Ilka Agricola, Thomas Friedrich Global Analysis Differential Forms in Analysis, Geometry, and Physics Graduate Studies in Mathematics, V. 52 PDF

Global Analysis
Differential Forms in Analysis,

Geometry and Physics
Ilka Agricola
Thomas Friedrich
Graduate Studies
in Mathematics
Volume 52
American Mathematical Society

Global Analysis
Global Analysis
Differential Forms in Analysis,
Geometry and Physics
Ilka Agricola
Thomas Friedrich
Translated by
Andreas Nestke
Graduate Studies
in Mathematics
Volume 52
American Mathematical Society

Providence, Rhode Island
Editorial Board
Walter Craig
Nikolai Ivanov
Steven G. Krantz
David Saltman (Chair)
2000 Mathematics Subject Classification. Primary 53-01;

Secondary 57-01, 58-01, 22-01, 74-01, 78-01, 80-01, 35-01.
This book was originally published in German by Friedr. Vieweg & Sohn Verlagsge-
sellschaft mbH, D-65189 Wiesbaden, Germany, as "Ilka Agricola and Thomas Friedrich:
Globale Analysis. 1. Auflage (1st edition)", ©F}iedr. Vieweg & Sohn Verlagsgesellschaft
mbH, Braunschweig/Wiesbaden, 2001
Translated from the German by Andreas Nestke
Library of Congress Cataloging-in-Publication Data

Agricola, I1ka, 1973-
(Globale Analysis. English)
Global analysis : differential forms in analysis, geometry, and physics / Ilka Agricola, Thomas
Ftiedrich ; translated by Andreas Nestke.
p. cm. - (Graduate studies in mathematics, ISSN 1065-7339 ; v. 52)
Includes bibliographical references and index.
ISBN 0-8218-2951-3 (alk. paper)
1. Differential forms. 2.Mathematical physics. 1. Friedrich, Thomas, 1949 ll. Title.
111. Series.
QA381.A4713 2002
514'.74- dc2l 2002027681
Copying and reprinting. Individual readers of this publication, and nonprofit libraries
acting for them, are permitted to make fair use of the material, such as to copy a chapter for use
in teaching or research. Permission is granted to quote brief passages from this publication in
reviews, provided the customary acknowledgment of the source is given.
Republication, systematic copying, or multiple reproduction of any material in this publication
is permitted only under license from the American Mathematical Society. Requests for such
permission should be addressed to the Acquisitions Department, American Mathematical Society,
201 Charles Street, Providence. Rhode Island 02904-2294, USA. Requests can also be made by
e-mail to reprint-peraissionlaes.org.
© 2002 by the American Mathematical Society. All rights reserved.
The American Mathematical Society retains all rights
except those granted to the United States Government.
Printed in the United States of America.
The paper used in this book is acid-free and falls within the guidelines
established to ensure permanence and durability.
Visit the AMS home page at http://wv.aas.org/
10987654321 070605040302
Preface
This book is intended to introduce the reader into the world of differential
forms and, at the same time, to cover those topics from analysis, differ-
ential geometry, and mathematical physics to which forms are particularly
relevant. It is based on several graduate courses on analysis and differen-
tial geometry given by the second author at Humboldt University in Berlin
since the beginning of the eighties. From 1998 to 2000 the authors taught
both courses jointly for students of mathematics and physics, and seized the
opportunity to work out a self-contained exposition of the foundations of
differential forms and their applications. In the classes accompanying the
course, special emphasis was put on the exercices, a selection of which the
reader will find at the end of each chapter. Approximately the first half of
the book covers material which would be compulsary for any mathematics
student finishing the first part of his/her university education in Germany.
The book can either accompany a course or be used in the preparation of
seminars.
We only suppose as much knowledge of mathematics as the reader would ac-

quire in one year studying mathematics or any other natural science. From
linear algebra, basic facts on multilinear forms are needed, which we briefly
recall in the first chapter. The reader is supposed to have a more extensive
knowledge of calculus. Here, the reader should be familiar with differential
calculus for functions of several variables in euclidean space R", the Rie-
mann integral and, in particular, the transformation rule for the integral,
as well as the existence and uniqueness theorem for solutions of ordinary
differential equations. It is a reader with these prerequisites that we have in
mind and whom we will accompany into the world of vector analysis, Pfaf flan
v
vi Preface
systems, the differential geometry of curves and surfaces in euclidean space,

Lie groups and homogeneous spaces, symplectic geometry and mechanics,
statistical mechanics and thermodynamics and, eventually, electrodynamics.
In Chapter 2 we develop the differential and integral calculus for differential

forms defined on open sets in euclidean space. The central result is Stokes'
formula turning the integral of the exterior derivative of a differential form
over a singular chain into an integral of the form itself over the bound-
ary of the chain. This is in fact a far-reaching generalization of the main
theorem of differential and integral calculus: differentiation and integration
are mutually inverse operations. At the end of a long historical develop-
ment mathematicians reached the insight that a series of important integral
formulas in vector analysis can be obtained by specialization from Stokes'
formula. We will show this in the second chapter and derive in this way
Green's first and second formula, Stokes' classical formula, and Cauchy's
integral formula for complex differentiable functions. Furthermore, we will
deduce Brouwer's fixed point theorem from Stokes' formula and the Weier-
strass approximation theorem.
In Chapter 3 we restrict the possible integration domains to "smooth"

chains. On these objects, called manifolds, a differential calculus; for func-
tions and forms can be developed. Though we only treat submanifolds of
euclidean space, this section is formulated in a way to hold for every Rie-
mannian manifold. We discuss the concept of orientation of a ma.nifold, its
volume form, the divergence of a vector field as well as the gradient and the
Laplacian for functions. We then deduce from Stokes' formula the remaining
classical integral formulas of Riemannian geometry (Gauss-Ostrogradski for-
mula, Green's first and second formula) as well as the Hairy Sphere theorem,
for which we decided to stick to its more vivid German name, `'Hedgehog
theorem". A separate section on the Lie derivative of a differential form
leads us to the interpretation of the divergence of a vector field as a mea-
sure for the volume distortion of its flow. We use the integral formulas to
solve the Dirichlet problem for the Laplace equation on the ball in euclidean
space and to study the properties of harmonic functions on R". For these
we prove, among other things, the maximum principle and Liouville's the-
orem. Finally we discuss the Laplacian acting on forms over a Riemannian
manifold, as well as the Hodge decomposition of a differential form. This
is a generalization of the splitting of a vector field with compact support in
R" into the sum of a gradient field and a divergence-free vector field, going
back to Helmholtz. In the final chapter we prove Helmholtz' theorem within
the framework of electrodynamics.
Preface vii
Apart from Stokes' theorem, the integrability criterion of Frobenius is one of

the fundamental results in the theory of differential forms. A geometric dis-
tribution (Pfafflan system) is defined by choosing a k-dimensional subspace
in each tangent space of an n-dimensional manifold in a smooth way. A
geometric distribution can alternatively be described as the zero set of a set
of linearly independent 1-forms. What one is looking for then is an answer
to the question of whether there exists a k-dimensional submanifold such
that, at each point, the tangent space coincides with the value of the given
geometric distribution. Frobenius' theorem gives a complete solution to this
question and provides a basic tool for the integration of certain systems of
first order partial differential equations. Therefore, Chapter 4 is devoted to
a self-contained and purely analytical proof of this key result, which, more-
over, will be needed in the sections on surfaces, symplectic geometry, and
completely integrable systems.
Chapter 5 treats the differential geometry of curves and surfaces in euclidean

space. We discuss the curvature and the torsion of a curve, Frenet's formu-
las, and prove the fundamental theorem of the theory of curves. We then
turn to some special types of curves and conclude this section by a proof of
Fenchel's inequality. This states that the total curvature of a closed space
curve is at least 27r. Surface theory is treated in Cartan's language of mov-
ing frames. First we describe the structural equations of a surface, and then
we prove the fundamental theorem of surface theory by applying Frobenius'
theorem. The latter is formulated with respect to a frame adapted to the
surface and the resulting 1-forms. Next we start the tensorial description of
surfaces. The first and second fundamental forms of a surface as well as the
relations between them as expressed in the Gauss and the Codazzi-Mainardi
equations are the central concepts here. We reformulate the fundamental
theorem in this tensorial description of surface theory. Numerous examples
(surfaces of revolution, general graphs and, in particular, reliefs, i.e. the
graph of the modulus of an analytic function, as well as the graphs of their
real and imaginary parts) illustrate the differential-geometric treatment of
surfaces in euclidean space. We study the normal map of a surface and are
thus lead to its Gaussian curvature, which by Gauss' Theorema Egregium
belongs to the inner geometry. Using Stokes' theorem, we prove the Gauss-
Bonnet formula and an analogous integral formula for the mean curvature
of a compact oriented surface, going back to Steiner and Minkowski. An im-
portant class of surfaces are minimal surfaces. Their normal map is always
conformal, and this observation leads to the so-called Weierstrass formu-
las. These describe the minimal surface locally by a pair of holomorphic
functions. Then we turn to the study of geodesic curves on surfaces, the in-
tegration of the geodesic flow using first integrals as well as the investigation
of maps between surfaces. Chapter 5 closes with an outlook on the geometry
of pseudo-Riemannian manifolds of higher dimension. In particular, we look
at Einstein spaces, as well as spaces of constant curvature.
Symmetries play a fundamental role in geometry and physics. Chapter 6

contains an introduction into the theory of Lie groups and homogeneous
spaces. We discuss the basic properties of a Lie group, its Lie algebra, and
the exponential map. Then we concentrate on proving the fact that every
closed subgroup of a Lie group is a Lie group itself, and define the structure
of a manifold on the quotient space. Many known manifolds arise as homo-
geneous spaces in this way. With regard to later applications in mechanics,
we study the adjoint representation of a Lie group.
Apart from Riemannian geometry, symplectic geometry is one of the es-

sential pillars of differential geometry, and it is particularly relevant to the
Hamiltonian formulation of mechanics. Examples of symplectic manifolds
arise as cotangent bundles of arbitrary manifolds or as orbits of the coadjoint
representation of a Lie group. We study this topic in Chapter 7. First we
prove the Darboux theorem stating that all symplectic manifolds are locally
equivalent. Then we turn to Noether's theorem and interpret it in terms of
the moment map for Hamiltonian actions of Lie groups on symplectic man-
ifolds. Completely integrable Hamiltonian systems are carefully discussed.
Using Frobenius' theorem, we demonstrate an algorithm for finding the ac-
tion and angle coordinates directly from the first integrals of the Hamilton
function. In §7.5, we sketch the formulations of mechanics according to
Newton, Lagrange, and Hamilton. In particular, we once again return to
Noether's theorem within the framework of Lagrangian mechanics, which
will be applied, among others, to integrate the geodesic flow of a pseudo-
Riemannian manifold. Among the exercises of Chapter 7, the reader will
find some of the best known mechanical systems.
In statistical mechanics, particles are described by their position probability

in space. Therefore one is interested in the evolution of statistical states of
a Hamiltonian system. In Chapter 8 we introduce the energy and informa-
tion entropy for statistical equilibrium states. Then we characterize Gibbs
states as those of maximal information entropy for fixed energy, and prove
that the microcanonical ensemble realizes the maximum entropy among all
states with fixed support. By means of the Gibbs states, we assign a ther-
modynamical system in equilibrium to a Hamiltonian system with auxiliary
parameters satisfying the postulates of thermodynamics. We discuss the
Preface ix
role of pressure and free energy. A series of examples, like the ideal gas,
solid bodies, and cycles, conclude Chapter 8.
Chapter 9 is devoted to electrodynamics. Starting from the Maxwell equa-

tions, formulated both for the electromagnetic field strengths and for the
dual 1-forms, we first deal with the static electromagnetic field. We prove
the solution formula. for the inhomogeneous Laplace equation in three-space
and obtain, apart from a description of the electric and the magnetic field
in the static case, at the same time a proof for Helmholtz' theorem as men-
tioned before. Next we turn to the vacuum electromagnetic field. Here we
prove the solution formula for the Cauchy problem of the wave equation in
dimensions two and three. The chapter ends with a relativistic formulation
of the Maxwell equations in Minkowski space, a discussion of the Lorentz
group. the Maxwell stress tensor and a thorough treatment of the Lorentz
force.
We are grateful to Ms. Heike Pahlisch for her extensive work on the prepara-
tion of the manuscript and the illustrations of the German edition. We also
thank the students in our courses from 1998 to 2000 for numerous comments
leading to additions and improvements in the manuscript. In particular,
Dipl.-Math. Uli Kriihmer pointed out corrections in many chapters. Not
least our thanks are due to M. A. Claudia Frank for her thourough reading
and correcting of the German manuscript with regard to language.
The English version at hand does not differ by much from the original Ger-
man edition. Besides small corrections and additions, we included a detailed
discussion of the Lorentz force and related topics in Chapter 9. Finally, we
thank Dr. Andreas Nestke for his careful translation.
Berlin, November 2000 and May 2002
Ilka Agricola
Thomas Friedrich
Contents
Preface v
Chapter 1. Elements of Multilinear Algebra 1
Exercises 8
Chapter 2. Differential Forms in R" 11
§2.1. Vector Fields and Differential Forms 11
§2.2. Closed and Exact Differential Forms 18

§2.3. Gradient, Divergence and Curl 23
§2.4. Singular Cubes and Chains 26
§2.5. Integration of Differential Forms and Stokes' Theorem 30
§2.6. The Classical Formulas of Green and Stokes 35
§2.7. Complex Differential Forms and Holomorphic Functions 36
§2.8. Brouwer's Fixed Point Theorem 38
Exercises 43
Chapter 3. Vector Analysis on Manifolds 47

§3.1. Submanifolds of lR' 47
§3.2. Differential Calculus on Manifolds 54
§3.3. Differential Forms on Manifolds 67
§3.4. Orientable Manifolds 69
§3.5. Integration of Differential Forms over Manifolds 76
§3.6. Stokes' Theorem for Manifolds 79
§3.7. The Hedgehog Theorem (Hairy Sphere Theorem) 81
xi
xii Contents
§3.8. The Classical Integral Formulas 82

§3.9. The Lie Derivative and the Interpretation of the Divergence 87
§3.10. Harmonic Functions 94
§3.11. The Laplacian on Differential Forms 100
Exercises 105
Chapter 4. Pfaffian Systems 111

§4.1. Geometric Distributions 111
§4.2. The Proof of Frobenius' Theorem 116
§4.3. Some Applications of Frobenius' Theorem 120
Exercises 126
Chapter 5. Curves and Surfaces in Euclidean 3-Space 129

§5.1. Curves in Euclidean 3-Space 129
§5.2. The Structural Equations of a Surface 141
§5.3. The First and Second Fundamental Forms of a Surface 147
§5.4. Gaussian and Mean Curvature 155
§5.5. Curves on Surfaces and Geodesic Lines 172
§5.6. Maps between Surfaces 180
§5.7. Higher-Dimensional Riemannian Manifolds 183
Exercises 198
Chapter 6. Lie Groups and Homogeneous Spaces 207

§6.1. Lie Groups and Lie Algebras 207
§6.2. Closed Subgroups and Homogeneous Spaces 215
§6.3. The Adjoint Representation 221
Exercises 226
Chapter 7. Symplectic Geometry and Mechanics 229

§7.1. Symplectic Manifolds 229
§7.2. The Darboux Theorem 236
§7.3. First Integrals and the Moment Map 238
§7.4. Completely Integrable Hamiltonian Systems 241
§7.5. Formulations of Mechanics 252
Exercises 264
Chapter 8. Elements of Statistical Mechanics and Thermodynam: .es 271

§8.1. Statistical States of a Hamiltonian System 271
Contents xiii
§8.2. Thermodynamical Systems in Equilibrium 283

Exercises 292
Chapter 9. Elements of Electrodynamics 295

§9.1. The Maxwell Equations 295
§9.2. The Static Electromagnetic Field 299
§9.3. Electromagnetic Waves 304
§9.4. The Relativistic Formulation of the Maxwell Equations 311
§9.5. The Lorentz Force 317
Exercises 325
Bibliography 333
Symbols 337
Index 339
Chapter 1
Elements of
Multilinear Algebra
Consider an n-dimensional vector space V over the field K of real or complex

numbers. Its dual space V' consists of all linear maps from V to K. More
generally, a multilinear and antisymmetric map,
wk: Vx...xV-+K.
depending on k vectors from the vector space V, is called an exterior (mul-
tilinear) form of degree k. The antisymmetry of wk means that, for all
k vectors vl, ... , vk from V and any permutation a E Sk of the numbers
{ 1, ... , k}, the following equation holds:
wk(VQ(1), ....Va(k)) = sgn(a) i.Jk(VI, -..,Vk).

Here sgn(a) denotes the sign of the permutation a. In particular, wk changes
sign under a transposition of the indices i and j:
wk(i1i ...,Vi, ...,Vj, ...,Vk) = -wk(v1, ...,v ....,VI.....Vk).
The vector space of all exterior k-forms will be denoted by Ak( V*). Fur-
thermore, we will use the conventions A°(V') = K and A1(V*) = V.
Fixing an arbitrary basis e1, ... , e in the n-dimensional vector space V, we

see that each exterior k-form wk is uniquely determined by its values on all
k-tuples of the form el,,... 441 where the indices are always supposed to
be strictly ordered, I = (i1 < ... < ik). On the other hand, a k-form can be
defined by arbitrarily prescribing its values on all ordered k-tuples of basis
vectors and extending it to all k-tuples of vectors in an antisymmetric and
I
2 1. Elements of 1llultilinear Algebra
multilinear way. The number of different k-tuples of n elements is equal to

(k) = k. Thus we conclude
Theorem 1. If k > n, Ak(V*) consists only of the zero map. For k < n
the dimension of the vector space Ak(V*) is equal to
dim (nk(V`)) _ (k)
Exterior forms can by multiplied, and the product is again an exterior form.
Definition 1. Let wk E Ak(V*) and 7/ E A1(V') be two exterior forms of
degrees k and 1. respectively. Then the exterior product wk n 171 is defined as
a (k + 1)-form by the formula
1
wknll!(t'1....L'k+l) = Esgn(v)wk(v,(1),...Vo(k))11l(z'o(k+l)....vn(k+l))
k!l! oESk+j
Obviously. wk n,1 is a multilinear and antisymmetric map acting on (k + 1)

vectors, i. e. of degeree k+l. The following theorem summarizes the algebraic
rules governing computations involving the exterior multiplication of forms.
Theorem 2. The exterior product has the following properties:
(1) (wi +w2)AY/=w1 nrft+w2n
(2)
(3) (awk)A1)l=wkA(ar/)=a(wkAr/)foranyaEIfs:
(4) (wkAgl)A m=wkA(r11A ');
(5) wk A 111 = (-1)k'711 n wk.
Proof. Only the last two formulas require a proof, in which we shall omit
some of the upper indices for better readability. First
(k+l)!m! (wk A11l) Aµ/'"(v1, ...,vk+l+m)
1: sgn(a)(wk A 711)(v,(1), ... , v,(k+1))1f (vn(k+1+1).... , vs(k+l+m))
f7E Sk+l+m
Decompose the permutation group Sk+l+m into residue classes with respect
to the subgroup Sk+l C Sk+1+m formed by all permutations acting as the
identity on the last m indices {k + 1 + 1, ..., k + l + m}. Each residue
class R thus consists of all permutations a E Sk+l+m with fixed values
Q(k + I + 1), ..., v(k + I + m). Fixing any permutation ao E R. all the
remaining elements a E R are parametrized by the elements in Sk+l:
or = a0 0 7r, 7r E Sk+l
1. Elements of Multilinear Algebra 3
Hence,
E sgn(o)(wk A rll)(Va(l), --- , Vo(k+l))11m(Va(k+1+l), - - - , Va(k+l+m))
oER
rsgn(oo)sgn(ir)(w A r/)(Vaoo,r(1), --Vooo,,(k+())1L(Vao(k+t+1), ..Vao(k+t+m))
,rESk+i
= sgn(ao)(k + l)!(w A 17)(vao(1), ... , Voo(k+1))1J(VCo(k+1+1), ... , Vao(k+1+m))-
Using now the definition of the exterior product wk A 711, this leads to the
formula
k! 1!
(k + l)! a ERsgn(a)(w A i))(va(1), vo(k+l+m))
= sgn(o)w(va(1), ... va(k))1T(t'a(k+1) ... V0(k+l))1-(Va(k+1+1), - - -Va(k+l+m))-

aER
In order to compute the sum over the whole group Sk+l+m, we sum over
all the residue classes R and, simplifying the scalar factors, we obtain the
equation
(k! l! m!) - (wk A ill) A pm (v1, ... , tk+l+m )
E sgn(a)w(va(1), ... Vo(k))17(Vo(k+1), -- Vo(k+l))1(Va(k+1+1), . Vo(k+l+m))-

Sk+I+m
This shows the associativity of the exterior multiplication of forms. The last
formula (5) is proved analogously. 0
Definition 2. The exterior algebra A(V') of the vector space V is formed
by the sum of all exterior forms
A(V*) = k=0
E
endowed with the exterior multiplication A of forms as multiplication.
Next we will construct an explicit basis of the vector spaces Ak(V*). To
do so, start from any basis el, ... e of V and denote by a, ... , o the
dual basis of the dual space V' _ A' (V*). For an ordered k-tuple of indices
(k-index for short) I = (i1 < ... < ik) let of denote the k-form defined by
the formula
of := ail A ... A o;k .
Obviously, for a fixed k-index J = (j1 < ... < jk),
ol(ejl, --.,eik)
J0 if 196 J,
t 1 if I = J.
In particular, the k-forms of are linearly independent in Ak(V'). For di-
mensional reasons, this immediately implies
4 1. Elements of Multilinear Algebra
Theorem 3. Let e1, ... , en be a basis of the vector space V, and denote by
ol.... , o its dual basis in the dual space V. Then the forms o,,1 = (i1 <
... < ik), are a basis for the vector space Ak(V*).
Exterior forms can be pulled back under a linear map. In fact, if L : W -+
V is a linear map from the vector space W to the vector space V, and
Wk E A(V') is an exterior k-form in V, then the formula
(L*wk)(wl, ...,Wk) := wk(L(wl),... , L(wk))
defines an exterior k-form (L'wk) E Ak(W`) in the vector space W. Passing
from the form wk to the induced form L*(wk) is compatible with all algebraic
operations. In particular, the following formula holds:
L' (wk n,/) = (L*wk) A (L5,1)

Furthermore, a vector can be inserted into an exterior form, and the result
is an exterior form of one degree less. Let wk E Ak(V*) be a k-form on V
and vo E V any vector. Define a (k- 1)-form (voJ wk) E Ak-1(V*) by the
formula
(voJ wk)(v1, ...,vk-1) := wk(vp,V1, .... Vk-1)
The (k -1)-form vo J wk is called the inner product of the vector vo with the
k form wk, and will also be denoted by The antisymmetry of the
k-form wk leads to the following relation for (k - 2)-forms:
vl J (vo J wk) = - vo J (vl J wk) .
From now on, let V be a real vector space equipped with a non-degenerate
scalar product g. This is a symmetric bilinear form,
g : V x V ---- R,
with the property that the linear map g# : V - V` from V to the dual
space V' defined by
g#(v)(w) := g(v,w)
is bijective. For a given basis e1, ... , e,, of V, the matrix
M(g) _ (g(e{, ei))ij-1
is symmetric and invertible. For brevity, its entries will be denoted by
gig := g(ei, e,), the entries of the inverse matrix (M(g))-1 by g'j. Recall the
following result, which goes back to Lagrange and Sylvester:
Theorem 4. Let g be a non-degenerate scalar product on the real vector
space V. Then there exists a basis e1, ...,en in V such that the matrix
1. Elements of Multilinear Algebra 5
M(g) is diagonal, i. e.
1
0
1
M(g) =
0
L -1 J
The number p of (+1)-entries as well as the number q of (-1)-entries in
this diagonal matrix are independent of the particular basis. The pair (p, q)
is called the signature of the scalar product g, and the number q is called its
index.
First we extend the scalar product g to the spaces Ak(V*) of k-forms, keep-
ing the same symbol for its extension. This continuation is done, relative to
an orthonormal basis, by means of the formula
k k = ilil
...g ikik wi(ei,,
k
...,eikw2(ei
k
...,eik)
g(wl,w2) - g
it <...<ik
It is not difficult to see that this defines a non-degenerate scalar product in all
the spaces Ak(V') which does not depend on the choice of the orthonormal
basis. The signature, however, changes. For example, if g is a scalar product
with signature (n - 1, 1) in V, then the induced scalar product in the space
Ak(V*) of exterior k-forms has signature ((°A 1), (k-i)) If the original basis
el, ... , en in V is orthonormal, the basis al = ail A ... A aik (i1 < ... < ik)
is an orthonormal basis in Ak(V'). Here a1, ...,an denotes the dual basis
in V*. Moreover, the length of the k-form al is computed by the formula
g(a/, al) = gu 11 .... g`k`k
In particular, the scalar product in the one-dimensional space An(V') is
positive or negative definite, depending on the parity of the index q of the
scalar product.
Apart from a scalar product g, we fix an orientation in the real vector

space V. Let us recall what this means. Consider the set B(V) of all
ordered bases B = (vl,...,vn) in the vector space V. For two ordered
bases, B = (v1, .. . , vn) and B` = (vi.....v), there exists an (n x n)
matrix A(93, 'Z*) = (aij) j=1 such that
n
vi = aijvj
j=1
We define an equivalence relation N on the set B(V) by requiring

B N'B' if and only if det(A(B, B')) > 0.
Obviously, there are precisely two equivalence classes.
Definition 3. An orientation of the real vector space V is the choice of one
of the two equivalence classes in the set B(V) of all bases in V.
In the plane R2, an orientation can be understood as a sense of rotation:
el el
Figure 1 Figure 2
Figure 1 illustrates R2 with the orientation determined by the basis (el, e2),
whereas Figure 2 depicts the plane R2 equipped with the orientation (e2, el).
The transition matrix between both bases is the matrix
A __ [00 1
0
with negative determinant, det(A) = -1. Hence the bases (er, e2) and
(e2, el) represent the two possible orientations of R2.
Example 1. Let (Vg) be a real vector space with fixed orientation. We
choose a basis el, ..., e,, in V in such a way that the matrix M(g) has the
diagonal format of Theorem 4, and (el, ... , e,) is positively oriented. Then
the formula
g(vl,el), ...,g(vn,el)
dV(vl, ...,vn) := det
g(vi, en), ... , g(vn, en)
defines an n-form dV E An(V') independent of the choice of the particular
basis e1, ... , e with the stated properties. This form dV is called the volume
form of the oriented vector space with the non-degenerate scalar product g.
By means of the dual basis al, ..., an, the volume form is represented as
dV = (-1)Q al A...Aa,,.
Here q denotes the index of the scalar product g. The length of the volume
form is, by definition,
g(dV, dV) = (-1)9
1. Elements of Multilinear Algebra
The volume form of the coordinate space R" with the euclidean scalar pro-
duct coincides with the determinant:
dR"(vl, ...,vn) = det(vi, ...,vn), v, E R".
In particular, the determinant turns out to be an n-form on the space R".
Now we introduce the *-operator (Hodge operator) assigning a k-form to
every (n - k)-form. Consider the given real vector space V together with a
scalar product g and a fixed orientation. For each k-form Sc E AV'), the
rule
An-k(V*) E) gn-k _ Wk A r)n-k E An(V')
determines alinear mapping from An-k(V*) to the 1-dimensional space
An(V*). The volume form dV is a basis vector in An(V*), and the vector
space An-k(V*) is equipped with a non-degenerate scalar product. Hence,
there exists exactly one (n - k)-form-to be denoted by *wk-such that
Wk A,fin-k = g(*Wke n-k)dV
71
holds for all (n - k)-forms 17n-k. Summarizing we defined a linear mapping

*: Ak(V*) -A n-k(V*),
....
the so-called Hodge operator.
Example 2. As above, let e1, ... , en be an orthonormal basis (g;i = fb;i)
of the vector space V representing the fixed orientation, and let al, ... , an
be the dual basis. For an ordered k-index I = (i1 < ... < ik), denote by
J = (j1 < ... < in-k) the "complementary" index, i. e. the ordered (n - k)-
index containing the numbers 11, ... , n}\{il, ... , ik}. The equations
l . n 1 . n
g(*aj,oj)dV = alAa = sgn IJo'1A...Aan = (-1)°sgn
IJ dV
and immediately imply the formula
*oj = (-1)Isgn (Il...Jn gijiI ...gin-yin-4a

The following theorem assembles the main properties of the Hodge *-operator
for a scalar product of arbitrary signature.
Theorem 5.
(1) For each k -form Sc, the twofold application of the Hodge operator
is given by * * wk = (-1)k(n-k)+gwk;
(2) for any k -forms wk, r7k, the following relations hold for the exterior
and the scalar product:
g(*wk, *rlk) = (-1)gg(Wk, rlk), Wk A *,nk = (-1)gg(Wk, rlk)dV .
Proof. To prove the first formula we first compute
= (-1)Qsgn
(i...n)9.,... gi..-kin-k * oJ
2y l...n l...n\
_ (-1) 5
1 J 9,0, ...9il-kie-ksgn (I J I g";, ...y;k,koi
_ (-1)9(-1)kin-kioi.
/
The other formulas are consequences of this first one and the definition of
the *-operator. We have
(*wk, *rik)dV = wk A *I?k = (-1)k(n-k) * flk Awk
(-1)k(n-k)9(*
= * r)k,wk)dV = (-1)99(r/k,wk)dV ,
and this implies the remaining identities. 0
Example 3. Let n = 2k be an even number. Then the Hodge operator
maps the vector space Ak (V *) to itself,
*:Ak(V*)-Ak(V*).
Moreover,
**talk = (-1)k+gWk,
and hence, in case k + q = 0 mod 2, the Hodge operator has the eigenvalues
± 1.
Exercises
1. Let a,, ..., an be a basis of V*, let wl = E aj o;, ril = F_ bi a; be two

arbitrary elements from V*, and let µ2 = c;j o; A aj be a 2-form.
a) Compute wI A ill and, in the case n = 3, explain in which sense the
exterior product generalizes the vector product;
b) compute wl A µ2 and discuss in case n = 3 the relation to the scalar
product.
2. Prove that each 2-form w2 E A2(V*) can be represented as

w2 = a, A a2 + ... + a2r_i A a2r
for a certain basis a,, - .. , on of V* - Prove, moreover, that the number r is
independent of the choice of the basis and is characterized by the condition
(w2)r 76 (w2)r+1 =0.
Exercises
3. Prove that k linear forms ol, ... , ok on V' are linearly independent if
and only if
olA...nok # 0-
4 (Cartan's Lemma). Let o1i ... , ok be linearly independent linear forms

from V' and let µl, ... µk be arbitrary elements of V. If o1 A pi + ... +
ok A Ilk = 0 holds, then the forms pi are linear combinations of the off,
k
J=1
Moreover, ail = aji.
5. Let e1, .... e,a be a basis of the vector space V and o1, ... , o the cor-
responding dual basis. Then the following formula holds for every k-form
Wk:
n
o,A(eiJ Wk) =
i=1
6. For a given form 0 96 wk E Ak(V*), define the subspace M(Wk) C V' by

{171
1tf(wk) = E A'(V*) : ql AWk = O } .
Prove the following statements.

a) The inequality dim M(wk) < k holds:
b) the equality dim M(wk) = k holds if and only if there exist k linear
forms Cl, .... ok such that wk = of A ... A ok.
Forms of the kind wk = of A ... A ok are called decomposable.
7. Prove the following statements.

a) A 2-form w2 = F_i.j wig oiAaj is decomposable (see the previous exercise)
if and only if
WijWpg - WipWjg + WigWjp = 0;
b) a 3-form w3 = Eij k Wijk of A of A ok is decomposable if and only if
WickWpgr - WijPWkgr + wi.jq Wkpr - wilt wA-M = 0-
8. The Hodge operator of a 4-dimensional vector space V maps the space

of 2-forms into itself,
* : A2(V') -A 2(V'),
If the index of the scalar product is even, q = 0, 2, then the Hodge operator
decomposes the real vector space A2(V') into the eigenspaces
A2 (V') = {w2 E A2(V') : * W2 = f w2
In the case of an odd index, q = 1, 3, the complexification A2(V') $ C
analogously decomposes into the ±i-eigenspaces. Compute the dimension
of the eigenspaces in both cases, and determine a basis of eigenforms.
Chapter ,2
Differential Forms
in '
2.1. Vector Fields and Differential Forms
Vectors in euclidean space R" can be understood as "free vectors" or as
"place-bound vectors" at a point in R". In the first case, we simply consider
R" as a euclidean vector space. The second point of view is based on the
concept of R" as a set or a metric space whose elements are called the points
of space. The "place-bound vectors" then form vector spaces of their own,
each one consisting of all those attached to a particular point. For example,
vectors at different points cannot be added. This second concept of a vector
in R" leads to the notion of the tangent space at a point in R".
Definition 1. Let p E R". The tangent space to R" at the point p is defined
to be the set
TpR" :_ {(p, v) : v E R" } .
An addition and a multiplication by scalars \ E R are introduced in this
set by (p, v) + (p, w) := (p, v + w) and A (p, v) :_ (p, \ v), respectively.
These operations endow each TpR" with the structure of a real n-dimensional
vector space.
The ordinary differential of a smooth map f from the space R" to the space
R'" at the point p, D fp, can now be interpreted as a linear map f.,p between
corresponding tangent spaces.
Definition 2. Let U C R" be open and f : U R' be a differentiable
map. For each point p in the set U, the linear map f.,p : TpR" -. Tf(p) Rm
11
12 2. Differential Forms in 1R"
is defined by the formula

(f (p), Dff(v))
Definition 3. A vector field defined on an open subset U of IR" assigns a
vector V(p) E TDR" in the corresponding tangent space to each point p E U.
If e1.... , e is the standard basis in euclidean space lR". the vector field
determined by p (p,ei) is usually denoted by 8/8x', i.e. (8/8x')(p) :=
(p, e,). (Compare Exercise 5 for an explanation of this notation.) Obviously,
every other vector field defined on U can be represented in the form
0 (P) + ... + V" (p) 81" (P)

V (P) = V, (P) iix-l
with certain functions V1..... V" defined on U. The vector field V is called
differentiable of class Ck if all its component functions V1, .... V" have con-
tinuous partial derivatives up to order k. The set of all vector fields of class
Ck is, on the one hand, a real vector space and, at the same time, a module
over the ring Ck(U) of all real-valued Ck-functions on U. Graphically, a
vector field can be depicted by drawing at each point p the corresponding
second component v of the vector from the tangent space. Consider, e.g.,
on the plane R2 with coordinates x, y the vector field
V(x, y) =
The following figure depicts this vector field:
y
A
tt
t
ttT
f
Each of the tangent spaces Tp R" is a real vector space. Hence we can
consider the dual space, 71R' := (TpR, "")'. as well as its exterior powers,
/"(R") := "k(T;R")
2.1. Vector Fields and Differential Forms 13
An element wk of the space Ap(R") is thus an antisymmetric multilinear

map with k arguments on the tangent space TplR":
wk : Tp1R" x ... x TpR" -p R.
In analogy with the notion of vector field, we will now introduce the notion
of differential form.
Definition 4. A k -form on the open subset U of lR" assigns to each point
p E U an element wk(p) E Ap(1R")
First we want to consider some examples of differential forms.

Example 1. Let f : U R be a smooth real-valued function, let p E U be
a fixed point, and let D fp : IR" -, IR be its differential at the point p. Then
the formula
df (p)(p, v) := Dfp(v)
obviously defines a 1-form df on the set U.
Example 2. A fixed basis el, ... , e" in IR" determines n coordinate func-
tions x1, ... , x" and hence their differentials dx', ..., dx". Thus for a tan-
gent vector (p, v) E TpR" we have the identity
dx`(p)(p,v) = v`,
where v` denotes the i-th component of v with respect to the basis eI, ... , e".
In addition, the 1-forms dxl (p), .... dx"(p) form a basis of the vector space
A' (R") = T R". Arbitrary exterior products dx" A ... A dx'k as well as
their linear combinations with functions as scalar coefficients lead to further
examples of k-forms. Conversely, each k-form wk on U can be represented
as
wk -= wjl....,tkdx
11
A ...ndx ik
;1 <...<ik
with certain functions w;,,
A differential k-form of class C' is defined as a form wk whose coefficient

functions w;,,,.,,;k all are of class C'. The set of all these forms is denoted by
Sti (U). Obviously, this is a real vector space as well as a module over the
ring C'(U). The 0-th exterior power of a vector space is its field of scalars.
Therefore, the 0-forms of class C' are taken to be the C'-functions,
1l?(U) := C'(U).
Example 3. Let f be a real-valued function of class C' on the open se
U C IR". Then df is a 1-form of class C". and
df = 209X LI dx'+...+'f dx".
14 2. Differential Forms in R"
In fact, at the point p E U the following equality holds for the vector (p. v):
n of
df(p)(p,v) = Dff(v) _ axi (P)v'
Replacing the vector components v' by dx'(p)(p, v) and omitting the argu-
ment (p, v), we arrive at the stated formula
n 9f
df (p) _ (P) . dx' (p)
The exterior product of multilinear forms can easily be extended to differ-

ential forms, defining for two forms wk, 7771 on U a (k + l)-form by
(wk An`)(P) := wk(P) A7l(p)
The rules known from the first chapter remain valid without change:
(wk+µk)A7j' = wk AY/ +µkA7/';
(1)
(2) (f wk) A = f wk A i/;
(3) wk A ii = (-1)kl7!property
A wk.
In particular, the last implies that the exterior product of a form
of odd degree with itself always vanishes, e.g., dx` A dx' = 0. Forms of even
degree do not in general have this property: A straightforward calculation
shows that for the forms on R4 below the following relations hold:
(dx' A dx2 + dx3 A dx4) A (dx' A dx2 + dx3 A dx4) = 2 dx' A dx2 A dx3 A dx4 .
For conceptual as well as computational reasons, it is important that differ-

ential forms can be "pulled back" by maps.
Definition 5. Let f : Ul --y U2 be a differentiable map between two open
subsets Ul C R" and U2 C R, and let, moreover, wk be a k-form on U2.
Then a k-form f' (wk) (the pullback or induced form) on Ul is defined by
f,(wk)((P,Vi),...,(p,vk)) := wk(fa,P(P,vl),...,f..P(P,vk))
The form f * (wk) obeys the following rules.
Theorem 1. Let f : Ul -, U2 be a differentiable map between the open sets
d,,,
Ui C R" and U2 C R' with component functions f'. Then
f`(dx`) _ " 8xi
Let, moreover, differential forms on U2 and a function g : U2 -p R be given.

Then
(1) f*(wi +w2) = f*(wk)+f*(w2);
(2) f' (g . wk) = (g ° f) f *(w k);

(3) f*(wk Arl1) = f'(wk) A f*(r/).
Furthermore, in the case n = m, the following additional equality holds:
Proof. The definition of f' implies

f*(dxt)(P)(P,v) = dx'(f(P))(f(P), Dfp(v))
But this is precisely the i-th component of Dfp(v). Hence
f(dx')(P)(P, v) = at (P)za 4(p)dx'(p)(p,v).

i=1 i=1
The three rules immediately follow from the definitions and will therefore
not be proved here. To derive the last equation, we use the second rule and
the identity just proved:
A...ndxn) = (guf)f*(dxl)n...A fs(dx")
afdxJI] fn n
= (g ° f) A ... A
8xi^
it=1 in=1
n afl (9f n
_ (g ° f) E aTi1 " ' 5; dxit n ... A dxin
it, ....i.,=1
Since the exterior square of a 1-form always vanishes, in this sum only
those terms remain whose n-index (j1.... ,j) is a permutation of (1, ..., n).
Using the antisymmetry of the multiplication in the exterior algebra, we
obtain the formula
(g°f)Esg11(1 J n)BxJi'...'8xj. dx1 n...ndx"

J
and thus the determinant of the differential D f . 0
Recall that a Ci-function f on U can be considered as a 0-form, so the
differential d is also a map turning a 0-form f into a 1-form df. Applying
this map d in a suitable way to the coefficient functions of a differential
form, the exterior derivative can be extended to act on k-forms in general.
Definition 6. Let wk be a k-form on the open set U,
k E wit, .ik dx
it
A ... A dxik
it <...<ik
16 2. Differential Forms in IR'
We define its exterior derivative dwk by the formula

CO = E d(wi1....,ik) dxi' A ... A dxik
i 1 <...<ik
n
wi1.... A.
axa dx a n dx'i n . . . A dxk:.
it <...<ik a=1
Hence dwk is a (k + 1)-form of class Cl-1, and d becomes a linear operator

between corresponding spaces,
d: nk(U) Qi+1(U).
Example 4. Consider on R2 the 1-form
w1 = sin x dy + sin y dx .
Since d(sin x) = cos x dx + 0 dy and d(sin y) = 0 dx + cosy dy, we have
dw' = cos x dx n dy + cos y dy n dx = (cos x - cos y) dx n dy.
Theorem 2. The exterior derivative obeys the following rules:
(1) d(wk +qk) = dwk+dqk forwk, 17k
E Qk(U);
(2) d(wk n 171) _ (dwk) Aril + (-1)kwk n (drll);
(3) d((Iwk) = 0 for wk E s22(U);
(4) f*(dwk) = d(f*wk) forwk E f21(U2) and f : U, - U2 c R"'.
Proof. The first identity is trivial. For the second, we use the multi-index
notation ! = (i1 < ... < ik) and set dx1 = dxi' n ... A dxik as well as
wA = > wl dxl , rlt = > qj dxJ .
I J
Thus
wk A rli = wlr,J dx' n dxJ ,
I,J
and, applying the exterior derivative, we obtain the formula
d(wk n q1) E a=
(8x° qJ + wI a7o
I dx° n dx1 n dxJ .
The first summand equals dwk A7/, and in the case of the second precisely
k transpositions lead to
d(w'nr71) = dWknr7'+(-1)kEEwI" dx'ndx"ndxJ

I,J a=1
= dwkAr71+(-1)kwkndr/.
Now we will show that dd = 0. Applying the derivative twice yields the
expression
3 n dx° n dxi
ddwk = EE 013x°
I 3=1 o=1
( a2à,'\ dx3 A dx° A dxl.
I a<3 8x3x° ax°x3
By assumption, the second partial derivatives of the functions w/ are con-
tinuous. Hence, Schwarz' lemma implies that the expression in brackets
vanishes f o r each multi-index I. Finally, w e show that pulling back differen-
tial forms commutes with the exterior derivative. T o this end, let y1, ... , yi1
be the coordinates on V, let the form wk = Ei wi dy' be expressed in these.
and let f 1, .... f' be the components of the map f : U1 - U2. From
dwk =
[ ` wI dy° A dy'
o=11
I
we obtain, for the pullback of the form,
f'( k) = 1: y° (f(x))df°ndf'1
! ww
A...Adf'k.
/ o=1
On the other hand,
f*wk = w,(f(x))df'1 A...Adf`k,
I
and its derivative is computed as follows:
n
aw,
d(f'wk) = axa(f(x))dx3ndj'1 n... ndf
13=1
(f (x)) af- dx3 A df" A ... A df'k

/ Q=1o=1
Making use of df° = E(af°/0x3) dx,3, this can be simplified:
d(f*wk) = .EaW°(f(x))dfoAdf'1 A...Adf'k = f*(dwk).

I °=1
Example 5. The following example illustrates that, for computational pur-
poses, the rules for f * and d formulated in Theorems 1 and 2 often suffice.
Consider, e.g., the map
f: R2 - R3, f (U, v)
= (u2, v3, uv)
18 2. Differential Forms in Rn
and the 1-form on R3 defined in the coordinates x, y, z by

w' = ydx+xdy+xyzdz.
Compute f'(w') as follows. First,
f*(wl) = (yof)f'(dx)+(xof)f'(dy)+(xyzof)f*(dz)
= v3 f' (dx) + u2 f' (dy) + u3v4 f' (dz) .
Then we use the fact that the exterior derivative commutes with the induced
map:
f'(dx) = d(f'x) = d(xof) = d(u2) = 2u du.
Similarly,
f*(dy) = 3v2dt,, f'(dz) = vdu+udv,
from which the result follows:
f* W) = (2uv3+u3v5)du+(3u2v2+u4v4)dv.
2.2. Closed and Exact Differential Forms
From the theory of the Riemann integral, it is well-known that every con-
tinuous function on R has a primitive function. In the language of forms,
this can be expressed by saying that for every 1-form µ' = g(x) dx with a
continuous coefficient function g : R - R, there exists a function f such that
df = p'. If, in addition, g is differentiable, then certainly dµ' = 0, since
each 2-form on R vanishes. Now we want to pose the analogous question of
whether any differential form has a "primitive form":
Let pk E fli (U) be a k-form. What are the conditions
guaranteeing the existence of a (k-1)-form gk-' E ft2 ' (U)
whose exterior derivative coincides with µk, dgk-1 = pk
The equality dd = 0 immediately implies that dpk = 0 is a necessary con-
dition, but, in general, it is not sufficient for the solvability of the equation
dq ' =,[k.
Definition 7.
(1) A k-form wk E f1k(U) is called closed if dwk = 0;
(2) a k-form wk E fZi (U) is called exact if there exists a (k - 1)-form
nk-' E f4-'(U) such that de-' = wk
The property dd = 0 states that each exact form is closed.
Example 6. Consider on the open set U = JR2-10) the winding form
Wi = -y x2+y2dx +x2+y2dy
x
2.2. Closed and Exact Differential Forms 19
and calculate its derivative dwi:

-dy y(2x dx + 2y dy)1 r dx x(2x dx + 2y dy) l y
x2 + y2 + (x2 + y2)2 J A dx + [x2 + y2 (x2 + y2)2 A d
- y2+x2-2y2+x2+y2-2x2dx
(x2 + y2)2 Ady = 0.
Hence, wl is closed; later we will see that wl is not exact (Example 10). The
integral of the winding form along a closed curve surrounding zero measures
how often it "turns around" the origin (Exercises 1 and 2).
In algebraic topology, it is common to describe the difference between closed
and exact forms by the so-called de Rham cohomology. The vector spaces
of "cycles" Zk(U) and "boundaries" Bk(U) are defined by
Zk(U) {wk E !1k (U) wk is closed},
Bk(U) {WA E S1;(U) WA is exact} .
Thus Bk(U) is a subspace of Zk(U), and the k-th de Rham cohomology of

U is defined as the quotient space
HDR(U) := Zk(U) / Bk(U).
The winding form, e.g., is a non-trivial element in H)DR(R2 - {0}) # 0, and
later (Exercise 14) we will show that
HDR(II 2 - {0}) = R.
The k-th de Rham cohomology only depends on the topological shape of the
set U. For example, the de Rham cohomology of a convex set vanishes. For
slightly more general sets this is the contents of Poincare's lemma, which
we are going to discuss now.
Definition 8. A subset U of IRn is called star-shaped (or a star-region if it
is open in addition) if there exists a point po E U with the property that
for every second point x E U the segment joining po with x is completely
contained in U. Obviously, star-regions are path-connected.
Theorem 3 (Poincare's Lemma). Let U be a star-shaped open set in lid".

Then
HDR(U) = 0
for every k = 1, ... , n.
In other words: For each closed k form wk E 1l (U) there exists a (k - 1)-
form rik-1
E Qk-1 such that drik-' = wk.
Proof. In order to show this, we assign to every k-form wk = wi dx' a

(k - 1)-form P(wk) satisfying the identity
wk = P(dwk) + dP(wk) .
For closed forms the first term vanishes, which proves the assertion. The
form P(wk) is defined as follows:
k rj'tk_1wI..ik(tx)dt] _
P(wk) _ (-1)°-1
x° dxi' A... Adxio A... A
°.
1j< ... <ik
I
Here the notation dxi° is intended to indicate that the corresponding factor
is omitted. Hence, P(wk) is of degree one less than wk, as claimed. Let
us compute the exterior derivative of P(wk). To do so, apply the product
rule to the coefficient functions: The integral has to be differentiated with
respect to x implicitly, whereas the derivative of xi° each time yields the
term
dxi° Adxi' A...Adxi° A...Adxik.
The sign (-1)°-1 is chosen in order to render this expression equal to dx1' A
... A dxik. Altogether, we thus obtain
dP(wk) =k JtlC_twjl..ik(tx)dt dx" A ... A dxik

l1<...<ik o
n.k
[]tk0';1k(tx)dt]
+ L
;.°=1,
(-1)°-1
xi°dx`jAdxi1A...Adxi°A...Adxik.
11G..<tk
Calculating, on the other hand, dwk as usual leads to

n
dwk - awîi...ik dxj A dx" A ... dxik .
it <...<ik 7=1
2.2. Closed and Exact Differential Forms 21
We apply the operator P to this result:

a
P(dwk) =EL [om
i1<...<ik.i-1
[ f't tk
&i1...ik
axi (tx)dt] xj dxi' A ... A dxik
-1-1 [J'tyk (tx)dt] x°dx Adxi'A...AdxioA...Adx.

n,k
- ll
i1 <.<ik j,a=1
P (k) + dP(wk) = k [1' wi1...ik(tx)dt] dxi' A ... A dxik

i1<..<ik J0 tk-l
n f1 1
a ik
+ tk (tx)dtJ x; dxi' A ... A dxik
Y- [J0 ax -7
i1 <...<ik .7=1
1d/
[dt
fo
(tkwl(tx)) dt] dxl
_ > (wj(x) - O . wl(0)) dxl = >2wl(x)dx'.
Poincare's lemma is but an existence result for the desired form 77k-1, and
does not claim its uniqueness. For example, if 77k-1 is a solution of the
equation d77k-1 = wk, the sum 77k-1 +dCk-2 also solves the equation for any
(k - 2)-form k-2. Conversely, for any two forms ii-i and ,jr' satisfying
d77i-1
= d772-1 = wk, we have
d(771-1
- 772 1) = 0.
-1-
Then, by Poincare's lemma, there exists a (k-2)-form do-2 such that ii
772-1 = duo-2. In other words,
77i-1
- 772-1 +
and we conclude that the general solution can always be expressed as the
sum of a particular solution and the derivative of an arbitrary (k - 2)-form.
Example 7. Consider on 1R3 the closed 2-form
w2 = xydxAdy+2xdyAdz+2ydxAdz.
We will determine, in two different ways, a 1-form whose derivative coincides
with w2. Let us first explain the Ansatz method to be used. The 1-form 771
will be taken as
771 = f (x, y, z) dx + g(x, y, z) dy + h(x, y, z) dz,
22 2. Differential Forms in IR"
where the functions f, g, h : R3 R still have to be determined. The

exterior derivative is easily computed:
ay] dx A dy + [82 az] dx A dz + I ah - ag] dy A dz.

d'l' [ax y
Hence the functions have to satisfy the conditions
ag
OX
of
ay X Y,
y' az 49Z
Oh - Of
2 y and
anayah - Og
Oz
2x
Integrating, e.g., the first two with respect to x yieldss
g= 2 x2y + ,l dx, h = 2xy + , J dx.

49Z
Inserting the result into the last condition, /
2x = 2x + / Oy dx - J a N dx,
we see that it is satisfied for function f; in particular, we may choose
f = 0. Then g = 2x2y, h = 2xy, and hence, using
21
171
= x2ydy+2xydz,
we obtain a solution, as is easily checked. The integration method computes
the "primitive form" by means of the map P(w2) introduced in the proof of
Poincare's lemma. In the example, this turns out to be the sum of 6 terms,
namely
f 1
P(w2) = + C
o t(tx)(ty)dt/J xdy - (J1 t(tx)(ty)dt) ydx ro
1 l
+2 (1 t(tx)dt J ydz - 2 (J 1 t(tx)dt I zdy
(jI
+2 Uo t(ty)dt I x dz - 2 t(ty)dt I z dx.
Each of the summands can easily be computed:

2- 1 2 I 2 2 2 2
P(w) = 4x ydy- 4xy2dx+ 2xydz - 2xzdy+ 2xydz - 2yzdx
1 2 2 2 2 4
(4xy + 2yz dx + 4x y- 2xz dy + 2xydz.
Which of these two methods leads to a solution more quickly depends on
the particular situation; since an integration has to be carried out in any
case, it might not be possible to find an explicit elementary solution (just.
as not every continuous function is elementary integrable).
2.3. Gradient, Divergence and Curl 23
2.3. Gradient, Divergence and Curl

Each tangent space TpRn of the coordinate space is an oriented, euclidean
vector space, and hence there is the volume form
dRn(p)
E Ap(R")
as well as the Hodge operator
/\p-k(Rn).
A (Rn) ---,
This allows us to associate with every differential form wk of degree k on
Rn a corresponding (n - k)-form *wk defined by applying the *-operator
pointwise, i.e., at each point p E Rn to wk(p). Consider an orthonormal
basis e1, . . . , en of the space Rn and the corresponding coordinate functions
x 1, ... , x". For a k-form wk = F w1dxr expressed in these coordinates the
associated form is thus determined by the following formula:
*W
k sgn l... n wt dxj.
J
Here, J = (jl < ... < jn_k) is the complementary multi-index to I = (i1 <
< ik). The volume form dRn itself is simply
dRn = dx1 A ... A dxn .
In addition, we have the possibility to pass from a vector field V to a 1-form
wv, and vice versa. This is accomplished by
Definition 9. For a vector field V the dual 1 -form wv is defined by the
equation
*wv := VJ dRn .
If the vector field V = E V'8/8x' is expressed in cartesian coordinates, then

the corresponding representation for w11 is obtained as follows:
wV = (-l)n-I *(V1dx2A...Adxn-V2dx1Adx3A...Adxnf...)
= V1dx1+...+Vndxn.
This transition from vector fields to 1-forms will be used now to introduce
the gradient of a function and the divergence and curl of a vector field in an
invariant way.
Definition 10. Let f : U --, R be a C'-function defined on an open subset
U C Rn. The vector field grad(f) associated with the 1-form df is called
the gradient of the function f. The defining equation for the vector field
grad(f) thus reads as follows:
' Wj1r.d(!) = grad(f) J dRn = *df .
In the chosen coordinates we have

grad(f) = aT1 Of 0-+ -Of a
art + ... + cle axn
Definition 11. The divergence of a C'-vector field V is the function deter-
mined by the equation
d(*;.,) = d(V J dRn) := div(V) - dRn .
The formula
n
div(V) =
10
ax=
expresses the divergence of V through its components.
The next theorem contains a few simple properties of this operation.

Theorem 4.
(1) Let f and g be C' -functions. Then
grad(f - 9) = f - grad(9) + 9 - grad(f) -
(2) For a function f and a vector field V of class C' the following
identity holds:
div(f - V) = f div(V) + df (V)
Proof. We only prove the second formula. By definition
div(f V) dRn = d(f (V J dRn)) = df A (V J dRn) + f d(V J dRn)
= df A *Wv + f div(V) dRn = [41(V) + f div(V)J dRn.
The last step makes use of the following equation, valid for every vector field
V and any 1-form n1:
171 A *4 = n1(V) (W.
Definition 12. Let f be a C2-function defined on an open subset U C Rn.
The divergence of the gradient of f is called the Laplacian o(f) of the
function f :
0(f) = div(grad(f)) -
In the chosen coordinates the Laplacian is computed by the formula

02f
o(f) _
An immediate consequence of the identities in the preceding theorem is the
formula
A(f -9) = 0(f)-9+f -A(9)+2(grad(f),grad(9))
2.3. Gradient, Divergence and Curl 25
In particular, there exists an additional operation acting on vector fields in

dimension n = 3, the so-called curl. For a vector field V and the correspond-
ing 1-form wv in R3, the form *d(wy) again is a 1-form and hence in turn
defines a vector field, the curl curl(V). In short, we have the following
Definition 13. The curl of the vector field V is the unique vector field
determined by the following condition:
dwy =: Curl(V) J dR3 = *'''curl(V)
The definition of the exterior derivative &-'y

dwy =
av2 _ avl dx1 ndx2+ Iav3 _ aV 1 dx1 ^d3+ [aV3 _ av21 dX AdX
a21 Ox a21 a23 a22 ax3
leads immediately to the formula
curl(V) = av3 _ av21 a [aV1 _ aV31 a [aV2 _ aV1 ]] a
a22 a23 J a21 + a23 all 5x2 + 5X1 aX2
The properties of the curl of a vector field are summarized in the following
theorem.
Theorem 5. If the function f and the vector field V are of class C2 and
defined on an open subset U C R3, then
(1) div(curl(V)) = 0 and curl(grad(f)) = 0;
(2) curl(f V) = f curl(V) + grad(f) x V, where x denotes the vector
product in R3;
(3) if the curl of V vanishes, curl(V) = 0, and U is star-shaped, then
there exists a function f such that V = grad(f);
(4) if the divergence of V vanishes, div(V) = 0, and U is star-shaped,
then there exists a vector field W such that V = curl(W).
Proof. The first equations immediately follow from the fact that the square
of the exterior derivative vanishes, dd = 0. In fact, we obtain
div(curl(V)) dIR3 = ddwv = 0
and
wcurl(grad(f)) _ *dwgrad(f) _ *ddf = 0
Claims (3) and (4) are special cases of Poincare's lemma. For curl(V) = 0 the
associated form wv is closed. By Poincare's lemma there exists a function f
such that wv = df. Hence V is the gradient of the function f. Property (4)
is proved similarly.
26 2. Differential Forms in It"
2.4. Singular Cubes and Chains

We intend to develop the integral calculus for differential forms, and to do
so, we need suitable sets as integration domains. First, in this chapter, we
will only allow for subsets of W' which are higher-dimensional analogues of
parametrized curves, so-called singular chains. The k-dimensional unit cube
will be denoted by [0, 1]k,
[0, 1]k = {(x'....,xk)ERk: 05 x1,...,xk<1).

Definition 14. A singular k-cube in U C R" is understood to be a C'-map
ck : [0,1]k -+ U.
Once and for all we agree to call a map defined on any (not necessarily open)
subset of Rk smooth if it has a smooth extension to an open neighborhood
of this subset.
Picture of the 2-cube [0, 1]2 B (x, y)' -* cos(x + 1)2 + sin(y - 3)2
Note that not only does a singular cube consist of the image set of ck, but
the notion also involves an explicit parametrization. For example, in case
k = 0 a singular 0-cube is a point in the set U. For k = 1, in general, it is
a parametrized curve, but it may also degenerate to a point, which explains
why these cubes are called "singular". In some situations, it can be helpful
to view a singular k-cube as a cube of some dimension larger than k. Notice
also that the boundary of the image set will, in general, have "comers".
2.4. Singular Cubes and Chains 27
Definition 15. A singular k-chain in U is a formal sum of singular k-cubes

cik in U with integer coefficients Ii E Z.
l1ci +... +lmCkm =: 3k.
k-chains are added in the obvious way, and with addition as composition
they form the abelian group of singular chains in U, which will be denoted
by Ck(U). For example, the inverse of the k-chain sk is its negative -sk.
Definition 16. The standard cube in 1Rk is defined to be the identity map
of the k-dimensional unit cube
Ik : [0, 1]k -+ Rk, Ik(x) = X.
Now we define a boundary for every k-chain. For each index i between
1 and k, we parametrize a part of the boundary of [0, 1]k by maps I(i,o)'
Ik.l) : [0, 1]k-1 -+ [0, 1]k which insert the value 0 or 1, respectively, at
position is
Ik
(io)[0,1]
k-1 k
flR, riio)(x,...,x
k k-1)-(x,...,x
1 - i-1 ,o,x,...,x
1 i k-1
I(i.l) : [0 1]k-1 Rk I(t 1)(x1 k-1) = (XI, i-1 i k-1
The boundary of the standard cube Ik is then taken to be the (k - 1)-chain

k
alk (I(i 0) - I(i,1) 1 .
i=1
signs
-1
-1 +1
+1
For an arbitrary singular k-cube ck : [0, 1]k - U C 1R we set

k
aCk
E(-1)i (ck o l(i 0) - Ck o l(i l) I
i=1
Finally, for a singular k-chain sk = l;cf we define the boundary operator

by
19Sk l;aC;
Example 8. We compute the boundary of the 2-dimensional standard cube

12. To this end, we label the 4 vertices by pi to p4 starting at the origin
and moving counterclockwise, and the 4 edges by sl to $4, again moving
counterclockwise and starting at pl. Then
012 = S1 + 82 + sg + s4,
and the boundary of this chain is
8812 = (p2 - pl) + (Ps - p2) + (p4 - Ps) + (pl - p4) = 0.
As in the case of the exterior derivative d, the square of the boundary

operator vanishes.
Theorem 6.
(1) The boundary operator a : Ck(U) -- Ck_1(U) is a group homomor-
phism;
(2) in the sequence
... 8 Ck+1(U) -' Ck(U) a Ck-1(U) - ... 8 . C1(U) a CO(U)

the equality 08 = 0 holds, i. e., for every k-chain sk E Ck(U) we
have
a(a sk) = 0.
Proof. The property 8(si + 82) = 8s1 + 8s2 immediately follows from the
definition of the boundary operator. Because of linearity, this obviously
suffices to prove the second statement for an arbitrary k-cube ck : [0, 1]k
U C R". By definition the boundary is determined by
k
\
ack = (-1)j
;_1
(ck o IV.o) ck o Ik 1i 1
2.4. Singular Cubes and Chains 29
Applying the boundary operator again, we obtain (omitting the composition

signs in the second line)
k
/
a(ack) = >(l) j 1 a(ck o I(j,o)) - a(ck o Ik 1))/l
j=1
k k-I
=E j:(-1)i+j [ckI(j
o)I(+.oj - C'(j,o)I(i 1) - C'(j,1)I(i:)
j=1 i=1 L
In order to be able to transform this expression, we introduce two additional

summation indices a,,3 = 0, 1, and rewrite it as
k k-1
(-1)i+j+a+0 l(ck o I(j.a) o
a(ack) _
j=1 i=1 a=o,1
0=o,1
For j < i and a,,3 = 0,1 we have

k
1U-) k-1 (x 1 , ... , x k-2 )
01 (i.li) k
= I(j,a) (x , ... , xi-1 , fl, xi , ... , x k-2 )
1
= (x1, ... , xj-1,

Similarly, one rewrites I(i+1 p) o 1(j,a) as
1(i+1.0) ° IU*(x1, ... , xk-2)

= I(i+1,0)(x1, ... , xj-1, a, x1, ... , xk-2)
1 1 i-1 i k-2
Together these yield the identity
I(j,a) ° I(i,8) = I( i+1,0) ° I(j,

)
In order to apply this, we split the sum appearing in the expression fo
O(Ock) into those terms for which j < i and the remaining ones:
a(ack) = Lam(-1)i+j+a+Qck o Ik o Ik-1
L.r
1<j<i<k-1 a.0
+ E E(-1)i+j+a+1Ck 0 I(j,a) 0 I(i,Q)
i<j a,0
In the second sum we replace i by j as well as j by i + 1:
a(8Ck) _ (-1)i+j+a+0ck o ik(,a) jk-1
1<j<i<k-1 a,(3
+ E(-1)i+j+l+a+Ock o j(i+1.a)
° I(j,
1<j<i+1 a,6 3)
= 0,
by the identity proved before. 0
In a similar way as the de Rham cohomology, we can define the k-th cubic
homology group of a set U C Rn:
Hk°''(U) := ker (8 : Ck(U) --' Ck-1(U)) / im (0: Ck+,(U) ~ Ck(U))
2.5. Integration of Differential Forms and Stokes' Theorem

Consider a singular k-cube ck : [0, 1]k _ U C R" of class C', as well as a
k-fonn wk on U. Then the induced differential form (ck)'wk is defined on
the unit cube [0,1]k, and, as such a multiple of dxl A A dx"
(ck)`wk = f (x) dxl A ... A dx'

for some function f : [0, 1]k - R. We can thus define the integral of wk over
the singular cube ck as follows:
Definition 17. Set
ck
wk
[01Jk
f(x)
and extend this definition linearly to any k-chain sk = >, lj in U by
rwk 1kWk
Example 9. For k = 1 a singular k-cube is simply a parametrized Cl-

curve c : (0, 1] -i U C R'. In this case, the integral of the 1-form wl =
p1dyl +... + p"dy" is called the line integnzl of wl along c. If cl, ..., c" are
the component functions of c, the pullback of the form is written as
c'w'(t) = pl(c(t)).
d dtt) dt + ... + pn(c(t)) - dcn

dtt) dt,
and hence, in this situation, we obtain the following general formula for the
line integral:
dctt)1
w' = 1 pi(c(t)). dt.
J J
If the 1-form wl is the differential of a smooth function f, then by
c*(wl) = c*(df) = d(f o c)
we obtain the following value for the integral over df :
J
df = jdt)dt =
In particular, the line integral of an exact 1-form depends only on the end
points of the curve and not on its shape. It vanishes for a closed curve.
2.5. Integration of Differential Forms and Stokes' Theorem 31
Example 10. We will use the last remark to show that the winding form,
cal = x2 +Y22 dx + x2 x y22 dy,

is not exact on R2 - {0}. To this end, consider the parametrization c(t) _
(cos 21rt, sin 21rt) of the circle by the interval [0.1]. Then
c'wl = 27r dt and jw1 = 27r.
The integral of a k-form over a singular k-cube is, up to sign. independent

of the parametrization of the cube. This is a consequence of a well-known
transformation rule for higher-dimensional integrals: Let y, : U -' V be a
diffeomorphism, and let f : V -+ R be integrable; /then
f(f o )(y) . I det(D(y))I dy

JV
f (x)dx .
Two parametrizations and 4 : [0, 1]A U of one and the same point
set in U differ by a diffeomorphisin cp : [0,11k [0, 1]k of the unit cube.
4 = ci o gyp. The determinant of the differential Dp(x) has constant sign,
which will be denoted by e(,p).
Theorem 7. In the above situation we have
Jk = f wk
cz"
Proof. By Theorem 1 the pullback of the form

F*(ci)"(wA)
( i)"wk = = cp`(f dxlA...Adx") = (foip)'4p'(dx'A...ndx")
coincides with
(c2)*wk = (f o cp) dxl A ... A dx".
Thus the statement follows from the quoted transformation rule for n-dimen-
sional integrals. 0
In case k = 1, changing the direction of a curve results in a change of sign
for the line integral.
Now we prove the essential result of this chapter, Stokes' theorem. This is
a far-reaching generalization of the Fundamental Theorem of Calculus and
comprises the classical integral formulas of the 19-th century as special cases.
The proof to follow will, however, reveal to the careful reader that the core
of the matter is the fact that integration and differentiation are mutually
inverse operations.
32 2. Differential Forms in 1R"
Theorem 8 (Stokes' Theorem). Let wk be a differential form defined on the

open subset U C IR", and let sk+l : [0, 1]k+l - U be a (k + 1)-chain. Then
wk = r dk
f.9$k+l Jgk+1
Proof. Since the integral is additive, it suffices to prove the formula for
singular (k+1)-cubes. First we consider the standard cubelk+l : [0, 1]k+l
R k+1 and represent the k-form wk on Rk+l as
k+l _
wl~ = fi d21 / ... / dxi A ... A dxk+1
i=1
The derivative is then
k+l
k = [(_1)1_1L]i dxl A ... A dxk+1
i=1
8xi
Hence, by the definition of the integral,
f k+1
dwk xi dx
J a
jk+1 i-1 [0,11k
On the other hand, applying the maps IV Ql parametrizing the different

parts of the boundary of the unit cube leads to the formula
( Ik+l
J.a ) >
*
wk-- f,(x , ..., X j
1 1
, a, Xj. . . ., xk) dx1 A . . . A dsk
from which we conclude that
J
(Ii I/ Ù)_ f ` J
k
[0.1[k 10111k
J J afi (xl, ...,

r fl
J-1, t, xJ, ... , xk)dt dxl ... dxk
dxjo
[o.1[k!
adx.
dxi
[O,1Jk+l
By the definition of the boundary of Ik+1, this implies that
ajk+l
Jk w= rf
kr+l
L L.
J-1 Q-o'1Ik+1
kr+1
j= l
f dfx)'dx =
fk
lk+1
[o.llk+l
li.Q)
and hence we have verified Stokes' formula for the standard cube. For an
arbitrary singular (k + 1)-cube ck+1 : [0, 1]k+1 U C 1R" we now use the
2.5. Integration of Differential Forms and Stokes' Theorem 33
fact that the exterior derivative commutes with the pullback of forms. We
have
dWk = f (ck+ll*( k) d((Ck+l)'Wk)
J
Ck+1 Jk+1 l J jkJ+1
l` J J
k+1
__ k+1 wk _ f k+1 ` Wk
(c
ajk+1
j=1 Q=0.1 wk
+1
.)
k+1
Jk.
wk =
j=1 a=0,1Ck+lolk+1 ask+l
U.u)
We already emphasized that the line integral of an exact 1-form is indepen-

dent of the particular shape of the curve. As a first application of Stokes'
theorem, we will prove that the line integrals of a closed 1-form along two
different curves coincide if there exists a continuously differentiable defor-
mation of one curve into the other leaving their initial and end points fixed.
This leads to the notion of homotopy, which is fundamental in topology.
Definition 18. Two Cl-curves co, cl : [0,11 U C R" are called homo-
topic if there exists a C1-map F : [0,1] x [0,11 - U with the properties
F(t, 0) = co(t), F(0, s) = co(0) = c(0).
F(t,l) = cl(t), F(1,s) = co(1) = cl(1).
The map F is called a homotopy between the curves co and cl.
Theorem 9. Let co, cl : [0,1] - U be two homotopic C1-curves, and let w1

be a closed 1 -form on U. Then the line integrals along co and cl coincide:
Lw1 = jwl
l
Proof. Choose any homotopy F : [0,1]2 -> U between co and c1. This map
F is, at the same time, a singular 2-cube, and hence the 2-form dw1 can be
34 2. Differential Forms in IR
integrated over F. By assumption we have dwl = 0, and the corresponding

integral vanishes:
dwl.
0=IF
On the other hand, by Stokes' theorem, we have for the right-hand side
r
JF
dw1 fJ w1 + J
F (t,O) (1,s)
_ J (t,I) _ I O,s)(
These four integrals are computed using the homotopy property of F. First,
F(t, a) = ca(t) immediately implies, for a = 0,1,
FlZt,a)w1 = cowl.
Moreover, F(a, s) is independent of s, and hence
Fl*(a,,)wl = 0.
Lastly, these relations combine to yield the equation
1 1
0= cowl - J0 cwl ,
J0
and by the definition of the integral this concludes the proof. 0
We will now generalize these observations concerning line integrals to the
higher-dimensional case. Let sk lj ch- be a singular k-chain with
image set A C IR", and suppose that Oak = 0. The set A can, e.g., be a k--
dimensional sphere Sk in R". Consider a smooth map f : A A defined on
a neighborhood of A which is homotopic to the identity IdA. This is again
supposed to mean that there is a smooth map defined on a neighborhood U
of A x (0.11 in R"+1
F : A x (0,11 -+ A such that F(a, 0) = f (a) and F(a,1) = a.
Then
Fo(skxIdlo,11) :=E
is a singular (k + 1)-chain in R", and, because Oak = 0, the boundary is
OF o (sk x Id10,11) = sk - f o sk.
For a k-form wk defined on an open neighborhood of the set A C W', the
following holds.
Theorem 10. If Osk = 0 anldf is homotopic to the identity, then
I
Jsk
k=J jock
wk
2.6. The Classical Formulas of Green and Stokes 35
Proof. We compute the difference using Stokes' theorem:

f Wk - k= fk = f k
J
sk fosk OFo(skxId1o,11) Fo(skxld[O,j])
The (k + 1)-form (sk x Id[o,I))`F*(dwk) vanishes. This follows from the

implicit function theorem together with the assumption that A is the image
of a k-dimensional chain. But this immediately implies the statement. 0
2.6. The Classical Formulas of Green and Stokes

In this section, we will discuss the classical two-dimensional special cases of
Stokes' formula. Let D C R2 be a subset of R2 which can be represented
as the image of a CI-map f defined on the standard cube [0, 112. By OD
we denote the boundary of this set, considered as a singular 1-chain. The
derivative of the 1-form w1 := (x dy - y dx)/2 is the volume form on R2.
Hence
vol(D) fdwl
2 aD
The formula transforms the calculation of a two-dimensional volume into
the evaluation of a line integral, and this turns out to be a special case of
Green's formula to be discussed now. Consider the 1-form
wI
and compute its derivative:
dw I = I - - J
dx A dy.
Thus we arrive at Green's first formula.

Theorem 11. Let P(x, y) and Q(x, y) be functions of class CI. Then
IOQ-8PJdxAdy.
I D fD L Ox ay
Suppose now that P(x, y) and Q(x, y) are of class C2, and consider the
1-forms
wl P [-LQ dx+ LQ dy] and qI := Q I-5y . dx + 5P dy]

The derivative of the difference wI - 771 is easily computed to be
d(w'-1 ) =
Applying Stokes' formula leads to Green's second formula.
36 2. Differential Forms in W'
Theorem 12. Let P(x, y) and Q(x, y) be functions of class C2. Then
[PO(Q) - QO(P)]dx n dy =f [QLP - P l dx + f P Q - Q

J L
Pl dy.
1
8D
Stokes' formula concerning certain surface integrals is, in a similarly simple

way, a special case of the above general integral formula. In fact, let F C R3
be a subset of IIt3, which can be represented as the image of a Cl-map f
defined on the standard cube [0,1]2 (a surface piece). By 8F we denote
the boundary of this set considered as a singular 1-chain in JR3. Consider a
vector field V defined on an open neighborhood of F. Its curl is defined by
the following condition:
dwy := curl(V) J dllt3 = *wcurl(V)
Integrating the 2-form sweurj(V) over the surface piece F, we obtain Stoke
theorem in its classical form.
Theorem 13. Let V be a smooth vector field on a neighborhood of F. Then
(V1dx1+V2dx2+V3dx3) = / &,;y =
J
OF F
8V2 8V11 8V11 IOV3 8V21
dx Adx 2+IIr8V3
1
dx ndx3+ axe - 'ax-3 dx2nda3.
axl - axe I axl - 5x3 I
F
Remark. Using the volume form dF of a regular surface piece yet to be
discussed, the classical Stokes formula can be written more concisely as
Lcun1'1) dF.
Here N denotes the normal vector to the surface in R3. We will return to
this in Chapter 3.
2.7. Complex Differential Forms and Holomorphic Functions

The complexification of the vector space of all real-valued forms is called the
space of complex-valued forms on an open subset of R". Such a form can be
split into its real and imaginary parts;
wk = wo + i Wk
and differentiation as well as integration are defined with respect to this
decomposition:
d,jk dwo + i dwi, LkjkOLkI .

