl (y ) = z, and so that
1 ) r, is the identity for 0 � t < E',
2) r, equals r for 1 - E' < t � 1, and
1
3 ) each r;- l (z) lies on the great circle from y to z, and hence is a
regular value of f.
Define the homotopy
F : M X [0 , 1] --+ SV
by F (x, t) = r,f(x) . For each t note that z is a regular value of the
composition
r, 0 f : NI --+ SP .
It follows a fortiori that z is a regular value for the mapping F. Hence
rJ (z) C M X [0 , 1 ]
is a fraIned manifold and provides a framed cobordism between the
framed manifolds r l (z) and (r o f) - I (Z ) = r l r� l (z) = r (y) . This
I
proves Lemma 2.

Lemma 3. If f and g are smoothly homotopic and y is a regular value

fOT both, then r 1 (y) is framed cobordant to g- ' (y) .
PROOF. Choose a homotopy F with
F(x , t) = f (x) o � t < E,
F(x , t) = g (x) l - E < t � l.
Choose a regular value z for F which is close enough to y so that r ' (z )
is framed cobordant to r l (y) and so that g- I (Z) is framed cobordant
to g-' (y) . Then F-1 (z) is a framed manifold and provides a framed
cobordism between r 1 (z) and g - I (Z) . This proves Lemma 3.
46 §7. Framed cobordisl1t

PROOF OF THEOREM A . Given any two regular values y and z for t,

we can choose rotations

so that ro is the identity and r l (y ) = z. Thus f is homotopic to r 0 fi

l
hence rl (z ) is framed cobordant to

(r l 0 f) -l (Z) = r'r � ' (z) = r 1 (y) .

This completes the proof of Theorem A.

The proof of Theorem C will be based on the following : Let N C M

be a framed submanifold of codimension p with framing o. Assume that
N is compact and that aN = aM = )25 .
Product Neighborhood Theorem. Some neighborhood of N in M is
diffeomorphic to the product N X RP• Furthermore the diffeomorphism
can be chosen so that each x E N corresponds to (x, 0) E N X RP and
so that each normal frame o (x ) corresponds to the standard basis for RP•
REMARK. Product neighborhoods do not exist for arbitrary submani­
folds . (Compare Figure 17. )

Figure 1 7. An unframable submanifold

p
First suppose that M is the euclidean space Rn+ • Consider
PROOF.
the mapping g : N X W -+ M, defined by
g(x i tl , . . . , tp) = x + tIVl (X) + . . . + tpvP(x) .
Clearly dg ex ; o , " " O ) is nonsingular; hence g maps some neighborhood
of (x, 0) E N X RP diffeomorphically onto an open set .
We will prove that g is one-one on the entire neighborhood N X U,
of N X 0, providing that E > 0 is sufficiently small ; where U, denotes
the E-neighborhood of 0 in RP• For otherwise there would exist pairs
(x, u) � (x', u') in N X RP with I l ul l and I lu' l l arbitrarily small and with
g(x, u) = g(x ' , u
'
).
The PontTyagin construction

Since N is compact, we could choose a sequence of such pairs with x

converging, say to Xo, with x' converging to X6, and with u -+ 0 and
u' -+ o. Then clearly Xo = Xb, and we have contradicted the statement
that g is one-one in a neighborhood of (x o, 0) .
Thus g maps N X U, diffeomorphically onto an open set . But U,
is diffeomorphic to the full euclidean space RP under the correspondence
u -+ u/(1 - I l u WN) .
Since g (x , 0) = x, and since dg (x , o) does what is expected of it, this
proves the Product Neighborhood Theorem for the special case
p
M = Rn+ •
For the general case it is necessary to replace straight lines in Rn+v
by geodesics in M. More precisely let g (x; t l , . . . , tp) be the endpoint
of the geodesic segment of length I l t l v l (x) + + tpv P (x ) I I in M which
starts at x with the initial velocity vector
t1v\x) + . . . + tpvP (x) / I I tlVl (X) + . . . + tpvP (x) I I .
The reader who IS familiar with geodesics will have no difficulty in
checking that
g : N X U, -+ M
is well defined and smooth, for E sufficiently small . The remainder of
the proof proceeds exactly as before.
PROOF OF THEOREM C. Let N C JJI be a compact, boundaryless,
framed submanifold . Choose a product representation
g : N X R P -+ V C M
for a neighborhood V of N, as above, and define the proj ection
7r : V -+ R P
by 7r (g (x, y» = y. (See Figure 1 8 . ) Clearly 0 is a regular value, and
7r- l (O ) is precisely N with its given framing.

