Universitext

Manfredo P. do Carmo

Differential Forms
and Applications
With 18 Figures

Springer-Verlag Berlin Heidelberg GmbH

Manfredo P. do Carmo
Instituto de Matematica Pura e Aplicada (IMPA)
Estrada Dona Castorina, 110
22460-320 Rio de Janeiro
Brazii

This is a transJation of the Portuguese book "Formas Diferenciais e Apli~6es", first published by
IMPA in 1971.

Mathematics Subject Classification (1991):

53-01, 53A05, 58AIO, 58Z05, 70Hxx

ISBN 978-3-540-57618-1

Ubrary of Congress Cataloging.in-Publication Data. Carmo, ManfredoPerdiga6 do. [Fonnas

diferenciais e aplica~6es. English) Differential fonns and applications I Manfredo P. do Canno.
p. cm. -- (Universitext) Includes bibliographical references aud index.
ISBN 978-3-540-57618-1 ISBN 978-3-642-57951-6 (eBook)
DOI 10.1007/978-3-642-57951-6
1. Differential fonns.1. Title. QA381.C2813 1994 515'.37--dc20 94-21965
This work is subject to copyright. AII rights are reserved, whether the whole or part of the material
is concerned, specifically the rights of translation, reprinting, reuse of iIlustrations, recitation,
broadcasting, reproduction on microfilms or in any olber way, and storage in data banks.
Duplication of this publication or parts thereof is permitted only under the provisions of the
German Copyright Law of September 9, 1965, in its current version, aud permission for use must
always be obtained from Springer-Verlag. Violations are Iiable for prosecution under the Gennan
Copyright Law.

© Springer-Verlag Berlin Heidelberg 1994

Originally published by Springer-Verlag Berlin Heidelberg New York in 1994
The ust: of general descriptive names, registered names, trademarks, etc. in this publication does
not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective law~ and regulations and therefore free for general use.
The final art for the drawings was prepared by Manfredo do Carmo Junior.
1Ypesetting: Camera-ready by the author using a Springer TEX macro package
41/3111- 5 4 - Printed on acid-free paper
To my friends around the world, with-
out whose help neither this book nor
its author would be seeing the light.
Preface

This is a free translation of a set of notes published originally in Portuguese in

1971. They were translated for a course in the College of Differential Geome-
try, ICTP, Trieste, 1989. In the English translation we omitted a chapter on
the Frobenius theorem and an appendix on the nonexistence of a complete
hyperbolic plane in euclidean 3-space (Hilbert's theorem). For the present
edition, we introduced a chapter on line integrals.
In Chapter 1 we introduce the differential forms in Rn. We only assume
an elementary knowledge of calculus, and the chapter can be used as a basis
for a course on differential forms for "users" of Mathematics.
In Chapter 2 we start integrating differential forms of degree one along
curves in Rn. This already allows some applications of the ideas of Chapter 1.
This material is not used in the rest of the book.
In Chapter 3 we present the basic notions of differentiable manifolds. It
is useful (but not essential) that the reader be familiar with the notion of a
regular surface in R3.
In Chapter 4 we introduce the notion of manifold with boundary and
prove Stokes theorem and Poincare's lemma.
Starting from this basic material, we could follow any of the possi-
ble routes for applications: Topology, Differential Geometry, Mechanics, Lie
Groups, etc. We have chosen Differential Geometry. For simplicity, we re-
stricted ourselves to surfaces.
Thus in Chapter 5 we develop the method of moving frames of Elie Cartan
for surfaces. We first treat immersed surfaces and next the intrinsic geometry
of surfaces.
Finally, in Chapter 6, we prove the Gauss-Bonnet theorem for compact
orient able surfaces. The proof we present here is essentially due to 8.8.Chern.
We also prove a relation, due to M. Morse, between the Euler characteristic
of such a surface and the critical points of a certain class of differentiable
functions on the surface.
As most authors, I am indebted to so many sources that it is hardly
possible to acknowledge them all. Let me at least mention that the first four
VIII Preface

chapters were strongly influenced by the writings of my friend and colleague

Elon Lima and the last two chapters bear the imprint of my teacher and
friend S.S. Chern.
For the present version I am indebted to my colleagues M. Dajczer, L.
Rodriguez and W. Santos for reading critically the manuscript and offering a
number of useful suggestions. Special thanks are due to Lucio Rodriguez for
his care in the camera ready presentation of the final text.

Rio de Janeiro, February 1994. Manfredo Perdigao do Carmo

Table of Contents

Preface ..................................................... vii

1. Differential Forms in R n ..••..•••..•••.••..••.••••..•••. 1
2. Line Integrals ........................................... 17
3. Differentiable Manifolds 33
4. Integration on Manifolds; Stokes Theorem and
Poincare's Lemma ...................................... 55
1. Integration of Differential Forms ........................ 55
2. Stokes Theorem ...................................... 60
3. Poincare's Lemma .................................... 66

5. Differential Geometry of Surfaces ....................... 77

1. The Structure Equations of R n . . • • . . . . . • • • • . • . • • . • • • . • • 77
2. Surfaces in R 3 . . . . • . . . . . . • . . . . . . . . . . . . . . . . . . . • . . . • . . . 82
3. Intrinsic Geometry of Surfaces .......................... 89

6. The Theorem of Gauss-Bonnet and

the Theorem of Morse .................................. 99
1. The Theorem of Gauss-Bonnet ......................... 99
2. The Theorem of Morse ................................ 106

References .................................................. 115

Index ....................................................... 117
1. Differential Forms in R n

The goal of this chapter is to define in R n "fields of alternate forms" that

will be used later to obtain geometric results.
In order to fix the ideas, we will work initially with the three-dimensional
space R3.
Let p be a point of R 3 . The set of vectors q - p, q E R 3 (that have origin
at p) will be called the tangent space of R3 at p and will be denoted by R!.
The vectors el = (1,0,0), e2 = (0,1,0), e3 = (0,0,1) of the canonical basis
of n.g will be identified with their translates (ed p, (e2)p, (e3)p at the point p.
A vector field in R 3 is a map v that associates to each point pER3 a
vector v(P) E R!. We can write v as

v(p) = al(p)el + a2(p)e2 + a3(p)e3,

thereby defining three functions ai: R3 --+ R, i = 1,2,3, that characterize
the vector field v. We say that v is differentiable if the functions ai are
differentiable.
To each tangent space R; we can associate its dual space (R;)* which is
the set of linear maps cp: R! --+ R. A basis for (R;)* is obtained by taking
(dxi)p, i = 1,2,3, where Xi: R3 --+ R is the map which assigns to each point
its ith -coordinate. The set

{(dxi)p; i = 1,2,3}
is in fact the dual basis of {( ei) p} since

( ) ( ) aXi
dXi p ej = aXj =
{O,1, if i # j
if i = j.

Definition 1. A field of linear forms (or an exterior form of degree 1) in

R3 is a map w that associates to each p E R3 an element w(p) E (R;)*; w
can be written as

or
2 1. Differential Forms in R n

3
w = Laidxi,
i=1
where ai are real functions in R 3 • If the functions ai are differentiable, w is
called a differential form of degree 1.

Now let A2(R!)* be the set of maps <P: R! x R! - t R that are bilinear
(i.e., <p is linear in each variable) and alternate (i.e., <P~Vl.V2) = -<p(v2,vd).
With the usual operations of functions, the set A2(Rp)* becomes a vector
space.
When <P1 and <P2 belong to (R!)*, we can obtain an element <P1 1\ <P2 E
A2(R!)* by setting
(<p11\ <P2)(V1, V2) = det(<pi(vj))
The element (dXi)p 1\ (dxj)p E A2(R!)* will be denoted by (dXi 1\ dXj)p' It is
easy to see that the set {(dXi I\dxj)p, i < j} is a basis for A2(R!t (this will
be proved in a more general setting in Proposition 1 below). Furthermore,

(dXi 1\ dXj)p = -(dxj 1\ dXi)p, i '" j,

and

Definition 2. A field of bilinear alternating forms or an exterior form of

degree 2 in R 3 is a correspondence w that associates to each pER3 an
element w(P) E A2(R!)*; w can be written in the form
w(P) = a12(P)(dx1 1\ dX2)p + a13(P)(dx1 1\ dX3)p + a23(P)(dx2 1\ dX3)p
or
w = Laijdxi 1\ dXj, i,j = 1,2,3,
i<j
where aij are real functions in R 3 • When the functions aij are differentiable,
w is a differential form of degree 2.
We will now generalize the notion of differential form to R n. Let pERn ,
R; the tangent space of R n at p and (R;)* its dual space. Let Ak(R;t be
the set of all k-linear alternating maps

<P: R; x ... x R; -t R
, I

"
k times

(alternating means that <P changes signs with the interchange of two con-
secutive arguments). With the usual operations, Ak(R;)* is a vector space.
Given <Pl. ... , <Pk E (R;)*, we can obtain an element <P1 1\ <P2 1\ ... 1\ <Pk of
Ak(R;)* by setting
 1. Differential Forms in R n 3

It follows from the properties of determinants that CPl/\ CP2/\ ••• /\ CPk is in fact
k-linear and alternate. In particular (dXh )p/\(dXh)p/\, .. /\ (dXik) E Ak(R;)*,
il, i2,' .. , ik = 1, ... , n. We will denote this element by (dXil /\ dXi2 /\ ... /\
dXik)P'

Proposition 1. The set

{(dXil /\ ... /\ dXik)P' i1 < i2 < ... < ik, ij E {I, ... , n}}

is a basis lor Ak(R;)*.

Proof. The elements of the set are linearly independent. For, if

L ai 1 ... ik dXil /\ ... /\ dXik == 0,

il <... <ik

is applied to

(ejt, ... ,ejk)' jl < ... <jk, jt E {1, ... ,n},

we obtain (Exercise 2)

L ail ... ik dXil /\ ... /\ dXik (ejt, ... , ejk) == ajt ... jk == O.
il <... <ik

We now show that if I E Ak (R;) *, then I is a linear combination of the

form

For that, set

Notice that 9 E Ak(R;)* and that

for all il, ... ,ik. It follows that I == g. Setting I(eip ... ,eik) == ail ... ik' we
obtain the above expression for f. 0

Definition 3. An exterior k-Iorm in R n is a map w that associates to each

p ERn an element w(p) E Ak(R;)*; by Proposition 1, w can be written as

w(P)= L ait ... dp)(dxil/\ .. ·/\dxik)P' i j E{I, ... ,n},

il <... <ik
4 1. Differential Forms in R n

where ah ... i" are real functions in R n. When the ail ... ;" are differentiable
functions, w is called a differential k-form.
For notational convenience, we will denote by I the k-upla (i l , ... ,i k ),
i l < ... < ik, ij E {I, ... , n}, and will use the following notation for w:

w = La/dx/.
/

We also set the convention that a differential O-form is a differentiable func-

tion f: R n --+ R.

Example 1. In R4 we have the following types of exterior forms (where ai, aij,
etc., are real functions in R4):
O-forms, functions in R 4 ,
I-forms, a1dxl + a2dx2 + a3dx3 + a4dx4,
2-forms, al2dxl /\ dX2 + al3dxl /\ dX3 + al4dxl /\ dX4 + a23dx2 /\ dX3 +
a24dx2 /\ dX4 + a34dx3 /\ dX4,
3-forms, al23dxl /\dX2 /\dX3 +al24dxl /\dX2 /\dX4 +al34dxl /\dX3 /\dX4 +
a234dx2 /\ dX3 /\ dX4,
4-forms, a1234dxl /\ dX2 /\ dX3 /\ dX4.
From now on, we will restrict ourselves to differential k-forms and we will
call them simply k-forms.
We are going to define some operations on k-forms in Rn.
First, if wand <p are two k-forms:

we can define their sum

w + <p = L(a/ + b/)dx/.

/

Next, if w is a k-form and <p is an s-form, we can define their exterior product
w /\ <p, which is an (s + k)-form, as follows.

Definition 4. Let

<p = LbJdxJ, Jt < ... < ]8'

By definition,
w /\ <p = L a/bJdx/ /\ dXJ.
/J
 1. Differential Forms in R n 5

Example 2. Let w = xldxl + x2dx2 + X3dx3 be a I-form in R3 and cP =

Xldxl 1\ dX2 + dXl 1\ dX3 be a 2-form in R3. Then, since dXi 1\ dXi = 0 and
dXi 1\ dXj = -dxj 1\ dXi, i '" j, we obtain

w 1\ cP = x2dx2 1\ dXl 1\ dX3 + X3Xl dX3 1\ dXl 1\ dX2

= (XlX3 - X2)dxl 1\ dX2 1\ dX3.

Remark 1. The definition of exterior product is made in such a way that if

CPl, ••• ,CPk are I-forms, then the exterior product CPl 1\ ... /\ CPk agrees with
the k-form previously defined by

This follows immediately from the definition and will be left as an exercise
(Exercise 3).
The exterior product of forms in R n has the following properties.

Proposition 2. Let w be a k-form, cp be an s-form and () be an r-form.

Then:
a) (w 1\ cp) 1\ () = w 1\ (cp 1\ ()),
b) (w 1\ cp) = (_I)k8(cp 1\ w),
c) wl\(cp+())=w/\cp+w/\(), ifr=s.

Proof. (a) and (c) are straightforward. To prove (b), we write

it < ... < js·

Then
w /\ cp =L a1bJdxil /\ ... /\ dXik /\ dXjt /\ ... /\ dXj.
IJ
= L bJaI( -l) dxi l /\ ... /\ dXik_l 1\ dXit 1\ dXik 1\ ... /\ dXj.
IJ

= L bJaI( -I)kdxit /\ dXi 1 /\ ••• /\ dXik 1\ dXh /\ ... /\ dXj •.

IJ
Since J has 8 elements, we obtain, by repeating the above argument for
each dXJt,jl E J,

w /\ cP = L bJaI( -1 )k8 dXjl /\ ... /\ dXj. /\ dXil /\ ... /\ dXik

JI
= (_l)kscp/\w. 0
6 1. Differential Forms in R n

Remark 2. Although dXi A dXi = 0, it is not true that for any form w Aw = O.
For instance, if

then
w A w = 2XIX2dxI /\ dX2 A dX3 A dX4.
See however Exercise 4.

One of the most important features of differential forms is the way they
behave under differentiable maps. Let f: R n -+ R m be a differentiable map.
Then f induces a map J* that takes k-forms in R m into k-forms in R n and is
defined as follows. Let w be a k-form in Rm. By definition, J*w is the k-form
in R n given by

Herep ERn, VI, ... ,Vk E R;, and dfp:R; -+ R/(p) is the differential of the
map f at p. We set the convention that if 9 is a O-form,
J*(g)=goJ.
We are going to show that the operation J* on forms is equivalent to
"substitution of variables". Before that, we need some properties of J*.

Proposition 3. Let f: R n -+ R m be a differentiable map, wand cP be k-forms

on R m and g:Rm -+ R be a O-form on Rm. Then:
a) J*(w + cp) = J*w + J*ep,
b) J*(gw) = J*(g)J*(w),
c) If CPI, ... , CPk are 1-forms in R m , J*(CPl A ... /\ CPk) = J*(cpd A ... A
J*(CPk).

Proof. The proofs are very simple. Let pERn and let VI, •.• , Vk E R;.
Then
(a) J*(w + CP)(P)(VI, ... , Vk) = (w + cp)U(p))(dfp(vd, .. ·, dfp(Vk)) =
U*W)(P)(VI"'" Vk) + U*cP)(p)(vI, ... , Vk) = U*w+ J*cp)(p)(vI, ... , Vk).
(b) J*(gw)(P)(vI, ... , Vk) = (gw)U(p))(dfp(Vl), .. " dfp(Vk)) = (g 0 J)(p) .
J*w(p)(Vl, ... , Vk) = J*g(p) . J*w(p)(Vl, ... , Vk).
(c) By omitting the indication of the point p, we obtain

J*(CPI A ... A CPk)(VI"", Vk) = (CPI A ... A CPk)(df(Vl), ... , df(Vk))

= det(cpi(df(vj)) = detU*CPi(Vj))
= U*CPl /\ ... /\ J*CPk)(V17 ... , Vk).

Remark 3. We will show below (See Proposition 4) that (c) holds not only
for 1-forms but for k-forms as well.
 1. Differential Forms in R n 7

We can now present the promised interpretation of r.

Let (Xl, ... , xn) be
coordinates in R n, (Yt. ... , Ym) be coordinates in R m and let f: R n -+ R m
be written as

Let w = L.1a[dy[ be a k-form in Rm. By using the above properties of r,

we obtain
rw = ~r(aI)(J*dYiJ /\ ... /\ (J*dYik)'
[

Since
!*(dYi)(V) = dYi(df(v)) = d(Yi 0 f)(v) = d/;(v),
we have
rw = ~ a[(h(xI,"" xn), ... , fm(Xl,"" xn))dh /\ ... /\ d/;k'
[

where /; and d/; are functions of Xj. Thus to apply r

to w is equivalent to
"substitute" in w the variables Yi and their differentials by the functions of
Xk and dXk obtained from (*).

Remark 4. In various situations, it is convenient to use differential forms

defined only on some open set U eRn and not on the entire R n. It is clear
that everything done so far extends trivially to this situation.

Example. (Polar coordinates). Let w be the I-form in R2 - {O,O} by

Y x
W=- 2 2dx + 2 2dy.
x +y x +y
Let U be the set in the plane (r, ()) given by
U = {r > 0; 0 < () < 27r}
and let f: U -+ R2 be the map

f(r ()) = {x = rcos()

, Y = rsen()

Let us compute rw. Since

dx = cos ()dr - r sen ()d(),
dy = sen ()dr + r cos ()d(),
we obtain
• r~() r~()
f w= - - - 2- (cos Odr
r
- r sen OdO) + - -r2- (sen Odr + r cos OdO)
= dO.
8 1. Differential Forms in R n

Notice that (a) of Proposition 3 states that the addition of differential

forms commutes with the "substitution of variables". We will now show that
the same holds for the exterior product.

Proposition 4. Let f: R n - t R m be a differentiable map. Then

(a) f*(w 1\ cp) = U*w) 1\ U*cp), where wand cp any two forms in Rm.
(b) U 0 g)*w = g*U*w), where g: RP - t R n is a differentiable map.

Proof. By setting (YI,. .. , Ym) = (It (Xl!' .. , xn), . .. , fm(xl!' .. , Xn)) E Rm,
(Xl! ... ,Xn) ERn, W = "'r.-rardYr, cp = "'r.-J bJdYJ, we obtain

f*(w 1\ cp) = f*(LarbJdYI 1\ dYJ)

rJ
= L ar(It, ... , fm)bJ(It,···, fm)dfr 1\ dh
rJ
= L ar(It,···, fm)dfr 1\ L bJ(It, ... , fm)dh
r J
= rw 1\ rcp.

b) U 0 g)*w = "'r.-r ar(U 0 gh,·.·, U 0 g)m)dU 0 g)r

= "'r.-r ar(ft(gl, . .. , gn), . .. , fm(gl, . .. ,gn»dfr(dg1 , • •• ,dgn)
= g*U*{w». 0
We are now going to define an operation on differential form that gen-
eralizes the differentiation of functions. Let g: R n - t R be a O-form (i.e., a
differentiable function). Then the differential
n 0
dg= LJ...dxi
i=1 OXi

is a I-form. We want to generalize this process by defining an operation that

takes k-forms into {k + I)-forms.

Definition 5. Let w = "'r.-ardxr be a k-form in Rn. The exterior differential

tkv of w is defined by

Example 4. Let w = xyzdx + yzdy + (x + z)dz and let us compute tkv:

tkv = d{xyz) 1\ dx + d{yz) 1\ dy + d(x + z) 1\ dz
= (yzdx + xzdy + xydz) 1\ dx + (zdy + ydz) 1\ dy + (dx + dz) 1\ dz
= -xzdx 1\ dy + (1 - xy)dx 1\ dz - ydy 1\ dz.
 1. Differential Forms in R n 9

We now present some properties of exterior differentiation. Item (c) is

probably the most important one and item (d) means that the operation d
commutes with substitution of variables.

Proposition 5.
a) d(WI + W2) = dwl + dw2, where WI and W2 are k-forms
b) d(w /\ <p) = dw /\ <p+ (_l)kw /\ d<p, where W is a k-form and <p is an s-form
c) d(dw) = d2w = O.
d) d(f*w) = f*(dw), where W is a k-form in R m and f: R n --+ R m is a
differentiable map.

Proof.
(a) is straightforward.
(b) Let W = '£r ardxr, <p = '£J bJdxJ. Then
= L d(arbJ) /\ dXr /\ dXJ
rJ
= L bJdar /\ dxr /\ dXJ + L ardbJ /\ dxr /\ dXJ
rJ rJ
= dw /\ <p + (_l)k L ardxr /\ dbJ /\ dXJ
rJ
= dw /\ <p + (_l)kw /\ d<p.
(c) Let us first assume that w is a O-form, i.e., w is a function f: R n --+ R
that associates to each (Xl, ... ,Xn ) ERn the value f(xl> ... ,xn ) E R. Then

·
Smce -.!f!.L - -.!f!.L
{}x;{}Xj - {}Xj{}x; an
d dXi /\ dXj -- - dXj /\ dXi, ~. r),
-I- . we 0 b ' t h at
tam

Now let w = ,£a/dx/. By (a), we can restrict ourselves to the case w =

ardxr with a[ =I O. By (b), we have that

dw = dar /\ dxr + ard(dx/).

But d(dx/) = d(l) /\ dx[ = O. Therefore,
10 1. Differential Forms in R n

d(dw) = d(da[ 1\ dx[) = d(da[) 1\ dx[ + da[ 1\ d(dx[) = 0,

°
since d(da[) = and d(dx[) = 0, which proves (c).
(d) We will first prove the result for a O-form. Let g: R m ~ R be a dif-
ferentiable function that associates to each (YI. ... , Ym) E R m the value
g(YI,"" Ym). Then

!*(dg) =!* (E . 0Yi

,
og dYi ) = E
'1
og oli dXj
.. 0Yi oxj
~ o(g 0 J)
= ~ O. dXj = d(g 0 f) = dU * g).
j Xl

Now, let <p = L:[ aldx[ be a k-form. By using the above, and the fact
that !* commutes with the exterior product, we obtain

dU*cp) = d(E !*(a[ )!*(dx[»

[

= EdU*(aI»I\!*(dx[» = E!*(daI)I\!*(dxI)
[ I

= !*(E daI 1\ dXI) = !*(dcp)

[

which proves (d). o

In the exercises that follow we will often use the canonical isomorphism be-
tween R; and its dual (R;)* that is established by the natural inner product
( , ) of R n. We recall that if {ei} is the canonical basis of R n and VI = L: aiei,
V2 = L: bi ei belong to (Rn ) p, then (VI, V2) = L: ai bi. The above canonical
isomorphism takes a vector V E R; to an element W E (R;)* given by
w(u) = (v,u), for all u E R;.
If we let the point p vary, this establishes a
one-to-one correspondence between vector fields in R n and exterior I-forms
in Rnj it is easily seen that this correspondence takes differentiable vector
fields into differential I-forms and conversely.

EXERCISES

1) Prove that a bilinear form cp: R 3 X R 3 ~ R is alternate if and only if

cp(v, v) = 0, for all V E R3.
2) Prove that if i 1 < i2 < ... < ik and jl < j2 < ... < jk, then
ifil =JI, ... ,ik =jk,
otherwise.
 1. Differential Forms in R n 11

Hint: Consider the determinant Q = IdXit(ejJI. If il > JI, then ik >

... > i2 > il > JI, that is, dXit(ej}) = 0, for e= 1, ... , k. If i 1 < JI, then
jk > .,. > h > jl > il, that is, dXi1 (ejJ = 0, for n = 1, ... , k. In any
case, Q = 0 if il :/= JI. Assume now that il = JI and i2 :/= h Proceeding
as before, we show that Q = O. The argument can be easily continued.
3) Prove the statement of Remark 1.
4) Let cp be an exterior k-form, where k is an odd integer. Show that cp/\cp =
O.
5) Let cp,1/1 and () the following forms in R3:

cp = xdx - ydy,
1/1 = zdx /\ dy + xdy /\ dz,
() = zdy.
Compute: cp /\ 1/1, () /\ cp /\ 1/1, dcp, d1/1, d(}.
6) Let f: U C R m -+ R n be a differentiable map. Assume that m < nand
let w be a k-form in Rn, with k > m. Show that j*w = O.
7) Let w be the 2-form in R 2n given by

w = dXl /\ dX2 + dX3 /\ dX4 + ... + dX2n-l /\ dX2n'

Compute the exterior product of n copies of w
8) Let f: R n -+ R n be a differentiable map given by

f(Xb ... , xn) = (Yb . .. , Yn),

and let w = dYI /\ ... /\ dYn' Show that

j*w = det(df)dxl /\ ... /\ dx n.

9) Let v the n-form in R n defined by

v(el, ... ,e n ) = 1,
where {e;}, i = 1, ... , n, is the canonical basis of R n. Show that:
a) If Vi = E aijej, then
V(Vl, ... , Vn) = det(aij) = vol(Vl, ... , Vn).
(the form v is called the volume element of R n)
b) v = dXl/\"'/\ dxn.
10) (Hodge star operation). Given a k-form w in R n we will define an (n- k)-
form *w by setting

*(dXil /\ ... /\ dXik) = (-1t(dxh /\ ... /\ dXjn_k)

12 1. Differential Forms in R n

and extending it linearly, where il < ... < ik, jl < ... < jn-k,
(il, ... ,ik,jl, ... ,jn-k) is a permutation of (I,2, ... ,n), and a is 0 or
1 according to the permutation is even or odd, respectively. Show that:
a) If w = a12dxl 1\ dX2 + a13dXl 1\ dX3 + a23dx2 1\ dX3 is a 2-form in R 3 ,
then
*w = a12dx3 - a13dx2 + a23dxl,
b) If w = a1dxl + a2dx2 is a I-form in R2, then

c) **w = (_I)k(n-k)w.
11) (The divergence). A differentiable vector field v in R n may be consid-
ered as a differentiable map v: R n - t R n. We will define a function
div v: R n - t R (to be called the divergence of v) as follows:
(div v)(P) = trace(dv)p, p ERn,
where (dv)p: R; - t R; is the differential of vat p. Show that:
a) If v = L: aiei, where {eil is the canonical basis of Rn, then

d. ,",oai
lV v = L...., OXi .

b) If w denotes the differential I-form obtained from v by the canonical

isomorphism induced by the inner product ( , ) and v is the volume
element of R n (Exercise 9), the divergence can be obtained as follows:
v --t W --t *w --t d(*w) = (div v)v,
where we have used the star operation introduced in Exercise 10.
12) (The gradient). Given a differentiable function I: R n - t R, define a vec-
tor field grad I in R n (the gradient of J) by
(grad I(p), u} = dlp(u), for all p ERn and all U E R;.
Notice that grad I is the vector field corresponding to the I-form dl in
the canonical isomorphism. Show that:
a) In the canonical basis {eil of R n:

grad 1= L :!i ei·

b) If pERn is such that grad I (p) :I 0, then grad I (P) is perpendicular
to the "level surface" {q ERn; I(q) = I(p)}.
c) The linear map dIp: R; - t R restricted to unit sphere with center at
p reaches its maximum for v = grad 1/ Igrad II.
13) (The Laplacian). Given a differentiable function I:R n - t R, we will
define the Laplacian !::./: R n - t R by
 1. Differential Forms in R n 13

b.f = div(grad f).