2.7. Complex Differential Forms and Holomorphic Functions 37
In a similar way we extend the exterior product to complex-valued forms:

W k A 771 := (WO A 170 - W I A 911) + i (,oo A 7l1 + W I A 770) .
Then the previous computational rules and Stokes' theorem still hold. Now
we want to apply complex-valued forms to study holomorphic functions and,
to do so, first identify the real vector space R2 with the complex numbers
C. From z := x + i y and z := x - i y we obtain the differential forms
dz := dz := and dzndz =
Let f be a complex-valued function with real and imaginary parts u and v of
class C1, f (z) = u + i v. Denote by u=, uy, vv, vy the partial derivatives with
respect to the corresponding variables. Then f (z) dz is a complex-valued
differential form. We compute its differential:
Now let f : U --+ C be a complex-differentiable function defined on an open

subset of C. Elementary complex analysis starts by proving that its real and
imaginary, parts are smooth functions in the sense of real analysis (Goursat's
theorem). Furthermore, the Cauchy-Riemann equations hold:
ux=vy and uy= -vx.
Theorem 14. If f (z) is a complex-differentiable function, then the 1-form
f (z) dz is closed,
d(f (z) dz) = 0.
An immediate consequence is Cauchy's theorem.
Theorem 15. Let U be an open subset of C, and let -y be a closed curve in
U, which is the boundary of a singular 2-cube. Then the integral
vanishes for each complex-differentiable function f .
In a similar way we derive Cauchy's integral formula. To do so, we assume

that f (z) is a complex-differentiable function on a neighborhood of the disc
K(zo,M) = {zEC:IIz-zoII<-M}.
Fix 0 < e < M. The function f (z)/(z - zo) is complex-differentiable in
K(zo, M)\K(zo, e), and, since d((f (z) dz)/(z - zo)) = 0, we have
f (z) dz = / f (z) dz f - f (z) dz
f.9K(zo,Al) Z - ZO
JaKi:o,E> Z - zo
Jax(zo,e) (z - zo) Ilz - zol12
= fK(zo (z - zo)f(z) dz.

e2 0
We compute the derivative of the 1-form (z - zo) f (z) dz:

2i -
Thus
= 2i.
f(x) dandy.
J K(zo,M) Z - ZO E2 K(zo.e)
The mean-value theorem of integral calculus states that there is a number
ze in K(zo, e) for which
f (z) dx n dy = f (ze)vol(K(zo, e)) = 7r `2 f (ze)
K(zo.e)
If e tends to zero, then f (ze) converges to f (zo), and we arrive at Cauchy's
integral formula:
Theorem 16.
1
f(zo) = 27ri JaK(zO,jvf) z - zo
This formula is fundamental for the theory of functions. It implies, e.g.,

that every function which is complex-differentiable in the neighborhood of
a point can be expanded into a power series (i.e. is an analytic function).
2.8. Brouwer's Fixed Point Theorem

A fixed point of a map f : X --+ X from a set to itself is defined to be a
point xo which is not moved by f, f (xo) = xo. In topology, several fixed
point theorems are known. They state that certain continuous maps from a
metric space to itself necessarily have at least one fixed point. If, e.g., X is
a complete metric space, and f : X -p X is a contracting map, then by the
Banach fixed point theorem the map f has at least one fixed point. There
are, of course, (non-contracting) maps from a complete metric space to itself
without fixed points; translations in R" are examples for this. A topological
space X is said to have the fixed point property if every continuous map
f : X - X from X to itself has a fixed point. This is obviously a topological
property, i. e., homeomorphic spaces either all have the fixed point property
or none of them has it. The unit circle X = S' is compact; nevertheless it
does not have the fixed point property; rotations are continuous maps from
the unit circle to itself without stationary points. Brouwer's fixed point
theorem states that the closed n-dimensional ball of radius R,
D"(R) = {xER": IIxii <-R},
is a metric space having the fixed point property. We will first prove this
theorem in the case of smooth maps from the ball to itself, and afterwards
extend the proof to continuous maps by means of an approximation argu-
ment.
2.8. Brouwer's Fixed Point Theorem 39
Theorem 17. Every Cl-map f : D" -+ D" from the n-dimensional ball to
itself has at least one fired point.
Proof. Suppose that f : D" - D" does not have any fixed points. Then
define the map F : D" Sn-' from the ball to its boundary by assigning
to every point x E D" the point of intersection of the ray from f (x) through
x with the sphere Sn-'. The formula for F,
F(x) = x
X - f(x)
2 (X' x - f(x)
_ IIxI12 + (X, x - f(x)
IIx - f(x)II 1- IIx - f(x)II> IIx - f(x)II
shows that F is smooth. Moreover. F acts on the boundary of the ball as
the identity, F(x) = x for all x E S. Let Fl.... , F" be the components
of F. Differentiating the following relation, which is valid for all x E Dn.
n
E(F'(x))2 = 1,
i=1
yields
2 (Fiaa*5) Idxi = 0,
i=1 ij=1
and hence for each index j
j:Fi(x)8F11'(x)
11T1
= 0.
i=1
Therefore, the system of equations
8P (x) = 0
Ox)
i=o
has a non-trivial solution (al, ... , an) = (Fl (x), ..., F"(x)) j4 (0.... , 0).
Hence the determinant of the matrix
det 0
vanishes. Now we apply this observationto the differential form

Wn-' = F1 A dF2 n ... A dF"
and conclude that its differential vanishes:
i \
&,n-1 = dFlAdF2A...AdF" = det(aa2 )dx'A...Adx" = 0.
40 2. Differential Forms in 1t"
By Stokes' theorem the integral of the form w' 1

over the boundary Sn-1
of the singular cube D" is equal to zero:
0= do-1 = n-1
- JD- jn-1
S"-1
On the other hand, F acts on the sphere as the identity; hence
wn-1ISn-I = x1 dx2 A ... A dxnlSn-i .
This implies
r
0=J x1 dx2 A ... A din = J dxl A ... A dx" = vol(D"),
n -I Dn
and we arrive at a contradiction. 0
Theorem 18 (Brouwer's Theorem). Every continuous map f : D" Dn
from the n-dimensional ball to itself has at least one fixed point.
Proof. The proof will be reduced to the case of a C1-map by applying the
Stone-Weierstrass approximation theorem (see, e.g., [Rudin, 19981, Theorem
7.32). The ball D" is compact. Consider the ring C°(D") of all real-valued
continuous functions on it, as well as the subring R of those functions which
are the restriction to D" of a C1-function with strictly larger, open domain
of definition. Obviously, the subring R contains the constant functions and
separates points. By the Stone-Weierstrass theorem, it is dense in C°(U).
Applying this to the components f 1, ... , f" of f, we conclude that for each
e > 0 there exists a C1-map
p : D' - R" such that IIf (x) - p(x)II < e for all x E D'.
Consider the renormalized map P(x) := p(x)/(1 + s). Since
IIP(x)II - IIf (x)II 5 IIP(x) - f (x)II < e and IIf (x)II < 1,
we have IIP(x)II 5 1 + e; hence IIP(x)II 5 1. Therefore, P is a map from the
ball to itself. Moreover, P can be estimated against f :
IIf (x) - P(x)II 5 IIf (x) - P(x)II + IIP(x) - P(x)II 5 e + IIP(x)II 11 - 1+

< e+(1+e)1+e < 2e.
Summarizing, we have proved that for each e > 0 there exists a C1-map
P : D" D" satisfying for every x E D" the estimate
IIf(x)-P(x)II <2e.
Now if the continuous map f : D" -+ D" had no fixed point, the number
xinf IIf (x) - xI I
2.8. Brouwer's Fixed Point Theorem 41
would be strictly positive. Choose for F = µ/2 a smooth approximation p

with the properties stated above; by Theorem 17 this map then has a fixed
point xo E Dn, for which in turn
IIf(xo) - Axo)II = Ilf(xo) - xoII < i
would have to hold. But this contradicts the definition of µ.
Brouwer's fixed point theorem can be viewed as an existence result for real,
non-linear systems of equations. We state a possible application, which plays
an important role in Galerkin's method.
Theorem 19. Let g1:... , gn : D' (R) -+ R be continuous functions de-
fined on the ball Dn(R) of radius R, and suppose that for all points x =
(x1, , xn) E S'-'(R) in the sphere Sn-1(R) the following inequality holds:
n
9i(x).xi
> 0.
i=1
Then the system of equations
91(x) = 92(X) _ ... = 9n (x) = 0

has at least one solution in D'(R).
Proof. We combine the functions to define a map g : Dn(R) --+ Rn, g(x) :=
(91(x), .. , 9n(x)). If g(x) 0 0 holds for all points x E Dn(R), we can
consider the map f : Dn(R) -p Sn-1(R),
f (x) := -R g(x)
119(x)11
whose image lies in the sphere Sn-1(R). By the fixed point theorem, f has
a fixed point in S"-'(R). Hence there exists a point xo E Sn-1(R) such
that
xo = -R 9(xo)
119(xo)11
This implies R2 II9(xo)II = -R (g(xo), xo), contradicting the assumption of
the theorem.
In particular, the assumption of the theorem is satisfied for gi(x) = xi +
hi(x) if the functions hi : R' -+ R grow more slowly than linear forms,
Ihi(x)I < Ci 11x11"
Under this condition the sum
n n
E 9i(x)xi = R2 + c` hi(x)x'
i=1 i=1
behaves like R2 on the sphere S"-'(R) and becomes positive for sufficiently
large radii.
Corollary 1. The system of equations
hl(x) = x', ..., h"(x) = x"
has at least one solution f o r arbitrary continuous functions bounded by I I I'
at infinity.
Example 11. The system of equations x = ' 1 + x2 + y'2. y = cos(x + y)
has in l 2 at least one solution, e.g.,
x = 1.2758079, y = 0.14722564.
Example 12. Consider in 1R2 the following system of equations:
= 0, 92(x, y) = y + e-(I-U)2 = 0.
e-(s+y)2
91(x, y) = x +
The picture below diaplays the graph of the function g, (x, y) x + 92(x. y) y
over the set [-2.2] x [-2,2]. It shows that this function is positive on the
circle S' (2) of radius 2. Hence the above system of equations has at least
one solution in the disc D2(2). A numerical computation of the solution
leads to the values
x = -0.303122, y = -0.789407.
Exercises 43
Exercises
1. Let f : 1R2 - {0} R2 - {0}, f (r, 0) = (r cos 0, r sin 0), be the polar
coordinate map on the "punctured" plane. Prove:
a) The winding form satisfies f dx) = d9;
b) the radial form x dx + y dy satisfies f ` (x dx + y dy) = r dr.
2. Consider on R2 - {0} the winding form wl = 114.-1., as well as the

following family of curves depending on the integer parameter n E Z:
c: [0, 11 - . R2 - {0}, (cos2ant,sin2nrnt).
Compute the line integral of the winding form along the curve c,,, and con-
clude that the curves c,l are not homotopic in R2 - {O} for different values
of the parameter n.
3. Compute the exterior derivative of the following differential forms:

a) xydxAdy+2xdyAdz+2ydxAdz;
b) z2dxAdy+(z2+2y)dxAdz;
c) 13 x dx + y2dy + xyz dz;
d) e' cos(y) dx - e' sin(y) dy;
e) xdyAdz+ydxAdz+zdxAdy.
4. Consider 1R2n with coordinates x1, ... , x2n and the following differential
form of degree 2:
w2 = dx' n dxn+l + dx2 A dxn+2 + ... + dxn n dx2n .
Prove:
a) The form w2 is closed;
b) the n-th exterior power of w2 is related to the volume form via the
formula
w2 n ... n w2 = (-1)n(n-1)/2n! dxl A ... A dx2n .
5. The subject of this exercise is to explain why it makes sense to denote

the vector fields x -+ (x, e;), defined using the standard basis e1, . . . , e,, of
R", by a/8x'. Writing these vector fields for the moment as E;, every vector
field can be written in the form
n
V=
For each function f : U R on an open set U of R" we define a new
function. the derivative of f in the direction of the vector field V, by
(V(f))(x) := (Df=)(V(x))
Prove the formula
n
V(f) _i=108xi (fl
which by omitting the argument f provides the explanation asked for.
6. Prove the following rules for vector fields on )1t3:

a) div(V1 x V2) _ (curl(Vl),V2) - (VI,curl(V2));
b) curl(curl(V)) = grad(div(V)) - 0(V), where the Laplacian is to be ap-
plied componentwise to V.
7. Compute the line integral
JC2
(x- 2xy)dx + (y2 - 2xy)dy
along the curve C = {(x, y) E R2 : x E [-1.1), y = x2}.
8. Compute the line-integral

sin(y)dx + sin(x)dy,
IC
where C is the segment joining the points (0, x) and (n. 0).
9. Consider on R3 the differential form w2 = y dxAdy. Determine all 1-forms

17 1 = p dx + q dy satisfying dal = w2.
10. Prove the following converse to one of the statements of Example 9: If

wl is a 1-form defined on the open set U, and if the line integral of wl is
independent of the curve, then wl is exact.
Hint. Prove this by explicitly constructing a "primitive function". In
physics, the vector field V corresponding to the 1-form wy is called con-
servative, if it does no work along any closed curve -y. 11 wy = 0. Thus, V is
conservative if and only if w1, is exact.
Exercises 45
11. Consider the singular 2-cube, f : [0, 2ir] x [0, 27r] -r S2 C R3 - {0},
f (u, v) _ (cos u sin v, sin u sin v, cos v),
as well as the 2-form w2 = (xdyAdz+ydz Adx+zdxndy)/r3 on 1R3 - {0},
where r = (x2 + y2 + z2)1/2 denotes the distance from the origin.
a) Prove that w2 is closed;
b) compute the integral of w2 over f ;
c) conclude from the properties just proved that w2 is not exact on 1R3-{0},
and that there is no singular 3-chain c3 in 1R3 - {0} whose boundary
equals f, 0c3 = f.
12. Prove that the integral defines a unique bilinear map

HDR(U) X Hkub(U) R, ([WI], [s']) -- Jgk
Wk.
13. Let w1 = f (x) dx be a 1-form on the interval [0, 1] with f (0) = f (1).
Prove that there exist a real number p and a function g with g(0) = g(1) by
means of which wl can be written as
W = pdx+dg.
1
14 (Continuation of 13). Let 771 be a closed 1-form on R2 - {0} and wl

the winding form. Prove that there exist a number p as well as a function
g : 1R2 - {0} -+ R for which
77 1 = pwl +dg.
Consequently, the winding form is the generating element of the first de
Rham cohomology of R2 - {0}.
Hint. Consider the polar coordinate map f from Exercise 1 and its pullback
f `wl. This can be written as f `wl = A(r, O)dr + B(r, B)dO; here B(r, O)dO
is a 1-form on [0, 27r] (depending on the parameter r) to which the previous
exercise applies.
15. Consider on 1R3 the following exact differential form known from 7:
W2 = xydxAdy+2xdyAdz+2ydxAdz,
and the upper half-ball A C S2:
A = {(x,y,z)EIR3: x2+y2+z2=1, z>0}.
Prove that the integral of w2 over A vanishes.
46 2. Differential Forms in R°
16. Let C be the circle in R2 with the equation x2 + (y - 1)2 = I in its

standard parametrization. Compute the line integral
f xy2 dy - yx2 dx
a) directly;
b) using Green's formula.
17. Let E be the ellipse with the equation x2/a2 + y2/b2 = 1 (a > 0, b > 0)
in its standard parametrization. Compute by means of Cauchy's integral
formula the integral
rJE dz
z
and obtain from this the value of the integral

12
dt
Jp a2 cos2(t) + b2 sin2(t)
18. Let 7-1 be an infinite-dimensional Hilbert space, and D = {x E l

I IxI I < 11 its unit ball. Does D have the fixed point property?
19. A subset A C X of a metric space X is called a retract of X, if there is

a continuous map r : X --. A such that r(a) = a for all points a E A. Prove
that if X has the fixed point property, then so does every retract A of X.
20. The set A = { (x, y) E [-1,1]2: xy = 0 } has the fixed point property.
Chapter 3
Vector Analysis on
Manifolds
3.1. Submanifolds of Rn
In Chapter 2 we introduced an integration method for differential forms over
sets which can be represented as images of singular chains. These sets, how-
ever, may be quite irregular, and it is rather difficult to develop a differential
calculus for functions defined on them. Further notions like tangent space,
vector field, etc., are not available either in their context. Hence we will now
restrict the possible subsets of Rn to a class for which a differential as well
as an integral calculus can be established in a satisfactory manner. These
sets are called manifolds, and they are-intuitively speaking-characterized
by the fact that their points can be defined locally in a continuous (differen-
tiable) way by finitely many real parameters, that is, locally these sets look
just like euclidean space. It was the fundamental idea of B. Riemann in his
Habilitationsvortrag (1854) to introduce the notion of a manifold as the new
basic concept of space into geometry. In physics, manifolds occur as config-
uration and phase spaces of particle systems as well as in field theory. The
precise description of what a submanifold of euclidean space is supposed to
be is the content of the following definition.
Definition 1. A subset M of Rn is called a k-dimensional submanifold

without boundary if, for each point x E M, there exist open sets x E U C R"
and V C R' as well as a diffeomorphism h : U V such that the image
h(U n M) is contained in the subspace IRk C R :
h(UnM) = Vn(IRkx{O}) = {yEV: yk+1- =yn=0}.

47
48 3. Vector Analysis on Manifolds
The set U* := UnM together with the map h' := hlu.: U* -- V' := VnRk
is called a chart around the point x of the manifold. The sets U* and V'
are open subsets of M and Rk, respectively (see next page).
V
Rn
v n (Rk x {o})
h-1
A family of charts covering the manifold M is called an atlas. The no-

tion "diffeomorphism" can be understood in the sense of an arbitrary C'-
regularity (1 > 1). Correspondingly, we have manifolds of regularity class
CL. For simplicity, we suppose in this chapter that all maps, manifolds, etc.,
are of class CO°, and for this reason we will simply talk about smooth maps,
manifolds, etc. Note that, without any change, all the statements also hold
assuming only C2-regularity.
For any two charts (h', U') and (hi, Ur) of the manifold for which the in-
tersection u* n u; is not empty, one can ask how they are related. The
map
h'o(hi)-1:
is called the chart transition from one chart to the other. The sets h1(U` n
Ul) and h' (U' n Ul) are open subsets of the coordinate space Rk, and the
transition function h' o (hi)-1 is obviously a diffeomorphism.
The notion of dimension also needs to be explained for manifolds. If a sub-

set of Rk is mapped homeomorphically onto an open subset of the space R,
the dimensions of both coincide, k = 1. Under the additional assumption of
3.1. Submanifolds of 1R 49
differentiability, which will always be made here, this is easy to prove: the
differential of a diffeomorphism at an arbitrary point is a linear isomorphism
between the tangent spaces T1IRk and TyR', and this immediately implies
k = 1. The corresponding fact for homeomorphisms is a deep topological
result going back to Brouwer (1910). In any case, the number k occurring
in the definition of a manifold is uniquely determined and will be called
the dimension of the manifold. Sometimes we will write the dimension of a
manifold as an upper index, i.e., denote the manifold also by Mk.
In the first theorem we will prove that, under certain conditions, subsets of
1R' defined by equations are submanifolds. This will give rise to plenty of
examples.
Theorem 1. Let U C lRn be an open subset, and let f : U - Rn-k be a
smooth map. Consider the set
M = {x E U : f (x) = 0} .
If the differential D f (x) has maximal rank (n - k) at each point x E M of
the set M, then M is a smooth, k-dimensional submanifold of 1R' without
boundary.
Proof. The proof is based on a straightforward application of the implicit

function theorem. If xo E M is a point from M, then there exist an open
neighborhood xo E Uxo C U and a diffeomorphism hxo : Uxo -' hxo(Uxo) C
Rn such that the map f o hzp : hxo (Uxo) - Rn-k is given by the formula
f oh=o (x1, ...,xn) = (xk+1, ...,x').
This implies Uxo n M = h=o (hxo (Uxa) n ]R' ), and hence the chart around the
point xo E M we asked for is constructed. 0
Example 1. Every open subset U C Rn is an n-dimensional manifold with-
out boundary.
Example 2. The sphere Sn = {x E Rn+1 : IIxUI = 1} is an n-dimensional
manifold. To see this, we consider the function f : Rn+1 -, R defined by
f (X) = I Ix1I2 - 1. Then we have D f (x) = (2x', ..., 2xn+1), and the rank of
the (1 x n)-matrix D f (x) on Sn is equal to 1. Theorem 1 implies that Sn is
an n-dimensional manifold.
Example 3. Consider a smooth map f : Rn ]Rm and its graph
G(f) = {(x, f (x)) : x E Rn} C 1Rn+m
G(f) is the zero set of the map 1 : IRn+m = lRn x 1R- 1Rm defined by
O (x, y) = f (x) - y. The differential D4 has maximal rank equal to m at
each point. Therefore, G(f) C Rn+m is an n-dimensional manifold.
Example 4. The torus of revolution is the surface in R3 described by the

equation (0 < r2 < rl )
( x2 + y2 - rl )2 = r2 - z2 .
A parametrization can be obtained by the formulas

x = (r1 + r2 cos cp) cos V,, y = (r1 + r2 cos cp) sin 0, z = r2 sin cp
with parameters 0 < cp, ' < 2ir. The partial derivatives of the function
f(x,y,z) = ( x2+y2-rl)2-r2+z2 are
Ox -
2x(1 - x-+y2 ), of - 2y(1 2), = 2z,
21
y VI'X21 +y Oz
and it is obvious that the vector D f does not vanish at any point of the
torus of revolution. Hence Theorem 1 applies and shows that the above
equation defines a manifold.
Example 5. Not every set defined by an equation is a manifold. For ex-
ample, consider the set in R2 described by the equation x4 = y2:
Near the point (0, 0) E ]R2 this set is not a manifold. In fact, after having
deleted this point, any neighborhood of the set splits into four components
and hence cannot be homeomorphic to an interval. Indeed, the assumption
of Theorem 1 concerning the differential of f (x, y) = x4 - y2 is not satisfied
at the point (0, 0) E R.
2
3.1. Submanifolds of R" 51
Example 6. On the other hand, there exist manifolds in R" which cannot be
described by systems of equations satisfying the assumptions of Theorem 1.
Later we will see that every manifold defined as in Theorem 1 has a particular
property-it is orientable. An example of a non-orientable manifold is the
so-called Mobius strip. One of its parametrizations is
x = cos(u)+vcos(u/2)cos(u), y = sin(u)+vcos(u/2)sin(u), z = vsin(u/2)
with parameters 0 < u < 2ir, -7r < v < 7r.
Apart from equations, manifolds can also be defined by prescribing their

local parametrizations (charts). Let us explain this construction principle.
Theorem 2. Let M be a subset of R" and assume that for each point x E M
there are an open set U, x E U C R", an open set W C Rk, and a smooth
map f : W U such that the following conditions are satisfied:
(1) f(W) = M n U;
(2) f is bijective;
(3) the differential Df(y) has rank k at each point y E W;
(4) f'1 : M n U W is continuous.
Then M is a k-dimensional submanifold without boundary in R.
Proof. For an arbitrary point x E M we choose a map f : W -i U with
the stated properties and denote by y the pre-image of x, f (y) = x. The
differential has rank k, and hence we can assume without loss of generality
that
1r
det 9k 0.
ak
4
Consider the map defined by g(a, b) := f (a) + (0, b) with g : W x R"-k -+ R".
Then the determinant of the differential of g coincides with the determinant
above, and hence, in particular, it is different from zero. Applying the
inverse function theorem of differential calculus, we obtain two open sets

V1 and V2. with (y, 0) E V1 and x E V2 in R", for which g : V1 V2
is a diffeomorphism. We invert this map and denote the resulting inverse
diffeomorphism by h := g-' : V2 -+ V1. By assumption f is continuous,
and hence there exists an open set 0 C R" such that
{f(a):(a,O)EV1} = f(w)no.
Consider now the sets V2 := V2 nO and V1 = g-'(V2). Then we have
V2nM = V2nOnM = {g(a,0):(a,0)EVII,
and thus we obtain
h(V2nAf) = g-1(V2 nM) = V1 n(Rk x 101)
Therefore, the condition to be satisfied for each point of a manifold holds
for A1.
Now we will extend the notion of manifold, taking into account also bound-
ary points. We confine ourselves to the case that the boundary itself is a
smooth manifold without boundary (no corners or edges). We define the
k-dimensional half space Hlik to be the set
Hk = {xERk:xk>O}.
Definition 2. A subset M of R" is called a k-dimensional submanifold
(with boundary) if for each point x E M one of the following conditions is
satisfied:
(1) There exist open sets U and V, with x E U C R", V C R", and a
diffeomorphism h : U -+ V such that
h(U n M) = V n (Rk x {0}) .
(2) There exist open sets U and V, with x E U C R", V C R", and a
diffeomorphism h : U -+ V such that
h(UnM) = Vn(Hk X {O})
and the k-th component hk of h vanishes at the point x, hk(x) = 0.
Conditions (1) and (2) cannot be satisfied at the same time for one and
the same point x E M, for otherwise there would exist diffeomorphisms
h1 :U1 -iV1, h2 : U2 --+ V2 such that
h1(UInM) = VlnRk and h2(U2nM) = V2nlEllk, h2(x)=0.
3.1. Subinanifolds of R" 53
The set hl (U1 n U2) then would be an open subset in IRk mapped diffeo-
morphically onto h2(Ul n U2) by the chart transition map h2 o h, 1. Since
h2(x) = 0, the set h2(U, nU2) would thus contain a point from the boundary
8Hk = Rk-4 of the half-space. Consequently, it could not be open in Rk.
Altogether this contradicts the inverse function theorem. This observation
justifies the following
Definition 3. Let M C 1[t" be a manifold. A boundary point of M is a
point x E M for which condition (2) of Definition 2 is satisfied. The set of
all boundary points is denoted by 8M and called the boundary of M.
Theorem 3. Let AEI be a k-dimensional manifold. Then its boundary OM
is either empty or a smooth (k-1)-dimensional manifold without boundary,
88M=0.
Proof. Fix a boundary point x E OM and choose open sets U C R n. X E U,
and V C Ht" with
h(U n m) = V n (Hk x (01).
For every other boundary point x' E U n OM the k-th component of h has
to vanish at x` by the preceding observation, hk(x*) = 0. Hence we have
h(U n 8M) = V n (IItk-' x {0}),
and thus (h IunaAf, U n OM) is a chart for the boundary 8M.
Example 7. The boundary of the Mobius strip is a closed space curve.
Example 8. The (n - 1)-dimensional sphere Sn-' is the boundary of the
n-dimensional ball D".
3.2. Differential Calculus on Manifolds

When a manifold is covered by charts, every chart range is an open subset
of R" and hence a set on which differential calculus is familiar from anal-
ysis. In this way it is possible to develop a differential calculus on manifolds.
As in the preceding section, we will now start from a k-dimensional mani-

fold Mk and a chart h : U V around a point x and denote by y := h(x)
the image of x under this chart map. Then h-1 : V -{ U is smooth and
(Dh-')y = (h-1).,y is a linear map between the tangent spaces to R" (com-
pare Definition 2, Ch. 2)
(h-1).,y: TyR" - T=R" .
Definition 4. The tangent space of the manifold Mk at the point x is
defined to be the image of TVRk under the map (h-1).,y:
Trllfk (h-1).,y(Ty %) C .,R" .
The tangent space T?Mk is a k-dimensional vector space, since the differ-
ential of the diffeomorphism h-1 is injective. Moreover, we have to check
that the tangent space of the manifold just defined does not depend on the
choice of the chart. But this is an immediate consequence of the equivalent
description for the tangent space that is to follow next.
Theorem 4. The tangent space TTMk consists of all vectors (x, v) E T=R"
for which there exists a smooth curve y : [0, ±E) - Mk C R" such that
-y(0) = x and y(0) = v.
Proof. A vector v = (x, v) E TXMk in the tangent space can be represented

as v = (Dh-'),(w) for a certain vector w E Rk. The image under h' of
the straight line in Rk through y in the direction of the vector w is the curve
7(t) we were looking for: the equality
-t(t) = h-1(y + tw)
immediately implies -y(O) = h-1(y) = x, and from the chain rule we obtain
for the tangent vector
dt (h-1(y + tw)) l t=o = (Dh-'),(w) = v.

ddtt) =
The converse is proved analogously.
If the manifold Mk is defined by (n - k) equations, and if, in addition,

the differentials of the defining functions are linearly independent, then the
tangent space has a simple description.
3.2. Differential Calculus on ?Manifolds 55
Theorem 5. Let fl, ... , fn_k : 1R" , 1R be smooth functions and suppose
that
df1 A ... A 00.
Then the tangent space TTMk of the manifold
Mk = {xER": f1(x)=...=f,,-k(:r.)=0}
consists of all vectors v E T11R" satisfying

df1(v) = ... = dfn-k(v) = 0.
In particular, the euclidean gradient fields grad(f1), ... , grad(fk) are per-
pendicular to the tangent space of the manifold at each point of :11k.
Proof. Taking a curve r) Alk in 11k and differentiating the

equation
f1('r(t)) = ... = f"-k(r(t)) = 0
with respect to the curve parameter t yields
df1(i(t)) _ ... = df"-k(i(t)) = 0.
The tangent space TA1k is thus the subspace of all those vectors v E T IR"
on which all the differentials df1, . df"-k vanish. Comparing the dinlen-
sions of these two vector spaces shows that they have to coincide.
The set of all tangent spaces to the manifold is called the tangent bundle of
Alk and denoted by TMk. It is a manifold of dimension 2k. In fact, at least
in the case that A1k is determined by equations, f1 = . . . = f"_k = 0, the
set TAlk is in turn defined as a subspace of 1R" x 1R" by the equations
fl (:r) = ... _ .fn-k(x) = 0 and df1(x. v) = ... = df"-k(x, v) = 0.
These are 2(n - k) conditions in R2". The corresponding functional determi-
nant (Jacobian) does not vanish, since the differentials df1 are linearly inde-
pendent. The formula (x, v) := x defines a projection 7r : TMk . Alk on
the tangent bundle of the manifold assigning to every vector its base point.
Example 9. Consider the sphere S' = {x E R"+1 :IIxII = 1}. The

differential of the function IIXI12-1 is 2(x1, xn+1) and hence the tangent
=
space to the sphere at any point consists of all vectors perpendicular to this
point:
TS" = {(x, v) E R"+1 x Rn+1: IIxII =1, (x, v) = 01.

Definition 5. Let Mk C R' and N' C R' be two manifolds, and let
f : Alk - N' be a continuous map. We call f a differentiable map if for
each chart h-1 : V - Mk of the manifold Mk the resulting map f o h-1
V -. N' C R' defined on the open subset V C Rk is differentiable.
As in euclidean space, the differential of a smooth map can be introduced as

a linear map between tangent spaces. For a tangent vector (x, v) E TZMk we
choose a curve y : [0, e) - Mk with 7(0) = x and y(O) = v. The composition
f o y(t) is a curve in N1 passing through f (x) E N', and its tangent vector
describes the result of applying the differential of f to the tangent vector
(x, v),
f.,x (X, V) := (f(x), I

dtf o y(0)
The differential of a smooth map between two manifolds has all the proper-
ties which are familiar from euclidean space.
Theorem 6. The differential f.,2 : TXMC -+ Tj(t)NI of a smooth map is

a linear map between the tangent spaces, and the differential of the super-
position of two smooth maps f and g is equal to the superposition of their
differentials,
(g o f).. = 9.,f1=) ° f*.x
The last formula is the generalized chain rule.
Definition 6. A vector field V on a manifold Mk assigns to every point

x E Alk a vector V(x) E ,,Mk in the corresponding tangent space.
If the map V : Mk - TRn = Rn x R" is smooth, then we will speak of a

smooth vector field on the manifold. Vector fields can again be added and
multiplied by functions, so that the vector space of smooth vector fields is
a module over the ring Coo(Mk) of all C°°-functions on Mk.
Example 10. The formula V(x) = (x, (x2,-x1,0)) defines a vector field
on the 2-dimensional sphere (see the figure on the next page).
3.2. Differential Calculus on Manifolds 57
For a chart map h : V Mk C R', which this time and and sometimes
also later will be denoted by h instead of h-1, h.(a/ayi) are vector fields
tangent to Mk defined on the subset h(V) C Mk, and they provide a basis
in each tangent space. For simplicity and as long as it is clear to which chart
we refer, these vector fields on the manifold will also be denoted by a/ayi.
On the subset h(V) C Mk. every other vector field V can be represented as
their linear combination
k
V(y) _ Vi(y)5
y
ii
i=1
Here V2(y) are functions defined on the set h(V); using the chart map. now
and then they will also be considered as functions on the parameter set V.
These functions are called the components of the vector field V with respect
to the fixed chart.
Example 11. In euclidean coordinates on 1R2, consider the vector field
V=x15x2-x25x1
depicted on the next page. Introducing in R2 - (0} polar coordinates by the

formula
h(r,p) = (r cos V, r sin so), 0 < r < oo, 0 < <p < 2ir,
we see that V corresponds to the vector field In fact, differentiating

the map h, we obtain the formulas
h.
ral
=
a
smV5X2
a_ 1, +
251
(1) ar ar r 7X1
57X2
(2)
49
hCl
ap a
-rsincp+ a
rcoscp ax2 _
-x2a ,a
axl +x 57X2
axl
Differentiable functions f : Mk -- R can be differentiated with respect to a

vector field V. At a fixed point x E Mk we choose a curve ry : [0, EJ Mk
satisfying the initial conditions y(0) = x and y(0) = V(x). The derivative
off at the point x in the direction of V(x) is now defined by the formula
V(f)(x) := tf o y(t)It=o
The result is a C30-function V(f) defined on the manifold Mk. In the next
theorem we stunmarize the properties of this differentiation:
Theorem 7.
(1) (V+W)(f) =V(f)+W(f);
(2) V(f1 + f2) = V(f1) + V(f2);
(3) V(f1 - f2) = V(fl) - f2 + f1 - V(f2);

(4) If the vector field V is represented as V = V'(y)818yi in some
local chart, then
k
V(f) _ Vi(y)8(foh)
i=1
Proof. We will prove (4); all the other claims follow immediately. If the
point x E Mk corresponds to the point y E V under the chain map h : V
Mk, then -y(t) = h(y + t(V 1(y), ... , Vk(y))) is a curve in Mk satisfying the
initial conditions -t(O) = x and ry' (0) = V(x). The formula for V(f)(x) to be
proved then follows from the chain rule:
k
V(f)(x) = d
dt f o9 (y+t(V1(y),...,Vk(y))) _ i=1' V'(y)a(f
h).
0
Next we will discuss the notion of Riemannian metric on a submanifold Mk

of euclidean space. The scalar product in R" is denoted by (v, w). We
restrict it to the tangent spaces of the submanifold.
Definition 7. Let M11k C R" be a submanifold. In each tangent space T, Mk
the formula
g= ((x. V), (x. w)) := (v, w)
defines a scalar product. The family {g.,} of all these scalar products is
called the Riemannian metric of A1k.
In a chart It : V _ Alk, the Riernannian metric is locally described by the

functions g,j defined on the set V,
a a ) ah,
= (ayi ah)
gij W = A(y) (ay' ayj
ami
Thus, the components gij(y) of the Riemannian metric are the scalar prod-
ucts of the partial derivatives of the chart map h considered as being vector-
valued. When it is clear to which chart we refer, h is often omitted (as in
the examples to follow). The (k x k) matrix (g,3) is symmetric and positive
definite for each y E V. By g(y) we will denote its determinant, which al-
ways is a positive function. Recall that we already agreed in Chapter 1 for
bilinear forms to denote by (g`J(y)) the inverse of the matrix (g,j(y)).
Example 12. In euclidean coordinates on R", the entries gij(y) = 6,j are
constant and equal to one for coinciding indices i = j. and vanish in the
other cases.
Example 13. In polar coordinates on R2 - {0}, equations (1) and (2)
immediately lead to the following coefficients for the usual euclidean metric:
9r. _ (ar
a s _
1,
_ d a_= 0,
, gw;v
/aa , a r
2
.
Or g,-" Or 8y7
In particular, in polar coordinates on the plane we have g(r, p) = r2.
Example 14. On the sphere S2 - {N. S} C R3 without north and south
pole, we consider spherical coordinates defined by
h(y^, y) = (cos f cos ri', sin, cos t,, sine) , 0 < V < 27r, -ir/2 < z, < r/2.
Computing the tangent vector fields yields
a a
(3) + cos cos y a-2 ,
a a a a
siny - sin yo sing + cosy
(4)
az- - Cos
axi 2 3.
For the coefficients of the Riemannian metric on the sphere we thus obtain
g 4_90b-)=cos29 - /(Y 0, 9vw= =1.

The determinantt of the metric on the sphere in these coordinates is equal
to g(v, cost 0; therefore, the spherical coordinates degenerate at the
poles, which is why we deleted them.
We will use the Riemannian metric g of a manifold in order to associate
with every smooth function a vector field, the so-called gradient. For a fixed
point x E Mk and a tangent vector v E TIMk we first choose a vector field
V such that V(x) = v. The assignment v - V(f)(x) determines a linear
functional TxMk - R on the tangent space, and hence there exists a vector
grad(f)(x) E TTMk such that the equality
V(f)(x) = gx (grad(f)(x), V(x))
holds for all vector fields. The vector field grad(f) is called the gradient of
the smooth function f : AIk -+ JR.
Theorem 8. Let f : AIk -+ JR be a smooth function, and let h : V _ Mk
be a chart. Denote the coefficients of the Riemannian metric in this chart
by gij. Then
k
grad(f) = e(f o h) ij a
ayi 9 ayi
i j=1
Proof. Inserting the right-hand side into the definition, we compute

k
a(f o h) 9ij a a
9 ayi ayi ,aym
ij=1
_ E a(fayio h) 9ij 9jm - kLa(fayi

o 11)bi
m = a(faym
o h)
i,j=1 i=1
According to Theorem 7, the last expression is precisely (8/aym)(f ), which
by the definition of the gradient implies the result.
Example 15. The formula for the gradient of a function defined on an open
subset U c R" of euclidean space expressed in cartesian coordinates is
Of
(5) grad(f) = n a-i aTi
a
i=1
Example 16. In polar coordinates on R2 - {0} the coefficients of the metric
are
grr = 1, 9rp = 0, gp' = 1
T2
For a function f in the variables r, gyp, the formula of the preceding theorem
yields
(6) (f) = Orof g ar + of ;, a

rr a of a 1 of a
app = ar ar + r2 aap aV
Example 17. In spherical coordinates on S2 - IN, S} we have
g'v;v = 95'" = 0, 1,
cos y)'
2
and for the gradient of a function f : S2 - {N, S} -R this leads to
grad(f) = cost
Of a
(7)
_ aV ao a0
For example, the height function f (x1, x2, x3) = x3 on the sphere may be
written in spherical coordinates as f (cp, ') = sin rP, and thus
grad(f) = cos z' .
Hence the vector field grad(f) on the sphere S2 vanishes at precisely two
points, at the north and the south pole.
Vector fields and the Riemannian metric can be represented in various co-
ordinates. Now we derive the transformation formulas for its components
with respect to different charts. To this end, fix two charts h : V -> Mk and
h : V -+ Mk and denote the points from V by y = (y1, ... , yk) and those
from V by z = (z1, . . . , zk). The chart transition maps will be denoted by
0 = (01, ... , 0k) := h-1 o h and t' = (VG1, ... , aiik) := h-1 o h, respectively.
Theorem 9.
(1) Let V be a vector field and denote by V 1, ... Vk its components
in the chart h and by 0, ..., Vk the components in the chart h.
Then, for each index I between 1 and k.
V, (Y) k ,
(4(y)) az; (b(y))
i=1
(2) Denote by g;j the coefficients of the Riemannian metric with respect
to the chart h and by g,; those referring to the chart h. Then they
are related by the formula
k 01 corn
gi;(y) _ ay;
-,
91m (0(y))
l,m= I
In particular, for the determinant of the Riemannian metric the
following holds:
2.
9(y) = det 9 (4(y))
Nr 1
Proof. The first formula is a consequence of the chain rule. In fact,

a
= k atvr a _ k awl a
(a) = ( ) (E E
az' h. az' h of az' a)h` ax' ayl = az' a-
This leads to the following formula, describing the vector field in the chart
h:
k ,a k
k a
i=1
a2 1=1 \ i-1 ax / a
The second formula follows directly from the first:
a a\ a a atll aij,m
ax'' axj 2k
1,m=1
811,1 awm
az' azi C ayl ' aym > = E
A
1,m=1
az' azi
if, in addition, the (formally completely equivalent) roles of h and h are
exchanged.
We will use these transformation formulas to define the divergence of a

vector field. This will be taken to be a function on the manifold determined
by a local formula. Therefore, it has to be proved that the corresponding
expression is independent of the choice of the chart.
Definition S. Let V = F; V'(y)a/ay' be a vector field on the manifold Mk
expressed with respect to the coordinates of the chart h : V Alk, and let
g(y) : V R be the determinant of the Riemannian metric in this chart.

Define a function div(V) : h(V) R. the divergence of V, by the formula
div(V)(h(y)) 1
Vg i=1
k
a (v/g V M)
ayi
Theorem 10. The defining formulas for the divergence of a vector field for
two arbitrary charts h : V Mk and h : V -- Mk of the manifold Mk
coincide at corresponding points:
a(,/g-V'(y))
1 = 1 - k a(vg= f7 '(z))
k
i=1 8yi azi
The proof uses a formula expressing the derivative of a determinant by its

minors. For the sake of completeness we state it here.
Lemma 1. Let k2 functions hj(x) be defined on an open subset V of lRc.
Denote by H(x) := det[hij] the determinant and by A;j(x) the minors of the
matrix hij(x). Then
k
_H(x) _ Ohij
Di (x) .
ax'' axr J
Proof. Differentiating the determinant and applying the Leibnitz rule yields
e h12 ... hlk h11 ... hl.k-1 a=te
OH(x)
ax''
det +...+det
8hk1 hk2 ... hkk hkl ... hk,k-I a8x'
Expanding each of the k resulting determinants with respect to the column

containing the derivatives leads to the stated formula. 0
Proof of Theorem 10. Applying the preceding formula, we obtain
1 k a( 9(y) V'(y)) = k aV'(y) k

+ V i(y) ai (in g(y))
9y) ;=1 ayi ;=1 ay i=1 aa/
i
= i=1
L yi j=1
My))
m
+ V'(y)ay; In 9(O(y)) +In Idet gyr
i=1
Now we compute the partial derivative of the first sum and simplify the
resulting expression, making use of the following formulas:
k
aoi
at' = alJ .
s i azj
i=1
This immediately leads to the expression
k- k-
1 a_ 9(1/) V`(}/)) = 1 O( 9(z) O(z))
syi
r--
9 5) ;=1 Vr9(z) azi
k 2 i 1 m 1
+ In Idet ayr I 0
i,j,t=1
'(4(y)) a )az, i + azi
Last, we have to show the following equation for each index 1 < j < k:
k ai l m
i,1=1
8 z1 Oyi + az,j In Idet 00
T
Io
= 0.
Using the lemma, the second summand in this equation can be written as
a -1 k a
jln
det amm I o V, = (det I o 10 (det 00m) OI`
J J
V,) -1 k a4,i
a2Q°
= (det syr J o syiaya 0° aZ;
L
L J i,a.0=1
The matrix [Ow'/ad] is the inverse of [490m/ay''], and, by Cramer's rule.

a-j0
= D&Q . (det I m 1 0 wl -1
8z°
This yields
aim OP a20°
a
In Idet
az"
rk
ay*
- ayiay,3 azi 19z° .
I
i,°,)9=1
and the equation to be proved becomes
k 0n2jpi 00° 0-20a atpi 0,03
kk
i,° =1
azjaz° ayi + ayiay'3 azi az° = 0
But this identity immediately follows from the fact that p and are mutually
inverse maps: Differentiate the equation
k n i
ww) a;
i=1
with respect to the variables z1, set a = 1, and finally take the sum. 0
Example 18. In cartesian coordinates on R", the formula for the divergence
becomes
"
aVi
(8) div(V) _
axi
Example 19. In polar coordinates on 1R2 - {0}, since g = r, for each vector
field V = V' 010r + V2a/0' the following formula holds:
1 a(rV') a(rV2) OV1 1 8V2
(9) div(V) = + = + -V1 +
r a l Or r aV
Example 20. In spherical coordinates on S2, the determinant fo the metric
is g = cost ly and thus for every vector field V = V1a/a<p + V2,9/a' the
corresponding formula becomes
(10)
1 a(cosip V1) a(cosii - V2) aV1 av2 2
div(V) =
cos y ( 0p + a ) - acp + a - tan tp V .
Theorem 11. Let V be a vector field and f a smooth function on Mk. Then
div(f V) = f div(V) + V(f).
Proof. It suffices to prove the formula in an arbitrary chart. This is done
by a straightforward computation:
div(f V)
1 8(,g- f Vi) _ 1 of a(Vfg- y')
agi 09 V+ f ay
" i=1 "" i=1 '
= f div(V)+EViaf = f div(V)+V(f). O
i=1
The last operator to be introduced in this chapter is the Laplacian 0; it

acts on functions and is a second order differential operator.
Definition 9. Let f : Mk --' R be a smooth function. Then the Laplacian
of f is defined as a function on Mk by the formula
O(f) := div(grad(f ))
Remark. The operator A : C°O(Mk) C- (Mk) is a linear operator act-
ing in the vector space of all C0°-functions on Mk. In mathematics, there
are-concerning the sign of 0-two differing conventions. Classically and
mainly in analysis, A is defined as above. On the other hand, in vari-
ous branches of geometry and global analysis it is common to introduce
the Laplacian via the formula A(f) = -div(grad(f)) and to call this the
Laplace-Beltrami operator. There are good reasons to do so, as we will see
in connection with the classical integral formulas. Thus, studying a text,

the reader has to be careful to detect which sign convention was chosen by
the author. In this book we decided to use the particular choice of sign for
the Laplacian stated in the definition.
Theorem 12. The Laplacian 0 : C-(A1k) -. C-(Mk) is a linear operator.
Moreover, for two functions fl and f2 the following formula holds:
A(fi - f2) = fi ' (f2) + f2 A(fl) + 2 (grad(fl ), grad(f2))
Proof. The linearity of 0 is obvious, and the second formula follows by
combining the corresponding rules for the operators grad and div:
A(fi f2) = div(grad(fi f2))
= div(fi grad(f2) + f2 grad(f1))
= fi div(grad(f2)) + f2 div(grad(f1)) + 2 (gr'ad(fi). grad(f2))

0
In local coordinates we immediately obtain from
grad(f)=Ya(foh)i;o
the formula
AM _ I- k
a
s i
a(foh)
g
Again, we write out the explicit form of the Laplacian in the examples we
have discussed so far.
Example 21. In cartesian coordinates on R",
(11) (OXi)2
.=1
Example 22. In polar coordinates on Ilt2,
(12) o(f) =
8 (Of a(r Of
a2f 1 of 102f
r ar lrar) + app \ r2 app)) are + r ar + r2
Example 23. In spherical coordinates on S2,
(13) f cost 0 OW2 + f,2

-tan f
In these coordinates the height function f (x1, x2, x3) = x3 on S2 can be
represented as f = sin ip. This leads to
A(f) = -2sing(, = -2f,
3.3. Differential Fornu; on Manifolds 67
i. e., f is an eigenfunction of the Laplace-Beltrami operator corresponding

to the eigenvalue -2. Other eigenfunctions of A on S2 are constructed
starting from a harmonic and homogeneous polynomial P(xl, x2. x3) in 1R3,
0R' P = 0, and restricting this to the sphere S2. The resulting function is
an eigenfunction of the Laplacian (see Exercise 25).
3.3. Differential Forms on Manifolds

Up to now, for every manifold Al" we considered its tangent space TAI'
and vector fields. Now we go one step further and form the exterior power
AA(T.Mm)
of all k-forms w= : TTM' x ... x T1Mm - IIt at the point x E Alm.
Definition 10. A k-form wk on a manifold /L1m is a family {wi} distin-
guishing a k-form wT E Ak(Mt) at each point x E AI'".
The differential of a smooth map f : N" -i M"' between two manifolds

allows to pull back k-forms from Mm to yield k-forms on N". This is
accomplished via the formula
(f'wk)(vi, ... , vk) = wk(f.(vl), ... ,
where vl, .... vk E TyN" are tangent vectors to N' at the point y E N",
and f.: TyN" Tf(y)Mm is the differential of f at this point. The k-form
f `wk is called the induced form or pullback of wk by the map f on N".
This construction can, in particular, be applied to a chart h : V Alm
of the manifold Alm. Hence, for a fixed chart., to every k-form wk on AI'"
there corresponds a k-form h' (wk) on the open set V in the space R' or
Mm, respectively. If y = (yi, ... , y'") are the associated coordinates, then
h'(wk) can be represented by means of the component functions w/ as
h*(wk) = Ewldy1.
I
The summation extends over all ordered multi-indices I = (il ... << ik),
where dy1 is a shorthand notation for the k-form dy1 := dy`' A ... A dy'k.
Definition 11. A k-form wk on the manifold Mm is called a differential
k-form or smooth k-form if for each chart h : V - Mm the coefficients w1
of the k-form h*(wk) are smooth functions on the subset V C IIt".
Differential forms can be added and multiplied by functions defined on A1m

in an obvious way. Hence the set S2k(Alm) of all C°°-forms of degree k is a
module over the ring C°O(AIm) of smooth functions on the manifold.
The exterior derivative of k-forms-familiar from the euclidean space ;,F"

and discussed in Chapter 2-can now be transferred to the situation of k-
forms on manifolds without difficulties, preserving all the known properties.
This proceeds as follows: For a k-form wk on Al' and a chart h : V - 11"'
we first consider the induced form h'(wk) and its differential d(h'(wk)). The
latter is a (k+ 1)-form on the set V, and its pullback under the inverse chart
map h-1 : It (17) -- V is a (k + 1)-form. This yields a (k + 1)-form
dwk := (h-1)`(d(h*w'))
defined on the open set h(V) C 1M'". The construction just described is in-
dependent of the particular choice of the chart, and hence it uniquely defines
a global (k + 1)-form dwk on the manifold M. In fact, for another chart
It, V1 -. Mm satisfying h(V) fl h1(V1) 0 0, we obtain on the intersection
:
h(V) n h1(V1) the equality

(h-')*(hi 1 o h)'(d((h-1 o hl)`h'(wk)))
(hl 1)'(d(hiwk)) =
= (h-1)d((h, 1 o h)'(h-1 o h1)*h'(wA))
= (h-1)d(h*wk) .
This computation relies on the fact that the exterior derivative commutes
with the diffeomorphism (hi 1 o h), a property discussed in Chapter 2.
Definition 12. The (k + 1)-form dwk defined starting from the k-form wk
on Al... is called the exterior derivative of wk.
All properties of the exterior derivative known from euclidean space remain
valid in the situation of a manifold. The next theorem summarizes them.
Theorem 13. For arbitrary forms wk, Wk' 17
I
on a manifold M.. the follow-
ing properties hold:
( 1 ) d (wk + wk) = dwk + dwi ;
(2) ddwk = 0;
(3) d(,,;k A tl') _ (&,k) A,/ + (-1)Awk A (d77');
(4) if f : N" -+ Mm is a smooth map, then f' commutes with the
exterior derivative, d(f*wk) = f'(dwk)
The purely algebraic operation of forming the inner product between a vector
field V and a k-form wk is also transferred into the situation that both objects
are defined over a manifold.
Definition 13. We define the inner product of a vector field V and a k-form
wk by
(V _J wk)(W1, ...,Wk-1) := wk(V,W1.....Wk-1).
3.4. Orientable Manifolds 69
3.4. Orientable Manifolds

An orientation of a real vector space is the choice of one of the two equiva-
lence classes in the set of its bases discussed in the first chapter. This can, in
particular, be applied to the tangent space T=M' of a manifold, and leads
to the notion of an orientation 0= at the point x E Alm. An orientation of
a manifold Mm consists in a "continuous" choice of orientations at each of
its points.
Definition 14. An orientation 0 of a manifold Alm is a family 0 = {Or} of
orientations in all tangent spaces TM m depending continuously on the point
x in the following sense: At each point x E Alm there exists a chart h : V
Mm containing this point such that the basis {h. (a/ay' ), ... , h. (a/aym) }
is compatible with the orientation Oh(y) for every point y E V.
Definition 15. A manifold Al' is called orientable if there exists at. least
one orientation on it.
First we state a necessary and at the same time also sufficient condition for
the orientability of a manifold in terms of the chart transition maps.
Theorem 14. Let 0 be an orientation on Mm. Then there exists a family
{(hi, V )liEI of charts with the following properties:
(1) The image sets hi (Vi) cover the manifold,
film = U hi(V) ;
iEI
(2) If the intersection hi (V) n h; (V3) 34 0 is non-empty, then the de-

terminant of the differential of the chart transition map h, 1 o hi is
positive,
det[D(hj 1 o hi)] > 0.
Conversely, if there exists a family of charts with these properties, then Mm
is orientable.
Proof. Let 0 be an orientation on Mm. Choose those of the charts h :

V Alm for which the basis {h.(a/ayl ), ... , h .(a/ay`)} is compatible
with the orientation of the manifold. By the definition of an orientation
these charts cover the manifold. For two of these charts, hi : V Afm and
h3 : Vj -+ Mm, with coordinates y = (y1, ...Iym) and z = (z'...., zm) we
have at mutually corresponding points
{hi.(a/ayl), ... , hi.(a/ay')} = Oh;(y) = Oh,(z)
_ {hj.(a/azl),...,hl.(a/azm)}.
70 3. Vector Analysis oil Manifolds
Therefore, these bases are compatibly oriented, the transition matrix be-
tween them is the differential D(h, o h;), and we obtain
1
det[D(hj 1 o h;)] > 0.

The converse is proved analogously.
If the manifold All C IR" is described by (n - m) independent equations,
then it is orientable. We will prove this fact now.
Theorem 15. Let f1, .... f"_,,, : U , R be smooth functions defined on an
open subset U C 1R". and assume that
... A dfn-in 1
df1 A T0
at each point. Then the manifold
Mm = {rEU: fl(x)_...=fn-n(x)=0l
is orientable.
Proof. Consider the euclidean gradients grad(fl ), ... , grad(fn_,,,) of the
functions. By assumption, these vector fields are linearly independent at
each point of the set U and, moreover, perpendicular to the tangent space
Tjlll"' at the points of the manifold (compare Theorem 5). An orientation
Ox in TiAfm is distinguished by requiring that a basis v1, ... , vm E TAf"'
is positively oriented if and only if
d]R"(grad(fl)(x) ....grad(fn-,.)(r),v1, ...,v, ) > 0.
Here dR = dx1 A ... A dx" denotes the volume form on R". It is not hard
to see that this condition determines all orientation on M"'.
An oriented submanifold Al'" c R" carries a distinguished differential form
of highest degree, the so-called volume form. Choose a basis e1.... , em E
TxM"' consisting of mutually perpendicular vectors of length one in the
fixed orientation Ox at any tangent space TIM . For arbitrary vectors
ill i .. , v,,, E Tx:'ll 'n we define
(vi,el),
dill'"(v1....,vm) = det
(v,, el),
This definition uniquely determines the form dAl". Every other basis
el, in the same orientation can be represented as a linear combi-
nation e; = F_ A,2ej with an orthogonal matrix A of positive determinant.
But then det(A) = 1.
Definition 16. The m-form dMm on the oriented manifold Mm C R" is
called the volume form.
Remark. The volume form dAlm is not the exterior derivative of an (m-1)-
form. Nevertheless, this form is traditionally denoted by dMm. The volume
form does not vanish at any point of the manifold. Evaluating dMm on any
orthogonal basis e1, ... , em in the orientation yields
dMm(e1, ..,em) = 1.
On the other hand, the orientation can be reconstructed from the volume
form. In fact, a basis v1, ... , vm is positively oriented if
dt11m(vl, ...,vm) > 0.
Changing the orientation of the manifold results in a change of sign for the
volume form.
Theorem 16. An m-dimensional manifold Mm is orientable if it carries a
nowhere vanishing differential form of degree in.
Proof. For an orientable manifold Alm the volume form dAlm has the nec-
essary property. Conversely, suppose that there is an m-form Wm on AI"' not
vanishing at any point. Then we call a basis v1, ... , vm positively oriented
if
Wm(vl, ...,v",) > 0.
This determines an orientation on Al m. 0
In a chart h the induced volume form h*(dM'") is proportional to dy' A
... A dy"', h' (dMm) = f (y)dy1 A ... A dym. We compute the function f (y)
as follows:
f2(y) = (h`(dMm)(a/ayl, ... , a/aym))2 = det2 [(h.(a/ay'), e,)]
= det [(h.(a/ay'),ej)J det [(ej,h.(a/ay'))]
= det [(h.(a/ay'), h.(.9/ay'))] = det[g;l]
Hence f2 (y) is equal to the determinant det[g;j (y)] of the Riemannian metric,
and we obtain the formula
h*(dM-) = 9(y)dy' A...Adym.
Example 24. For a surface piece A12 C R" with a parametrization h
][1;2 , AI2 C R", the following classical notation is frequently used:
/ ah ah \
E = 911(x,y) = ate, Y-
ah ah
G = 922(X, Y)
( -.9Y 09-Y
ah ah
F = 912(X, y) = a-, 5
The formula for the volume form is then

dM2 = EG-F'2dxAdy.
Example 25. The volume form of 1R' in cartesian coordinates is
(14) d1W = dx1 A... Adx".
Example 26. In polar coordinates on 1R2 - {0} we have g(r, gyp) = r2, and
hence the volume form is
(15) dIR2 = r . dr n dip .
Example 27. In spherical coordinates on the 2-sphere, 1P) = cost v/i,
and thus the volume form is
(16) dS2 = cos t dip A diP.
In the case of an oriented manifold, the divergence of a vector field V can

be expressed in terms of the exterior derivative and the volume form.
Theorem 17. Let Mm C R" be an oriented manifold, let dMm be its volume
form, and let V be a vector field with divergence div(V). Then
d(V J dMm) = div(V) dM'".
Proof. In local coordinates we start from
m
V= Vs(y)a ii and dMm = /dyl A...Adyt.
i=1 y
This implies
m
V-i dMm = ,Fg (-1)'-1V'dy' n...AdyiA...dy'",
i=1
and the formula
d(V.j dMm) = m
a(/ V t) dy' A ... A dym = div(V) dMm
i=1
immediately follows from the definition of the divergence of a vector field.
0
Remark. As in the euclidean space R. for every (oriented) manifold M'"
the 1-form w , dual to a vector field V can be described using the volume
form and the Hodge operator:
*wv := V i dM' .
The divergence formula can then be written as d(*wl,) = div(V) dM'".
Non-orientable manifolds exist; proving non-orientability for a particular

manifold, however, is sometimes a little more difficult than to show ori-
entability. We state a simple criterion for the non-orientability of a manifold.
This can be used, e.g., to prove that the Nlobius strip is non-orientable.
Theorem 18. Suppose that. for a manifold Mm, there exist two charts
h:V Mm and h : V M' with the following properties:
(1) V and f7 are connected sets, and the intersection h(V) n h(V) is
not connected;
(2) the determinant det(D(h-' o h)) has opposite sign at two points in
h-'(h(V) n h(V)).
Then M' is a non-orientable manifold.
Proof. Assume that MI is orientable. The sets V and V are open subsets of
]R'", and hence orientable. Thus, without loss of generality, we may assume
that the chart transition maps h and h preserve the orientations. Consider
the volume form dMt, and represent it in both coordinate systems:
h* (dM-) = J dy' A ... A dyt , h' (dM-) = f dz' A ... A dz' .
Note that in both these cases the sign is "+", since the chart transition
maps h, h preserve the orientation. Because
(h-1 o h)'(dz' A ... Adz') = det(D(h-' o h)) dz' A ... A dzm,
we have
f det(D(h-'oh)) = f.
But this is a contradiction, since f and Vg' are always positive. 0
The orientation of a manifold induces a unique orientation on its boundary.
To define it, we make use of the exterior normal vector field. Its definition
relies on the following observation.
Lemma 2. Let U C ]H' be an open subset of the half-space. let x E U n 8HHm
be a point in the boundary OHHt, and let f : U - V be a dieèomorphism from
U to an open subset V C H. F o r a vector v = (x, (v', ... , v' )) E T=H"'
at the point x with non-positive m-th component, v"' _< 0, the image vector
MV) = (f (x), (w1, ..., w"')) also has non-positive m-th component w"' < 0.
f. (v)
Proof. Choose a straight line -y(t) = x + t v and note that, because of
v"' < 0. the point y(t) belongs to U fl HH"' for sufficiently small negative
values of the parameter t. Then we have y(0) = x, y(0) = v. and the in-th
component of f.(v) becomes negative:
_dtf"'(x+tv)
d
= lim f"'(x+tv)-0 < 0. 0
t-o- t
At a boundary point x E 8M' C M' of a manifold there are two tangent

spaces. On the one hand, there is the tangent space TxMm of Al'. and,
on the other hand, the boundary determines its own (in - 1)-dimensional
tangent space, T=(8M'") C TXM'". At each boundary point x E 8M'. we
define a unique tangent vector N(x) E TxMm by the following conditions:
(1) N(x) is perpendicular to T
(2) N(x) has length 1;
(3) for a chart h : V C Him - Mm around the point x E Al"' the m-th
component of the vector (h)-1(N(x)) is negative.
Definition 17. The vector field N constructed along the boundary is called
the (exterior) unit normal vector field of the boundary.
We fix an orientation of the boundary OMm of an oriented manifold M' by

calling a basis vl, .... vm_1 E TT(OMm) positively oriented if the extended
basis A((x), vi, is positively oriented in T2A1m. A simple argument
shows that this condition actually determines an orientation of the boundary.
-
For the corresponding volume forms we have the important formula:
Theorem 19. The volume form of the boundary of a manifold is the inner
product of the exterior normal vector field N with dull'":
d(aMm) = Ni dM'n
Proof. Choose an orthonormal basis in the orientation of the vector space
T,,111 consisting of the normal vector el := N(x) together with additional
vectors e2, ... , e,n. For arbitrary tangent vectors v1.... , 0m-I to the bound-
ary we obtain
(Ari ditlm)(t'1, ...,v1) = d111"'(el,vl, ...,vrn-1)
(vi, e2) ... (vl,em)
= det
(vm-1, e2) ... (vm-1, em)
= d(OMm)(i'1, .. , l'm-1) ,
since the vectors vl, ...,v,,,_1 are perpendicular to el. 0
In the subsequent sections of this book the boundary of a manifold will
always be oriented this way.
Example 28. The exterior unit normal vector field of the sphere Sn-I(R)
of radius R, considered as the boundary of the ball Dn(R), has the form
N(x)
1r(la +...+x nal
= R(x ax-1 ax's
The volume form of Sn-I(R) can also be computed by means of dSn-1(R) _
N.J dR', yielding the formula
(17) dSi-l(R) = 1 E(-1)'-Ix'dxlA...A dx'A ...Adx".

i=1
Hence the volume form of the sphere S"-'(R) can be represented via the
embedding i : Sr-1(R) -+ IR" as the form induced from the following (n-1)-
form defined on Rn:
,,n-1 = -1x'dxl A ... Adx' A ... Adxn'
i=1
i.e., the following equation holds:
dS"-1(R) = i.(wn-l)
Computing the exterior derivative of wi-1 in R' leads to the formula

dwn-1 = n dlRn ,
R
and by applying Stokes' theorem from Chapter 2 we obtain the integral of
the volume form:
J n-l(R) dSn-1(R) = JD"(R)

dwn-1 = Rvol(D°(R)) .
3.5. Integration of Differential Forms over Manifolds

The integral of an m-form over an m-dimensional manifold will be defined
by dividing the manifold into small subsets which are diffeomorphic to open
subsets of R' or Hn', respectively, and integrating the given form using its
chart representatives one by one. The sum of the resulting values is then the
total integral of the form over the manifold. From the very beginning we will
confine ourselves to compact manifolds in order to have to deal with finitely
many summands only and thus to avoid convergence questions for series
arising otherwise. A detailed exposition of this definition of the integral
requires a so-called partition of unity in order not to count contributions
from overlapping charts twice. This is a special family of smooth functions to
be discussed first. Recall that the support of a function cp : Mn' - R
is the closure of the set {x E Mm : V(x) 710).
Theorem 20. Let Mm be a compact manifold. Then there exist smooth
functions and charts
cpi : A f '- 1R and hi : V - Mm (1 < i < l)
with the following properties:
(1) The support of the function vi is contained in hi (V ),
supp(wi) C hi(V)
(2) The functions Wi are non-negative, and their sum is equal to one:
Ws(x) 1.
i=1
Proof. We choose a chart h., : Vr - M"' around each point x E M n

satisfying h,,(0) = x. Here, the Vi are open subsets of Rn or H'n. Choose,
moreover, non-negative functions 1G:: R'n, H'n R such that
(1) z(0) = 1;
(2) supp(r!.?) C V.
3.5. Integration of Differential Forms over Manifolds 77
Setting
0 if y V hx(VV) ,
;Px(y) = if y E hx(VV),
1Gx(hx 1(y))
these functions can be transferred to form a family of non-negative smooth
functions bx : M' --+ R on M. By construction we have c3x(x) = 1 and
supp(,px) C hx(VV). Taking advantage of the compactness of Mm, we finally
obtain finitely many functions 31, ... , yet and charts (hl, V1). ... , (h1, VI)
satisfying
t
Ar = I {xEMm:yi(x)j4 01
ii=11
and supp(cpi) C hi(Vi). The sum i(, < is positive at each point, and
the functions cpi we are looking for result from normalizing these functions
'A' Vi = cpi/v.
Now we will define the integral of an m-form w' over an oriented and
compact manifold M. To this end, we choose charts hl, .. . , ht and a
partition of unity, Cpl, .....pt subordinate to them satisfying the properties
formulated in the preceding theorem. Furthermore, we suppose that the
chart maps hi : Vi -+ M'" preserve orientation. Then
fi(y)'dylA...Adym
is an rn-form on V with compact support, and we eventually define the
integral:
t t
wm jhWm) _ Jf(y).dy1...dym.
J t=1 t=1 i
We show that this definition of the integral is independent of the chosen

partition of unity. Consider another atlas (h1, V1), ..., (h,., V,.) with subor-
dinate partition of unity 'p1, ... , ;p,.. If the intersection of the chart ranges
is not empty, hi(V) f1 ha(V0) 0, then
fi(0pWm) = jhQ.Jm).
This follows immediately from Theorem 7 in Chapter 2, since the deter-
minant of the differential D(h= 1 o ha) is positive. Summing now over the
index i, 1 < i < 1, as well as the index a, 1 < a < r, we obtain from
E Bpi =_ E cpa = 1 the relation
i a
gym) _ f hi (vi ' gym)

a=1 V i=1 V,
Definition 18. The m-dimensional volume of a compact and oriented sub-

manifold Mm C lR" is the number
f
vol(Mm) := J dM'".
M^'
Consider a curve, viewed as a 1-dimensional submanifold M1 C IR" of R".

If It : (a. b) - M1 is a parametrization of the curve, then
jdM'
b
= Za
and we recover the length of the curve. More generally, if It : V - If'
is a parametrization of Mm, then, using the coefficients of the Riemannian
metric, the volume of M' can be written as
vol(M'n) = det[g,.1 (y)] dy.
JV
Example 29. The last formula in §3.4 shows that the (n - 1)-dimensional
volume of the sphere Sn-'(R) and the n-dimensional volume of the ball
D"(R), both of radius R, are related by
vol(S"-1(R)) = Rvol(D"(R)).
Example 30. The coefficients of the Riemannian metric on the torus of
revolution discussed in Example 4 are
(ri + r2 cos V)2, 9vv = 0, g = r2 .
Hence f = r2(rl + r2 cos V), and we obtain

21r 2ff
vol(T2) = fJr2(ri + r2 cos cp)dcpdi' = 47,2r1 r2 .

0 0
We will conclude this section by a remark concerning measure theory on

manifolds. Let CO(Mm) be the ring of all continuous functions defined on
the compact and oriented manifold Mm. Setting
µ(f) := f dM-
JMm
defines a linear functional µ : CO(Mm) - lR which is positive and monoton-
ously continuous in the sense of the following properties:
(1) If f E C0(lbfm) is a non-negative function, then µ(f) > 0.
(2) If f,, is a monotonously increasing sequence converging to a function
f E CO(Mm), then
lim
n -.a o µ(f") = µ(f)
3.6. Stokes' Theorem for Manifolds 79
This turns the pair (CO(Mm), p) into a so-called Daniell-Stone functional

on the set M. Within the framework of general measure theory, one first
associates with every, D-S functional p on a set X an outer measure p' on
X. The Caratheodory construction then leads to a a-algebra of subsets in
X on which p` becomes a measure. By applying these general principles
to a compact and oriented manifold, one constructs the so-called Lebesgue
measure on MI. It is defined on a a-algebra containing the Borel sets, and
every bounded and Borel-measurable function turns out to be integrable.
For the purposes of vector analysis we do not need this extension of the
notions of measure and integral to manifolds, since the functions occurring
are, as a rule, at least continuous. It is, however, interesting to note that
the notion of integral treated here fits into the general theory of measure
and integration as sketched above. The interested reader may refer to the
literature for details (see the book of K. Maurin in the bibliography).
3.6. Stokes' Theorem for Manifolds

In the preceding sections we collected all the notions necessary for the for-
mulation of Stokes' theorem on manifolds.
Theorem 21 (Stokes' Theorem). Let Mk be a compact, oriented manifold,

and suppose that the boundary OMk is endowed with the induced orientation.
Then, for every (k - 1) -form wk-1 on Mk,
f8Mk wk-1 = f
JAlk
dwk-1
If, in particular, Mk has no boundary, then for every (k - 1) form wk-1
dwk-1
= 0.
AIk
Proof. In Theorem 20, it was shown that we can choose finitely many charts
(V1, h1), ... , (Vr, hr) covering the manifold 11Ik, together with a subordinate
partition of unity O 1, ... , cpr : Mk R. We label these charts in such a
way that, on the one hand, for each index i less than a certain index ro the
set V intersects the boundary, V n 8Mk 34 0, and, on the other hand, for
all indices i > ro the set V is disjoint from the boundary, V n OMk = 0.
From >i vi = 1 we obtain >i dcpi = 0 and use this to rewrite the exterior
wk-1:
derivative of
k-1
= k-1 - Pidw k-1 + dipi n wk-1
i=1 i=1 i=1 i=1

This implies the equation

r
JM' dwk-1 = d (hi (Piwk-1))
V is an open subset of Rk for rp + 1 < i < r, and h; is a k-form on

V1 whose support is completely contained in V. For each of these indices i
we choose a k-chain c in Rk for which
supp h, (Wiwk-1) C Int ck C ck C V j.
Applying now Stokes' theorem for chains (Theorem 8, Chapter 2) to ck, we
obtain
wA-1))
Jd(h(iwk_1)) =
f h' (Viwk-1) = 0,
1.d (ht = Jask
since the form vanishes on the boundary of the chain. Now we consider the
indices i between 1 and ro. For any of these, V is an open subset of the half-
space Hk, and as before we obtain chains ck in lHlk with the same properties
of the supports with respect to h; Applying Stokes' theorem to
these as well, we obtain
1v;d (hi J
hi (ca1wk-1) .
Now hi (,p1wk-1) does not necessarily vanish any more on Ock n (Rk- l x {0} ),
but only at the points of 8c; belonging to the interior of Hk:
V;
d (h; (Vjwk-1))
_
I
VflRk-1
hi
The pairs (V n Rk-1, h1IRk) with i = 1, ..., ro form a covering of the bound-
(`Piwk-l).
ary BAIk. Hence, by the definition of the integral,

ro
wk-1 h i (ViwR-1) ,
Jamk -
t-1 knRl- 1
showing the equation we set out to prove. O
In the sections to follow we will be dealing with various applications of

Stokes' theorem. As a generalization of the discussion in §2.5, we will first
study line integrals and prove an analogue-only for 1-forms, however-of
Poincare's lemma. This holds for manifolds in which every closed path can
be contracted to a point.
Definition 19. A connected manifold Mk is called simply connected if every
two C'-curves co, cl : [0, 1] , Mk with coinciding initial and end points are
homotopic.
3.7. The Hedgehog Theorem (Hairy Sphere Theorem) 81
By Theorem 9 in §2.5, whose proof immediately carries over to the case of a

manifold, on a simply connected manifold the line integral of a closed 1-form
w1 depends exclusively on the end points of the curve. Having fixed a point
xo E Mk, the line integral along a curve joining the points xo and x,
w
f(x) =
lox
uniquely defines a function on the manifold. Its differential df coincides with
w1, and we obtain
Theorem 22. Every closed 1-form on a simply connected manifold is exact.
Example 31. The winding form defined on 1R2 - {0} is closed, but not
exact. This shows that R2 - {0} is not simply connected.
3.7. The Hedgehog Theorem (Hairy Sphere Theorem)

Consider two oriented compact manifolds Mk and Nk without boundary
and of equal dimension. Two maps fo, fl : Mk - Nk between them are
called homotopic if there exists a smooth map
F : Mk x [0,1] -+ Nk such that F(x, 0) = fo(x) and F(x,1) = fl (x).
We prove
Theorem 23. Let wk be a k-form on Nk and let fo, fl : Mk Nk be
homotopic maps. Then
JMk0 (wk) = fm k fl (wk) .