7T
..

Figure 18. Constructing a map with given Pontryagin manifold

48 §7. Framed cobordism

Now choose a smooth map <p : RP --7 SP which maps every x with
I l x l l ;?: 1 into a base point So, and maps the open unit ball in RP diffeomor­
phically* onto SP - so. Define

by
f(x) = <p (7r (x)) for x (: V
f(x) = So for x ¢ V.
Clearly f i s smooth, and the point <p (0) i s a regular value o f f. Since
the corresponding Pontryagin manifold
r1(<p (0)) = 7r-1 (0)
is precisely equal to the framed manifold N, this completes the proof
of Theorem C.
In order to prove Theorem B we must first show that the Pontryagin
manifold of a map determines its homotopy class . Let f, g : M --7 SP
be smooth maps with a common regular value y.

Lemma 4. If the framed manifold (f- l (y) , f*\.J) is equal to (g - l (y) , g*\.J) ,
then f is smoothly homotopic to g.
PROOF. It will be convenient to set N = r 1 (y) . The hypothesis
that f* \.J = g*\.J means that dfx = dgx for all x (: N.
First suppose that f actually coincides with g throughout an entire
neighborhood V of N. Let h : SP - y --7 R" be stereo graphic proj ection .
Then the homotopy
F(x , t) = f(x) for x € V
F(x , t) = h- 1 [t · h(f (x) ) + ( 1 - t) · h(g(x)) ] for x (: M - N
proves that f is smoothly homotopic to g.
Thus is suffices to deform f so that it coincides with g in some small
neighborhood of N, being careful not to map any new points into y
during the deformation. Choose a product representation
N X RP --7 VC M
for a neighborhood V of N, where V is small enough so that f ( V ) and

* For example, ",(x) h-'(x/A( l lx I 1 2» , where h is the stereographic proj ection

from So and where A is a smooth monotone decreasing function with A(t) > 0 for
=

t < 1 and A(t) 0 for t 2: 1 .

=
The Pontryagin construction 49
g (V) do not contain the antipode y of y. Identifying V with N X R"
and identifying SP - Y with RP, we obtain corresponding mappings

with

and with
dF ex , o) = dGex ,o) = (proj ection to RP)
for all x 1: N .
We will first find a constant c so that

F(x , u) ' u > 0, G(x , u) ' 1l > °

for x 1: N and 0 < I l u l l < c. That is, the points F(x , u ) and G (x, u )
belong to the same open half-space in RP• So the homotopy

(1 - t) F(x , u) + tG(x , u)
between F and G will not map any new points into 0, at least for I l u l l < c ·
By Taylor's theorem

I IF(x , u) - u l l :::; c , 1 1 1l 1 1 2 , for I l u l l :::; 1.

Hence
I (F(x , u) - u) ' u l :::; c , I l u W
and
F(x , u) · u 2: I lu l l " - c, I l u l l " > °
for ° < I l u l l < Min (c;" 1 ) , with a similar inequality for G.
To avoid moving distant points we select a smooth map "A : RP � R
with
"A(u) = 1 for I l u I I :::; c /2
"A(u) = 0 for I l u l l 2: c ,
Now the homotopy
F , (x , u) = [1 - "A (u) t] F(x , u) + "A (u) tG(x , u)
deforms F = Fo into a mapping F, that ( 1 ) coincides with G in the
region I l u l l < c/2, (2) coincides with F for I l u l l 2: c, and (3 ) has no
new zeros , Making a corresponding deformation of the original mapping
f, this clearly completes the proof of Lemma 4 .
50 §7. Framed c obordism

PROOF OF THEOREM B. If f and g are smoothly homotopic, then

Lemma 3 asserts that the Pontryagin manifolds r 1 (y) and g - l (y)
are framed cobordant . Conversely, given a framed cobordism (X, In)
between r l (y) and g - l (y) , an argument completely analogous to the
proof of Theorem C constructs a homotopy