Show that:
",{Pf
a } l::"f = L..JP' ,
b} l::,,(fg} = fl::"g + gl::"f + 2(gradf,gradg),
c} d * (df) = (l::"f)v, where v is the volume element of Rn.
14) (The rotationaO. Let v be a differentiable vector field in R n. The rota-
tional rot v is the (n - 2}-form defined by:

v ~ w --+ dw --+ *(dw) = rot v,

where v 1-+ W is the correspondence between I-forms and vector fields
induced by the inner product.
a) Prove that rot (grad f) = O.
b} In the particular case when n = 3, the I-form rot v corresponds to a
vector field which is also denoted by rot v. Show that, for n = 3:
bl}

b2} div (rot v) = O.

15} (A geometric definition of the * operation). An element cP E Ak(R;}*
is called decomposable if cP = CPI /\ ••• /\ CPk, where CPi, i = 1, ... , k, are
linearly independent elements of AI(R;)* ~ (R;)*. Prove that:
a) Ifcpi = Eiaiif3i, i = 1, ... ,k, f3i E (R;)*, and det(aii) = 1, then
CPI/\ •• . /\ CPk = 131/\" ./\ 13k; thus, a decomposable element may have
more than one representation.
b) If CPI /\ ..• /\ CPk = f3I /\ ... /\ 13k = cP are two representations of cP, then
CPi = Eiaiif3i, with det(aii) = 1.
Hint: Extend the f3i into a basis f3I.' .. ,f3k, f3k+l,' .. f3n of (R;)* and
write

CPi = L aijf3j + L bi7)f37) , fJ = k + 1, ... ,n.

j 7)

Notice that 131 /\ •.. /\ 13k /\ CPi = CPI /\ ... /\ CPk /\ CPi = O. This implies
that
L bi7)f3I /\ ... /\ 13k /\ (37) = 0,
7)

and since 131 /\ ... /\ 13k /\ f3T/ are linearly independent, biT/ = O.
14 1. Differential Forms in R n

c) If 1,0 = 1,01/\ ... /\l,Ok is decomposable, the vectors VI, . .. ,Vk E R;,where
Vi +-+ l,Oi is the correspondence induced by the inner product ( , ) of
R n , are linearly independent, and the subspace of R n generated by
them does not depend on the representation of 1,0. This will be called
the subspace of 1,0.
d) If 1,0 = 1,01/\' .. /\ I,Ok is decomposable, the k-volume of the solid gener-
ated by VI,' .. ,Vk does not depend on the representation of 1,0. This
will be called the volume of 1,0.
e) If 1,0 = 1,01 /\ ••• /\ I,Ok is decomposable, define *1,0 as an element of
/\n-k(R;)* with the following properties:
i) the subspace of *1,0 is perpendicular to the subspace of 1,0.
ii) the volume of *1,0 is equal to the volume of 1,0.
iii) 1,0 /\ *1,0 is positive, i.e., its value in a positive basis ofR; is positive.
Prove: That *1,0 is well defined.
Hint: Let Vi +-+ l,Oi, i = 1, ... , k, as in (c) and let W be the subspace
of R; generated by the v~s. In W 1. consider an orthonormal basis
ek+1, ... ,en so that the basis VI, ... ,Vr , ek+1, ... ,en of R; is positive.
Define 1,0j E (R;)*, j = k + 1, ... ,n, by the correspondence 1,0j +-+ ej
mentioned in (c), and let). > 0 be the volume of W. Check that
).1,Ok+1. /\ ••. /\ I,On satisfies (i), (ii) and (iii).
f) Let VI, V2 be two vectors in R 3 and let 1,01 +-+ VI, 1,02 +-+ V2 be the one-
forms given by the correspondence in (c). Define the vector product
VI x V2 +-+ *(1,01 /\ 1,02) and describe geometrically the vector VI x V2.
g) A k-form w in R n is decomposable if w(P) is decomposable for each
pERn. Every k- form in R n is a linear combination of decomposable
k-forms of the type dXil /\ ... /\ dXik' Show that with the above defini-
tion, *(dXil /\ ... /\ dXik) gives the same expression as in Exercise 10.
16) (Poincare's lemma for I-forms). Let

w = a(x, y, z)dx + b(x, y, z)dy + c(x, y, z)dz

be a differentiable I-form in R3 such that dw = O. Define f: R3 -+ R by

!(x, y, z) = 10 1 (a(tx, ty, tz)x + b(tx, ty, tz)y + c(tx, ty, tz)z) dt.
Show that df = w (This means that the condition dw = 0, which is
satisfied when w = df, is in R3 also sufficient for the existence of an !
such that w = df. We have used R 3 for notational convenience but the
result holds for R n. Also, if w is only defined in an open set U eRn,
the result still holds in a neighborhood of each p E U).
Hint: Notice that dw = 0 implies that
ab aa ac aa ab ac
ax = ay' ax = az' az = ay
and use the identity
 1. Differential Forms in R n 15

(I d
a(x, y, z) = Jo dt (a(tx, ty, tz)t)dt

= 11 a(tx, ty, tz)dt + 11 t(alX + a2Y + a3z)dt,

where aI, a2 and a3 denote the partial derivatives of a(tx, ty, tz) relative
to the first, second and third argument, respectively.
17) We say that a differentiable vector field v defined in an open set U eRn
derives locally from a potential if for each p E U then exists a neighbor-
hood V :3 p, V C U, and a differentiable function g: V -+ R (to be called
the potential) such that v = grad g.
a) Let v be as above and let w be the I-form corresponding to v, i.e.,
w(u) = (v,u), for all u ERn. Show that v derives locally from a
potential if and only if dw = 0.
Hint: Use the local form of Exercise 16.
b) Show that v derives locally from a potential if and only if rot v = 0.
c) Let v be the vector field (electric attraction):
1
v(P) = - (x2 + y2 + z2)3/2 (x, y, z), P E R3 - {O,O,O}

Show that v derives locally from a potential g:

I
9 = (2
x +y +z 2)1/2
2 + const.
and that t::,g = 0.
18) A function 9 : R3 -+ R is said to be homogeneous of degree k if
g(tx,ty,tz) = tkg(x,y,z), t > 0, (x,y,z) E R3. Prove that:
a) If 9 is differentiable and homogeneous of degree k, then (Euler's re-
lation)
xgx + ygy + zgz = kg.
Hint: Differentiate g(tx, ty, tz) = tkg(x, y, z) in t and set t = 1.
b) If the differential form
w = adx + bdy + cdz
is such that a, band c are homogeneous of degree k and dw = 0, then
w = d/, where
/ = xa + yb + zc

°
k+I
Hint: Notice that dw = implies that
ab aa ac aa ab ac
ax = ay' ax = az' az = ay'
and apply Euler's relation
16 1. Differential Forms in R n

c) If the differential form

(7 = ady /I. dz + bdz /I. dx + cdx /I. dy

is such that a, b and c are homogeneous of degree k and d(7 = 0, then
(7 = d"Y, where

(zb - yc)dx + (xc - za)dy + (ya - xb)dz

"Y= k+2 .
 2. Line Integrals*

Differential forms are to be integrated. We will do that soon (Chapter 4) after

some preliminaries on the natural "habitat" of differential forms (Chapter 3).
However, the special case of integration of forms of degree one along curves
(the so called line integrals) is so simple that it can be treated independently
of the general theory. We will do that in this cha.pter.
Although we restrict ourselves to curves in R n, proofs are organized in
such a way that will be valid in a more general setting to be considered later.
Let w = L aidxi be a differential form of degree one defined in an open
set U C R and let c: [a, bJ ~ U be a piecewise differentiable curve in U;
n

we recall that c is piecewise differentiable if c is continuous and there ex-

ists a partition a = to, h, ... ,tic, t1c+l = b of [a, bJ such that the restriction
cl[tj, tj+lJ = Cj is differentiable, j = 0,1, ... ,k. Notice that in each interval
[tj, tj+lJ, cjw is a form in R given by

cjw = L. ai(xl (t), .. . ,xn(t» ddxit dt.

I

where c(t) = (Xl(t), ... , xn(t». Define

1 c(t)
w= Li
j tj
ti+l
cjw = lb (L
a i
dx')
ai(t)-d• dt.
t

A change of parametrization of c: [a, bJ --t U is a differentiable homeomor-

phism cp: [c, dJ --t [a, bJ. We say that cp preserves orientation if cp is increasing;
otherwise, it reverses orientation. If t = cp(T) and cp is increasing, we obtain,
using the formula of change of variables for integrals,

* This chapter is not used in the rest of the book and can be omitted on a first
reading.
18 2. Line Integrals

which shows that 1w is invariant by a change of parametrization that pre-

serves orientation; similarly, if c.p changes orientation, 1 w changes sign.

We will denote by c the trace of c(t) with a given orientation and by -c
Ie
the same curve with the opposite orientation. Then W is well defined and

1w=-/w.
-e e

We will say that w is closed if dw = 0, and that w is exact in V c U if

there exist a differentiable function f: V --. R such that w = df in V. Notice
that if w is exact in V, and c: [a, bJ --. V is a curve,

1 = 1 = lb
w df c*(df) = f(c(b)) - f(c(a)),

that is, 1 w depends only on the end points of c. It also follows that if w is

exact in V and c is a closed curve in V, then 1 w = 0.

Actually, these three properties are equivalent. From now on, w =
L ajdxj is a differential form defined in an open set U eRn.
Proposition 1. The following are equivalent:
1) w is exact in a connected open set V C U.

2) 1 w depends only on the end points of c for all c C V.

9) 1 w = 0, for all closed curves c C v.

Proof. We have already shown that (1) :::} (2) :::} (3). That (3) :::} (2) is
immediate. It remains to show that (2) :::} (l).

1
Let us assume (2) and fix a point p E V. For each x E V, let c be a
piecewise differentiable curve joining p to x. Define f: V --. R by f(x) = w.
By (2), f is well defined. We claim that df = w, which will conclude the proof.
Since df = '" ~f dXi,
L....- uX'
we must show that if(x)
x,
= ai(x), i = 1, ... , n.
i '
Let ei = (0, ... ,0,1,0, ... ,0), where 1 is in i-th place, and consider the curve

Ci:t--.x+tei, tE(-£,£),
that joins x to x + tei and is contained in V for t small. Then
 2. Line Integrals 19

~f (x) = t--+o
UXi
lim ~{J(x + tei) -
t
f(x)}

= 1Im-
· 1{1 w- 1} = li
w m-11 w

lit
t--+O t c+c, c t--+O t c,

= lim -
t--+O t 0
ai(s)ds = ai(O) = ai(X).
o
Example 1. Consider the form

wo = - 2y
x +y
2dx + 2 2dy
x
x +y
defined in U = R2 - {(O, On. A direct computation shows that dwo = O.
Rather than doing this computation, we prefer a geometric approach. Choose
a half-line L issuing from the origin 0 = {(O, On and consider polar coordi-
nates (p, (J) in R2 -L. Since x = pcos((J+(Jo), y = psin((J+(Jo), we obtain that
wo = d(J in R2 - Lj here (Jo is the angle from Ox to L. Since L is arbitrary,
dwo = d2(J = 0, hence wo is closed. In addition, this shows that the form wo
is locally (that is, in a neighborhood V of each point of U) the differential of
-+
a function (J that measures the positive angle of Op with L, p E V. We call (J
an angle function and wo the element of angle (with respect to the origin 0).
Although the form wo is closed and locally exact, it is not exact in R2 -
{O}. To see this, consider the unit circle c(t) = (cos t, sin t), t E [0,211"], around
the origin. Then c: [0,211"] -- U, and

1 c(t)
wo =1
0
2w c·Wo = 12w dt = 211".
0

It follows by (3) of Proposition 1 that Wo is not exact in R2 - {O}. This means

that it is not possible to patch together the various an~le functions locally
defined to make an angle function globally defined in R - {O}.

Remark 1. (The winding number) Although Wo is not exact in R2 - {O}, it

determines, along a given curve 'Y: [0, 1] -- R2 - {O}, a well defined "angle
function" cp(t), t E [0,1], given by

cp(t) = 11 xy' - yx'

2
o x +y
2 dt + CPo, 'Y(t) = (x(t), y(t».
Setting
x Y
a(t) = b(t) =
Jx2 +y2
(t),
Jx2 +y2 (t),
it is easily checked that 'Y·wo = (ab' - ba')dt, where a2 + b2 = 1. It can be
shown that if CPo is such that
20 2. Line Integrals

cos CPo = a(O), sin CPo = b(O),

then cos cp(t) = aCt), sin cp(t) = bet) (for details, see M. do Carmo [de],
Lemma 1, p. 250, or Lemma 5 of Chapter 5 below). Thus cp(t) is a continuous
determination of the angle that -y(t) makes with -y(to). If -y is closed (i.e.,
-y(1) = -y(0», then cp(1) =f cp(O), but since coscp(1) = coscp(O) and sincp(I) =
sin cp(O) , we obtain that cp(I) - cp(O) is an integral multiple of 211'. This integer
is usually called the winding number of -y around O.

The fact that the closed form Wo in Example 1 is locally exact is a general
fact. More precisely.

Theorem 1. (Poincare's Lemma for 1-forms) Let w = L

aidxi be defined
in an open set U eRn. Then dw = 0 if and only if for each p E U there is a
neighborhood V C U of p and a differentiable function f: V --+ R with df = w
(i.e., w is locally exact) .
Proof. If w is locally exact, clearly dw = O. Let us now assume that dw = O.
For simplicity of notation, let us restrict "'.lrselves to the case where w =
adx + bdy + cdz is defined in U C R3. For· :1 p E U, let B(P) be a ball
of center p = (xo, Yo, zo) contained U.: or each q E B(p), q = (x, y, z), let
(3(t) = p + t(q - p), t E [0, IJ, be the line that joins p to q. Since B(p) is a
ball, (3(t) C B(p). Define

f(q) = {
J(jet)
w= 11
0
{a«(3(t»(x - Xo) + b«(3(t»(y - Yo) + e«(3(t»(z - zo)}dt

We want to show that df = w, that is,

of of of
ax (q) = a(q), oy (q) = b(q), oz (q) = c(q).
To see this, notice that the condition that dw = 0 is equivalent to:
oa ob oa oc
oy = ax' oz = ax'
Let us consider first the case ~ = a. Differentiating f and using the first
two identities above, we obtain
of {I {oa ob oe }
ax (q) = Jo ax t(x - xo) + a + ax t(y - Yo) + ax t(z - zo) dt =

= 1
o
1 {(oa
ax oy
oa - zo) ) t + a } dt =
oa - Yo) + -(z
-(x - xo) + -(y
oz

= 11 { (:t(a({3(t)))) t + a} dt =
{I d
= Jo dt (a«(3(t»t)dt = a«(3(I» = a(q).
 2. Line Integrals 21

In a similar way, we can prove that

of of
oy (q) = b(q), oz (q) = c(q).

This completes the proof of Theorem 1. o

One of the interesting applications of Theorem 1 is to extend the definition

of the integral of a closed form to curves that are merely continuous. To do
that, observe first that if w is defined in U c R n is closed and c: [0, I] -+ U
is a differentiable curve, we can choose a partition

°= to < tt < ... < tk < tk+l =1

of [0,1] in such a way that the restriction Cl[ti' ti+1] = Ci, i = 0,1, ... ,k, is
contained in a ball Bi where w is exact, i.e., there exists a function Ii: Bi -+ R
with dJ; = w. Then

1 1;
w= L
•
w = L[J;(ti+d - J;(ti)].
•
(1)

1
Now, if c: [0, I] -+ U is only continuous, such a partition still exists and
we define w by (1).
We must show that this is independent of the chosen subdivision. This is
a standard argument and we will present it here for completeness.
Given a partition P, a refinement of P is a new partition obtained from
P by adding new points to it. If we add a point t' E (ti' ti+1), we have that
c(t) E B j • Since
(Ji(ti+d - J;(t')] + (Ji(t') - !i(ti)] = [J;(ti+d - !i(ti)],
the value of the integral for this new partition does not change. It follows
that the integral remains the same under refinements of a given partition.
Now, given two distinct partitions, we can form a third partition, by adding
to the first all the points of the second. This is a common refinement, the
integral of which must be equal to the integrals of the first and of the second
partitions, which are therefore equal. This shows the required independence.
Theorem 1 also raises the following question. We know that a closed form
is locally exact. When is it globally exact? We know that we need some
restriction on the domain of the closed form, since by Example 1, the form
element of angle is closed in R2 - {o} but it is not exact there.
The answer to this question will depend on how the integral of a closed
form along a continuous curve varies when we deform the curve continuously.
To make this notion precise, we introduce the notion of homotopy.
22 2. Line Integrals

Definition 1. Two continuous curves CO,Cl: [a,bj-+ U c R n with the same

end points Co (a) = Cl (a) and Co (b) = Cl (b) are homotopic if there exists a
continuous map

H: [a, bJ x [O,IJ -+ U, (s, t) E [a, bJ x [O,IJ,

such that:
H(s,O) = co(s), H(s, 1) = Cl(S), (2)
H(a, t) = co(a) = cl(a), H(b, t) = co(b) = cl(b). (3)
Thus the homotopy H(s, t) = Ht(s) is a continuous family of curves
parametrized by t E [O,IJ that deforms the curve Ho(s) = co(s) into the
curve Hl(S) = Cl(S) (condition (2)), by keeping fixed the end points Ht(a)
and Ht(b) (condition (3)).
Sometimes it is convenient to drop condition (3) in the definition of ho-
motopy and to let the end points vary; in this case, we say that H is a free
homotopy between Co and Cl.
It turns out that line integrals of closed forms are invariant under homo-
topies. More precisely.

Theorem 2. Let w be a closed 1-)<-1171 ... ejined in an open set U eRn. Let
CO,Cl: [a,bJ -+ U be two continuous hutnotopic curves in U. Then

1w=l w.
Co Cl
(4)

Proof. Since dw = 0, w is locally exact. Let H be a homotopy between

Co and Cl. and let {Bi} be a covering of H([a, bJ x [0,1]) C U by balls Bi
where w is exact. Since [a, bJ x [O,IJ = R is compact, the covering of R by
{H-1(B i )} = Wi has a Lebesgue number d (that is, each subset of R with
diameter < d is contained in some Wi). Divide R into subrectangles Rjk by
the lines s = const., t = const., in such a way that the diameter of each
Rjk is smaller than d. By local exactness, r
J8Rjk
w = 0. Therefore, if Rjk has
sides Ojk, fJj,k+l' OJ+l,k' fJjk, with the orientations given by increasing sand
increasing t, we obtain (Fig. 2.1).

0=2:: r w=2::{j w+ r w-j - r w}.

jk J 8Rjk jk Q;k J{3j.k+1 Qj+l.k J{3;k

But the sides of Rjk that are interior to R appear twice with opposite orien-
tations in the above sum. So the corresponding integrals cancel out, and we
have
 2. Line Integrals 23

O!j+l,k
t = tj+ 1
t
/3jk Rjk /3j,Hl

t = tj
1

o a b 8

Fig. 2.1

o= 1 + J[ -1
co
w
f3b
w
Cl
w - [ w,
Jf3a
((5))

where /3a is the curve H(a, t) and /3b is the curve H(b, t). Since these curves
reduce to points, the corresponding integrals vanish, and we are left with

1w-j w
Co - Cl '

as we wished. o
We should observe that if Co and Cl are closed curves (i.e., co(a) = co(b)
and Cl (a) = Cl (b)) that are freely homotopic, then, although the curves /3a
and f3b do not reduce to points, they are equal. Therefore, by (5), we still have
the equality in (4). For future reference, we will sum up the above discussion
in a proposition.

1
Proposition 2. Let w be a closed 1-form defined in U, and let Co and Cl be
two closed curves freely homotopic in U. Then w =j w; in particular,

if Co is freely homotopic to a point, 1 = o.

Co
w
Co Cl

We will say that a connected open set U eRn is simply-connected if

every continuous closed curve in U is freely homotopic to a point in U. This
is a topological property of a domain U eRn. For example, the space R n
itself, the balls of R n and its homeomorphic images are all simply connected
24 2. Line Integrals

domains in Rn. On the other hand, as we will show in a moment, R2 -

{O} is not simply-connected. This will follow from Proposition 3 below and
Example 1. The notion of simple connectivity yields a sufficient condition on
a domain for a closed form to be exact.

Proposition 3. Let w be a closed form defined in a simply-connected domain.

Then w is exact.

Proof. By Proposition 2, 1= w 0 for every closed curve c in U. By Propo-

sition 1, w is exact. 0

A further application of the invariance under free homotopy of the integral

of a closed form along a closed curve can be obtained as follows.
Let F: U C R2 -+ R2 be a differentiable map. We say that p E U is a
zero of F if F(P) = O. If there exists a neighborhood V of p such that V
contains no zero of F other than p, then p is called an isolated zero. If the
differential dF(P) of Fat p is nonsingular, we say that p is a simple zero of
F. By the inverse function theorem, F is one-to-one in a neighborhood of a
simple zero, hence a simple zero is isolated.
Now let F(x,y) = (J(x,y),g(x,y» and let D cUbe a closed disk such
that the boundary aD = C contains no zeroes of F. Consider the differential
form
() = fdg - gd!
j2 +92
defined at the points of U where j2 + g2 "I- O. We will call

~ r () = n(FjD)
211" lOD
the index of F in D. We claim that n(Fj D) is an integer.
To see that, let u = f(x,y), v = g(x,y), and let Wo = u:rt~~U be the
element of angle in the plane (u, v) relative to the origin (0,0). Then () = F*wo
and
n(Fj D) = -2 1
11"
r
(J = -2
lc
1
11"
r
F*wo = 21
lc
wo,
11"
r
lFoC
that is, n(Fj D) is the winding number (see Remark 1) of the closed curve
Foe around the origin in the plane (u, v). This shows that n( Fj D) is an
integer.
A consequence of the invariance under free homotopy of Jc () is the fol-
lowing existence theorem for solutions of the equation F = o.

Proposition 4. If n(Fj D) "I- 0, there exists some point qED such that
F(q) = O.
 2. Line Integrals 25

Proof. Assume that no such q exists, and let p be the center of the disk D.
Then the map H: [0,211"] x [0,1] --* R2 - {o} given by

H(s, t) = F((l - t)C(s) + tp)

is a free homotopy between the curve F 0 C and the constant curve F(p).
Therefore,
1 [ () = 21 [
n(F; D) = -2
11"
Wo =
18D 11" l F oc
°
which is a contradiction. o

It is clear that if we take another disk Dl ~ D such that F has no zeroes

in Dl - int D then n(F; Dd = n(F; D). This follows because there is an
obvious free homotopy between F(aD) and F(aD 1 ).
Furthermore, if Ft, F2 : U C R2 --* R2 are two differentiable maps with no
zeroes in C = aD, D c U, and there exists a continuous map H: C x [0,1] --*
R2 - {o} with H(q,O) = Fl (q), H(q,l) = F2(q), H(q, t) # 0, for all q E C,
t E [0,1]' then

This follows because the curves FlOC, F2 0 C are freely homotopic by

K: [0,211"] x [0,1] --* R2 - {O}, K(s, t) = H(C(s), t).

A most interesting application of the index is the formula below that

is attributed to Kronecker (cf. [PIC] pp. 103-105; the proof below is taken
from [LIM 1] pp. 229-230). We say that a simple zero p of F is positive if
°
det(dF) > and is negative otherwise. Since simple zeroes are isolated, there
are only finitely many of them in a compact set.

Theorem 3. Assume that F: U C R2 --* R2 has only simple zeroes in a disk

Dc U none of which is in aD. Then
n(F;D) = P - N,
where P is the number of positive zeroes in D and N is the number of negative
zeroes in D.

For the proof, we need the following lemma.

Lemma. Assume that F has a unique simple zero p E D CU. Then

n(F; D) = ±1 according to det(dFp) > 0 or det(dFp) < O.

Proof of the Lemma. We can assume that p = (0,0). By Taylor's formula,

F(q) = Tq + R(q) Iql , q-+O

lim R(q) = 0,
26 2. Line Integrals

where T = dFp • Consider the map H = U x [O,IJ -t R2

H(q, t) = Tq + (1 - t)R(q) Iql , q E U, t E [0,1].

If we show that, for D sufficiently small, H(q, t)f:. 0, qED, we will have
that H(G(s),t) will be a free homotopy between the curves FoG and ToG,
hence n(F; D) = n(T; D). Since T is one-to-one, the image of a circle around
P will turn at most once. Thus

n(F; D) = n(T; D) = ±1,

as we wished to prove.