Proof. The oriented manifold Mk x [0, 1] has boundary
8(Mk x [0,1]) = Mk x {1} - Mk x (0},
where the minus sign indicates that the orientation is reversed. Therefore,
Stokes' theorem implies
f(wk) = F*(wk) = dF(wk) .

Jf k
fl (w) -
JMk 8J(Mkx[O,1]) JMk x [OI]
But the form dF*(wk) = F'(dwk) = 0 vanishes, since the k-dimensional
manifold Nk carries no non-trivial (k + 1)-forms. 0
Theorem 24. The antipodal map from the sphere to itself, A : Sn - Sn,
A(x) = -x, is homotopic to the identity Ids. only for odd dimensions n.
Proof. Consider on Rn+1 the form
n+1 _
wn = -1x' dx1 A ... A dx' A ... A dx"+1
i=1
whose restriction to the sphere Sn is its volume form dSn (Example 28,
equation (17)). If A is homotopic to the identity Ids.., then the previous
theorem implies
A*(wn) = / n w" = vol(Sn) .
Jis. S
The induced form A* (,n) = (-1)n+lwn is proportional to the form n.
Thus we obtain the condition
(_1)n+lvol(Sn) = vol(S"),
i. e., (n + 1) has to be an odd number.
Theorem 25 (Hedgehog Theorem). A sphere S2k of even dimension has
no nowhere vanishing, continuous tangent vector field.
Proof. Suppose that there exists such a vector field on the n-dimensional
sphere S". We first approximate this vector field by a smooth vector field
V (Stone-Weierstrass theorem), and then normalize it so that the vector
V(x) has length one at each point. Next we consider the resulting smooth
tangent vector field as a vector-valued function V : Sn - Rn+I satisfying
the following two conditions:
(x, V(x)) = 0, IIV(x)II = 1.
Define the homotopy F : Sn x [0, 1] - Sn from the sphere to itself by the
formula
F(x, t) = cos(irt) x + sin(irt) . V(x).
The length of F(x, t) is equal to one everywhere, since x and V(x) are per-
pendicular. Moreover, F(x, 0) = x and F(x, 1) = -x, i. e., F is a homotopy
between the identity and the antipodal map of the sphere Sn. But then the
dimension n has to be an odd number.
In the German mathematical literature, this result is known as the "Hedge-
hog Theorem" (,,Satz vom Igel"), since its contents can be expressed figu-
ratively by saying that a hedgehog cannot be combed in a continuous way.
Because it is so vivid, we prefer this to the name "Hairy Sphere Theorem",
which seems to be more common in the Anglo-Saxon world.
3.8. The Classical Integral Formulas

Now we will discuss the classical integral formulas already treated in §2.6 for
chains. Compared to the preceding case, in this new formulation we benefit
from having notions like divergence, gradient and Laplacian as developed
in the differential calculus on manifolds at our disposal. At the same time,
the orientation of the manifold and the exterior unit normal vector field on
the boundary will play a special role. We will prove these integral formulas
3.8. The Classical Integral Formulas 83
for arbitrary compact and oriented manifolds (in R'). This will be the final
formulation of the classical integral formulas as they are needed in many
branches of mathematics as well as theoretical electrodynamics and hydro-
dynamics. We start with the Ostrogradski formula relating the divergence
of a vector field to its flow across the surface.
Theorem 26 (Ostrogradski Formula). Let Mk be an oriented, compact
manifold, and let N be the exterior unit normal vector field to its boundary.
Then, for every vector field V : Mk Tlllk on Mk,
div(V)dMk = (V, N) d(OMk).
lL nik
Proof. We know from Theorem 17 that the divergence and its inner product
with the volume form are related by the formula
div(V) dMk = d(V J dMk) .
A straightforward application of Stokes' theorem implies
J div(V)dMk = d(V i dMk) = f V
lk J k nl k
Let x E 8Mk be a point of the boundary. We decompose the vector V(x)

into one part that is proportional to the exterior normal vector, and a vector
W(x) that belongs to the tangent space T,ZBMk to the boundary:
V(x) = (V(x), N(x)) N(x) + W(x) .
Moreover, note that the restriction of the inner product W J dMk to the
boundary 8Mk vanishes identically. This implies that for the inner product
of V with the volume form dMk. on the boundary 8Mk
V _j dMk = (V, N) N j dMk + W i dMk = (V, Al) N(x) _j dMk .
Hence, we obtainJ
ik
div(V)dMk =
f Mk
(V, N) Ni dll'ik .
By Theorem 19, the inner product N(x) . dMk coincides with the volume
form of the boundary, d(8Jik). 0
As a direct application of the Ostrogradski formula we obtain Gauss' theo-
rem.
Theorem 27 (Gauss' Theorem). Let V be a vector field, let f be a function
on the oriented, compact manifold Mk, and let N be the exterior unit normal
vector field of the boundary. Then
J l (V, grad(f )) dMk + f l f . div(V)dMk = J (V, N) d(81bik) .

alk f
k
Proof. This equation immediately follows from the Ostrogradski formula

together with the rule from Theorem 11
div(f V) = f div(V) + V(f) = f div(V) + (V, grad(f)) .
In a similar way we derive Green's formulas in versions that are not confined
to 12. First we generalize Green's first formula.
Theorem 28 (Green's First Formula). Let f, g : Mk R be smooth func-
tions on the compact, oriented manifold Mk. Then
f f'O(g) dMk +
A1k
g'ad(g)) dMk = ff
aMk
(grad(.9)N) d(8Mk).
Proof. By the definition of the Laplacian, we have
J %rk
f O(g) dMk = f
'k f div(g'ad(g)) dAik .
Now apply Gauss' theorem.

Applying Green's first formula twice leads to Green's second formula.
Theorem 29 (Green's Second Formula). Let f, g : Mk - III be two smooth
functions. Then
1 [g. (f)-f. (g)] dMk = f N) ] d(aMk).

Ark aMk
1-1
Remark. The scalar product (grad(f ), N) defined only on the boundary is
iaN aN
often denoted by the symbol Of ION, since it is the derivative of the function
f in the direction of the exterior normal vector. This leads to a different
formulation of Green's second formula:
f [g . (f) - f . %(g)] dMk = fMk

[g.-f. ]
Corollary 1. Let Mk be a compact, oriented manifold without boundary.
Then
(1) f div(V) dMk = 0 for every vector field V;
Mk
(2) f go(f) dMk = f fo(g) dMk = - f (grad(f ). grad(g))

Mk
dMk
Alk Mk
for any two functions f, g E CO°(Mk).
0
3.8. The Classical Integral Formulas 85
Hence the Laplacian is symmetric with respect to the L2-scalar product.

Moreover, the choice of sign we adopted implies that it is non-positive.
Corollary 2 (Hopf's Theorem). Let Mk be a compact, connected, oriented
manifold without boundary and assume that the function f : Mk R sat-
isfies at each point the condition 0(f)(x) > 0. Then the function f is
constant.
Proof. Integrating the assumption s(f)(x) > 0 over Mk and applying the
symmetry of the Laplacian just proved, we first obtain
0< f 1-o(f)-dMk = J f-j(1)-dMk = 0,

fk Mk
i.e., the Laplacian of f vanishes identically, A(f) = 0. Inserting f = g
into Green's first formula (Theorem 28) and taking into account that the
boundary integral vanishes by the assumption concerning Mk, this implies
J grad(f)I2-dMk
= -Jntkf 0,
and hence grad(f) = 0. Thus f is constant, since Mk is connected. 0

Concluding this section, we formulate Stokes' theorem in its classical form
on 1R3. Contrary to the preceding theorems involving the generalizations of
divergence and gradient to manifolds as introduced in the second section of
this chapter, this only involves the notion of curl on open subsets of 1R3 from
§2.3.
Theorem 30 (Stokes' Theorem-Classical Version). Let M2 C R3 be a

compact, oriented surface, let V be a vector field defined on an open subset
M2 C U C 1R3, let N : M2 S2 be the exterior unit normal field to the
surface M2, and let T : aM2 - T(aM2) be the unit tangent vector field on
the boundary curve aM2 with the induced orientation. Then
f (curl(V), N) dM2 = f MZ (V, T) d(0M2).

Proof. Consider the 1-form
4 := V1dx1 + V2dx2 + V3dx3
on U associated with the vector field V = V la/axl + V2a/axe + Via/ax3

as explained in §2.3 and its derivative
d`4 =
av2 aV1
ax, - avl
aV3 aV3 aV2
, ,
I dx ndx2 + ] dx ndx3 + [ J dx2 ndx 3.
axl - axe ax3 ax2 - ax3
Recall that the curl of V corresponds to the 1-form *dwv. If, on the other
hand. It : W -+ M2 is a parametrization of the surface with components
h', h2, h3 and coordinates yl, y2 from W, then for the exterior normal vector
N to the surface we have the relation
09 h x 09h
yl y2 / II A
09y1 y2
x ah ll
Two arbitrary vectors v, w E R3 satisfy the identity

1Iv (V, V) (v, w) 1
x wII2 = det
(v, w) (w, w) J
and hence the preceding equation implies
x f.
ayl aye II
II
For the first component of the normal vector N written in the coordinates
y1, y2 this reads, e.g., as
N'dM2 = N' f dy' n dye = 1 ahe Oh3 _ah2 ah3

[ay' aye aye ayl
, dy' A dye
.
On the other hand, it is easy to compute the pullback by h of the forms dx2
and dx3:
OhY12
h*(dx2) = + A2dy2, h`(dx3) = ldy' + Oh3dy2.
A direct comparison implies the following formula, which is independent of

the coordinates y', y2:
N'dM2 = dx2 A dx3 .
Similarly one proves
N2dM2 = - dx' A dx3, N3dM2 = dx' A dx2 .
The scalar product of the curl of V with the unit normal field N multiplied
by the volume form dM2 is thus simply the differential of wv:
(curl(V), N) dM2
= [ aV3 aV' 9V3 aV2
09x1, N2 dM2 + 09x1 - axe, N 3dltil 2
aV2 aV'
2
] N dM + 109x3
1
= axe - 09x3
dwv. L L
Therefore, Stokes' theorem can be applied in the format
J (curl(V),N) dM2 = &4 = y=J[V1dx1 + V2dx2 + V3dx3].

J 8M2
%f2 Af2 8M2
3.9. The Lie Derivative and the Interpretation of the Divergence 87
If, however, T is the unit tangent vector field to the curve 9M2, then
d(8M2)(T) = 1, and hence
T'd(aM2) = dxl, T2d(8M2) = dx2, T3d(8M2) = dx3.
Now we can rewrite the line integral above as
J [V ldxl + V2dx2 +V 3dX3] = J1L12

(V, T) d(8M2),
and, summarizing, we arrive at Stokes' classical integral formula. 0
3.9. The Lie Derivative and the Interpretation of the

Divergence
The aim of this section is to interpret the divergence of a vector field geo-
metrically as the infinitesimal volume distortion of its flow. First we recall
some results from the local theory of ordinary differential equations and in-
troduce the flow on a manifold as well as the Lie derivative of forms. Then
we compute this Lie derivative by means of the exterior derivative, which,
in a special case, leads to the interpretation of the divergence mentioned in
the title.
Let V be a vector field on the manifold Mk. An integral curve of V is a

curve ry : (a, b) Mk whose tangent vector -y(t) = ry,(8/8t) at each point
coincides with the value of the vector field there:
?(t) = V('Y(t))
The well-known existence theorem for autonomous differential equations
states that for every initial point x E Mk there exists a maximal integral
curve ryx : (ax, bx) - Mk defined on an interval, containing the number

0 E R, satisfying
'Y.,(0) = X.
Moreover, this maximal integral curve is uniquely determined by the initial
condition. Denote by EV the set
EV = {(t,x)E]RxMk:ax<t<bx}
and call the map -0: EV Mk defined by the formula
,Dt(x) = '(t,x) := 7x(t)
the flow of the vector field V . If the maximal integral curves are defined for
all values t E R of the parameter (i.e., if Ev = R x Mk), then the vector field
V is called complete. A complete vector field V determines a one-parameter
group 4Pt : Mk ---+ Mk of diffeomorphisms from the manifold to itself:
4)to+tl = 4t0 O 46tl .
This relation between diffeomorphisms is a direct consequence of the unique-
ness of the integral curve for a fixed initial condition. Conversely, if a one-
parameter group of difeomorphisms is given, then we obtain a vector field
by looking at the tangents to the trajectories +t(7) of a point:
V (X) dt (bt(x))1t=o
Example 32. The one-parameter transformation group in R2 of the vector
field
V = -x2871 +x1872
is determined by the differential equations
i1 = -x2 th2 = x1
and coincides with the group of all rotations in the plane.
Example 33. The one-parameter transformation group of the vector field
V +...+x"ban
= x1871
in ]R" is determined by the differential equations
.i1 = x1, ...,
in = 7n
and coincides with the group of all dilatations in Ilt".

Suppose that the manifold Mk is a closed subset of R. Then Mk is a com-
plete metric space: Every Cauchy sequence in Mk converges. For bounded
vector fields we will prove
Theorem 31. Every bounded vector field on a complete manifold without

boundary is a complete vector field. In particular, every vector field on a
compact manifold without boundary is complete.
Proof. Denote the maximal length of the vector field V by K,

K := sup{IIV(x)JI: xEMk}.
Let : (a, b) , Mk be a maximal integral curve of the vector field. and as-
sume that b < oc. We apply the mean value theorem of differential calculus
to the vector-valued function y : (a, b) - AIk C IR". It states that for every
two values of the parameter, a < t1 < t2 < b, there exists one more value,
tl < t' < t2, such that
Ily(tl)-y(t2)II <_
It2-tll Ily(t*)II = It2-tll'IIV(y(t'))II <-
Since b is finite, any sequence t; E (a, b) converging to b is a Cauchy sequence
in R. The last estimate shows that the image sequence -y(ti) E Mk is also a
Cauchy sequence. By assumption, Mk is a complete manifold, and hence the
sequence -y(ti) converges in Mk. This observation shows that the left-hand
limit,
lim y(t) _: x`,
tb
exists. The point x` E Mk is an inner point of the manifold, since there are
no boundary points by assumption. We choose a chart h : Rk -+ Mk around
the point x' = h(O) and represent the vector field V in the corresponding
coordinates,
V = Vla/eyl + ... + Vka/Oyk .
The solution y(t) of the differential equations
y1 = V 1(yl, .. . yk), .. yk = V k(yl, ... , yk)
,
converges for t - b to the point 0 E Rk. Hence the integral curve y(t) can
be extended beyond the parameter b. This contradicts the maximality of b
resulting from the assumption b < oc. 0
Now we turn to the Lie derivative of a differential form with respect to a
vector field. If w' is an i-form, then. pulling it back by the flow of the vector
field, we obtain a 1-parameter family of forms, 4 (w'). The derivative of that
with respect to the parameter at t = 0 is the so-called Lie derivative. This
concept involves the flow of the vector field only in a small neighborhood of
zero and is thus well-defined for arbitrary vector fields.
Definition 20. The Lie derivative CV (w') of an i-form w' is the form
wt
CV(w`)
li o
Theorem 32. The Lie derivative of a differential form is expressed in terms

of the exterior derivative and the inner product as
'CV (d) = d(VJ w')+VJ (dw`)
Proof. The exterior product w' A 171 is bilinear, and the product rule is
proved as usual,
Cv(w' Ar?) = Cv(w') AT? +W' ACv(T?)
Furthermore, the exterior derivative d commutes with the induced maps 4)i,
and hence we have
Cv(dw') = d(Cv(w))
Denote the right-hand side of the equation to be proved by C4(J). Then
Cv(dw') = d(V J dw') + V J (ddwl) = d(V J dw')
= d(VJdw'+d(VJw')) = d(C*(w')).
The first, purely algebraic equation,
VJ (w'Ar7') _ (VJ w')Ar7+(-1)'w'A(VJ r'),
leads to the identity
Cv(w'Ar7') = d(VJ (w'Arl'))+VJ d(w'Ar1')
= d((VJ w')Aq'+(-1)'w'A(VJ r?))+VJ (dw'Ar7'+(-1)'w'Adr7')
= d(VJ w')Ar7'+(-1)'-'(VJw')Adr7'+(-1)'dw'A(VJ r))
+w'Ad(VJ
+(-1)'(V J w') A dr7' + w' A (V J dri)
= (d(VJ w')+VJ r7')+VJ dr7')
= Lv(w')Arl'+w'ACv(r')
Lastly. C and C coincide for functions,
Cv(f) = V(f) = VJ df = G`v(f),
and thus the formula is proved. 0
We apply the formula of the preceding theorem to the volume form of an
oriented manifold Alk. From the definition of the divergence of a vector field
we obtain
Cv(dMk) = d(V J dAlk) = div(V) dAfk.
Fixing a point x E Afk and choosing a neighborhood U(x, e) C Mk of radius
EJiv(V).dM'
> 0, we obtain by integration
1
= I l=
(l,e) /
Dividing both sides by the volume of the set U(x, e) and taking the limit for
E - 0. we see that the mean value theorem of integral calculus immediately
implies the formula
div(V)(x) = lim (vol(t(U(x.e)))) .

vol(U(x, r)) u=o
This formula allows a geometric interpretation of the divergence of a vector
field: It is the volume distortion of its flow in infinitesimally small time and
on infinitesimally small domains surrounding the point x.
Corollary 3. The divergence of a vector field vanishes if and only if its flow
4t : AIk _ Mk consists of volume-preserving diffeomorphisms.
Proof. For a measurable set A C Afk we denote by hA(t) := vol((Dt(A)) the

volume of the set -(Dt(A). The derivative of the volume change is computed
as follows:
d (hA(t 61t =
dtd
I fd(A) dAfk)
=dt =
L1A)
f t (4)
Gv(dMk) = 41
(A)
div(V) dAlk.
From this we see that the flow 4bt preserves the volume if and only if the
divergence of the vector field V vanishes identically. 0
The Lie derivative of a vector field W with respect to another vector field V
is defined just like the Lie derivative for forms; nevertheless, it is called the
commutator of both vector fields.
Definition 21. Let V and W be two vector fields, and denote by the
flow of the vector field V. The commutator [V, WI is defined as the vector
field
W
[V, W1 = lim
t-0 t
We prove an important formula relating the commutator of two vector fields
to the exterior derivative of 1-forms.
Theorem 33. Let wl be a 1-form, and let V, W be two vector fields. Then
dwl(V,W) = V(wl(W)) - W(w'(V)) -w'([V, WI)
Here V(wl(W)) is the derivative of the function w1(W) in the direction of
the vector field V.
Proof. If x E Mk is a fixed point, then by the definition of the commutator

[V, W]
[V, W](x) = Iio x))) - W(x) .

The formula to be proved then follows from Theorem 32 by means of the
following computation:
wl([V W])(x) = lim w'(W(x))

' t--o t
= li (4'-t)'(f') - w' (W(t(x)) - W(x))l
o (W(4'c(x))) + u ow'
_ -(Cvw')(W)(x) + V(w'(W))(x)
_ -W(w'(V))(x) - dw'(V,W)(x)+V(w'(W))(x). 0
If, in particular, wl is a closed form, then we obtain
w'([V, W]) = V(w'(W)) - W(w'(V))
In local coordinates the vector fields are represented by their components,
V=via k
Applying the above formula to the closed form dyt gives us
dy'([V,W]) _
k
V;aWl - k W1.
i=1 i==11
This leads to the following local expression for the commutator of two vector
fields:
k
(Viewt
[V , W] _ - Wi avt) l a
J \- 1 1 ay' I ay'
Example 34. The commutator of the vector fields V = y a/ax, W = 0/4
defined on 1R2 is the vector field [V, W] = -8/0x.
The algebraic properties of the commutator of vector fields are summarized
in the following theorem.
Theorem 34. Let U, V, W be vector fields, and let f be a smooth function
on the manifold Mk. Then the following identities hold:
(1) [U + V, W] _ [U, W] + [V, W];
(2) [V, W] _ -[W, V];
(3) [U, [V, W]] + [V, [W.U]] + [W, [U, VJ] = 0 (Jacobi identity);
(4) [f V,W] =f
Proof. These formulas immediately result from the local expression for the
commutator. We present the calculation for the last of the formulas:
[f v, w] _ E
k
-l
(i.v'
taw,aw f-VI)l a
vi
ayt - ayt J'
f
0=1
vi awl
ayi
wi av a
ayi ) ay,
_ (F ay,
wt
i=1
of_) ((vi)
E a
k
1=1
-
OV
In the following sense, the commutator is a natural operation between vector

fields.
Theorem 35. For a diffeomorphism 4' : Mk -+ Mk and vector fields V, W

on the manifold Mk the differential (P. is compatible with the commutator:
'F, ([V, W]) _
Proof. Note first that for every 1-form w' the functions 4'(w')(W) and
w' ('. (W)) are related by the formula
'DO(wl)(w) = w'(=(w)) 04-' -
Theorem 33 then implies

wl ((I).[V+w]) = ('F*wl) ([V, W]) O'D
= V ((''wl)(W)) o 4)- W o 4)- d(4)`wl)(V, W) o 4)
_ ('F.V) (w'(4'.W)) - ('.w) (U" (4%v)) - (dw') ('.v, 4.W)
= w' (['.(V),4'.(W)])
This equation holds for every 1-form w', and hence the theorem follows.
Remark. Requiring that 'F be a diffeomorphism is often too strong. The

compatibility of the commutator with the differential can sometimes be
proved under weaker assumptions. For example, if the map 4) : M - N is
smooth, the vector field V on M is said to be -t-related to the vector field
W on N if at each point m E M
$.,m (V(m)) = W('F(m)).
If one can now show that, together with the pairs (VI, WI) and (V2, W2),
the commutator pair ([V1, V2], [WI, W2]) is necessarily also fi-related, the
same compatibility property follows immediately. To see this, apply the
commutator [WI, W2] to a function g : N --+ R:
[WI, W2]$(m)(9) = WI(4i(m))(W29) - W2 (4)(m)) (WI 9)

= 4'.,m(VI(m))(W29) - 4)4,m(V2(m))(WI9)
= VI (M) ((W2 9) 0 4') - V2 (M) ((WI 9) 0 4?).
On the other hand, (W1 g) o 4? = Vi(g o 4?), since
V+(n)(9 o 41) = 41.,n(Vi(n))(9) = Wi(4)(n))(9) = (Wi 9)(4'(n))

Thus we obtain
[W1, W24(m)(9) = [Vl, V2]m(9 0 4>) = 4).,m[Vl, V2]m(9) ,
i.e., the vector fields [WI, W2] and [VI, V2] are, in fact, 4)-related.
The commutator of two vector fields measures the extent to which their
flows do or do not commute. This explains the name for the vector field
[V, W].
Theorem 36. Let V and W be two complete vector fields on the manifold
Mk and denote by 4't and X8, respectively, their flows. Then the commutator
[V, W] vanishes if and only if 4it o 4' = %P, o tt for all -oo < s, t < oo.
Proof. Because
d
= tim (4'-h-h).W - (4'-t1) / = (4'-t1).([V, W]),
the commutator [V, W] vanishes if and only if the vector field W is invariant
under the flow 4it, (4)1).W = W. This condition is in turn equivalent to the
commutativity 4't o T, = T. o 4it of the diffeomorphism 4't with the flow ID,
of W.
3.10. Harmonic Functions

A function f : Mk - R is called harmonic if it is a solution of the homo-
geneous Laplace equation 0(f) = 0. As a special case of Hopf's theorem
(Corollary 2), we have the following:
Theorem 37. Every harmonic function on a compact, connected, and ori-
ented manifold without boundary is constant.
If the boundary of the manifold Mk is not empty, there exist two particu-
larly important boundary value problems for harmonic functions.
3.10. Harmonic Functions 05
The Dirichlet Problem: Assume that a function V : OMk - IR is given. We

look for a harmonic function f : Mk IR whose values on the boundary
aMk coincide with those of cp:
0(f) = 0 in Mk and f IaaMk = cp .
The Neumann Problem: Assume that a function cp : OMk -+ IR is given. We
look for a harmonic function f : Aik R whose normal derivative on the
boundary coincides with yp:
0(f) = 0 in Mk and eiv = cp on aMk.

A solution of the Neumann problem is never unique. For each solution f
the sum f + C is, for an arbitrary constant C, a solution of this problem,
too. This is the only degree of freedom, since we have
Theorem 38. Let Mk be a compact, connected, and oriented manifold in
R", and let cp : aMk -+ IR be a smooth function.
(1) The Dirichlet problem has at most one solution.
(2) If f,, f2 both are solutions of the Neumann problem, then f, - f2
is constant.
(3) The vanishing of the mean value of cp is a necessary condition for
the solvability of the Neumann problem:
p d(aMk) = 0 .
JOMk
Proof. If fl, f2 both are solutions of the Dirichlet problem, then the differ-
ence u := f, - f2 satisfies the equations
0(u) = 0 and u IBMk = 0.
From the first Green formula we obtain
0 = J fk -Lk 85Jk
U. a"
Nd(aMk),
aN
and hence the gradient of u vanishes. Since we have u IBMk= 0, the func-
tion u has to vanish identically. In the case of the Neumann problem the
argument runs along the same lines. i.e., starting from the equations
au
0(u) = 0 and = 0 on aMk
and the Green formula, we again conclude that grad(u) = 0. Thus the
difference u = fl - f2 is constant. If the Neumann problem has at least one
solution for any given function cp :8Mk R, then, applying the classical
integral formulas, we obtain
0 f O(f) dMA = r L . d(Wk) = J

Mk
160 .
d(OMk).
The Dirichlet and the Neumann problems have, for a given boundary condi-
tion cp : 8Mk R (satisfying f cp = 0 in the case of the Neumann problem)
a solution. We will not prove this existence result here for general mani-
folds, but confine ourselves to the case of the ball D"(R) C R' of radius
R > 0. For these spaces, there is an explicit classical solution formula which
will be derived here. For the sake of simplicity we only discuss the case of
dimensions n > 3 and leave it to the reader to complete the discussion in
dimension n = 2, which differs only slightly from the one below. We start
with a few preparations. By
r= (x1)2 + ... + (xn)2
we denote the distance from the point (x', ... , x") E Rn to the point 0 E Rn.
Lemma 3. Let u(x) be a harmonic function defined on the set Rn - {0},
and assume that u depends only on the radius r. Then there exist constants
C1 and C2 such that
C2
u(x) = C1 + -n-2
Proof. By assumption there exists a smooth function h : (0, oo) -' R satis-
fying u(x) = h(r(x)). Differentiating this, we obtain the formulas
Ou x' OZU n( )(' i( )r2 - (xi)2
112
8x
h(r)r, ax = h r \r/ r +h
r3
and the latter implies the following differential equation for the function
h(r):
n _
0 O(u)
52x1
= lh(r).
For n > 3 the function h(r) = C1 + C2 r2-n is the solution of this ordinary
differential equation.
Let y E Rn be a fixed point. The function u(x) = JJx - y112-n defined on

the set Rn - {y} is a translate of r2-n, and hence a harmonic function. On
the other hand, the partial derivatives of a harmonic function are harmonic
functions as well. Thus
au I Ix112 I lyl l2
u(x) - 2 y
.=1 ax' Ilx yl In
is a harmonic function defined on iRn - {y}. We will use this family of

harmonic functions in the observations to follow.
Lemma 4. Let IIxii < 1 be any point in the interior of the unit ball D".
Then
I n 1 IIx
IIXII" dSn-1(y) = vol(Sn-').
Proof. First we prove that the function

h(x) . f n-' IIx_ IIyIIn 2
dSni-1(J)
depends only on the radius r = r(x). In fact, for a linear orthogonal map T :
1R" -+ R" we obtain, from I det(DT)I = I det(T) I = 1 and the corresponding
transformation formula for the volume form of the sphere,
T*(dSn-1) = T*(NJ d1Rn) = NJ (T'(d1Rn)) = NJ dRn = dSn-1,
the property to be proved:
TII'(Iy)Iin
h(Tx) = J dSn-1(y)
= J n-s lixl dSn-1(y)
n-1 IITxIITyIIn
_ I 1- IIXII2
. dSn-1(z) = h(x).
Jsn-' IIx - zlln
If, in addition, h(x) is a harmonic function, then
A. (1_ilx2
z (h)(x)
= - f
IS, 1
yIIn
dSn-1(y))
= 0.
The first lemma implies that there are constants C1, C2 with h(x) = C1 +
C2 r2-". At the point x = 0 the function h(x) is a regular function and
satisfies h(O) = vol(Sn-1). Thus the constant C2 vanishes, and C1 is equal
to vol(Sn-').
The solution of the Dirichlet problem for the unit ball D' C ]R" together
with an explicit formula are the subjects of the next theorem.
1
AX) = vol(Sn-1)
_ 12
Theorem 39. For every continuous function cp : S"-1 - 1R, the formula
II-
yII" cp(y) dS"-'(y)

is., IIx
defines a harmonic function in the interior of the unit ball, and at a boundary
point z E Sn-1
lim f (t z) = cp(z).
t-1-
Proof. The function (1 - IIxII2)IIx - is harmonic with respect to the

yll-,,
variable z, and hence the function f (x) is harmonic, too. We prove that
f(z) coincides with ;p(z) on the boundary Sn-1 in the way stated above.
Denote by in := y E Sn-1} the maximum of the modulus of p
and fix a positive number e > 0. The continuous function p is uniformly
continuous on the compact set Sn-'. Hence there exists a number b > 0
such that for any two points y, z E Sn-1 in the sphere, i i - z < 6 implies
the estimate I :p(y) - cp(z) I < e. We decompose the sphere Sn-1 = D1 u D2
into the parts
D1 = {y E S" : IIy - ziI < b}, D2 = {y E Sn-1 - zII > a} . :
II
For y E D2 and 0 < t < 1 we estimate the distance from y to the line
segment between 0 and z E S"-':
Z)2
fly - tz1I2 = 1 - 2t(y,z) + t2 = (t - (y,z))2 + 1 - (y,z)2 > 1 - (y.
1 1 b2
> 1 - (y, z) = 2-(2- 2(y, z)) = -Ily - zII2 > .
2
We then split the difference
_ 2
P ' z) - Y (z) = vol(Sn-1) Y dSr (?/)

J$^ Iltz
into the integrals over D1 and D2. The modulus of the first integral can be
increased to
1-IItzII2 n
dS" 1 1
Di S^-' Iltz - ylln

We treat the second integral using the inequality obtained before:
a2)n
< 2m - (1-t2)( vol(Sn-').

D2
In summary, for every positive number E > 0 there exists a number 6 > 0
such that for all 0 < t < 1 the following inequality holds:
E+2mb - t2).
Hence the upper limit is bounded by E,
lim suplf(t' z) - V(z)I < E
The last inequality holds for all positive numbers E > 0, and this in turn
implies
0
tt-
lim f (t z) = p(z) .
Next we will discuss several consequences of the solution formula for the
Dirichlet problem. For a harmonic function f : D"(R) IR defined on the
closed ball of radius R > 0, the function j (z) := f (R z) is also harmonic
on the unit ball. Applying the previous theorem and returning to the vari-
able r = R z E D"(R) finally leads to the Poisson formula for harmonic
functions:
2 2 2
-
f(x) = vo1Sn-1) J "-' IIR- RI yl l1"f(R y) dS"-1(y)
Evaluating the Poisson formula at the point x = 0 leads to Gauss' mea
value theorem for harmonic functions.
Theorem 40. The value of a harmonic function f at the center of the ball
coincides with the mean value of the harmonic function on the boundary of
the ball:
We will use Gauss' mean value theorem in the proof of the maximum prin-
ciple for harmonic functions.
Theorem 41. Every harmonic function f on the connected manifold Mn C
R" attaining its maximal value in the interior of Mn is constant.
Proof. Denote by m the maximal value of f and by Q = {x E Al"\8AI" :

f (x) = m} the set of all inner points of M" at which f attains this maximal
value. By assumption 1 is a non-empty closed set in M"\811". Hence it
suffices to prove that 1 is an open subset of M"\8M". Choose a point
xo E S2 and a radius Ro such that the ball D"(xo, R0) with center x0 and
radius Ra is completely contained in Mn\8M". By Gauss' mean value
theorem
VOl(S1
f (x0) = n-1) Jsn_l f (x0 + Ro y) dSn-1(y) m. = .f (xo)
This implies that f is constant and equal to m on the sphere with center
x0 and radius Ra. This observation can also be applied to each radius
RR < Ro below R0. Together this implies that f - m is constant on the ball
D"(x0i R0), i. e., D" (x0, Ra) is contained in Q. Thus Il is an open subset. 0
Eventually we will prove one more application of the Poisson formula, Li-
ouville's theorem for harmonic functions.
Theorem 42. Every harmonic function f : 1[P" R bounded from below
(above) is constant.
Proof. Changing f, if necessary, by adding a constant, we can suppose

without loss of generality that the function f is non-negative. Fix a point
xo E Rn and choose the radius R such that xo lies in the ball D"(0, R). By
the Poisson formula, Gauss' mean value theorem, and the assumption f > 0
we have
Rn 2 R' - II 2
f (xo) = . f (R
y) dS"-' (y)
vol(Sn-1) Jsn IIo-R yll In
Rn-2 R2 - I IxOI I2
< vol(Sn-1) Jn-i IIIxoII - IIR ylll
n .f (R y) dSn- ' (y)
_ 2( 2 - II oIIRIn
2
f(R y) dSi-1(y)
vo Sn-RIIlxoll Isn
-2(R2 _
- iRI" f(0)
RSn )IIIxoII
Taking the limit for R -* oo yields for all points xoi E R" the estimate
f(xo) <- f(0),
and the maximum principle for harmonic functions implies that f is con-
stant.
3.11. The Laplacian on Differential Forms

This section is a supplement to the vector analysis on manifolds as explained
before. We will discuss the Laplacian acting on forms and its properties.
To this end, we will start from an oriented manifold Mm of dimension m
together with a family of non-degenerate scalar products
g=:TXM' xTIMR
and assume that these scalar products depend smoothly on the point. This
is understood to mean that for any two smooth vector fields V and W on
M"' the function g(V, W) is smooth. Such a family {gx)xEAfm is called a
pseudo-Riemannian metric on M. In general, it has to be neither positive
definite nor the restriction of the scalar product on the ambient euclidean
space.
In this situation the tangent space is a real oriented vector space. and we
denote by k the index of the scalar product. According to Chapter 1, at each
point there exist a volume form dMm(x) E A ,'(M') and a Hodge operator
* : A (Mm) - Ax'-'(Mm) '
Together we obtain an m-form dMm on the manifold, as well as a *-operator
associating with every i-form a smooth (m - i)-form. The algebraic rules
3.11. The Laplacian on Differential Forms 101
from Chapter 1 can still be applied. We will use these and Stokes' formula
to determine the operator b : S i+1(Mm) -+ 1'l (Mm) adjoint to the exterior
derivative d : S1'(MTh) - S1'11 (M') acting on i-forms. As is well-known,
the adjoint operator is characterized by the requirement
(w',17'+1) . dMm = ! (wi bi7i+1) . dMm .
J m f)fm
Here wi and rli+1 are forms with compact support disjoint from the boundary
of M', and (dwi, gi+1) denotes the scalar product induced on forms. By
Theorem 5 in Chapter 1 we have
(dwi dMm = (-1)k&' A (*1]i+1)
= (-1)kd(wi A (*,ni+1)) + (_1)k+i+lwi A (d * 17i+1)
Integrating this equation and once again applying the algebraic rules, we
obtain
(dwi,,gi+') , dIim = (_I)k+i+1(-1)i(m-i)+k
wi A * * (d * 37i+l)
fm - J Mm
= (_1)$(m-i)+i+1+k (wi *d * qi+1) , dMm.
nMm
Thus we have proved

Theorem 43. The operator 6 adjoint to the exterior derivative d : S2i(Mm) --,
11i+1(Mm) on a pseudo-Riemannian manifold of index k is given by
b(lli+1) = (-1)k+mi+l * d * ni+1
Definition 22. Let Mm be a pseudo-Riemanniani manifold. The Hodge-

Laplace operator acting on i-forms, A: S2i(M') 1'(M'), is defined by
°db+bd.
Example 35. For a function wO = f, in the sense of this definition,
°f = bdf = (-1)k+l *d*df,
and, in the case of a positive definite scalar product, ° coincides with the
Laplace-Beltrami operator (see the remark following Theorem 11, §3.2).
Considering, on the other hand, the pseudo-euclidean scalar product of sig-
sion
°f = i 2f f
nature (n, 1) in Rn+1, we obtain for the Hodge-Laplace operator the expres-
(92x1 +''' + a2xn

a2f
a2xn+1
Rn+1 with this pseudo-Riemannian metric is called Minkowski space. Hence,
its Laplacian on functions is the wave operator (also known as the d'Alem-
bertian).
Definition 23. An i-form w' defined on a pseudo-Riemannian manifold is

called harmonic if 0(w') = 0; it is called co-closed if bw' = 0.
In the following theorem we collect the properties of the Hodge-Laplace
operator immediately resulting from the definition.
Theorem 44.
(1) Let w', if be forms with compact support disjoint from the boundary
of the manifold. Then
(A (u;'), rl') dMm = r (w',0(rl')) dJV1-
Im-
(2) Aod=doA and Dob=boo.
Finally, consider a compact manifold without boundary endowed with a
Riemannian metric (k = 0). Then the second order differential equation
0(w') = 0 is equivalent to two first order differential equations.
Theorem 45. Let MI be a compact, oriented manifold without boundary,
endowed with a positive definite metric. Then the following conditions are
equivalent:
(1) w' is a harmonic form, A(w') = 0;
(2) w' is a closed and co-closed form, dw' = 0 = bw'.
Proof. If w' is harmonic, then
f
0= r f'^ J 5fr
J M'"
This implies dw' = 0 = bw', since the metric is positive definite. 0
Corollary 4. Let Mm be a compact, oriented manifold without boundary,
endowed with a positive definite Riemannian metric. If every closed 1 -form
on M' is exact, then there exist no non-trivial harmonic forms.
Proof. In particular, a harmonic 1-form wl is closed, and hence wl can be
represented as the differential of a function, wl = df. Since
0 = bwl = bdf = 0(f) ,
f is a harmonic function, and by Hopf's theorem (Corollary 2) it is constant
on each connected component of M. Therefore, its differential vanishes,
w'=df=0. 0
Example 36. The assumption of this corollary is, for example, satisfied for
manifolds Mm which can be represented as the union M' = U U V of two
open sets U, V C M'", where
3.11. The Laplacian on Differential Forms 103
(1) the intersection U fl V is connected,

(2) U and V are diffeomorphic to star-shaped domains in R"'.
In fact, Poincare's lemma implies the existence of two functions ft, : U - R

and fi : V R such that wl = dfu over U and wl = dfv over V. On the
intersection we then have
d(fu - fv) = 0,
and hence fuu - fi, = C is constant. Now
f(x) _ I fl(x)+C
ifxE V.
fu (x) if x E U.
is a uniquely defined function on AP" satisfying wl = df. For n > 2 the
sphere S' allows a decomposition of the required shape, and hence S" has
no harmonic 1-form with respect to any Riemannian metric.
Example 37. Consider the 2-dimensional torus in 1R'1.
T2 = {(xi, s2, x3. xi) E 1R4 : (x1)2 + (x2)2 = 1, (x3)2 + (x4)2 = 1 }
and denote by i : T2 R4 its embedding into 1R4. The 1-forms induced on
T2
wl:= i'(-x2dx1 + xldx2) and n1 := i*(-x4dx3 + x3dxa)
are harmonic. To prove this, we use the parametrization h : [0,27r] x
[0.2a[ -+ T2 defined by the formula
h(,jp, z') = (cos p. sin yp. cos i;'', sin r') .
In this chart,
h`(wl) = dp and h'(gl) = dv',
and hence wl and 171 are closed forms on T2. Computing the coefficients of
the Riemannian metric leads to the matrix
0
9iv 0 1 '
From this, it is easy to see that wi and ill correspond to each other via the
Hodge operator of the torus T2:
But then, wl and 91 are also co-closed, bwi = *d * wl = *dij' = 0.

Harmonic forms play an essential role in topology and global differential ge-
ometry. Here we formulate the main result of Hodge theory; a proof would
go beyond the scope of this elementary text. For a compact, oriented Rie-
mannian manifold without boundary, (Mm. g), we introduce the following
vector spaces:
(1) the space 7-l'(Mr,g) of all harmonic i-forms,

x' (M', g) = {w' : 0(w') = 0} ;
(2) the space E(Mr) of exact differential forms,
E'(M') = {w' : there exists an (i -1)-form n`-1 such that d7ji-1
= w'} ;
(3) the space CE'(Mm, g) of co-exact differential forms
CE' (M'", g) = {w' : there exists an (i + 1)-form tl'+1 such that 67i'+1 = w' }
Theorem 46 (Hodge's Theorem).
(1) The space lf(Mm,g) is a finite-dimensional vector space. It is
isomorphic to the de Rham cohomology HDR(Mm). For a fixed
metric, each cohomology class of closed forms contains precisely
one harmonic form.
(2) With respect to the L2-scalar product, the spaces W, E' and CE' are
orthogonal subspaces of the space Q'(M) of all square-integrable
i-forms and decompose it into the sum
fl' = W ®E' ®CE'.
From the Hodge decomposition we immediately obtain Helmholtz's theo-
rem for 3-dimensional compact manifolds without boundary. Geometrically
speaking, it states that each vector field can be represented as the sum of
a gradient field, a curl field, and a harmonic field. In particular, the vector
field is the sum of a curl-free and a divergence-free vector field. In the final
chapter we will discuss and prove an analogous result for vector fields on
R3 which decrease sufficiently rapidly at infinity within the framework of
electrodynamics in detail.
Corollary 5. For each vector field V on an oriented, compact, 3-dimensional
manifold M3 without boundary, there exist a function f , a vector field W
and a harmonic 1 -form ill such that
wsd(1) + wcurl(W) + 171
In particular, wy is the sum of a closed and a co-closed 1 -form.
Proof. Applying Hodge's theorem to the 1-form associated with the vector
field V, we conclude that there exist a function f, a 2-form (32 and a harmonic
1-form 771, together satisfying the equation
wv = df +0132+R1.
From §2.3, we know that the gradient of a function corresponds to the
differential 4f,
df = wsd(I)
1
Exercises 105
Let W be the vector field corresponding to the 1-form *i3`,

2 1
=: WW
The curl of the vector field W is the vector field determined by the equation
du) l 1
W = *wcurl(w)
But this implies
a2
= *d * = * W = * * Wcurl(W) = wcurl(W)
I 1 1
since in dimension m = 3 for a positive definite Riemannian metric the

relation ** = Id holds on the space of 1-forms. This completes the proof. 0
Exercises
1. a) The equation
(x1)2
a12
+...+ (xann2 = 1, where al, ...,a, > 0 are constants,
defines an (n - 1)-dimensional submanifold of R', the ellipsoid.
b) The equation x2 + y2 = r2 (r > 0 fixed) defines a two-dimensional
submanifold of R3, the cylinder.
c) Consider for every c E R the subset of R3 defined by
M, = {(x,y,z)ER3: x2+y2-z2=c}.
For which values of the parameter c is Al, a two-dimensional submanifold
of 1R3?
2. Prove or disprove the converse of Theorem 1: Let U C 1Rn be open, let

f : U - Rn-k be smooth, and denote by Al the zero-set of f, Al := {x E
U : f (x) = 0}. If the estimate rk D f (x) < n - k holds for at least one point
x E M, then M is not a submanifold.
3. Prove that the formula

x2 + y2 = a2 cosh 2(z/a), a > 0 constant
defines a two-dimensional submanifold of R3, which is, moreover, parame-
trized by
h(v',v2) = (acosh(vl/a)cosv2. acosh(v'/a)sinv2, vl)
This surface is called the catenoid in differential geometry.
4. Let f : C C be a complex polynomial f = Ek0 aiz` without double

zeroes. Consider for every natural number I > 2 the set
M = {(z,w)EC2: w'- f(z) = 0).
Prove that Al is a two-dimensional submanifold of C2 = ]R4. The assignment
(z, w) w uniquely defines a map G : M C satisfying G'(z, w) = f (z).
The set M is a so-called Riemann surface. The l-th root of f (z) is uniquely
defined on it.
5. Consider the special linear group SL(2, R) of all real-valued (2 x 2) ma-

trices of determinant 1:
SL(2,1R) = JA E M2(IR) : det A= 1 } .
Prove that SL(2, R) is a three-dimensional manifold, and that its tangent

space at the neutral element E is determined by
TESL(2,R) = {AEM2(R): trA=0}.
Hint: Apply Theorem 4 and the (well-known) formula det(exp A) = etr.a
6. Prove that the orthogonal group O(n,1R) consisiting of all orthogonal

(n x n) matrices
O(n, R) = {A E A A' = E }
is a submanifold of 1Rn2, and compute its dimension. Is the subgroup
SO(n.IR) of all orthogonal matrices of determinant I also a submanifold?
If it is, compute its dimension, and describe its relation to O(n, R).
7. The set of all unit tangent vectors T1S2 to the sphere S2,
T,S2 = {(x, v) E IR3 X 1R3 : I
= IIvII =1, (x, v) = 0 } .
is a submanifold of R3 x 1R3 which is diffeomorphic to the manifold

{(z'., z 3) E C3: Iz1I2
+ Iz2I2 + 1Z312 = 1, (z')2 + (z2)2 + (z3)2 = 0 } .
8. The product of a manifold M without boundary with an arbitrary mani-

fold N again is a manifold, and it has the boundary 8 (M x N) = Al x 8N.
9. The tangent bundle TM of every manifold M is always an orientable

manifold.
Exercises 107
10. The topological boundary Fr(A) of a subset A C R" is the intersection

of the closure of A with the closure of its complement,
Fr(A) := A n R' \A .
Prove:
a) If M' C iR' is a closed m-dimensional submanifold of Rm, then OM' _
Fr(M"').
b) If MI C 1R" (m < n) is a submanifold with or without boundary of
strictly smaller dimension, then always Fr(Mm) = Mm.
11. Find three vector fields of length one on the sphere S3 which are mu-
tually othogonal at each point.
12. Construct a vector field of length one on each sphere of odd dimension.
13. Compute the representation of the following vector field in polar coor-
dinates:
V (X, y)
= x2 + y2 (8/ax + 49/ay) .
14. Consider on the sphere S2 the vector field

V (X' y, z) = ((x, y, z), (-y, x, 0))
Prove that V cannot be the gradient field of any function f : S2 R. On
which more general fact is this example based?
15. Compute the local coefficients (gij) of the Riemannian metric as well as
the volume form of the following submanifolds in the respective charts:
a) for the sphere S" in the chart determined by the stereographic projection
h : 1R" -> S":
2 yi 2 y" 1y12 - 1
h(y', ..., y") ( Jyi2 + 1' ... +
JyJ2 + 1' 1y12 + 1) ,
b) for the pseudosphere, i.e., the surface of revolution defined by the trac-
trix in the parametrization
h(u',u2) = (asinu' cosu2, asinu'sinu2, a(cosu' +lntan(u'/2)))
16. Compute the formulas for the gradient of a function, for the divergence
of a vector field, and for the Laplacian on the sphere S2 in the coordinates
defined by stereographic projection.
108 3. Vector Analysis on manifolds
17. On an open subset of R3 (to be determined!) we introduce coordinates

by the chart
(rcosVcosv,rsin;pcos ,rsin;p)
on the parameter domain r E (0, oo), E [0, 27r], , E (-t'/2, o/2). Com-
pute the coefficients of the Riemannian metric, and the formulas for the
gradient, the divergence, and the Laplacian in these coordinates.
18. Let 111' and N' be k-dimensional manifolds. The map f : 111' , N'
is called angle-preserving or conformal if, for two arbitrary vectors w1, w2 E
TZM' and their image vectors vi = (W;) E TjiyiNA, the respective angles
coincide:
(y1, V2) _ (wl, w2)
]Iv1 11 -1111211 11w111. IIw211
If, in particular, (h, U) is a chart of the manifold Alk, the chart h is angle-
preserving if and only if the representation of the Riemannian metric in the
corresponding coordinates is a multiple of the k-dimensional unit matrix E:
(gtj (y)) = A(y) . E, ,\(y) > 0.
19. Prove the orientability of the sphere S", using stereographic projection
and Theorem 14.
20. Let AI' be a manifold, h : U _ M' a chart, and -y [a, b] --. h(U) C 111'
:
a curve in Afk completely contained in the image set of h. Represent the

curve 7 in the coordinates (h, U) as h-1 o ry(t) = (y1(t), ... , y'(t)). Prove
that the length of the curve can be computed by means of the coefficients
of the Riemannian metric g1, in the chart (h, U) via the formula
b 1/2
l(7) = f 9ti(7(t))d dtt) d dtt)

dl.
a
21. Let f : [a, b] , R+ be a positive function and let Ale = {(x. y, z) E R3 :

y2 + z2 = f2(x)} be the corresponding surface of revolution in R3. Prove
the volume formula
/b
vol(M2) = 2n / f (x) 1 + (f'(x))2 dx.
a
22. Compute the following surface integrals:

a) for M2 = (x, y, z E R3) : x2+y2+:2 = a2. z > 01:
ty -
Exercises 109
b) / (x2 + y2) 012.
where M2 is the boundary of the subset of 1R3

hr
described by the inequality x2 + y2 < z < 1.
23. Compute the following surface integrals:
a) f2(R) (xdyAdz+ydzAdx+zdxAdy);
b) [(y - z)dy A dz + (z - x)dz A dx + (x - y)dx A dy], where M2 is the
nJt2
boundary of the subset of R3 described by the inequalities x2 + y2 <
z2,0<z<1.
24. Let the 1-form w1 on the Riemannian manifold Mk be given in the chart
h : V - Mk by wl = Ej f, dy'. Prove the following formula for the adjoint
operator:
k
1 a
b(w) =
t ay, (119-9*3fi)
25. Let Mk be a compact, oriented manifold without boundary. A function
f : Mk -+ R is called an eigenfunction of the Laplacian with respect to the
eigenvalue A E R, if 0 f= A f. Prove:
a) If fl and f2 are eigenfunctions with respect to different eigenvalues,
1\1 A2, then
(fl, f2) L2 = JM k fl(x)f2(x) dMk = 0.

b) Let S2 C R3 be the sphere of radius 1, AR3 the Laplacian on 1R3, A
the Laplacian on S2, f : 1R3 -+ R a smooth function, and set r =
x2 + y2 + z2. Then
2
(A`3 Of
f)Is2 _ (arj +2 +°S2(fIS2)
c) Let P : IIt3 -y 1R be a homogeneous polynomial of degree m which
is harmonic, i.e. P satisfies OR'P(x) = 0. Prove that PIS2 is an
eigenfunction of the Laplacian on S2, and determine the corresponding
eigenvalue.
26. Let Mk C 1R" be a compact, oriented manifold, and assume that f

Mk x [0, oo) -+ 1R is smooth. The heat equation is
t)
ozf(x, t) =
af(x,
at
Prove that if f is a solution of the heat equation satisfying f (x, 0) = 0 for

all x E Mk and f (y, t) = 0 for all boundary points y E 8Mk and all times
t E [0, to], then f vanishes identically on the set Mk x [0, to].
27. Let Al"-1 be the boundary of a compact n-dimensional submanifold of

1R". and N(x) the exterior unit normal vector field. If V E 1R" is a constant
vector, then
f(N(x), V) dM"-1 = 0.
28 . Prove the following commutator relations for the Lie derivative:

[Cu, iw] = i(V,w] and [Gv, Lw] = G[v,wl .
Chapter 4
Pfaffian Systems
4.1. Geometric Distributions

For smooth functions f 1, ... , f,"_k : R' - R with linearly independent
differentials, the equations
1Vrk....,c,,,-k = {T E R' : J1 l2) = C1, ..., f,,,-k(X) = Cm_k I
define a smooth k-dimensional manifold. Linearizing this, in general non-
linear, system of equations by passing to the tangent bundle. we see that
this manifold is described by the system
Tblc,..... c,,,-k = {v E T lR" : df1(v) = 0, ... , dfm-k(v) = 0 1.
These equations determine at each point of IR"' a k-dimensional subspace
of the tangent space to II2'. The resulting family of subspaces can also be
described by different systems of 1-forms w1.... , wm-k. For example, if (hj)
is a matrix of functions with nowhere vanishing determinant, the 1-forms
w; = E ik hj - dfj satisfy
TMk.....cn,_k = {v E TIRt : w1 (v) = 0, ..., 0}.
The level surfaces, however, cannot be recovered from the knowledge of the
forms w, alone. In general, the problem arises under which conditions lin-
early independent 1-forms wl, ... , wm-k describe a family of k-dimensional
manifolds via the system of equations (Pfaffian system)
W1 = ... = wm-k = 0.
Frobenius' theorem provides a complete answer to this question. This chap-
ter is devoted to its proof and the discussion of various applications. In
111
112 4. Pfafhan Systems
Chapter 5. we will apply it to prove the fundamental theorems of curve and

surface theory.
Definition 1. A k-dimensional geometric distribution (or Pfaffian system)
on Al'" is a family Ek = {Ek(x)} consisting of k-dimensional subspaces
Ek(x) C Tj A1"' in the tangent spaces to Mm depending smoothly on the
points in the following sense: For each point xo E M"'. there exist a neigh-
borhood xo E U C Al' and vector fields VI, ..., Vk defined on U such that
Ek(x) coincides with the linear hull of the vectors VI (x), .... Vk(x) at every
point x E U.
Example 1. Every nowhere vanishing vector field V on AI' induces a one-
dimensional distribution El(x) formed by all multiples of the vector V(x).
Conversely, every one-dimensional distribution El is locally determined by
a nowhere vanishing vector field.
Example 2. The linearly independent 1-forms wl, ... , w,n_k on Al' deter-
mine a k-dimensional distribution by
Ek(x) = {vETAf' : wl(v)=...=W,,,_k(v)=0}.
In analogy to the integral curve for a vector field, we now introduce the
notion of an integral manifold for a distribution.
Definition 2. Let Ek be a k-dimensional distribution on M'T'. A k-dimen-
sional submanifold Nk C Mm is called an integral manifold of Ek if the
tangent spaces of Nk coincide with the spaces of the distribution:
T,XNk = Ek(x) for all x E Nk .
Definition 3. The k-dimensional distribution Ek is called integrable if it
admits at least one integral manifold through each point x E Al'.
Example 3. For a distribution Ek defined by linearly independent 1-forms
W I , . . . , W,, -k as in Example 2, a submanifold i : Nk --+ Al"' is an integral
manifold of Ek if and only if the restrictions of the forms to Nk vanish,
i*(wl) = ... = i*(wm-k) = 0.
The local existence theorem for solutions of ordinary differential equations
can be formulated as follows:
Proposition 1. Every one-dimensional geometric distribution is integrable.
Example 4. Consider the nowhere vanishing 1-form w = x dy + dz on 1R3
together with the 2-dimensional distribution determined by w,
E2 = {vETR3: w(v)=0}.
We show that this distribution is not integrable.
4.1. Geometric Distributions 113
Proof. Suppose that E2 is integrable. Then there have to exist an open

set W C 1R2 and a smooth map h : W R3 such that h'(W) = 0 and
rank (D(h)) _- 2. For example, one can choose for h a chart of the integral
manifold. In the coordinates of R3, the map h = (h', h2, h3) consists of
three functions, and the condition h`(W) = 0 is expressed on W C R2 by the
equation
0 = h' dh2 + dh3 .
Differentiating this expression and forming the exterior product with the
1-form dh2, we obtain
0 = dh' A dh2, 0 = dh2 A dh3 .
We then multiply 0 = h'dh2 + dh3 once again by dh' and take into account
the equation we already derived, dh' A dh.2 = 0, to arrive at
dh' A dh3 = 0.
In summary, all twofold products vanish, dhi Adhj = 0, and this contradicts
the assumption that the differential D(h) of the map h : W -. R3 has the
maximal rank two. Hence the 2-dimensional distribution in 1R3 defined by
w = x dy + dz cannot have an integral manifold.
For higher-dimensional distributions on a manifold the problem thus arises:
Under which conditions do they turn out to be integrable? The answer to
this question forms the content of Frobenius' theorem. In order to formulate
it, we also need the notion of an involutive distribution.
Definition 4. A distribution Ek on the manifold Mm is called involutive if,
for every pair of vector fields V, Won Mm whose values V(x), W(x) E Ek(x)
at each point belong to the distribution, the commutator [V, W](x) E Ek(x)
again has values in Ek.
Theorem 1 (Frobenius's Theorem). Let Ek be a k-dimensional distribution
on the manifold Mm defined by the (m - k) linearly independent 1 -forms
W1.... , Wm-k:
Ek = {VETMm: Wj(v)=...=Wm-k(V)=0}.
Then the following conditions are equivalent:
(1) Ek is integrable;
(2) Ek is involutive;
(3) for every point xo E Mm, there exist a neighborhood xo E U C Mm
and 1 forms Oij defined on U such that
m-k
dwi = ei j A W j
j=1
114 4. Pfaffian Systems
for1<i<m-k;
(4) for all indices 1 < i < m-k, the following exterior products vanish:
dwi A (Wl A ...
A Wm-k) = 0 -
Remark. The fourth condition occurring in Frobenius's theorem is called
the integrability condition for the geometric distribution or the correspond-
ing Pfaffian system.
Remark. A one-dimensional distribution is determined by m - 1 1-forms
W1, .... and then dWi n (Wl A ... A Wm-1) is an (m + 1)-form on the
m-dimensional manifold. This has to be zero for trivial reasons, and the
integrability condition of Frobenius' theorem is automatically satisfied; see
Proposition 1.
We first prove some of the simpler equivalences in Frobenius' theorem and
proceed as follows: (2) (3) . (4) and the implication (1) (4).
The next section will be concerned with the proof of the central assertion of
Frobenius' Theorem, i. e. the implication (3) (1).
Proof of the equivalence (3) (4). Suppose that there exist local 1-
m-k
forms Oij such that dwi = E 9ij A wj. Then
j=1
m-k
dWi A (WI A ... AWr_k) _ E Oij AWj A (WI A ... A Wm_k) = 0,
j=1
since the exterior square of any 1-form vanishes. Conversely, assume that
condition (4) is satisfied. In a neighborhood U C MI of the point xo we
extend the family of linearly independent 1-forms Wl, ... , Wm-k by adding
1-forms r1 j, ..., r1k so that the combined family, {wl, , Wm-k, 111, . . . , nk ),
forms a basis f o r A (Mm) at each point x of U. The 2-form dwi (1 < i <
m - k) can thus be represented as
m-k m-k k k
dwi = E CQ0 WQ A Wp + E E Da j - Wa A nj + E Eji - rlj ^ nl
Q,3=1 0=1 j=1 j,1=1
with functions CQo, DQj and Ejl. The condition dwi A (W1 A ... A Wm_k) = 0
implies
k
E Ejl-njA171A(W1A...AWm_k) = 0,
jd=1
and hence the coefficients vanish, Ejl = 0. If we introduce the 1-forms
m-k k
3=1 j=1
4.1. Geometric Distributions 115
the exterior derivative dwi takes the desired shape,

m-k
dwi = E Bia n wa . 11
a=1
Proof of the implication (3) (2). For any two vector fields V and W
with values in the distribution £k, we have wi(V) = wi(W) = 0. From
dwi(V. W) = V(wi(W)) - W(wi(V)) - wi([V. W]) we obtain
dwi(V,W) = -wi([V,W]).
If now condition (3) holds for the distribution £k, then
m-k
dwi(V,W) _ > Oji A wj(V, W) = 0.
j=1
and hence all 1-forms w1, ... , w,,,_k vanish on the commutator [V, WJ. There-
fore, this vector field takes values in £k, i.e.. the distribution £k is involu-
tive.
Proof of the implication (2) ; (4). Let £k be a k-dimensional involu-

tive distribution. The form dwi A (wl n ... Aw,,,_k) has degree (m - k + 2).
Inserting (m - k + 2) vector fields into this form, we can assume, without
loss of generality, that two of these vector fields have values in £k. Denote
them by V and W. Because of the involutivity of £k, we have
w;(V) = w;(W) = 0 and dwi(V,W) = -wi([V,W]) = 0.
Therefore, the exterior product dwi A (wl A ... A wm-k) vanishes on every
(m - k + 2)-tuple of vector fields.
Proof of the implication (1) : (4). The proof of this implication pro-
ceeds like the one before. Let £k be an integrable distribution, xo E All
a fixed point, and h : W -+ All a parametrization of an integral manifold
through this point. Then we have h*(wi) = 0 (1 < i < m - k), and this
implies h`(dwi) = 0. Thus the 2-form dwi vanishes on the k-dimensional
subspace £k(xo),
dwi IEk(zo)XEk(zo) = 0.
Inserting again (m - k + 2) vectors into the form dwi A (wi A ... Awm_k), we
conclude that at least two of these vectors-call them V and W-lie in the
subspace Ek(xo). Hence wj (V) = wj (W) = 0 and &&,i (V, W) = 0. As above,
we conclude that the form dwi A (wl A ... A Wm_k) = 0 vanishes.
4.2. The Proof of Frobenius' Theorem

The implication (3) = (1), which is the core of Fobenius' theorem, is a
local statement. Hence, without loss of generality, we can suppose that the
manifold Mm is an open subset of R. We will prove a slightly more general
result from which the proof of this implication immediately follows.
Theorem 2. Let w1i ... , w,,,_k be linearly independent 1 forms on an open
subset M"' C Rm such that
m-k
dwi = E B.ij A wj
j=1
for certain 1 forms Oij. Then there exist at each point x E Mm a neighbor-
hood U C Mm of x and functions hj and fj defined on the set U satisfying
m-k
w= =
j=1
Proof of the implication (3) (1). Let £k be a distribution with the

property stated in condition (3). By Theorem 2, we can represent the forms
w; in a neighborhood U of an arbitrary point zo E M' as
m-k
j=1
for certain functions. By assumption, the 1-forms w1, . . . , w,,,_k are linearly
independent. Thus the differentials dfli ... , df,,,_k are linearly independent
as well, and the set
Nk = (X E U : f1(x) = fi(xo), ..., fm-k(X) = fm-k(XO) }
is a submanifold containing the point xo E M. At an arbitrary point
x E Nk, we determine the tangent space:
TxNk = {v E TM' : df1(v) dfm_k(v) = 0 }
C {v E TM' : w1(v) _ ... = w,,,-k(v) = 0) = Ek(x).
For dimensional reasons, the vector spaces coincide, i.e., Nk is an integral
manifold of the distribution £k through the point xo E Mm, and thus the
integrability of the distribution £k is proved. 0
Proof of Theorem 2. We proceed in two steps. First we reduce the proof
to the case of a system of 1-forms w1, ... , wm_k in a special normal form.
To this end, we represent the euclidean space R n as a product RI = Rk X
am-k and denote its points accordingly by y = (y', ... , yk) E Rk and z =
4.2. The Proof of Frobenius' Theorem 117
(z1,... , zm-k) E Rm-k. It is sufficient to prove the claim in Theorem 2 for

1-forms of the following special type:
k
wi = dz`-1: Aii(y,z)'dy' (1<i<rn-k).
j=1
In fact, representing the linearly independent 1-forms wi in coordinates,
in
wi = E Pia da ° ,
a=1
we may first assume that the determinant of the part of the matrix (pij)
containing the first (m - k) columns and rows does not vanish. Let (qjj) be
the inverse of this matrix, and consider the 1-forms
m-k m-k in in mk
-
w{ _ gia'wa = giaPa3d23 = dr'- E d.. 3 giapa3
0=1 a=1 3=1 3=m-k+1 0=1
Thus the forms w; are of the special type above. Moreover,
m-k rn-k m-k m-k
dwi = E dgia A wa + qia . d41a = dgia A wa + Y, qia ' Ba 3 A W3
a=1 a=1 a=1 a,3=1
m-k (M- k m-k
_EE
a=1 3=1
dgia + E gi03ae3ry
3.y=1
A wa,
and the system of forms wi, ... , wm-k also satisfies the condition
m-k
dw; _ B a A UJ
a=1
for certain 1-forms O . If Theorem 2 now holds for this system of forms, we
obtain functions h,j and fj* such that

= m-k
w; > h,j dfj .
j=1
Rewriting this leads to
m-k m-k m-k
Wt = Pia ' wa = E E 11aj ' Pia ' dfi
a=1 a=1 j=1
Summarizing, it is sufficient to prove Theorem 2 for forms of the type
k
wi =
j=1
118 4. Pfafan Systems
Without loss of generality, we may choose the origin, x = 0 in Rm, as the

point x = (y, z) in whose neighborhood we want to represent the forms w;
in this special way. For fixed parameters (y', ...,y") E Rk near 0 E Rk, we
consider the ordinary differential equation with initial conditions
k
(t) _ > A;j(t - y, z(t)) yt, z'(0) = z', 1 < i < m - k.
j=1
Denote by F` (t, y, z) the solution of this differential equation that is uniquely
determined near zero. These are m - k smooth functions in the variables
t E R, y E 1Rk and z E ]R"'-k defined on a neighborhood of zero in R x RI.
Moreover, for small values of the parameters we have F'(µ t, y, z) _
F'(t, p y, z), since the functions
G'(t,y,z) F'(µ't, y, z)
satisfy the initial conditions G'(0, y, z) = z', and they are also solutions of
the differential equation corresponding to the vector p y,
k
Y, Z) = p'F"(p.t,y,z) _ A:,(lA't_
j=1
We introduce a coordinate transformation near zero in R- = ]Rk x ]R--k by
means of the equations
u .= y, F(1,u,v) := z.
The Jacobian of this transformation is
ay,
a(u,v)
zIo 0 E0lJ
((
LE
since, by the differential equation and the initial condition,
az' I = OF' (1, O, v) I = 8v'
0v j
8 .7 &0 0 = .
o 0
This observation shows that both these equations determine a local diffeo-
morphism from the space RI to itself near zero. We represent the 1-forms
WI, ... , wm-k in the {u, v}-coordinates:
m-k k
w; _ Bi j (u, v) dut + j:Pij (u, v) - duj .
j=1 j=1
Obviously it suffices to prove that the functions P;j - 0 vanish identically.
First we derive the identity
k
0.
j=1
4.2. The Proof of Frobenius' Theorem 119
To this end, consider the map 4i : R' x Rk(t, y) -p Rk x R' -k(u, v) from the
space R1 x Rk with coordinates (t, y) to the space Rk x R' -k with coordinates
(u, v) defined for a fixed point v E R'"-k by the formula 40(t. y) := (ty, v).
Then
k
V (wi) = E Pig (t y, v)(t dyi + y dt) .
j=1
On the other hand, with respect to the (y, z)-coordinates on R, the map
4i : R' x Rk(t, y) - Rk x R"'-k(y, z) is determined by
4'(t, y) = (t y, F(1, ty, v)) = (t y, F(t, y, v))
From the normal form,
k
wi = dz' - Aid 91,
j=1
we obtain for the induced form V(wi) the new expression
k OF'
V (w;) =
OF'
at
dt+E-y k
j=1 j=1
Comparing in 4' (wi) the coefficient at dt and taking the differential equation
into account leads to the identity
k k
0=
.i=1 i=1
In the final step of the proof of Frobenius' theorem, we now show that the
functions Pij = 0 vanish. To do so, we again resort to the map 4'(t, y, v) :_
(t y, v), but consider it this time as a map from R x Rk x RI-k 9 (t, y, v)
to Rk x R°'-k 3 (u, v) (v is no longer assumed to be constant). The identity
already proved implies the formula
m-k k
4)*(wi) =
j=1 j=1
We denote by Pi*j the function

P*j(t. Y, v) := t Pip(t y, u).
Then
k
4 (wi) _ E P'j(t, y, v) dyt + (terms in dv'},
i=1
120 4. Pfafan Systems
and for the exterior derivative we obtain the expression

k 8P
d-P`(wi) _ dt n dyj + {terms without dt}.
j=1
at
m-k
By assumption, there exist 1-forms satisfying dwi = F, 9,1 A wj, and we
j=1
represent these as V (9ij) := H,jdt + {terms containing dy', dvj}. This
leads to a homogeneous differential equation for the functions we are inter-
ested in:
ps m-k
E H.PQ>
Q=1
with the initial condition P,j(0, y, v) = 0. The only solution of this homoge-
neous differential equation with the given initial condition is P;j (t, y, v) - 0,
and this immediately implies Pij(t y, v) = 0. Thus the proof of Frobenius'
theorem is completed. 0
4.3. Some Applications of Frobenius' Theorem
The simplest case is that of an (m-1)-dimensional distribution £'"-1 on an
m-dimensional manifold Mm. If Eii-1 is defined by one nowhere vanishing
1-form w, the integrability of the distribution reduces to the condition that
the 3-form dw n w vanishes,
dwAw = 0.
The method to explicitly integrate this (m - 1)-dimensional Pfaffian system
is based on looking for a so-called integrating factor and an application of
Poincare's lemma.
Definition 5. An integrating factor for the 1-form w is a nowhere vanishing
function f : Mm IR such that the 1-form f w is closed,
d(f . w) = 0.
Theorem 3. Let w be a nowhere vanishing 1 form on the manifold Mm.
(1) If there exists an integrating factor for w, then dw n w = 0. In this
case the distribution Em-1 is integrable.
(2) If dw Aw = 0, then there exists an integrating factor for the 1-form
w in a neighborhood of each point in M'".
(3) Locally, the integral manifolds of the distribution £'n-1 are the level
surfaces of the function g determined from the integrating factor f
via the equations
d(f w) = 0, f -w = dg.
4.3. Some Applications of Frobenius' Theorem 121
Proof. The equation d(f w) = 0 implies df A w + f dw = 0. Multiplying

this equation once again by the 1-form w leads to f dw A w = 0. Since
f # 0, we obtain dw A w = 0 as a necessary condition for the existence of an
integrating factor. If, on the other hand, dw A w = 0, then the existence of
an integrating factor follows immediately from Theorem 2, §4.2.
In dimension m = 2 the 3-form vanishes, dw A w = 0, for purely algebraic

reasons. Thus
Corollary 1. Every nowhere vanishing 1-form on a 2-dimensional manifold
locally has an integrating factor.
Example 5. Consider in R2 the differential equation
P(t, X) + Q(t, x) i = 0.
Near a point (to,xo) E 1122 at which P and Q do not vanish simultaneously,
we have the 1-form
w=
and its integrating factor f (t, x). The equivalent differential equation
(f P) (t, x) + (f Q) (t, x)± = 0
is called the total differential equation, and the solution curves are implicitly
determined by the equation
g(t, x) = const
with dg = f w. Frobenius' theorem now claims that it is always possible
to solve the original differential equation by using the outlined method. It
does not, however, provide an algorithm for finding the integrating factor.
In simple cases this may be computed directly. If we can find, e. g., functions
F(t) and G(x) depending only on the variables t, x and satisfying
OP(t, x) OQ(t, x) = Q(t, x)F(t) - P(t, x)G(x) ,
Ox - at
then f (t, x) = of F(t)dtef G(x)dx is an integrating factor.
Example 6. Consider the differential equation
(2t2 + 3tx - 4t)± + (3x - 2tx - 3x2) = 0.
We can choose F(t) = 2/t and G(x) = -5/x and obtain the integrating
factor f = t2x-5. The solutions of this differential equation computed using
this integrating factor are the curves described by the equation
t3x-4 - 2 t4x-4 - t3x-3 = coast
(compare the figure on the next page).
-a .4 -0.2 0 0.4
The curves t3 - zt2

- tax = cx4 for different values of c
Assume now that on the manifold M'" we are given a Riemannian metric
(, ) as well as an (m -1)-dimensional integrable distribution E'-1 described
by the 1-form w. Denote the vector field associated with the 1-form via the
Riemannian metric by W. This is uniquely determined by either of the two
equivalent equations
*w = W j dM' or w(V) = (V, W) .
Using the vector field W, the distribution E'"-1 can be described as

E"'_1
= {v E TM' : (v, W) = 0) ,
and hence W is orthogonal to each integral manifold Nin-1 of the distribu-
tion. Normalizing the length of the vector field W to one, the volume form
of each integral manifold is given by the formula
dN'"-1
= -
W i dM'" .
We study the behavior of the integral manifold N'"-1 C Mm of the dis-

tribution under the flow 4b:: M'" -i Mm of the vector field W. Compute
first the Lie derivative of the 1-form w with respect to the associated vector
field W. Since w(W) = 11W112, the formula £w(w) = d(W J w) + W i (dw)
implies
(Gw(w))(V) = W(w(V)) - w([W, V))
In particular, for a vector field V tangent. to the distribution this formula

simplifies to
(,Cw(w))(V) = - w([W, V1) (W, [W, V]) ,
and we arrive at
Theorem 4. The flow 4it of the vector field W maps an integral manifold
of the distribution E"`-1 to another integral manifold if and only if
V(IIW112)+dw(W,V) = -(W, [W,V]) = 0
for every vector field V on M with values in £'"-1
Corollary 2. Let the distribution Em-1 be defined by the closed 1 -form w.