F : M X [0 , 1 ] -? sp
whose Pontryagin manifold (F- l (y ) , F*v) is precisely equal to (X, In) .
Setting F, (x) F(x, t) , note that the maps Fo and f have exactly the
=

same Pontryagin manifold . Hence Fo "" f by Lemma 4; and similarly

F "" g. Therefore f "" g, which completes the proof of Theorem B .
l
REMARKS . Theorems A , B, and C can easily b e generalized so as
to apply to a manifold M with boundary . The essential idea is to
consider only mappings which carry the boundary into a base point
So. The homotopy classes of such mappings

are m one-one correspondence with the cobordism classes of framed

submanifolds
N C Interior (M)

of codimension p. If p 2: tm + 1, then this set of homotopy classes

can be given the structure of an abelian group, called the p-th co homotopy
group 1l'P (M, aM) . The composition operation in 1l'P (M, aM) corresponds
to the union operation for disj oint framed submanifolds of Interior (M) .
(Compare §8, Problem 1 7 . )

THE HOPF THEOREM

As an example, let M be a connected and oriented manifold of dimen­

sion m = p. A framed submanifold of codimension p is j ust a finite set
of points with a preferred basis at each . Let sgn (x) equal + 1 or - 1
according as the preferred basis determines the right or wrong orien­
tation . Then L sgn (x) is clearly equal to the degree of the associated
map M -? sm. But it is not difficult to see that the framed cobordism
class of the O-manifold is uniquely determined by this integer L sgn (x) .
Thus we have proved the following .
The H opt theorem 51

Theorem of Hopf. If M is connected, oriented, and boundaryless,

then two maps M � sm are smoothly homotopic if and only if they have
the same degree.
On the other hand, suppose that M is not orientable. Then given a
basis for TMx we can slide x around M in a closed loop so as to transform
the given basis into one of opposite orientation. An easy argument
then proves the following :

Theorem. If M is connected but nonorientable, then two maps M � sm

are homotopic if and only if they have the same mod 2 degree.
The theory of framed cobordism was introduced by Pontryagin in
order to study homotopy classes of mappings

with m > p. For example if m = p + 1 � 4, there are precisely two

homotopy classes of mappings sm � SP. Pontryagin proved this result
by classifying framed I -manifolds in s m . With considerably more
difficulty he was able to show that there are j ust two homotopy classes
also in the case m = p + 2 � 4, using framed 2-manifolds . However,
for m - p > 2 this approach to the problem runs into manifold diffi­
culties.
It has since turned out to be easier to enumerate homotopy classes
of mappings by quite different, more algebraic methods . * Pontryagin's
construction is, however, a double-edged tool. It not only allows us to
translate information about manifolds into homotopy theory ; it con­
versely enables us to translate any information about homotopy into
manifold theory. Some of the deepest work in modern topology has
come from the interplay of these two theories . Rene Thorn's work on
cobordism is an important example of this. (References [36] , [2 1 ] . )

* See for example S .-T. Hu, Homotopy Theory.

 §8. EXERCISES

IN CONCLUSION here are some problems for the reader.

PnOBLEM 1 . Show that the degree of a composition g o f is equal to

the product (degree g ) (degree f) .

PROBLEM 2. Show that every complex polynomial of degree n gives

2
rise to a smooth map from the Gauss sphere S to itself of degree n.

PnOBLEM 3 . If two maps f and g from X to SP satisfy I lf (x) - g(x) 1 1 < 2

for all x, prove that f is homotopic to g, the homotopy being smooth if f
and g are smooth .

PROBLEM 4. If X is compact, show that every continuous map X -? SP

can be uniformly approximated by a smooth map . If two smooth maps
X -? SP are continuously homotopic, show that they are smoothly
homotopic .

PROBLEM 5 . If rn < p, show that every map Mffl -? SP is homotopic

to a constant .

PROBLE M 6. (Brouwer) . Show that any map sn -? sn with degree

different from ( - 1 r+ 1 must have a fixed point.

PROBLEM 7. Show that any map sn -? sn of odd degree must carry

some pair of antipodal points into a pair of antipodal points .

PROBLEM 8 . Given smooth manifolds M C R k and N C R l , show

that the tangent space T (M X N) cx , y ) is equal to TMx X TNy .