° °
It remains to show that there exists c > so that H(q, t) f:. for all q f:. 0,
Iql < c, and all t E [0, 1J. To do this, set G = 1/ IT-II and observe that, for
all q,
Iql = IT-ITql ~ IT-IIITql = l/G ITql ,
hence ITql ~ G Iql. Now, take c > 0 so that in the disk D of radius c we have
IR(q)1 ~ G/2 (recall that lim R(q) = 0). Then, if q f:. 0,
q--+O

IH(q,t)1 = ITq + (1 - t)R(q) Iqll

G
~ ITql- (1 - t) IR(q)llql ~ G Iql- "2 lql > 0,
as we claimed. o

Fig. 2.2

Proof of Theorem 3. Let Pb ... ,Pk be the zeroes of Fin D. By homotopy we

can change slightly D; thus we can assume that no radius of D contains more
than one zero. We divide D into k sectors Si, i = 1, ... , k, so that each such
 2. Line Integrals 27

sector contains exactly one zero (Fig. 2.2). Let C i be a circle around Pi E Si
so small that is contained in Si. There is an obvious free homotopy taking
Ci in aSi. Therefore n(Fj D) is given by

as we wished to prove. o

EXERCISES

1) Show that the form w = 2xy3 dx + 3x 2 y2 dy is closed and compute 1 w,

where c is the arc of the parabola y = x 2 from (0,0) to (x, y).
2) a) Show that if w is a differential1-form defined in U eRn, c: [a, bj ---+ U
is a differentiable curve and Iw(c(t»1 ~ M, for all t E [a, bj, then

where L is the length of c.

b) Let w be a closed 1-form in R2 - {o}. Assume that w is bounded
(that is, its coefficients are bounded) in a disk of center O. Show that
w is exact in R2 - {O}.
c) Show that the result of item (b) still holds if we only suppose that
dw = 0 and that

3) Consider the form

w = x 2 eX
+y
2 {(x cos y + y sin y)dy + (xsiny - ycosy)dx}

defined in R2 - {o}.
a) Show that w can be written as
w = eX cos y Wo + eX sin y d(log r),

J
where Wo is the element of angle at 0 and r = x 2 + y2 j check by
computation that dw = O.
b) Show that w - Wo satisfies the condition of Exercise 2(c), hence is
exact.
28 2. Line Integrals

c) Compute 1w, where c is a simple (i.e., without self intersections)

closed curve in R2 - {o}.

1
4) Let w be a I-form defined in an open set U eRn. Assume that for each
closed differentiable curve c in U, w is a rational number. Prove that
w is closed.
5) Let U, VeRn be simply connected open sets such that Un V is con-
nected. Let w be a closed I-form so that w is exact in U and w is exact
in V. Show that w is exact in U U V.
6) (Applications to complex junctions). Line integrals are quite useful in
the study of complex functions J: C -+ C. Here the complex plane C
is identified with R2 by setting z = x + iy, z E C, (x,y) E R2. It is
convenient to introduce the complex differential form dz = dx + idy and
to write
J{z) = u{x, y) + iv(x, y) = u + iv.
Then the complex form

J{z)dz = (u + iv){dx + idy) = (udx - vdy) + i(udy + vdx)

has udx - vdy as its real part and udy + vdx as its imaginary part. Define

1 J{z)dz = l(UdX - vdy) + i l{UdY + vdx)

Assume that u and v are Cl. Recall that J is holomorphic if and only if
(Cauchy-Riemann equations.)

Show that:
a) J is holomorphic if and only if the real and imaginary parts of J{z)dz
are closed.

1
b) (Cauchy's theorem). If J is holomorphic in a simply-connected domain
U c C and c is a closed curve in U, then J{z)dz = O.
c) If J is holomorphic, the function J'(z) (the derivative of J in z) given
by the equation dJ d~f du + idv = J'(z)dz is well defined and J'(z) =
Ux - iu ll •
d) If J is holomorphic in U c C and J'{z) '# 0, z E U, then all zeroes of
J are simple and positive; furthermore, if D c U is a disk such that
there are no zeroes in aD, then

# of zeroes of J in D 1. [
= -2 dlf .
7I'Z laD J
Hint. That the zeroes of J are simple and positive follows from
 2. Line Integrals 29

det(df) = u; + u; = 1/'(z)1 2 > 0.

Now, a computation shows that
dl du + idv udv + vdu . udv - vdu
- = = + t -"...----".-
1 u + iv u2 + v2 u2 + v2
1 2 2 . udv - vdu
= 2d (log(u +v ))+2 U 2 +V 2 .

Thus, by Kronecker's formula,

1
-.
2n
1 -Idf J
IJD
= -
1
271"
udv - vdu
u v
2 2 =
.
# of zeroes of 1 III D.

7) Consider the form

2(x 2 - y2 - l)dy - 4xydx
(x 2 + y2 _ 1)2 + 4y2
w=~-:--;:,.....::--::---'-...,..:::-----;:,--

defined in R2 - {pi U P2}, Pi = (1,0), pz = (-1,0). Let Dl and D2

be disks centered in Pi and pz, respectively, and so small that pz ~ D 1 ,
Pi ~ D 2 •
a) Show that the integrals

~ f
271" JIJDl
w = +1 -1
271"
18D2
w--l
- ,

where aDl and aD2 are oriented counterclockwise.

Hint. Set F = (/,g) = (x 2 +y2 -1, 2y) and observe that w = 11q~:~/.
Notice that Pi and P2 are the only zeroes of F, where Pi is a positive
zero and P2 is a negative zero.
b) Conclude by homotopy that the integral of w along the curve C below
(Fig. 2.3), with the indicated orientation, is 471".
8. Show that if F: U C R2 --+ R2 satisfies F( -q) = -F(q) for all qED c
U, where D is a disk centered at (0,0), and F has no zeroes in aD, then
n(F; D) is an odd integer; in particular n(F; D) ::J 0, and there exists a
zero of Fin D.
9. (Line integrals 01 vector fields). Let v be a differentiable vector field
defined in an open set U eRn. At the end of Chapter 1, we have
associated to v a differentiable I-form w by w(u) = (v, u), for all u ERn.
Let c: [a, b] --+ U be a piecewise differentiable curve. By defining

we can translate the results of the present chapter into properties of

integrals of vector fields along curves. For instance:
30 2. Line Integrals

Y C

Fig. 2.3

a) Assume that n = 3, that U c R3 is simply-connected, and that

rot v = 0 (see the definition of rot v in Exercise 14, Chapter 1).
Prove that there exists a function f: U - t R such that v = grad f
(see Exercise 12, Chapter 1); furthermore for any curve c joining the

1
points Pl,P2 E U,
v = f(P2) - f(PI)

(If v is a field of forces, f or rather - f, is called the potential energy

of v and the expression above means that the work done by v along
c is equal to change of potential energy between PI and P2).
b) Let the situation be as in (a) and assume, in addition, that div v =
o (see Exercise 11, Chapter 1). Show that the potential is then a
harmonic function, i.e., fxx + fyy + fzz = O.
10. Let w be a differentiable I-form defined in U C R2. A local integrating
factor at P for w is a function g: V - t R defined in a neighborhood
V C U of P such that the form gw is exact in V, i.e., there exists a
function f: V - t R with gw = df.
a) Show that if w(P) =I 0 there exists a local integrating factor at p.
Hint. The condition w(v) = 0 determines a vector field v, in a neigh-
borhood of p, which is nowhere zero. By the fundamental theorem
of ordinary differential equations, there exists a neighborhood V of
P and a function f: V - t R (the so-called first integral of f) so that
f = const. along the trajectories of v. Thus
df(v) = 0 = w(v).
It follows that df = gw.
b) Prove that if g: V - t R is a local integrating factor at P E V, i.e.,
df = gw and B: R - t R is any differential function, then g: V - t R
defined by g(p) = dB(f(P» . g(P) is still an integrating factor.
 2. Line Integrals 31

Hint. We use the notation of the hint in (a). O(J) is still constant
along the trajectories of v. Thus d(O(J)) = gw or dO· gw = gw, where
9 is a new integrating factor. Since w :f; 0, dO· 9 = g.
3. Differentiable Manifolds

Differential forms were introduced in the first chapter as objects in Rn; how-
ever, they, as everything else that refers to differentiability, live naturally in
a differentiable manifold, a concept that we will develop presently.
We will start with the most familiar example of a differentiable manifold,
namely a regular surface in R 3 . In what follows, we will use as a reference
"M. do Carmo, Differential Geometry of Curves and Surfaces, Prentice-Hall,
1976", to be quoted as [dC]. Let us recall (d. [dC), Chap. 2 §2.2) that a
subset S c R 3 is a regular surface if, for each pES, there exist a neighbor-
hood V of pin R3 and a map 10;: Uo; C R2 - t V n S of an open set Uo; in R2
onto V n S such that:
1) 10; is a differentiable homeomorphism.
2) the differential (dlo;)q: Tq(Uo;) - t R3 is injective for each q E Uo;.
The map 10;: Uo; - t S is called a parametrization of S around p.
The most important consequence of the above definition is the fact that
the change of parameters is a diffeomorphism. More precisely, if 10;: Uo; - t S
and 1{3: U{3 - t S are two parametrizations such that 10;(U0;) n 1{3(U{3) = W =I-
¢, then (cf. [dC], Theorem of §2.5) the maps
IiI o/o;:/;;I(W) -t R2,
1;;10 1{3: lil(W) -t R2
are differentiable. It follows that on a regular surface it makes sense to talk
about differentiable functions and to apply the methods of Differential Cal-
culus.
The most serious problem with the above definition is its dependence on
R3. Indeed the natural idea of an abstract surface is that of an object that
is, in a certain sense, two-dimensional, and to which we can apply, locally,
the Differential Calculus in R2. Such an idea of an abstract surface (Le.,
with no reference to an ambient space) has been foreseen since Gauss. It
took, however, about a century for the definition to reach the definitive form
that we present below. One of the reasons for this delay was that, even for
surfaces in R 3 , the fundamental role of the change of parameters was not
clearly understood (see Remark 1).
Since nothing is gained in limiting ourselves to dimension two, we will
present the definition for dimension n.
34 3. Differentiable Manifolds

Definition 1. An n-dimensional differentiable manifold is a set M together

with a family of injective maps fa: Uo. C R n -+ M of open sets Uo. in R n
into M such that:
1) U/o.(Uo.) = M.
0.

2) For each pair a,/3, with fo.(Uo.) n f/3(U/3) = W ::I </>, the sets f~l(W)
and l;l(W) are open sets in R n and the maps f;l 0 10., f~l 0 f/3 are
differentiable (Fig. 3.1).

-
Fig. 3.1

3) The family {(Uo., fa)} is maximal relative to (1) and (2).

The pair (Uo., fa) with p E fo.(Uo.) is called a parametrization (or a coordi-
nate system) of Mat p; fo.(Uo.) is then called a coordinate neighborhood of p.
A family (fa, Uo.) satisfying the properties (1) and (2) is called a differentiable
structure on M.

Condition 3 is a technical condition. Actually, we can always extend a dif-

ferentiable structure into a maximal one, by adjoining to the given structure
all parametrizations that together with some parametrization of the given
structure satisfy condition 2. Thus, with a certain abuse of language, we
can say that a differentiable manifold is a set endowed with a differentiable
structure, the extension to the maximal one being assumed whenever needed.

Remark 1. A comparison between the definition of a regular surface in R 3

and the definition of a differentiable manifold shows that the crucial point
was to introduce the fundamental property of change of parameters (which
is a theorem for surfaces) as an axiom in Definition 1. As we will soon see,
 3. Differentiable Manifolds 35

this is what allows to transport to differentiable manifolds all the notions of

Differential Calculus in R n •

Remark 2. A differentiable structure on a set M induces in a natural way a

topology in M. It suffices to define that A c M is an open set if f;;l(A n
fo.(Uo.» is an open set in R n, for all Q. It is easily checked that this defines
a topology in M in which the sets fo.(Uo.) are open and the maps fa are
continuous.

Remark 9. The natural topology of a differentiable manifold can be quite

strange. In particular, it can happen that one (or both) of the following
axioms do not hold:
a) Axiom of Hausdorff. Given two distinct points of M there exist neigh-
borhoods of these points that do not intersect.
b) Axiom of countable basis. M can be covered by a countable number of
coordinate neighborhoods (we say then that M has a countable basis)
Axiom (a) is essential to prove that the limit of a converging sequence
is unique, and axiom (b) is essential for the existence of a partition of unity
(cf. Chapter 3) which is an almost indispensable tool for the study of the
topology of manifolds.
We shall assume, from now on, that all the manifolds to be considered
are Hausdorff and have a countable basis.

Of course, a regular surface is an example of a 2-dimensional differentiable

manifold. A trivial example of an n-dimensional differentiable manifold is the
euclidean space R n with the differentiable structure given by the identity.
Less trivial examples are the following ones.

Example 1. The real projective plane p2(R). We will denote by P2(R) the
set of all lines in R3 that pass through the origin (0,0,0) of R 3 , i.e., p2(R)
is the set of "directions" in R 3 .
We want to introduce a differentiable structure in p2(R). For that, let
(x, y, z) E R 3, and notice that p2(R) is the quotient space of R3 - {(O, 0, O)}
by the equivalence relation rv:

(x,y,Z) '" (AX, AY, AZ), A E R, A =1= O.

The points of P2(R) will be denoted by [x, y, z].

We now define sets Vi, V2, V3 in p2(R) by:

Vl = {[x, y, z]; x =1= O},

V2 = {[x, y, z]; y # O},
V3 = {[x,y,z];z # O},
36 3. Differentiable Manifolds

and maps Ii: R2 --. Vi, i = 1,2,3, by

/t(u,v) = [1,u,vj, /2(u,v) = [u,l,vj, !a(u,v) = [u,v,l],
where (u, v) E R 2 • Geometrically V2, for instance, is the set of lines of R 3
that pass through the origin and do not belong to the plane xOz. We claim
that the family {(Ii, R2)} is a differentiable structure for p2(R).
Each Ii, i = 1,2,3, is clearly bijective and U
li(R2) = p2(R). It remains
i
to be shown that li-l(Vi n Vj) is open in R2 and I j- l 0 Ii is differentiable.
Let us consider the case i = 1, j = 2; the other cases are entirely analogous.
The points of III (VI n V2 ) are of the form (u, v) with u -::j:. O. Thus
III (VI n Vi) is open in R2, and

is clearly differentiable, as we claimed.

Example 1 '. (The real projective space). Example 1 can easily be generalized.
Let (Xl. . .. ,xn+t> E R n +1 and define the n-dimensional real projective space
pn(R) as the quotient space of R n +1 - {OJ by the equivalence relation:

(Xl. ... , Xn +1) ,...., (AXb"" AX n +1), A E R, A -::j:. 0;

points in pn(R) will be denoted by [Xl, .•• , Xn+lJ.
Define subsets Vi c pn(R), i = 1, ... ,n + 1, by
Vi = {[Xl, ••. , X n +1J; Xi -::j:. OJ,
and maps Ii: R n --. Vi by

li(Yl.· . . ,Yn) = [Yl.· .. ,Yi-l, 1, Yi,' .. ,YnJ.

Proceeding as in Example 1, one easily checks that the family {(Ii, R n)}
is a differentiable structure in pn(R). (See Exercise 1).

Example 2. (The Klein bottle). The Klein bottle is the subset of R4 defined
as follows (Fig. 3.2). Let Ox, Oy, Oz, Ow be the four coordinate axis in R4.
Let S be a circle with radius r, contained in the plane xOz, with center C
in the axis Ox so that C is at a distance a > r from O. The Klein bottle
is generated by rotating this circle around 0 z in such a way that when the
center C has described a rotation of an angle u in the plane xOy, the plane of
S has described a rotation of angle u/2 around OC in the 3-space OCOzOw
(this is possible because we are in R 4 ).
Let u and v be as in Fig. 3.1, and let Ul C R2 be given by
Ul = {(u,v) E R2;0 < u < 27r,0 < V < 27r}
 3. Differentiable Manifolds 37

Fig. 3.2

Now define a map /1: Ul -+ R4 by

X= (r cosv + a) cos u
(r cos v + a) sin u
/1(u, v) = ~ =:
{ y=
r sin v cos u/2
r sin v sin u/2.

It is clear that /1 (Ud contains the points ofthe Klein bottle that are not
on the circles u = 0 and v = O. We claim that /1 is injective.
To see that, let us first assume that z =I O. Then both sin v and cos u/2
are nonzero. Since 0 < u/2 < 7T, the expression -; = tan ~ determines u. By
knowing u, we can use that

. w +v'X2 +y2 - a
smv= -'-u' cos v = -~---"'--­
rsmI r

to determine v. This proves our claim if z =I O. If z = 0 then v = 7T or u = 7T,

and again injectivity is easily checked.
It can be shown that if we change the origin of u and the origin of v, we
can cover the whole Klein bottle by the images of maps similar to the above.
For instance, if we define a map 12: U2 -+ R4 given by
X= -(rcosv + a) sin u
{ y- (rcosv + a) cosu
12 z= . -
rsmvcos (tiI + "411') '
w= rsinvsin(¥ + 1)
(geometrically, this means that we measure u from Oy), we see that h(U2)
includes the points of the Klein bottle with u = O. It is easily checked that 12
38 3. Differentiable Manifolds

is injective. Notice that b(U1 ) n !2(U2) = W is not connected but has two
connected components:
7r 7r
WI = {b(u,v); 2" < u < 27r}, W2 = {b(u,v);O < u < 2"}.
The change of coordinates is given by:
'it - u - 7r/2
1:;10 b { _-
v= v

which is clearly differentiable.

In a similar way, we can find an injective map fa: U3 -+ R4 whose image
covers the Klein bottle minus the circle v = 0 (this amounts to change the
origin of v). It also is not difficult to check that the changes of coordinates
I j- 1 0 h i,j = 1,2,3, are differentiable, and this shows that the family (h U;)
is a differentiable structure on the Klein bottle.

Remark 4. The Klein bottle can be thought of as a twisted torus in the

following sense. The torus is obtained from a rectangle by identifying the
opposite sides. In the Klein bottle, one of the sides is reflected across its
center before performing the identification (Fig. 3.3). It can be shown that
the Klein bottle is not a regular surface in R3: the model of Fig. 3.3 has
self-intersections.

Before presenting further examples, we need to extend to differentiable

manifolds the local Differential Calculus that holds for R n.
From now on, when we denote a differentiable manifold by Mn, the upper
index n will denote the dimension of M = Mn.

Definition 2. Let Mf and Mf be differentiable manifolds. A map cp: Ml -+

M2 is differentiable at a point p E Ml if given a parametrization g: V c
R m -+ M2 around cp(p), there exists a parametrization I: U C R n -+ Ml
around p such that cp(J(U» c g(V) and the map
g-1 0 cp 0 I: U eRn -+ R m
is differentiable at 1- 1 (P). The map cp is differentiable in an open set of Ml
if it is differentiable at all points of this set.

The map g-1 0 cp 0 I is the expression of cp in the parametrizations I

and g. Since the change of parameters is differentiable, the fact that cp is
differentiable does not depend on the choice of parametrizations.
 3. Differentiable Manifolds 39

(J • • (.~
Torus

B
0
A
A
0
B - 0J (]
~

Klein bottle

Fig. 3.3

In particular, it follows from the above that we can talk about differen-
tiable functions (<p: M n - t R) and differentiable curves (<p: 1 c R - t Mn) on
a differentiable manifold (1 c R will always denote an open interval of the
real line containing the origin 0 E R).

Now we would like to define the notion of a tangent vector to a differ-

entiable curve on a differentiable manifold. For the case of a differentiable
curve 0: 1 C R - t S C R3 on a regular surface in R 3, the tangent vector
o'{t) is merely the speed of 0 as a vector in R3. Since we do not have the
support of a nice ambient space, we must choose a characteristic property of
the tangent vector that does not depend on the ambient space.
For that, let 0: (-c, c) - t R n be a differentiable curve in Rn, with 0(0) =
p ERn, and write

o(t) = (Xl(t), ... ,xn(t», t E (-c,c), (Xl, ... ,Xn ) ERn.

Then 0'(0) = (x~(O), ... ,x~(O» = VERn. Now let <p be a real function in
R n, differentiable in a neighborhood of p. Then the derivative of <p along v
at p is given by

d (
dt <p 0 0
)\
t=O =~
i=l
a<p dXi \
a
(~ '( ) a )
L..J ax- dt t=O = L..J Xi 0 ax-
i=l a
<po
40 3. Differentiable Manifolds

Thus the "directional derivative along v" is an operator over differentiable

functions which depends only on v. This is the characteristic property of
vectors that we will use to extend them to differentiable manifolds.

Definition 3. Let a: I -+ M be a differentiable curve on a differentiable

manifold M, with a(O) = p E M, and let D be the set of functions of M
which are differentiable at p. The tangent vector to the curve a at p is the
map a'(O): D -+ R given by

a'(O)1;? = :t (I;? 0 a)jt=o' I;? E D.

A tangent vector at p E M is the tangent vector of some differentiable curve

a: I -+ M with a(O) = p.

We want to show that the set of tangent vectors at a point p E Mn makes

up an n-dimensional real vector space. For that, choose a parametrization
I:U c R n -+ M around p = 1(0, ... ,0). Then a curve a:I -+ M and a
function I;? E D can be written as:
1-10 a(t) = (Xl (t), . .. , Xn(t»,
I;? 0 I(q) = I;?(Xl, ••. , Xn),

respectively. Thus

a'(O)1;? = :t(1;? 0 a)lt=o = :tl;?(Xl(t), ... ,xn(t»lt=o

= (~X~(O) (a~Jo) I;?,

so that the tangent vector a'(O) at p can be written as

a'(O) = t x~(O) (aa.) .

i=l X. 0
(1)

Notice that (~) 0 is the tangent vector at p to the "coordinate curve"

Xi -+ I(O, ... ,O,xi'O, ... ,O).

Now let T, be the vector space generated by { (~) J, i = 1, ... , n.

Lemma 1. The set TpM 01 tangent vectors to M at p is equal to T,.

Proof. We have just shown that TpM C T,. Conversely, if VET" then
v = Ei Ai (Ix;)o' Let a: I -+ M be given in the parametrization I by
Xi = Ait. Then a'(O) = v, i.e., v E TpM. 0
 3. Differentiable Manifolds 41

It follows that TpM is a vector space. Furthermore, the choice of a

parametrization 1 determines a basis {( Ix; ) o} for TpM. Thus TpM is an
n-dimensional vector space which is called the tangent space of M at p. The
basis {(Ix; )o} is called the associated basis to the parametrization I·

With the notion of a tangent space we can define the notions of differential
of a map tp: Mf -4 Mr'

Definition 4. Let Mf and Mr' be differentiable manifolds and let tp: Ml -4

M2 be a differentiable map. For each p E M, the differential 01 tp at p is
the linear map dtpp: TpMl -4 T<p(p)M2 which associates to each v E TpMl
the vector dtpp(v) E T<p(p)M2 defined as follows: Choose a differentiable curve
a: (-€,€) -4 Ml. with a(O) = p, a'(O) = v; then dtpp(v) = (tp 0 a)'(O).
For the definition to make sense, it is necessary to show that (tp 0 a)'(O)
does not depend on the choice of a and that dtpp is in fact a linear map. This
is proved by taking parametrizations around p and tp(p) thus reducing the
question to the case of a map from R n to R m , where the properties are well
known (or can easily be proved); for the case n = 2, m = 3, see, for instance,
[de] §2.4.

Definition 5. Let Ml and M2 be differentiable manifolds. A map tp: Ml -4

M2 is a diffeomorphism if it is differentiable, bijective, and its inverse tp-l is
also differentiable. The map tp is a local diffeomorphism at p E M if there exist
neighborhoods U of p and V of tp(p) such that tp: U -4 V is a diffeomorphism.

The linear map dtpp may be thought as a first order approximation of the
map t.p around p. Probably the most important local theorem in Calculus is the
inverse function theorem which states that if dtpp is an isomorphism then tp
is a local diffeomorphism at p. Being a local theorem, it extends immediately
to differentiable manifolds.

Example 3. (The tangent bundle). Let Mn be a differentiable manifold and

let
TM = {(p,v);p E M,v E TpM},
i.e., T M is the set of all tangent vectors to M. We will introduce in T M
a differentiable structure (of dimension 2n); with such an structure, T M is
called the tangent bundle of M.
Let 10.: Uo. eRn -4 M be a parametrization of M with (x?, ... , x~) E Uo..
For W E Tf~(q)M, q E Uo., we can write

W = '~Yi
" 0. 8
8x~'
i •
Define a map Fa: Uo. X R n --+ TM by
42 3. Differentiable Manifolds

F.(xf,···,x:,yf,···,y:) ~ (f.(Xf' ... 'x:),~y~ a~r)

We claim that if {(Uar, far)} is a differentiable structure for M then {(Uar x
R n , Far)} is a differentiable structure for T M. Geometrically, this means that
we take as coordinates of a point (p, v) E T M the coordinates of p together
with the coordinates of v in the associated basis.
Let

that is,

F;;l 0 Far (qar, var) = F;;l(far(qar), dfar(var»

= (fi l 0 fO;(qar), d(fi l 0 fO;)(var».

Since fil 0 far is differentiable, so is d(fi 1 0 far). It follows that Fi l 0 Far is

differentiable. This proves condition (2) of Definition 1. Since condition (1)
is immediate, our claim follows.

Definition 6. Let Mm and Nn be differentiable manifolds. A differentiable

map cp: M --+ N is an immersion if dcpp: TpM --+ T<p(p)N is injective for
all p E M. If, in addition, cp is a homeomorphism onto cp(M) c N, where
cp(M) has the topology induced by N, cp is an embedding. If MeN and the
inclusion i: MeN is an embedding, we say that M is a sub manifold of N.

Example 4.
a) The curve a: R --+ R2 given by a(t) = (t 3 , t 2 ) is a differentiable map
but not an immersion. In fact, the condition of immersion is this case is
equivalent to the fact that a'(t) :f 0, which does not hold for t = 0.
b) The curve a(t) = (t 3 -4t, t 2 -4) is an immersion that is not an embedding,
since it has a self-intersection at (0,0) (for t = 2, t = -2).
c) The curve (Fig. 3.4)

=(0,-(t+2», tE(-3,-I)
a(t) { = a regular curve as in Fig. 3.2, t E (-1, -1/11")
= (-t,-sin(l/t», t E (-1/11",0)
is an immersion a: (-3,0) --+ R2 without self-intersections. However, a
is not an embedding. For, in the topology of R 2 , a neighborhood of a
point p in the vertical part of the curve has infinitely many connected
components, where as in the topology induced by a it is a connected
interval.
 3. Differentiable Manifolds 43

Fig. 3.4

d) The inclusion i: 8 2 C R3 of a regular surface in R3 is an embedding. This

follows from conditions (1) and (2) of the definition of a regular surface
in the beginning of this chapter.