Then the flow of the dual vector field W transforms integral manifolds into
integral manifolds if the length I I W I is constant on every connected integral
I
manifold.
Corollary 3. If the distribution E'"-1 is defined by a 1 form w of constant
length. and if, moreover, the flow of the dual vector field W transforms
integral manifolds into integral manifolds. then dw = 0. In this case 6'-1
locally consists of level surfaces of a function whose gradient has constant
length.
Proof. First, Theorem 4 implies W . dw = 0. At a point x E Alm, we

choose an orthogonal basis e1, ... , e,,, in the tangent space so that W is
proportional to el . W = a el. Denoting by al, ... . o. the dual basis, the
form w = a of is proportional to al. Represent the 2-form
du; = Eb;j o1Aa
t<J
in this basis. Since 0 = dw A w, the 2-form dw only contains the in - 1

summands dw = b12 Q1 A (72 +... + bi 2al A a,,,. But then the condition
0 = W-i dw = a(b12a2+...+bjmam)
means that w is a closed form.
We compute the infinitesimal volume change of a compact integral manifold
N'"' 1 under the flow 4)t of the vector field W. The Lie derivative of the
volume form dN1-1 is
,Cw(dA"-1) = £w(III .WJdAm) = WJd(IIWII.WJdMm)
_ (div(W) - 2 W(ln IIWII2)) dN'"-1
and hence we obtain

Theorem 5. The derivative of the volume change of a compact integral

manifold N` of the distribution Em-' under the flow of the vector field
W is given by the formula
dt
(vol(Ft(Nm-1)))
It=o =
f (div(W) - 2 - W(ln IIWII2)) - dNm-'
If, in particular, the distribution Em-' consists of the level surfaces of a

function f : Mm -' IR, and we choose the gradient of this function as the
vector field, W = grad(f), then div(W) = 0(f) is the Laplacian of f.
Corollary 4. The volume change of a compact level surface N` of the
function f : Mm R under the flow of the gradient vector field is given by
the formula
d (Vol(4 (Nm-1))) 1, =0 = JN^ (°(f) - -grad(f) In ilgrad(f)I12) -dNm-'.

1
Remark. In all these formulas the Laplacian, the divergence, and the gra-
dient are taken with respect to the manifold.
Example 7. Consider a function f : Mm - R and assume that there is
another function µ : Mm -+ IR such that
d(Ilgrad(f)112) = 2p . df
By Theorems 4 and 5, the flow of the gradient vector field grad(f) maps
level surfaces of f to level surfaces, and the volume change is described by
the formula
(0(f) - µ)dNm-1
Wt
(vol (41(Nm-1))) It=o =
J vm
For example, the spheres S'-'(R) C R" are the level surfaces of the func-
tion f (x) = IIxII2, and we obtain Ilgrad(f )II2 = 411x112 as well as 0(f) = 2m.
Hence 0(f) - µ =_ 2(m - 1) is constant, and for the flow 4 (x) = e2t . x
we obtain the following differential equation describing the evolution of the
volume:
wt-
The second application of Frobenius' theorem will play an important role in

the chapter devoted to surface theory.
Theorem 6. Let 0 = (wig) be a (k x k) matrix of 1-forms defined on a
neighborhood of 0 E Rm, and let AO be an invertible (k x k) matrix. In a
connected neighborhood 0 E V there exists a (k x k) matrix A = (fib) of
functions satisfying
Il = dA - A-' and A(O) = Ao
if and only if
d12 = 12 n Q.
In this case, the function matrix A is uniquely determined. If, in addition,
11 is an anti-symmetric matrix (12+12' = 0), and AO is an orthogonal matrix
(Ao Aa = E), then the solution A(x) is also orthogonal at each point of the
set V.
Proof. First, the condition d12 = 1Z A 12 is necessary for the solvability of

the equation Q = dA A-1. In fact, dA = SZ A implies
0 = ddA = d(f?-A) = (dIl)-A-I1AdA =
The matrix A is invertible, and hence d12 = fZA11. Uniqueness of the solution
is also easy to see. For two solutions A(x) and B(x) we have, e. g.,
d(B-1) = -B-1
(dB) - B-',
and hence the differential d(B-1 . A) vanishes:
B-1
d(B-'
- A) = d(B-1) - A+ B-1dA = -B-1
(dB) A+ B-1 - 12 A
_ 0.
Thus B-1 A is constant, and at the point x = 0 it is equal to the unit
matrix. This implies A(x) = B(x) for all points x E V. Now we prove
the existence of a solution under the condition d11 = 12 A Q. To this end,
consider the following (k x k) matrix of 1-forms on the space Rm x IItk2 with
coordinates (x1, ... , x"', z') ):
k-
A=
r=1
From d12 =12 A f? we obtain

dA = ddZ-d1ZAZ+flAdZ = -12A12AZ+SlAdZ
= -1lAQAZ+1lA(A+12AZ) = fIAA.
k
The system of forms dz'j - E Wjrzr3 is linearly independent in IR x Rk2,
r=1
and by Frobenius' theorem, there exists an m-dimensional integral manifold
Mm c IRt x Ht k2 through the point (0, A0) E Ht'" x Rk2. The tangent space
T(o,Ao)Mm to this integral manifold has only the null vector in common with
Rk2
ittk2
TiO,AaiMm n = {0} .
This follows directly from the shape of the form A, since the tangent space to
Mm is determined by the equation A = 0. Then the integral manifold Mm
is the graph of a map A : W -+ Rk2 defined on an open set 0 E W C R'"

satisfying the initial condition A(O) = Ao. From
A'(A) =
we see that the (k x k) matrix A is the solution of the differential equation
we looked for. The remaining statements of the theorem follow from the
formula
d(At A) = (dA)t A + At dA = At 52t A + At f2 - A .
In fact, if f2 is an anti-symmetric matrix, and dA = f2 A is a solution of the

differential equation, then we immediately conclude that d(At - A) = 0. O
Exercises
1. Consider on R3 the 1-form

w= Idx+Idy+1dz.
yz xz xy
Prove that the distribution defined by w = 0 is integrable. Find, moreover,
an integrating factor for w, i.e. , a function f such that d(f w) = 0.
2. Let P and Q be functions in the variables x, y. u, v. Consider on RR the

1-forms
wi = w-2 =
a) What are the conditions for the distribution wi = w2 = 0 to be inte-
grable?
b) Let the function f : C2 --+ C be defined in the complex variables z
x + iy, w := u + iv by f (z, w) = P + i Q. Prove that the integrability
conditions for the system wi = w2 = 0 are, in particular, satisfied when
f is a holomorphic function.
3. Prove that in R3, a distribution defined by the form

w=
P Q R
det NU =0
[PQRJ
Exercises 127
vanishes. This formula is to be interpreted in the following way: Write the

determinant out and apply each of the resulting differential operators to the
respective function.
4. Solve the initial value problem i = ex cost, x(O) = xo

a) by separation of variables;
b) by finding an integrating factor.
5. The method of solving a differential equation by finding an integrating

factor is of interest, in particular, if separation of variables is impossible.
As an example, solve the differential equation i(x3 - tx) = 1, x(0) = xo.
How could this equation, nevertheless, be solved by a suitable separation of
variables?
6. Solve the differential equations

a)
b) sin t + ez + cost i = 0 ;
c) 4t3+6tx3 + (3t3x2 +3)i = 0 .
7. Let f = (w23) be a (k x k) matrix of 1-forms defined on an open neighbor-

hood of 0 E R'. Prove that locally there exists a (k x k) matrix A consisting
of functions satisfying
dA =
if and only if there exists a 1-form a such that the matrix E := SZ - a E
solves the equation dE = E A E .
Chapter 5
Curves and Surfaces in

Euclidean 3-Space
5.1. Curves in Euclidean 3-Space

The notion of a curve in 3-space R3 is not as simple as it may seem at first
glance. For a long time in the 19-th century, the common understanding
was that a curve is a subset A C 1[13 described in a continuous way by a real
parameter. In this sense, a curve was the surjective image of a continuous
map f : [0, 1] - A C R3. To require the injectivity of f is too strong, in
so far as it excluded curves with self-intersections. Apart from the fact that
the assumption of continuity is too weak to render such a "curve" accessible
to differential calculus. a much more essential reason shattered this notion
of a curve.
Theorem 1 (Peano 1890). There exists a continuous. surjective map from

the interval onto the square.
Proof. We construct the map f : [0, 1] - 10, 1] x [0, 1] as the limit of a

uniformly converging sequence of continuous maps f,. In the first step, we
divide the interval [0, 1] into nine subintervals of equal length and decompose
the square accordingly. The map f, is then to map each of these subintervals,
[(i - 1)/9, i/9], to the diagonal of the i-th square continuously and bijectively
(see the figure). The map f2 arises in the same way when we further divide
each subinterval [(i - 1)/9,i/9] as well as each subsquare. This results in
81 intervals and squares, and f2 maps the small intervals continuously and
bijectively onto the diagonals of the small squares
129
130 5. Curves and Surfaces in Euclidean 3-Space
7 8 9
5 4
1 2 3
0 1/9 9/9
so that f2([(i - 1)/9, i/9]) lies in the i-th square of the first step of the
decomposition. By construction, f1(i/9) = f2(i/9). This way we obtain a
sequence of continuous maps satisfying
(1) II f.+1(t)II < V 2-13n for all t E [0, 11;
(2) f,,(i/3n+1) = fit+1(i/3n+1) = fn+2(Z/3n+1) for all n and i < 3n+1.
61 6 63 64 65 78 80 81
60 /5 58 6 68 s 78 7 76
55 57 7 71 73 7 75
54
49
52
51
< 3 ( 38
41
3 36
4 31
35 34
3 33
48 4 46 4 44 4 30 29 28
7 8 9 11 25 26 27
6 5 4 14 24 23 22
1 2 3 17 19 2 21
The estimate in (1) implies the Cauchy condition,

1
11f
n (t) - rf (till <
- V/2- .3n-1'
for all t E [0, 1] and in > n. Thus, the sequence fn converges uniformly to
a continuous map f : [0, 1] [0, 1] x 10, 1]. By construction, all the points
(s , 3^') belong to the image of the map f. The interval [0, 1] is compact,
and hence the image set f([0,1]) is again a compact subset of the square.
But the subset of the image of f mentioned before is dense in the square
5.1. Curves in Euclidean 3-Space 131
[0, 1] x [0,1], and hence f ([0, 1]) = [0,1] x [0. 1], i.e. , f is continuous and
surjective.
This example had an essential impact on general topology. In fact. it shows
that the "dimension" of a topological space can increase under a continuous
map, and that this notion has to be made more precise (topological dimen-
sion theory). We will not deal with this problem here, and instead confine
ourselves to the case of smooth curves in R3. We could try to consider these
as smooth, one-dimensional submanifolds of 1183, but that would exclude self-
intersections of curves again. For this reason, we formulate the notion of
curve slightly more generally.
Definition 1. A (parametrized) curve is a differentiable map y' : [a, b] 1<83
from an interval to 1183 whose derivative vanishes nowhere.

It is easy to see that sufficiently small image sets of a parametrized curve
are submanifolds. Since dy(t)/dt 0 0, for a fixed initial point, the length
function
L(t) := fat II do (u) II dp.
is monotonously increasing. A point on the curve is uniquely determined

by the parameter describing the length of the curve segment from the ini-
tial point to the point considered. In other words, inverting the length
function, we obtain a parametrization -y o L-1 [0, L(y)] : 1R3 of the
curve y, and the tangent vector with respect to the length parameters has
length one, IIdy/dsIj = 1. This kind of parametrization is called the nat-
ural parametrization of the curve. The preceding observation shows that
the natural parametrization can be constructed starting from an arbitrary
one. We agree on the following: If the curve is given in any parametrization
y(t), then we denote by y(t), etc., the derivative with respect to the param-
eter t; if the curve is given in its natural parametrization, then -y(s) is the
derivative with respect to the length parameter s of the curve.
Example 1. Consider the helix in the parametrization
-y(t) = (a cos t, a sin t. b t) .
Since jjy(t)Ij = a' + 2 , the length function is L(t) = a + b2 t. From
this, we obtain the natural parametrization by passing from the parameter
t to the parameter s = L(t):
s s
(s) _
a2+b2 a2 +61
To every curve, we shall assign three linearly independent vectors at each
point of general type. The first of these vectors is the normalized tangent
vector.
132 S. Curves and Surfaces in Euclidean 3-Space
Definition 2. If y(s) is a curve in its natural parametrization. the vector

t(s) := y'(s) is called the unit tangent vector to the curve.
The curvature of a curve at a point measures the angle variation of the
tangent vector per length unit. More precisely:
Definition 3. Let C C R3 be a curve, and let p E C be a point on the curve.
The curvature K(p) of the curve C at the point p is the limit
tc(p) = lim 4
q-p pq
where o is the angle between the tangent vectors at the points p and q. and
pq denotes the length of the curve segment between these points.
Theorem 2. In the natural parametrization of the curve, :(s) = IIy"(s)II.
Proof. Setting p = y(s) and q = y(s + h), we obtain Tiq = h in the natural
parametrization of the curve, and the angled is computed using the formula
for the angle in an isosceles triangle,
sin(0/2) = IIt(s+h) - t(s)II/2.
Thus
0 0/2 2 sin(¢/2) = lim Il t(s + h) - t(s)II
lim = lim = I Iti (s)II
q--p pq h-.0 sin(0/2) h-0 h h-0 h
In an arbitrary parametrization of the curve, a straightforward parameter

transformation yields the following formula (see Exercise 1):
K(t) = II''(t) x y(t)II
Ily(t)II3
Definition 4. The principal normal vector h to a curve at a point of non-
vanishing curvature is the normalized derivative of the unit tangent vector,
h(s) := 1 t'(s) = y(S)

n(s) 11-y" (s)II
The vector product
b(s) := t(s) x h(s)
is called the binormal vector to the curve.
Thus, at each point of the curve with non-vanishing curvature, there exist
three mutually orthogonal vectors F(s), hh(s), and bb(s) all of length one, the
so-called Frenet frame of the curve. Finally, we introduce a last geometric
characteristic of a space curve, the torsion.
Definition 5. The torsion of a curve C c R3 of class C3 at a point of

non-vanishing curvature is the scalar product
d6(s)
r(s) _ -( ds
The formula for the torsion in an arbitrary parametrization is the subject

of Exercise 1.
The structural equations of a curve in euclidean space are the Prenet formu-
las expressing the derivatives of the P enet frame through this same frame.
Theorem 3 (Fundamental Theorem of Curve Theory). Consider a curve of

class C3 with nowhere vanishing curvature in the natural parametrization.
Then, in the above notation,
d t(s) 0 0 t'(s)
= K(s)
d h(s) -ac(s) 0 r(s) h(s)
b(s) 0 -r(s) 0 b(s)
Let y, [0, L] R3 be two curves with coinciding curvature and torsion

in their natural parametrizations. Then there exists a euclidean motion A :
1R3 - R3 such that y`(s) = A o y(s). Let ac(s) be any positive function, and
let r(s) be an arbitrary function, both defined on the interval [0, L]. Then
there exists a curve whose curvature and torsion are the given functions.
Proof. By the definition of the principal normal vector, we have t'(s) =

K(s) K(s). Since 11h(s)II __ 1, the derivative k(s) is orthogonal to K(s).
Hence it can be represented as a linear combination of t(s) and 9(s). The
computation
(dh(s), t(s)) _ (h(s), t(s)> - (h(s), j.f1> = -K(s)

s
and the definition of torsion imply the second of the Frenet formulas. But
then
d9(s) _ O(s) dh(s)
x h(s) + t x = _r(s) h(s).
ds ds ds
Viewing the vectors of the Frenet frame as row vectors,
t(s)
A := h(s)
b(s)
defines a (3 x 3) matrix A of functions on [0, L] satisfying St, where
the entries of the skew-symmetric matrix Q of 1-forms are the curvature and
the torsion of the curve:

0 a(s) 0
Q= -K(S) 0 r(s) ds.
0 -r(s) 0
Existence and uniqueness of a curve forgiven curvature and torsion functions

thus follow immediately from Chapter 4, Theorem 6, since the integrability
condition. dil - Q A Q = 0, is satisfied for trivial reasons. Note that the
solution of the system of linear differential equations, dA = Q A. is defined
on the whole interval on which the curvature ac(s) and the torsion r(s) are
given. Having first determined the matrix A(s) for the prescribed functions
K(s). r(s) from the equation dA = Q A, we obtain the curve -y(s) by one
more integration.
7(s) = Jt(s).ds.
Remark. The plane spanned by the tangent vector f and the principal
normal vector h at a given point is called the osculating plane of the curve
at that point. The binormal vector 6 is perpendicular to this plane. The
third Frenet formula implies Ir(s)I = 110(s)II, and hence the absolute value
Ir(p)l of the torsion of a curve at a point p can be described as the limit
IrQ)I = v m
where v denotes the angle between the osculating planes to the curve at the
points p and q (compare the proof of Theorem 2). Thus the absolute value
of the torsion measures how much the curve "winds out" of the osculating
plane.
Next we discuss some curves with special properties, and explain how the
Frenet formulas can be applied to study them.
Definition 6. A curve whose tangents form a constant angle with a fixed
direction in R3 is called a slope line'.
Straight lines and helices are slope lines. These can be characterized by the
fact that the quotient r(s)/tc(s) is constant.
Theorem 4 (Lancret, 1802). A curve of class C3 with nowhere vanishing
curvature is a slope line if and only if r(s)ln(s) is constant.
Proof. If there exists a vector a E R3 such that (t(s), a) is constant, then,
differentiating this equation, we obtain that the scalar product (h(s). a)
1 In German. such a curve is called Boschungslinie.
vanishes, since K(s) # 0. Differentiating once again leads, using the Frenet
formulas, to
T(s) (6(s), a = c(s) .
The vector d lies in the if, 9}-plane, and (t, a) is constant. Hence (9, d) is
also constant, and thus the quotient T/rc is constant. Conversely, if r/K is
constant, then consider the vector
T t(s) + 6(s).
K
The Frenet formulas imply
T h - T h = 0,
ds I£ ds + ds =
and hence a is a constant vector. The scalar product (t(s), d) = T/rc is also
constant, and the curve is thus a slope line. O
Curves lying on a sphere in 3-space can be described by a similar relation

between curvature and torsion.
Theorem 5. A curve of class C4 with nowhere vanishing curvature and
torsion, K, T 96 0, lies on a sphere of radius R > 0 if and only if it satisfies
,i
_I + 2
= R2
K2 rc2T
in the natural parametrization.
Proof. Differentiating the equation II-'(S)II2 ° R2, we obtain (t(s), -Y(s)) _

0, and hence y(s) is a linear combination of the vectors h(s) and 6(s),
y(s) = a(s) h(s) +,13(s) b(s) .
Furthermore, IIy(S)112 - R2 implies a2(s) + f32(s) R2. Differentiating
(t(s), y(s)) - 0 again leads to K(s) (h(s), y(s)) + 1 0, and thus
a(s) =
We differentiate the equation c(s) (h(s), y(s)) + 1 - 0 and obtain the
following relation by a simple transformation:
a(3)
2s)
ic'(
I£ (S)T(S)
The asserted necessary condition for a spherical curve then immediately

follows:
r 2
R2 = a2(s) + $2(s) =
1
( K2(3)T(S)
a2(s) + /
If, conversely, this relation between curvature and torsion holds for a C4-
curve, then we first differentiate it and obtain the equation
r(s) d KG(s) _
0.
/G(S) ds !GZ(s)r(s)
Now consider the vector
.6
a(s) := 7(s) + h(s) (s).
;R-S) ,z() (s)

Using the Frenet formulas and the preceding relation between curvature and
torsion, we compute the derivative of the latter, and find that
id(s) = 0.
Hence d:= a(s) is constant, and
11-y(S)
- aI1z =
az(s) + \az(s)(s)/z = Rz'
i. e., the curve -y(s) lies on the sphere of radius R with center d. 0
Next we turn to plane curves. Note first that these can be described as the
curves with vanishing torsion.
Theorem 6. A curve of class C3 with nowhere vanishing curvature, k(s)
0, lies in a plane in R3 if and only if its torsion r(s) - 0 vanishes identically.
Proof. Let a' be a vector perpendicular to the plane in iR3 containing the
curve. All the tangent vectors t(s) lie in this plane; hence (t'(s), a) - 0.
Since ac(s) # 0, we immediately obtain (h'(s), a") - 0 by differentiating this
equation. Thus d coincides with the binormal vector 6(s). In other words,
the binormal vector 6(s) is constant. Then 0 = 6'(s) = -r(s) h(s) implies
that the torsion vanishes, r(s) - 0. The converse is proved analogously. 0
The curvature of a plane curve can be ascribed a sign. In fact, the principal
normal vector is proportional to the vector obtained by rotating the tangent
vector through the angle 7r/2 in the positive sense. The curvature of a plane
curve is ascribed the positive sign if the corresponding factor is positive.
Identifying R2 with the complex numbers, the rotation through 7r/2 in the
positive sense corresponds to multiplication by the number i E C. Using the
multiplication of complex numbers, this leads to
Definition 7. Let -y : [0, L] - C = R2 be a plane curve in its natural
parametrization. The plane curvature k(s) is the function k : [0, L) -' R
defined by the equation
ast(s) = k(s) i t(s) .
The absolute value jk(s)j of the plane curvature coincides with the curvature
K(s) of the curve viewed as a space curve.
Example 2. -y(t) = (t, ±t2) .
A closed curve i' : [0, L] - C = R2 is one that starts and ends at the same
point and whose tangent vectors at this point coincide as well, i'(0) = -Y(L)
and t(0) = F(L).
Theorem 7. Let [0, L] C be a closed curve. Then the integral
r ILL
2 J7 k(s)ds = 2
is an integer, called the winding number of the closed curve.
Proof. Consider the map t' : [0, L] C defined by
t* (s) = exp [k(u)du]

Then dt'(s)/ds = i k(s) - t*(s). and hence (t(s)/t*(s))' = 0. The tangent
vectors t(s) are thus described by the formula
F(s) = C exp Lk(u)duj
fL
for a certain constant C. Since t(0) = t(L), the number J k(s)ds is an
o
integral multiple of 27r. 0
We conclude the section on curves with the discussion of the Fenchel in-
equality, which claims that the total curvature of a closed space curve is
bounded from below by 2;r. We start by considering plane curves, and then
generalize the result to space curves. First we need an auxiliary observation.
Lemma 1. Let cp : [a, b] lR be a real function of class C', and suppose that
the function f (t) := ew<<l" joins the numbers ±1, i. e., f (a) = 1, f (b) = -1.
Then
(1) f If(u)Idµ
b 2: ir;
a
(2) in this estimate, equality holds if and only if the derivative cp does
not change sign, and, moreover, IV(b) - p(a)I = r.
Proof. From j (p) = i we obtain
f If (j)I dµ ? I f
b b
(p)dui = Iw(b) - w(a)I.
Since f (a) = 1 and f (b) = -1, there exist integers k, I such that W(a) = 2kir
and yp(b) = 21;r + tr. This implies
f bIf(i)Idy > irI1+2(k-l)I > a.

a
The case of equality follows immediately. 0

Theorem 8. Let -y : [0, L] - R2 = C be a closed plane curve. The total
curvature of the curve is at least 27r:
jL
jIk(s)Ids = > 2ir.
Equality in this estimate occurs if and only if the closed curve -y bounds a
convex region.
Proof. To every point of the curve we assign the tangent vector to the curve.
This determines a map from the interval to the unit circle S' satisfying
F(0) = F(L). The image set F([0, L]) C S' is a compact and connected subset
of S', hence an arc. This arc contains at least one pair ±zo of opposite
points. Otherwise, there would exist a number zo E S' such that neither
+za nor -zo would belong to the arc. But then o F(s) would have an
imaginary part which would have constant sign, and from
J:Im Ga F(s)1 -
ds = Im
\ Za
f L F(s)ds) = Im
\ Z;
(y(L) - y(0))) = 0
we would obtain a contradiction. Without loss of generality, we may assume
zo = I and r(O) = F(L) = 1. Moreover, there exists a parameter value
Si E [0, L] such that r(s1) _ -1. Applying the lemma above, we obtain
f 0sl 81
ik(s)Ids = f IZ'(µ)Idi
0
>- 7r, f
81
L
Ik(s)Ids = f L
81
71,
L
and, altogether, Ik(s)Ids >_ 2a. The inequality for plane curves is thus
J
proved. In the case of equality, the lemma implies that the plane curvature
k(s) of the curve does not change sign on the interval [0, s1], and
/081 k(s)dsI = it = 14LL k(s)dsl .
Now we exclude the case that k(s) has different signs on the intervals [0, s1]
and [s1, L]. If this were the case, then
k(0) = k(si) = k(L) = 0,
and k(s) would be positive on the interval [0, s1] and negative on [s1, L].
But then the inequality
0< f k(µ)dp < 7r
would hold for all parameters s E [0, L], and the tangent vector t(s) _
y
exp k(p)dµ] would stay in the upper half-plane (z E C : Im
hence we derive a contradiction to
L
t(s)ds = -t(L) - ^t(O) = 0.
The plane curvature k(s) of the curve thus has constant sign, k(s) > 0. In
this case, the whole curve completely lies on one side of the tangent to the
curve at an arbitrary point. In fact, without loss of generality, we can choose
the point as that corresponding to the parameter value s = 0. Consider the
function /3(s) measuring the height of the curve point -y(s) over the tangent
through -t(0),
Q(s) := Re (('Y(s) - Y(0)) (i t(0))

The derivative of Q(s) is easily computed:
(j8
13'(s) = Re 00) exp I i f e k(u)du] Ti t(0))) = sin k(u)du)
L o
The curvature is non-negative, and thus /3(s) is negative for parameter values
close to s = 0. If /3(s) had any negative value in the interval [0, L], then
there would exist a value snin such that
/3(smin) < 0 and Q'(smin) = 0.
/L
Since k(u) du = 21r, this value s is determined by the equation
J0
5min
k(u) du = 7r,
and, for s E [0, smin], the derivative of the function 0(s) would be negative.
Therefore, $(s) would not be decreasing in [0, smi,,], contradicting /3(smio) <
0. Summarizing, we obtain that the height function is non-negative, and
thus the curve lies completely on one side of each of its tangents. Consider
now the half-planes determined by all the tangents. Their intersection is a
convex domain whose boundary coincides with the curve ry(s). 0
We generalize the inequality of the preceding theorem to space curves. To
do so, we need a preparation.
Lemma 2. If D C S2 is a closed curve in the sphere of radius one whose
length does not exceed 2a, then V is contained in a hemisphere.
Proof. Choose two points P and Q on D which divide the curve into two
segments of equal length, V = D1 UD2. Let the north pole N = (0, 0, 1) lie
on the shorter segment of the great circle through P and Q on the sphere.
Denote by Di the curve segment obtained from Dl by rotation around the
z-axis through the angle a. The union Dl U Di is a closed curve on the
sphere whose length coincides with that of the original curve D. If the curve
segment Dl intersects the equator S' = {(x, y, 0) : x2+y2 = 1} of S2, then
the curve Dl U DI contains opposite points on the sphere. Hence its length
has to be at least 27r. By assumption this length cannot exceed 27r. Thus 7)i
does not intersect the equator S' at all, if the length of the curve D is smaller
than 27r. Otherwise, D, or V2 is a semi-circle on the equator, respectively.
In both cases, the curve D is contained in the upper hemisphere. 0
Q
5.2. The Structural Equations of a Surface 141
Theorem 9 (Fenchel Inequality). The total curvature of a closed curve

C C R3 is at least 27r,
K(s) ds > 27r.
Jc
Equality holds if and only if C is a plane curve bounding a convex domain.
Proof. Let a non-plane curve be given. Consider its tangent map t :

[0, L] -+ S2 as a "curve" in the sphere S2. The length of the latter co-
incides with the total curvature of the original curve,
fL rL
L(t) = I
0 Jo
Note first that the image of t cannot be contained completely in a hemisphere
of S2. If this were the case, then there would be a vector a' E S2 such that
(a, F(s)) _> 0 for all values of the parameter. But the integral of t(s)
vanishes, since we started from a closed curve. Hence (d, F(s)) - 0, and
the curve C would be a plane curve. The assertions of the theorem now
follow from the previous lemma together with Theorem 8.
5.2. The Structural Equations of a Surface

A surface is a two-dimensional submanifold of 1R3. As in the case of curves,
we want to allow for self-intersections and will also consider parametrized
surfaces. This is meant to denote a map F : U --+ R3 defined on an open
subset U of R2 (or, more generally, on a two-dimensional manifold U) with
the property that the differential D(F) has rank two at each point. As in
the case of a two-dimensional submanifold A12 C l3, we denote by M :
M2 - R3 the inclusion into 3-space and view this map as a vector-valued
function. Locally, we choose an orthonormal tangent frame el, e2 to the
surface and denote by e3 := el x e2 the normal vector to the surface. We
sometimes consider-without introducing a new notation-el, e2 and e3 also
as vector-valued functions on the manifold 1112. Denote by al and o2 the
frame of 1-forms dual to {el, e2}. The differential dM of the identity map on
the surface is the identity map on each tangent space to the surface. Hence
the following equation holds:
dM =01 - el +
expressing nothing but the decomposition of a tangent vector with respect
to the basis e1, e2. The exterior product of A o2 is a 2-form on the surface
which is independent of the choice of the orthonormal frame el, e2 for a fixed
orientation and coincides with the volume form,
012 = of A a2.
The scalar products (ei, ej) of the vector-valued functions ei are constant
functions on M2. By differentiation, we obtain
(dei, ej) + (ei, dej) = 0
and denote the 1-forms by wij := (dei, ej). Now combine the differential
forms into an antisymmetric (3 x 3) matrix Cl:
0 w12 W13
= w21 0 W23
W31 W32 0
Using the forms just introduced, the differentials of the vector-valued func-
tions ei can be represented as
3
de. = E wij ej
j=1
Altogether, we constructed five differential forms of degree one, the forms

of and wij. These are, however, not independent; the so-called structural
equations of a surface express the relations among them. The fact that the
form w12 is completely determined by 01 and 02 will turn out to be of special
importance. Later on, this will imply the fact that the Gaussian curvature
belongs to the inner geometry of a surface (Theorema Egregium, Gauss,
1827).
Theorem 10 (Structural Equations of a Surface).

(1) d01 = W12 A C2, do2 = w21 A o1 .
The form W12 is completely determined by the forms of and o2 .
(2) a1 A w13 + 02 A w23 = 0 .
(3) dC = n A Cl .
Proof. Differentiate the function M twice,

0 = ddM = d(o 1 - el + 02 e2) = (dal - w12 A 02) el
+ (do2 - w21 A 01) e2 - (a A w13 + 02 A w23) e3 .
This yields the equations stated in (1) and (2). The identity for Cl follows
using 0 = ddei = Ek (dwik - Ej wij A wjk) ek. The form w12 can be
computed from al and 02 as follows. Represent the differentials
d01 = A 01 A o2i d02 = B 01 A o2
as multiples of the 2-form of A 02. The first structural equation implies
w12 = Aa1+Bo2.
Let a parametrization F : U - M2 C R3 of a surface by a a domain

U C R2 be given; then the 1-forms ai and wij can be pulled back to U. The
resulting forms will be denoted by a, := F'(ai) and wij := F'(wij). Then
the structural equations of the surface hold as well:
2
do,; _ wij A aj*, al A W13 + a2 n w23 = 0, dtl' = St' n it .
j=1
Now we prove a first formulation of the fundamental theorem of surface
theory. This allows to determine a surface locally in 3-space from the system
of associated 1-forms satisfying the structural equations relating them.
Theorem 11 (Fundamental Theorem of Surface Theory-First Formula-
tion). On a simply connected, open subset U of R2 let four differential forms
01, a2, 013, 023 of degree one be given. Suppose that the forms 01 and 02 are
linearly independent at each point. Define the 1 -form 012 by the equations
d'1 = 012/02, d02 = -012A01.
Extending them antisymmetrically, wji := -wij, assume that the given sys-
tem satisfies the structural equations
U1 A 01$ + a2 A 023 = 0, dfl = NA N.
Then there exist a parametrized surface F : U -' R3 and an orthonormal
frame of tangent vector fields such that the induced forms a, and wij coincide
with the original forms aFi and wij, respectively. The surface together with
its orthonormal frame is uniquely determined up to a euclidean motion of
R3.
Proof. The (3 x 3) matrix S2 is skew-symmetric and satisfies the integrability

condition dSl = SE A li. For a given initial condition Ap E SO(3, R), there
exists a matrix q of functions such that St A = dA by §4.3, Theorem 6. The
rows of this matrix are orthogonal and have length one. Thus they define
three vector-valued functions e1, e2, e3 : U - R3, and
3
dei =
j=1
Consider the 1-form
0 := 01.91 + 02.92
Computing the differential of 0 yields
dO = 0.
By Poincares' lemma (§3.6, Theorem 22) there exists a map F : U - R3 such
that dF = ¢, = Q1 e1 +02 e2i and hence we obtain the parametrized surface
together with the orthonormal frame. Twice in this proof we integrated a
differential equation, and thus there remains a certain degree of freedom

with respect to the initial condition. First, the solution q of the differential
equation Il A = dA depends on the prescribed orthogonal matrix Ao E
SO(3, R). In the second step, the function F : U -' R3 is determined by
the differential equation dF = ¢ only up to a constant vector a E 1R3.
Summarizing, we see that the surface is, for prescribed forms of and wtj,
determined up to a euclidean motion.
Example 3. We discuss the case of a surface of revolution in R3. Consider
a plane curve in the (x, z)-plane with component functions r(s) > 0 and
z(s). The surface F arises by letting this generating curve revolve around
the z-axis, and thus it has the following parametrization defined on U =
1[2 x (0, 27r):
F(s,cp) = (r(s)sincp, r(s)cosco, z(s)).
For simplicity, we suppose that the generating curve of the surface F is
parametrized by arc-length, i.e., at each point the following condition is
satisfied:
(*) (r')2 + (Z,)2 = 1.
(Compare Exercise 15 for the general case.) First we compute two tangent
vectors to the surface,
a
as
= (r sin co, r cos co, z'),
a = (r cos V, -r sin tp, 0),
as well as their scalar products,
a
as' as
s
=
(r,)2
+
(z,)2
= 1,
a
as 5 = 0,
,
s a s--
app asp
r2
Normalizing the second tangent vector correspondingly, we obtain the or-

thonormal frame
el
a
_ as, e2
_ 1a
r Op
and the basis dual to it,
01 = ds, 02 = r dip.
The unit normal vector to the surface F is the vector product of el and e2,
z'sin,p
e3 := el x e2 = z'cos
-r'
We compute the differential of el = (r'sin gyp, r' cos cp, z') componentwise,
del = (r" sin cp ds + r' cos cp dcp, r" cos pds - r'sin pdcp, z"ds),
and check that

del = r' dcp e2
+r ds e3 .
zI
In fact, inserting e2 and e3 into this linear combination, we obtain
r' dcp e2 + ; ds e3
z
= (r' cos cp dcp, -r'sin cp dcp, 0) + z, (z'sin cp dcp, z' cos cp ds, -r' ds)
= (r" sin cpds + r' cos cp dcp, r" cos cp ds - r'sin cp dcp, -r" r'/z' ds) .
But, since differenting the relation (*) implies rr' + z'z" = 0, this is equal
to del. Similarly, one proves that
-r
de2 = -r'4 el - z' dcp e3, de3 = ds el + z' dcp e2,
z
and, therefore, we computed the matrix f of 1-forms w, completely:
0 W12 w13 0 r dcp Z' ds
-w12 0 w23 = -r dcp 0 -z'dp
-w13 -w23 0 : ds z' dcp 0
In addition, we show in this example how to compute the form w12 directly
from al and a2. Starting from dal = 0 and dal = r'-dsAdcp = r/r-al Aa2,
we obtain the formula
W12 = r -a2 = r - dtp

using the method described in the proof of Theorem 10. Computing w12
directly from the definition turns out to be even easier:
r'
w12 = (del, e2) = (r'dcp e2 + z' ds e3, e2) r' dcp -
Example 4. Let h : U R be a real function in two variables on the open

set U C R2, and denote by F its graph,
F(x,y) = (x, y, h(x,y)).
We indicate partial derivatives by lower indices. Starting from the two

tangent vectors
a
ax = (1, 0, h=) and ay = (0, 1, hy),
we determine an orthonormal frame with the same orientation,
a y 1+ by a- a
e, =
-+h2 x+
1
e2 _
h2
hZhy
1 + h2 + h2 l h,hy ay ax lJ
as well as the 1-forms dual to these vectors,
hxhy VT T h2 + h2
of = 1+h2dx+ Q2 dy.
1+h2 dy' = l+h2
The corresponding volume element is
of Aa2 = 1+hz+hndxndy.
The computation of the whole matrix fl is more extensive. Because of its
geometric relevance, we give the form w12 for later use:
hu dl
W12 = (del, e2) =
(1+h2 1+h2+h2
In the discussion of curvature properties of surfaces, we will see that partic-
ularly interesting graphs arise from holomorphic functions. Therefore, these
are discussed next.
Example 5. Let f : U C be a holomorphic function. The modular
surface of f is the graph of the modulus function h = if I = (fl) 112 of f.
The relations between the derivatives with respect to the complex parameter
z and the (real) partial derivatives are expressed in the following formulas:
8 1 a 2.) , 8 1 8 8
8z - 8x - z by 8z - 2 ex + i FY)
Hence it is possible to express hx and by by the z-derivatives of f and f.
The latter are denoted by '. It is easy to see that
8h _ f'f 8h _ ff'
8z ]F h-' 8z 2h
Thus the derivatives we wanted to compute are
hz = 2h(f'f +ff) and by =

Rewriting this yields, e. g. ,
1+h2+h2 = 1+1992,
and inserting these expressions into the formula of the previous example we
immediately obtain an explicit expression for the corresponding form w12
depending on the complex derivatives of the function f. Indicating the lines
of constant modulus as well as those of constant argument on its modular
surface leads to a clear picture of the behavior of the function in the complex
domain. As an example, we consider the function f (z) = 1/ sin(z) outside
its poles, shaded according to the argument and showing the level lines of
the modulus:
5.3. The First and Second Fundamental Forms of a Surface 147
Example 6. Decompose a holomorphic function f : U - C into its real

and imaginary parts, f = u + iv. Taking in Example 4 for h the functions
h = u and h = v, respectively, we obtain the graphs of the real and the
imaginary part of f. Starting from the Cauchy-Riemann equations,
ux = Vy, uy = -vx ,
and first using the formula for 8182 derived in the preceding example, f _
ux - iuy = vy + ivx, we obtain
1+uz+uy = 1+vv+v22 = l+If'I2.
At this point we omit the explicit formula for w12.
5.3. The First and Second Fundamental Forms of a Surface

In classical differential geometry, it is common to describe a surface by two
square matrices, the so-called first and second fundamental forms. The
first of these is already familiar from Chapter 3, since the first fundamental
form is just the Riemannian metric. The second fundamental form depends
on the particular realization of the surface in euclidean space. To be able
to define the first fundamental form in the language of differential forms,
we introduce the symmetric product of 1-forms, in addition to the exteror

product we already know.
Definition 8. Let M' be a manifold, and let a, µ be two 1-forms on Mn.
Define a symmetric bilinear form a 0 p : TM" x TM" IR by
a 0 µ(V, W) := 2 (a(V)µ(W) + a(W)µ(V)) .
In the literature, it is common to omit the symbol 0. The symmetric square
of a 1-form or is often denoted by a2.
The definition immediately implies that, on a surface with orthonormal

frame el, e2 and corresponding dual basis of 1-forms al, a2, the following
equation holds:
a: O al(ej, ee) = ai(ei)a1(et) = bijb k
For the basis e1, e2 the bilinear form
I := al 0ol+a2Oa2
is thus precisely the usual euclidean scalar product. It is called the first
fundamental form of the surface; it coincides with the Riemannian metric
defined in Chapter 3. Geometric quantities defined for the surface depending
only on the first fundamental form are called quantities belonging to the
inner geometry.
Example 7. For the surface of revolution from Example 3, we had al = ds
and a2 = r dip; thus the first fundamental form is
I = ds O ds + r2dcp O dcp.
This is also written as I = ds2 + r2dc 2.
Example 8. The first fundamental form of a graph (Example 4) is
I = (1 + h2) dx2 + 2 hhxhy dx dy + (1 + h2) dy2 .
Example 9. On the upper half-plane x2 = {(x,y) E R2 : y > 0}, we
consider the 1-forms
al = dxy-, a2 =dyy-.
They determine the Riemannian metric
dx2 + dye
y2
The half-plane N2 endowed with this metric is called the hyperbolic plane.
The form w12 depending only on al and a2 is equal to al.
If V and W are two tangent vector fields to the surface M2, then, viewing
W as a vector-valued function and differentiating it as such with respect to
the vector field V, we denote the resulting vector-valued function by V(W).
In general, V(W) is not tangent to the surface. However,
Theorem 12. Let V and W be two tangent vector fields on a surface. Then
the difference V(W) - W(V) is also tangent to the surface, and it coincides
with the commutator of the vector fields,
V(W) - W(V) = [V, W] .
Proof. In the proof we make use of the structural equations of the surface.
V and W are tangent vector fields, and hence the scalar products (V, e3) _
(W, e3) = 0 vanish identically. This implies
(V(W) - W(V), e3)
_ - (W, de3(V)) + (V, de3(W))
- (W, w31(V)el +w32(V)e2) + (V, w31(W)el +w32(V)e2)
al A w31 (V, W) + a2 A W32 (V, W)
Because of the structural equation a1 A w31 + a2 A w32 = 0, the difference
V(W) - W(V) is a tangent vector field on the surface. Moreover, from §3.9
and the structural equation, we conclude for i = 1, 2 that
a;(V(W) - ),V(V)) = (V(W) - W(V), e;)
= V((W,e1)) - (W,de1(V)) - W((V,e1)) + (V,de;(W))
2
= V((W,e;)) - W((V,e:)) +1: aj Awii(V,W)

i=1
= V(a;(W)) - W(a2(V)) -dai(V,W) = a+([v,WI).
Thus V(W) - W(V) coincides with the commutator [V, W]. 0
We decompose the vector-valued function V(W) into its tangent and its
normal part. The tangent part is denoted by V W, whereas the normal
part will give rise to the so-called second fundamental form II(V, W) of the
surface.
Definition 9. Let V and W be tangent vector fields on the surface M2 C R3.
The tangent part of V(W),
V W := V(W) - (V(W), e3) e3 = V(W) + (W, de3(V)) e3,
is called the covariant derivative of the vector field W with respect to the
vector field V. The normal part of the vector-valued function V(W) will be
denoted by II(V, W),
II(V, W) := (W, de3(V)) :
it is called the second fundamental form of the surface.

Remark. The second fundamental form of a surface depends on the ori-
entation of M2. Changing it reverses the normal vector e3, and hence II
changes sign, too.
First we collect the properties of the covariant derivative V.

Theorem 13.
(1) Vv(W1 + W2) = Vv(W1) + Vv(W2);
(2) Vv,+y,W = Vy,W+Vv,W;
(3) If f : M2 R is a smooth function, then
Vv(f W) = df (V) W + f VyW and V1.yW = f VyW;
(4) V (W1, W2) = (VVW1, W2) + (W1, VvW2);
(5) VyW - VwV = [V, W).
Proof. The additivity of VvW is a straightforward consequence of the defi-

nition. The last formula follows from the previous theorem. All the remain-
ing assertions are easily checked, e. g.,
Vv(f - W) = V(f W)+(f - W, VyW.
Representing the vector field W in an orthonormal frame e1, e2 on the sur-

face, W = W' e1 + W2 e2, and taking into account that
VVeI = del(V) + (el,de3(V)) e3
= w12(V) . e2 + w13(V) e3 + wsi(V) e3 = w12(V) e2,
we obtain the important formula
VvW = (dW' (V) + w21(V) W2) - el + (dW2(V) + w12(V) W') e2 .
The form w12 depends only on a,, a2i and hence the covariant derivative
VvW also depends exclusively on quantities belonging to the inner geom-
etry of the surface. The second fundamental form lI(V, W) is a symmetric
bilinear form on the tangent bundle of the surface.
Theorem 14. Let V and W1, W2 be tangent vector fields on a surface, and
let f1, f2 be smooth functions. Then
(1) the second fundamental form is symmetric, II(V, W1) = II(Wi, V);
(2) II(V, f1 Wi + f2 W2) = f1 II(V, Wi) + f2 II(V, W2).
Proof. We prove the symmetry of the second fundamental form; the re-
maining property immediately follows from the definition. With V and W,
the commutator [V, W] = V(W) - W(V) is a tangent vector field as well.
Thus
(V, de3(W)) (W(V), e3) = - (V(W), e3) + (V(W) - W(V), e3)
_ -(V(W), e3) = (W, de3(V)),
i.e. , II(V, W) is symmetric.
Because of the structural equation de3 = W31 el + W32 e2, the second
fundamental form can be written as a symmetric product of the following
1-forms:
II = W31 O Q1 + W32 O v2.
Hence, in the orthonormal frame el, e2, the second fundamental form is
represented as the symmetric matrix
II(e1, el) II(el, e2)1
II = I II(e1,e2) II(e2,e2)
w31(el) (31(e2)
w32(el) w32(e2)]
If, on the other hand, F : U - M2 C R3 is a parametrization of the surface,

and we choose the normal vector
_ F X OF - 1 (OF 8F
X Oy2)
OF Ig-
e3 II e-Fr i Oy1
57 11
then the second fundamental form is given by the symmetric (2 x 2) matrix

b_ b11 b12
b21 b22,
with coefficients
_ 82F
, e3) - 1 82F OF
I
OF
W2/
So for each surface M2 C R3, there exist two symmetric bilinear forms, I and
II, of which the first is positive definite at each point. These two fundamental
forms are not independent; the structural equations of the surface will lead
to a pair of differential equations relating them to each other. In order to
formulate these equations, we introduce the curvature tensor of the surface.
This is a transformation R : TM2 x TM2 x TM2 - TM2 associating with
every triple of vectors a tangent vector.
Definition 10. Let U, V, W be tangent vector fields on the surface. The
curvature tensor R is defined by the following formula:
R(U, V)W := VuvyW - VvvuW - vu,vlW .
The following relations for the curvature tensor are a formal consequence
of the properties of the covariant derivative VvW for vector fields stated in
Theorem 13. We leave their verification to the reader.
Theorem 15. Let U, V, W be tangent vector fields, and let f be a junction.
Then
(1) R(U,V)(f'W) = f.R(U.V)W.

(2) R(U, V)W = -R(V,U)W,
(3) I(R(U, V)W1, W2) = -I(R(U, V)W2, W1),
(4) I(R(U, V)W1, W2) = I(R(W1, W2)U, V),
(5) 1(U.V)W+7Z(V,W)U+7Z(W,U)V = 0.
Recalling that the covariant derivative VvW of two tangent vectors only
depends on the forms o1, o2 belonging to the inner geometry, the same also
holds for the curvature tensor. We prove a local formula for R(U, V)W
specifying this fact.
Theorem 16. Let e1, e2 be an orthonormal frame on M2. Then
R(U,V)W = dw12(U, V) (°t(W) e2 - o2(W) e1) .
Proof. Because of the symmetry properties of the curvature tensor stated

in the preceding theorem, it suffices to prove this formula for the basis fields
U = Cl. V = e2, W = e1. In this case,
1Z(e1,e2)e1 = Vei(Ve2e1) - Ve2(Veiel) - V[eI.e2]e1
V'. (w12(e2) . e2) - Ve2(u12(el) ' e2) - W12([el e2]) - e2
(el(w12(e2)) - e2(w12(el)) - W12([el,e2]))'e2

+ (wt2(e2)W21(e1) - w12(et)w21(e2))'el
dw12(el,e2)'e2,
and this last expression coincides with the right-hand side of the equation
to be proved.
The first fundamental form is positive definite. Hence the symmetric second
fundamental form can be represented by a symmetric endomorphism S
T A,12 - TAP defined by
II(V, W) = I(V, S(W)) ,
the so-called Weingarten map of the surface (see §5.4). The equation II(V, W)
(V. de3(W)) implies that S can be rewritten as
S = de3 = W31 ' el + w32 e2 .
The covariant derivative of an arbitrary endomorphism E : TM2 - TM2

of the tangent bundle is a 2-form VE : TM2 x TM2 TM2 with values in
TM2. This is defined by the equation
VE(V, W) :_ Vv(E(W)) - Vw(E(V)) - E([V. W]).
Theorem 17 (Gauss and Codazzi-Mainardi Equations).
(1) VS = 0 (Codazzi-Mainardi equation);
(2) IZ(U, V)W = II(V, (Gauss equation).
Proof. Using the formula for the exterior derivative of a 1-form, the equa-
tion dS = dde3 = 0 first implies
0 = V(S(W)) - W(S(V)) - S([V, W]).
Looking at the tangent parts and taking into account that S([V, W]) is
already tangent, we obtain
0 = v (S(W)) - Vw(S(V)) - S([V, W]).
Hence VS = 0 is proved. Recalling that the second fundamental form
is II(V. W) = (V, de3(W)), we obtain S(U) = de3(U). Using now de3 =
w31 el +W32 - e2 and then applying the preceding theorem proves the Gauss
equation:
II(V, W) S(U) - II(U, W) S(V) = w13 A w32(U, V) (o l (W) e2 - el )
= dw12(U,V) (al (W) e2 - o2 (W) e1)
= R(U, V)W. 0
Next we will prove that prescribing a positive definite first fundamental form
as well as a second fundamental form uniquely determines a surface up to a
euclidean motion in R3, provided that both fundamental forms satisfy the
Gauss and the Codazzi-Mainardi equations.
Theorem 18 (Fundamental Theorem of Surface Theory-Second Formu-
lation). Let two symmetric bilinear forms 1, II : TU x TU -+ R be given on
a simply connected, open subset U of R2, and suppose that the first, 1, is
positive definite at each point. Starting from 1, the equation
21(VUV, W) = U(1(V, W)) + V(1(U, W)) - W(I(U, V)) + 1([U, V]. W)
+1(V, [W, 141) - I(U, [V, W])
defines a covariant derivative V for vector fields with respect to vector fields,
and the curvature transformation is determined by
R(U,V)W := VuvvW - VvVuW - v[U,V]W.
If the symmetric endomorphism S : TU TU defined from II by II(U, V) _
I(U, S(V)) satisfies
(1) V S = 0,
(2) R(U, V)W = II(V, W) -3(U) - II(U, W) S(V),

then there exists a parametrized surface F : U -+ R3 such that the induced
first and second fundamental forms of this surface coincide with I and II,
1 = F'(I), II = F'(II) .
This parametrized surface is uniquely determined up to a euclidean motion
of 1113.
Proof. The proof is based on a straightforward application of the funda-

mental theorem of surface theory in its first formulation (Theorem 11, §5.2).
To apply this, we choose a frame {ei, e2} of vector fields on the open set
U C JR2 which is orthonormal with respect to the bilinear form I, and de-
note by 01, 02 the corresponding dual frame of 1-forms. Define additional
1-forms by the equations
w12((V) := Wvei, e2),
013(V) 11(V, e1) = 1(S-(V), el),
i23(V) := R(V, e2) = 1(3(V), e2)-
This leads to a system {ai, Q2, "12, W13, X23} of 1-forms, which is extended
by requiring antisymmetry, wji := -Oil. The equations we supposed to
hold imply that the integrability conditions of the fundamental theorem of
surface theory in its first formulation are satisfied. To see this, note first
that the equation defining the covariant derivative V immediately yields the
formulas
U(I(V, W)) = I(VuV, W) +I(V, VuW), Vuv - Vvu = [u, V] .
From these, we obtain

doi(U,V) = U(o1(V)) - V(aF1(U)) -aFi([U,v1)
= U(I(V,e1)) -V((U,e1)) -I([U,V],ei)
= I(VuV - VvU - (U, V), ei)+1(V, Duel) - I(U, Vve1)
= w12(U) . a2(V) - w12(V) U2(U) = 012 AQ2(U,V)
Hence dv1 = 012 A o2, and similarly one verifies that dv2 = 021 A U1. In
the next step, we show that v1 A 013 + Q2 A 023 = 0. This follows from
01 A 013(U, V) + Q2 A O23(U, V) = II(V,U) - II(U, V) = 0,
since the second bilinear form is symmetric. For the computations to follow,
note that I(ei,ei) e 1 implies I(Vu'el,'e1) = 0, and hence Duel is parallel
to e2. Thus for arbitrary vector fields U, V we always have
IMue1, Ove2) = 0.
5.4. Gaussian and Mean Curvature 155
Using this equation and the Gauss equation, we compute the form dw12:
dw12lu,V) = U(i(vvel,e2)) - V((VUel,e2)) - I(V[u,v]el,e2)
= I(R(U, V)el, e2) = II(V, el) II(U, E2) - II(U, E1) . II(V, e2)
= w13Aw32(U,V)
From this equation we get the structural equation to be satisfied, d012 =
w13 A 032. The relation for d013 is derived in a similar way using the
Codazzi-Mainardi equation:
d0134 V) = U(I(S(V),el)) -V((S(U),el)) -1(3([U,V]),El)
= I(V S(u, V), el) + I(S(V), duel) - I(S(U), vvel )
= I(S(V), F2) I(e2, 7ue1) - I(S(U), e2) I(e2, vvel)
= 012A023(U,V)
Thus d013 = 012 A 023, and an analogous computation yields d O23 =
w21 A W13. Together, the system of forms {Q1, a2, w12, 013, w23} satisfies
the integrability conditions of Theorem 11, §5.2, and hence the existence of
a parametrized surface F : U 1R3 such that F'(I) = I and F'(II) = II
is proved. The uniqueness statement is a consequence of the corresponding
uniqueness result of that theorem. 0
5.4. Gaussian and Mean Curvature
Consider a surface M2 C R3 and its first fundamental form, the induced
Riemannian metric. Choosing a local orthonormal frame {el, e2} with dual
frame al, a2, we know that the 1-form w12 = (del, e2) = (Vel, e2) is com-
pletely determined by the forms al, 02. Since its exterior differential dw12 is
a 2-form, there exists a function G on the surface such that
d012 =
The function G is independent of the choice of the local frame of vector
fields. Every other frame {ei, e2} can be represented as
ei = e2,
e2 = -µ sinap el +coscp e2,
where µ = ±1 is constant, and V is a function on that part of M2 where
both orthonormal frames are defined. The dual frames are thus related by
d = µ cos 92 of + sin cp o,2,
a2 = -µ sin p a2 + cos a2,
and we obtain ai A o2 = µ of A a2. On the other hand, the form w12 is equal
to w12 = (dei, e2) = µ w12 + dip, leading to the formula &.012 = µ &012-
This observation shows that the function G is uniquely defined.
Definition 11. The Gaussian curvature G : M2 -* R is the function defined

by the equation
dw12 = - not.
Remark. The Gaussian curvature depends exclusively on the first funda-
mental form of the surface. If this is given, we can choose an orthonormal
frame of 1-forms al, a2. From the equations
dal = A- al na2 and da2 = B - al n ag,
we determine two functions A, B and then introduce the 1-form
wit =
Finally, the defining equation for the Gaussian curvature is
dw12 =
This way, the Gaussian curvature can be computed in practice.
The formula for the curvature tensor of the surface stated earlier can be
simplified by means of the Gaussian curvature:
R(U, V)W = G - (a2(W) el - al (W) e2) dM2(U, V),
and, conversely, G can be expressed by the curvature tensor.
Theorem 19. The Gaussian curvature of a surface is
G = (R(e1, e2)e2, el) ,
where 1Z is the curvature tensor, and lei, e2} is an orthonormal frame of
vector fields.
Now we turn to the invariants of the second fundamental form, which we

view, by means of the Riemannian metric, as a symmetric endomorphism
S : TM2 - TM2,
the Weingarten map of the surface introduced in the previous section. Its
normalized trace is the mean curvature of the surface.
Definition 12. The mean curvature H of a surface is the function
H := tr (S)12.
The mean curvature of a surface can also be determined as the divergence
(in R3) of the normal vector field.
Theorem 20. Let M2 be a surface, and let N be a vector field of length one
defined on an open neighborhood of M2 in R3. Assume that the restriction
NJnr2 of N to M2 is perpendicular to the surface. Then
H= 1 div(N),
where the divergence is taken with respect to the euclidean metric on R3.
Proof. From (N, dAl) = 0, we obtain

2H = (et, dN(e1)) + (e2, dN(e2)) + (N, dN(N))
The last sum, however, is independent of the particular choice of the or-
thonormal basis in IR3. This implies
2H = y'-, +( ay j + (', Ar = div(N) . 0

Apart from the trace, the second fundamental form S has one more invariant,
the determinant. But this does not lead to a new geometric quantity, as we
shall now see.
Theorem 21. The determinant of the second fundamental form is the
Gaussian curvature,
det(S) = G.
Proof. Let x E M2 be an arbitrary point of the surface, and choose an
orthonormal basis {e1, e2} consisting of eigenvectors of the symmetric endo-
morphism in the tangent space TXM2. If 1C1i K2 are the eigenvalues, then,
with respect to this basis, S has the matrix
rcl 0
0 K2
The assertion now follows directly from the Gauss equation,
G = (1.(e1,e2)e2,e1) = (II(e2,e2)S(el), e1) - (II(el,e2)S(e2), el)
= Ic1 - 1£2 = det(S). 0
Remark. The preceding theorem allows to compute the Gaussian curva-
ture, belonging to the inner geometry of a surface, from the second funda-
mental form. Take an arbitrary (not necessarily orthonormal) basis v1i v2 in
the tangent space TXM2 at a point of the surface, and denote the matrices
of the first and second fundamental forms by g and b, respectively,
b12
9
__ 911 912
9ij = I(v;,ZJ), b= b11
b21
bij = II(v{,vj)
921 922 ' b22J ,
The Gaussian and the mean curvature can then be computed using the
formulas
det(g)
G
and H = 2tr (b g-1) .
Their proof is a consequence of a more general observation. A symmetric

bilinear form B on a real vector space V without any additional structure
has only a single invariant, the signature. If, however, a non-degenerate
scalar product G is given in addition, then this induces an isomorphism
G : V' - V between the vector space V and its dual space V. Thus B
can be viewed as a G-symmetric endomorphism, S := B o G-t : V -. V
in the vector space V, and all coefficients of its characteristic polynomial
are invariants of B with respect to G. These, in turn, can be computed by
means of an arbitrary basis, and the result does not depend on the particular
basis. Two of these invariants of the characteristic polynomial are the trace
and the determinant.
Example 10. For the hyperbolic plane 7{2 from Example 9, a straight-
forward computation shows that d&12 = a1 A a2. Therefore, the Gaussian
curvature is -1. A complete realization of H2 in 3-space does not exist
(Hilbert, 1901). Certain open subsets of the hyperbolic plane, however, can
be realized as surfaces. The related mean curvature is not uniquely deter-
mined; it depends on the particular realization as a surface in R3. One of
them will be presented at the end of the discussion concerning surfaces of
revolution to follow.
Example 11. For the surface of revolution from Example 3, we had

r"
de3 = -z,,
4
This enables us to compute the Weingarten map explicitly:
S(a)
8s 8s
i-j
S O'p JJ = z e2 =
x'
r 8V
8
Thus, in the basis 8/8s, 8/Ow, the Weingarten map is described by the
matrix
S= r"/z' 0
0 z'/r
From this, the Gaussian and the mean curvature are immediately computed:
G=-r" H= 1
(z' r"
r' 2 r z')
Examples for surfaces of revolution with constant Gaussian curvature are the
cylinder (C = 0, r = 1, z = a), the sphere (G = 1, r = sins, z = cos s), or
the pseudo-sphere (G = -1): The latter is an example of a (necessarily non-
compact) surface of constant negative curvature, and hence a 2-dimensional
model of hyperbolic geometry. It is defined as the surface obtained as the
surface of revolution generated by the tractrix with the parametrization
(e_', ja 1 - e-2tdt I = (e-', -VI - e-2' + arctanh( 1 - e-28)) .

Example 12. The Gaussian curvature of the graph of the function h(x, y)
(Example 4) is preferably computed using the differential of the form w12
The resulting expression is
h== hyy - hiy
(1 +h2
To compute the mean curvature, it is convenient to describe the graph
the zero set of the function g : U -+ R, g(x, y, z) := h(x, y) - z. From
Theorem 5, §3.2, we then know that the normalized gradient vector field,
grad(g) 1
(ham, hy, -1)1
IIgrad(g)f J+ h.2 + h2 Y
is perpendicular to the surface at each point. Hence the mean curvature

can be computed as the divergence of this vector field (Theorem 20), and
we obtain
(1 + hy)hy= - 2h,hyhy + (1 + h2)hyy
H =
2(1 + hZ + h2)3/2
Example 13. In order to apply this formula to the computation of the

Gaussian curvature for a modular surface (Example 5), we still have to
determine the second derivatives of h = If 1. From
=
h_ 8h 8h
and _= Z (8h - 8h)
ez + e 8z 8z J
we first conclude formally that
h=
_ a2h 82h 82h
+2
8z2 8z8z + 0z2 '
_ .92h
+2
82h _ 82h
h,,,
102h - 82h
huy 8z2 8- 822' =
Z
8z2 8z2
After expanding, the numerator of the Gaussian curvature turns out to be

2
- a2h 0h
_ (92h
h.= hba
z
- h :y 4 (idzoz) 022
- a52 .
We compute the derivatives:

a2h fll.f (f1f)2 f fN 92h f, f,
O2h (fp)2
az2 2h 4 h3 ' azz 2h 4 h3 ' azaz 4h
from which we deduce the formula for the Gaussian curvature:
f,z
G = (I+lf,12)2 Ref f"-1 .
In the example discussed before, f (z) = 1/ sin z, the Gaussian curvature has
the form
12 [Re
Cost z + cost z
2 1
- 1J
sill z 1+ cu52 z J
G =
1212
+ cos z
sine z
Of particular interest is the curve on the surface separating the points of pos-
itive Gaussian curvature from those of negative Gaussian curvature. Here,
it is implicitly described by the equation
Cost
Re z =1
1 + cost z '
and is indicated as a dashed line in the figure on the previous page showing
the modular surface shaded according to the Gaussian curvature.
Example 14. We now continue the discussion of the surfaces defined by the
real and the imaginary part of a holomorphic function, f = it + iv : U C
(Example 6). There we already derived from the Cauchy-Riemann equations
the relation
l+uZ+uy = l+vi+vb = l+If'I2.
A straightforward calculation yields the second derivative of f :
f" = uz2 - iu"V = v=y + iv==.
The formula for the Gaussian curvature of a graph thus leads to the remark-
able fact that the surfaces defined by the real and the imaginary parts have
the same Gaussian curvature,

fui2
G = (1 -I
+ I f,I2)2
The picture on the previous page illustrates this fact for the function f (z) _
z2, again shaded according to the values of the Gaussian curvature. The
surface becoming pointed at the comers belongs to Im z2, and the saddle-
like surface corresponds to Re z2. The Gaussian curvature is
-4
(1 +4IzI2)2
On the other hand, the mean curvatures of these surfaces differ:

-8 xy
H(Re z2) = 4 (y2 - x2) , H(Im z2) =
(1 + 4x2 + 4y2)3/2 (1 + 4x2 + 4y2)3/2
Now we want to interpret the formula for the Gaussian curvature express-
ing it as the determinant of the second fundamental form in a differential-
geometric way. To this end, consider the normal vector e3 as a map on the
surface M2 with values in the unit sphere S2 C R3. The volume form dS2
of the sphere evaluated on two vectors V, W E T=S2 at the point x E S2 is
the vector product
dS2(V.W) = (VxW,x).
This formula is a special case of the general one for the volume form of the
sphere S"'l C R" stated in §3.4, Example 28. From this, the 2-form e3(dS2)
induced on M2 is easily computed:
e3(dS2)(V, W) = (de3(V) x de3(W), e3)
= ((w3l(V)el +w32(V)e2) x (w31(W)el +w32(W)e2),e3)
= - w13 A w32(V, W) = -dwi2(V, W) = G dM2(V, W).
We summarize the result in
Theorem 22. Let e3 : M2 S2 be the unit normal map of the surface,
and let G be its Gaussian curvature. Then the induced 2 -form e3(dS2) is
equal to the volume form multiplied by the Gaussian curvature:
e3(dS2) = G - dM2.
Fix a point x E M2 on the surface and denote by D(x, E) the set of all points
on the surface whose distance from x is less than E. Then the two-dimen-
sional surface area of the set D*(x, e) := e3(D(x, E)) C S2 can be computed
using the transformation formula for the integral as follows:
volS2 (D'(x,E)) = dS2 = ± e3(dS2) = ± G C. dM2.

J J J
D(x,e) D(x,e) D(x,e)
Dividing both sides by the surface area, volM2(D(x,e)), and applying the
mean value theorem of integral calculus for continuous functions, we obtain
the formula
volss(D*(x,e))
IG(x)I = £1ovolA,2(D(x,e))
This was the original geometric definition of the curvature for a surface in 3-
space. It is-similarly to the curvature of a curve-the infinitesimal volume
distortion of the normal map of the surface. For this approach to curvature,
it is at first not obvious at all that the Gaussian curvature depends only on
the first fundamental form of the surface. This fact forms the contents of
the Theorema Egregium (Gauss, 1827). Proceeding, however, like we did
here, this becomes immediate.
The total Gaussian curvature of a compact, oriented surface without bound-

ary depends only on the topology of the manifold. We prove the so-called
Gauss-Bonnet formula, expressing this fact explicitly. To do so, we need a
preparation.
Lemma 3. Let V be a vector field of length one on the oriented surface M2,
and let {el,e2} be an orthonormal frame in this orientation. Representing
V as a linear combination, V = cos r el + sin r e2, the 1 form
Xv w12 + dr
is independent of the chosen frame. The exterior derivative of Xv is
dXV = - G dM2.
Proof. Every other frame el, e2 can be written as
el =
e2 = -
and we already proved the formula w12 = w12 + dip. Moreover, r' = r - cp,
and hence w12 + dr* = w12 + dr . 0
Stokes' theorem now immediately implies
Theorem 23. Let M2 be an oriented, compact manifold without boundary,
and let V be a vector field of length one defined outside finitely many points
x1, ... , xk E M2. If D(xi, E) C M2 are disk neighborhoods of the points xi
of radius E > 0, then
r /k
J 12
G dM2 = lim I J 1
XvJ
Hence we see that the integral of the Gaussian curvature is independent

of the geometry of the surface. We evaluate the last formula explicitly
by choosing special vector fields. To this end, consider a smooth function
f : Aft -+ IR all of whose critical points are non-degenerate, and denote the
number of its critical points with corresponding index by
rno (f) := the number of minima of f ,
nz z (f) the number of saddle points off,
M2(f) := the number of maxima off.
The vector field V = grad(f)/11grad(f) II is defined outside the set of critical
points. Computing the limits on the right-hand side of the formula, we
obtain
2zr if x; is a minimum or maximum;
lim Xv = 5 -2a if x; is a saddle point.
0 JOD(z,.e) l
To prove this, choose coordinates x, y with respect to which the function f
has normal form, f (x, y) = (± x2 ± y2). We are only interested in the limit
of these integrals for a 0,2 and hence we may, without loss of generality,
take the metric to be flat. From
gradf =f x a y t a
11grad(f)II x2 + y2 ax 7X=2=+7y2 ay
we obtain
x
cosy , sin r y
x2 -+y 2 VI_X2 -+y2
Computing the differential dr leads to the formula
dr = f x2 + y2
Here we have to take the positive sign in the cases of a minimum or a
maximum, and the negative sign for a saddle point. Since
dr = ±21r,
JJJSI
we obtain the limit we wanted to compute, and hence we have proved the
Gauss-Bonnet theorem.
Theorem 24 (Gauss-Bonnet Formula). Let M2 be a compact. oriented sur-
face without boundary, and let f : M2 - R be a smooth function all of whose
critical points are non-degenerate. Then
G dM2 = 21r (rno(f) - ml(f) + rn2(f)) .
ff2
where 1110(f), MI M, m2(f) denote the number of minima. saddle points, and
maxima of f. respectively.
Corollary 1. Let M2 be a compact, oriented surface without boundary.

(1) The integral of the Gaussian curvature depends only on the topology
of the surface, but does not depend on its fundamental forms.
(2) The alternating sum mo(f) - MI (f) + M2 (f) is independent of the
choice of the function f all of whose critical points are non degen-
erate.
Definition 13. The Euler characteristic X(M2) of a compact surface with-
out boundary is the number
X(M12) := mo(f) - MI(f) + m2(f)
Using this number, the Gauss-Bonnet formula for compact, oriented surfaces
without boundary can be written as
G - dM2 = X012)
2n n fExample
z
15. The Euler characteristic of the sphere S2 is 2. In fact. the

restriction of the height function h(x, y, z) = z to S2 = {(x, y, z) E 1R3 :
x2 + y2 + z2 = 1) has one minimum, one maximum, and no saddle points.
Thus
1 C d1112 = 2
27r i' a
for every surface M2 which is diffeomorphic (but not necessarily isometric)

to S2.
Example 16. The Euler characteristic of the torus T2 is zero. It is not
difficult to find, e.g., on the torus of revolution (see §3.1), a function with
one minimum, one maximum, and two saddle points. Thus
G d1112 = 0
jAf2
for every surface which is diffeomorphic to the torus.
The Gauss-Bonnet formula in its general form comprises an additional re-
lation for every vector field V of length one with finitely many singularities.
First we note that the limit
lim
e-o JD(X.'0 Xv
can also be computed, along the same lines as for gradient fields, for an
arbitrary vector field. Start by choosing coordinates y', y2 around the point
xi E A12 such that this point corresponds to y' = 0, y2 = 0. Then represent
the vector field V in these coordinates,
V = V' + V2y2 .
'
As before, we may assume, without loss of generality, that the metric is flat
in this neighborhood of the point, since we do not want to compute the
integral itself, but only its limit. Then (Vl)2+(V2)2 = 1, and hence cosy =
V 1, sin T = V2. From this we obtain - sin r dr = dV 1, cos r dr = dV2,
and thus the form Xv = dr is given by
Xv = dr =
Therefore,
lim
E-O J Xv = J (V1 dV2 - V2 dVl) := 2a Ind(V, xi).
(x i,e) 1
This number is called the index of the vector field V at the singular point
x; E M2. Notice that integrating V1 dV1 + V2 dV2 along S1 yields the same
result. The geometric meaning of the index is illustrated in the figure above.
The picture on the left shows some vector field (the circle is only drawn for
greater clarity and has no direct relation to the vector field itself). Starting
at A, move around the circle once, and assign to each point on the circle the
normalized vector in S' pointing into the direction of the vector field at that
point. In the case shown in this figure, one full turn in the positive direction
on the left corresponds to one full turn on the right, but in the opposite
direction. Hence the index of this vector field at the indicated point is -1.
From the Gauss-Bonnet theorem, we obtain
Theorem 25 (Hopf-Poincard). Let M2 be a compact, oriented surface urith-
out boundary, and let V be a vector field of length one defined outside
{x1, ... , xk}. Then the sum
k
Ind(V, X,) = X(M2)
i=1
is independent of the choice of the vector field.
A further integral formula deals with the integral of the mean curvature. It
goes back to Steiner (1840) and Minkowski (1900), and plays an important
role in the theory of ovaloids.
Theorem 26 (Minkowski-Steiner). Let M2 C 1R3 be a compact, oriented
surface without boundary, and let e3 be a unit normal field. Then
H(x) dM2(x) e3(x)) G(x) dM2(x) .