PROBLEM 9. The graph r of a smooth map f : M -? N is defined to be

the set of all (x, y) 1: M X N with f (x) = y. Show that r is a smooth
Exercises 53
manifold and that the tangent space
T r (x , y ) C TMx X TNy
is equal to the graph of the linear map dtx'
PROBLEM 10 . Given M C Rk, show that the tangent bundle space

TM = { (x , v) c M X Rk I v c TMx }
is also a smooth manifold. Show that any smooth map t : M -? N
gives rise to a smooth map
df : TM -? TN
where
d (identity) = identity, d (g 0 f) = (dg) 0 (dt) .
PROBLEM 1 1 . Similarly show that the normal bundle space
E = { (x , v) c ivI X Rk I v ..1 TMx }
is a smooth manifold. If M is compact and boundaryless, show that
the correspondence
(x , v) f-+ :1: +v
from E to Rk maps the t-neighborhood of M X 0 in E diffeomorphically
onto the t-neighborhood N, of M in Rk. (Compare the Product Neighbor­
hood Theorem in §7. )

PROBLEM 1 2 . Define r : N, -? M by r (x + v ) = x. Show that r ex + v)

is closer to x + v than any other point of M. Using this retraction r,
prove the analogue of Problem 4 in which the sphere SP is replaced
by a manifold M.

PROBLEM 1 3 . Given disj oint manifolds M, N C Rk + \ the linking map

A : M X N -? S k
is defined by A (X , y) = (x - y ) / l l x - y l l . If M and N are compact,
oriented, and boundaryless, with total dimension m + n = k, then
the degree of A is called the linking number l (M, N ) . Prove that
leN, M) = ( _ I ) (m + l ) (n+ l ) l (M , N) .
If M bounds an oriented manifold X disj oint from N, prove that
l (M, N) = O . Define the linking number for disj oint manifolds in
sm+n + l .
the sphere
54 §8. Exercises

PROBLEM 14, THE HOPF INVARIANT . If Y � z are regular values for a

map t : S21;- 1
SP, then the manifolds r ' (y), r ' (z) can be oriented
-?

as in §5; hence the linking number l (r ' (y) , r ' (z) ) is defined .
a) Prove that this linking number is locally constant as a function
of y.
b) If y and z are regular values of g also, where

I l t (x) - g(x) 1 1 < I I Y - z l l

for all x, prove that
l (f-' (y) , r' (z)) = l e g- , (y) , r ' (z)) = l(g- l (y) , g- , (z)) .
c) Prove that l (r ' (y) , r ' (z) ) depends only on the homotopy class
of t, and does not depend on the choice of y and z.
This integer H (f) = l (r ' (y), r ' (z) ) is called the H o pt invariant of f.
(Reference [ 1 5] . )
PROBLEM 1 5 . I f the dimension p i s odd, prove that H (f) = O. For
a composition

prove that H (g 0 f) is equal to H (f) mUltiplied by the square of the

degree of g.
The H oPt fibration 7r : S3 -? S 2 is defined by

7r(x" x 2 , x3 , X4) = h - 1 ((x + iX ) / (x3 + ix4))

, 2
where h denotes stereographic proj ection to the complex plane. Prove
that H (7r) = l.
PROBLEM 1 6 . Two submanifolds N and N' of M are said to intersect
transversally if, for each x t N (\ N', the subspaces TNx and TN:,
together generate TMx. (If n + n' < m this means that N (\ N' = 95 . )
I f N i s a framed submanifold, prove that i t can b e deformed slightly
so as to intersect a given N' transversally. Prove that the resulting
intersection is a smooth manifold.
PROBLEM 17. Let llV(M) denote the set of all framed cobordism
classes of codimension p in 11,11 . Use the transverse intersection operation
to define a correspondence

If p 2:: !m + 1 , use the disjoint union operation to make ll P (M) into

an abelian group . ( Compare p. 50.)
 AP P END I X

CLAS S IFYING I - MANIFOLDS

WE WILL prove the following result, which has been assumed in the
text. A brief discussion of the classification problem for higher dimen­
sional manifolds will also be given .

Theorem. Any smooth, connected I-dimensional manifold is diffeo­

morphic either to the circle 81 or to some interval of real numbers.
(An interval is a connected subset of R which is not a point . It may
be finite or infinite; closed, open, or half-open . )
Since any interval is diffeomorphic * either to [0, 1 ] , (0, 1 ] , or (0, 1 ) ,
it follows that there are only four distinct connected I-manifolds .
The proof will make use of the concept of "arc-length. " Let I be an
interval .

DEFINITION. A map t : I -----* M is a parametrization by arc-length if f

maps I diffeomorphically onto an open subset t of M, and if the " velocity
vector" df. (I ) t TMf(.) has unit length, for each s t I.
Any given local parametrization I' -----* M can be transformed into
a parametrization by arc-length by a straightforward change of variables .