Example 5. (Regular surfaces of R n). The natural generalization of the idea

of a regular surface in R 3 is the notion of a k-dimensional surface in R n ,
k < n. A subset Mk eRn is a k-dimensional regular surface if for each p E M
there exist a neighborhood V of p in R n and a map f: U c R k ~ M n V of
an open set U of R k onto M n V such that:
1) f is a differentiable homeomorphism,
2) (df)q:Rk ~ R n is injective for all q E U.
Except for the dimensions involved, the definition is the same as the one
given for a regular surface of R 3 .
It can be proved that given two parametrizations II: U1 C Rk ~ M,
h: U2 C Rk ~ M, with II(U1) n h(U2) = W i ¢, the change of parameters
11-10 12: 121(w}~ Ill(W) is a diffeomorphism. The proof is the same as
for surfaces (see (de] pg. 71).

Example 6. It can be shown that the maps /;, i = 1,2,3, given in Example 2,
have injective differentials. This shows that the Klein bottle is a 2-surface of
R4.

Example 7. (The projective plane revisited). The set of straight lines of R3

that pass through the origin may be looked upon as the quotient space of
the unit sphere S2 = {p E R 3 ; Ipi = I} by the equivalence relation that
identifies p with the antipodal point -po For each straight line through the
origin determines in S2 two antipodal points, and the correspondence thus
obtained is clearly bijective.
Taking this into account, let us introduce another differentiable structure
in p2(R). For that, notice that we can cover the regular surface 8 2 c R3 by
a family of parametrizations:
44 3. Differentiable Manifolds

where
Ul = {(Xl,X2, X3) E R3; Xl = 0, X~ + X~ < 1}
fi(X2,X3) = (D ll X2,X3), Dl = V~1---(-X-~-+-X-~-)
fl(x2,x3) = (-D ll X2,X3);
U2 = {(XllX2,X3) E R 3 ;X2 = O,X~ +X~ < 1}
ii(XllX3) = (xllD2,X3), D2 = J~1---(-x-i-+-X-5-)

ii(Xl, X3) = (Xl, -D2, X3);

and similarly for i = 3 (Fig. 3.5). It is immediate to check that this is in fact
a differentiable structure for 8 2 •

Fig. 3.5

Let 7r: 8 2 - p2(R) be the canonical projection, that is, 7r(p) = {p, -p}
and notice that 7r(Ji+(Ui» = 7r(Ji-(Ui». We will define gi: Ui - P2(R) by

gi = 7r 0 it
Since the restriction of 7r to it(Ui) is injective, we obtain that

gil 0 gj = (7r 0 it)-l 0 (7r 0 If) = (Jt)-l 0 i t

It follows that gil ogj is differentiable for all i,j = 1,2,3, hence {(Ui, gi)}
is a differentiable structure for p2(R).
 3. Differentiable Manifolds 45

Actually the above structure and that of Example 1 will determine the
same maximal structure. For the coordinate neighborhoods are the same and
the coordinate changes are given (for i = 1, say) by

which is clearly differentiable. Notice also that it follows from the above that
the canonical projection 11": 8 2 - t p2(R} is a local diffeomorphism at each
point of 8 2 •

Remark 5. It is possible to prove that the projective plane cannot be em-

bedded in R3. Fig. 3.6 describes a differentiable map of the projective plane
into R3 that has a line segment of self-intersection and two singular points at
the endpoints of this segment. With a more elaborate construction, one can
describe an immersion of p2(R} into R3 (the so-called Boy's surface which,
of course, has self-intersections). For more details, see Gerd Fischer, [FISC],
Commentary, Chapter 6 (by U. Pinkall).

A A

AI---~~---IA

Fig. 3.6

Example 8. (Immersion of the projective plane in R 4). Let <p: R3 -t R4 be

the map given by
46 3. Differentiable Manifolds

tp(x,y,z) = (X 2 - y2,xy,xz,yZ), (X,y,Z) E R3.

Let 8 2 C R3 be the unit sphere and let 11": 8 2 - P2(R) be the canonical
projection of 8 2 onto the real projective plane (see Example 7). Notice that
tp(P) = tp( -p), and define a map (J: p2(R) _ R4 by
(J({p, -p}) = tp(P).

Since 11" is a local diffeomorphism, to show that (J is an immersion, it

suffices to show that the restriction of tp to the sphere 8 2 is an immersion.
To see that, choose the parametrization of 8 2 given in Example 7. Then It,
for instance, is given by
It (x, y) = (x, y, JI - (x 2 + y2»
and
tpolt(x,y,z) = (x 2 -y2,xy,xD,yD).
To show that d( tp 0 It) is injective, it suffices to show that the rank of the
matrix
( 2x Y D + xDx YDx)
-2y x xDII D+yD II
is equal to two, and this is easily checked. Similarly, we can check that the
same holds for all other parametrizations and this completes the example.
The immersion (J is actually an embedding. The details will be left as an
exercise (Exerc. 5).

Remark 6. A natural question in the theory of differentiable manifolds is to

know whether a given differentiable manifold can be immersed or embedded
in some euclidean space. A fundamental theorem due to Whitney states that:
every differentiable manilold (Hausdorff and with countable basis) 01 dimen-
sion n can be immersed in R 2n and embedded in R2n+1. See, e.g., M. Hirsh,
[HIRJ.

We now extend for differentiable manifolds the notion of a differential

form of Chapter 1. Given a vector space V, we will denote by Ak(V) the set
of alternate, k-linear maps w: V x ... x V - R, where V x ... x V contains
k factors.

Definition 7. Let M n be a differentiable manifold. An exterior k-Iorm w in

M is the choice, for every p EM, of an element w(p) of the space Ak(TpM)*
of alternate k-linear forms of t.he tangent space TpM.

Given an exterior k-form wand a parametrization lOt: Uot - Mn, around

p E IOt(UOt ), we define the representation of w in this parametrization as the
exterior k-form Wot in Uot C R n given by
 3. Differentiable Manifolds 47

w,Avl, ... , Vk) = w(dfOt(vd,.·., dfOt(Vk)), Vl, •• " Vk ERn.

If we change coordinates to f{3: U{3 --t Mn, p E f{3(U{3), we obtain that

(fii I 0 fOt)*w{3(vl, ... , Vk) = W{3(d(fii l 0 fOt)(vt} , ... ,d(fii l 0 fOt)(Vk))

= w((df{3 0 d(fi l 0 fOt))(vd, ... , (df{3 0 d(fi l 0 fOt))(Vk)
= wOt(Vl, ... , Vk),

that is, (fi l 0 fOt)*w{3 = WOt .

Definition 8. A differential form of order k (or a differential k-form) in a

differentiable manifold Mn is an exterior k-form such that, in some coordinate
system (hence, in all), its representation is differentiable.

From the above, it follows that a differential k-form in Mn is the choice, for
each parametrization (UOt, fOt) of M, of a differential k-form WOt in Uot in such
a way that for another parametrization (U{3, f(3), with fOt(UOt) n f{3(U{3) "I 1;,
we have Wot = (fi l 0 fOt)*w{3.
It is an important fact that all the operations defined for differential
forms in R n can be extend to differential forms in Mn through their local
representations. For instance, if W is a differential form in M, dw is the
differential form in M whose local representation is dw Ot . Since

dwOt = d(fi l 0 fOt)*w{3 = (fi l 0 fOt)*dw{3,

dw is a well defined differential form on M.
Closely associated with differential forms is the notion of a vector field.

Definition 9. A vector field X on a differentiable manifold M is a corre-

spondence that associates to each point p E M a vector X(p) E TpM. The
vector field X is differentiable if for every differentiable function cp: M --t R,
X cp is again a differentiable function.

Let JOt.: UOt. --t Mn be a parametrization of M and Xi = i = 1, ... , n, k,

the basis associated to the parametrization. Then a vector field X can be
written in fOt(UOt ) as

Since a vector field X on M is an operation on the space D of differentiable

functions of M, we can take the iterates of this operation. For instance, if
X and Y are differentiable vector fields and cp: M --t R is a differentiable
function, we can consider the functions Y (X cp) and X (Y cp). In general, such
iterated operations do not lead to vector fields, since they involve derivatives
of order higher than the first. However the following holds.
48 3. Differentiable Manifolds

Lemma 2. Let X and Y be differentiable vector fields on a differentiable

manilold M. Then there exists a unique vector field Z on M such that, lor
each cp E D, Zcp = (XY - YX)cp.

Prool. We first prove that if such a Z exists, then it is unique. For that, let
I: U --. M be a parametrization, and let

be the expressions of X and Y, respectively, in the parametrization I. Then

hence
(XY - Y X)cp =~
~ (~( ab 0 aa 0) a)
aXi - bi-aXi' -aXj cpo
~ ai-'
) ,o 0

It follows that if a Z exists with the required property, it must be expressed

as above in any coordinate system, hence it is unique
To prove existence, just define Za in each coordinate neighborhood
la(Ua ) c M by the above expression. By uniqueness, Za = Z{3 in !a(Ua ) n
1{3(U{3), hence Z is well defined on M. 0

Definition 10. The vector field determined by the above lemma is called
the bracket [X, Y] = XY - YX of X and Y, and it is clearly differentiable.

The bracket operation has the following properties:

Proposition 1. Let X, Y and Z be differentiable vector fields, a and b be

real numbers, and cp and 0 be differentiable lunctions. Then
a) [X,Y] = -[Y,XJ,
b) [aX + bY, Z] = a[X, Z] + b[Y, Zl,
c) [[X, Yl, Zl + [[V, Zl, Xl + [[Z, Xl, Yl = 0 (Jacobi's identity),
d) [OX, cpYl = Ocp[X, Y] + O· X(cp)Y - cp. Y(O)X.

Prool. (a) and (b) are immediate. To prove (c), we observe that
[[X,Y],Z] = [XV - YX,Z] =XYZ - YXZ- ZXY +ZYX
= [X, [Y, ZII + [V, [Z, XII,
 3. Differentiable Manifolds 49

and use (a) to obtain (c). The proof of (d) is a direct and simple computation.
o
There exists an interesting relation between exterior differentiation of dif-
ferential forms and the bracket operation. For the case of I-forms, this relation
is as follows.

Proposition 2. Let W be a differentiable I-form on a differentiable manifold

M and let X and Y be differentiable vector fields on M. Then

dw(X, Y) = Xw(Y) - Yw(X) - w([X, YJ) (2)

Proof. Set f: U -+ M be a parametrization of M and let

a a
X=L:a
. i -ax,., Y=L:bj-a.
. x J
J J

be the expressions of X and Y in this parametrization. We first observe that

if (2) holds for Xi and l'j, then it also holds for EiXi and EiY;. Next, we
claim that if (2) holds for X and Y, it also holds for OX and tpY, when 0 and
tp are differentiable functions.
To see that, we first notice that, by hypothesis,

dw(OX,tpY) = Otpdw(X, Y) = Otp{Xw(Y) - Yw(X) - w([X, YJ)}.

By using (d) of Proposition 1, we obtain

(OX)w(tpY) - (tpY)w(OX) - w([OX, tpYJ)

= OX(tp)w(Y) + (OtpX)w(Y) - tpY(O)w(X) - (tpOY)w(X)
- OtpW([X, Y]) - OX(tp)w(Y) + tpY(O)w(X)
=Otp{Xw(Y) - Yw(X) - w([X, Y])}
= dw(OX, tpY)
and this proves our claim.
It follows that it suffices to prove (2) for the vectors k, /;;. Since
[k, /;;] = 0, it suffices to prove that
dw (a~i' a~j) = a~i w (a~j) - a~j w (a~J· (2')

Notice that if (2') holds for WI and W2, it also holds for WI + W2. Thus, it
suffices to show that
50 3. Differentiable Manifolds

where 0: is a differentiable function. The above reduces to

(do: /\ dXk) (~,~) = oki 00: - Oki 00:

OXi oXi OXi oXi
which holds by the very definition of exterior product. o
Remark 7. With essentially the same proof, one can show the following gener-
alization of Proposition 2. Let w be a differentiable k-form and XI.' .. ,Xk+l
differentiable vector fields. Then

i=l

+ 2:(-l)i+i w([Xi,Xil,XI. ... ,,,t,··· ,Xi"" ,Xk+d

i<i

where Xi means that Xi is missing.

We will conclude this chapter with the global notion of orientability for
manifolds.

Definition 11. A differentiable manifold M is orientable if M has a differen-

tiable structure {(Ua , la)} such thatfor each pair 0:, f3 with la (Ua )nlf3 (Uf3) =f
<p, the differential of the change of coordinates r;; 1 0 I a has positive determi-
nant. Otherwise, M is called nonorientable.
If M is orientable, the choice of a differentiable structure satisfying the
above is called an orientation for M.
Examples of orient able and nonorientable manifolds are given in the Ex-
ercises.

Remark 8. It can be shown that a compact regular surface in R3 is orientable.

A beautiful and simple proof of this fact can be found in Elon Lima [LIM 21.
As we will see in Exercises 10 and 11, the projective plane and the Klein bottle
are not orient able; hence, they cannot be embedded in R 3 , a fact mentioned
earlier in Remarks 4 and 5 .

EXERCISES

1) Show with details that the real projective space pn(R) is a differentiable
manifold.
 3. Differentiable Manifolds 51

2) Let M and N be differentiable manifolds where {(Ua , fa)} is a differen-

tiable structure for M and {(V/3,9/3)} is a differentiable structure for N.
Consider the cartesian product M x N and the maps ha/3: Ua x V/3 -+
M x N given by

Show that {(Ua x V/3, ha/3)} is a differentiable structure for M x N which

is then called the product manifold of M and N. Describe the product
manifold 8 1 x 8 1 of two circles, where 8 1 has the usual differentiable
structure.
3) Let c.p: M -+ N be a differentiable map. Show that the definition of the
differential dc.pp: TpM -+ Tcp(p)N of c.p at p (Definition 4) does not depend
on the choice of the curve and that dc.pp is a linear map.
4) Let c.p: M -+ N be an immersion and let p be a point in M. Show that
there exists a neighborhood V C M of p such that the restriction c.p1V of
c.p to V is an embedding (This means that every immersion is locally an
embedding).
5) Prove that the immersion of p2(R) into R4 given in Example 8 is an
embedding.
Hint. To show the injectivity of () set

x 2 - y2 = a, xy = b, xz = c, yz = d. (*)
It suffices to check that, under the condition x 2 + y2 + Z2 = 1, the
above equations have only two solutions which are of the form (x, y, z)
and (-x, -y, -z). The three last equations give:

If b, c, d are all zero, the equation (*) shows that at least two of the
coordinates x, y, z are zero, the remaining one being ±1, since x 2 +
y2 + Z2 = 1. If one of the values b, c, d is non zero the equation (**)
together with x 2+y2 + z2 = 1 will determine x 2, y2, z2 . From equations
(*), we see that the choice of a sign for one of the x, y, z determines
the signs of the remaining two.
6) Consider the cylinder C = {(x,y,z) E R3;x 2 +y2 = I} and identify the
point (x, y, z) with (-x, -y, -z). Show that the quotient space of C by
this equivalence relation can be given a differentiable structure (infinite
Mobius band).
7) Show that the tangent bundle of a differentiable manifold is orientable
(even if the manifold is not so).
8) Let M be a differentiable manifold that can be covered by two coordinate
neighborhoods VI and V2 in such a way that the intersection VI n V2 is
connected. Show that M is orientable.
9) Show that the sphere 8 n = {p ERn+!; Ipi = I} is orientable.
52 3. Differentiable Manifolds

10) Show that the real projective plane p 2 (R) is not orientable.
Hint. Show that if a manifold M is orientable, any open set in M is an
orient able manifold. Notice that p2(R) contains an open Mobius band
that is not orient able (cf. [de], §2.6, Example 3).
11) Show that the Klein bottle is not orientable.
12) A field of planes in an open set U C R3 is a correspondence P that
associates to each p E U a plane P(p) passing through p. The field P is
differentiable if the coefficients of the equation of P(p) are differentiable
functions in p. A integral surface of P is a surface S c R3 such that for
each q E S, we have TqS = P(q), i.e., S is tangent at each of its points
to the plane of the field passing there.
Let W be a differentiable I-form in U C R3 with w(q) =F 0, q E U. Show
that:
(a) W determines a differentiable plane field P by the condition
v E P(p) C R3 ~ wp(v) = 0.
(b) If S is an integral surface of P passing through p, then i*(w) = 0,
where i: S C R3 is the inclusio' .
(c) If there exists an integral' iacf' J of P passing through p, for all
p E U, there exists a I-form (f ill a neighborhood V C U of p such
that dw = w /I. (1.
Hint for (c): Consider two I-forms W2, W3 in such that w = WI, W2, W3 are
linearly independent in V, and write
dw = aW2 /I. W3 + {3w3 /I. WI + ')'WI /I. W2
°
By using the fact that d(i*w) = i*dw = and that the 2-form W2 /I. W3 is
nonzero, one concludes that a = 0, hence

(d) If there exists an integral surface of P for all p E U and W = adx +

bdy + cdz, then

( ac _ ab) a + (aa _ ac) b + (ab _ aa) c =

ay az az ax ax ay
°
°
Hint for (d): One concludes from (c) that dw /I. W = and writes this
equation in coordinates.
13) Set W = xdx + ydy + zdz, and let P be the field of planes in R3 - {o}
determined by w. Show that the integral surface of P passing through
p = (x, y, z) is the sphere with center in the origin (0,0,0) and passing
through p.
14) Set w = zdx + xdy + ydz. Show that the plane field determined by w has
no integral surface.
 3. Differentiable Manifolds 53

15) Consider in the real line R the two following differentiable structures: 1)
(R,fd where fI(x) = x. 2) (R,h), where h(x) = x 3 • Show that:
a) The identity map i: (R, ft} -+ (R, h), i(x) = x, is not a diffeomor-
phism (thus the maximal structures determined by (R, ft} and (R, h)
are distinct)
b) The map t.p: (R, fI) -+ (R, h) given by t.p(x) = x 3 is a diffeomorphism
(thus, although the differentiable structures are distinct, they define
differentiable manifolds that are diffeomorphic).
16) (The orientable double covering). Let M be a connected differentiable
manifold. For each p E M, denote by Op the quotient space of the set
of all bases of TpM under the following equivalence relation: two bases
are equivalent if they are related by a matrix with positive determinant.
Clearly Op has two elements, and each element Op of Op is called an
orientation at p. Now let

and let fa: Ua -+ M be a parametrization of M with p E fa(Ua ). Define

ia:Ua -+ if by
- a a
fa(XI, ... ,Xn) = (ja(XI. ... ,xn)'[-a '···'-a ])
Xl Xn

where (XI, ... ,Xn ) E Ua , and [/xt, ... ,kl denotes the element ofOp
determined by this basis. Show that:
a) If {(Ua , fa)} is a differentiable structure in M, then {(Ua , j~)} is a
differentiable structure in if which is orient able (even if M is not).
b) The map 7r: if -+ M given by 7r(p,Op) = P is differentiable, sur-
jective, and each point p E M has a neighborhood V whose inverse
image 7r- 1 (V) is the disjoint union of two open sets each of which is
applied by 7r diffeomorphically onto V. By this reason, if is called
the orientable double covering of M.
c) Mis orient able if and only if if is not connected.
Hint. Notice that if M is orientable, the sets Ml = {(p, orient. of M in
pH, M2 = {(p, orient. opposite to that of M in p)}, are nonempty, dis-
joint, open subsets of if. For the converse, one has to show first that the
image 7r(F) of a closed set F c if is a closed set in M; this follows from
the fact that for each p E M, the inverse image 7r- I (p) has two points.
Now, if if is not connected, denote by C a connected component of if.
Then 7r(C) is an open, closed and nonvoid subset of M, hence 7r(C) = M.
Therefore, if is the disjoint union of two connected components and 7l"
applies diffeomorphically each such component onto M. It follows that
M is orient able.
17) (A non-Hausdorff manifold). Let S be the set given by the disjoint union
of R2 with a point p*. Let fI and 12 maps of R2 in S defined by
54 3. Differentiable Manifolds

/1(u, v) = h(u, v), if (u, v) :I (0,0), (u, v) E R2,

/1(0,0) = (0,0),
12(0,0) = p*.

Show that (R2 , Ii), i = 1,2, is a differentiable structure in S the topology

of which does not satisfy the axiom of Hausdorff.
4. Integration on Manifolds;
Stokes Theorem and Poincare's Lemma

1. Integration of Differential Forms

In this section we will define the integral of a differential n-form on an n-

dimensional differentiable manifold. We will start with the case of R n.
Let W be a differential form defined in an open set U C Mn. The support
K of w is the closure of the set
A = {p E Mnj w(P)"# O}.
Let W be a n-form and Mn = Rn. Then

w = a(xl, . .. ,Xn)dXl /I. ••• /I. dx n .

Assume that the support K of w is compact and contained in U. We define

L= w [adxl ... dxn,

where the right hand side is the usual multiple integral in R n.

We will now proceed to the definition of the integral of an n-form on
Mn. To avoid convergence problems, it is convenient to assume that M is
compactj then the support K of w, being a closed set in a compact space, is
also compact. As we will see in a while, it is also necessary to assume that M
is orientable, i.e., that M is covered by a family of coordinate neighborhoods
{Va} such that the coordinate changes have positive jacobians.
Let us assume initially that K is contained in some coordinate neighbor-
hood Va = fa(Uo} Then, if the local representation Wa of win Ua is

we define
1 1 J
M
w=
v",
Wa =
U",
aadXl, ... ,dxn,

where the right hand side is an integral in R n.

It may happen that K is contained in another coordinate neighborhood
V,B = f,B(U,B) of the same family, and we must show that the above definition
is independent of the choice of the coordinate neighborhood.
56 4. Integration on Manifolds; Stokes Theorem and Poincare's Lemma

For that, we can assume, by contracting UOl and U/3 if necessary, that
VOl = V/3' Let the change of coordinates
1 = 1;1 01/3: U/3 -+ UOl
be given by
Xi = J;(Yl,"" Yn), i = 1, ... , n,
(XlI""X n) E UOl , (Yl, ... ,Yn) E U/3'
Since w/3 = /*(wOl ), we obtain that
w/3 = det(df)a/3dYI 1\ •.• 1\ dYn>
where
a/3 =aOl (/l(Yl, .. ·,Yn),···,ln(Yl,··.,Yn))'
On the other hand, by the formula of change of variables for multiple integrals
in R n, we obtain that

Thus, since det(df) > 0,

J J
Va
WOl =
Vp
W/3,
hence the asserted independence.
Notice that without the hypothesis of orientability for M, the sign of the
integral of W is not well defined. The choice of an orientation for M fixes a
sign for the integral of W which changes with the change of orientation.
Let us now consider the case in which the support K of W is contained
in no coordinate neighborhood. For that, we need some preliminaries, and
before going into details, we will present a sketch of what we intend to do.
Given a covering {VOl} of a compact differentiable manifold M, we will
construct a finite family of differentiable functions cpt, ... , CPm such that:
m

a) LCPi = 1,
i=1

b) 0:5 CPi :5 1, and the support of CPi is contained in some VOl; = Vi.
The family {CPi} is called a differentiable partition 01 unity subordinate to
the covering {VOl}' (When M is orientable, we choose {VOl} compatible with
the orientation).
Let us assume, for the time being, the existence of such a family. Assume
furthermore that M is orientable. We will define the integral of an n-form W
on M n as follows.
The support of the form CPiW is contained in Vi. By a previous definition,
it makes sense to write
 1. Integration of Differential Forms 57

[
1M
W = fi=11M[ CPiW ,

the only question being whether this definition is independent of the choices
made.
Consider another covering {W.a} of M which determines on M the same
orientation as {Va}, and let {""i}, j = 1, ... ,s, be a partition of unity sub-
ordinate to {W.a}. Then {Va n W.a} will be a covering for M and the family
CPi""i will be a partition of unity subordinate to {Va n W.a}. Thus

where in the last equality it was used that, for each i, the functions CPi""i are
defined in Vi. Similarly,

which proves the required independence.

Remark 1. What we have done was essentially the following. We observed

that the integral of a differential form whose domain is contained in a coor-
dinate neighborhood reduces to a multiple integral. To integrate differential
forms in more complicated domains we can proceed in either of the two fol-
lowing ways: We divide the complicated domain in simpler domains, and add
up the results, or we decompose the form into forms that are zero outside
simple domains and add up the results. Since it is easier to work with func-
tions than with domains, the second alternative is usually preferred, and it
was the one used here.

We now go into the proof of existence of a differentiable partition of unity

subordinate to a given covering by coordinate neighborhoods of a compact
differentiable manifold Mn (we will require no orient ability for that).
In what follows, Br(O) = {p ERn; Ipl < r}.

Lemma 1. There exists a differentiable /unction cP: B3(0) -+ R such that:

a) cP(p) = 1, if p E B1 (0)
b) 0 < cP(p) $ 1, if p E B 2 (O)
c) cP(p) = 0, if p E B3(O) - B2(O).

Proof. We first consider the function 0:: R -+ R given by (Fig. 1 (a»

o:(t) = e-(t+l)(1+2), tE(-2,-I).
o:(t) = 0 t~(-2,-I).
58 4. Integration on Manifolds; Stokes Theorem and Poincare's Lemma

Notice that the function 0: is a simple modification of the well known

function e-(I/X2 }, and the point is that it is Coo everywhere.
Now take the integral

'Y(t) = J~oo o:(s)ds

to obtain a differentiable function 'Y (Fig. 1, (b)) whose maximum value (at
t = -1) is given by J:21 0:(s)ds = A. Then, by setting (3(t) = 'Y(t)/A, we
obtain a differentiable function with the following properties:
(3(t) = 0, if t -2,
°< (3(t)
~

~ 1, if tE(-2,-I),
(3(t) = 1, if t~-l.