J f2 = Jtz (x,
Proof. Consider the three functions p2(x) = (M(x), e;(x)) on the surface
M2 and compute their differentials using the structural equations:
dpi = al +w12 P2+w13P3,
dP2 = a2 + w21 P1 + w23 P3,
dP3 = W31 P1 + W32 P2
Note also that the 1-form
w := P1 W23 +P2 w31
does not depend on the choice of the tangent frame el, e2. Thus w is a
globally defined form on M2. We compute the differential of w, making use
of the structural equations as well as the formula for the differentials dpi.
This yields
Stokes' theorem now implies the Minkowski-Steiner integral formula.
The eigenvalues K1, K2 of the second fundamental form are called the princi-
pal curvatures of the surface at the point. Their arithmetic mean is the mean
curvature of the surface, whereas their product coincides with the Gaussian
curvature,
H= 2(i+K2), G=
The points of a surface are divided into several types, depending on the signs
of their principal curvatures. A point of the surface is called
(1) elliptic if Kl K2 > 0 ;
(2) hyperbolic if K1 K2 < 0 ;
(3) parabolic if Kl K2 = 0 and '4 + K2 > 0 ;
(4) flat if K1 = K2 = 0 .
Umbilic points are the points of the surface for which the principal curvatures
coincide, Kl = K2. An umbilic point is either elliptic or flat.
Example 17. The line of points with vanishing Gaussian curvature indi-
cated in the modular surface of the function f (z) = 1/ sin z (see p. 160)
consists of parabolic points. For this reason, it is called the parabolic curve.
By definition, it separates the hyperbolic from the elliptic points.
Theorem 27. If the surface M2 C R3 consists exclusively of umbilic points,
it is part of a plane or a sphere.
Proof. By assumption, the Weingarten map is a multiple of the identity,

S = x Id. The Codazzi equation,
0 = VS = V(K Id) = V(rc) Id,
implies that K is constant. In case K = 0, the normal vector e3 is also
constant. In fact, de3 = w31 el +w32 e2 = 0. Thus M2 is part of a plane. If
K # 0, we obtain from r.-Id = S = de3 the equation d(e3-KM) = 0.
Hence e3 - KM is a constant vector a. Thus,
Ila+KMII2 = 1,
i. e., M2 is part of a sphere.
Theorem 28. A surface M2 C R3 with vanishing Gaussian curvature, G
0, and constant mean curvature, H 0 0, is part of a cylinder.
Proof. The first principal curvature vanishes, r.1 = 0, and the second princi-
pal curvature is constant, K2 0 0. In an orthonormal frame el i e2 consisting
of eigenvectors of the second fundamental form, we obtain the formulas
W 13 = 0 and w23 = K2 a2.
Differentiate both equations:
0 = d1)13 = W12Aw23 = K2W12Aa2,
0 = w21 A w13 = dw23 = Kra da2 .
Since K2 54 0, on the one hand, w12 is proportional to a2, and a2 is a closed
form, dal = 0. But then dal = W12 A a2 = 0, so that altogether we obtain
dal = 0, dal = 0 and w12 = 0.
The structural equations imply
de1 = 0, de2 = r-2'0`2-e3, de3 = K2 a2 e2 .
Hence e 1 is constant, and the surface M2 lies on a cylinder whose axis points
in the direction of el.
Theorem 29. There exists no piece of a surface M2 C R3 satisfying G =
-1 andH =0.
Proof. Let r£ 1i rc2 be the principal curvatures. Then 0 H= (rcl + K2)

and -1 = G = rclrc2 immediately imply rcl = -k2 = 1. Thus we
2 obtain,
with respect to a tangent orthonormal frame consisting of eigenvectors of
the second fundamental form, the equations
w13 = al and W23 = - 02 .
Differentiating these equations yields
dal = d013 = w12 A w23 = - w12 A a2 ,
dal = - dw23 = - W21 A w13 = w12 A al .
On the other hand, the form w12 = A al + B a2 is a linear combination of
al, 0`2 with coefficients A and B still to be determined from
dal = A al A a2 and dal = B al A a2.
But then
A al A a2 = dal = -w12 A a2 = -A al A a2,
B al A a2 = dal = w12 A o1 = -B a1 A a2,
and hence A = B = 0. The form w12 = 0 vanishes, contradicting G =
-1. 0
Definition 14. A surface M2 c ]R3 with vanishing mean curvature, H =_ 0,
is called a minimal surface.
The preceding theorem states that there does not exist any minimal surface
with constant, non-vanishing Gaussian curvature. Consider, more generally,
the normal map e3 : lbi2 - S2 of a minimal surface. From i1 = -rc2 =
v G, we obtain the formulas
W13 = Gal, w23 = -V-v02.
The relation de3 = w31 el+w32 e2 evaluated on tangent vectors V, W E TM2
immediately implies
(de3(V), de3(W)) G (V, W).
If and IM2 denote the fundamental forms of the sphere S2 and the surface
M2, respectively, then the induced fundamental form (Riemannian metric)
e3(IS2) is proportional to IM2:
e3(IS2) _ - G IM2 .
Thus we have proved the following theorem.

Theorem 30. The normal map e3 : M2 -* S12 of a minimal surface pre-
serves angles, and the Riemannian metric g' := - G Ir112 defined on M2 is
a metric of constant positive Gaussian curvature.
The vector-valued position function M : M2 -R 3 of a minimal surface

gives rise to harmonic functions.
Theorem 31. Let M : M2 --+ 1R3 be a surface, and let 0 be its Laplacian
acting on functions. Then M2 is a minimal surface if and only if the vector-
valued function M is harmonic, 0(141) = 0.
Proof. The Laplacian can be expressed by the exterior derivative and the
Hodge operator (see §3.11). Using the structural equations, we compute as
follows:
O(M) dM2 = d * d(M) = d * (al el + a2 e2) = d (a2 el - ai e2)

= del Aa2+do2 el -de2Aal -dal e2
= Aal - el
-(w21 - el -w12Aa2-e2
=
Corollary 2. There does not exist a compact minimal surface M2 C R3

without boundary.
Proof. Otherwise, by Hopf's theorem (see §3.8), the components of the

position function would have to be constant, a contradiction.
The normal map e3 : 1412 -+ S2 of a minimal surface with non-vanishing

Gaussian curvature is a local conformal diffeomorphism. The stereographic
projection from S2\{north pole} to 1R2 is a conformal diffeomorphism, too.
Hence, by inverting the normal map e3 : M2 -+ S2 IIt2, every minimal
surface can locally be parametrized by a map F : U -+ M2 such that
(*) OF, 49F = /8F
OF, 8F\
and
\8F
OF,OF)
ayl aye = 0.
ayl , ayl J àye aye J
Coordinates like these are generally called isothermic coordinates. The min-
imality of the surface expressed in these coordinates is equivalent to the
requirement that F = F(yl, y2) is harmonic with respect to the Laplacian
A = 82/(ayl)2 + O2/(oy2)2 on R2,
A(F) = 0.
Identifying R2 with the complex numbers C, every harmonic map F : U -+
1R3 can locally be represented as the real part of a holomorphic map 4
U -+ C3; this is a direct consequence of the Cauchy-Riemann equations and
Poincare's lemma. The real partial derivatives of 4) can be expressed by the
complex derivative (z = yl + iy2), since 4' is holomorphic:

a4; a4 a4 a4
ayl = az and ay2 = i az
Denoting the derivative 84'/az by W and writing 4< _ ('I'i,'1'2,'P3) with
respect to the coordinates in C3, condition (*) is equivalent to the quadratic
equation
'I,j+*2+q13 = 0.
The solutions of this equation are described by two arbitrary holomorphic
functions f (z) and g(z):
4'1(x) = f z) (1 - 92(z)), 'I2(z) = i f 2(1 + 92(x)). 13x= f(z)9(z)

Summarizing, we obtain
Theorem 32 (Weierstrass Representation of a Minimal Surface). Every
minimal surface M2 C 1R3 with non-vanishing Gaussian curvature can locally
be represented in terms of two holomorphic functions f (z) and g(z). A
particular parametrization is provided by the formula
F = Re(J f 2z) (1 - g2(z))dz, i J L_ z) (1 +92(z))dz, f f (z)9(z)dz).
Remark. A simply connected minimal surface always occurs as one min-
imal surface within an S'-parameter family; it can thus be deformed. In
fact, if 4i := 4'R + i 4i1 : U C3 is holomorphic, then the equation
Re(e7Q4i) = COs(a) 4iR - sin(s) 4i1
defines a family of minimal surfaces, since the quadratic equation 4 + %F2 +
'p3 = 0 is homogeneous. i
Example 18. Very nice non-trivial minimal surfaces already arise by insert-
ing relatively simple functions. Consider, for example, the following triple
of holomorphic functions:
1-Z2 1+z2 1
2z2 ,
2iz2 ' az)
satisfying the condition 'I' +'y2 +'I'3 = 0, and their complex primitives,
F = (Ft, F2, F3) = (_(z + 1/z), (z - 1/z), In z)

2
The real and the imaginary parts of these functions determine a pair of
minimal surfaces which are called adjoint to one another. Setting In z =
g + ir, these are described by the formulas
FR := Re(F) _ (- sinh a sin r, sinh a cos r, r),
Fj := Im (F) _ (- cosh p cos r, - cosh p sin r, q) .
Hence FQ = cos a FR+sin a F1 determines a 1-parameter family of minimal

surfaces joining FR to Fl. The surface FR is a helicoid, whereas F, is the
surface of revolution generated by the catenary 2, the catenoid. The sequence
of pictures above illustrates the deformation of both these minimal surfaces
into one another. Further examples of classical minimal surfaces can be
found in the exercises.
5.5. Curves on Surfaces and Geodesic Lines

Consider an oriented surface M2 C R3 in 3-space, and on it a curve 7 :
[a, b] M2, which we assume to be naturally parametrized. The tangent
vector t(s) to the curve is also tangent to the surface, and will be denoted
by el(s). If e3(s) is the normal vector to the surface AI2 along the curve,
the equation el(s) x e2(s) = e3(s) determines a third vector, which is again
tangential to M2 along the curve. This way, we construct at each point of
the curve an orthonormal frame, the so-called Darboux frame of the curve
2We recall that this is the curve formed by a perfectly flexible inextensible chain hanging
from two supports.
5.5. Curves on Surfaces and Geodesic Lines 173
'y with respect to the surface M2. Unlike the Frenet frame, the Darboux
frame of a curve in a surface is defined at every point of the curve, even at
those points where its curvature vanishes. Now we decompose the curvature
vector k(s) := K(s) h(s) of the curve y in R3 into the part kg(s) which is
tangential and the part kn(s) which is perpendicular to the surface.
Definition 15. The vector field kg(s) is called the geodesic curvature vector,
the vector field kn(s) the normal curvature vector of the curve with respect
to the surface,
k(s) = kg(s) +
The vector icn(s) e3 is proportional to the normal vector of

the surface with a factor Kn(s), called the normal curvature. Analogously,
kg(s) := -Kg(s) e2 is proportional to e2, since the principal normal vector
h(s) is orthogonal to the tangent vector. From this we obtain the geodesic
curvature Kg(s) of the curve, and the square of the curvature K(s) of the
curve is the sum
K2(S) = K92 (S) + K2(S).
Rewriting the geodesic curvature,

Kg(S) = (kg, e2) = (k, e2) = -(del(el), e2) = -I(Veiel,e2),
we are led to the following theorem.
Theorem 33. The geodesic curvature of a curve in a surface depends only
on the first fundamental form of the surface. In the Darboux frame.
K9(s) = - I(Velel, e2)
Similarly, the normal curvature Kn(s) of a curve can be expressed by the
second fundamental form of the surface and its tangent vector:
K. = (kn, e3) e3) = (e3 _ (de1(eI), e3)

ds /
Theorem 34. If y and y` are any two curves tangent at a point of the
surface M2, their normal curvatures at this point coincide.
Now we consider a variation of the curve y : [a, b] 1412 within the surface
1112 with fixed initial and end points. We may imagine this as a family of
curves,
yE (s) := y(s) + e h(s) e2(s) + O(e2),
determined by a function h vanishing at the ends of the interval, h(a) _
h(b) = 0. The length of the curve yE is
L((s)) = fbIIs)+ed(s).e2(s)+E.h(s).(s)+0(e2)II,
174 5. Curves and Surfaces in Euclideai 3-Space
and computing the derivative with respect to the parameter e yields

jb
= e2(s) + h(s) d 2 (s)) ds .
(s)
The formulas (t(s), ez(s)) = 0 and
(t(s), p(s)) = (e1, ve,ez) = w21(el) = ng(s)
then, eventually, imply
de (L(7£(s))) Lo = f a
b
h(s) sg(s)ds.
Theorem 35. Let 7 : [a, b] - M2 be a curve in the surface, realizing the
shortest distance between the points y(a) and 7(b) within the surface. Then
the geodesic curvature of the curve vanishes,
r..9(s) = 0.
Definition 16. A curve in the surface is called a geodesic line if its geodesic
curvature vanishes. This condition is equivalent to VF= 0.
If the geodesic curvature Kg = 0 vanishes, the curvature vector
ac(s) h(s) = k(s) = k' (s)
is proportional to the normal vector e3 to the surface. Hence a curve with
non-vanishing curvature, K(s) # 0, is a geodesic line in the surface.bfz if and
only if the principal normal vector h(s) coincides with the normal vector e3
to the surface,
h(s) = ±e3(7(s))
Geodesic lines realize the shortest distance between two points of the surface
only if the points lie sufficiently close to one another. A longer geodesic line
does not necessarily have to be the shortest path between its end points
within the surface. Examples-on S2 or on the cylinder-are easily con-
structed (see Example 21). We represent the curve in local coordinates
y1,y2 on the surface as y(s) = (y'(s),yz(s)). If g3(y',yz) are the coeffi-
cients of the Riemannian metric and s is the natural parameter of the curve,
then
2 ;
1: gti(y'(s),yz(s)) dy dp' 1.
ds ds =
i j=1
The Christoffel symbols r occur in the representation of the covariant
derivatives of the basis vector fields 8/8y` as linear combinations of them-
selves:
z
ey y7 Erayk
k=1
Since
2
8 dyi k 8
ds - ayk ,
i.k=1
the following formula holds for the tangent vector t :
°t{ 2 k i
I
- k=1
ds2 + 1.3=1
ds ds ayk
We summarize this observation in the following theorem:

Theorem 36. Let -y (s) = (y1(s), y2(s)) be a curve represented in local coor-
dinates. It is a geodesic line if and only if the following system of differential
equations is satisfied (1¢ = 1, 2):
d2yk 2
ds2
+ E rjasa =o.
i j=1
This is a second order system of differential equations. The general existence

and uniqueness result from the theory of ordinary differential equations en-
sures the existence of precisely one geodesic line through any fixed point in
a given direction.
Corollary 3. For each point x E M2 and every tangent vector V E T,M2 of
length one on a surface without boundary, there exists exactly one geodesic
line -y : (-e, e) -+ M2 satisfying the initial conditions y(0) = x, Y(0) = V.
Example 19. Let -y(s) be a geodesic line on the two-dimensional sphere
S2 of radius R > 0. Since x9 = 0, the normal curvature sn and the spatial
curvature a of the curve coincide. Hence
= Ian l = II(t, t) = R
since the sphere consists exclusively of umbilic points. The normal vector
to the sphere is e3(x) = x. Thus, on the one hand, we obtain
11
ds2) s(k(s)) = ds(lcn(s)) = -R ds(e3('y(s))

1 dry (s) 1
ds = - R2 r(s)
R2
and, on the other hand, the Frenet formulas for the curve imply
d2t(s) d 1
t(s) + R r b(s) .
ds2 R dss) R2
Consequently, the torsion r of the curve vanishes identically. The curve -y(s)
is thus a plane curve in S2 with constant curvature 1/R, i.e., a principal
circle in S2.
Example 20. The relation K 2 = n9 + I£,2a immediately implies that every

curve with vanishing curvature is a geodesic line. Formulated differently, if
a straight line of R3 lies in a surface, then it is necessarily a geodesic line in
this surface.
Example 21. Consider the cylinder M2 in 1R3 whose axis is parallel to the
unit vector a5. The normal vector at a point x E M2 is proportional to the
projection of x onto the plane perpendicular to a,
x - (x, a)a
e3(x) =
lix - (x, a)611
For a geodesic line -y(s) on the cylinder, we immediately obtain from h&(s) =
e3(y(s)) the equations (h(s), = 0 and
C := (h(s), 'Y(s)) = (e3('Y(s)), 'Y(s)) = hh'Y(s) - ('Y(s), a)d1l
Here C is positive and constant, since y(s) lies on the cylinder. As dt(s)/ds =
K(s) hs(s) and db(s)/ds = -r(s) hs(s), the equation (h(s), a) = 0 implies
that
A := (t(s), a) and B := (b(s), a-)
are constant. In case A = 0, the curve lies in an affine plane perpendicular
to d and on the cylinder; hence it is a circle. For A 34 0, we differentiate
the equation (hh(s), a) = 0 once, and then the Frenet formulas imply the
relation
-A Ic(s) + B r(s) = 0.
Differentiating the constant C2 twice, we deduce that
C ic(s) + 1 = A2.
The curvature K(s) as well as the torsion r(s) are thus constant, and hence
the curve y(s) is a helix (see Exercise 6). In summary, the geodesic lines on
the cylinder are helices, circles, and the straight lines parallel to the 5-axis.
Two points on the cylinder may be joined by infinitely many geodesic lines,
only one of which is the shortest path between them:
The equations characterizing geodesics are non-linear differential equations

of second order, and hence, in general, difficult to solve. Things become a lit-
tle easier if, because of the symmetry of the surface, one can find sufficiently
many functions which are constant along geodesics:
Definition 17. A function f : TM -+ llt is called a first integral of the
geodesic flow if f is constant on all tangent curves y'(t) of geodesics.
The length of y' itself is always one non-trivial first integral, which is inter-
preted as energy. It expresses nothing but the fact that geodesic lines are
parametrized by arc length. Further first integrals are obtained by Noether's
theorem starting from isometries of the surface.
Theorem 37 (Noether's Theorem). Let V be a tangent vector field on the
surface M2 whose flow 4bt : M2 -+ M2 consists only of isometries. Then the
function
fv : TM -i R, fv(W) = l(W, V),
is a first integral of the geodesic flow.
Proof. Consider a geodesic line y(s) on M2 in its natural parametrization.

Setting W = y', the function defined in the theorem has the following de-
rivative with respect to s:
ds
(Y'(s), V(7 (s)) _ (7'(s), V(Y'(s)) + y'(s), dV(y'(s)))
The vector -y" is nothing but the curvature vector of the curve in R3, and
for a geodesic line this is always perpendicular to the surface. Hence the
first summand vanishes. On the other hand, the flow Ot was supposed to
consist of isometries,
(W1, W2) = (d't(Wi), d0t(W2))

Differentiating this identity and evaluating the result at t = 0, we obtain
(dV(W1), W2) + (dV(W2), WI) = 0,
and the second term vanishes, too, since it has to be antisymmetric in y'(s).
O
Example 22. We want to find all geodesic lines on a surface of revolution.

For simplicity, assume as before that the generating curve is given in natural
parametrization. With respect to the coordinates s and V on the surface, a
geodesic line is parametrized as y(t) _ (s(t), ap(t)) and has the first integral
E := I1-r'(t)II2 = s 2 + r(s)2V2
Since the rotation through the angle V around the z-axis is an isometry of
the surface, the vector field 8/8W satisfies the assumptions of the Noether
theorem and gives rise to a further first integral,
M := (y', 8/8V) = r2c .
The existence of this second invariant is known as Clairaut's theorem (1731)
and has the following geometric interpretation. Consider a geodesic line
through a point of the surface, and also the meridian 77(t) = (so, W(t))
through this same point. The angle a formed by the geodesic and the
meridian is computed as
_ ('y', W) r2Vt2 rI
cos(a)
11Y11 ' 11741 E r2Vp' E
Therefore, M is constant if and only if r(s) cos(a) is constant. For the
qualitative discussion of the geodesic lines, we take E = 1. Inserting yY _
M/r2 into E = 1, we obtain
s 2r2 = r2 - M2, thus 0 < M2 < r2 .
If M vanishes identically, then gyp' = 0, and p has to be constant. This
corresponds precisely to the generating curve of the surface of revolution;
each profile curve of F is thus a geodesic. In the other extremal case,
M2 = r2, we have s' = 0, i. e., a has to be constant. This curve is a
meridian on F; note, however, that in this case differentiating M2 = r2
immediately leads to r r' = 0, which, in turn, implies r' = 0. A meridian
of F is a geodesic if and only if the radius function r has an extremal point
there. All that remains is the generic case, 0 < M2 < r2. The condition
E = 1 is equivalent to
a' = flr r2-M2.
In order to eliminate the curve parameter t, we view s as a function of gyp;
hence ds/dcp = s'/V,
d,p
= f M r2 - M2,
and thus
rSOP) 1 r,v
MJ
r(s) r(s) - M2
ds=J do =fop.
,o 0
This provides a parametrization of the geodesic lines in the case of surfaces

for which the integral on the left-hand side is elementarily integrable and
can be solved for s.
Example 23. Let 7.(2 be the hyperbolic plane (see Example 9). We again
parametrize the geodesic ry as 'y(s) = (x(s), y(s)). Then the condition that
-1 is parametrized by arc length takes the form
xi2 i2
y2 + y2 = 1 .
Since the metric does not depend on x, translations in the x-direction are
isometries; the associated vector field is 8/8x. Hence
M :_ (y', 3/ex) = lye
is a first integral. If M vanishes, x has to be constant, and the geodesic
lines are precisely the open straight half-lines parallel to the y-axis. In the
general case, rewrite the first condition, inserting M. A short calculation
leads to
y = ±y l - M2y2.
Again, we view y as a function of x and set r:= 1/jMI,
dy
= y 1 r2 _ y2
dx x' y
Directly integrating this equation and denoting the integration constant by
a yields
- r2 - y2 = (x - a) hence (x - a)2 + y2 = r2.
The trace of the geodesic line is thus a semi-circle whose center lies on the
x-axis.
Once again we now turn to the Gauss-Bonnet formula. Let M2 be an ori-

ented, compact surface with boundary 9M2, and let V be a vector field of
length one defined on M2 except at finitely many points xi, ... , xk in the
interior of M2. Assume, moreover, that V is perpendicular to 9M2 at the
boundary 3M2 and outwardly directed. Consider the 1-form Xv introduced
in §5.4. From the formulas proved there, we deduce
aM2 k
dXv = Xv - 27r Ind (V, xi).
J J
Near the boundary 8M2, we choose an orthonormal frame el, e2 such that
el = V. Then Xv = w12, and e2 is the unit tangent vector to the boundary.
Since W12(e2) = I(Dezel, e2) = -I(el, V2e2) = K9, we obtain the Gauss-
Bonnet formula for manifolds with boundary.
Theorem 38 (Gauss-Bonnet Formula). Let M2 be a compact, oriented sur-
face with boundary, and let V be a vector field of length one defined on M2
except at finitely many points xl, ... , xk in the interior of M2. Assume,
moreover, that at the boundary V is perpendicular to the boundary and out-

wardly directed. Then
f f k
G dM2 + icy d(8M2) = 2a Ind(V, xi).
J
,112 J0M2 i=1
This has similar consequences as in the case of surfaces without boundary.

On the one hand, the sum of the integrals
IMZ f1112 K9.d(aM2)

8
is independent of the first fundamental form. On the other hand, the sum
of the indices of a vector field does not depend on the vector field.
5.6. Maps between Surfaces

The length IIVII = I(V, V) of a tangent vector can be determined by
measuring the length of curves. To see this, choose a curve -y on the surface
J112 in arbitrary parametrization whose tangent at t = 0 coincides with V,
y(O) = V. The length of the curve segment y([0, t]) is defined as
L(t) = f0 I("Y(Ft),'Y(p))dp ,
and computing the derivative at t = 0 yields

dL(t)
I(V, V) .
dt t=o
The first fundamental form of the surface is thus completely determined by
the lengths of curve segments, and vice versa. Hence each map f : M2 M2
between two surfaces which does not alter the lengths of curve segments has
a differential df : TM2 -+ TM2 preserving the first fundamental forms I,1f2
and IAt2,
I (IT12) = IJ12.
Definition 18. An isometry f : M2 M2 between two surfaces is a
length-preserving smooth map.
The Gaussian curvature of a surface is a geometric quantity depending on
the first fundamental form. This immediately implies
Theorem 39. Let f : M2 -+ M2 be an isometry. Then the Gaussian
curvatures G,2, Gh12 at corresponding points coincide:
GM2(x) = Gh12(f(x))
Corollary 4. There does not exist an isometric map from an open subset
of S2 onto an open subset of the euclidean space R2.
5.6. Maps between Surfaces 181
For this reason, cartography was forced to look for other suitable maps. Es-
pecially important are conformal (angle-preserving) and volume-preserving
maps. Let us first turn to the former ones. A map f : M2 AI2 is called
conformal (or angle-preserving) if the angle between two intersecting curves
11,12 in M2 is the same as that of the image curves f o 11, f o y2 in R12.
These maps can be characterized using the first fundamental form.
Theorem 40. A map f : M2 11 I2 is conformal if and only if there exists

a positive function h : M2 - (0, oc) such that f `(ITf2) = h I tif2.
Proof. This is an easily proved fact from linear algebra. Let ( , ) 1 and 2
be two positive definite scalar products on a two-dimensional vector space
satisfying the equation
(V, W)1 (V, W)2
IIVII11IWII1 IIVII2IIWI12
for all vectors V. W. Then there exists a positive number h such that
(V, W)2 = h (V, W)1 .
We already encountered conformal maps. The normal map e3 : M2 _ S2

of a minimal surface in R3 is conformal (Theorem 30). Complex analytic
functions are conformal maps from the plane R2 to itself. In fact, if f =
(u, v) = u + i v is a map f : 1R2 1R2, then the following formula holds for
the first fundamental form IR2 = dx2 + dy2:
f'(Ia2) = (utdx+ul,dy)2+(v.dx+vydy)2
= (u2 + v2)dx2 + 2(uruy + v=vy)dx O dy + (v2 + vy)dy2.
The Cauchy-Riemann equations, ur = vy, uy = -vi, imply that, for the

induced form,
f(1R2) = If,(z)12.Ia2.
Hence every complex-differentiable function f = u + iv is a conformal map,

and the square of the absolute value of its derivative, I f'(x)12, appears as the
conformal factor in the sense of two-dimensional geometry. The best-known
conformal map is the stereographic projection from the two-dimensional
sphere to the plane. This shows that conformal maps from R2 to S2 ex-
ist, and so angle-preserving geographic maps can be drawn.
Example 24. In coordinates, the stereographic projection f : R2 S2 is
the map
2x 2y 1 - x2 - y2
(l+x2+y2' 1+x2+y2' 1+x2+y2).
The partial derivative vectors of f are
Of 2 (1 - x2 + y2, -2xy, -2x) ,

Ox - (1+x2+y2)2
Of 2
(I + x2 + y2)2 (-2xy, 1 + x2 - y2, -2y)
-
and hence
Of Of _ 4 _ Of Of 8f Of
and 0.
\ 8x' 8x (1 + x2 + y2)2 8y' ay 8x' 8y _
These formulas show that the stereographic projection is conformal.
Example 25 (Mercator projection). Let (gyp, tai) be spherical coordinates
on the sphere S2\{N, S} outside the north and the south pole (see §3.2,
Example 14). Define a map from the sphere S2 to R2 by the formula
log (tan(4 + 2111

(V,
This turns out to be a conformal map with an additional interesting prop-

erty: curves of constant azimuth (i.e. forming constant angles with the
meridians on S2) are mapped to straight lines under f. For this reason, the
Mercator projection (1569) has been one of the most widely used methods
for the creation of nautical maps for centuries. The reader may find an
extensive discussion of further important map projections in the book by
K. Strubecker (Volume 2) cited in the bibliography.
Another interesting class of maps between surfaces are those preserving the
two-dimensional volume (surface area) of measurable subsets of the surface.
Such a map f : M2 --+ M2 is characterized by the property that the volume
form is preserved,
f*(dM2) = dM2.
We will call such an f an area-preserving map. Isometries are the maps
which are both conformal and area preserving.
Theorem 41. A map f : M2 M2 is an isometry if and only if f is
conformal and area-preserving.
Proof. The relation f*(IM2) = h. IM2 implies f*(dM2) = h2d.112' and

hence also the assertion.
The sphere S2 (or pieces of it) thus cannot be mapped to the plane in
an angle- and area-preserving way. There exist, however, angle-preserving
maps, as the stereographic projection shows. We now want to describe an
example for an area-preserving map from S2 to 1R2.
5.7. Higher-Dimensional Riemannian Manifolds 183
Example 26. Delete the north and the south pole from the sphere S2, join
them by a straight line, and consider the cylinder whose axis this line is:
Z = {(x,y,z) E R3: x2+y2 = 1}.
To each point different from the north and the south pole, P 0 N, S, there
corresponds one straight line containing the point P which is perpendicular
to the fixed axis. This line intersects the cylinder in two points, and we
denote the one closer to P by f (P). This defines the so-called Lambert
projection,
f : S2\{N,S} -> Z.
Composing this with an isometry from the cylinder to the plane, we obtain
an area-preserving map from S2 to R2.
To see this, let el, e2 and ei, e2 be the orthonormal frames on S2 and Z.
respectively, indicated in the above picture. Then an elementary geometric
argument immediately leads to the relation
df(ei) = rei, df(e2) = Tee
Here r is the radius of the meridian of S2 tangent to e2. These formulas
prove that f is an area-preserving map.
5.7. Higher-Dimensional Riemannian Manifolds

Following §3.11, we will conclude this chapter by a brief survey of higher-
dimensional Riemannian manifolds. Only the basic facts are discussed, and
no comprehensive introduction into Riemannian geometry is intended. As
in the case of surfaces, every pseudo-Riemannian manifold (11I'", g) can be
endowed with a covariant derivative V of vector fields with respect to vector
fields which is uniquely determined by two conditions. This formalizes the

notion of parallel displacement of vectors on a manifold. In fact, Levi-
Civita's idea (1916) that every Riemannian space admits a unique parallel
displacement preserving the metric turned out to be fundamental for the
further development of geometry.
Theorem 42. Every pseudo-Riemannian manifold has precisely one covari-
ant derivative V W of vector fields with the following properties:
(1) V (W1 + W2) = V (Wi) + V (W2)
(2) Vv, +v2 W = Vv, W + Vv, W ;
(3) if f is a smooth function, then
Vv(f-W) = and V f.vW = f VvW ;
(4) Vg(WI, W2) = g(VvW1, W2) +g(W1, VvW2)

(5) VvW - VwV = [V, W).
The vector field VvW is determined by the expression
2g(VuV, W) = U(g(V,W))+V(g(U,W))-W(g(U,V))+g([U,V}, W)
+g(V, [W, 141) - g(U, [V, W])
Proof. We prove uniqueness of the covariant derivative and derive in par-

ticular the stated formula for VvW. By assumption, we have
U(g(v, W)) = g(VuV, W) + g(V, V W) ,
V(g(U, W)) = g(VvU, W) + g(U, V W),
W(g(U, V)) = g(VwU, V) + g(U, V WV).
Because of the properties of V, we may introduce the commutators of the
vector fields into these equations:
W(g(U,V)) = g(VuW,V)+g(U,VvW)+g([W,U],V)-g(U, [V, W]),
V(g(U,W)) = g(VuV,W)+g(U, VvW)-g(W, [U,VI)
From the first, fourth, and fifth equation we obtain the formula for the scalar
product g(VuV, W). Hence there exists at most one covariant derivative
with the required properties. Conversely, the formula determines the vector
VvW, and an analogous computation shows that the operation thus defined
is a covariant derivative with all the properties needed.
Definition 19. The covariant derivative V W is called the Levi-Civita
connection of the pseudo-Riemannian manifold.
Let V1, .. . , Vm be an arbitrary frame of vector fields defined on an open

subset of Mm. Representing the covariant derivative of the basis vector
fields as
EU]
OV,Vj 17
13
- Vk,
k=1
we introduce the Christoffel symbols of the Levi-Civita connection. The
symmetry properties of the functions r strongly depend on the chosen
frame. If the vector fields Vi mutually commute, [Vi, Vj] = 0, and if gij =
g(Vi, Vj) are the coefficients of the Riemannian metric, then formula (5)
from Theorem 42 simplifies, and we obtain the expression
m
rj = 9>9'(Vi(9j0)+Vj(9ta)-Va(9ij)).
a=1
In this case, the Christoffel symbols are symmetric in the lower indices,
r =r (for commuting frames).
If the frame V1 := el, ..., Vm := em consists of mutually orthogonal vector
fields of constant length ±1, then the equality
0 = Vi(g(Vj,Vk)) = 9(VV,Vj,Vk)+9(Vj,VV.Vk)
implies the following symmetry property of the Christoffel symbols:
r = -rk (for orthonormal frames).
Example 27. The first fundamental form of a surface of revolution can be
written (see Example 3) as
I = ds2 + r(s)2 dcp2 .
With respect to the commuting frame V1 = 8/8s, V2 = 8/8cp we obtain, for
the "classical" Christoffel symbol rte,
I'l
22
r,g aQ
p9 + a-g ,0 -
8 -r
2 I as9"''J 2 [ as )2 J r'
=8.;
and similarly rig = r21 = r'/r. The other Christoffel symbols vanish.
From Example 3 we take the formula W12 = (r'/r) 02 with respect to the
orthonormal frame el = 8/8s, e2 = (1/r) 8/8cp and its dual frame ?1,02
considered there. Together with the definition of the "Cartan" Christoffel
symbols,
W12 = r11 ai + r21 o2 = 0.01 + -a2,
this yields r2 1l = 0 and r2, = r'/r. Similarly we have
W21 = r11
12 o1+r1
22 a2
2 =
r
implying r12 = 0, r22 = -r21.

Again denoting by Q1, the 1-forms dual to the frame e1.... , c,,,, we
introduce the local connection forms
wij := Ej -9(Vei,ej) = oj(Vei) with Ej :=g(ej,ej)
By definition, they satisfy
m
wij = rQi - va and Ei - wij + Ej wji = 0 .
a=1
N%,e compute the exterior derivative doi using the properties of the Levi-
Civita connection:
dui(V,W) = V(Qi(W)) - W(oi(V)) -oi([V,WI)
M
_ Eej-{9(Vvei,ea).a.(W)-9(Vwei,ea)-a.(V)}
a=1
M
_ 1: aQAw,,i(V,W)
a=I
This leads to the first structural equation of a pseudo-Riemannian space.
In order to formulate the second structural equation, denote by SZ := (wi j )
the antisymmetric (m x m) matrix composed of the 1-forms 4;ij. This addi-
tional structural equation computes the matrix d!2 - S1 A S2 of 2-forms, and
to formulate it we need the curvature of the space. The curvature tensor
7 (U. V)W := VuVVW-VVOuW-V1U,VIW of a pseudo-Riemannian man-
ifold is introduced generalizing the case of surfaces, and all the identities of
Theorem 15 in §5.3 remain valid. In particular, the first Bianchi identity
holds:
R(u, V)W + R(V,W)u + R(W,u)V = 0.
However, in the higher-dimensional case the curvature tensor does not re-
duce to a single function. By definition, it is a (3,1)-tensor. This can be
transformed into a (4, 0)-tensor, the Riemannian curvature tensor:
R(U,V, W, Wi) := 9(R(U, V)W, W1) .
Referring to an orthonormal frame e1, ... , e,,, of basis vector fields of lengths
±1, we obtain from
m
R(ei, ej )ek Rijk el
1=1
the components Rijk of the curvature tensor. These can be computed by
the formula
Rtijk = El . R(ei, ej, ek, el)
from the Riemannian curvature tensor.
Theorem 43 (Structural Equations for Pseudo-Riemannian Manifolds).

Let el, ... , em be a local orthonormal frame on a pseudo-Riernannian man-
ifold, and let o1, ... , an be the dual frame. Then the following equations
hold:
Lul
(1) dai = 1: a. A wai ;

Q=1
m 1 m
(2) dwij = win A waj + as A a3 .
2
a=1 a,J3=1
Proof. We already derived the first equation. To prove the second, we

use the properties of the Levi-Civita connection and obtain, after several
elementary steps,
dwij(V,W) = a (R(V, W)ei) +Ej - (g(Vwe;, Vvej) - g(Vvei, Vwej))
71'
_ aj(R(V, W)ei) + ejEa(Qa(Vwei)Qa(Ovej) - aa(Vwej)an(Vvei))

a=1
Moreover,
ej - e -an(Vvej) = -wnj(V)
and, similarly, ej en aa(Vwej) _ (W). Altogether, this completes
the proof,
m
dwij(V,W) _ winAwaj(V.W)+oj(R.(V.W)ei)
a=1
_ wiaA waj(V,W)+RVWi 0
a=1
Corollary 5. If the curvature tensor of a pseudo-Riemannian manifold X11"'
vanishes, then in a neighborhood of each point there exists a chart in which
the coefficients of the Riemannian metric are constant,
g = diag(±1, ....±1).
Proof. First choose a local orthonormal frame e1, ... e.. and consider the
matrix n = (wij) of connection forms. Then
dfl = SZ A Q,
since the curvature tensor vanishes. By Theorem 6 in §4.3. there exists a
(pseudo-)orthogonal matrix A of functions such that
ft = dA A-1 and A(O) = diag(± 1.... , ±1).
Denote by r := (a1 ... a,,,) A a special system of 1-forms determined by

a1, . . . , o,,, and the matrix A. The structural equations imply that this form
is closed.
dr = 0.
Hence, by Poincare's lemma, each 1-form r, can locally be represented as the
differential of a function fi. This way, we construct local coordinates on the
manifold, and for the pseudo-Riemannian metric g we obtain the expression
g = ±ai + ... t a2, = ±rl + ... f T'2 = ±(dfl )Z + ... f (dfm)Z.
Definition 20. A pseudo-Riemannian manifold with vanishing curvature'
tensor is called flat.
The preceding corollary thus states that every flat pseudo-Rieinannian space
is locally isometric to the standard space Rm.k with the following metric of
index k:
dxl &I +... + dxm-k
- ,
dXin-k dxm-k+l dXm-k+l - ... - dx"' ) dx'n.
Consider a curve -y(t) in the pseudo-Riemannian manifold (111',g) and a

vector field W(t) along this curve. In addition, we may suppose that locally
the latter is the restriction of a vector field W defined on Mm. In the neigh-
borhood of a point on the curve we choose the coordinates yl.... , y'n and.
represent the vector field W and the tangent vector i(t) in these coordinates:.
m m
W = ` yi. ^'y(t) = E^"`(t)
yi
i=1 i=1
m A
d
E1'(t) ayi ('Y(t))
i=1
Then
dWA(7(t)) + ['
ii(t) . W'(7(t)) . q(7m) 8
(07(r)W)(y(t)) = > dt 8yk
k=1
Along the curve y(t), this vector field is independent of the particular (lo-cal)
extension of the vector field W(t). originally defined only on y(t), to
a vector field W. The formula thus uniquely defines a vector field VW/dt,l
the covariant derivative of W(t).
Definition 21. A vector field W(t) is called parallel along y(t) if
vW
dt 0.
In coordinates, this condition is equivalent to the following system of ordi-

nary differential equations:
dWk
14#
o.
The existence and uniqueness theorem for ordinary differential equations

ensures the existence of a unique parallel vector field W (t) for every initial
vector W. E TAi a) at the initial point y(a) on the curve,
vw
dt = 0. W(a) = Wa.
The value LV(b) at the end point ';f(b) of the curve is called the parallel
displacement P., of the vector W. along the curve y(t),
P,(lV) := IV(b).
Because of the linearity of the differential equation and the property
U(g(V,W)) = g(V V,W)+g(V,vuW)
of the Levi-Civita connection, the parallel displacement
Try(b)A.lm
P.7:T.(,.)Alm
is a linear isometric map between tangent spaces. The parallel displacement
P.y,..Y2 along a curve formed by combining two curves y1 and y2, for which
the end point of the first coincides with the initial point of the second, into
a single curve y1 * 72 leads to the same result as the superposition of the
single parallel displacements,
P12 o P,., = P'Yl *'12
Consider a surface (1112, g) with a positive definite Riemannian metric and
a closed curve y in M2 bounding a domain N2 C 1112. Parallel displacement
along y then is a rotation in the tangent space T.,(o)M2. The corresponding
rotation angle can be computed from the Gaussian curvature.
Theorem 44. If the closed curve y bounds a domain N2 in a surface AI2
with Gaussian curvature G, the parallel displacement along the curve is the
rotation through the angle
0=
JN2 r
Proof. Choose an orthonormal frame e1,e2 along -y in which e1 is tangent

to the curve. In any parametrization y : [0, L] Ail of the curve, a vector
field V = V1(s) e1(s) + V2(s) e2(s) is parallel if
dV1 dV2
= w12(e1) . V2 and = -w12(el) V1.
ds ds
Consider the complex-valued function z : [0, L] C defined by z(s) :=

Vl(s) + i. V2(s). Then we can write this system of ordinary differential
equations in the form
dz
Z /
= -2 W12(el) z.
The solution of this is
z(8) = z(0) exp l _ i e Wl2(el) dsJ

J0
and, in particular, for the rotation angle of the parallel displacement we
obtain
0 = - fr w12 = JN2 d'012 = JG.dM2.
Example 28. It is a peculiar property of the hyperbolic space 7.12 that
the hyperbolic angle coincides with the usual euclidean angle. In general,
the angle does not change under parallel displacement. Hence the parallel
displacement of a vector along a geodesic line in a surface is the vector at
the end point of the line forming the same angle with the tangent vector as
the original vector did at the initial point. The points PQR in the picture
below are the vertices of a geodesic triangle in the hyperbolic plane (see
Example 23). Start at P with any tangent vector (1), translate it to the
point Q (2), then along the upper geodesic arc to R (3), and finally along
the straight segment RP back to P (4). The resulting vector obviously does
not coincide with the original one.
Parallel displacement allows one to generalize the covariant derivative V VT

with respect to vector fields to arbitrary tensors T. We explain this for ten-
sors of type (k, 0). Assume that at a point a E Mm, the vectors W1.... W. E
TXM' are given, and let y(t) be the integral curve of the vector field V pass-
ing at time t = 0 through the point x E Mm. Then we define
d
(VyT)(WI, ...,)4)(x) := at(Ty(t)(Py(WI),
...,P7(Wk))It=o'
Here Py denotes parallel displacement along the curve y from the point
x = y(0) to the curve point -f(t). If, for example, T is a (1, 0)-tensor and W
a vector field, we obtain the formula
(VyT)(W)(x) = V(T(W))(x) - T(VyW)(x).
Theorem 45. The covariant derivative VyT of a (k, 0)-tensor is given by
(VvT)(Wi, ..., Wk) = V(T(Wi, ... , Wk)) - T(VvW1, W2, ... , Wk)
- ... - T(W1i ... , Wk-1, VVWk) .
Remark. The metric g, viewed as a (2, 0)-tensor, is parallel:
(Vy9)(Wl, W2) = V(9(Wi, W2)) - 9(VvW1, W2) - g(WI, VvW22) = 0.
A completely elementary but somewhat lengthy computation, using the
properties of the Levi-Civita connection and the definition of the curvature
tensor, leads to an identity for the covariant derivative VTZ of the curvature
tensor, the so-called second Bianchi identity. We omit the simple proof and
formulate only the result.
Theorem 46 (Second Bianchi Identity). Let U, V, W, W1, W2 be five vector
fields on the pseudo-Riemannian manifold (M', g). Then
(VuR)(V, W, W1, W2)+(Vy7)(W,U, W1, W2)+(VWR)(U, V, W1, W2) = 0.
Define the length of a curve y : [a, b] -+ Mm in a pseudo-Riemannian space
by the integral
L(y) :=1 b I9( (t), ry(t)) I dt .

a
For a variation yE : [a, b] - Mm of the original curve y(t) with fixed initial
and end points, y. (a) = y(a),-yy(b) = y(b), the field of the variation,
W (t) (yE (t)) I E=o,
is a vector field defined along the curve y(t) vanishing at the end points. Its
covariant derivative VW/dt along the curve y(t) coincides with the covariant
derivative Vyy(t)/de of the vector field ye(t) defined along the curve i7(c)
-Y" (t),
V t)
dt (t) - de L-0
This formula is an immediate consequence of the local expressions for the

two covariant derivatives. Parametrizing the curve y(t) by are length. we
obtain the following formula for the variation of the length of the curve:
d(L(,E))IE_o =
J
g(y t)
1==0
,y(t))dt
lb g (W(t), 7- (t))dt.
= jbg(dw(t),ry(t))dt =
Theorem 47. If the curve y : [a, b] Mm is a critical point of the length

functional defined on the set of all curves with fixed initial and end points.
then the tangent vector y: (t) is parallel along the curve y(t).
V7
0.
dt =
Definition 22. A curve y(t) in the pseudo-Riemannian space (M"'.g)
whose tangent, vector field is parallel is called a geodesic line.
Remark. If the curve y(t) = (y' (t), ..., ym(t)) is given in local coordinates,
the equation for it to be a geodesic line is
d2yk M dy' dy'
dt2 + E dt -;it
r = 0.
*a=1
This is an ordinary differential equation of second order, and Corollary 3

from §5.5 holds correspondingly:
Corollary 6. To each point x E M' and every vector V E TIMtm on
a pseudo- Riemannian manifold without boundary there exists precisely one
geodesic line y,; : (-e, e) -+ Mm such that -y, (0) = x and i,(0) = V.
The geodesic line y (t) thereby determined is not necessarily defined for
all values of the parameter t E R. The subset Ex C TxA1t of all vectors
V E TxMt for which exists for t = 1 is an open neighborhood of the
zero vector in TTM". The exponential map at the point x of the pseudo-
Riemannian manifold maps Ez to the base space,
Expx : Ex -p Mm, Expx(V) = yv(1)
The exponential map is smooth, and its differential d(Expx) : To(T1M'") _
T,Al"' -. TZMm is the identity,
dExpx(W) = dt(Exp(tW))It=o = dt(y'.w(1))It=" = W.
Hence the exponential map Expx is a diffeomorphism from an open neigh-
borhood 0 E Ux C TZMm of the zero vector in the tangent space onto an
open neighborhood x E Vx C M^' of the point in the base space.
Simpler curvature invariants of the pseudo-Riemannian manifold arise from

the curvature tensor by contraction (taking the trace), the most important
ones being the Ricci tensor and the scalar curvature. The sectional curvature
appears as the Gaussian curvature of subplanes of the tangent space. l e
shall now discuss these curvature quantities.
Definition 23. The Ricci tensor Ric is the first contraction of the curvature
tensor,
m
Ric(U, V) := F'g'"R(ei, U, V, e3) .
ij=1
The algebraic identities of the curvature tensor imply that Ric is a symmetric
(2,0)-tensor,
Ric(U, V) = Ric(V,U).
Taking the trace of the Ricci tensor with respect to the pseudo-Riemannian
metric g (§5.4) leads to the scalar curvature r of the space.
Definition 24. The scalar curvature of a Riemannian manifold is the func-
tion
T ._ ij9k1R(ei, ek, el, e3)
i j.k.l=1
Contrary to the surface case (m = 2), the curvature tensor is no longer de-
termined by the scalar curvature for manifolds of dimension at least three.
Thus the Ricci tensor and the scalar curvature do not contain all the cur-
vature information about a manifold. The last of the curvature quantities
is the sectional curvature. Apart from the point of the manifold it depends,
in addition, on a non-degenerate 2-plane in the tangent space. This is a
two-dimensional subspace E2 C ,,Mm whose orthogonal complement.
(E2)1 = {W E T Al- : g(V,W) = 0 for all V E E2},
is complementary to E2, E2 e (E2)1 = TXM. Intersecting E2 with an
open set 0 E U= C T=Ai"` on which the exponential map is defined and a
diffeomorphism, we obtain a surface in M"',
M2(E2) := Expr(E2 n U:) .
Its tangent plane at the point x is E2, and, by restriction, the pseudo-
Riemannian metric g of M'" induces a metric on M2(E2). The Gaussian
curvature G(x) of the latter at the point x E h12(E2) is called the sectional
curvature K(x. E2) of the pseudo-Riemannian manifold at the point x in
the direction of the 2-plane E2. It can be computed using the curvature
tensor of Af '.
Theorem 48. Let V1, V2 be a basis of the non-degenerate plane E2. Then
K(x, E2) -- 1Z(V1, V2, V2, Vl)

g(Vl, V1) g(V2, V2) - g2 (VI, V2)
Proof. For the proof, we assume that the metric of MI is positive definite.
This is no essential restriction, but some of the formulas simplify. Note first
that the right-band side of the formula for K(x, E2) does not depend on
the choice of the basis V1, V2; thus it suffices to prove this formula for an
orthonormal basis in E2. Fix an orthonormal basis e1(x), e2(x), ... , e,,,(x)
at the point x E C Al" such that the first two vectors belong to E2.
We extend this basis to a local orthonormal frame by parallel displacement
along the geodesic straight segments {Expx(t W) : W E TXMm}. At the
point x this frame is parallel,
Vei(x) = 0.
Furthermore, on 1612(E2) the vector fields el, e2 are tangent to this surface,
since M2(E2) consists of geodesic straight segments with initial vectors in
E2. Let an, w,j be the forms dual to the frame e1, ..., e,,, and the connection
forms of the Riemannian manifold Mm, respectively. The embedding f :
M2(E2) Mm allows us to restrict these forms to the surface,
A. := J'(aa), ti.i f'(wij)
Then µ3 = p4 = ... = um = 0, and the structural equations of M'" imply,
after restriction to M2(E2), that
d{ll = ll2 t21, dA2 = n t12, X12 = t In A Cat + n121 pI A 142-
Sln
a
The pair µ1,/l2 is, within the surface M2(E2), the frame dual to el, e2, and
the last equations show that the connection form w12 of M2(E2) Coincides
with t;12, wit = e12. The Gaussian curvature a of M2(E2) is determined
by
- G - JAI A P2 = &w12 = d512 = tlnAta2+R2121 p1 AP2
At the point x E M2(E2), however, we have = g(De,e1, ea) = 0, and

hence we obtain
- G(x) = R2121 = g(R(el, e2) el, e2) = - g(7Z(el, e2) e2, e1).
Special Riemannian manifolds are the so-called Einstein spaces and spaces
of constant sectional curvature.
Definition 25. A pseudo-Riemannian manifold is called an Einstein space
if the Ricci tensor is proportional to the metric, Ric = h g.
Remark. Contracting the equation Ric = h g leads to r = m h, i. e., the

factor of proportionality is uniquely determined by the scalar curvature of
the manifold,
T
Ric = m
_19.
We prove that this factor is constant.
Theorem 49. If (Mm. g) is a connected Einstein space of dimension m > 3.
then the scalar curvature T is constant.
Proof. Choose an orthonormal frame el, ...,en and an arbitrary vector

field U. After we insert the vector fields V = W2 := ei, W := ej, and W1
ej into the second Bianchi identity and sum over all indices 1 < i, j < m,
an elementary computation leads to the equation
> ej ((Vu)Ric(ej, ej) - 2(V Ric)(U, ei)) = 0.

i
If the Ricci tensor is proportional to the metric, Ric = h g, the relation
Vg = 0 implies VuRic = U(h) g, and the last equation becomes
m U(h) - 2E ejej (h)g(U,ei) = 0.
Considering the vector fields U := ek, this yields (m - 2) dh(ek) = 0, and,
in case m > 2, the differential dh of the function h vanishes. 0
Example 29. In spherical coordinates, the standard metric of the two-
dimensional sphere is the quadratic form
gJs2 = COS2(71') djl2 + d/12 .
Choose a pseudo-Riemannian metric h in the product R+ x R and a function

f : R+ x R R. On the four-dimensional manifold (1R3 -0) x lR = S2 x R+ x R
we can then consider the rotationally symmetric metric
g := h (B e2f(r,t) . gJs2 .
A longer computation reveals that g is a four-dimensional Einstein metric

with vanishing scalar curvature if and only if the following pair of differential
equations is satisfied:
e-2f = A(f)+2lgrad(f)I2 and Odf.
Here, A and grad are the Laplacian and the gradient of the two-dimensional
pseudo- Riemannian manifold (R+ x Ift, h), G is its Gaussian curvature, and
Hess h(f)(V, W) := h(Vy(grad(f )), W)
is the symmetric Hessian form of the function f with respect to the metric
h. Choosing, for example, the metric
2A1 r 2Af 11
h= Ll + dr2 + c dr C dt - c2 I 1 - dt2
r J r 111
r J
and the function f (r, t) = log(r) gives rise to the Schwarzschild-Eddington
metric on (R3 - 0) x R, which, in Einstein's general theory of relativity,
models the gravitational field generated by a mass concentrated at the point
x = 0. Further solutions of this system of differential equations describe the
gravitational field of a rotating mass (Kerr metric).
Definition 26. A pseudo-Riemannian manifold (Mm,g) is called a space
of constant sectional curvature (or space form) if there exists a function
K' : 111"' -+ R such that for every 2-plane E2 C TXMm
K(x, E2) = K'(x).
Theorem 50. The sectional curvature K = K' of every space Mm
of dimension m > 3 is constant, and the curvature tensor is algebraically
determined by the metric,
R.(U,V)W = K'(g(V,W) -U -g(U,W) -V).
Proof. Consider the tensor
S(U, V, W1, W2)
= R(U, V, W1, W2) - K* (9(V, Wi)9(U, W2) - 9(U, W1)9(V, W2)),
and note that S has the same symmetry properties as the curvature tensor:
(1) S(U,V,W1,W2) = -S(V,U,W1,W2).
(2) S(U, V, W1, W2) = -S(V,U, W2, W1) .
(3) S(U,V,W1,W2)+S(V,W1,U,W2)+S(W1,U,V,W2) = 0.
(4) S(U, V, W1i W2) = S(W1, W2,U, V) .
An additional property of the tensor S results from the formula for the
sectional curvature:
(5) S(U, V, V, W) = 0 .
A calculation only making use of equations (1) - (5) shows that this tensor
has to vanish, S - 0. In fact, from
0 = S(U. V + W2, V + W2iU)
= S(U. V. V, U) + S(U, V, W2, U) + S(U, W2, V, U) + S(U. W2, W2, U)
= 2 - S(U, V, W2, U)
we first conclude that S(U, V, W2, U) = 0. Hence we have succeeded in

replacing the third argument. V in the identity (5) by an arbitrary vector
W2. We repeat this step once more, starting from S(U, V, W2, U) = 0. Thus,
as the fourth entry another independent vector W1 can be substituted for
the vector U, that is.. S vanishes, and the curvature tensor of a space form
is determined by
R(U, V)W = K`(g(V, W) U - g(U, W) V).
Contracting the curvature tensor, the resulting Ricci tensor is proportional
to the metric,
Ric = (m-1) K*.g.
Hence M"' is an Einstein space, and K* is a constant function. 0
The components Ra31 of the curvature tensor of a space form with positive
definite metric (for simplicity) are
Ra3i = K' - (b3idaj - 63j6ai),
and the structural equations simplify-,
m
do'i = Qa A Wai dwij = E w1a A ula j - K* ai A o'j.
a=1 a=1
In the middle of the 19th century. Riemann gave the complete local descrip-
tion of these spaces.
Theorem 51 (Riemann, 1854). Each space form is locally isometric to the
spherical space S'", the euclidian space R"', or the hyperbolic space 1.tm.
Exercises
1. Let y(s) be a curve in its natural parametrization, and let p(t) be this
same curve in an arbitrary parametrization, which hence is related to 7 via
Ea(t) = (s(t))-
a) Prove the following formula for the second derivative of p:
d2µ(t) _ d27 ds 2 dy d2s =
(t) (s)2 h + s t
µ(t) = dt2 182 (dt+ ds dt2
.
b) Deduce from this the formula for the curvature in a general parametri-
zation:
mo(t) = IIi(t) X µ(011
IIi (t)113
c) Prove that the torsion can be written in the Frenet frame as
det (7', 7", -y"')
r(t) (Fx h', h) = det (t, h, h') = K(t)2
d) Derive in an analogous way to a) a formula for -ii (t), and conclude that
the torsion is given in an arbitrary parametrization by
det (µ(t), µ(t), i (t))
r(t) =
IIi (t) x A(t)112
2. Let C C R3 be a curve all of whose tangents pass through one and the
same point. Prove that C is part of a straight line.
3. Let C C 1123 be a curve all of whose tangents are parallel to one and the
same straight line. Prove that C is part of a straight line.
4. Show that the tractrix (Example 11) is the curve passing through the
point (1.0) on the horizontal axis with the property that the length of the
segment of the tangent line from any point on the curve to the vertical axis
is constant. The following intuitive interpretation gave it its German name
"Schleppkurve"(tow curve): It is the path that a dog pulling on a leash to
the west is constrained to take when his master walks along a north-south
path.
5. Compute the curvature and the torsion of the following curves:

a) y(t) = (t. a(e'/° + e-t/a)/2, 0) (catenary);
Exercises 199
b) -y(t) = (a cost, a sin t, b t) (helix);

c) y(t) = (a(t - sint), a(l - cost), bt);
d) y(t) = (t, t2, t3).
6. Prove that y(t) = (at, b t2, t3) with a, b > 0 is a slope line if and only if
2b2 = 3a.
7. Prove that a C3-curve C is a helix if and only if its curvature r, > 0 and
its torsion r are constant.
8. Determine all plane curves C C R2 satisfying

a) K(s) = const;
b) c(s) = 1
c) K(s) =
9. The Darboux vector of a curve with non-vanishing curvature is the vecto

C T:= r Prove that the Frenet formulas can be written in the following
form:
ds = d x t, ds = j -X hh,
ds
= d *X 6.
10. Consider the spherical curve y*(s) := t(s) S2 C iR3 and

: [0, L]
suppose that the curvature K(s) of the original curve y(s) does not vanish.
Prove that 'y' (s) is a regular curve, and compute its curvature K' and its
torsion r'. Which curve y*(s) arises in the case of the helix y(s)?
11. A slope line with slope angle a lies on a sphere of radius R if and only
if its curvature K(s) and its arc length s satisfy the relation
2
S
_ R2
tan2 a + K2 (S)
12. Let y(s) : [0, L] -> R2 = C be a closed plane curve, and let p be a point
in the plane not belonging to the curve. Prove that the integral
1j
L
t(s)
ds
21ri -Y(s) - P
is an integer. This is called the index or winding number of the curve y(s)
around the point p.
13 (Jacobi, 1842). Let -y C R3 be a closed curve with non-vanishing curva-

ture, and suppose that the curve of principal normal vectors h is a simple
closed curve in the sphere S2. Prove that h divides the sphere into two parts
of equal area.
14. Classify all surfaces of revolution with constant Gaussian curvature us-
ing the results from Examples 3 and 11. Altogether there are 9 geometric
types. Which of these can be compact manifolds without boundary?
15. Let F : U -* 1R3 be a parametrized surface such that the tangent vectors
r9/au and a/av are orthogonal at each point (u,v) E U. Set
a
E_ \ 49u, au. > , C= \
8v49
' Ov
and choose the orthonormal frame
_ 1 8 a
au' e2 _ 1
e av
a) In the dual frame al = /Edu, 02 = /dv, the connection form w12 is

given by
1 aV G
wl2 = al +
vE - av &
b) The Gaussian curvature is equal to
K = - EG ava a (8,/-G/8u)j
+OU
E
c) As an application, determine the Gaussian curvature of a surface of
revolution whose generating curve is not necessarily parametrized by
arc length.
16. Let 1112 C 11 P3 be a surface with non-vanishing Gaussian curvature.

Prove:
a) There exist two orthogonal vector fields V and W with lengths IIVII =
1 and IIWII = IHI whose flows are volume-preserving, Gy(dM2) =
Gw(dM2) = 0.
b) The integral curves of the vector field W/IH) are geodesic lines.
c) Let B : M2 - R be the function measuring the geodesic curvature of
the integral curve of the vector field V at a point. Then W(H) = H2 B.
d) The following commutator formula holds: [W/IHI, V) = B V.
Exercises 201
17. Discuss the geometric properties of the helicoid and the catenoid, and
study the map transforming the one into the other.
18. Enneper's minimal surface is parametrized by (z = a + ib)

F(a, b) = (Re(3z - z3), Re i(3z + z3), Re(-3z2)) .
Compute its Gaussian curvature. Prove, moreover, that its curvature lines
(i. e., the curves on the surface obtained by fixing one of the two parameters
a or b) all lie in a plane through the y- and z-axis and have polynomial arc
length. Determine the equation for the tangent plane to a point x E F and
transform it into
(p - q, x) = (p, p) - (q, q)
This is the equation of a plane with respect to which the two points p and
q are mirror images of one another. The curves p(a), q(b) are parabolas in
the perpendicular planes x = 0, y = 0, and the summit of each of them
coincides with the focus of the other (so-called focal parabolas). Illustrate
these properties using Maple or Mathematica.
19. The minimal surface studied by Bour (1862) arises from the Weierstrass
representation by inserting f (z) = czm, g(z) = z. Prove that it can be
mapped to a surface of revolution in a length-preserving way.
20. Prove the following properties of the modular surface of a holomorphic

function f : U -* C:
a) The modular surface is hyperbolically curved in a sufficiently small
neighborhood of a finite pole of f.
b) Let zo be a finite zero of f . If this zero is multiple, the modular surface is
elliptically curved in a sufficiently small neighborhood of zo. If the zero
is simple, every sufficiently small neighborhood of zo decomposes into
2k - 2 sector-like pieces, in which the modular surface is alternatingly
curved elliptically and hyperbolically. Here k is the exponent of the
non-vanishing term succeeding the summand a, (z - zo) in the series
expansion of f.
In the next two exercises it will be helpful to use the position functions pi
from the proof of the Minkowski-Steiner theorem.
21. Let M2 be a surface with normal vector e3. Prove that the 1-form
77 1= (x, de3) defined on the surface is closed, and conclude that, for a
curve ry C M2, the relation
j(xde3) = 0
holds if one of the following two conditions is satisfied:

a) The curve y is the boundary of a two-dimensional submanifold G2 C M2.
b) 111 is simply connected.
22. Let A12 be a compact surface without boundary. Prove that there exists
a point ro E A!2 with the following properties:
a) The normal vector e3 is parallel to the position vector rO at the point
ro.
b) The Gaussian curvature is positive at xo, G(xo) > 0.
Hint: The function f : M2 -+ R, f (x) = (x, x), has to have a critical point
on Ate.
23. Let the group G = SL(2,R) act on the hyperbolic plane ?{2 by
_ a b .z _ az+b
6F
cd cz+d'
Verify that the image point g z actually does belong to ?{2, and that super-
position of two of these transformations corresponds to matrix multiplication
in G. Show. moreover, that each g E G leaves the metric invariant; hence G
is an 3-dimensional group of isometries from N2 onto itself.
24. Let AI'; = 1R3 and denote by DyX the directional derivative of the
vector-valued function X : R3 R3 in the direction of the vector X. Prove
that
VXY := DXY+2X xY
defines a covariant derivative having all the properties (1)-(4) from Theorem
42. but violating property (5).
25. Consider on the set {(x,y) E R2 : -7r/2 < y < 7r/2} the pseudo-
Riemannian metric
1 _ dx2-dye
COS2
x'2 - y'2 x'

a) Prove that E _2(y) and P =2(y) are first integrals of the
cos
geodesic flow, satisfying, in addition, the inequality p2 - E > 0.
b) Discuss the geodesic lines on M2. To do so, assume that y is a function
of r and integrate the resulting ordinary differential equation.
c) On Al" there exist points that cannot be joined by a geodesic line.
Exercises 203
26. Prove that every three-dimensional Einstein space is a space of constant

curvature.
27. Let M' be a non-flat Einstein space (for example, S"'). Show that
MI x M"' with the product metric is an Einstein space, but not a space of
constant curvature.
28. It is well-known that a symmetric bilinear form h.(x, y) on a vector

space is completely determined by its quadratic form q(x) := h(x, x) (via the
polarization formula: 2h(x, y) = q(x + y) - q(x) - q(y)). Prove, in a similar
way, that the Riemannian curvature tensor 1 (U. V, W1, W2) is completely
determined by the quadratic form K(U, V) := R(U, V, V, U) corresponding
to the sectional curvature.
29. Prove that a four-dimensional Riemannian manifold is an Einstein space

if and only if for every 2-plane E2 C TM4 and its orthogonal complement
(E2)1 the corresponding sectional curvatures coincide, K(E2) = K((E2)1).
30. Because of its symmetry properties, the curvature tensor of a pseudo-

Riemannian manifold can be interpreted as a transformation R : A2(M'') -
A2 (M7°) on 2-forms,
n
1
R(ai Aaj) = 2 > Ro3i . a, Aa,3.
Q,:3=1
In dimension m = 4, this gives rise to an endomorphism R : A2(A14)

A2(M4). On the other hand, the Hodge operator * : A2(A14) A2(M4)
acts on A2(M4), and its square depends only on the index k of the metric (see
Theorem 5, Chapter 1): ** = (-1)k. In the cases k = 0 (positive definite
metric) and k = 2 (neutral metric), the Hodge operator decomposes the real
bundle A2(M4) into the corresponding eigenspaces A2 (M4) (see Exercise
8, Chapter 1). Prove that the block representation of the curvature tensor.
R _ R++ R-+
R+- R--
with respect to this decomposition of A2(M4) has the following properties:
a) 7Z is symmetric, i.e., R++ = R++, R__ = R__ and R+_ = R_+.
b) The traces of R++ and R__ coincide, tr(R++) = tr(R_-) = -r/12.
c) M4 is an Einstein space if and only if R+_ vanishes.
Literature: Th. Friedrich, Self-duality of Riemannian manifolds and connec-

tions, in: Riemannian geometry and instantons, Teubner-Verlag, Leipzig,
1981, 56-104.
The Einstein equation in the general theory of relativity combines the geo-
metric curvature quantities of a four-dimensional pseudo-Riemannian man-
ifold of signature (1, 3) with its physical properties encoded in the energy-
momentum tensor T,
Ric- rcT.
Here Ric. g. r are the Ricci tensor, the metric, and the scalar curvature;
K is a constant depending on the chosen system of units. Already for the
vacuum, T = 0, there are non-trivial, physically very interesting solutions of
this equation, among others the Schwarzschild metric to be discussed now.
Obviously, a vacuum solution of the Einstein equation has to be an Einstein
space with vanishing Ricci tensor in the sense of the definition given before.
31 (Schwarzschild metric). On a spherically symmetric and static space-time
manifold M4 (this is a pseudo-Riemannian manifold of signature (1, 3) with
isometry group SO(3, R) and a distinguished time direction), it is always
possible to introduce coordinates from R x R+ x S2 with respect to which
the metric can be written as
g = e2a(r) dt2 - [e 2b(r) dr2 + r2 (d92 + sin2 O dV2) J .
Here, the functions a(r) and b(r) asymptotically tend to zero for r - co
(the metric is "asymptotically flat"). We introduce the following basis of
1-forms:
ao = ea dt, Cl = eb dr, a2 = rdO, 03 = r sin 9 dcp .
a) Check that the metric satisfies g = 0,02 - 0i - 02 - a3 .
b) Compute the forms dai and show, using the first structural equation,
that the connection forms are given by
wo, = -a'e-bcO, WO2 = W03 = 0,
e-b a-b cot 9
W12 = -r 0`2, w13 = r- a3, w23 = r a3-
c) We introduce the notation lid := 2 E.4 Rtasia0 A0,3. (f is the so-called

curvature form). Show, using the second structural equation, that
110 = e-2b(a'b'-a"-a2)aoAal, f20 = -a'er"aoAa2,
6a1
A02,
-a'er6
03O = ao A 03, 1121 = bier
5231 = b'erbal A a3, X32 = '-'-'012 A a3,

Exercises 205
and compute from these formulas the components R kl of the curvature

tensor in this basis.
d) Compute the components of the Einstein tensor G := Ric - 1g ,r:
Goo= I e-2b (13 _ 2b
- ) , Gi l = ,1-s _e-2b(77 +),
7- F
G22 = G33 = -e-2b (a'2 - a'N + a-+ =d)

r
e) Solve the vacuum equation by means of this Ansatz. The result is (M
is a constant of integration)
1_1
r 1 r 2M
g = 11- 2M dt2 -
r J
11-
I r J
dr2 - r2 (d92 + sin2 0 d 2)
32. Restrict the Schwarzschild metric g to the two-dimensional submanifold
defined by 0 = x/2 and t = const. Prove that this yields the metric of
a paraboloid of revolution (a "lying" parabola!) with the equation z2 =
8M(r - 2M) (see the picture).
33. Light moves along geodesic lines whose tangent vectors have length zero.
Making use of the fact that the equations defining a geodesic are precisely
the Euler-Lagrange equations of the length function G = ta gjjx'ia, prove
the following assertions:
a) A particle moving at t = 0 in the equatorial plane 0 = x/2 stays there
forever.
b) The quantities L := r2cp and E := t(1-2M/r) are first integrals. More-
over, the second Kepler law holds: The orbit ray covers equal areas in
equal times.
c) Set r = r(W) and derive the equation describing the motion of light rays.
Result: It is reasonable to set u := 1/r: u" + u = 3Mu2.
d) Solve this differential equation approximately up to second order in V.
Result:
1
uo = sin V + [1 + cos 2W .
Interpret the constant of integration

2 0 ro as 3the scattering length. Which
asymptotic value arises for cp if r tends to oo and sp is supposed to stay
small? Twice this value, denoted by 5 in the picture, is the relativistic
deviation of light in the gravitational field of a very large miss, which
can be described by the Schwarzschild metric.
Chapter 6
Lie Groups and

Homogeneous Spaces
6.1. Lie Groups and Lie Algebras

In the preceding chapter Noether's theorem showed that symmetry consider-
ations simplify the study of geometric problems, and sometimes it is only by
symmetry considerations that a solution is possible at all. In fact, beginning
in the 1870s, the conviction grew that the basic principle organizing geom-
etry ought to be the study of its symmetry groups. In his inaugural lecture
at the University of Erlangen, which later became known as the "Erlanger
Programm", Felix Klein said, in 1872,
"Es ist eine Mannigfaltigkeit and in derselben eine Trans-
formationsgruppe gegeben; man soil die der Mannigfaltigkeit
angehorigen Gebilde hinsichtlich solcher Eigenschaften un-
tersuchen, die durch die Transformationen der Gruppe
nicht geandert werdenl.
One has to distinguish whether the groups under consideration are dis-
crete (for example permutation groups) or continuous (for example, one-
parameter groups of isometries). The latter were systematically introduced
by the Norwegian mathematician Sophus Lie (1842-1899). which is why they
bear his name today.
1 "Let a manifold and a transformation group in it be given; the objects belonging to the
manifold ought to be studied with respect to those properties which are not changed by the trans-
formations of the group."-quoted from F. Klein, Des Erlanger Programm, Ostwalds Klassiker
der exakten Wissenschaften, Band 253. Verlag H. Deutsch, Frankfurt a. M., 1995, p. 34.
207
208 6. Lie Groups and Homogeneous Spaces
The fundamental idea of a Lie group is a very simple one. It ought to be a

group which is at the same time a manifold, and hence allows a differential
calculus. Moreover, the manifold structure has to be compatible with the
group structure, i. e., the product is a differentiable map.
Definition 1. A Lie group is a group G which, at the same time, is a
differentiable manifold such that the map
is (infinitely often) differentiable.