Lemma. Let f : I -----* M and g : J -----* M be parametrizations by arc-length .

Then f (l) n g (J) has at most two components. It it has only one component,
then f can be extended to a parametrization by arc-length ot the union
f (l) U g (J) . If it has two components, then M must be diffeomorphic to 81•

* For example, use a diffeomorphism of the form

J(t) = a tanh (t) + b.

t Thus I can have boundary points only if M has boundary points.

56 A ppendix

PROOF. Clearly g- 1 0 f maps some relatively open subset of I diffeomor­

phically onto a relatively open subset of J. Furthermore the derivative
of g - 1 0 f is equal to ± 1 everywhere.
Consider the graph r e I X J, consisting of all (8, t) with f (8) = g (t) .
Then r is a closed subset of I X J made up of line segments of slope ± l .
Since r is closed and g - 1 0 f is locally a diffeomorphism, these line
segments cannot end in the interior of I X J, but must extend to the
boundary. Since g - 1 0 f is one-one and single valued, there can be at
most one of these segments ending on each of the four edges of the
rectangle I X J . Hence r has at most two components. (See Figure 19 .)
Furthermore, if there are two components, the two must have the same
slope.

a -

8 -
L-
y _ __ -.L.�

a b c d

Figure 1 9. Three of the possibilities for r

If r is connected, then g - 1 0 f extends to a linear map L : R -? R.

Now f and g 0 L piece together to yield the required extension
F : I U L- \J) -? f(1) U g(J) .
If r has two components, with slope say + 1, they must be arranged as
in the left-hand rectangle of Figure 19. Translating the interval
J = (-y, (3) if necessary, we may assume that 'Y = c and 15 = d, so that
a < b � c < d � a < (3 .
Now setting 8 = 27rt/ (a - a) , the required diffeomorphism
h : S1 -? M
is defined by the formula

h ( cos 8 , sin 8) = t(t) for a < t < d,

= g(t) for c < t < {3 .
Classifying 1 -manifolds 57

The image h ( 8 1 ) , being compact and open in M , must be the entire

manifold M. This proves the lemma .
PRom' OF CLA S SIFICATION TH EO RJ<� M . Any parametrization by arc­
length can be extended to one
f :/�M
which is maximal in the sense that f cannot be extended over any
larger interval as a parametrization by arc-length : it is only necessary
to extend f as far as possible to the left and then as far as possible to
the right .
If M is not diffeomorphic to 81, we will prove that f is onto, and
hence is a diffeomorphism. For if the open set f (/) were not all of M,
there would be a limit point x of f (/) in M - f (/ ) . Parametrizing a
neighborhood of x by arc-length and applying the lemma, we would
see that f can be extended over a larger interval. This contradicts the
assumption that t is maximal and hence completes the proof.
REMARKS . For manifolds of higher dimension the classification
problem becomes quite formidable. For 2-dimensional manifolds, a
thorough exposition has been given by Kerekj a,rt6 [17]. The study of
3-dimensional manifolds is very much a topic of current research.
(See Papakyriakopoulos [26] . ) For compact manifolds of dimension � 4
the classification problem is actually unsolvable. * But for high dimen­
sional simply connected manifolds there has been much progress in
recent years, as exemplified by the work of Smale [3 1] and Wall [37] .

* See Markov [19J ; and also a forthcoming paper by Boone, Haken, and Po€maru
in Pundamenta Mathematicae.
 B I B LI O GRAP HY

THE following is a miscellaneous list consisting of original sources and of recom­

mended textbooks. For the reader who wishes to pursue the study of differential
topology, let me recommend Milnor [22] , Munkres [25] , and Pontryagin [28] .
The survey articles [23] and [32] should also prove useful. For background
knowledge in closely related fields, let me recommend Hilton and Wylie [1 1],
Hu [16] , Lang [ 1 8] , de Rham [29], Steenrod [34] , and Sternberg [35] .

[ 1 ] Allendoerfer, C. B . , "The Euler number of a Riemann manifold, " A mer.