The required function i.p: B3(0) -+ R is obtained by i.p(P) = (3( -lpD, p E

B 3 (0)j for the case of R2 it has the form of Fig. 1 (c). 0

- 1

-2 o -2 -1 o
lea) /l(b)

l(c)

Fig.4.1(a) Fig.4.1(b) Fig.4.1(c)

Lemma 2. Let Mn be a differentiable manifold, let p E M and let g: U c

R n -+ M be a parametrization around p. Then, it is possible to obtain a
parametrization f: B3(0) -+ M around p in such a way that f(B3(0)) c g(U)
and that f-l(P) = (0, ... ,0).

°
Proof. Let (x~, ... ,x~) E U be such that g(x~, ... ,x~) = p. Since U is open,
there exists an r > such that Br(x~, ... , x~) c U. Let T the translation
in R n that takes (x~, ... , x~) to (0, ... ,0), and let H: R n -+ R n be the map
 1. Integration of Differential Forms 59

that to each p E R n associates the point ~p. Then HoT takes Br(xY, . .. ,x~)
to B3(O).

We define the parametrization f: B3(O) -+ M by

f = 9 0 T- I 0 H- I
which is easily seen to satisfy the required conditions. o
Proposition 1. (Existence of a differentiable partition of unity). Let M be
a compact manifold and let {Va} be a covering of M by coordinate neighbor-
hoods. Then there exist differentiable junctions 'PI, ... ,'Pm such that;
m

a) L'Pi =1
i=l
b) o $ 'Pi $ 1, and the support of 'Pi is contained in some Val of the covering
{Va}.

Proof. For each p E M consider the parametrization fp: B 3(O) -+ M given

by Lemma 2 with fp(B3(O) = Vp C Va, for some Va of the covering {Va}. Set
Wp = fp(Bl(O)) C Vp.

The family {Wp } is an open covering of M. Since M is compact, we can

select from it a finite covering WI,.'" Wm. The corresponding VI,"" Vm
will make up a covering of M.
Let us define functions (Ji: M -+ R, i = 1, ... , m, by
(Ji = 'P 0 f i- l in Vii (Ji = 0 in M - Vi,
where 'P: B3(O) -+ R is the function given by Lemma 1. The functions (Ji are
differentiable and the support of (Ji is contained in Vi.
Finally define 'Pi by
(Ji(P)
'Pi(P) = L:j=l (Jj(P) , p E M.

It is immediate to check that the functions 'Pi so constructed satisfy the

conditions (a) and (b). 0

Remark 2. The existence of a differentiable partition of unity is one of the

most useful facts for the study of global questions on differentiable manifolds.
Proposition 1 still holds for noncompact manifolds (with countable basis) by
considering locally finite countable coverings (locally finite means that each
point of the manifold meets only a finite number of members of the covering),
and the proof is essentially the same. That every manifold has a locally finite
countable covering can be found, for instance, in F. Warner, [WAR].
60 4. Integration on Manifolds; Stokes Theorem and Poincare's Lemma

2. Stokes Theorem

In this section we intend to establish Stokes theorem. For that, we will need a
series of definitions that will make it possible to state precisely the theorem;
once stated, the proof of the theorem is relatively simple.
For the two-dimensional case, a rough description of the theorem is as
follows.
Let w be a differential I-form defined on a two-dimensional, oriented,
manifold M2, and let dw be its exterior differential. Consider a region R of
M2 bounded by a closed regular curve C = oR. The orientation of R induces
an orientation on C, and the inclusion i: C ---. M allows us to consider the
restriction i*w of w to C. Under these conditions, Stokes theorem states that
the integral of the 2-form dw in R is equal to the integral of i*w in oR = C.
So, in a certain sense, the operators d (applied to forms) and 0 (applied to
smooth domains) are dual to each other.
If, in particular, M2 = R2 and w = Pdx + Qdy, Stokes theorem reduces
to Gauss theorem:

Jl (~~ -~:) dxdy = L Pdx+Qdy.

We now start to present the definitions we need and which are useful in
other contexts. The first one is an extension of the notion of a manifold to
include manifolds with "boundary". The definition of a manifold does not
include, for instance, the set M, given by
M = {(x, y, z) E R 3 ; Z = x 2 + y2, z:::; Zo, Zo > O}
(M is the closed set of the rotation paraboloid bounded above by Z = zo),
because the intersection VnM of any neighborhood V of a point p = (x, y, zoJ
in the "boundary" of M with M is not homeomorphic to an open set of R
(Fig. 4.2). Notice, however, that V n M is homeomorphic to an open set of
the closed half-space {(Xl,X2) E R2; Xl :::; OJ, whereas points of M that are
not in the boundary behave as points in a 2-manifold. This suggests a new
definition that will include the above situation.
A hall-space 01 R n is the set
H n = {(xt. ... ,Xn) ERn; Xl :::; OJ.
An open set of Hn is the intersection with Hn of an open U of R n.
We say that a function I: V ---. R defined in an open set V of Hn is
differentiable if there exists an open set U :::> V and a differentiable function
7 in U such that the restriction of 7 to V is equal to I. In this case, the
differential dIp, p E V, of I at p is defined to be dIp = d7 p.
When V does not contain points of the form (0, X2, ... , Xn), V is an open
set of R n and the definition of dIp agrees with the usual one. If p is of the form
 2. Stokes Theorem 61

-
(X,1I,Zo)

-11

Fig. 4.2

(0,X2,'" ,xn ), dIp is defined for all tangent vector of curves in U passing
through p, i.e., for all vectors in R n with origin in p. Using such curves it is
easy to show that the definition of dIp is independent of the extension 1 of I.
In a similar way, we define a differentiable map I: V -+ R n •

Definition 1. An n-dimensional differentiable manifold with (regular) bound-

ary is a set M and a family of injective maps fa: Ua C Hn -+ M of open sets
of Hn into M such that:
1) Ufa(Ua ) = M.
a
2) For all pairs a,/3 with la(Ua) n IfJ(UfJ) = w of ¢ the sets f;-1(W) and
lil(W) are open sets in Hn and the maps fil 0/01, f;;1 0 f{3 are differ-
entiable.
3) The family {(Ua , fa)} is maximal relative to (1) and (2).

A point p E M is said to be a point in the boundary of M if for some

parametrization f: U C Hn -+ M around p we have that f(O, X2,"" xn) = p.

Lemma 3. The definition of point in the boundary does not depend on

parametrizations.

Proof. Let h: UI -+ M be a parametrization around p such that h(q) = p,

q = (0, X2,
••• , x n ). /_C>

Assume, by contradiction, that for some parametrization 12: U2 -+ M

around p we have li 1 (P) = q2 = (Xl. ... , xn) with Xl of 0. (Fig. 4.3)
Let W = h(UI) n h(U2}. The map
111012: fi1(W} -+ Ill(W)
62 4. Integration on Manifolds; Stokes Theorem and Poincare's Lemma

'"
q

q2

;---
11 012

Fig. 4.3

is a diffeomorphism. Since Xl # 0, there exists a neighborhood U of Q2,

U c l;l(W), that does not intersect the xl-axis. Restricting III 012 to U,
we will have a differentiable map

III 012: U - t H n
such that the determinant of d(j-l oh)ql is nonzero. By the inverse function
theorem, 11-1 012 will take a neighborhood V C U of Q2 diffeomorphically
onto III 0 h(V). But then f1l 0 h(v) would contain points of the form
(Xl, ... , xn) with Xl > 0 which are not in Hn. This yields a contradiction
and completes the proof. 0

The set of points in the boundary of M is therefore well defined; it is

called the boundary of M and denoted by 8M. If 8M = cp, Definition I
agrees with the definition of a differentiable manifold given in Chapter 2.
The definitions of differentiable functions, tangent space, orient ability, etc,
for manifolds with boundary are introduced in exactly the same way as the
corresponding definitions for differentiable manifolds, with the additional care
of replacing R n by Hn.

Proposition 2. The boundary 8M of an n-dimensional differentiable man-

ifold M with boundary is an (n - I)-differentiable manifold. Furthermore, if
M is orientable, an orientation for M induces an orientation for 8M.

Proof. Let p EM be a point in the boundary of M and let fo:: Uo: C Hn - t

Mn be a parametrization around p. Then f;;l(P) = Q = (0, X2,"" xn) E Uo:.

Let
 2. Stokes Theorem 63

By identifying the set {(Xl. ... ,Xn) E Rnj Xl = O} with Rn-l, we see that
Ua is an open set in R n - 1 . By denoting by lathe restriction of fa to U a,
we see, by Lemma 3, that 101 (U a) C 8M. Finally, by letting p run in the
points of 8M, we easily check that the family {( U a,]a)} is a differentiable
structure for 8M. This proves the first part of the Proposition.
To prove the second part, assume that M is orient able and choose an
orientation for M, i.e., a differentiable structure {(Ua , fa)} such that the
changes of coordinates have positive jacobian. Consider the elements of
the family that satisfy the condition fa(Ua) n 8M i: ¢. Then the family
{(U a,fa)} described in the first part is a differentiable structure for 8M. We
want to show that if 1a (U a) n 1{3 (U {3) i: ¢, the change of coordinates has
positive jacobian, i.e., that
--1 -
det(d(J a 0 f {3)q) > 0,
for all q whose image, by some parametrization, is in the boundary.
Observe that the change of coordinates fa 0 til takes a point of the form
(0, xg, ... , x~) into a point of the form (0, x~, ... , x~). Thus, for a point q
whose image is in the boundary,
1 8x~ --1-
det(d(J;: 0 f{3) = -(J det(d(J a 0 f {3))'
8X1

But ~8xQ > 0, because

Xl
xl = 0 in q = (0, x~, . .. , x~), and both x~ and xf are
negative in a neighborhood of p. Since det(d(J;:l 0 f{3)) > 0, by hypothesis,
--1 -
we conclude that det(d(J a 0 f {3)) > 0, as we wished. 0

We can now state and prove Stokes theorem.

Theorem 1. Let Mn be a differentiable manifold with boundary, compact

and oriented. Let w be a differential (n - I)-form on M, and let i: 8M -+ M
be the inclusion map of the boundary 8M into M. Then

r
JaM
i*w = r dw.
JM
Proof. Let K be the support of w. We will consider the following cases:
A) K is contained in some coordinate neighborhood V = f(U) of a
parametrization f: U c Hn --t M. In U,
n
W= L ajdX1 /\ ... /\ dXj_1 /\ dXj+l/\"'/\ dXn,
j=l

where aj = aj(x1, ... , xn) is a differentiable function on U. Thus

64 4. Integration on Manifolds; Stokes Theorem and Poincare's Lemma

AI) Assume first that f(U)noM = ¢J. Then w is zero in oM and i·w = O.
Thus
f i·w = O.
IBM
We will show that

For that, extend the functions aj to Hn by setting

aj(xI, ... ,xn ) = aj(xl, .. . ,xn ), if (x}, . .. ,x n ) E U

aj(xI, ... ,xn ) = 0, if (Xl, •.• ,xn ) E Hn - U
Since f-I(K) C U, the functions aj so defined are differentiable in HU. Now
let Q C H n be a parallelepiped given by x} ~ x j ~ x~, j = 1, ... , n, and
containing f-I(K) in its interior (Fig. 4.4).

Fig. 4.4
 2, Stokes Theorem 65

Then

1("{, )F'-I x,
u,
~ -1 aaj ) d Xl ' .. dX n = "
-a, ~( -1 ) '
,.
'-Ii
Q
aaj,dXl'" dX n
-a
x,

'''-If
,= ~(-IF [aj{xt. ... ,Xj-I,Xj,Xj+I,
0 ... Xn)

- aj{xt. ... ,Xj-l, x}, Xj+t. ..• ,Xn)]dXI ... dXj-1 dXj+1 ••. dXn = 0,

since aj(X1, .. ' ,x~, ... ,xn) = aj(xI,'" ,x}, ... ,xn ) = 0, for all j.
A2 ) Assume now that f{U) n aM :F ¢. Then the inclusion map i can be
written as: Xl = 0, Xj = Xj' Thus, using the induced orientation on the
boundary,
i*w = a1(0, X2,.'" Xn)dX2 1\ •.• 1\ dxn.
As in case (AI)' we will extend the functions aj to Hn, and will consider the
parallelepiped Q given by
j = 2, ... ,n

and such that the union of the interior of Q with the hyperplane
contains f-I(K), Then
Xl = °
1 dw = ~()'
~ -1'- 11 aaj
-.dXI···dxn
Q ax,

k
M j=l

= [a1 (0, X2,' .. ,Xn ) - al (xL X2,'" , Xn )]dX2,'" ,dxn

+~
j=2
.11
L...,,( -1)'-
Q
0
[aj(Xl,"" Xj"", I
Xn) - aj(XI,"', Xj"", Xn)]

dXl'" dXj-l dXj+l'" dxn.

Since aj(xt. . .. ,x~,. '. ,Xn ) = aj(xt., " ,x}" " ,Xn ) = 0, for j = 2, ... ,n,
and al(xi,x2,"" xn) = 0, we obtain

B) Let us now consider the general case. Let {Va} be a covering of

M by coordinate neighborhoods compatible with the orientation, and let
CPI, ••• ,l{Jm be a differentiable partition of unity subordinate to Va' The forms
Wj = I{JjW, j = 1, ... ,m satisfy the conditions of case A. Furthermore, since
E j dl{Jj = 0, we have
66 4. Integration on Manifolds; Stokes Theorem and Poincare's Lemma

Therefore,

o
Example. Let M be a bounded region of R3 such that the boundary 8M
of M is a regular hypersurface of R 3 j M is then a compact 3-dimensional
manifold with boundary 8M. Let v be a differentiable vector field in R 3 , and
let w be the I-form in R3 dual to v in the natural inner product of R3. Then
(cf. Exercise 11, Chap. 1) d(*w) = (div v)v, where v is the volume element
ofR3 •
Now choose an orientation for R3 and let N be the unit normal vector of
8M in the induced orientation. Finally, let a be the area element of 8M.
Consider, in a neighborhood U c R 3 of p EM, differentiable orthonormal
fields el, e2, N such that, in the points of 8M, el and e2 are tangent to 8M.
Then
i- * w(eb e2) = w(N) =< v, N >,
i.e., i-( *w) = (v, N)a. Thus, in this case, Stokes theorem

can be written as
[ div v v = [ < v, N > a
1M IBM
which is the well known divergence theorem in Analysis.

Remark. The divergence theorem is a fundamental tool in Analysis (see, for

instance, the beautiful account in O. Kellog, [KELLJ). Although we have
proved it for regular boundaries and smooth functions, it can be generalized
considerably. See, for instance, the article of D. Figueiredo, [FIG].

3. Poincare's Lemma

Let Mn be a differentiable manifold. A differential k-form w is said to be

exact if there exists a (k - I)-form {3 such that d{3 == Wj w is said to be closed
if dw = 0. Since d2 = 0, an exact form is closed.
The converse of the above fact does not hold in general. For instance, let
w = x~,~:gx be defined in R2 - {(O, On = U. It is easily checked that dw = 0,
 3. Poincare's Lemma 67

i.e., W is closed, but there exists no differentiable function 9 in U such that

dg = w; otherwise, by Stokes theorem,

1 =1
e
w
e
dg ={
Joe 9 = 0,
Ie
and this contradicts the fact, easily computable, that W = 211'. It is possible,
however, to show that for each p E U there exists a neighborhood V C U of
p and a differentiable function gv in V such that dgv = w.
In this section, we will show that the situation of this example is com-
pletely general, that is, that the condition dw = 0 is a sufficient condition for
w to be locally exact (Those who are familiar with the material of Chapter 2
will notice that we are generalizing Theorem 1 ofthat Chapter). Actually, we
will prove the result in a form slightly more general which is more convenient
for the applications.

Definition 2. A differentiable manifold M is contractible(to some point Po E

M) if there exists a differentiable map H: M x R ---+ M, H (p, t) EM, P EM,
t E R such that

H(p, 1) = p, H(p,O) = Po, for all p E M.

It is easy to see that R n is contractible to an arbitrary point Po ERn; it

suffices to define H(p, t) = Po + (p - Po)t. The same argument shows that the
ball Br(O) = {p ERn; Ipi < r} is contractible to the origin O. It follows that
any differentiable manifold is locally contractible.

Theorem 2 (Poincare's lemma) Let M be a contractible differentiable

manifold, and let w be a differentiable k-form in M with dw = O. Then w is
exact, i.e., there exists a (k - I)-form 0 in M such that do = w.

Proof. Let 11': M x R ---+ M be the projection 1I'(p, t) = p, and let w be the
k-form on M x R given by w = H*w, where H is the map given in the
definition of contract ability. We will need the following lemma.

Lemma 4. Every k-form win M x R can be written uniquely as

w= Wl + dt 1\ "I, (1)

where Wl is a k-form on M x R with the property that Wl (Vb .. . , Vk) = 0,

if some Vi, i = 1, ... , k, belongs to the kernel of d1l', and "I is a (k - I)-form
with a similar property.

Proof of Lemma 4. Let p E M and let f : U ---+ M be a parametrization

around p. Then feU) x R is a coordinate neighborhood of M x R, with
coordinates, say, (Xl, . .. , x n , t). In feU) x R, w can be written as
68 4. Integration on Manifolds; Stokes Theorem and Poincare's Lemma

+ dt /\ L bh ... j"_1 dXh

(2)
/\ ... /\ dXj"_l
= wI /\ dt /\ 11.
It is clear that WI and 11 have the required properties. Furthermore, if the
decomposition (I) holds in all of M, it has to be locally of the form (2),
hence it is unique. To prove existence, we define WI and 11 on each coordinate
neighborhood by (2). In the intersection of two such neighborhoods, the def-
initions agree by uniqueness, thus WI and 11 can be extended to the whole M
satisfying (I). This proves Lemma 4.

Now let it : M --+ M x R the map given by it(P) = (p, t); it is the inclusion
of Minto M x R at the "level" t. We will define a map I that takes k-forms of
M x R into {k -I)-forms of M as follows: If p EM and VI! V2, ••• , Vk E TpM,
then at p,

(Iw) {VI , ... , Vk-t} = 101 {11(P, t)(dit{vt}, ... , dit(Vk-t})}dt,

where 11 is given by the decomposition w = WI + dt /\ 11 of Lemma 4.
The crucial point of the theorem is contained in the following lemma.

Lemma 5. iiw - iiiw = d(IW) + I{dW).

Prool 01 Lemma 5. Let p E M. We will use the coordinate system (Xl>""
X n , t)
introduced in Lemma 4. We first notice that the operation I is additive,
i.e., I(wl + W2) = I(wt) + I(w2). It follows that it suffices to consider the
following two cases: a) w = I dXil /\ ... /\ dXik j b) w = I dt /\ dXil 1\ ... /\
dxi"_l'
Case (a). Ifw = Idxil /\ ... /\ dXi" , then dW = ¥tdt /\ dXil /\ ... /\ dXi" + terms
without dt.
Notice that in the coordinate systems (XI! ... , Xn> t) the operation I
amounts to integrate the local representations of walong the second factor t.
Therefore,
.I(dW)(P) = ( 10fl of )
at dt dXil /\ ... /\ dXi"
= (f(P, 1) - f(P, O»dXil /\ ... /\ dXik
= iiw(P) - i~w(P).

Since Iw = 0, we conclude the lemma in case (a).

Case (b). If w = fdt 1\ dXil 1\ ... /\ dXik_l" then iiw = 0 = iiiw. On the other
hand,
 3. Poincare's Lemma 69

Therefore,

and

which completes the Case (b), and the proof of the lemma. o
Now we can complete the proof of the theorem (notice that so far we have
not used that dw = 0). Since M is contractible,
H 0 il = identity, H 0 io = const. = Po E M.
Thus
w = (H 0 iI)*w = ii{H*w) = i~w,
0= (H 0 io)*w= io(H*w) = iow.
Now, since dw = 0, we obtain that dW = H*dw = O. It follows by Lemma 5
that
w = iiw = d(Iw) = do:,
where 0: = lW. o

EXERCISES

1) Let f : R3 -+ R be a differentiable function and let w be the 2-form in

R3 given by
f:rdy "dz + f'/ldz " dx + fzdx "dy
W= .
Jf~ + f~ + f~
It is well known that if a E R is a regular value of f (that is, for all
P E f-l(a), the map dip is surjective) then

M2 = ((x,y,z) E R3if(x,y,z) = a}
70 4. Integration on Manifolds; Stokes Theorem and Poincare's Lemma

is a regular orient able surface in R3 (cf. [de], §2.2). Show that the
restriction of W to M2 is the element of area of M2.
Hint: We want to show that if {VbV2} is a positive basis of Tp(M),
p E M then W(Vb V2) = area of the parallelogram made up by VI and V2.
Choose a parametrization 9( u, v) of M around p, compatible with the
orientation, and notice that

dy /I. dz + dz /I. dx + dx /I. dy = (Z)9u /I. 9v)i)du /I. dv,

i

where (9u /l.9v)i is the i-th coordinate ofthe vector product 9u /l.9v in the
canonical basis ofR3. Since I(x,y, z) = const., the vector (Ix, I", Iz) = A
lies along the positive normal of M. Thus

W=
(A,9u
IAI/I. 9v) du /I. dv = I9u /I. 9v Idu /I. dv.
It is now easy to check that W(9u,9v) = area (9u,9v), hence W(Vl,V2) =
area (Vb V2)'
2) a) Let W = xdy - ydx and j: M <-+ R2 the inclusion of a bounded
region with regular boundary 8M. Show that the area of M is given by
(1/2) JOM rw.
b) Let W = xdy /I. dz - ydx /I. dz + zdx /I. dy and j: MeR3 the inclusion
of a bounded region with regular boundary 8M. Show that the volume
of M is given by (1/3) JOM rw.
c) Generalize the above for R n.
3) Let
xdy /I. dz + ydz /I. dx + zdx /I. dy
(x 2 + y2 + z2)3/2
w=~~~~~~~~~--~

a 2-form defined in R3 - {O}, and let M2 C R3 be an oriented surface

that does not pass through the origin 0 = (0,0,0). Show that:
a) The restriction of w to M2 is equal to
cosO
-2-a ,
r
where a is the area element of M2, r is the distance from 0 to a
point p E M2 and 0 is the positive angle from Op to the unit positive
normal N to M2 at p.
Hint: Proceed is in Exercise 1. Set p = (x, y, z), r2 = x 2 + y2 + Z2,
p/r = v. One obtains:
1 1
w = 3(P,9u /I. 9v)du /I. dv = "2(v,N) 19u /I. 9vl du /I. dv
r r
cosO
=--a.
r2
 3. Poincare's Lemma 71

b) Define the solid angle under which M2 is seen from 0 as

(for justification sake, notice that if p E M2, cos () ¥- 0, and LlM is a

small neighborhood around p, then (l/r 2) cos () (area LlM) is the area
of the region of the unit sphere with center 0 that is determined by
the rays that join 0 to points of LlM; this area is usually called the
solid angle under which LlM is see from 0). Now let M be a bounded
region in R3 with regular boundary 8M such that 0 ¢ 8M, and let
n be the solid angle under which 8M is seen from o. Show that

n = 0, if 0 ¢ M, and n = 471", if 0 E M.

Hint: Notice that if 0 ¢ M, dw = 0, and apply Stokes theorem.

4) Let cp: R3 ----t R a differentiable function, homogenous of degree k (that
is, cp(tx,ty,tz) = tkcp(x,y,z». Show that:
a) If B = {p E R3; Ipi :::; I} is the region bounded by the unit sphere 8 2 ,
then

JBf Ll cp dx 1\ dy 1\ dz = J5f kcp u,

2
2

where u is the area element of 8 2 and Ll2cp = cpzz + cpyy + cpzz is the
Laplacian of cpo
Hint: Notice that by Euler's relation for homogeneous functions (cf.
Exercise 18, Chapter 1) xcpz+ycpy+zcpz = kcp, and use the divergence
theorem.
b) Let cp = alx4 + a2y4 + a3z4 + 3a4x2y2 + 3aSy2z2 + 3a6x2z2, then

1 52
cp U
471"
= -5 La;.
6

;=1

5) Let 9 : R3 ----t R, I : R3 ----t R be differentiable functions, and let M3 C

R3 be a compact differentiable manifold with boundary 8M2 • Prove that:
a) (first Green's identity)

f (grad/,gradg}lI+ f ILl g 2 f l(gradg,N}u,

JM JM 11=
JaM
where 11 and u are, respectively, the volume element of M and the
area element of 8M, and N is the unit normal of 8M.
Hint: Set v = I grad 9 in the divergence theorem.
b) (second Green's identity)

JMf (J Ll 9 - gLl2 J)II = f (J (grad g, N) - g(grad I, N} )u.