Remark. Obviously, the last condition is equivalent to requiring that the
product map and the inversion
(9, h) -. g . h, g'-' g-1 ,
be differentiable maps.
Example 1. Every finite-dimensional vector space V is an abelian Lie group
whose group product is exactly the addition of vectors (v, w) H v + w.
Example 2. The unit circle S' = {z E C : IzI = 1} is an abelian Lie group
with the usual multiplication of complex numbers as product.
Example 3. Most Lie groups can be realized as matrix groups. The set of
all invertible matrices with entries in K = R or C is an open subset of IK"z
and hence a manifold. Endowed with matrix multiplication as product, it
forms a Lie group, the general linear group
GL(n,K) := {A E .M"(K) : det A 540} .
More generally, GL(V) is meant to denote the group of invertible endomor-
phisms of the vector space V.
Example 4. The vector space K" and the general linear group GL(n, K)
combine to form a new Lie group, the affine group Aff (K") := GL(n, K) x IIS"
with multiplication
(A, v) (B, w) := (AB, Aw + v).
This product rule arises in a natural way by defining an action of the affine
group on the vector space K" through
Aff(K") x K' V. (A, v)x := Ax + v.
and then applying the transformations determined by (B, w) and (A, v) on
a vector x one after the other.
Each Lie group G acts on itself by means of the left and the right translation
with a fixed element g E G,
L9, R9 : G -. G, L9(h) = g h and R9(h) = h g.
6.1. Lie Groups and Lie Algebras 209
The corresponding differentials are the following maps in the tangent bundle
of G:
(dLg)h: ThG - TghG, (dR9)h : ThG Th9G.
To avoid double indices, in this chapter we will use the notation (df)h instead
of f.,h for the differential of a map f.
Definition 2. A vector field X on G is called left-invariant (or right-
invariant, respectively) if it is transformed into itself by dL9 (or dR9, i.e.,
dL9X = X. At a point h E G, this means
(dLg)hX(h) = X(g - h).
Since left translation is obviously a diffeomorphism of G, Theorem 35 in
§3.9 can be applied to yield, for the commutator of two left-invariant vector
fields, the formula
dLg[X,Y] = [dL9X,dLgy] = [X,Y1.
This property, together with the fact that vector fields satisfy the Jacobi
identity (Theorem 34, §3.9), endows the vector space of left-invariant vector
fields with the structure of a Lie algebra, the Lie algebra g of the Lie group
G.
Theorem 1. The vector space of left-invariant vector fields on a Lie group

G is canonically isomorphic to the tangent space at the neutral element,
g='TeG.
Proof. With each left-invariant vector field X, we associate its evaluation
at the neutral element e, X - X(e) =: X E TOG. Conversely, every element
X E TeG determines a vector field Xx on G by setting
Mx(g) := (dL9)e(X)
This satisfies the relation
Xx(gh) = (dLgh)e(X) = (dLg)h(dLh)e(X) = (dLg)hXX(h)
hence Xx is left-invariant. 0
Remark. Because of this fact, we will no longer distinguish between the
Lie algebra of left-invariant vector fields and the tangent space to the group
at the neutral element. Its elements will be denoted by upper case Latin
letters X, Y, ... E 9.
Choosing a basis X1, ..., Xr of 9, we can again write their commutators as
linear combinations of the basis elements,
r
[Xi,Xj] = &Xk-
k=1
The antisymmetry of the commutator and the Jacobi identity imply that
the constants C have to satisfy the relations
Cij - - Cii ' CijCk,n + Cj' "ki + CrniCkj = 0.
Following E. Cartan, the numbers C are called the structure constants of
the Lie group G, since Cartan's structural equations are simple to formulate
in their terms. To see this, we agree to call a differential form w on G
left-invariant if it satisfies the condition
L*9w = w.
Following the argument in the case of vector fields, it is easy to see that
the r-dimensional vector space g* of left-invariant 1-forms is canonically
isomorphic to Te G. Now let al, ...,o-, be the basis of g* dual to X1. ..., Xr.
Theorem 2 (Maurer-Cartan Equations). Let C be the structure constants
of a Lie group G with respect to the basis X1, ... , Xr of its Lie algebra g.
Then the exterior derivatives of the forms in the dual basis al, ... , or of g*
are given by
do, _ -r k of Aak.
j<k
Proof. From Theorem 33, §3.9, we know the general formula for the differ-
ential of a 1-form oi:
do (Xj, Xk) = Xj (0t(Xk)) - Xk(ai(Xj)) - ai([Xj, Xk))
Since the forms ai are dual to the basis vectors Xj, the first two terms on
the right-hand side vanish. Now
dn'i(Xj,Xk) = -ai([Xj,Xk]) = -ai(E,nC7kXm) = -Cjik,
which is precisely the assertion.
We introduce a 1-form e : T(G) -+ g defined by the formula

e(t) := dL9-1(t)
for every tangent vector t E T9(G) at the element g E G. The form e
shifts the tangent vectors of the Lie group via left translations to the space
Te(G) = g. It is the so-called Maurer-Cartan form of the Lie group. In
general, if el, e2 are two 1-forms with values in the Lie algebra g, we define
its "exterior" product as a 2-form with values in the Lie algebra by the
formula
[ei,e21(tl,t2) [el(tl),e2(t2)] - [el(t2),e2(tl)l
In particular,
[e, el(tl, t2) = 2 - [9(tl ), e(t2)],
and the Maurer-Cartan equations can be formulated equivalently as

0.
Example 5. The special linear group in dimension two,
SL(2, R) = {A E M2 (R) : det A = 1 } ,
is a 3-dimensional Lie group with Lie algebra (see Exercise 5, Chapter 3)
s1(2, R) = {X E M2(R) : tr A = 0 } .
The matrices
H =
10
011' E [0 01' F [1 0,
form a basis of sl(2, lR) and satisfy the comniutator relations
[H, E] = 2E. [H, F] = -2F, [E, F] = H.
If we denote the forms of the dual basis of sl(2, I8)` by ax, 0E, and aF, the
Maurer-Cartan structural equations read as
dvy = - CE A QF, dOE = -2 aH A O'E, doF = 2 otl A aF .
Example 6. The Lie group GL(n, K) is an open subset of M. (K). Its Lie
algebra is therefore
gl(n,K) '=' Te GL(n, K) = M,. (K) .
Theorem 3. Let X be a left-invariant vector field on G and denote by 4x (t)
the maximal integral curve of X such that 4x (0) = e. Then
(1) 4x(t) is a group homomorphism. 4x(t1+t2) = 4'x(tt) ° 4>x(t2)
(2) the vector field X is complete, i. e.. 4b,y(1) is defined on all of R.
Proof. Let 4bx(t) be defined on the interval to < t < tb. We start by
showing the first assertion for two values tl, t2 such that they together with
their sum t t + t2 belong to the domain of definition of 4x (t). To do so, fix
t 1, set. g 4,x (t 1), and consider the curve
z1(s) = g 4x(s) on the interval (ta, tb)
as well as the integral curve obtained by shifting the argument by tt,
zit (s) = Ix(tt + s) on the interval (t,i - tt, tb - tt) .
From their derivatives,
dzi ,r fd4x 1 r /-'- , \ v/ . , ". v/ , %%
d
and
dill __ d4bx(ti +s)
ds ds
= X (tx(tt + s)) = X (zlt (s)),
we conclude that q and ill both are integral curves of X with equal initial
conditions.
fi(0) = 'x(ti) = g = 9(0)
Hence'i and r11 have to coincide on (ta, tb) fl (ta - t1, tb - t1). One easily
checks that t2 has to belong to this intersection, i. e..
' x(tl+t2) = 0 o4x(t2)
Now we show that tb = 00; to prove ta = -oo one proceeds analogously.
Suppose that tb is finite. Then,
r)(s) _ (px (2) x (s - )
2
is a curve defined on [ta + tb/2, 3 tb/2)) with tangent vector
ds = 2) = X (sx() 4x(s - 2 = X(gl(s)).

Hence ()(.s) is an integral curve of X defined on [ta + tb/2, 3 tb/2) such that
q(s) = 4 x (s) for s E (ta, tb) fl [ta + tb/2, 3 tb/2). Thus the maximal integral
curve has to be defined at least on 3 tb/2, a contradiction to the maximality
of tb.
This theorem enables us to define the exponential map for Lie groups.
Definition 3. The exponential map exp : g -. G is the evaluation at t = 1
of the integral curve 4 x of X E g satisfying 4x (0) = e,
exp(X) 4x(l)
Theorem 4. The exponential exp : g -y G is a smooth map with the fol-
lowing properties:
(1) exp(O) = e;
(2) exp(-X) = [exp(X)]-1;

(3) exp ((s + t)X) = exp(sX) exp(tX).
In addition, it is a diffeomorphism from an open neighborhood V of zero in
g onto an open neighborhood of the neutral element in G.
Proof. For the smoothness claim, we have to refer to the fact that the
solution of a system of differential equations depends smoothly on the initial
conditions. Under scalar multiplication, the flow transforms as
Op (t) = 4Dx(A-t),
since
d4x(At)
dt
= a X('Px(,\t))
together with'x(A 0) = e implies that cx(A t) is an integral curve of

A X. Restricting the exponential map to a straight line through the origin
yields a one-parameter subgroup of G. This proves property (3):
exp ((s+t)X) = 1D(s+t)x(1) = 4x(s+t) =
4tx(1) = exp(tX) exp(tX).
Property (2) is an immediate consequence of (3). and property (1) is trivial.
The final assertion follows from the inverse function theorem if we can show
that the differential of the exponential map at the origin,
d(exp)o: Tog=9- TG=9.
is invertible. But this is just the identity map: Consider the curve -Y(s) _
s X in g with tangent vector 7(0) = X. Then
dexp(sX) = d48x(1) _ dqbx(s)
d(exp)o(s.X) =
ds 18=o ds I3=0 ds 8=0 = X(e).
I
Since we already know that the evaluation of a left-invariant vector field at

the neutral element e is an isomorphism, the proof is completed.
It turns out that all one-parameter subgroups of a Lie group G have the
form exp(tX) for some X E 9.
Theorem 5. Every continuous group homomorphism ' : R G has the
form
11(t) = exp(tX)
for a certain X E 9. In particular, every continuous group homomorphism
1R G is smooth.
Proof. Let 41 : 1R G be a continuous group homomorphism, and let

V C g be a neighborhood of the origin so small that exp : g - G is a
diffeomorphism on 2V. Let 11' = exp(V) be the image of V in G. Since %F
is continuous, there exists an e > 0 such that 'P([-e, e]) C W. In particular,
fle) has to have a pre-image X E V under the exponential map, %P(e) _
exp(X). The smooth group homomorphism
f: G, f(t) = exp (ix).

coincides with 'F(e) at the point . Consider the set
K = {tEIR: 'F(t)= f(t)}.
This is a closed subgroup of IR, since ' and f are continuous group homo-
morphisms. Therefore, K is equal either to R or to a discrete subgroup of it.
The latter case will be excluded, implying the equality of f and i. Assume
for a contradiction that K is generated by a number a > 0,
K = In-a: nEZ}.
As c already belongs to K, a has to be less than e, and '+(a) belongs to W.
«'e show a/2 E K to obtain the desired contradiction. To do so. start from
the Ansatz
f (a/2)2 = f(a) = 'I'(a) = 'I' (a/2)2 .
Then, applying the inverse of the exponential map,
2exp-' (f (a/2)) = 2exp 1('' (a/2))
and noting that'P(a/2) belongs to IV, we conclude that 2exp-' (1P (a/2)) E
2V. On 2V. the exponential map is a diffeomorphism, and hence
f (a/2) = IP (a/2) .
This means that a/2 E K. 0
Definition 4. Let G1, C2 be two Lie groups, and let 0 : G1 -. C2 be a
group homomorphism. We define the differential of io as a map between the
Lie algebras of left-invariant vector fields on GI and G2 as follows: Consider
for X E g i the one-parameter subgroup exp(tX) E G1. By assumption,
its image curve '(exp(tX)) is a one-parameter subgroup of G2, which by
Theorem 5 has to have the form exp(tY) for some Y E 92. Now we set
u' : 91 - 92, y.(X) := Y.
This definition immediately implies that the following diagram commutes:
92
jex
C2
Identifying 91 - TeG1 as well as 92 ^_' TeG2, we see that this definition
is compatible with the usual one of the differential at the neutral element,
di'e : TeG1 -+ TE.G2. More precisely, the diagram extended by d,', remains
commutative:
v.
91
exp
G1
6.2. Closed Subgroups and Homogeneous Spaces 215
To see this, take any left-invariant vector field X E gl and compute

_ dexp(tO.(X)) _ dV,(exp(tX)) _
.(X)(e) dt I _ dt dVe(X (e))
=o t=o
A further important property of the differential V. is its compatibility with
the commutator.
Theorem 6. The differential of a homomorphism between Lie groups, :
GI -+ G2. is a homomorphism of Lie algebras, i. e..

14'.[X.Y] = [?P.X,V'.YJ-
Proof. The commutativity of the diagram just discussed is equivalent to
the following identity, holding for every left-invariant vector field X E g:
iP.(X)(e) = dl/'e(X(e)).
From this, we deduce the analogous identity holding at each point g E GI,
d,;"9(X (g)) = dt''g(dLg)e(X (e)) = d(it' o L9)e(X(e)) = (dLg9))ed ke(X (e))
_ (dL, ls))e (X) (e) = w'.(X)(vI(g)).
This means that the vector fields X and ,b.X are ?P-related in the sense of
the remark following Theorem 35, §3.9. In particular, this remark implies
the assertion.
6.2. Closed Subgroups and Homogeneous Spaces

The exponential map of Lie groups does not satisfy the functional equation
familiar from calculus; it turns out that a second order correction has to be
made.
Lemma 1. Let G be a Lie group and let exp : g G be the exponential map.
Then, for two arbitrary left-invariant vector fields X, Y E g, the following
identity holds:
exp(tX) exp(tY) = exp (t(X + Y) + 0(t2))
Here 0(t2) indicates that 0(t2)/t2 stays bounded for t - 0.
Proof. Choose a neighborhood V C g, 0 E V. such that exp : V - W' C
G is a chart around the neutral element e E G, and also choose a basis
XI, ... , Xr of g. By means of
(XI , ... , xr) _ exp(x1Xi +... + xrXr) E G,
we introduce coordinates in W. Let U C W be so small that U U C W.
(By " " we denote the product in the group G.) On U x U C G x G. we
then have the coordinates
y' = x'opri, z' = xtopre
and, because U U C W, we can restrict the group multiplication to obtain

a mapp:UxU-+W. Let
i i 1 r 1 r
be the coordinate representation of p. Then

f`(0, ...,0,z1, ...,zr) = z', f'(yl, ...,yr,0, ...,0) = y',
and hence f' has the Taylor expansion
i i
ft(y..) = f`(0,0)+ k(0,0)yk+ ayl(0,0)zl+R,
where the remainder R consists of second order terms. Inserting tY =
y'Xi. tZ = z'Xi into the Taylor expansion, we now obtain
exp(tY + tZ + R) = exp(tY) exp(tZ),
and the term R tends to zero like O(t2). O
The following example shows that not every continuous subgroup of a Lie
group has to be a Lie group itself.
Example 7. Let T2 = Sl x S' be the two-dimensional torus, parametrized
by (t. s) E [0, 1] x [0, 1]. For every real number q, the formula yq : R -
T2, yq(t) = (t, q t) defines a curve in T2, which is isomorphic to the additive
group R of real numbers. We want to show that this curve is dense in T2 for
irrational q. and hence provides an example of a connected, non-closed one-
parameter subgroup of a Lie group. Let (to, so) E [0,1]2 be an arbitrarily
given point on T2, which we want to approximate by a sequence of points
in yq. For each integer n E Z, we have
,y(to+n) = (to+n,qto+qn),
and, as an angle coordinate in S1, to + n is equivalent to to. Similarly,
qto + qn + m is equivalent to qto+qn for every integer m; hence it corresponds
to the same point on yq and on T2. If we are able to show now that qZ + Z
is dense in R. then we could conclude the existence of a sequence of pairs
(n;, m,) E Z x Z for which
limgni+mi = so-qto.
i-00
But then q(to + ni) + mi converges to so for i - oc, and
lim ryq(to + nt) = 'Y(to, so),
as claimed. Thus it remains to prove the following lemma.
Lemma 2. For every irrational number q, the set Z + qZ is dense in R.
Proof. Like the set Z+qZ, its closure is a subgroup of R. The group Z+qZ
cannot be cyclic; if it were, it would be generated by an element a > 0, and
there would exist integers k and I such that
I= q = Va.
But this would imply q = Ilk E Q. Hence we arrive at a contradiction. As
the closed subgroups of IR are precisely the cyclic groups and R itself, only
this last possibility remains for Z + qZ.
In Example 4, §3.1. we encountered a parametrization of a torus of revolution
by means of S' x S'. The following pictures show the trace of a curve yy
in this parametrization for two close rational and irrational values of q,
respectively. In Example 8, §7.4, we show that the motion of a spherical
pendulum can be parametrized precisely by such a curve on a torus.
The main objective of this section is to prove that a closed subgroup H of a

Lie group G is a submanifold of it. Consequently, it is itself a Lie group, and
the quotient space C/H carries the structure of a manifold with a smooth
G-action. As a preparation, we need a few technical lemmas. Let II - II be
any norm on the vector space g.
Lemma 3. Let H be a closed subgroup of G. and let Xn # 0 form a sequence
converging to zero in 9 such that exp(X,,) belongs to H and Xn/IIXnII tends
to an element X E g. Then
exp(tX) E H for all t E R.
Proof. For a fixed t > 0 we define a sequence of natural numbers m by
mn := max{kEN:
Then the following estimate holds:
mnIIXnII < t < (mn + 1)IIXnII = mnIIXnII + IIXnII
But on the right-hand side, the sequence IIXnII tends to zero, and hence
lim mnIIXnII = t.
n-or,
This implies
limn rnXn = hn-oo m m HIX,,II
IlXnll = tX,
and, since the exponential map is continuous,
Iim
n-oo
exp(t X).
Each term in the sequence exp(m,,Xn) = [exp(X,,)]"'" is an element of H.
As H is closed by assumption, the limit of this sequence, exp(t X), has to
lie in H, too. The proof for the case t < 0 proceeds along the same lines.
Lemma 4. For every closed subgroup H C G, the set
b :_ {XEg: exp(tX)EHforalltElR}
is a linear subspace of g.
Proof. For any vector X E h, its scalar multiples a X also belong to h. It
thus suffices to show that h is closed under addition. To this end, let X, Y
be elements of h, and suppose that X + Y 36 0. In any case, the product
exp(tX) exp(tY) belongs to the subgroup H, and for sufficiently small t we
have, by Lemma 1,
exp(tX) exp(tY) = exp(t(X +Y) + 0(t2)).
Then Z(t) := O(t2)/t apparently converges to zero for t - 0, and we can
rewrite the preceding equation as
exp(tX) exp(tY) = exp (t(X +Y + Z(t))) E H.
Choose a sequence of positive numbers that converges to zero, t - 0, and
define Xn := tn(X + Y + Z(tn)). Each term exp(XX) lies in H, and
--:in = lim X+Y+Z(t,) X+Y
lim
11X-11 n-- IIX + Y + Z(t,a)[I = Fix -+Y11
Obviously, we have Xn # 0 and Xn -' 0. Thus, Lemma 3 applies, and we
can conclude that for every t E R
exp(t
IIX+YII
is an element of H. Hence X + Y belongs to Fj.
Suppose that H is, in addition, a closed subgroup of G, and let h be defined
as in the preceding lemma. We then choose any linear complement h' of h
in 9,
9 = h+h'.
Lemma 5. There exists a neighborhood V' C 4' of 0 such that, for every
X' 34 0 in V', the element exp(X') does not belong to H.
Proof. If the assertion were false, there would exist a sequence X;, E h'
converging to zero and satisfying exp(X,,) E H. Now consider the compact
set
K :_ {X' E h': 1 < JJX'II < 2}
and choose natural numbers p,, such that p ,,X,, E K. Since K is compact,
we may assume that the sequence p,X;, converges to some 0 9& X' E K.
Again, [exp(X;,)]P is an element of H and
pnX' _ X'
lim
11p-Xnii 11X'6
Then Lemma 3 implies that X'/IIX'II E h, contradicting 0 0 X' E '. 0
Now we can turn to the main theorem of this section.
Theorem 7. Let G be a Lie group, and let H be a closed subgroup. Then:
(1) H is a submanifold of G and thus itself a Lie group.
(2) There exists precisely one differential structure on G/H such that
(a) the projection 7r : G - G/H is smooth,
(b) for every p E G/H there exist a neighborhood Wp C G/H of p
and a smooth map w : Wp G such that 7r o p = Idw,,,
(c) the action of G on G/H defined by (g. kH) gkH is smooth.
Proof. It obviously suffices to show that there exists a neighborhood W C G

of e for which H n W is a submanifold (left translation is a diffeomorphism
of G). As before, ddecompose the Lie algebra into g = 1) + h' and consider
the map corresponding to this decomposition,
4D : g = h + 4' ---i G, 4(X + X') = exp(X) exp(X').
In h' we choose a neighborhood V' C h' as in Lemma 5. as well as a subset
V C h so small that the exponential map still is a diffeomorphism on V +V'.
The image W of V + V' under -ID is an open neighborhood of e E G, and
HnW=
by the definition of b and Lemma 5. The set H fl W is thus parametrized
by the chart (V, -D Iv+(o}), and hence a submanifold of G.
Now we turn to the proof of the second assertion. Let 7r : G - G/H denote
the projection. We define a topology on G/H by the condition
A C G/H is open :a 7r-' (A) C G is open.
It is called the quotient topology on G/H, and it is designed to render the
map 7r continuous. Endowed with this topology, G/H is a Hausdorff space
(see Duistermaat/Koik, Lemma 1.11.3). To verify the properties a manifold
has to satisfy, consider the distinguished point xo := e H E G/H together

with the sets V, V' introduced in the first part of the proof. The map
' : V' - G/H, X' a(exp X') ,
is continuous and maps V' onto an open neighborhood U of xo. Moreover, V,

is injective, since v(X') = O(Y') implies the existence of an element h E H
such that exp(X') = exp(Y') h. Hence
h = exp(X') exp(-Y') = b(0 + (X' - Y')) .
Thus h also belongs to the set W, which was defined as the image of V + V
under C Since we already proved H fl W = 4(V + {0}), this implies
X' = Y'. In summary, the map 1/i : V' U is continuous and bijective.
For an arbitrary point gH E G/H, consider the left translation by g E G on
G/H.
L9 : G/H G/H, kH gkH
and introduce a chart around gH E G/H by
U9H L9(U), 09H: V' - UgH, Z'gH := Lg o TV .
For two points gH and kH, the chart transition can be rewritten as follows,
kH o y9H = v-1 o Lk-1 o L9 0 V1 = exp-1 o(ir-I o 4-1 0 L9 0 a) o exp
= exp-1 oLk-1h oexp .
Therefore. as a superposition of smooth maps, the chart transition is also
smooth. Hence we have proved that G/H is a differentiable manifold, the
projection ;r : G - G/H is smooth, and C acts smoothly from the left on
G/H. It remains to show (b). For the distinguished coset p = .ro = e H,
define y; for each x in U =: Wp by
cp(x) = exp(Vi-1(x)) = 7r-1(x).
For an arbitrary point p = gH one again uses the left translation L9. 0
Definition 5. The action of a Lie group G on a manifold Al is called
transitive if, for two arbitrarily given points x and y in Al. the one can
always be written as the image of the other under the action of G. i. e.,
there exists a g E G such that y = g x. An equivalent formulation of this
requirement is to say that M consists of a single G-orbit, G x = Al. A
manifold together with a transitive group action is also called a homogeneous
space.
Obviously, the left translation on the quotient Al = G/H is a transitive

group action, and thus G/H is a homogeneous space. Theorem 7 can be
applied to show that some well-known matrix groups are Lie groups: The
following groups apparently are closed subgroups of GL(n, K).
6.3. The Adjoint Representation 221
Example 8. The subgroup of GL(n, K) consisting of all matrices with de-

terminant 1 is a Lie group, the special linear group,
SL(n,K) :_ {A E Mn(K) : det A = 1} .
Example 9. Let H I =: h u, v E c } be the vector space of

{[. V
Hamilton's quaternions with standard basis
111
110 1
E= 11' 1= oil' J
LO Lof OJ, K = Lo i 01
and norm N(h) := uu + vv. The group of all quaternions with norm 1 is
isomorphic to the Lie group
SU(2) := JA E GL(n, IC) : AAt = 12 and det(A) = 1).
Example 10. The preceding example can be generalized as follows. The
unitary group is embedded into the space of complex matrices as
U(n) :_ {A E AAt = 1n} .
The condition AAt = 1n immediately implies I det Al = 1, hence det A E Sl;

the special unitary group is defined as the group of all unitary matrices A
satisfying det A = 1:
SU(n) := {A E U(n) : det A = 1} .
Example 11. The orthogonal group O(n, K) consists of the matrices A E
M ,,(K) leaving the euclidean standard scalar product of Kn invariant,
(Ax, Ay) _ (x, y)
Realizing the scalar product as (x, y) = xty, we see that this condition is
equivalent to AAt = ln. Hence we obtain
O(n, K) = JA E Mn(K) : AA' = 1n} .
Obviously, an orthogonal matrix has determinant + 1 or -1. The subgroup
of all orthogonal matrices with determinant +1 is called the special orthog-
onal group SO(n, K),
SO(n,K) = {AEMn(K): AAt=lnand detA=1}.
6.3. The Adjoint Representation

Definition 6. Let G be a Lie group with Lie algebra g, and let V be a
finite-dimensional vector space.
(1) A representation of the Lie group C on V is a smooth group ho-

momorphism e : G GL(V), i.e., a smooth map compatible with
the group structure,
e(g h) = e(g) e(h)
(2) A representation of the Lie algebm g on V is a homomorphism of
Lie algebras, p : g - gl(V), i. e., a linear map compatible with the
commutator,
e([X, Yl) = [e(X ), e(Y)] = e(X) e(Y) - e(Y), e(X)
Sometimes, V is then also called a G-module or a g-module, respectively.
Example 12. The trivial representation of a Lie group G is the group
homomorphism that maps every element g E G to the neutral element in
GL(V): p(g) = 1V; the trivial representation of g associates the zero map
with every element X, o(X) = 0v.
Example 13. Matrix groups are defined by means of one of their represen-
tations, often called the defining representation. In fact, we introduced the
groups GL(n, R), SL(n, R) and SO(n, R) in a way endowing them naturally
with a representation on R". A simple example illustrates that these matrix
groups and their Lie algebras have many more representations. The Lie al-
gebra sl(2, R), for example, has representations in all dimensions: For every
natural number n, define e : s((2, R) - gl(n + 1, R) by
g(H) = diag(n, n - 2, ..., -(n - 2), -n),
0 1 0
n 0
0 2
e(E) e(F) = n-1 0
0 n-
L
1 0
These matrices satisfy the commutator relations of sl(2, R),
[e(H), e(E)1 = 2,o (E), [e (H), e(F)l = -2e(F), [e(E), e(F)] = e(H),
and hence form an (n + 1)-dimensional representation of sl(2, R). Properties
that cannot be expressed by the Lie bracket do not have to be preserved
under a representation: For example, we have E2 = 0, but e(E)2 0 0.
Nevertheless, the property that g(E) is a nilpotent matrix is preserved.
There is also a a representation of the Lie group SL(2, R) corresponding to
this representation of the Lie algebra; this will be the subject of Exercise 4.
Apart from left and right translation, there is a third remarkable action of
a given Lie group G on itself, the so-called conjugation action,
ag : G - G, a9(h) := ghg-' = L9R9-, h.
It is smooth and satisfies ag(e) = e, and in contrast to left and right trans-
lation, it is far from being transitive. In the case G = GL(V), it decomposes
the invertible matrices precisely into their similarity classes. In addition,
the relation ag(e) = e implies that its differential at e is a map from g to g,
d(ag)e: TG 5--- g ----+ TG 2--- 9,
which is obviously invertible, since d(Lg)e and d(Rg-i)e are invertible. We
define the adjoint representation of G on g by
Ad : G ---+ GL(g), Ad(g) = d(ag)e E GL(g).
Before we verify that this actually is a representation, recall the definition
of the center of a group. It consists of those elements which commute with
all the others:
ZG = {gEG: gh=hgdhEG}.
Theorem 8. The map Ad : G - GL(g) is a representation of G on the
vector space g. The center ZG of G is contained in its kernel,
ZG C ker Ad,
and equality holds if and only if G is connected.
Proof. First we check the homomorphism property:

Ad(gh) = d(LghRgh )e = d(LgLhRh-, Rg-i)e = d(agah)e
= d(ag)ed(ah)e = Ad(g)Ad(h).
Let z belong to the center Z. Then aZ = IdG, and hence Ad(z) = IdGL(9),
i.e., z is in the kernel of Ad. Now suppose that G is connected and that
Ad(g) = Ide. Since ag : G - G is a group homomorphism, the map
t ' -+ ag (exp tX) is a one-parameter subgroup for every X E g, and, by
Theorem 5, there exists an element Y E g such that
exp(tY) = ag(exptX).
Differentiating this equation with respect to t, we obtain
Y(e) = dt (ag(exp tX )) I e=o = Ad(g) (X (e)) = X (e) ,

which proves X = Y. For the one-parameter group defined above this means
that
exp(tX) = ag(exptX)
for all t E R and X E 9. The exponential map is a local diffeomorphism
g - G; hence ag = IdG on an open neighborhood W of e. For a connected
Lie group G this implies ay = IdG, since G can be represented as the union
of all powers of W (with respect to the group products),
00
G = UW'.
As a9 = Idc is equivalent to g E ZZ, everything is proved. 0
The differential of the adjoint representation of G (in the sense of Definition
4) is a representation of the Lie algebra g which can now be expressed by
the commutator.
Theorem 9. The differential ad := Ad. : g -. gl(g) of the adjoint
representation is a homomorphism of Lie algebras determined by the formula
ad(X)(Y) = [X, YJ.
Proof. By the definition of the differential, we have
Ad(exptX) = exp(tAd.(X)) = 1 + tAd.(X) + ... ;
hence
Xd
Ad(exptX)(Y) - Y
Ad.(X)(Y) =
li.o
The flow corresponding to the vector field -X is (bt = R P(_tX), since
d4ie(e) dexp(-tX) I
LO dt t=o
Applied to a left-invariant vector field Y, however, its differential coincides
with Ad(exptX)(Y),
Ad(exptX)(Y) = dL P(tx)dR P(-ex)(Y) = dR p(_tx)(Y) = d4it(Y).
Thus the original identity can be rewritten as
`ht(Y) - Y
Ad.(X)(Y) = li o
The right-hand side is precisely the definition of the commutator [Y, -X] _
[X, Y]. O
This representation will also be called the adjoint representation (this time
of the Lie algebra g). In case of doubt, the context has to decide whether
the representation of the Lie group or that of its Lie algebra is meant.
Remark. The definition of the differential immediately implies the identity
Ad(expX) = exp(adX).
This has to be understood as an identity of operators. Applied to an element

Y, it means
ad(X)3i3(Y)
exp(X) Y exp(-X) = 1 + ad(X)(Y) + + ...
= 1 + [X, Y1 + [X, [2I Y11 + [X, [X 3'XI Y111 + ... .
Example 14. Let g be the three-dimensional Lie algebra which is abstractly

defined by the following commutator relations for a basis el, e2, e3:
[el, e21 = e3, [e2, e31 = el, [e3, ei1 = e2.
The representing matrices of the adjoint representation with respect to this

basis can be computed from them. For the operator ad(ei ), we obtain
0 l i
ad(ei) e2 e3 = rO 0 1 e2 Ll e2
e3 -e2 -1 0 e3 e3
and similarly for the other two operators,

0-1 0 10
L2 := ad(e2) 0 0 , L3 := ad(e3) _ 1 0 0
rol 0 0 0 00
Let us look, on the other hand, more closely at the orthogonal group O(n, R).
It was defined as the set of matrices satisfying f (A) = AAt - 1, = 0. The
differential of this map at the point X is
df(A)x = AXt+XAt,
and hence, according to Theorem 5 in §3.2, the Lie algebra of O(n, R) is
o(n, R) = Te O(n, R) = {A E M,(R) : A + At = 01.
The Lie algebra of the orthogonal group consists precisely of the skew-
symmetric matrices, which for n = 3 is apparently spanned by Li, L2 and L3.
This proves that the three-dimensional defining representation of o(3, R) is
isomorphic to the adjoint representation. In higher dimensions, this fact no
longer holds as a simple dimensional consideration shows: A skew-symmetric
matrix has exactly as many degrees of freedom as entries above the diagonal.
Therefore,
- n = n(n2 1)
22
dim o(n, R) =
and this is equal to n only for n = 3.
Exercises
1. The exponential map exp : si(2, R) SL(2, R) is not surjective. Hint:

What are the values that tr exp(A) can attain for A E al(2, R)?
2. Hamilton's quaternions H (Example 9) form not only a vector space,

but also a (non-abelian) division algebra, i. e., an associative algebra in
which each non-trivial element is invertible. Prove that the standard basis
E, I, J, K of Hamilton's quaternions obeys the following algebra relations:
I.K=-J,
and compute the inverse of the quaternion h = 0.
3. Identify the quaternions of trace 0,

Ho = {xi . I + x2 J + x3 K j xi, x2, x3 E R},
with the 3-dimensional euclidean space R3. Prove that, for every U E SU(2),
the map
Ho - Ho, x UxU-1,
defines a special orthogonal transformation of R3. The resulting map e
SU(2) - SO(3,R) is a representation which is not injective ("faithful").
4. The defining representation of SL(2, R) on R2 is the usual matrix action

on vectors,
gV
Ic d] [y] [cx + dy]
Let V,, be the (n + 1)-dimensional vector space of homogeneous polynomials
of degree n in the variables x and y. Define an action of g E SL(2, R) on the
polynomial p E V,, by
P(g) ' P ([X])

= P\g_1 [;]).
Prove that B is a representation of SL(2, R).
5. Let G be a Lie group, and let H be a discrete, normal subgroup of G.

Prove that H is necessarily contained in the center of G.
Exercises 227
6. Let (p, V) be a representation of the group G. A subspace W of the

representation space V is called invariant if. for every g E G. the relation
p(g)W C W holds. For trivial reasons, the subspaces W = V and W = {0}
are invariant: if the representation has no further invariant subspaces, it is
called irreducible. Consider the following two-dimensional representation of
the additive group R:
rrI
LOW = L0
tl 1
11
Prove that this representation is not irreducible. Does the invariant subspace
have an invariant complement?
7. By Theorem 6. the differential of a representation (p. V) of the Lie group

G is a representation (Lo., V) of its Lie algebra g. The tensor product of two
representations (p, V) and (µ, W) of G is defined by
(p ®Fr)(9)(v $ w) := p(9)v (9 p(g)w, g E G. V E V. W E W .
Prove that this determines a representation of G on V O W with differential
(e® i),(X)(v ®w) := p,(X )v ®w + v ®p.(X)w, X E g.
8. In order to describe the hyperbolic plane as a homogeneous space. it is

useful to introduce a new model for it, the open unit disc.
a) Let D = {z E C I lzj < 1} be the open unit disc with metric
9 = (1_1212)2 I0 i
I.
Show that the Cayley transform. `
-i`+i=:x+iy,
c: c(z) =
z-i
is an isometry between D with the metric above and the upper half-plane
?{2 with the fourfold of the hyperbolic metric.
b) Let the Lie group
SU(1.1) :_ {
lb b : 1a12 - 1b12 = 1}
act on D via the formula
a b z _ az+b
b a az+b
Prove that this action is transitive, and that the isotropy group of zero,
Go := {g E SU(1,1) : g 0 = 0},
is isomorphic to SO(2,IR). Hence D 5 SU(1,1)/SO(2,1R).
Chapter 7
Symplectic Geometry
and Mechanics
7.1. Symplectic Manifolds

Riemannian geometry is the geometry of a symmetric, bilinear form depend-
ing on the point of a manifold. The curvature is a measure of how far two
symmetric bilinear forms differ locally. Contrary to this, symplectic geom-
etry is that of an antisymmetric bilinear form depending on the point of
a manifold-hence the geometry of a 2-form w. It turns out that all sym-
plectic manifolds are locally equivalent: there cannot be any concept similar
to curvature in the sense of Riemannian geometry. Symplectic structures
differ, if at all, only globally. Historically, the formulation of mechanics in
the sense of Hamilton led to symplectic geometry, hence its essential role in
modern mathematical physics.
Definition 1. A symplectic manifold is a pair (1112., w) consisting of a

manifold 1112ni of even dimension together with a closed non-degenerate 2-
form w,
d w = 0 and A
called the symplectic form or symplectic structure.
By Theorem 16 in §3.4, every symplectic manifold is orientable. The volume

form is understood to be the 2m-form
-(--1)/2
dM2m = (-1) ' win .
m!
229
230 7. Symplectic Geometry and Mechanics
Example 1. In 1R2"' with coordinates {q1, ..., q,", pi, ..., p,,,), the formula
m
E dpi A dqi
i=1
defines a symplectic form with highest power
w"' = m! - (-1)-(--1)/2
- dpl A ... A dpm A dq1 A ... A dq,,, .
The volume form in the sense of symplectic geometry is the ordinary volume
form of ]R2i'. This symplectic structure is called the canonical symplectic
structure.
Example 2. Define a 1-form 0-the so-called Liouville form--in the cotan-
gent bundle T'X"' of an arbitrary m-dimensional manifold as follows: Let
V E T,, (T' X ') be a tangent vector at q E T' X' and represent it by a curve
V : (-e, e) - T* X' such that
V(0) = q, V(0) = V.
Project this curve first by means of the projection 1f : T'X"' - X"' to the
manifold, and, after that, apply the 1-form q to the tangent vector of the
projected curve:
d
9(V) := n
d (T ° V (t)) I=o
The 2-form w := dO is a symplectic structure on T'X'. Any system,
{q1, ... , qm }, of coordinates in X' determines-representing a 1-form q
as q = >2 pi dqi-coordinates {qt, ... , 9m, pi, ... , p,n } in TX". By the
definition of the Liouville form 0, we have
m
0= pi dqi
and the 2-form

m
w=d9=EdpiAdgi
i=1
is non-degenerate. In particular, the (co-)tangent bundle of every manifold
is an orientable manifold (see Exercise 9 in Chapter 3).
Further examples of symplectic manifolds arise as the orbits of the coadjoint

representation of a Lie group G. Starting from the adjoint representation,
Ad : G GL(g), of the group G and passing to the dual of the linear
operator, Ad'(g) := (Ad(g-1))* : g' - g', we obtain a representation
Ad` : G -b GL(g')
7.1. Symplectic Manifolds 231
of the group G in the dual space g' of the vector space g. Through each
functional F E g' passes an orbit
01(F) := {Ad*(g)F : g E G},
on which the group G acts transitively. The isotropy group
GF := {g E G : Ad'(g)F = F)
is a closed subgroup of G, and 0 *(F) is diffeomorphic to the homogeneous
space G/GF. Its Lie algebra can be characterized by a similar condition:
Theorem 1. The Lie algebra OF C g of the isotropy group GF C G is equal
to
OF = {XEg :F([X,Y])=0 for all YEg}.
Proof. Suppose that F([X,YJ) = 0 holds for all elements Y E g. Then
from
(Ad*(exp(t X))F)(Y) = F(Ad(exp(- t X))Y) = F(exp(- t ad(X))Y)
z
= F(Y - t
we immediately obtain Ad*(exp(t X))F = F. The one-parameter group
exp(t X) is a subgroup of GF, and hence its tangent X belongs to the Lie
algebra OF. This proves one inclusion, the converse is proved analogously.
0
If a Lie group G acts smoothly on a manifold Mm, we can associate with
each element X E g of the Lie algebra the unique vector field k on Mm
whose integral curves coincide with the trajectories of the one-parameter
transformation group exp(t X):
X(x) _ d
The vector field X is called the fundamental vector field corresponding to

the element X E g of the Lie algebra. If G acts transitively on Mm, every
tangent vector V E TXM'" at a fixed point x E Mm can be realized by a
fundamental vector field.
This general construction will now be applied to an orbit O' of the coadjoint
representation. First, realize a given vector V E TFG' as the value of a
fundamental vector field, f(F) = V. For a further element Y E g of the
Lie algebra such that k(F) = V, the equality (X-- Y)(F) = 0 immediately
implies
0.
e=0
Theorem 1 implies X - Y E OF, and the resulting map is injective,

9/9F3X- X(F)ETFO'.
For dimensional reasons, it is bijective: the tangent space TFO* to an orbit
0* at F E O' can be identified with the vector space 9/9F. We now define
a symplectic structure wo on each orbit O' C g'.
Definition 2. Let V, W E TFO' be two tangent vectors to the orbit at
F E O', and choose elements X, Y E g with f ((F) = V, Y(F) = W. The
value of the Kirillov form wo on the vectors V, W is determined by the
formula
wo (V, W) := F([X, Y])
Theorem 2. The pair (O', is a symplectic manifold, and the 2 form
wo. is G-invariant.
Proof. Note first that the 2-form wo is uniquely defined. If the elements
X, X1 E g realize the vector V at F, then the difference X - X1 belongs to
the Lie algebra OF, and Theorem 1 implies
F([X,Y1) = F([X -X1,Y])+F([X1,Y]) = F([X1,Y]).
Moreover, wo is a non-degenerate 2-form. If, in fact, wo (V, W) = 0 for
every tangent vector W E TFO', then we obtain F([X,Y]) = 0 for all
elements Y E g. By Theorem 1, X lies in the Lie algebra OF, and hence
V = f(F) = 0. It remains to show that wo is a closed form. For two
elements X, Y E 9, the function wo (X, k) : O' R is determined by the
formula
F([X,YI)
Differentiate this relation in the direction of a third fundamental vector field:
2(wo.(X,Y))(F) = d [Ad'(exp(-t.Z)F)[X,Y]]It_o
= dtF(Ad(exp(tZ))([X,Y]))It=o = F([Z, [X,YII)
Then the expression for the exterior derivative dwo of the 2-form vanishes
identically:
dwo. (X, Y, Z) = X (wo (Y, Z)) - Y(wo (X, Z)) + Z(wo (X, f))
-wo.([Y,Z1,X),
since it reduces to the Jacobi identity of the Lie algebra g. 0
Corollary 1. Each orbit O' C g' of the coadjoint representation of a Lie
group is a manifold of even dimension.
7.1. Sylnplectic Manifolds 233
Example 3. The affine group of R has the matrix representation

G= { [0 1 1
a>0,bER}
with Lie algebra
9 = 0
1[Y l
0
:x ,y ER} .
J11
The computation
r[0 0]'[100a -b/a 1 _ 10 ay0bx1
Ad
1] [0 0] [0 1] [0
implies that g has one-dimensional orbits. To determine the orbits of the
coadjoint representation, write any element of g' as a pair (a, 0) of real
numbers, whose evaluation on the element (x, y) E g is
((a, /3), (x, y)) = ax + /3y .
By definition, the group element g = 10 1] acts as follows:
(Ad* (g-1)(a, 3), (x, y)) = ((a, R), Ad(g)(x, y)) = ((a, Q), (x, ay - bx))
= ax + /3(ay - bx) = ((a - fib, /3a), (x, y)) .
Summarizing, we have Ad*(g-1)(a,0) = (a - f3b,,Oa), and hence for 3 96 0
the coadjoint orbit through (a,,3) is two-dimensional. This example shows
that the adjoint and the coadjoint representation of a group G are, in general,
not equivalent.
After having discussed examples of symplectic manifolds, we now want to
introduce the symplectic gradient, which is the analogue of the gradient of
a function on a Riemannian manifold. In this situation, we will make use
of the fact that the non-degenerate 2-form w also provides a linear bijection
between the tangent bundle TA12m and the cotangent bundle T* M2",.
Definition 3. Let H : M2, IR be a smooth function on a symplectic
manifold. The symplectic gradient s-grad(H) is the vector field on M2in
defined by
w(V,s-grad(H)) := dH(V).
Example 4. Let {q1, ..., q,,,, p1i ... , pm } be coordinates on M2'" such that
the symplectic form w can be written as w = E dpi A dq;. Then
aH a
s-grad(H) = (aH 0
This formula immediately follows from the equation defining the symplectic
gradient. A curve -y(t) in the symplectic manifold M2rn, represented in the
fixed coordinates y(t) = {q1(t), .... q,n(t), pl (t), ... ,p,n(t)}, is thus an inte-
gral curve of the vector field s-grad(H) if and only if the so-called Hamilton
equations hold:
aH aH
and Pi=--
9i = api
aqi
Theorem 3 (Liouville's Theorem). Let H be a function on a symplectic
manifold (M2m, w), and suppose that s-grad(H) is a complete vector field
with flow 4Dt : M2m , M2m. Then:
(1) The Lie derivative of w vanishes,
Gs-g,ad(H)(w) = 0.
(2) The flow 't preserves the symplectic volume,
dM2m = dm2m.
J J
Proof. From dw = 0 and Theorem 32 in §3.9, we conclude that
Gggrad(H)(w) = d(s-grad(H) J w) = - dd H = 0.
,11m
Hence the diffeomorphisms 4)t : A f2m do not alter the symplectic
structure, i.e., 4 (w) = w, and so both assertions are proved. 0
The existence of this invariant measure has consequences for the dynamics
of symplectic gradient fields.
Theorem 4 (Poincare's Return Theorem). Let (M2,, w) be a symplectic
manifold of finite volume, and let 4bt : M2m _ hf2m be the flow of the
symplectic gradient of a smooth function H. For any set A C M2m of
positive measure, the set
B = {x E A : tn(x) 0 A for all n = 1, 2, ...}
has measure zero.
Proof. Note first that the intersections 4>_n(B) fl B are empty. Any point
xE fl B would be a point x E B such that 4n(x) E B C A,
contradicting the definition of the set B. This immediately also implies that
the intersections ' _n(B) fl 4'_m(B) are empty for n m, and, from the
invariance of the measure, we obtain
x oc
Evol(B) = Evol((Dn(B)) < Vol(M2m) < x,

n=1 n=1
i. e., the measure of the set B has to vanish. 0
This result has several famous generalizations, as the only example of which
we quote Birkhoff's ergodicity theorem (without proof).
?.1. Symplectic Manifolds 235
Theorem 5 (Birkhoff's Ergodicity Theorem). Let f be an integrable func-

tion on the symplectic manifold (M2,, w), and let Ot be the flow of a sym-
plectic gradient. Then the following limit exists almost everywhere:
lim 1 t f o .0,(x) =: f` (x) .

t Jo
Furthermore, the function f' is also integrable, and its integral coincides
with that of f. Lastly, f* is invariant under the flow fit.
Definition 4. The Poisson bracket of two functions f and g on a symplectic
manifold is the function
{f,9} = w(s-grad(f),s-grad(9)) = dg(s-grad(f)) = -df(s-grad(g)).
Example 5. In the {q, p}-coordinates, we have
{f, 9} _ m Of a9 _ Of a9
i=1 \ apt 5q1 aq1 Opi
In the next theorem, we summarize the properties of the Poisson bracket.

Theorem 6. The ring C'°(M2,) endowed with the Poisson bracket is a
Lie-Poisson algebra:
(1) for constants cl, c2 E R;
(2) { f, g} = -{g, f };
(3) { f, {g, h}} + {g, {h, f }} + {h, { f,g}} = 0 (Jacobi identity);
(4) {f,g-h} =g {f,h}+h- {f - g};
(5) s-grad({f, 9}) = [s-grad(f ), s-grrd(9)J.
Proof. The two first identities result immediately from the definition of
the Poisson bracket, and the fourth follows from d(g - h) = g - dh + h dg.
We prove (5). We insert the vector fields V := s-grad(f ), W := s-grad(g)
together with an additional vector field Y into the equation defining the
3-form dw,
0 = dw(V,W,Y) = V(w(W, Y)) - W(w(V, Y)) + Y(w(V, W))
- w([V, W], Y) + w([V, YJ, W) - w([W, Y), V) .
Applying the definition of the symplectic gradient as well as that of the
Poisson bracket, we can rewrite this equation as
0 = -V (Y(9)) + W (Y (f )) + Y({ f, 9}) - w([V, W], Y)
+ [V,Y)(9)-[W,Yj(f) = -Y({f,9})+w(Y,[V,WI)
Hence s-grad({ f, g}) = [V, W] = [s-grad(f ), s-grad(g)J. The Jacobi identity

is a consequence of formula (5). In fact, we obtain
{f. {g,h}}+{g.{h, f}}+ {h.{f,g}}
= s-grad(f)(s-grad(g)(h)) - s-grad(g)(s-grad(f)(h)) -s-grad({f,g})(h)
= [s-grad(f ), s-grad(g)] (h) - [s-grad(f ), s-grad(g)] (h) = 0.
A Hamiltonian system consists of a symplectic manifold (111211, W, H) to-

gether with a function H. The integration of the corresponding Hamil-
ton equation relies on determining the integral curves of the vector field
s-grad(H). For this, there exists an analogous notion of first integrals as in
the Riemannian case.
Definition 5. A function f : 1112" -. R is called a first integral of the
Hamilton function if it is constant on each integral curve of the vector field
s-grad(H).
Theorem 7.
(1) A function f is a first integral of the Hamilton function H if and
only if its Poisson bracket with H vanishes,
{ f, H} = 0.
(2) The set of all first integrals of a Hamilton function is a Lie-Poisson
algebra.
Proof. We compute the derivative of a function f along an integral curve

y(t) of s-grad(H):
dt f o -y(t) = df (j(t)) = df (s-grad(H)) = {H, f } .

This implies the first assertion. The second follows from the Jacobi identity
for the Poisson bracket.
7.2. The Darboux Theorem

The Darboux theorem states that all symplectic manifolds are locally equiv-
alent.
Theorem 8 (Darboux Theorem). Near each point x E 1112n of a symplectic
manifold (M2i/.w), there exists a chart h : U C 1112m - R2m in which the
symplectic form w is the pullback of the usual symplectic form.
w = h* dpi n dqi)
e=1
7.2. The Darboux Theorem 237
Coordinates with these properties are called symplectic (canonical) coordi-

nates.
Proof. In the cotangent space TTM2," to the manifold at the point x E

M2",, we choose a basis al, ... , o , ILl...... m in which the symplectic
form at this point is represented in normal form,
Iii
w(x) = E ai A pi .
i=1
Consider, moreover, a chart (D : V - 1R 2»i around x such that
4t(x) = 0 and w(x) = 4D` (dPAdQi(0))
Denoting the corresponding symplectic form on V by
Wi := fi' I dpi A dqi I ,

i=1
we see that there exists a neighborhood U C V of x such that for all param-
eters t E [0,1] the form
wt :_ (1 - t)w + twl
is a symplectic structure on U. In fact, dwt = 0, and since wt(x) = w(x) for
all t, a compactness argument shows that all the forms wt (t E [0,11) do not
degenerate at the same time in a neighborhood of x. Since d(wl - wo) = 0,
Poincare's lemma shows the existence of a 1-forma such that the difference
w, -wo = da
is the exterior derivative of this 1-form. By subtracting locally, if necessary,
a 1-form with constant coefficients from a, we may assume that a vanishes
at the point x, a(x) = 0. Dualizing a by means of the symplectic forms wt,
we obtain a family Wt of vector fields on U parametrized by t,
wt W, Wt) = a(V) .
Let cp(y, t) E Mgr" be the solution of the (non-autonomous) differential
equation
Ve(t) = Wt('p(t)), V(0) = y
All the vector fields W1(x) 0 vanish at the point x, and the solution
corresponding to the initial condition x is constant, p(x, t) - x. Hence
there exists a neighborhood U1 C U of x such that, for every initial condition
y E U1, the corresponding solution p(y, t) is at least defined in the interval
[0, 1]. Let t : Ul -+ A12°' be the corresponding map. The formula for the
Lie derivative of a differential form (Theorem 32, §3.9) implies
dt 4 (wt) _ 'pi
(L) + 0i ('Ca , ,a (Wt)) = V P1 - w) + Vi (Gww, ((WO)
= ipi (wi - w + d(Wt J wt) + W1 J dwt) = tipi (wt - w - da) = 0.
Thus Spi (wl) = pp(w) = w, and 4 o cpl is the chart we were looking for,
(DoVI)" Edpindyi w. O
7.3. First Integrals and the Moment Map

As in the Riemannian case, some first integrals can be derived from symme-
try considerations. The isometries (which are not available on symplectic
manifolds) giving rise to these first integrals will be replaced by symplectic
diffeomorphisms which, however, satisfy a compatibilty condition with re-
spect to the Hamilton function under consideration. We will describe this
in detail in the case of an exact symplectic manifold, i. e. , a symplectic man-
ifold whose symplectic form is the exterior derivative of a 1-form, w = d9.
Suppose that a Lie group G acts from the left on M2" in such a way that
each diffeomorphism 1g : M2i' -+ M2ni leaves the form 9 invariant,
l9(0) = 9.
These diffeomorphisms 1g are then symplectic, i.e., they preserve the sym-
plectic structure w. Now let X E g be an element of the Lie algebra of
the group G, and let k be the fundamental vector field corresponding to
X under the G-action on M2in. The evaluation of the 1-form 9 on X is a
function,
.6(X) := 0(.k).
This construction determines a linear Map '1 : g _ Coo(M2,n) from the Lie
algebra g to the space of functions COD(M2in) on the symplectic manifold.
Its properties are the subject of the following symplectic variant of Noether's
theorem.
Theorem 9 (Noether's Theorem).
(1) $ : g - C, (M2,) is a homomorphism of Lie algebras,
'DQX, Y]) = (4'(X ), CY)} .
(2) s-grado4' corresponds to the transition to fundamental vector fields,
s-grad(qD(X)) = f(.
7.3. First Integrals and the Moment Map 239
(3) If the Hamilton function H is G-invariant, then is a first

integral of H,
0.
Proof. Fix an element X E g in the Lie algebra and consider the one-
parameter group of diffeomorphisms corresponding to the group elements
exp(-t X). Its generating vector field is the fundamental vector field X.
The relation l9 (0) = 0 implies that the Lie derivative of 0 with respect to X
vanishes,
0 = cX(0) = X-jd0+d(XJ0) = X_jw+d(-t(X)).
Thus, for every vector field V, we have the equation
-w(X,V) = V(4(X)) = w(V,s-grad(4'(X)))
as well as X = s-grad(4(X)). Using this formula, we compute the difference
{4(X),4(Y)} - .0([X, Y]) = w(X,Y) - 4,([X, Y])
= dO(X,Y) - 4,([X, Y])
= X(4(Y)) -Y(,t(X)) -24([X,Y])
= 2({'F(X), 4'(Y)} - -NX,Y])) ,
and this yields
{4(X), F(Y)} = C[X, Y]) .
If, finally, the Hamilton function is G-invariant, we obtain
{H,fi(X)} = -X(H) = 0,
i.e., 4(X) is a first integral. 0
The elements of the Lie algebra g provide first integrals for every G-invariant
Hamilton function. These first integrals can be combined into a single
vector-valued first integral by passing to the dual space g'.
Definition 6. The moment map of a Hamiltonian system with symmetry
group G is the map IF : M2m -+ g' from the symplectic manifold to the dual
of the Lie algebra defined by
IF(m)(X) CX)(m)
Theorem 10.
(1) The map %P is Ad'-equivariant, i. e., the following diagram com-
mutes:
M2m 19 M2.
(2) %V is a first integral of H.

Proof. The fundamental vector field X of a G-action has the following
invariance property:
d
X (1,(x)) = dt [exp(-tX) g x] Jt_o
_ d
d lg [exp(-t Ad(g-1)X) xJ I t-0
dt
= dlg(Ad(g-1)X(x)).
N
The 1-form 0 is G-invariant by assumption, and from this we obtain
'P(lgx)X = 8(X(lg.x)) = B(Ad(g-1)X(x)) = '(x)(Ad(g-1)X). 0
N
In Exercise 11, we discuss the case M'' = T'R3 with symmetry group
SO(3, R) and its usual representation on R3. In particular, it is shown that
the moment map P : T`R3 --+ so(3, R) = R3 coincides with classical angular
momentum, hence justifying its name. Closely related to this situation is
the following example:
Example 6. Consider the 2-dimensional representation of G = SL(2, R) on
M2 = V = R2. Its cotangent bundle is T'M = V x V' ^' V x V, since
the representation V is self-dual. A group element g E SL(2, R) acts on an
element (p, q) E V x V of the cotangent bundle by
g - (p, q) = (gR gq)-
We call two elements (p, q) and (p', q') equivalent if they lie in the same
G-orbit, (p, q) - (p', q'). Let p = (pl, p2) and q = (ql, q2) be the components
of the vectors p and q, respectively. One easily computes that the moment
map is given by
V x V - + s1(2, R), (q, p) -- 2 (g1P2 + g2P1) 1 -gipi

[ q2P2 - 2(g1p2 + g2Pl)
In particular, this map is equivariant with respect to the adjoint action of

SL(2, R) on sl(2, R). The moment map is best studied by examining its
action on SL(2, R)-orbits. For this, observe that the quantity
det(q, p) := det [qi pi = g1P2 - 92P1

g2 P2
is SL(2, R)-invariant. It thus allows to parametrize the orbit space; we omit
the easy proof here'.
1For details on this SL(2,R)-action, we refer to §1.4. of the book by Hanspeter Kraft,
Ceometnsche Afethoden in der Invariantentheorie. Vieweg, 1985.
7.4. Completely Integrable Hamiltonian Systems 241
Theorem 11.
(1) If det(q,p) _: A 56 0, then
(q, p) -
( (1)1(0))
0 A
(2) det(q,p) = 0 if and only if p and q are linearly dependent. In this

case, there exist infinitely many G-orbits, for which one can choose
the following representatives:
with AERR.
( (0)1(0)),
0 0 ((1)1(0)),
0 0 ((,U)
0 W)
0
Thus, the moment map acts on G-orbits as follows:

((01),(A0)) LA02
-A/2J' \(0)' 0// ~ [0 0 J
((0) ' (1)) ' (CO) ' Co)) [0 0

In particular, the generic orbits with parameter A 54 0 are mapped to semi-
simple elements of the Lie algebra sl(2, R). Their orbits are 2-dimensional
closed submanifolds of sl(2, J).
7.4. Completely Integrable Hamiltonian Systems

In this section we will make use of the following fact concerning the structure
of discrete subgroups IF of the additive group IRA.
Theorem 12. Let t C ]RI be a discrete subgroup. Then there exist linearly
independent vectors vl, ... , vk such that
k
r= m; vi : m; an integer
i. e., F is the lattice generated by the vectors vi, .... vk.

Proof. If I' # {0} is not trivial, we choose a vector ryl E r such that
I I7i I I 0 0 and consider the ball D" (0; I I7i I I) The intersection D" (0; I I71 I I) nr
is a compact and discrete subset of IR^, hence finite. Thus, on the straight
line generated by 71i there exists a vector 7i E D° (0; 117, 1I) n r realizing the
minimum of the distance to 0 E lR'. For this vector we have
Ht 7i n r = {m 7j : m an integer),
since any vector x 36 m7i belonging to the intersection (1R 7i) n r would
have to lie in one of the segments and then (m+1)7i -x
would be a vector on the line through 7t with a smaller distance to 0 than 7l*.
If the group r contains only integer multiples of -y,*, the proof is completed.
242 7. Syrnplectic Geometry and Mechanics
Otherwise, there has to exist a vector 1'2 E r\{m - ryi : man integer}. We
project ry2 orthogonally to the straight line passing through ry1 and denote
by Y2 the resulting vector. It lies in one of the half-closed segments y2 E
[m - 'y , (m + 1) ryi ). Let E be the cylinder with axis [m' . (m + 1) - 7i)
whose radius is equal to the distance from ry2 to the line through 'Y1. In this
cylinder, there are again only finitely many elements of the group r. Let rye
be the vector in r n E whose distance to the axis of the cylinder is minimal
and which is not a multiple of ryi . Then we have
2
r n {R 7i ED R7;) mi ry, : mi an integer .
i=1
In fact, if there were a point x 0 miel + m2e2 in r n {Rryi ® R-y }, then x

would belong to the interior of a parallelogram in the {'y , 7s }-plane. Taking
the difference with a vertex of this parallelogram, we obtain a vector in r
lying closer than -y; to the axis of the cylinder. Repeating this construction
finitely many times proves the assertion. 0
Corollary 2. Let F C R" be a discrete subgroup. Then R"/r is diffeomor-
phic to the product of a k-dimensional torus Tk with R"-k,
R"/r Tk x Rn-k
Theorem 13 (Arnold-Liouville Theorem). Let (M2m, w, H) be a Hamil-

tonian system, and let fl = H, f2, ..., fm be m functions with the following
properties:
(1) all functions fi are first integrals of H:
(2) the functions fi commute, { fi, f;} = 0;
(3) the differentials dfl, ..., df,,, are linearly independent at each point;
(4) the symplectic gradients s-grad(fi) are complete vector fields.
For a given point c = (cl, ... , c,,) E R'", we consider the level manifold
Al, = {x E M2m : fl(x) = C1, ..., fm(x) = C,n}.
Then:
R'-k
a) The connected components of Al, are diffeomorphic to Tk x
b) The vector field s-grad(H) is tangent to Mc. In particular, each
integral curve of this vector field is completely contained in one of
the level manifolds.
c) If Mc is compact and connected, angle coordinates y91, ... , cp,,, can
be introduced in Mc -- T' so that the integral curves of s-grad(H)
are described by the system of differential equations

y , = vi, v; = constant.
Proof. Consider the flows 4i , ... , 4if :
M2m M2, of the symplectic
gradients s-grad(fi). Since
0 = s-grad{fi,f3) = [s-grad(fi),s-grad(f3)1,
all these flows commute with one another (Theorem 36, §3.9). This deter-
mines an action of the additive group R' on the manifold M2'":
(tl, ... , tm) x = .01 0 ...
The orbits of this R'-action coincide with the connected components of the
level manifolds. In fact, since
0 = {ff,fi} = s-grad(fi)(f3),
each function fi is constant on every orbit. Hence, the orbits of the R'"-
action are contained in the level manifolds. On the other hand, both are
m-dimensional submanifolds of M2in, since the differentials dfl, ..., dfm are
linearly independent. The isotropy group r(xo) = {t E Rm : t xo = xo}
of a point xo E M2rn for the R'-action is discrete. Hence each component
of a level manifold is diffeomorphic to the product of a torus and euclidean
space:
Rm/r(xo) = T" x ][ -
Assertions a) and b) are proved, so we turn to the remaining one. Suppose
that a level manifold Mc is compact and connected. Choose a point xo E Mf
and denote by v1, ... , v,1 E Rm a basis of the isotropy group r(xo). Repre-
senting the basis vectors {v;) of the vector space I(tm as linear combinations
of the vectors in the standard basis e1, ... , em of R..
M
vi = 1: ajaeo ,
a=1
we obtain a quadratic matrix A := (aid). Let
m
r(m) _ {ni.ei : ni an integer
i=1
denote the standard integral lattice in Rm. Then
m m
0: Rm/r(m) -. Rm/r(xo) _ - o Exi. ei) = E xi . Vi,
i=1 i=1
defines a diffeomorphism. The inverse of this map,
4D-1 : MM -+ Rm/r(m) = S1 x ... x S',
as well as its components, 4b-1 = (WI, ..., cp"), lead to the angle coordinates
for the level manifold M, In fact, if yl, ... , y' are the coordinates in
R"'/r(xo) = MM determined by
1 m 1 M. VM,
then, by the construction of the R'-action on Af,
s-grad(fi) = y;
With respect to the gyom}-coordinates, this yields
m a
s-grad(fi) = aj,
a=1 awa
In particular. the symplectic gradient s-grad(H) is a vector field with con-

stant coefficients on the torus MM = T', and the third assertion results by
taking v. := alb.
W e want to discuss more closely h o w the angle coordinates ( 1, ... , m) of

a compact and connected level manifold can be determined directly from
the commuting first integrals. This will lead to an explicit algorithm for
the integration of a Hamiltonian system (M2m, w, H) provided with suffi-
ciently many commuting first integrals. Because of this procedure, these
systems are called completely integrable (or integrable by quadrature). De-
note by wl (c), ... , ul,,, (c) the frame of 1-forms on Al, dual to the vector fields
s-grad(fl ), ..., s-grad(f,,,). The representation of the vector fields s-grad( f;)
in terms of the vector fields 8/&pj immediately implies the following formula
for the differentials:
M
d'pi = aia wa(c)
=1
Let ik be the closed curve in MM corresponding to the parameter values
rpm=0. Then
m
aik = f dcpi = F, ai0 f wn(c)
'Y k a=1 k
Hence, first the coefficients aid and then the angle coordinates can be com-
puted directly from the first integrals. We summarize this in the form of an
algorithm comprising five steps.
Step 1. Fix c = (cl, ... , c,,,) E IIt"' and let Al, be compact and connected.
Choose a homology basis y', ... , y", for the first homology group H, (Al,: 7L).
Step 2. Compute the symplectic gradients s-grad(fi) of the first integrals

f1=H,f2.....fm
Step 3. Determine the frame of 1-forms w1 (c), ..., wm(c) dual to the frame
of vector fields s-grad(f l ), ... , s-grad (f.. ) on M.
Step 4. Compute the line integrals frk w0(c), and invert the resulting
(m x m) matrix. This yields the matrix A = (aij(c)).
Step 5. Compute the angle coordinates (pi(c) on the level set MM from the
equations
m
dvi = E ai0(c) wa(c), 1 < i < m.
Q=1
This procedure computes the angle coordinates Vi(c) on one level manifold
Mc. Note that, according to Step 5, these are only determined up to con-
stants. As we vary the parameters c = (cl, ... , c,,,), the Cpl, ..., cp,,, become
functions on an open neighborhood of a level manifold Mc C M2m. Since
the symplectic gradients are tangent to Mc, the Poisson bracket with the
original functions f1, . . . , fm is computed by
IVi,fjI = dpi(s-gradfj) = aij(fl, . . . , fm).
Moreover, it is a function exclusively depending on fl, . . . , fm. Similarly,
we prove that the functions {Vi, cpj } are also constant on the level sets.
Lemma 1. The Poisson brackets {cpj,Vj} = bij(fl, ...,fm) are functions
depending only on fl, ... , fm. In particular, they are constant on each level
set 't1c.
Proof. We compute the derivative of {vi, wj } with respect to the vector

field s-grad(fk):
{{Vi,'j},fk} _ {Wj,fk},'Pi} - {{fk,'ci},'pj}
_ -{aik(fl, ...,fm),'Pi} + {aik /lfl, ...,fm),pj}
m COajk m 49 }
Oya
a=1 a_I
m
8ajk
as - aaEk
fta aj.)
Thus all the derivatives cpj} (1 < k < m) are constant on ev-
ery torus T' = Afc. But then the Poisson bracket {Vi, oj} itself is constant
on every level manifold Mc. 0
Now we alter the angle coordinates, which up to now were considered only
on a single level manifold, by a suitable constant on adjacent level manifolds.
The aim of choosing these constants of integration for the angle coordinates
is to obtain functions cpi commuting on M2n'.
Lemma 2. There exist functions Bl (yl, ... , B," (yl, ... , y') such
that the angle coordinates
Bpi := Wi+Bi(fl,...,fm)
commute on the symplectic manifold M2m, {(pi , Vj*) = 0.
Proof. Using the notation {cpi, f;} = aij and {Wi,co } = bij, we apply
the Jacobi identity to the triples (cpi, ip,, fk) and (ipi, Wj, cpk), and take into
account the fact that the Poisson brackets { fi, fl) vanish. Thus we obtain
the relations m
aaik Oa,j
E a;a- aia = 0,
m
ab;k 1
E Obi;
y u 'aka +
ft la
' afa +8bki
d ajQ J = 0 .
Inserting also the coefficients air of the inverse of the matrix A = (aid), we
consider the 2-form
m m
Il :_ > bijai°a'p dy° A dyO.
i,j=1 a,p=1
The above relations say that fl is a closed 2-form. In fact, the first relation
means that the 1-forms
m
of E a° dy°
a=1
are closed, doi = 0. Hence we can choose coordinates z1, ... , z'" such that
of = dzi, and 1 becomes
M
fl = > bij dzi Adzj .
i,j=1
Since
m
_ ay°
Obi;
Ozk 1 Obi,
Oya 8zk
a= 1
8bij
or aka
or
the second relation then precisely expresses the vanishing of the exterior
derivative dfl = 0. The angle coordinates we set out to find are now taken
to have the form m
Bpi Wi + aiaBa
O=1
with functions Bl*, ..., Bm depending only on fl, ... , f,,,. Then
m
aY ai. B.
a;l_ L syv
M
aB;
+ a (aiaaj3 - ajasid)
a.8=1
Taking into account the first of the relations above, the condition
0 turns out to be equivalent to
bij = 1:
a.3=1
".
a 33
(aiflaja - ai.aj,3)
This system of differential equations can be reformulated as

aB M -
a r3
m
[: bijaiaajZ,
8ii - aya -
i;=1
and, by Poincare's lemma, it has a solution, since the 2-form f) considered
above is closed.
Thus we can determine the constants of integration occurring in the transi-
tion from the differentials
m
d'pi(c) = E afa(r) wa(c)

a=1
to the angle coordinates on the individual level manifolds in such a way that
the functions Vi, defined in a neighborhood of a level manifold, commute
on M2'". Now we add so-called action coordinates J1, ..., J,, and thereby
bring the symplectic structure near a level manifold into normal form. The
Hamilton function H = f, is itself constant on the level manifolds and only
depends on the action variables, H = H(J1, ...,J,,,).
Theorem 14 (Action and Angle Coordinates). In a neighborhood of any
compact, connected level manifold M,: C At" of a Hamiltonian system de-
termined by m commuting integrals fl, ... , fm, there exist angle coordinates
Cpl, ... , v,,, and first integrals J1, .... J,,, such that
in
w = dindJi.
i=1
In particular, this implies Vj} = 0 = {Ji, Jj} and {y'i, Jj} = bij.
Proof. First, we determine the angle coordinates near the compact, con-
nected level manifold MM such that
J(pj, ypj} = 0 and {tpi, fj} = aij(fi, fm)
We look for the functions J1, ..., J,,, using the Ansatz J, = Ai(fl,
and compute the Poisson bracket
m
8Aj ai.
OY.
The condition {,..1j} = 8ij leads to the system of equations

OA = aij
ayj
where a'j is the inverse of the matrix aij. By Poincare's lemma, the inte-
grability condition is
8aij 8aik
8yk = $yjj
On the other hand, we obtain from the Jacobi identity
0 = {`r'k,{iPi,.fj}}+{Vi.{fj, 7k}}+{.f .{Y"'k,Y^'i}},
and, taking into account {j,9i.k} = 0, this immediately yields
8aq = L 8ukj
E fta aka
Q=1 a=1
8 a aip..
y
A simple computation shows that this relation is equivalent to the integra-

bility condition for the coefficients aij of the inverse matrix. 0
Example 7 (Two-dimensional Hamiltonian System). Consider in R2 with
the symplectic structure w = dp A dq a Hamilton function H(q, p) for which
the level curves
(q, P) E 1R2 : H(q. p) = c}
are closed. The action variable J = J(H) is a function of H, and, together
with the angle coordinate , we have
dpAdq = dV AV =
Applying the Hodge operator * of R2, we see that dy: is proportional to the
1-form *dH,
dV * dH .
J'(H) IIdHII2
The integral of d,o over each level curve is an integer, which we take to
be equal to -1. This condition is called the classical Bohr-Sommerfeld
condition. The equation
J'(c) 1 * dH
nor IIdHII2
uniquely determines the action variable J = J(H) in terms of the Hamilton

function H. Consider the domain bounded by the level curve A4,,
Q, = {(q, p) E R2 : H(9, p) 5 C1.
The vector field W := grad(H)/Ilgrad(H)II2 satisfies W(H) - 1; hence its
flow +t maps the set flc onto tl +t. We compute the change of the area of
the domain:
d
(vol(S2c)) =
tli o
t (J$2)
k
t -J st4
2) =
Js24 Gbb'(dR2)
r
d(W i dlt2) = dH = X(c)
Ji1c J lo
W I dR2 =
If ,1 4 IIdHII2 *
.
Thus the action variable J = J(H) can be interpreted as follows: J(c) is

the volume vol(Q,) of the domain bounded by the level curve hf, _ {(q, p) E
R2 : H(q,p) = c}.
Example 8 (Spherical Pendulum). Consider spherical coordinates on the

sphere S2\{N, S} with the north and south pole deleted,
h(4,) = (cos cp cos ti, sin yp cos 0 , sin g) .
The Riemannian metric of the sphere is then described by the matrix (see
Example 14 in §3.2)
_ 1 0
9 O cost'
L
Denote the coordinates in the cotangent bundle by (cp, o, pw, p v) and con-
sider the Hamilton function
H = 2v+2
H describes the motion of a pendulum of length one suspended in the center
of the sphere. The meaning of the angles tp and t'} can be seen in the picture
to follow. For simplicity, the gravitational constant was taken to be one.
The variable (p does not explicitly occur in the Hamilton function; hence
P:= p,, = Ocos2 iii is a first integral, {H, P} = 0. The Hamiltonian system
(T*S2, H) is thus completely integrable. The level manifold
M2(ci,c2) :_ {pw,po) E T* S2 : P = c1, H = C2}
is empty for negative values of the parameter c2, and it consists of the
south pole remaining at rest in the case c2 = 0. Hence we suppose that
the parameter is positive, c2 > 0. The equations describing the manifold
A12(cl,c2) are
C1 = pr,, c2 = 2 + 2 cs2 i + 1 + sin ?P.

The relation cl = 0 implies that cp has to be constant. In this case, the
second equation of motion reduces to that of a planar pendulum, so we will
henceforth exclude this case. Depending upon the sign of c1, the function
is monotone increasing or decreasing, i. e., the pendulum does not change its
direction of motion. Rewriting the second equation and setting z = sin 4i,
we obtain
p= C- 2
c2-1-z- 2(1 z2) U(z)

and see that the function U(z) thus defined cannot be negative. The limiting
case p = 0 corresponds to the pendulum moving in a fixed plane, hence
on a meridian. To see where the function U(z) can be strictly positive, we
multiply it by the denominator 1- z2 and look for the zeroes of the resulting
cubic polynomial,
c2
V(z) = (1 - z2)U(z) = (c2 - 1 - z)(1 - z2) - 2 .
At the boundaries of the interval, V(±1) = -c/2 < 0 is negative, and at

+oe the function V diverges to +oc. Hence one of the three zeroes has to
lie above 1, and, since it cannot correspond to any angle 1;i, this zero has
no physical relevance. In the generic case, the other zeroes belong to the
interval (-1, 1); between them V, and hence also U, is positive. We conclude
that V(z) has the qualitative behavior of the graph on the previous page.
The mass point can only move between the two meridians corresponding to
the zeroes in the interval (-1, 1). The boundary values t ,'4'2 defined this
way are actually reached at the end of every up or down swing. Summarizing,
the level manifold can be parametrized by the two parameters 4z = V and
0= via
( P Cl! cos - 2 - 2sini ) .