Jour. Math. 62 ( 1940) , 243-248.
[2] Apostol, T. M., Mathematical A nalysis. Reaiing, Mass. : Addison-Wesley,
1957.
[3] Auslander, L., and R. MacKenzie, Introduction to Differentiable Manifolds.
New York : McGraw-Hill, 1963 .
[4] Brouwer, L. E. J., " Uber Abbildung von Mannigfaltigkeiten", Math.
A nnalen 71 ( 1 9 1 2) , 97-1 15.
[5] Brown, A. B . , "Functional dependence, " Trans. A mer. Math. Soc. 38
( 1935) , 379-394. (See Theorem 3-III.)
[6] Chern, S . S . , "A simple intrinsic proof of the Gauss-Bonnet formula for
closed Riemannian manifolds, " A nnals of Math. 45 ( 1944) , 747-752.
[7] Dieudonne, J., Foundations of Modern Analysis. New York : Academic
Press, 1960.
[8] Dubovickil, A. Ya., "On differentiable mappings of an n-dimensional cube
into a k-dimensional cube," Mat. Sbornik N . S . 32 (74) , ( 1953) , 443-464.
(In Russian.)
[9] Fenchel, W., "On total curvatures of Riemannian manifolds, " Jour.
London Math. Soc . 15 ( 1940) , 1 5-22.
[ 1 0] Goffman, C . , Calculus of Several Variables. New York : Harper & Row,
1965.
[ 1 1 ] Hilton, P . , and S . Wylie, Homology Theory . Cambridge Univ. Press, 1960 .
 (JO Bibliograph y

[ 12] Hirsch, M., "A proof of the nonretractibility of a cell onto its boundary, "
Proc . A mer. Math. Soc. 14 ( 1963) , 364-365.
[ 13] Hopf, H . , "Abbildungsklassen n-dimensionaler Mannigfaltigkeiten, " illath.
A nnalen 96 ( 1926) , 209-224 .
[ 1 4] -- , "Vektorfelder in n-dimensionalen Mannigfaltigkeiten, " Math.
A nnalen 96 ( 1926) , 225-250.
[ 1.'5] -- , "Uber die Abbildungen von Spharen auf Spharen niedrigerer
Dimension, " Fundamenta Mathematicae 25 ( 1935) , 427-440.
[ 1 6] Hu, S . -T., Homotopy Theory . New York : Academic Press, 1959.
[ 1 7] Kerekj art6, B . v., Vorlesungen uber Topologie. Berlin : Springer, 1923 .
[ 1 8] Lang, S . , Introdliction to Differentiable illanifolds. Xew York : Interscience ,
1962.
[ 19] Markov, A. A., "Insolubility of the problem of homeomorphy, " Pro­
ceedings Intern. Congress of l1lath. 1 958, Cambridge Univ. Press, 1960,
pp. 300-306. (In Russian.)
[20] Milnor, J., "On manifolds homeomorphic to the 7 -sphere, " A nnals of
Math . 64 ( 1956) , 399-405 .
. [21] -- , "A survey of cobordism theory, " L'Enseignement math. 8 ( 1962) ,
1 6-23 .
[22] --- , Morse Theory . (Annals Studies 5 1 .) Princeton Univ. Press, 1B63 .
[23] --- , "Differential topology," Lectures o n illodern Mathematics, II, e d .
T . L . Saaty, X e w York : Wiley, 1964, pp. 165-183.
[24] Morse, A. P., "The behavior of a function on its critical set, " A nnals of
Math. 40 ( 1939) , 62-70 .
[25] Munkres, J. R., Elementary Differential Topology. (Annals Studies 54) .
Princeton Univ. Press, 1963 .
[26] Papakyriakopoulos, C. D . , "The theory of three-dimensional manifolds
since 1950, " Proceedings Intern. Congress of 1lIath. 1 958, Cambridge
Univ. Press, 1 960, pp. 433-440.
[27] Pontryagin, L. S . , "A classification of continuous transformations of a
complex into a sphere, " Doklady A kad. Nauk. S.S.S.R. (Comptes Rendues)
19 ( 1938) , 147-149.
[28] -- , "Smooth manifolds and their applications in homotopy theory, "
A mer. Math. Soc. Translations, Ser. 2, II ( 1959) , 1-1 14. (Translated from
Trudy Inst. Steklov 45 (1955) .)
[29] Rham, G. dc, Varieles differentiables. Paris : Hermann, 1955.
[30] Sard, A., "The measure of the critical points of differentiable maps, "
Bull. A mer. Math. Soc. 48 ( 1942) , 883-890.
[3 1] Smale, S . , "Generalized Poincare's conj ecture in dimensions greater than
four, " A nnals of Math. 74 ( 1961) , 391-406.
[32] -- , "A survey of some recent developments in differential topology, "
Bull. A mer. Math. Soc . 69 ( 1 963) , 13 1-145.
[33] Spivak, M., Calculus on Manifolds . New York : Benj amin, 1965.
[34] Steenrod, X., The Topology of Fibre Bundles. Princeton Univ. Press, 1 95 1 .
Bibliography 61