2
JaM
72 4. Integration on Manifolds; Stokes Theorem and Poincare's Lemma

6) Can one find a three-dimensional orientable differentiable manifold M3

whose boundary is the real projective plane?
7) Let WI and W2 be differential forms on a differentiable manifold M. As-
sume that WI and W2 are closed and that W2 is exact. Show that WI 1\ W2
is closed and exact.
8) Let MR be a compact orientable manifold without boundary (i.e., 8M =
0) and let W be a differential (n - I)-form on MR. Show that there exists
a point p E M such that dw(P) = O.
9) Show that there exists no immersion f : 8 1 -+ R of the unit circle into
the real line R.
Hint: Use Exercise 8.
10) Let M2 C R3 be a compact, oriented, regular surface with regular bound-
ary 8M, and let v be a differentiable vector field in an open set of R 3
containing M2.
a) Show that
[ (rot v, N}u = [ (v, t}ds
1M2 10M
where N is the unit normal field, u the area element of M2, t
the
unit tangent vector to 8M, and ds the element of arc of 8M.
Hint: Notice that rot v = *(dw), where W is the I-form dual to v
in the natural inner product of R3 (cf. Exercise 14, Chapter 1). By
choosing local orthonormal fields el, e2, N such that el and e2 are
tangent to M and el is tangent to 8M, we obtain
dw(ebe2) = (*dw)(N) = (rot v,N),
w(et} = (v, el) = (v, t),

that is, dw = (rot v, N}u and i*w = (v, t}ds. Now apply Stokes
theorem.
b) Let p E R 3 , ~ a unit vector in R! and P the plane normal to ~,
passing through p and whose orientation together with ~ gives the
orientation of R3. Consider a disk D c P, with center p, and apply
(a) to the surface made up of D and its boundary 8D to obtain

(rot v, ~}(p) = lim ~D [ (v, t}ds,

D--.p area 10 D

where the limit is taken when D runs in a family of concentric disks

that approach p.
11) (Introduction to potential theory in R 3 )
A differentiable function 9 : R3 -+ R is said to be harmonic in a subset
Be R3 if £l2 g = 0 for all p E B. Let M C R3 be a bounded region with
regular boundary 8M. Prove that:
 3. Poincare's Lemma 73

a) H 91 and 92 are harmonic in M and 91 = 92 in oM, then 91 = 92 in

M.
Hint: Use Green's first identity (Exercise 5a) with f = 9 = 91 - 92

:t
b) H 9 is harmonic in M and

~f (grad9,N) = 0
in oM, where N is the unit normal vector of oM, then 9 = const. in
M.
Hint: Use Green's first identity with f = 9.
c) If 91 and 92 are harmonic in M and
091 092
oN = oN
in oM, then 91 = 92+ const. in M.
d) If 9 is harmonic in M, then

f a9 u = 0
iOM oN
e) The function (Z:a+II:a~%:aP/2 is harmonic in R3 - {o}.
f) (Mean value theorem). Let f be harmonic in the region

Br = {p E R 3 j Ip - Pol 2 $ r2}
whose boundary is the sphere Sr with center in Po. Then

f(Po) = 4:r2l.. fu
Hint: Use Green's second identity in the region D = Br - Bp, P < r,
with f = f and 9 = l/r. Since 9 and f are harmonic,

isf (f~(!)
aN r
-!raNaf ) u = is..
f (f~(!) _! af ) u.
aN r raN
p

Since !N(l/r) = f,,(l/r) = -1/r2, we obtain from (d),

4:p2lp fu = 4:r2l.. fu.

Now let P -+ 0 to obtain the desired conclusion.
g) (The maximum principle). Let f be a nonconstant harmonic function
in a closed bounded region Me R3 (i.e., M is the union of a bounded
connected open set with its boundary which is not necessarily regu-
lar). Then f reaches the maximum and the minimum in the boundary
aMofM.
74 4. Integration on Manifolds; Stokes Theorem and Poincare's Lemma

Hint: Assume that !(P) is maximum, p E M - 8M and consider a

ball B C M - 8M with center p and such that !(P) ~ !(q), for all
q E B. Show this contradicts (f).
12) Let M" be a compact differentiable manifold without boundary. Show
that M is orient able if and only if there exists a differential n-form w
defined on M and which is everywhere nonzero.
Hint: For the "only if" part use a partition of unity to construct a
nonzero n-form globally defined on M.
13) Let M be a compact, orientable, differentiable manifold without bound-
ary. Show that M is not contractible to a point.
Hint: Use Exercise 12, Poincare's lemma and Stokes theorem.
14) Let A, Band C be differentiable functions in R3 and consider the differ-
ential system

{ 7Jii
8R_~=A
Oz
op _ ~ =B
~ of. - C
7f%-F,i-
where P, Q and R are unknown functions in R3.
a) Show that a necessary and sufficient condition for a solution to the
above system to exist is that
8A 8B 8C_ O
8x + 8y + 8z - .

Hint: Consider in R3 the differential form

w = AdyAdz+ BdzAdx+ CdxAdy

and notice that dw = (~+ ~ + !!iz)

dx A dy A dz. By Poincare's
lemma, dw = 0 if and only if there exists a form a = Pdx + Qdy + Rdz
with da = Wi this last condition is precisely the above system.
b) Assume the above condition to be satisfied and determine the func-
tions P, Q, R.
Hint: Consider the contraction H(P, t) = tp of R3 to (0,0,0). Then
w= H*w = A(tx, ty, tz)(ytdt Adz - ztdt A dy) + ... +
+ terms without dt.
Thus the form a of (a) is given by

a = rw = (fal A(tx, ty, tz )tdt) (ydz - zdy) + ...

15) Let v be a differentiable vector field in R3. Prove that:

a) If div v = 0, then there exists a vector field u in R3 such that rot
u=v.
 3. Poincare's Lemma 75

b) If rot v =0, then there exists a function f in R 3 such that grad f = v.

16) {Brouwer fixed point theorem}.
a) Let Mn be a compact, orient able, differentiable manifold with bound-
ary 8M 'f: q,. Show that there exists no differentiable map f : M -
8M such that the restriction fl8M is the identity.
Hint (following E. Lima): Assume the existence of such an f, and let
w be the nonzero (n - I)-form on aM given by Exercise 12. Clearly
dU*w) = !*(dw) = 0, hence

0= [ d(/*w) = [ i* /*w = [ w 'f: 0,

1M IBM IBM
and this is a contradiction.
b) Prove the Brouwer fixed point theorem: Let B eRn be the ball
{p ERn; Ipi S I}. Every differentiable map g: B - B has a fixed
point, i.e., there exists q E B such that g(q) = q.
Hint: If g(p) 'f: p, for all p E B, the half-line starting in g(p) and
passing through p intersects aB in a unique point, say q = f(P). The
map f: B - aB so defined satisfy the conditions of (a) and yields a
contradiction.
5. Differential Geometry of Surfaces

1. The Structure Equations of R n

We now apply our knowledge of differential forms to study some differential

geometry. We start with a few definitions.
A Riemannian manifold is a differentiable manifold M and a choice, for
each point p E M, of a positive definite inner product C. )p in TpM which
varies differentiably with p in the following sense: If X and Y are differentiable
vector fields in M, the function p ....... (X, Y)p is differentiable in M. The inner
product ( , ) is usually called a Riemannian metric on M.
The notion of equivalence between Riemannian manifolds is the notion of
isometry. A diffeomorphism <p: M -+ M' between Riemannian manifolds M
and M' is an isometry if for all p and all pairs x, Y E TpM, we have

The importance of the notion of Riemannian manifold is that we can de-

fine on it the usual metric notions (length, area, angles, etc.) of the euclidean
geometry. Actually, euclidean geometry is just the study of metric notions in
the simplest Riemannian geometry, namely, R n endowed with the following
inner product: If x = (Xl!"" xn) and Y = (Yl,'" ,Yn) are vectors in R n , one
defines
(x, y) = XIYl + ... + XnYn'
Although R n is the simplest Riemannian manifold, it is, in a certain sense,
the universal Riemannian manifold. We hope to make this clearer later.
We will begin, therefore, by establishing the so-called structure equations
ofRn.
Let U eRn be an open set and let el, ... ,en be n differentiable vector
fields such that for each p E U, (ei,ej)p = 6ij, where 6ij = 0 if i # j and
6ij = 1 if i = j. Such a set of vector fields is called an orthonormal moving
frame. From now on, we will omit the adjective orthonormal.
Given the moving frame lei}, i = 1, ... , n, we can define differential 1-
forms Wi by the condition wi(ej) = 6ij, j = 1, ... , nj in other words, at each
p, the basis {(Wi)p} is the dual basis of {(ei)p}. The set offorms {Wi} is called
the coframe associated to lei}.
78 5. Differential Geometry of Surfaces

Each vector field ei is a differentiable map ei: U eRn - t R n. The differ-

ential at p E U, (dei)p: R n - t R n, is a linear map. Thus, for each p and each
vERn we can write

(dei)p(V) = :~:)Wij)p(v)ej.
j

It is easily checked that the expressions (Wij)p(V), above defined, depend

linearly on v. Thus (Wij)p is a linear form in R n and, since ei is a differentiable
vector field, Wij is a differential I-form. Keeping this in mind, we write the
above as
dei = LWijej. (1)
]

The n2
forms Wij so defined are called the connection forms of R n in the
moving frame {ei}.
Not all of the forms Wij are independent. If we differentiate (ei' ej) = Dij,
we obtain
0= (dei, ej) + (ei' dej) = Wij + Wji,
that is, the connection forms Wij = -Wji are antisymmetric in the indices
i,j.
The crucial point in the method of moving frames is that the forms Wi
and Wij satisfy the structure equations of Elie Cart an.

Proposition 1. (The structure equations of R n ). Let {ei} be a moving frame

in an open set U eRn. Let {Wi} be the coframe associated to {ei} and Wi]
the connection forms of U in the frame {ei}. Then

(2)

dwij = LWik 1\ Wkj, i,j, k = 1, ... ,no (3)

k

Proof. Let al = (1, ... ,0), ... ,an = (0, ... ,1) the canonical basis ofRn, and
let Xi: U - t R be the function that assigns to the point (Xl! ••• , xn) its ith_
coordinate. Then dXi is a differential I-form on U and, since dXj(aj) = Dij,
we conclude that {dXi} is the coframe associated to {ail.
Now write
(4)
]

where !3ij is a differentiable function on U and, for each p E U, the matrix

(!3ij(P» is an orthogonal matrix. Since wi(ej) = Dij,

Wi = L !3ijdxj. (5)
j
 1. The Structure Equations of R n 79

We first prove that df3ij = Ek wikf3kj. In fact,

de; = LWikek = L Wik(Lf3kjaj) = LW;kf3kjaj,

k k j jk

and since from (4), de; = Edf3;jaj, we obtain by comparison,

df3ij = L w;kf3kj. (6)

k

To obtain the first structure equation (2), we differentiate (5) and use (6):

dw; =L df3ij 1\ dXj =L wikf3kj 1\ dXj =L Wk 1\ Wki·

j jk k

For the second structure equation (3), we differentiate (6), obtaining

0= L dw ikf3kj - L Wik 1\ df3kj'

k k

that is
L dwikf3kj = L Wik 1\ L wksf3sj,
k k s

or, finally, multiplying by the inverse matrix of (f3kj)

dw ie = L Wik 1\ Wke,
k

as we wished. o
Remark 1. If we denote by x: U '-+ R n the inclusion map, to say that the
forms Wi are dual to the frame {ei} is equivalent to saying that dx = EWiei.
Intuitively, the expressions that define Wi and Wij, that is,

describe how the moving frame x, el, ... ,en varies as we move (along a curve
x(t» in U. This was how Elie Cartan introduced the method of moving
frames. The structure equations were then consequences of the "necessary"
relations:
d(dx) = 0,
For instance, the first structure equation can be obtained as follows:

0= d(dx) = L dwiei - L Wi 1\ dei

i

= Ldwjej - LWi 1\ LWijej = L(dWj - LWji I\wi)ej,

j j j i
80 5. Differential Geometry of Surfaces

hence
dw j = L Wi" Wij·
i

The second equation can be obtained similarly.

The main idea of Cart an's method to study the geometry of submanifolds
of RN can be described as follows. Let x: Mn -+ Rn+k be an immersion of a
differentiable manifold Mn into the euclidean space R n+k. It is a consequence
of the inverse function theorem that for p E M there exists a neighborhood
U C M of p such that the restriction xlU C M -+ Rn+k is an embedding.
(See Exercise 4 of Chap. 3).
Let V C Rn+k be a neighborhood of x(P) in R n+k such that V n
M = x(U). Assume that V is such that there exists a moving frame
{el' ... , en, en+l, ... , eq } in V with the property that, when restricted to
x(U), the vectors el, ... , en are tangent to x(U)j such a moving frame is said
to be an adapted frame.
In V we have, associated to the frame {ei}, the coframe forms Wi and the
connection forms Wij which satisfy the structure equations (2) and (3). The
map x: U C M -+ V C Rn+k induce forms X*(Wi), X*(Wij) in U. Since x*
commutes with exterior derivation and exterior products, such forms in U
satisfy again the structure equations (2) and (3). It turns out that the local
metric geometry of U C M is all contained in the structure equations, and
this reBects the "universal character" of R n ,
In the next section we will apply the method of moving frames to a simple
but important case, namely, surfaces in R 3 . For that, we will need a few
preliminary lemmas that we establish now.

Lemma 1. (Cartan's lemma). Let vn be a vector space of dimension n,

and let WI, ... , Wr: vn -+ R, r ~ n, be linear forms in V that are lin-
early independent. Assume that there exist forms 01 , •.• , Or: V -+ R such
that E~=1 Wi " 0i = o. Then

Oi =L aijWj, with aij = aji.

j

Proof. We complete the forms Wi into a basis Wl. •.. , Wn Wr+l, ... , Wn of V*
and we write
Oi = LaijWj + Lbi/w/, 1= r + 1, ... ,n.
j I

By using the hypothesis, we obtain

 1. The Structure Equations of R n 81
n
o= L Wi 1\ Oi = L aijWi 1\ Wj +L bilWi 1\ WI
i=l ij I

= L(aij - aji)Wi 1\ Wj +L bilWi 1\ WI·

i<j i<1

Since Wk 1\ W., k < s, k, s = 1, ... , n, are linearly independent, we conclude

that bil = 0 and aij = aji' 0

Lemma 2. Let U eRn and let Wl, ••• ,Wn be linearly independent differential
l-forms in U. Assume that there exists a set of differential l-forms {Wij},
i,j = 1, ... ,n that satisfy the conditions:

dw j = L Wk 1\ Wkj'

Then such a set is unique.

Proof. Suppose the existence of another set Wij with

dw j = LWk I\ wkj'
k

Then
LWk 1\ (Wkj - Wkj) = 0,
k

and, by Cartan's lemma,

j -Bj
B ki - ik'

Notice that

Wkj - Wkj =L BtWi = -(Wjk - Wjk) =- L Bjiwi,

i

and, since the Wi are linearly independent, Bt = -Bji' By using the above
symmetries, we obtain finally that
k . . . . k k
Bji = -B'i = -Blk = Bjk = Bkj = -Bij = -Bji = 0,
that is, Wkj = Wkj. o
82 5. Differential Geometry of Surfaces

2. Surfaces in R3

We now apply the method of moving frames to the special case of surfaces
in R3. Let x: M2 -+ R3 be an immersion of a two-dimensional differentiable
manifold in R3. For each point p E M2, an inner product { , }p is defined in
TpM by the rule:

{VllV2}p = (dXp(Vl),dxp(V2)}x(p),
where the inner product in the right hand side is the canonical inner product
of R3. It is straightforward to check that ( , }p is differentiable and defines a
Riemannian metric in M2 to be called the metric induced by the immersion x.
We will study the local geometry of M2 around a point p E M2. Let
U c M be a neighborhood of p such that the restriction xlU is an embedding.
Let V C R3 be a neighborhood of pis R3 such that V n x(M) = x(U), and
that it is possible to choose in V an adapted moving frame ell e2, e3i this
means that, when restricted to x(U), el and e2 are tangent to x(U) (hence
e3 is normal to x(U».
In V we have, associated to the frame {ei}, the coframe forms Wi and
the connection forms Wij = -Wji, i,j = 1,2,3, which satisfy the structure
equations:
dwl = w21\ W2l + w31\ W3l,

dw2 = Wl 1\ Wl2 + W3 1\ W32,

dw3 = Wl 1\ Wl3 + W2 1\ W23,
dwl2 = W13 1\ W32,
dwl3 = Wl2 1\ W23,
dw23 = W2l 1\ W13.
The immersion x: U C M -+ V C R3 induces forms X*(Wi), X*(Wij) in
U. Since x* commutes with d and 1\, such forms still satisfy the structure
equations. Notice that X*(W3) = 0, since for all q E U and all v E TqM,
X*(W3)(V) = w3(dx(v» = w3(alel + a2e2) = 0,
where v = alel + a2e2.
With a slight abuse of notation, we will write

This amounts to look upon U as a subset of R3 by the inclusion x: U -+ R3

(notice that xlU is an embedding), and to look upon the forms Wi and Wij as
restricted to U. These restricted forms satisfy the above structure equations
with the additional relation W3 = 0.
Since W3 = 0,
 2. Surfaces in R3 83

hence, by Cartan's lemma,

W13 = hUWl + h12W2,

W23 = h21Wl + h22W2,
where h ij = h ji are differentiable functions in U.
We want to obtain a geometric interpretation of the functions h ij • For
that, observe that the map ea: U -+ R a takes its values in the unit sphere
8 2 C R a , since leal = 1. By fixing orientations of U and R a , we can choose
the frame {ed in such a way that, for each q E U, {ell e2} is in the orientation
of U and {el' e2, ea} is in the orientation of R a. In this case, ea: U -+ 8 2 C R 3
is well defined, does not depend on the choice of the frame, and it is called
the Gauss (normal) map in U (Fig. 5.1).

The Gauss map

Fig. 5.1

We notice the important fact that if M is oriented, the Gauss map can
be defined globally on M.
Now, since dea = Walel + W32e2, we obtain, for all q E U and all v =
aIel + a2e2 E TQm,

de3(v) = _ (hll
h21 h12)
h22
(al) ,
a2

that is, (-h ij ) is the matrix of the differential of the Gauss map ea: U -+ 8 2
in the basis {e 11 e2} which is the geometric interpretation we were looking
for.
Since the matrix (hij) is symmetric, we conclude immediately that the
differential dea: T M -+ T 8 2 of the Gauss map e3: U -+ 8 2 is a selfadjoint
linear map. From a well known result in Linear Algebra, we know that such a
84 5. Differential Geometry of Surfaces

linear map can be diagonalized with real eigenvalues - AI, - A2 and orthogonal
eigenvectors.
It is usual to define the Gaussian curvature K of M in p by

K = det(de3)p = AIA2 = hllh22 - h~2

and the mean curvature H of M at p by

H -- - 2"1 (t race d) _ Al + A2 _ hll + h22

e3 p - 2 - 2 '
where the functions involved are computed at p. Clearly K and H do not
depend on the choice of the moving frame. Notice that H changes sign with
a change of orientation but K remains the same under such a change. The
expressions of K and H in terms of a moving frame are immediately obtained:

dwl2 = Wl3 /\ W32 = -(hllh22 - h~2)WI /\ W2 = -KWI /\ W2,

Wl3 /\ W2 + WI /\ W23 = (h ll + h2 2)WI/\ W2 = 2Hwi /\ W2.
The expression dw l2 = - K WI /\ W2 ,liows us to prove one of the most
important theorems in the theory of c,urfal' " I R3.
r .

Theorem 1. (Gauss). K only depends on the induced metric of M2; that is,
if x, x': M2 -+ R3 are two immersions with the same induced metrics, then
K(P) = K'(P), p E M, where K and K' are the Gaussian curvatures of the
immersions x and x', respectively.

Proof. Let U C M be a neighborhood of p and consider a moving frame

{el,e2} in U, orthonormal in the induced metric. The set {dx(ed,dx(e2)}
can be extended into an adapted frame in V :::> x(U) and, similarly, the set
{tlx'(el), dx'(e2)} can be extended into an adapted frame in V' :::> x'(U).
Let us denote by a prime the entities that refer to the immersion x'. Then,
WI = w~, W2 = w~, by duality. By the uniqueness of Lemma 2, Wl2 = wb. It
follows that

hence K = K'. o
Gauss theorem means that the Gaussian curvature, the definition of which
made use of the ambient space R 3, only depends on measurements made
on the surface. This led Gauss, around 1827, to imagine the existence of
geometries that were independent of the ambient space. Because he lacked
adequate tools (in particular, the notion of a differentiable manifold), Gauss
did not develop these ideas which were later (1852) taken up by Riemann.
In general, geometric entities on M that can be computed from WI, W2 and
Wl2 depend only on the induced metric in the sense above described, and we
 2. Surfaces in R3 85

ought to be able to define them with no mention to the immersion x. We will

come back to that in the next section.

Example 1. Consider the immersion x: U C R2 -+ R 3 , where U is

U = {(s,v) E R2j -00 < s < 00,0 < v < 211"},
and x is given by
x(x, v) = (h(s}sinv,h(s}cosv,g(s)).
Here h(s} "" 0 and 9(S} are differentiable functions that satisfy

The image x(U} is a rotation surface with axis Oz whose generating curve
y = h(s}, z = 9(S} is parametrized by the arc length s (Fig. 5.2).

z = h(s}
= g(s}

Fig. 5.2

We want to show that the Gaussian curvature of this surface is K =

-(h" Ih), where prime denotes derivative relative to s.
Observe that vlh measures the length of the parallel x(const., v} . Thus,

e2 = dx (!~)
hGV
86 5. Differential Geometry of Surfaces

are orthonormal vectors tangent to x(U). Together with a unit vector e3

normal to x(U), they constitute a frame adapted to the immersion x.
It is immediate to check that Wl = ds, W2 = hdv. On the other hand, if
we set Wl2 = ads + bdv, we obtain
bds /\ dv = ds /\ Wl2 = Wl/\ wl2 = dw2 = dh /\ dv = h'ds /\ dv
and
ahds /\ dv = Wl2 /\ hdv = Wl2 /\ W2 = dwl = O.
It follows that Wl2 = h'dv, and
h"
dwl2 = h"ds /\ dv = TWl /\ W2 = - K Wl /\ W2,

and this yields the required expression. In particular, for the sphere of radius
R, we have h(s) = Rcos(s/R), hence K = I/R2.
As we mentioned before, to a given immersion x: M2 --+ R 2 we associate
two quadratic forms at each TpM, p EM, which are defined as follows.
The first quadratic form Ip is merely t h" :J.uadratic form associated to the
bilinear form ( , }p, that is,

In an adapted frame {ei}, i = 1,2,3, the expression of the quadratic form

I is given by
(7)
where WlWl, for instance, is the symmetric product (not the exterior product)
of Wl with Wl, that is, WlWl(V) = Wl(V) . Wl(V). To check (7), we write v =
alel + a2e2. Then

Thus, the first quadratic form is given by

I = wi +w~.
where, as usually, we dropped the indication of p.
The second quadratic form is defined in an adapted moving frame by

IIp(v) = (Wl3Wl + W23W2) (v) = L hijWiWj, i,j = 1,2,

ij

where again we are considering symmetric products of differential forms. In

this case, we have to prove that II does not depend on the choice of frames.
This is actually so, since this is the quadratic form associated to minus the
differential of the Gauss map, that is,
 2. Surfaces in R a 87

It is convenient to describe still another interpretation of I I that can be

obtained as follows. Let a: (-€, €) - t M be a curve in M parametrized by the
arclength s, with a(O) = p, a'(O) = v E TpM. Then, by writing xoa(s) = x(s)
and ea 0 a(s) = ea(s), we obtain
dx
(ds' ea(s») = 0,
hence
d2x
(ds2 ,ea(s))
I8=0= -(ds'
dx dea I
Ts) 8=0= -(dx(v), de 3(V»)p
= (Wl el + W2e2,w31el + W32e2) (V)
= (WIWI3 + W2W23)(V) = IIp(v).
On the other hand, denoting by k(s) the curvature of the curve a(s) and by
n(s) the principal normal of a(s), we obtain that
d2 x
(dS"(0), e3(0») = k(O)(n(O), e3(0»).

The expression k(n,ea)(p) is called the normal curvature kn(v) of the surface
in the direction v = a'(O) at the point p. Since IIp(v) = kn(v), we have that
kn(v) is the same for all curves a(s) with the same tangent vector v at p.
Thus, collecting the two interpretations, we conclude that
IIp(v) = -(dea(v), v)p = kn(v).
It is known from Linear Algebra that the maximum and minimum of IIp(v),
as v runs the unit circle 8 1 C Tp8, are the eigenvalues -AI, -A2 of (-de3) and
the corresponding vectors generate the eigenspaces of (-dea). The extremal
normal curvatures (-Ad = kb (-A2) = k2 are called the principal curvatures
at p and the corresponding directions are called the principal directions at p.
The importance of the first and second fundamental forms is that they
determine the local geometry of surfaces in R 3 .
In the same vein, we can work out the whole local geometry of surfaces in
R3 (See, for instance, [de] Chapter 3). We stop here, however, in the hope
that the method is sufficiently clear. It only remains to be explained what is
meant by the statement that I and I I determine entirely the local geometry
of a surface in R 3 . This is the content of the next theorem, and the corollary
following it.

Theorem 2. Let U and U' be connected submanifolds of dimension two in

R3. Assume that there exist adaptedjrames {ei} in U, tea in U', i = 1,2,3,
and a diffeomorphism f: U - t U' such that
88 5. Differential Geometry of Surfaces

Then there exists a rigid motion p: R3 -t R3 such that the restriction plU =
f·
Proof. Let p E U and f(P) E U' . Let T be the translation of R3 that takes
p to p' = f (P) and let R be the rotation of R 3 that takes ei to e~. Set
p = RoT. We will show that 9 = f 0 p-l: p(U) - t U' is the identity in U' ,
and this implies the statement of the theorem.
Since p is an isometry of R 3 , consider the orthonormal moving frame
ej = dp(ej) in p(U). We will denote with an upper index", the entities
associated to the frame {ej} in p(U). By definition, for all q E p(U) and all
v E Tq(p(U»,
(dej)q(v) = L(Wjj)q(v)(ej)q.
j

Define e~ 0 9 by (ei 0 g)(q) = ei(g(q». Then

d(e~ 0 g)q(v) = (deDg(q)(dg(v» = L(w;j)g(q)(dg(v»(ej)g(q)

j

= L(g*w~j)q(v)(ej 0 g)q = L(Wij)q(v)(ej 0 g)q,

j j

where the last equality follows from the fact that

9 • Wij
I = (f 0 P-1)* Wij
I = (-l)*f*
P WijI = (-1)*
P Wij = Wij'
-
Because q and v are arbitrary, it follows that ei - e~ 0 9 satisfies the system
of ordinary differential equations

d(ei - e~ 0 g) = L wjj(ej - ej 0 g),

j

with initial conditions at the point p(P) given by:

(ei - e~ 0 g)(p(P» = 0.
By the uniqueness theorem for ordinary differential equations, ej = ei 0 p.
In a similar way, we can show that

d(x - x' 0 g) = L wj(ej - ei 0 g) = 0,

j

where x: p(U) C R3 and x': U' C R3 are the respective inclusions and the
last equality comes from what we just proved. Since the initial conditions in
p(P) are: (x - x' 0 g)(p(P» = 0, we conclude that x = x' 0 g. Since x and x'
are inclusions, this implies that 9 is the identity, as we wish~d. 0

Corollary. Let U and U' be connected submanifolds of dimension two in

R3. Assume that there exists a diffeomorphism f: U - t U' which preserves
the first and second quadratic forms, that is,
 3. Intrinsic Geometry of Surfaces 89

Jp(V, v) = J/(p) (df(v), df(v)), IIp (v, v) = lI/(p) (df(v), df(v))

lor all p E U and all v E TpU. Then, there exists a rigid motion p: R 3 -t R3
such that plU = I.