It consists of 2 two-dimensional tori, where ' only takes values in (>G1, 021.
The corresponding coordinate vector fields are
a a a a apti; a
a = TT' a = a + o ' ap ,
We express the symplectic gradients of the first integrals in the coordinates

of the level manifold:
P;, a a d P,2. a
s-grad(H)
= cost ;, a + p av - a (2 cost V, + 1 + sin V) aplo
Pro a a
= cost' ap +Pva +Pv a i "PV
pw a a
_ 2 Zi app. + Pv a,.
s-grad(P) = a = ate, .
The dual forms w1(cl , c2) and w2(cl, c2) are thus
1 C12
wl(Cl,C2) = de*, w2(cl,c2) = dw - p o Cos '+G dip .
PO
We compute the periods with respect to the homology cycle 'rl which is
parametrized in M2(c1i C2) by 't()'. The factor 2 is a consequence of the
fact that the boundaries, y1,'Y2, of the interval correspond to the minimum
and the maximum of the motion, whereas a cycle is meant to be a motion
between two extremal points of the same kind:
b1l = f wl = 2
ti Jc>>
%
Py
b12 = f7i
w2 = -2c1
r 02 diP
Jv Pb oos2
Let, similarly, 72 denote the homology cycle determined by 0 < gyp` < 2Tr.
Then
b21 = Jwi = 0, b22 = Jw2 = 27r .
The matrix occurring in the algorithm for computing the angle coordinates
is now easily calculated:
-l 1 -j w2
- [a21 f7twl 2vfnwl
[b21 b221 a22]
0 1/27r
The resulting quotient of the basic frequencies is
V1 =all =_1f1-'2= C1 112 di&

21r 7r cost tai ' PW
V2 a12 , .//ol
This is nothing but the perihelion precession, i.e. the total variation of the
angle V for a complete cycle:
02
d` d1/b = 2
O +a
J 'rI
dye = 2
J d
.1,1 fo , fo, I
1
. dpi = 2c1
COS2t' ' PG = it d 27r L1 .
In general, the motion is quasi-periodic. The trajectory is closed if and

only if vl /v2 is rational; otherwise, it is dense on the torus. The transition
to action angle coordinates allows us to determine the physically relevant
basic frequencies of the system without having to explicitly perform the
integration. This accounts for their importance in astronomic perturbation
theory.
7.5. Formulations of Mechanics

Newton's equations describe the motion of a mechanical system under the
impact of a force. The latter is understood as a vector field depending upon
position and velocity, and is central for Newton's formulation of mechanics.
7.5. Formulations of Mechanics 253
During the 18th century, this view changed in that Lagrange considered the
action integral as the fundamental quantity for the description of dynam-
ics. Newton as well as Lagrange formulated mechanics within the tangent
bundle of configuration space. In the 19th century, by transition to the
cotangent bundle, Hamilton succeeded in formulating the dynamics of me-
chanical systems within the framework of symplectic geometry. The aim
of this section is to explain these fundamental ideas of mechanics and the
related mathematical structures.
Newton (1643-1727) Lagrange (1736-1813) Hamilton (1805-1865)
Newtonian systems Lagrangian systems
Hamiltonian systems
i 1
Newtonian systems hyper-regular Legendre

with potential energy Lagrangian systems transformations
Mathematical Contents:
Riemannian geometry Finsler geometry symplectic geometry
Formulation of Mechanics According to Newton
In Newtonian mechanics, the state of a mechanical system is described by

finitely many real parameters. This leads to the notion of configuration
space. which is a smooth and finite-dimensional manifold Mm. A motion of
the mechanical system is a curve 'y : (a, b) - M' in configuration space.
Its tangent-the velocity-is then a curve ' : (a, b) - TM' in the tangent
bundle. and this space is called the phase space. According to Newton,
the forces acting on the mechanical system are described by vector fields
depending only on position and velocity, that is, vector fields X on TM"'.
However, not all vector fields are allowed, since a force can act only in space.
The force vector field X has to satisfy the condition

drr o X = Id.
Here 7r : TM" - Mm denotes the projection of the tangent bundle, and
dir : TTM' --+ TM' is its differential. Summarizing, we arrive at the
notion of a Newtonian system.
Definition 7. An autonomous Newtonian system is a triple (Mm, g, X )
consisting of a manifold M'", a Riemannian metric g, and a vector field on
the space TM'" such that
d7r o X = IdTMm .
The function T : TM'" -+ R defined by
T(v) := 2 g(v,v)
is called the kinetic energy.

Definition 8. A motion of the Newtonian system (M"', g, X) is understood
to be a curve y : (a, b) - M' in configuration space whose curve of tangents
ry : (a, b) - TM'" is an integral curve of X,
'1'(t) = X MW
This is an invariant formulation of Newton's equation.
Example 9. Consider R with the coordinate x and identify TR = R2 with
R2. Here the coordinates are denoted by {x, i}. The vector field
8 1 2 0
X= wax+m(-k x-P)8i
on TR has the required projection property, and a curve x(t) in R is a motion

in this Newtonian system if x(t) is a solution of the oscillator equation
ml(t) = -k2x(t) - pe(t) .
Example 10. Let (Mm,g) be a Riemannian manifold. We define a vec-
tor field S : TM°1 - TTMm on its tangent bundle-the so-called geodesic
spray-as follows: If v E TAM' is a tangent vector, then there exists pre-
cisely one geodesic line y,,,,, : (-e, c) -> Mm such that
7'r,.(0) = X, % AO) = v.
Consider its tangent curve, %,v : (-e, e) - TM'", and set
S(v) dt (%"(t)) t=o

The relation 7r o ryx,,, (t) = ry v (t) implies dir o S = IdTMm. Hence (M', g, S)
is a Newtonian system, and the motions of this system are the geodesic lines
of the Riemannian manifold (Mm, g). In the coordinates {x', ii} of the
tangent bundle, the geodesic spray is given by the formula
S= x'
axi
- E r;k x'xA axi
i=1 i,j,k=1
a straightforward consequence of the system of differential equations de-
scribing geodesic lines-see §5.7.
Many Newtonian systems have a potential energy. This is a smooth function
V : Mm --+ R defined on configuration space. The gradient grad(V) is a
vector field on Mm locally determined by
grad(V)
'" av a
- ì,j=1
g'' (x) axi axj
In the sequel we will need, however, a different vector field, denoted by
grad(V). This will be a vector field on phase space. At the point v E TM'",
it is defined by the following equation:
d
grad (v + t - grad(V)(ir(v))) e-o
(V) (v) = dt -
In the manifold Mm, the curve v + t - grad(V)(a(v)) projects to the base

point rr(v) E M' of the vector v E TM'. Hence the vector field grad(V)
projects to zero under the differential dir,
d7r o grad(V) = 0,
and, for any potential energy V, the vector field X := S - grad(V) is an
admissible vector field on TM'" in the sense of Newtonian mechanics. In
local coordinates, we obtain the formula
in
av a
E e(x) axi aij
i,j=1
Definition 9. A Newtonian system with potential energy is a triple (M, g, V)
consisting of a Riemannian metric g and a potential energy V : M' -+ R.
The corresponding force vector field is
X = S-grad(V).
A motion in a Newtonian system with potential energy is defined to be a
curve 7 : (a, b) Mm whose tangential curve (a, b) -+ TM'" is an
integral curve of X.
Definition 10. The energy of a Newtonian system (Mm, g, V) with poten-
tial energy is the sum of the kinetic and the potential energy,
E:TM' R, E=T+Voir.
Theorem 15 (Energy Conservation for Newtonian Systems). Let (Mm, g, V)

be a Newtonian system with potential energy, and let X = S - grad(V) be
the force vector field. Then
dE(X) = 0.
In particular, E(y(t)) is constant for every motion-y(t) of the system.
Proof. In local coordinates, the energy E and the force vector field are
given by the formulas
I M
E = 9 E 9ij(x)xY +V(x),
i,j=1
X = x' a ii - I'v (x) +

i=1 k=1 i=1
8i 9ikW I 82k
.
Using the expression for the Christoffel symbols riki from §5.7,
gkQ (x) (Lgii (x) + 8x (x) -ij (X)) ,

2 axQ
we obtain dE(X) = X (E) = 0 by an elementary calculation. 0

Motions with large energy in a Newtonian system (M'", g, V) with potential
energy are-up to a change of parametrization-geodesic lines with respect
to a new Riemannian metric. This construction will lead to the Maupertuis-
Jacobi principle. Suppose that the potential energy V : Mm - R is bounded
from above by Eo,
sup{V(x): xEMm} < Eo.
Then g' = (Eo-V) -g is a Riemannian metric on the manifold Mm. Consider
a motion y : (a, b) - MI of the Newtonian system (MI, g, V) with energy
Eo,
Eo = 29('Y(t),'Y(t))+V(7(t))
Since Eo > V(y(t)), the tangent vector y(t) vanishes nowhere, and the
function
s(t) := f r' (Eo - V(7(µ)))dp
a
becomes a strictly monotone function a : (a, b) - (0, b* ). We invert this
function and thus view t E (a, b) as a function of the parameter s E (0, b*),
t = t(s). Let the curve .y*(s) be the initial curve y in this new parametriza-
tion.
Theorem 16 (Maupertuis-Jacobi principle). Let -y(t) be a motion of the

Newtonian system (Mm, g, V) with energy E0. Then ry' (s) is a geodesic line
in Mm with respect to the Riemannian metric g' = (Eo - V) g.
Proof. The Christoffel symbols1) r and 't V of the metrics g and g' are, in
local coordinates, related by the formula
k k 1 1 _ 8V 8V 8V ak
= -&k + 9 9ij
rtj r'3 + 2 (Eo - V) ajk C7xi 02-a
We write the motion -y(t) _ (x' (t), ... , x'(t)) in local coordinates. Then
dxk _ dsk dt _ 1 1 dxk
ds dt Ws- 72= (Eo - V) dt
_ dxk m 8V dxa
d2xk
ds2
1
2(Eo - V)2 dt2

d2xk 1
2(Eo - V)3 dt Ox- dt ' 1

and, using the equation of motion,
m
d2xk m rk Ix' dx' OV ka
dt2 -- i,j=1
ij dt dt - a=1 8xa 9
as well as the energy condition
m
dxi dxj
E
i,j=i gij dt dt
2(Eo - V),
we obtain the claimed result:

d2xk m k dx' dx' 1 8V ka
WS-2 + I'i ds ds =
ij=1 -2(Eo - V)2 a=1 8xa9
m in
1 8V ka dx' dx-I
+ `1(Ec, __V) 3 = 0.
a=1 8xa9 ij=1 go dt dt
Formulation of Mechanics According to Lagrange
The transition to Lagrangian mechanics proceeds by considering the La-

grange function of a Newtonian system with potential energy.
Definition 11. Let (M'", g, V) be a Newtonian system with potential en-
ergy. The Lagrange function (or Lagrangian) L : TM' R is the difference
of kinetic and potential energy,
L = T-Voir.
Theorem 17 (d'Alembert-Lagrange). A curve 7 : (a, b) -+ Mm is a motion

of the Newtonian system (Mm, g, V) with potential energy if and only if the
Euler-Lagrange equations hold:
dt (ex (y(t))) = 8x (7(t))

Proof. We prove this theorem again using local coordinates. The curve
7(t) _ (x1(t), .. . , xm (t)) is a motion of the Newtonian system with potential
energy if and only if it solves the system of differential equations
»i m
ik =-
ij=1
I jiìj - a=1
E Va9ak
For brevity, we denoted by V. the partial derivative of the potential function,

4G, := 8V/8x°. The Lagrange function is
L('y) = gijii2j - V(1'),

2
i,7=1
and this leads to the difference

m m
d 8L 8L d9ia 1 99. Q axa + 9iaya+ V; .
dt (8zi a,;3=1
ax-*l 2 axi) a=1
Multiplying the Euler-Lagrange equations by gik and summing over the

index i, this system of equations turns out to be equivalent to
m m
a9ia 109.e
E
0 = xk + M V 'k +
i-l i9
M
E ` 8x3
i=1 a.3= 1
2 axi 9
The claim now immediately follows from the formula for the Christoffel
symbols r 13.. o
Thus the equations of motion of Newtonian mechanics are, in the case of a
potential force, equivalent to the Euler-Lagrange equations. For the latter,
it does not matter that the Lagrange function arises as the difference of a
kinetic and a potential energy. Hence we define:
Definition 12. An autonomous Lagrangian system is a pair (Mm, L) con-
sisting of a manifold Al" and a smooth function L : TMm - R. A La-
grangian motion is a curve -y : (a, b) -- M" which solves the system of
Euler-Lagrange equations
Wt
(er ('i (t))) = dL ('Y(t))
Example 11. Let (M'", g) be a pseudo-Riemannian manifold and A a 1-

form on it. The Lagrange function
L(v) = g(v, v) - A(v)

2
generalizes, in a sense to be discussed in Chapter 9, the motion of a charged
particle of mass m under the influence of the Lorentz force of the electro-
magnetic field generated by A.
Motions in a Lagrangian system can be understood from the point of view

that they are critical points of the action integral L. This measures for a
curve -y : [a. b] -p Mm the mean value of the Lagrange function on this
curve,
b
L(y) :=
Ja
Theorem 18 (Principle of Least Action). A curve -y : [a. b] Mm is a
motion of the Lagrangian system (Mm, L) if and only if the variation of the
(jb)
action integral vanishes for every variation y1, of the curve with fixed initail
and end points, y1,(a) = y(a), -y. (b) = y(b):
d
0.
dµ LO =
Proof. We compute the derivative of the action integral with respect to
the parameter p in coordinates -t,, (t) = (x1(µ, t), ... , x'"(µ, t)) by partial
integration:
[aL((t))
=Jab
d (,C(-t,.)) 1,0 - d (a (7(t))) (0, t)dt.
The functions 8x'(0, t)/8µ are arbitrary functions vanishing at the end
points of the interval [a, b], and hence the Euler-Lagrange equations are
equivalent to the condition
dµ (L(yµ)) Iµ=o = 0.
For general Lagrangian systems, there exists a notion of energy which, on

the one hand, generalizes the energy of a Newtonian system with potential
energy, and is, on the other hand, a preserved quantity. This Lagrangian
energy is obtained by first introducing the Legendre transformation. Let
L : TM' R be a Lagrange function, and let v E .,Mm be a vector at
the point x E Mm. Now restrict L to the tangent space .,Aim and consider
the differential D(LIT=SIm)(v) at the point v E TTMm. This is a linear map
TTll1"' IR. hence a covector in T *M'.
Definition 13. The Legendre transformation C : TM' T'M' of an

arbitrary Lagrangian system is the map
,C(v) := D(LIT=Mm)(v).
Example 12. If the Lagrange function L = Zg - V is the difference of a
kinetic and a potential energy, the following relation holds for v, w E TM':
G(v)(w) = 2 . D(g)(v)(w) = &,w).

Hence the Legendre transformation L : TM"' T* M' is simply the iden-
tification of the tangent bundle with the cotangent bundle via the metric.
Definition 14. The energy of a Lagrangian system (MI, L) is the function
E : TM' -' R on the tangent bundle defined by
E(v) = L(v)(v) - L(v).
In the case of a Newtonian system with potential energy, we have
E(v) = L(v)(v) - L(v) 2g(v, v) + V(r(v)) .

This shows that the energy in the sense of Lagrangian mechanics coincides
with the Newtonian energy.
Example 13. We compute the Legendre transformation for the Lagrange
function from Example 11. Let (x, y) E TTM" with local coordinates
xl... , xm and y', ... , y'. Then,
m m
...,y"') = 2 Egjjy'1!r - EAiy',
1,3 i
and, as an element of T,M, its differential is
m m
D(LITTMm)(yl,...,ym) = mEgify'dhe
-EAidy'.
i,j i
We evaluate this map at (x, v), v = (v', ... , v'"):

G(y)(v) = D(LI T:MO1) (y)(v) = m . g(y, v) - A(v) .
By definition, this yields for the energy the value
E(v) = C(v)(v) - L(v) =
2 g(v, v),
which has the remarkable property of not depending on A. A more careful
physical analysis shows that the energy of a charged particle indeed only
depends on the electric, not on the magnetic field. This is related to the
fact that the magnetic field does no work on the particle (see §9.5).
Remark. Denoting the coordinates on TM' by {x', , X-, x', ...'e }

and the coordinates on T` Mm by {qj, ... , qm. pi, .... p,"}. we see that the
Legendre transformation f- is given by
aL
qi = x' and pi =5p,
and the expression for the energy E takes the following form:
E(x,
_ "' OL xi - L(x, x) .
x) - ==Y 8ii
Theorem 19 (Conservation of Energy for Lagrangian Systems). The energy
E(y' (t)) of each motion ti(t) of a Lagrangian system (Mm, L) is constant.
Proof. The energy of a curve is
8L dxi
E(7(t)) _ f72i dt
- L(ti(t))
t-1
and by differentiation we obtain
d "' 8L dx' dx- 8L dx' 1 ' &L dx'
dt E(Y(t)) _ ij=1 C 8ii8xj dt dt + C7x'&i dt dtZ J - i=Y
dxi dt
Using the Euler-Lagrange equation
m
i3L c7L dx' "` OL d2xi
Ox-. = EY O±'Oxi dt + axiaxj dt2
ij= i,7=Y
we immediately obtain the assertion. 0
Thus the energy is a first integral for any motion in a Lagrangian system. As
in §7.3. further first integrals can be derived from symmetries of the Lagrange
function. To this end, consider a one-parameter group of diffeomorphisms
,P6 : Al' M' of the configuration space as well as its generating vector
field,
d
V(x) := d('ts(x))j8=0
on Al'. The differentials d(4 Q : TM' TM"' are diffeomorphisms of the
phase space TM"' into itself.
Theorem 20 (Noether's Theorem). Let the Lagrange function L be invari-
ant under the action of a one-parameter group of diffeomorphisms,
L(d(4s)(v)) = L(v). Then the function fv : TM' --+ R defined by
fv(w) = lim
µ-.O µ
is a constant of motion for the Lagrangian system (Mm, L).
Proof. Let the one-parameter group of diffeomorphisms $t be determined

in coordinates by
4tt(xil ...,xm) = ('Ft(x1, ...,X"), .... Dm(x'....,xm
The invariance of the Lagrange function implies that
m aL a(De
m
aL O4 dx'
= 0,
ax= as I + E axj . ax=as dt
and the vector field V has the components
m
a
V=E a3
s
s=IUC7-
x '
i=1
Thus we obtain
fv(7(t)) = 1 8x as Is.o'
and from the Euler-Lagrange equation we conclude that
m
d m IL 81' aL 02V dxj
d ax= WS=0 I8=0 + ax' ax_as o dt = o. o
Example 14. To each transformation group'Ft : Mm - Mm preserving
the metric g and the potential function V there corresponds a first integral,
fv(w) = g(w,V),
which is linear in every fiber of the tangent bundle Tlblt. We made use
of this first integral in Theorem 37, Chapter 5, to integrate the geodesic
flow on surfaces of revolution (Clairaut's theorem). Hence first integrals of
the geodesic flow which are linear in the fibers arise from isometries of the
metric.
Example 15. If 't preserves the pseudo-Riemannian metric g as well as
the 1-form A from Example 11, then the Lagrangian of a charged particle
in an electromagnetic field is invariant, and hence Noether's theorem can be
applied to yield the invariant
fv(w) = m g(w, V) - A(V).
Formulation of Mechanics According to Hamilton
We will arrive at the formulation of mechanics according to Hamilton by

starting from a Lagrangian system (M, L) with bijective Legendre trans-
formation C : TM' T'Mm. These Lagrangian systems are called hyper-
regular. A regular Lagrangian system is one whose Legendre transformation
is locally invertible. For simplicity, we confine ourselves here to the case
of a globally invertible Legendre transformation. For example, Newtonian

systems with potential energy or the system of a charged particle moving in
an electromagnetic field (Example 11) are always hyper-regular.
Definition 15. Let (Mm, L) be a hyper-regular Lagrangian system with
Legendre transformation C : TM' --+ T* M' and energy E : TM' --+ R.
The function
H := EoL 1
defined on the cotangent bundle is called the Hamilton function (or Hamil-
tonian) of the system.
Example 16. In the case of a Newtonian system (Mm, g, V) with potential
energy, the Legendre transformation is determined by the formulas
m
(9L
qi = xi and pi= axi = E gi.±
a=1
Inverting this transformation leads to
m
x' = qi and i' = E gtnPa ,
U=1
and the formulas for the energy E, the Lagrange function L, and the Hamil-
ton function H read as follows:
(1) E = 2 gijx'i +V(xl, ...,x'"),
i j=1
m
(2) L = 2 E gijz'xi - V(xl, ...,xm),
ij=1
(3) H = 2 E g''rpipj+V(gl,...,gm).
ij=1
Through this change of phase space-i. e., replacing the tangent bundle
TM"' by the cotangent bundle T'M'-we enter the realm of symplectic
geometry, since T*M'" always carries the symplectic form w = d9.
Theorem 21 (Hamilton's Theorem). Let (M'", L) be a hyper-regular La-
grangian system. A curve -y : (a, b) -' AM"' is a Lagrangian motion if and
only if the curve G(ry) : (a, b) -. T' M"' is an integral curve of the symplectic
gradient s-grad(H) of the Hamilton function.
Proof. The Legendre transformation C is defined by
i OL
qi = x, Pi = axi .
For its inverse map, we introduce the notation
x' = qi and i = I (ql, ... , qm, Pi, ... , pm).
The Hamilton function is then

m
H=
a=1
and we compute its partial derivatives:
+ M
apt 1: (P. -
a=l
OH
a E51-aL aV _
m
a aqt
OL
axt =
_ OL
axt
.
s-grad(H) = E
t=l
(i+)
Thus we obtain a formula for the symplectic gradient:
A curve
G(?'(t)) = 1 21(t), ... , 2m(t), 'oil (fi(t)), -- , OL

is an integral curve of the vector field s-grad(h) if and only if
it = V and (5p
IL
Wt (ti(t))) = axt ('Y(t))
The first equation is trivialy satisfied by setting it = fit, and this proves the
assertion. 0
Exercises
1. In R4 with the symplectic structure w = dxl Adx3 +dx2 Adx4. we choose

the following four diffeomorphisms a, b, c, d:
a(xl,x2,x3,x4) = (21,x2+1,x3,x4),
b(xl, x2, x3, x4) = (xl, x2, x3, x4 + 1),
C(xl, x2, x3, x4) = (xl + 1,x2, x3, x4) ,
d(xl, x2, x3, x4) = (x1, 22 + 24,x3 + 1,x4 ),
and denote by r the group of motions of R4 generated by them. Prove
that w is a F-invariant 2-form. Hence w induces a symplectic form on the
manifold M4 = R4/F. Prove that M4 is compact. Finally, denote by [F, F)
the commutator group of F. Then F/[t, rJ is a free abelian group of rank
three (Thurston, 1971).
Exercises 265
2. Prove that the symplectic form w of a compact symplectic manifold M2m

can never be an exact differential form. Hence the second de Rham cohomol-
ogy HDR(M2m) is non-trivial. In particular, for m 1, even-dimensional
spheres St' have no symplectic structure.
3. Consider M = R2\{0} with the symplectic form w = dx n dy and the

vector field
V_ x 8 y 8
x2+y2 8x+x2+y2 8y
a) Let cp = arctan(y/x) be the polar angle defined in every sufficiently
small neighborhood of (x, y) (0, 0). Compute grad(W) and
b) Conclude from this that V is no Hamiltonian vector field on all of M.
4 (Continuation of Example 3). Prove that the Liouville form on the two-
dimensional coadjoint orbit through the element (a, Q) E g` (3 54 0) is given
by the expression
w= daAdf .
Hint: Show first that the fundamental vector fields corresponding to the Lie
algebra elements (1, 0) and (0, 1) are
aa, and - Qom.

5. Let V be a vector field on a manifold Mm and 1 = V(x) the associated
differential equation. A first integral of V is a smooth function h : Mm - R
which is constant along every solution of the differential equation.
a) Prove that h : Mm -p JR is a first integral of V if and only if dh(V) = 0.
b) Prove that the set Cv (M'n; R) of all first integrals of V is a subring of
C' (M'; R).
c) Show that h(x, y) = y2 - 2x2 + x9 is a first integral of the vector field
V = 2y 8/8x + (4x - 4x3) 8/ay
in R2. Describe the geometric shape of the integral curves in the
(x, y)-plane by means of h.
d) Find a vector field in the plane which has no non-trivial first integral.
6. Let f = a + b i : U -p C be a holomorphic function on an open subset

U of JR2 ^_' C. Consider the vector field V = grad (a).
a) Prove that b : U -. R is a first integral for V.
b) Describe the geometric shape of the integral curves of this vector field
for the functions f(z) = zk and f(z) = z + 1/z.
7. Consider on ]R3 a vector field B (a magnetic field) and its 2-form

B=
as well as the symplectic form on the phase space R3 x JR3 with coordinates
(9, ii) = (x, y, z, Vx, Vy, Vz),
wB = m(dxAdvx+dyAdvy+dzAdvz) - - B.
C
As Hamilton function, we choose the kinetic energy
Ho = 2 (vZ+vy2+vZ).
a) Show that the defining equation for the Hamiltonian vector field associ-
ated with Ho is equivalent to the Lorentz equation
dv _ e
c.vxB.
(*) mdt
b) If B = dA is exact and A denotes the associated vector field, show that
the map
f: 1R3 x JR3 -R 3 x 1R3, (4> ) ' (q, mii + A) (q, p)
is a canonical transformation, i.e., for the canonical symplectic form wo
and the Hamilton function HB,
wo = dxndpx+dyAdpy+dzndpzi HB = -lip- -All
the following equations hold:
f*wo = 'B, f*HB = Ho,
and equation (*) does not change.
c) If B is constant, any particle moves on a helix. Hint: Interpret torsion
and curvature as first integrals. If both these quantities are constant for
a curve, then this curve has to be a helix (Exercise 6 in Chapter 5).
8 (Plane Toda Lattice). Consider in JR2 the second order differential equa-
tions
i= -ex-y, y= ex-y
which in phase space (x, y, i, y) E JR4 are equivalent to t; = X with
a a
X(x,y,i,y) = a TX + y ay - ex-y a +
ai
ex-y
ay
The energy E = (i2 + y2)/2 + ex-y is a first integral.
Exercises 267
a) Show that this system has further first integrals, for example P = i + y
or K = (i - 2y)(y - 2i)/9 - e---y. These quantities are related by
E + K = 5P2/18.
b) The set
M2 (E, P) (x, y, i, y) E 11 P4 : (i2 + y2)/2 + eZ-v = E, ±+y=P}
is not empty if and only if 4E - P2 > 0. In this case M2(E, P) is a
smooth two-dimensional submanifold of R4 without boundary . This
leads to a decomposition of ]R4 into a family of submanifolds, and every
integral curve of X lies completely in one of them.
c) Show that each of the submanifolds M2(E, P) lies in an affine subspace
of dimension three and is diffeomorphic to R2.
d) If {(t) = (x(t), y(t), i(t), y(t)) is an integral curve of X in M2(E, P),
then
x y
i+y = P and f 4E-P -4ez-V = t
Show that (A > 0)
dz _= 1
In
vA- A -Bee
A -Bez v/-A ( I + A -Bee '
and use this formula to integrate the equations for the integral curves
in M2(E, P) completely.
9 (Euler Equations). Let I : ]R3 - R3 be a symmetric positive definite

operator. Consider the differential equation I(i) = 1(w) x w, where x
denotes the vector product.
a) Prove that this differential equation has two first integrals, the energy
2E = (I(w),w) and the momentum M2 = (I(m), I(w)).
b) Conclude from this that the integral curves of the differential equation
are intersections of two ellipsoids with center 0 E IR3 (I 0 IdR3). In
particular, all integral curves are closed. Let Il be the smallest and 13
the largest eigenvalue of I. Then there exists an integral curve only for
2EIl < M2 < 2E13. .
10 (Motion in a Central Force Field). Let a potential force F act on a point

of mass one in R3,
dU
F drr) r -grad(U(r))
with a function U(r) depending on the radius r = IIxil only. The energy
E = 11±112/2 + U(r) is a first integral.
a) Show that M = x x . is a further first integral (M is the angular

momentum), and conclude from this that every trajectory of the point
lies in a plane in R3.
b) Determine for which values of the parameter M E JR3 the level surface
A3(M) = {(x,x)ER6:xx:i=M}
is a three-dimensional submanifold of phase space R6.
c) Let r(t) be the distance to the origin of a motion whose an;;ular mo-
mentum is M. Then
d2r _ _dr
dU 1IM112
dt2 + _73-
i.e., r(t) describes the motion of a point in R1 under the effective force
F2 = -grad(V) with effective potential V(r) = U(r) + IIMII2/2r2. The
energy of this motion is
I I2r2Iz
E* = 2 + U(r) +
Prove that this energy E* coincides with the energy E (for fixed angular
momentum M).
Hence, if x(t) E R3 is the trajectory of a point moving under a force F and
r(t) is its distance to 0 E K3, then
r
V2E -
IIMII2 - 2U(r) = r', i.e. -dt = t
J
11 (Classical Moment Map). Consider K3 with the defining representation
of SO(3, K).
a) Prove that this representation is equivalent to the adjoint representation
of SO(3, K) on its Lie algebra so(3, K), if R3 and so(3, K) are identified
via the map
vl 0 -V3 v::
R3 so(3, f8), v = v2 u.- v = v3 0 -V1
V3 -V2 vi 0
Moreover, the adjoint and the coadjoint representation of SO(3, K) are
also equivalent.
b) Prove that this identification satisfies the equation
v(w) = v x w, [v, w] = [v, w], (v, w) Ztr (vv-) .
Exercises 269
c) By a), the moment map' : T'R3 - so(3,R)* can be interpreted as a

map from T*R3 to R3. Show that it can then be written in the form
W(9,P) = 9XP,
and hence it coincides with the classical angular momentum.
Chapter 8
Elements of Statistical
Mechanics and
Thermodynamics
8.1. Statistical States of a Hamiltonian System

The Hamiltonian formulation of mechanics starts from a configuration space
X"' and makes use of the phase space T'X' with its canonical symplectic
structure. A state of the mechanical system under consideration is a point in
the phase space T`X'", and the motions of states are the integral curves of
the symplectic gradient s-grad(H) of a Hamilton function H : T* X' --+ R.
In this formulation of mechanics, the only essential data to be given are a
symplectic manifold M2rn and a function H. The point of view of statistical
mechanics is based on the idea that, for instance because of the size of the
mechanical system or as a consequence of the imprecision of measurements,
the state of the mechanical system cannot be determined precisely by fixing
2m real parameters. Instead, to each open set U C M2"', we can only
ascribe the probability p(U) that the state belongs to the set U. This leads
to the concept of considering mechanical states no longer as points in the
phase space M2, but as probability measures on M2'".
Definition 1. Let a Hamiltonian system (M2"', w, H) consisting of a sym-

plectic manifold and a Hamilton function be given. A statistical state is a
probability measure p defined on the a-algebra of all Borel sets on M2'".
271
272 8. Elements of Statistical Mechanics and Thermodynamics
Example 1. For each point x E M2n', consider the measure 62 concentrated

at this point,
1 0 ifx¢U,
a:(U) l 1 if a E U.
Therefore, classical mechanical states are particular statistical states.
Let -tt M2" -+ M2in be the flow of the symplectic gradient s-grad(H).
:
The motion of the classical state x E MZ"' is the trajectory 0t(x). For the
probability measure b+,(s) corresponding to the point it(x), we have
60&)(U) = 1 0, xE
1 , x E
t (U) 1 = bx(-tt I(u)),
t 1(U )
and this formula leads to the following definition.

Definition 2. Let a Hamiltonian system (M2-, w, H) and a statistical state
p be given. The motion of µ under the impact of the Hamiltonian system
is the curve At of measures
At (U) := µ(4 1(U))

Definition 3. An equilibrium state of a Hamiltonian system (M2m,W, H)
is a state A which does not change in the course of the motion of s-grad(H),
At = P.
Definition 4. A statistical state p of a Hamiltonian system (M2in,W, H)
has a stationary terminal distribution if the equation
'U-(U) := lim
t 00 MO)
defines a Borel measure on M2"'.
Theorem 1. A stationary terminal distribution µ«, is always an equilibrium
state.
Consider the volume form

(-1)m(m-1U2
dM2m _ Wm
m!
of the symplectic manifold and the Borel measure induced from it (see the
final remark at the end of §3.5). If the measure p is absolutely continuous
with respect to the volume measure,
µ(U) := je(x).dM2m(x),
then a is called the density function of the state µ.
8.1. Statistical States of a Hamiltonian System 273
Theorem 2 (Liouville's Equation). Let p = be a state with density

function, and let at be the motion of this state in a Hamiltonian system.
Then the measures pt = pt dM2" are also states with density function, and
d
dtet =
-{H,e}o$_t.
Proof. The flow ibt consists of symplectic transformations and preserves, in

particular, the symplectic volume form dM2in, 4 (dM2i`) = dM2m. Trans-
forming the integral, we see that pt = P o &_t is the density function of the
state µt. This implies
dt of = dt
P o D_t = do o dt -t = - o -t = -{H, p} o O_t.
0
Corollary 1. A statistical state µ with density function P is an equilibrium
state if and only if a is a first integral of the Hamiltonian H, (H, g} = 0.
If x E M2m is a classical state in Hamiltonian mechanics, the value H(x) of
the Hamilton function H is the energy E of the state,
E(x) = H(x) = JM2rn
The definition of the energy of a statistical state generalizes this relation:
Definition 5. The energy of a statistical state p is the integral
E(µ) = H(x) dp(x),
JMom
if this integral exists.
Theorem 3 (Conservation of Energy in Statistical Mechanics). If u t is the
motion of a statistical state in a Hamiltonian system, the energy E(pt) is
constant.
Proof. After transformation of/ the integral, we immediately obtain
E(pt) = l H o I 1(x) dtt(x),

J M2m
and differentiating this with respect to the parameter t yields the formula
d E(µt) = fM2m {H, H}(x) du(x) = 0. 0

Now we want to study how the probability 1At(N2m) that the state µ =
B dM2ri is in the compact subset N2ni C M21 at time t changes in time.
The following lemma serves as a preparation.
Lemma 1. Let (M2i', w) be a symplectic manifold. Then, for any two

functions f, g : lbl2in --+ R, the following formula holds:
df AdgAw,,,1 = 1 If,
M
Proof. To prove this formula, we choose on 1112»' local syinplectic coordi-

nates (ql, ..,P.),
w= dp0 A dqq
,r= I
Setting Ai := dpi Adgi, we see that the exterior product satisfies the relations
AiAAj = AjAAi and AiAAi = 0.
Since the forms Ai commute with one another, we can use the binomial
formula to compute the exterior power wk,
m > 1
m.
wn, _ (A1)' Aa' A A A"
Because Ai A Ai = 0, only summands with rri < 1 occur in this sum, and
hence it reduces to a single term:
writ = in!.AlA...AAna = m!.(dpiAdgi)A...A(dpmAdq»,)
We compute WI-1 in a similar way, and obtain
",
(m-1)! AlA...A,;...A
Thus, the following exterior products vanish:

(1) dpi Adpj Aw"'-1 = 0 = deli Adgj Awl"-
(2) dpi A dqj A w' = 0 if i # j

and we obtain
fit
df A dg Awrr-1 = (.!m- dg dp, A dqi + 2f ag dq, Adpi) w"'
api Oqi OCli dpi
rn
('m - 1 ) !
Of yg Of 09 A I A ... A A,,,
pi Tqi- di CdR
{f,g} wr
arr.-1 .
mi
The relations m. d(f dg A w'"-1) = in df A dg A w"'- I = {f, gj w"' together

with Stokes' theorem imply, after integration.
Theorem 4. Let (A12mw) be a compact symplectic manifold with boundary.

Then
{f,g} w"r f.dg nwm-1
J12in i)At 2° .
In particular, the integral
J r2" {f,9} w"` = 0

vanishes for any compact symplectic manifold without boundary.
Corollary 2. If the Poisson bracket if, g} of two functions defined on a
compact symplectic manifold 1112" without boundary does not change sign,
then it vanishes identically, If, g} = 0.
We will apply these formulas to a statistical state with density function e.
Definition 6. The (2m. - 1)-form
At) (m-1)!
A'dHAw"'-1
is called the probability current of the statistical state it = e dM2"' in the

symplectic manifold (Al2m,w) with Hamilton function H.
Theorem 5. Let lit be the motion of the statistical state p = e dA12"' in
the Hamiltonian system (1112"', W, H), and let N2", C AI2si be a compact
submanifold. Then
(µr(N2-)) (r=o
Wt = JNFIJ
Proof. We compute the derivative with respect to time at t = 0 and use
Lionville's equation as well as the preceding integral formulas:
d N2",
o- _J dt r=e dM2", {He} dM2"rJ
N2m N2s'
(m - 1)! J
UN2"4
e dH n w"'-i =
aN2"
f j(Ic) .
The probability current j(µ) is a hypersurface measure on all (2m - 1)-

dimensional submanifolds of the symplectic manifold M21. It expresses
the infinitesimal change of the probability for the state it to be in the set
N2i' C A12"' as an integral over the boundary of N2'".
Example 2. If the subinanifold N2i" C M2" is described by inequalities of
the Hamilton function,
N2n = {x E M2" : C1 < H(x) < C21,
then the differential dH vanishes on the tangent bundle T(8N2m) of the

boundary, and so does the form j(p). We hence obtain, for the motion At
of every statistical state p = p dM2,,
dtpt({x E M2in : C1 < H(x) < C2}) = 0.

Thus the probability for a state p to be in N2in at time t is constant.
Example 3. In case m = 1, the formulas for the probability current of a
two-dimensional Hamiltonian system (M2, w, H) simplify:
j(p) = e dH and dpt(N2)

t = JaN2 p dH.
Now we turn to the notion of information entropy for a statistical state.
As a motivation, we first recall the notion of information of an event in-
troduced by C. E. Shannon (1948). The heuristics is the following: If, in a
series of experiments, an event occurs with probability close to one, then the
"information" contents of this event is small. Conversely, if the probability
of this event is close to zero, the occurrence of this event contains a large
amount of "information". Modeling the probability computation by a triple
(i, 21, p) consisting of a set i, a or-algebra 21 of subsets of S2, and a measure
p defined on 21 such that A(fl) = 1, we arrive at the following definition for
the Shannon information.
Definition 7. Let (S2, 21,,u) be a probability space. The amount of infor-
mation contained in an event A E 21 is
1(A) := - log(p(A)).
For a finite set S2, one forms the mean of these information amounts and
gets in this way the so-called information entropy.
Definition 8. Let (12, p) be a finite probability space. The information
entropy of this space is
S(Q, p) f t
1(w) . dp(w) = - E pi
i=1
log(pi) ,
where the set 11 consists of the elements {w1, ...,wn}, and pi := p({wi}) is
the probability for the event wi.
The information entropy is a measure for the uncertainty of the probability

space. In fact, if St = {wl, ... , w } consists of n points and we denote by p'
the measure corresponding to the equidistribution, p'(wi) := 1/n, then the
following holds.
Theorem 6. The information entropy of a finite probability space (S),µ)

does not exceed the information entropy of the equidistribution:
S(Q, Ft) <_ S(n, µ')
Equality occurs if and only if u = µ' is the equidistribution.
Proof. The proof is based on the fact that the function f (x) := x log(x)
is convex on the half-line (0, oo), since
f'(x) = 1 + log(x) and f"(x) = x > 0.
The convexity of f then implies the inequality

n f (pi) n
f pi <_
n = n E pi log(pi )
i=1 i=1 i=1
n
Using the relation pi = 1, one immediately deduces from it that
i=1
S(1, lt) > pi log(pi) < -log (n) = S(S2,

i=1
The formula S(12,,u) = - E pi log(pi) leads to the notion of information

entropy for a statistical state with density function.
Definition 9. Let (M2,, w, H) be a Hamiltonian system, and let a = e
dM2m be a statistical state with density function. We define its information
entropy as the integral
S(u) - JM2m P log(P) dM2m,
if this integral exists.

Remark. This notion of entropy is not without problems in statistical
physics. The following result concerning the conservation of entropy claims
that the information entropy does not change in the course of the motion of
a state. This is a good assumption for reversible processes (and processes
which can be approximated by reversible ones), but cannot be applied to
the vast range of irreversible processes, in which the entropy increases. For
this reason, we will apply the notion of information entropy S(µ) only to
equilibrium states.
Theorem 7 (Conservation of Information Entropy). The information en-
tropy S(pt) of a statistical state with density function is constant under the
motion of a Hamiltonian system.
Proof. We differentiate the function S(Pt) with respect to the parameter t

and use Liouville's equation:
(d-tL-log(owL dM2-
dt S(At) JM2m
_ ({HQ}l(Q)+{HQ}) dM2-.
M2m
But since {H, log(e)} = {H, a}/e, we have
{H,e}+{H,g}.log(e),
and this yields
log(e)} dM2m
dt S(µt)It_o =
JM2m{H, e
(-1)m(m-1)/2
m J
M
(-1)m(m-1)/2
dA/ d(e log(e)) A w"`-1
( m-1 ) JMs"`
Since M2m has no boundary, Stokes' theorem immediately implies the as-
sertion.
Remark. In the case of a manifold M2"' with boundary we obtain
d (-1)m(m-1)/2
S({pt)It e log(e) dH n
dt ( IOM2.
w, H) has several special equilibrium states.

These are characterized as realizing the maximum of information entropy in
a particular class of statistical states. We will discuss two of these states,
the so-called Gibbs state (or the canonical ensemble) and the microcanoni-
cal ensemble. We will, in general, make the following three assumptions for
the Hamiltonian system (M2m, w, H):
Assumption 1. The Hamilton function H is non-negative,

H(x) > 0 for all x E M2s` .
Assumption 2. For every positive number B > 0, the following integral
exists:
Z(B) := r e-N(x)/B. dM2m(x)
JM
The function Z(B) is called the partition function of the Hamiltonian sys-
tem.
Assumption 3. For every positive number B > 0, the following integral

exists:
H(x) . e-H(=)lB . dM2m(W ) .
bl2m
Remark. The parameter B will later be identified with absolute temper-
ature (multiplied by the Boltzmann constant k). Its appearance here indi-
cates the exceptional role that temperature plays among all thermodynamic
parameters.
The function
E(B) := I H(z)e-H(:)/B , dA12-(z)
Z(B) M2m
is called the inner energy of the Hamiltonian system (M2m, w, H); its inner
entropy is
E(B)
S(B) := log(Z(B)) +
B
Finally, the free energy F(B) is defined by
F(B) := -B log(Z(B)) = E(B) - B - S(B).
First we note that the inner energy E(B) is a non-decreasing function on
the interval (0, oo). Its derivative is
dE _ 1
TB Z2(B) B [(I
M2m
e- dM2m(x)) (ML H2(
z)e-dM2m(x))
_ l2
- (ML H(x )e dM2m(x))

and the Cauchy-Schwarz inequality shows that this derivative is positive
(H j4 const). Denote by En and Em., respectively, the bounds of the
range of the inner energy E : (0, oo) - R. We calculate the derivative of
the partition function in a similar way:
dZ 1
BZ Z(B) E(B).
dB = B2 JM2m
Hence the inner energy as well as the inner entropy can be expressed in
terms of the partition function. We summarize these formulas.
Theorem 8 (Simple Equation of State for a Hamiltonian System).
(1) E(B) = B2gR (log(Z(B)));
(2) S(B) = W(B log(Z(B)));

(3) in the sense of 1 forms on the one-dimensional manifold (0, oo),
dS = B dE.
Proof. The third equation is a straightforward consequence of the first two.
a
The Gibbs state (canonical ensemble) is distinguished by the property that
it realizes the maximum of the information entropy among all stag of fixed
energy.
Theorem 9 (Gibbs State, Canonical Ensemble). Let an energy value Eo
between the minimum and the maximum of the energy function E(B) for
the Hamiltonian system (M2m, w, H) be given. Among all statistical states
p = P dM2' with density function of energy E(p) = Eo, there exists
precisely one state, PGibbs = PGibbs dM2m, of maximal information entropy
S(pcibb.). The density function of this state is
1
PGibbs(x) = Z(Bo)e -H(x)/Bo ,
where the parameter Bo is determined as a solution of the equation E(Bo) _

Eo. The value of the maximal entropy S(pCibbs) is
S(pGibbs) = S(Bo) = log(Z(Bo)) + E(BO) o)
Proof. Choose Bo as a solution of the equation E(Bo) = E0, and )et
pGibbs = 1 e-H(x)/Bo dM2m

Z(Bo)
be the corresponding Gibbs state. Its inner energy is
E(pcibb.) = Z(B0) H(x)e-H(-)/Bo

. dM2m = E(Bo) = Eo,
JM2m
and hence UGibbe is a statistical state of energy Eo. For every oti er state
p = P dM2m with energy
E(IL) = fM2m H(x)P(x) dM2'n = E0,

we consider the function
f (t) = S((i - t)pGibbs + tp) ,
the information entropy of the statistical state (1 - t)pGibbs + tp. Differen-

tiating f (t) with respect to the parameter t, we obtain
d2f(t) (e - PGib.)2 dM2' <
dt2
- _ JU2- (1 - t)PGib, + tP
0.
Moreover,
df
dt
=
f
y2" e - log(eGibbs) -
dM2m - S(PGibbs),
and calculating the integral using E(p) = Eo leads to

log(Z(Bo))) dM
J M2"'
e log(gGibbB) - dM2"' =
Jim g( go -
_ -log(Z(Bo)) - BoEo = -S(Bo)
We conclude that the derivative of f vanishes at t = 0. Hence the derivative
is non-positive, and f (t) is decreasing in [0, 1]. Since f (0) > f (1), we obtain
S(P) < S(lpGibbs) -
If S(µ) = S(PGibb.), the first derivative of f (t) vanishes identically. Hence

the second derivative of this function is also zero, and from the formula
stated before we conclude that e = eGibbs The assertions of the theorem
are proved. 0
Example 4 (Maxwell and Boltzmann Distribution). In classical statistics,
the Hamilton function of a particle with potential energy depending only on
the coordinates q can be written as
2 2 2
P1
H(q, p) = 2m + U(q) = +2m+ 3 + U(q).
Thus, the partition function is equal to
Z(B) = I exp [_i? ] dpiAdp2ndp3 f exp I

r
L
qJ1
which, using the integral fR a-QS2dx = r/a, can also be written as

Z(B) = (27rmB)3/2 - ZZ(B) ;
Zq(B) denotes the value of the integral in which the unknown potential
energy occurs. In the same way, the Gibbs state factors into the probability
measure Pp of momentum values and that of coordinate values Pq,
(27rmB)3/2e-p2/2mBdP Zqe-U/Bdq
PGibbs = lip 'A q =
Going from the momenta to the velocities, we arrive at the so-called Maxwell
distribution,
3/2exp [m(vl +v3)] dv1dv2dv3,
Pp
2M ) 2B
which again is the product of probability distributions for every single ve-
locity component. The Boltzmann ensemble is defined to be the probability
282 8. Elements of Statistical Mechanics and Thermoc.ynalnics
measure p. and is proportional to the particle counting measure v. For

example, for the potential U(x, y, z) = ingz of a. homogeneous gravitational
field which is parallel to the z-axis, this yields the barometric formula
V = 1/0 t,-inyz/'idx. d y dz.
The second natural equilibrium distribution of a Hamiltonian systc n starts

from a subset A C A12ni of the symplectic manifold which is invariant under
the flow 4Dt of the Hamilton function H and has positive finite volme,
4bt(A) C A, 0 < vol(A) < oc.
Consider the density function
j 1/vol(A) if x E A,
`°A (x) = l 0 if .r E A,
as well as the statistical state ILA = eA 012m. This is an equilibrium state,
since A is invariant under the flow 4bt. Its information entropy is
S(,A) = - f ll,PA(x) - log(PAW) d12i'(x) = log(vol(A)).
Theorem 10 (Microcanonical Ensemble). Let A C M2"' be an invariant

subset for the flow of the Hamiltonian system (A12", w, H) with positive
finite measure. If it = p dM2"' is a statistical state whose density function
has support in A, supp(p) C A. then its information entropy is always
bounded from above by the information entropy of p.A,
S(Ez) < S(tA) .
Equality only occurs for it = AA.
Hence the microcanonical ensemble IAA realizes the maximum of the infor-
mation entropy among all statistical states concentrated in the set A.
Proof. The proof proceeds in complete analogy to that of the preceding

theorem. Again, we consider the function f(t) = S((1 - t)IA 4 tit) and
calculate its second derivative:
d2f (e - UA)2
. dA12,n < 0.
dt2 = -(1-t)eA+p
fA
The derivative of f at t = 0 vanishes:
df I
dt t_tt
=- f
A
(e- -log(eA) f
A
0,
since It is a probability measure with support in A. 0

8.2. Thermodynamical Systems in Equilibrium 283
8.2. Thermodynamical Systems in Equilibrium

First we generalize the simple equations of state of a Hamiltonian system
by keeping a fixed symplectic manifold (M2",, w) but considering a family of
Hamilton functions H(x, y) depending on an additional parameter y E N'.
Let the parameter space also be a smooth manifold, and suppose that the
Hamilton function H(x, yo) satisfies the three assumptions of the preceding
section for any fixed value of the parameter yo E NT. Then the partition
function Z(B, y), the inner energy E(B, y), the inner entropy S(B, y), and
the free energy F(B, y) become functions on the manifold (0, oo) x Nk,
Z,E,S,F: (0,oo) x Nk -4 R,
and, of course, the equation of state B dBS = dBE holds for each fixed
value of the parameter yo E The general equation of state of a Hainil-
tonian system depending on parameters is a relation between the complete
differentials considered as 1-forms on (0, oo) x N".
Theorem 11 (General Equation of State for Parameter-Dependent Hamil-
tonian Systems). In the sense of an equality for I-forms defined on the
manifold (0, oo) x N", one has
dE =
B
Proof. By the definition of inner energy, we have
B dS + S dB = log(Z) dB + B d(log(Z)) + dE,

and again inserting S, yields
dE.
Remark. If we allow the until now independent variable B to be a function
B : N" -> (0, oo) on the parameter space N", we may consider all the
functions Z, E, S and F as functions on N' by means of the map
N' (0, oo) x N", y -' (B(y), y).
The exterior derivative commutes with the pullback of forms and thus im-
plies on the manifold N" the equation
dE = B dS + E dB - B d(log(Z)).
The macroscopic description of thermodynamic systems starts from a few

assumptions. First, all thermodynamic systems should be in mechanical
equilibrium. Furthermore, it is supposed that no particle exchange takes
place during the change of states. Lastly, the macroscopic state of the ther-
modynamic system is to be completely described by a parameter y E Nk,
which is an element of a finite-dimensional manifold. If this holds, the
following phenomenologically observed laws are historically known as the
fundamental theorems of thermodynamics. In a modern account of theoret-
ical thermodynamics, they are taken as axioms:
Zeroth Fundamental Theorem of Thermodynamics. There exists a

positive function T : N'' - IR, called the temperature.
First Fundamental Theorem of Thermodynamics. There exist two

(not necessarily exact) 1-forms dA and dQ on the manifold Nr such that
their sum
dA + dQ = dE
is an exact form. The function E is called the energy of the thermodynamic
system, dA is called the work form, and dQ is called the heat form.
Second Fundamental Theorem of Thermodynamics. The 1 -form
dQ = dS
T
is an exact form, and the resulting function S is called the inner entropy of
the thermodynamic system.
At this level of abstraction, a thermodynamic system in its macroscopic

description is a 4-tuple (NT, T, dA, dQ) consisting of a parameter manifold
N'', a function T : N' -+ R, as well as two 1-forms dA and dQ such that
the 1-forms
dA + dQ and dQ
T
are exact. Of course, the following relation holds:
dE = dA + T dS ,
and this is Gibbs' fundamental equation of thermodynamics.
Now, with each parameter-dependent Hamiltonian system a thermodynamic

system can be associated in a canonical way. In fact, let a symplectic mani-
fold (M2n', w) and an additional manifold N'' be given. Fix two non-negative
functions,
B : N' (0, oo) and H : M2m x N' --10, oo) ,
and consider H as a family of Hamilton functions.on the symplectic manifold.
Using the Gibbs state,
e-H(z.b)lB(b) . dM21n(x)
=
pGibt)s
we define the function T : N' --+ R1 and the 1-forms dA and dQ on N' as
follows:
(1) The temperature of the thermodynamic system is the function B
divided by the Boltzmann constant k (= 1.380.10-23W/K):
T(y) := B(y)
k
(2) The energy of the thermodynamic system is the inner energy of the
Gibbs state uGG ib)
bbs'
H(x, y)e-H(x,b)/B(b)
E(y) = E(B(y)) = . dM2"'(x) .
Z(B(y)) M2'"
(3) The entropy of the thermodynamic system is the inner entropy of

the Gibbs stateuB(b)
Gibbs*
S(y) = k S(iz Y) = k - log(Z(B(y))) + k E(B(y))

B(y)
(4) The heat form dQ is the product of temperature with the differen-
tial of entropy,
dQ :=
(5) The work form dA is
E
dA = dB - B d(log(Z)).
The First Fundamental Theorem of Thermodynamics,
dA + dQ = dE,
is, in the model just described, an automatic consequence of the general
equation of state for parameter-dependent Hamiltonian systems. Hence, by
means of the Gibbs states, a thermodynamic system is associated with every
such Hamiltonian system. Let V be a vector field defined on the manifold
N. Then
dA(V) = B Z
and, since
V(Z) = J V(-H/B)e-H/BdM2m = - JV(H)e/I5dM2m+Y-E.Z,
,%12m M2m
we obtain an explicit formula for the work form,
dA(V) = Jwzm V(H) - e-B dM2m .
Z
The pressure pi, of the thermodynamic system generated by the vector field
V on the parameter space Nr is the function
PV (Y) - I
Z(B(y)) M2"
V(H(x, y)) e H(=.v)/B(y) . dM2-(x) .
Thus the relation dA(V) = - pv holds by definition, and the work form
can be represented with respect to a frame V1, ... , Vr of vector fields of the
manifold Nr in the form
r
dA = - PV. Qi,
i=1
where a,, ... , ar is the dual frame of 1-forms on Nr.
Example 5. The pressure p as used in thermodynamics is obtained if the
parameter manifold N' is one-dimensional and describes only the volume
V, and V = O/8V is chosen as the vector field. It is known that then the
only contribution to the form dA comes from the mechanical work of volume
change,
dA = -p - dV.
Considering in the continuum limit for a large number of particles their num-
ber as a continuous variable, we see that the amounts of matter nl, ..., nr
of a system consisting of r different chemical components can be chosen as
parameters for the manifold Nr. In this case, the chemical work is tradi-
tionally expressed by the so-called chemical potentials µl, ... , µr (which the
reader should not confuse with statistical states):
r
dA = pi dni.
i=1
The chemical potential µi of the i-th matter component is the negative

of the pressure associated with the vector field O/Oni. Similar formulas
express the electromagnetic work in the presence of magnetic or electric
dipol-momentum densities.
Consider, in particular, the case that there exists a vector field V on the
parameter space Nr such that
V(T) - 1 and V(H) - 0.
All the thermodynamic quantities are then computable starting from the
partition function
Z e-H/kTdM2m
M2m
or the free energy derived from it,
F = -k T log(Z) .
Theorem 12. Suppose that there exists a vector field V on the parameter
space Nr such that V(T) = 1 and V(H) = 0. Then:
(1) S = -V(F) = -dF(V) = -V i dF.
(2) E = F-T-V(F) =
(3) dQ = - T Lv(dF).
(4) dA =
(5) The pressure generated by an arbitrary vector field W is
pw = - dF(W) + dF(V) dT(W) .
In particular, the vector field V generates no pressure, pv - 0.
Proof. Differentiating the partition function Z with respect to the vector
field V, we obtain
He-H/AT . dM2m =
E-Z
V(Z) k T2 M2m J k.T2
This implies the following formula for the derivative of the free energy,
V(F) = -k log(Z) - k T T2 = -k (log(Z) + T) = - S,

k k
and the first of the assertions is proved. The relation
S = -V(F) = -(V J dF)
implies
dS = -d(V J dF) = -,Cv(dF).
The expression for the heat form is calculated from its definition,
dQ = - T Lv(dF) .
The energy E is determined by the free energy F as well as the entropy,
E= F-T(VJdF).
Lastly, we calculate the work form:
dA = dE - T dS = d(F - T(V J dF)) + TLv(dF)
= dF - (V J dF)dT.
Example 6 (Ideal Gas). We view an ideal gas as n points of equal mass

M in a volume region V C 1R3 which do not interact with one another.
The configuration space of the mechanical system consists of the product
of n copies of the set V factored by the action of the permutation group
consisting of n! elements. If we want to avoid this group identification, we
can simply consider the phase space M := T'V x ... x T'V equipped with
a measure which already takes into account this identification:
3n
dM :=
i=1
The Hamilton function of the mechanical system is

i
X(91,Q2,43).
2111
Here X denotes the characteristic function of the set V C R3. Calculating

the partition function of the thermodynamic system depending only on the
two parameters T and V leads to the formula
3n
Z(T V) = je_hh1TdcAdp = 1J el2MkTdp/I1
n
m ` oo
3n/2
;j (27rkTM) .
We apply Stirling's formula log(n!) n (log(n) - 1) and calculate all the

thermodynamic quantities for the ideal gas using this approximation. The
free energy is
F = - k T log(Z) nkT (log(V/n) + 1 + 3 log(2akTM)) .

We obtain the entropy and the energy from the free energy according to
Theorem 12, since the Hamilton function does not depend on the tempera-
ture:
OF 3
S=- = nk (log(V/n) + log(T) + coast) ,
2
OF 3n
E= O
The pressure p of the ideal gas corresponds to that of the vector field de-
scribed by the volume, and is determined by
OF n
P=- 8V =
Here n/V is the particle density of the ideal gas.
Example 7 (Solid Body). A solid body consists of n points oscillating

around a point. If we take this point to be the origin, the Hamilton function
is
n
H
- ` \(p3i-2 +i-1
i=1
2Mi
+ 3i + K2 3i-2
2 + q3i-1
2
2
+ q3i
2
Introducing the basic frequencies

Ki
v
2,r Mi'
we obtain as the value of the partition function
Z (k T) 3n 1
v3 .... yn
The free energy is
F = - k T log(Z) kT (n log(n/e) + 3n log(kT) + const) .
Starting from this, one can calculate the other thermodynamic quantities,
for example the energy:
E = F-T 5T _ - T2 5T 49
(7-F,) =
This formula says that every degree of freedom of an atom carries the mean
inner energy k T, a result known as the Dulong-Petit rule. However, at
temperatures close to absolute zero, this rule no longer holds, since the
model for a solid body as being a combination of harmonic oscillators is not
adequate anymore.
The inner and the free energy are examples of so-called thermodynamic
potentials. Classical thermodynamics usually adopts the point of view that
the choice of a particular potential already determines the quantities by
which the system is to be described. In this sense, the inner energy E is
understood to be a function depending upon entropy and volume, the free
energy F is a function of temperature and volume, etc. Consider a Hamilton
function not depending on temperature. As in the two preceding examples,
the free and the inner energy are related to the vector field V dual to the
temperature via the relation (Theorem 12)
E= O
Hence the inner energy E appears to be the Legendre transformation with

respect to temperature of the free energy chosen as Lagrange function, L :_
-F (see §7.5). By means of Legendre transformations with respect to other
parameters, one obtains the other thermodynamic potentials, enthalpy and
free enthalpy, which we will not discuss in detail here. In the case that only
the mechanical work of the volume change contributes to the work form, the
First Fundamental Theorem reads as follows:
dE =
OE OE
T = and =
aS p -VV
If the energy is even twice continuously differentiable, the interchangeability
of second partial derivatives is equivalent to a so-called thermodynamic or
Maxwell relation
02E OT Op
avas = aV = OS
Beginning with the free energy,
dF =
we similarly obtain the equations
S , p= -aV,
as well as the thermodynamic relation
aS ap
aV aT
Eventually, we want to discuss the Carnot cycle briefly, which will lead us
to the notion of universal thermal efficiency. Historically this was essential
for the understanding of the First and Second Fundamental Theorems of
thermodynamics.
We use an arbitrary homogeneous substance and suppose that its state is

determined exclusively by the two mechanical variables, the pressure p and
the volume V, from which the temperature T can be computed by means
of a general equation of state. A Carnot cycle is understood to be a closed
path ry in the (V, p)-plane consisting of two isotherms and two adiabatics.
More precisely:
(1) From A to B, the substance is expanded up to the volume VB
through contact with a heat reservoir of temperature Ti while keep-
ing the temperature fixed ("isothermally"). During this ?rocess it
absorbs the quantity of heat Q1.
(2) The substance is taken out of the bath Tr at B and expanded to
the point C without absorbing or emitting heat ("adiabatically").
The temperature drops to the value T2.
(3) The material is put into a heat reservoir of temperature T2 and
compressed isothermally from C to D, emitting the quantity of
heat Q2.
(4) The material is taken out of the second heat reservoir and adia-
batically compressed from D to A to reach the temperature T1
again.
It has to be stressed that we are supposing that the idealized processes
discussed throughout this chapter are reversible. In the way explained here,
the Carnot cycle describes a power engine (the temperature difference of
the heat reservoirs is used to do work); reversing the process, it provides
a thermal engine (work is done to further cool down the colder of the two
heat reservoirs). By assumption, work is done exclusively in its mechanical
form; the total work A is computed as
A := jdA = -7 ipdV,
and, since -d(pdV) = dVAdp, this is equal to the area of the region bounded
by the curve y in the (V,p)-diagram by Stokes' theorem. We integrate the
First Fundamental Theorem
dE = dA + dQ
over the closed path -y. The line integral of the exact form dE along 7
vanishes, since the inner energy again reaches the initial value at A:
0= jdE
ry
_ 5dA+ ¢ dQ.
'T y
The thermal efficiency q is defined as the ratio of the work done and the
added heat, q := A/Q1. But the line integral along dQ is precisely Q2 - Qi;
hence we can write the thermal efficiency also as
q=1-Q1
As the Carnot cycle was supposed to be reversible, the entropy does not
change, and the Second Fundamental Theorem implies the identity
Q2 Qi
Jry T T2 T1
The thermal efficiency only depends on the temperatures of both heat reser-
voirs.
T1 - T2
n =
T2
and not on the particular kind of substance or the particular construction
of the power engine.
Exercises
1 (Gay-Lussac Experiment, 1807). Consider a fixed amount of an ideal gas

which at time 1 has pressure pl and volume V1. The gas is heated while
keeping the volume fixed until it reaches at time 2 the pressure p2. Then
the gas is allowed to expand adiabatically up to the volume V3 ("overflow").
At the point 3. the gas reaches again the initial pressure pl. Prove that the
entropy change between the states 2 and 3 is equal to
AS = S3-S2 = n-k-log
V2
independently of the way in which the physical process is realized in practice.
2. Derive from the Maxwell distribution the probability distribution for the
speed lit' I I = (vi + v2 + V32)112 using spherical coordinates.
3. Consider two containers filled with the same ideal gas having equal tem-
perature To and particle number N, but different volumes V1 and V2. The
containers are then connected.
a) Calculate the equilibrium temperature of the connected containers from
the condition that the entropy remains constant during the connection
process.
b) Explain why the work done is equal to the difference of the energies
before and after the connection, and calculate the latter.
Exercises 293
4. Determine the quantity of heat Q absorbed by an ideal gas in the course

of the Carnot cycle. Hint: One of the intermediate results is
Q = (T2-T1)(S2-S1)
5 (Van der W4aals Equation). The free energy of a real gas is heuristically
defined as the sum of the free energy of the ideal gas F;d and an additional
term depending on two constants a and b to be determined experimentally:
F = F;d - nkT log ( V

V - nb n2a\
F2
)
a) Explain why this Ansatz tends for large volumes to the free energy of
an ideal gas, and why, on the other hand, it prevents the gas from being
compressed unboundedly.
b) Derive the equation of state (the so-called van der Waals equation)
n2a
(p + (V - nb) = nkT
V
as well as the formulas for entropy
) and energy. How can the result for
the energy be interpreted physically?
6. The Hamilton function of an ideal gas in the ultra-relativistiiyc limit is
(c ist the velocity of light). Calculate the free energy F, the entropy S, and
the energy E.
7 (Planck's Radiation Law). An ideal photon gas in thermal equilibrium

emits black-body radiation, which is described by Planck's law. We describe
each photon by a quantum mechanical oscillator and assume as a known
fact that the energy levels of an oscillator with basic frequency w are
En = hw(n + 1/2). n E N .
For photons, it may be supposed that these are distributed within each
energy level (up to irrelevant factors of proportionality) according to the
Gibbs distribution Qn = e-E^1kT
a) Calculate the mean energy,
x x
On - En On
n=1 n=1
and conclude that the number n = [eF 1kT - 11-1 can be interpreted as
the mean occupation number.
b) The frequency w is related to the momentum of the photons by E =

clIpIl = hw. Quantum statistics postulates that in the Maxwell distri-
bution j 4p, the Gibbs factor e-`tIPII/kT is to be replaced by the mean
QA[
occupation number n. Conclude that the probability measure po of
I IpI I, up to some factors, is
w2
uPQaf - et-1k]
the spectral energy density c is defined as energy per volume unit, hence
it is obtained after again multiplying by hw:
hw 3
e= efiW/kT _ 1
This is Planck's radiation law (Planck, 1900).
c) Discuss the behavior of the curve c depending on the variable w and the
parameter T; describe, in particular, how the position of the maximum
changes with T.
d) Discuss the limiting cases hw << kT (Rayleigh-Jeans law) and hw >> kT
(Wien's law).
Chapter 9
Elements of
Electrodynamics
9.1. The Maxwell Equations

The Maxwell equations describe the impact of an electromagnetic field on a
distribution of electrical charges in space as well as the interaction between
the electric field E and the magnetic field B. These vector fields are time-
dependent vector fields defined on a domain 11 C R3,
E: 1lxR---+R3, B: IlxR-.R3,
and the electric charge is described by a time-dependent density function
e: QxR -*R.
The electromagnetic field induces a current of the electric charges, and hence
also a time-dependent current density vector
J:S2xR-bR3.
The operations of divergence and curl are supposed to be applied to the
spatial coordinates of the time-dependent vector fields.
First Group of the Maxwell Equations:

curl(E)
-
Second Group of the Maxwell- Equations:
, div(B) = 0.
4c
curl(B) = c + J, div(E) = 47rg.
- 295
296 9. Elements of Electrodynamics
Here c denotes the constant velocity of light. Since div(curl(B)) = 0 (see

§2.3, Theorem 5), the continuity equation
div(J) + = 0
is an immediate consequence of the second group of the Maxwell equations.
Now we want to write the Maxwell equations involving differential forms on

R3. For this, we use the transition from a vector field V to the associated
1-form wv discussed in Chapters 2 and 3. On a manifold Mm, wv is the
1-form defined by the equation (see §§2.3 and 3.4)
*wv = V J dMm,
and the gradient and divergence satisfy the identities
Wgrad(f) = df, d(*wv) = div(V) - dAIm.
In JR3, there is an additional relation involving the curl of a vector field V
(§3.11):
wcurl(V) = *dwv.
Passing now from the time-dependent vector fields E and B to the corre-
sponding time-dependent 1-forms WE and WE, we can restate the Maxwell
equations in the following equivalent form:
First Group of the Maxwell Equations:

1 (9
*dwE _ (we). d(*wB) = 0.
.
c 8t
Second Group of the Maxwell Equations:
le 47r
*dWB = - Cat (WE) + - wJ, d(*wE) = 41rgo - dR3.
In these equations, the exterior derivative exclusively refers to the spatial

coordinates, and not to the time coordinate. By applying Poincare's lem-
ma, we can represent the closed 2-form *wB locally as the differential of a
time-dependent 1-form A,
*wB = dA.
Definition 1. The (locally) defined time-dependent 1-form A is called the
magnetic potential.
The first Maxwell equation can be rewritten equivalently as
d(wE + c (A)) = 0.
9.1. The Maxwell Equations 297
Again, Poincare's lemma shows the local existence of a time-dependent real-

valued function 0 with
WE
c at
A -do.
Definition 2. The (locally) defined time-dependent function ¢ is called the
electric potential.
The essence of the electromagnetic field is not the same in classical elec-
trodynamics and in quantum mechanics. In classical field theory, only the
measurable field strengths E and B are physically relevant; the potentials
A and ¢ are auxiliary mathematical functions which turn out to be useful.
In quantum mechanics, however, the electric and the magnetic potential ac-
quire a physical meaning on their own; we illustrate this briefly by discussing
the Aharonov-Bohm effect.
Remark (Aharonov-Bohm Effect). An infinitely extended coil S. through
which a current flows, induces a magnetic field which is almost homogeneous
in its interior and vanishes outside of the coil. Consider a closed curve -y
around the coil and denote by 0 the surface bounded by ry. Imagine that
an electron is prevented from moving inside the coil by an infinitely high
potential well. Then the remaining region ci - S of the plane is, from the
point of view of the electron, no longer simply connected. The vanishing of
the magnetic field B in this region implies that the 1-form A of the magnetic
potential is closed in 11 - S,
0 = *wB = dA.
but nevertheless the line integral is not path-independent, since, for example,
the curve segments ryl and rye of ry cannot be deformed into one another by a
homotopy (see Theorem 9, §2.5). Instead, by Stokes' theorem, the integral
1A= JA = fdA = f
is equal to the magnetic flux through the entire region Q. Quantum me-
chanics describes the states of the electron by solutions of the Schrodinger
equation called wave functions. Let V51 and v/2 be the wave functions of the
electron along -ti and rye, respectively, while the current is switched off (i.e..
B = 0 in S). The covariance principle for the Schrodinger equation with
respect to gauge transformations implies that, if the current is switched on,
the wave functions have to become
01 exp I AI , V2 = 02 exp [ , I AJ
L 7i rr
We put a screen at Si and form the linear superposition of the two wave
functions:
B
+Gi + 1Pz = 'Pt exp [Z 2 ] + 02) eXp A]
As we vary the enclosed magnetic flux 4'O, the absolute value IV)*, + t21
changes. This corresponds to a shift in the interference picture on the screen,
and provides experimental evidence for the intrinsic meaning of the potential
A.
A theoretically rigorous explanation of this effect (A does not exist globally
on the region fZ!) can only be obtained within the theory of connections on
principal S1-fiber bundles. The integral of A over a closed curve can then
be defined correctly and measures the holonomy of the connection'.
If we put the potentials A and 0 at the beginning of our study, the electric
and magnetic fields are determined by the equations
10
WE
c a (A) -do, wB = *dA,
and the first group of the Maxwell equations is automatically satisfied. In
these formulas, we view the quantities as time-dependent functions or forms
in 3-space, and the Hodge operator * and the exterior derivative d refer to
that space. We use the adjoint operator 5 of the exterior derivative d and
note that in R3 the following formulas hold:
8 = -*d* for 1-forms and 8 = *d * for 2-forms .
Then we can write the second group of the Maxwell equations as
1 8 2 (A) - 1 8 (do) + a
8dA w.J,
1 8A) O = 1aB
c2 8t2 c cat c c tit (
The Laplacian A on functions will be used in this chapter analogously to
Chapter 2. Hence we have Ao = *d * do = -8d¢.
1As a physics reference we recommend: Y. Aharonov and D. Bohm, Phys. Rev. 115 (1959),
485-491, as well as the more detailed discussion of this effect in the textbook by F Schwabl,
Quantum Mechanics, Springer, 2nd rev. ed. 1995.
9.2. The Static Electromagnetic Field 299
9.2. The Static Electromagnetic Field

If the electric and magnetic fields are time-independent, the Maxwell equa-
tions simplify:
*dWE = 0, d(*WB) = 0,
4a
*dwB - WJ, d(*WE) = 4ap d1R3 .
C
Hence the charge density P and the current density vector J do not depend
on time, and J is a divergence-free vector field. This system of equations
decomposes into a pair of partial differential equations for the electric field
and another similar pair of equations for the magnetic field. We begin by
solving the equations for the electric field. Using the locally defined electric
potential WE = -do, this system reduces to the inhomogeneous Laplace
equation
-AO = 47re .
From now on we suppose that the support of the charge density is compact.
Then it is to be expected that the electric field E(x) generated by this charge
distribution tends to zero at infinity. Under this condition, the electric field
is uniquely determined.
Theorem 1. Let the support of the charge density p be compact. Then there
exists precisely one electric field E such that
div(E) = 4ae, curl(E) = 0
and IjE(x)II - 0 for Jlxii oc. This electric field is given by E = -grad(4)
with the electric potential
0(y)
OW dy,
11x - Y11
R3
which is also called the Newton potential for the density e.
Proof. Consider two electric fields El = -grad(01), E2 = -grad(O) gen-

erated by one and the same charge density p. The difference u := 01 - 02 of
the electric potentials is a harmonic function with a gradient vanishing at
infinity,
Au = 0, 1Igrad(u)(x)iI - 0 for IIxil -+ oo.
Since i (Ou/Ox') = 8(Du)/ax' = 0, the partial derivatives au/8x(1<
i < 3) are harmonic functions on 1[t3 also vanishing at infinity. Liouville's
theorem for harmonic functions (see Theorem 42, §3.10) immediately implies
that all partial derivatives vanish identically. Hence u is a constant function,
and the electric fields El =- E2 coincide. There exists at most one electric
field with the required properties, and it remains to be proved that the
electric field E = -grad(O) with potential
OW = -
f
R3
P(y)
lix - yII
dy
is a solution for the problem. To see this, we have to show that O(x) is a
smooth function in R3, that it solves the inhomogeneous Laplace equation
-0O = 41rp, and that the gradient l grad(O)II -' 0 at infinity in R3. We
introduce spherical coordinates around a fixed point x = (x1, x2, x3) E R3
(0 < cp < 27r, -a/2 < ip < 7r/2):
y1 = x 1 + r cos cp cos ? / i , y2 = x2 + r sin c p cost , y3 = x3 + r sin Eli .
The volume form of euclidean space is described in these coordinates by the
following formulas:
dR3 = dy1 A dy2 A dy3 = r2 cos /i dr A dcp A d/i.
For every positive number e > 0, the function
1
Ay) := yii3_e
IIx -
is integrable in a neighborhood of its singular point. Integration over the
ball D3(x,1) with center x E R3 and radius 1 yields
1 2,r K/2
J
r
D3(x,1)
f (y) dy = ffJ r3
0 0 -7r/2
-E cos i,i dr dcp dzb = 4a/e.
Because of the compactness of the support of the charge density g(y), the
function c(x) and its first and second partial derivatives are smooth,
ao x' y` e
ax' -
(x) -
/'
P(y) dy = - /' P(y) ayi (ox
IIxyII3 J YII) dy
= f ay= (y) 1 dy
II x yII
R3
Differentiating once again leads to the formula

1
- AO(x) = lim lim f 0(P)(y) ' dy.
MM-00 E-0 IIx yII
D3(x.M)\D3(x,e)
Applying Green's second formula for sufficiently large radius M, we can
rewrite the integral over the spherical shell D3(x, M)\D3(x, E):
f o(P)(y)
FIX 1 yII f P(y) ' b IIx 1 y1I
D3(x,M)\D3(x,e) D3(x,M)\D3(x,M)
- f [(y)(gy1 1 Y11,
N(y)) - IIx 1 YII