[ 35] Sternberg, S . , Lectures on Differential Geometry. New York : Prentice­

Hail, 1 964.
[36] Thorn, R., "Quelques proprietes globales des varietes differentiables, "
Commentarii Math. Helvet. 2/\ ( 1 954) , 1 7-86.
[37] Wall, C . T. C., " Classification of (n - I ) -connected 2n-rnanifolds, "
Annals of l\Jath. 75 ( 962) , 1 63- 189.
[ 3/\] Whitney, H . , "A function not constant on a connected set of critical
points , " Duke l\!Jath. J01ir. 1 ( 1 935) , 514-5 1 7 .
 PAGE-BARBOUR LECTURE SERIES

The Page-Barbour Lecture Foundation was founded in 1 907 by a gift from

Mrs . Thomas Nelson Page (nee Barbour) and the Honorable Thomas Nelson
Page for the purpose of bringing to the University of Virginia each session a
series of lectures by an eminent person in some field of scholarly endeavor. The
materials in this volume were presented by Professor John W. Milnor in Decem­
ber, 1 963, as the forty-seventh series sponsored by the Foundation.
 I N D EX

antipodal map, 30, 52 homotopy, 20, 21, 52

boundary, 1 2 Hopf, H., 31, 35, 36, 40, 5 1 , 54
Brouwer, L . E . J . , 35, 52 index (of a zero of a vector field) , 32-
degree, 20, 28 34
fixed point theorem, 14 index sum, 35-4 1
Brown, A . B . , 1 0, 1 1 inverse function theorem, 4, 8
chain rule for derivatives, 3, 7 inward vector, 26
cobordism, 42, 43, 5 1 isotopy, 2 1 , 22, 34
codimension, 4 2 linking, 53
cohomotopy, 50 Morse, A. P., 1 0
coordinate system, 1 Morse, M., 40
corresponding vector fields, 32 nondegenerate zero (of a vector field)
critical point, 8, 1 1 37, 40
critical value, 8, 1 1 normal bundle, 53
degree of a map, 28 normal vectors, 1 1 , 12, 42
mod two, 20 orientation, 26
derivative of a map, 2-7 of a boundary, 27
diffeomorphism, 1 of F-l (y) , 28
differential topology, 1, 59 outward vector, 26
dimension of a manifold, 1 , 5, 7 parametrization, 1, 2
disk, 13, 14 by arc-length, 55
Euler number, 35, 36, 40 Poincare, H., 35
framed cobordism, 43 Pontryagin, L., 1 6, 42
framed submanifold, 42, 44, 46 Pontryagin manifold, 43
Fubini theorem, 1 7 product neighborhood theorem, 46
fundamental theorem of algebra, 8 regular point, 7
Gauss-Bonnet theorem, 38 regular value, 8, 1 1 , 1 3, 14, 20, 27, 40, 43
Gauss mapping, 35, 38 Sard, A., 10, 1 6
half-space, 12 smooth manifolds, 1
Hirsch, M . , 14 with boundary, 12
 I n a ex

smooth manifolds (cont.) stereographic proj ection, 9, 48

the classification problem, 57 tangent bundle, 53
of dimension zero, 2 tangent space, 2-5
of dimension one, 14, 55 at a boundary point, 1 2, 26
oriented, 26 tangent vector, 2
smooth maps ( = smooth mappings) , 1 vector fields, 30, 32-4 1
sphere, 2 ·Weierstrass approximation theorem, 1 4

INDEX OF SYM BOLS

go
deg (/ ; y) , 27 t, 1
dtx , 2-7 H m, 12
D m, 14 R k, 1
ax, 1 2 Sn-l , 2
t*\), 43 TJIx, 2-5
#f-1(y) , 8 I lx l ! ' 14
TOPOLOGY FROM THE DIFFERENTIABLE VIEWPO I�T

was composed and printed by

The William Byrd Press, Inc., Richmond, Virginia
and bound by Russell Rutter Co., Inc., New York, N . Y.
The types are Modern Number 8 and Century Schoolbook,
and the paper is Oxford's Bookbuilders Plate.
Design is by Edward Foss .