Prool. Consider in U an adapted frame {ei} and define in U' a frame

{ea = {dl(ei)}' Since I preserves inner products, this is again an adapted
frame, and rWi =
Because the second fundamental forms are preserved,
Wi.
(h ij ) = (h~j 0 f). Thus,
rw'ta = Wl3 and rW23 = W23' Finally, by Lemma
2 (uniqueness of connection forms), we see that rWl2 = Wl2. We can now
apply the theorem to obtain the conclusion.

3. Intrinsic Geometry of Surfaces

In the study of surfaces M2 in R 3 we have seen that certain geometric entities,

for instance, the Gaussian curvature, only depend on the first fundamental
form, that is to say, on the Riemannian metric of M2. A surprising number
of geometric properties of surfaces are in the same situation as the Gaussian
curvature, i.e., they only depend on the first fundamental form, and they
constitute the intrinsic geometry of surfaces. In this section, we will present
a more systematic study of such properties by using the method of moving
frames.
Our starting point is a two-dimensional differentiable manifold M2 to-
gether with a Riemannian metric ( , ). For each point p EM, choose a neigh-
borhood U C M of p such that one can define orthonormal vector fields el
and e2 on U. From this moving frame {eb e2}, we can define a corresponding
coframe {WbW2} by the condition wi(ej) = Oij, i,j = 1,2. The question now
is whether we can define differential forms that play the role of connection
forms.
The choice we make below can be motivated by the following consider-
ations. If U could be isometrically embedded (that is, in such a way that
the Riemannian inner product ( , ) in M2 is induced by R 3 ), we would ob-
tain a moving frame el, e2, e3 in a open set V ::> U of R 3 that extends the
frame el, e2 in U. From the forms Wb W2,Wl2,Wl3, W23, and from the structure
equations
dw l = Wl2 1\ W2,
dw2 = W2l 1\ Wl,
dw l 2 = Wl3 1\ W32,
dw l3 = Wl2 1\ W23,
dw23 = W2l 1\ Wl3,
only the forms Wl,W2,Wl2 and the first two equations do not contain elements
related to the "external" vector e3. It is thus reasonable to expect that there
90 5. Differential Geometry of Surfaces

exists in U a unique form Wl2 = -W2l such that the two first equations hold.
This is indeed the case.

Lemma 3. (Theorem of Levi-Civitta). Let M2 be a Riemannian (two-

dimensional) manifold. Let U c M be an open set where a moving orthonor-
mal frame {el' e2} is defined, and let {WI, W2} be the associated coframe. Then
there exists a unique i-form Wl2 = -W21 such that

Proof. Uniqueness has already been proved in Lemma 2 of Section 1. To

prove existence, just define

Wl2(ed = dwl (el,e2),

W12(e2) = dw2(el, e2),
and check the required properties: For instance,

The problem now is to obtain geometric entities (that is, independent of

the choice of frame) from the forms WI,W2,W12. For that, it is convenient to
see how such forms change under a change of frame.
Let {el,e2} be another frame in U. If {el,e2} has the same orientation
as {eI, e2}, we obtain
el = Jel + ge2
e2 = -gel + fe2,
where f and 9 and differentiable functions in U, and J2 +g 2 = 1; on the other
hand, if the orientations of {el, e2} and {el, e2} are opposite, we obtain

el = fel + ge2
e2 = gel - Je2.

Lemma 4. If {el, e2} and {el, e2} have the same orientation, then
W12 = W12 - T,

where T = f dg - gdf. If the above orientations are opposite,

W12 = -W12 - T.

Proof. If the orientations are the same, we obtain that

(1)
 3. Intrinsic Geometry of Surfaces 91

Differentiating (1), we obtain

dwl = dJ A WI + /djih - dg A W2 - gdW2.

By using the structure equations for dWl and dW2, and the fact that W12 =
-W2l' it follows that
dwl = W12 A W2 + (fdJ + gdg) A WI + (gdJ - fdg) A W2.
Since f2 + g2 = 1, fdJ + gdg = O. Thus,

dwl = W12 A W2 - T A W2 = (W12 - T) A W2.

Similarly, by differentiating (2), we obtain
dw2 = -(WI2 - T) A WI.
By the uniqueness of the connection form, we conclude finally that

W12 = W12 - T,

and this proves the first part of the lemma. The case in which the orientations
are opposite is analogous. 0

A geometric interpretation for the 1-form T is given below and asserts

that, along a curve in U, T is the differential of the "angle function" between
el and el along the curve; actually, what. we will do is to show that it is
possible to define such a function in a way that it is differentiable.

Lemma 5. Let p E U c M be a point and let 'Y: I --+ U be a curve such that
'Y(to) = p. Let CPo = angle (el(P), el(P». Then

cp(t) = it to
(f d9 - 9 df ) dt + CPo
dt dt
is a differentiable junction such that:
coscp(t) = f, sencp(t) = g, cp(to) = CPo, dcp = 'Y· T •

Proof. We first show that

f(t) cos cp(t) + g(t)sencp(t) == 1. (3)
To see that, notice that from the definition of cp, we have that cp' = fg' - gf'.
Thus,
(f cos cp + 9 sen cp)' = f' cos cp - f sen cpcp' + g' sen cp + 9 cos cpcp'
=(g' + fgf' - f2 g/) sen cp + (f' - g2 f' + gfg') cos cp = 0,
92 5. Differential Geometry of Surfaces

where in the last equality we have used that, since f2 + g2 = 1, f f' + gg' = 0.
Therefore, f cOS!{J + 9 sen!(J = const., and since
f(to)cos!{J(to) + g(to)sen!{J(to) = (f2 + g2)(tO) = 1
we conclude (3).
It follows that
(f - COS!{J)2 + (g - sen!{J)2 = f2 + l- 2f cOS!{J - 2g sen!(J + 1 = 0,
hence
cos!{J(t) = f(t),sen!{J(t) = g(t),
and the lemma follows immediately. o
We are now in a position to develop the intrinsic geometry of surfaces.
The first observation is that from (1) and (2) it follows that in a oriented
surface the 2-form
WI /I. W2 = CA:h /I. W2 = u

does not depend on the choice of frames and is therefore globally defined in
M2. The geometric meaning of the form u is obtained as follows. If VI =
aUel + a12e2, V2 = a2lel + a22e2 are linearly independent vectors at a point
p EM, then
u(VI. V2) = det(aij) = area (VI. V2),
where (VI, V2) denotes the parallelogram generated by VI and V2. Because of
that, u is called the area element of M.
The next object of intrinsic geometry is motivated by Gauss theorem.

Proposition 2. Let M2 be a Riemannian manifold of dimension two. For

each p EM, we define a number K (P) by choosing a moving frame {eb e2}
around p and setting

Then K(P) does not depend on the choice of frames, and it is called the
Gaussian curvature of M at p.

Proof. Let {eb e2} be another moving frame around p. Assume first that
the orientations of the two moving frames are the same. Then

W12 = W12 - r.
Since r = fdg - = 0, hence dw12 = dW12. It follows that
gdf, dr
-KWI /I. W2 = dw12 = dW12 = -K(;h /I. W2 = -KWI /I. W2
hence K = K, as we wished.
If the orientations an opposite, we obtain
 3. Intrinsic Geometry of Surfaces 93

and the same conclusion holds. 0

Another entity that does not depend on the choice of frames is the (co-
variant) derivative of vectors.

Definition 1. Let M2 be a Riemannian manifold and let Y be a differentiable

vector field on M. Let p E M, x E TpM, and consider a curve a: (-c, c) - t M
with a(O) = p, a'(O) = x. To define the covariant derivative (V xY)(P) of Y
relative to x in p, we choose a moving frame {ei} around p, express Y(a(t»
in this frame
Y(a(t» = LYi(t)ej, i = 1,2,
and set

(V.Y)(P) = ~ ( ~; (0) + ~ W;;(X)y,(O») e;, i, j = 1,2,

where the convention is made that Wi; = O.

Lemma 6. The covariant derivative does not depend on the choice of frames.

Proof. Let {el' e2} and {e1' e2} be two orthonormal frames around p. Assume
that they have the same orientation. Then

{
Y1 = f'iiI - g'ih
(5)
Y2 = g'iiI + f'jh
where Y(a(t» = E Yi(t)ej = E y;(t)ei, and f, 9 are differentiable functions
with P + 9 2 = 1. By definition,
VxY = (d~l +W21(X)Y2) el + (d~2 +W12(X)Y1) e2
where the functions are taken at t = O. By using (5), and the facts that
W12 = W12 - T and f f' + gg' = 0, we arrive after a long but straightforward
computation that

V xY = (~1 + W21(X)Y2) e1 + (~2 + W12(X)Y1) e2

which proves the lemma in this case.
When the orientations of the frames are opposite, the proof is similar. 0

The notion of covariant derivative can be used to give a geometric inter-

pretation of the connection form W12 associated to a moving frame {el. e2}.
In fact, since e1 = 1· e1 + Oe2, we obtain V xe1 = w12(x)e2, hence
94 5. Differential Geometry of Surfaces

Thus the form W12 applied to a vector x is the e2-component of the covariant
derivative V xel'

Remark. The covariant derivative was introduced by Levi-Civitta in 1916.

For the induced metric of surfaces M2 C R 3 , it can be shown (See Exercise
7) that the covariant derivative V xY is just the projection onto the tangent
plane of M of the usual derivative in R 3 of Y along a curve tangent to
x. Thus, on a certain sense, V xY is the derivative of Y as "seen from the
surface".

Starting from the covariant derivative, we can develop all concepts of

the Riemannian geometry in dimension two (parallelism, geodesics, geodesic
curvature, etc; see, for instance, [dC] Chapter 4). In what follows, we will
present a short exposition of these ideas. M2 will always be a two-dimensional
Riemannian manifold.

Definition 2. A vector field Y along a curve 0: 1-. M2 is said to be parallel

along 0 if Va'(t)Y = 0, for all tEl.

Definition 3. A curve 0: I -. M2 is a geodesic if o'(t) is a parallel field

along o.

Definition 4. Assume that M2 is oriented, and let 0: I -. M be a differ-

entiable curve parametrized by the arc length s with o'(s) "I 0, s E I. In
a neighborhood of a point o(s) E M, consider a moving frame {el,e2} in
the orientation of M such that, restricted to 0, el(s) = o'(s). The geodesic
curvature kg of 0 in M is defined by

kg = (0*W12)(:s)'

where til is the canonical basis of R.

Proposition 3. Let M2 and {el' e2} be as in Definition .4 (here we
0: I -+
don't need to assume that M2 is orientable, so that there are two possible
choices for e2). Then el is parallel along 0 if and only if 0*W12 = 0.

Proof. el is parallel along 0 if and only if Vel el = O. Since (Vel el, el) = 0,
the last statement is equivalent to

0= (Ve1el,e2) = W12(et},
or to 0*W12 = 0, as we wished. o
 3. Intrinsic Geometry of Surfaces 95

Corollary. A differentiable curve a: I - t M2 is a geodesic if and only if its

geodesic curvature vanishes everywhere.

A geometric interpretation of the geodesic curvature is given below.

Proposition 4. Let M2 be oriented and let a: I - t M be a differentiable

curve parametrized by the arc length s with a'(s) ::I 0, s E I. Let V be a
parallel vector field along a and let cp = ang(V, a' (s)), where the angle is
measured in the given orientation. Then

Proof. Choose two frames {el,e2} and {el,e2} around a(s) as follows: el =
VI IVI and e2 is normal to el in the positive direction; el = a'(s), and e2
is normal to el in the positive direction. As usual, they are first defined
along a small interval of the curve a about a(s), and then extended to a
neighborhood of a(s) in M. Denote by Wl2 and Wl2 the connection forms
associated to {el, e2} and {el, e2}, respectively.
Now cp is the angle from el to el; cp is only defined up to a constant, but
dcp is well defined, and
dcp = a*wl2 - a*wl2'
Since el is a parallel field along a, a·w12 = O. Also, since el = a'(s), we have
that

as we wished. o
The proof of the above proposition also contains the following interpreta-
tion of the Gaussian curvature in terms of parallel transport. Let p E M2, and
D c M be a neighborhood of p homeomorphic to a disk with smooth bound-
ary aD. Let q E aD and Vo E TqM, lVol = 1, and transport V parallelly
around the closed curve aD. When V returns to q, it makes an angle cp with
the initial position Vo. Parametrizing aD as a(s), where s is the arc length
of aD, and using the frames {el(s) = a'(s),e2(s)}, {el(S) = V(S),e2(S)} as
in the above proof, we obtain

- [ a*(w12) = [ dcp = cp.

laD laD
On the other hand, by Stokes theorem,

cp= - [ a*(w12) =- [ dw l 2 = [ Ku.

laD lD lD
It follows, by the mean value theorem of integral calculus, that
96 5. Differential Geometry of Surfaces

K(P) = lim
D-tp
~D'
area
that is, the Gaussian curvature at p measures how different from the identity
is parallel transport along small circles about p.

EXERCISES

1) (The flat torus). Let f: R2 -+ R4 be given by

f(x,y) = (cosx,sinx,cosy,siny), (x,y) E R2.
Prove that:
a) f is an immersion and f(R2) is homeomorphic to a torus,
b) The frame el = ~, e2 = U
in f(R2) C R4 is orthonormal in the
metric of f(R2) induced by R4. Compute Wl, W2, Wl2,
c) The Gaussian curvature of the induced metric is identically zero.
2) (The hyperbolic plane). Let H2 be the upper half-plane, that is,
H2 = {(x,y) E R2;y > O}
Consider in H2 the following inner product: If (x, y) E H2 and u, v E
TpH2, then
u·v
(u,v)p = -2'
Y
where U· v is the canonical inner product of R2. Prove that this is a
Riemannian metric in H2 whose Gaussian curvature is K == -1; with this
Riemannian metric H2 is called the hyperbolic plane.
Hint: Choose the orthonormal frame el = 7'
e2 = ~,where {al,a2} is
the canonical frame of R2.
3) Let M2 be a Riemannian manifold of dimension two. Let f: U C R2 -+ M
be a parametrization of M2 such that fu = df(f..) and fv = df(-/;) ,
(u, v) E U, are orthogonal. Set E = (JuJu) and G = (Jv, fv). Choose an
orthonormal frame el = fu/VE, e2 = fv/,jG in U. Show that:
a) The associated coframe is given by

Wl = vEdu, W2 = VGdv.
b) The connection form is given by

_ (VE)vd + (v'G)ud
Wl2 - - ,jG u VE v.

Hint: Use the fact that wl2(ei) = dwi(el, e2), i = 1,2.

 3. Intrinsic Geometry of Surfaces 97

c) The Gaussian curvature of M2 is

K= _.J_
1 {(..;E)V) + (.,fG)u) }.
EG.,fG v
..;E u
4) Let 8 2 = {(x, y, z) E R 3 j x 2 + y2 + Z2 = I}. Prove that there exists no
differentiable nonzero vector field X on 8 2 •
Hint: Assume the existence of such a field X. Let el = XI IXI e consider
the orthonormal oriented frame {el.e2}. Then dwl 2 = - K WI 1\ W2 = -a,
hence
area 8 2 = [ a = - [ dwl2 = - [ W12 = 0,
JS2 JS2 Jas 2 •
which is a contradiction.
5) Consider R2 with the following inner product: If p = (x,y)
U,v E 1'pR2, then
u·v
(u,v)p = (g(p»2'
where u· v is the canonical inner product of R2 and g: R2 --+ R is a
differentiable positive function. Prove that the Gaussian curvature of this
metric is
K = g(gxx + gyy) - (g; + g~).
6) Let M2 C R 3 be a surface with the induced metric. Let p E M2, x E 1'pM2
and Y be a vector field tangent to M2 . Show that

(V xY)(P) = projection onto 1'pM of (dY~(S») (0),

where a: I --+ M is a differentiable curve, S E I, and ~~ is the usual
derivative of vectors in R 3 . Conclude that a curve 'Y( s) in M, parametrized
by the arc length s, is a geodesic in M if and only if the "acceleration"
vector ~ in R3 is everywhere perpendicular to M.
7) Let 8 2 = {(x,y,z) E R3;X 2 + y2 + Z2 = I} be the unit sphere with the
metric induced from R3. Show that:
a) The geodesics of 8 2 are its great circles,
b) The antipodal map A: 8 2 --+ 8 2 given by A(x, y, z) = (-x, -y, -z) is
an isometry,
c) The projective plane P2(R) (cf. Example 7 of Chapter 2) can be given a
Riemannian metric such that the canonical projection 71': 8 2 --+ p 2 (R}
is a local isometry (that is, each p E 8 2 has a neighborhood V such
that the restriction 71'1 V is an isometry).
8) Let M2 be a Riemannian manifold (of dimension two). The goal of the
exercise is to show that the Gaussian curvature K of M is identically zero
if and only if M is locally euclidean, that is, there exist local coordinates
98 5. Differential Geometry of Surfaces

(u, v) around any point such that the first fundamental form I = du 2 +dv 2 •
Clearly if I is as above, K = O. To prove the converse, proceed as follows:
a) Choose a frame {ell e2} around p EM. Since dw I2 = - K WI /I. W2 = 0,
by Poincare's Lemma, there exists a function 0 defined in a neighbor-
hood V of such that dO = W12,
b) Choose another frame {ell e2} by setting ang (ell ex) = O. Show that
the connection form W12 of this frame vanishes identically,
c) Show that WI2 = 0 implies that dW I = dW2 = 0 and use again
Poincare's Lemma to obtain the required local coordinates.
6. The Theorem of Gauss-Bonnet
and the Theorem of Morse

1. The Theorem of Gauss-Bonnet

The considerations of the last chapter were strictly local. However, one of the
most interesting features of differential geometry is the connection between
local properties and properties that depend on the entire surface. One of the
most striking of such properties is the so-called Gauss-Bonnet theorem which
we intend to prove in this section.
In his fundamental work (Considerations on curved surfaces, 1827), Gauss
proved the special case of this theorem for geodesic triangles and foresaw its
importance for the development of differential geometry. The theorem for
more general regions is due to O. Bonnet (Jour. Ecole Polytech. 19 (1848),
1-146). With the advent of Topology, it became soon clear that a global for-
mulation of the Gauss-Bonnet theorem would be an important link between
Geometry and Topology. The extension of this result to higher dimensions
became then an important mathematical problem. After some preliminary
work by Allendoerfer and Weil, a satisfactory solution was obtained in 1944
by 8.8. Chern, as an application of the method of moving frames. We will
come back to that in Remark 2 of this section.
Before starting, we want to make the general remark that any differen-
tiable manifold Mn (Hausdorff and with countable basis) can be given a
Riemannian metric. The proof depends on the existence of a partition of
unity. For the compact case (which is the only one we will use), it suffices
to define arbitrarily an inner product (,}O on each coordinate neighborhood
r(UO) of a finite differentiable structure of Mn, and to set

where <Po is a differentiable partition of unit subordinate to the (finite) cov-

ering !o(Uo ).
From now on, M will denote a compact, oriented, differentiable manifold
of dimension two. Let X be a differentiable vector field on M. A point P EM
is a singular point of X if X (P) = OJ the singular point p is isolated if there
exists a neighborhood V C M of p which contains no singular point other
than p. In what follows, it will be convenient to choose V homeomorphic to
100 6. The Theorem of Gauss-Bonnet and the Theorem of Morse

an open disk in the plane. Notice that the number of isolated singular points
is finite, since M is compact.
To each isolated singular point of X, we are going to associate an integer
to be called the index of X at p, as follows. First, choose a Riemannian
metric on M, and consider the moving frame {el,e2}, where el = X/IXI and
e2 is a unit vector field orthogonal to el and in the orientation of M. This
determines differential forms Wl,W2, W12 in V - {pl. Next, we choose another
moving frame {el' e2}, in the same orientation as before, defined throughout
V, thus obtaining forms WI, W2, W12 in V. The difference

W12 - W12 = T

is defined in V - {p}.
Now consider a simple closed curve C that bounds a compact region of V
containing p in its interior; C will be oriented as the boundary of this region.
By Lemma 5 of Chapter 5, the restriction of T to C is the differential of the
angle tp(t} between el and el along C. Thus

11 T = dtp = 27rl.

The integer I is called the index of X at p.

Notice that in the definition of index we made various choices, namely,
the choice of a Riemannian metric, the choice of a frame {el' e2}, and the
choice of a curve C. It is clearly necessary to show that the index I does not
depend on these choices.
Before that, let us look at some examples of singularities of vector fields in
the plane. Intuitively, the index is the number of "turns" given by the vector
field as we go along a simple closed curve around an isolated singularity. The
figure below (Fig. 6.1) displays some vector fields in the plane (described by
their trajectories) with isolated singularities, and their respective indices.

Lemma 1. The definition of I does not depend on the curve C.

Proof. Let Ct and C2 be two simple closed curves around p, as in the def-
inition of index. Assume first that Ct and C2 do not intersect and consider
the annular region b. bounded by C1 and C2. Let 11 be the index computed
with C 1 and 12 be the index computed with C2. By Stokes theorem, and the
fact that dT = 0,

11 - 12 = 2. [ T - 2. [ T = 2. [ dT = 0,
211' lCI 211' l C2 211' 1D.
and this proves the Lemma in this case. If C1 and C2 intersect, we choose a
curve Ca that does not intersect both C 1 and C2 • By applying the above, we
conclude that It = Ia =1 2 • 0
 1. The Theorem of Gauss-Bonnet 101

~I~
• 0 ill(

~I\' 1=1

1=2

Fig. 6.1

Lemma 2. The definition of I does not depend on the choice of the frame
{el,e2}' More precisely, let Sr = 8B r be the boundary of a disk of mdius
T and center p, and consideT the frame {el' e2} of the definition. Then, the
limit
1
lim -2 i:ih2 = I
r->O 71'
1 - Sr

exists, and j = I.

Proof. Let Sri' Sr2 be concentric circles, T2 < T1, and let 6. be the annular
region bounded by Sri and Sr2' By Stokes theorem,

1 -1Sri
W12
Sr2
W12 =
J6
f ciW12 ~ 0, as T},T2 ~ 0. (1)

Notice that W12 is not defined in B r2 ; however, ciW12 = - K (j is certainly

defined everywhere. It follows that any sequence

with {Tn} ~ 0, is a Cauchy sequence, hence converges. Thus the limit

102 6. The Theorem of Gauss-Bonnet and the Theorem of Morse

lim -1
r-+O 271'
1 -
Sr
W12 = I

exists, and we want to show that 1 = I.

In (1), fix rl and let r2 go to zero. Then

1 Sri
W12 - 271'1 = 1Bq
diJ 12 = -1Brl
KWII\ W2. (2)

On the other hand, since W12 = Wl2 + T, we have

(3)

By (2) and (3), we conclude that I = 1, as we wished. o

Lemma 3. The index does not depend on the metric.

Proof. Let (,)0 and (, h be two Riemannian metrics on M. Let, for t E [0,11,

( ,)t = t(, h + (1 - t)( , )0.

Then (,)t is easily seen be a positive definite inner product on M which
varies differentiably with p. Thus (,)t is a one-parameter family of metrics
on M that starts with (,)0 and ends with (, h. Let 10, It and It be the
corresponding indices. It is easily checked, by using Lemmas 1 and 2, that
It is a continuous function of t. Being an integer, It = const., t E [0,11.
Thus 10 = II, as we wished. 0

We can now state and prove the Gauss-Bonnet Theorem in the following
form.

Theorem 1. Let M2 be a two-dimensional compact oriented differentiable

manifold. Let X be a differentiable vector field on M with isolated singulari-
ties PI, ... , Pk whose indices are II, ... , I k. Then, for any Riemannian metric
onM,

1M
K(7 = 271'Lh
k

i=l

where K is the Gaussian curvature of the metric and (7 is its element of area.

Proof. Consider in M2 -l!{Pi}

, the frame {il = X/lXI, e2}, where e2 is a
unit vector field orthogonal to i l in the orientation of M. Let us denote by
 1. The Theorem of Gauss-Bonnet 103

Bi a ball with center Pi which is such that it contains no singular point other
than Pi. From Stokes theorem, we have that

[
1M-UBi
Klih /\ W2 = - [
JM-UBi
dW12 = [
JU(lJB;)
W12 = L[
i JlJB;
W12,

• •
where OBi has the orientation induced by Bi (this is the opposite of the
orientation of M - B i , hence the change of sign in the second equality). Now,
take the limit of the above, as the radii of Bi go to zero, and use Lemma 2
to obtain that

as we wished. o
Notice that the right hand side does not depend on the vector field X
and the left hand side does not depend on the metric. Thus we obtain the
striking conclusion that L: Ii is the same for all vector fields with isolated
singularities, and JM K (J is the same for all Riemannian metrics on M.
k
The number L Ii is called the Euler-Poincare characteristic of M and is
i=l
also denoted by X(M). By the above, X(M) is invariant by diffeomorphisms
and 2~ JM K (J does not depend on the metric on M and it is equal to this
invariant.

Remark 1. Another way of introducing X(M), for a compact surface M, is to

decompose M into a finite number of curvilinear triangles in such a way that
the intersection of two such triangles be either empty, a common edge, or a
common vertex of the triangles (such a decomposition is called a triangulation
of M and it is possible to prove that it always exists.)