\grady(y), N(y))] dy.
8D3(x,e)
The function 1/IIx - yUI is harmonic in the spherical shell: hence the first
integral vanishes. Moreover, the normal vector field of the sphere 8D3(x, e)
of radius a is given by the formula N(y) = (y - x)/E, which immediately
implies
1 1
(grads V(y))
IIx - Y1I , E2
Inserting these expressions, we obtain
f OP(y)
II x-
1 y1I
dy = 2 f y(y)dy+ 2
e
J (grad e(y), y-r)dy .
D3(x,A1)\D3(x.e) aD3(Xx) aD3(x,e)
The second integral on the right-hand side is bounded by
(1 /E2) - e - max{ Ilgrad oII } vol(8D3(x, E)) = 4;r - e max{ Ilgrad oII } ,
and hence it does not contribute in the limit as I 0. Together, this yields
-O0(x) = limo F J B(y)dy = 4irp(x),
0D3(x.e)
where we used the mean value theorem of integral calculus in the last step.
We estimate the length of the gradient of the electric potential using the
formula for the partial derivatives 8o/0x` and for points lying far out. To
do so we choose a radius R > 0 such that the ball with center 0 E R3
contains the support of the charge density. If now IIxII > R, then
2
Ilgrad6(x)II2 =
3
i=1
f 0(y)Ily-x113dy
D3(0,R)
x{ _ yi
1
<
3-max{p2(y):yER3}.4rr
3 dist2(x. D3(0; R))

We conclude that the length of the gradient IIgrad 4(x)II decreases like 1/IIxII
for IIxII oo
Remark 1. Obviously, the preceding theorem and its proof remain valid
for a charge density p which no longer has compact support, but, together
with its derivatives, vanishes sufficiently fast at infinity.
Example 1. Consider for a fixed point xo E R3 a charge density e which is
concentrated in a small ball around it,
ee(y) _ fe (vol(D3(xo,E)))-1 if y E D3(xo,E),

0 otherwise.
We calculate the electric potential:
dy = -e
(x)
R3
Lot(Y)
IIx - Y11 vol(D3(xo, e) I
D3(xo,e)
IIx - Y11
dy,
and in the limit as e -+ 0 we obtain

-e
lim , (x) -
E-o ` IIx - x011
In this way, the Coulomb potential arises naturally as the field generated
by an electric point charge. Let, similarly, n points xi, ... , x,, with charges
e1, ... , e be given. Then they generate the potential
n
O(x) = - E IIx ej- xcll
i=i
Now we turn to the solution of the two partial differential equations for
the static magnetic field B(x) and consider again only the case that the
divergence-free current density vector has compact support (see the remark
above).
Theorem 2. Let J be a divergence free vector field with compact support in
R3. Then there exists precisely one magnetic field B such that
4a
div(B) = 0, curl(B) =
c
J
and IIB(x)II - 0 for IIxii - oo. This magnetic field is given by B = curl(A)
with the magnetic potential
A(x) l dy.
cJ IIx -Y11
R3
Proof. Let B1 and B2 be two magnetic fields with the stated properties.
Their difference V := BI - B2 is a divergence- and curl-free vector field,
div(V) = 0, curl(V) = 0,
whose length II V(x) 11 tends to zero for IIx11 -+ oo. By Poincare's lemma, such
a vector field can be represented as the gradient of a harmonic function f ,
V = grad(f) and Of = 0.
The partial derivatives Of /8x' (1 < i < 3) are thus bounded harmonic func-
tions on the entire space R3, and by Liouville's theorem they are constant.
This immediately implies V = Bl - B2 = 0, and this observation shows that
there can be at most one magnetic field with the required properties. By
the same arguments as in the proof of the preceding theorem, the magnetic
potential
A(x) c AY) dy
J
R3
11X - Y11
is smooth. It remains to be shown that it is divergence-free. The partial

derivatives of the components A = (Al, A2, A3) of the magnetic potential
are
ON P J`(y) (x' -1/')
dy,,
c llx-yll3
R3
and hence
div(A) = Z' (J(y), )dy = I f (J(y), grad, (IIx
J
)dy.
R3
IIx - y113 1 yIl)
R3
Using the formula
divy(lix YIIdiv(J)
1 Y11 J(y)) = (J(y),g'ady(llx 1 y1I)) + I1x 1
and the fact that J was supposed to be divergence-free, we obtain
div(A) = f divy(llx 1 YIIJ(v))dy = 0

R3
c
by Green's formula, since the vector field J(y) has compact support. The
curl of the vector field B = curl(A) can now easily be calculated (see Exercise
6.b, Chapter 2):
curl(B) = curl(curl(A)) = grad(div(A)) - 0(A) = -0(A)
Here the Laplacian is to be applied component-wise to the vector field A,
and the proof of the preceding theorem yields the result
47r
curl(B) = -i(A) = c
J.
Example 2. Let the current density vector J be constant in a domain
ft C H3 and zero outside this domain. The magnetic potential and the
magnetic field B = curl(A) are then determined by the formulas
A(x)
c (fn II X 'YII) J' B(x)
c J " f II(x - 11 dy.
n
The magnetic field induced by an electric charge distribution with constant
current density vector is perpendicular to the direction of the current flow.
This fact is called the Biot-Savart law.
304 9. Elements of Elect rodvnamics
Finally, we interpret both theorems as the Hodge decomposition of a 1-form

on R3 (see Corollary 5, §3.11). Let wl be a 1-form on II23 with compact
support or, more generally, with rapidly decreasing coefficients. Applying
Theorem 1 to the function p = *d(*wl), we obtain a closed 1-form wE with
dwE = 0 and * d(*wE) _ *d(*w') .
The form wE also tends to zero at infinity, and hence, by Theorem 2, the
difference J = w' - WE can be represented as
*dw' = J = w' - wE
with a 1-form wa vanishing at infinity. This observation leads to
Theorem 3 (Helmholtz's Theorem, Hodge's Theorem). Every smooth 1-
form w' in R3 with compact support, or, more generally, with rapidly de-
creasing coefficients, can uniquely be represented as the sum of a closed and
a coclosed form.
w' = WI +*dwg, L4 = 0.
wF and 4 both tend to zero at infinity.
Remark 2. Comparing this result with Corollary 5 in §3.11, we stress that,
in I3. the harmonic part does not occur in this decomposition. The reason
for this is that there are no harmonic 1-forms in 1R3 having compact support
or tending to zero sufficiently fast at infinity. If w' were such a 1-form, then,
by Poincare's lemma, we could represent w' as the derivative of a harmonic
function, w' = df,zf, = 0. The partial derivatives Of 149x' would then be
bounded harmonic functions, and thus constant. If then the length of w'
tended to zero at infinity, then these partial derivatives would vanish and
we could conclude that wl = 0.
9.3. Electromagnetic Waves

An electromagnetic wave is an electromagnetic field in the vacuum (p = 0.
.1 = 0). The Maxwell equations in this case read as
div(B) = 0, div(E) = 0,
149E
curl(B) = 149E
- , curl(E)
c 8t
,
8t
which implies that the components of the electric and magnetic fields have
to be solutions of the wave equation
AB = 1 02B AE _ 1 492E
c2 &2 ' cz 8t2 .
9.3. Electromagnetic Waves 305
If the wave is known at a certain time t = 0 and we want to study its

propagation in time, we have to solve the Cauchy initial value problem for
the wave equation:
102
Du = c2 at2 ,
u(x, 0) = uo(x), - (x, 0) = ui (x) .
For smooth initial conditions uo and ul, this problem has precisely one
solution, which we are going to describe now. To do so we need the spherical
mean of a function V : 1R3 --i R. This is understood to be the function
IV : 1R3 x 1R+ -+ 1R which computes the mean value of cp on an arbitrary
2-dimensional sphere S2(x; r) with center x E R3 and radius r > 0,
(1V)(x,r) 4Trr2 f p(y)dy

S2(x,r)
Theorem 4 (Poisson Formula for the Wave Equation). For any two smooth
initial conditions uo, ul : 1R3 - R, the Cauchy initial value problem for the
wave equation
1 2
Du = c2 at2 '
0)
u(x, = uo(x), at
(x, 0) = ui (x)
has precisely one solution on the space 1R3 x 1R+. Moreover, this solution can
be computed explicitely:
u(x, t) = (t (Iuo)(x, c t)) + t (Iul)(x, c t).
ttII
to+ A (xo, to)
R3
r- (xo, to)
Remark 3. The value u(x, t) of the solution for the wave equation depends
only on the behavior of the initial values on the boundary of the base of the
backward light cone,
r (xo, to) := {(x, t) E 1R3 x 1R+ : Iix - roil < c It - toe, 0 < t < to} ,
which is a two-dimensional sphere. Therefore, every wave in 1R3 generated by
initial conditions with compact support has a forward as well as a backward
wave-front (Huygens' principle).
Proof of Uniqueness for the solution. We introduce the notation

1 012
A- cz
for the wave operator and
Eu(x,t) := 1(a l+(cat

3 z Jz
19U
J J)
for the energy. Then the following easy identity holds:

au 3 a au au a
E( u)
Now let u be a solution of the wave equation with initial conditions u(x, 0) _
0 and au(x, 0)/at = 0, and denote by FT (xo, to) the truncated cone described
by the condition 0 < t < r. By integrating the stated identity, we obtain
a r 3 a au au
at
E(u) = 2 t axe
J axe
rr r,
Gauss' theorem (Theorem 27, §3.8) allows us to transform the volume inte-
grals into surface integrals:
3
f E(u) ' (N, atj = 2 at axj \N' axj
arr er, '=1
The boundary 81'7 consists of three parts. On the first part (t = 0), the
energy E(u) and the derivative au/at vanish; this term does not contribute
to either side of the equation. On the second boundary piece (t = r), the
normal vector N is parallel to a/at. Hence the integral on the right-hand
side of the equation vanishes, whereas the integral on the left-hand side is
non-negative. On the third part of the boundary, the lateral surface of the
truncated cone, we calculate the scalar products
a zo_
xJ.
- ( a _ C
CN'ax) Ilx - xoll 1 + \N' i+c

Thus we can transform the difference of the left-hand and the right-hand
integrand into an expression which is non-negative, too:
2
( a au au a c au x) - xo On
E
\N' at / - 2
at axi \N' axi / - i+ F- cat l I x - xo l l ax)
All in all, the integral formula implies that the energy E(u) vanishes on
the upper boundary (t = r) of the truncated cone E. The height r of
the truncated cone can be varied, and thus E(u) vanishes identically in the
interior of the cone r- (xo, to). There the function u(x, t) is constant, and,
looking at the initial conditions, identically zero.
Proof of the existence of a solution. We calculate the derivatives of the
spherical mean with respect to time:
(Iul)(x,ct) = Jui(y)dy = f u1(x+ct z)dz,

4nc2t2 41r
S2(x,Ct) 52
where S2(x, Ct) is the sphere with center x E R3 and radius ct. Denote by
S2 the unit sphere. Then Green's formula implies
s
8(lul) r C?u l
(x, ct) = c J z
axjj (x + ct. z)dz
ax-
at 4w
j=1
2
47rct
2 f N(a), axj ' axJ
a"1(a)da = l 2J
41rrt
Aul (y)dy
S2(x.ct) D3(x,ct)
Introducing spherical coordinates (r, a) E (0, oo) x S2 in the three-dimen-

sional ball D3(x, ct) and applying the formula d]R3 = r2 dr A da, we obtain
ct
a(at 1) (x, ct) =

4x 2
jr2jAU,(X+ra)da-dr.
0 S2
This leads to
Ct
(t (Iul)(x,ct)) = (Iul)(x,ct) + 41rct f r2

f Dul(x+ra)da dr,
0 S2
and, after further differentiation,

2
2
(t. (Iu1)(x, ct)) =
4t
JAu,(x+ct.a)da = c20(t' (Iul)(x, ct))
S2
Thus w(x, t) := t- (1ul) (x, ct) is a solution for the wave equation with initial
conditions
2
w(x, 0) = 0 = 02 x, 0) and at(x,0) = (Iul)(x,0) = ul(x).
The derivative Ow/8t is also a solution of the wave equation, and it provides
the second summand in the asserted formula.
We have thus solved the wave equation for a scalar function on R3. The
expression for the vector valued functions E and B follows from a tedious
but routine computation. The explicit formula may be found in Exercise 3.
Now we turn to the inhomogeneous wave equation u = f with a smooth
function f : JR3 x R+ -s R, and will solve this for the initial condition
ud = 0 = ul. Combining the resulting solution with a solution for the
homogeneous wave equation with arbitrary initial condition constructed in
the preceding theorem, we will arrive at the solution for the inhomogeneous
wave equation Du = f with arbitrary initial condition u0, u1. The retarded
potential of a smooth function f : R3 x lR --> JR is defined by the integral
Rf(x,t) := - I f(y,t-r/c) dy
41r r
11x-Y11 <d
with the distance function r:= uix - y1l.

Theorem 5. Let f : R3 x R+ - R be a smooth function. Then the re-
tarded potential R f is a solution of the Cauchy initial value problem for the
inhomogeneous wave equation,
Du = f, u(x,0) = 0 = 8 (x,0).
Proof. Passing to spherical coordinates with center x E R3, we can repre-

sent the retarded potential as
Rf(x,t) _
fJ s2
Introducing the parameter r := t - 1: and the function
p(x,t,r) :=
t47rJ
32
we obtain from p(x, t, t) = 0 the following formulas for the retarded potential
and its derivatives:
tr tr
Rf(x,t) =-c2J p(x,t,r)dr, 8t(Rf)(x,t) = -c2J (x,t.r)dr,

0 0
2
5jZ(Rf)(x,t) = -c2 J ate(x,t,r)dr-c2L(x,t,t).
0
Calculating the derivative of p with respect to t yields
5'(x, t, t) = Irf f (x, t) da = f (x, t) ,
S2
and, summarizing, we obtain

t
a 22(Rf)(x,t) = -c2 (x,t,T)dT-c2 f(x,t) .
Applying the wave operator to the retarded potential leads to the formula
t
D(Rf)(x,t) = f(x,t)-c2 f Dp(x,t,T)dr.

0
If the parameter T is fixed, the function
p(x, t, T0) = t 4i 0 f f (x + c(t - TO)a, T0) d

S2
is a solution of the homogeneous wave equation. This immediately follows

from the preceding theorem, since p(x, t, TO) is defined as a spherical mean.
Hence the retarded potential solves the inhomogeneous wave equation. 0
The inhomogeneous wave equation Du = f may be interpreted as follows.

For simplicity, suppose that f describes the current generated by a charged
particle moving along the world line y. At the point A, the particle pro-
duces an electromagnetic field which moves at the speed of light along the
surface of the light cone with vertex A. Hence the field in B depends on the
motion of the particle in A. In this way, the principle of causality appears
as a mathematical consequence of the inhomogeneous wave equation, and
its retarded solutions were first described by the physicists Lienard (1898)
and Wiechert (1900).
x
The solution of the two-dimensional wave equation can be constructed start-

ing from that of the three-dimensional wave equation. To this end we trans-
form the spherical mean of a function ¢(x', x2) depending only on the first
two spatial variables into a two-dimensional integral by parametrizing the
sphere S2 (x, ct) C R3 through
yl = xl + a1, Y2 = x2 + a2, y3 = x3 + c2t2 - (al )2 - (a2)2,

where (al)2 + (a2)2 < c2t2. A straightforward calculation shows that the
volume form of the sphere is described by the formula
dS2 = t2 _ (1) - (a2)2 dal n dal,

from which we obtain the following expression for the spherical mean:
I (xl x2 t) = 1
1J _(x1 - a1, x2 - a2) dal n dal.
29rct c2t2 - (al)2 - (a2)2
D2(O,d)
We treat the retarded potential of a function only depending on the first

two spatial variables and time analogously f (x1, x2, t):
_ C
f xl - al, x2 - a2, r
Rf (xl, x2, t) dal nda2ndr .
21r J (t - r)2 - (a1)2 - (a2)2
Inserting the resulting expressions into the formulas of the preceding theo-
rem, we obtain a representation of the solution of the inhomogeneous two-
dimensional wave equation with prescribed initial conditions uo(xl,x2) and
u1(x',x2) as an integral over the interior of a disk in R2. From this we
conclude that a wave in R2 generated by initial conditions with compact
support has a forward front, but it does not have a backward front. This
effect is well-known, for example in the case of water waves. Stated differ-
ently, this means that Huygens' principle (see Remark 3) does not hold in
two dimensions (in fact, it can be shown to fail in all even dimensions).
Now we turn to the question of which conditions rotating an electromagnetic
wave lead to another electromagnetic wave. Let an electromagnetic wave
E(x, t), B(x, t) be given. We rotate these vector fields in the 2-plane spanned
by the two vectors through the angle 1/i(x, t):
E` (x, t) = cos 1Ji(x, t) E(x, t) + sin '(x, t) B(x, t) ,
B'(x, t) = - sin 1'(x, t) E(x, t) + cos?i(x, t) B(x, t) .
We discuss the conditions to be satisfied so that the pair (E', B') becomes
an electromagnetic wave. Necessarily, ip has to be a solution of the eikonal
9.4. The Relativistic Formulation of the Maxwell Equations 311
equation
1
8t-) - Ilg'ad('+6)fl2 = 0.
(2
c2
Theorem 6. (E*, B') is an electromagnetic wave if and only if the phase

function 0 satisfies the following conditions:
(1) (grad(tG), B) = 0 = (grad(tG), E);
(2) - 00 ' B = -grad(O) x E;
(3) E = grad(O) x B.
c
Proof. The vector fields E and B are divergence-free, and hence the diver-
gences of E' and B` are easily calculated:
div(E*) = - sin i' (grad(,O), E) + dcos'. (grad('+G), B) ,
div(B') = - cos lO (grad(t/i), E) - sin O (grad(O), B) .

Thus the vanishing of these divergences div(E') = div(B') = 0 is equivalent
to the condition (grad(t1), E) = 0 = (grad(Vi), B). Making use of the formula
curl(f V) = grad(f) x V + f - curl(V),
which generally holds in R3, we write down the Maxwell equations for the
pair (E', B'). This leads to the following pair of equations:
sin ,p grad(O) x E - c B l = cos 10 _ - E - grad(O) x B)

-6T 5T
sin t/' (_grad(1iL) x B + . E) = cos 7[i ( B + grad(t/') x E I

-C -N
which are equivalent to the second and third conditions of the theorem. 0
9.4. The Relativistic Formulation of the Maxwell Equations

Consider Minkowski space R3.1 with coordinates (x, y, z, t) as well as the
pseudo-Riemannian metric of index 1 determined by
g :_ (dx1)2 + (dx2)2 + (dx3)2 - c2dt2,
the so-called Minkowski metric (see Example 35, section 3.11). The 1-forms
j1 , dx2, dz3 and c dt form an orthonormal frame. Computing the Hodge
operator * of 183.1 on 2-forms leads to the following result:

*(dx' A dx2) _ -dx3 A c dt, *(dx1 A dx3) = dx2 A c dt,
*(dxl A c dt) = dx2 A dx3 , *(dx2 A dx3) = -dx1 A c dt,
*(dx2 A c dt) = -dx1 A dx3 , *(dx3 A c dt) = dx' A dx2 .
If w is a 1-form on 1R3, the Hodge operators * of euclidean space R3 and *
of Minkowski space 183.1, respectively, are related by
*w.
In R33 we define the 2-form called field strength form by
F :=
and the 1-form called density form by
1
J :=
C
wr-
Theorem 7. The 2-form F has the following properties:
(1) IIFI12 = IIBI12-IIE112.
(2) (F,*F) = 2(E, B).

(3) The first group of the Maxwell equations is equivalent to
dF = 0.
(4) The second group of the Maxwell equations is equivalent to
*d*F =
Proof. We calculate the exterior derivative dF of F in 4-space:
dF =
(*&B)+a( B) Adt
18(*OtwB))
= (d(wE)+
and see that dF = 0 is equivalent to the pair of equations
8
dR3 (*wB) = 0 and e (WE) C (*wB)
Similarly, we obtain 5i
d*F=d(*w E - WB A cdt) = dR3 (*wE) + (*wE) A dt - 0 (wB) A cdt,

and, after a slight reformulation,
18
*d*F = *dR3 (wB)-c8t(WE)
Summarizing, * d * F = 4a J is equivalent to
d(*WE) = 47rp and * d(wB) =
18 4c
wj.
c 8t (wE) +
Remark 4. The 2-form F is a closed form in ]R3.1. Hence by Poincare's
lemma there exists a 1-form A such that
dA = -eF.
c
The 1-form A comprises the electric as well as the magnetic potential:
F=
c 5i
dR3.'(A
=
and thus A = -(e/c)A + e 0 dt.
The isometry group of the Minkowski metric g is 0(3, 1), the Lorentz group.
It consists of all linear transformations L :183,1 -+ R3,1 which leave g invari-
ant. If we represent g by the matrix
A=
0 -C2
then 0(3, 1) can be identified with the (4 x 4) matrices L satisfying
A.
Its Lie algebra o(3, 1) is spanned by the matrices X with
Lt = 0.
Of course, 0(3, 1) contains the 3-dimensional group of rotations of euclidean
space R3 as a subgroup, embedded as the upper left 3 x 3 matrices
0 0
1 M E 0(3,R)
{ I. 1
and generated by the skew-symmetic matrices L1, L2, L3 described in §6.3,
Example 14. The missing part of the Lie algebra o(3, 1) is spanned by the
three elements
1 f 0
0
B2 :=
0 L B3 :=
0 0
B' 0 ' 0
000 0 00J' 00G 0
with commutator relations

[B1, B2]= c2L3, [B1, B3[ _ -c2L2, [B2, B3[ = c2Lg,
[L1, B2] _ -B3, [L1, B3] = B2, [L2, B1] = B3, [L2, B3] _ -B1,
[L3, B1] _ -B2, [L3, B2] = B1, [Li, B,] = 0.
In the same way as L 1, L2, L3 generate ordinary rotations, the elements
B1, B2, B3 induce hyperbolic rotations, for example,
cosh(c8) 0 0 sinh(cO)/c
e OBI
- 0 1 0 0
0 0 1 0
c sinh(c 8) 00 cosh(c 8)
The defining condition for L E 0(3, 1) implies that L has determinant 1 or
-1. The hyperbolic rotation e8Bi and the time reflection diag(1,1,1, -1)
realize these values of the determinant. Hence 0(3, 1) has two connectivity
components.
Theorem 8.
(1) The inhomogeneous Maxwell equations dF = 0 and bF = 41rJ are
invariant under isometries of the Minskowski metric, the Lorentz
transformations.
(2) The homogeneous Maxwell equations dF = 0 and d * F = 0 are
invariant under conformal changes of the metric.
(3) The quantities I IFI I2 and (F, *F) are invariant under Lorentz trans-
formations.
Proof. An element eox E 0(3,1) acts on differential forms and vectors

by its differential X E o(3, 1). The 2-form F transforms into its pullback
X*(F). But since pulling back commutes with exterior differentiation, dF =
0 is equivalent to d(X*F) = 0, which establishes the Lorentz invariance
of the first Maxwell equation. For the second equation, observe that the
volume form dM of any pseudo-Riemannian manifold M is invariant under
isometries of the metric; hence the adjoint of d, the coderivative 0, also
commutes with pullbacks by isometries. Thus, OF = 47rJ is equivalent
to X*(bF) = 47rX*(J) and b(X*F) = 4irX*(J). This shows the Lorentz
invariance of the second Maxwell equation.
Now consider a conformal change of the metric on a pseudo-R.iemannian
manifold M, y = e2f g for some function f. The exterior derivative does
not depend on the metric; the first Maxwell equation is thus trivially con-
formally equivalent. An orthonormal basis el,... , e with dual basis of
1-forms Qn transforms into
e-fel,...,e-fe, and efv1,...,efa,,.
If M has even dimension n = 2k, we have the remarkable effect that the
Hodge operators * and * with respect to the metrics g and g coincide on
differential forms in half the dimension of M. For example,
*(a,A...Aak) = -'ak+1A...Aa2k, *(efaiA...Aefak) = E'efak+1A...Aefa2k

for some irrelevant sign e = f1, so that k factors of of can be cancelled
in the action of *, which implies the claim. Its follows for the 2-form F on
4-dimensional Minkowski space that d * F = 0 is conformally invariant. 0
The conformal group can be identified with O(4, 2), whis has dimension 15.
The fact that only the homogeneous Maxwell equations are conformally in-
variant was the main reason why Lorentz invariance has been established as
the fundamental invariance property of electrodynamics. However, the full
conformal invariance is useful when dealing with homogeneous situations.
The invariance of the scalar quantities I IF112 and (F, *F) is of practical im-
portance, too; it means. for example, that E and B stay orthogonal if they
were orthogonal in some frame of reference.
As a 2-form, F is an element of 112(R3"). There exists a purely algebraic,
Lorentz invariant map from 112(R3.') into the space of symmetric (2,0)-
tensors. Apart from normalizations and an additional multiple of the metric,
the image of F under this map is known as the Maxwell stress tensor, which
we will now discuss.
So let F be any 2-form on Minkowski space, and e1, ... , e4 an orthonormal
basis, i.e., satisfying g(e;, ej) = ej d;j with 1 = el = s2 = E3 = -64. We
define
TF :=
where O denotes the symmetric product of 1-forms (see 5.3). This is an

SO(3,1)-invariant, symmetric (2, 0)-tensor whose trace is a multiple of IIF112.
In the example at hand, we may choose
el = axl,
e2 = axe' e3 = ax3
and e4 = .
The 2-form F is given by

F = WE A c dt + Bl dx2 A dx3 - B2 dx' A dx3 + B3 dx' A dx2 .
Hence one computes (x4 := cdt)
TF = (Eicdt - B2dx3 + B3dx2)2 + (E2cdt + B1dx3 - B3dx1)2
+ (E3cdt + Bldx2 - B2dx1)2 - (Eldxl + E2dx2 + E3dxa)2
4
_: 2 dxÒdxJ.
In order to obtain a trace-free endomorphism, one typically defines as the

Maxwell stress tensor the combination
Tit :_ -2TF + (F, F) (dx' C dx' + dx2 ^ dx2 + dx3 ®dx3 - c2dt O dt).
Its importance stems from the fact that its components have a physical
interpretation. For example, its pure time component
T4'4' = -(11E112 + 1IBI12)c2dt2
is the energy density of the electromagnetic field, and the mixed contribu-
tions (i = 1.....3)
T"' = 4(ExB)idx'0,cdt
are the components of the Poynting vector E x B. which has the dimension of
an energy flux density. Since Ta' contains basically the same information as
F, it is plausible that one can rewrite the Maxwell equations in terms of TMM
solely. For their derivation, we refer to classical textbooks in eletrodynamics.
The electromagnetic potential A is a real-valued 1-form defined on Minkowski

space. We now view the real numbers as the Lie algebra of the compact one-
dimensional group U(1). in which the only quantity describing the electron
which is relevant in Maxwell's theory is encoded---its electric charge. Pass-
ing now to the description of more complex elementary particles which carry
not just electric charge, but in addition other characteristics (color, ... ), we
have to replace the one-dimensional group U(1) by a higher-dimensional
compact Lie group G. If we also want to include gravitation in the model,
as understood in the sense of general relativity, flat Minkowski space has
to be replaced by a pseudo-Riemannian manifold. Combining all this, we
are led to define a "generalized electromagnetic field with gravitation" as
a 4-tuple (Al, g, G, A) consisting of a pseudo-Riemannian manifold, a Lie
group G, and a 1-form A : T(M) -+ g with values in the Lie algebra g
of G. The non-commutativity of the group G will become essential in the
sequel. The field strength FA is described by a g-valued 2-form, and to the
Maxwell equation in the vacuum there corresponds the so-called Yang-Mills
equation,
FA := dA + 2 [A, A], DA(* FA) := d * FA + [A, *FA] = 0 .

The possibly complicated topological structure of the base space Al may
lead to a situation in which the generalized electromagnetic potential A
is not a globally defined 1-form on M. This leads to topics studied in the
differential geometry of principal fiber bundles and the theory of connections.
9.5. The Lorentz Force 317
With their help, the models discussed in the physics of elementary particles
can be formulated and studied rigorously.2
9.5. The Lorentz Force

In this section, we discuss the equation of motion for the Lagrangian of a
massive particle in an electromagnetic field as introduced in Example 11
of Chapter 7. We show that it is a natural generalization of the Lorentz
equation, and discuss its solution for a few electromagnetic fields, i.e., a
time-independent homogeneous field and a Dirac monopole. Frenet's for-
mulation of curve theory will prove to be a suitable tool for the description
of the particle's motion.
Let (Mm, g) be a pseudo-Riemannian manifold, A a 1-form on it, and con-

sider the Lagrange function
L(v) = 2g(v, v) - A(v) .
Theorem 9 (Generalized Lorentz equation). A curve e(s) : R - Mm is a

motion of the Lagrange system (Mm, L) if and only if it satisfies
mds = V(eJdA),
where the right hand side is meant to denote the vector field associated with
the 1 form e J dA relative to the Minkowski metric.
Proof. We use local coordinates x1, ... , xm on Mm. The Lagrangian can
then be rewritten as
m
L(xl,...,xm,2....... m)
2 i,j=1 i=1
By definition, e(s) : ]R -p All is a motion of the Lagrange system (Mm, L) if

and only if it fulfills for all k between 1 and m the Euler-Lagrange equation
ds (8i (e(9))) = a k (O(s)).

Since 6(s) = (dx1 /ds, ... , dxm/ds), the left and right sides can be computed
to be
OL m No j dxi dxj 8A; dxi
8xk W M = 2 8xk ds ds 8xk ds
to t
2See Thomas Friedrich, Dirac Operators in Riemannian Geometry, Grad. Studies in Math.
vol. 25. AMS, Providence, 2000.
and
d aL agile dxi dx' d2xi aAk dxi
(aik (9(s)) I = m>2
ds axle ds ds + m>9ik ds2 i
axi ds
Equating them and rearranging some terms, we obtain the equation
[aAk aAi dx
(*) axi - axk ] ds
r99ik 1 agi; dx' dx'

e m 9ikd2xi
+m
i
ds2
i,j
ax; 2 axk I ds ds
Hence it remains to interpret both sides geometrically, as done in the formu-

lation of the theorem. We begin by recalling the definition of the covariant
derivative as discussed in §5.7, Definition 21. From there, we know that the
covariant derivative may be expressed in terms of the Christoffel symbols
associated to the chosen local coordinates as
M ye
_ d2xi a dxi dx' a
ds = m E d82 axi +m ds ds axi
But the Christoffel symbols can by computed from the metric (§5.7, Theorem
42 ff.):
k
r+ - 1 [: km agim a97m 89ii
2
9 8xk + Oxi - Oxm
m
hence one checks that the right-hand side of the equation of motion (*) is
equal to
r ve a l
g 'n ds ' axle
Similarly,
dA = ez dxi A dxx
ij
implies that
8 aAk aAi dxi
dA (e,
axk) i axi axk, d
which is just the left-hand side of (*). We conclude that equation
equivalent to
8 a
g md,8xk) = dA (k, axk 1
Since this has to hold for all k, the/claim is proved. 0

The most important situation where this result can be applied is the case
that Mm is the flat Minkowski space R3.1 with its pseudo-Riemannian met-
ric. As 1-form on R3'1, we choose the form A introduced in the previous
section, which is composed of the magnetic 1-form A on the euclidean space
R3 and the electric potential ¢:
A = -eA+ecdt = -e(A1dx1+A2dx2+A3dx3)+eodt,
and it satisfies dA = (-e/c) F. We emphasize that time t is viewed as the
fourth coordinate of space, whereas the curve parameter s of p is interpreted
as the proper time of the particle. The curve p(s) = (x1(8), x2(s), x3(s), t(s))
is said to be parametrized in proper time if its tangent vector is normalized
to length -c2,
2: =
c2
d- 11
In classical mechanics as discussed in Chapter 7 (the Newtonian limit), all

velocities were so small compared to the velocity of light c that it was safe to
identify s with time. We shall henceforth abstain from denoting derivatives
with dots or primes, in order to avoid confusion. By velocities, we mean the
time derivatives vi := dx`/dt with euclidean length v2 := vi + v2 + v32. The
parameters s and t are related through the relativistic -y-factor. To compute
it, we express the length of do(s)/ds as
2 3 [i)2 []2 = 3 dxt dt 2-c2 dt 2
= (v2 -c2) [ds]
[dt]2
d oII = ` ds -c2
ds [ dt ds]
ds
and deduce from its length normalization that

dt 1 _ ds
. 7, and = 1/ry .
ds = 1 - v2/c2 dt
Thus, it is comfortable to rewrite do/ds as
do r a
Y v1ax1 +V2
a a
+v3a7X3 + at
al
ds - 57X2
The right-hand side of the generalized Lorentz equation now reads, using
the definition of F,
WS J c TS
j F = +-_r [_(v. E)cdt + (cE1 + (v x B)i)dx`
where v E denotes the euclidean scalar product and v x B the vector product
in R3. Its dual vector is
a
V(doJdA) = ry
ds c2 at
+(eE;+e(vxB)i)
c ax'
Since the left-hand side of the generalized Lorentz equation is

ddg _ ddp
md-ds = mrydt ds
we conclude that its spacelike components yield after simplification the clas-
sical Lorentz equation
(1) mdt(ryv) = e E+ _v x B,
and its time component leads to mdry/dt = +(e/c2)v E. This equation
describes in fact the change in time of the kinetic energy e = mc2-y of the
particle,
(2) t = e(v E).

This proves that only the electric field exercises power on the particle, as
claimed in Example 13 of 7.5.
We shall now describe the motion of a charged particle in some special

electromagnetic fields. Consider the case of a vanishing electric field (E = 0).
Equation (2) then implies that the kinetic energy is constant; hence the
-y-factor and, in particular, the total velocity v2 are constant, too. The
spatial part of the particle's motion e(s) = (x' (s), x2(s), x3(s)) satisfies
IldWds112 = v2 = const; hence the reparametrized curve q(s) := o(s/v) is
given in its natural parametrization. The Lorentz equation for this curve
can be written as
mryv2 = cdd q x B ,
and it is natural to use Frenet's frame to describe the curve q(s) in euclidean
space R3. It is safe again to denote derivatives with respect to the natural
parameter s by dots.
Example 3 (Static and homogeneous magnetic field). Suppose that the
magnetic field B is static and homogeneous, i. e., constant in space and
time. Recall that t, h, b denote the tangent, the principal normal and the
binormal vector, respectively, and that c and r are the curvature and the
torsion of q. The Lorentz equation is thus geometrically equivalent to
e-.
-txB.
c
We differentiate this equation with respect to s and use Frenet's formulas
for the derivative of h and t:
eKh'
mryv(id +/c(-Kt+'rb)) = x B.
C
Since the vectors t, h', 6 constitute, at every point, an orthonormal frame, we

conclude that k = 0. We compute the value of the curvature:
e eIIBII eIIBII
K = 110 = jI x BII = sin(i),B) = 1 - (4.B) 1IIBII
m-yvc mryvc m-yvc
In fact, on can check directly that (4, B) has to be constant, for its derivative
is 2 (ii, B); but by the Lorentz equation, ij is perpendicular to B. Alterna-
tively, we could also have derived it from Noether's theorem. In order to
determine the curve's torsion, we first compute its binormal vector,
b = Fxh = xij,
as well as its derivative,
e
(h x B) = (h(t' B(h, ryB) h'.
b = -K x 11 = ym vcq x y mVc
e B) -
The torsion is then
h) _ -e(t' B)
rymvc '
and again constant. From Chapter 5, Exercice 7 we know that a curve whose
torsion and curvature are constant is necessarily a helix, the direction of
which is given by B. In the special case that (F, B) = 0, the curve is a circle
in the plane perpendicular to B with radius of curvature r and synchrotron
frequency w:
1 mryvc v _ eliBII
r= K
=
eIIBII'
=
r rymc
Example 4 (Dirac monopole3). Although their physical existence could

never be established, Dirac monopoles have proved important models in
theoretical physics. A monopole at the origin of R3 induces the field
c
B
= 2eIIiiII3'1
The Lorentz equation is thus
ij 27mvIMII3J7x g
(*) =
We are going to prove that the particle moves along a geodesic of a cone
with vertex the origin. As in the integration of the geodesic flow or the
abstract discussion of Lagrange systems, the proof makes crucial use of a
clever invariant of motion. But the invariant of the Dirac monopole differs
from all invariants encountered so far in the fact that it is not linear in ;
3This discussion has been published by Katharina Habermann (formerly Neitzke) in her
article K. Neitkze, Die Lorentz-Kraft auf pseudo-Riemannschen Mannigfaltigkeiten, Math. Nachr.
149 (1990), 183-214.
hence its existence does not follow from Noether's theorem, but only from
a direct calculation. First, we observe that
d !n, (n, 7i) + (n, n) = 0 + 1,

because n is perpendicular to 6j by the Lorentz equation (*), and so there
exists a constant a E R with (n, rl) = s +a. It also implies that the following
derivative vanishes:
ds 0,
which yields the invariant we were looking for. By the Cauchy-Schwarz
inequality, it has to be non-negative; hence there exists a constant k > 0
such that
Ilnl12 - (n,
)2 = k2 and 11,7112 = k2+(S+ a)2.
A short calculation for the curvature of the particle's curve of motion yields
1117 x ells sin(n,il) 1 -oos (2, _
K = Itrill = = In11s
=
2TmvJ lot I 2'rmvl 27mvI InI
1-ln,n>2/1117112 __ 111711'-117,17)2 __ k
2 ymvllnl l2 2'rrmll,II3 2-tmvl ln113
If k = 0, the curvature vanishes and the curve is a straight line, i. e., there
exist vectors v", w' E R3 such that n(s) = s v + v7. The equation of motion
(*) then implies v' x u7 = 0, i. e., v' and 0 are linearly dependent, so thet we
finally obtain
n(s) = (s + A)t7, A E It ,
which is the equation of a line through the origin. Let us assume from now
on that k > 0. We shall derive the expression for the torsion of the curve.
By Frenet's formulas, the principal normal vector hh is
h=-=- k7xil. 1
The binormal vector then becomes

b = ,7xh = -i4x(rlx1),
with derivative
n
dsb = -kn x (n x ii) = 2kmv7II,II3 ))
Using Grassmann's identity u7 x (u' x ) = u (v, w-) - v' (u, m'), one shows that
6 x (u" x (i x ii)) = - (u, v3) i x *7, so that the former expression can further
be simplified to
dg _ (n, n) __
ds (-k r) X rl
2mvryIIi11I3h.
2mvry11i1113
The torsion is thus equal to
r= tail)
(b, h) _ -2mvryll71113
__ s+a
k
+(a+a)23
2mvry
In particular, the quotient of torsion and curvature satisfies the simple iden-
tity r/ic = -(s + a)/k. We prove next that the curve lies on a cone, that is,
that there exists a constant vector ii whose angle p with 71 is also constant.
Define the functions

_ s+a _ 1
f(s) 2km yv k + (a + a) ' g(s) 2m yv k + -(s+ a)2
They are primitives of K and -T and have the same quotient,

d f (s) dg(s) f (s) _ k
ds = K' ds g(s)
r
Frenet's formulas thus imply immediately that the vector
iZ :=
is constant. To compute its angle V with rl, we derive an exact expression
for q by applying Grassmann's identity to the binormal,
1 1
b = -krl x (71 x n) _ -k [rl -il(s+a)], q = (s+a)t-k6,
and remark that ii can be rewritten 11= rl/2mkryvllr7l l + h. For the opening
angle of the cone we obtain
cos() = (fi, u') - 1
110 141 1 + (2mkyv)

By definition, a curve is a geodesic on a given surface if and only if its

principal normal vector is orthogonal to the tangent plane of the surface at
any of its points. Denoting by 7P the angle between h and i , we get
(h, u)
- 1 _ 2mkvy
Iluli 1 + (2mkvy)2
and we see that this is equal to 1 - cos2(cp) = sin(e). Hence, W and V are
related through V-V = it/2, and we conclude that h is indeed perpendicular
to the tangent plane of the curve. The particle moves along a geodesic of
the cone, as claimed.
Exercises 325
Exercises
1 (Kirchhoff Formula). Let u(x, t) be a function defined on JR3 x Ht and

S C JR3 a compact domain with smooth boundary. Prove, for each point
(Xo, to) E SZ x JR, the Kirchhoff formula:
u(1l, to - r/c) r f au (or, to -

u(xo, to) = Jsl r dy + J IL r 8N
Ou(a,to- or or(a,to) 8r-1(a,to)
+ cr 8t c eN - u(a'to C
OM
where r := Ilxo - yII is the distance to the spatial point xo E R3, N is

the normal vector to the boundary 00, and = Ox - 1/c28it is the wave
operator. Deduce from this the solution formulas for the Cauchy problem
of the wave equation by choosing for H a 3-dimensional ball.
2. Under the assumptions of Helmholtz' theorem, the electric and the mag-
netic field E and B, as well as the current density field J can be written as
the sum of a divergence-free and a curl-free vector field,
E = Ediv + Ecuri, B = Bdiv + Bcuri and J = Jdiv + Jcurl
a) Prove that the Maxwell equations can then be written as follows:
18Bdiv
curl(Ed;v) = - 8t
chv(Bcuri) = 0
1 8Ed;v 47r
curl(Bd;Y) =
at +
Jai,,, div(Ecuri) = 47r LO.
c
The continuity equation then becomes
div(Jcuri) + OLO = 0.
Hint: In the proof of Theorem 2, we proved that, under the assumptions
made here, any vector field which is at the same time divergence- and
curl-free, has to vanish identically.
b) On the other hand, for the electric field E we already know the decom-
position
E_ c +grad(-O).
Prove that this is precisely the Helmholtz decomposition of E if the

magnetic potential satisfies the condition called Coulomb gauge:
div(A) = 0.
3. Deduce from Theorem 4 that if Eo(x) and Bo(x) are C3-functions on R3

such that divE0 = divB0 = 0, then the Cauchy initial value problem for an
electromagnetic wave
curl(B) = c 5 , curl(E)
c at ,
div(B) = 0, div(E) = 0, E(x,O) = Eo(x), B(x, 0) = Bo(x)
has the unique solution
E(x, ct)
4act f
S2(x,d)
curl Bo(y)dy +
4ac 8t f Eo(y)dy
S2(x,d)
,
B(x, ct)
4act f
S2(x,d)
curl Eo(y)dy + 41
8t f Bo(y)dy
1S2(x,ct)
4. Using Theorems 1 and 2, determine the electric and the magnetic field
in 1R3 which is generated in the following situations:
a) a homogeneously charged ball of radius R with constant charge density
P;
b) a charged spherical shell of radius R with constant surface charge density
a;
c) an infinitely extended straight wire of radius R through which a current
with constant current density j flows.
5. Describe the solution of the classical Lorentz equation for a homogeneous,

static electric field (E = const) and vanishing magnetic field (B = 0). What
happens in the non-relativistic limit of small velocities v << c?
6. The radiation rate of a Lorentz electron is proportional to the second

derivative of the vector p(s) E R3,1:
R
211
Compute the general expression for R and discuss the case of planar circular
synchrotron radiation.
Exercises 327
7. Show that the Lie algebra o(3,1) of the Lorentz group is not simple-more
precisely, that it is the sum of two 3-dimensional ideals.
8 (Infinitely Extended String with Large Oscillations). Consider the Cauchy

problem for the one-dimensional wave equation determined by two functions
up(x) E C2(R), u1(x) E C1(It):
attu = c2atxU, t > 0,
u(x,0) = uo(x), atu(x,0) = u1(x)
Part 1. General Shape of the Solution. Prove that there exist two C2-
functions f and g satisfying
u(x,t) = f(x+ct)+g(x-ct).
The solution is hence the superposition of a wave traveling to the left and
one traveling to the right. Hint: Introduce the new coordinates xt = x ± ct
and show that the wave equation is equivalent to the differential equation
a2u
0.
ax+(9x- -
Part 2. Solution of the Cauchy Problem. With the above Ansatz for the
solution, the following relations have to hold:
uo(x) = f(x) + g(x), u1 (x) = c(f'(x) - g'(x))
By integration of the second equation over the interval [0, x], prove that the
general solution has to be given by
Ed
u(x,t) = 2[uo(x + ct) + ua(x - ct)) + 2cu1(s)ds.
In particular, this argument shows the uniqueness of the solution. How can
the result be understood qualitatively by means of the light-cone ?
9 (One-sided Infinitely Extended Oscillating String). An oscillating string

extending infinitely in the positive x-direction is modeled by the one-dimen-
sional wave equation,
attu = c2axxu, t > 0, x > 0.
In addition to the initial conditions,
u(x, O) = uo(x), 8tu(x, 0) = u, (x),
one has to impose boundary condition which describe the behavior of th
"wall" at x = 0 for all times t:
u(0, t) = cp(t).
Here uo, 4o E C2(R+) and ul E C'(R+) are required. Prove (using a similar
Ansatz as in the preceding exercise) that the solution is
x+d
[uo(x + ct) + uo(x - ct)] + -L 2cjx-d ui(s)ds,
u(x,t) = d+x
2 [uo(x + Ct) - up(ct - x)] + p(t - x/C) + ZJ ul (s) ds,
d-x
where the upper line is to be taken for points (x, t) in the region I, i. e. below
the line x = ct, and the lower line for points (x, t) in the region II, above
the line x = ct. What can be said concerning the behavior of the light-
cone, in particular in II? Describe, moreover, the regularity properties of
the solution on the line x = ct, and explain carefully under which additional
conditions it is of class C2 there.
x=ct
/1
10 (Oscillating String Fixed on Either Side). For the wave equation on

bounded domains a separation Ansatz going back to Bernoulli proved to be
successful, in that it reduces the problem to one for Fourier series. We are
looking for a solution of the one-dimensional wave equation on the interval
[0, 1],
attu = c28xxu, t>0, 0<x<1.
The initial and boundary conditions are
u(x, 0) = uo(x), atu(x, 0) = ul (x), u(0, t) = 0, u(1, t) = 0.
Exercises 329
The explanation for the Fourier method and the related regularity properties
to be required of uo and u1 will be treated separately in the next exercise.
Start from the Ansatz
u(x, t) = T(t) X(x) .
a) Prove that there exists a constant A, such that
X" _ T" _
X c2T
b) Show that the cases A = 0 and A > 0 necessarily lead to the trivial
solution u = 0; in the case A < 0 there has to exist a natural number k
such that A = -k27r2 and
uk(x, t) = Xk(x)Tk(t) = sin kirx (ak sin klrct + bk cos kirct) .
The general solution is then obtained as the series
00
u(x, t) _ uk(x, t) _ E sin kirx (ak sin k7rct + bk cos klrct) .
k=1 k=1
c) Find an integral formula for the coefficients ak, bk depending on the
initial conditions uo and u1. Solution:
1 1
2
ul (x) sin k7rx dx, bk = 2 fo uo(x) sin k7rx dx .
ak lore 0
11 (Validity of the Fourier Method). By a theorem of Dirichlet, the Fourier

series of a function f E C1([O,1]) satisfying f (0) = f (1) = 0 is uniformly
convergent and tends pointwise to f. Prove the following lemma:
For a function f E Ck([O,1]) such that f(0), ..., f(k-1) vanish at 0 and 1,
there exists a constant A for which the Fourier coefficients of f satisfy the
following inequality:
j f(x)sinn7rxdx< A
Wk
Formulate necessary conditions for uo and u1, under which the solution
u(x, t) constructed in the preceding exercise is twice differentiable with re-
spect to x and t and satisfies the initial conditions.
12 (Inhomogeneous Wave Equation).

Part 1. The inhomogeneous wave equation is understood to be the equation
(*) attu = c28x2u + f(x, t)
with a given function f. We consider this equation on the interval [0, 1] with
the initial conditions
u(x,0) = uo(x), i31u(x,0) = ni(x)
as well as, first, trivial boundary conditions,
u(O,t) = 0, u(l,t) = 0.
Show that the Ansatz
u(x, t) _ an (t) sin(nirx)

n=I
satisfies the boundary conditions, and prove that the functions an(t) are
uniquely determined. To do so, prove that they solve an ordinary differen-
tial equation of second order, and that the initial values an(0),a'n(0) can be
computed from the initial conditions. Hint: The Fourier coefficier..ts of uo,
ul and f play an important role here.
Part 2. We are looking for a solution u(x, t) of the inhomogeneous wave

equation (*) with arbitrarily given boundary conditions,
u(0, t) = spo(t), u(1, t) = VIM-
Show that the solution of this equation can be reduced to the solution of
an inhomogeneous wave equation with trivial boundary conditions (and
a different inhomogeneity f!). Hint: Let 4(x, t) be any function in two
variables satisfying 4(0,t) = spo(t) and fi(l, t) = spi(t), e. g. 4-(x, t) _
(1 - x)Wo(t) +xW1(t). Then consider the function v = u - 4D.
13 (Hadamard Example). This exercise intends to illustrate that the Cauchy

problem is not posed correctly for every differential equation, i. e., the so-
lution does not necessarily depend "continuously" on the initial conditions.
Consider the following Cauchy problem for the Laplace equation: Datermine
the solution of the differential equation
d2u
dt2
+ -dx2= 0
d2u
which at t = 0 satisfy the conditions
u(0,x) = 0, dJu(0.x) = n sinnx,

where k and n are positive integers. Suppose, moreover, that x JR and
t>0.
Exercises 331
a) Verify that this problem has the solution

ent - e-nt
u(t, x) = 2 nk+l sin nx.
How do (8tu(0, x) I and u(t, x) behave for arbitrarily small t, if n is suf-
ficiently large?
b) Now we suppose that we have found the solution 2i(t, x) for the initial
conditions
u(0, x) = uo(x), 8tu(0, x) = u1 (x).
What is then the solution of the Cauchy problem with the initial con-
ditions
1
U(O, x) = uO(x), 8tu(0, x) = u1(x) + sinnx?
Conclude that the Cauchy problem for the two-dimensional Laplace
equation is ill-posed.
Bibliography
Textbooks on Analysis on Manifolds

M. P. do Carmo, Differential forms and applications, Universitext, Springer,
Berlin, 1994.
H. Flanders, Differential forms. With applications to the physical sciences,
Academic Press, New York-London, 1963.
0. Forster, Analysis 3. Integralrechnung im Rn mit Anwendungen, Vieweg-
StudiuYn, Bd. 52, 2. Auflage, Vieweg-Verlag, Braunschweig-Wiesbaden, 1989.
K. Maurin, Analysis 11, Reidel/PWN, Warsaw, 1980.
W. Rudin, Principles of mathematical analysis, 3rd ed., McGraw-Hill, New
York, 1976.
M. Spivak, Calculus on manifolds, Addison-Wesley, Reading, MA, 1965.
Textbooks on Differential Geometry

W. Blaschke, H. Reichardt, Einfiihrung in die Differentialgeometrie, Grundl.
der math. Wiss., Bd. 58, Springer-Verlag, Berlin, 1960.
M. P. do Carmo, Riemannian geometry, Birkhauser, Basel, 1992.
M. P. do Carmo, Differential geometry of curves and surfaces, Prentice-Hall,
Englewood Cliffs, NJ, 1976.
C. Godbillon, Geometrie differentielle et mecanique analytique, Hermann,
Paris, 1969.
333
334 Bibliography
A. Gray, Modern differential geometry of curves and surfaces with. Mathe-

matics, 2nd ed., CRC Press, Boca Raton, FL, 1998.
M. Spivak, Differential geometry I- V, Publish or Perish Inc., since 1975.
K. Strubecker, Differentialgeometrie I-III, Sammlung Goschen, W. de
Gruyter Verlag, Berlin, 1969.
R. Sulanke, P. Wintgen, Differentialgeometrie and Faserbi ndel, VEB Dt.
Verlag der Wiss., Berlin, 1972.
A. Svec, Global differential geometry of surfaces, VEB Dt. Verlag der Wiss.,
Berlin, 1981.
Textbooks on Lie Groups and Lie Algebras

J. F. Adams, Lectures on Lie groups, Benjamin, New York, and Univ. of
Chicago Press, Chicago, IL, 1969.
Th. Brocker, T. torn Dieck, Representations of compact Lie groups, GTM
98, Springer-Verlag, Berlin, 1985.
J. J. Duistermaat, J. A. C. Kolk, Lie groups, Universitext, Springer, Berlin,
2000.
W. Fulton, J. Harris, Representation theory - a first course, GTM 129,
Springer, Berlin, 1991.
Textbooks on Symplectic Geometry and Mechanics

V. I. Arnold, Mathematical methods of classical mechanics, GTM 60. Spring-
er, Berlin, 1989.
A. T. Fomenko, Symplectic geometry, Advanced Studies in Contemporary
Mathematics 5, Gordon and Breach Publ., Amsterdam, 1995.
L. D. Landau, E. M. Lifschitz, Course of theoretical physics, vol. I: Mechan-
ics, Pergamon Press, Oxford, 3rd corr. ed., 1994.
A. Sommerfeld, Lectures on theoretical physics, vol. I. Mechanics, Academic
Press, New York, 1952.
W. Thirring, A course in mathematical physics, vol. 1: Classical dynamical
systems, Springer, Berlin, 1978.
Bibliography 335
Textbooks on Statistical Mechanics and Thermodynamics
R. Becker, Theorie der Wdrme, Heidelb. Taschenb. Bd. 10, Springer-Verlag,

Berlin, 1985.
L. D. Landau, E. M. Lifschitz, Course of theoretical physics, vol V. Statistical
physics, Pergamon Press, Oxford, 3rd rev. ed., 1996.
A. Sommerfeld, Lectures on theoretical physics, vol. V. Thermodynamics
and statistical mechanics, Academic Press, New York, 1956.
D. N. Zubarev, V. Morozov, G. Roepke, Statistical mechanics of non-equili-
brium processes, vols. 1 and 2, Akademie-Verlag, Berlin, 1996, 1997.
Textbooks on Electrodynamics
L. D. Landau, E. M. Lifschitz, Course of theoretical physics, vol. II: The
classical theory of fields, 4th rev. ed., Butterworth-Heinemann, Oxford,
1996.
A. Sommerfeld, Lectures on theoretical physics, vol. III: Electrodynamics,
Academic Press, New York, 1952.
A. Sommerfeld, Partial differential equations in physics, Academic Press,
New York, 1949.
W. Thirring, A course in mathematical physics, vol. 2: Classical field theory,
Springer, Berlin, 1986.
Symbols
J ................ 4, 4,68 GF ................231 0(t2) .............. 215
* .................7, 7,23 g .................. 209 Wk ...................1
[, ] ..................91 GL(n,K)...........208 *wy ................23
{, ) ................235 23,60
grad(f) .........23, wi1 .................142
O ..................148 grad(V)............ 255 wo................. 232
V ..................149 I' .................185 52 .................. 142
0/dt ............... 188 h(s) ...............132 R9 ................. 208
27,53
0 ................27, HkR(U) ............ 19 Rf(x,t)............308
0/0xi ..............12 Hk ..................52 curl(V) .............25
0(f) ....... 24, 65, 101 El ..................221 R(U, V)W ......... 151
0(f) ...............306 712 .................148 SL(n, R).. 106, 211, 221
Aff(K') ............208 Ik ...................27 SO(n, K) ......106,
106,221
Ad .................223 I(A) ............... 276 SU(1,1)............227
Ad* ................230 IV ................. SU(n) ..............221
305
ad := Ad........... 224 I ................... 148 s-grad(H) .......... 233
bid .................151 11 .................. 150 supp(ip) ............. 76
b(s) ............... 132 7(p) ............... 275 S(f?, IA) ............ 276
Bk(U) .............. 19 K*(x) ............. 196 ai ..................141
ck ...................26 K(z, M) ............ 37 t( s) ................ 132
d ................ 16,68 c(s) ............... 132 TpRn ................11
5 ...................101 Lg .................208 TXMk ...............54
dA, dQ ............ 284 Ev ..................89 TMk ................55
24,63
div(V)........... 24, C(v) ............... 260 T*Mk ............. 230
D"(R) .............. 38 L*(wk) .............. 4 r(s) ................133
dMk ................70 Mk .................47 U(n) ...............221
dM2m ............. 229 (M2m, w) .......... 229 12,56
V(p) ............12,
dV ...................6 (M2m, w, H) ....... 236 vol(Mm) ............ 78
Ek .................112 mi(f ), i = 1, 2, 3 ... 164 X(M2) ............. 165
Exp. ...............192 µt ..................272 Zk(U) .............. 19
f*., .............11,
11,56 N(x) ............... 74 Z(B) .............. 278
f*(wk) ..........14,
14,67 0,0 ...............69 Zc ................ 223
..................88 0*(F) ............. 231
9ij, 9ii .............. 59 O(n, K) ....... 106, 221
337
Index
action coordinates, 247 ensemble, 280

action integral, 259 symplectic structure, 230
adiabatic, 290 Carathdodory construction, 79
adjoint Carnot cycle, 290
minimal surfaces, 171 catenary, 172, 198
operator, 101 catenoid, 105, 172, 201
representation, 223, 224, 230 Cauchy problem
Aff(K^), 208 for the Laplace equation, 330
affine group, 208, 233, 265 for the wave equation, 305
Aharonov-Bohm effect, 297 Cauchy's
angle coordinates, 247 integral formula, 38
angle-preserving map, 108, 181 theorem, 37
angular momentum, 268 Cauchy-Riemann equations, 37
area-preserving map, 182 Cayley transformation, 227
Arnold-Liouville theorem, 242 center, 223
atlas, 48 central force, 267
chart, 48
barometric formula, 282 chart transition, 48
Bianchi identity Christoffel symbols, 174, 185
first, 186 Clairaut's theorem, 178, 262
second, 191 closed, 18
binormal vector, 132 coadjoint representation, 230
Biot-Savart law, 303 coclosed, 102
Birkhoff's ergodicity theorem, 235 Codazzi-Mainardi equation, 153
black-body radiation, 293 commutator, 91
Bohr-Sommerfeld condition, 248 completely integrable system, 244
Boltzmann distribution, 281 configuration space, 253
Boltzmnann's constant, 285 conformal group, 315
boundary conformal map, 108, 181
of a chain, 27 conjugation action, 222
of a manifold, 53 connection form, 142, 186
Bour's minimal surface, 201 conservation of energy
Brouwer's theorem, 40 for Lagrangian systems, 261
for Newtonian systems, 256
canonical in statistical mechanics, 273
coordinates, 236 conservation of information entropy, 277
339
340 Index
continuity equation, 296 for Newtonian systems, 255

Coulomb gauge, 326 free, 279, 283
Coulomb potential, 302 inner, 279, 283
covariant derivative kinetic, 254
on manifolds, 184, 188, 190 of a statistical state, 273
on surfaces, 149, 153 of a thermodynamic system, 284
cubic homology group, 30 energy density, 316
curl, 25 Enneper's minimal surface, 201
current density vector, 295 entropy, inner, 279, 283, 284
curvature equation of state, 279
form, 204 general, 283
Gaussian, 156 equidistribution, 276
geodesic, 173 equilibrium state, 272
lines, 201 Erlanger Programm, 207
mean, 156 Enter characteristic, 165
normal, 173 Enter equations, 267
of a curve, 132, 198 Euler-Lagrange equations, 258
principal, 167 exact, 18
scalar, 193 exponential map, 192, 212
sectional, 193 exterior
tensor, 151, 186 algebra, 3
cylinder, 158, 168, 176, 183 derivative, 15, 68
form, I
d'Alembert-Lagrange theorem, 258 normal vector field, 74
d'Alembertian, 101 product, 2
Daniell-Stone functional, 79
Darboux frame, 172 Fenchel inequality, 137, 141
Darboux theorem, 236 field
Darboux vector, 199 electric, 295
de Rham cohomology, 19 magnetic, 295
density form, 312 field strength form, 312
density function, 272 first Bianchi identity, 186
of an electric charge, 295 first fundamental form, 148
diffeomorphism group, 88 first integral, 177, 236, 265
differential form, 13, 67 fixed point, 38
closed, 18 fixed point property, 38
enclosed, 102 flow (of a vector field), 88
exact, 1S form
harmonic, 102 differential, 13
left-invariant, 210 exterior, 1
dimension, 49, 131 Frenet formulas, 133
Dirac monopole, 321 Frenet frame, 132
Dirichlet problem, 95 Frobenius' theorem, 113
distribution, 112 fundamental theorem
integrable, 112 of curve theory, 133
involutive, 113 of surface theory, 143, 153
divergence, 24, 62 fundamental theorems of
geometric interpretation, 91 thermodynamics, 284
dual 1-form (of a vector field), 23, 72
Dulong-Petit rule, 289 Calerkin's method, 41
y-factor, 319
effective potential, 267 gas
eikonal equation, 311 ideal, 288
Einstein equation, 204 real, 293
Einstein space, 194 Gauss equation, 153
energy, 267 Gauss'
for Lagrangian systems, 260 mean value theorem, 99
Index 341
theorem, 83 of a vector field, 166

Gauss-Bonnet formula, 164, 179 induced
Gaussian curvature, 156 differential form, 14, 67
Gay-Lussac experiment, 292 exterior form, 4
geodesic information, 276
curvature, 173 information entropy
line, 174, 192 of a probability measure, 276
spray, 254 of a statistical state, 277
Gibbs state, 280 inner entropy, 279, 283, 284
Gibbs' fundamental equation, 284 inner product, 4, 68
GL(n,K), 208 integrability condition, 114
gradient, 23. 60 integral curve, 87
symplectic, 233 integral manifold, 112
graph, 49, 145, 148, 159 integrating factor, 120
Green's formula irreversible process, 277
first, 35, 84 isometry, 180
second, 36, 84 isotherm, 290
group of diffeomorphisms, 88 isothermic coordinates, 170
isotropy group, 231
Hadamard. J., 330
hairy sphere theorem, 82 Jacobi, 200
half-space, 52
Jacobi identity, 93, 235
Hamilton equations, 234
Hamilton function, 263
Hamilton's theorem, 263 Kerr metric, 196
Hamiltonian quaternions, 221, 226 Kirchhoff formula, 325
Hamiltonian system, 236 Kirillov form, 232
harmonic differential form, 102 Klein, Felix, 207
harmonic function, 94
Liouville's theorem, 99 Lagrange function, 257
maximum principle, 99 Lagrange's theorem, 4
mean value theorem, 99 Lagrangian system, 258
Poisson formula, 99 hyper-regular, 262
heat form, 284 Lambert projection, 183
hedgehog theorem, 82 Lancret's theorem, 134
helicoid, 172, 201 Laplace-Beltrami operator, 65
helix, 131, 176, 198, 266 Laplacian, 24, 65
Helmholtz's theorem, 104, 304 Lebesgue measure, 79
Hessian form, 196 left-invariant, 209, 210
Hilbert, David, 158 Legendre transformation, 260
Hodge operator, 7, 23 Levi-Civita connection, 184
Hodge's theorem, 104, 304 Lie algebra, 209
Hodge-Laplace operator, 101 Lie derivative, 89
homogeneous space, 220 Lie group, 208
homotopy, 33, 81 Lie, Sophus, 207
Hopf's theorem, 85 line integral, 30
Hopf-Poincare theorem, 166 Liouville form, 230, 265
Huygens' principle, 305, 310 Liouville's equation, 273
hyperbolic plane, 148, 158, 179, 190, 202, Liouville's theorem
227 for harmonic functions, 99
for symplectic manifolds, 234
ideal gas, 288 Lienard, 309
Igel, Satz vom", 82 Lorentz equation, 266
imaginary part surface, 147, 161 classical, 320
index generalized, 317
of a curve, 199 Lorentz force, 259, 260, 262
of a scalar product, 5 Lorentz group, 313
342 Index
manifold Pfaffian system, 112

flat, 188 phase space, 253
orientable, 69 Planck's radiation law, 293
simply connected, 80 Poincar4's lemma, 20, 81
symplectic, 229 Poincar6's return theorem, 234
with boundary, 52 point
without boundary, 47 elliptic, 167
Maupertuis-Jacobi principle, 257 flat, 167
Maurer-Cartan equations, 210 hyperbolic, 167
Maurer-Cartan form, 210 parabolic, 167
maximum principle, 99 umbilic, 167
Maxwell distribution, 281, 292 Poisson bracket, 235
Maxwell equations Poisson formula
in classical formulation, 295 for harmonic functions, 99
relativistic version, 312 for the wave equation, 305
Maxwell relations, 290 polar coordinates, 57
Maxwell stress tensor, 315 potential
mean curvature, 156 chemical, 286
Mercator projection, 182 electric, 297
microcanonical ensemble, 282 magnetic, 296
minimal surface, 169 Newton's, 299
Bour's, 201 retarded, 308
Enneper's, 201 thermodynamic, 289
Minkowski space, 101, 311 potential energy, 255
Minkowski-Steiner theorem, 167 Poynting vector, 316
modular surface, 146, 159, 201 pressure, 286
module, 222 principal curvature, 167
Mobius strip, 51 principal normal vector, 132
moment map, 239, 268 principle of least action, 259
momentum, 267 probability current, 275
motion, 254 proper time, 319
pseudo-Riemannian metric, 100
natural parametrization, 131 pseudosphere, 107, 158
Neumann problem, 95 pullback
Newton potential, 299 of a differential form, 14, 67
Newton's equation, 254 of an exterior form, 4
Newtonian system, 254
with potential energy, 255 quaternions, 221, 226
Noether's theorem, 177, 238, 261
normal curvature vector, 173 Rayleigh-Jeans law, 294
normal vector field, 74 real gas, 293
real part surface, 147, 161
O(n, R), 106, 221 representation
orientation adjoint, 230
induced, 75 coadjoint, 230, 265
of a manifold, 69 defining, 222
of a vector space, 6 irreducible, 227
osculating plane, 134 of a Lie algebra, 221
Ostrogradski formula, 83 of a Lie group, 221
trivial, 222
parallel displacement, 189 retarded potential, 308
parallel vector field, 188 retarded solutions, 309
parametrized curve, 131 reversible process, 277
partition function, 278, 283 Ricci tensor, 193
partition of unity, 76 Riemann surface, 106
Peano curve, 129 Riemann, Bernhard, 197
perihelion precession, 252 Riemannian curvature tensor, 186
Index 343
Riemannian metric, 59 tangent space

of Rn, 11
scalar curvature, 193 of a manifold, 54
scalar product, 4 temperature, 284, 285
Schrudinger equation, 297 tensor product (of representations), 227
Schwarzschild-Eddington metric, 196, 204 Theorems Egregium, 142, 163
second Bianchi identity, 191 thermal efficiency, 291
second fundamental form, 150 thermodynamic relations, 290
sectional curvature, 193 Toda lattice, 266
Shannon, C. E., 276 torsion, 133, 198
signature (of a scalar product), 5 torus, 50
simply connected, 80 total differential equation, 121
singular tractrix, 107, 158
chain. 27 transitive action, 220
cube, 26
SL(2, R), 106, 211 U(n), 221
SL(n, K), 221 unit cube, 26
slope line, 134 unit normal vector field, 74
smooth map, 26, 34, 39
solid body, 289 van der Waals equation, 293
SO(n, R), 106, 221 vector field, 12, 56
space form, 196 complete, 88
space of constant sectional curvature, 196 components of, 57
spherical coordinates, 59 flow of, 88
spherical mean, 305 fundamental, 231
spherical pendulum, 249 left-invariant, 209
standard cube, 27 normal, 74
star-shaped set, 19 parallel, 188
stationary terminal distribution, 272 related, 93
statistical state, 271 velocity of light, 296
stereographic projection, 107, 181 volume, 78
Stokes' theorem, 32, 79 volume form, 6, 70, 229
classical version, 36, 85
structural equations wave (electromagnetic), 304
of a Lie group, 210 wave equation, 304
of a manifold, 187 Cauchy problem for the, 305
of a surface, 133, 142 Poisson formula for the, 305
SU(1,1), 227 wave function, 297
SU(n), 221 wave operator, 101, 306
support, 76 Weierstrass representation, 171
surface of revolution, 144, 148, 158, 177, 185, Weingarten map, 152, 156
200 Wiechert, 309
Sylvester's theorem, 4 Wien's law, 294
symmetric product, 148 winding form, 18
symplectic winding number, 137, 199
coordinates, 236 work form, 284
diffeomorphism, 238
form, 229 Yang-Mills equation, 316
gradient, 233
manifold, 229
structure, 229
volume form, 229
synchrotron frequency, 321
synchrotron radiation, 326
system, completely integrable, 244
tangent bundle, 55
From a Review of the German Edition:
Drawing on his great experience in research, writing books, teaching, and working with
students, Friedrich presents once more a clearly written, smoothly readable self
contained textbook The mathematical material and approaches are well motivated,
enriched by valuable considerations and reflections. Proofs are elegant, not too tech-
nical and carefully performed ... Each chapter finishes with exercises designed to
increase comprehension ... For any student who has passed the linear algebra course
and calculus, this book offers on excellent opportunity to learn global analysis and its
applications to mathematical physics.
-Mathematical Reviews
This book introduces the reader to the world of differential forms and their
uses in geometry, analysis. and mathematical physics. It begins with a few basic
topics, partly as review, then moves on to vector analysis on manifolds and the
study of curves and surfaces in 3-space. Lie groups and homogeneous spaces are
discussed, providing the appropriate framework for introducing symmetry in
both mathematical and physical contexts.The final third of the book applies the
mathematical ideas to important areas of physics: Hamiltonian mechanics, statis-
tical mechanics, and electrodynamics.
There are many classroom-tested exercises and examples with excellent figures
throughoutThe book is ideal as a text for a first course in differential geometry,
suitable for advanced undergraduates or graduate students in mathematics or
physics.
\\1ti !.r !h: \\cII

www.ams.org

Ilka Agricola, Thomas Friedrich Global Analysis Differential Forms in Analysis, Geometry, and Physics Graduate Studies in Mathematics, V. 52 PDF

Uploaded by

Copyright:

Available Formats

Ilka Agricola, Thomas Friedrich Global Analysis Differential Forms in Analysis, Geometry, and Physics Graduate Studies in Mathematics, V. 52 PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Ilka Agricola, Thomas Friedrich Global Analysis Differential Forms in Analysis, Geometry, and Physics Graduate Studies in Mathematics, V. 52 PDF

Uploaded by

Copyright:

Available Formats

What topics does the book cover?

What topics does the book cover?

What mathematical concepts are discussed?

What mathematical concepts are discussed?

Global Analysis

Differential Forms in Analysis,

American Mathematical Society

American Mathematical Society

2000 Mathematics Subject Classification. Primary 53-01;

Library of Congress Cataloging-in-Publication Data

We only suppose as much knowledge of mathematics as the reader would ac-

systems, the differential geometry of curves and surfaces in euclidean space,

In Chapter 2 we develop the differential and integral calculus for differential

In Chapter 3 we restrict the possible integration domains to "smooth"

Apart from Stokes' theorem, the integrability criterion of Frobenius is one of

Chapter 5 treats the differential geometry of curves and surfaces in euclidean

Symmetries play a fundamental role in geometry and physics. Chapter 6

Apart from Riemannian geometry, symplectic geometry is one of the es-

In statistical mechanics, particles are described by their position probability

Chapter 9 is devoted to electrodynamics. Starting from the Maxwell equa-

Berlin, November 2000 and May 2002

Chapter 1. Elements of Multilinear Algebra 1

Chapter 2. Differential Forms in R" 11

§2.1. Vector Fields and Differential Forms 11

§2.2. Closed and Exact Differential Forms 18

Chapter 3. Vector Analysis on Manifolds 47

§3.8. The Classical Integral Formulas 82

Chapter 4. Pfaffian Systems 111

Chapter 5. Curves and Surfaces in Euclidean 3-Space 129

Chapter 6. Lie Groups and Homogeneous Spaces 207

Chapter 7. Symplectic Geometry and Mechanics 229

Chapter 8. Elements of Statistical Mechanics and Thermodynam: .es 271

§8.2. Thermodynamical Systems in Equilibrium 283

Chapter 9. Elements of Electrodynamics 295

Consider an n-dimensional vector space V over the field K of real or complex

wk(VQ(1), ....Va(k)) = sgn(a) i.Jk(VI, -..,Vk).

Fixing an arbitrary basis e1, ... , e in the n-dimensional vector space V, we

multilinear way. The number of different k-tuples of n elements is equal to

Obviously. wk n,1 is a multilinear and antisymmetric map acting on (k + 1)

= sgn(o)w(va(1), ... va(k))1T(t'a(k+1) ... V0(k+l))1-(Va(k+1+1), - - -Va(k+l+m))-

E sgn(a)w(va(1), ... Vo(k))17(Vo(k+1), -- Vo(k+l))1(Va(k+1+1), . Vo(k+l+m))-

L' (wk n,/) = (L*wk) A (L5,1)

Apart from a scalar product g, we fix an orientation in the real vector

We define an equivalence relation N on the set B(V) by requiring

holds for all (n - k)-forms 17n-k. Summarizing we defined a linear mapping

and immediately imply the formula

*oj = (-1)Isgn (Il...Jn gijiI ...gin-yin-4a

Proof. To prove the first formula we first compute

1. Let a,, ..., an be a basis of V*, let wl = E aj o;, ril = F_ bi a; be two

2. Prove that each 2-form w2 E A2(V*) can be represented as

4 (Cartan's Lemma). Let o1i ... , ok be linearly independent linear forms

Moreover, ail = aji.

6. For a given form 0 96 wk E Ak(V*), define the subspace M(Wk) C V' by

Prove the following statements.

7. Prove the following statements.

8. The Hodge operator of a 4-dimensional vector space V maps the space

is defined by the formula

0 (P) + ... + V" (p) 81" (P)

The following figure depicts this vector field:

d(fwk) = .EaW°(f(x))dfoAdf'1 A...Adf'k = f(dwk).