Let us denote by F the total number of triangles, V the total number of

vertices and A the total number of edges of such a triangulation. By definition
X(M) =V - A + F.
We will show that, when M is orientable, this definition agrees with the
previous one. For that, choose a triangulation, and consider the vector field
in M indicated by its trajectories in Fig. 6.2 below (actually we only display
the field in two triangles ofthe triangulation). The indices ofthe vector fields
at the points B,G and D are, respectively, 1,1 and -1. Thus the total sum
of the indices is

Since L:i Ii = X(M) does not depend on the chosen field, we conclude that
V - A +F = X(S).
104 6. The Theorem of Gauss-Bonnet and the Theorem of Morse

Fig. 6.2

Example 1. Let us compute X(M2) for the sphere 8 2 and the torus T.
For the sphere, we choose the metric induced by R3 on 8 2 = {p E
R 3 j Ipi = I} for which K == 1 (cf. Example 1 of Chap. 4). Thus Is Ku =
area 8 2 = 411", hence X(8 2 ) = 2. As a consequence, we see that every tangent
vector field on 8 2 has at least one singularity.
For the torus T, we know that it is possible to introduce in T a metric
with K == 0 (cf. Exercise 1 in Chap. 4). Since X(T) does not depend on the
metric, X(T) = O.

Remark 2. The proof of the Gauss-Bonnet Theorem presented here is essen-

tially due to S.S. Chern. The crucial point of the proof is the existence of a
form W12 in M - U{Pi} that depends on the vector field X but whose differ-
ential ciWl2 = -Ku does not depend on X and is globally defined on M. The
idea that this was the crucial part of the proof lead Chern to give a proof
of a generalized Gauss-Bonnet theorem [CHERJ. Furthermore, such an idea
opened the way for other Gauss-Bonnet type theorems and is in the basis of
the creation of the so-called Chern classes.

Remark 3. It can be shown that a surface M (connected, compact and ori-

entable) is homeomorphic to M' if and only if X(M) = X(M'). A proof can
be found in W. Massey, [MAS].

We now go into the Gauss-Bonnet Theorem for surfaces with boundary.

Theorem 2. Let M be an oriented, compact, two-dimensional differentiable

manifold with boundary aM, and let X be a differentiable vector field on M
such that it is transversal to aM (that is, X is nowhere tangent to aM).
Assume that the singularities Pl, ... ,Pk of X are isolated, do not belong to
 1. The Theorem of Gauss-Bonnet 105

8M and denote their indices by II, ... , Ik. Then, for any Riemannian metric
onM,
f
JM
Ka + f
J OM
kgds = 271" Lh
where kg is the geodesic curvature of 8M and ds is the arc element of 8M.

Proof. Choose a Riemannian metric on M and consider in M the orthonor-

mal oriented frame el = X/lXI, e2. Choose, in a neighborhood V c M
of 8M, another oriented frame {el,e2} such that, restricted to 8M, el is
tangent to 8M. Then
i·W12 = i·W12 + d<p,

where i: 8M ---t M is the inclusion map and <p is the angle between el and
el along 8M.

Let Bi be a ball of center Pi, i = 1, ... ,k, so that Bi contains no singular

point other than Pi. Then

f
JM-UBi
KWI AW2 =- fJ M - UBi
dW 12 =! U8Bi
W12 - f
J OM
i·W12,

hence

But, by the definition of geodesic curvature,

f i·W12 = f i·W12 +f d<p = f kgds + f d<p.

J 8M J OM J OM J OM J 8M

Since el = X/IXI is nowhere tangent to 8M, JOM dcp = O. Therefore, by

taking the limit when the radius of Bi goes to zero, we obtain

f K a + f kgds =
JM JOM
271" Lh
as we wished. 0
The number l: Ii is again an invariant by diffeomorphisms of the (com-
pact, oriented) surface M with boundary 8M, and is again called the Euler-
Poincare characteristic X(M) of M. Let us compute some examples:

Example 2. Consider the segment of a right circular cylinder M bounded by

two parallel circles. The metric induced by R3 on M has vanishing Gaussian
curvature and the two boundary circles that constitute 8M are geodesics in
M. Thus
106 6. The Theorem of Gauss-Bonnet and the Theorem of Morse

Example 3. The Euler-Poincare characteristic of a plane disk D of radius r

is equal to one. In fact, taking the canonical metric in the plane of the disk,
we obtain that K == 0 and that the geodesic curvature of aD is kg = 1/r.

i l= l
Therefore,
x(D) = "L.Jli = -2
1
11" 8D
kgds -2 -211"r = 1.
11" r

Remark 4. The Gauss-Bonnet Theorem still holds for compact surfaces with
boundaries and corners, that is, those surfaces for which the boundary fails
to be regular at finitely many points which are called comers. Each corner
qj, j = 1, ... , n, gives rise to an external angle aj (which is the positive
angle made by the tangents at the corner) and this must be added to the
total geodesic curvature, so that the theorem now reads:

The proof involves a certain number of technicalities and we will not enter
into that here.

2. The Theorem of Morse

Closely related with the Gauss-Bonnet theorem is a relation, due to Marston

Morse, between critical points of a certain class of functions on a compact
surface M2 and the topology of M2. In this section we intend to establish
this relation.
The class of functions that we have in mind can be described as follows.
As always, M2 will denote an oriented compact differentiable manifold of
dimension two.
Let I: M2 ~ R be a differentiable function on M2. The point p E M is
a critical point of I if dIp = O. If we choose any metric on M2 and define a
vector field grad I by
(grad/(P),v) = dlp(v), for any v E TpM2,
we obtain that p is a critical point of I if and only if, for any metric on M2,
p is a singular point of grad f.
A critical point p of I: M2 ~ R is said to be nondegenerate if for some
parametrization g: U C R2 ~ M2 around p = g(O, 0), we have that det(A) =/;
0, where A is the matrix

( ~ 8 2 (fog) )
"(f;lfjJ
A= (0,0),
8~(fog) 82U~g)
Ty8x
 2. The Theorem of Morse 107

°
Since at a critical point the first derivatives vanish, it is easily checked that
the fact that det A :I does not depend on the parametrization g.
Nondegenerate critical points are the simplest type of critical points. The
behavior of the function fog = h in a neighborhood of such a point can be
easily described using Taylor's expansion:

d = h(x, y) - h(O, 0) = '12 {({Ph)

ox2 0
( f ) xy + (0oy2f) y2 }
()2
x 2 + 2 oxoy 0
2
0

+ terms of higher order .

In fact, if x 2 + y2 is small enough, and det A :I 0, the sign of d is controled
by the sign of the quadratic form in the right hand side. Thus if det A > 0,
we have one of the following alternatives:
°
a) d < in some neighborhood of pj p is then called a point of maximum

°
for f.
b) d> in some neighborhood of pj p is then a point of minimum for f.
On the other hand, if det A < 0, there exist exactly two distinct directions for
which the quadratic form vanishesj for all other directions, either d is positive
(a direction of minimum) or negative (a direction of maximum). Such a point
is called a saddle point.
Now we can state the relation we want to prove

Theorem 3. (Morse) Let f: M2 -+ R be a differentiable function on a com-

pact oriented surface M2 such that all its critical points are nondegenerate.
Let us denote by M, m, and s the number of points of maximum, minimum
and saddle, respectively, of f. Then M - s + m does not depend on f; more
precisely,

Before going into the proof, we need some preliminary considerations.

From now on, assume that we have chosen a Riemannian metric on M2. As
we have seen, p is a critical point of f if and only if the vector field grad f
has a singularity at p. How does the fact that p is a nondegenerate critical
point reflects on the field grad f?
To answer that question, we will need a definition.
Let X be a differentiable vector field on M2, p be a singular point of X
and 9 = U C R2 -+ M a parametrization around p = g(O"O). In the basis
{Ix, Iv}, (x, y) E U, associated to the parametrization g, we can write X
as
o 0
X = a(x, y) ox + ,8(x, y) oy'

where a and ,8 are differentiable functions in U, and, since p is a singular

point, a(O, 0) = ,8(0,0) = 0. Let Ag be the matrix of the linear part of X,
that is,
108 6. The Theorem of Gauss-Bonnet and the Theorem of Morse

We say that p is a simple singularity of X if det Ag # 0. This definition does

not depend on the parametrization g, since for another parametrization g,
we easily obtain
Ag = dh 0 Ag 0 (dh)-l,
where h is the change of coordinates.
We can now answer our question.

Proposition 1. Let p E M2 be a critical point of a differentiable function

f: M2 -+ R on a Riemannian manifold M2. Then p is a nondegenemte
critical point of f if and only if p is a simple singularity of grad f.

Proof. Let g: U C R2 -+ M2 be a parametrization around p = g(O, 0) such

that (/x, /y) = 0, for all (x,y) E Uj such a parametrization always exist
(cf.[dC], §3.4). Let us compute grad f in this parametrization.

Let grad f = a Ix + /3 /y, and let Y = Yl tx + Y2 t y be an arbitrary

vector of Tg(x'II)M2. By definition,
(grad f, Y) = df(Y),
hence
a a a a ) au 0 g) au 0 g)
aYl (ax' ax) + /3Y2 ( ay' ay = ax Yl + ay Y2,

for all pairs (Yl, Y2). Thus setting (Ix, Ix)

= g11 and ( /y , /y) = g22, we
obtain
grad f = au 0 g) .2... ~ + au 0 g) .2... ~.
ax g11 ax ay g22 ay
Observe now that it is possible to choose the parametrization 9 so that,
at p, 911 (P) = 922 (P) = 1. Thus the linear part of grad f is given, in this
parametrization, by

Ag = ( ~~)
(P(fog) lj2(fog)
(0,0).
~ or-
It now suffices to observe that both conditions in the statement of the Propo-
sition are equivalent to det(Ag) # 0. 0
It turns out that simple singularities of vector fields are all isolated.

Lemma 4. Let p E M2 be a simple singularity of a differentiable vector field

X on M. Then p is an isolated singular point of X.
 2. The Theorem of Morse 109

Proof. Let g: U C R2 -+ M2 be a parametrization around p = g(O, 0), and

let
a +(3-
x = 0:- a
ax ay
be the expression of X in this parametrization.

Consider the map <p: U C R2 --+ R2 given by <p(x, y) = (o:(x, y), (3(x, y)).
Since p is a simple singularity,

det(d<po) = det Ag '" 0, at p.

By the inverse function theorem, there exists a neighborhood V C U of
(0,0) where <p is bijective. Thus if o:(x, y) = (3(x, y) = 0, (x, y) E V, then
x = y = 0, that is, in g(V) there is no singular point other than p. 0

Corollary. Nondegenerate critical points of a differentiable functions f =

M2 -+ R are isolated.

From Lemma 4 it follows that it makes sense to talk about the index of
a simple singularity. We will now show that they are easily computable.

Proposition 2. Let p E M2 be a simple singular point of a differentiable

vector field X on M2. Then the index of X at p is either + 1 (if the deter-
minant of the linear part of X is positive) or -1 (if the determinant of the
linear part of X is negative).

Proof. Since the index is local and does not depend on the choice of a
metric, we can take M2 = R 2 with the canonical metric, and assume that

°
p = (0,0) = 0. Thus X: R2 --+ R2 is a differentiable map with X(O) = 0,
and the linear part (the differential) of X at is given by

Let us define a map F: R2 x I --+ R2 by

F(p,t) = {
~
t ' if t '"°
°
(p,t) E R2 x I.
dXo(P), if t =

We claim that F is continuous. To see that, we need the following fact

from Calculus: If f: R n --+ R is differentiable and f(O, 0, ... ,0) = 0, we can
write
110 6. The Theorem of Gauss-Bonnet and the Theorem of Morse

-11
f( Xl, ... ,X n ) - df(txl, ... ,txn) dt
d

= LXi
n
o

i=l
11 0
t
of
a.(txl, ... ,txn)dt;
x,
then, by setting,

we obtain that hi is differentiable, hi(O, ... ,0) = it(O, ... , 0), and

f(xl,' .. ,xn) = L xihi(Xl, .. . ,xn).

i

It follows, by setting

X(XllX2) = (a(xllx2), .B(Xl,X2)),

that we can write:
a(xl, X2) = L xihli(Xl, X2),
i

i = 1,2,
where hij, i,j = 1,2, are differentiable functions. Since, even for t = 0,

F«Xl,X2),t) = (LXihli(txlltX2), LXih2i(txl,tX2)),

i i

we conclude that F is continuous, as we claimed.

F maps continuously the vector field dXo = F(P,O) to the vector field
X = F(p, 1). Thus the index of X at 0 is equal to the index of dXo at 0, and
it suffices to compute the latter.
For that, consider the circle C = {(Xl, X2) E R 2; xi +X~ = I}. If ql, q2 E
C and ql ;;f. q2, since dXo is non-singular, dXo(qt} ;;f. dXO(q2). If det(dXo) >
0, the index I of dXo is positive and, by what we have just seen, cannot
be greater than one; thus I = 1. If det(dXo) < 0, the index I of dXo is
negative and, by the same token, cannot be smaller than -1; Thus I = -1.
This proves the Proposition. 0

Remark. The above is just another proof of the Lemma used to prove Theo-
rem 3 in Chapter 2.

With all these preliminaries, the proof of Morse Theorem is almost im-
mediate.
 2. The Theorem of Morse 111

Prool 01 Theorem 3. Choose a Riemannian metric on M. Since the critical

points of I are nondegenerate, the singularities of grad I are isolated and
simple. Thus the index of grad I is 1, at a point where I is either maximum
or minimum, or the index of grad I is -1, at a saddle point of I. It follows
that M - 8 + m is equal to the sum of the indices of the singularities of
grad I. By the Gauss Bonnet theorem, such a sum does not depend on the
chosen metric or on the field grad I, and it is equal to X(M 2 ). 0

Remark. Theorem 2 is only a sample of the deep relations established by M.

Morse between the topology of differentiable manifolds and the critical points
of certain classes of differentiable functions. A beautiful introduction to the
subject is J. Milnor [MILN].

EXERCISES

1) Compute the Euler-Poincare characteristic of

a) an ellipsoid,
b) M = {(x,y,z) E R 3 j x 2 + y4 + z6 = I}.
2) Prove that there exists no Riemannian metric on a torus T such that K
is nonzero and does not change sign on T.
3) Let M2 be a connected compact orient able manifold of dimension two.
Prove that the following statements are equivalent (Assume that if X(M2)
-2 2 -2
= X(M ) then M is homeomorphic to M ):
a) These exists a nowhere zero differentiable vector field on M2.
b) X(M2) = o.
c) M2 is homeomorphic to a torus.
4) Let M2 C R3 be a regular surface in R3. Assume that M2 is compact,
oriented and not homeomorphic to a sphere. Show that there exist points
in M2 for which the Gaussian curvature is positive, negative and zero.
5) Let M2 be a connected, compact, oriented Riemannian manifold of dimen-
sion two such that the Gaussian curvature K is always positive. Prove that
two simple closed geodesics in M2 have a common point.
6) Let I: R2 ---t R be given by (the monkey saddle) I(x, y) = x 3 - 3xy2. Let
P = (0,0) E R2. Show that:
a) p is an isolated critical point of f.
b) p is a degenerate critical point.
c) The index of grad f at p is equal to -2.
7) Let x: M2 ---t R3 be an immersion of a two-dimensional differentiable
manifold M2 into R 3 (Le., a surface in R 3 ), and let hv: M ---t R be the
height junction, hv(p) = (x(p),v),p E M, of x relative to a fixed unit
vector v E R3(h v (P) measures the "height" of x(P) relative to a plane
through the origin and perpendicular to v.).
112 6. The Theorem of Gauss-Bonnet and the Theorem of Morse

Fig. 6.3

a) Show that p EM is a critical point of hI' if and only if TpM .lv.

b) Show that a critical point p of hI) is nondegenerate if and only the
Gaussian cUnJature K(P) of the surface at p is nonzero.
Hint. Choose a parametrization g: U C R2 -+ M2 around p so that
g(O,O) = p. The condition that p is nondezenerate is given by det A :1= 0,
where A is the matrix (fl.";!} (0») ,i,j = 1,2. Since
[J2(hog) (0) = / 02 (xog) (0), v(P») ,
OXiOXj \ OXiOXj
A is the matrix of the second fundamental of the surface at p in the
parametrization g. But then det A = ±K(P)
c) For this part assume the following version of Sard's theorem. Let
I: Mn-+ Nn be a differentiable map and let q E Nj we call q a regular
value of 1 if at all points p E 1-1(q) the differential djp is nonsingular
(notice that if q ¢ I(M), then q is a regular value of f). Sard's theorem
asserts that the set 01 regular values ollis an open and dense subset
01 N. Use Sard's theorem and part (b) to show that then exists an
open and dense set U of the unit sphere of R 3 such that if v E U, all
critical points of hI' are nondegenerate
8) The n-torus (a torus with n holes) is a compact ,orientable surface M
diffeomorphic to the surface that appears in Fig. 6.3 Consider the height
function hI' of M (see Exercise 7) and choose v so that all critical points of
hI' are nondegenerate (Exercise 7(e». Use Morse's theorem'to show that
the Euler characteristic of the n-torus is 2 - 2n.
9) Let M2 C R3 be a regular surface in R 3, let q E R 3, q ¢ M2, and let
I: M2 -+ R be the function given by Iq(P) = distance from p to q, P E
M2. Show that:
 2. The Theorem of Morse 113

a) /q is differentiable
b) P E M2 is a critical point of /q if and only the straight line pq is normal
to M2.
c) The critical point p is degenerate if and only /q(P) = =
l/ki ,i 1,2,
where kl and k2 are the principal curvatures of M2 at p relative to the
normal pq
References

[CHER] Chern, S.S., A simple intrinsic proof of the Gauss-Bonnet formula for
closed Riemannian manifolds, Annals of Math. 45(1944), 747-752.
[dC] do Carmo, M., Differential Geometry of Curves and Surfaces, Prentice-
Hall,1976.
[FIG] Figueiredo, D., A simplified proof of the divergence theorem, American
Math. Monthly, 71(1964), 619-622.
[FISC) Fischer, G., Mathematical Models, Vieweg, Wiesbaden, 1986.
[H1R] Hirsh, M., Differential Topology, Springer-Verlag, Berlin, 1976.
[KELL] Kellog, 0., Foundations of Potential Theory, Dover Publications, New
York,1954.
[LIM 1] Lima, E., Curso de An8.lise, vol. 2, Projeto Euclides, IMPA, Rio de Janeiro,
2.9. ed. 1985 (in Portuguese).
[LIM 2] Lima, E., Orientability of smooth hypersurfaces and the Jordan Brower
separation theorem, Expo Math. 5 (1987), 283-286.
[MAS] Massey, W., Algebraic Topology, an Introduction, Harcourt Brace, 1968.
[MILN] Milnor, J., Morse Theory, Princeton Press, 1963.
[PIC] Picard, E., Traite d'Analyse, Toure I, Gauthier-Villars, Paris, 3eme ed.
1922.
(WAR] Warner, F., Foundations of Diferentiable Manifolds and Lie Groups, 2nd
edition, Springer, 1986.
Index

adapted frame 80 exterior differential 8

angle function 19 exterior product 4
associated basis 41
associated coframe 77 first Green's identity 71
Axiom of countable basis 35 first quadratic form 86
Axiom of Hausdorff 35 formula of Kronecker 25
- a non-Hausdorff manifold 53
Gauss map 83
boundary 62 Gauss theorem 84
bracket of a vector field 48 Gaussian curvature 84
Brouwer fixed point theorem 75 gradient 12

Cart an's lemma 80 half-space 60

change of parametrization for a curve harmonic 72
17 Hodge star operation 11
closed differential form 18 - a geometric definition of the star
complex functions 28 operation 13
connection forms of R n 78 homotopic curves 22
contractible manifold 67
immersion 42
coordinate neighborhood 34
immersion of the projective plane in R 4
coordinate system 34
45
index of a map 24
diffeomorphism 41
isolated zero of a map 24
differentiable manifold 34
isometry 77
differentiable manifold with boundary
61 Laplacian 12
differentiable map 38 Line integrals of vector fields 29
differentiable partition of unity 56 local diffeomorphism 41
differentiable structure 34 local integrating factor 30
differential k-form 4 locally exact differential form 20
differential form on a manifold 47
differential of a map 41 maximum principle for harmonic
divergence 12 functions 73
mean curvature 84
element of angle 19
embedding 42 nonorientable 50
Euler's relation 15 normal curvature 87
exact differential form 18
exterior k-form 3 open set 35
118 Index

orientable 50 second Green's identity 71

orientable double covering 53 second quadratic form 86
orthonormal moving frame 77 simple zero of a map 24
simply-connected 23
parametrization 34
piecewise differentiable curve 17 solid angle 71
Poincare's lemma 67 structure equations of R n 78
Poincare's lemma for I-forms 14 submanifold 42
potential theory in R 3 72 support of a differential form 55
principal curvatures 87
principal directions 87 tangent bundle 41
product manifold 51
tangent space of a manifold 41
real projective plane 35
real projective space 36 vector field 47
regular surface in R 3 33 volume element 11
regular surfaces of R n 43
Riemannian manifold 77 winding number 19
rotation surface 85
rotational 13 zero of a map 24
Universitext
Aksoy, A., Khamsi, M. A.: Methods in Fixed Point Theory
Aupetit, B.: A Primer on Spectral Theory
Bachem, A., Kern, W.: Linear Programming Duality
Benedetti, R, Petronio, C.: Lectures on Hyperbolic Geometry
Berger, M.: Geometry I
Berger, M.: Geometry II
Bliedtner, J., Hansen, w.: Potential Theory
Booss, B., Bleecker, D. D.: Topology and Analysis
Carleson, L., Gamelin, T.: Complex Dynamics
Carmo, M. P. do: Differential Forms and Applications
Cecil, T. E.: Lie Sphere Geometry: With Applications of Submanifolds
Chandrasekharan, K.: Classical Fourier Transforms
Charlap, L. S.: Bieberbach Groups and Flat Manifolds
Chern, S.: Complex Manifolds without Potential Theory
Chorin, A. 1., Marsden, J. E.: Mathematical Introduction to Fluid Mechanics
Cohn, H.: A Classical Invitation to Algebraic Numbers and Class Fields
Curtis, M. L.: Abstract Linear Algebra
Curtis, M. L.: Matrix Groups
Dalen, D. van: Logic and Structure
Das, A.: The Special Theory of Relativity: A Mathematical Exposition
Dev lin, K. J.: Fundamentals of Contemporary Set Theory
DiBenedetto, E.: Degenerate Parabolic Equations
Dimca, A.: Singularities and Topology of Hypersurfaces
Edwards, R E.: A Formal Background to Higher Mathematics I a, and I b
Edwards, R E.: A Formal Background to Higher Mathematics II a, and II b
Emery, M.: Stochastic Calculus in Manifolds
Foulds, L. R: Graph Theory Applications
Frauenthal, J. C.: Mathematical Modeling in Epidemiology
Fuks, D. B., Rokhlin, V. A.: Beginner's Course in Topology
Gallot, S., Hulin, D., Lafontaine, J.: Riemannian Geometry
Gardiner, C. F.: A First Course in Group Theory
Gfu-ding, L., Tambour, T.: Algebra for Computer Science
Godbillon, c.: Dynamical Systems on Surfaces
Goldblatt, R: Orthogonality and Spacetime Geometry
Gouvea, F. Q.: p-Adic Numbers
Hahn, A. J.: Quadratic Algebras, Clifford Algebras, and Arithmetic Witt Groups
Hajek, P., Havranek, T.: Mechanizing Hypothesis Formation
Hlawka, E., SchoiBengeier, J., Taschner, R.: Geometric and Analytic Number Theory
Holmgren, R A.: A First Course in Discrete Dynamical Systems
Howe, R, Tan, E. Ch.: Non-Abelian Harmonic Analysis
Humi, M., Miller, w.: Second Course in Ordinary Differential Equations
for Scientists and Engineers
Hurwitz, A., Kritikos, N.: Lectures on Number Theory
Iversen, B.: Cohomology of Sheaves
Kelly, P., Matthews, G.: The Non-Euclidean Hyperbolic Plane
Kempf, G.: Complex Abelian Varieties and Theta Functions
Universitext
Kloeden, P. E., Platen E., Schurz, H.: Numerical Solution of SDE
Through Computer Experiments
Kostrikin, A. I.: Introduction to Algebra
Krasnoselskii, M. A., Pokrovskii, A. Y.: Systems with Hysteresis
Luecking, D. H., Rubel, L. A.: Complex Analysis. A Functional Analysis Approach
Ma, Zhi-Ming, Roeckner, M.: Dirichlet Forms
Mac Lane, S., Moerdijk, I.: Sheaves in Geometry and Logic
Marcus, D. A.: Number Fields
McCarthy, P. 1.: Introduction to Arithmetical Functions
Meyer, R. M.: Essential Mathematics for Applied Fields
Meyer-Nieberg, P.: Banach Lattices
Mines, R., Richman, E, Ruitenberg, W.: A Course in Constructive Algebra
Moise, E. E.: Introductory Problem Courses in Analysis and Topology
Montesinos,1. M.: Classical Tessellations and Three-Manifolds
Nikulin, V. V., Shafarevich, I. R.: Geometries and Groups
Oden, J. J., Reddy, 1. N.: Variational Methods in Theoretical Mechanics
0ksendal, B.: Stochastic Differential Equations
Porter,1. R., Woods, R. G.: Extensions and Ahsolut("c ,< t1ausdorff Spaces
Rees, E. G.: Notes on Geometry
Reisel, R. B.: Elementary Theory of Metric Sr,dces
Rey, W. 1. J.: Introduction to Robust and Quasi-Robust Statistical Methods
Rickart, C. E.: Natural Function Algebras
Rotman, J. J.: Galois Theory
Rybakowski, K. P.: The Homotopy Index and Partial Differential Equations
Samelson, H.: Notes on Lie Algebras
Schiff, J. L.: Normal Families of Analytic and Mesomorphic Functions
Shapiro, J. H.: Composition Operators and Classical Function Theory
Smith, K. T.: Power Series from a Computational Point of View
Smoryriski, C.: Logical Number Theory I: An Introduction
Smoryriski, c.: Self-Reference and Modal Logic
Stanisic, M. M.: The Mathematical Theory ofThrbulence
Stillwell, J.: Geometry of Surfaces
Stroock, D. W.: An Introduction to the Theory of Large Deviations
Sunada, T.: The Fundamental Group and the Laplacian (to appear)
Sunder, V. S.: An Invitation to von Neumann Algebras
Tamme, G.: Introduction to Etale Cohomology
Tondeur, P.: Foliations on Riemannian Manifolds
Verhulst, E: Nonlinear Differential Equations and Dynamical Systems
Zaanen, A. C.: Continuity, Integration and Fourier Theory